[PATCH v11 00/26] Speculative page faults

[PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
page fault without holding the mm semaphore [1].

The idea is to try to handle user space page faults without holding the
mmap_sem. This should allow better concurrency for massively threaded
process since the page fault handler will not wait for other threads memory
layout change to be done, assuming that this change is done in another part
of the process's memory space. This type page fault is named speculative
page fault. If the speculative page fault fails because of a concurrency is
detected or because underlying PMD or PTE tables are not yet allocating, it
is failing its processing and a classic page fault is then tried.

The speculative page fault (SPF) has to look for the VMA matching the fault
address without holding the mmap_sem, this is done by introducing a rwlock
which protects the access to the mm_rb tree. Previously this was done using
SRCU but it was introducing a lot of scheduling to process the VMA's
freeing operation which was hitting the performance by 20% as reported by
Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
limiting the locking contention to these operations which are expected to
be in a O(log n) order. In addition to ensure that the VMA is not freed in
our back a reference count is added and 2 services (get_vma() and
put_vma()) are introduced to handle the reference count. Once a VMA is
fetched from the RB tree using get_vma(), it must be later freed using
put_vma(). I can't see anymore the overhead I got while will-it-scale
benchmark anymore.

The VMA's attributes checked during the speculative page fault processing
have to be protected against parallel changes. This is done by using a per
VMA sequence lock. This sequence lock allows the speculative page fault
handler to fast check for parallel changes in progress and to abort the
speculative page fault in that case.

Once the VMA has been found, the speculative page fault handler would check
for the VMA's attributes to verify that the page fault has to be handled
correctly or not. Thus, the VMA is protected through a sequence lock which
allows fast detection of concurrent VMA changes. If such a change is
detected, the speculative page fault is aborted and a *classic* page fault
is tried.  VMA sequence lockings are added when VMA attributes which are
checked during the page fault are modified.

When the PTE is fetched, the VMA is checked to see if it has been changed,
so once the page table is locked, the VMA is valid, so any other changes
leading to touching this PTE will need to lock the page table, so no
parallel change is possible at this time.

The locking of the PTE is done with interrupts disabled, this allows
checking for the PMD to ensure that there is not an ongoing collapsing
operation. Since khugepaged is firstly set the PMD to pmd_none and then is
waiting for the other CPU to have caught the IPI interrupt, if the pmd is
valid at the time the PTE is locked, we have the guarantee that the
collapsing operation will have to wait on the PTE lock to move forward.
This allows the SPF handler to map the PTE safely. If the PMD value is
different from the one recorded at the beginning of the SPF operation, the
classic page fault handler will be called to handle the operation while
holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
the lock is done using spin_trylock() to avoid dead lock when handling a
page fault while a TLB invalidate is requested by another CPU holding the
PTE.

In pseudo code, this could be seen as:
    speculative_page_fault()
    {
	    vma = get_vma()
	    check vma sequence count
	    check vma's support
	    disable interrupt
		  check pgd,p4d,...,pte
		  save pmd and pte in vmf
		  save vma sequence counter in vmf
	    enable interrupt
	    check vma sequence count
	    handle_pte_fault(vma)
		    ..
		    page = alloc_page()
		    pte_map_lock()
			    disable interrupt
				    abort if sequence counter has changed
				    abort if pmd or pte has changed
				    pte map and lock
			    enable interrupt
		    if abort
		       free page
		       abort
		    ...
    }
    
    arch_fault_handler()
    {
	    if (speculative_page_fault(&vma))
	       goto done
    again:
	    lock(mmap_sem)
	    vma = find_vma();
	    handle_pte_fault(vma);
	    if retry
	       unlock(mmap_sem)
	       goto again;
    done:
	    handle fault error
    }

Support for THP is not done because when checking for the PMD, we can be
confused by an in progress collapsing operation done by khugepaged. The
issue is that pmd_none() could be true either if the PMD is not already
populated or if the underlying PTE are in the way to be collapsed. So we
cannot safely allocate a PMD if pmd_none() is true.

This series add a new software performance event named 'speculative-faults'
or 'spf'. It counts the number of successful page fault event handled
speculatively. When recording 'faults,spf' events, the faults one is
counting the total number of page fault events while 'spf' is only counting
the part of the faults processed speculatively.

There are some trace events introduced by this series. They allow
identifying why the page faults were not processed speculatively. This
doesn't take in account the faults generated by a monothreaded process
which directly processed while holding the mmap_sem. This trace events are
grouped in a system named 'pagefault', they are:
 - pagefault:spf_vma_changed : if the VMA has been changed in our back
 - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
 - pagefault:spf_vma_notsup : the VMA's type is not supported
 - pagefault:spf_vma_access : the VMA's access right are not respected
 - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
   back.

To record all the related events, the easier is to run perf with the
following arguments : 
$ perf stat -e 'faults,spf,pagefault:*' <command>

There is also a dedicated vmstat counter showing the number of successful
page fault handled speculatively. I can be seen this way: 
$ grep speculative_pgfault /proc/vmstat

This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
on x86, PowerPC and arm64.

---------------------
Real Workload results

As mentioned in previous email, we did non official runs using a "popular
in memory multithreaded database product" on 176 cores SMT8 Power system
which showed a 30% improvements in the number of transaction processed per
second. This run has been done on the v6 series, but changes introduced in
this new version should not impact the performance boost seen.

Here are the perf data captured during 2 of these runs on top of the v8
series:
		vanilla		spf
faults		89.418		101.364		+13%	
spf                n/a		 97.989

With the SPF kernel, most of the page fault were processed in a speculative
way.

Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
it a try on an android device. He reported that the application launch time
was improved in average by 6%, and for large applications (~100 threads) by
20%.

Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
MSM845 (8 cores) with 6GB (the less is better):

Application				4.9	4.9+spf	delta
com.tencent.mm 				416	389 	-7%
com.eg.android.AlipayGphone 		1135 	986 	-13%
com.tencent.mtt 			455	454	0%
com.qqgame.hlddz 			1497	1409	-6%
com.autonavi.minimap 			711	701	-1%
com.tencent.tmgp.sgame 			788	748	-5%
com.immomo.momo 			501	487	-3%
com.tencent.peng 			2145	2112	-2%
com.smile.gifmaker 			491	461	-6%
com.baidu.BaiduMap 			479	366	-23%
com.taobao.taobao 			1341	1198	-11%
com.baidu.searchbox 			333	314 	-6%
com.tencent.mobileqq 			394	384	-3%
com.sina.weibo 				907	906	0%
com.youku.phone 			816	731	-11%
com.happyelements.AndroidAnimal.qq 	763	717	-6%
com.UCMobile 				415	411	-1%
com.tencent.tmgp.ak			1464	1431	-2%
com.tencent.qqmusic			336	329	-2%
com.sankuai.meituan			1661	1302	-22%
com.netease.cloudmusic			1193	1200	1%
air.tv.douyu.android			4257	4152	-2%

------------------
Benchmarks results

Base kernel is v4.17.0-rc4-mm1
SPF is BASE + this series

Kernbench:
----------
Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
kernel (kernel is build 5 times):

Average	Half load -j 8
		 Run	(std deviation)
		 BASE			SPF
Elapsed	Time	 1448.65 (5.72312)	1455.84	(4.84951)	0.50%
User	Time	 10135.4 (30.3699)	10148.8	(31.1252)	0.13%
System	Time	 900.47	 (2.81131)	923.28	(7.52779)	2.53%
Percent	CPU	 761.4	 (1.14018)	760.2	(0.447214)	-0.16%
Context	Switches 85380	 (3419.52)	84748	(1904.44)	-0.74%
Sleeps		 105064	 (1240.96)	105074	(337.612)	0.01%
						
Average	Optimal	load -j	16
		 Run	(std deviation)
		 BASE			SPF
Elapsed	Time	 920.528 (10.1212)	927.404	(8.91789)	0.75%
User	Time	 11064.8 (981.142)	11085	(990.897)	0.18%
System	Time	 979.904 (84.0615)	1001.14	(82.5523)	2.17%
Percent	CPU	 1089.5	 (345.894)	1086.1	(343.545)	-0.31%
Context	Switches 159488	 (78156.4)	158223	(77472.1)	-0.79%
Sleeps		 110566	 (5877.49)	110388	(5617.75)	-0.16%


During a run on the SPF, perf events were captured:
 Performance counter stats for '../kernbench -M':
         526743764      faults                   
               210      spf                      
                 3      pagefault:spf_vma_changed
                 0      pagefault:spf_vma_noanon 
              2278      pagefault:spf_vma_notsup 
                 0      pagefault:spf_vma_access 
                 0      pagefault:spf_pmd_changed

Very few speculative page faults were recorded as most of the processes
involved are monothreaded (sounds that on this architecture some threads
were created during the kernel build processing).

Here are the kerbench results on a 80 CPUs Power8 system:

Average	Half load -j 40
		 Run	(std deviation)
		 BASE			SPF
Elapsed	Time	 117.152 (0.774642)	117.166	(0.476057)	0.01%
User	Time	 4478.52 (24.7688)	4479.76	(9.08555)	0.03%
System	Time	 131.104 (0.720056)	134.04	(0.708414)	2.24%
Percent	CPU	 3934	 (19.7104)	3937.2	(19.0184)	0.08%
Context	Switches 92125.4 (576.787)	92581.6	(198.622)	0.50%
Sleeps		 317923	 (652.499)	318469	(1255.59)	0.17%
						
Average	Optimal	load -j	80
		 Run	(std deviation)
		 BASE			SPF
Elapsed	Time	 107.73	 (0.632416)	107.31	(0.584936)	-0.39%
User	Time	 5869.86 (1466.72)	5871.71	(1467.27)	0.03%
System	Time	 153.728 (23.8573)	157.153	(24.3704)	2.23%
Percent	CPU	 5418.6	 (1565.17)	5436.7	(1580.91)	0.33%
Context	Switches 223861	 (138865)	225032	(139632)	0.52%
Sleeps		 330529	 (13495.1)	332001	(14746.2)	0.45%

During a run on the SPF, perf events were captured:
 Performance counter stats for '../kernbench -M':
         116730856      faults
                 0      spf
                 3      pagefault:spf_vma_changed
                 0      pagefault:spf_vma_noanon
               476      pagefault:spf_vma_notsup
                 0      pagefault:spf_vma_access
                 0      pagefault:spf_pmd_changed

Most of the processes involved are monothreaded so SPF is not activated but
there is no impact on the performance.

Ebizzy:
-------
The test is counting the number of records per second it can manage, the
higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
consistent result I repeated the test 100 times and measure the average
result. The number is the record processes per second, the higher is the
best.

  		BASE		SPF		delta	
16 CPUs x86 VM	742.57		1490.24		100.69%
80 CPUs P8 node 13105.4		24174.23	84.46%

Here are the performance counter read during a run on a 16 CPUs x86 VM:
 Performance counter stats for './ebizzy -mTt 16':
           1706379      faults
           1674599      spf
             30588      pagefault:spf_vma_changed
                 0      pagefault:spf_vma_noanon
               363      pagefault:spf_vma_notsup
                 0      pagefault:spf_vma_access 
                 0      pagefault:spf_pmd_changed

And the ones captured during a run on a 80 CPUs Power node:
 Performance counter stats for './ebizzy -mTt 80':
           1874773      faults
           1461153      spf
            413293      pagefault:spf_vma_changed
                 0      pagefault:spf_vma_noanon
               200      pagefault:spf_vma_notsup
                 0      pagefault:spf_vma_access
                 0      pagefault:spf_pmd_changed 

In ebizzy's case most of the page fault were handled in a speculative way,
leading the ebizzy performance boost.

------------------
Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
 - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
   and Minchan Kim, hopefully.
 - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
   __do_page_fault().
 - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
   instead
   of aborting the speculative page fault handling. Dropping the now
useless
   trace event pagefault:spf_pte_lock.
 - No more try to reuse the fetched VMA during the speculative page fault
   handling when retrying is needed. This adds a lot of complexity and
   additional tests done didn't show a significant performance improvement.
 - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.

[1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
[2] https://patchwork.kernel.org/patch/9999687/


Laurent Dufour (20):
  mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
  x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
  powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
  mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
  mm: make pte_unmap_same compatible with SPF
  mm: introduce INIT_VMA()
  mm: protect VMA modifications using VMA sequence count
  mm: protect mremap() against SPF hanlder
  mm: protect SPF handler against anon_vma changes
  mm: cache some VMA fields in the vm_fault structure
  mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
  mm: introduce __lru_cache_add_active_or_unevictable
  mm: introduce __vm_normal_page()
  mm: introduce __page_add_new_anon_rmap()
  mm: protect mm_rb tree with a rwlock
  mm: adding speculative page fault failure trace events
  perf: add a speculative page fault sw event
  perf tools: add support for the SPF perf event
  mm: add speculative page fault vmstats
  powerpc/mm: add speculative page fault

Mahendran Ganesh (2):
  arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
  arm64/mm: add speculative page fault

Peter Zijlstra (4):
  mm: prepare for FAULT_FLAG_SPECULATIVE
  mm: VMA sequence count
  mm: provide speculative fault infrastructure
  x86/mm: add speculative pagefault handling

 arch/arm64/Kconfig                    |   1 +
 arch/arm64/mm/fault.c                 |  12 +
 arch/powerpc/Kconfig                  |   1 +
 arch/powerpc/mm/fault.c               |  16 +
 arch/x86/Kconfig                      |   1 +
 arch/x86/mm/fault.c                   |  27 +-
 fs/exec.c                             |   2 +-
 fs/proc/task_mmu.c                    |   5 +-
 fs/userfaultfd.c                      |  17 +-
 include/linux/hugetlb_inline.h        |   2 +-
 include/linux/migrate.h               |   4 +-
 include/linux/mm.h                    | 136 +++++++-
 include/linux/mm_types.h              |   7 +
 include/linux/pagemap.h               |   4 +-
 include/linux/rmap.h                  |  12 +-
 include/linux/swap.h                  |  10 +-
 include/linux/vm_event_item.h         |   3 +
 include/trace/events/pagefault.h      |  80 +++++
 include/uapi/linux/perf_event.h       |   1 +
 kernel/fork.c                         |   5 +-
 mm/Kconfig                            |  22 ++
 mm/huge_memory.c                      |   6 +-
 mm/hugetlb.c                          |   2 +
 mm/init-mm.c                          |   3 +
 mm/internal.h                         |  20 ++
 mm/khugepaged.c                       |   5 +
 mm/madvise.c                          |   6 +-
 mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
 mm/mempolicy.c                        |  51 ++-
 mm/migrate.c                          |   6 +-
 mm/mlock.c                            |  13 +-
 mm/mmap.c                             | 229 ++++++++++---
 mm/mprotect.c                         |   4 +-
 mm/mremap.c                           |  13 +
 mm/nommu.c                            |   2 +-
 mm/rmap.c                             |   5 +-
 mm/swap.c                             |   6 +-
 mm/swap_state.c                       |   8 +-
 mm/vmstat.c                           |   5 +-
 tools/include/uapi/linux/perf_event.h |   1 +
 tools/perf/util/evsel.c               |   1 +
 tools/perf/util/parse-events.c        |   4 +
 tools/perf/util/parse-events.l        |   1 +
 tools/perf/util/python.c              |   1 +
 44 files changed, 1161 insertions(+), 211 deletions(-)
 create mode 100644 include/trace/events/pagefault.h

-- 
2.7.4

[PATCH v11 01/26] mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT

[Laurent Dufour]

This configuration variable will be used to build the code needed to
handle speculative page fault.

By default it is turned off, and activated depending on architecture
support, ARCH_HAS_PTE_SPECIAL, SMP and MMU.

The architecture support is needed since the speculative page fault handler
is called from the architecture's page faulting code, and some code has to
be added there to handle the speculative handler.

The dependency on ARCH_HAS_PTE_SPECIAL is required because vm_normal_page()
does processing that is not compatible with the speculative handling in the
case ARCH_HAS_PTE_SPECIAL is not set.

Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Suggested-by: David Rientjes <rientjes@google.com>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 mm/Kconfig | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/mm/Kconfig b/mm/Kconfig
index 1d0888c5b97a..a38796276113 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -761,3 +761,25 @@ config GUP_BENCHMARK
 
 config ARCH_HAS_PTE_SPECIAL
 	bool
+
+config ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
+       def_bool n
+
+config SPECULATIVE_PAGE_FAULT
+       bool "Speculative page faults"
+       default y
+       depends on ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
+       depends on ARCH_HAS_PTE_SPECIAL && MMU && SMP
+       help
+         Try to handle user space page faults without holding the mmap_sem.
+
+	 This should allow better concurrency for massively threaded process
+	 since the page fault handler will not wait for other threads memory
+	 layout change to be done, assuming that this change is done in another
+	 part of the process's memory space. This type of page fault is named
+	 speculative page fault.
+
+	 If the speculative page fault fails because of a concurrency is
+	 detected or because underlying PMD or PTE tables are not yet
+	 allocating, it is failing its processing and a classic page fault
+	 is then tried.
-- 
2.7.4

[PATCH v11 02/26] x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT

[Laurent Dufour]

Set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT which turns on the
Speculative Page Fault handler when building for 64bit.

Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 arch/x86/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 47e7f582f86a..603f788a3e83 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -32,6 +32,7 @@ config X86_64
 	select SWIOTLB
 	select X86_DEV_DMA_OPS
 	select ARCH_HAS_SYSCALL_WRAPPER
+	select ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
 
 #
 # Arch settings
-- 
2.7.4

[PATCH v11 03/26] powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT

[Laurent Dufour]

Set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT for BOOK3S_64. This enables
the Speculative Page Fault handler.

Support is only provide for BOOK3S_64 currently because:
- require CONFIG_PPC_STD_MMU because checks done in
  set_access_flags_filter()
- require BOOK3S because we can't support for book3e_hugetlb_preload()
  called by update_mmu_cache()

Cc: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 arch/powerpc/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
index be7aca467692..75f71b963630 100644
--- a/arch/powerpc/Kconfig
+++ b/arch/powerpc/Kconfig
@@ -232,6 +232,7 @@ config PPC
 	select OLD_SIGACTION			if PPC32
 	select OLD_SIGSUSPEND
 	select SPARSE_IRQ
+	select ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT if PPC_BOOK3S_64
 	select SYSCTL_EXCEPTION_TRACE
 	select VIRT_TO_BUS			if !PPC64
 	#
-- 
2.7.4

[PATCH v11 04/26] arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT

[Laurent Dufour]

From: Mahendran Ganesh <opensource.ganesh@gmail.com>

Set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT for arm64. This
enables Speculative Page Fault handler.

Signed-off-by: Ganesh Mahendran <opensource.ganesh@gmail.com>
---
 arch/arm64/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
index 4759566a78cb..c932ae6d2cce 100644
--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig
@@ -147,6 +147,7 @@ config ARM64
 	select SWIOTLB
 	select SYSCTL_EXCEPTION_TRACE
 	select THREAD_INFO_IN_TASK
+	select ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
 	help
 	  ARM 64-bit (AArch64) Linux support.
 
-- 
2.7.4

[PATCH v11 05/26] mm: prepare for FAULT_FLAG_SPECULATIVE

[Laurent Dufour]

From: Peter Zijlstra <peterz@infradead.org>

When speculating faults (without holding mmap_sem) we need to validate
that the vma against which we loaded pages is still valid when we're
ready to install the new PTE.

Therefore, replace the pte_offset_map_lock() calls that (re)take the
PTL with pte_map_lock() which can fail in case we find the VMA changed
since we started the fault.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>

[Port to 4.12 kernel]
[Remove the comment about the fault_env structure which has been
 implemented as the vm_fault structure in the kernel]
[move pte_map_lock()'s definition upper in the file]
[move the define of FAULT_FLAG_SPECULATIVE later in the series]
[review error path in do_swap_page(), do_anonymous_page() and
 wp_page_copy()]
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 mm/memory.c | 87 ++++++++++++++++++++++++++++++++++++++++---------------------
 1 file changed, 58 insertions(+), 29 deletions(-)

diff --git a/mm/memory.c b/mm/memory.c
index 14578158ed20..a55e72c8e469 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2298,6 +2298,13 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
 }
 EXPORT_SYMBOL_GPL(apply_to_page_range);
 
+static inline bool pte_map_lock(struct vm_fault *vmf)
+{
+	vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
+				       vmf->address, &vmf->ptl);
+	return true;
+}
+
 /*
  * handle_pte_fault chooses page fault handler according to an entry which was
  * read non-atomically.  Before making any commitment, on those architectures
@@ -2487,25 +2494,26 @@ static int wp_page_copy(struct vm_fault *vmf)
 	const unsigned long mmun_start = vmf->address & PAGE_MASK;
 	const unsigned long mmun_end = mmun_start + PAGE_SIZE;
 	struct mem_cgroup *memcg;
+	int ret = VM_FAULT_OOM;
 
 	if (unlikely(anon_vma_prepare(vma)))
-		goto oom;
+		goto out;
 
 	if (is_zero_pfn(pte_pfn(vmf->orig_pte))) {
 		new_page = alloc_zeroed_user_highpage_movable(vma,
 							      vmf->address);
 		if (!new_page)
-			goto oom;
+			goto out;
 	} else {
 		new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma,
 				vmf->address);
 		if (!new_page)
-			goto oom;
+			goto out;
 		cow_user_page(new_page, old_page, vmf->address, vma);
 	}
 
 	if (mem_cgroup_try_charge(new_page, mm, GFP_KERNEL, &memcg, false))
-		goto oom_free_new;
+		goto out_free_new;
 
 	__SetPageUptodate(new_page);
 
@@ -2514,7 +2522,10 @@ static int wp_page_copy(struct vm_fault *vmf)
 	/*
 	 * Re-check the pte - we dropped the lock
 	 */
-	vmf->pte = pte_offset_map_lock(mm, vmf->pmd, vmf->address, &vmf->ptl);
+	if (!pte_map_lock(vmf)) {
+		ret = VM_FAULT_RETRY;
+		goto out_uncharge;
+	}
 	if (likely(pte_same(*vmf->pte, vmf->orig_pte))) {
 		if (old_page) {
 			if (!PageAnon(old_page)) {
@@ -2601,12 +2612,14 @@ static int wp_page_copy(struct vm_fault *vmf)
 		put_page(old_page);
 	}
 	return page_copied ? VM_FAULT_WRITE : 0;
-oom_free_new:
+out_uncharge:
+	mem_cgroup_cancel_charge(new_page, memcg, false);
+out_free_new:
 	put_page(new_page);
-oom:
+out:
 	if (old_page)
 		put_page(old_page);
-	return VM_FAULT_OOM;
+	return ret;
 }
 
 /**
@@ -2627,8 +2640,8 @@ static int wp_page_copy(struct vm_fault *vmf)
 int finish_mkwrite_fault(struct vm_fault *vmf)
 {
 	WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED));
-	vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd, vmf->address,
-				       &vmf->ptl);
+	if (!pte_map_lock(vmf))
+		return VM_FAULT_RETRY;
 	/*
 	 * We might have raced with another page fault while we released the
 	 * pte_offset_map_lock.
@@ -2746,8 +2759,11 @@ static int do_wp_page(struct vm_fault *vmf)
 			get_page(vmf->page);
 			pte_unmap_unlock(vmf->pte, vmf->ptl);
 			lock_page(vmf->page);
-			vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
-					vmf->address, &vmf->ptl);
+			if (!pte_map_lock(vmf)) {
+				unlock_page(vmf->page);
+				put_page(vmf->page);
+				return VM_FAULT_RETRY;
+			}
 			if (!pte_same(*vmf->pte, vmf->orig_pte)) {
 				unlock_page(vmf->page);
 				pte_unmap_unlock(vmf->pte, vmf->ptl);
@@ -2954,11 +2970,15 @@ int do_swap_page(struct vm_fault *vmf)
 
 		if (!page) {
 			/*
-			 * Back out if somebody else faulted in this pte
-			 * while we released the pte lock.
+			 * Back out if the VMA has changed in our back during
+			 * a speculative page fault or if somebody else
+			 * faulted in this pte while we released the pte lock.
 			 */
-			vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
-					vmf->address, &vmf->ptl);
+			if (!pte_map_lock(vmf)) {
+				delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
+				ret = VM_FAULT_RETRY;
+				goto out;
+			}
 			if (likely(pte_same(*vmf->pte, vmf->orig_pte)))
 				ret = VM_FAULT_OOM;
 			delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
@@ -3011,10 +3031,13 @@ int do_swap_page(struct vm_fault *vmf)
 	}
 
 	/*
-	 * Back out if somebody else already faulted in this pte.
+	 * Back out if the VMA has changed in our back during a speculative
+	 * page fault or if somebody else already faulted in this pte.
 	 */
-	vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
-			&vmf->ptl);
+	if (!pte_map_lock(vmf)) {
+		ret = VM_FAULT_RETRY;
+		goto out_cancel_cgroup;
+	}
 	if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte)))
 		goto out_nomap;
 
@@ -3092,8 +3115,9 @@ int do_swap_page(struct vm_fault *vmf)
 out:
 	return ret;
 out_nomap:
-	mem_cgroup_cancel_charge(page, memcg, false);
 	pte_unmap_unlock(vmf->pte, vmf->ptl);
+out_cancel_cgroup:
+	mem_cgroup_cancel_charge(page, memcg, false);
 out_page:
 	unlock_page(page);
 out_release:
@@ -3144,8 +3168,8 @@ static int do_anonymous_page(struct vm_fault *vmf)
 			!mm_forbids_zeropage(vma->vm_mm)) {
 		entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address),
 						vma->vm_page_prot));
-		vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
-				vmf->address, &vmf->ptl);
+		if (!pte_map_lock(vmf))
+			return VM_FAULT_RETRY;
 		if (!pte_none(*vmf->pte))
 			goto unlock;
 		ret = check_stable_address_space(vma->vm_mm);
@@ -3180,14 +3204,16 @@ static int do_anonymous_page(struct vm_fault *vmf)
 	if (vma->vm_flags & VM_WRITE)
 		entry = pte_mkwrite(pte_mkdirty(entry));
 
-	vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
-			&vmf->ptl);
-	if (!pte_none(*vmf->pte))
+	if (!pte_map_lock(vmf)) {
+		ret = VM_FAULT_RETRY;
 		goto release;
+	}
+	if (!pte_none(*vmf->pte))
+		goto unlock_and_release;
 
 	ret = check_stable_address_space(vma->vm_mm);
 	if (ret)
-		goto release;
+		goto unlock_and_release;
 
 	/* Deliver the page fault to userland, check inside PT lock */
 	if (userfaultfd_missing(vma)) {
@@ -3209,10 +3235,12 @@ static int do_anonymous_page(struct vm_fault *vmf)
 unlock:
 	pte_unmap_unlock(vmf->pte, vmf->ptl);
 	return ret;
+unlock_and_release:
+	pte_unmap_unlock(vmf->pte, vmf->ptl);
 release:
 	mem_cgroup_cancel_charge(page, memcg, false);
 	put_page(page);
-	goto unlock;
+	return ret;
 oom_free_page:
 	put_page(page);
 oom:
@@ -3305,8 +3333,9 @@ static int pte_alloc_one_map(struct vm_fault *vmf)
 	 * pte_none() under vmf->ptl protection when we return to
 	 * alloc_set_pte().
 	 */
-	vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address,
-			&vmf->ptl);
+	if (!pte_map_lock(vmf))
+		return VM_FAULT_RETRY;
+
 	return 0;
 }
 
-- 
2.7.4

[PATCH v11 06/26] mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE

[Laurent Dufour]

When handling page fault without holding the mmap_sem the fetch of the
pte lock pointer and the locking will have to be done while ensuring
that the VMA is not touched in our back.

So move the fetch and locking operations in a dedicated function.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 mm/memory.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/mm/memory.c b/mm/memory.c
index a55e72c8e469..fa0d9493acac 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2298,6 +2298,13 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
 }
 EXPORT_SYMBOL_GPL(apply_to_page_range);
 
+static inline bool pte_spinlock(struct vm_fault *vmf)
+{
+	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
+	spin_lock(vmf->ptl);
+	return true;
+}
+
 static inline bool pte_map_lock(struct vm_fault *vmf)
 {
 	vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
@@ -3814,8 +3821,8 @@ static int do_numa_page(struct vm_fault *vmf)
 	 * validation through pte_unmap_same(). It's of NUMA type but
 	 * the pfn may be screwed if the read is non atomic.
 	 */
-	vmf->ptl = pte_lockptr(vma->vm_mm, vmf->pmd);
-	spin_lock(vmf->ptl);
+	if (!pte_spinlock(vmf))
+		return VM_FAULT_RETRY;
 	if (unlikely(!pte_same(*vmf->pte, vmf->orig_pte))) {
 		pte_unmap_unlock(vmf->pte, vmf->ptl);
 		goto out;
@@ -4008,8 +4015,8 @@ static int handle_pte_fault(struct vm_fault *vmf)
 	if (pte_protnone(vmf->orig_pte) && vma_is_accessible(vmf->vma))
 		return do_numa_page(vmf);
 
-	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
-	spin_lock(vmf->ptl);
+	if (!pte_spinlock(vmf))
+		return VM_FAULT_RETRY;
 	entry = vmf->orig_pte;
 	if (unlikely(!pte_same(*vmf->pte, entry)))
 		goto unlock;
-- 
2.7.4

[PATCH v11 07/26] mm: make pte_unmap_same compatible with SPF

[Laurent Dufour]

pte_unmap_same() is making the assumption that the page table are still
around because the mmap_sem is held.
This is no more the case when running a speculative page fault and
additional check must be made to ensure that the final page table are still
there.

This is now done by calling pte_spinlock() to check for the VMA's
consistency while locking for the page tables.

This is requiring passing a vm_fault structure to pte_unmap_same() which is
containing all the needed parameters.

As pte_spinlock() may fail in the case of a speculative page fault, if the
VMA has been touched in our back, pte_unmap_same() should now return 3
cases :
	1. pte are the same (0)
	2. pte are different (VM_FAULT_PTNOTSAME)
	3. a VMA's changes has been detected (VM_FAULT_RETRY)

The case 2 is handled by the introduction of a new VM_FAULT flag named
VM_FAULT_PTNOTSAME which is then trapped in cow_user_page().
If VM_FAULT_RETRY is returned, it is passed up to the callers to retry the
page fault while holding the mmap_sem.

Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/mm.h |  4 +++-
 mm/memory.c        | 39 ++++++++++++++++++++++++++++-----------
 2 files changed, 31 insertions(+), 12 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 338b8a1afb02..113b572471ca 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1249,6 +1249,7 @@ static inline void clear_page_pfmemalloc(struct page *page)
 #define VM_FAULT_NEEDDSYNC  0x2000	/* ->fault did not modify page tables
 					 * and needs fsync() to complete (for
 					 * synchronous page faults in DAX) */
+#define VM_FAULT_PTNOTSAME 0x4000	/* Page table entries have changed */
 
 #define VM_FAULT_ERROR	(VM_FAULT_OOM | VM_FAULT_SIGBUS | VM_FAULT_SIGSEGV | \
 			 VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE | \
@@ -1267,7 +1268,8 @@ static inline void clear_page_pfmemalloc(struct page *page)
 	{ VM_FAULT_RETRY,		"RETRY" }, \
 	{ VM_FAULT_FALLBACK,		"FALLBACK" }, \
 	{ VM_FAULT_DONE_COW,		"DONE_COW" }, \
-	{ VM_FAULT_NEEDDSYNC,		"NEEDDSYNC" }
+	{ VM_FAULT_NEEDDSYNC,		"NEEDDSYNC" },	\
+	{ VM_FAULT_PTNOTSAME,		"PTNOTSAME" }
 
 /* Encode hstate index for a hwpoisoned large page */
 #define VM_FAULT_SET_HINDEX(x) ((x) << 12)
diff --git a/mm/memory.c b/mm/memory.c
index fa0d9493acac..75163c145c76 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2319,21 +2319,29 @@ static inline bool pte_map_lock(struct vm_fault *vmf)
  * parts, do_swap_page must check under lock before unmapping the pte and
  * proceeding (but do_wp_page is only called after already making such a check;
  * and do_anonymous_page can safely check later on).
+ *
+ * pte_unmap_same() returns:
+ *	0			if the PTE are the same
+ *	VM_FAULT_PTNOTSAME	if the PTE are different
+ *	VM_FAULT_RETRY		if the VMA has changed in our back during
+ *				a speculative page fault handling.
  */
-static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
-				pte_t *page_table, pte_t orig_pte)
+static inline int pte_unmap_same(struct vm_fault *vmf)
 {
-	int same = 1;
+	int ret = 0;
+
 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
 	if (sizeof(pte_t) > sizeof(unsigned long)) {
-		spinlock_t *ptl = pte_lockptr(mm, pmd);
-		spin_lock(ptl);
-		same = pte_same(*page_table, orig_pte);
-		spin_unlock(ptl);
+		if (pte_spinlock(vmf)) {
+			if (!pte_same(*vmf->pte, vmf->orig_pte))
+				ret = VM_FAULT_PTNOTSAME;
+			spin_unlock(vmf->ptl);
+		} else
+			ret = VM_FAULT_RETRY;
 	}
 #endif
-	pte_unmap(page_table);
-	return same;
+	pte_unmap(vmf->pte);
+	return ret;
 }
 
 static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
@@ -2922,10 +2930,19 @@ int do_swap_page(struct vm_fault *vmf)
 	pte_t pte;
 	int locked;
 	int exclusive = 0;
-	int ret = 0;
+	int ret;
 
-	if (!pte_unmap_same(vma->vm_mm, vmf->pmd, vmf->pte, vmf->orig_pte))
+	ret = pte_unmap_same(vmf);
+	if (ret) {
+		/*
+		 * If pte != orig_pte, this means another thread did the
+		 * swap operation in our back.
+		 * So nothing else to do.
+		 */
+		if (ret == VM_FAULT_PTNOTSAME)
+			ret = 0;
 		goto out;
+	}
 
 	entry = pte_to_swp_entry(vmf->orig_pte);
 	if (unlikely(non_swap_entry(entry))) {
-- 
2.7.4

[PATCH v11 08/26] mm: introduce INIT_VMA()

[Laurent Dufour]

Some VMA struct fields need to be initialized once the VMA structure is
allocated.
Currently this only concerns anon_vma_chain field but some other will be
added to support the speculative page fault.

Instead of spreading the initialization calls all over the code, let's
introduce a dedicated inline function.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 fs/exec.c          |  2 +-
 include/linux/mm.h |  5 +++++
 kernel/fork.c      |  2 +-
 mm/mmap.c          | 10 +++++-----
 mm/nommu.c         |  2 +-
 5 files changed, 13 insertions(+), 8 deletions(-)

diff --git a/fs/exec.c b/fs/exec.c
index 6fc98cfd3bdb..7e134a588ef3 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -311,7 +311,7 @@ static int __bprm_mm_init(struct linux_binprm *bprm)
 	vma->vm_start = vma->vm_end - PAGE_SIZE;
 	vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
 	vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
-	INIT_LIST_HEAD(&vma->anon_vma_chain);
+	INIT_VMA(vma);
 
 	err = insert_vm_struct(mm, vma);
 	if (err)
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 113b572471ca..35ecb983ff36 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1303,6 +1303,11 @@ struct zap_details {
 	pgoff_t last_index;			/* Highest page->index to unmap */
 };
 
+static inline void INIT_VMA(struct vm_area_struct *vma)
+{
+	INIT_LIST_HEAD(&vma->anon_vma_chain);
+}
+
 struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
 			     pte_t pte, bool with_public_device);
 #define vm_normal_page(vma, addr, pte) _vm_normal_page(vma, addr, pte, false)
diff --git a/kernel/fork.c b/kernel/fork.c
index 744d6fbba8f8..99198a02efe9 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -458,7 +458,7 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
 		if (!tmp)
 			goto fail_nomem;
 		*tmp = *mpnt;
-		INIT_LIST_HEAD(&tmp->anon_vma_chain);
+		INIT_VMA(tmp);
 		retval = vma_dup_policy(mpnt, tmp);
 		if (retval)
 			goto fail_nomem_policy;
diff --git a/mm/mmap.c b/mm/mmap.c
index d2ef1060a2d2..ceb1c2c1b46b 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -1709,7 +1709,7 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
 	vma->vm_flags = vm_flags;
 	vma->vm_page_prot = vm_get_page_prot(vm_flags);
 	vma->vm_pgoff = pgoff;
-	INIT_LIST_HEAD(&vma->anon_vma_chain);
+	INIT_VMA(vma);
 
 	if (file) {
 		if (vm_flags & VM_DENYWRITE) {
@@ -2595,7 +2595,7 @@ int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
 	/* most fields are the same, copy all, and then fixup */
 	*new = *vma;
 
-	INIT_LIST_HEAD(&new->anon_vma_chain);
+	INIT_VMA(new);
 
 	if (new_below)
 		new->vm_end = addr;
@@ -2965,7 +2965,7 @@ static int do_brk_flags(unsigned long addr, unsigned long request, unsigned long
 		return -ENOMEM;
 	}
 
-	INIT_LIST_HEAD(&vma->anon_vma_chain);
+	INIT_VMA(vma);
 	vma->vm_mm = mm;
 	vma->vm_start = addr;
 	vma->vm_end = addr + len;
@@ -3184,7 +3184,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
 		new_vma->vm_pgoff = pgoff;
 		if (vma_dup_policy(vma, new_vma))
 			goto out_free_vma;
-		INIT_LIST_HEAD(&new_vma->anon_vma_chain);
+		INIT_VMA(new_vma);
 		if (anon_vma_clone(new_vma, vma))
 			goto out_free_mempol;
 		if (new_vma->vm_file)
@@ -3327,7 +3327,7 @@ static struct vm_area_struct *__install_special_mapping(
 	if (unlikely(vma == NULL))
 		return ERR_PTR(-ENOMEM);
 
-	INIT_LIST_HEAD(&vma->anon_vma_chain);
+	INIT_VMA(vma);
 	vma->vm_mm = mm;
 	vma->vm_start = addr;
 	vma->vm_end = addr + len;
diff --git a/mm/nommu.c b/mm/nommu.c
index 4452d8bd9ae4..ece424315cc5 100644
--- a/mm/nommu.c
+++ b/mm/nommu.c
@@ -1212,7 +1212,7 @@ unsigned long do_mmap(struct file *file,
 	region->vm_flags = vm_flags;
 	region->vm_pgoff = pgoff;
 
-	INIT_LIST_HEAD(&vma->anon_vma_chain);
+	INIT_VMA(vma);
 	vma->vm_flags = vm_flags;
 	vma->vm_pgoff = pgoff;
 
-- 
2.7.4

[PATCH v11 09/26] mm: VMA sequence count

[Laurent Dufour]

From: Peter Zijlstra <peterz@infradead.org>

Wrap the VMA modifications (vma_adjust/unmap_page_range) with sequence
counts such that we can easily test if a VMA is changed.

The calls to vm_write_begin/end() in unmap_page_range() are
used to detect when a VMA is being unmap and thus that new page fault
should not be satisfied for this VMA. If the seqcount hasn't changed when
the page table are locked, this means we are safe to satisfy the page
fault.

The flip side is that we cannot distinguish between a vma_adjust() and
the unmap_page_range() -- where with the former we could have
re-checked the vma bounds against the address.

The VMA's sequence counter is also used to detect change to various VMA's
fields used during the page fault handling, such as:
 - vm_start, vm_end
 - vm_pgoff
 - vm_flags, vm_page_prot
 - anon_vma
 - vm_policy

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>

[Port to 4.12 kernel]
[Build depends on CONFIG_SPECULATIVE_PAGE_FAULT]
[Introduce vm_write_* inline function depending on
 CONFIG_SPECULATIVE_PAGE_FAULT]
[Fix lock dependency between mapping->i_mmap_rwsem and vma->vm_sequence by
 using vm_raw_write* functions]
[Fix a lock dependency warning in mmap_region() when entering the error
 path]
[move sequence initialisation INIT_VMA()]
[Review the patch description about unmap_page_range()]
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/mm.h       | 44 ++++++++++++++++++++++++++++++++++++++++++++
 include/linux/mm_types.h |  3 +++
 mm/memory.c              |  2 ++
 mm/mmap.c                | 31 +++++++++++++++++++++++++++++++
 4 files changed, 80 insertions(+)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 35ecb983ff36..18acfdeee759 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1306,6 +1306,9 @@ struct zap_details {
 static inline void INIT_VMA(struct vm_area_struct *vma)
 {
 	INIT_LIST_HEAD(&vma->anon_vma_chain);
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	seqcount_init(&vma->vm_sequence);
+#endif
 }
 
 struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
@@ -1428,6 +1431,47 @@ static inline void unmap_shared_mapping_range(struct address_space *mapping,
 	unmap_mapping_range(mapping, holebegin, holelen, 0);
 }
 
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+static inline void vm_write_begin(struct vm_area_struct *vma)
+{
+	write_seqcount_begin(&vma->vm_sequence);
+}
+static inline void vm_write_begin_nested(struct vm_area_struct *vma,
+					 int subclass)
+{
+	write_seqcount_begin_nested(&vma->vm_sequence, subclass);
+}
+static inline void vm_write_end(struct vm_area_struct *vma)
+{
+	write_seqcount_end(&vma->vm_sequence);
+}
+static inline void vm_raw_write_begin(struct vm_area_struct *vma)
+{
+	raw_write_seqcount_begin(&vma->vm_sequence);
+}
+static inline void vm_raw_write_end(struct vm_area_struct *vma)
+{
+	raw_write_seqcount_end(&vma->vm_sequence);
+}
+#else
+static inline void vm_write_begin(struct vm_area_struct *vma)
+{
+}
+static inline void vm_write_begin_nested(struct vm_area_struct *vma,
+					 int subclass)
+{
+}
+static inline void vm_write_end(struct vm_area_struct *vma)
+{
+}
+static inline void vm_raw_write_begin(struct vm_area_struct *vma)
+{
+}
+static inline void vm_raw_write_end(struct vm_area_struct *vma)
+{
+}
+#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
+
 extern int access_process_vm(struct task_struct *tsk, unsigned long addr,
 		void *buf, int len, unsigned int gup_flags);
 extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index 54f1e05ecf3e..fb5962308183 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -335,6 +335,9 @@ struct vm_area_struct {
 	struct mempolicy *vm_policy;	/* NUMA policy for the VMA */
 #endif
 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	seqcount_t vm_sequence;
+#endif
 } __randomize_layout;
 
 struct core_thread {
diff --git a/mm/memory.c b/mm/memory.c
index 75163c145c76..551a1916da5d 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1499,6 +1499,7 @@ void unmap_page_range(struct mmu_gather *tlb,
 	unsigned long next;
 
 	BUG_ON(addr >= end);
+	vm_write_begin(vma);
 	tlb_start_vma(tlb, vma);
 	pgd = pgd_offset(vma->vm_mm, addr);
 	do {
@@ -1508,6 +1509,7 @@ void unmap_page_range(struct mmu_gather *tlb,
 		next = zap_p4d_range(tlb, vma, pgd, addr, next, details);
 	} while (pgd++, addr = next, addr != end);
 	tlb_end_vma(tlb, vma);
+	vm_write_end(vma);
 }
 
 
diff --git a/mm/mmap.c b/mm/mmap.c
index ceb1c2c1b46b..eeafd0bc8b36 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -701,6 +701,30 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 	long adjust_next = 0;
 	int remove_next = 0;
 
+	/*
+	 * Why using vm_raw_write*() functions here to avoid lockdep's warning ?
+	 *
+	 * Locked is complaining about a theoretical lock dependency, involving
+	 * 3 locks:
+	 *   mapping->i_mmap_rwsem --> vma->vm_sequence --> fs_reclaim
+	 *
+	 * Here are the major path leading to this dependency :
+	 *  1. __vma_adjust() mmap_sem  -> vm_sequence -> i_mmap_rwsem
+	 *  2. move_vmap() mmap_sem -> vm_sequence -> fs_reclaim
+	 *  3. __alloc_pages_nodemask() fs_reclaim -> i_mmap_rwsem
+	 *  4. unmap_mapping_range() i_mmap_rwsem -> vm_sequence
+	 *
+	 * So there is no way to solve this easily, especially because in
+	 * unmap_mapping_range() the i_mmap_rwsem is grab while the impacted
+	 * VMAs are not yet known.
+	 * However, the way the vm_seq is used is guarantying that we will
+	 * never block on it since we just check for its value and never wait
+	 * for it to move, see vma_has_changed() and handle_speculative_fault().
+	 */
+	vm_raw_write_begin(vma);
+	if (next)
+		vm_raw_write_begin(next);
+
 	if (next && !insert) {
 		struct vm_area_struct *exporter = NULL, *importer = NULL;
 
@@ -911,6 +935,7 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 			anon_vma_merge(vma, next);
 		mm->map_count--;
 		mpol_put(vma_policy(next));
+		vm_raw_write_end(next);
 		kmem_cache_free(vm_area_cachep, next);
 		/*
 		 * In mprotect's case 6 (see comments on vma_merge),
@@ -925,6 +950,8 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 			 * "vma->vm_next" gap must be updated.
 			 */
 			next = vma->vm_next;
+			if (next)
+				vm_raw_write_begin(next);
 		} else {
 			/*
 			 * For the scope of the comment "next" and
@@ -971,6 +998,10 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 	if (insert && file)
 		uprobe_mmap(insert);
 
+	if (next && next != vma)
+		vm_raw_write_end(next);
+	vm_raw_write_end(vma);
+
 	validate_mm(mm);
 
 	return 0;
-- 
2.7.4

[PATCH v11 10/26] mm: protect VMA modifications using VMA sequence count

[Laurent Dufour]

The VMA sequence count has been introduced to allow fast detection of
VMA modification when running a page fault handler without holding
the mmap_sem.

This patch provides protection against the VMA modification done in :
	- madvise()
	- mpol_rebind_policy()
	- vma_replace_policy()
	- change_prot_numa()
	- mlock(), munlock()
	- mprotect()
	- mmap_region()
	- collapse_huge_page()
	- userfaultd registering services

In addition, VMA fields which will be read during the speculative fault
path needs to be written using WRITE_ONCE to prevent write to be split
and intermediate values to be pushed to other CPUs.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 fs/proc/task_mmu.c |  5 ++++-
 fs/userfaultfd.c   | 17 +++++++++++++----
 mm/khugepaged.c    |  3 +++
 mm/madvise.c       |  6 +++++-
 mm/mempolicy.c     | 51 ++++++++++++++++++++++++++++++++++-----------------
 mm/mlock.c         | 13 ++++++++-----
 mm/mmap.c          | 22 +++++++++++++---------
 mm/mprotect.c      |  4 +++-
 mm/swap_state.c    |  8 ++++++--
 9 files changed, 89 insertions(+), 40 deletions(-)

diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 597969db9e90..7247d6d5afba 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -1137,8 +1137,11 @@ static ssize_t clear_refs_write(struct file *file, const char __user *buf,
 					goto out_mm;
 				}
 				for (vma = mm->mmap; vma; vma = vma->vm_next) {
-					vma->vm_flags &= ~VM_SOFTDIRTY;
+					vm_write_begin(vma);
+					WRITE_ONCE(vma->vm_flags,
+						   vma->vm_flags & ~VM_SOFTDIRTY);
 					vma_set_page_prot(vma);
+					vm_write_end(vma);
 				}
 				downgrade_write(&mm->mmap_sem);
 				break;
diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index cec550c8468f..b8212ba17695 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -659,8 +659,11 @@ int dup_userfaultfd(struct vm_area_struct *vma, struct list_head *fcs)
 
 	octx = vma->vm_userfaultfd_ctx.ctx;
 	if (!octx || !(octx->features & UFFD_FEATURE_EVENT_FORK)) {
+		vm_write_begin(vma);
 		vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
-		vma->vm_flags &= ~(VM_UFFD_WP | VM_UFFD_MISSING);
+		WRITE_ONCE(vma->vm_flags,
+			   vma->vm_flags & ~(VM_UFFD_WP | VM_UFFD_MISSING));
+		vm_write_end(vma);
 		return 0;
 	}
 
@@ -885,8 +888,10 @@ static int userfaultfd_release(struct inode *inode, struct file *file)
 			vma = prev;
 		else
 			prev = vma;
-		vma->vm_flags = new_flags;
+		vm_write_begin(vma);
+		WRITE_ONCE(vma->vm_flags, new_flags);
 		vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
+		vm_write_end(vma);
 	}
 	up_write(&mm->mmap_sem);
 	mmput(mm);
@@ -1434,8 +1439,10 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx,
 		 * the next vma was merged into the current one and
 		 * the current one has not been updated yet.
 		 */
-		vma->vm_flags = new_flags;
+		vm_write_begin(vma);
+		WRITE_ONCE(vma->vm_flags, new_flags);
 		vma->vm_userfaultfd_ctx.ctx = ctx;
+		vm_write_end(vma);
 
 	skip:
 		prev = vma;
@@ -1592,8 +1599,10 @@ static int userfaultfd_unregister(struct userfaultfd_ctx *ctx,
 		 * the next vma was merged into the current one and
 		 * the current one has not been updated yet.
 		 */
-		vma->vm_flags = new_flags;
+		vm_write_begin(vma);
+		WRITE_ONCE(vma->vm_flags, new_flags);
 		vma->vm_userfaultfd_ctx = NULL_VM_UFFD_CTX;
+		vm_write_end(vma);
 
 	skip:
 		prev = vma;
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index d7b2a4bf8671..0b28af4b950d 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1011,6 +1011,7 @@ static void collapse_huge_page(struct mm_struct *mm,
 	if (mm_find_pmd(mm, address) != pmd)
 		goto out;
 
+	vm_write_begin(vma);
 	anon_vma_lock_write(vma->anon_vma);
 
 	pte = pte_offset_map(pmd, address);
@@ -1046,6 +1047,7 @@ static void collapse_huge_page(struct mm_struct *mm,
 		pmd_populate(mm, pmd, pmd_pgtable(_pmd));
 		spin_unlock(pmd_ptl);
 		anon_vma_unlock_write(vma->anon_vma);
+		vm_write_end(vma);
 		result = SCAN_FAIL;
 		goto out;
 	}
@@ -1080,6 +1082,7 @@ static void collapse_huge_page(struct mm_struct *mm,
 	set_pmd_at(mm, address, pmd, _pmd);
 	update_mmu_cache_pmd(vma, address, pmd);
 	spin_unlock(pmd_ptl);
+	vm_write_end(vma);
 
 	*hpage = NULL;
 
diff --git a/mm/madvise.c b/mm/madvise.c
index 4d3c922ea1a1..e328f7ab5942 100644
--- a/mm/madvise.c
+++ b/mm/madvise.c
@@ -184,7 +184,9 @@ static long madvise_behavior(struct vm_area_struct *vma,
 	/*
 	 * vm_flags is protected by the mmap_sem held in write mode.
 	 */
-	vma->vm_flags = new_flags;
+	vm_write_begin(vma);
+	WRITE_ONCE(vma->vm_flags, new_flags);
+	vm_write_end(vma);
 out:
 	return error;
 }
@@ -450,9 +452,11 @@ static void madvise_free_page_range(struct mmu_gather *tlb,
 		.private = tlb,
 	};
 
+	vm_write_begin(vma);
 	tlb_start_vma(tlb, vma);
 	walk_page_range(addr, end, &free_walk);
 	tlb_end_vma(tlb, vma);
+	vm_write_end(vma);
 }
 
 static int madvise_free_single_vma(struct vm_area_struct *vma,
diff --git a/mm/mempolicy.c b/mm/mempolicy.c
index 9ac49ef17b4e..898d325c9fea 100644
--- a/mm/mempolicy.c
+++ b/mm/mempolicy.c
@@ -380,8 +380,11 @@ void mpol_rebind_mm(struct mm_struct *mm, nodemask_t *new)
 	struct vm_area_struct *vma;
 
 	down_write(&mm->mmap_sem);
-	for (vma = mm->mmap; vma; vma = vma->vm_next)
+	for (vma = mm->mmap; vma; vma = vma->vm_next) {
+		vm_write_begin(vma);
 		mpol_rebind_policy(vma->vm_policy, new);
+		vm_write_end(vma);
+	}
 	up_write(&mm->mmap_sem);
 }
 
@@ -554,9 +557,11 @@ unsigned long change_prot_numa(struct vm_area_struct *vma,
 {
 	int nr_updated;
 
+	vm_write_begin(vma);
 	nr_updated = change_protection(vma, addr, end, PAGE_NONE, 0, 1);
 	if (nr_updated)
 		count_vm_numa_events(NUMA_PTE_UPDATES, nr_updated);
+	vm_write_end(vma);
 
 	return nr_updated;
 }
@@ -657,6 +662,7 @@ static int vma_replace_policy(struct vm_area_struct *vma,
 	if (IS_ERR(new))
 		return PTR_ERR(new);
 
+	vm_write_begin(vma);
 	if (vma->vm_ops && vma->vm_ops->set_policy) {
 		err = vma->vm_ops->set_policy(vma, new);
 		if (err)
@@ -664,11 +670,17 @@ static int vma_replace_policy(struct vm_area_struct *vma,
 	}
 
 	old = vma->vm_policy;
-	vma->vm_policy = new; /* protected by mmap_sem */
+	/*
+	 * The speculative page fault handler accesses this field without
+	 * hodling the mmap_sem.
+	 */
+	WRITE_ONCE(vma->vm_policy,  new);
+	vm_write_end(vma);
 	mpol_put(old);
 
 	return 0;
  err_out:
+	vm_write_end(vma);
 	mpol_put(new);
 	return err;
 }
@@ -1614,23 +1626,28 @@ COMPAT_SYSCALL_DEFINE4(migrate_pages, compat_pid_t, pid,
 struct mempolicy *__get_vma_policy(struct vm_area_struct *vma,
 						unsigned long addr)
 {
-	struct mempolicy *pol = NULL;
+	struct mempolicy *pol;
 
-	if (vma) {
-		if (vma->vm_ops && vma->vm_ops->get_policy) {
-			pol = vma->vm_ops->get_policy(vma, addr);
-		} else if (vma->vm_policy) {
-			pol = vma->vm_policy;
+	if (!vma)
+		return NULL;
 
-			/*
-			 * shmem_alloc_page() passes MPOL_F_SHARED policy with
-			 * a pseudo vma whose vma->vm_ops=NULL. Take a reference
-			 * count on these policies which will be dropped by
-			 * mpol_cond_put() later
-			 */
-			if (mpol_needs_cond_ref(pol))
-				mpol_get(pol);
-		}
+	if (vma->vm_ops && vma->vm_ops->get_policy)
+		return vma->vm_ops->get_policy(vma, addr);
+
+	/*
+	 * This could be called without holding the mmap_sem in the
+	 * speculative page fault handler's path.
+	 */
+	pol = READ_ONCE(vma->vm_policy);
+	if (pol) {
+		/*
+		 * shmem_alloc_page() passes MPOL_F_SHARED policy with
+		 * a pseudo vma whose vma->vm_ops=NULL. Take a reference
+		 * count on these policies which will be dropped by
+		 * mpol_cond_put() later
+		 */
+		if (mpol_needs_cond_ref(pol))
+			mpol_get(pol);
 	}
 
 	return pol;
diff --git a/mm/mlock.c b/mm/mlock.c
index 74e5a6547c3d..c40285c94ced 100644
--- a/mm/mlock.c
+++ b/mm/mlock.c
@@ -445,7 +445,9 @@ static unsigned long __munlock_pagevec_fill(struct pagevec *pvec,
 void munlock_vma_pages_range(struct vm_area_struct *vma,
 			     unsigned long start, unsigned long end)
 {
-	vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
+	vm_write_begin(vma);
+	WRITE_ONCE(vma->vm_flags, vma->vm_flags & VM_LOCKED_CLEAR_MASK);
+	vm_write_end(vma);
 
 	while (start < end) {
 		struct page *page;
@@ -568,10 +570,11 @@ static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev,
 	 * It's okay if try_to_unmap_one unmaps a page just after we
 	 * set VM_LOCKED, populate_vma_page_range will bring it back.
 	 */
-
-	if (lock)
-		vma->vm_flags = newflags;
-	else
+	if (lock) {
+		vm_write_begin(vma);
+		WRITE_ONCE(vma->vm_flags, newflags);
+		vm_write_end(vma);
+	} else
 		munlock_vma_pages_range(vma, start, end);
 
 out:
diff --git a/mm/mmap.c b/mm/mmap.c
index eeafd0bc8b36..add13b4e1d8d 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -852,17 +852,18 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 	}
 
 	if (start != vma->vm_start) {
-		vma->vm_start = start;
+		WRITE_ONCE(vma->vm_start, start);
 		start_changed = true;
 	}
 	if (end != vma->vm_end) {
-		vma->vm_end = end;
+		WRITE_ONCE(vma->vm_end, end);
 		end_changed = true;
 	}
-	vma->vm_pgoff = pgoff;
+	WRITE_ONCE(vma->vm_pgoff, pgoff);
 	if (adjust_next) {
-		next->vm_start += adjust_next << PAGE_SHIFT;
-		next->vm_pgoff += adjust_next;
+		WRITE_ONCE(next->vm_start,
+			   next->vm_start + (adjust_next << PAGE_SHIFT));
+		WRITE_ONCE(next->vm_pgoff, next->vm_pgoff + adjust_next);
 	}
 
 	if (root) {
@@ -1793,13 +1794,15 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
 out:
 	perf_event_mmap(vma);
 
+	vm_write_begin(vma);
 	vm_stat_account(mm, vm_flags, len >> PAGE_SHIFT);
 	if (vm_flags & VM_LOCKED) {
 		if (!((vm_flags & VM_SPECIAL) || is_vm_hugetlb_page(vma) ||
 					vma == get_gate_vma(current->mm)))
 			mm->locked_vm += (len >> PAGE_SHIFT);
 		else
-			vma->vm_flags &= VM_LOCKED_CLEAR_MASK;
+			WRITE_ONCE(vma->vm_flags,
+				   vma->vm_flags & VM_LOCKED_CLEAR_MASK);
 	}
 
 	if (file)
@@ -1812,9 +1815,10 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
 	 * then new mapped in-place (which must be aimed as
 	 * a completely new data area).
 	 */
-	vma->vm_flags |= VM_SOFTDIRTY;
+	WRITE_ONCE(vma->vm_flags, vma->vm_flags | VM_SOFTDIRTY);
 
 	vma_set_page_prot(vma);
+	vm_write_end(vma);
 
 	return addr;
 
@@ -2443,8 +2447,8 @@ int expand_downwards(struct vm_area_struct *vma,
 					mm->locked_vm += grow;
 				vm_stat_account(mm, vma->vm_flags, grow);
 				anon_vma_interval_tree_pre_update_vma(vma);
-				vma->vm_start = address;
-				vma->vm_pgoff -= grow;
+				WRITE_ONCE(vma->vm_start, address);
+				WRITE_ONCE(vma->vm_pgoff, vma->vm_pgoff - grow);
 				anon_vma_interval_tree_post_update_vma(vma);
 				vma_gap_update(vma);
 				spin_unlock(&mm->page_table_lock);
diff --git a/mm/mprotect.c b/mm/mprotect.c
index 625608bc8962..83594cc68062 100644
--- a/mm/mprotect.c
+++ b/mm/mprotect.c
@@ -375,12 +375,14 @@ mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev,
 	 * vm_flags and vm_page_prot are protected by the mmap_sem
 	 * held in write mode.
 	 */
-	vma->vm_flags = newflags;
+	vm_write_begin(vma);
+	WRITE_ONCE(vma->vm_flags, newflags);
 	dirty_accountable = vma_wants_writenotify(vma, vma->vm_page_prot);
 	vma_set_page_prot(vma);
 
 	change_protection(vma, start, end, vma->vm_page_prot,
 			  dirty_accountable, 0);
+	vm_write_end(vma);
 
 	/*
 	 * Private VM_LOCKED VMA becoming writable: trigger COW to avoid major
diff --git a/mm/swap_state.c b/mm/swap_state.c
index c6b3eab73fde..2ee7198df281 100644
--- a/mm/swap_state.c
+++ b/mm/swap_state.c
@@ -572,6 +572,10 @@ static unsigned long swapin_nr_pages(unsigned long offset)
  * the readahead.
  *
  * Caller must hold down_read on the vma->vm_mm if vmf->vma is not NULL.
+ * This is needed to ensure the VMA will not be freed in our back. In the case
+ * of the speculative page fault handler, this cannot happen, even if we don't
+ * hold the mmap_sem. Callees are assumed to take care of reading VMA's fields
+ * using READ_ONCE() to read consistent values.
  */
 struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
 				struct vm_fault *vmf)
@@ -665,9 +669,9 @@ static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
 				     unsigned long *start,
 				     unsigned long *end)
 {
-	*start = max3(lpfn, PFN_DOWN(vma->vm_start),
+	*start = max3(lpfn, PFN_DOWN(READ_ONCE(vma->vm_start)),
 		      PFN_DOWN(faddr & PMD_MASK));
-	*end = min3(rpfn, PFN_DOWN(vma->vm_end),
+	*end = min3(rpfn, PFN_DOWN(READ_ONCE(vma->vm_end)),
 		    PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
 }
 
-- 
2.7.4

[PATCH v11 11/26] mm: protect mremap() against SPF hanlder

[Laurent Dufour]

If a thread is remapping an area while another one is faulting on the
destination area, the SPF handler may fetch the vma from the RB tree before
the pte has been moved by the other thread. This means that the moved ptes
will overwrite those create by the page fault handler leading to page
leaked.

	CPU 1				CPU2
	enter mremap()
	unmap the dest area
	copy_vma()			Enter speculative page fault handler
	   >> at this time the dest area is present in the RB tree
					fetch the vma matching dest area
					create a pte as the VMA matched
					Exit the SPF handler
					<data written in the new page>
	move_ptes()
	  > it is assumed that the dest area is empty,
 	  > the move ptes overwrite the page mapped by the CPU2.

To prevent that, when the VMA matching the dest area is extended or created
by copy_vma(), it should be marked as non available to the SPF handler.
The usual way to so is to rely on vm_write_begin()/end().
This is already in __vma_adjust() called by copy_vma() (through
vma_merge()). But __vma_adjust() is calling vm_write_end() before returning
which create a window for another thread.
This patch adds a new parameter to vma_merge() which is passed down to
vma_adjust().
The assumption is that copy_vma() is returning a vma which should be
released by calling vm_raw_write_end() by the callee once the ptes have
been moved.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/mm.h | 24 +++++++++++++++++++-----
 mm/mmap.c          | 53 +++++++++++++++++++++++++++++++++++++++++------------
 mm/mremap.c        | 13 +++++++++++++
 3 files changed, 73 insertions(+), 17 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 18acfdeee759..3f8b2ce0ef7c 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -2253,18 +2253,32 @@ void anon_vma_interval_tree_verify(struct anon_vma_chain *node);
 
 /* mmap.c */
 extern int __vm_enough_memory(struct mm_struct *mm, long pages, int cap_sys_admin);
+
 extern int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 	unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
-	struct vm_area_struct *expand);
+	struct vm_area_struct *expand, bool keep_locked);
+
 static inline int vma_adjust(struct vm_area_struct *vma, unsigned long start,
 	unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert)
 {
-	return __vma_adjust(vma, start, end, pgoff, insert, NULL);
+	return __vma_adjust(vma, start, end, pgoff, insert, NULL, false);
 }
-extern struct vm_area_struct *vma_merge(struct mm_struct *,
+
+extern struct vm_area_struct *__vma_merge(struct mm_struct *mm,
+	struct vm_area_struct *prev, unsigned long addr, unsigned long end,
+	unsigned long vm_flags, struct anon_vma *anon, struct file *file,
+	pgoff_t pgoff, struct mempolicy *mpol,
+	struct vm_userfaultfd_ctx uff, bool keep_locked);
+
+static inline struct vm_area_struct *vma_merge(struct mm_struct *mm,
 	struct vm_area_struct *prev, unsigned long addr, unsigned long end,
-	unsigned long vm_flags, struct anon_vma *, struct file *, pgoff_t,
-	struct mempolicy *, struct vm_userfaultfd_ctx);
+	unsigned long vm_flags, struct anon_vma *anon, struct file *file,
+	pgoff_t off, struct mempolicy *pol, struct vm_userfaultfd_ctx uff)
+{
+	return __vma_merge(mm, prev, addr, end, vm_flags, anon, file, off,
+			   pol, uff, false);
+}
+
 extern struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *);
 extern int __split_vma(struct mm_struct *, struct vm_area_struct *,
 	unsigned long addr, int new_below);
diff --git a/mm/mmap.c b/mm/mmap.c
index add13b4e1d8d..2450860e3f8e 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -689,7 +689,7 @@ static inline void __vma_unlink_prev(struct mm_struct *mm,
  */
 int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 	unsigned long end, pgoff_t pgoff, struct vm_area_struct *insert,
-	struct vm_area_struct *expand)
+	struct vm_area_struct *expand, bool keep_locked)
 {
 	struct mm_struct *mm = vma->vm_mm;
 	struct vm_area_struct *next = vma->vm_next, *orig_vma = vma;
@@ -805,8 +805,12 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 
 			importer->anon_vma = exporter->anon_vma;
 			error = anon_vma_clone(importer, exporter);
-			if (error)
+			if (error) {
+				if (next && next != vma)
+					vm_raw_write_end(next);
+				vm_raw_write_end(vma);
 				return error;
+			}
 		}
 	}
 again:
@@ -1001,7 +1005,8 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 
 	if (next && next != vma)
 		vm_raw_write_end(next);
-	vm_raw_write_end(vma);
+	if (!keep_locked)
+		vm_raw_write_end(vma);
 
 	validate_mm(mm);
 
@@ -1137,12 +1142,13 @@ can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags,
  * parameter) may establish ptes with the wrong permissions of NNNN
  * instead of the right permissions of XXXX.
  */
-struct vm_area_struct *vma_merge(struct mm_struct *mm,
+struct vm_area_struct *__vma_merge(struct mm_struct *mm,
 			struct vm_area_struct *prev, unsigned long addr,
 			unsigned long end, unsigned long vm_flags,
 			struct anon_vma *anon_vma, struct file *file,
 			pgoff_t pgoff, struct mempolicy *policy,
-			struct vm_userfaultfd_ctx vm_userfaultfd_ctx)
+			struct vm_userfaultfd_ctx vm_userfaultfd_ctx,
+			bool keep_locked)
 {
 	pgoff_t pglen = (end - addr) >> PAGE_SHIFT;
 	struct vm_area_struct *area, *next;
@@ -1190,10 +1196,11 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm,
 							/* cases 1, 6 */
 			err = __vma_adjust(prev, prev->vm_start,
 					 next->vm_end, prev->vm_pgoff, NULL,
-					 prev);
+					 prev, keep_locked);
 		} else					/* cases 2, 5, 7 */
 			err = __vma_adjust(prev, prev->vm_start,
-					 end, prev->vm_pgoff, NULL, prev);
+					   end, prev->vm_pgoff, NULL, prev,
+					   keep_locked);
 		if (err)
 			return NULL;
 		khugepaged_enter_vma_merge(prev, vm_flags);
@@ -1210,10 +1217,12 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm,
 					     vm_userfaultfd_ctx)) {
 		if (prev && addr < prev->vm_end)	/* case 4 */
 			err = __vma_adjust(prev, prev->vm_start,
-					 addr, prev->vm_pgoff, NULL, next);
+					 addr, prev->vm_pgoff, NULL, next,
+					 keep_locked);
 		else {					/* cases 3, 8 */
 			err = __vma_adjust(area, addr, next->vm_end,
-					 next->vm_pgoff - pglen, NULL, next);
+					 next->vm_pgoff - pglen, NULL, next,
+					 keep_locked);
 			/*
 			 * In case 3 area is already equal to next and
 			 * this is a noop, but in case 8 "area" has
@@ -3184,9 +3193,20 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
 
 	if (find_vma_links(mm, addr, addr + len, &prev, &rb_link, &rb_parent))
 		return NULL;	/* should never get here */
-	new_vma = vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
-			    vma->anon_vma, vma->vm_file, pgoff, vma_policy(vma),
-			    vma->vm_userfaultfd_ctx);
+
+	/* There is 3 cases to manage here in
+	 *     AAAA            AAAA              AAAA              AAAA
+	 * PPPP....      PPPP......NNNN      PPPP....NNNN      PP........NN
+	 * PPPPPPPP(A)   PPPP..NNNNNNNN(B)   PPPPPPPPPPPP(1)       NULL
+	 *                                   PPPPPPPPNNNN(2)
+	 *                                   PPPPNNNNNNNN(3)
+	 *
+	 * new_vma == prev in case A,1,2
+	 * new_vma == next in case B,3
+	 */
+	new_vma = __vma_merge(mm, prev, addr, addr + len, vma->vm_flags,
+			      vma->anon_vma, vma->vm_file, pgoff,
+			      vma_policy(vma), vma->vm_userfaultfd_ctx, true);
 	if (new_vma) {
 		/*
 		 * Source vma may have been merged into new_vma
@@ -3226,6 +3246,15 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
 			get_file(new_vma->vm_file);
 		if (new_vma->vm_ops && new_vma->vm_ops->open)
 			new_vma->vm_ops->open(new_vma);
+		/*
+		 * As the VMA is linked right now, it may be hit by the
+		 * speculative page fault handler. But we don't want it to
+		 * to start mapping page in this area until the caller has
+		 * potentially move the pte from the moved VMA. To prevent
+		 * that we protect it right now, and let the caller unprotect
+		 * it once the move is done.
+		 */
+		vm_raw_write_begin(new_vma);
 		vma_link(mm, new_vma, prev, rb_link, rb_parent);
 		*need_rmap_locks = false;
 	}
diff --git a/mm/mremap.c b/mm/mremap.c
index 049470aa1e3e..8ed1a1d6eaed 100644
--- a/mm/mremap.c
+++ b/mm/mremap.c
@@ -302,6 +302,14 @@ static unsigned long move_vma(struct vm_area_struct *vma,
 	if (!new_vma)
 		return -ENOMEM;
 
+	/* new_vma is returned protected by copy_vma, to prevent speculative
+	 * page fault to be done in the destination area before we move the pte.
+	 * Now, we must also protect the source VMA since we don't want pages
+	 * to be mapped in our back while we are copying the PTEs.
+	 */
+	if (vma != new_vma)
+		vm_raw_write_begin(vma);
+
 	moved_len = move_page_tables(vma, old_addr, new_vma, new_addr, old_len,
 				     need_rmap_locks);
 	if (moved_len < old_len) {
@@ -318,6 +326,8 @@ static unsigned long move_vma(struct vm_area_struct *vma,
 		 */
 		move_page_tables(new_vma, new_addr, vma, old_addr, moved_len,
 				 true);
+		if (vma != new_vma)
+			vm_raw_write_end(vma);
 		vma = new_vma;
 		old_len = new_len;
 		old_addr = new_addr;
@@ -326,7 +336,10 @@ static unsigned long move_vma(struct vm_area_struct *vma,
 		mremap_userfaultfd_prep(new_vma, uf);
 		arch_remap(mm, old_addr, old_addr + old_len,
 			   new_addr, new_addr + new_len);
+		if (vma != new_vma)
+			vm_raw_write_end(vma);
 	}
+	vm_raw_write_end(new_vma);
 
 	/* Conceal VM_ACCOUNT so old reservation is not undone */
 	if (vm_flags & VM_ACCOUNT) {
-- 
2.7.4

[PATCH v11 12/26] mm: protect SPF handler against anon_vma changes

[Laurent Dufour]

The speculative page fault handler must be protected against anon_vma
changes. This is because page_add_new_anon_rmap() is called during the
speculative path.

In addition, don't try speculative page fault if the VMA don't have an
anon_vma structure allocated because its allocation should be
protected by the mmap_sem.

In __vma_adjust() when importer->anon_vma is set, there is no need to
protect against speculative page faults since speculative page fault
is aborted if the vma->anon_vma is not set.

When calling page_add_new_anon_rmap() vma->anon_vma is necessarily
valid since we checked for it when locking the pte and the anon_vma is
removed once the pte is unlocked. So even if the speculative page
fault handler is running concurrently with do_unmap(), as the pte is
locked in unmap_region() - through unmap_vmas() - and the anon_vma
unlinked later, because we check for the vma sequence counter which is
updated in unmap_page_range() before locking the pte, and then in
free_pgtables() so when locking the pte the change will be detected.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 mm/memory.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/mm/memory.c b/mm/memory.c
index 551a1916da5d..d0b5f14cfe69 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -624,7 +624,9 @@ void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
 		 * Hide vma from rmap and truncate_pagecache before freeing
 		 * pgtables
 		 */
+		vm_write_begin(vma);
 		unlink_anon_vmas(vma);
+		vm_write_end(vma);
 		unlink_file_vma(vma);
 
 		if (is_vm_hugetlb_page(vma)) {
@@ -638,7 +640,9 @@ void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
 			       && !is_vm_hugetlb_page(next)) {
 				vma = next;
 				next = vma->vm_next;
+				vm_write_begin(vma);
 				unlink_anon_vmas(vma);
+				vm_write_end(vma);
 				unlink_file_vma(vma);
 			}
 			free_pgd_range(tlb, addr, vma->vm_end,
-- 
2.7.4

[PATCH v11 13/26] mm: cache some VMA fields in the vm_fault structure

[Laurent Dufour]

When handling speculative page fault, the vma->vm_flags and
vma->vm_page_prot fields are read once the page table lock is released. So
there is no more guarantee that these fields would not change in our back.
They will be saved in the vm_fault structure before the VMA is checked for
changes.

In the detail, when we deal with a speculative page fault, the mmap_sem is
not taken, so parallel VMA's changes can occurred. When a VMA change is
done which will impact the page fault processing, we assumed that the VMA
sequence counter will be changed.  In the page fault processing, at the
time the PTE is locked, we checked the VMA sequence counter to detect
changes done in our back. If no change is detected we can continue further.
But this doesn't prevent the VMA to not be changed in our back while the
PTE is locked. So VMA's fields which are used while the PTE is locked must
be saved to ensure that we are using *static* values.  This is important
since the PTE changes will be made on regards to these VMA fields and they
need to be consistent. This concerns the vma->vm_flags and
vma->vm_page_prot VMA fields.

This patch also set the fields in hugetlb_no_page() and
__collapse_huge_page_swapin even if it is not need for the callee.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/mm.h | 10 ++++++++--
 mm/huge_memory.c   |  6 +++---
 mm/hugetlb.c       |  2 ++
 mm/khugepaged.c    |  2 ++
 mm/memory.c        | 50 ++++++++++++++++++++++++++------------------------
 mm/migrate.c       |  2 +-
 6 files changed, 42 insertions(+), 30 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 3f8b2ce0ef7c..f385d721867d 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -373,6 +373,12 @@ struct vm_fault {
 					 * page table to avoid allocation from
 					 * atomic context.
 					 */
+	/*
+	 * These entries are required when handling speculative page fault.
+	 * This way the page handling is done using consistent field values.
+	 */
+	unsigned long vma_flags;
+	pgprot_t vma_page_prot;
 };
 
 /* page entry size for vm->huge_fault() */
@@ -693,9 +699,9 @@ void free_compound_page(struct page *page);
  * pte_mkwrite.  But get_user_pages can cause write faults for mappings
  * that do not have writing enabled, when used by access_process_vm.
  */
-static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
+static inline pte_t maybe_mkwrite(pte_t pte, unsigned long vma_flags)
 {
-	if (likely(vma->vm_flags & VM_WRITE))
+	if (likely(vma_flags & VM_WRITE))
 		pte = pte_mkwrite(pte);
 	return pte;
 }
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 323acdd14e6e..6bf5420cc62e 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1194,8 +1194,8 @@ static int do_huge_pmd_wp_page_fallback(struct vm_fault *vmf, pmd_t orig_pmd,
 
 	for (i = 0; i < HPAGE_PMD_NR; i++, haddr += PAGE_SIZE) {
 		pte_t entry;
-		entry = mk_pte(pages[i], vma->vm_page_prot);
-		entry = maybe_mkwrite(pte_mkdirty(entry), vma);
+		entry = mk_pte(pages[i], vmf->vma_page_prot);
+		entry = maybe_mkwrite(pte_mkdirty(entry), vmf->vma_flags);
 		memcg = (void *)page_private(pages[i]);
 		set_page_private(pages[i], 0);
 		page_add_new_anon_rmap(pages[i], vmf->vma, haddr, false);
@@ -2168,7 +2168,7 @@ static void __split_huge_pmd_locked(struct vm_area_struct *vma, pmd_t *pmd,
 				entry = pte_swp_mksoft_dirty(entry);
 		} else {
 			entry = mk_pte(page + i, READ_ONCE(vma->vm_page_prot));
-			entry = maybe_mkwrite(entry, vma);
+			entry = maybe_mkwrite(entry, vma->vm_flags);
 			if (!write)
 				entry = pte_wrprotect(entry);
 			if (!young)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 129088710510..d7764b6568f5 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -3718,6 +3718,8 @@ static int hugetlb_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
 				.vma = vma,
 				.address = address,
 				.flags = flags,
+				.vma_flags = vma->vm_flags,
+				.vma_page_prot = vma->vm_page_prot,
 				/*
 				 * Hard to debug if it ends up being
 				 * used by a callee that assumes
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index 0b28af4b950d..2b02a9f9589e 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -887,6 +887,8 @@ static bool __collapse_huge_page_swapin(struct mm_struct *mm,
 		.flags = FAULT_FLAG_ALLOW_RETRY,
 		.pmd = pmd,
 		.pgoff = linear_page_index(vma, address),
+		.vma_flags = vma->vm_flags,
+		.vma_page_prot = vma->vm_page_prot,
 	};
 
 	/* we only decide to swapin, if there is enough young ptes */
diff --git a/mm/memory.c b/mm/memory.c
index d0b5f14cfe69..9dc455ae550c 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1822,7 +1822,7 @@ static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
 out_mkwrite:
 	if (mkwrite) {
 		entry = pte_mkyoung(entry);
-		entry = maybe_mkwrite(pte_mkdirty(entry), vma);
+		entry = maybe_mkwrite(pte_mkdirty(entry), vma->vm_flags);
 	}
 
 	set_pte_at(mm, addr, pte, entry);
@@ -2482,7 +2482,7 @@ static inline void wp_page_reuse(struct vm_fault *vmf)
 
 	flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
 	entry = pte_mkyoung(vmf->orig_pte);
-	entry = maybe_mkwrite(pte_mkdirty(entry), vma);
+	entry = maybe_mkwrite(pte_mkdirty(entry), vmf->vma_flags);
 	if (ptep_set_access_flags(vma, vmf->address, vmf->pte, entry, 1))
 		update_mmu_cache(vma, vmf->address, vmf->pte);
 	pte_unmap_unlock(vmf->pte, vmf->ptl);
@@ -2558,8 +2558,8 @@ static int wp_page_copy(struct vm_fault *vmf)
 			inc_mm_counter_fast(mm, MM_ANONPAGES);
 		}
 		flush_cache_page(vma, vmf->address, pte_pfn(vmf->orig_pte));
-		entry = mk_pte(new_page, vma->vm_page_prot);
-		entry = maybe_mkwrite(pte_mkdirty(entry), vma);
+		entry = mk_pte(new_page, vmf->vma_page_prot);
+		entry = maybe_mkwrite(pte_mkdirty(entry), vmf->vma_flags);
 		/*
 		 * Clear the pte entry and flush it first, before updating the
 		 * pte with the new entry. This will avoid a race condition
@@ -2624,7 +2624,7 @@ static int wp_page_copy(struct vm_fault *vmf)
 		 * Don't let another task, with possibly unlocked vma,
 		 * keep the mlocked page.
 		 */
-		if (page_copied && (vma->vm_flags & VM_LOCKED)) {
+		if (page_copied && (vmf->vma_flags & VM_LOCKED)) {
 			lock_page(old_page);	/* LRU manipulation */
 			if (PageMlocked(old_page))
 				munlock_vma_page(old_page);
@@ -2660,7 +2660,7 @@ static int wp_page_copy(struct vm_fault *vmf)
  */
 int finish_mkwrite_fault(struct vm_fault *vmf)
 {
-	WARN_ON_ONCE(!(vmf->vma->vm_flags & VM_SHARED));
+	WARN_ON_ONCE(!(vmf->vma_flags & VM_SHARED));
 	if (!pte_map_lock(vmf))
 		return VM_FAULT_RETRY;
 	/*
@@ -2762,7 +2762,7 @@ static int do_wp_page(struct vm_fault *vmf)
 		 * We should not cow pages in a shared writeable mapping.
 		 * Just mark the pages writable and/or call ops->pfn_mkwrite.
 		 */
-		if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
+		if ((vmf->vma_flags & (VM_WRITE|VM_SHARED)) ==
 				     (VM_WRITE|VM_SHARED))
 			return wp_pfn_shared(vmf);
 
@@ -2809,7 +2809,7 @@ static int do_wp_page(struct vm_fault *vmf)
 			return VM_FAULT_WRITE;
 		}
 		unlock_page(vmf->page);
-	} else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
+	} else if (unlikely((vmf->vma_flags & (VM_WRITE|VM_SHARED)) ==
 					(VM_WRITE|VM_SHARED))) {
 		return wp_page_shared(vmf);
 	}
@@ -3088,9 +3088,9 @@ int do_swap_page(struct vm_fault *vmf)
 
 	inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
 	dec_mm_counter_fast(vma->vm_mm, MM_SWAPENTS);
-	pte = mk_pte(page, vma->vm_page_prot);
+	pte = mk_pte(page, vmf->vma_page_prot);
 	if ((vmf->flags & FAULT_FLAG_WRITE) && reuse_swap_page(page, NULL)) {
-		pte = maybe_mkwrite(pte_mkdirty(pte), vma);
+		pte = maybe_mkwrite(pte_mkdirty(pte), vmf->vma_flags);
 		vmf->flags &= ~FAULT_FLAG_WRITE;
 		ret |= VM_FAULT_WRITE;
 		exclusive = RMAP_EXCLUSIVE;
@@ -3115,7 +3115,7 @@ int do_swap_page(struct vm_fault *vmf)
 
 	swap_free(entry);
 	if (mem_cgroup_swap_full(page) ||
-	    (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
+	    (vmf->vma_flags & VM_LOCKED) || PageMlocked(page))
 		try_to_free_swap(page);
 	unlock_page(page);
 	if (page != swapcache && swapcache) {
@@ -3173,7 +3173,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
 	pte_t entry;
 
 	/* File mapping without ->vm_ops ? */
-	if (vma->vm_flags & VM_SHARED)
+	if (vmf->vma_flags & VM_SHARED)
 		return VM_FAULT_SIGBUS;
 
 	/*
@@ -3197,7 +3197,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
 	if (!(vmf->flags & FAULT_FLAG_WRITE) &&
 			!mm_forbids_zeropage(vma->vm_mm)) {
 		entry = pte_mkspecial(pfn_pte(my_zero_pfn(vmf->address),
-						vma->vm_page_prot));
+						vmf->vma_page_prot));
 		if (!pte_map_lock(vmf))
 			return VM_FAULT_RETRY;
 		if (!pte_none(*vmf->pte))
@@ -3230,8 +3230,8 @@ static int do_anonymous_page(struct vm_fault *vmf)
 	 */
 	__SetPageUptodate(page);
 
-	entry = mk_pte(page, vma->vm_page_prot);
-	if (vma->vm_flags & VM_WRITE)
+	entry = mk_pte(page, vmf->vma_page_prot);
+	if (vmf->vma_flags & VM_WRITE)
 		entry = pte_mkwrite(pte_mkdirty(entry));
 
 	if (!pte_map_lock(vmf)) {
@@ -3428,7 +3428,7 @@ static int do_set_pmd(struct vm_fault *vmf, struct page *page)
 	for (i = 0; i < HPAGE_PMD_NR; i++)
 		flush_icache_page(vma, page + i);
 
-	entry = mk_huge_pmd(page, vma->vm_page_prot);
+	entry = mk_huge_pmd(page, vmf->vma_page_prot);
 	if (write)
 		entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
 
@@ -3502,11 +3502,11 @@ int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg,
 		return VM_FAULT_NOPAGE;
 
 	flush_icache_page(vma, page);
-	entry = mk_pte(page, vma->vm_page_prot);
+	entry = mk_pte(page, vmf->vma_page_prot);
 	if (write)
-		entry = maybe_mkwrite(pte_mkdirty(entry), vma);
+		entry = maybe_mkwrite(pte_mkdirty(entry), vmf->vma_flags);
 	/* copy-on-write page */
-	if (write && !(vma->vm_flags & VM_SHARED)) {
+	if (write && !(vmf->vma_flags & VM_SHARED)) {
 		inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
 		page_add_new_anon_rmap(page, vma, vmf->address, false);
 		mem_cgroup_commit_charge(page, memcg, false, false);
@@ -3545,7 +3545,7 @@ int finish_fault(struct vm_fault *vmf)
 
 	/* Did we COW the page? */
 	if ((vmf->flags & FAULT_FLAG_WRITE) &&
-	    !(vmf->vma->vm_flags & VM_SHARED))
+	    !(vmf->vma_flags & VM_SHARED))
 		page = vmf->cow_page;
 	else
 		page = vmf->page;
@@ -3799,7 +3799,7 @@ static int do_fault(struct vm_fault *vmf)
 		ret = VM_FAULT_SIGBUS;
 	else if (!(vmf->flags & FAULT_FLAG_WRITE))
 		ret = do_read_fault(vmf);
-	else if (!(vma->vm_flags & VM_SHARED))
+	else if (!(vmf->vma_flags & VM_SHARED))
 		ret = do_cow_fault(vmf);
 	else
 		ret = do_shared_fault(vmf);
@@ -3856,7 +3856,7 @@ static int do_numa_page(struct vm_fault *vmf)
 	 * accessible ptes, some can allow access by kernel mode.
 	 */
 	pte = ptep_modify_prot_start(vma->vm_mm, vmf->address, vmf->pte);
-	pte = pte_modify(pte, vma->vm_page_prot);
+	pte = pte_modify(pte, vmf->vma_page_prot);
 	pte = pte_mkyoung(pte);
 	if (was_writable)
 		pte = pte_mkwrite(pte);
@@ -3890,7 +3890,7 @@ static int do_numa_page(struct vm_fault *vmf)
 	 * Flag if the page is shared between multiple address spaces. This
 	 * is later used when determining whether to group tasks together
 	 */
-	if (page_mapcount(page) > 1 && (vma->vm_flags & VM_SHARED))
+	if (page_mapcount(page) > 1 && (vmf->vma_flags & VM_SHARED))
 		flags |= TNF_SHARED;
 
 	last_cpupid = page_cpupid_last(page);
@@ -3935,7 +3935,7 @@ static inline int wp_huge_pmd(struct vm_fault *vmf, pmd_t orig_pmd)
 		return vmf->vma->vm_ops->huge_fault(vmf, PE_SIZE_PMD);
 
 	/* COW handled on pte level: split pmd */
-	VM_BUG_ON_VMA(vmf->vma->vm_flags & VM_SHARED, vmf->vma);
+	VM_BUG_ON_VMA(vmf->vma_flags & VM_SHARED, vmf->vma);
 	__split_huge_pmd(vmf->vma, vmf->pmd, vmf->address, false, NULL);
 
 	return VM_FAULT_FALLBACK;
@@ -4082,6 +4082,8 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
 		.flags = flags,
 		.pgoff = linear_page_index(vma, address),
 		.gfp_mask = __get_fault_gfp_mask(vma),
+		.vma_flags = vma->vm_flags,
+		.vma_page_prot = vma->vm_page_prot,
 	};
 	unsigned int dirty = flags & FAULT_FLAG_WRITE;
 	struct mm_struct *mm = vma->vm_mm;
diff --git a/mm/migrate.c b/mm/migrate.c
index 8c0af0f7cab1..ae3d0faf72cb 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -240,7 +240,7 @@ static bool remove_migration_pte(struct page *page, struct vm_area_struct *vma,
 		 */
 		entry = pte_to_swp_entry(*pvmw.pte);
 		if (is_write_migration_entry(entry))
-			pte = maybe_mkwrite(pte, vma);
+			pte = maybe_mkwrite(pte, vma->vm_flags);
 
 		if (unlikely(is_zone_device_page(new))) {
 			if (is_device_private_page(new)) {
-- 
2.7.4

[PATCH v11 14/26] mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()

[Laurent Dufour]

migrate_misplaced_page() is only called during the page fault handling so
it's better to pass the pointer to the struct vm_fault instead of the vma.

This way during the speculative page fault path the saved vma->vm_flags
could be used.

Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/migrate.h | 4 ++--
 mm/memory.c             | 2 +-
 mm/migrate.c            | 4 ++--
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/include/linux/migrate.h b/include/linux/migrate.h
index f2b4abbca55e..fd4c3ab7bd9c 100644
--- a/include/linux/migrate.h
+++ b/include/linux/migrate.h
@@ -126,14 +126,14 @@ static inline void __ClearPageMovable(struct page *page)
 #ifdef CONFIG_NUMA_BALANCING
 extern bool pmd_trans_migrating(pmd_t pmd);
 extern int migrate_misplaced_page(struct page *page,
-				  struct vm_area_struct *vma, int node);
+				  struct vm_fault *vmf, int node);
 #else
 static inline bool pmd_trans_migrating(pmd_t pmd)
 {
 	return false;
 }
 static inline int migrate_misplaced_page(struct page *page,
-					 struct vm_area_struct *vma, int node)
+					 struct vm_fault *vmf, int node)
 {
 	return -EAGAIN; /* can't migrate now */
 }
diff --git a/mm/memory.c b/mm/memory.c
index 9dc455ae550c..cb6310b74cfb 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -3904,7 +3904,7 @@ static int do_numa_page(struct vm_fault *vmf)
 	}
 
 	/* Migrate to the requested node */
-	migrated = migrate_misplaced_page(page, vma, target_nid);
+	migrated = migrate_misplaced_page(page, vmf, target_nid);
 	if (migrated) {
 		page_nid = target_nid;
 		flags |= TNF_MIGRATED;
diff --git a/mm/migrate.c b/mm/migrate.c
index ae3d0faf72cb..884c57a16b7a 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -1945,7 +1945,7 @@ bool pmd_trans_migrating(pmd_t pmd)
  * node. Caller is expected to have an elevated reference count on
  * the page that will be dropped by this function before returning.
  */
-int migrate_misplaced_page(struct page *page, struct vm_area_struct *vma,
+int migrate_misplaced_page(struct page *page, struct vm_fault *vmf,
 			   int node)
 {
 	pg_data_t *pgdat = NODE_DATA(node);
@@ -1958,7 +1958,7 @@ int migrate_misplaced_page(struct page *page, struct vm_area_struct *vma,
 	 * with execute permissions as they are probably shared libraries.
 	 */
 	if (page_mapcount(page) != 1 && page_is_file_cache(page) &&
-	    (vma->vm_flags & VM_EXEC))
+	    (vmf->vma_flags & VM_EXEC))
 		goto out;
 
 	/*
-- 
2.7.4

[PATCH v11 15/26] mm: introduce __lru_cache_add_active_or_unevictable

[Laurent Dufour]

The speculative page fault handler which is run without holding the
mmap_sem is calling lru_cache_add_active_or_unevictable() but the vm_flags
is not guaranteed to remain constant.
Introducing __lru_cache_add_active_or_unevictable() which has the vma flags
value parameter instead of the vma pointer.

Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/swap.h | 10 ++++++++--
 mm/memory.c          |  8 ++++----
 mm/swap.c            |  6 +++---
 3 files changed, 15 insertions(+), 9 deletions(-)

diff --git a/include/linux/swap.h b/include/linux/swap.h
index f73eafcaf4e9..730c14738574 100644
--- a/include/linux/swap.h
+++ b/include/linux/swap.h
@@ -338,8 +338,14 @@ extern void deactivate_file_page(struct page *page);
 extern void mark_page_lazyfree(struct page *page);
 extern void swap_setup(void);
 
-extern void lru_cache_add_active_or_unevictable(struct page *page,
-						struct vm_area_struct *vma);
+extern void __lru_cache_add_active_or_unevictable(struct page *page,
+						unsigned long vma_flags);
+
+static inline void lru_cache_add_active_or_unevictable(struct page *page,
+						struct vm_area_struct *vma)
+{
+	return __lru_cache_add_active_or_unevictable(page, vma->vm_flags);
+}
 
 /* linux/mm/vmscan.c */
 extern unsigned long zone_reclaimable_pages(struct zone *zone);
diff --git a/mm/memory.c b/mm/memory.c
index cb6310b74cfb..deac7f12d777 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2569,7 +2569,7 @@ static int wp_page_copy(struct vm_fault *vmf)
 		ptep_clear_flush_notify(vma, vmf->address, vmf->pte);
 		page_add_new_anon_rmap(new_page, vma, vmf->address, false);
 		mem_cgroup_commit_charge(new_page, memcg, false, false);
-		lru_cache_add_active_or_unevictable(new_page, vma);
+		__lru_cache_add_active_or_unevictable(new_page, vmf->vma_flags);
 		/*
 		 * We call the notify macro here because, when using secondary
 		 * mmu page tables (such as kvm shadow page tables), we want the
@@ -3105,7 +3105,7 @@ int do_swap_page(struct vm_fault *vmf)
 	if (unlikely(page != swapcache && swapcache)) {
 		page_add_new_anon_rmap(page, vma, vmf->address, false);
 		mem_cgroup_commit_charge(page, memcg, false, false);
-		lru_cache_add_active_or_unevictable(page, vma);
+		__lru_cache_add_active_or_unevictable(page, vmf->vma_flags);
 	} else {
 		do_page_add_anon_rmap(page, vma, vmf->address, exclusive);
 		mem_cgroup_commit_charge(page, memcg, true, false);
@@ -3256,7 +3256,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
 	inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
 	page_add_new_anon_rmap(page, vma, vmf->address, false);
 	mem_cgroup_commit_charge(page, memcg, false, false);
-	lru_cache_add_active_or_unevictable(page, vma);
+	__lru_cache_add_active_or_unevictable(page, vmf->vma_flags);
 setpte:
 	set_pte_at(vma->vm_mm, vmf->address, vmf->pte, entry);
 
@@ -3510,7 +3510,7 @@ int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg,
 		inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
 		page_add_new_anon_rmap(page, vma, vmf->address, false);
 		mem_cgroup_commit_charge(page, memcg, false, false);
-		lru_cache_add_active_or_unevictable(page, vma);
+		__lru_cache_add_active_or_unevictable(page, vmf->vma_flags);
 	} else {
 		inc_mm_counter_fast(vma->vm_mm, mm_counter_file(page));
 		page_add_file_rmap(page, false);
diff --git a/mm/swap.c b/mm/swap.c
index 26fc9b5f1b6c..ba97d437e68a 100644
--- a/mm/swap.c
+++ b/mm/swap.c
@@ -456,12 +456,12 @@ void lru_cache_add(struct page *page)
  * directly back onto it's zone's unevictable list, it does NOT use a
  * per cpu pagevec.
  */
-void lru_cache_add_active_or_unevictable(struct page *page,
-					 struct vm_area_struct *vma)
+void __lru_cache_add_active_or_unevictable(struct page *page,
+					   unsigned long vma_flags)
 {
 	VM_BUG_ON_PAGE(PageLRU(page), page);
 
-	if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
+	if (likely((vma_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED))
 		SetPageActive(page);
 	else if (!TestSetPageMlocked(page)) {
 		/*
-- 
2.7.4

[PATCH v11 16/26] mm: introduce __vm_normal_page()

[Laurent Dufour]

When dealing with the speculative fault path we should use the VMA's field
cached value stored in the vm_fault structure.

Currently vm_normal_page() is using the pointer to the VMA to fetch the
vm_flags value. This patch provides a new __vm_normal_page() which is
receiving the vm_flags flags value as parameter.

Note: The speculative path is turned on for architecture providing support
for special PTE flag. So only the first block of vm_normal_page is used
during the speculative path.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/mm.h | 18 +++++++++++++++---
 mm/memory.c        | 21 ++++++++++++---------
 2 files changed, 27 insertions(+), 12 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index f385d721867d..bcebec117d4d 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1317,9 +1317,21 @@ static inline void INIT_VMA(struct vm_area_struct *vma)
 #endif
 }
 
-struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
-			     pte_t pte, bool with_public_device);
-#define vm_normal_page(vma, addr, pte) _vm_normal_page(vma, addr, pte, false)
+struct page *__vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
+			      pte_t pte, bool with_public_device,
+			      unsigned long vma_flags);
+static inline struct page *_vm_normal_page(struct vm_area_struct *vma,
+					    unsigned long addr, pte_t pte,
+					    bool with_public_device)
+{
+	return __vm_normal_page(vma, addr, pte, with_public_device,
+				vma->vm_flags);
+}
+static inline struct page *vm_normal_page(struct vm_area_struct *vma,
+					  unsigned long addr, pte_t pte)
+{
+	return _vm_normal_page(vma, addr, pte, false);
+}
 
 struct page *vm_normal_page_pmd(struct vm_area_struct *vma, unsigned long addr,
 				pmd_t pmd);
diff --git a/mm/memory.c b/mm/memory.c
index deac7f12d777..cc4e6221ee7b 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -780,7 +780,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
 }
 
 /*
- * vm_normal_page -- This function gets the "struct page" associated with a pte.
+ * __vm_normal_page -- This function gets the "struct page" associated with
+ * a pte.
  *
  * "Special" mappings do not wish to be associated with a "struct page" (either
  * it doesn't exist, or it exists but they don't want to touch it). In this
@@ -821,8 +822,9 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
  * PFNMAP mappings in order to support COWable mappings.
  *
  */
-struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
-			     pte_t pte, bool with_public_device)
+struct page *__vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
+			      pte_t pte, bool with_public_device,
+			      unsigned long vma_flags)
 {
 	unsigned long pfn = pte_pfn(pte);
 
@@ -831,7 +833,7 @@ struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
 			goto check_pfn;
 		if (vma->vm_ops && vma->vm_ops->find_special_page)
 			return vma->vm_ops->find_special_page(vma, addr);
-		if (vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP))
+		if (vma_flags & (VM_PFNMAP | VM_MIXEDMAP))
 			return NULL;
 		if (is_zero_pfn(pfn))
 			return NULL;
@@ -863,8 +865,8 @@ struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
 
 	/* !CONFIG_ARCH_HAS_PTE_SPECIAL case follows: */
 
-	if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
-		if (vma->vm_flags & VM_MIXEDMAP) {
+	if (unlikely(vma_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
+		if (vma_flags & VM_MIXEDMAP) {
 			if (!pfn_valid(pfn))
 				return NULL;
 			goto out;
@@ -873,7 +875,7 @@ struct page *_vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
 			off = (addr - vma->vm_start) >> PAGE_SHIFT;
 			if (pfn == vma->vm_pgoff + off)
 				return NULL;
-			if (!is_cow_mapping(vma->vm_flags))
+			if (!is_cow_mapping(vma_flags))
 				return NULL;
 		}
 	}
@@ -2753,7 +2755,8 @@ static int do_wp_page(struct vm_fault *vmf)
 {
 	struct vm_area_struct *vma = vmf->vma;
 
-	vmf->page = vm_normal_page(vma, vmf->address, vmf->orig_pte);
+	vmf->page = __vm_normal_page(vma, vmf->address, vmf->orig_pte, false,
+				     vmf->vma_flags);
 	if (!vmf->page) {
 		/*
 		 * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
@@ -3863,7 +3866,7 @@ static int do_numa_page(struct vm_fault *vmf)
 	ptep_modify_prot_commit(vma->vm_mm, vmf->address, vmf->pte, pte);
 	update_mmu_cache(vma, vmf->address, vmf->pte);
 
-	page = vm_normal_page(vma, vmf->address, pte);
+	page = __vm_normal_page(vma, vmf->address, pte, false, vmf->vma_flags);
 	if (!page) {
 		pte_unmap_unlock(vmf->pte, vmf->ptl);
 		return 0;
-- 
2.7.4

[PATCH v11 17/26] mm: introduce __page_add_new_anon_rmap()

[Laurent Dufour]

When dealing with speculative page fault handler, we may race with VMA
being split or merged. In this case the vma->vm_start and vm->vm_end
fields may not match the address the page fault is occurring.

This can only happens when the VMA is split but in that case, the
anon_vma pointer of the new VMA will be the same as the original one,
because in __split_vma the new->anon_vma is set to src->anon_vma when
*new = *vma.

So even if the VMA boundaries are not correct, the anon_vma pointer is
still valid.

If the VMA has been merged, then the VMA in which it has been merged
must have the same anon_vma pointer otherwise the merge can't be done.

So in all the case we know that the anon_vma is valid, since we have
checked before starting the speculative page fault that the anon_vma
pointer is valid for this VMA and since there is an anon_vma this
means that at one time a page has been backed and that before the VMA
is cleaned, the page table lock would have to be grab to clean the
PTE, and the anon_vma field is checked once the PTE is locked.

This patch introduce a new __page_add_new_anon_rmap() service which
doesn't check for the VMA boundaries, and create a new inline one
which do the check.

When called from a page fault handler, if this is not a speculative one,
there is a guarantee that vm_start and vm_end match the faulting address,
so this check is useless. In the context of the speculative page fault
handler, this check may be wrong but anon_vma is still valid as explained
above.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/rmap.h | 12 ++++++++++--
 mm/memory.c          |  8 ++++----
 mm/rmap.c            |  5 ++---
 3 files changed, 16 insertions(+), 9 deletions(-)

diff --git a/include/linux/rmap.h b/include/linux/rmap.h
index 988d176472df..a5d282573093 100644
--- a/include/linux/rmap.h
+++ b/include/linux/rmap.h
@@ -174,8 +174,16 @@ void page_add_anon_rmap(struct page *, struct vm_area_struct *,
 		unsigned long, bool);
 void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
 			   unsigned long, int);
-void page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
-		unsigned long, bool);
+void __page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
+			      unsigned long, bool);
+static inline void page_add_new_anon_rmap(struct page *page,
+					  struct vm_area_struct *vma,
+					  unsigned long address, bool compound)
+{
+	VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
+	__page_add_new_anon_rmap(page, vma, address, compound);
+}
+
 void page_add_file_rmap(struct page *, bool);
 void page_remove_rmap(struct page *, bool);
 
diff --git a/mm/memory.c b/mm/memory.c
index cc4e6221ee7b..ab32b0b4bd69 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -2569,7 +2569,7 @@ static int wp_page_copy(struct vm_fault *vmf)
 		 * thread doing COW.
 		 */
 		ptep_clear_flush_notify(vma, vmf->address, vmf->pte);
-		page_add_new_anon_rmap(new_page, vma, vmf->address, false);
+		__page_add_new_anon_rmap(new_page, vma, vmf->address, false);
 		mem_cgroup_commit_charge(new_page, memcg, false, false);
 		__lru_cache_add_active_or_unevictable(new_page, vmf->vma_flags);
 		/*
@@ -3106,7 +3106,7 @@ int do_swap_page(struct vm_fault *vmf)
 
 	/* ksm created a completely new copy */
 	if (unlikely(page != swapcache && swapcache)) {
-		page_add_new_anon_rmap(page, vma, vmf->address, false);
+		__page_add_new_anon_rmap(page, vma, vmf->address, false);
 		mem_cgroup_commit_charge(page, memcg, false, false);
 		__lru_cache_add_active_or_unevictable(page, vmf->vma_flags);
 	} else {
@@ -3257,7 +3257,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
 	}
 
 	inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
-	page_add_new_anon_rmap(page, vma, vmf->address, false);
+	__page_add_new_anon_rmap(page, vma, vmf->address, false);
 	mem_cgroup_commit_charge(page, memcg, false, false);
 	__lru_cache_add_active_or_unevictable(page, vmf->vma_flags);
 setpte:
@@ -3511,7 +3511,7 @@ int alloc_set_pte(struct vm_fault *vmf, struct mem_cgroup *memcg,
 	/* copy-on-write page */
 	if (write && !(vmf->vma_flags & VM_SHARED)) {
 		inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES);
-		page_add_new_anon_rmap(page, vma, vmf->address, false);
+		__page_add_new_anon_rmap(page, vma, vmf->address, false);
 		mem_cgroup_commit_charge(page, memcg, false, false);
 		__lru_cache_add_active_or_unevictable(page, vmf->vma_flags);
 	} else {
diff --git a/mm/rmap.c b/mm/rmap.c
index 6db729dc4c50..42d1ebed2b5b 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1136,7 +1136,7 @@ void do_page_add_anon_rmap(struct page *page,
 }
 
 /**
- * page_add_new_anon_rmap - add pte mapping to a new anonymous page
+ * __page_add_new_anon_rmap - add pte mapping to a new anonymous page
  * @page:	the page to add the mapping to
  * @vma:	the vm area in which the mapping is added
  * @address:	the user virtual address mapped
@@ -1146,12 +1146,11 @@ void do_page_add_anon_rmap(struct page *page,
  * This means the inc-and-test can be bypassed.
  * Page does not have to be locked.
  */
-void page_add_new_anon_rmap(struct page *page,
+void __page_add_new_anon_rmap(struct page *page,
 	struct vm_area_struct *vma, unsigned long address, bool compound)
 {
 	int nr = compound ? hpage_nr_pages(page) : 1;
 
-	VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
 	__SetPageSwapBacked(page);
 	if (compound) {
 		VM_BUG_ON_PAGE(!PageTransHuge(page), page);
-- 
2.7.4

[PATCH v11 18/26] mm: protect mm_rb tree with a rwlock

[Laurent Dufour]

This change is inspired by the Peter's proposal patch [1] which was
protecting the VMA using SRCU. Unfortunately, SRCU is not scaling well in
that particular case, and it is introducing major performance degradation
due to excessive scheduling operations.

To allow access to the mm_rb tree without grabbing the mmap_sem, this patch
is protecting it access using a rwlock.  As the mm_rb tree is a O(log n)
search it is safe to protect it using such a lock.  The VMA cache is not
protected by the new rwlock and it should not be used without holding the
mmap_sem.

To allow the picked VMA structure to be used once the rwlock is released, a
use count is added to the VMA structure. When the VMA is allocated it is
set to 1.  Each time the VMA is picked with the rwlock held its use count
is incremented. Each time the VMA is released it is decremented. When the
use count hits zero, this means that the VMA is no more used and should be
freed.

This patch is preparing for 2 kind of VMA access :
 - as usual, under the control of the mmap_sem,
 - without holding the mmap_sem for the speculative page fault handler.

Access done under the control the mmap_sem doesn't require to grab the
rwlock to protect read access to the mm_rb tree, but access in write must
be done under the protection of the rwlock too. This affects inserting and
removing of elements in the RB tree.

The patch is introducing 2 new functions:
 - vma_get() to find a VMA based on an address by holding the new rwlock.
 - vma_put() to release the VMA when its no more used.
These services are designed to be used when access are made to the RB tree
without holding the mmap_sem.

When a VMA is removed from the RB tree, its vma->vm_rb field is cleared and
we rely on the WMB done when releasing the rwlock to serialize the write
with the RMB done in a later patch to check for the VMA's validity.

When free_vma is called, the file associated with the VMA is closed
immediately, but the policy and the file structure remained in used until
the VMA's use count reach 0, which may happens later when exiting an
in progress speculative page fault.

[1] https://patchwork.kernel.org/patch/5108281/

Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/mm.h       |   1 +
 include/linux/mm_types.h |   4 ++
 kernel/fork.c            |   3 ++
 mm/init-mm.c             |   3 ++
 mm/internal.h            |   6 +++
 mm/mmap.c                | 115 +++++++++++++++++++++++++++++++++++------------
 6 files changed, 104 insertions(+), 28 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index bcebec117d4d..05cbba70104b 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1314,6 +1314,7 @@ static inline void INIT_VMA(struct vm_area_struct *vma)
 	INIT_LIST_HEAD(&vma->anon_vma_chain);
 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
 	seqcount_init(&vma->vm_sequence);
+	atomic_set(&vma->vm_ref_count, 1);
 #endif
 }
 
diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index fb5962308183..b16ba02f7fd6 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -337,6 +337,7 @@ struct vm_area_struct {
 	struct vm_userfaultfd_ctx vm_userfaultfd_ctx;
 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
 	seqcount_t vm_sequence;
+	atomic_t vm_ref_count;		/* see vma_get(), vma_put() */
 #endif
 } __randomize_layout;
 
@@ -355,6 +356,9 @@ struct kioctx_table;
 struct mm_struct {
 	struct vm_area_struct *mmap;		/* list of VMAs */
 	struct rb_root mm_rb;
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	rwlock_t mm_rb_lock;
+#endif
 	u32 vmacache_seqnum;                   /* per-thread vmacache */
 #ifdef CONFIG_MMU
 	unsigned long (*get_unmapped_area) (struct file *filp,
diff --git a/kernel/fork.c b/kernel/fork.c
index 99198a02efe9..f1258c2ade09 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -907,6 +907,9 @@ static struct mm_struct *mm_init(struct mm_struct *mm, struct task_struct *p,
 	mm->mmap = NULL;
 	mm->mm_rb = RB_ROOT;
 	mm->vmacache_seqnum = 0;
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	rwlock_init(&mm->mm_rb_lock);
+#endif
 	atomic_set(&mm->mm_users, 1);
 	atomic_set(&mm->mm_count, 1);
 	init_rwsem(&mm->mmap_sem);
diff --git a/mm/init-mm.c b/mm/init-mm.c
index f0179c9c04c2..228134f5a336 100644
--- a/mm/init-mm.c
+++ b/mm/init-mm.c
@@ -17,6 +17,9 @@
 
 struct mm_struct init_mm = {
 	.mm_rb		= RB_ROOT,
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	.mm_rb_lock	= __RW_LOCK_UNLOCKED(init_mm.mm_rb_lock),
+#endif
 	.pgd		= swapper_pg_dir,
 	.mm_users	= ATOMIC_INIT(2),
 	.mm_count	= ATOMIC_INIT(1),
diff --git a/mm/internal.h b/mm/internal.h
index 62d8c34e63d5..fb2667b20f0a 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -40,6 +40,12 @@ void page_writeback_init(void);
 
 int do_swap_page(struct vm_fault *vmf);
 
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+extern struct vm_area_struct *get_vma(struct mm_struct *mm,
+				      unsigned long addr);
+extern void put_vma(struct vm_area_struct *vma);
+#endif
+
 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
 		unsigned long floor, unsigned long ceiling);
 
diff --git a/mm/mmap.c b/mm/mmap.c
index 2450860e3f8e..54d298a67047 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -169,6 +169,27 @@ void unlink_file_vma(struct vm_area_struct *vma)
 	}
 }
 
+static void __free_vma(struct vm_area_struct *vma)
+{
+	if (vma->vm_file)
+		fput(vma->vm_file);
+	mpol_put(vma_policy(vma));
+	kmem_cache_free(vm_area_cachep, vma);
+}
+
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+void put_vma(struct vm_area_struct *vma)
+{
+	if (atomic_dec_and_test(&vma->vm_ref_count))
+		__free_vma(vma);
+}
+#else
+static inline void put_vma(struct vm_area_struct *vma)
+{
+	__free_vma(vma);
+}
+#endif
+
 /*
  * Close a vm structure and free it, returning the next.
  */
@@ -179,10 +200,7 @@ static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
 	might_sleep();
 	if (vma->vm_ops && vma->vm_ops->close)
 		vma->vm_ops->close(vma);
-	if (vma->vm_file)
-		fput(vma->vm_file);
-	mpol_put(vma_policy(vma));
-	kmem_cache_free(vm_area_cachep, vma);
+	put_vma(vma);
 	return next;
 }
 
@@ -402,6 +420,14 @@ static void validate_mm(struct mm_struct *mm)
 #define validate_mm(mm) do { } while (0)
 #endif
 
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+#define mm_rb_write_lock(mm)	write_lock(&(mm)->mm_rb_lock)
+#define mm_rb_write_unlock(mm)	write_unlock(&(mm)->mm_rb_lock)
+#else
+#define mm_rb_write_lock(mm)	do { } while (0)
+#define mm_rb_write_unlock(mm)	do { } while (0)
+#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
+
 RB_DECLARE_CALLBACKS(static, vma_gap_callbacks, struct vm_area_struct, vm_rb,
 		     unsigned long, rb_subtree_gap, vma_compute_subtree_gap)
 
@@ -420,26 +446,37 @@ static void vma_gap_update(struct vm_area_struct *vma)
 }
 
 static inline void vma_rb_insert(struct vm_area_struct *vma,
-				 struct rb_root *root)
+				 struct mm_struct *mm)
 {
+	struct rb_root *root = &mm->mm_rb;
+
 	/* All rb_subtree_gap values must be consistent prior to insertion */
 	validate_mm_rb(root, NULL);
 
 	rb_insert_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
 }
 
-static void __vma_rb_erase(struct vm_area_struct *vma, struct rb_root *root)
+static void __vma_rb_erase(struct vm_area_struct *vma, struct mm_struct *mm)
 {
+	struct rb_root *root = &mm->mm_rb;
 	/*
 	 * Note rb_erase_augmented is a fairly large inline function,
 	 * so make sure we instantiate it only once with our desired
 	 * augmented rbtree callbacks.
 	 */
+	mm_rb_write_lock(mm);
 	rb_erase_augmented(&vma->vm_rb, root, &vma_gap_callbacks);
+	mm_rb_write_unlock(mm); /* wmb */
+
+	/*
+	 * Ensure the removal is complete before clearing the node.
+	 * Matched by vma_has_changed()/handle_speculative_fault().
+	 */
+	RB_CLEAR_NODE(&vma->vm_rb);
 }
 
 static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
-						struct rb_root *root,
+						struct mm_struct *mm,
 						struct vm_area_struct *ignore)
 {
 	/*
@@ -447,21 +484,21 @@ static __always_inline void vma_rb_erase_ignore(struct vm_area_struct *vma,
 	 * with the possible exception of the "next" vma being erased if
 	 * next->vm_start was reduced.
 	 */
-	validate_mm_rb(root, ignore);
+	validate_mm_rb(&mm->mm_rb, ignore);
 
-	__vma_rb_erase(vma, root);
+	__vma_rb_erase(vma, mm);
 }
 
 static __always_inline void vma_rb_erase(struct vm_area_struct *vma,
-					 struct rb_root *root)
+					 struct mm_struct *mm)
 {
 	/*
 	 * All rb_subtree_gap values must be consistent prior to erase,
 	 * with the possible exception of the vma being erased.
 	 */
-	validate_mm_rb(root, vma);
+	validate_mm_rb(&mm->mm_rb, vma);
 
-	__vma_rb_erase(vma, root);
+	__vma_rb_erase(vma, mm);
 }
 
 /*
@@ -576,10 +613,12 @@ void __vma_link_rb(struct mm_struct *mm, struct vm_area_struct *vma,
 	 * immediately update the gap to the correct value. Finally we
 	 * rebalance the rbtree after all augmented values have been set.
 	 */
+	mm_rb_write_lock(mm);
 	rb_link_node(&vma->vm_rb, rb_parent, rb_link);
 	vma->rb_subtree_gap = 0;
 	vma_gap_update(vma);
-	vma_rb_insert(vma, &mm->mm_rb);
+	vma_rb_insert(vma, mm);
+	mm_rb_write_unlock(mm);
 }
 
 static void __vma_link_file(struct vm_area_struct *vma)
@@ -655,7 +694,7 @@ static __always_inline void __vma_unlink_common(struct mm_struct *mm,
 {
 	struct vm_area_struct *next;
 
-	vma_rb_erase_ignore(vma, &mm->mm_rb, ignore);
+	vma_rb_erase_ignore(vma, mm, ignore);
 	next = vma->vm_next;
 	if (has_prev)
 		prev->vm_next = next;
@@ -932,16 +971,13 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 	}
 
 	if (remove_next) {
-		if (file) {
+		if (file)
 			uprobe_munmap(next, next->vm_start, next->vm_end);
-			fput(file);
-		}
 		if (next->anon_vma)
 			anon_vma_merge(vma, next);
 		mm->map_count--;
-		mpol_put(vma_policy(next));
 		vm_raw_write_end(next);
-		kmem_cache_free(vm_area_cachep, next);
+		put_vma(next);
 		/*
 		 * In mprotect's case 6 (see comments on vma_merge),
 		 * we must remove another next too. It would clutter
@@ -2199,15 +2235,11 @@ get_unmapped_area(struct file *file, unsigned long addr, unsigned long len,
 EXPORT_SYMBOL(get_unmapped_area);
 
 /* Look up the first VMA which satisfies  addr < vm_end,  NULL if none. */
-struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
+static struct vm_area_struct *__find_vma(struct mm_struct *mm,
+					 unsigned long addr)
 {
 	struct rb_node *rb_node;
-	struct vm_area_struct *vma;
-
-	/* Check the cache first. */
-	vma = vmacache_find(mm, addr);
-	if (likely(vma))
-		return vma;
+	struct vm_area_struct *vma = NULL;
 
 	rb_node = mm->mm_rb.rb_node;
 
@@ -2225,13 +2257,40 @@ struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
 			rb_node = rb_node->rb_right;
 	}
 
+	return vma;
+}
+
+struct vm_area_struct *find_vma(struct mm_struct *mm, unsigned long addr)
+{
+	struct vm_area_struct *vma;
+
+	/* Check the cache first. */
+	vma = vmacache_find(mm, addr);
+	if (likely(vma))
+		return vma;
+
+	vma = __find_vma(mm, addr);
 	if (vma)
 		vmacache_update(addr, vma);
 	return vma;
 }
-
 EXPORT_SYMBOL(find_vma);
 
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+struct vm_area_struct *get_vma(struct mm_struct *mm, unsigned long addr)
+{
+	struct vm_area_struct *vma = NULL;
+
+	read_lock(&mm->mm_rb_lock);
+	vma = __find_vma(mm, addr);
+	if (vma)
+		atomic_inc(&vma->vm_ref_count);
+	read_unlock(&mm->mm_rb_lock);
+
+	return vma;
+}
+#endif
+
 /*
  * Same as find_vma, but also return a pointer to the previous VMA in *pprev.
  */
@@ -2599,7 +2658,7 @@ detach_vmas_to_be_unmapped(struct mm_struct *mm, struct vm_area_struct *vma,
 	insertion_point = (prev ? &prev->vm_next : &mm->mmap);
 	vma->vm_prev = NULL;
 	do {
-		vma_rb_erase(vma, &mm->mm_rb);
+		vma_rb_erase(vma, mm);
 		mm->map_count--;
 		tail_vma = vma;
 		vma = vma->vm_next;
-- 
2.7.4

[PATCH v11 19/26] mm: provide speculative fault infrastructure

[Laurent Dufour]

From: Peter Zijlstra <peterz@infradead.org>

Provide infrastructure to do a speculative fault (not holding
mmap_sem).

The not holding of mmap_sem means we can race against VMA
change/removal and page-table destruction. We use the SRCU VMA freeing
to keep the VMA around. We use the VMA seqcount to detect change
(including umapping / page-table deletion) and we use gup_fast() style
page-table walking to deal with page-table races.

Once we've obtained the page and are ready to update the PTE, we
validate if the state we started the fault with is still valid, if
not, we'll fail the fault with VM_FAULT_RETRY, otherwise we update the
PTE and we're done.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>

[Manage the newly introduced pte_spinlock() for speculative page
 fault to fail if the VMA is touched in our back]
[Rename vma_is_dead() to vma_has_changed() and declare it here]
[Fetch p4d and pud]
[Set vmd.sequence in __handle_mm_fault()]
[Abort speculative path when handle_userfault() has to be called]
[Add additional VMA's flags checks in handle_speculative_fault()]
[Clear FAULT_FLAG_ALLOW_RETRY in handle_speculative_fault()]
[Don't set vmf->pte and vmf->ptl if pte_map_lock() failed]
[Remove warning comment about waiting for !seq&1 since we don't want
 to wait]
[Remove warning about no huge page support, mention it explictly]
[Don't call do_fault() in the speculative path as __do_fault() calls
 vma->vm_ops->fault() which may want to release mmap_sem]
[Only vm_fault pointer argument for vma_has_changed()]
[Fix check against huge page, calling pmd_trans_huge()]
[Use READ_ONCE() when reading VMA's fields in the speculative path]
[Explicitly check for __HAVE_ARCH_PTE_SPECIAL as we can't support for
 processing done in vm_normal_page()]
[Check that vma->anon_vma is already set when starting the speculative
 path]
[Check for memory policy as we can't support MPOL_INTERLEAVE case due to
 the processing done in mpol_misplaced()]
[Don't support VMA growing up or down]
[Move check on vm_sequence just before calling handle_pte_fault()]
[Don't build SPF services if !CONFIG_SPECULATIVE_PAGE_FAULT]
[Add mem cgroup oom check]
[Use READ_ONCE to access p*d entries]
[Replace deprecated ACCESS_ONCE() by READ_ONCE() in vma_has_changed()]
[Don't fetch pte again in handle_pte_fault() when running the speculative
 path]
[Check PMD against concurrent collapsing operation]
[Try spin lock the pte during the speculative path to avoid deadlock with
 other CPU's invalidating the TLB and requiring this CPU to catch the
 inter processor's interrupt]
[Move define of FAULT_FLAG_SPECULATIVE here]
[Introduce __handle_speculative_fault() and add a check against
 mm->mm_users in handle_speculative_fault() defined in mm.h]
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/hugetlb_inline.h |   2 +-
 include/linux/mm.h             |  30 ++++
 include/linux/pagemap.h        |   4 +-
 mm/internal.h                  |  16 +-
 mm/memory.c                    | 340 ++++++++++++++++++++++++++++++++++++++++-
 5 files changed, 385 insertions(+), 7 deletions(-)

diff --git a/include/linux/hugetlb_inline.h b/include/linux/hugetlb_inline.h
index 0660a03d37d9..9e25283d6fc9 100644
--- a/include/linux/hugetlb_inline.h
+++ b/include/linux/hugetlb_inline.h
@@ -8,7 +8,7 @@
 
 static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
 {
-	return !!(vma->vm_flags & VM_HUGETLB);
+	return !!(READ_ONCE(vma->vm_flags) & VM_HUGETLB);
 }
 
 #else
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 05cbba70104b..31acf98a7d92 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -315,6 +315,7 @@ extern pgprot_t protection_map[16];
 #define FAULT_FLAG_USER		0x40	/* The fault originated in userspace */
 #define FAULT_FLAG_REMOTE	0x80	/* faulting for non current tsk/mm */
 #define FAULT_FLAG_INSTRUCTION  0x100	/* The fault was during an instruction fetch */
+#define FAULT_FLAG_SPECULATIVE	0x200	/* Speculative fault, not holding mmap_sem */
 
 #define FAULT_FLAG_TRACE \
 	{ FAULT_FLAG_WRITE,		"WRITE" }, \
@@ -343,6 +344,10 @@ struct vm_fault {
 	gfp_t gfp_mask;			/* gfp mask to be used for allocations */
 	pgoff_t pgoff;			/* Logical page offset based on vma */
 	unsigned long address;		/* Faulting virtual address */
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	unsigned int sequence;
+	pmd_t orig_pmd;			/* value of PMD at the time of fault */
+#endif
 	pmd_t *pmd;			/* Pointer to pmd entry matching
 					 * the 'address' */
 	pud_t *pud;			/* Pointer to pud entry matching
@@ -1415,6 +1420,31 @@ int invalidate_inode_page(struct page *page);
 #ifdef CONFIG_MMU
 extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
 		unsigned int flags);
+
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+extern int __handle_speculative_fault(struct mm_struct *mm,
+				      unsigned long address,
+				      unsigned int flags);
+static inline int handle_speculative_fault(struct mm_struct *mm,
+					   unsigned long address,
+					   unsigned int flags)
+{
+	/*
+	 * Try speculative page fault for multithreaded user space task only.
+	 */
+	if (!(flags & FAULT_FLAG_USER) || atomic_read(&mm->mm_users) == 1)
+		return VM_FAULT_RETRY;
+	return __handle_speculative_fault(mm, address, flags);
+}
+#else
+static inline int handle_speculative_fault(struct mm_struct *mm,
+					   unsigned long address,
+					   unsigned int flags)
+{
+	return VM_FAULT_RETRY;
+}
+#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
+
 extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
 			    unsigned long address, unsigned int fault_flags,
 			    bool *unlocked);
diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
index b1bd2186e6d2..6e2aa4e79af7 100644
--- a/include/linux/pagemap.h
+++ b/include/linux/pagemap.h
@@ -456,8 +456,8 @@ static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
 	pgoff_t pgoff;
 	if (unlikely(is_vm_hugetlb_page(vma)))
 		return linear_hugepage_index(vma, address);
-	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
-	pgoff += vma->vm_pgoff;
+	pgoff = (address - READ_ONCE(vma->vm_start)) >> PAGE_SHIFT;
+	pgoff += READ_ONCE(vma->vm_pgoff);
 	return pgoff;
 }
 
diff --git a/mm/internal.h b/mm/internal.h
index fb2667b20f0a..10b188c87fa4 100644
--- a/mm/internal.h
+++ b/mm/internal.h
@@ -44,7 +44,21 @@ int do_swap_page(struct vm_fault *vmf);
 extern struct vm_area_struct *get_vma(struct mm_struct *mm,
 				      unsigned long addr);
 extern void put_vma(struct vm_area_struct *vma);
-#endif
+
+static inline bool vma_has_changed(struct vm_fault *vmf)
+{
+	int ret = RB_EMPTY_NODE(&vmf->vma->vm_rb);
+	unsigned int seq = READ_ONCE(vmf->vma->vm_sequence.sequence);
+
+	/*
+	 * Matches both the wmb in write_seqlock_{begin,end}() and
+	 * the wmb in vma_rb_erase().
+	 */
+	smp_rmb();
+
+	return ret || seq != vmf->sequence;
+}
+#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
 
 void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
 		unsigned long floor, unsigned long ceiling);
diff --git a/mm/memory.c b/mm/memory.c
index ab32b0b4bd69..7bbbb8c7b9cd 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -769,7 +769,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
 	if (page)
 		dump_page(page, "bad pte");
 	pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
-		 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
+		 (void *)addr, READ_ONCE(vma->vm_flags), vma->anon_vma,
+		 mapping, index);
 	pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
 		 vma->vm_file,
 		 vma->vm_ops ? vma->vm_ops->fault : NULL,
@@ -2306,6 +2307,118 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
 }
 EXPORT_SYMBOL_GPL(apply_to_page_range);
 
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+static bool pte_spinlock(struct vm_fault *vmf)
+{
+	bool ret = false;
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	pmd_t pmdval;
+#endif
+
+	/* Check if vma is still valid */
+	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
+		vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
+		spin_lock(vmf->ptl);
+		return true;
+	}
+
+again:
+	local_irq_disable();
+	if (vma_has_changed(vmf))
+		goto out;
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	/*
+	 * We check if the pmd value is still the same to ensure that there
+	 * is not a huge collapse operation in progress in our back.
+	 */
+	pmdval = READ_ONCE(*vmf->pmd);
+	if (!pmd_same(pmdval, vmf->orig_pmd))
+		goto out;
+#endif
+
+	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
+	if (unlikely(!spin_trylock(vmf->ptl))) {
+		local_irq_enable();
+		goto again;
+	}
+
+	if (vma_has_changed(vmf)) {
+		spin_unlock(vmf->ptl);
+		goto out;
+	}
+
+	ret = true;
+out:
+	local_irq_enable();
+	return ret;
+}
+
+static bool pte_map_lock(struct vm_fault *vmf)
+{
+	bool ret = false;
+	pte_t *pte;
+	spinlock_t *ptl;
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	pmd_t pmdval;
+#endif
+
+	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
+		vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
+					       vmf->address, &vmf->ptl);
+		return true;
+	}
+
+	/*
+	 * The first vma_has_changed() guarantees the page-tables are still
+	 * valid, having IRQs disabled ensures they stay around, hence the
+	 * second vma_has_changed() to make sure they are still valid once
+	 * we've got the lock. After that a concurrent zap_pte_range() will
+	 * block on the PTL and thus we're safe.
+	 */
+again:
+	local_irq_disable();
+	if (vma_has_changed(vmf))
+		goto out;
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+	/*
+	 * We check if the pmd value is still the same to ensure that there
+	 * is not a huge collapse operation in progress in our back.
+	 */
+	pmdval = READ_ONCE(*vmf->pmd);
+	if (!pmd_same(pmdval, vmf->orig_pmd))
+		goto out;
+#endif
+
+	/*
+	 * Same as pte_offset_map_lock() except that we call
+	 * spin_trylock() in place of spin_lock() to avoid race with
+	 * unmap path which may have the lock and wait for this CPU
+	 * to invalidate TLB but this CPU has irq disabled.
+	 * Since we are in a speculative patch, accept it could fail
+	 */
+	ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
+	pte = pte_offset_map(vmf->pmd, vmf->address);
+	if (unlikely(!spin_trylock(ptl))) {
+		pte_unmap(pte);
+		local_irq_enable();
+		goto again;
+	}
+
+	if (vma_has_changed(vmf)) {
+		pte_unmap_unlock(pte, ptl);
+		goto out;
+	}
+
+	vmf->pte = pte;
+	vmf->ptl = ptl;
+	ret = true;
+out:
+	local_irq_enable();
+	return ret;
+}
+#else
 static inline bool pte_spinlock(struct vm_fault *vmf)
 {
 	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
@@ -2319,6 +2432,7 @@ static inline bool pte_map_lock(struct vm_fault *vmf)
 				       vmf->address, &vmf->ptl);
 	return true;
 }
+#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
 
 /*
  * handle_pte_fault chooses page fault handler according to an entry which was
@@ -3208,6 +3322,14 @@ static int do_anonymous_page(struct vm_fault *vmf)
 		ret = check_stable_address_space(vma->vm_mm);
 		if (ret)
 			goto unlock;
+		/*
+		 * Don't call the userfaultfd during the speculative path.
+		 * We already checked for the VMA to not be managed through
+		 * userfaultfd, but it may be set in our back once we have lock
+		 * the pte. In such a case we can ignore it this time.
+		 */
+		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
+			goto setpte;
 		/* Deliver the page fault to userland, check inside PT lock */
 		if (userfaultfd_missing(vma)) {
 			pte_unmap_unlock(vmf->pte, vmf->ptl);
@@ -3249,7 +3371,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
 		goto unlock_and_release;
 
 	/* Deliver the page fault to userland, check inside PT lock */
-	if (userfaultfd_missing(vma)) {
+	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE) && userfaultfd_missing(vma)) {
 		pte_unmap_unlock(vmf->pte, vmf->ptl);
 		mem_cgroup_cancel_charge(page, memcg, false);
 		put_page(page);
@@ -3994,13 +4116,22 @@ static int handle_pte_fault(struct vm_fault *vmf)
 
 	if (unlikely(pmd_none(*vmf->pmd))) {
 		/*
+		 * In the case of the speculative page fault handler we abort
+		 * the speculative path immediately as the pmd is probably
+		 * in the way to be converted in a huge one. We will try
+		 * again holding the mmap_sem (which implies that the collapse
+		 * operation is done).
+		 */
+		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
+			return VM_FAULT_RETRY;
+		/*
 		 * Leave __pte_alloc() until later: because vm_ops->fault may
 		 * want to allocate huge page, and if we expose page table
 		 * for an instant, it will be difficult to retract from
 		 * concurrent faults and from rmap lookups.
 		 */
 		vmf->pte = NULL;
-	} else {
+	} else if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
 		/* See comment in pte_alloc_one_map() */
 		if (pmd_devmap_trans_unstable(vmf->pmd))
 			return 0;
@@ -4009,6 +4140,9 @@ static int handle_pte_fault(struct vm_fault *vmf)
 		 * pmd from under us anymore at this point because we hold the
 		 * mmap_sem read mode and khugepaged takes it in write mode.
 		 * So now it's safe to run pte_offset_map().
+		 * This is not applicable to the speculative page fault handler
+		 * but in that case, the pte is fetched earlier in
+		 * handle_speculative_fault().
 		 */
 		vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
 		vmf->orig_pte = *vmf->pte;
@@ -4031,6 +4165,8 @@ static int handle_pte_fault(struct vm_fault *vmf)
 	if (!vmf->pte) {
 		if (vma_is_anonymous(vmf->vma))
 			return do_anonymous_page(vmf);
+		else if (vmf->flags & FAULT_FLAG_SPECULATIVE)
+			return VM_FAULT_RETRY;
 		else
 			return do_fault(vmf);
 	}
@@ -4128,6 +4264,9 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
 	vmf.pmd = pmd_alloc(mm, vmf.pud, address);
 	if (!vmf.pmd)
 		return VM_FAULT_OOM;
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	vmf.sequence = raw_read_seqcount(&vma->vm_sequence);
+#endif
 	if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
 		ret = create_huge_pmd(&vmf);
 		if (!(ret & VM_FAULT_FALLBACK))
@@ -4161,6 +4300,201 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
 	return handle_pte_fault(&vmf);
 }
 
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+/*
+ * Tries to handle the page fault in a speculative way, without grabbing the
+ * mmap_sem.
+ */
+int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
+			       unsigned int flags)
+{
+	struct vm_fault vmf = {
+		.address = address,
+	};
+	pgd_t *pgd, pgdval;
+	p4d_t *p4d, p4dval;
+	pud_t pudval;
+	int seq, ret = VM_FAULT_RETRY;
+	struct vm_area_struct *vma;
+#ifdef CONFIG_NUMA
+	struct mempolicy *pol;
+#endif
+
+	/* Clear flags that may lead to release the mmap_sem to retry */
+	flags &= ~(FAULT_FLAG_ALLOW_RETRY|FAULT_FLAG_KILLABLE);
+	flags |= FAULT_FLAG_SPECULATIVE;
+
+	vma = get_vma(mm, address);
+	if (!vma)
+		return ret;
+
+	seq = raw_read_seqcount(&vma->vm_sequence); /* rmb <-> seqlock,vma_rb_erase() */
+	if (seq & 1)
+		goto out_put;
+
+	/*
+	 * Can't call vm_ops service has we don't know what they would do
+	 * with the VMA.
+	 * This include huge page from hugetlbfs.
+	 */
+	if (vma->vm_ops)
+		goto out_put;
+
+	/*
+	 * __anon_vma_prepare() requires the mmap_sem to be held
+	 * because vm_next and vm_prev must be safe. This can't be guaranteed
+	 * in the speculative path.
+	 */
+	if (unlikely(!vma->anon_vma))
+		goto out_put;
+
+	vmf.vma_flags = READ_ONCE(vma->vm_flags);
+	vmf.vma_page_prot = READ_ONCE(vma->vm_page_prot);
+
+	/* Can't call userland page fault handler in the speculative path */
+	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING))
+		goto out_put;
+
+	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP)
+		/*
+		 * This could be detected by the check address against VMA's
+		 * boundaries but we want to trace it as not supported instead
+		 * of changed.
+		 */
+		goto out_put;
+
+	if (address < READ_ONCE(vma->vm_start)
+	    || READ_ONCE(vma->vm_end) <= address)
+		goto out_put;
+
+	if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
+				       flags & FAULT_FLAG_INSTRUCTION,
+				       flags & FAULT_FLAG_REMOTE)) {
+		ret = VM_FAULT_SIGSEGV;
+		goto out_put;
+	}
+
+	/* This is one is required to check that the VMA has write access set */
+	if (flags & FAULT_FLAG_WRITE) {
+		if (unlikely(!(vmf.vma_flags & VM_WRITE))) {
+			ret = VM_FAULT_SIGSEGV;
+			goto out_put;
+		}
+	} else if (unlikely(!(vmf.vma_flags & (VM_READ|VM_EXEC|VM_WRITE)))) {
+		ret = VM_FAULT_SIGSEGV;
+		goto out_put;
+	}
+
+#ifdef CONFIG_NUMA
+	/*
+	 * MPOL_INTERLEAVE implies additional checks in
+	 * mpol_misplaced() which are not compatible with the
+	 *speculative page fault processing.
+	 */
+	pol = __get_vma_policy(vma, address);
+	if (!pol)
+		pol = get_task_policy(current);
+	if (pol && pol->mode == MPOL_INTERLEAVE)
+		goto out_put;
+#endif
+
+	/*
+	 * Do a speculative lookup of the PTE entry.
+	 */
+	local_irq_disable();
+	pgd = pgd_offset(mm, address);
+	pgdval = READ_ONCE(*pgd);
+	if (pgd_none(pgdval) || unlikely(pgd_bad(pgdval)))
+		goto out_walk;
+
+	p4d = p4d_offset(pgd, address);
+	p4dval = READ_ONCE(*p4d);
+	if (p4d_none(p4dval) || unlikely(p4d_bad(p4dval)))
+		goto out_walk;
+
+	vmf.pud = pud_offset(p4d, address);
+	pudval = READ_ONCE(*vmf.pud);
+	if (pud_none(pudval) || unlikely(pud_bad(pudval)))
+		goto out_walk;
+
+	/* Huge pages at PUD level are not supported. */
+	if (unlikely(pud_trans_huge(pudval)))
+		goto out_walk;
+
+	vmf.pmd = pmd_offset(vmf.pud, address);
+	vmf.orig_pmd = READ_ONCE(*vmf.pmd);
+	/*
+	 * pmd_none could mean that a hugepage collapse is in progress
+	 * in our back as collapse_huge_page() mark it before
+	 * invalidating the pte (which is done once the IPI is catched
+	 * by all CPU and we have interrupt disabled).
+	 * For this reason we cannot handle THP in a speculative way since we
+	 * can't safely indentify an in progress collapse operation done in our
+	 * back on that PMD.
+	 * Regarding the order of the following checks, see comment in
+	 * pmd_devmap_trans_unstable()
+	 */
+	if (unlikely(pmd_devmap(vmf.orig_pmd) ||
+		     pmd_none(vmf.orig_pmd) || pmd_trans_huge(vmf.orig_pmd) ||
+		     is_swap_pmd(vmf.orig_pmd)))
+		goto out_walk;
+
+	/*
+	 * The above does not allocate/instantiate page-tables because doing so
+	 * would lead to the possibility of instantiating page-tables after
+	 * free_pgtables() -- and consequently leaking them.
+	 *
+	 * The result is that we take at least one !speculative fault per PMD
+	 * in order to instantiate it.
+	 */
+
+	vmf.pte = pte_offset_map(vmf.pmd, address);
+	vmf.orig_pte = READ_ONCE(*vmf.pte);
+	barrier(); /* See comment in handle_pte_fault() */
+	if (pte_none(vmf.orig_pte)) {
+		pte_unmap(vmf.pte);
+		vmf.pte = NULL;
+	}
+
+	vmf.vma = vma;
+	vmf.pgoff = linear_page_index(vma, address);
+	vmf.gfp_mask = __get_fault_gfp_mask(vma);
+	vmf.sequence = seq;
+	vmf.flags = flags;
+
+	local_irq_enable();
+
+	/*
+	 * We need to re-validate the VMA after checking the bounds, otherwise
+	 * we might have a false positive on the bounds.
+	 */
+	if (read_seqcount_retry(&vma->vm_sequence, seq))
+		goto out_put;
+
+	mem_cgroup_oom_enable();
+	ret = handle_pte_fault(&vmf);
+	mem_cgroup_oom_disable();
+
+	put_vma(vma);
+
+	/*
+	 * The task may have entered a memcg OOM situation but
+	 * if the allocation error was handled gracefully (no
+	 * VM_FAULT_OOM), there is no need to kill anything.
+	 * Just clean up the OOM state peacefully.
+	 */
+	if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
+		mem_cgroup_oom_synchronize(false);
+	return ret;
+
+out_walk:
+	local_irq_enable();
+out_put:
+	put_vma(vma);
+	return ret;
+}
+#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
+
 /*
  * By the time we get here, we already hold the mm semaphore
  *
-- 
2.7.4

[PATCH v11 20/26] mm: adding speculative page fault failure trace events

[Laurent Dufour]

This patch a set of new trace events to collect the speculative page fault
event failures.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/trace/events/pagefault.h | 80 ++++++++++++++++++++++++++++++++++++++++
 mm/memory.c                      | 57 ++++++++++++++++++++++------
 2 files changed, 125 insertions(+), 12 deletions(-)
 create mode 100644 include/trace/events/pagefault.h

diff --git a/include/trace/events/pagefault.h b/include/trace/events/pagefault.h
new file mode 100644
index 000000000000..d9438f3e6bad
--- /dev/null
+++ b/include/trace/events/pagefault.h
@@ -0,0 +1,80 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM pagefault
+
+#if !defined(_TRACE_PAGEFAULT_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_PAGEFAULT_H
+
+#include <linux/tracepoint.h>
+#include <linux/mm.h>
+
+DECLARE_EVENT_CLASS(spf,
+
+	TP_PROTO(unsigned long caller,
+		 struct vm_area_struct *vma, unsigned long address),
+
+	TP_ARGS(caller, vma, address),
+
+	TP_STRUCT__entry(
+		__field(unsigned long, caller)
+		__field(unsigned long, vm_start)
+		__field(unsigned long, vm_end)
+		__field(unsigned long, address)
+	),
+
+	TP_fast_assign(
+		__entry->caller		= caller;
+		__entry->vm_start	= vma->vm_start;
+		__entry->vm_end		= vma->vm_end;
+		__entry->address	= address;
+	),
+
+	TP_printk("ip:%lx vma:%lx-%lx address:%lx",
+		  __entry->caller, __entry->vm_start, __entry->vm_end,
+		  __entry->address)
+);
+
+DEFINE_EVENT(spf, spf_vma_changed,
+
+	TP_PROTO(unsigned long caller,
+		 struct vm_area_struct *vma, unsigned long address),
+
+	TP_ARGS(caller, vma, address)
+);
+
+DEFINE_EVENT(spf, spf_vma_noanon,
+
+	TP_PROTO(unsigned long caller,
+		 struct vm_area_struct *vma, unsigned long address),
+
+	TP_ARGS(caller, vma, address)
+);
+
+DEFINE_EVENT(spf, spf_vma_notsup,
+
+	TP_PROTO(unsigned long caller,
+		 struct vm_area_struct *vma, unsigned long address),
+
+	TP_ARGS(caller, vma, address)
+);
+
+DEFINE_EVENT(spf, spf_vma_access,
+
+	TP_PROTO(unsigned long caller,
+		 struct vm_area_struct *vma, unsigned long address),
+
+	TP_ARGS(caller, vma, address)
+);
+
+DEFINE_EVENT(spf, spf_pmd_changed,
+
+	TP_PROTO(unsigned long caller,
+		 struct vm_area_struct *vma, unsigned long address),
+
+	TP_ARGS(caller, vma, address)
+);
+
+#endif /* _TRACE_PAGEFAULT_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
diff --git a/mm/memory.c b/mm/memory.c
index 7bbbb8c7b9cd..30433bde32f2 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -80,6 +80,9 @@
 
 #include "internal.h"
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/pagefault.h>
+
 #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST)
 #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
 #endif
@@ -2324,8 +2327,10 @@ static bool pte_spinlock(struct vm_fault *vmf)
 
 again:
 	local_irq_disable();
-	if (vma_has_changed(vmf))
+	if (vma_has_changed(vmf)) {
+		trace_spf_vma_changed(_RET_IP_, vmf->vma, vmf->address);
 		goto out;
+	}
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 	/*
@@ -2333,8 +2338,10 @@ static bool pte_spinlock(struct vm_fault *vmf)
 	 * is not a huge collapse operation in progress in our back.
 	 */
 	pmdval = READ_ONCE(*vmf->pmd);
-	if (!pmd_same(pmdval, vmf->orig_pmd))
+	if (!pmd_same(pmdval, vmf->orig_pmd)) {
+		trace_spf_pmd_changed(_RET_IP_, vmf->vma, vmf->address);
 		goto out;
+	}
 #endif
 
 	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
@@ -2345,6 +2352,7 @@ static bool pte_spinlock(struct vm_fault *vmf)
 
 	if (vma_has_changed(vmf)) {
 		spin_unlock(vmf->ptl);
+		trace_spf_vma_changed(_RET_IP_, vmf->vma, vmf->address);
 		goto out;
 	}
 
@@ -2378,8 +2386,10 @@ static bool pte_map_lock(struct vm_fault *vmf)
 	 */
 again:
 	local_irq_disable();
-	if (vma_has_changed(vmf))
+	if (vma_has_changed(vmf)) {
+		trace_spf_vma_changed(_RET_IP_, vmf->vma, vmf->address);
 		goto out;
+	}
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 	/*
@@ -2387,8 +2397,10 @@ static bool pte_map_lock(struct vm_fault *vmf)
 	 * is not a huge collapse operation in progress in our back.
 	 */
 	pmdval = READ_ONCE(*vmf->pmd);
-	if (!pmd_same(pmdval, vmf->orig_pmd))
+	if (!pmd_same(pmdval, vmf->orig_pmd)) {
+		trace_spf_pmd_changed(_RET_IP_, vmf->vma, vmf->address);
 		goto out;
+	}
 #endif
 
 	/*
@@ -2408,6 +2420,7 @@ static bool pte_map_lock(struct vm_fault *vmf)
 
 	if (vma_has_changed(vmf)) {
 		pte_unmap_unlock(pte, ptl);
+		trace_spf_vma_changed(_RET_IP_, vmf->vma, vmf->address);
 		goto out;
 	}
 
@@ -4329,47 +4342,60 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
 		return ret;
 
 	seq = raw_read_seqcount(&vma->vm_sequence); /* rmb <-> seqlock,vma_rb_erase() */
-	if (seq & 1)
+	if (seq & 1) {
+		trace_spf_vma_changed(_RET_IP_, vma, address);
 		goto out_put;
+	}
 
 	/*
 	 * Can't call vm_ops service has we don't know what they would do
 	 * with the VMA.
 	 * This include huge page from hugetlbfs.
 	 */
-	if (vma->vm_ops)
+	if (vma->vm_ops) {
+		trace_spf_vma_notsup(_RET_IP_, vma, address);
 		goto out_put;
+	}
 
 	/*
 	 * __anon_vma_prepare() requires the mmap_sem to be held
 	 * because vm_next and vm_prev must be safe. This can't be guaranteed
 	 * in the speculative path.
 	 */
-	if (unlikely(!vma->anon_vma))
+	if (unlikely(!vma->anon_vma)) {
+		trace_spf_vma_notsup(_RET_IP_, vma, address);
 		goto out_put;
+	}
 
 	vmf.vma_flags = READ_ONCE(vma->vm_flags);
 	vmf.vma_page_prot = READ_ONCE(vma->vm_page_prot);
 
 	/* Can't call userland page fault handler in the speculative path */
-	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING))
+	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING)) {
+		trace_spf_vma_notsup(_RET_IP_, vma, address);
 		goto out_put;
+	}
 
-	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP)
+	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP) {
 		/*
 		 * This could be detected by the check address against VMA's
 		 * boundaries but we want to trace it as not supported instead
 		 * of changed.
 		 */
+		trace_spf_vma_notsup(_RET_IP_, vma, address);
 		goto out_put;
+	}
 
 	if (address < READ_ONCE(vma->vm_start)
-	    || READ_ONCE(vma->vm_end) <= address)
+	    || READ_ONCE(vma->vm_end) <= address) {
+		trace_spf_vma_changed(_RET_IP_, vma, address);
 		goto out_put;
+	}
 
 	if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
 				       flags & FAULT_FLAG_INSTRUCTION,
 				       flags & FAULT_FLAG_REMOTE)) {
+		trace_spf_vma_access(_RET_IP_, vma, address);
 		ret = VM_FAULT_SIGSEGV;
 		goto out_put;
 	}
@@ -4377,10 +4403,12 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
 	/* This is one is required to check that the VMA has write access set */
 	if (flags & FAULT_FLAG_WRITE) {
 		if (unlikely(!(vmf.vma_flags & VM_WRITE))) {
+			trace_spf_vma_access(_RET_IP_, vma, address);
 			ret = VM_FAULT_SIGSEGV;
 			goto out_put;
 		}
 	} else if (unlikely(!(vmf.vma_flags & (VM_READ|VM_EXEC|VM_WRITE)))) {
+		trace_spf_vma_access(_RET_IP_, vma, address);
 		ret = VM_FAULT_SIGSEGV;
 		goto out_put;
 	}
@@ -4394,8 +4422,10 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
 	pol = __get_vma_policy(vma, address);
 	if (!pol)
 		pol = get_task_policy(current);
-	if (pol && pol->mode == MPOL_INTERLEAVE)
+	if (pol && pol->mode == MPOL_INTERLEAVE) {
+		trace_spf_vma_notsup(_RET_IP_, vma, address);
 		goto out_put;
+	}
 #endif
 
 	/*
@@ -4468,8 +4498,10 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
 	 * We need to re-validate the VMA after checking the bounds, otherwise
 	 * we might have a false positive on the bounds.
 	 */
-	if (read_seqcount_retry(&vma->vm_sequence, seq))
+	if (read_seqcount_retry(&vma->vm_sequence, seq)) {
+		trace_spf_vma_changed(_RET_IP_, vma, address);
 		goto out_put;
+	}
 
 	mem_cgroup_oom_enable();
 	ret = handle_pte_fault(&vmf);
@@ -4488,6 +4520,7 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
 	return ret;
 
 out_walk:
+	trace_spf_vma_notsup(_RET_IP_, vma, address);
 	local_irq_enable();
 out_put:
 	put_vma(vma);
-- 
2.7.4

[PATCH v11 21/26] perf: add a speculative page fault sw event

[Laurent Dufour]

Add a new software event to count succeeded speculative page faults.

Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/uapi/linux/perf_event.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index b8e288a1f740..e2b74c055f51 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -112,6 +112,7 @@ enum perf_sw_ids {
 	PERF_COUNT_SW_EMULATION_FAULTS		= 8,
 	PERF_COUNT_SW_DUMMY			= 9,
 	PERF_COUNT_SW_BPF_OUTPUT		= 10,
+	PERF_COUNT_SW_SPF			= 11,
 
 	PERF_COUNT_SW_MAX,			/* non-ABI */
 };
-- 
2.7.4

[PATCH v11 22/26] perf tools: add support for the SPF perf event

[Laurent Dufour]

Add support for the new speculative faults event.

Acked-by: David Rientjes <rientjes@google.com>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 tools/include/uapi/linux/perf_event.h | 1 +
 tools/perf/util/evsel.c               | 1 +
 tools/perf/util/parse-events.c        | 4 ++++
 tools/perf/util/parse-events.l        | 1 +
 tools/perf/util/python.c              | 1 +
 5 files changed, 8 insertions(+)

diff --git a/tools/include/uapi/linux/perf_event.h b/tools/include/uapi/linux/perf_event.h
index b8e288a1f740..e2b74c055f51 100644
--- a/tools/include/uapi/linux/perf_event.h
+++ b/tools/include/uapi/linux/perf_event.h
@@ -112,6 +112,7 @@ enum perf_sw_ids {
 	PERF_COUNT_SW_EMULATION_FAULTS		= 8,
 	PERF_COUNT_SW_DUMMY			= 9,
 	PERF_COUNT_SW_BPF_OUTPUT		= 10,
+	PERF_COUNT_SW_SPF			= 11,
 
 	PERF_COUNT_SW_MAX,			/* non-ABI */
 };
diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
index 4cd2cf93f726..088ed45c68c1 100644
--- a/tools/perf/util/evsel.c
+++ b/tools/perf/util/evsel.c
@@ -429,6 +429,7 @@ const char *perf_evsel__sw_names[PERF_COUNT_SW_MAX] = {
 	"alignment-faults",
 	"emulation-faults",
 	"dummy",
+	"speculative-faults",
 };
 
 static const char *__perf_evsel__sw_name(u64 config)
diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c
index 2fb0272146d8..54719f566314 100644
--- a/tools/perf/util/parse-events.c
+++ b/tools/perf/util/parse-events.c
@@ -140,6 +140,10 @@ struct event_symbol event_symbols_sw[PERF_COUNT_SW_MAX] = {
 		.symbol = "bpf-output",
 		.alias  = "",
 	},
+	[PERF_COUNT_SW_SPF] = {
+		.symbol = "speculative-faults",
+		.alias	= "spf",
+	},
 };
 
 #define __PERF_EVENT_FIELD(config, name) \
diff --git a/tools/perf/util/parse-events.l b/tools/perf/util/parse-events.l
index a1a01b1ac8b8..86584d3a3068 100644
--- a/tools/perf/util/parse-events.l
+++ b/tools/perf/util/parse-events.l
@@ -308,6 +308,7 @@ emulation-faults				{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_EM
 dummy						{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_DUMMY); }
 duration_time					{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_DUMMY); }
 bpf-output					{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_BPF_OUTPUT); }
+speculative-faults|spf				{ return sym(yyscanner, PERF_TYPE_SOFTWARE, PERF_COUNT_SW_SPF); }
 
 	/*
 	 * We have to handle the kernel PMU event cycles-ct/cycles-t/mem-loads/mem-stores separately.
diff --git a/tools/perf/util/python.c b/tools/perf/util/python.c
index 863b61478edd..df4f7ff9bdff 100644
--- a/tools/perf/util/python.c
+++ b/tools/perf/util/python.c
@@ -1181,6 +1181,7 @@ static struct {
 	PERF_CONST(COUNT_SW_ALIGNMENT_FAULTS),
 	PERF_CONST(COUNT_SW_EMULATION_FAULTS),
 	PERF_CONST(COUNT_SW_DUMMY),
+	PERF_CONST(COUNT_SW_SPF),
 
 	PERF_CONST(SAMPLE_IP),
 	PERF_CONST(SAMPLE_TID),
-- 
2.7.4

[PATCH v11 23/26] mm: add speculative page fault vmstats

[Laurent Dufour]

Add speculative_pgfault vmstat counter to count successful speculative page
fault handling.

Also fixing a minor typo in include/linux/vm_event_item.h.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/vm_event_item.h | 3 +++
 mm/memory.c                   | 3 +++
 mm/vmstat.c                   | 5 ++++-
 3 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h
index 5c7f010676a7..a240acc09684 100644
--- a/include/linux/vm_event_item.h
+++ b/include/linux/vm_event_item.h
@@ -111,6 +111,9 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
 		SWAP_RA,
 		SWAP_RA_HIT,
 #endif
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+		SPECULATIVE_PGFAULT,
+#endif
 		NR_VM_EVENT_ITEMS
 };
 
diff --git a/mm/memory.c b/mm/memory.c
index 30433bde32f2..48e1cf0a54ef 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4509,6 +4509,9 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
 
 	put_vma(vma);
 
+	if (ret != VM_FAULT_RETRY)
+		count_vm_event(SPECULATIVE_PGFAULT);
+
 	/*
 	 * The task may have entered a memcg OOM situation but
 	 * if the allocation error was handled gracefully (no
diff --git a/mm/vmstat.c b/mm/vmstat.c
index a2b9518980ce..3af74498a969 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -1289,7 +1289,10 @@ const char * const vmstat_text[] = {
 	"swap_ra",
 	"swap_ra_hit",
 #endif
-#endif /* CONFIG_VM_EVENTS_COUNTERS */
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	"speculative_pgfault",
+#endif
+#endif /* CONFIG_VM_EVENT_COUNTERS */
 };
 #endif /* CONFIG_PROC_FS || CONFIG_SYSFS || CONFIG_NUMA */
 
-- 
2.7.4

[PATCH v11 24/26] x86/mm: add speculative pagefault handling

[Laurent Dufour]

From: Peter Zijlstra <peterz@infradead.org>

Try a speculative fault before acquiring mmap_sem, if it returns with
VM_FAULT_RETRY continue with the mmap_sem acquisition and do the
traditional fault.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>

[Clearing of FAULT_FLAG_ALLOW_RETRY is now done in
 handle_speculative_fault()]
[Retry with usual fault path in the case VM_ERROR is returned by
 handle_speculative_fault(). This allows signal to be delivered]
[Don't build SPF call if !CONFIG_SPECULATIVE_PAGE_FAULT]
[Handle memory protection key fault]
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 arch/x86/mm/fault.c | 27 +++++++++++++++++++++++++--
 1 file changed, 25 insertions(+), 2 deletions(-)

diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c
index fd84edf82252..11944bfc805a 100644
--- a/arch/x86/mm/fault.c
+++ b/arch/x86/mm/fault.c
@@ -1224,7 +1224,7 @@ __do_page_fault(struct pt_regs *regs, unsigned long error_code,
 	struct mm_struct *mm;
 	int fault, major = 0;
 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
-	u32 pkey;
+	u32 pkey, *pt_pkey = &pkey;
 
 	tsk = current;
 	mm = tsk->mm;
@@ -1314,6 +1314,27 @@ __do_page_fault(struct pt_regs *regs, unsigned long error_code,
 		flags |= FAULT_FLAG_INSTRUCTION;
 
 	/*
+	 * Do not try to do a speculative page fault if the fault was due to
+	 * protection keys since it can't be resolved.
+	 */
+	if (!(error_code & X86_PF_PK)) {
+		fault = handle_speculative_fault(mm, address, flags);
+		if (fault != VM_FAULT_RETRY) {
+			perf_sw_event(PERF_COUNT_SW_SPF, 1, regs, address);
+			/*
+			 * Do not advertise for the pkey value since we don't
+			 * know it.
+			 * This is not a matter as we checked for X86_PF_PK
+			 * earlier, so we should not handle pkey fault here,
+			 * but to be sure that mm_fault_error() callees will
+			 * not try to use it, we invalidate the pointer.
+			 */
+			pt_pkey = NULL;
+			goto done;
+		}
+	}
+
+	/*
 	 * When running in the kernel we expect faults to occur only to
 	 * addresses in user space.  All other faults represent errors in
 	 * the kernel and should generate an OOPS.  Unfortunately, in the
@@ -1427,8 +1448,10 @@ __do_page_fault(struct pt_regs *regs, unsigned long error_code,
 	}
 
 	up_read(&mm->mmap_sem);
+
+done:
 	if (unlikely(fault & VM_FAULT_ERROR)) {
-		mm_fault_error(regs, error_code, address, &pkey, fault);
+		mm_fault_error(regs, error_code, address, pt_pkey, fault);
 		return;
 	}
 
-- 
2.7.4

[PATCH v11 25/26] powerpc/mm: add speculative page fault

[Laurent Dufour]

This patch enable the speculative page fault on the PowerPC
architecture.

This will try a speculative page fault without holding the mmap_sem,
if it returns with VM_FAULT_RETRY, the mmap_sem is acquired and the
traditional page fault processing is done.

The speculative path is only tried for multithreaded process as there is no
risk of contention on the mmap_sem otherwise.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 arch/powerpc/mm/fault.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/arch/powerpc/mm/fault.c b/arch/powerpc/mm/fault.c
index ef268d5d9db7..d7b5742ffb2b 100644
--- a/arch/powerpc/mm/fault.c
+++ b/arch/powerpc/mm/fault.c
@@ -465,6 +465,21 @@ static int __do_page_fault(struct pt_regs *regs, unsigned long address,
 	if (is_exec)
 		flags |= FAULT_FLAG_INSTRUCTION;
 
+	/*
+	 * Try speculative page fault before grabbing the mmap_sem.
+	 * The Page fault is done if VM_FAULT_RETRY is not returned.
+	 * But if the memory protection keys are active, we don't know if the
+	 * fault is due to key mistmatch or due to a classic protection check.
+	 * To differentiate that, we will need the VMA we no more have, so
+	 * let's retry with the mmap_sem held.
+	 */
+	fault = handle_speculative_fault(mm, address, flags);
+	if (fault != VM_FAULT_RETRY && (IS_ENABLED(CONFIG_PPC_MEM_KEYS) &&
+					fault != VM_FAULT_SIGSEGV)) {
+		perf_sw_event(PERF_COUNT_SW_SPF, 1, regs, address);
+		goto done;
+	}
+
 	/* When running in the kernel we expect faults to occur only to
 	 * addresses in user space.  All other faults represent errors in the
 	 * kernel and should generate an OOPS.  Unfortunately, in the case of an
@@ -565,6 +580,7 @@ static int __do_page_fault(struct pt_regs *regs, unsigned long address,
 
 	up_read(&current->mm->mmap_sem);
 
+done:
 	if (unlikely(fault & VM_FAULT_ERROR))
 		return mm_fault_error(regs, address, fault);
 
-- 
2.7.4

[PATCH v11 26/26] arm64/mm: add speculative page fault

[Laurent Dufour]

From: Mahendran Ganesh <opensource.ganesh@gmail.com>

This patch enables the speculative page fault on the arm64
architecture.

I completed spf porting in 4.9. From the test result,
we can see app launching time improved by about 10% in average.
For the apps which have more than 50 threads, 15% or even more
improvement can be got.

Signed-off-by: Ganesh Mahendran <opensource.ganesh@gmail.com>

[handle_speculative_fault() is no more returning the vma pointer]
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 arch/arm64/mm/fault.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c
index 91c53a7d2575..fb9f840367f9 100644
--- a/arch/arm64/mm/fault.c
+++ b/arch/arm64/mm/fault.c
@@ -411,6 +411,16 @@ static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
 
 	/*
+	 * let's try a speculative page fault without grabbing the
+	 * mmap_sem.
+	 */
+	fault = handle_speculative_fault(mm, addr, mm_flags);
+	if (fault != VM_FAULT_RETRY) {
+		perf_sw_event(PERF_COUNT_SW_SPF, 1, regs, addr);
+		goto done;
+	}
+
+	/*
 	 * As per x86, we may deadlock here. However, since the kernel only
 	 * validly references user space from well defined areas of the code,
 	 * we can bug out early if this is from code which shouldn't.
@@ -460,6 +470,8 @@ static int __kprobes do_page_fault(unsigned long addr, unsigned int esr,
 	}
 	up_read(&mm->mmap_sem);
 
+done:
+
 	/*
 	 * Handle the "normal" (no error) case first.
 	 */
-- 
2.7.4

Re: [PATCH v11 01/26] mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT

[Randy Dunlap]

Hi,

On 05/17/2018 04:06 AM, Laurent Dufour wrote:
> This configuration variable will be used to build the code needed to
> handle speculative page fault.
> 
> By default it is turned off, and activated depending on architecture
> support, ARCH_HAS_PTE_SPECIAL, SMP and MMU.
> 
> The architecture support is needed since the speculative page fault handler
> is called from the architecture's page faulting code, and some code has to
> be added there to handle the speculative handler.
> 
> The dependency on ARCH_HAS_PTE_SPECIAL is required because vm_normal_page()
> does processing that is not compatible with the speculative handling in the
> case ARCH_HAS_PTE_SPECIAL is not set.
> 
> Suggested-by: Thomas Gleixner <tglx@linutronix.de>
> Suggested-by: David Rientjes <rientjes@google.com>
> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
> ---
>  mm/Kconfig | 22 ++++++++++++++++++++++
>  1 file changed, 22 insertions(+)
> 
> diff --git a/mm/Kconfig b/mm/Kconfig
> index 1d0888c5b97a..a38796276113 100644
> --- a/mm/Kconfig
> +++ b/mm/Kconfig
> @@ -761,3 +761,25 @@ config GUP_BENCHMARK
>  
>  config ARCH_HAS_PTE_SPECIAL
>  	bool
> +
> +config ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> +       def_bool n
> +
> +config SPECULATIVE_PAGE_FAULT
> +       bool "Speculative page faults"
> +       default y
> +       depends on ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> +       depends on ARCH_HAS_PTE_SPECIAL && MMU && SMP
> +       help
> +         Try to handle user space page faults without holding the mmap_sem.
> +
> +	 This should allow better concurrency for massively threaded process

	                                                             processes

> +	 since the page fault handler will not wait for other threads memory

	                                                      thread's

> +	 layout change to be done, assuming that this change is done in another
> +	 part of the process's memory space. This type of page fault is named
> +	 speculative page fault.
> +
> +	 If the speculative page fault fails because of a concurrency is

	                                     because a concurrency is

> +	 detected or because underlying PMD or PTE tables are not yet
> +	 allocating, it is failing its processing and a classic page fault

	 allocated, the speculative page fault fails and a classic page fault

> +	 is then tried.


Also, all of the help text (below the "help" line) should be indented by
1 tab + 2 spaces (in coding-style.rst).


-- 
~Randy

Re: [PATCH v11 01/26] mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT

[Matthew Wilcox]

On Thu, May 17, 2018 at 09:36:00AM -0700, Randy Dunlap wrote:
> > +	 If the speculative page fault fails because of a concurrency is
> 
> 	                                     because a concurrency is

While one can use concurrency as a noun, it sounds archaic to me.  I'd
rather:

	If the speculative page fault fails because a concurrent modification
	is detected or because underlying PMD or PTE tables are not yet

> > +	 detected or because underlying PMD or PTE tables are not yet
> > +	 allocating, it is failing its processing and a classic page fault
> 
> 	 allocated, the speculative page fault fails and a classic page fault
> 
> > +	 is then tried.

Re: [PATCH v11 01/26] mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT

[Randy Dunlap]

On 05/17/2018 10:19 AM, Matthew Wilcox wrote:
> On Thu, May 17, 2018 at 09:36:00AM -0700, Randy Dunlap wrote:
>>> +	 If the speculative page fault fails because of a concurrency is
>>
>> 	                                     because a concurrency is
> 
> While one can use concurrency as a noun, it sounds archaic to me.  I'd
> rather:
> 
> 	If the speculative page fault fails because a concurrent modification
> 	is detected or because underlying PMD or PTE tables are not yet

Yeah, OK.

>>> +	 detected or because underlying PMD or PTE tables are not yet
>>> +	 allocating, it is failing its processing and a classic page fault
>>
>> 	 allocated, the speculative page fault fails and a classic page fault
>>
>>> +	 is then tried.


-- 
~Randy

Re: [PATCH v11 01/26] mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT

[Laurent Dufour]

On 17/05/2018 19:19, Matthew Wilcox wrote:
> On Thu, May 17, 2018 at 09:36:00AM -0700, Randy Dunlap wrote:
>>> +	 If the speculative page fault fails because of a concurrency is
>>
>> 	                                     because a concurrency is
> 
> While one can use concurrency as a noun, it sounds archaic to me.  I'd
> rather:
> 
> 	If the speculative page fault fails because a concurrent modification
> 	is detected or because underlying PMD or PTE tables are not yet

Thanks Matthew, I'll do that.

> 
>>> +	 detected or because underlying PMD or PTE tables are not yet
>>> +	 allocating, it is failing its processing and a classic page fault
>>
>> 	 allocated, the speculative page fault fails and a classic page fault
>>
>>> +	 is then tried.
> 

Re: [PATCH v11 01/26] mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT

[Laurent Dufour]

On 17/05/2018 18:36, Randy Dunlap wrote:
> Hi,
> 
> On 05/17/2018 04:06 AM, Laurent Dufour wrote:
>> This configuration variable will be used to build the code needed to
>> handle speculative page fault.
>>
>> By default it is turned off, and activated depending on architecture
>> support, ARCH_HAS_PTE_SPECIAL, SMP and MMU.
>>
>> The architecture support is needed since the speculative page fault handler
>> is called from the architecture's page faulting code, and some code has to
>> be added there to handle the speculative handler.
>>
>> The dependency on ARCH_HAS_PTE_SPECIAL is required because vm_normal_page()
>> does processing that is not compatible with the speculative handling in the
>> case ARCH_HAS_PTE_SPECIAL is not set.
>>
>> Suggested-by: Thomas Gleixner <tglx@linutronix.de>
>> Suggested-by: David Rientjes <rientjes@google.com>
>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>> ---
>>  mm/Kconfig | 22 ++++++++++++++++++++++
>>  1 file changed, 22 insertions(+)
>>
>> diff --git a/mm/Kconfig b/mm/Kconfig
>> index 1d0888c5b97a..a38796276113 100644
>> --- a/mm/Kconfig
>> +++ b/mm/Kconfig
>> @@ -761,3 +761,25 @@ config GUP_BENCHMARK
>>  
>>  config ARCH_HAS_PTE_SPECIAL
>>  	bool
>> +
>> +config ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>> +       def_bool n
>> +
>> +config SPECULATIVE_PAGE_FAULT
>> +       bool "Speculative page faults"
>> +       default y
>> +       depends on ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>> +       depends on ARCH_HAS_PTE_SPECIAL && MMU && SMP
>> +       help
>> +         Try to handle user space page faults without holding the mmap_sem.
>> +
>> +	 This should allow better concurrency for massively threaded process
> 
> 	                                                             processes
> 
>> +	 since the page fault handler will not wait for other threads memory
> 
> 	                                                      thread's
> 
>> +	 layout change to be done, assuming that this change is done in another
>> +	 part of the process's memory space. This type of page fault is named
>> +	 speculative page fault.
>> +
>> +	 If the speculative page fault fails because of a concurrency is
> 
> 	                                     because a concurrency is
> 
>> +	 detected or because underlying PMD or PTE tables are not yet
>> +	 allocating, it is failing its processing and a classic page fault
> 
> 	 allocated, the speculative page fault fails and a classic page fault
> 
>> +	 is then tried.
> 
> 
> Also, all of the help text (below the "help" line) should be indented by
> 1 tab + 2 spaces (in coding-style.rst).

Thanks, Randy for reviewing my miserable English grammar.
I'll fix that and the indentation.

[FIX PATCH v11 01/26] mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT

[Laurent Dufour]

This configuration variable will be used to build the code needed to
handle speculative page fault.

By default it is turned off, and activated depending on architecture
support, ARCH_HAS_PTE_SPECIAL, SMP and MMU.

The architecture support is needed since the speculative page fault handler
is called from the architecture's page faulting code, and some code has to
be added there to handle the speculative handler.

The dependency on ARCH_HAS_PTE_SPECIAL is required because vm_normal_page()
does processing that is not compatible with the speculative handling in the
case ARCH_HAS_PTE_SPECIAL is not set.

Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Suggested-by: David Rientjes <rientjes@google.com>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 mm/Kconfig | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)

diff --git a/mm/Kconfig b/mm/Kconfig
index 1d0888c5b97a..d958fd8ce73a 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -761,3 +761,25 @@ config GUP_BENCHMARK
 
 config ARCH_HAS_PTE_SPECIAL
 	bool
+
+config ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
+       def_bool n
+
+config SPECULATIVE_PAGE_FAULT
+	bool "Speculative page faults"
+	default y
+	depends on ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
+	depends on ARCH_HAS_PTE_SPECIAL && MMU && SMP
+	help
+	  Try to handle user space page faults without holding the mmap_sem.
+
+	  This should allow better concurrency for massively threaded processes
+	  since the page fault handler will not wait for other thread's memory
+	  layout change to be done, assuming that this change is done in
+	  another part of the process's memory space. This type of page fault
+	  is named speculative page fault.
+
+	  If the speculative page fault fails because a concurrent modification
+	  is detected or because underlying PMD or PTE tables are not yet
+	  allocated, the speculative page fault fails and a classic page fault
+	  is then tried.
-- 
2.7.4

RE: [PATCH v11 00/26] Speculative page faults

[Song, HaiyanX]

Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
tested on Intel 4s Skylake platform.

The regression result is sorted by the metric will-it-scale.per_thread_ops.
Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
Commit id:
    base commit: d55f34411b1b126429a823d06c3124c16283231f
    head commit: 0355322b3577eeab7669066df42c550a56801110
Benchmark suite: will-it-scale
Download link:
https://github.com/antonblanchard/will-it-scale/tree/master/tests
Metrics:
    will-it-scale.per_process_ops=processes/nr_cpu
    will-it-scale.per_thread_ops=threads/nr_cpu
test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
THP: enable / disable
nr_task: 100%

1. Regressions:
a) THP enabled:
testcase                        base            change          head       metric
page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops

b) THP disabled:
testcase                        base            change          head       metric
page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops

2. Improvements:
a) THP enabled:
testcase                        base            change          head       metric
malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops

b) THP disabled:
testcase                        base            change          head       metric
malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops

Notes: for above values in column "change", the higher value means that the related testcase result
on head commit is better than that on base commit for this benchmark.


Best regards
Haiyan Song

________________________________________
From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
Sent: Thursday, May 17, 2018 7:06 PM
To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
Subject: [PATCH v11 00/26] Speculative page faults

This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
page fault without holding the mm semaphore [1].

The idea is to try to handle user space page faults without holding the
mmap_sem. This should allow better concurrency for massively threaded
process since the page fault handler will not wait for other threads memory
layout change to be done, assuming that this change is done in another part
of the process's memory space. This type page fault is named speculative
page fault. If the speculative page fault fails because of a concurrency is
detected or because underlying PMD or PTE tables are not yet allocating, it
is failing its processing and a classic page fault is then tried.

The speculative page fault (SPF) has to look for the VMA matching the fault
address without holding the mmap_sem, this is done by introducing a rwlock
which protects the access to the mm_rb tree. Previously this was done using
SRCU but it was introducing a lot of scheduling to process the VMA's
freeing operation which was hitting the performance by 20% as reported by
Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
limiting the locking contention to these operations which are expected to
be in a O(log n) order. In addition to ensure that the VMA is not freed in
our back a reference count is added and 2 services (get_vma() and
put_vma()) are introduced to handle the reference count. Once a VMA is
fetched from the RB tree using get_vma(), it must be later freed using
put_vma(). I can't see anymore the overhead I got while will-it-scale
benchmark anymore.

The VMA's attributes checked during the speculative page fault processing
have to be protected against parallel changes. This is done by using a per
VMA sequence lock. This sequence lock allows the speculative page fault
handler to fast check for parallel changes in progress and to abort the
speculative page fault in that case.

Once the VMA has been found, the speculative page fault handler would check
for the VMA's attributes to verify that the page fault has to be handled
correctly or not. Thus, the VMA is protected through a sequence lock which
allows fast detection of concurrent VMA changes. If such a change is
detected, the speculative page fault is aborted and a *classic* page fault
is tried.  VMA sequence lockings are added when VMA attributes which are
checked during the page fault are modified.

When the PTE is fetched, the VMA is checked to see if it has been changed,
so once the page table is locked, the VMA is valid, so any other changes
leading to touching this PTE will need to lock the page table, so no
parallel change is possible at this time.

The locking of the PTE is done with interrupts disabled, this allows
checking for the PMD to ensure that there is not an ongoing collapsing
operation. Since khugepaged is firstly set the PMD to pmd_none and then is
waiting for the other CPU to have caught the IPI interrupt, if the pmd is
valid at the time the PTE is locked, we have the guarantee that the
collapsing operation will have to wait on the PTE lock to move forward.
This allows the SPF handler to map the PTE safely. If the PMD value is
different from the one recorded at the beginning of the SPF operation, the
classic page fault handler will be called to handle the operation while
holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
the lock is done using spin_trylock() to avoid dead lock when handling a
page fault while a TLB invalidate is requested by another CPU holding the
PTE.

In pseudo code, this could be seen as:
    speculative_page_fault()
    {
            vma = get_vma()
            check vma sequence count
            check vma's support
            disable interrupt
                  check pgd,p4d,...,pte
                  save pmd and pte in vmf
                  save vma sequence counter in vmf
            enable interrupt
            check vma sequence count
            handle_pte_fault(vma)
                    ..
                    page = alloc_page()
                    pte_map_lock()
                            disable interrupt
                                    abort if sequence counter has changed
                                    abort if pmd or pte has changed
                                    pte map and lock
                            enable interrupt
                    if abort
                       free page
                       abort
                    ...
    }

    arch_fault_handler()
    {
            if (speculative_page_fault(&vma))
               goto done
    again:
            lock(mmap_sem)
            vma = find_vma();
            handle_pte_fault(vma);
            if retry
               unlock(mmap_sem)
               goto again;
    done:
            handle fault error
    }

Support for THP is not done because when checking for the PMD, we can be
confused by an in progress collapsing operation done by khugepaged. The
issue is that pmd_none() could be true either if the PMD is not already
populated or if the underlying PTE are in the way to be collapsed. So we
cannot safely allocate a PMD if pmd_none() is true.

This series add a new software performance event named 'speculative-faults'
or 'spf'. It counts the number of successful page fault event handled
speculatively. When recording 'faults,spf' events, the faults one is
counting the total number of page fault events while 'spf' is only counting
the part of the faults processed speculatively.

There are some trace events introduced by this series. They allow
identifying why the page faults were not processed speculatively. This
doesn't take in account the faults generated by a monothreaded process
which directly processed while holding the mmap_sem. This trace events are
grouped in a system named 'pagefault', they are:
 - pagefault:spf_vma_changed : if the VMA has been changed in our back
 - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
 - pagefault:spf_vma_notsup : the VMA's type is not supported
 - pagefault:spf_vma_access : the VMA's access right are not respected
 - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
   back.

To record all the related events, the easier is to run perf with the
following arguments :
$ perf stat -e 'faults,spf,pagefault:*' <command>

There is also a dedicated vmstat counter showing the number of successful
page fault handled speculatively. I can be seen this way:
$ grep speculative_pgfault /proc/vmstat

This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
on x86, PowerPC and arm64.

---------------------
Real Workload results

As mentioned in previous email, we did non official runs using a "popular
in memory multithreaded database product" on 176 cores SMT8 Power system
which showed a 30% improvements in the number of transaction processed per
second. This run has been done on the v6 series, but changes introduced in
this new version should not impact the performance boost seen.

Here are the perf data captured during 2 of these runs on top of the v8
series:
                vanilla         spf
faults          89.418          101.364         +13%
spf                n/a           97.989

With the SPF kernel, most of the page fault were processed in a speculative
way.

Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
it a try on an android device. He reported that the application launch time
was improved in average by 6%, and for large applications (~100 threads) by
20%.

Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
MSM845 (8 cores) with 6GB (the less is better):

Application                             4.9     4.9+spf delta
com.tencent.mm                          416     389     -7%
com.eg.android.AlipayGphone             1135    986     -13%
com.tencent.mtt                         455     454     0%
com.qqgame.hlddz                        1497    1409    -6%
com.autonavi.minimap                    711     701     -1%
com.tencent.tmgp.sgame                  788     748     -5%
com.immomo.momo                         501     487     -3%
com.tencent.peng                        2145    2112    -2%
com.smile.gifmaker                      491     461     -6%
com.baidu.BaiduMap                      479     366     -23%
com.taobao.taobao                       1341    1198    -11%
com.baidu.searchbox                     333     314     -6%
com.tencent.mobileqq                    394     384     -3%
com.sina.weibo                          907     906     0%
com.youku.phone                         816     731     -11%
com.happyelements.AndroidAnimal.qq      763     717     -6%
com.UCMobile                            415     411     -1%
com.tencent.tmgp.ak                     1464    1431    -2%
com.tencent.qqmusic                     336     329     -2%
com.sankuai.meituan                     1661    1302    -22%
com.netease.cloudmusic                  1193    1200    1%
air.tv.douyu.android                    4257    4152    -2%

------------------
Benchmarks results

Base kernel is v4.17.0-rc4-mm1
SPF is BASE + this series

Kernbench:
----------
Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
kernel (kernel is build 5 times):

Average Half load -j 8
                 Run    (std deviation)
                 BASE                   SPF
Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
Sleeps           105064  (1240.96)      105074  (337.612)       0.01%

Average Optimal load -j 16
                 Run    (std deviation)
                 BASE                   SPF
Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%


During a run on the SPF, perf events were captured:
 Performance counter stats for '../kernbench -M':
         526743764      faults
               210      spf
                 3      pagefault:spf_vma_changed
                 0      pagefault:spf_vma_noanon
              2278      pagefault:spf_vma_notsup
                 0      pagefault:spf_vma_access
                 0      pagefault:spf_pmd_changed

Very few speculative page faults were recorded as most of the processes
involved are monothreaded (sounds that on this architecture some threads
were created during the kernel build processing).

Here are the kerbench results on a 80 CPUs Power8 system:

Average Half load -j 40
                 Run    (std deviation)
                 BASE                   SPF
Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
Sleeps           317923  (652.499)      318469  (1255.59)       0.17%

Average Optimal load -j 80
                 Run    (std deviation)
                 BASE                   SPF
Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
Context Switches 223861  (138865)       225032  (139632)        0.52%
Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%

During a run on the SPF, perf events were captured:
 Performance counter stats for '../kernbench -M':
         116730856      faults
                 0      spf
                 3      pagefault:spf_vma_changed
                 0      pagefault:spf_vma_noanon
               476      pagefault:spf_vma_notsup
                 0      pagefault:spf_vma_access
                 0      pagefault:spf_pmd_changed

Most of the processes involved are monothreaded so SPF is not activated but
there is no impact on the performance.

Ebizzy:
-------
The test is counting the number of records per second it can manage, the
higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
consistent result I repeated the test 100 times and measure the average
result. The number is the record processes per second, the higher is the
best.

                BASE            SPF             delta
16 CPUs x86 VM  742.57          1490.24         100.69%
80 CPUs P8 node 13105.4         24174.23        84.46%

Here are the performance counter read during a run on a 16 CPUs x86 VM:
 Performance counter stats for './ebizzy -mTt 16':
           1706379      faults
           1674599      spf
             30588      pagefault:spf_vma_changed
                 0      pagefault:spf_vma_noanon
               363      pagefault:spf_vma_notsup
                 0      pagefault:spf_vma_access
                 0      pagefault:spf_pmd_changed

And the ones captured during a run on a 80 CPUs Power node:
 Performance counter stats for './ebizzy -mTt 80':
           1874773      faults
           1461153      spf
            413293      pagefault:spf_vma_changed
                 0      pagefault:spf_vma_noanon
               200      pagefault:spf_vma_notsup
                 0      pagefault:spf_vma_access
                 0      pagefault:spf_pmd_changed

In ebizzy's case most of the page fault were handled in a speculative way,
leading the ebizzy performance boost.

------------------
Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
 - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
   and Minchan Kim, hopefully.
 - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
   __do_page_fault().
 - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
   instead
   of aborting the speculative page fault handling. Dropping the now
useless
   trace event pagefault:spf_pte_lock.
 - No more try to reuse the fetched VMA during the speculative page fault
   handling when retrying is needed. This adds a lot of complexity and
   additional tests done didn't show a significant performance improvement.
 - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.

[1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
[2] https://patchwork.kernel.org/patch/9999687/


Laurent Dufour (20):
  mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
  x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
  powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
  mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
  mm: make pte_unmap_same compatible with SPF
  mm: introduce INIT_VMA()
  mm: protect VMA modifications using VMA sequence count
  mm: protect mremap() against SPF hanlder
  mm: protect SPF handler against anon_vma changes
  mm: cache some VMA fields in the vm_fault structure
  mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
  mm: introduce __lru_cache_add_active_or_unevictable
  mm: introduce __vm_normal_page()
  mm: introduce __page_add_new_anon_rmap()
  mm: protect mm_rb tree with a rwlock
  mm: adding speculative page fault failure trace events
  perf: add a speculative page fault sw event
  perf tools: add support for the SPF perf event
  mm: add speculative page fault vmstats
  powerpc/mm: add speculative page fault

Mahendran Ganesh (2):
  arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
  arm64/mm: add speculative page fault

Peter Zijlstra (4):
  mm: prepare for FAULT_FLAG_SPECULATIVE
  mm: VMA sequence count
  mm: provide speculative fault infrastructure
  x86/mm: add speculative pagefault handling

 arch/arm64/Kconfig                    |   1 +
 arch/arm64/mm/fault.c                 |  12 +
 arch/powerpc/Kconfig                  |   1 +
 arch/powerpc/mm/fault.c               |  16 +
 arch/x86/Kconfig                      |   1 +
 arch/x86/mm/fault.c                   |  27 +-
 fs/exec.c                             |   2 +-
 fs/proc/task_mmu.c                    |   5 +-
 fs/userfaultfd.c                      |  17 +-
 include/linux/hugetlb_inline.h        |   2 +-
 include/linux/migrate.h               |   4 +-
 include/linux/mm.h                    | 136 +++++++-
 include/linux/mm_types.h              |   7 +
 include/linux/pagemap.h               |   4 +-
 include/linux/rmap.h                  |  12 +-
 include/linux/swap.h                  |  10 +-
 include/linux/vm_event_item.h         |   3 +
 include/trace/events/pagefault.h      |  80 +++++
 include/uapi/linux/perf_event.h       |   1 +
 kernel/fork.c                         |   5 +-
 mm/Kconfig                            |  22 ++
 mm/huge_memory.c                      |   6 +-
 mm/hugetlb.c                          |   2 +
 mm/init-mm.c                          |   3 +
 mm/internal.h                         |  20 ++
 mm/khugepaged.c                       |   5 +
 mm/madvise.c                          |   6 +-
 mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
 mm/mempolicy.c                        |  51 ++-
 mm/migrate.c                          |   6 +-
 mm/mlock.c                            |  13 +-
 mm/mmap.c                             | 229 ++++++++++---
 mm/mprotect.c                         |   4 +-
 mm/mremap.c                           |  13 +
 mm/nommu.c                            |   2 +-
 mm/rmap.c                             |   5 +-
 mm/swap.c                             |   6 +-
 mm/swap_state.c                       |   8 +-
 mm/vmstat.c                           |   5 +-
 tools/include/uapi/linux/perf_event.h |   1 +
 tools/perf/util/evsel.c               |   1 +
 tools/perf/util/parse-events.c        |   4 +
 tools/perf/util/parse-events.l        |   1 +
 tools/perf/util/python.c              |   1 +
 44 files changed, 1161 insertions(+), 211 deletions(-)
 create mode 100644 include/trace/events/pagefault.h

--
2.7.4

RE: [PATCH v11 00/26] Speculative page faults

[Song, HaiyanX]

=0A=
Some regression and improvements is found by LKP-tools(linux kernel perform=
ance) on V9 patch series=0A=
tested on Intel 4s Skylake platform.=0A=
=0A=
The regression result is sorted by the metric will-it-scale.per_thread_ops.=
=0A=
Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch se=
ries)=0A=
Commit id:=0A=
    base commit: d55f34411b1b126429a823d06c3124c16283231f=0A=
    head commit: 0355322b3577eeab7669066df42c550a56801110=0A=
Benchmark suite: will-it-scale=0A=
Download link:=0A=
https://github.com/antonblanchard/will-it-scale/tree/master/tests=0A=
Metrics:=0A=
    will-it-scale.per_process_ops=3Dprocesses/nr_cpu=0A=
    will-it-scale.per_thread_ops=3Dthreads/nr_cpu=0A=
test box: lkp-skl-4sp1(nr_cpu=3D192,memory=3D768G)=0A=
THP: enable / disable=0A=
nr_task: 100%=0A=
=0A=
1. Regressions:=0A=
a) THP enabled:=0A=
testcase                        base            change          head       =
metric=0A=
page_fault3/ enable THP         10092           -17.5%          8323       =
will-it-scale.per_thread_ops=0A=
page_fault2/ enable THP          8300           -17.2%          6869       =
will-it-scale.per_thread_ops=0A=
brk1/ enable THP                  957.67         -7.6%           885       =
will-it-scale.per_thread_ops=0A=
page_fault3/ enable THP        172821            -5.3%        163692       =
will-it-scale.per_process_ops=0A=
signal1/ enable THP              9125            -3.2%          8834       =
will-it-scale.per_process_ops=0A=
=0A=
b) THP disabled:=0A=
testcase                        base            change          head       =
metric=0A=
page_fault3/ disable THP        10107           -19.1%          8180       =
will-it-scale.per_thread_ops=0A=
page_fault2/ disable THP         8432           -17.8%          6931       =
will-it-scale.per_thread_ops=0A=
context_switch1/ disable THP   215389            -6.8%        200776       =
will-it-scale.per_thread_ops=0A=
brk1/ disable THP                 939.67         -6.6%           877.33    =
will-it-scale.per_thread_ops=0A=
page_fault3/ disable THP       173145            -4.7%        165064       =
will-it-scale.per_process_ops=0A=
signal1/ disable THP             9162            -3.9%          8802       =
will-it-scale.per_process_ops=0A=
=0A=
2. Improvements:=0A=
a) THP enabled:=0A=
testcase                        base            change          head       =
metric=0A=
malloc1/ enable THP               66.33        +469.8%           383.67    =
will-it-scale.per_thread_ops=0A=
writeseek3/ enable THP          2531             +4.5%          2646       =
will-it-scale.per_thread_ops=0A=
signal1/ enable THP              989.33          +2.8%          1016       =
will-it-scale.per_thread_ops=0A=
=0A=
b) THP disabled:=0A=
testcase                        base            change          head       =
metric=0A=
malloc1/ disable THP              90.33        +417.3%           467.33    =
will-it-scale.per_thread_ops=0A=
read2/ disable THP             58934            +39.2%         82060       =
will-it-scale.per_thread_ops=0A=
page_fault1/ disable THP        8607            +36.4%         11736       =
will-it-scale.per_thread_ops=0A=
read1/ disable THP            314063            +12.7%        353934       =
will-it-scale.per_thread_ops=0A=
writeseek3/ disable THP         2452            +12.5%          2759       =
will-it-scale.per_thread_ops=0A=
signal1/ disable THP             971.33          +5.5%          1024       =
will-it-scale.per_thread_ops=0A=
=0A=
Notes: for above values in column "change", the higher value means that the=
 related testcase result=0A=
on head commit is better than that on base commit for this benchmark.=0A=
=0A=
=0A=
Best regards=0A=
Haiyan Song=0A=
=0A=
________________________________________=0A=
From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laur=
ent Dufour [ldufour@linux.vnet.ibm.com]=0A=
Sent: Thursday, May 17, 2018 7:06 PM=0A=
To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kir=
ill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Mat=
thew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; =
benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Glei=
xner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.s=
enozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi=
; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan K=
im; Punit Agrawal; vinayak menon; Yang Shi=0A=
Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.=
com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; =
Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org=0A=
Subject: [PATCH v11 00/26] Speculative page faults=0A=
=0A=
This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle=
=0A=
page fault without holding the mm semaphore [1].=0A=
=0A=
The idea is to try to handle user space page faults without holding the=0A=
mmap_sem. This should allow better concurrency for massively threaded=0A=
process since the page fault handler will not wait for other threads memory=
=0A=
layout change to be done, assuming that this change is done in another part=
=0A=
of the process's memory space. This type page fault is named speculative=0A=
page fault. If the speculative page fault fails because of a concurrency is=
=0A=
detected or because underlying PMD or PTE tables are not yet allocating, it=
=0A=
is failing its processing and a classic page fault is then tried.=0A=
=0A=
The speculative page fault (SPF) has to look for the VMA matching the fault=
=0A=
address without holding the mmap_sem, this is done by introducing a rwlock=
=0A=
which protects the access to the mm_rb tree. Previously this was done using=
=0A=
SRCU but it was introducing a lot of scheduling to process the VMA's=0A=
freeing operation which was hitting the performance by 20% as reported by=
=0A=
Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is=0A=
limiting the locking contention to these operations which are expected to=
=0A=
be in a O(log n) order. In addition to ensure that the VMA is not freed in=
=0A=
our back a reference count is added and 2 services (get_vma() and=0A=
put_vma()) are introduced to handle the reference count. Once a VMA is=0A=
fetched from the RB tree using get_vma(), it must be later freed using=0A=
put_vma(). I can't see anymore the overhead I got while will-it-scale=0A=
benchmark anymore.=0A=
=0A=
The VMA's attributes checked during the speculative page fault processing=
=0A=
have to be protected against parallel changes. This is done by using a per=
=0A=
VMA sequence lock. This sequence lock allows the speculative page fault=0A=
handler to fast check for parallel changes in progress and to abort the=0A=
speculative page fault in that case.=0A=
=0A=
Once the VMA has been found, the speculative page fault handler would check=
=0A=
for the VMA's attributes to verify that the page fault has to be handled=0A=
correctly or not. Thus, the VMA is protected through a sequence lock which=
=0A=
allows fast detection of concurrent VMA changes. If such a change is=0A=
detected, the speculative page fault is aborted and a *classic* page fault=
=0A=
is tried.  VMA sequence lockings are added when VMA attributes which are=0A=
checked during the page fault are modified.=0A=
=0A=
When the PTE is fetched, the VMA is checked to see if it has been changed,=
=0A=
so once the page table is locked, the VMA is valid, so any other changes=0A=
leading to touching this PTE will need to lock the page table, so no=0A=
parallel change is possible at this time.=0A=
=0A=
The locking of the PTE is done with interrupts disabled, this allows=0A=
checking for the PMD to ensure that there is not an ongoing collapsing=0A=
operation. Since khugepaged is firstly set the PMD to pmd_none and then is=
=0A=
waiting for the other CPU to have caught the IPI interrupt, if the pmd is=
=0A=
valid at the time the PTE is locked, we have the guarantee that the=0A=
collapsing operation will have to wait on the PTE lock to move forward.=0A=
This allows the SPF handler to map the PTE safely. If the PMD value is=0A=
different from the one recorded at the beginning of the SPF operation, the=
=0A=
classic page fault handler will be called to handle the operation while=0A=
holding the mmap_sem. As the PTE lock is done with the interrupts disabled,=
=0A=
the lock is done using spin_trylock() to avoid dead lock when handling a=0A=
page fault while a TLB invalidate is requested by another CPU holding the=
=0A=
PTE.=0A=
=0A=
In pseudo code, this could be seen as:=0A=
    speculative_page_fault()=0A=
    {=0A=
            vma =3D get_vma()=0A=
            check vma sequence count=0A=
            check vma's support=0A=
            disable interrupt=0A=
                  check pgd,p4d,...,pte=0A=
                  save pmd and pte in vmf=0A=
                  save vma sequence counter in vmf=0A=
            enable interrupt=0A=
            check vma sequence count=0A=
            handle_pte_fault(vma)=0A=
                    ..=0A=
                    page =3D alloc_page()=0A=
                    pte_map_lock()=0A=
                            disable interrupt=0A=
                                    abort if sequence counter has changed=
=0A=
                                    abort if pmd or pte has changed=0A=
                                    pte map and lock=0A=
                            enable interrupt=0A=
                    if abort=0A=
                       free page=0A=
                       abort=0A=
                    ...=0A=
    }=0A=
=0A=
    arch_fault_handler()=0A=
    {=0A=
            if (speculative_page_fault(&vma))=0A=
               goto done=0A=
    again:=0A=
            lock(mmap_sem)=0A=
            vma =3D find_vma();=0A=
            handle_pte_fault(vma);=0A=
            if retry=0A=
               unlock(mmap_sem)=0A=
               goto again;=0A=
    done:=0A=
            handle fault error=0A=
    }=0A=
=0A=
Support for THP is not done because when checking for the PMD, we can be=0A=
confused by an in progress collapsing operation done by khugepaged. The=0A=
issue is that pmd_none() could be true either if the PMD is not already=0A=
populated or if the underlying PTE are in the way to be collapsed. So we=0A=
cannot safely allocate a PMD if pmd_none() is true.=0A=
=0A=
This series add a new software performance event named 'speculative-faults'=
=0A=
or 'spf'. It counts the number of successful page fault event handled=0A=
speculatively. When recording 'faults,spf' events, the faults one is=0A=
counting the total number of page fault events while 'spf' is only counting=
=0A=
the part of the faults processed speculatively.=0A=
=0A=
There are some trace events introduced by this series. They allow=0A=
identifying why the page faults were not processed speculatively. This=0A=
doesn't take in account the faults generated by a monothreaded process=0A=
which directly processed while holding the mmap_sem. This trace events are=
=0A=
grouped in a system named 'pagefault', they are:=0A=
 - pagefault:spf_vma_changed : if the VMA has been changed in our back=0A=
 - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.=0A=
 - pagefault:spf_vma_notsup : the VMA's type is not supported=0A=
 - pagefault:spf_vma_access : the VMA's access right are not respected=0A=
 - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our=0A=
   back.=0A=
=0A=
To record all the related events, the easier is to run perf with the=0A=
following arguments :=0A=
$ perf stat -e 'faults,spf,pagefault:*' <command>=0A=
=0A=
There is also a dedicated vmstat counter showing the number of successful=
=0A=
page fault handled speculatively. I can be seen this way:=0A=
$ grep speculative_pgfault /proc/vmstat=0A=
=0A=
This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional=
=0A=
on x86, PowerPC and arm64.=0A=
=0A=
---------------------=0A=
Real Workload results=0A=
=0A=
As mentioned in previous email, we did non official runs using a "popular=
=0A=
in memory multithreaded database product" on 176 cores SMT8 Power system=0A=
which showed a 30% improvements in the number of transaction processed per=
=0A=
second. This run has been done on the v6 series, but changes introduced in=
=0A=
this new version should not impact the performance boost seen.=0A=
=0A=
Here are the perf data captured during 2 of these runs on top of the v8=0A=
series:=0A=
                vanilla         spf=0A=
faults          89.418          101.364         +13%=0A=
spf                n/a           97.989=0A=
=0A=
With the SPF kernel, most of the page fault were processed in a speculative=
=0A=
way.=0A=
=0A=
Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave=
=0A=
it a try on an android device. He reported that the application launch time=
=0A=
was improved in average by 6%, and for large applications (~100 threads) by=
=0A=
20%.=0A=
=0A=
Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom=0A=
MSM845 (8 cores) with 6GB (the less is better):=0A=
=0A=
Application                             4.9     4.9+spf delta=0A=
com.tencent.mm                          416     389     -7%=0A=
com.eg.android.AlipayGphone             1135    986     -13%=0A=
com.tencent.mtt                         455     454     0%=0A=
com.qqgame.hlddz                        1497    1409    -6%=0A=
com.autonavi.minimap                    711     701     -1%=0A=
com.tencent.tmgp.sgame                  788     748     -5%=0A=
com.immomo.momo                         501     487     -3%=0A=
com.tencent.peng                        2145    2112    -2%=0A=
com.smile.gifmaker                      491     461     -6%=0A=
com.baidu.BaiduMap                      479     366     -23%=0A=
com.taobao.taobao                       1341    1198    -11%=0A=
com.baidu.searchbox                     333     314     -6%=0A=
com.tencent.mobileqq                    394     384     -3%=0A=
com.sina.weibo                          907     906     0%=0A=
com.youku.phone                         816     731     -11%=0A=
com.happyelements.AndroidAnimal.qq      763     717     -6%=0A=
com.UCMobile                            415     411     -1%=0A=
com.tencent.tmgp.ak                     1464    1431    -2%=0A=
com.tencent.qqmusic                     336     329     -2%=0A=
com.sankuai.meituan                     1661    1302    -22%=0A=
com.netease.cloudmusic                  1193    1200    1%=0A=
air.tv.douyu.android                    4257    4152    -2%=0A=
=0A=
------------------=0A=
Benchmarks results=0A=
=0A=
Base kernel is v4.17.0-rc4-mm1=0A=
SPF is BASE + this series=0A=
=0A=
Kernbench:=0A=
----------=0A=
Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15=0A=
kernel (kernel is build 5 times):=0A=
=0A=
Average Half load -j 8=0A=
                 Run    (std deviation)=0A=
                 BASE                   SPF=0A=
Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%=0A=
User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%=0A=
System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%=0A=
Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%=0A=
Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%=0A=
Sleeps           105064  (1240.96)      105074  (337.612)       0.01%=0A=
=0A=
Average Optimal load -j 16=0A=
                 Run    (std deviation)=0A=
                 BASE                   SPF=0A=
Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%=0A=
User    Time     11064.8 (981.142)      11085   (990.897)       0.18%=0A=
System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%=0A=
Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%=0A=
Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%=0A=
Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%=0A=
=0A=
=0A=
During a run on the SPF, perf events were captured:=0A=
 Performance counter stats for '../kernbench -M':=0A=
         526743764      faults=0A=
               210      spf=0A=
                 3      pagefault:spf_vma_changed=0A=
                 0      pagefault:spf_vma_noanon=0A=
              2278      pagefault:spf_vma_notsup=0A=
                 0      pagefault:spf_vma_access=0A=
                 0      pagefault:spf_pmd_changed=0A=
=0A=
Very few speculative page faults were recorded as most of the processes=0A=
involved are monothreaded (sounds that on this architecture some threads=0A=
were created during the kernel build processing).=0A=
=0A=
Here are the kerbench results on a 80 CPUs Power8 system:=0A=
=0A=
Average Half load -j 40=0A=
                 Run    (std deviation)=0A=
                 BASE                   SPF=0A=
Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%=0A=
User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%=0A=
System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%=0A=
Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%=0A=
Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%=0A=
Sleeps           317923  (652.499)      318469  (1255.59)       0.17%=0A=
=0A=
Average Optimal load -j 80=0A=
                 Run    (std deviation)=0A=
                 BASE                   SPF=0A=
Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%=0A=
User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%=0A=
System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%=0A=
Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%=0A=
Context Switches 223861  (138865)       225032  (139632)        0.52%=0A=
Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%=0A=
=0A=
During a run on the SPF, perf events were captured:=0A=
 Performance counter stats for '../kernbench -M':=0A=
         116730856      faults=0A=
                 0      spf=0A=
                 3      pagefault:spf_vma_changed=0A=
                 0      pagefault:spf_vma_noanon=0A=
               476      pagefault:spf_vma_notsup=0A=
                 0      pagefault:spf_vma_access=0A=
                 0      pagefault:spf_pmd_changed=0A=
=0A=
Most of the processes involved are monothreaded so SPF is not activated but=
=0A=
there is no impact on the performance.=0A=
=0A=
Ebizzy:=0A=
-------=0A=
The test is counting the number of records per second it can manage, the=0A=
higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get=0A=
consistent result I repeated the test 100 times and measure the average=0A=
result. The number is the record processes per second, the higher is the=0A=
best.=0A=
=0A=
                BASE            SPF             delta=0A=
16 CPUs x86 VM  742.57          1490.24         100.69%=0A=
80 CPUs P8 node 13105.4         24174.23        84.46%=0A=
=0A=
Here are the performance counter read during a run on a 16 CPUs x86 VM:=0A=
 Performance counter stats for './ebizzy -mTt 16':=0A=
           1706379      faults=0A=
           1674599      spf=0A=
             30588      pagefault:spf_vma_changed=0A=
                 0      pagefault:spf_vma_noanon=0A=
               363      pagefault:spf_vma_notsup=0A=
                 0      pagefault:spf_vma_access=0A=
                 0      pagefault:spf_pmd_changed=0A=
=0A=
And the ones captured during a run on a 80 CPUs Power node:=0A=
 Performance counter stats for './ebizzy -mTt 80':=0A=
           1874773      faults=0A=
           1461153      spf=0A=
            413293      pagefault:spf_vma_changed=0A=
                 0      pagefault:spf_vma_noanon=0A=
               200      pagefault:spf_vma_notsup=0A=
                 0      pagefault:spf_vma_access=0A=
                 0      pagefault:spf_pmd_changed=0A=
=0A=
In ebizzy's case most of the page fault were handled in a speculative way,=
=0A=
leading the ebizzy performance boost.=0A=
=0A=
------------------=0A=
Changes since v10 (https://lkml.org/lkml/2018/4/17/572):=0A=
 - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran=
=0A=
   and Minchan Kim, hopefully.=0A=
 - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in=0A=
   __do_page_fault().=0A=
 - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails=0A=
   instead=0A=
   of aborting the speculative page fault handling. Dropping the now=0A=
useless=0A=
   trace event pagefault:spf_pte_lock.=0A=
 - No more try to reuse the fetched VMA during the speculative page fault=
=0A=
   handling when retrying is needed. This adds a lot of complexity and=0A=
   additional tests done didn't show a significant performance improvement.=
=0A=
 - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.=0A=
=0A=
[1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-spec=
ulative-page-faults-tt965642.html#none=0A=
[2] https://patchwork.kernel.org/patch/9999687/=0A=
=0A=
=0A=
Laurent Dufour (20):=0A=
  mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT=0A=
  x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT=0A=
  powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT=0A=
  mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE=0A=
  mm: make pte_unmap_same compatible with SPF=0A=
  mm: introduce INIT_VMA()=0A=
  mm: protect VMA modifications using VMA sequence count=0A=
  mm: protect mremap() against SPF hanlder=0A=
  mm: protect SPF handler against anon_vma changes=0A=
  mm: cache some VMA fields in the vm_fault structure=0A=
  mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()=0A=
  mm: introduce __lru_cache_add_active_or_unevictable=0A=
  mm: introduce __vm_normal_page()=0A=
  mm: introduce __page_add_new_anon_rmap()=0A=
  mm: protect mm_rb tree with a rwlock=0A=
  mm: adding speculative page fault failure trace events=0A=
  perf: add a speculative page fault sw event=0A=
  perf tools: add support for the SPF perf event=0A=
  mm: add speculative page fault vmstats=0A=
  powerpc/mm: add speculative page fault=0A=
=0A=
Mahendran Ganesh (2):=0A=
  arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT=0A=
  arm64/mm: add speculative page fault=0A=
=0A=
Peter Zijlstra (4):=0A=
  mm: prepare for FAULT_FLAG_SPECULATIVE=0A=
  mm: VMA sequence count=0A=
  mm: provide speculative fault infrastructure=0A=
  x86/mm: add speculative pagefault handling=0A=
=0A=
 arch/arm64/Kconfig                    |   1 +=0A=
 arch/arm64/mm/fault.c                 |  12 +=0A=
 arch/powerpc/Kconfig                  |   1 +=0A=
 arch/powerpc/mm/fault.c               |  16 +=0A=
 arch/x86/Kconfig                      |   1 +=0A=
 arch/x86/mm/fault.c                   |  27 +-=0A=
 fs/exec.c                             |   2 +-=0A=
 fs/proc/task_mmu.c                    |   5 +-=0A=
 fs/userfaultfd.c                      |  17 +-=0A=
 include/linux/hugetlb_inline.h        |   2 +-=0A=
 include/linux/migrate.h               |   4 +-=0A=
 include/linux/mm.h                    | 136 +++++++-=0A=
 include/linux/mm_types.h              |   7 +=0A=
 include/linux/pagemap.h               |   4 +-=0A=
 include/linux/rmap.h                  |  12 +-=0A=
 include/linux/swap.h                  |  10 +-=0A=
 include/linux/vm_event_item.h         |   3 +=0A=
 include/trace/events/pagefault.h      |  80 +++++=0A=
 include/uapi/linux/perf_event.h       |   1 +=0A=
 kernel/fork.c                         |   5 +-=0A=
 mm/Kconfig                            |  22 ++=0A=
 mm/huge_memory.c                      |   6 +-=0A=
 mm/hugetlb.c                          |   2 +=0A=
 mm/init-mm.c                          |   3 +=0A=
 mm/internal.h                         |  20 ++=0A=
 mm/khugepaged.c                       |   5 +=0A=
 mm/madvise.c                          |   6 +-=0A=
 mm/memory.c                           | 612 +++++++++++++++++++++++++++++-=
----=0A=
 mm/mempolicy.c                        |  51 ++-=0A=
 mm/migrate.c                          |   6 +-=0A=
 mm/mlock.c                            |  13 +-=0A=
 mm/mmap.c                             | 229 ++++++++++---=0A=
 mm/mprotect.c                         |   4 +-=0A=
 mm/mremap.c                           |  13 +=0A=
 mm/nommu.c                            |   2 +-=0A=
 mm/rmap.c                             |   5 +-=0A=
 mm/swap.c                             |   6 +-=0A=
 mm/swap_state.c                       |   8 +-=0A=
 mm/vmstat.c                           |   5 +-=0A=
 tools/include/uapi/linux/perf_event.h |   1 +=0A=
 tools/perf/util/evsel.c               |   1 +=0A=
 tools/perf/util/parse-events.c        |   4 +=0A=
 tools/perf/util/parse-events.l        |   1 +=0A=
 tools/perf/util/python.c              |   1 +=0A=
 44 files changed, 1161 insertions(+), 211 deletions(-)=0A=
 create mode 100644 include/trace/events/pagefault.h=0A=
=0A=
--=0A=
2.7.4=0A=
=0A=

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

On 28/05/2018 07:23, Song, HaiyanX wrote:
> 
> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
> tested on Intel 4s Skylake platform.

Hi,

Thanks for reporting this benchmark results, but you mentioned the "V9 patch
series" while responding to the v11 header series...
Were these tests done on v9 or v11 ?

Cheers,
Laurent.

> 
> The regression result is sorted by the metric will-it-scale.per_thread_ops.
> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
> Commit id:
>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>     head commit: 0355322b3577eeab7669066df42c550a56801110
> Benchmark suite: will-it-scale
> Download link:
> https://github.com/antonblanchard/will-it-scale/tree/master/tests
> Metrics:
>     will-it-scale.per_process_ops=processes/nr_cpu
>     will-it-scale.per_thread_ops=threads/nr_cpu
> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
> THP: enable / disable
> nr_task: 100%
> 
> 1. Regressions:
> a) THP enabled:
> testcase                        base            change          head       metric
> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
> 
> b) THP disabled:
> testcase                        base            change          head       metric
> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
> 
> 2. Improvements:
> a) THP enabled:
> testcase                        base            change          head       metric
> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
> 
> b) THP disabled:
> testcase                        base            change          head       metric
> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
> 
> Notes: for above values in column "change", the higher value means that the related testcase result
> on head commit is better than that on base commit for this benchmark.
> 
> 
> Best regards
> Haiyan Song
> 
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Thursday, May 17, 2018 7:06 PM
> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: [PATCH v11 00/26] Speculative page faults
> 
> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
> page fault without holding the mm semaphore [1].
> 
> The idea is to try to handle user space page faults without holding the
> mmap_sem. This should allow better concurrency for massively threaded
> process since the page fault handler will not wait for other threads memory
> layout change to be done, assuming that this change is done in another part
> of the process's memory space. This type page fault is named speculative
> page fault. If the speculative page fault fails because of a concurrency is
> detected or because underlying PMD or PTE tables are not yet allocating, it
> is failing its processing and a classic page fault is then tried.
> 
> The speculative page fault (SPF) has to look for the VMA matching the fault
> address without holding the mmap_sem, this is done by introducing a rwlock
> which protects the access to the mm_rb tree. Previously this was done using
> SRCU but it was introducing a lot of scheduling to process the VMA's
> freeing operation which was hitting the performance by 20% as reported by
> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
> limiting the locking contention to these operations which are expected to
> be in a O(log n) order. In addition to ensure that the VMA is not freed in
> our back a reference count is added and 2 services (get_vma() and
> put_vma()) are introduced to handle the reference count. Once a VMA is
> fetched from the RB tree using get_vma(), it must be later freed using
> put_vma(). I can't see anymore the overhead I got while will-it-scale
> benchmark anymore.
> 
> The VMA's attributes checked during the speculative page fault processing
> have to be protected against parallel changes. This is done by using a per
> VMA sequence lock. This sequence lock allows the speculative page fault
> handler to fast check for parallel changes in progress and to abort the
> speculative page fault in that case.
> 
> Once the VMA has been found, the speculative page fault handler would check
> for the VMA's attributes to verify that the page fault has to be handled
> correctly or not. Thus, the VMA is protected through a sequence lock which
> allows fast detection of concurrent VMA changes. If such a change is
> detected, the speculative page fault is aborted and a *classic* page fault
> is tried.  VMA sequence lockings are added when VMA attributes which are
> checked during the page fault are modified.
> 
> When the PTE is fetched, the VMA is checked to see if it has been changed,
> so once the page table is locked, the VMA is valid, so any other changes
> leading to touching this PTE will need to lock the page table, so no
> parallel change is possible at this time.
> 
> The locking of the PTE is done with interrupts disabled, this allows
> checking for the PMD to ensure that there is not an ongoing collapsing
> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
> valid at the time the PTE is locked, we have the guarantee that the
> collapsing operation will have to wait on the PTE lock to move forward.
> This allows the SPF handler to map the PTE safely. If the PMD value is
> different from the one recorded at the beginning of the SPF operation, the
> classic page fault handler will be called to handle the operation while
> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
> the lock is done using spin_trylock() to avoid dead lock when handling a
> page fault while a TLB invalidate is requested by another CPU holding the
> PTE.
> 
> In pseudo code, this could be seen as:
>     speculative_page_fault()
>     {
>             vma = get_vma()
>             check vma sequence count
>             check vma's support
>             disable interrupt
>                   check pgd,p4d,...,pte
>                   save pmd and pte in vmf
>                   save vma sequence counter in vmf
>             enable interrupt
>             check vma sequence count
>             handle_pte_fault(vma)
>                     ..
>                     page = alloc_page()
>                     pte_map_lock()
>                             disable interrupt
>                                     abort if sequence counter has changed
>                                     abort if pmd or pte has changed
>                                     pte map and lock
>                             enable interrupt
>                     if abort
>                        free page
>                        abort
>                     ...
>     }
> 
>     arch_fault_handler()
>     {
>             if (speculative_page_fault(&vma))
>                goto done
>     again:
>             lock(mmap_sem)
>             vma = find_vma();
>             handle_pte_fault(vma);
>             if retry
>                unlock(mmap_sem)
>                goto again;
>     done:
>             handle fault error
>     }
> 
> Support for THP is not done because when checking for the PMD, we can be
> confused by an in progress collapsing operation done by khugepaged. The
> issue is that pmd_none() could be true either if the PMD is not already
> populated or if the underlying PTE are in the way to be collapsed. So we
> cannot safely allocate a PMD if pmd_none() is true.
> 
> This series add a new software performance event named 'speculative-faults'
> or 'spf'. It counts the number of successful page fault event handled
> speculatively. When recording 'faults,spf' events, the faults one is
> counting the total number of page fault events while 'spf' is only counting
> the part of the faults processed speculatively.
> 
> There are some trace events introduced by this series. They allow
> identifying why the page faults were not processed speculatively. This
> doesn't take in account the faults generated by a monothreaded process
> which directly processed while holding the mmap_sem. This trace events are
> grouped in a system named 'pagefault', they are:
>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>  - pagefault:spf_vma_access : the VMA's access right are not respected
>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>    back.
> 
> To record all the related events, the easier is to run perf with the
> following arguments :
> $ perf stat -e 'faults,spf,pagefault:*' <command>
> 
> There is also a dedicated vmstat counter showing the number of successful
> page fault handled speculatively. I can be seen this way:
> $ grep speculative_pgfault /proc/vmstat
> 
> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
> on x86, PowerPC and arm64.
> 
> ---------------------
> Real Workload results
> 
> As mentioned in previous email, we did non official runs using a "popular
> in memory multithreaded database product" on 176 cores SMT8 Power system
> which showed a 30% improvements in the number of transaction processed per
> second. This run has been done on the v6 series, but changes introduced in
> this new version should not impact the performance boost seen.
> 
> Here are the perf data captured during 2 of these runs on top of the v8
> series:
>                 vanilla         spf
> faults          89.418          101.364         +13%
> spf                n/a           97.989
> 
> With the SPF kernel, most of the page fault were processed in a speculative
> way.
> 
> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
> it a try on an android device. He reported that the application launch time
> was improved in average by 6%, and for large applications (~100 threads) by
> 20%.
> 
> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
> MSM845 (8 cores) with 6GB (the less is better):
> 
> Application                             4.9     4.9+spf delta
> com.tencent.mm                          416     389     -7%
> com.eg.android.AlipayGphone             1135    986     -13%
> com.tencent.mtt                         455     454     0%
> com.qqgame.hlddz                        1497    1409    -6%
> com.autonavi.minimap                    711     701     -1%
> com.tencent.tmgp.sgame                  788     748     -5%
> com.immomo.momo                         501     487     -3%
> com.tencent.peng                        2145    2112    -2%
> com.smile.gifmaker                      491     461     -6%
> com.baidu.BaiduMap                      479     366     -23%
> com.taobao.taobao                       1341    1198    -11%
> com.baidu.searchbox                     333     314     -6%
> com.tencent.mobileqq                    394     384     -3%
> com.sina.weibo                          907     906     0%
> com.youku.phone                         816     731     -11%
> com.happyelements.AndroidAnimal.qq      763     717     -6%
> com.UCMobile                            415     411     -1%
> com.tencent.tmgp.ak                     1464    1431    -2%
> com.tencent.qqmusic                     336     329     -2%
> com.sankuai.meituan                     1661    1302    -22%
> com.netease.cloudmusic                  1193    1200    1%
> air.tv.douyu.android                    4257    4152    -2%
> 
> ------------------
> Benchmarks results
> 
> Base kernel is v4.17.0-rc4-mm1
> SPF is BASE + this series
> 
> Kernbench:
> ----------
> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
> kernel (kernel is build 5 times):
> 
> Average Half load -j 8
>                  Run    (std deviation)
>                  BASE                   SPF
> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
> 
> Average Optimal load -j 16
>                  Run    (std deviation)
>                  BASE                   SPF
> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
> 
> 
> During a run on the SPF, perf events were captured:
>  Performance counter stats for '../kernbench -M':
>          526743764      faults
>                210      spf
>                  3      pagefault:spf_vma_changed
>                  0      pagefault:spf_vma_noanon
>               2278      pagefault:spf_vma_notsup
>                  0      pagefault:spf_vma_access
>                  0      pagefault:spf_pmd_changed
> 
> Very few speculative page faults were recorded as most of the processes
> involved are monothreaded (sounds that on this architecture some threads
> were created during the kernel build processing).
> 
> Here are the kerbench results on a 80 CPUs Power8 system:
> 
> Average Half load -j 40
>                  Run    (std deviation)
>                  BASE                   SPF
> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
> 
> Average Optimal load -j 80
>                  Run    (std deviation)
>                  BASE                   SPF
> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
> Context Switches 223861  (138865)       225032  (139632)        0.52%
> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
> 
> During a run on the SPF, perf events were captured:
>  Performance counter stats for '../kernbench -M':
>          116730856      faults
>                  0      spf
>                  3      pagefault:spf_vma_changed
>                  0      pagefault:spf_vma_noanon
>                476      pagefault:spf_vma_notsup
>                  0      pagefault:spf_vma_access
>                  0      pagefault:spf_pmd_changed
> 
> Most of the processes involved are monothreaded so SPF is not activated but
> there is no impact on the performance.
> 
> Ebizzy:
> -------
> The test is counting the number of records per second it can manage, the
> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
> consistent result I repeated the test 100 times and measure the average
> result. The number is the record processes per second, the higher is the
> best.
> 
>                 BASE            SPF             delta
> 16 CPUs x86 VM  742.57          1490.24         100.69%
> 80 CPUs P8 node 13105.4         24174.23        84.46%
> 
> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>  Performance counter stats for './ebizzy -mTt 16':
>            1706379      faults
>            1674599      spf
>              30588      pagefault:spf_vma_changed
>                  0      pagefault:spf_vma_noanon
>                363      pagefault:spf_vma_notsup
>                  0      pagefault:spf_vma_access
>                  0      pagefault:spf_pmd_changed
> 
> And the ones captured during a run on a 80 CPUs Power node:
>  Performance counter stats for './ebizzy -mTt 80':
>            1874773      faults
>            1461153      spf
>             413293      pagefault:spf_vma_changed
>                  0      pagefault:spf_vma_noanon
>                200      pagefault:spf_vma_notsup
>                  0      pagefault:spf_vma_access
>                  0      pagefault:spf_pmd_changed
> 
> In ebizzy's case most of the page fault were handled in a speculative way,
> leading the ebizzy performance boost.
> 
> ------------------
> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>    and Minchan Kim, hopefully.
>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>    __do_page_fault().
>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>    instead
>    of aborting the speculative page fault handling. Dropping the now
> useless
>    trace event pagefault:spf_pte_lock.
>  - No more try to reuse the fetched VMA during the speculative page fault
>    handling when retrying is needed. This adds a lot of complexity and
>    additional tests done didn't show a significant performance improvement.
>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
> 
> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
> [2] https://patchwork.kernel.org/patch/9999687/
> 
> 
> Laurent Dufour (20):
>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>   mm: make pte_unmap_same compatible with SPF
>   mm: introduce INIT_VMA()
>   mm: protect VMA modifications using VMA sequence count
>   mm: protect mremap() against SPF hanlder
>   mm: protect SPF handler against anon_vma changes
>   mm: cache some VMA fields in the vm_fault structure
>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>   mm: introduce __lru_cache_add_active_or_unevictable
>   mm: introduce __vm_normal_page()
>   mm: introduce __page_add_new_anon_rmap()
>   mm: protect mm_rb tree with a rwlock
>   mm: adding speculative page fault failure trace events
>   perf: add a speculative page fault sw event
>   perf tools: add support for the SPF perf event
>   mm: add speculative page fault vmstats
>   powerpc/mm: add speculative page fault
> 
> Mahendran Ganesh (2):
>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>   arm64/mm: add speculative page fault
> 
> Peter Zijlstra (4):
>   mm: prepare for FAULT_FLAG_SPECULATIVE
>   mm: VMA sequence count
>   mm: provide speculative fault infrastructure
>   x86/mm: add speculative pagefault handling
> 
>  arch/arm64/Kconfig                    |   1 +
>  arch/arm64/mm/fault.c                 |  12 +
>  arch/powerpc/Kconfig                  |   1 +
>  arch/powerpc/mm/fault.c               |  16 +
>  arch/x86/Kconfig                      |   1 +
>  arch/x86/mm/fault.c                   |  27 +-
>  fs/exec.c                             |   2 +-
>  fs/proc/task_mmu.c                    |   5 +-
>  fs/userfaultfd.c                      |  17 +-
>  include/linux/hugetlb_inline.h        |   2 +-
>  include/linux/migrate.h               |   4 +-
>  include/linux/mm.h                    | 136 +++++++-
>  include/linux/mm_types.h              |   7 +
>  include/linux/pagemap.h               |   4 +-
>  include/linux/rmap.h                  |  12 +-
>  include/linux/swap.h                  |  10 +-
>  include/linux/vm_event_item.h         |   3 +
>  include/trace/events/pagefault.h      |  80 +++++
>  include/uapi/linux/perf_event.h       |   1 +
>  kernel/fork.c                         |   5 +-
>  mm/Kconfig                            |  22 ++
>  mm/huge_memory.c                      |   6 +-
>  mm/hugetlb.c                          |   2 +
>  mm/init-mm.c                          |   3 +
>  mm/internal.h                         |  20 ++
>  mm/khugepaged.c                       |   5 +
>  mm/madvise.c                          |   6 +-
>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>  mm/mempolicy.c                        |  51 ++-
>  mm/migrate.c                          |   6 +-
>  mm/mlock.c                            |  13 +-
>  mm/mmap.c                             | 229 ++++++++++---
>  mm/mprotect.c                         |   4 +-
>  mm/mremap.c                           |  13 +
>  mm/nommu.c                            |   2 +-
>  mm/rmap.c                             |   5 +-
>  mm/swap.c                             |   6 +-
>  mm/swap_state.c                       |   8 +-
>  mm/vmstat.c                           |   5 +-
>  tools/include/uapi/linux/perf_event.h |   1 +
>  tools/perf/util/evsel.c               |   1 +
>  tools/perf/util/parse-events.c        |   4 +
>  tools/perf/util/parse-events.l        |   1 +
>  tools/perf/util/python.c              |   1 +
>  44 files changed, 1161 insertions(+), 211 deletions(-)
>  create mode 100644 include/trace/events/pagefault.h
> 
> --
> 2.7.4
> 
> 

Re: [PATCH v11 00/26] Speculative page faults

[Haiyan Song]

Hi Laurent,

Yes, these tests are done on V9 patch.


Best regards,
Haiyan Song

On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
> On 28/05/2018 07:23, Song, HaiyanX wrote:
> > 
> > Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
> > tested on Intel 4s Skylake platform.
> 
> Hi,
> 
> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
> series" while responding to the v11 header series...
> Were these tests done on v9 or v11 ?
> 
> Cheers,
> Laurent.
> 
> > 
> > The regression result is sorted by the metric will-it-scale.per_thread_ops.
> > Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
> > Commit id:
> >     base commit: d55f34411b1b126429a823d06c3124c16283231f
> >     head commit: 0355322b3577eeab7669066df42c550a56801110
> > Benchmark suite: will-it-scale
> > Download link:
> > https://github.com/antonblanchard/will-it-scale/tree/master/tests
> > Metrics:
> >     will-it-scale.per_process_ops=processes/nr_cpu
> >     will-it-scale.per_thread_ops=threads/nr_cpu
> > test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
> > THP: enable / disable
> > nr_task: 100%
> > 
> > 1. Regressions:
> > a) THP enabled:
> > testcase                        base            change          head       metric
> > page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
> > page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
> > brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
> > page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
> > signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
> > 
> > b) THP disabled:
> > testcase                        base            change          head       metric
> > page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
> > page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
> > context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
> > brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
> > page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
> > signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
> > 
> > 2. Improvements:
> > a) THP enabled:
> > testcase                        base            change          head       metric
> > malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
> > writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
> > signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
> > 
> > b) THP disabled:
> > testcase                        base            change          head       metric
> > malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
> > read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
> > page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
> > read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
> > writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
> > signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
> > 
> > Notes: for above values in column "change", the higher value means that the related testcase result
> > on head commit is better than that on base commit for this benchmark.
> > 
> > 
> > Best regards
> > Haiyan Song
> > 
> > ________________________________________
> > From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> > Sent: Thursday, May 17, 2018 7:06 PM
> > To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
> > Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> > Subject: [PATCH v11 00/26] Speculative page faults
> > 
> > This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
> > page fault without holding the mm semaphore [1].
> > 
> > The idea is to try to handle user space page faults without holding the
> > mmap_sem. This should allow better concurrency for massively threaded
> > process since the page fault handler will not wait for other threads memory
> > layout change to be done, assuming that this change is done in another part
> > of the process's memory space. This type page fault is named speculative
> > page fault. If the speculative page fault fails because of a concurrency is
> > detected or because underlying PMD or PTE tables are not yet allocating, it
> > is failing its processing and a classic page fault is then tried.
> > 
> > The speculative page fault (SPF) has to look for the VMA matching the fault
> > address without holding the mmap_sem, this is done by introducing a rwlock
> > which protects the access to the mm_rb tree. Previously this was done using
> > SRCU but it was introducing a lot of scheduling to process the VMA's
> > freeing operation which was hitting the performance by 20% as reported by
> > Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
> > limiting the locking contention to these operations which are expected to
> > be in a O(log n) order. In addition to ensure that the VMA is not freed in
> > our back a reference count is added and 2 services (get_vma() and
> > put_vma()) are introduced to handle the reference count. Once a VMA is
> > fetched from the RB tree using get_vma(), it must be later freed using
> > put_vma(). I can't see anymore the overhead I got while will-it-scale
> > benchmark anymore.
> > 
> > The VMA's attributes checked during the speculative page fault processing
> > have to be protected against parallel changes. This is done by using a per
> > VMA sequence lock. This sequence lock allows the speculative page fault
> > handler to fast check for parallel changes in progress and to abort the
> > speculative page fault in that case.
> > 
> > Once the VMA has been found, the speculative page fault handler would check
> > for the VMA's attributes to verify that the page fault has to be handled
> > correctly or not. Thus, the VMA is protected through a sequence lock which
> > allows fast detection of concurrent VMA changes. If such a change is
> > detected, the speculative page fault is aborted and a *classic* page fault
> > is tried.  VMA sequence lockings are added when VMA attributes which are
> > checked during the page fault are modified.
> > 
> > When the PTE is fetched, the VMA is checked to see if it has been changed,
> > so once the page table is locked, the VMA is valid, so any other changes
> > leading to touching this PTE will need to lock the page table, so no
> > parallel change is possible at this time.
> > 
> > The locking of the PTE is done with interrupts disabled, this allows
> > checking for the PMD to ensure that there is not an ongoing collapsing
> > operation. Since khugepaged is firstly set the PMD to pmd_none and then is
> > waiting for the other CPU to have caught the IPI interrupt, if the pmd is
> > valid at the time the PTE is locked, we have the guarantee that the
> > collapsing operation will have to wait on the PTE lock to move forward.
> > This allows the SPF handler to map the PTE safely. If the PMD value is
> > different from the one recorded at the beginning of the SPF operation, the
> > classic page fault handler will be called to handle the operation while
> > holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
> > the lock is done using spin_trylock() to avoid dead lock when handling a
> > page fault while a TLB invalidate is requested by another CPU holding the
> > PTE.
> > 
> > In pseudo code, this could be seen as:
> >     speculative_page_fault()
> >     {
> >             vma = get_vma()
> >             check vma sequence count
> >             check vma's support
> >             disable interrupt
> >                   check pgd,p4d,...,pte
> >                   save pmd and pte in vmf
> >                   save vma sequence counter in vmf
> >             enable interrupt
> >             check vma sequence count
> >             handle_pte_fault(vma)
> >                     ..
> >                     page = alloc_page()
> >                     pte_map_lock()
> >                             disable interrupt
> >                                     abort if sequence counter has changed
> >                                     abort if pmd or pte has changed
> >                                     pte map and lock
> >                             enable interrupt
> >                     if abort
> >                        free page
> >                        abort
> >                     ...
> >     }
> > 
> >     arch_fault_handler()
> >     {
> >             if (speculative_page_fault(&vma))
> >                goto done
> >     again:
> >             lock(mmap_sem)
> >             vma = find_vma();
> >             handle_pte_fault(vma);
> >             if retry
> >                unlock(mmap_sem)
> >                goto again;
> >     done:
> >             handle fault error
> >     }
> > 
> > Support for THP is not done because when checking for the PMD, we can be
> > confused by an in progress collapsing operation done by khugepaged. The
> > issue is that pmd_none() could be true either if the PMD is not already
> > populated or if the underlying PTE are in the way to be collapsed. So we
> > cannot safely allocate a PMD if pmd_none() is true.
> > 
> > This series add a new software performance event named 'speculative-faults'
> > or 'spf'. It counts the number of successful page fault event handled
> > speculatively. When recording 'faults,spf' events, the faults one is
> > counting the total number of page fault events while 'spf' is only counting
> > the part of the faults processed speculatively.
> > 
> > There are some trace events introduced by this series. They allow
> > identifying why the page faults were not processed speculatively. This
> > doesn't take in account the faults generated by a monothreaded process
> > which directly processed while holding the mmap_sem. This trace events are
> > grouped in a system named 'pagefault', they are:
> >  - pagefault:spf_vma_changed : if the VMA has been changed in our back
> >  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
> >  - pagefault:spf_vma_notsup : the VMA's type is not supported
> >  - pagefault:spf_vma_access : the VMA's access right are not respected
> >  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
> >    back.
> > 
> > To record all the related events, the easier is to run perf with the
> > following arguments :
> > $ perf stat -e 'faults,spf,pagefault:*' <command>
> > 
> > There is also a dedicated vmstat counter showing the number of successful
> > page fault handled speculatively. I can be seen this way:
> > $ grep speculative_pgfault /proc/vmstat
> > 
> > This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
> > on x86, PowerPC and arm64.
> > 
> > ---------------------
> > Real Workload results
> > 
> > As mentioned in previous email, we did non official runs using a "popular
> > in memory multithreaded database product" on 176 cores SMT8 Power system
> > which showed a 30% improvements in the number of transaction processed per
> > second. This run has been done on the v6 series, but changes introduced in
> > this new version should not impact the performance boost seen.
> > 
> > Here are the perf data captured during 2 of these runs on top of the v8
> > series:
> >                 vanilla         spf
> > faults          89.418          101.364         +13%
> > spf                n/a           97.989
> > 
> > With the SPF kernel, most of the page fault were processed in a speculative
> > way.
> > 
> > Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
> > it a try on an android device. He reported that the application launch time
> > was improved in average by 6%, and for large applications (~100 threads) by
> > 20%.
> > 
> > Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
> > MSM845 (8 cores) with 6GB (the less is better):
> > 
> > Application                             4.9     4.9+spf delta
> > com.tencent.mm                          416     389     -7%
> > com.eg.android.AlipayGphone             1135    986     -13%
> > com.tencent.mtt                         455     454     0%
> > com.qqgame.hlddz                        1497    1409    -6%
> > com.autonavi.minimap                    711     701     -1%
> > com.tencent.tmgp.sgame                  788     748     -5%
> > com.immomo.momo                         501     487     -3%
> > com.tencent.peng                        2145    2112    -2%
> > com.smile.gifmaker                      491     461     -6%
> > com.baidu.BaiduMap                      479     366     -23%
> > com.taobao.taobao                       1341    1198    -11%
> > com.baidu.searchbox                     333     314     -6%
> > com.tencent.mobileqq                    394     384     -3%
> > com.sina.weibo                          907     906     0%
> > com.youku.phone                         816     731     -11%
> > com.happyelements.AndroidAnimal.qq      763     717     -6%
> > com.UCMobile                            415     411     -1%
> > com.tencent.tmgp.ak                     1464    1431    -2%
> > com.tencent.qqmusic                     336     329     -2%
> > com.sankuai.meituan                     1661    1302    -22%
> > com.netease.cloudmusic                  1193    1200    1%
> > air.tv.douyu.android                    4257    4152    -2%
> > 
> > ------------------
> > Benchmarks results
> > 
> > Base kernel is v4.17.0-rc4-mm1
> > SPF is BASE + this series
> > 
> > Kernbench:
> > ----------
> > Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
> > kernel (kernel is build 5 times):
> > 
> > Average Half load -j 8
> >                  Run    (std deviation)
> >                  BASE                   SPF
> > Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
> > User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
> > System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
> > Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
> > Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
> > Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
> > 
> > Average Optimal load -j 16
> >                  Run    (std deviation)
> >                  BASE                   SPF
> > Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
> > User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
> > System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
> > Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
> > Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
> > Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
> > 
> > 
> > During a run on the SPF, perf events were captured:
> >  Performance counter stats for '../kernbench -M':
> >          526743764      faults
> >                210      spf
> >                  3      pagefault:spf_vma_changed
> >                  0      pagefault:spf_vma_noanon
> >               2278      pagefault:spf_vma_notsup
> >                  0      pagefault:spf_vma_access
> >                  0      pagefault:spf_pmd_changed
> > 
> > Very few speculative page faults were recorded as most of the processes
> > involved are monothreaded (sounds that on this architecture some threads
> > were created during the kernel build processing).
> > 
> > Here are the kerbench results on a 80 CPUs Power8 system:
> > 
> > Average Half load -j 40
> >                  Run    (std deviation)
> >                  BASE                   SPF
> > Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
> > User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
> > System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
> > Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
> > Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
> > Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
> > 
> > Average Optimal load -j 80
> >                  Run    (std deviation)
> >                  BASE                   SPF
> > Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
> > User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
> > System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
> > Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
> > Context Switches 223861  (138865)       225032  (139632)        0.52%
> > Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
> > 
> > During a run on the SPF, perf events were captured:
> >  Performance counter stats for '../kernbench -M':
> >          116730856      faults
> >                  0      spf
> >                  3      pagefault:spf_vma_changed
> >                  0      pagefault:spf_vma_noanon
> >                476      pagefault:spf_vma_notsup
> >                  0      pagefault:spf_vma_access
> >                  0      pagefault:spf_pmd_changed
> > 
> > Most of the processes involved are monothreaded so SPF is not activated but
> > there is no impact on the performance.
> > 
> > Ebizzy:
> > -------
> > The test is counting the number of records per second it can manage, the
> > higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
> > consistent result I repeated the test 100 times and measure the average
> > result. The number is the record processes per second, the higher is the
> > best.
> > 
> >                 BASE            SPF             delta
> > 16 CPUs x86 VM  742.57          1490.24         100.69%
> > 80 CPUs P8 node 13105.4         24174.23        84.46%
> > 
> > Here are the performance counter read during a run on a 16 CPUs x86 VM:
> >  Performance counter stats for './ebizzy -mTt 16':
> >            1706379      faults
> >            1674599      spf
> >              30588      pagefault:spf_vma_changed
> >                  0      pagefault:spf_vma_noanon
> >                363      pagefault:spf_vma_notsup
> >                  0      pagefault:spf_vma_access
> >                  0      pagefault:spf_pmd_changed
> > 
> > And the ones captured during a run on a 80 CPUs Power node:
> >  Performance counter stats for './ebizzy -mTt 80':
> >            1874773      faults
> >            1461153      spf
> >             413293      pagefault:spf_vma_changed
> >                  0      pagefault:spf_vma_noanon
> >                200      pagefault:spf_vma_notsup
> >                  0      pagefault:spf_vma_access
> >                  0      pagefault:spf_pmd_changed
> > 
> > In ebizzy's case most of the page fault were handled in a speculative way,
> > leading the ebizzy performance boost.
> > 
> > ------------------
> > Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
> >  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
> >    and Minchan Kim, hopefully.
> >  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
> >    __do_page_fault().
> >  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
> >    instead
> >    of aborting the speculative page fault handling. Dropping the now
> > useless
> >    trace event pagefault:spf_pte_lock.
> >  - No more try to reuse the fetched VMA during the speculative page fault
> >    handling when retrying is needed. This adds a lot of complexity and
> >    additional tests done didn't show a significant performance improvement.
> >  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
> > 
> > [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
> > [2] https://patchwork.kernel.org/patch/9999687/
> > 
> > 
> > Laurent Dufour (20):
> >   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
> >   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> >   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> >   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
> >   mm: make pte_unmap_same compatible with SPF
> >   mm: introduce INIT_VMA()
> >   mm: protect VMA modifications using VMA sequence count
> >   mm: protect mremap() against SPF hanlder
> >   mm: protect SPF handler against anon_vma changes
> >   mm: cache some VMA fields in the vm_fault structure
> >   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
> >   mm: introduce __lru_cache_add_active_or_unevictable
> >   mm: introduce __vm_normal_page()
> >   mm: introduce __page_add_new_anon_rmap()
> >   mm: protect mm_rb tree with a rwlock
> >   mm: adding speculative page fault failure trace events
> >   perf: add a speculative page fault sw event
> >   perf tools: add support for the SPF perf event
> >   mm: add speculative page fault vmstats
> >   powerpc/mm: add speculative page fault
> > 
> > Mahendran Ganesh (2):
> >   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> >   arm64/mm: add speculative page fault
> > 
> > Peter Zijlstra (4):
> >   mm: prepare for FAULT_FLAG_SPECULATIVE
> >   mm: VMA sequence count
> >   mm: provide speculative fault infrastructure
> >   x86/mm: add speculative pagefault handling
> > 
> >  arch/arm64/Kconfig                    |   1 +
> >  arch/arm64/mm/fault.c                 |  12 +
> >  arch/powerpc/Kconfig                  |   1 +
> >  arch/powerpc/mm/fault.c               |  16 +
> >  arch/x86/Kconfig                      |   1 +
> >  arch/x86/mm/fault.c                   |  27 +-
> >  fs/exec.c                             |   2 +-
> >  fs/proc/task_mmu.c                    |   5 +-
> >  fs/userfaultfd.c                      |  17 +-
> >  include/linux/hugetlb_inline.h        |   2 +-
> >  include/linux/migrate.h               |   4 +-
> >  include/linux/mm.h                    | 136 +++++++-
> >  include/linux/mm_types.h              |   7 +
> >  include/linux/pagemap.h               |   4 +-
> >  include/linux/rmap.h                  |  12 +-
> >  include/linux/swap.h                  |  10 +-
> >  include/linux/vm_event_item.h         |   3 +
> >  include/trace/events/pagefault.h      |  80 +++++
> >  include/uapi/linux/perf_event.h       |   1 +
> >  kernel/fork.c                         |   5 +-
> >  mm/Kconfig                            |  22 ++
> >  mm/huge_memory.c                      |   6 +-
> >  mm/hugetlb.c                          |   2 +
> >  mm/init-mm.c                          |   3 +
> >  mm/internal.h                         |  20 ++
> >  mm/khugepaged.c                       |   5 +
> >  mm/madvise.c                          |   6 +-
> >  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
> >  mm/mempolicy.c                        |  51 ++-
> >  mm/migrate.c                          |   6 +-
> >  mm/mlock.c                            |  13 +-
> >  mm/mmap.c                             | 229 ++++++++++---
> >  mm/mprotect.c                         |   4 +-
> >  mm/mremap.c                           |  13 +
> >  mm/nommu.c                            |   2 +-
> >  mm/rmap.c                             |   5 +-
> >  mm/swap.c                             |   6 +-
> >  mm/swap_state.c                       |   8 +-
> >  mm/vmstat.c                           |   5 +-
> >  tools/include/uapi/linux/perf_event.h |   1 +
> >  tools/perf/util/evsel.c               |   1 +
> >  tools/perf/util/parse-events.c        |   4 +
> >  tools/perf/util/parse-events.l        |   1 +
> >  tools/perf/util/python.c              |   1 +
> >  44 files changed, 1161 insertions(+), 211 deletions(-)
> >  create mode 100644 include/trace/events/pagefault.h
> > 
> > --
> > 2.7.4
> > 
> > 
> 

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

On 28/05/2018 10:22, Haiyan Song wrote:
> Hi Laurent,
> 
> Yes, these tests are done on V9 patch.

Do you plan to give this V11 a run ?

> 
> 
> Best regards,
> Haiyan Song
> 
> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>
>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>> tested on Intel 4s Skylake platform.
>>
>> Hi,
>>
>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>> series" while responding to the v11 header series...
>> Were these tests done on v9 or v11 ?
>>
>> Cheers,
>> Laurent.
>>
>>>
>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>> Commit id:
>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>> Benchmark suite: will-it-scale
>>> Download link:
>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>> Metrics:
>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>> THP: enable / disable
>>> nr_task: 100%
>>>
>>> 1. Regressions:
>>> a) THP enabled:
>>> testcase                        base            change          head       metric
>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>
>>> b) THP disabled:
>>> testcase                        base            change          head       metric
>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>
>>> 2. Improvements:
>>> a) THP enabled:
>>> testcase                        base            change          head       metric
>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>
>>> b) THP disabled:
>>> testcase                        base            change          head       metric
>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>
>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>> on head commit is better than that on base commit for this benchmark.
>>>
>>>
>>> Best regards
>>> Haiyan Song
>>>
>>> ________________________________________
>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Thursday, May 17, 2018 7:06 PM
>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>
>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>> page fault without holding the mm semaphore [1].
>>>
>>> The idea is to try to handle user space page faults without holding the
>>> mmap_sem. This should allow better concurrency for massively threaded
>>> process since the page fault handler will not wait for other threads memory
>>> layout change to be done, assuming that this change is done in another part
>>> of the process's memory space. This type page fault is named speculative
>>> page fault. If the speculative page fault fails because of a concurrency is
>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>> is failing its processing and a classic page fault is then tried.
>>>
>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>> which protects the access to the mm_rb tree. Previously this was done using
>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>> freeing operation which was hitting the performance by 20% as reported by
>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>> limiting the locking contention to these operations which are expected to
>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>> our back a reference count is added and 2 services (get_vma() and
>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>> fetched from the RB tree using get_vma(), it must be later freed using
>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>> benchmark anymore.
>>>
>>> The VMA's attributes checked during the speculative page fault processing
>>> have to be protected against parallel changes. This is done by using a per
>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>> handler to fast check for parallel changes in progress and to abort the
>>> speculative page fault in that case.
>>>
>>> Once the VMA has been found, the speculative page fault handler would check
>>> for the VMA's attributes to verify that the page fault has to be handled
>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>> allows fast detection of concurrent VMA changes. If such a change is
>>> detected, the speculative page fault is aborted and a *classic* page fault
>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>> checked during the page fault are modified.
>>>
>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>> so once the page table is locked, the VMA is valid, so any other changes
>>> leading to touching this PTE will need to lock the page table, so no
>>> parallel change is possible at this time.
>>>
>>> The locking of the PTE is done with interrupts disabled, this allows
>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>> valid at the time the PTE is locked, we have the guarantee that the
>>> collapsing operation will have to wait on the PTE lock to move forward.
>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>> different from the one recorded at the beginning of the SPF operation, the
>>> classic page fault handler will be called to handle the operation while
>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>> page fault while a TLB invalidate is requested by another CPU holding the
>>> PTE.
>>>
>>> In pseudo code, this could be seen as:
>>>     speculative_page_fault()
>>>     {
>>>             vma = get_vma()
>>>             check vma sequence count
>>>             check vma's support
>>>             disable interrupt
>>>                   check pgd,p4d,...,pte
>>>                   save pmd and pte in vmf
>>>                   save vma sequence counter in vmf
>>>             enable interrupt
>>>             check vma sequence count
>>>             handle_pte_fault(vma)
>>>                     ..
>>>                     page = alloc_page()
>>>                     pte_map_lock()
>>>                             disable interrupt
>>>                                     abort if sequence counter has changed
>>>                                     abort if pmd or pte has changed
>>>                                     pte map and lock
>>>                             enable interrupt
>>>                     if abort
>>>                        free page
>>>                        abort
>>>                     ...
>>>     }
>>>
>>>     arch_fault_handler()
>>>     {
>>>             if (speculative_page_fault(&vma))
>>>                goto done
>>>     again:
>>>             lock(mmap_sem)
>>>             vma = find_vma();
>>>             handle_pte_fault(vma);
>>>             if retry
>>>                unlock(mmap_sem)
>>>                goto again;
>>>     done:
>>>             handle fault error
>>>     }
>>>
>>> Support for THP is not done because when checking for the PMD, we can be
>>> confused by an in progress collapsing operation done by khugepaged. The
>>> issue is that pmd_none() could be true either if the PMD is not already
>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>> cannot safely allocate a PMD if pmd_none() is true.
>>>
>>> This series add a new software performance event named 'speculative-faults'
>>> or 'spf'. It counts the number of successful page fault event handled
>>> speculatively. When recording 'faults,spf' events, the faults one is
>>> counting the total number of page fault events while 'spf' is only counting
>>> the part of the faults processed speculatively.
>>>
>>> There are some trace events introduced by this series. They allow
>>> identifying why the page faults were not processed speculatively. This
>>> doesn't take in account the faults generated by a monothreaded process
>>> which directly processed while holding the mmap_sem. This trace events are
>>> grouped in a system named 'pagefault', they are:
>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>    back.
>>>
>>> To record all the related events, the easier is to run perf with the
>>> following arguments :
>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>
>>> There is also a dedicated vmstat counter showing the number of successful
>>> page fault handled speculatively. I can be seen this way:
>>> $ grep speculative_pgfault /proc/vmstat
>>>
>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>> on x86, PowerPC and arm64.
>>>
>>> ---------------------
>>> Real Workload results
>>>
>>> As mentioned in previous email, we did non official runs using a "popular
>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>> which showed a 30% improvements in the number of transaction processed per
>>> second. This run has been done on the v6 series, but changes introduced in
>>> this new version should not impact the performance boost seen.
>>>
>>> Here are the perf data captured during 2 of these runs on top of the v8
>>> series:
>>>                 vanilla         spf
>>> faults          89.418          101.364         +13%
>>> spf                n/a           97.989
>>>
>>> With the SPF kernel, most of the page fault were processed in a speculative
>>> way.
>>>
>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>> it a try on an android device. He reported that the application launch time
>>> was improved in average by 6%, and for large applications (~100 threads) by
>>> 20%.
>>>
>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>> MSM845 (8 cores) with 6GB (the less is better):
>>>
>>> Application                             4.9     4.9+spf delta
>>> com.tencent.mm                          416     389     -7%
>>> com.eg.android.AlipayGphone             1135    986     -13%
>>> com.tencent.mtt                         455     454     0%
>>> com.qqgame.hlddz                        1497    1409    -6%
>>> com.autonavi.minimap                    711     701     -1%
>>> com.tencent.tmgp.sgame                  788     748     -5%
>>> com.immomo.momo                         501     487     -3%
>>> com.tencent.peng                        2145    2112    -2%
>>> com.smile.gifmaker                      491     461     -6%
>>> com.baidu.BaiduMap                      479     366     -23%
>>> com.taobao.taobao                       1341    1198    -11%
>>> com.baidu.searchbox                     333     314     -6%
>>> com.tencent.mobileqq                    394     384     -3%
>>> com.sina.weibo                          907     906     0%
>>> com.youku.phone                         816     731     -11%
>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>> com.UCMobile                            415     411     -1%
>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>> com.tencent.qqmusic                     336     329     -2%
>>> com.sankuai.meituan                     1661    1302    -22%
>>> com.netease.cloudmusic                  1193    1200    1%
>>> air.tv.douyu.android                    4257    4152    -2%
>>>
>>> ------------------
>>> Benchmarks results
>>>
>>> Base kernel is v4.17.0-rc4-mm1
>>> SPF is BASE + this series
>>>
>>> Kernbench:
>>> ----------
>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>> kernel (kernel is build 5 times):
>>>
>>> Average Half load -j 8
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>
>>> Average Optimal load -j 16
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>
>>>
>>> During a run on the SPF, perf events were captured:
>>>  Performance counter stats for '../kernbench -M':
>>>          526743764      faults
>>>                210      spf
>>>                  3      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>               2278      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> Very few speculative page faults were recorded as most of the processes
>>> involved are monothreaded (sounds that on this architecture some threads
>>> were created during the kernel build processing).
>>>
>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>
>>> Average Half load -j 40
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>
>>> Average Optimal load -j 80
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>
>>> During a run on the SPF, perf events were captured:
>>>  Performance counter stats for '../kernbench -M':
>>>          116730856      faults
>>>                  0      spf
>>>                  3      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                476      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> Most of the processes involved are monothreaded so SPF is not activated but
>>> there is no impact on the performance.
>>>
>>> Ebizzy:
>>> -------
>>> The test is counting the number of records per second it can manage, the
>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>> consistent result I repeated the test 100 times and measure the average
>>> result. The number is the record processes per second, the higher is the
>>> best.
>>>
>>>                 BASE            SPF             delta
>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>
>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>  Performance counter stats for './ebizzy -mTt 16':
>>>            1706379      faults
>>>            1674599      spf
>>>              30588      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                363      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> And the ones captured during a run on a 80 CPUs Power node:
>>>  Performance counter stats for './ebizzy -mTt 80':
>>>            1874773      faults
>>>            1461153      spf
>>>             413293      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                200      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>> leading the ebizzy performance boost.
>>>
>>> ------------------
>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>    and Minchan Kim, hopefully.
>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>    __do_page_fault().
>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>    instead
>>>    of aborting the speculative page fault handling. Dropping the now
>>> useless
>>>    trace event pagefault:spf_pte_lock.
>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>    handling when retrying is needed. This adds a lot of complexity and
>>>    additional tests done didn't show a significant performance improvement.
>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>
>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>
>>>
>>> Laurent Dufour (20):
>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>   mm: make pte_unmap_same compatible with SPF
>>>   mm: introduce INIT_VMA()
>>>   mm: protect VMA modifications using VMA sequence count
>>>   mm: protect mremap() against SPF hanlder
>>>   mm: protect SPF handler against anon_vma changes
>>>   mm: cache some VMA fields in the vm_fault structure
>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>   mm: introduce __vm_normal_page()
>>>   mm: introduce __page_add_new_anon_rmap()
>>>   mm: protect mm_rb tree with a rwlock
>>>   mm: adding speculative page fault failure trace events
>>>   perf: add a speculative page fault sw event
>>>   perf tools: add support for the SPF perf event
>>>   mm: add speculative page fault vmstats
>>>   powerpc/mm: add speculative page fault
>>>
>>> Mahendran Ganesh (2):
>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   arm64/mm: add speculative page fault
>>>
>>> Peter Zijlstra (4):
>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>   mm: VMA sequence count
>>>   mm: provide speculative fault infrastructure
>>>   x86/mm: add speculative pagefault handling
>>>
>>>  arch/arm64/Kconfig                    |   1 +
>>>  arch/arm64/mm/fault.c                 |  12 +
>>>  arch/powerpc/Kconfig                  |   1 +
>>>  arch/powerpc/mm/fault.c               |  16 +
>>>  arch/x86/Kconfig                      |   1 +
>>>  arch/x86/mm/fault.c                   |  27 +-
>>>  fs/exec.c                             |   2 +-
>>>  fs/proc/task_mmu.c                    |   5 +-
>>>  fs/userfaultfd.c                      |  17 +-
>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>  include/linux/migrate.h               |   4 +-
>>>  include/linux/mm.h                    | 136 +++++++-
>>>  include/linux/mm_types.h              |   7 +
>>>  include/linux/pagemap.h               |   4 +-
>>>  include/linux/rmap.h                  |  12 +-
>>>  include/linux/swap.h                  |  10 +-
>>>  include/linux/vm_event_item.h         |   3 +
>>>  include/trace/events/pagefault.h      |  80 +++++
>>>  include/uapi/linux/perf_event.h       |   1 +
>>>  kernel/fork.c                         |   5 +-
>>>  mm/Kconfig                            |  22 ++
>>>  mm/huge_memory.c                      |   6 +-
>>>  mm/hugetlb.c                          |   2 +
>>>  mm/init-mm.c                          |   3 +
>>>  mm/internal.h                         |  20 ++
>>>  mm/khugepaged.c                       |   5 +
>>>  mm/madvise.c                          |   6 +-
>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>  mm/mempolicy.c                        |  51 ++-
>>>  mm/migrate.c                          |   6 +-
>>>  mm/mlock.c                            |  13 +-
>>>  mm/mmap.c                             | 229 ++++++++++---
>>>  mm/mprotect.c                         |   4 +-
>>>  mm/mremap.c                           |  13 +
>>>  mm/nommu.c                            |   2 +-
>>>  mm/rmap.c                             |   5 +-
>>>  mm/swap.c                             |   6 +-
>>>  mm/swap_state.c                       |   8 +-
>>>  mm/vmstat.c                           |   5 +-
>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>  tools/perf/util/evsel.c               |   1 +
>>>  tools/perf/util/parse-events.c        |   4 +
>>>  tools/perf/util/parse-events.l        |   1 +
>>>  tools/perf/util/python.c              |   1 +
>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>  create mode 100644 include/trace/events/pagefault.h
>>>
>>> --
>>> 2.7.4
>>>
>>>
>>
> 

RE: [PATCH v11 00/26] Speculative page faults

[Wang, Kemi]

Full run would take one or two weeks depended on our resource available. Could you pick some ones up, e.g. those have performance regression?

-----Original Message-----
From: owner-linux-mm@kvack.org [mailto:owner-linux-mm@kvack.org] On Behalf Of Laurent Dufour
Sent: Monday, May 28, 2018 4:55 PM
To: Song, HaiyanX <haiyanx.song@intel.com>
Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox <willy@infradead.org>; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner <tglx@linutronix.de>; Ingo Molnar <mingo@redhat.com>; hpa@zytor.com; Will Deacon <will.deacon@arm.com>; Sergey Senozhatsky <sergey.senozhatsky@gmail.com>; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli <aarcange@redhat.com>; Alexei Starovoitov <alexei.starovoitov@gmail.com>; Wang, Kemi <kemi.wang@intel.com>; Daniel Jordan <daniel.m.jordan@oracle.com>; David Rientjes <rientjes@google.com>; Jerome Glisse <jglisse@redhat.com>; Ganesh Mahendran <opensource.ganesh@gmail.com>; Minchan Kim <minchan@kernel.org>; Punit Agrawal <punitagrawal@gmail.com>; vinayak menon <vinayakm.list@gmail.com>; Yang Shi <yang.shi@linux.alibaba.com>; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen <tim.c.chen@linux.intel.com>; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
Subject: Re: [PATCH v11 00/26] Speculative page faults

On 28/05/2018 10:22, Haiyan Song wrote:
> Hi Laurent,
> 
> Yes, these tests are done on V9 patch.

Do you plan to give this V11 a run ?

> 
> 
> Best regards,
> Haiyan Song
> 
> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>
>>> Some regression and improvements is found by LKP-tools(linux kernel 
>>> performance) on V9 patch series tested on Intel 4s Skylake platform.
>>
>> Hi,
>>
>> Thanks for reporting this benchmark results, but you mentioned the 
>> "V9 patch series" while responding to the v11 header series...
>> Were these tests done on v9 or v11 ?
>>
>> Cheers,
>> Laurent.
>>
>>>
>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 
>>> patch series) Commit id:
>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>> Benchmark suite: will-it-scale
>>> Download link:
>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>> Metrics:
>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>> THP: enable / disable
>>> nr_task: 100%
>>>
>>> 1. Regressions:
>>> a) THP enabled:
>>> testcase                        base            change          head       metric
>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>
>>> b) THP disabled:
>>> testcase                        base            change          head       metric
>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>
>>> 2. Improvements:
>>> a) THP enabled:
>>> testcase                        base            change          head       metric
>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>
>>> b) THP disabled:
>>> testcase                        base            change          head       metric
>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>
>>> Notes: for above values in column "change", the higher value means 
>>> that the related testcase result on head commit is better than that on base commit for this benchmark.
>>>
>>>
>>> Best regards
>>> Haiyan Song
>>>
>>> ________________________________________
>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf 
>>> of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Thursday, May 17, 2018 7:06 PM
>>> To: akpm@linux-foundation.org; mhocko@kernel.org; 
>>> peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; 
>>> dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; 
>>> khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; 
>>> benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; 
>>> Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey 
>>> Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; 
>>> Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; 
>>> Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak 
>>> menon; Yang Shi
>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; 
>>> haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; 
>>> paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; 
>>> x86@kernel.org
>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>
>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to 
>>> handle page fault without holding the mm semaphore [1].
>>>
>>> The idea is to try to handle user space page faults without holding 
>>> the mmap_sem. This should allow better concurrency for massively 
>>> threaded process since the page fault handler will not wait for 
>>> other threads memory layout change to be done, assuming that this 
>>> change is done in another part of the process's memory space. This 
>>> type page fault is named speculative page fault. If the speculative 
>>> page fault fails because of a concurrency is detected or because 
>>> underlying PMD or PTE tables are not yet allocating, it is failing its processing and a classic page fault is then tried.
>>>
>>> The speculative page fault (SPF) has to look for the VMA matching 
>>> the fault address without holding the mmap_sem, this is done by 
>>> introducing a rwlock which protects the access to the mm_rb tree. 
>>> Previously this was done using SRCU but it was introducing a lot of 
>>> scheduling to process the VMA's freeing operation which was hitting 
>>> the performance by 20% as reported by Kemi Wang [2]. Using a rwlock 
>>> to protect access to the mm_rb tree is limiting the locking 
>>> contention to these operations which are expected to be in a O(log 
>>> n) order. In addition to ensure that the VMA is not freed in our 
>>> back a reference count is added and 2 services (get_vma() and
>>> put_vma()) are introduced to handle the reference count. Once a VMA 
>>> is fetched from the RB tree using get_vma(), it must be later freed 
>>> using put_vma(). I can't see anymore the overhead I got while 
>>> will-it-scale benchmark anymore.
>>>
>>> The VMA's attributes checked during the speculative page fault 
>>> processing have to be protected against parallel changes. This is 
>>> done by using a per VMA sequence lock. This sequence lock allows the 
>>> speculative page fault handler to fast check for parallel changes in 
>>> progress and to abort the speculative page fault in that case.
>>>
>>> Once the VMA has been found, the speculative page fault handler 
>>> would check for the VMA's attributes to verify that the page fault 
>>> has to be handled correctly or not. Thus, the VMA is protected 
>>> through a sequence lock which allows fast detection of concurrent 
>>> VMA changes. If such a change is detected, the speculative page 
>>> fault is aborted and a *classic* page fault is tried.  VMA sequence 
>>> lockings are added when VMA attributes which are checked during the page fault are modified.
>>>
>>> When the PTE is fetched, the VMA is checked to see if it has been 
>>> changed, so once the page table is locked, the VMA is valid, so any 
>>> other changes leading to touching this PTE will need to lock the 
>>> page table, so no parallel change is possible at this time.
>>>
>>> The locking of the PTE is done with interrupts disabled, this allows 
>>> checking for the PMD to ensure that there is not an ongoing 
>>> collapsing operation. Since khugepaged is firstly set the PMD to 
>>> pmd_none and then is waiting for the other CPU to have caught the 
>>> IPI interrupt, if the pmd is valid at the time the PTE is locked, we 
>>> have the guarantee that the collapsing operation will have to wait on the PTE lock to move forward.
>>> This allows the SPF handler to map the PTE safely. If the PMD value 
>>> is different from the one recorded at the beginning of the SPF 
>>> operation, the classic page fault handler will be called to handle 
>>> the operation while holding the mmap_sem. As the PTE lock is done 
>>> with the interrupts disabled, the lock is done using spin_trylock() 
>>> to avoid dead lock when handling a page fault while a TLB invalidate 
>>> is requested by another CPU holding the PTE.
>>>
>>> In pseudo code, this could be seen as:
>>>     speculative_page_fault()
>>>     {
>>>             vma = get_vma()
>>>             check vma sequence count
>>>             check vma's support
>>>             disable interrupt
>>>                   check pgd,p4d,...,pte
>>>                   save pmd and pte in vmf
>>>                   save vma sequence counter in vmf
>>>             enable interrupt
>>>             check vma sequence count
>>>             handle_pte_fault(vma)
>>>                     ..
>>>                     page = alloc_page()
>>>                     pte_map_lock()
>>>                             disable interrupt
>>>                                     abort if sequence counter has changed
>>>                                     abort if pmd or pte has changed
>>>                                     pte map and lock
>>>                             enable interrupt
>>>                     if abort
>>>                        free page
>>>                        abort
>>>                     ...
>>>     }
>>>
>>>     arch_fault_handler()
>>>     {
>>>             if (speculative_page_fault(&vma))
>>>                goto done
>>>     again:
>>>             lock(mmap_sem)
>>>             vma = find_vma();
>>>             handle_pte_fault(vma);
>>>             if retry
>>>                unlock(mmap_sem)
>>>                goto again;
>>>     done:
>>>             handle fault error
>>>     }
>>>
>>> Support for THP is not done because when checking for the PMD, we 
>>> can be confused by an in progress collapsing operation done by 
>>> khugepaged. The issue is that pmd_none() could be true either if the 
>>> PMD is not already populated or if the underlying PTE are in the way 
>>> to be collapsed. So we cannot safely allocate a PMD if pmd_none() is true.
>>>
>>> This series add a new software performance event named 'speculative-faults'
>>> or 'spf'. It counts the number of successful page fault event 
>>> handled speculatively. When recording 'faults,spf' events, the 
>>> faults one is counting the total number of page fault events while 
>>> 'spf' is only counting the part of the faults processed speculatively.
>>>
>>> There are some trace events introduced by this series. They allow 
>>> identifying why the page faults were not processed speculatively. 
>>> This doesn't take in account the faults generated by a monothreaded 
>>> process which directly processed while holding the mmap_sem. This 
>>> trace events are grouped in a system named 'pagefault', they are:
>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our 
>>> back
>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>  - pagefault:spf_vma_access : the VMA's access right are not 
>>> respected
>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>    back.
>>>
>>> To record all the related events, the easier is to run perf with the 
>>> following arguments :
>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>
>>> There is also a dedicated vmstat counter showing the number of 
>>> successful page fault handled speculatively. I can be seen this way:
>>> $ grep speculative_pgfault /proc/vmstat
>>>
>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is 
>>> functional on x86, PowerPC and arm64.
>>>
>>> ---------------------
>>> Real Workload results
>>>
>>> As mentioned in previous email, we did non official runs using a 
>>> "popular in memory multithreaded database product" on 176 cores SMT8 
>>> Power system which showed a 30% improvements in the number of 
>>> transaction processed per second. This run has been done on the v6 
>>> series, but changes introduced in this new version should not impact the performance boost seen.
>>>
>>> Here are the perf data captured during 2 of these runs on top of the 
>>> v8
>>> series:
>>>                 vanilla         spf
>>> faults          89.418          101.364         +13%
>>> spf                n/a           97.989
>>>
>>> With the SPF kernel, most of the page fault were processed in a 
>>> speculative way.
>>>
>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel 
>>> and gave it a try on an android device. He reported that the 
>>> application launch time was improved in average by 6%, and for large 
>>> applications (~100 threads) by 20%.
>>>
>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a 
>>> Qcom
>>> MSM845 (8 cores) with 6GB (the less is better):
>>>
>>> Application                             4.9     4.9+spf delta
>>> com.tencent.mm                          416     389     -7%
>>> com.eg.android.AlipayGphone             1135    986     -13%
>>> com.tencent.mtt                         455     454     0%
>>> com.qqgame.hlddz                        1497    1409    -6%
>>> com.autonavi.minimap                    711     701     -1%
>>> com.tencent.tmgp.sgame                  788     748     -5%
>>> com.immomo.momo                         501     487     -3%
>>> com.tencent.peng                        2145    2112    -2%
>>> com.smile.gifmaker                      491     461     -6%
>>> com.baidu.BaiduMap                      479     366     -23%
>>> com.taobao.taobao                       1341    1198    -11%
>>> com.baidu.searchbox                     333     314     -6%
>>> com.tencent.mobileqq                    394     384     -3%
>>> com.sina.weibo                          907     906     0%
>>> com.youku.phone                         816     731     -11%
>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>> com.UCMobile                            415     411     -1%
>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>> com.tencent.qqmusic                     336     329     -2%
>>> com.sankuai.meituan                     1661    1302    -22%
>>> com.netease.cloudmusic                  1193    1200    1%
>>> air.tv.douyu.android                    4257    4152    -2%
>>>
>>> ------------------
>>> Benchmarks results
>>>
>>> Base kernel is v4.17.0-rc4-mm1
>>> SPF is BASE + this series
>>>
>>> Kernbench:
>>> ----------
>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 
>>> 4.15 kernel (kernel is build 5 times):
>>>
>>> Average Half load -j 8
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>
>>> Average Optimal load -j 16
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>
>>>
>>> During a run on the SPF, perf events were captured:
>>>  Performance counter stats for '../kernbench -M':
>>>          526743764      faults
>>>                210      spf
>>>                  3      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>               2278      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> Very few speculative page faults were recorded as most of the 
>>> processes involved are monothreaded (sounds that on this 
>>> architecture some threads were created during the kernel build processing).
>>>
>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>
>>> Average Half load -j 40
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>
>>> Average Optimal load -j 80
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>
>>> During a run on the SPF, perf events were captured:
>>>  Performance counter stats for '../kernbench -M':
>>>          116730856      faults
>>>                  0      spf
>>>                  3      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                476      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> Most of the processes involved are monothreaded so SPF is not 
>>> activated but there is no impact on the performance.
>>>
>>> Ebizzy:
>>> -------
>>> The test is counting the number of records per second it can manage, 
>>> the higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. 
>>> To get consistent result I repeated the test 100 times and measure 
>>> the average result. The number is the record processes per second, 
>>> the higher is the best.
>>>
>>>                 BASE            SPF             delta
>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>
>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>  Performance counter stats for './ebizzy -mTt 16':
>>>            1706379      faults
>>>            1674599      spf
>>>              30588      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                363      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> And the ones captured during a run on a 80 CPUs Power node:
>>>  Performance counter stats for './ebizzy -mTt 80':
>>>            1874773      faults
>>>            1461153      spf
>>>             413293      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                200      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> In ebizzy's case most of the page fault were handled in a 
>>> speculative way, leading the ebizzy performance boost.
>>>
>>> ------------------
>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>    and Minchan Kim, hopefully.
>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>    __do_page_fault().
>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>    instead
>>>    of aborting the speculative page fault handling. Dropping the now 
>>> useless
>>>    trace event pagefault:spf_pte_lock.
>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>    handling when retrying is needed. This adds a lot of complexity and
>>>    additional tests done didn't show a significant performance improvement.
>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>
>>> [1] 
>>> http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-s
>>> peculative-page-faults-tt965642.html#none
>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>
>>>
>>> Laurent Dufour (20):
>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>   mm: make pte_unmap_same compatible with SPF
>>>   mm: introduce INIT_VMA()
>>>   mm: protect VMA modifications using VMA sequence count
>>>   mm: protect mremap() against SPF hanlder
>>>   mm: protect SPF handler against anon_vma changes
>>>   mm: cache some VMA fields in the vm_fault structure
>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>   mm: introduce __vm_normal_page()
>>>   mm: introduce __page_add_new_anon_rmap()
>>>   mm: protect mm_rb tree with a rwlock
>>>   mm: adding speculative page fault failure trace events
>>>   perf: add a speculative page fault sw event
>>>   perf tools: add support for the SPF perf event
>>>   mm: add speculative page fault vmstats
>>>   powerpc/mm: add speculative page fault
>>>
>>> Mahendran Ganesh (2):
>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   arm64/mm: add speculative page fault
>>>
>>> Peter Zijlstra (4):
>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>   mm: VMA sequence count
>>>   mm: provide speculative fault infrastructure
>>>   x86/mm: add speculative pagefault handling
>>>
>>>  arch/arm64/Kconfig                    |   1 +
>>>  arch/arm64/mm/fault.c                 |  12 +
>>>  arch/powerpc/Kconfig                  |   1 +
>>>  arch/powerpc/mm/fault.c               |  16 +
>>>  arch/x86/Kconfig                      |   1 +
>>>  arch/x86/mm/fault.c                   |  27 +-
>>>  fs/exec.c                             |   2 +-
>>>  fs/proc/task_mmu.c                    |   5 +-
>>>  fs/userfaultfd.c                      |  17 +-
>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>  include/linux/migrate.h               |   4 +-
>>>  include/linux/mm.h                    | 136 +++++++-
>>>  include/linux/mm_types.h              |   7 +
>>>  include/linux/pagemap.h               |   4 +-
>>>  include/linux/rmap.h                  |  12 +-
>>>  include/linux/swap.h                  |  10 +-
>>>  include/linux/vm_event_item.h         |   3 +
>>>  include/trace/events/pagefault.h      |  80 +++++
>>>  include/uapi/linux/perf_event.h       |   1 +
>>>  kernel/fork.c                         |   5 +-
>>>  mm/Kconfig                            |  22 ++
>>>  mm/huge_memory.c                      |   6 +-
>>>  mm/hugetlb.c                          |   2 +
>>>  mm/init-mm.c                          |   3 +
>>>  mm/internal.h                         |  20 ++
>>>  mm/khugepaged.c                       |   5 +
>>>  mm/madvise.c                          |   6 +-
>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>  mm/mempolicy.c                        |  51 ++-
>>>  mm/migrate.c                          |   6 +-
>>>  mm/mlock.c                            |  13 +-
>>>  mm/mmap.c                             | 229 ++++++++++---
>>>  mm/mprotect.c                         |   4 +-
>>>  mm/mremap.c                           |  13 +
>>>  mm/nommu.c                            |   2 +-
>>>  mm/rmap.c                             |   5 +-
>>>  mm/swap.c                             |   6 +-
>>>  mm/swap_state.c                       |   8 +-
>>>  mm/vmstat.c                           |   5 +-
>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>  tools/perf/util/evsel.c               |   1 +
>>>  tools/perf/util/parse-events.c        |   4 +
>>>  tools/perf/util/parse-events.l        |   1 +
>>>  tools/perf/util/python.c              |   1 +
>>>  44 files changed, 1161 insertions(+), 211 deletions(-)  create mode 
>>> 100644 include/trace/events/pagefault.h
>>>
>>> --
>>> 2.7.4
>>>
>>>
>>
> 

RE: [PATCH v11 00/26] Speculative page faults

[Wang, Kemi]
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RE: [PATCH v11 00/26] Speculative page faults

[Song, HaiyanX]

Hi Laurent,

Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
V9 patch serials.

The regression result is sorted by the metric will-it-scale.per_thread_ops.
branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
commit id:
  head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
  base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
Benchmark: will-it-scale
Download link: https://github.com/antonblanchard/will-it-scale/tree/master

Metrics:
  will-it-scale.per_process_ops=processes/nr_cpu
  will-it-scale.per_thread_ops=threads/nr_cpu
  test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
THP: enable / disable
nr_task:100%

1. Regressions:

a). Enable THP
testcase                          base           change      head           metric
page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops

b). Disable THP
page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops

Notes: for the above  values of test result, the higher is better.

2. Improvement: not found improvement based on the selected test cases.


Best regards
Haiyan Song
________________________________________
From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
Sent: Monday, May 28, 2018 4:54 PM
To: Song, HaiyanX
Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
Subject: Re: [PATCH v11 00/26] Speculative page faults

On 28/05/2018 10:22, Haiyan Song wrote:
> Hi Laurent,
>
> Yes, these tests are done on V9 patch.

Do you plan to give this V11 a run ?

>
>
> Best regards,
> Haiyan Song
>
> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>
>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>> tested on Intel 4s Skylake platform.
>>
>> Hi,
>>
>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>> series" while responding to the v11 header series...
>> Were these tests done on v9 or v11 ?
>>
>> Cheers,
>> Laurent.
>>
>>>
>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>> Commit id:
>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>> Benchmark suite: will-it-scale
>>> Download link:
>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>> Metrics:
>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>> THP: enable / disable
>>> nr_task: 100%
>>>
>>> 1. Regressions:
>>> a) THP enabled:
>>> testcase                        base            change          head       metric
>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>
>>> b) THP disabled:
>>> testcase                        base            change          head       metric
>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>
>>> 2. Improvements:
>>> a) THP enabled:
>>> testcase                        base            change          head       metric
>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>
>>> b) THP disabled:
>>> testcase                        base            change          head       metric
>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>
>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>> on head commit is better than that on base commit for this benchmark.
>>>
>>>
>>> Best regards
>>> Haiyan Song
>>>
>>> ________________________________________
>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Thursday, May 17, 2018 7:06 PM
>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>
>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>> page fault without holding the mm semaphore [1].
>>>
>>> The idea is to try to handle user space page faults without holding the
>>> mmap_sem. This should allow better concurrency for massively threaded
>>> process since the page fault handler will not wait for other threads memory
>>> layout change to be done, assuming that this change is done in another part
>>> of the process's memory space. This type page fault is named speculative
>>> page fault. If the speculative page fault fails because of a concurrency is
>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>> is failing its processing and a classic page fault is then tried.
>>>
>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>> which protects the access to the mm_rb tree. Previously this was done using
>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>> freeing operation which was hitting the performance by 20% as reported by
>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>> limiting the locking contention to these operations which are expected to
>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>> our back a reference count is added and 2 services (get_vma() and
>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>> fetched from the RB tree using get_vma(), it must be later freed using
>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>> benchmark anymore.
>>>
>>> The VMA's attributes checked during the speculative page fault processing
>>> have to be protected against parallel changes. This is done by using a per
>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>> handler to fast check for parallel changes in progress and to abort the
>>> speculative page fault in that case.
>>>
>>> Once the VMA has been found, the speculative page fault handler would check
>>> for the VMA's attributes to verify that the page fault has to be handled
>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>> allows fast detection of concurrent VMA changes. If such a change is
>>> detected, the speculative page fault is aborted and a *classic* page fault
>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>> checked during the page fault are modified.
>>>
>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>> so once the page table is locked, the VMA is valid, so any other changes
>>> leading to touching this PTE will need to lock the page table, so no
>>> parallel change is possible at this time.
>>>
>>> The locking of the PTE is done with interrupts disabled, this allows
>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>> valid at the time the PTE is locked, we have the guarantee that the
>>> collapsing operation will have to wait on the PTE lock to move forward.
>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>> different from the one recorded at the beginning of the SPF operation, the
>>> classic page fault handler will be called to handle the operation while
>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>> page fault while a TLB invalidate is requested by another CPU holding the
>>> PTE.
>>>
>>> In pseudo code, this could be seen as:
>>>     speculative_page_fault()
>>>     {
>>>             vma = get_vma()
>>>             check vma sequence count
>>>             check vma's support
>>>             disable interrupt
>>>                   check pgd,p4d,...,pte
>>>                   save pmd and pte in vmf
>>>                   save vma sequence counter in vmf
>>>             enable interrupt
>>>             check vma sequence count
>>>             handle_pte_fault(vma)
>>>                     ..
>>>                     page = alloc_page()
>>>                     pte_map_lock()
>>>                             disable interrupt
>>>                                     abort if sequence counter has changed
>>>                                     abort if pmd or pte has changed
>>>                                     pte map and lock
>>>                             enable interrupt
>>>                     if abort
>>>                        free page
>>>                        abort
>>>                     ...
>>>     }
>>>
>>>     arch_fault_handler()
>>>     {
>>>             if (speculative_page_fault(&vma))
>>>                goto done
>>>     again:
>>>             lock(mmap_sem)
>>>             vma = find_vma();
>>>             handle_pte_fault(vma);
>>>             if retry
>>>                unlock(mmap_sem)
>>>                goto again;
>>>     done:
>>>             handle fault error
>>>     }
>>>
>>> Support for THP is not done because when checking for the PMD, we can be
>>> confused by an in progress collapsing operation done by khugepaged. The
>>> issue is that pmd_none() could be true either if the PMD is not already
>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>> cannot safely allocate a PMD if pmd_none() is true.
>>>
>>> This series add a new software performance event named 'speculative-faults'
>>> or 'spf'. It counts the number of successful page fault event handled
>>> speculatively. When recording 'faults,spf' events, the faults one is
>>> counting the total number of page fault events while 'spf' is only counting
>>> the part of the faults processed speculatively.
>>>
>>> There are some trace events introduced by this series. They allow
>>> identifying why the page faults were not processed speculatively. This
>>> doesn't take in account the faults generated by a monothreaded process
>>> which directly processed while holding the mmap_sem. This trace events are
>>> grouped in a system named 'pagefault', they are:
>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>    back.
>>>
>>> To record all the related events, the easier is to run perf with the
>>> following arguments :
>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>
>>> There is also a dedicated vmstat counter showing the number of successful
>>> page fault handled speculatively. I can be seen this way:
>>> $ grep speculative_pgfault /proc/vmstat
>>>
>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>> on x86, PowerPC and arm64.
>>>
>>> ---------------------
>>> Real Workload results
>>>
>>> As mentioned in previous email, we did non official runs using a "popular
>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>> which showed a 30% improvements in the number of transaction processed per
>>> second. This run has been done on the v6 series, but changes introduced in
>>> this new version should not impact the performance boost seen.
>>>
>>> Here are the perf data captured during 2 of these runs on top of the v8
>>> series:
>>>                 vanilla         spf
>>> faults          89.418          101.364         +13%
>>> spf                n/a           97.989
>>>
>>> With the SPF kernel, most of the page fault were processed in a speculative
>>> way.
>>>
>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>> it a try on an android device. He reported that the application launch time
>>> was improved in average by 6%, and for large applications (~100 threads) by
>>> 20%.
>>>
>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>> MSM845 (8 cores) with 6GB (the less is better):
>>>
>>> Application                             4.9     4.9+spf delta
>>> com.tencent.mm                          416     389     -7%
>>> com.eg.android.AlipayGphone             1135    986     -13%
>>> com.tencent.mtt                         455     454     0%
>>> com.qqgame.hlddz                        1497    1409    -6%
>>> com.autonavi.minimap                    711     701     -1%
>>> com.tencent.tmgp.sgame                  788     748     -5%
>>> com.immomo.momo                         501     487     -3%
>>> com.tencent.peng                        2145    2112    -2%
>>> com.smile.gifmaker                      491     461     -6%
>>> com.baidu.BaiduMap                      479     366     -23%
>>> com.taobao.taobao                       1341    1198    -11%
>>> com.baidu.searchbox                     333     314     -6%
>>> com.tencent.mobileqq                    394     384     -3%
>>> com.sina.weibo                          907     906     0%
>>> com.youku.phone                         816     731     -11%
>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>> com.UCMobile                            415     411     -1%
>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>> com.tencent.qqmusic                     336     329     -2%
>>> com.sankuai.meituan                     1661    1302    -22%
>>> com.netease.cloudmusic                  1193    1200    1%
>>> air.tv.douyu.android                    4257    4152    -2%
>>>
>>> ------------------
>>> Benchmarks results
>>>
>>> Base kernel is v4.17.0-rc4-mm1
>>> SPF is BASE + this series
>>>
>>> Kernbench:
>>> ----------
>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>> kernel (kernel is build 5 times):
>>>
>>> Average Half load -j 8
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>
>>> Average Optimal load -j 16
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>
>>>
>>> During a run on the SPF, perf events were captured:
>>>  Performance counter stats for '../kernbench -M':
>>>          526743764      faults
>>>                210      spf
>>>                  3      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>               2278      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> Very few speculative page faults were recorded as most of the processes
>>> involved are monothreaded (sounds that on this architecture some threads
>>> were created during the kernel build processing).
>>>
>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>
>>> Average Half load -j 40
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>
>>> Average Optimal load -j 80
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>
>>> During a run on the SPF, perf events were captured:
>>>  Performance counter stats for '../kernbench -M':
>>>          116730856      faults
>>>                  0      spf
>>>                  3      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                476      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> Most of the processes involved are monothreaded so SPF is not activated but
>>> there is no impact on the performance.
>>>
>>> Ebizzy:
>>> -------
>>> The test is counting the number of records per second it can manage, the
>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>> consistent result I repeated the test 100 times and measure the average
>>> result. The number is the record processes per second, the higher is the
>>> best.
>>>
>>>                 BASE            SPF             delta
>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>
>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>  Performance counter stats for './ebizzy -mTt 16':
>>>            1706379      faults
>>>            1674599      spf
>>>              30588      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                363      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> And the ones captured during a run on a 80 CPUs Power node:
>>>  Performance counter stats for './ebizzy -mTt 80':
>>>            1874773      faults
>>>            1461153      spf
>>>             413293      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                200      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>> leading the ebizzy performance boost.
>>>
>>> ------------------
>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>    and Minchan Kim, hopefully.
>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>    __do_page_fault().
>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>    instead
>>>    of aborting the speculative page fault handling. Dropping the now
>>> useless
>>>    trace event pagefault:spf_pte_lock.
>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>    handling when retrying is needed. This adds a lot of complexity and
>>>    additional tests done didn't show a significant performance improvement.
>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>
>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>
>>>
>>> Laurent Dufour (20):
>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>   mm: make pte_unmap_same compatible with SPF
>>>   mm: introduce INIT_VMA()
>>>   mm: protect VMA modifications using VMA sequence count
>>>   mm: protect mremap() against SPF hanlder
>>>   mm: protect SPF handler against anon_vma changes
>>>   mm: cache some VMA fields in the vm_fault structure
>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>   mm: introduce __vm_normal_page()
>>>   mm: introduce __page_add_new_anon_rmap()
>>>   mm: protect mm_rb tree with a rwlock
>>>   mm: adding speculative page fault failure trace events
>>>   perf: add a speculative page fault sw event
>>>   perf tools: add support for the SPF perf event
>>>   mm: add speculative page fault vmstats
>>>   powerpc/mm: add speculative page fault
>>>
>>> Mahendran Ganesh (2):
>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   arm64/mm: add speculative page fault
>>>
>>> Peter Zijlstra (4):
>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>   mm: VMA sequence count
>>>   mm: provide speculative fault infrastructure
>>>   x86/mm: add speculative pagefault handling
>>>
>>>  arch/arm64/Kconfig                    |   1 +
>>>  arch/arm64/mm/fault.c                 |  12 +
>>>  arch/powerpc/Kconfig                  |   1 +
>>>  arch/powerpc/mm/fault.c               |  16 +
>>>  arch/x86/Kconfig                      |   1 +
>>>  arch/x86/mm/fault.c                   |  27 +-
>>>  fs/exec.c                             |   2 +-
>>>  fs/proc/task_mmu.c                    |   5 +-
>>>  fs/userfaultfd.c                      |  17 +-
>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>  include/linux/migrate.h               |   4 +-
>>>  include/linux/mm.h                    | 136 +++++++-
>>>  include/linux/mm_types.h              |   7 +
>>>  include/linux/pagemap.h               |   4 +-
>>>  include/linux/rmap.h                  |  12 +-
>>>  include/linux/swap.h                  |  10 +-
>>>  include/linux/vm_event_item.h         |   3 +
>>>  include/trace/events/pagefault.h      |  80 +++++
>>>  include/uapi/linux/perf_event.h       |   1 +
>>>  kernel/fork.c                         |   5 +-
>>>  mm/Kconfig                            |  22 ++
>>>  mm/huge_memory.c                      |   6 +-
>>>  mm/hugetlb.c                          |   2 +
>>>  mm/init-mm.c                          |   3 +
>>>  mm/internal.h                         |  20 ++
>>>  mm/khugepaged.c                       |   5 +
>>>  mm/madvise.c                          |   6 +-
>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>  mm/mempolicy.c                        |  51 ++-
>>>  mm/migrate.c                          |   6 +-
>>>  mm/mlock.c                            |  13 +-
>>>  mm/mmap.c                             | 229 ++++++++++---
>>>  mm/mprotect.c                         |   4 +-
>>>  mm/mremap.c                           |  13 +
>>>  mm/nommu.c                            |   2 +-
>>>  mm/rmap.c                             |   5 +-
>>>  mm/swap.c                             |   6 +-
>>>  mm/swap_state.c                       |   8 +-
>>>  mm/vmstat.c                           |   5 +-
>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>  tools/perf/util/evsel.c               |   1 +
>>>  tools/perf/util/parse-events.c        |   4 +
>>>  tools/perf/util/parse-events.l        |   1 +
>>>  tools/perf/util/python.c              |   1 +
>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>  create mode 100644 include/trace/events/pagefault.h
>>>
>>> --
>>> 2.7.4
>>>
>>>
>>
>

RE: [PATCH v11 00/26] Speculative page faults

[Song, HaiyanX]

Hi Laurent,

Regression test for v11 patch serials have been run, some regression is fou=
nd by LKP-tools (linux kernel performance)
tested on Intel 4s skylake platform. This time only test the cases which ha=
ve been run and found regressions on
V9 patch serials.

The regression result is sorted by the metric will-it-scale.per_thread_ops.
branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
commit id:
  head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
  base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
Benchmark: will-it-scale
Download link: https://github.com/antonblanchard/will-it-scale/tree/master

Metrics:
  will-it-scale.per_process_ops=3Dprocesses/nr_cpu
  will-it-scale.per_thread_ops=3Dthreads/nr_cpu
  test box: lkp-skl-4sp1(nr_cpu=3D192,memory=3D768G)
THP: enable / disable
nr_task:100%

1. Regressions:

a). Enable THP
testcase                          base           change      head          =
 metric
page_fault3/enable THP           10519          -20.5%        836      will=
-it-scale.per_thread_ops
page_fault2/enalbe THP            8281          -18.8%       6728      will=
-it-scale.per_thread_ops
brk1/eanble THP                 998475           -2.2%     976893      will=
-it-scale.per_process_ops
context_switch1/enable THP      223910           -1.3%     220930      will=
-it-scale.per_process_ops
context_switch1/enable THP      233722           -1.0%     231288      will=
-it-scale.per_thread_ops

b). Disable THP
page_fault3/disable THP          10856          -23.1%       8344      will=
-it-scale.per_thread_ops
page_fault2/disable THP           8147          -18.8%       6613      will=
-it-scale.per_thread_ops
brk1/disable THP                   957           -7.9%        881      will=
-it-scale.per_thread_ops
context_switch1/disable THP     237006           -2.2%     231907      will=
-it-scale.per_thread_ops
brk1/disable THP                997317           -2.0%     977778      will=
-it-scale.per_process_ops
page_fault3/disable THP         467454           -1.8%     459251      will=
-it-scale.per_process_ops
context_switch1/disable THP     224431           -1.3%     221567      will=
-it-scale.per_process_ops

Notes: for the above  values of test result, the higher is better.

2. Improvement: not found improvement based on the selected test cases.


Best regards
Haiyan Song
________________________________________
From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laur=
ent Dufour [ldufour@linux.vnet.ibm.com]
Sent: Monday, May 28, 2018 4:54 PM
To: Song, HaiyanX
Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kir=
ill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Mat=
thew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; =
benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Glei=
xner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.s=
enozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi=
; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan K=
im; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; l=
inux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora=
@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs=
.org; x86@kernel.org
Subject: Re: [PATCH v11 00/26] Speculative page faults

On 28/05/2018 10:22, Haiyan Song wrote:
> Hi Laurent,
>
> Yes, these tests are done on V9 patch.

Do you plan to give this V11 a run ?

>
>
> Best regards,
> Haiyan Song
>
> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>
>>> Some regression and improvements is found by LKP-tools(linux kernel per=
formance) on V9 patch series
>>> tested on Intel 4s Skylake platform.
>>
>> Hi,
>>
>> Thanks for reporting this benchmark results, but you mentioned the "V9 p=
atch
>> series" while responding to the v11 header series...
>> Were these tests done on v9 or v11 ?
>>
>> Cheers,
>> Laurent.
>>
>>>
>>> The regression result is sorted by the metric will-it-scale.per_thread_=
ops.
>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patc=
h series)
>>> Commit id:
>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>> Benchmark suite: will-it-scale
>>> Download link:
>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>> Metrics:
>>>     will-it-scale.per_process_ops=3Dprocesses/nr_cpu
>>>     will-it-scale.per_thread_ops=3Dthreads/nr_cpu
>>> test box: lkp-skl-4sp1(nr_cpu=3D192,memory=3D768G)
>>> THP: enable / disable
>>> nr_task: 100%
>>>
>>> 1. Regressions:
>>> a) THP enabled:
>>> testcase                        base            change          head   =
    metric
>>> page_fault3/ enable THP         10092           -17.5%          8323   =
    will-it-scale.per_thread_ops
>>> page_fault2/ enable THP          8300           -17.2%          6869   =
    will-it-scale.per_thread_ops
>>> brk1/ enable THP                  957.67         -7.6%           885   =
    will-it-scale.per_thread_ops
>>> page_fault3/ enable THP        172821            -5.3%        163692   =
    will-it-scale.per_process_ops
>>> signal1/ enable THP              9125            -3.2%          8834   =
    will-it-scale.per_process_ops
>>>
>>> b) THP disabled:
>>> testcase                        base            change          head   =
    metric
>>> page_fault3/ disable THP        10107           -19.1%          8180   =
    will-it-scale.per_thread_ops
>>> page_fault2/ disable THP         8432           -17.8%          6931   =
    will-it-scale.per_thread_ops
>>> context_switch1/ disable THP   215389            -6.8%        200776   =
    will-it-scale.per_thread_ops
>>> brk1/ disable THP                 939.67         -6.6%           877.33=
    will-it-scale.per_thread_ops
>>> page_fault3/ disable THP       173145            -4.7%        165064   =
    will-it-scale.per_process_ops
>>> signal1/ disable THP             9162            -3.9%          8802   =
    will-it-scale.per_process_ops
>>>
>>> 2. Improvements:
>>> a) THP enabled:
>>> testcase                        base            change          head   =
    metric
>>> malloc1/ enable THP               66.33        +469.8%           383.67=
    will-it-scale.per_thread_ops
>>> writeseek3/ enable THP          2531             +4.5%          2646   =
    will-it-scale.per_thread_ops
>>> signal1/ enable THP              989.33          +2.8%          1016   =
    will-it-scale.per_thread_ops
>>>
>>> b) THP disabled:
>>> testcase                        base            change          head   =
    metric
>>> malloc1/ disable THP              90.33        +417.3%           467.33=
    will-it-scale.per_thread_ops
>>> read2/ disable THP             58934            +39.2%         82060   =
    will-it-scale.per_thread_ops
>>> page_fault1/ disable THP        8607            +36.4%         11736   =
    will-it-scale.per_thread_ops
>>> read1/ disable THP            314063            +12.7%        353934   =
    will-it-scale.per_thread_ops
>>> writeseek3/ disable THP         2452            +12.5%          2759   =
    will-it-scale.per_thread_ops
>>> signal1/ disable THP             971.33          +5.5%          1024   =
    will-it-scale.per_thread_ops
>>>
>>> Notes: for above values in column "change", the higher value means that=
 the related testcase result
>>> on head commit is better than that on base commit for this benchmark.
>>>
>>>
>>> Best regards
>>> Haiyan Song
>>>
>>> ________________________________________
>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of =
Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Thursday, May 17, 2018 7:06 PM
>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org;=
 kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz;=
 Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.c=
om; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas =
Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; serg=
ey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, =
Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minch=
an Kim; Punit Agrawal; vinayak menon; Yang Shi
>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.=
ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.c=
om; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>
>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to han=
dle
>>> page fault without holding the mm semaphore [1].
>>>
>>> The idea is to try to handle user space page faults without holding the
>>> mmap_sem. This should allow better concurrency for massively threaded
>>> process since the page fault handler will not wait for other threads me=
mory
>>> layout change to be done, assuming that this change is done in another =
part
>>> of the process's memory space. This type page fault is named speculativ=
e
>>> page fault. If the speculative page fault fails because of a concurrenc=
y is
>>> detected or because underlying PMD or PTE tables are not yet allocating=
, it
>>> is failing its processing and a classic page fault is then tried.
>>>
>>> The speculative page fault (SPF) has to look for the VMA matching the f=
ault
>>> address without holding the mmap_sem, this is done by introducing a rwl=
ock
>>> which protects the access to the mm_rb tree. Previously this was done u=
sing
>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>> freeing operation which was hitting the performance by 20% as reported =
by
>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>> limiting the locking contention to these operations which are expected =
to
>>> be in a O(log n) order. In addition to ensure that the VMA is not freed=
 in
>>> our back a reference count is added and 2 services (get_vma() and
>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>> fetched from the RB tree using get_vma(), it must be later freed using
>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>> benchmark anymore.
>>>
>>> The VMA's attributes checked during the speculative page fault processi=
ng
>>> have to be protected against parallel changes. This is done by using a =
per
>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>> handler to fast check for parallel changes in progress and to abort the
>>> speculative page fault in that case.
>>>
>>> Once the VMA has been found, the speculative page fault handler would c=
heck
>>> for the VMA's attributes to verify that the page fault has to be handle=
d
>>> correctly or not. Thus, the VMA is protected through a sequence lock wh=
ich
>>> allows fast detection of concurrent VMA changes. If such a change is
>>> detected, the speculative page fault is aborted and a *classic* page fa=
ult
>>> is tried.  VMA sequence lockings are added when VMA attributes which ar=
e
>>> checked during the page fault are modified.
>>>
>>> When the PTE is fetched, the VMA is checked to see if it has been chang=
ed,
>>> so once the page table is locked, the VMA is valid, so any other change=
s
>>> leading to touching this PTE will need to lock the page table, so no
>>> parallel change is possible at this time.
>>>
>>> The locking of the PTE is done with interrupts disabled, this allows
>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then=
 is
>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd =
is
>>> valid at the time the PTE is locked, we have the guarantee that the
>>> collapsing operation will have to wait on the PTE lock to move forward.
>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>> different from the one recorded at the beginning of the SPF operation, =
the
>>> classic page fault handler will be called to handle the operation while
>>> holding the mmap_sem. As the PTE lock is done with the interrupts disab=
led,
>>> the lock is done using spin_trylock() to avoid dead lock when handling =
a
>>> page fault while a TLB invalidate is requested by another CPU holding t=
he
>>> PTE.
>>>
>>> In pseudo code, this could be seen as:
>>>     speculative_page_fault()
>>>     {
>>>             vma =3D get_vma()
>>>             check vma sequence count
>>>             check vma's support
>>>             disable interrupt
>>>                   check pgd,p4d,...,pte
>>>                   save pmd and pte in vmf
>>>                   save vma sequence counter in vmf
>>>             enable interrupt
>>>             check vma sequence count
>>>             handle_pte_fault(vma)
>>>                     ..
>>>                     page =3D alloc_page()
>>>                     pte_map_lock()
>>>                             disable interrupt
>>>                                     abort if sequence counter has chang=
ed
>>>                                     abort if pmd or pte has changed
>>>                                     pte map and lock
>>>                             enable interrupt
>>>                     if abort
>>>                        free page
>>>                        abort
>>>                     ...
>>>     }
>>>
>>>     arch_fault_handler()
>>>     {
>>>             if (speculative_page_fault(&vma))
>>>                goto done
>>>     again:
>>>             lock(mmap_sem)
>>>             vma =3D find_vma();
>>>             handle_pte_fault(vma);
>>>             if retry
>>>                unlock(mmap_sem)
>>>                goto again;
>>>     done:
>>>             handle fault error
>>>     }
>>>
>>> Support for THP is not done because when checking for the PMD, we can b=
e
>>> confused by an in progress collapsing operation done by khugepaged. The
>>> issue is that pmd_none() could be true either if the PMD is not already
>>> populated or if the underlying PTE are in the way to be collapsed. So w=
e
>>> cannot safely allocate a PMD if pmd_none() is true.
>>>
>>> This series add a new software performance event named 'speculative-fau=
lts'
>>> or 'spf'. It counts the number of successful page fault event handled
>>> speculatively. When recording 'faults,spf' events, the faults one is
>>> counting the total number of page fault events while 'spf' is only coun=
ting
>>> the part of the faults processed speculatively.
>>>
>>> There are some trace events introduced by this series. They allow
>>> identifying why the page faults were not processed speculatively. This
>>> doesn't take in account the faults generated by a monothreaded process
>>> which directly processed while holding the mmap_sem. This trace events =
are
>>> grouped in a system named 'pagefault', they are:
>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>    back.
>>>
>>> To record all the related events, the easier is to run perf with the
>>> following arguments :
>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>
>>> There is also a dedicated vmstat counter showing the number of successf=
ul
>>> page fault handled speculatively. I can be seen this way:
>>> $ grep speculative_pgfault /proc/vmstat
>>>
>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functi=
onal
>>> on x86, PowerPC and arm64.
>>>
>>> ---------------------
>>> Real Workload results
>>>
>>> As mentioned in previous email, we did non official runs using a "popul=
ar
>>> in memory multithreaded database product" on 176 cores SMT8 Power syste=
m
>>> which showed a 30% improvements in the number of transaction processed =
per
>>> second. This run has been done on the v6 series, but changes introduced=
 in
>>> this new version should not impact the performance boost seen.
>>>
>>> Here are the perf data captured during 2 of these runs on top of the v8
>>> series:
>>>                 vanilla         spf
>>> faults          89.418          101.364         +13%
>>> spf                n/a           97.989
>>>
>>> With the SPF kernel, most of the page fault were processed in a specula=
tive
>>> way.
>>>
>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and g=
ave
>>> it a try on an android device. He reported that the application launch =
time
>>> was improved in average by 6%, and for large applications (~100 threads=
) by
>>> 20%.
>>>
>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>> MSM845 (8 cores) with 6GB (the less is better):
>>>
>>> Application                             4.9     4.9+spf delta
>>> com.tencent.mm                          416     389     -7%
>>> com.eg.android.AlipayGphone             1135    986     -13%
>>> com.tencent.mtt                         455     454     0%
>>> com.qqgame.hlddz                        1497    1409    -6%
>>> com.autonavi.minimap                    711     701     -1%
>>> com.tencent.tmgp.sgame                  788     748     -5%
>>> com.immomo.momo                         501     487     -3%
>>> com.tencent.peng                        2145    2112    -2%
>>> com.smile.gifmaker                      491     461     -6%
>>> com.baidu.BaiduMap                      479     366     -23%
>>> com.taobao.taobao                       1341    1198    -11%
>>> com.baidu.searchbox                     333     314     -6%
>>> com.tencent.mobileqq                    394     384     -3%
>>> com.sina.weibo                          907     906     0%
>>> com.youku.phone                         816     731     -11%
>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>> com.UCMobile                            415     411     -1%
>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>> com.tencent.qqmusic                     336     329     -2%
>>> com.sankuai.meituan                     1661    1302    -22%
>>> com.netease.cloudmusic                  1193    1200    1%
>>> air.tv.douyu.android                    4257    4152    -2%
>>>
>>> ------------------
>>> Benchmarks results
>>>
>>> Base kernel is v4.17.0-rc4-mm1
>>> SPF is BASE + this series
>>>
>>> Kernbench:
>>> ----------
>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>> kernel (kernel is build 5 times):
>>>
>>> Average Half load -j 8
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>
>>> Average Optimal load -j 16
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>
>>>
>>> During a run on the SPF, perf events were captured:
>>>  Performance counter stats for '../kernbench -M':
>>>          526743764      faults
>>>                210      spf
>>>                  3      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>               2278      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> Very few speculative page faults were recorded as most of the processes
>>> involved are monothreaded (sounds that on this architecture some thread=
s
>>> were created during the kernel build processing).
>>>
>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>
>>> Average Half load -j 40
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>
>>> Average Optimal load -j 80
>>>                  Run    (std deviation)
>>>                  BASE                   SPF
>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>
>>> During a run on the SPF, perf events were captured:
>>>  Performance counter stats for '../kernbench -M':
>>>          116730856      faults
>>>                  0      spf
>>>                  3      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                476      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> Most of the processes involved are monothreaded so SPF is not activated=
 but
>>> there is no impact on the performance.
>>>
>>> Ebizzy:
>>> -------
>>> The test is counting the number of records per second it can manage, th=
e
>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>> consistent result I repeated the test 100 times and measure the average
>>> result. The number is the record processes per second, the higher is th=
e
>>> best.
>>>
>>>                 BASE            SPF             delta
>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>
>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>  Performance counter stats for './ebizzy -mTt 16':
>>>            1706379      faults
>>>            1674599      spf
>>>              30588      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                363      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> And the ones captured during a run on a 80 CPUs Power node:
>>>  Performance counter stats for './ebizzy -mTt 80':
>>>            1874773      faults
>>>            1461153      spf
>>>             413293      pagefault:spf_vma_changed
>>>                  0      pagefault:spf_vma_noanon
>>>                200      pagefault:spf_vma_notsup
>>>                  0      pagefault:spf_vma_access
>>>                  0      pagefault:spf_pmd_changed
>>>
>>> In ebizzy's case most of the page fault were handled in a speculative w=
ay,
>>> leading the ebizzy performance boost.
>>>
>>> ------------------
>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahend=
ran
>>>    and Minchan Kim, hopefully.
>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>    __do_page_fault().
>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>    instead
>>>    of aborting the speculative page fault handling. Dropping the now
>>> useless
>>>    trace event pagefault:spf_pte_lock.
>>>  - No more try to reuse the fetched VMA during the speculative page fau=
lt
>>>    handling when retrying is needed. This adds a lot of complexity and
>>>    additional tests done didn't show a significant performance improvem=
ent.
>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>
>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-=
speculative-page-faults-tt965642.html#none
>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>
>>>
>>> Laurent Dufour (20):
>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>   mm: make pte_unmap_same compatible with SPF
>>>   mm: introduce INIT_VMA()
>>>   mm: protect VMA modifications using VMA sequence count
>>>   mm: protect mremap() against SPF hanlder
>>>   mm: protect SPF handler against anon_vma changes
>>>   mm: cache some VMA fields in the vm_fault structure
>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>   mm: introduce __vm_normal_page()
>>>   mm: introduce __page_add_new_anon_rmap()
>>>   mm: protect mm_rb tree with a rwlock
>>>   mm: adding speculative page fault failure trace events
>>>   perf: add a speculative page fault sw event
>>>   perf tools: add support for the SPF perf event
>>>   mm: add speculative page fault vmstats
>>>   powerpc/mm: add speculative page fault
>>>
>>> Mahendran Ganesh (2):
>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>   arm64/mm: add speculative page fault
>>>
>>> Peter Zijlstra (4):
>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>   mm: VMA sequence count
>>>   mm: provide speculative fault infrastructure
>>>   x86/mm: add speculative pagefault handling
>>>
>>>  arch/arm64/Kconfig                    |   1 +
>>>  arch/arm64/mm/fault.c                 |  12 +
>>>  arch/powerpc/Kconfig                  |   1 +
>>>  arch/powerpc/mm/fault.c               |  16 +
>>>  arch/x86/Kconfig                      |   1 +
>>>  arch/x86/mm/fault.c                   |  27 +-
>>>  fs/exec.c                             |   2 +-
>>>  fs/proc/task_mmu.c                    |   5 +-
>>>  fs/userfaultfd.c                      |  17 +-
>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>  include/linux/migrate.h               |   4 +-
>>>  include/linux/mm.h                    | 136 +++++++-
>>>  include/linux/mm_types.h              |   7 +
>>>  include/linux/pagemap.h               |   4 +-
>>>  include/linux/rmap.h                  |  12 +-
>>>  include/linux/swap.h                  |  10 +-
>>>  include/linux/vm_event_item.h         |   3 +
>>>  include/trace/events/pagefault.h      |  80 +++++
>>>  include/uapi/linux/perf_event.h       |   1 +
>>>  kernel/fork.c                         |   5 +-
>>>  mm/Kconfig                            |  22 ++
>>>  mm/huge_memory.c                      |   6 +-
>>>  mm/hugetlb.c                          |   2 +
>>>  mm/init-mm.c                          |   3 +
>>>  mm/internal.h                         |  20 ++
>>>  mm/khugepaged.c                       |   5 +
>>>  mm/madvise.c                          |   6 +-
>>>  mm/memory.c                           | 612 ++++++++++++++++++++++++++=
+++-----
>>>  mm/mempolicy.c                        |  51 ++-
>>>  mm/migrate.c                          |   6 +-
>>>  mm/mlock.c                            |  13 +-
>>>  mm/mmap.c                             | 229 ++++++++++---
>>>  mm/mprotect.c                         |   4 +-
>>>  mm/mremap.c                           |  13 +
>>>  mm/nommu.c                            |   2 +-
>>>  mm/rmap.c                             |   5 +-
>>>  mm/swap.c                             |   6 +-
>>>  mm/swap_state.c                       |   8 +-
>>>  mm/vmstat.c                           |   5 +-
>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>  tools/perf/util/evsel.c               |   1 +
>>>  tools/perf/util/parse-events.c        |   4 +
>>>  tools/perf/util/parse-events.l        |   1 +
>>>  tools/perf/util/python.c              |   1 +
>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>  create mode 100644 include/trace/events/pagefault.h
>>>
>>> --
>>> 2.7.4
>>>
>>>
>>
>

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Hi Haiyan,

I don't have access to the same hardware you ran the test on, but I give a try
to those test on a Power8 system (2 sockets, 5 cores/s, 8 threads/c, 80 CPUs 32G).
I run each will-it-scale test 10 times and compute the average.

test THP enabled		4.17.0-rc4-mm1	spf		delta
page_fault3_threads		2697.7		2683.5		-0.53%
page_fault2_threads		170660.6	169574.1	-0.64%
context_switch1_threads		6915269.2	6877507.3	-0.55%
context_switch1_processes	6478076.2	6529493.5	0.79%
brk1				243391.2	238527.5	-2.00%

Test were launched with the arguments '-t 80 -s 5', only the average report is
taken in account. Note that page size is 64K by default on ppc64.

It would be nice if you could capture some perf data to figure out why the
page_fault2/3 are showing such a performance regression.

Thanks,
Laurent.

On 11/06/2018 09:49, Song, HaiyanX wrote:
> Hi Laurent,
> 
> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
> V9 patch serials.
> 
> The regression result is sorted by the metric will-it-scale.per_thread_ops.
> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
> commit id:
>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
> Benchmark: will-it-scale
> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
> 
> Metrics:
>   will-it-scale.per_process_ops=processes/nr_cpu
>   will-it-scale.per_thread_ops=threads/nr_cpu
>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
> THP: enable / disable
> nr_task:100%
> 
> 1. Regressions:
> 
> a). Enable THP
> testcase                          base           change      head           metric
> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
> 
> b). Disable THP
> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
> 
> Notes: for the above  values of test result, the higher is better.
> 
> 2. Improvement: not found improvement based on the selected test cases.
> 
> 
> Best regards
> Haiyan Song
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Monday, May 28, 2018 4:54 PM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
> 
> On 28/05/2018 10:22, Haiyan Song wrote:
>> Hi Laurent,
>>
>> Yes, these tests are done on V9 patch.
> 
> Do you plan to give this V11 a run ?
> 
>>
>>
>> Best regards,
>> Haiyan Song
>>
>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>
>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>> tested on Intel 4s Skylake platform.
>>>
>>> Hi,
>>>
>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>> series" while responding to the v11 header series...
>>> Were these tests done on v9 or v11 ?
>>>
>>> Cheers,
>>> Laurent.
>>>
>>>>
>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>> Commit id:
>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>> Benchmark suite: will-it-scale
>>>> Download link:
>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>> Metrics:
>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>> THP: enable / disable
>>>> nr_task: 100%
>>>>
>>>> 1. Regressions:
>>>> a) THP enabled:
>>>> testcase                        base            change          head       metric
>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>
>>>> b) THP disabled:
>>>> testcase                        base            change          head       metric
>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>
>>>> 2. Improvements:
>>>> a) THP enabled:
>>>> testcase                        base            change          head       metric
>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>
>>>> b) THP disabled:
>>>> testcase                        base            change          head       metric
>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>
>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>> on head commit is better than that on base commit for this benchmark.
>>>>
>>>>
>>>> Best regards
>>>> Haiyan Song
>>>>
>>>> ________________________________________
>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>
>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>> page fault without holding the mm semaphore [1].
>>>>
>>>> The idea is to try to handle user space page faults without holding the
>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>> process since the page fault handler will not wait for other threads memory
>>>> layout change to be done, assuming that this change is done in another part
>>>> of the process's memory space. This type page fault is named speculative
>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>> is failing its processing and a classic page fault is then tried.
>>>>
>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>> freeing operation which was hitting the performance by 20% as reported by
>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>> limiting the locking contention to these operations which are expected to
>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>> our back a reference count is added and 2 services (get_vma() and
>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>> benchmark anymore.
>>>>
>>>> The VMA's attributes checked during the speculative page fault processing
>>>> have to be protected against parallel changes. This is done by using a per
>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>> handler to fast check for parallel changes in progress and to abort the
>>>> speculative page fault in that case.
>>>>
>>>> Once the VMA has been found, the speculative page fault handler would check
>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>> checked during the page fault are modified.
>>>>
>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>> leading to touching this PTE will need to lock the page table, so no
>>>> parallel change is possible at this time.
>>>>
>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>> different from the one recorded at the beginning of the SPF operation, the
>>>> classic page fault handler will be called to handle the operation while
>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>> PTE.
>>>>
>>>> In pseudo code, this could be seen as:
>>>>     speculative_page_fault()
>>>>     {
>>>>             vma = get_vma()
>>>>             check vma sequence count
>>>>             check vma's support
>>>>             disable interrupt
>>>>                   check pgd,p4d,...,pte
>>>>                   save pmd and pte in vmf
>>>>                   save vma sequence counter in vmf
>>>>             enable interrupt
>>>>             check vma sequence count
>>>>             handle_pte_fault(vma)
>>>>                     ..
>>>>                     page = alloc_page()
>>>>                     pte_map_lock()
>>>>                             disable interrupt
>>>>                                     abort if sequence counter has changed
>>>>                                     abort if pmd or pte has changed
>>>>                                     pte map and lock
>>>>                             enable interrupt
>>>>                     if abort
>>>>                        free page
>>>>                        abort
>>>>                     ...
>>>>     }
>>>>
>>>>     arch_fault_handler()
>>>>     {
>>>>             if (speculative_page_fault(&vma))
>>>>                goto done
>>>>     again:
>>>>             lock(mmap_sem)
>>>>             vma = find_vma();
>>>>             handle_pte_fault(vma);
>>>>             if retry
>>>>                unlock(mmap_sem)
>>>>                goto again;
>>>>     done:
>>>>             handle fault error
>>>>     }
>>>>
>>>> Support for THP is not done because when checking for the PMD, we can be
>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>
>>>> This series add a new software performance event named 'speculative-faults'
>>>> or 'spf'. It counts the number of successful page fault event handled
>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>> counting the total number of page fault events while 'spf' is only counting
>>>> the part of the faults processed speculatively.
>>>>
>>>> There are some trace events introduced by this series. They allow
>>>> identifying why the page faults were not processed speculatively. This
>>>> doesn't take in account the faults generated by a monothreaded process
>>>> which directly processed while holding the mmap_sem. This trace events are
>>>> grouped in a system named 'pagefault', they are:
>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>    back.
>>>>
>>>> To record all the related events, the easier is to run perf with the
>>>> following arguments :
>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>
>>>> There is also a dedicated vmstat counter showing the number of successful
>>>> page fault handled speculatively. I can be seen this way:
>>>> $ grep speculative_pgfault /proc/vmstat
>>>>
>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>> on x86, PowerPC and arm64.
>>>>
>>>> ---------------------
>>>> Real Workload results
>>>>
>>>> As mentioned in previous email, we did non official runs using a "popular
>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>> which showed a 30% improvements in the number of transaction processed per
>>>> second. This run has been done on the v6 series, but changes introduced in
>>>> this new version should not impact the performance boost seen.
>>>>
>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>> series:
>>>>                 vanilla         spf
>>>> faults          89.418          101.364         +13%
>>>> spf                n/a           97.989
>>>>
>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>> way.
>>>>
>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>> it a try on an android device. He reported that the application launch time
>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>> 20%.
>>>>
>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>
>>>> Application                             4.9     4.9+spf delta
>>>> com.tencent.mm                          416     389     -7%
>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>> com.tencent.mtt                         455     454     0%
>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>> com.autonavi.minimap                    711     701     -1%
>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>> com.immomo.momo                         501     487     -3%
>>>> com.tencent.peng                        2145    2112    -2%
>>>> com.smile.gifmaker                      491     461     -6%
>>>> com.baidu.BaiduMap                      479     366     -23%
>>>> com.taobao.taobao                       1341    1198    -11%
>>>> com.baidu.searchbox                     333     314     -6%
>>>> com.tencent.mobileqq                    394     384     -3%
>>>> com.sina.weibo                          907     906     0%
>>>> com.youku.phone                         816     731     -11%
>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>> com.UCMobile                            415     411     -1%
>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>> com.tencent.qqmusic                     336     329     -2%
>>>> com.sankuai.meituan                     1661    1302    -22%
>>>> com.netease.cloudmusic                  1193    1200    1%
>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>
>>>> ------------------
>>>> Benchmarks results
>>>>
>>>> Base kernel is v4.17.0-rc4-mm1
>>>> SPF is BASE + this series
>>>>
>>>> Kernbench:
>>>> ----------
>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>> kernel (kernel is build 5 times):
>>>>
>>>> Average Half load -j 8
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>
>>>> Average Optimal load -j 16
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>
>>>>
>>>> During a run on the SPF, perf events were captured:
>>>>  Performance counter stats for '../kernbench -M':
>>>>          526743764      faults
>>>>                210      spf
>>>>                  3      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>               2278      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> Very few speculative page faults were recorded as most of the processes
>>>> involved are monothreaded (sounds that on this architecture some threads
>>>> were created during the kernel build processing).
>>>>
>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>
>>>> Average Half load -j 40
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>
>>>> Average Optimal load -j 80
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>
>>>> During a run on the SPF, perf events were captured:
>>>>  Performance counter stats for '../kernbench -M':
>>>>          116730856      faults
>>>>                  0      spf
>>>>                  3      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>                476      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>> there is no impact on the performance.
>>>>
>>>> Ebizzy:
>>>> -------
>>>> The test is counting the number of records per second it can manage, the
>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>> consistent result I repeated the test 100 times and measure the average
>>>> result. The number is the record processes per second, the higher is the
>>>> best.
>>>>
>>>>                 BASE            SPF             delta
>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>
>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>            1706379      faults
>>>>            1674599      spf
>>>>              30588      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>                363      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>            1874773      faults
>>>>            1461153      spf
>>>>             413293      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>                200      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>> leading the ebizzy performance boost.
>>>>
>>>> ------------------
>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>    and Minchan Kim, hopefully.
>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>    __do_page_fault().
>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>    instead
>>>>    of aborting the speculative page fault handling. Dropping the now
>>>> useless
>>>>    trace event pagefault:spf_pte_lock.
>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>    additional tests done didn't show a significant performance improvement.
>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>
>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>
>>>>
>>>> Laurent Dufour (20):
>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>   mm: make pte_unmap_same compatible with SPF
>>>>   mm: introduce INIT_VMA()
>>>>   mm: protect VMA modifications using VMA sequence count
>>>>   mm: protect mremap() against SPF hanlder
>>>>   mm: protect SPF handler against anon_vma changes
>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>   mm: introduce __vm_normal_page()
>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>   mm: protect mm_rb tree with a rwlock
>>>>   mm: adding speculative page fault failure trace events
>>>>   perf: add a speculative page fault sw event
>>>>   perf tools: add support for the SPF perf event
>>>>   mm: add speculative page fault vmstats
>>>>   powerpc/mm: add speculative page fault
>>>>
>>>> Mahendran Ganesh (2):
>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>   arm64/mm: add speculative page fault
>>>>
>>>> Peter Zijlstra (4):
>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>   mm: VMA sequence count
>>>>   mm: provide speculative fault infrastructure
>>>>   x86/mm: add speculative pagefault handling
>>>>
>>>>  arch/arm64/Kconfig                    |   1 +
>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>  arch/x86/Kconfig                      |   1 +
>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>  fs/exec.c                             |   2 +-
>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>  fs/userfaultfd.c                      |  17 +-
>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>  include/linux/migrate.h               |   4 +-
>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>  include/linux/mm_types.h              |   7 +
>>>>  include/linux/pagemap.h               |   4 +-
>>>>  include/linux/rmap.h                  |  12 +-
>>>>  include/linux/swap.h                  |  10 +-
>>>>  include/linux/vm_event_item.h         |   3 +
>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>  kernel/fork.c                         |   5 +-
>>>>  mm/Kconfig                            |  22 ++
>>>>  mm/huge_memory.c                      |   6 +-
>>>>  mm/hugetlb.c                          |   2 +
>>>>  mm/init-mm.c                          |   3 +
>>>>  mm/internal.h                         |  20 ++
>>>>  mm/khugepaged.c                       |   5 +
>>>>  mm/madvise.c                          |   6 +-
>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>  mm/mempolicy.c                        |  51 ++-
>>>>  mm/migrate.c                          |   6 +-
>>>>  mm/mlock.c                            |  13 +-
>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>  mm/mprotect.c                         |   4 +-
>>>>  mm/mremap.c                           |  13 +
>>>>  mm/nommu.c                            |   2 +-
>>>>  mm/rmap.c                             |   5 +-
>>>>  mm/swap.c                             |   6 +-
>>>>  mm/swap_state.c                       |   8 +-
>>>>  mm/vmstat.c                           |   5 +-
>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>  tools/perf/util/evsel.c               |   1 +
>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>  tools/perf/util/python.c              |   1 +
>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>
>>>> --
>>>> 2.7.4
>>>>
>>>>
>>>
>>
> 

Re: [PATCH v11 00/26] Speculative page faults

[Haiyan Song]

[-- Attachment #1: Type: text/plain, Size: 31691 bytes --]

On Mon, Jun 11, 2018 at 05:15:22PM +0200, Laurent Dufour wrote:

Hi Laurent,

For perf date tested on Intel 4s Skylake platform, here attached the compare result
between base and head commit in attachment, which include the perf-profile comparision information.

And also attached some perf-profile.json captured from test result for page_fault2 and page_fault3 for
checking the regression, thanks.


Best regards,
Haiyan Song



> Hi Haiyan,
> 
> I don't have access to the same hardware you ran the test on, but I give a try
> to those test on a Power8 system (2 sockets, 5 cores/s, 8 threads/c, 80 CPUs 32G).
> I run each will-it-scale test 10 times and compute the average.
> 
> test THP enabled		4.17.0-rc4-mm1	spf		delta
> page_fault3_threads		2697.7		2683.5		-0.53%
> page_fault2_threads		170660.6	169574.1	-0.64%
> context_switch1_threads		6915269.2	6877507.3	-0.55%
> context_switch1_processes	6478076.2	6529493.5	0.79%
> rk1				243391.2	238527.5	-2.00%
> 
> Test were launched with the arguments '-t 80 -s 5', only the average report is
> taken in account. Note that page size is 64K by default on ppc64.
> 
> It would be nice if you could capture some perf data to figure out why the
> page_fault2/3 are showing such a performance regression.
> 
> Thanks,
> Laurent.
> 
> On 11/06/2018 09:49, Song, HaiyanX wrote:
> > Hi Laurent,
> > 
> > Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
> > tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
> > V9 patch serials.
> > 
> > The regression result is sorted by the metric will-it-scale.per_thread_ops.
> > branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
> > commit id:
> >   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
> >   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
> > Benchmark: will-it-scale
> > Download link: https://github.com/antonblanchard/will-it-scale/tree/master
> > 
> > Metrics:
> >   will-it-scale.per_process_ops=processes/nr_cpu
> >   will-it-scale.per_thread_ops=threads/nr_cpu
> >   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
> > THP: enable / disable
> > nr_task:100%
> > 
> > 1. Regressions:
> > 
> > a). Enable THP
> > testcase                          base           change      head           metric
> > page_fault3/enable THP           10519          -20.5%       8368      will-it-scale.per_thread_ops
> > page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
> > brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
> > context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
> > context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
> > 
> > b). Disable THP
> > page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
> > page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
> > brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
> > context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
> > brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
> > page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
> > context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
> > 
> > Notes: for the above  values of test result, the higher is better.
> > 
> > 2. Improvement: not found improvement based on the selected test cases.
> > 
> > 
> > Best regards
> > Haiyan Song
> > ________________________________________
> > From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> > Sent: Monday, May 28, 2018 4:54 PM
> > To: Song, HaiyanX
> > Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> > Subject: Re: [PATCH v11 00/26] Speculative page faults
> > 
> > On 28/05/2018 10:22, Haiyan Song wrote:
> >> Hi Laurent,
> >>
> >> Yes, these tests are done on V9 patch.
> > 
> > Do you plan to give this V11 a run ?
> > 
> >>
> >>
> >> Best regards,
> >> Haiyan Song
> >>
> >> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
> >>> On 28/05/2018 07:23, Song, HaiyanX wrote:
> >>>>
> >>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
> >>>> tested on Intel 4s Skylake platform.
> >>>
> >>> Hi,
> >>>
> >>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
> >>> series" while responding to the v11 header series...
> >>> Were these tests done on v9 or v11 ?
> >>>
> >>> Cheers,
> >>> Laurent.
> >>>
> >>>>
> >>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
> >>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
> >>>> Commit id:
> >>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
> >>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
> >>>> Benchmark suite: will-it-scale
> >>>> Download link:
> >>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
> >>>> Metrics:
> >>>>     will-it-scale.per_process_ops=processes/nr_cpu
> >>>>     will-it-scale.per_thread_ops=threads/nr_cpu
> >>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
> >>>> THP: enable / disable
> >>>> nr_task: 100%
> >>>>
> >>>> 1. Regressions:
> >>>> a) THP enabled:
> >>>> testcase                        base            change          head       metric
> >>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
> >>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
> >>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
> >>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
> >>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
> >>>>
> >>>> b) THP disabled:
> >>>> testcase                        base            change          head       metric
> >>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
> >>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
> >>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
> >>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
> >>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
> >>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
> >>>>
> >>>> 2. Improvements:
> >>>> a) THP enabled:
> >>>> testcase                        base            change          head       metric
> >>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
> >>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
> >>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
> >>>>
> >>>> b) THP disabled:
> >>>> testcase                        base            change          head       metric
> >>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
> >>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
> >>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
> >>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
> >>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
> >>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
> >>>>
> >>>> Notes: for above values in column "change", the higher value means that the related testcase result
> >>>> on head commit is better than that on base commit for this benchmark.
> >>>>
> >>>>
> >>>> Best regards
> >>>> Haiyan Song
> >>>>
> >>>> ________________________________________
> >>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> >>>> Sent: Thursday, May 17, 2018 7:06 PM
> >>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
> >>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> >>>> Subject: [PATCH v11 00/26] Speculative page faults
> >>>>
> >>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
> >>>> page fault without holding the mm semaphore [1].
> >>>>
> >>>> The idea is to try to handle user space page faults without holding the
> >>>> mmap_sem. This should allow better concurrency for massively threaded
> >>>> process since the page fault handler will not wait for other threads memory
> >>>> layout change to be done, assuming that this change is done in another part
> >>>> of the process's memory space. This type page fault is named speculative
> >>>> page fault. If the speculative page fault fails because of a concurrency is
> >>>> detected or because underlying PMD or PTE tables are not yet allocating, it
> >>>> is failing its processing and a classic page fault is then tried.
> >>>>
> >>>> The speculative page fault (SPF) has to look for the VMA matching the fault
> >>>> address without holding the mmap_sem, this is done by introducing a rwlock
> >>>> which protects the access to the mm_rb tree. Previously this was done using
> >>>> SRCU but it was introducing a lot of scheduling to process the VMA's
> >>>> freeing operation which was hitting the performance by 20% as reported by
> >>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
> >>>> limiting the locking contention to these operations which are expected to
> >>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
> >>>> our back a reference count is added and 2 services (get_vma() and
> >>>> put_vma()) are introduced to handle the reference count. Once a VMA is
> >>>> fetched from the RB tree using get_vma(), it must be later freed using
> >>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
> >>>> benchmark anymore.
> >>>>
> >>>> The VMA's attributes checked during the speculative page fault processing
> >>>> have to be protected against parallel changes. This is done by using a per
> >>>> VMA sequence lock. This sequence lock allows the speculative page fault
> >>>> handler to fast check for parallel changes in progress and to abort the
> >>>> speculative page fault in that case.
> >>>>
> >>>> Once the VMA has been found, the speculative page fault handler would check
> >>>> for the VMA's attributes to verify that the page fault has to be handled
> >>>> correctly or not. Thus, the VMA is protected through a sequence lock which
> >>>> allows fast detection of concurrent VMA changes. If such a change is
> >>>> detected, the speculative page fault is aborted and a *classic* page fault
> >>>> is tried.  VMA sequence lockings are added when VMA attributes which are
> >>>> checked during the page fault are modified.
> >>>>
> >>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
> >>>> so once the page table is locked, the VMA is valid, so any other changes
> >>>> leading to touching this PTE will need to lock the page table, so no
> >>>> parallel change is possible at this time.
> >>>>
> >>>> The locking of the PTE is done with interrupts disabled, this allows
> >>>> checking for the PMD to ensure that there is not an ongoing collapsing
> >>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
> >>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
> >>>> valid at the time the PTE is locked, we have the guarantee that the
> >>>> collapsing operation will have to wait on the PTE lock to move forward.
> >>>> This allows the SPF handler to map the PTE safely. If the PMD value is
> >>>> different from the one recorded at the beginning of the SPF operation, the
> >>>> classic page fault handler will be called to handle the operation while
> >>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
> >>>> the lock is done using spin_trylock() to avoid dead lock when handling a
> >>>> page fault while a TLB invalidate is requested by another CPU holding the
> >>>> PTE.
> >>>>
> >>>> In pseudo code, this could be seen as:
> >>>>     speculative_page_fault()
> >>>>     {
> >>>>             vma = get_vma()
> >>>>             check vma sequence count
> >>>>             check vma's support
> >>>>             disable interrupt
> >>>>                   check pgd,p4d,...,pte
> >>>>                   save pmd and pte in vmf
> >>>>                   save vma sequence counter in vmf
> >>>>             enable interrupt
> >>>>             check vma sequence count
> >>>>             handle_pte_fault(vma)
> >>>>                     ..
> >>>>                     page = alloc_page()
> >>>>                     pte_map_lock()
> >>>>                             disable interrupt
> >>>>                                     abort if sequence counter has changed
> >>>>                                     abort if pmd or pte has changed
> >>>>                                     pte map and lock
> >>>>                             enable interrupt
> >>>>                     if abort
> >>>>                        free page
> >>>>                        abort
> >>>>                     ...
> >>>>     }
> >>>>
> >>>>     arch_fault_handler()
> >>>>     {
> >>>>             if (speculative_page_fault(&vma))
> >>>>                goto done
> >>>>     again:
> >>>>             lock(mmap_sem)
> >>>>             vma = find_vma();
> >>>>             handle_pte_fault(vma);
> >>>>             if retry
> >>>>                unlock(mmap_sem)
> >>>>                goto again;
> >>>>     done:
> >>>>             handle fault error
> >>>>     }
> >>>>
> >>>> Support for THP is not done because when checking for the PMD, we can be
> >>>> confused by an in progress collapsing operation done by khugepaged. The
> >>>> issue is that pmd_none() could be true either if the PMD is not already
> >>>> populated or if the underlying PTE are in the way to be collapsed. So we
> >>>> cannot safely allocate a PMD if pmd_none() is true.
> >>>>
> >>>> This series add a new software performance event named 'speculative-faults'
> >>>> or 'spf'. It counts the number of successful page fault event handled
> >>>> speculatively. When recording 'faults,spf' events, the faults one is
> >>>> counting the total number of page fault events while 'spf' is only counting
> >>>> the part of the faults processed speculatively.
> >>>>
> >>>> There are some trace events introduced by this series. They allow
> >>>> identifying why the page faults were not processed speculatively. This
> >>>> doesn't take in account the faults generated by a monothreaded process
> >>>> which directly processed while holding the mmap_sem. This trace events are
> >>>> grouped in a system named 'pagefault', they are:
> >>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
> >>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
> >>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
> >>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
> >>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
> >>>>    back.
> >>>>
> >>>> To record all the related events, the easier is to run perf with the
> >>>> following arguments :
> >>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
> >>>>
> >>>> There is also a dedicated vmstat counter showing the number of successful
> >>>> page fault handled speculatively. I can be seen this way:
> >>>> $ grep speculative_pgfault /proc/vmstat
> >>>>
> >>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
> >>>> on x86, PowerPC and arm64.
> >>>>
> >>>> ---------------------
> >>>> Real Workload results
> >>>>
> >>>> As mentioned in previous email, we did non official runs using a "popular
> >>>> in memory multithreaded database product" on 176 cores SMT8 Power system
> >>>> which showed a 30% improvements in the number of transaction processed per
> >>>> second. This run has been done on the v6 series, but changes introduced in
> >>>> this new version should not impact the performance boost seen.
> >>>>
> >>>> Here are the perf data captured during 2 of these runs on top of the v8
> >>>> series:
> >>>>                 vanilla         spf
> >>>> faults          89.418          101.364         +13%
> >>>> spf                n/a           97.989
> >>>>
> >>>> With the SPF kernel, most of the page fault were processed in a speculative
> >>>> way.
> >>>>
> >>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
> >>>> it a try on an android device. He reported that the application launch time
> >>>> was improved in average by 6%, and for large applications (~100 threads) by
> >>>> 20%.
> >>>>
> >>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
> >>>> MSM845 (8 cores) with 6GB (the less is better):
> >>>>
> >>>> Application                             4.9     4.9+spf delta
> >>>> com.tencent.mm                          416     389     -7%
> >>>> com.eg.android.AlipayGphone             1135    986     -13%
> >>>> com.tencent.mtt                         455     454     0%
> >>>> com.qqgame.hlddz                        1497    1409    -6%
> >>>> com.autonavi.minimap                    711     701     -1%
> >>>> com.tencent.tmgp.sgame                  788     748     -5%
> >>>> com.immomo.momo                         501     487     -3%
> >>>> com.tencent.peng                        2145    2112    -2%
> >>>> com.smile.gifmaker                      491     461     -6%
> >>>> com.baidu.BaiduMap                      479     366     -23%
> >>>> com.taobao.taobao                       1341    1198    -11%
> >>>> com.baidu.searchbox                     333     314     -6%
> >>>> com.tencent.mobileqq                    394     384     -3%
> >>>> com.sina.weibo                          907     906     0%
> >>>> com.youku.phone                         816     731     -11%
> >>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
> >>>> com.UCMobile                            415     411     -1%
> >>>> com.tencent.tmgp.ak                     1464    1431    -2%
> >>>> com.tencent.qqmusic                     336     329     -2%
> >>>> com.sankuai.meituan                     1661    1302    -22%
> >>>> com.netease.cloudmusic                  1193    1200    1%
> >>>> air.tv.douyu.android                    4257    4152    -2%
> >>>>
> >>>> ------------------
> >>>> Benchmarks results
> >>>>
> >>>> Base kernel is v4.17.0-rc4-mm1
> >>>> SPF is BASE + this series
> >>>>
> >>>> Kernbench:
> >>>> ----------
> >>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
> >>>> kernel (kernel is build 5 times):
> >>>>
> >>>> Average Half load -j 8
> >>>>                  Run    (std deviation)
> >>>>                  BASE                   SPF
> >>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
> >>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
> >>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
> >>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
> >>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
> >>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
> >>>>
> >>>> Average Optimal load -j 16
> >>>>                  Run    (std deviation)
> >>>>                  BASE                   SPF
> >>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
> >>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
> >>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
> >>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
> >>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
> >>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
> >>>>
> >>>>
> >>>> During a run on the SPF, perf events were captured:
> >>>>  Performance counter stats for '../kernbench -M':
> >>>>          526743764      faults
> >>>>                210      spf
> >>>>                  3      pagefault:spf_vma_changed
> >>>>                  0      pagefault:spf_vma_noanon
> >>>>               2278      pagefault:spf_vma_notsup
> >>>>                  0      pagefault:spf_vma_access
> >>>>                  0      pagefault:spf_pmd_changed
> >>>>
> >>>> Very few speculative page faults were recorded as most of the processes
> >>>> involved are monothreaded (sounds that on this architecture some threads
> >>>> were created during the kernel build processing).
> >>>>
> >>>> Here are the kerbench results on a 80 CPUs Power8 system:
> >>>>
> >>>> Average Half load -j 40
> >>>>                  Run    (std deviation)
> >>>>                  BASE                   SPF
> >>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
> >>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
> >>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
> >>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
> >>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
> >>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
> >>>>
> >>>> Average Optimal load -j 80
> >>>>                  Run    (std deviation)
> >>>>                  BASE                   SPF
> >>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
> >>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
> >>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
> >>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
> >>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
> >>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
> >>>>
> >>>> During a run on the SPF, perf events were captured:
> >>>>  Performance counter stats for '../kernbench -M':
> >>>>          116730856      faults
> >>>>                  0      spf
> >>>>                  3      pagefault:spf_vma_changed
> >>>>                  0      pagefault:spf_vma_noanon
> >>>>                476      pagefault:spf_vma_notsup
> >>>>                  0      pagefault:spf_vma_access
> >>>>                  0      pagefault:spf_pmd_changed
> >>>>
> >>>> Most of the processes involved are monothreaded so SPF is not activated but
> >>>> there is no impact on the performance.
> >>>>
> >>>> Ebizzy:
> >>>> -------
> >>>> The test is counting the number of records per second it can manage, the
> >>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
> >>>> consistent result I repeated the test 100 times and measure the average
> >>>> result. The number is the record processes per second, the higher is the
> >>>> best.
> >>>>
> >>>>                 BASE            SPF             delta
> >>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
> >>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
> >>>>
> >>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
> >>>>  Performance counter stats for './ebizzy -mTt 16':
> >>>>            1706379      faults
> >>>>            1674599      spf
> >>>>              30588      pagefault:spf_vma_changed
> >>>>                  0      pagefault:spf_vma_noanon
> >>>>                363      pagefault:spf_vma_notsup
> >>>>                  0      pagefault:spf_vma_access
> >>>>                  0      pagefault:spf_pmd_changed
> >>>>
> >>>> And the ones captured during a run on a 80 CPUs Power node:
> >>>>  Performance counter stats for './ebizzy -mTt 80':
> >>>>            1874773      faults
> >>>>            1461153      spf
> >>>>             413293      pagefault:spf_vma_changed
> >>>>                  0      pagefault:spf_vma_noanon
> >>>>                200      pagefault:spf_vma_notsup
> >>>>                  0      pagefault:spf_vma_access
> >>>>                  0      pagefault:spf_pmd_changed
> >>>>
> >>>> In ebizzy's case most of the page fault were handled in a speculative way,
> >>>> leading the ebizzy performance boost.
> >>>>
> >>>> ------------------
> >>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
> >>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
> >>>>    and Minchan Kim, hopefully.
> >>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
> >>>>    __do_page_fault().
> >>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
> >>>>    instead
> >>>>    of aborting the speculative page fault handling. Dropping the now
> >>>> useless
> >>>>    trace event pagefault:spf_pte_lock.
> >>>>  - No more try to reuse the fetched VMA during the speculative page fault
> >>>>    handling when retrying is needed. This adds a lot of complexity and
> >>>>    additional tests done didn't show a significant performance improvement.
> >>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
> >>>>
> >>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
> >>>> [2] https://patchwork.kernel.org/patch/9999687/
> >>>>
> >>>>
> >>>> Laurent Dufour (20):
> >>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
> >>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> >>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> >>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
> >>>>   mm: make pte_unmap_same compatible with SPF
> >>>>   mm: introduce INIT_VMA()
> >>>>   mm: protect VMA modifications using VMA sequence count
> >>>>   mm: protect mremap() against SPF hanlder
> >>>>   mm: protect SPF handler against anon_vma changes
> >>>>   mm: cache some VMA fields in the vm_fault structure
> >>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
> >>>>   mm: introduce __lru_cache_add_active_or_unevictable
> >>>>   mm: introduce __vm_normal_page()
> >>>>   mm: introduce __page_add_new_anon_rmap()
> >>>>   mm: protect mm_rb tree with a rwlock
> >>>>   mm: adding speculative page fault failure trace events
> >>>>   perf: add a speculative page fault sw event
> >>>>   perf tools: add support for the SPF perf event
> >>>>   mm: add speculative page fault vmstats
> >>>>   powerpc/mm: add speculative page fault
> >>>>
> >>>> Mahendran Ganesh (2):
> >>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> >>>>   arm64/mm: add speculative page fault
> >>>>
> >>>> Peter Zijlstra (4):
> >>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
> >>>>   mm: VMA sequence count
> >>>>   mm: provide speculative fault infrastructure
> >>>>   x86/mm: add speculative pagefault handling
> >>>>
> >>>>  arch/arm64/Kconfig                    |   1 +
> >>>>  arch/arm64/mm/fault.c                 |  12 +
> >>>>  arch/powerpc/Kconfig                  |   1 +
> >>>>  arch/powerpc/mm/fault.c               |  16 +
> >>>>  arch/x86/Kconfig                      |   1 +
> >>>>  arch/x86/mm/fault.c                   |  27 +-
> >>>>  fs/exec.c                             |   2 +-
> >>>>  fs/proc/task_mmu.c                    |   5 +-
> >>>>  fs/userfaultfd.c                      |  17 +-
> >>>>  include/linux/hugetlb_inline.h        |   2 +-
> >>>>  include/linux/migrate.h               |   4 +-
> >>>>  include/linux/mm.h                    | 136 +++++++-
> >>>>  include/linux/mm_types.h              |   7 +
> >>>>  include/linux/pagemap.h               |   4 +-
> >>>>  include/linux/rmap.h                  |  12 +-
> >>>>  include/linux/swap.h                  |  10 +-
> >>>>  include/linux/vm_event_item.h         |   3 +
> >>>>  include/trace/events/pagefault.h      |  80 +++++
> >>>>  include/uapi/linux/perf_event.h       |   1 +
> >>>>  kernel/fork.c                         |   5 +-
> >>>>  mm/Kconfig                            |  22 ++
> >>>>  mm/huge_memory.c                      |   6 +-
> >>>>  mm/hugetlb.c                          |   2 +
> >>>>  mm/init-mm.c                          |   3 +
> >>>>  mm/internal.h                         |  20 ++
> >>>>  mm/khugepaged.c                       |   5 +
> >>>>  mm/madvise.c                          |   6 +-
> >>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
> >>>>  mm/mempolicy.c                        |  51 ++-
> >>>>  mm/migrate.c                          |   6 +-
> >>>>  mm/mlock.c                            |  13 +-
> >>>>  mm/mmap.c                             | 229 ++++++++++---
> >>>>  mm/mprotect.c                         |   4 +-
> >>>>  mm/mremap.c                           |  13 +
> >>>>  mm/nommu.c                            |   2 +-
> >>>>  mm/rmap.c                             |   5 +-
> >>>>  mm/swap.c                             |   6 +-
> >>>>  mm/swap_state.c                       |   8 +-
> >>>>  mm/vmstat.c                           |   5 +-
> >>>>  tools/include/uapi/linux/perf_event.h |   1 +
> >>>>  tools/perf/util/evsel.c               |   1 +
> >>>>  tools/perf/util/parse-events.c        |   4 +
> >>>>  tools/perf/util/parse-events.l        |   1 +
> >>>>  tools/perf/util/python.c              |   1 +
> >>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
> >>>>  create mode 100644 include/trace/events/pagefault.h
> >>>>
> >>>> --
> >>>> 2.7.4
> >>>>
> >>>>
> >>>
> >>
> > 
> 

[-- Attachment #2: compare-result.txt --]
[-- Type: text/plain, Size: 185207 bytes --]

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/always/page_fault3/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
         44:3          -13%          43:3     perf-profile.calltrace.cycles-pp.error_entry
         22:3           -6%          22:3     perf-profile.calltrace.cycles-pp.sync_regs.error_entry
         44:3          -13%          44:3     perf-profile.children.cycles-pp.error_entry
         21:3           -7%          21:3     perf-profile.self.cycles-pp.error_entry
         %stddev     %change         %stddev
             \          |                \  
     10519 +-  3%     -20.5%       8368 +-  6%  will-it-scale.per_thread_ops
    118098           +11.2%     131287 +-  2%  will-it-scale.time.involuntary_context_switches
 6.084e+08 +-  3%     -20.4%  4.845e+08 +-  6%  will-it-scale.time.minor_page_faults
      7467            +5.0%       7841        will-it-scale.time.percent_of_cpu_this_job_got
     44922            +5.0%      47176        will-it-scale.time.system_time
   7126337 +-  3%     -15.4%    6025689 +-  6%  will-it-scale.time.voluntary_context_switches
  91905646            -1.3%   90673935        will-it-scale.workload
     27.15 +-  6%      -8.7%      24.80 +- 10%  boot-time.boot
   2516213 +-  6%      +8.3%    2726303        interrupts.CAL:Function_call_interrupts
    388.00 +-  9%     +60.2%     621.67 +- 20%  irq_exception_noise.softirq_nr
     11.28 +-  2%      -1.9        9.37 +-  4%  mpstat.cpu.idle%
     10065 +-140%    +243.4%      34559 +-  4%  numa-numastat.node0.other_node
     18739           -11.6%      16573 +-  3%  uptime.idle
     29406 +-  2%     -11.8%      25929 +-  5%  vmstat.system.cs
    329614 +-  8%     +17.0%     385618 +- 10%  meminfo.DirectMap4k
    237851           +21.2%     288160 +-  5%  meminfo.Inactive
    237615           +21.2%     287924 +-  5%  meminfo.Inactive(anon)
   7917847           -10.7%    7071860        softirqs.RCU
   4784181 +-  3%     -14.5%    4089039 +-  4%  softirqs.SCHED
  45666107 +-  7%     +12.9%   51535472 +-  3%  softirqs.TIMER
 2.617e+09 +-  2%     -13.9%  2.253e+09 +-  6%  cpuidle.C1E.time
   6688774 +-  2%     -12.8%    5835101 +-  5%  cpuidle.C1E.usage
 1.022e+10 +-  2%     -18.0%  8.376e+09 +-  3%  cpuidle.C6.time
  13440993 +-  2%     -16.3%   11243794 +-  4%  cpuidle.C6.usage
     54781 +- 16%     +37.5%      75347 +- 12%  numa-meminfo.node0.Inactive
     54705 +- 16%     +37.7%      75347 +- 12%  numa-meminfo.node0.Inactive(anon)
     52522           +35.0%      70886 +-  6%  numa-meminfo.node2.Inactive
     52443           +34.7%      70653 +-  6%  numa-meminfo.node2.Inactive(anon)
     31046 +-  6%     +30.3%      40457 +- 11%  numa-meminfo.node2.SReclaimable
     58563           +21.1%      70945 +-  6%  proc-vmstat.nr_inactive_anon
     58564           +21.1%      70947 +-  6%  proc-vmstat.nr_zone_inactive_anon
  69701118            -1.2%   68842151        proc-vmstat.pgalloc_normal
 2.765e+10            -1.3%  2.729e+10        proc-vmstat.pgfault
  69330418            -1.2%   68466824        proc-vmstat.pgfree
    118098           +11.2%     131287 +-  2%  time.involuntary_context_switches
 6.084e+08 +-  3%     -20.4%  4.845e+08 +-  6%  time.minor_page_faults
      7467            +5.0%       7841        time.percent_of_cpu_this_job_got
     44922            +5.0%      47176        time.system_time
   7126337 +-  3%     -15.4%    6025689 +-  6%  time.voluntary_context_switches
     13653 +- 16%     +33.5%      18225 +- 12%  numa-vmstat.node0.nr_inactive_anon
     13651 +- 16%     +33.5%      18224 +- 12%  numa-vmstat.node0.nr_zone_inactive_anon
     13069 +-  3%     +30.1%      17001 +-  4%  numa-vmstat.node2.nr_inactive_anon
    134.67 +- 42%     -49.5%      68.00 +- 31%  numa-vmstat.node2.nr_mlock
      7758 +-  6%     +30.4%      10112 +- 11%  numa-vmstat.node2.nr_slab_reclaimable
     13066 +-  3%     +30.1%      16998 +-  4%  numa-vmstat.node2.nr_zone_inactive_anon
      1039 +- 11%     -17.5%     857.33        slabinfo.Acpi-ParseExt.active_objs
      1039 +- 11%     -17.5%     857.33        slabinfo.Acpi-ParseExt.num_objs
      2566 +-  6%      -8.8%       2340 +-  5%  slabinfo.biovec-64.active_objs
      2566 +-  6%      -8.8%       2340 +-  5%  slabinfo.biovec-64.num_objs
    898.33 +-  3%      -9.5%     813.33 +-  3%  slabinfo.kmem_cache_node.active_objs
      1066 +-  2%      -8.0%     981.33 +-  3%  slabinfo.kmem_cache_node.num_objs
      1940            +2.3%       1984        turbostat.Avg_MHz
   6679037 +-  2%     -12.7%    5830270 +-  5%  turbostat.C1E
      2.25 +-  2%      -0.3        1.94 +-  6%  turbostat.C1E%
  13418115           -16.3%   11234510 +-  4%  turbostat.C6
      8.75 +-  2%      -1.6        7.18 +-  3%  turbostat.C6%
      5.99 +-  2%     -14.4%       5.13 +-  4%  turbostat.CPU%c1
      5.01 +-  3%     -20.1%       4.00 +-  4%  turbostat.CPU%c6
      1.77 +-  3%     -34.7%       1.15        turbostat.Pkg%pc2
 1.378e+13            +1.2%  1.394e+13        perf-stat.branch-instructions
      0.98            -0.0        0.94        perf-stat.branch-miss-rate%
 1.344e+11            -2.3%  1.313e+11        perf-stat.branch-misses
 1.076e+11            -1.8%  1.057e+11        perf-stat.cache-misses
 2.258e+11            -2.1%   2.21e+11        perf-stat.cache-references
  17788064 +-  2%     -11.9%   15674207 +-  6%  perf-stat.context-switches
 2.241e+14            +2.4%  2.294e+14        perf-stat.cpu-cycles
 1.929e+13            +2.2%  1.971e+13        perf-stat.dTLB-loads
      4.01            -0.2        3.83        perf-stat.dTLB-store-miss-rate%
 4.519e+11            -1.3%  4.461e+11        perf-stat.dTLB-store-misses
 1.082e+13            +3.6%  1.121e+13        perf-stat.dTLB-stores
  3.02e+10           +23.2%  3.721e+10 +-  3%  perf-stat.iTLB-load-misses
 2.721e+08 +-  8%      -8.8%  2.481e+08 +-  3%  perf-stat.iTLB-loads
 6.985e+13            +1.8%  7.111e+13        perf-stat.instructions
      2313           -17.2%       1914 +-  3%  perf-stat.instructions-per-iTLB-miss
 2.764e+10            -1.3%  2.729e+10        perf-stat.minor-faults
 1.421e+09 +-  2%     -16.4%  1.188e+09 +-  9%  perf-stat.node-load-misses
 1.538e+10            -9.3%  1.395e+10        perf-stat.node-loads
      9.75            +1.4       11.10        perf-stat.node-store-miss-rate%
 3.012e+09           +14.1%  3.437e+09        perf-stat.node-store-misses
 2.789e+10            -1.3%  2.753e+10        perf-stat.node-stores
 2.764e+10            -1.3%  2.729e+10        perf-stat.page-faults
    760059            +3.2%     784235        perf-stat.path-length
    193545 +- 25%     -57.8%      81757 +- 46%  sched_debug.cfs_rq:/.MIN_vruntime.avg
  26516863 +- 19%     -49.7%   13338070 +- 33%  sched_debug.cfs_rq:/.MIN_vruntime.max
   2202271 +- 21%     -53.2%    1029581 +- 38%  sched_debug.cfs_rq:/.MIN_vruntime.stddev
    193545 +- 25%     -57.8%      81757 +- 46%  sched_debug.cfs_rq:/.max_vruntime.avg
  26516863 +- 19%     -49.7%   13338070 +- 33%  sched_debug.cfs_rq:/.max_vruntime.max
   2202271 +- 21%     -53.2%    1029581 +- 38%  sched_debug.cfs_rq:/.max_vruntime.stddev
      0.32 +- 70%    +253.2%       1.14 +- 54%  sched_debug.cfs_rq:/.removed.load_avg.avg
      4.44 +- 70%    +120.7%       9.80 +- 27%  sched_debug.cfs_rq:/.removed.load_avg.stddev
     14.90 +- 70%    +251.0%      52.31 +- 53%  sched_debug.cfs_rq:/.removed.runnable_sum.avg
    205.71 +- 70%    +119.5%     451.60 +- 27%  sched_debug.cfs_rq:/.removed.runnable_sum.stddev
      0.16 +- 70%    +237.9%       0.54 +- 50%  sched_debug.cfs_rq:/.removed.util_avg.avg
      2.23 +- 70%    +114.2%       4.77 +- 24%  sched_debug.cfs_rq:/.removed.util_avg.stddev
    573.70 +-  5%      -9.7%     518.06 +-  6%  sched_debug.cfs_rq:/.util_avg.min
    114.87 +-  8%     +14.1%     131.04 +- 10%  sched_debug.cfs_rq:/.util_est_enqueued.avg
     64.42 +- 54%     -63.9%      23.27 +- 68%  sched_debug.cpu.cpu_load[1].max
      5.05 +- 48%     -55.2%       2.26 +- 51%  sched_debug.cpu.cpu_load[1].stddev
     57.58 +- 59%     -60.3%      22.88 +- 70%  sched_debug.cpu.cpu_load[2].max
     21019 +-  3%     -15.1%      17841 +-  6%  sched_debug.cpu.nr_switches.min
     20797 +-  3%     -15.0%      17670 +-  6%  sched_debug.cpu.sched_count.min
     10287 +-  3%     -15.1%       8736 +-  6%  sched_debug.cpu.sched_goidle.avg
     13693 +-  2%     -10.7%      12233 +-  5%  sched_debug.cpu.sched_goidle.max
      9976 +-  3%     -16.0%       8381 +-  7%  sched_debug.cpu.sched_goidle.min
      0.00 +- 26%     +98.9%       0.00 +- 28%  sched_debug.rt_rq:/.rt_time.min
      4230 +-141%    -100.0%       0.00        latency_stats.avg.trace_module_notify.notifier_call_chain.blocking_notifier_call_chain.do_init_module.load_module.__do_sys_finit_module.do_syscall_64.entry_SYSCALL_64_after_hwframe
     28498 +-141%    -100.0%       0.00        latency_stats.avg.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4065 +-138%     -92.2%     315.33 +- 91%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
      0.00       +3.6e+105%       3641 +-141%  latency_stats.avg.down.console_lock.console_device.tty_lookup_driver.tty_open.chrdev_open.do_dentry_open.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +2.5e+106%      25040 +-141%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +3.4e+106%      34015 +-141%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_get_acl.get_acl.posix_acl_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open
      0.00       +4.8e+106%      47686 +-141%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
      4230 +-141%    -100.0%       0.00        latency_stats.max.trace_module_notify.notifier_call_chain.blocking_notifier_call_chain.do_init_module.load_module.__do_sys_finit_module.do_syscall_64.entry_SYSCALL_64_after_hwframe
     28498 +-141%    -100.0%       0.00        latency_stats.max.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4065 +-138%     -92.2%     315.33 +- 91%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
      4254 +-134%     -88.0%     511.67 +- 90%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
     43093 +- 35%     +76.6%      76099 +-115%  latency_stats.max.blk_execute_rq.scsi_execute.ioctl_internal_command.scsi_set_medium_removal.cdrom_release.[cdrom].sr_block_release.[sr_mod].__blkdev_put.blkdev_close.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64
     24139 +- 70%    +228.5%      79285 +-105%  latency_stats.max.blk_execute_rq.scsi_execute.scsi_test_unit_ready.sr_check_events.[sr_mod].cdrom_check_events.[cdrom].sr_block_check_events.[sr_mod].disk_check_events.disk_clear_events.check_disk_change.sr_block_open.[sr_mod].__blkdev_get.blkdev_get
      0.00       +3.6e+105%       3641 +-141%  latency_stats.max.down.console_lock.console_device.tty_lookup_driver.tty_open.chrdev_open.do_dentry_open.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +2.5e+106%      25040 +-141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +3.4e+106%      34015 +-141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_get_acl.get_acl.posix_acl_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open
      0.00       +6.5e+106%      64518 +-141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
      4230 +-141%    -100.0%       0.00        latency_stats.sum.trace_module_notify.notifier_call_chain.blocking_notifier_call_chain.do_init_module.load_module.__do_sys_finit_module.do_syscall_64.entry_SYSCALL_64_after_hwframe
     28498 +-141%    -100.0%       0.00        latency_stats.sum.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4065 +-138%     -92.2%     315.33 +- 91%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
     57884 +-  9%     +47.3%      85264 +-118%  latency_stats.sum.blk_execute_rq.scsi_execute.ioctl_internal_command.scsi_set_medium_removal.cdrom_release.[cdrom].sr_block_release.[sr_mod].__blkdev_put.blkdev_close.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64
      0.00       +3.6e+105%       3641 +-141%  latency_stats.sum.down.console_lock.console_device.tty_lookup_driver.tty_open.chrdev_open.do_dentry_open.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +2.5e+106%      25040 +-141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +3.4e+106%      34015 +-141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_get_acl.get_acl.posix_acl_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open
      0.00       +9.5e+106%      95373 +-141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
     11.70           -11.7        0.00        perf-profile.calltrace.cycles-pp.__do_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
     11.52           -11.5        0.00        perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
     10.44           -10.4        0.00        perf-profile.calltrace.cycles-pp.shmem_getpage_gfp.shmem_fault.__do_fault.__handle_mm_fault.handle_mm_fault
      9.83            -9.8        0.00        perf-profile.calltrace.cycles-pp.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault.__handle_mm_fault
      9.55            -9.5        0.00        perf-profile.calltrace.cycles-pp.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      9.35            -9.3        0.00        perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      6.81            -6.8        0.00        perf-profile.calltrace.cycles-pp.page_add_file_rmap.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault
      7.71            -0.3        7.45        perf-profile.calltrace.cycles-pp.find_get_entry.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault
      0.59 +-  7%      -0.2        0.35 +- 70%  perf-profile.calltrace.cycles-pp.smp_apic_timer_interrupt.apic_timer_interrupt.__do_page_fault.do_page_fault.page_fault
      0.59 +-  7%      -0.2        0.35 +- 70%  perf-profile.calltrace.cycles-pp.apic_timer_interrupt.__do_page_fault.do_page_fault.page_fault
     10.41            -0.2       10.24        perf-profile.calltrace.cycles-pp.native_irq_return_iret
      7.68            -0.1        7.60        perf-profile.calltrace.cycles-pp.swapgs_restore_regs_and_return_to_usermode
      0.76            -0.1        0.70        perf-profile.calltrace.cycles-pp.down_read_trylock.__do_page_fault.do_page_fault.page_fault
      1.38            -0.0        1.34        perf-profile.calltrace.cycles-pp.do_page_fault
      1.05            -0.0        1.02        perf-profile.calltrace.cycles-pp.trace_graph_entry.do_page_fault
      0.92            +0.0        0.94        perf-profile.calltrace.cycles-pp.find_vma.__do_page_fault.do_page_fault.page_fault
      0.91            +0.0        0.93        perf-profile.calltrace.cycles-pp.vmacache_find.find_vma.__do_page_fault.do_page_fault.page_fault
      0.65            +0.0        0.67        perf-profile.calltrace.cycles-pp.set_page_dirty.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      0.62            +0.0        0.66        perf-profile.calltrace.cycles-pp.page_mapping.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault
      4.15            +0.1        4.27        perf-profile.calltrace.cycles-pp.page_remove_rmap.unmap_page_range.unmap_vmas.unmap_region.do_munmap
     10.17            +0.2       10.39        perf-profile.calltrace.cycles-pp.munmap
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.munmap
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.munmap
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.unmap_region.do_munmap.vm_munmap.__x64_sys_munmap.do_syscall_64
      9.54            +0.2        9.76        perf-profile.calltrace.cycles-pp.unmap_page_range.unmap_vmas.unmap_region.do_munmap.vm_munmap
      9.54            +0.2        9.76        perf-profile.calltrace.cycles-pp.unmap_vmas.unmap_region.do_munmap.vm_munmap.__x64_sys_munmap
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.do_munmap.vm_munmap.__x64_sys_munmap.do_syscall_64.entry_SYSCALL_64_after_hwframe
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.vm_munmap.__x64_sys_munmap.do_syscall_64.entry_SYSCALL_64_after_hwframe.munmap
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.__x64_sys_munmap.do_syscall_64.entry_SYSCALL_64_after_hwframe.munmap
      0.00            +0.6        0.56 +-  2%  perf-profile.calltrace.cycles-pp.lock_page_memcg.page_add_file_rmap.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +0.6        0.59        perf-profile.calltrace.cycles-pp.page_mapping.set_page_dirty.fault_dirty_shared_page.handle_pte_fault.__handle_mm_fault
      0.00            +0.6        0.60        perf-profile.calltrace.cycles-pp.current_time.file_update_time.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +0.7        0.68        perf-profile.calltrace.cycles-pp.___might_sleep.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault
      0.00            +0.7        0.74        perf-profile.calltrace.cycles-pp.unlock_page.fault_dirty_shared_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +0.8        0.80        perf-profile.calltrace.cycles-pp.set_page_dirty.fault_dirty_shared_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +0.9        0.88        perf-profile.calltrace.cycles-pp._raw_spin_lock.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +0.9        0.91        perf-profile.calltrace.cycles-pp.__set_page_dirty_no_writeback.fault_dirty_shared_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +1.3        1.27        perf-profile.calltrace.cycles-pp.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +1.3        1.30        perf-profile.calltrace.cycles-pp.file_update_time.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +2.8        2.76        perf-profile.calltrace.cycles-pp.fault_dirty_shared_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +6.8        6.81        perf-profile.calltrace.cycles-pp.page_add_file_rmap.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +9.4        9.39        perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +9.6        9.59        perf-profile.calltrace.cycles-pp.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +9.8        9.77        perf-profile.calltrace.cycles-pp.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault
      0.00           +10.4       10.37        perf-profile.calltrace.cycles-pp.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault
      0.00           +11.5       11.46        perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00           +11.6       11.60        perf-profile.calltrace.cycles-pp.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00           +26.6       26.62        perf-profile.calltrace.cycles-pp.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      7.88            -0.3        7.61        perf-profile.children.cycles-pp.find_get_entry
      1.34 +-  8%      -0.2        1.16 +-  2%  perf-profile.children.cycles-pp.hrtimer_interrupt
     10.41            -0.2       10.24        perf-profile.children.cycles-pp.native_irq_return_iret
      0.38 +- 28%      -0.1        0.26 +-  4%  perf-profile.children.cycles-pp.tick_sched_timer
     11.80            -0.1       11.68        perf-profile.children.cycles-pp.__do_fault
      0.55 +- 15%      -0.1        0.43 +-  2%  perf-profile.children.cycles-pp.__hrtimer_run_queues
      0.60            -0.1        0.51        perf-profile.children.cycles-pp.pmd_devmap_trans_unstable
      0.38 +- 13%      -0.1        0.29 +-  4%  perf-profile.children.cycles-pp.ktime_get
      7.68            -0.1        7.60        perf-profile.children.cycles-pp.swapgs_restore_regs_and_return_to_usermode
      5.18            -0.1        5.12        perf-profile.children.cycles-pp.trace_graph_entry
      0.79            -0.1        0.73        perf-profile.children.cycles-pp.down_read_trylock
      7.83            -0.1        7.76        perf-profile.children.cycles-pp.sync_regs
      3.01            -0.1        2.94        perf-profile.children.cycles-pp.fault_dirty_shared_page
      1.02            -0.1        0.96        perf-profile.children.cycles-pp._raw_spin_lock
      4.66            -0.1        4.61        perf-profile.children.cycles-pp.prepare_ftrace_return
      0.37 +-  8%      -0.1        0.32 +-  3%  perf-profile.children.cycles-pp.current_kernel_time64
      5.26            -0.1        5.21        perf-profile.children.cycles-pp.ftrace_graph_caller
      0.66 +-  5%      -0.1        0.61        perf-profile.children.cycles-pp.current_time
      0.18 +-  5%      -0.0        0.15 +-  3%  perf-profile.children.cycles-pp.update_process_times
      0.27            -0.0        0.26        perf-profile.children.cycles-pp._cond_resched
      0.16            -0.0        0.15 +-  3%  perf-profile.children.cycles-pp.rcu_all_qs
      0.94            +0.0        0.95        perf-profile.children.cycles-pp.vmacache_find
      0.48            +0.0        0.50        perf-profile.children.cycles-pp.__mod_node_page_state
      0.17            +0.0        0.19 +-  2%  perf-profile.children.cycles-pp.__unlock_page_memcg
      1.07            +0.0        1.10        perf-profile.children.cycles-pp.find_vma
      0.79 +-  3%      +0.1        0.86 +-  2%  perf-profile.children.cycles-pp.lock_page_memcg
      4.29            +0.1        4.40        perf-profile.children.cycles-pp.page_remove_rmap
      1.39 +-  2%      +0.1        1.52        perf-profile.children.cycles-pp.file_update_time
      0.00            +0.2        0.16        perf-profile.children.cycles-pp.__vm_normal_page
      9.63            +0.2        9.84        perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
      9.63            +0.2        9.84        perf-profile.children.cycles-pp.do_syscall_64
      9.63            +0.2        9.84        perf-profile.children.cycles-pp.unmap_page_range
     10.17            +0.2       10.39        perf-profile.children.cycles-pp.munmap
      9.56            +0.2        9.78        perf-profile.children.cycles-pp.unmap_region
      9.56            +0.2        9.78        perf-profile.children.cycles-pp.do_munmap
      9.56            +0.2        9.78        perf-profile.children.cycles-pp.vm_munmap
      9.56            +0.2        9.78        perf-profile.children.cycles-pp.__x64_sys_munmap
      9.54            +0.2        9.77        perf-profile.children.cycles-pp.unmap_vmas
      1.01            +0.2        1.25        perf-profile.children.cycles-pp.___might_sleep
      0.00            +1.6        1.59        perf-profile.children.cycles-pp.pte_map_lock
      0.00           +26.9       26.89        perf-profile.children.cycles-pp.handle_pte_fault
      4.25            -1.0        3.24        perf-profile.self.cycles-pp.__handle_mm_fault
      1.42            -0.3        1.11        perf-profile.self.cycles-pp.alloc_set_pte
      4.87            -0.3        4.59        perf-profile.self.cycles-pp.find_get_entry
     10.41            -0.2       10.24        perf-profile.self.cycles-pp.native_irq_return_iret
      0.37 +- 13%      -0.1        0.28 +-  4%  perf-profile.self.cycles-pp.ktime_get
      0.60            -0.1        0.51        perf-profile.self.cycles-pp.pmd_devmap_trans_unstable
      7.50            -0.1        7.42        perf-profile.self.cycles-pp.swapgs_restore_regs_and_return_to_usermode
      7.83            -0.1        7.76        perf-profile.self.cycles-pp.sync_regs
      4.85            -0.1        4.79        perf-profile.self.cycles-pp.trace_graph_entry
      1.01            -0.1        0.95        perf-profile.self.cycles-pp._raw_spin_lock
      0.78            -0.1        0.73        perf-profile.self.cycles-pp.down_read_trylock
      0.36 +-  9%      -0.1        0.31 +-  4%  perf-profile.self.cycles-pp.current_kernel_time64
      0.28            -0.0        0.23 +-  2%  perf-profile.self.cycles-pp.__do_fault
      1.04            -0.0        1.00        perf-profile.self.cycles-pp.find_lock_entry
      0.30            -0.0        0.28 +-  3%  perf-profile.self.cycles-pp.fault_dirty_shared_page
      0.70            -0.0        0.67        perf-profile.self.cycles-pp.prepare_ftrace_return
      0.44            -0.0        0.42        perf-profile.self.cycles-pp.do_page_fault
      0.16            -0.0        0.14        perf-profile.self.cycles-pp.rcu_all_qs
      0.78            -0.0        0.77        perf-profile.self.cycles-pp.shmem_getpage_gfp
      0.20            -0.0        0.19        perf-profile.self.cycles-pp._cond_resched
      0.50            +0.0        0.51        perf-profile.self.cycles-pp.set_page_dirty
      0.93            +0.0        0.95        perf-profile.self.cycles-pp.vmacache_find
      0.36 +-  2%      +0.0        0.38        perf-profile.self.cycles-pp.__might_sleep
      0.47            +0.0        0.50        perf-profile.self.cycles-pp.__mod_node_page_state
      0.17            +0.0        0.19 +-  2%  perf-profile.self.cycles-pp.__unlock_page_memcg
      2.34            +0.0        2.38        perf-profile.self.cycles-pp.unmap_page_range
      0.78 +-  3%      +0.1        0.85 +-  2%  perf-profile.self.cycles-pp.lock_page_memcg
      2.17            +0.1        2.24        perf-profile.self.cycles-pp.__do_page_fault
      0.00            +0.2        0.16 +-  3%  perf-profile.self.cycles-pp.__vm_normal_page
      1.00            +0.2        1.24        perf-profile.self.cycles-pp.___might_sleep
      0.00            +0.7        0.70        perf-profile.self.cycles-pp.pte_map_lock
      0.00            +1.4        1.42 +-  2%  perf-profile.self.cycles-pp.handle_pte_fault

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/never/context_switch1/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
           :3           33%           1:3     dmesg.WARNING:at#for_ip_interrupt_entry/0x
          2:3          -67%            :3     kmsg.pstore:crypto_comp_decompress_failed,ret=
          2:3          -67%            :3     kmsg.pstore:decompression_failed
         %stddev     %change         %stddev
             \          |                \  
    224431            -1.3%     221567        will-it-scale.per_process_ops
    237006            -2.2%     231907        will-it-scale.per_thread_ops
 1.601e+09 +- 29%     -46.9%  8.501e+08 +- 12%  will-it-scale.time.involuntary_context_switches
      5429            -1.6%       5344        will-it-scale.time.user_time
  88596221            -1.7%   87067269        will-it-scale.workload
      6863 +-  6%      -9.7%       6200        boot-time.idle
    144908 +- 40%     -66.8%      48173 +- 93%  meminfo.CmaFree
      0.00 +- 70%      +0.0        0.00        mpstat.cpu.iowait%
    448336 +- 14%     -34.8%     292125 +-  3%  turbostat.C1
      7684 +-  6%      -9.5%       6957        uptime.idle
 1.601e+09 +- 29%     -46.9%  8.501e+08 +- 12%  time.involuntary_context_switches
      5429            -1.6%       5344        time.user_time
  44013162            -1.7%   43243125        vmstat.system.cs
    207684            -1.1%     205485        vmstat.system.in
   2217033 +- 15%     -15.8%    1866876 +-  2%  cpuidle.C1.time
    451218 +- 14%     -34.7%     294841 +-  2%  cpuidle.C1.usage
     24839 +- 10%     -19.9%      19896        cpuidle.POLL.time
      7656 +- 11%     -38.9%       4676 +-  8%  cpuidle.POLL.usage
      5.48 +- 49%     -67.3%       1.79 +-100%  irq_exception_noise.__do_page_fault.95th
      9.46 +- 21%     -58.2%       3.95 +- 64%  irq_exception_noise.__do_page_fault.99th
     35.67 +-  8%   +1394.4%     533.00 +- 96%  irq_exception_noise.irq_nr
     52109 +-  3%     -16.0%      43784 +-  4%  irq_exception_noise.softirq_time
     36226 +- 40%     -66.7%      12048 +- 93%  proc-vmstat.nr_free_cma
     25916            -1.0%      25659        proc-vmstat.nr_slab_reclaimable
     16279 +-  8%   +2646.1%     447053 +- 82%  proc-vmstat.pgalloc_movable
   2231117           -18.4%    1820828 +- 20%  proc-vmstat.pgalloc_normal
   1109316 +- 46%     -86.9%     145207 +-109%  numa-numastat.node1.local_node
   1114700 +- 45%     -84.5%     172877 +- 85%  numa-numastat.node1.numa_hit
      5523 +-140%    +402.8%      27768 +- 39%  numa-numastat.node1.other_node
     29013 +- 29%   +3048.1%     913379 +- 73%  numa-numastat.node3.local_node
     65032 +- 13%   +1335.1%     933270 +- 70%  numa-numastat.node3.numa_hit
     36018           -44.8%      19897 +- 75%  numa-numastat.node3.other_node
     12.79 +- 21%   +7739.1%       1002 +-136%  sched_debug.cpu.cpu_load[1].max
      1.82 +- 10%   +3901.1%      72.92 +-135%  sched_debug.cpu.cpu_load[1].stddev
      1.71 +-  4%   +5055.8%      88.08 +-137%  sched_debug.cpu.cpu_load[2].stddev
     12.33 +- 23%   +9061.9%       1129 +-139%  sched_debug.cpu.cpu_load[3].max
      1.78 +- 10%   +4514.8%      82.18 +-137%  sched_debug.cpu.cpu_load[3].stddev
      4692 +- 72%    +154.5%      11945 +- 29%  sched_debug.cpu.max_idle_balance_cost.stddev
     23979            -8.3%      21983        slabinfo.kmalloc-96.active_objs
      1358 +-  6%     -17.9%       1114 +-  3%  slabinfo.nsproxy.active_objs
      1358 +-  6%     -17.9%       1114 +-  3%  slabinfo.nsproxy.num_objs
     15229           +12.4%      17119        slabinfo.pde_opener.active_objs
     15229           +12.4%      17119        slabinfo.pde_opener.num_objs
     59541 +-  8%     -10.1%      53537 +-  8%  slabinfo.vm_area_struct.active_objs
     59612 +-  8%     -10.1%      53604 +-  8%  slabinfo.vm_area_struct.num_objs
 4.163e+13            -1.4%  4.105e+13        perf-stat.branch-instructions
 6.537e+11            -1.2%  6.459e+11        perf-stat.branch-misses
 2.667e+10            -1.7%  2.621e+10        perf-stat.context-switches
      1.21            +1.3%       1.22        perf-stat.cpi
    150508            -9.8%     135825 +-  3%  perf-stat.cpu-migrations
      5.75 +- 33%      +5.4       11.11 +- 26%  perf-stat.iTLB-load-miss-rate%
 3.619e+09 +- 36%    +100.9%  7.272e+09 +- 30%  perf-stat.iTLB-load-misses
 2.089e+14            -1.3%  2.062e+14        perf-stat.instructions
     64607 +- 29%     -50.5%      31964 +- 37%  perf-stat.instructions-per-iTLB-miss
      0.83            -1.3%       0.82        perf-stat.ipc
      3972 +-  4%     -14.7%       3388 +-  8%  numa-meminfo.node0.PageTables
    207919 +- 25%     -57.2%      88989 +- 74%  numa-meminfo.node1.Active
    207715 +- 26%     -57.3%      88785 +- 74%  numa-meminfo.node1.Active(anon)
    356529           -34.3%     234069 +-  2%  numa-meminfo.node1.FilePages
    789129 +-  5%     -19.8%     633161 +- 12%  numa-meminfo.node1.MemUsed
     34777 +-  8%     -48.2%      18010 +- 30%  numa-meminfo.node1.SReclaimable
     69641 +-  4%     -20.7%      55250 +- 12%  numa-meminfo.node1.SUnreclaim
    125526 +-  4%     -96.3%       4602 +- 41%  numa-meminfo.node1.Shmem
    104419           -29.8%      73261 +- 16%  numa-meminfo.node1.Slab
    103661 +- 17%     -72.0%      29029 +- 99%  numa-meminfo.node2.Active
    103661 +- 17%     -72.2%      28829 +-101%  numa-meminfo.node2.Active(anon)
    103564 +- 18%     -72.0%      29007 +-100%  numa-meminfo.node2.AnonPages
    671654 +-  7%     -14.6%     573598 +-  4%  numa-meminfo.node2.MemUsed
     44206 +-127%    +301.4%     177465 +- 42%  numa-meminfo.node3.Active
     44206 +-127%    +301.0%     177263 +- 42%  numa-meminfo.node3.Active(anon)
      8738           +12.2%       9805 +-  8%  numa-meminfo.node3.KernelStack
    603605 +-  9%     +27.8%     771554 +- 14%  numa-meminfo.node3.MemUsed
     14438 +-  6%    +122.9%      32181 +- 42%  numa-meminfo.node3.SReclaimable
      2786 +-137%   +3302.0%      94792 +- 71%  numa-meminfo.node3.Shmem
     71461 +-  7%     +45.2%     103771 +- 29%  numa-meminfo.node3.Slab
    247197 +-  4%      -7.8%     227843        numa-meminfo.node3.Unevictable
    991.67 +-  4%     -14.7%     846.00 +-  8%  numa-vmstat.node0.nr_page_table_pages
     51926 +- 26%     -57.3%      22196 +- 74%  numa-vmstat.node1.nr_active_anon
     89137           -34.4%      58516 +-  2%  numa-vmstat.node1.nr_file_pages
      1679 +-  5%     -10.8%       1498 +-  4%  numa-vmstat.node1.nr_mapped
     31386 +-  4%     -96.3%       1150 +- 41%  numa-vmstat.node1.nr_shmem
      8694 +-  8%     -48.2%       4502 +- 30%  numa-vmstat.node1.nr_slab_reclaimable
     17410 +-  4%     -20.7%      13812 +- 12%  numa-vmstat.node1.nr_slab_unreclaimable
     51926 +- 26%     -57.3%      22196 +- 74%  numa-vmstat.node1.nr_zone_active_anon
   1037174 +- 24%     -57.0%     446205 +- 35%  numa-vmstat.node1.numa_hit
    961611 +- 26%     -65.8%     328687 +- 50%  numa-vmstat.node1.numa_local
     75563 +- 44%     +55.5%     117517 +-  9%  numa-vmstat.node1.numa_other
     25914 +- 17%     -72.2%       7206 +-101%  numa-vmstat.node2.nr_active_anon
     25891 +- 18%     -72.0%       7251 +-100%  numa-vmstat.node2.nr_anon_pages
     25914 +- 17%     -72.2%       7206 +-101%  numa-vmstat.node2.nr_zone_active_anon
     11051 +-127%    +301.0%      44309 +- 42%  numa-vmstat.node3.nr_active_anon
     36227 +- 40%     -66.7%      12049 +- 93%  numa-vmstat.node3.nr_free_cma
      0.33 +-141%  +25000.0%      83.67 +- 81%  numa-vmstat.node3.nr_inactive_file
      8739           +12.2%       9806 +-  8%  numa-vmstat.node3.nr_kernel_stack
    696.67 +-137%   +3299.7%      23684 +- 71%  numa-vmstat.node3.nr_shmem
      3609 +-  6%    +122.9%       8044 +- 42%  numa-vmstat.node3.nr_slab_reclaimable
     61799 +-  4%      -7.8%      56960        numa-vmstat.node3.nr_unevictable
     11053 +-127%    +301.4%      44361 +- 42%  numa-vmstat.node3.nr_zone_active_anon
      0.33 +-141%  +25000.0%      83.67 +- 81%  numa-vmstat.node3.nr_zone_inactive_file
     61799 +-  4%      -7.8%      56960        numa-vmstat.node3.nr_zone_unevictable
    217951 +-  8%    +280.8%     829976 +- 65%  numa-vmstat.node3.numa_hit
     91303 +- 19%    +689.3%     720647 +- 77%  numa-vmstat.node3.numa_local
    126648           -13.7%     109329 +- 13%  numa-vmstat.node3.numa_other
      8.54            -0.1        8.40        perf-profile.calltrace.cycles-pp.dequeue_task_fair.__schedule.schedule.pipe_wait.pipe_read
      5.04            -0.1        4.94        perf-profile.calltrace.cycles-pp.__switch_to.read
      3.43            -0.1        3.35        perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.write
      2.77            -0.1        2.72        perf-profile.calltrace.cycles-pp.reweight_entity.enqueue_task_fair.ttwu_do_activate.try_to_wake_up.autoremove_wake_function
      1.99            -0.0        1.94        perf-profile.calltrace.cycles-pp.copy_page_to_iter.pipe_read.__vfs_read.vfs_read.ksys_read
      0.60 +-  2%      -0.0        0.57 +-  2%  perf-profile.calltrace.cycles-pp.find_next_bit.cpumask_next_wrap.select_idle_sibling.select_task_rq_fair.try_to_wake_up
      0.81            -0.0        0.78        perf-profile.calltrace.cycles-pp.___perf_sw_event.__schedule.schedule.pipe_wait.pipe_read
      0.78            +0.0        0.80        perf-profile.calltrace.cycles-pp.__fdget_pos.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe.write
      0.73            +0.0        0.75        perf-profile.calltrace.cycles-pp.__fget_light.__fdget_pos.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.92            +0.0        0.95        perf-profile.calltrace.cycles-pp.check_preempt_wakeup.check_preempt_curr.ttwu_do_wakeup.try_to_wake_up.autoremove_wake_function
      2.11            +0.0        2.15        perf-profile.calltrace.cycles-pp.security_file_permission.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      7.00            -0.1        6.86        perf-profile.children.cycles-pp.syscall_return_via_sysret
      5.26            -0.1        5.14        perf-profile.children.cycles-pp.__switch_to
      5.65            -0.1        5.56        perf-profile.children.cycles-pp.reweight_entity
      2.17            -0.1        2.12        perf-profile.children.cycles-pp.copy_page_to_iter
      2.94            -0.0        2.90        perf-profile.children.cycles-pp.update_cfs_group
      3.11            -0.0        3.07        perf-profile.children.cycles-pp.pick_next_task_fair
      2.59            -0.0        2.55        perf-profile.children.cycles-pp.load_new_mm_cr3
      1.92            -0.0        1.88        perf-profile.children.cycles-pp._raw_spin_lock_irqsave
      1.11            -0.0        1.08 +-  2%  perf-profile.children.cycles-pp.find_next_bit
      0.59            -0.0        0.56        perf-profile.children.cycles-pp.finish_task_switch
      0.14 +- 15%      -0.0        0.11 +- 16%  perf-profile.children.cycles-pp.write@plt
      1.21            -0.0        1.18        perf-profile.children.cycles-pp.set_next_entity
      0.85            -0.0        0.82        perf-profile.children.cycles-pp.___perf_sw_event
      0.13 +-  3%      -0.0        0.11 +-  4%  perf-profile.children.cycles-pp.timespec_trunc
      0.47 +-  2%      -0.0        0.45        perf-profile.children.cycles-pp.anon_pipe_buf_release
      0.38 +-  2%      -0.0        0.36        perf-profile.children.cycles-pp.file_update_time
      0.74            -0.0        0.73        perf-profile.children.cycles-pp.copyout
      0.41 +-  2%      -0.0        0.39        perf-profile.children.cycles-pp.copy_user_enhanced_fast_string
      0.32            -0.0        0.30        perf-profile.children.cycles-pp.__x64_sys_read
      0.14            -0.0        0.12 +-  3%  perf-profile.children.cycles-pp.current_kernel_time64
      0.91            +0.0        0.92        perf-profile.children.cycles-pp.touch_atime
      0.40            +0.0        0.41        perf-profile.children.cycles-pp._cond_resched
      0.18 +-  2%      +0.0        0.20        perf-profile.children.cycles-pp.activate_task
      0.05            +0.0        0.07 +-  6%  perf-profile.children.cycles-pp.default_wake_function
      0.24            +0.0        0.27 +-  3%  perf-profile.children.cycles-pp.rcu_all_qs
      0.60 +-  2%      +0.0        0.64 +-  2%  perf-profile.children.cycles-pp.update_min_vruntime
      0.42 +-  4%      +0.0        0.46 +-  4%  perf-profile.children.cycles-pp.probe_sched_switch
      1.33            +0.0        1.38        perf-profile.children.cycles-pp.__fget_light
      0.53 +-  2%      +0.1        0.58        perf-profile.children.cycles-pp.entry_SYSCALL_64_stage2
      0.31            +0.1        0.36 +-  2%  perf-profile.children.cycles-pp.generic_pipe_buf_confirm
      4.35            +0.1        4.41        perf-profile.children.cycles-pp.switch_mm_irqs_off
      2.52            +0.1        2.58        perf-profile.children.cycles-pp.selinux_file_permission
      0.00            +0.1        0.07 +- 11%  perf-profile.children.cycles-pp.hrtick_update
      7.00            -0.1        6.86        perf-profile.self.cycles-pp.syscall_return_via_sysret
      5.26            -0.1        5.14        perf-profile.self.cycles-pp.__switch_to
      0.29            -0.1        0.19 +-  2%  perf-profile.self.cycles-pp.ksys_read
      1.49            -0.1        1.43        perf-profile.self.cycles-pp.dequeue_task_fair
      2.41            -0.1        2.35        perf-profile.self.cycles-pp.__schedule
      1.46            -0.0        1.41        perf-profile.self.cycles-pp.select_task_rq_fair
      2.94            -0.0        2.90        perf-profile.self.cycles-pp.update_cfs_group
      0.44            -0.0        0.40        perf-profile.self.cycles-pp.dequeue_entity
      0.48            -0.0        0.44        perf-profile.self.cycles-pp.finish_task_switch
      2.59            -0.0        2.55        perf-profile.self.cycles-pp.load_new_mm_cr3
      1.11            -0.0        1.08 +-  2%  perf-profile.self.cycles-pp.find_next_bit
      1.91            -0.0        1.88        perf-profile.self.cycles-pp._raw_spin_lock_irqsave
      0.78            -0.0        0.75        perf-profile.self.cycles-pp.___perf_sw_event
      0.14 +- 15%      -0.0        0.11 +- 16%  perf-profile.self.cycles-pp.write@plt
      0.37            -0.0        0.35 +-  2%  perf-profile.self.cycles-pp.__wake_up_common_lock
      0.20 +-  2%      -0.0        0.17 +-  2%  perf-profile.self.cycles-pp.__fdget_pos
      0.47 +-  2%      -0.0        0.44        perf-profile.self.cycles-pp.anon_pipe_buf_release
      0.87            -0.0        0.85        perf-profile.self.cycles-pp.copy_user_generic_unrolled
      0.13 +-  3%      -0.0        0.11 +-  4%  perf-profile.self.cycles-pp.timespec_trunc
      0.41 +-  2%      -0.0        0.39        perf-profile.self.cycles-pp.copy_user_enhanced_fast_string
      0.38            -0.0        0.36        perf-profile.self.cycles-pp.__wake_up_common
      0.32            -0.0        0.30        perf-profile.self.cycles-pp.__x64_sys_read
      0.14 +-  3%      -0.0        0.12 +-  3%  perf-profile.self.cycles-pp.current_kernel_time64
      0.30            -0.0        0.28        perf-profile.self.cycles-pp.set_next_entity
      0.28 +-  3%      +0.0        0.30        perf-profile.self.cycles-pp._cond_resched
      0.18 +-  2%      +0.0        0.20        perf-profile.self.cycles-pp.activate_task
      0.17 +-  2%      +0.0        0.19        perf-profile.self.cycles-pp.__might_fault
      0.05            +0.0        0.07 +-  6%  perf-profile.self.cycles-pp.default_wake_function
      0.17 +-  2%      +0.0        0.20        perf-profile.self.cycles-pp.ttwu_do_activate
      0.66            +0.0        0.69        perf-profile.self.cycles-pp.write
      0.24            +0.0        0.27 +-  3%  perf-profile.self.cycles-pp.rcu_all_qs
      0.67            +0.0        0.70        perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
      0.60 +-  2%      +0.0        0.64 +-  2%  perf-profile.self.cycles-pp.update_min_vruntime
      0.42 +-  4%      +0.0        0.46 +-  4%  perf-profile.self.cycles-pp.probe_sched_switch
      1.33            +0.0        1.37        perf-profile.self.cycles-pp.__fget_light
      1.61            +0.0        1.66        perf-profile.self.cycles-pp.pipe_read
      0.53 +-  2%      +0.1        0.58        perf-profile.self.cycles-pp.entry_SYSCALL_64_stage2
      0.31            +0.1        0.36 +-  2%  perf-profile.self.cycles-pp.generic_pipe_buf_confirm
      1.04            +0.1        1.11        perf-profile.self.cycles-pp.pipe_write
      0.00            +0.1        0.07 +- 11%  perf-profile.self.cycles-pp.hrtick_update
      2.00            +0.1        2.08        perf-profile.self.cycles-pp.switch_mm_irqs_off

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/never/page_fault3/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
          1:3          -33%            :3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=file_update_time/0x
           :3           33%           1:3     stderr.mount.nfs:Connection_timed_out
         34:3         -401%          22:3     perf-profile.calltrace.cycles-pp.error_entry.testcase
         17:3         -207%          11:3     perf-profile.calltrace.cycles-pp.sync_regs.error_entry.testcase
         34:3         -404%          22:3     perf-profile.children.cycles-pp.error_entry
          0:3           -2%           0:3     perf-profile.children.cycles-pp.error_exit
         16:3         -196%          11:3     perf-profile.self.cycles-pp.error_entry
          0:3           -2%           0:3     perf-profile.self.cycles-pp.error_exit
         %stddev     %change         %stddev
             \          |                \  
    467454            -1.8%     459251        will-it-scale.per_process_ops
     10856 +-  4%     -23.1%       8344 +-  7%  will-it-scale.per_thread_ops
    118134 +-  2%     +11.7%     131943        will-it-scale.time.involuntary_context_switches
 6.277e+08 +-  4%     -23.1%  4.827e+08 +-  7%  will-it-scale.time.minor_page_faults
      7406            +5.8%       7839        will-it-scale.time.percent_of_cpu_this_job_got
     44526            +5.8%      47106        will-it-scale.time.system_time
   7351468 +-  5%     -18.3%    6009014 +-  7%  will-it-scale.time.voluntary_context_switches
  91835846            -2.2%   89778599        will-it-scale.workload
   2534640            +4.3%    2643005 +-  2%  interrupts.CAL:Function_call_interrupts
      2819 +-  5%     +22.9%       3464 +- 18%  kthread_noise.total_time
     30273 +-  4%     -12.7%      26415 +-  5%  vmstat.system.cs
      1.52 +-  2%     +15.2%       1.75 +-  2%  irq_exception_noise.__do_page_fault.99th
    296.67 +- 12%     -36.7%     187.67 +- 12%  irq_exception_noise.softirq_time
    230900 +-  3%     +30.3%     300925 +-  3%  meminfo.Inactive
    230184 +-  3%     +30.4%     300180 +-  3%  meminfo.Inactive(anon)
     11.62 +-  3%      -2.2        9.40 +-  5%  mpstat.cpu.idle%
      0.00 +- 14%      -0.0        0.00 +-  4%  mpstat.cpu.iowait%
   7992174           -11.1%    7101976 +-  3%  softirqs.RCU
   4973624 +-  2%     -12.9%    4333370 +-  2%  softirqs.SCHED
    118134 +-  2%     +11.7%     131943        time.involuntary_context_switches
 6.277e+08 +-  4%     -23.1%  4.827e+08 +-  7%  time.minor_page_faults
      7406            +5.8%       7839        time.percent_of_cpu_this_job_got
     44526            +5.8%      47106        time.system_time
   7351468 +-  5%     -18.3%    6009014 +-  7%  time.voluntary_context_switches
 2.702e+09 +-  5%     -16.7%  2.251e+09 +-  7%  cpuidle.C1E.time
   6834329 +-  5%     -15.8%    5756243 +-  7%  cpuidle.C1E.usage
 1.046e+10 +-  3%     -19.8%  8.389e+09 +-  4%  cpuidle.C6.time
  13961845 +-  3%     -19.3%   11265555 +-  4%  cpuidle.C6.usage
   1309307 +-  7%     -14.8%    1116168 +-  8%  cpuidle.POLL.time
     19774 +-  6%     -13.7%      17063 +-  7%  cpuidle.POLL.usage
      2523 +-  4%     -11.1%       2243 +-  4%  slabinfo.biovec-64.active_objs
      2523 +-  4%     -11.1%       2243 +-  4%  slabinfo.biovec-64.num_objs
      2610 +-  8%     -33.7%       1731 +- 22%  slabinfo.dmaengine-unmap-16.active_objs
      2610 +-  8%     -33.7%       1731 +- 22%  slabinfo.dmaengine-unmap-16.num_objs
      5118 +- 17%     -22.6%       3962 +-  9%  slabinfo.eventpoll_pwq.active_objs
      5118 +- 17%     -22.6%       3962 +-  9%  slabinfo.eventpoll_pwq.num_objs
      4583 +-  3%     -14.0%       3941 +-  4%  slabinfo.sock_inode_cache.active_objs
      4583 +-  3%     -14.0%       3941 +-  4%  slabinfo.sock_inode_cache.num_objs
      1933            +2.6%       1984        turbostat.Avg_MHz
   6832021 +-  5%     -15.8%    5754156 +-  7%  turbostat.C1E
      2.32 +-  5%      -0.4        1.94 +-  7%  turbostat.C1E%
  13954211 +-  3%     -19.3%   11259436 +-  4%  turbostat.C6
      8.97 +-  3%      -1.8        7.20 +-  4%  turbostat.C6%
      6.18 +-  4%     -17.1%       5.13 +-  5%  turbostat.CPU%c1
      5.12 +-  3%     -21.7%       4.01 +-  4%  turbostat.CPU%c6
      1.76 +-  2%     -34.7%       1.15 +-  2%  turbostat.Pkg%pc2
     57314 +-  4%     +30.4%      74717 +-  4%  proc-vmstat.nr_inactive_anon
     57319 +-  4%     +30.4%      74719 +-  4%  proc-vmstat.nr_zone_inactive_anon
     24415 +- 19%     -62.2%       9236 +-  7%  proc-vmstat.numa_hint_faults
  69661453            -1.8%   68405712        proc-vmstat.numa_hit
  69553390            -1.8%   68297790        proc-vmstat.numa_local
      8792 +- 29%     -92.6%     654.33 +- 23%  proc-vmstat.numa_pages_migrated
     40251 +- 32%     -76.5%       9474 +-  3%  proc-vmstat.numa_pte_updates
  69522532            -1.6%   68383074        proc-vmstat.pgalloc_normal
 2.762e+10            -2.2%  2.701e+10        proc-vmstat.pgfault
  68825100            -1.5%   67772256        proc-vmstat.pgfree
      8792 +- 29%     -92.6%     654.33 +- 23%  proc-vmstat.pgmigrate_success
     57992 +-  6%     +56.2%      90591 +-  3%  numa-meminfo.node0.Inactive
     57916 +-  6%     +56.3%      90513 +-  3%  numa-meminfo.node0.Inactive(anon)
     37285 +- 12%     +36.0%      50709 +-  5%  numa-meminfo.node0.SReclaimable
    110971 +-  8%     +22.7%     136209 +-  8%  numa-meminfo.node0.Slab
     23601 +- 55%    +559.5%     155651 +- 36%  numa-meminfo.node1.AnonPages
     62484 +- 12%     +17.5%      73417 +-  3%  numa-meminfo.node1.Inactive
     62323 +- 12%     +17.2%      73023 +-  4%  numa-meminfo.node1.Inactive(anon)
    109714 +- 63%     -85.6%      15832 +- 96%  numa-meminfo.node2.AnonPages
     52236 +- 13%     +22.7%      64074 +-  3%  numa-meminfo.node2.Inactive
     51922 +- 12%     +23.2%      63963 +-  3%  numa-meminfo.node2.Inactive(anon)
     60241 +- 11%     +21.9%      73442 +-  8%  numa-meminfo.node3.Inactive
     60077 +- 12%     +22.0%      73279 +-  8%  numa-meminfo.node3.Inactive(anon)
     14093 +-  6%     +55.9%      21977 +-  3%  numa-vmstat.node0.nr_inactive_anon
      9321 +- 12%     +36.0%      12675 +-  5%  numa-vmstat.node0.nr_slab_reclaimable
     14090 +-  6%     +56.0%      21977 +-  3%  numa-vmstat.node0.nr_zone_inactive_anon
      5900 +- 55%    +559.4%      38909 +- 36%  numa-vmstat.node1.nr_anon_pages
     15413 +- 12%     +14.8%      17688 +-  4%  numa-vmstat.node1.nr_inactive_anon
     15413 +- 12%     +14.8%      17688 +-  4%  numa-vmstat.node1.nr_zone_inactive_anon
     27430 +- 63%     -85.6%       3960 +- 96%  numa-vmstat.node2.nr_anon_pages
     12928 +- 12%     +20.0%      15508 +-  3%  numa-vmstat.node2.nr_inactive_anon
     12927 +- 12%     +20.0%      15507 +-  3%  numa-vmstat.node2.nr_zone_inactive_anon
      6229 +- 10%    +117.5%      13547 +- 44%  numa-vmstat.node3
     14669 +- 11%     +19.6%      17537 +-  7%  numa-vmstat.node3.nr_inactive_anon
     14674 +- 11%     +19.5%      17541 +-  7%  numa-vmstat.node3.nr_zone_inactive_anon
     24617 +-141%    -100.0%       0.00        latency_stats.avg.io_schedule.nfs_lock_and_join_requests.nfs_updatepage.nfs_write_end.generic_perform_write.nfs_file_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      5049 +-105%     -99.4%      28.33 +- 82%  latency_stats.avg.call_rwsem_down_write_failed.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
    152457 +- 27%    +233.6%     508656 +- 92%  latency_stats.avg.max
      0.00       +3.9e+107%     390767 +-141%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_openat
     24617 +-141%    -100.0%       0.00        latency_stats.max.io_schedule.nfs_lock_and_join_requests.nfs_updatepage.nfs_write_end.generic_perform_write.nfs_file_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4240 +-141%    -100.0%       0.00        latency_stats.max.call_rwsem_down_write_failed.do_unlinkat.do_syscall_64.entry_SYSCALL_64_after_hwframe
      8565 +- 70%     -99.1%      80.33 +-115%  latency_stats.max.call_rwsem_down_write_failed.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
    204835 +-  6%    +457.6%    1142244 +-114%  latency_stats.max.max
      0.00       +5.1e+105%       5057 +-141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_access.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_openat.do_filp_open
      0.00         +1e+108%     995083 +-141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_openat
     13175 +-  4%    -100.0%       0.00        latency_stats.sum.io_schedule.__lock_page_or_retry.filemap_fault.__do_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault.page_fault
     24617 +-141%    -100.0%       0.00        latency_stats.sum.io_schedule.nfs_lock_and_join_requests.nfs_updatepage.nfs_write_end.generic_perform_write.nfs_file_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4260 +-141%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_write_failed.do_unlinkat.do_syscall_64.entry_SYSCALL_64_after_hwframe
      8640 +- 70%     -97.5%     216.33 +-108%  latency_stats.sum.call_rwsem_down_write_failed.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      6673 +- 89%     -92.8%     477.67 +- 74%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
      0.00       +4.2e+105%       4228 +-130%  latency_stats.sum.io_schedule.__lock_page_killable.__lock_page_or_retry.filemap_fault.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault.page_fault
      0.00       +7.5e+105%       7450 +- 98%  latency_stats.sum.io_schedule.__lock_page_or_retry.filemap_fault.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault.page_fault
      0.00       +1.3e+106%      13050 +-141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_access.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_openat.do_filp_open
      0.00       +1.5e+110%  1.508e+08 +-141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_openat
      0.97            -0.0        0.94        perf-stat.branch-miss-rate%
 1.329e+11            -2.6%  1.294e+11        perf-stat.branch-misses
 2.254e+11            -1.9%   2.21e+11        perf-stat.cache-references
  18308779 +-  4%     -12.8%   15969618 +-  5%  perf-stat.context-switches
      3.20            +1.8%       3.26        perf-stat.cpi
 2.233e+14            +2.7%  2.293e+14        perf-stat.cpu-cycles
      4.01            -0.2        3.83        perf-stat.dTLB-store-miss-rate%
  4.51e+11            -2.2%   4.41e+11        perf-stat.dTLB-store-misses
  1.08e+13            +2.6%  1.109e+13        perf-stat.dTLB-stores
 3.158e+10 +-  5%     +16.8%  3.689e+10 +-  2%  perf-stat.iTLB-load-misses
      2214 +-  5%     -13.8%       1907 +-  2%  perf-stat.instructions-per-iTLB-miss
      0.31            -1.8%       0.31        perf-stat.ipc
 2.762e+10            -2.2%  2.701e+10        perf-stat.minor-faults
 1.535e+10           -11.2%  1.362e+10        perf-stat.node-loads
      9.75            +1.1       10.89        perf-stat.node-store-miss-rate%
 3.012e+09           +10.6%  3.332e+09 +-  2%  perf-stat.node-store-misses
 2.787e+10            -2.2%  2.725e+10        perf-stat.node-stores
 2.762e+10            -2.2%  2.701e+10        perf-stat.page-faults
    759458            +3.2%     783404        perf-stat.path-length
    246.39 +- 15%     -20.4%     196.12 +-  6%  sched_debug.cfs_rq:/.load_avg.max
      0.21 +-  3%      +9.0%       0.23 +-  4%  sched_debug.cfs_rq:/.nr_running.stddev
     16.64 +- 27%     +61.0%      26.79 +- 17%  sched_debug.cfs_rq:/.nr_spread_over.max
     75.15           -14.4%      64.30 +-  4%  sched_debug.cfs_rq:/.util_avg.stddev
    178.80 +-  3%     +25.4%     224.12 +-  7%  sched_debug.cfs_rq:/.util_est_enqueued.avg
      1075 +-  5%     -12.3%     943.36 +-  2%  sched_debug.cfs_rq:/.util_est_enqueued.max
   2093630 +- 27%     -36.1%    1337941 +- 16%  sched_debug.cpu.avg_idle.max
    297057 +- 11%     +37.8%     409294 +- 14%  sched_debug.cpu.avg_idle.min
    293240 +- 55%     -62.3%     110571 +- 13%  sched_debug.cpu.avg_idle.stddev
    770075 +-  9%     -19.3%     621136 +- 12%  sched_debug.cpu.max_idle_balance_cost.max
     48919 +- 46%     -66.9%      16190 +- 81%  sched_debug.cpu.max_idle_balance_cost.stddev
     21716 +-  5%     -16.8%      18061 +-  7%  sched_debug.cpu.nr_switches.min
     21519 +-  5%     -17.7%      17700 +-  7%  sched_debug.cpu.sched_count.min
     10586 +-  5%     -18.1%       8669 +-  7%  sched_debug.cpu.sched_goidle.avg
     14183 +-  3%     -17.6%      11693 +-  5%  sched_debug.cpu.sched_goidle.max
     10322 +-  5%     -18.6%       8407 +-  7%  sched_debug.cpu.sched_goidle.min
    400.99 +-  8%     -13.0%     348.75 +-  3%  sched_debug.cpu.sched_goidle.stddev
      5459 +-  8%     +10.0%       6006 +-  3%  sched_debug.cpu.ttwu_local.avg
      8.47 +- 42%    +345.8%      37.73 +- 77%  sched_debug.rt_rq:/.rt_time.max
      0.61 +- 42%    +343.0%       2.72 +- 77%  sched_debug.rt_rq:/.rt_time.stddev
     91.98           -30.9       61.11 +- 70%  perf-profile.calltrace.cycles-pp.testcase
      9.05            -9.1        0.00        perf-profile.calltrace.cycles-pp.__do_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      8.91            -8.9        0.00        perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      8.06            -8.1        0.00        perf-profile.calltrace.cycles-pp.shmem_getpage_gfp.shmem_fault.__do_fault.__handle_mm_fault.handle_mm_fault
      7.59            -7.6        0.00        perf-profile.calltrace.cycles-pp.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault.__handle_mm_fault
      7.44            -7.4        0.00        perf-profile.calltrace.cycles-pp.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      7.28            -7.3        0.00        perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      5.31            -5.3        0.00        perf-profile.calltrace.cycles-pp.page_add_file_rmap.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault
      8.08            -2.8        5.30 +- 70%  perf-profile.calltrace.cycles-pp.native_irq_return_iret.testcase
      5.95            -2.1        3.83 +- 70%  perf-profile.calltrace.cycles-pp.find_get_entry.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault
      5.95            -2.0        3.93 +- 70%  perf-profile.calltrace.cycles-pp.swapgs_restore_regs_and_return_to_usermode.testcase
      3.10            -1.1        2.01 +- 70%  perf-profile.calltrace.cycles-pp.__perf_sw_event.__do_page_fault.do_page_fault.page_fault.testcase
      2.36            -0.8        1.55 +- 70%  perf-profile.calltrace.cycles-pp.___perf_sw_event.__perf_sw_event.__do_page_fault.do_page_fault.page_fault
      1.08            -0.4        0.70 +- 70%  perf-profile.calltrace.cycles-pp.do_page_fault.testcase
      0.82            -0.3        0.54 +- 70%  perf-profile.calltrace.cycles-pp.trace_graph_entry.do_page_fault.testcase
      0.77            -0.3        0.50 +- 70%  perf-profile.calltrace.cycles-pp.ftrace_graph_caller.__do_page_fault.do_page_fault.page_fault.testcase
      0.59            -0.2        0.37 +- 70%  perf-profile.calltrace.cycles-pp.down_read_trylock.__do_page_fault.do_page_fault.page_fault.testcase
     91.98           -30.9       61.11 +- 70%  perf-profile.children.cycles-pp.testcase
      9.14            -3.2        5.99 +- 70%  perf-profile.children.cycles-pp.__do_fault
      8.20            -2.8        5.40 +- 70%  perf-profile.children.cycles-pp.shmem_getpage_gfp
      8.08            -2.8        5.31 +- 70%  perf-profile.children.cycles-pp.native_irq_return_iret
      6.08            -2.2        3.92 +- 70%  perf-profile.children.cycles-pp.find_get_entry
      6.08            -2.1        3.96 +- 70%  perf-profile.children.cycles-pp.sync_regs
      5.95            -2.0        3.93 +- 70%  perf-profile.children.cycles-pp.swapgs_restore_regs_and_return_to_usermode
      4.12            -1.4        2.73 +- 70%  perf-profile.children.cycles-pp.ftrace_graph_caller
      3.65            -1.2        2.42 +- 70%  perf-profile.children.cycles-pp.prepare_ftrace_return
      3.18            -1.1        2.07 +- 70%  perf-profile.children.cycles-pp.__perf_sw_event
      2.34            -0.8        1.52 +- 70%  perf-profile.children.cycles-pp.fault_dirty_shared_page
      0.80            -0.3        0.50 +- 70%  perf-profile.children.cycles-pp._raw_spin_lock
      0.76            -0.3        0.50 +- 70%  perf-profile.children.cycles-pp.tlb_flush_mmu_free
      0.61            -0.2        0.39 +- 70%  perf-profile.children.cycles-pp.down_read_trylock
      0.48 +-  2%      -0.2        0.28 +- 70%  perf-profile.children.cycles-pp.pmd_devmap_trans_unstable
      0.26 +-  6%      -0.1        0.15 +- 71%  perf-profile.children.cycles-pp.ktime_get
      0.20 +-  2%      -0.1        0.12 +- 70%  perf-profile.children.cycles-pp.perf_exclude_event
      0.22 +-  2%      -0.1        0.13 +- 70%  perf-profile.children.cycles-pp._cond_resched
      0.17            -0.1        0.11 +- 70%  perf-profile.children.cycles-pp.page_rmapping
      0.13            -0.1        0.07 +- 70%  perf-profile.children.cycles-pp.rcu_all_qs
      0.07            -0.0        0.04 +- 70%  perf-profile.children.cycles-pp.ftrace_lookup_ip
     22.36            -7.8       14.59 +- 70%  perf-profile.self.cycles-pp.testcase
      8.08            -2.8        5.31 +- 70%  perf-profile.self.cycles-pp.native_irq_return_iret
      6.08            -2.1        3.96 +- 70%  perf-profile.self.cycles-pp.sync_regs
      5.81            -2.0        3.84 +- 70%  perf-profile.self.cycles-pp.swapgs_restore_regs_and_return_to_usermode
      3.27            -1.6        1.65 +- 70%  perf-profile.self.cycles-pp.__handle_mm_fault
      3.79            -1.4        2.36 +- 70%  perf-profile.self.cycles-pp.find_get_entry
      3.80            -1.3        2.53 +- 70%  perf-profile.self.cycles-pp.trace_graph_entry
      1.10            -0.5        0.57 +- 70%  perf-profile.self.cycles-pp.alloc_set_pte
      1.24            -0.4        0.81 +- 70%  perf-profile.self.cycles-pp.shmem_fault
      0.80            -0.3        0.50 +- 70%  perf-profile.self.cycles-pp._raw_spin_lock
      0.81            -0.3        0.51 +- 70%  perf-profile.self.cycles-pp.find_lock_entry
      0.80 +-  2%      -0.3        0.51 +- 70%  perf-profile.self.cycles-pp.__perf_sw_event
      0.61            -0.2        0.38 +- 70%  perf-profile.self.cycles-pp.down_read_trylock
      0.60            -0.2        0.39 +- 70%  perf-profile.self.cycles-pp.shmem_getpage_gfp
      0.48            -0.2        0.27 +- 70%  perf-profile.self.cycles-pp.pmd_devmap_trans_unstable
      0.47            -0.2        0.30 +- 70%  perf-profile.self.cycles-pp.file_update_time
      0.34            -0.1        0.22 +- 70%  perf-profile.self.cycles-pp.do_page_fault
      0.22 +-  4%      -0.1        0.11 +- 70%  perf-profile.self.cycles-pp.__do_fault
      0.25 +-  5%      -0.1        0.14 +- 71%  perf-profile.self.cycles-pp.ktime_get
      0.21 +-  2%      -0.1        0.12 +- 70%  perf-profile.self.cycles-pp.finish_fault
      0.23 +-  2%      -0.1        0.14 +- 70%  perf-profile.self.cycles-pp.fault_dirty_shared_page
      0.22 +-  2%      -0.1        0.14 +- 70%  perf-profile.self.cycles-pp.prepare_exit_to_usermode
      0.20 +-  2%      -0.1        0.12 +- 70%  perf-profile.self.cycles-pp.perf_exclude_event
      0.16            -0.1        0.10 +- 70%  perf-profile.self.cycles-pp._cond_resched
      0.13            -0.1        0.07 +- 70%  perf-profile.self.cycles-pp.rcu_all_qs
      0.07            -0.0        0.04 +- 70%  perf-profile.self.cycles-pp.ftrace_lookup_ip

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/always/context_switch1/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
           :3           33%           1:3     dmesg.WARNING:at#for_ip_interrupt_entry/0x
           :3           33%           1:3     dmesg.WARNING:at#for_ip_ret_from_intr/0x
           :3           67%           2:3     kmsg.pstore:crypto_comp_decompress_failed,ret=
           :3           67%           2:3     kmsg.pstore:decompression_failed
         %stddev     %change         %stddev
             \          |                \  
    223910            -1.3%     220930        will-it-scale.per_process_ops
    233722            -1.0%     231288        will-it-scale.per_thread_ops
 6.001e+08 +- 13%     +31.4%  7.887e+08 +-  4%  will-it-scale.time.involuntary_context_switches
     18003 +-  4%     +10.9%      19956        will-it-scale.time.minor_page_faults
  1.29e+10            -2.5%  1.258e+10        will-it-scale.time.voluntary_context_switches
  87865617            -1.2%   86826277        will-it-scale.workload
   2880329 +-  2%      +5.4%    3034904        interrupts.CAL:Function_call_interrupts
   7695018           -23.3%    5905066 +-  8%  meminfo.DirectMap2M
      0.00 +- 39%      -0.0        0.00 +- 78%  mpstat.cpu.iowait%
      4621 +- 12%     +13.4%       5241        proc-vmstat.numa_hint_faults_local
    715714           +27.6%     913142 +- 13%  softirqs.SCHED
    515653 +-  6%     -20.0%     412650 +- 15%  turbostat.C1
  43643516            -1.2%   43127031        vmstat.system.cs
   2893393 +-  4%     -23.6%    2210524 +- 10%  cpuidle.C1.time
    518051 +-  6%     -19.9%     415081 +- 15%  cpuidle.C1.usage
     23.10           +22.9%      28.38 +-  9%  boot-time.boot
     18.38           +23.2%      22.64 +- 12%  boot-time.dhcp
      5216            +5.0%       5478 +-  2%  boot-time.idle
    963.76 +- 44%    +109.7%       2021 +- 34%  irq_exception_noise.__do_page_fault.sum
      6.33 +- 14%    +726.3%      52.33 +- 62%  irq_exception_noise.irq_time
     56524 +-  7%     -18.8%      45915 +-  4%  irq_exception_noise.softirq_time
 6.001e+08 +- 13%     +31.4%  7.887e+08 +-  4%  time.involuntary_context_switches
     18003 +-  4%     +10.9%      19956        time.minor_page_faults
  1.29e+10            -2.5%  1.258e+10        time.voluntary_context_switches
      1386 +-  7%     +15.4%       1600 +- 11%  slabinfo.scsi_sense_cache.active_objs
      1386 +-  7%     +15.4%       1600 +- 11%  slabinfo.scsi_sense_cache.num_objs
      1427 +-  5%      -8.9%       1299 +-  2%  slabinfo.task_group.active_objs
      1427 +-  5%      -8.9%       1299 +-  2%  slabinfo.task_group.num_objs
     65519 +- 12%     +20.6%      79014 +- 16%  numa-meminfo.node0.SUnreclaim
      8484           -11.9%       7475 +-  7%  numa-meminfo.node1.KernelStack
      9264 +- 26%     -33.7%       6146 +-  7%  numa-meminfo.node1.Mapped
      2138 +- 61%    +373.5%      10127 +- 92%  numa-meminfo.node3.Inactive
      2059 +- 61%    +387.8%      10046 +- 93%  numa-meminfo.node3.Inactive(anon)
     16379 +- 12%     +20.6%      19752 +- 16%  numa-vmstat.node0.nr_slab_unreclaimable
      8483           -11.9%       7474 +-  7%  numa-vmstat.node1.nr_kernel_stack
      6250 +- 29%     -42.8%       3575 +- 24%  numa-vmstat.node2
      3798 +- 17%     +63.7%       6218 +-  5%  numa-vmstat.node3
    543.00 +- 61%    +368.1%       2541 +- 91%  numa-vmstat.node3.nr_inactive_anon
    543.33 +- 61%    +367.8%       2541 +- 91%  numa-vmstat.node3.nr_zone_inactive_anon
 4.138e+13            -1.1%   4.09e+13        perf-stat.branch-instructions
 6.569e+11            -2.0%  6.441e+11        perf-stat.branch-misses
 2.645e+10            -1.2%  2.613e+10        perf-stat.context-switches
      1.21            +1.2%       1.23        perf-stat.cpi
    153343 +-  2%     -12.1%     134776        perf-stat.cpu-migrations
 5.966e+13            -1.3%  5.889e+13        perf-stat.dTLB-loads
 3.736e+13            -1.2%   3.69e+13        perf-stat.dTLB-stores
      5.85 +- 15%      +8.8       14.67 +-  9%  perf-stat.iTLB-load-miss-rate%
 3.736e+09 +- 17%    +161.3%   9.76e+09 +- 11%  perf-stat.iTLB-load-misses
 5.987e+10            -5.4%  5.667e+10        perf-stat.iTLB-loads
 2.079e+14            -1.2%  2.054e+14        perf-stat.instructions
     57547 +- 18%     -62.9%      21340 +- 11%  perf-stat.instructions-per-iTLB-miss
      0.82            -1.2%       0.81        perf-stat.ipc
  27502531 +-  8%      +9.5%   30122136 +-  3%  perf-stat.node-store-misses
      1449 +- 27%     -34.6%     948.85        sched_debug.cfs_rq:/.load.min
    319416 +-115%    -188.5%    -282549        sched_debug.cfs_rq:/.spread0.avg
    657044 +- 55%     -88.3%      76887 +- 23%  sched_debug.cfs_rq:/.spread0.max
  -1525243           +54.6%   -2357898        sched_debug.cfs_rq:/.spread0.min
    101614 +-  6%     +30.6%     132713 +- 19%  sched_debug.cpu.avg_idle.stddev
     11.54 +- 41%     -61.2%       4.48        sched_debug.cpu.cpu_load[1].avg
      1369 +- 67%     -98.5%      20.67 +- 48%  sched_debug.cpu.cpu_load[1].max
     99.29 +- 67%     -97.6%       2.35 +- 26%  sched_debug.cpu.cpu_load[1].stddev
      9.58 +- 38%     -55.2%       4.29        sched_debug.cpu.cpu_load[2].avg
      1024 +- 68%     -98.5%      15.27 +- 36%  sched_debug.cpu.cpu_load[2].max
     74.51 +- 67%     -97.3%       1.99 +- 15%  sched_debug.cpu.cpu_load[2].stddev
      7.37 +- 29%     -42.0%       4.28        sched_debug.cpu.cpu_load[3].avg
    600.58 +- 68%     -97.9%      12.48 +- 20%  sched_debug.cpu.cpu_load[3].max
     43.98 +- 66%     -95.8%       1.83 +-  5%  sched_debug.cpu.cpu_load[3].stddev
      5.95 +- 19%     -28.1%       4.28        sched_debug.cpu.cpu_load[4].avg
    325.39 +- 67%     -96.4%      11.67 +- 10%  sched_debug.cpu.cpu_load[4].max
     24.19 +- 65%     -92.5%       1.81 +-  3%  sched_debug.cpu.cpu_load[4].stddev
    907.23 +-  4%     -14.1%     779.70 +- 10%  sched_debug.cpu.nr_load_updates.stddev
      0.00 +- 83%    +122.5%       0.00        sched_debug.rt_rq:/.rt_time.min
      8.49 +-  2%      -0.3        8.21 +-  2%  perf-profile.calltrace.cycles-pp.dequeue_task_fair.__schedule.schedule.pipe_wait.pipe_read
     57.28            -0.3       57.01        perf-profile.calltrace.cycles-pp.read
      5.06            -0.2        4.85        perf-profile.calltrace.cycles-pp.select_task_rq_fair.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_common_lock
      4.98            -0.2        4.78        perf-profile.calltrace.cycles-pp.__switch_to.read
      3.55            -0.2        3.39 +-  2%  perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.read
      2.72            -0.1        2.60        perf-profile.calltrace.cycles-pp.reweight_entity.enqueue_task_fair.ttwu_do_activate.try_to_wake_up.autoremove_wake_function
      2.67            -0.1        2.57 +-  2%  perf-profile.calltrace.cycles-pp.reweight_entity.dequeue_task_fair.__schedule.schedule.pipe_wait
      3.40            -0.1        3.31        perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.write
      3.77            -0.1        3.68        perf-profile.calltrace.cycles-pp.select_idle_sibling.select_task_rq_fair.try_to_wake_up.autoremove_wake_function.__wake_up_common
      1.95            -0.1        1.88        perf-profile.calltrace.cycles-pp.copy_page_to_iter.pipe_read.__vfs_read.vfs_read.ksys_read
      2.19            -0.1        2.13        perf-profile.calltrace.cycles-pp.__switch_to_asm.read
      1.30            -0.1        1.25        perf-profile.calltrace.cycles-pp.update_curr.reweight_entity.enqueue_task_fair.ttwu_do_activate.try_to_wake_up
      1.27            -0.1        1.22 +-  2%  perf-profile.calltrace.cycles-pp.update_curr.reweight_entity.dequeue_task_fair.__schedule.schedule
      2.29            -0.0        2.24        perf-profile.calltrace.cycles-pp.load_new_mm_cr3.switch_mm_irqs_off.__schedule.schedule.pipe_wait
      0.96            -0.0        0.92        perf-profile.calltrace.cycles-pp.__calc_delta.update_curr.reweight_entity.dequeue_task_fair.__schedule
      0.85            -0.0        0.81 +-  3%  perf-profile.calltrace.cycles-pp.cpumask_next_wrap.select_idle_sibling.select_task_rq_fair.try_to_wake_up.autoremove_wake_function
      1.63            -0.0        1.59        perf-profile.calltrace.cycles-pp.native_write_msr.read
      0.72            -0.0        0.69        perf-profile.calltrace.cycles-pp.copyout.copy_page_to_iter.pipe_read.__vfs_read.vfs_read
      0.65 +-  2%      -0.0        0.62        perf-profile.calltrace.cycles-pp._raw_spin_lock_irqsave.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_common_lock
      0.61            -0.0        0.58 +-  2%  perf-profile.calltrace.cycles-pp.find_next_bit.cpumask_next_wrap.select_idle_sibling.select_task_rq_fair.try_to_wake_up
      0.88            -0.0        0.85        perf-profile.calltrace.cycles-pp.touch_atime.pipe_read.__vfs_read.vfs_read.ksys_read
      0.80            -0.0        0.77 +-  2%  perf-profile.calltrace.cycles-pp.___perf_sw_event.__schedule.schedule.pipe_wait.pipe_read
      0.82            -0.0        0.79        perf-profile.calltrace.cycles-pp.prepare_to_wait.pipe_wait.pipe_read.__vfs_read.vfs_read
      0.72            -0.0        0.70        perf-profile.calltrace.cycles-pp.mutex_lock.pipe_write.__vfs_write.vfs_write.ksys_write
      0.56 +-  2%      -0.0        0.53        perf-profile.calltrace.cycles-pp.update_rq_clock.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_common_lock
      0.83            -0.0        0.81        perf-profile.calltrace.cycles-pp.__wake_up_common_lock.pipe_read.__vfs_read.vfs_read.ksys_read
     42.40            +0.3       42.69        perf-profile.calltrace.cycles-pp.write
     31.80            +0.4       32.18        perf-profile.calltrace.cycles-pp.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
     24.35            +0.5       24.84        perf-profile.calltrace.cycles-pp.pipe_wait.pipe_read.__vfs_read.vfs_read.ksys_read
     20.36            +0.6       20.92 +-  2%  perf-profile.calltrace.cycles-pp.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_common_lock.pipe_write
     22.01            +0.6       22.58        perf-profile.calltrace.cycles-pp.schedule.pipe_wait.pipe_read.__vfs_read.vfs_read
     21.87            +0.6       22.46        perf-profile.calltrace.cycles-pp.__schedule.schedule.pipe_wait.pipe_read.__vfs_read
      3.15 +- 11%      +1.0        4.12 +- 14%  perf-profile.calltrace.cycles-pp.ttwu_do_wakeup.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_common_lock
      1.07 +- 34%      +1.1        2.12 +- 31%  perf-profile.calltrace.cycles-pp.tracing_record_taskinfo_sched_switch.__schedule.schedule.pipe_wait.pipe_read
      0.66 +- 75%      +1.1        1.72 +- 37%  perf-profile.calltrace.cycles-pp.trace_save_cmdline.tracing_record_taskinfo.ttwu_do_wakeup.try_to_wake_up.autoremove_wake_function
      0.75 +- 74%      +1.1        1.88 +- 34%  perf-profile.calltrace.cycles-pp.tracing_record_taskinfo.ttwu_do_wakeup.try_to_wake_up.autoremove_wake_function.__wake_up_common
      0.69 +- 76%      +1.2        1.85 +- 36%  perf-profile.calltrace.cycles-pp.trace_save_cmdline.tracing_record_taskinfo_sched_switch.__schedule.schedule.pipe_wait
      8.73 +-  2%      -0.3        8.45        perf-profile.children.cycles-pp.dequeue_task_fair
     57.28            -0.3       57.01        perf-profile.children.cycles-pp.read
      6.95            -0.2        6.70        perf-profile.children.cycles-pp.syscall_return_via_sysret
      5.57            -0.2        5.35        perf-profile.children.cycles-pp.reweight_entity
      5.26            -0.2        5.05        perf-profile.children.cycles-pp.select_task_rq_fair
      5.19            -0.2        4.99        perf-profile.children.cycles-pp.__switch_to
      4.90            -0.2        4.73 +-  2%  perf-profile.children.cycles-pp.update_curr
      1.27            -0.1        1.13 +-  8%  perf-profile.children.cycles-pp.fsnotify
      3.92            -0.1        3.83        perf-profile.children.cycles-pp.select_idle_sibling
      2.01            -0.1        1.93        perf-profile.children.cycles-pp.__calc_delta
      2.14            -0.1        2.06        perf-profile.children.cycles-pp.copy_page_to_iter
      1.58            -0.1        1.51        perf-profile.children.cycles-pp._raw_spin_unlock_irqrestore
      2.90            -0.1        2.84        perf-profile.children.cycles-pp.update_cfs_group
      1.93            -0.1        1.87        perf-profile.children.cycles-pp._raw_spin_lock_irqsave
      2.35            -0.1        2.29        perf-profile.children.cycles-pp.__switch_to_asm
      1.33            -0.1        1.27 +-  3%  perf-profile.children.cycles-pp.cpumask_next_wrap
      2.57            -0.1        2.52        perf-profile.children.cycles-pp.load_new_mm_cr3
      1.53            -0.1        1.47 +-  2%  perf-profile.children.cycles-pp.__fdget_pos
      1.11            -0.0        1.07 +-  2%  perf-profile.children.cycles-pp.find_next_bit
      1.18            -0.0        1.14        perf-profile.children.cycles-pp.update_rq_clock
      0.88            -0.0        0.83        perf-profile.children.cycles-pp.copy_user_generic_unrolled
      1.70            -0.0        1.65        perf-profile.children.cycles-pp.native_write_msr
      0.97            -0.0        0.93 +-  2%  perf-profile.children.cycles-pp.account_entity_dequeue
      0.59            -0.0        0.56        perf-profile.children.cycles-pp.finish_task_switch
      0.91            -0.0        0.88        perf-profile.children.cycles-pp.touch_atime
      0.69            -0.0        0.65        perf-profile.children.cycles-pp.account_entity_enqueue
      2.13            -0.0        2.09        perf-profile.children.cycles-pp.mutex_lock
      0.32 +-  3%      -0.0        0.29 +-  4%  perf-profile.children.cycles-pp.__sb_start_write
      0.84            -0.0        0.81 +-  2%  perf-profile.children.cycles-pp.___perf_sw_event
      0.89            -0.0        0.87        perf-profile.children.cycles-pp.prepare_to_wait
      0.73            -0.0        0.71        perf-profile.children.cycles-pp.copyout
      0.31 +-  2%      -0.0        0.28 +-  3%  perf-profile.children.cycles-pp.__list_del_entry_valid
      0.46 +-  2%      -0.0        0.44        perf-profile.children.cycles-pp.anon_pipe_buf_release
      0.38            -0.0        0.36 +-  3%  perf-profile.children.cycles-pp.idle_cpu
      0.32            -0.0        0.30 +-  2%  perf-profile.children.cycles-pp.__x64_sys_read
      0.21 +-  2%      -0.0        0.20 +-  2%  perf-profile.children.cycles-pp.deactivate_task
      0.13            -0.0        0.12 +-  4%  perf-profile.children.cycles-pp.timespec_trunc
      0.09            -0.0        0.08        perf-profile.children.cycles-pp.iov_iter_init
      0.08            -0.0        0.07        perf-profile.children.cycles-pp.native_load_tls
      0.11 +-  4%      +0.0        0.12        perf-profile.children.cycles-pp.tick_sched_timer
      0.08 +-  5%      +0.0        0.10 +-  4%  perf-profile.children.cycles-pp.finish_wait
      0.38 +-  2%      +0.0        0.40 +-  2%  perf-profile.children.cycles-pp.file_update_time
      0.31            +0.0        0.33 +-  2%  perf-profile.children.cycles-pp.smp_apic_timer_interrupt
      0.24 +-  3%      +0.0        0.26 +-  3%  perf-profile.children.cycles-pp.rcu_all_qs
      0.39            +0.0        0.41        perf-profile.children.cycles-pp._cond_resched
      0.05            +0.0        0.07 +-  6%  perf-profile.children.cycles-pp.default_wake_function
      0.23 +-  2%      +0.0        0.26 +-  3%  perf-profile.children.cycles-pp.current_time
      0.30            +0.0        0.35 +-  2%  perf-profile.children.cycles-pp.generic_pipe_buf_confirm
      0.52            +0.1        0.58        perf-profile.children.cycles-pp.entry_SYSCALL_64_stage2
      0.00            +0.1        0.08 +-  5%  perf-profile.children.cycles-pp.hrtick_update
     42.40            +0.3       42.69        perf-profile.children.cycles-pp.write
     31.86            +0.4       32.26        perf-profile.children.cycles-pp.__vfs_read
     24.40            +0.5       24.89        perf-profile.children.cycles-pp.pipe_wait
     20.40            +0.6       20.96 +-  2%  perf-profile.children.cycles-pp.try_to_wake_up
     22.30            +0.6       22.89        perf-profile.children.cycles-pp.schedule
     22.22            +0.6       22.84        perf-profile.children.cycles-pp.__schedule
      0.99 +- 36%      +0.9        1.94 +- 32%  perf-profile.children.cycles-pp.tracing_record_taskinfo
      3.30 +- 10%      +1.0        4.27 +- 13%  perf-profile.children.cycles-pp.ttwu_do_wakeup
      1.14 +- 31%      +1.1        2.24 +- 29%  perf-profile.children.cycles-pp.tracing_record_taskinfo_sched_switch
      1.59 +- 46%      +2.0        3.60 +- 36%  perf-profile.children.cycles-pp.trace_save_cmdline
      6.95            -0.2        6.70        perf-profile.self.cycles-pp.syscall_return_via_sysret
      5.19            -0.2        4.99        perf-profile.self.cycles-pp.__switch_to
      1.27            -0.1        1.12 +-  8%  perf-profile.self.cycles-pp.fsnotify
      1.49            -0.1        1.36        perf-profile.self.cycles-pp.select_task_rq_fair
      2.47            -0.1        2.37 +-  2%  perf-profile.self.cycles-pp.reweight_entity
      0.29            -0.1        0.19 +-  2%  perf-profile.self.cycles-pp.ksys_read
      1.50            -0.1        1.42        perf-profile.self.cycles-pp._raw_spin_unlock_irqrestore
      2.01            -0.1        1.93        perf-profile.self.cycles-pp.__calc_delta
      1.93            -0.1        1.86        perf-profile.self.cycles-pp._raw_spin_lock_irqsave
      1.47            -0.1        1.40        perf-profile.self.cycles-pp.dequeue_task_fair
      2.90            -0.1        2.84        perf-profile.self.cycles-pp.update_cfs_group
      1.29            -0.1        1.23        perf-profile.self.cycles-pp.do_syscall_64
      2.57            -0.1        2.52        perf-profile.self.cycles-pp.load_new_mm_cr3
      2.28            -0.1        2.23        perf-profile.self.cycles-pp.__switch_to_asm
      1.80            -0.1        1.75        perf-profile.self.cycles-pp.select_idle_sibling
      1.11            -0.0        1.07 +-  2%  perf-profile.self.cycles-pp.find_next_bit
      0.87            -0.0        0.83        perf-profile.self.cycles-pp.copy_user_generic_unrolled
      0.43            -0.0        0.39 +-  2%  perf-profile.self.cycles-pp.dequeue_entity
      1.70            -0.0        1.65        perf-profile.self.cycles-pp.native_write_msr
      0.92            -0.0        0.88 +-  2%  perf-profile.self.cycles-pp.account_entity_dequeue
      0.48            -0.0        0.44        perf-profile.self.cycles-pp.finish_task_switch
      0.77            -0.0        0.74        perf-profile.self.cycles-pp.___perf_sw_event
      0.66            -0.0        0.63        perf-profile.self.cycles-pp.account_entity_enqueue
      0.46 +-  2%      -0.0        0.43 +-  2%  perf-profile.self.cycles-pp.anon_pipe_buf_release
      0.32 +-  3%      -0.0        0.29 +-  4%  perf-profile.self.cycles-pp.__sb_start_write
      0.31 +-  2%      -0.0        0.28 +-  3%  perf-profile.self.cycles-pp.__list_del_entry_valid
      0.38            -0.0        0.36 +-  3%  perf-profile.self.cycles-pp.idle_cpu
      0.19 +-  4%      -0.0        0.17 +-  2%  perf-profile.self.cycles-pp.__fdget_pos
      0.50            -0.0        0.48        perf-profile.self.cycles-pp.__atime_needs_update
      0.23 +-  2%      -0.0        0.21 +-  3%  perf-profile.self.cycles-pp.touch_atime
      0.31            -0.0        0.30        perf-profile.self.cycles-pp.__x64_sys_read
      0.21 +-  2%      -0.0        0.20 +-  2%  perf-profile.self.cycles-pp.deactivate_task
      0.21 +-  2%      -0.0        0.19        perf-profile.self.cycles-pp.check_preempt_curr
      0.40            -0.0        0.39        perf-profile.self.cycles-pp.autoremove_wake_function
      0.40            -0.0        0.38        perf-profile.self.cycles-pp.copy_user_enhanced_fast_string
      0.27            -0.0        0.26        perf-profile.self.cycles-pp.pipe_wait
      0.13            -0.0        0.12 +-  4%  perf-profile.self.cycles-pp.timespec_trunc
      0.22 +-  2%      -0.0        0.20 +-  2%  perf-profile.self.cycles-pp.put_prev_entity
      0.09            -0.0        0.08        perf-profile.self.cycles-pp.iov_iter_init
      0.08            -0.0        0.07        perf-profile.self.cycles-pp.native_load_tls
      0.11            -0.0        0.10        perf-profile.self.cycles-pp.schedule
      0.12 +-  4%      +0.0        0.13        perf-profile.self.cycles-pp.copyin
      0.08 +-  5%      +0.0        0.10 +-  4%  perf-profile.self.cycles-pp.finish_wait
      0.18            +0.0        0.20 +-  2%  perf-profile.self.cycles-pp.ttwu_do_activate
      0.28 +-  2%      +0.0        0.30 +-  2%  perf-profile.self.cycles-pp._cond_resched
      0.24 +-  3%      +0.0        0.26 +-  3%  perf-profile.self.cycles-pp.rcu_all_qs
      0.05            +0.0        0.07 +-  6%  perf-profile.self.cycles-pp.default_wake_function
      0.08 +- 14%      +0.0        0.11 +- 14%  perf-profile.self.cycles-pp.tracing_record_taskinfo_sched_switch
      0.51            +0.0        0.55 +-  4%  perf-profile.self.cycles-pp.vfs_write
      0.30            +0.0        0.35 +-  2%  perf-profile.self.cycles-pp.generic_pipe_buf_confirm
      0.52            +0.1        0.58        perf-profile.self.cycles-pp.entry_SYSCALL_64_stage2
      0.00            +0.1        0.08 +-  5%  perf-profile.self.cycles-pp.hrtick_update
      1.97            +0.1        2.07 +-  2%  perf-profile.self.cycles-pp.switch_mm_irqs_off
      1.59 +- 46%      +2.0        3.60 +- 36%  perf-profile.self.cycles-pp.trace_save_cmdline

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/never/brk1/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
           :3           33%           1:3     kmsg.pstore:crypto_comp_decompress_failed,ret=
           :3           33%           1:3     kmsg.pstore:decompression_failed
         %stddev     %change         %stddev
             \          |                \  
    997317            -2.0%     977778        will-it-scale.per_process_ops
    957.00            -7.9%     881.00 +-  3%  will-it-scale.per_thread_ops
     18.42 +-  3%      -8.2%      16.90        will-it-scale.time.user_time
 1.917e+08            -2.0%  1.879e+08        will-it-scale.workload
     18.42 +-  3%      -8.2%      16.90        time.user_time
      0.30 +- 11%     -36.7%       0.19 +- 11%  turbostat.Pkg%pc2
     57539 +- 51%    +140.6%     138439 +- 31%  meminfo.CmaFree
    410877 +- 11%     -22.1%     320082 +- 22%  meminfo.DirectMap4k
    343575 +- 27%     +71.3%     588703 +- 31%  numa-numastat.node0.local_node
    374176 +- 24%     +63.3%     611007 +- 27%  numa-numastat.node0.numa_hit
   1056347 +-  4%     -39.9%     634843 +- 38%  numa-numastat.node3.local_node
   1060682 +-  4%     -39.0%     646862 +- 35%  numa-numastat.node3.numa_hit
     14383 +- 51%    +140.6%      34608 +- 31%  proc-vmstat.nr_free_cma
    179.00            +2.4%     183.33        proc-vmstat.nr_inactive_file
    179.00            +2.4%     183.33        proc-vmstat.nr_zone_inactive_file
    564483 +-  3%     -38.0%     350064 +- 36%  proc-vmstat.pgalloc_movable
   1811959           +10.8%    2008488 +-  5%  proc-vmstat.pgalloc_normal
      7153 +- 42%     -94.0%     431.33 +-119%  latency_stats.max.pipe_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      6627 +-141%    +380.5%      31843 +-110%  latency_stats.max.call_rwsem_down_write_failed_killable.do_mprotect_pkey.__x64_sys_mprotect.do_syscall_64.entry_SYSCALL_64_after_hwframe
     15244 +- 31%     -99.9%      15.00 +-141%  latency_stats.sum.call_rwsem_down_read_failed.__do_page_fault.do_page_fault.page_fault.__get_user_8.exit_robust_list.mm_release.do_exit.do_group_exit.get_signal.do_signal.exit_to_usermode_loop
      4301 +-117%     -83.7%     700.33 +-  6%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
     12153 +- 28%     -83.1%       2056 +- 70%  latency_stats.sum.pipe_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      6772 +-141%   +1105.8%      81665 +-127%  latency_stats.sum.call_rwsem_down_write_failed_killable.do_mprotect_pkey.__x64_sys_mprotect.do_syscall_64.entry_SYSCALL_64_after_hwframe
 2.465e+13            -1.3%  2.434e+13        perf-stat.branch-instructions
 2.691e+11            -2.1%  2.635e+11        perf-stat.branch-misses
 3.402e+13            -1.4%  3.355e+13        perf-stat.dTLB-loads
 1.694e+13            +1.4%  1.718e+13        perf-stat.dTLB-stores
      1.75 +- 50%      +4.7        6.45 +- 11%  perf-stat.iTLB-load-miss-rate%
 4.077e+08 +- 48%    +232.3%  1.355e+09 +- 11%  perf-stat.iTLB-load-misses
  2.31e+10 +-  2%     -14.9%  1.965e+10 +-  3%  perf-stat.iTLB-loads
 1.163e+14            -1.6%  1.144e+14        perf-stat.instructions
    346171 +- 36%     -75.3%      85575 +- 11%  perf-stat.instructions-per-iTLB-miss
 6.174e+08 +-  2%      -9.5%  5.589e+08        perf-stat.node-store-misses
    595.00 +- 10%     +31.4%     782.00 +-  3%  slabinfo.Acpi-State.active_objs
    595.00 +- 10%     +31.4%     782.00 +-  3%  slabinfo.Acpi-State.num_objs
      2831 +-  3%     -14.0%       2434 +-  5%  slabinfo.avtab_node.active_objs
      2831 +-  3%     -14.0%       2434 +-  5%  slabinfo.avtab_node.num_objs
    934.00           -10.9%     832.33 +-  5%  slabinfo.inotify_inode_mark.active_objs
    934.00           -10.9%     832.33 +-  5%  slabinfo.inotify_inode_mark.num_objs
      1232 +-  4%     +13.4%       1397 +-  6%  slabinfo.nsproxy.active_objs
      1232 +-  4%     +13.4%       1397 +-  6%  slabinfo.nsproxy.num_objs
    499.67 +- 12%     +24.8%     623.67 +- 10%  slabinfo.secpath_cache.active_objs
    499.67 +- 12%     +24.8%     623.67 +- 10%  slabinfo.secpath_cache.num_objs
     31393 +- 84%    +220.1%     100477 +- 21%  numa-meminfo.node0.Active
     31393 +- 84%    +220.1%     100477 +- 21%  numa-meminfo.node0.Active(anon)
     30013 +- 85%    +232.1%      99661 +- 21%  numa-meminfo.node0.AnonPages
     21603 +- 34%     -85.0%       3237 +-100%  numa-meminfo.node0.Inactive
     21528 +- 34%     -85.0%       3237 +-100%  numa-meminfo.node0.Inactive(anon)
     10247 +- 35%     -46.4%       5495        numa-meminfo.node0.Mapped
     35388 +- 14%     -41.6%      20670 +- 15%  numa-meminfo.node0.SReclaimable
     22911 +- 29%     -82.3%       4057 +- 84%  numa-meminfo.node0.Shmem
    117387 +-  9%     -22.5%      90986 +- 12%  numa-meminfo.node0.Slab
     68863 +- 67%     +77.7%     122351 +- 13%  numa-meminfo.node1.Active
     68863 +- 67%     +77.7%     122351 +- 13%  numa-meminfo.node1.Active(anon)
    228376           +22.3%     279406 +- 17%  numa-meminfo.node1.FilePages
      1481 +-116%   +1062.1%      17218 +- 39%  numa-meminfo.node1.Inactive
      1481 +-116%   +1062.0%      17216 +- 39%  numa-meminfo.node1.Inactive(anon)
      6593 +-  2%     +11.7%       7367 +-  3%  numa-meminfo.node1.KernelStack
    596227 +-  8%     +18.0%     703748 +-  4%  numa-meminfo.node1.MemUsed
     15298 +- 12%     +88.5%      28843 +- 36%  numa-meminfo.node1.SReclaimable
     52718 +-  9%     +21.0%      63810 +- 11%  numa-meminfo.node1.SUnreclaim
      1808 +- 97%   +2723.8%      51054 +- 97%  numa-meminfo.node1.Shmem
     68017 +-  5%     +36.2%      92654 +- 18%  numa-meminfo.node1.Slab
    125541 +- 29%     -64.9%      44024 +- 98%  numa-meminfo.node3.Active
    125137 +- 29%     -65.0%      43823 +- 98%  numa-meminfo.node3.Active(anon)
     93173 +- 25%     -87.8%      11381 +- 20%  numa-meminfo.node3.AnonPages
      9150 +-  5%      -9.3%       8301 +-  8%  numa-meminfo.node3.KernelStack
      7848 +- 84%    +220.0%      25118 +- 21%  numa-vmstat.node0.nr_active_anon
      7503 +- 85%    +232.1%      24914 +- 21%  numa-vmstat.node0.nr_anon_pages
      5381 +- 34%     -85.0%     809.00 +-100%  numa-vmstat.node0.nr_inactive_anon
      2559 +- 35%     -46.4%       1372        numa-vmstat.node0.nr_mapped
      5727 +- 29%     -82.3%       1014 +- 84%  numa-vmstat.node0.nr_shmem
      8846 +- 14%     -41.6%       5167 +- 15%  numa-vmstat.node0.nr_slab_reclaimable
      7848 +- 84%    +220.0%      25118 +- 21%  numa-vmstat.node0.nr_zone_active_anon
      5381 +- 34%     -85.0%     809.00 +-100%  numa-vmstat.node0.nr_zone_inactive_anon
      4821 +-  2%     +30.3%       6283 +- 15%  numa-vmstat.node1
     17215 +- 67%     +77.7%      30591 +- 13%  numa-vmstat.node1.nr_active_anon
     57093           +22.3%      69850 +- 17%  numa-vmstat.node1.nr_file_pages
    370.00 +-116%   +1061.8%       4298 +- 39%  numa-vmstat.node1.nr_inactive_anon
      6593 +-  2%     +11.7%       7366 +-  3%  numa-vmstat.node1.nr_kernel_stack
    451.67 +- 97%   +2725.6%      12762 +- 97%  numa-vmstat.node1.nr_shmem
      3824 +- 12%     +88.6%       7211 +- 36%  numa-vmstat.node1.nr_slab_reclaimable
     13179 +-  9%     +21.0%      15952 +- 11%  numa-vmstat.node1.nr_slab_unreclaimable
     17215 +- 67%     +77.7%      30591 +- 13%  numa-vmstat.node1.nr_zone_active_anon
    370.00 +-116%   +1061.8%       4298 +- 39%  numa-vmstat.node1.nr_zone_inactive_anon
    364789 +- 12%     +62.8%     593926 +- 34%  numa-vmstat.node1.numa_hit
    239539 +- 19%     +95.4%     468113 +- 43%  numa-vmstat.node1.numa_local
     71.00 +- 28%     +42.3%     101.00        numa-vmstat.node2.nr_mlock
     31285 +- 29%     -65.0%      10960 +- 98%  numa-vmstat.node3.nr_active_anon
     23292 +- 25%     -87.8%       2844 +- 19%  numa-vmstat.node3.nr_anon_pages
     14339 +- 52%    +141.1%      34566 +- 32%  numa-vmstat.node3.nr_free_cma
      9151 +-  5%      -9.3%       8299 +-  8%  numa-vmstat.node3.nr_kernel_stack
     31305 +- 29%     -64.9%      10975 +- 98%  numa-vmstat.node3.nr_zone_active_anon
    930131 +-  3%     -35.9%     596006 +- 34%  numa-vmstat.node3.numa_hit
    836455 +-  3%     -40.9%     493947 +- 44%  numa-vmstat.node3.numa_local
     75182 +- 58%     -83.8%      12160 +-  2%  sched_debug.cfs_rq:/.load.max
      6.65 +-  5%     -10.6%       5.94 +-  6%  sched_debug.cfs_rq:/.load_avg.avg
      0.16 +-  7%     +22.6%       0.20 +- 12%  sched_debug.cfs_rq:/.nr_running.stddev
      5.58 +- 24%    +427.7%      29.42 +- 93%  sched_debug.cfs_rq:/.nr_spread_over.max
      0.54 +- 15%    +306.8%       2.19 +- 86%  sched_debug.cfs_rq:/.nr_spread_over.stddev
      1.05 +- 25%     -65.1%       0.37 +- 71%  sched_debug.cfs_rq:/.removed.load_avg.avg
      9.62 +- 11%     -50.7%       4.74 +- 70%  sched_debug.cfs_rq:/.removed.load_avg.stddev
     48.70 +- 25%     -65.1%      17.02 +- 71%  sched_debug.cfs_rq:/.removed.runnable_sum.avg
    444.31 +- 11%     -50.7%     219.26 +- 70%  sched_debug.cfs_rq:/.removed.runnable_sum.stddev
      0.47 +- 13%     -60.9%       0.19 +- 71%  sched_debug.cfs_rq:/.removed.util_avg.avg
      4.47 +-  4%     -46.5%       2.39 +- 70%  sched_debug.cfs_rq:/.removed.util_avg.stddev
      1.64 +-  7%     +22.1%       2.00 +- 13%  sched_debug.cfs_rq:/.runnable_load_avg.stddev
     74653 +- 59%     -84.4%      11676        sched_debug.cfs_rq:/.runnable_weight.max
   -119169          -491.3%     466350 +- 27%  sched_debug.cfs_rq:/.spread0.avg
    517161 +- 30%    +145.8%    1271292 +- 23%  sched_debug.cfs_rq:/.spread0.max
    624.79 +-  5%     -14.2%     535.76 +-  7%  sched_debug.cfs_rq:/.util_est_enqueued.avg
    247.91 +- 32%     -99.8%       0.48 +-  8%  sched_debug.cfs_rq:/.util_est_enqueued.min
    179704 +-  3%     +30.4%     234297 +- 16%  sched_debug.cpu.avg_idle.stddev
      1.56 +-  9%     +24.4%       1.94 +- 14%  sched_debug.cpu.cpu_load[0].stddev
      1.50 +-  6%     +27.7%       1.91 +- 14%  sched_debug.cpu.cpu_load[1].stddev
      1.45 +-  3%     +30.8%       1.90 +- 14%  sched_debug.cpu.cpu_load[2].stddev
      1.43 +-  3%     +36.1%       1.95 +- 11%  sched_debug.cpu.cpu_load[3].stddev
      1.55 +-  7%     +43.5%       2.22 +-  7%  sched_debug.cpu.cpu_load[4].stddev
     10004 +-  3%     -11.6%       8839 +-  3%  sched_debug.cpu.curr->pid.avg
      1146 +- 26%     +52.2%       1745 +-  7%  sched_debug.cpu.curr->pid.min
      3162 +-  6%     +25.4%       3966 +- 11%  sched_debug.cpu.curr->pid.stddev
    403738 +-  3%     -11.7%     356696 +-  7%  sched_debug.cpu.nr_switches.max
      0.08 +- 21%     +78.2%       0.14 +- 14%  sched_debug.cpu.nr_uninterruptible.avg
    404435 +-  3%     -11.8%     356732 +-  7%  sched_debug.cpu.sched_count.max
      4.17            -0.3        3.87        perf-profile.calltrace.cycles-pp.kmem_cache_alloc.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      2.40            -0.2        2.17        perf-profile.calltrace.cycles-pp.vma_compute_subtree_gap.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk
      7.58            -0.2        7.36        perf-profile.calltrace.cycles-pp.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     15.00            -0.2       14.81        perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.brk
      7.83            -0.2        7.66        perf-profile.calltrace.cycles-pp.unmap_vmas.unmap_region.do_munmap.__x64_sys_brk.do_syscall_64
     28.66            -0.1       28.51        perf-profile.calltrace.cycles-pp.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      2.15            -0.1        2.03        perf-profile.calltrace.cycles-pp.vma_compute_subtree_gap.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.07            -0.1        0.99        perf-profile.calltrace.cycles-pp.memcpy_erms.strlcpy.perf_event_mmap.do_brk_flags.__x64_sys_brk
      1.03            -0.1        0.95        perf-profile.calltrace.cycles-pp.kmem_cache_free.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64
      7.33            -0.1        7.25        perf-profile.calltrace.cycles-pp.unmap_page_range.unmap_vmas.unmap_region.do_munmap.__x64_sys_brk
      0.76            -0.1        0.69        perf-profile.calltrace.cycles-pp.__vm_enough_memory.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     11.85            -0.1       11.77        perf-profile.calltrace.cycles-pp.unmap_region.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.64            -0.1        1.57        perf-profile.calltrace.cycles-pp.strlcpy.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64
      1.06            -0.1        0.99        perf-profile.calltrace.cycles-pp.__indirect_thunk_start.brk
      0.73            -0.1        0.67        perf-profile.calltrace.cycles-pp.sync_mm_rss.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      4.59            -0.1        4.52        perf-profile.calltrace.cycles-pp.security_vm_enough_memory_mm.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      2.82            -0.1        2.76        perf-profile.calltrace.cycles-pp.selinux_vm_enough_memory.security_vm_enough_memory_mm.do_brk_flags.__x64_sys_brk.do_syscall_64
      2.89            -0.1        2.84        perf-profile.calltrace.cycles-pp.down_write_killable.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      3.37            -0.1        3.32        perf-profile.calltrace.cycles-pp.get_unmapped_area.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.99            -0.0        1.94        perf-profile.calltrace.cycles-pp.cred_has_capability.selinux_vm_enough_memory.security_vm_enough_memory_mm.do_brk_flags.__x64_sys_brk
      2.32            -0.0        2.27        perf-profile.calltrace.cycles-pp.perf_iterate_sb.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64
      1.88            -0.0        1.84        perf-profile.calltrace.cycles-pp.security_mmap_addr.get_unmapped_area.do_brk_flags.__x64_sys_brk.do_syscall_64
      0.77            -0.0        0.73        perf-profile.calltrace.cycles-pp._raw_spin_lock.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      1.62            -0.0        1.59        perf-profile.calltrace.cycles-pp.memset_erms.kmem_cache_alloc.do_brk_flags.__x64_sys_brk.do_syscall_64
      0.81            -0.0        0.79        perf-profile.calltrace.cycles-pp.___might_sleep.down_write_killable.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.66            -0.0        0.64        perf-profile.calltrace.cycles-pp.arch_get_unmapped_area_topdown.brk
      0.72            +0.0        0.74        perf-profile.calltrace.cycles-pp.do_munmap.brk
      0.90            +0.0        0.93        perf-profile.calltrace.cycles-pp.___might_sleep.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      4.40            +0.1        4.47        perf-profile.calltrace.cycles-pp.find_vma.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.96            +0.1        2.09        perf-profile.calltrace.cycles-pp.vmacache_find.find_vma.do_munmap.__x64_sys_brk.do_syscall_64
      0.52 +-  2%      +0.2        0.68        perf-profile.calltrace.cycles-pp.__vma_link_rb.brk
      0.35 +- 70%      +0.2        0.54 +-  2%  perf-profile.calltrace.cycles-pp.find_vma.brk
      2.20            +0.3        2.50        perf-profile.calltrace.cycles-pp.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     64.62            +0.3       64.94        perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.brk
     60.53            +0.4       60.92        perf-profile.calltrace.cycles-pp.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
     63.20            +0.4       63.60        perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      3.73            +0.5        4.26        perf-profile.calltrace.cycles-pp.vma_link.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00            +0.6        0.56        perf-profile.calltrace.cycles-pp.free_pgtables.unmap_region.do_munmap.__x64_sys_brk.do_syscall_64
     24.54            +0.6       25.14        perf-profile.calltrace.cycles-pp.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      0.00            +0.6        0.64        perf-profile.calltrace.cycles-pp.put_vma.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64
      0.71            +0.6        1.36        perf-profile.calltrace.cycles-pp.__vma_rb_erase.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00            +0.7        0.70        perf-profile.calltrace.cycles-pp._raw_write_lock.__vma_rb_erase.do_munmap.__x64_sys_brk.do_syscall_64
      3.10            +0.7        3.82        perf-profile.calltrace.cycles-pp.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk.do_syscall_64
      0.00            +0.8        0.76        perf-profile.calltrace.cycles-pp._raw_write_lock.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk
      0.00            +0.8        0.85        perf-profile.calltrace.cycles-pp.__vma_merge.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      5.09            -0.5        4.62        perf-profile.children.cycles-pp.vma_compute_subtree_gap
      4.54            -0.3        4.21        perf-profile.children.cycles-pp.kmem_cache_alloc
      8.11            -0.2        7.89        perf-profile.children.cycles-pp.perf_event_mmap
      8.05            -0.2        7.85        perf-profile.children.cycles-pp.unmap_vmas
     15.01            -0.2       14.81        perf-profile.children.cycles-pp.syscall_return_via_sysret
     29.20            -0.1       29.06        perf-profile.children.cycles-pp.do_brk_flags
      1.11            -0.1        1.00        perf-profile.children.cycles-pp.kmem_cache_free
     12.28            -0.1       12.17        perf-profile.children.cycles-pp.unmap_region
      7.83            -0.1        7.74        perf-profile.children.cycles-pp.unmap_page_range
      0.87 +-  3%      -0.1        0.79        perf-profile.children.cycles-pp.__vm_enough_memory
      1.29            -0.1        1.22        perf-profile.children.cycles-pp.__indirect_thunk_start
      1.81            -0.1        1.74        perf-profile.children.cycles-pp.strlcpy
      4.65            -0.1        4.58        perf-profile.children.cycles-pp.security_vm_enough_memory_mm
      3.08            -0.1        3.02        perf-profile.children.cycles-pp.down_write_killable
      2.88            -0.1        2.82        perf-profile.children.cycles-pp.selinux_vm_enough_memory
      0.73            -0.1        0.67        perf-profile.children.cycles-pp.sync_mm_rss
      3.65            -0.1        3.59        perf-profile.children.cycles-pp.get_unmapped_area
      2.26            -0.1        2.20        perf-profile.children.cycles-pp.cred_has_capability
      1.12            -0.1        1.07        perf-profile.children.cycles-pp.memcpy_erms
      0.39            -0.0        0.35        perf-profile.children.cycles-pp.__rb_insert_augmented
      2.52            -0.0        2.48        perf-profile.children.cycles-pp.perf_iterate_sb
      2.13            -0.0        2.09        perf-profile.children.cycles-pp.security_mmap_addr
      0.55 +-  2%      -0.0        0.52        perf-profile.children.cycles-pp.unmap_single_vma
      1.62            -0.0        1.59        perf-profile.children.cycles-pp.memset_erms
      0.13 +-  3%      -0.0        0.11 +-  4%  perf-profile.children.cycles-pp.__vma_link_file
      0.80            -0.0        0.77        perf-profile.children.cycles-pp._raw_spin_lock
      0.43            -0.0        0.41        perf-profile.children.cycles-pp.strlen
      0.07 +-  6%      -0.0        0.06 +-  8%  perf-profile.children.cycles-pp.should_failslab
      0.43            -0.0        0.42        perf-profile.children.cycles-pp.may_expand_vm
      0.15            +0.0        0.16        perf-profile.children.cycles-pp.__vma_link_list
      0.45            +0.0        0.47        perf-profile.children.cycles-pp.rcu_all_qs
      0.81            +0.1        0.89        perf-profile.children.cycles-pp.free_pgtables
      6.35            +0.1        6.49        perf-profile.children.cycles-pp.find_vma
      2.28            +0.2        2.45        perf-profile.children.cycles-pp.vmacache_find
     64.66            +0.3       64.98        perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
      2.42            +0.3        2.76        perf-profile.children.cycles-pp.remove_vma
     61.77            +0.4       62.13        perf-profile.children.cycles-pp.__x64_sys_brk
     63.40            +0.4       63.79        perf-profile.children.cycles-pp.do_syscall_64
      1.27            +0.4        1.72        perf-profile.children.cycles-pp.__vma_rb_erase
      4.02            +0.5        4.53        perf-profile.children.cycles-pp.vma_link
     25.26            +0.6       25.89        perf-profile.children.cycles-pp.do_munmap
      0.00            +0.7        0.70        perf-profile.children.cycles-pp.put_vma
      3.80            +0.7        4.53        perf-profile.children.cycles-pp.__vma_link_rb
      0.00            +1.2        1.24        perf-profile.children.cycles-pp.__vma_merge
      0.00            +1.5        1.51        perf-profile.children.cycles-pp._raw_write_lock
      5.07            -0.5        4.60        perf-profile.self.cycles-pp.vma_compute_subtree_gap
      0.59            -0.2        0.38        perf-profile.self.cycles-pp.remove_vma
     15.01            -0.2       14.81        perf-profile.self.cycles-pp.syscall_return_via_sysret
      3.15            -0.2        2.96        perf-profile.self.cycles-pp.do_munmap
      0.98            -0.1        0.87        perf-profile.self.cycles-pp.__vma_rb_erase
      1.10            -0.1        0.99        perf-profile.self.cycles-pp.kmem_cache_free
      0.68            -0.1        0.58        perf-profile.self.cycles-pp.__vm_enough_memory
      0.42            -0.1        0.33        perf-profile.self.cycles-pp.unmap_vmas
      3.62            -0.1        3.53        perf-profile.self.cycles-pp.perf_event_mmap
      1.41            -0.1        1.34        perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
      1.29            -0.1        1.22        perf-profile.self.cycles-pp.__indirect_thunk_start
      0.73            -0.1        0.66        perf-profile.self.cycles-pp.sync_mm_rss
      2.96            -0.1        2.90        perf-profile.self.cycles-pp.__x64_sys_brk
      3.24            -0.1        3.19        perf-profile.self.cycles-pp.brk
      1.11            -0.0        1.07        perf-profile.self.cycles-pp.memcpy_erms
      0.53 +-  3%      -0.0        0.49 +-  2%  perf-profile.self.cycles-pp.vma_link
      0.73            -0.0        0.69        perf-profile.self.cycles-pp.unmap_region
      1.66            -0.0        1.61        perf-profile.self.cycles-pp.down_write_killable
      0.39            -0.0        0.35        perf-profile.self.cycles-pp.__rb_insert_augmented
      1.74            -0.0        1.71        perf-profile.self.cycles-pp.kmem_cache_alloc
      0.55 +-  2%      -0.0        0.52        perf-profile.self.cycles-pp.unmap_single_vma
      1.61            -0.0        1.59        perf-profile.self.cycles-pp.memset_erms
      0.80            -0.0        0.77        perf-profile.self.cycles-pp._raw_spin_lock
      0.13            -0.0        0.11 +-  4%  perf-profile.self.cycles-pp.__vma_link_file
      0.43            -0.0        0.41        perf-profile.self.cycles-pp.strlen
      0.07 +-  6%      -0.0        0.06 +-  8%  perf-profile.self.cycles-pp.should_failslab
      0.81            -0.0        0.79        perf-profile.self.cycles-pp.tlb_finish_mmu
      0.15            +0.0        0.16        perf-profile.self.cycles-pp.__vma_link_list
      0.45            +0.0        0.47        perf-profile.self.cycles-pp.rcu_all_qs
      0.71            +0.0        0.72        perf-profile.self.cycles-pp.strlcpy
      0.51            +0.1        0.56        perf-profile.self.cycles-pp.free_pgtables
      1.41            +0.1        1.48        perf-profile.self.cycles-pp.__vma_link_rb
      2.27            +0.2        2.44        perf-profile.self.cycles-pp.vmacache_find
      0.00            +0.7        0.69        perf-profile.self.cycles-pp.put_vma
      0.00            +1.2        1.23        perf-profile.self.cycles-pp.__vma_merge
      0.00            +1.5        1.50        perf-profile.self.cycles-pp._raw_write_lock

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/always/brk1/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
           :3           33%           1:3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=schedule_tail/0x
           :3           33%           1:3     kmsg.DHCP/BOOTP:Reply_not_for_us_on_eth#,op[#]xid[#]
         %stddev     %change         %stddev
             \          |                \  
    998475            -2.2%     976893        will-it-scale.per_process_ops
    625.87            -2.3%     611.42        will-it-scale.time.elapsed_time
    625.87            -2.3%     611.42        will-it-scale.time.elapsed_time.max
      8158            -1.9%       8000        will-it-scale.time.maximum_resident_set_size
     18.42 +-  2%     -11.9%      16.24        will-it-scale.time.user_time
  34349225 +- 13%     -14.5%   29371024 +- 17%  will-it-scale.time.voluntary_context_switches
 1.919e+08            -2.2%  1.877e+08        will-it-scale.workload
      1639 +- 23%     -18.4%       1337 +- 30%  meminfo.Mlocked
     17748 +- 82%    +103.1%      36051        numa-numastat.node3.other_node
  33410486 +- 14%     -14.8%   28449258 +- 18%  cpuidle.C1.usage
    698749 +- 15%     -18.0%     573307 +- 20%  cpuidle.POLL.usage
   3013702 +- 14%     -15.1%    2559405 +- 17%  softirqs.SCHED
  54361293 +-  2%     -19.0%   44044816 +-  2%  softirqs.TIMER
  33408303 +- 14%     -14.9%   28447123 +- 18%  turbostat.C1
      0.34 +- 16%     -52.0%       0.16 +- 15%  turbostat.Pkg%pc2
      1310 +- 74%    +412.1%       6710 +- 58%  irq_exception_noise.__do_page_fault.samples
      3209 +- 74%    +281.9%      12258 +- 53%  irq_exception_noise.__do_page_fault.sum
    600.67 +-132%     -96.0%      24.00 +- 23%  irq_exception_noise.irq_nr
     99557 +-  7%     -24.0%      75627 +-  7%  irq_exception_noise.softirq_nr
     41424 +-  9%     -24.6%      31253 +-  6%  irq_exception_noise.softirq_time
    625.87            -2.3%     611.42        time.elapsed_time
    625.87            -2.3%     611.42        time.elapsed_time.max
      8158            -1.9%       8000        time.maximum_resident_set_size
     18.42 +-  2%     -11.9%      16.24        time.user_time
  34349225 +- 13%     -14.5%   29371024 +- 17%  time.voluntary_context_switches
    988.00 +-  8%     +14.5%       1131 +-  2%  slabinfo.Acpi-ParseExt.active_objs
    988.00 +-  8%     +14.5%       1131 +-  2%  slabinfo.Acpi-ParseExt.num_objs
      2384 +-  3%     +21.1%       2888 +- 11%  slabinfo.pool_workqueue.active_objs
      2474 +-  2%     +20.4%       2979 +- 11%  slabinfo.pool_workqueue.num_objs
    490.33 +- 10%     -19.2%     396.00 +- 11%  slabinfo.secpath_cache.active_objs
    490.33 +- 10%     -19.2%     396.00 +- 11%  slabinfo.secpath_cache.num_objs
      1123 +-  7%     +14.2%       1282 +-  3%  slabinfo.skbuff_fclone_cache.active_objs
      1123 +-  7%     +14.2%       1282 +-  3%  slabinfo.skbuff_fclone_cache.num_objs
      1.09            -0.0        1.07        perf-stat.branch-miss-rate%
 2.691e+11            -2.4%  2.628e+11        perf-stat.branch-misses
  71981351 +- 12%     -13.8%   62013509 +- 16%  perf-stat.context-switches
 1.697e+13            +1.1%  1.715e+13        perf-stat.dTLB-stores
      2.36 +- 29%      +4.4        6.76 +- 11%  perf-stat.iTLB-load-miss-rate%
  5.21e+08 +- 28%    +194.8%  1.536e+09 +- 10%  perf-stat.iTLB-load-misses
    239983 +- 24%     -68.4%      75819 +- 11%  perf-stat.instructions-per-iTLB-miss
   3295653 +-  2%      -6.3%    3088753 +-  3%  perf-stat.node-stores
    606239            +1.1%     612799        perf-stat.path-length
      3755 +- 28%     -37.5%       2346 +- 52%  sched_debug.cfs_rq:/.exec_clock.stddev
     10.45 +-  4%     +24.3%      12.98 +- 18%  sched_debug.cfs_rq:/.load_avg.stddev
      6243 +- 46%     -38.6%       3831 +- 78%  sched_debug.cpu.load.stddev
    867.80 +-  7%     +25.3%       1087 +-  6%  sched_debug.cpu.nr_load_updates.stddev
    395898 +-  3%     -11.1%     352071 +-  7%  sched_debug.cpu.nr_switches.max
    -13.33           -21.1%     -10.52        sched_debug.cpu.nr_uninterruptible.min
    395674 +-  3%     -11.1%     351762 +-  7%  sched_debug.cpu.sched_count.max
     33152 +-  4%     -12.8%      28899        sched_debug.cpu.ttwu_count.min
      0.03 +- 20%     +77.7%       0.05 +- 15%  sched_debug.rt_rq:/.rt_time.max
     89523            +1.8%      91099        proc-vmstat.nr_active_anon
    409.67 +- 23%     -18.4%     334.33 +- 30%  proc-vmstat.nr_mlock
     89530            +1.8%      91117        proc-vmstat.nr_zone_active_anon
   2337130            -2.2%    2286775        proc-vmstat.numa_hit
   2229090            -2.3%    2178626        proc-vmstat.numa_local
      8460 +- 39%     -75.5%       2076 +- 53%  proc-vmstat.numa_pages_migrated
     28643 +- 55%     -83.5%       4727 +- 58%  proc-vmstat.numa_pte_updates
   2695806            -1.8%    2646639        proc-vmstat.pgfault
   2330191            -2.1%    2281197        proc-vmstat.pgfree
      8460 +- 39%     -75.5%       2076 +- 53%  proc-vmstat.pgmigrate_success
    237651 +-  2%     +31.3%     312092 +- 16%  numa-meminfo.node0.FilePages
      8059 +-  2%     +10.7%       8925 +-  7%  numa-meminfo.node0.KernelStack
      6830 +- 25%     +48.8%      10164 +- 35%  numa-meminfo.node0.Mapped
      1612 +- 21%     +70.0%       2740 +- 19%  numa-meminfo.node0.PageTables
     10772 +- 65%    +679.4%      83962 +- 59%  numa-meminfo.node0.Shmem
    163195 +- 15%     -36.9%     103036 +- 32%  numa-meminfo.node1.Active
    163195 +- 15%     -36.9%     103036 +- 32%  numa-meminfo.node1.Active(anon)
      1730 +-  4%     +33.9%       2317 +- 14%  numa-meminfo.node1.PageTables
     55778 +- 19%     +32.5%      73910 +-  8%  numa-meminfo.node1.SUnreclaim
      2671 +- 16%     -45.0%       1469 +- 15%  numa-meminfo.node2.PageTables
     61537 +- 13%     -17.7%      50647 +-  3%  numa-meminfo.node2.SUnreclaim
     48644 +- 94%    +149.8%     121499 +- 11%  numa-meminfo.node3.Active
     48440 +- 94%    +150.4%     121295 +- 11%  numa-meminfo.node3.Active(anon)
     11832 +- 79%     -91.5%       1008 +- 67%  numa-meminfo.node3.Inactive
     11597 +- 82%     -93.3%     772.00 +- 82%  numa-meminfo.node3.Inactive(anon)
     10389 +- 32%     -43.0%       5921 +-  6%  numa-meminfo.node3.Mapped
     33704 +- 24%     -44.2%      18792 +- 15%  numa-meminfo.node3.SReclaimable
    104733 +- 14%     -25.3%      78275 +-  8%  numa-meminfo.node3.Slab
    139329 +-133%     -99.8%     241.67 +- 79%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
      5403 +-139%     -97.5%     137.67 +- 71%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
    165968 +-101%     -61.9%      63304 +- 58%  latency_stats.avg.max
     83.00        +12810.4%      10715 +-140%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_access.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat.filename_lookup
    102.67 +-  6%  +18845.5%      19450 +-140%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    136.33 +- 16%  +25043.5%      34279 +-141%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.__lookup_slow.lookup_slow.walk_component.path_lookupat.filename_lookup
     18497 +-141%    -100.0%       0.00        latency_stats.max.call_rwsem_down_write_failed_killable.vm_mmap_pgoff.ksys_mmap_pgoff.do_syscall_64.entry_SYSCALL_64_after_hwframe
    140500 +-131%     -99.8%     247.00 +- 78%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
      5403 +-139%     -97.5%     137.67 +- 71%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
     87.33 +-  5%  +23963.0%      21015 +-140%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_access.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat.filename_lookup
    136.33 +- 16%  +25043.5%      34279 +-141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.__lookup_slow.lookup_slow.walk_component.path_lookupat.filename_lookup
    149.33 +- 14%  +25485.9%      38208 +-141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
     18761 +-141%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_write_failed_killable.vm_mmap_pgoff.ksys_mmap_pgoff.do_syscall_64.entry_SYSCALL_64_after_hwframe
     23363 +-114%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_read_failed.__do_page_fault.do_page_fault.page_fault.__get_user_8.exit_robust_list.mm_release.do_exit.do_group_exit.get_signal.do_signal.exit_to_usermode_loop
    144810 +-125%     -99.8%     326.67 +- 70%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
      5403 +-139%     -97.5%     137.67 +- 71%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
     59698 +- 98%     -78.0%      13110 +-141%  latency_stats.sum.call_rwsem_down_read_failed.do_exit.do_group_exit.get_signal.do_signal.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
    166.33        +12768.5%      21404 +-140%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_access.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat.filename_lookup
    825.00 +-  6%  +18761.7%     155609 +-140%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    136.33 +- 16%  +25043.5%      34279 +-141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.__lookup_slow.lookup_slow.walk_component.path_lookupat.filename_lookup
     59412 +-  2%     +31.3%      78021 +- 16%  numa-vmstat.node0.nr_file_pages
      8059 +-  2%     +10.7%       8923 +-  7%  numa-vmstat.node0.nr_kernel_stack
      1701 +- 25%     +49.1%       2536 +- 35%  numa-vmstat.node0.nr_mapped
    402.33 +- 21%     +70.0%     684.00 +- 19%  numa-vmstat.node0.nr_page_table_pages
      2692 +- 65%    +679.5%      20988 +- 59%  numa-vmstat.node0.nr_shmem
    622587 +- 36%     +37.7%     857545 +- 13%  numa-vmstat.node0.numa_local
     40797 +- 15%     -36.9%      25757 +- 32%  numa-vmstat.node1.nr_active_anon
    432.00 +-  4%     +33.9%     578.33 +- 14%  numa-vmstat.node1.nr_page_table_pages
     13944 +- 19%     +32.5%      18477 +-  8%  numa-vmstat.node1.nr_slab_unreclaimable
     40797 +- 15%     -36.9%      25757 +- 32%  numa-vmstat.node1.nr_zone_active_anon
    625073 +- 26%     +29.4%     808657 +- 18%  numa-vmstat.node1.numa_hit
    503969 +- 34%     +39.2%     701446 +- 23%  numa-vmstat.node1.numa_local
    137.33 +- 40%     -49.0%      70.00 +- 29%  numa-vmstat.node2.nr_mlock
    667.67 +- 17%     -45.1%     366.33 +- 15%  numa-vmstat.node2.nr_page_table_pages
     15384 +- 13%     -17.7%      12662 +-  3%  numa-vmstat.node2.nr_slab_unreclaimable
     12114 +- 94%    +150.3%      30326 +- 11%  numa-vmstat.node3.nr_active_anon
      2887 +- 83%     -93.4%     190.00 +- 82%  numa-vmstat.node3.nr_inactive_anon
      2632 +- 30%     -39.2%       1600 +-  5%  numa-vmstat.node3.nr_mapped
    101.00           -30.0%      70.67 +- 29%  numa-vmstat.node3.nr_mlock
      8425 +- 24%     -44.2%       4697 +- 15%  numa-vmstat.node3.nr_slab_reclaimable
     12122 +- 94%    +150.3%      30346 +- 11%  numa-vmstat.node3.nr_zone_active_anon
      2887 +- 83%     -93.4%     190.00 +- 82%  numa-vmstat.node3.nr_zone_inactive_anon
    106945 +- 13%     +17.4%     125554        numa-vmstat.node3.numa_other
      4.17            -0.3        3.82        perf-profile.calltrace.cycles-pp.kmem_cache_alloc.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     15.02            -0.3       14.77        perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.brk
      2.42            -0.2        2.18        perf-profile.calltrace.cycles-pp.vma_compute_subtree_gap.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk
      7.60            -0.2        7.39        perf-profile.calltrace.cycles-pp.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      7.79            -0.2        7.63        perf-profile.calltrace.cycles-pp.unmap_vmas.unmap_region.do_munmap.__x64_sys_brk.do_syscall_64
      0.82 +-  9%      -0.1        0.68        perf-profile.calltrace.cycles-pp.__vm_enough_memory.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      2.13            -0.1        2.00        perf-profile.calltrace.cycles-pp.vma_compute_subtree_gap.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.05            -0.1        0.95        perf-profile.calltrace.cycles-pp.kmem_cache_free.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64
      7.31            -0.1        7.21        perf-profile.calltrace.cycles-pp.unmap_page_range.unmap_vmas.unmap_region.do_munmap.__x64_sys_brk
      0.74            -0.1        0.67        perf-profile.calltrace.cycles-pp.sync_mm_rss.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      1.06            -0.1        1.00        perf-profile.calltrace.cycles-pp.memcpy_erms.strlcpy.perf_event_mmap.do_brk_flags.__x64_sys_brk
      3.38            -0.1        3.33        perf-profile.calltrace.cycles-pp.get_unmapped_area.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.05            -0.0        1.00 +-  2%  perf-profile.calltrace.cycles-pp.__indirect_thunk_start.brk
      2.34            -0.0        2.29        perf-profile.calltrace.cycles-pp.perf_iterate_sb.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64
      1.64            -0.0        1.59        perf-profile.calltrace.cycles-pp.strlcpy.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64
      1.89            -0.0        1.86        perf-profile.calltrace.cycles-pp.security_mmap_addr.get_unmapped_area.do_brk_flags.__x64_sys_brk.do_syscall_64
      0.76            -0.0        0.73        perf-profile.calltrace.cycles-pp._raw_spin_lock.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      0.57 +-  2%      -0.0        0.55        perf-profile.calltrace.cycles-pp.selinux_mmap_addr.security_mmap_addr.get_unmapped_area.do_brk_flags.__x64_sys_brk
      0.54 +-  2%      +0.0        0.56        perf-profile.calltrace.cycles-pp.do_brk_flags.brk
      0.72            +0.0        0.76 +-  2%  perf-profile.calltrace.cycles-pp.do_munmap.brk
      4.38            +0.1        4.43        perf-profile.calltrace.cycles-pp.find_vma.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.96            +0.1        2.04        perf-profile.calltrace.cycles-pp.vmacache_find.find_vma.do_munmap.__x64_sys_brk.do_syscall_64
      0.53            +0.2        0.68        perf-profile.calltrace.cycles-pp.__vma_link_rb.brk
      2.21            +0.3        2.51        perf-profile.calltrace.cycles-pp.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     64.44            +0.5       64.90        perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.brk
     63.04            +0.5       63.54        perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
     60.37            +0.5       60.88        perf-profile.calltrace.cycles-pp.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      3.75            +0.5        4.29        perf-profile.calltrace.cycles-pp.vma_link.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00            +0.6        0.57        perf-profile.calltrace.cycles-pp.free_pgtables.unmap_region.do_munmap.__x64_sys_brk.do_syscall_64
      0.00            +0.6        0.64        perf-profile.calltrace.cycles-pp.put_vma.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64
      0.72            +0.7        1.37        perf-profile.calltrace.cycles-pp.__vma_rb_erase.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     24.42            +0.7       25.08        perf-profile.calltrace.cycles-pp.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      0.00            +0.7        0.71        perf-profile.calltrace.cycles-pp._raw_write_lock.__vma_rb_erase.do_munmap.__x64_sys_brk.do_syscall_64
      3.12            +0.7        3.84        perf-profile.calltrace.cycles-pp.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk.do_syscall_64
      0.00            +0.8        0.77        perf-profile.calltrace.cycles-pp._raw_write_lock.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk
      0.00            +0.9        0.85        perf-profile.calltrace.cycles-pp.__vma_merge.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      5.10            -0.5        4.60        perf-profile.children.cycles-pp.vma_compute_subtree_gap
      4.53            -0.3        4.18        perf-profile.children.cycles-pp.kmem_cache_alloc
     15.03            -0.3       14.77        perf-profile.children.cycles-pp.syscall_return_via_sysret
      8.13            -0.2        7.92        perf-profile.children.cycles-pp.perf_event_mmap
      8.01            -0.2        7.81        perf-profile.children.cycles-pp.unmap_vmas
      0.97 +- 14%      -0.2        0.78        perf-profile.children.cycles-pp.__vm_enough_memory
      1.13            -0.1        1.00        perf-profile.children.cycles-pp.kmem_cache_free
      7.82            -0.1        7.70        perf-profile.children.cycles-pp.unmap_page_range
     12.23            -0.1       12.13        perf-profile.children.cycles-pp.unmap_region
      0.74            -0.1        0.67        perf-profile.children.cycles-pp.sync_mm_rss
      3.06            -0.1        3.00        perf-profile.children.cycles-pp.down_write_killable
      0.40 +-  2%      -0.1        0.34        perf-profile.children.cycles-pp.__rb_insert_augmented
      1.29            -0.1        1.23        perf-profile.children.cycles-pp.__indirect_thunk_start
      2.54            -0.1        2.49        perf-profile.children.cycles-pp.perf_iterate_sb
      3.66            -0.0        3.61        perf-profile.children.cycles-pp.get_unmapped_area
      1.80            -0.0        1.75        perf-profile.children.cycles-pp.strlcpy
      0.53 +-  2%      -0.0        0.49 +-  2%  perf-profile.children.cycles-pp.cap_capable
      1.57            -0.0        1.53        perf-profile.children.cycles-pp.arch_get_unmapped_area_topdown
      1.11            -0.0        1.08        perf-profile.children.cycles-pp.memcpy_erms
      0.13            -0.0        0.10        perf-profile.children.cycles-pp.__vma_link_file
      0.55            -0.0        0.52        perf-profile.children.cycles-pp.unmap_single_vma
      1.47            -0.0        1.44        perf-profile.children.cycles-pp.cap_vm_enough_memory
      2.14            -0.0        2.12        perf-profile.children.cycles-pp.security_mmap_addr
      0.32            -0.0        0.30        perf-profile.children.cycles-pp.userfaultfd_unmap_complete
      1.25            -0.0        1.23        perf-profile.children.cycles-pp.up_write
      0.50            -0.0        0.49        perf-profile.children.cycles-pp.userfaultfd_unmap_prep
      0.27            -0.0        0.26        perf-profile.children.cycles-pp.tlb_flush_mmu_free
      1.14            -0.0        1.12        perf-profile.children.cycles-pp.__might_sleep
      0.07            -0.0        0.06        perf-profile.children.cycles-pp.should_failslab
      0.72            +0.0        0.74        perf-profile.children.cycles-pp._cond_resched
      0.45            +0.0        0.47        perf-profile.children.cycles-pp.rcu_all_qs
      0.15 +-  3%      +0.0        0.17 +-  4%  perf-profile.children.cycles-pp.__vma_link_list
      0.15 +-  5%      +0.0        0.18 +-  5%  perf-profile.children.cycles-pp.tick_sched_timer
      0.05 +-  8%      +0.1        0.12 +- 17%  perf-profile.children.cycles-pp.perf_mux_hrtimer_handler
      0.80            +0.1        0.89        perf-profile.children.cycles-pp.free_pgtables
      0.22 +-  7%      +0.1        0.31 +-  9%  perf-profile.children.cycles-pp.__hrtimer_run_queues
      0.00            +0.1        0.11 +- 15%  perf-profile.children.cycles-pp.clockevents_program_event
      6.34            +0.1        6.47        perf-profile.children.cycles-pp.find_vma
      2.27            +0.1        2.40        perf-profile.children.cycles-pp.vmacache_find
      0.40 +-  4%      +0.2        0.58 +-  5%  perf-profile.children.cycles-pp.apic_timer_interrupt
      0.40 +-  4%      +0.2        0.58 +-  5%  perf-profile.children.cycles-pp.smp_apic_timer_interrupt
      0.37 +-  4%      +0.2        0.54 +-  5%  perf-profile.children.cycles-pp.hrtimer_interrupt
      0.00            +0.2        0.19 +- 12%  perf-profile.children.cycles-pp.ktime_get
      2.42            +0.3        2.77        perf-profile.children.cycles-pp.remove_vma
     64.49            +0.5       64.94        perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
      1.27            +0.5        1.73        perf-profile.children.cycles-pp.__vma_rb_erase
     61.62            +0.5       62.10        perf-profile.children.cycles-pp.__x64_sys_brk
     63.24            +0.5       63.74        perf-profile.children.cycles-pp.do_syscall_64
      4.03            +0.5        4.56        perf-profile.children.cycles-pp.vma_link
      0.00            +0.7        0.69        perf-profile.children.cycles-pp.put_vma
     25.13            +0.7       25.84        perf-profile.children.cycles-pp.do_munmap
      3.83            +0.7        4.56        perf-profile.children.cycles-pp.__vma_link_rb
      0.00            +1.2        1.25        perf-profile.children.cycles-pp.__vma_merge
      0.00            +1.5        1.53        perf-profile.children.cycles-pp._raw_write_lock
      5.08            -0.5        4.58        perf-profile.self.cycles-pp.vma_compute_subtree_gap
     15.03            -0.3       14.77        perf-profile.self.cycles-pp.syscall_return_via_sysret
      0.59            -0.2        0.39        perf-profile.self.cycles-pp.remove_vma
      0.72 +-  7%      -0.1        0.58        perf-profile.self.cycles-pp.__vm_enough_memory
      1.12            -0.1        0.99        perf-profile.self.cycles-pp.kmem_cache_free
      3.11            -0.1        2.99        perf-profile.self.cycles-pp.do_munmap
      0.99            -0.1        0.88        perf-profile.self.cycles-pp.__vma_rb_erase
      3.63            -0.1        3.52        perf-profile.self.cycles-pp.perf_event_mmap
      3.26            -0.1        3.17        perf-profile.self.cycles-pp.brk
      0.41 +-  2%      -0.1        0.33        perf-profile.self.cycles-pp.unmap_vmas
      0.74            -0.1        0.67        perf-profile.self.cycles-pp.sync_mm_rss
      1.75            -0.1        1.68        perf-profile.self.cycles-pp.kmem_cache_alloc
      0.40 +-  2%      -0.1        0.34        perf-profile.self.cycles-pp.__rb_insert_augmented
      1.29 +-  2%      -0.1        1.23        perf-profile.self.cycles-pp.__indirect_thunk_start
      0.73            -0.0        0.68 +-  2%  perf-profile.self.cycles-pp.unmap_region
      0.53            -0.0        0.49        perf-profile.self.cycles-pp.vma_link
      1.40            -0.0        1.35        perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
      5.22            -0.0        5.18        perf-profile.self.cycles-pp.unmap_page_range
      0.53 +-  2%      -0.0        0.49 +-  2%  perf-profile.self.cycles-pp.cap_capable
      1.11            -0.0        1.07        perf-profile.self.cycles-pp.memcpy_erms
      1.86            -0.0        1.82        perf-profile.self.cycles-pp.perf_iterate_sb
      1.30            -0.0        1.27        perf-profile.self.cycles-pp.arch_get_unmapped_area_topdown
      0.13            -0.0        0.10        perf-profile.self.cycles-pp.__vma_link_file
      0.55            -0.0        0.52        perf-profile.self.cycles-pp.unmap_single_vma
      0.74            -0.0        0.72        perf-profile.self.cycles-pp.selinux_mmap_addr
      0.32            -0.0        0.30        perf-profile.self.cycles-pp.userfaultfd_unmap_complete
      1.13            -0.0        1.12        perf-profile.self.cycles-pp.__might_sleep
      1.24            -0.0        1.23        perf-profile.self.cycles-pp.up_write
      0.50            -0.0        0.49        perf-profile.self.cycles-pp.userfaultfd_unmap_prep
      0.27            -0.0        0.26        perf-profile.self.cycles-pp.tlb_flush_mmu_free
      0.07            -0.0        0.06        perf-profile.self.cycles-pp.should_failslab
      0.45            +0.0        0.47        perf-profile.self.cycles-pp.rcu_all_qs
      0.71            +0.0        0.73        perf-profile.self.cycles-pp.strlcpy
      0.15 +-  3%      +0.0        0.17 +-  4%  perf-profile.self.cycles-pp.__vma_link_list
      0.51            +0.1        0.57        perf-profile.self.cycles-pp.free_pgtables
      1.40            +0.1        1.49        perf-profile.self.cycles-pp.__vma_link_rb
      2.27            +0.1        2.39        perf-profile.self.cycles-pp.vmacache_find
      0.00            +0.2        0.18 +- 12%  perf-profile.self.cycles-pp.ktime_get
      0.00            +0.7        0.69        perf-profile.self.cycles-pp.put_vma
      0.00            +1.2        1.24        perf-profile.self.cycles-pp.__vma_merge
      0.00            +1.5        1.52        perf-profile.self.cycles-pp._raw_write_lock

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/always/page_fault2/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
           :3           33%           1:3     dmesg.WARNING:at#for_ip_native_iret/0x
          1:3          -33%            :3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=__schedule/0x
           :3           33%           1:3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=__slab_free/0x
          1:3          -33%            :3     kmsg.DHCP/BOOTP:Reply_not_for_us_on_eth#,op[#]xid[#]
          3:3         -100%            :3     kmsg.pstore:crypto_comp_decompress_failed,ret=
          3:3         -100%            :3     kmsg.pstore:decompression_failed
          2:3            4%           2:3     perf-profile.calltrace.cycles-pp.sync_regs.error_entry
          5:3            7%           5:3     perf-profile.calltrace.cycles-pp.error_entry
          5:3            7%           5:3     perf-profile.children.cycles-pp.error_entry
          2:3            3%           2:3     perf-profile.self.cycles-pp.error_entry
         %stddev     %change         %stddev
             \          |                \  
      8281 +-  2%     -18.8%       6728        will-it-scale.per_thread_ops
     92778 +-  2%     +17.6%     109080        will-it-scale.time.involuntary_context_switches
  21954366 +-  3%      +4.1%   22857988 +-  2%  will-it-scale.time.maximum_resident_set_size
  4.81e+08 +-  2%     -18.9%  3.899e+08        will-it-scale.time.minor_page_faults
      5804           +12.2%       6512        will-it-scale.time.percent_of_cpu_this_job_got
     34918           +12.2%      39193        will-it-scale.time.system_time
   5638528 +-  2%     -15.3%    4778392        will-it-scale.time.voluntary_context_switches
  15846405            -2.0%   15531034        will-it-scale.workload
   2818137            +1.5%    2861500        interrupts.CAL:Function_call_interrupts
      3.33 +- 28%     -60.0%       1.33 +- 93%  irq_exception_noise.irq_time
      2866           +23.9%       3552 +-  2%  kthread_noise.total_time
   5589674 +- 14%     +31.4%    7344810 +-  6%  meminfo.DirectMap2M
     31169           -16.9%      25906        uptime.idle
     25242 +-  4%     -14.2%      21654 +-  6%  vmstat.system.cs
      7055           -11.6%       6237        boot-time.idle
     21.12           +19.3%      25.19 +-  9%  boot-time.kernel_boot
     20.03 +-  2%      -3.7       16.38        mpstat.cpu.idle%
      0.00 +-  8%      -0.0        0.00 +-  4%  mpstat.cpu.iowait%
   7284147 +-  2%     -16.4%    6092495        softirqs.RCU
   5350756 +-  2%     -10.9%    4769417 +-  4%  softirqs.SCHED
     42933 +- 21%     -28.2%      30807 +-  7%  numa-meminfo.node2.SReclaimable
     63219 +- 13%     -16.6%      52717 +-  6%  numa-meminfo.node2.SUnreclaim
    106153 +- 16%     -21.3%      83525 +-  5%  numa-meminfo.node2.Slab
    247154 +-  4%      -7.6%     228415        numa-meminfo.node3.Unevictable
     11904 +-  4%     +17.1%      13945 +-  8%  numa-vmstat.node0
      2239 +- 22%     -26.6%       1644 +-  2%  numa-vmstat.node2.nr_mapped
     10728 +- 21%     -28.2%       7701 +-  7%  numa-vmstat.node2.nr_slab_reclaimable
     15803 +- 13%     -16.6%      13179 +-  6%  numa-vmstat.node2.nr_slab_unreclaimable
     61788 +-  4%      -7.6%      57103        numa-vmstat.node3.nr_unevictable
     61788 +-  4%      -7.6%      57103        numa-vmstat.node3.nr_zone_unevictable
     92778 +-  2%     +17.6%     109080        time.involuntary_context_switches
  21954366 +-  3%      +4.1%   22857988 +-  2%  time.maximum_resident_set_size
  4.81e+08 +-  2%     -18.9%  3.899e+08        time.minor_page_faults
      5804           +12.2%       6512        time.percent_of_cpu_this_job_got
     34918           +12.2%      39193        time.system_time
   5638528 +-  2%     -15.3%    4778392        time.voluntary_context_switches
   3942289 +-  2%     -10.5%    3528902 +-  2%  cpuidle.C1.time
    242290           -14.2%     207992        cpuidle.C1.usage
  1.64e+09 +-  2%     -15.7%  1.381e+09        cpuidle.C1E.time
   4621281 +-  2%     -14.7%    3939757        cpuidle.C1E.usage
 2.115e+10 +-  2%     -18.5%  1.723e+10        cpuidle.C6.time
  24771099 +-  2%     -18.0%   20305766        cpuidle.C6.usage
   1210810 +-  4%     -17.6%     997270 +-  2%  cpuidle.POLL.time
     18742 +-  3%     -17.0%      15559 +-  2%  cpuidle.POLL.usage
      4135 +-141%    -100.0%       0.00        latency_stats.avg.x86_reserve_hardware.x86_pmu_event_init.perf_try_init_event.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
     33249 +-129%    -100.0%       0.00        latency_stats.max.call_rwsem_down_read_failed.m_start.seq_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4135 +-141%    -100.0%       0.00        latency_stats.max.x86_reserve_hardware.x86_pmu_event_init.perf_try_init_event.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
     65839 +-116%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_read_failed.m_start.seq_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4135 +-141%    -100.0%       0.00        latency_stats.sum.x86_reserve_hardware.x86_pmu_event_init.perf_try_init_event.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      8387 +-122%     -90.9%     767.00 +- 13%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    263970 +- 10%     -68.6%      82994 +-  3%  latency_stats.sum.do_syslog.kmsg_read.proc_reg_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      6173 +- 77%    +173.3%      16869 +- 98%  latency_stats.sum.pipe_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
    101.33            -4.6%      96.67        proc-vmstat.nr_anon_transparent_hugepages
     39967            -1.8%      39241        proc-vmstat.nr_slab_reclaimable
     67166            -2.4%      65522        proc-vmstat.nr_slab_unreclaimable
    237743            -3.9%     228396        proc-vmstat.nr_unevictable
    237743            -3.9%     228396        proc-vmstat.nr_zone_unevictable
 4.807e+09            -2.0%   4.71e+09        proc-vmstat.numa_hit
 4.807e+09            -2.0%   4.71e+09        proc-vmstat.numa_local
 4.791e+09            -2.1%   4.69e+09        proc-vmstat.pgalloc_normal
 4.783e+09            -2.0%  4.685e+09        proc-vmstat.pgfault
 4.807e+09            -2.0%  4.709e+09        proc-vmstat.pgfree
      1753            +4.6%       1833        turbostat.Avg_MHz
    239445           -14.1%     205783        turbostat.C1
   4617105 +-  2%     -14.8%    3934693        turbostat.C1E
      1.40 +-  2%      -0.2        1.18        turbostat.C1E%
  24764661 +-  2%     -18.0%   20297643        turbostat.C6
     18.09 +-  2%      -3.4       14.74        turbostat.C6%
      7.53 +-  2%     -17.1%       6.24        turbostat.CPU%c1
     11.88 +-  2%     -19.1%       9.61        turbostat.CPU%c6
      7.62 +-  3%     -20.8%       6.04        turbostat.Pkg%pc2
    388.30            +1.5%     393.93        turbostat.PkgWatt
    390974 +-  8%     +35.8%     530867 +- 11%  sched_debug.cfs_rq:/.min_vruntime.stddev
  -1754042           +75.7%   -3081270        sched_debug.cfs_rq:/.spread0.min
    388140 +-  8%     +36.2%     528494 +- 11%  sched_debug.cfs_rq:/.spread0.stddev
    542.30 +-  3%     -10.0%     488.21 +-  3%  sched_debug.cfs_rq:/.util_avg.min
     53.35 +- 16%     +48.7%      79.35 +- 12%  sched_debug.cfs_rq:/.util_est_enqueued.avg
     30520 +-  6%     -15.2%      25883 +- 12%  sched_debug.cpu.nr_switches.avg
    473770 +- 27%     -37.4%     296623 +- 32%  sched_debug.cpu.nr_switches.max
     17077 +-  2%     -15.1%      14493        sched_debug.cpu.nr_switches.min
     30138 +-  6%     -15.0%      25606 +- 12%  sched_debug.cpu.sched_count.avg
    472345 +- 27%     -37.2%     296419 +- 32%  sched_debug.cpu.sched_count.max
     16858 +-  2%     -15.2%      14299        sched_debug.cpu.sched_count.min
      8358 +-  2%     -15.5%       7063        sched_debug.cpu.sched_goidle.avg
     12225           -13.6%      10565        sched_debug.cpu.sched_goidle.max
      8032 +-  2%     -16.0%       6749        sched_debug.cpu.sched_goidle.min
     14839 +-  6%     -15.3%      12568 +- 12%  sched_debug.cpu.ttwu_count.avg
    235115 +- 28%     -38.3%     145175 +- 31%  sched_debug.cpu.ttwu_count.max
      7627 +-  3%     -15.9%       6413 +-  2%  sched_debug.cpu.ttwu_count.min
    226299 +- 29%     -39.5%     136827 +- 32%  sched_debug.cpu.ttwu_local.max
      0.85            -0.0        0.81        perf-stat.branch-miss-rate%
 3.675e+10            -4.1%  3.523e+10        perf-stat.branch-misses
 4.052e+11            -2.3%  3.958e+11        perf-stat.cache-misses
 7.008e+11            -2.5%  6.832e+11        perf-stat.cache-references
  15320995 +-  4%     -14.3%   13136557 +-  6%  perf-stat.context-switches
      9.16            +4.8%       9.59        perf-stat.cpi
  2.03e+14            +4.6%  2.124e+14        perf-stat.cpu-cycles
     44508            -1.7%      43743        perf-stat.cpu-migrations
      1.30            -0.1        1.24        perf-stat.dTLB-store-miss-rate%
 4.064e+10            -3.5%  3.922e+10        perf-stat.dTLB-store-misses
 3.086e+12            +1.1%  3.119e+12        perf-stat.dTLB-stores
 3.611e+08 +-  6%      -8.5%  3.304e+08 +-  5%  perf-stat.iTLB-loads
      0.11            -4.6%       0.10        perf-stat.ipc
 4.783e+09            -2.0%  4.685e+09        perf-stat.minor-faults
      1.53 +-  2%      -0.3        1.22 +-  8%  perf-stat.node-load-miss-rate%
 1.389e+09 +-  3%     -22.1%  1.083e+09 +-  9%  perf-stat.node-load-misses
 8.922e+10            -1.9%   8.75e+10        perf-stat.node-loads
      5.06            +1.7        6.77 +-  3%  perf-stat.node-store-miss-rate%
 1.204e+09           +29.3%  1.556e+09 +-  3%  perf-stat.node-store-misses
 2.256e+10            -5.1%  2.142e+10 +-  2%  perf-stat.node-stores
 4.783e+09            -2.0%  4.685e+09        perf-stat.page-faults
   1399242            +1.9%    1425404        perf-stat.path-length
      1144 +-  8%     -13.6%     988.00 +-  8%  slabinfo.Acpi-ParseExt.active_objs
      1144 +-  8%     -13.6%     988.00 +-  8%  slabinfo.Acpi-ParseExt.num_objs
      1878 +- 17%     +29.0%       2422 +- 16%  slabinfo.dmaengine-unmap-16.active_objs
      1878 +- 17%     +29.0%       2422 +- 16%  slabinfo.dmaengine-unmap-16.num_objs
      1085 +-  5%     -24.1%     823.33 +-  9%  slabinfo.file_lock_cache.active_objs
      1085 +-  5%     -24.1%     823.33 +-  9%  slabinfo.file_lock_cache.num_objs
     61584 +-  4%     -16.6%      51381 +-  5%  slabinfo.filp.active_objs
    967.00 +-  4%     -16.5%     807.67 +-  5%  slabinfo.filp.active_slabs
     61908 +-  4%     -16.5%      51713 +-  5%  slabinfo.filp.num_objs
    967.00 +-  4%     -16.5%     807.67 +-  5%  slabinfo.filp.num_slabs
      1455           -15.4%       1232 +-  4%  slabinfo.nsproxy.active_objs
      1455           -15.4%       1232 +-  4%  slabinfo.nsproxy.num_objs
     84720 +-  6%     -18.3%      69210 +-  4%  slabinfo.pid.active_objs
      1324 +-  6%     -18.2%       1083 +-  4%  slabinfo.pid.active_slabs
     84820 +-  5%     -18.2%      69386 +-  4%  slabinfo.pid.num_objs
      1324 +-  6%     -18.2%       1083 +-  4%  slabinfo.pid.num_slabs
      2112 +- 18%     -26.3%       1557 +-  5%  slabinfo.scsi_sense_cache.active_objs
      2112 +- 18%     -26.3%       1557 +-  5%  slabinfo.scsi_sense_cache.num_objs
      5018 +-  5%      -7.6%       4635 +-  4%  slabinfo.sock_inode_cache.active_objs
      5018 +-  5%      -7.6%       4635 +-  4%  slabinfo.sock_inode_cache.num_objs
      1193 +-  4%     +13.8%       1358 +-  4%  slabinfo.task_group.active_objs
      1193 +-  4%     +13.8%       1358 +-  4%  slabinfo.task_group.num_objs
     62807 +-  3%     -14.4%      53757 +-  3%  slabinfo.vm_area_struct.active_objs
      1571 +-  3%     -12.1%       1381 +-  3%  slabinfo.vm_area_struct.active_slabs
     62877 +-  3%     -14.3%      53880 +-  3%  slabinfo.vm_area_struct.num_objs
      1571 +-  3%     -12.1%       1381 +-  3%  slabinfo.vm_area_struct.num_slabs
     47.45           -47.4        0.00        perf-profile.calltrace.cycles-pp.alloc_pages_vma.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
     47.16           -47.2        0.00        perf-profile.calltrace.cycles-pp.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault.handle_mm_fault.__do_page_fault
     46.99           -47.0        0.00        perf-profile.calltrace.cycles-pp.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault.handle_mm_fault
     44.95           -44.9        0.00        perf-profile.calltrace.cycles-pp._raw_spin_lock.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault
      7.42 +-  2%      -7.4        0.00        perf-profile.calltrace.cycles-pp.copy_page.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      6.32 +- 10%      -6.3        0.00        perf-profile.calltrace.cycles-pp.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      6.28 +- 10%      -6.3        0.00        perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +0.9        0.85 +- 11%  perf-profile.calltrace.cycles-pp._raw_spin_lock.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +0.9        0.92 +-  4%  perf-profile.calltrace.cycles-pp.__list_del_entry_valid.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault
      0.00            +1.1        1.13 +-  7%  perf-profile.calltrace.cycles-pp.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault
      0.00            +1.2        1.19 +-  7%  perf-profile.calltrace.cycles-pp.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault
      0.00            +1.2        1.22 +-  5%  perf-profile.calltrace.cycles-pp.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +1.3        1.34 +-  7%  perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +1.4        1.36 +-  7%  perf-profile.calltrace.cycles-pp.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +4.5        4.54 +- 19%  perf-profile.calltrace.cycles-pp.pagevec_lru_move_fn.__lru_cache_add.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +4.6        4.64 +- 19%  perf-profile.calltrace.cycles-pp.__lru_cache_add.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +6.6        6.64 +- 15%  perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +6.7        6.68 +- 15%  perf-profile.calltrace.cycles-pp.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +7.5        7.54 +-  5%  perf-profile.calltrace.cycles-pp.copy_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00           +44.6       44.55 +-  3%  perf-profile.calltrace.cycles-pp._raw_spin_lock.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault
      0.00           +46.6       46.63 +-  3%  perf-profile.calltrace.cycles-pp.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault.__handle_mm_fault
      0.00           +46.8       46.81 +-  3%  perf-profile.calltrace.cycles-pp.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00           +47.1       47.10 +-  3%  perf-profile.calltrace.cycles-pp.alloc_pages_vma.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00           +63.1       63.15        perf-profile.calltrace.cycles-pp.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      0.39 +-  3%      +0.0        0.42 +-  3%  perf-profile.children.cycles-pp.radix_tree_lookup_slot
      0.21 +-  3%      +0.0        0.25 +-  5%  perf-profile.children.cycles-pp.__mod_node_page_state
      0.00            +0.1        0.06 +-  8%  perf-profile.children.cycles-pp.get_vma_policy
      0.00            +0.1        0.08 +-  5%  perf-profile.children.cycles-pp.__lru_cache_add_active_or_unevictable
      0.00            +0.2        0.18 +-  6%  perf-profile.children.cycles-pp.__page_add_new_anon_rmap
      0.00            +1.4        1.35 +-  5%  perf-profile.children.cycles-pp.pte_map_lock
      0.00           +63.2       63.21        perf-profile.children.cycles-pp.handle_pte_fault
      1.40 +-  2%      -0.4        1.03 +- 10%  perf-profile.self.cycles-pp._raw_spin_lock
      0.56 +-  3%      -0.2        0.35 +-  6%  perf-profile.self.cycles-pp.__handle_mm_fault
      0.22 +-  3%      -0.0        0.18 +-  7%  perf-profile.self.cycles-pp.alloc_set_pte
      0.09            +0.0        0.10 +-  4%  perf-profile.self.cycles-pp.vmacache_find
      0.39 +-  2%      +0.0        0.41 +-  3%  perf-profile.self.cycles-pp.__radix_tree_lookup
      0.18            +0.0        0.20 +-  6%  perf-profile.self.cycles-pp.mem_cgroup_charge_statistics
      0.17 +-  2%      +0.0        0.20 +-  7%  perf-profile.self.cycles-pp.___might_sleep
      0.33 +-  2%      +0.0        0.36 +-  6%  perf-profile.self.cycles-pp.handle_mm_fault
      0.20 +-  2%      +0.0        0.24 +-  3%  perf-profile.self.cycles-pp.__mod_node_page_state
      0.00            +0.1        0.05        perf-profile.self.cycles-pp.finish_fault
      0.00            +0.1        0.05        perf-profile.self.cycles-pp.get_vma_policy
      0.00            +0.1        0.08 +- 10%  perf-profile.self.cycles-pp.__lru_cache_add_active_or_unevictable
      0.00            +0.2        0.25 +-  5%  perf-profile.self.cycles-pp.handle_pte_fault
      0.00            +0.5        0.49 +-  8%  perf-profile.self.cycles-pp.pte_map_lock

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/never/page_fault2/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
          1:3          -33%            :3     kmsg.DHCP/BOOTP:Reply_not_for_us_on_eth#,op[#]xid[#]
           :3           33%           1:3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=sched_slice/0x
          1:3          -33%            :3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=schedule_tail/0x
          1:3           24%           2:3     perf-profile.calltrace.cycles-pp.sync_regs.error_entry
          3:3           46%           5:3     perf-profile.calltrace.cycles-pp.error_entry
          5:3           -9%           5:3     perf-profile.children.cycles-pp.error_entry
          2:3           -4%           2:3     perf-profile.self.cycles-pp.error_entry
         %stddev     %change         %stddev
             \          |                \  
      8147           -18.8%       6613        will-it-scale.per_thread_ops
     93113           +17.0%     108982        will-it-scale.time.involuntary_context_switches
 4.732e+08           -19.0%  3.833e+08        will-it-scale.time.minor_page_faults
      5854           +12.0%       6555        will-it-scale.time.percent_of_cpu_this_job_got
     35247           +12.1%      39495        will-it-scale.time.system_time
   5546661           -15.5%    4689314        will-it-scale.time.voluntary_context_switches
  15801637            -1.9%   15504487        will-it-scale.workload
      1.43 +- 11%     -59.7%       0.58 +- 28%  irq_exception_noise.__do_page_fault.min
      2811 +-  3%     +23.7%       3477 +-  3%  kthread_noise.total_time
    292776 +-  5%     +39.6%     408829 +- 21%  meminfo.DirectMap4k
     19.80            -3.7       16.12        mpstat.cpu.idle%
     29940           -14.5%      25593        uptime.idle
     24064 +-  3%      -8.5%      22016        vmstat.system.cs
     34.86            -1.9%      34.19        boot-time.boot
     26.95            -2.8%      26.19 +-  2%  boot-time.kernel_boot
   7190569 +-  2%     -15.2%    6100136 +-  3%  softirqs.RCU
   5513663           -13.8%    4751548        softirqs.SCHED
     18064 +-  2%     +24.3%      22461 +-  7%  numa-vmstat.node0.nr_slab_unreclaimable
      8507 +- 12%     -16.8%       7075 +-  4%  numa-vmstat.node2.nr_slab_reclaimable
     18719 +-  9%     -19.6%      15043 +-  4%  numa-vmstat.node3.nr_slab_unreclaimable
     72265 +-  2%     +24.3%      89855 +-  7%  numa-meminfo.node0.SUnreclaim
    115980 +-  4%     +22.6%     142233 +- 12%  numa-meminfo.node0.Slab
     34035 +- 12%     -16.8%      28307 +-  4%  numa-meminfo.node2.SReclaimable
     74888 +-  9%     -19.7%      60162 +-  4%  numa-meminfo.node3.SUnreclaim
     93113           +17.0%     108982        time.involuntary_context_switches
 4.732e+08           -19.0%  3.833e+08        time.minor_page_faults
      5854           +12.0%       6555        time.percent_of_cpu_this_job_got
     35247           +12.1%      39495        time.system_time
   5546661           -15.5%    4689314        time.voluntary_context_switches
 4.792e+09            -1.9%  4.699e+09        proc-vmstat.numa_hit
 4.791e+09            -1.9%  4.699e+09        proc-vmstat.numa_local
     40447 +- 11%     +13.2%      45804 +-  6%  proc-vmstat.pgactivate
 4.778e+09            -1.9%  4.688e+09        proc-vmstat.pgalloc_normal
 4.767e+09            -1.9%  4.675e+09        proc-vmstat.pgfault
 4.791e+09            -1.9%  4.699e+09        proc-vmstat.pgfree
    230178 +-  2%     -10.1%     206883 +-  3%  cpuidle.C1.usage
 1.617e+09           -15.0%  1.375e+09        cpuidle.C1E.time
   4514401           -14.1%    3878206        cpuidle.C1E.usage
 2.087e+10           -18.5%  1.701e+10        cpuidle.C6.time
  24458365           -18.0%   20045336        cpuidle.C6.usage
   1163758           -16.1%     976094 +-  4%  cpuidle.POLL.time
     17907           -14.6%      15294 +-  4%  cpuidle.POLL.usage
      1758            +4.5%       1838        turbostat.Avg_MHz
    227522 +-  2%     -10.2%     204426 +-  3%  turbostat.C1
   4512700           -14.2%    3873264        turbostat.C1E
      1.39            -0.2        1.18        turbostat.C1E%
  24452583           -18.0%   20039031        turbostat.C6
     17.85            -3.3       14.55        turbostat.C6%
      7.44           -16.8%       6.19        turbostat.CPU%c1
     11.72           -19.3%       9.45        turbostat.CPU%c6
      7.51           -21.3%       5.91        turbostat.Pkg%pc2
    389.33            +1.6%     395.59        turbostat.PkgWatt
    559.33 +- 13%     -17.9%     459.33 +- 20%  slabinfo.dmaengine-unmap-128.active_objs
    559.33 +- 13%     -17.9%     459.33 +- 20%  slabinfo.dmaengine-unmap-128.num_objs
     57734 +-  3%      -5.7%      54421 +-  4%  slabinfo.filp.active_objs
    905.67 +-  3%      -5.6%     854.67 +-  4%  slabinfo.filp.active_slabs
     57981 +-  3%      -5.6%      54720 +-  4%  slabinfo.filp.num_objs
    905.67 +-  3%      -5.6%     854.67 +-  4%  slabinfo.filp.num_slabs
      1378           -12.0%       1212 +-  7%  slabinfo.nsproxy.active_objs
      1378           -12.0%       1212 +-  7%  slabinfo.nsproxy.num_objs
    507.33 +-  7%     -26.8%     371.33 +-  2%  slabinfo.secpath_cache.active_objs
    507.33 +-  7%     -26.8%     371.33 +-  2%  slabinfo.secpath_cache.num_objs
      4788 +-  5%      -8.3%       4391 +-  2%  slabinfo.sock_inode_cache.active_objs
      4788 +-  5%      -8.3%       4391 +-  2%  slabinfo.sock_inode_cache.num_objs
      1431 +-  8%     -12.3%       1255 +-  3%  slabinfo.task_group.active_objs
      1431 +-  8%     -12.3%       1255 +-  3%  slabinfo.task_group.num_objs
      4.27 +- 17%     +27.0%       5.42 +-  7%  sched_debug.cfs_rq:/.runnable_load_avg.avg
     13.44 +- 62%     +73.6%      23.33 +- 24%  sched_debug.cfs_rq:/.runnable_load_avg.stddev
    772.55 +- 21%     -32.7%     520.27 +-  4%  sched_debug.cfs_rq:/.util_est_enqueued.max
      4.39 +- 15%     +29.0%       5.66 +- 11%  sched_debug.cpu.cpu_load[0].avg
    152.09 +- 72%     +83.9%     279.67 +- 33%  sched_debug.cpu.cpu_load[0].max
     13.84 +- 58%     +78.7%      24.72 +- 29%  sched_debug.cpu.cpu_load[0].stddev
      4.53 +- 14%     +25.8%       5.70 +- 10%  sched_debug.cpu.cpu_load[1].avg
    156.58 +- 66%     +76.6%     276.58 +- 33%  sched_debug.cpu.cpu_load[1].max
     14.02 +- 55%     +72.4%      24.17 +- 28%  sched_debug.cpu.cpu_load[1].stddev
      4.87 +- 11%     +17.3%       5.72 +-  9%  sched_debug.cpu.cpu_load[2].avg
      1.58 +-  2%     +13.5%       1.79 +-  6%  sched_debug.cpu.nr_running.max
     16694           -14.6%      14259        sched_debug.cpu.nr_switches.min
     31989 +- 13%     +20.6%      38584 +-  6%  sched_debug.cpu.nr_switches.stddev
     16505           -14.8%      14068        sched_debug.cpu.sched_count.min
     32084 +- 13%     +19.9%      38482 +-  6%  sched_debug.cpu.sched_count.stddev
      8185           -15.0%       6957        sched_debug.cpu.sched_goidle.avg
     12151 +-  2%     -13.5%      10507        sched_debug.cpu.sched_goidle.max
      7867           -15.7%       6631        sched_debug.cpu.sched_goidle.min
      7595           -16.1%       6375        sched_debug.cpu.ttwu_count.min
     15873 +- 13%     +21.2%      19239 +-  6%  sched_debug.cpu.ttwu_count.stddev
      5244 +- 17%     +17.0%       6134 +-  5%  sched_debug.cpu.ttwu_local.avg
     15646 +- 12%     +21.5%      19008 +-  6%  sched_debug.cpu.ttwu_local.stddev
      0.85            -0.0        0.81        perf-stat.branch-miss-rate%
 3.689e+10            -4.6%  3.518e+10        perf-stat.branch-misses
     57.39            +0.6       58.00        perf-stat.cache-miss-rate%
 4.014e+11            -1.2%  3.967e+11        perf-stat.cache-misses
 6.994e+11            -2.2%   6.84e+11        perf-stat.cache-references
  14605393 +-  3%      -8.5%   13369913        perf-stat.context-switches
      9.21            +4.5%       9.63        perf-stat.cpi
 2.037e+14            +4.6%   2.13e+14        perf-stat.cpu-cycles
     44424            -2.0%      43541        perf-stat.cpu-migrations
      1.29            -0.1        1.24        perf-stat.dTLB-store-miss-rate%
 4.018e+10            -2.8%  3.905e+10        perf-stat.dTLB-store-misses
 3.071e+12            +1.4%  3.113e+12        perf-stat.dTLB-stores
     93.04            +1.5       94.51        perf-stat.iTLB-load-miss-rate%
 4.946e+09           +19.3%  5.903e+09 +-  5%  perf-stat.iTLB-load-misses
 3.702e+08            -7.5%  3.423e+08 +-  2%  perf-stat.iTLB-loads
      4470           -15.9%       3760 +-  5%  perf-stat.instructions-per-iTLB-miss
      0.11            -4.3%       0.10        perf-stat.ipc
 4.767e+09            -1.9%  4.675e+09        perf-stat.minor-faults
      1.46 +-  4%      -0.1        1.33 +-  9%  perf-stat.node-load-miss-rate%
      4.91            +1.7        6.65 +-  2%  perf-stat.node-store-miss-rate%
 1.195e+09           +32.8%  1.587e+09 +-  2%  perf-stat.node-store-misses
 2.313e+10            -3.7%  2.227e+10        perf-stat.node-stores
 4.767e+09            -1.9%  4.675e+09        perf-stat.page-faults
   1399047            +2.0%    1427115        perf-stat.path-length
      8908 +- 73%    -100.0%       0.00        latency_stats.avg.call_rwsem_down_read_failed.m_start.seq_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      3604 +-141%    -100.0%       0.00        latency_stats.avg.call_rwsem_down_write_failed.do_unlinkat.do_syscall_64.entry_SYSCALL_64_after_hwframe
     61499 +-130%     -92.6%       4534 +- 16%  latency_stats.avg.expand_files.__alloc_fd.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4391 +-138%     -70.9%       1277 +-129%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
     67311 +-112%     -48.5%      34681 +- 36%  latency_stats.avg.max
      3956 +-138%    +320.4%      16635 +-140%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    164.67 +- 30%   +7264.0%      12126 +-138%  latency_stats.avg.flush_work.fsnotify_destroy_group.inotify_release.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +5.4e+105%       5367 +-141%  latency_stats.avg.call_rwsem_down_write_failed.unlink_file_vma.free_pgtables.exit_mmap.mmput.flush_old_exec.load_elf_binary.search_binary_handler.do_execveat_common.__x64_sys_execve.do_syscall_64.entry_SYSCALL_64_after_hwframe
     36937 +-119%    -100.0%       0.00        latency_stats.max.call_rwsem_down_read_failed.m_start.seq_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      3604 +-141%    -100.0%       0.00        latency_stats.max.call_rwsem_down_write_failed.do_unlinkat.do_syscall_64.entry_SYSCALL_64_after_hwframe
     84146 +-107%     -72.5%      23171 +- 31%  latency_stats.max.expand_files.__alloc_fd.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4391 +-138%     -70.9%       1277 +-129%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
      5817 +- 83%     -69.7%       1760 +- 67%  latency_stats.max.pipe_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      6720 +-137%   +1628.2%     116147 +-141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    164.67 +- 30%   +7264.0%      12126 +-138%  latency_stats.max.flush_work.fsnotify_destroy_group.inotify_release.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +1.2e+106%      12153 +-141%  latency_stats.max.call_rwsem_down_write_failed.unlink_file_vma.free_pgtables.exit_mmap.mmput.flush_old_exec.load_elf_binary.search_binary_handler.do_execveat_common.__x64_sys_execve.do_syscall_64.entry_SYSCALL_64_after_hwframe
    110122 +-120%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_read_failed.m_start.seq_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      3604 +-141%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_write_failed.do_unlinkat.do_syscall_64.entry_SYSCALL_64_after_hwframe
  12078828 +-139%     -99.3%      89363 +- 29%  latency_stats.sum.expand_files.__alloc_fd.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
    144453 +-120%     -80.9%      27650 +- 19%  latency_stats.sum.poll_schedule_timeout.do_sys_poll.__x64_sys_poll.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4391 +-138%     -70.9%       1277 +-129%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
      9438 +- 86%     -68.4%       2980 +- 35%  latency_stats.sum.pipe_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
     31656 +-138%    +320.4%     133084 +-140%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    164.67 +- 30%   +7264.0%      12126 +-138%  latency_stats.sum.flush_work.fsnotify_destroy_group.inotify_release.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +8.8e+105%       8760 +-141%  latency_stats.sum.msleep_interruptible.uart_wait_until_sent.tty_wait_until_sent.tty_port_close_start.tty_port_close.tty_release.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +1.3e+106%      12897 +-141%  latency_stats.sum.tty_wait_until_sent.tty_port_close_start.tty_port_close.tty_release.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +3.2e+106%      32207 +-141%  latency_stats.sum.call_rwsem_down_write_failed.unlink_file_vma.free_pgtables.exit_mmap.mmput.flush_old_exec.load_elf_binary.search_binary_handler.do_execveat_common.__x64_sys_execve.do_syscall_64.entry_SYSCALL_64_after_hwframe
     44.43 +-  3%     -44.4        0.00        perf-profile.calltrace.cycles-pp.alloc_pages_vma.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
     44.13 +-  3%     -44.1        0.00        perf-profile.calltrace.cycles-pp.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault.handle_mm_fault.__do_page_fault
     43.95 +-  3%     -43.9        0.00        perf-profile.calltrace.cycles-pp.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault.handle_mm_fault
     41.85 +-  4%     -41.9        0.00        perf-profile.calltrace.cycles-pp._raw_spin_lock.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault
      7.74 +-  8%      -7.7        0.00        perf-profile.calltrace.cycles-pp.copy_page.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      7.19 +-  4%      -7.2        0.00        perf-profile.calltrace.cycles-pp.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      7.15 +-  4%      -7.2        0.00        perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      5.09 +-  3%      -5.1        0.00        perf-profile.calltrace.cycles-pp.__lru_cache_add.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault
      4.99 +-  3%      -5.0        0.00        perf-profile.calltrace.cycles-pp.pagevec_lru_move_fn.__lru_cache_add.alloc_set_pte.finish_fault.__handle_mm_fault
      0.93 +-  6%      -0.1        0.81 +-  2%  perf-profile.calltrace.cycles-pp.find_get_entry.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault
      0.00            +0.8        0.84        perf-profile.calltrace.cycles-pp._raw_spin_lock.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +0.9        0.92 +-  3%  perf-profile.calltrace.cycles-pp.__list_del_entry_valid.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault
      0.00            +1.1        1.08        perf-profile.calltrace.cycles-pp.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault
      0.00            +1.1        1.14        perf-profile.calltrace.cycles-pp.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault
      0.00            +1.2        1.17        perf-profile.calltrace.cycles-pp.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +1.3        1.29        perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +1.3        1.31        perf-profile.calltrace.cycles-pp.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
     61.62            +1.7       63.33        perf-profile.calltrace.cycles-pp.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault.page_fault
     41.73 +-  4%      +3.0       44.75        perf-profile.calltrace.cycles-pp.native_queued_spin_lock_slowpath._raw_spin_lock.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma
      0.00            +4.6        4.55 +- 15%  perf-profile.calltrace.cycles-pp.pagevec_lru_move_fn.__lru_cache_add.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +4.6        4.65 +- 14%  perf-profile.calltrace.cycles-pp.__lru_cache_add.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +6.6        6.57 +- 10%  perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +6.6        6.61 +- 10%  perf-profile.calltrace.cycles-pp.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +7.2        7.25 +-  2%  perf-profile.calltrace.cycles-pp.copy_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
     41.41 +- 70%     +22.3       63.67        perf-profile.calltrace.cycles-pp.handle_mm_fault.__do_page_fault.do_page_fault.page_fault
     42.19 +- 70%     +22.6       64.75        perf-profile.calltrace.cycles-pp.__do_page_fault.do_page_fault.page_fault
     42.20 +- 70%     +22.6       64.76        perf-profile.calltrace.cycles-pp.do_page_fault.page_fault
     42.27 +- 70%     +22.6       64.86        perf-profile.calltrace.cycles-pp.page_fault
      0.00           +44.9       44.88        perf-profile.calltrace.cycles-pp._raw_spin_lock.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault
      0.00           +46.9       46.92        perf-profile.calltrace.cycles-pp.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault.__handle_mm_fault
      0.00           +47.1       47.10        perf-profile.calltrace.cycles-pp.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00           +47.4       47.37        perf-profile.calltrace.cycles-pp.alloc_pages_vma.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00           +63.0       63.00        perf-profile.calltrace.cycles-pp.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      0.97 +-  6%      -0.1        0.84 +-  2%  perf-profile.children.cycles-pp.find_get_entry
      1.23 +-  6%      -0.1        1.11        perf-profile.children.cycles-pp.find_lock_entry
      0.09 +- 10%      -0.0        0.07 +-  6%  perf-profile.children.cycles-pp.unlock_page
      0.19 +-  4%      +0.0        0.21 +-  2%  perf-profile.children.cycles-pp.mem_cgroup_charge_statistics
      0.21 +-  2%      +0.0        0.25        perf-profile.children.cycles-pp.__mod_node_page_state
      0.00            +0.1        0.05 +-  8%  perf-profile.children.cycles-pp.get_vma_policy
      0.00            +0.1        0.08        perf-profile.children.cycles-pp.__lru_cache_add_active_or_unevictable
      0.00            +0.2        0.18 +-  2%  perf-profile.children.cycles-pp.__page_add_new_anon_rmap
      0.00            +1.3        1.30        perf-profile.children.cycles-pp.pte_map_lock
     63.40            +1.6       64.97        perf-profile.children.cycles-pp.__do_page_fault
     63.19            +1.6       64.83        perf-profile.children.cycles-pp.do_page_fault
     61.69            +1.7       63.36        perf-profile.children.cycles-pp.__handle_mm_fault
     63.19            +1.7       64.86        perf-profile.children.cycles-pp.page_fault
     61.99            +1.7       63.70        perf-profile.children.cycles-pp.handle_mm_fault
     72.27            +2.2       74.52        perf-profile.children.cycles-pp.native_queued_spin_lock_slowpath
     67.51            +2.4       69.87        perf-profile.children.cycles-pp._raw_spin_lock
     44.49 +-  3%      +3.0       47.45        perf-profile.children.cycles-pp.alloc_pages_vma
     44.28 +-  3%      +3.0       47.26        perf-profile.children.cycles-pp.__alloc_pages_nodemask
     44.13 +-  3%      +3.0       47.12        perf-profile.children.cycles-pp.get_page_from_freelist
      0.00           +63.1       63.06        perf-profile.children.cycles-pp.handle_pte_fault
      1.46 +-  7%      -0.5        1.01        perf-profile.self.cycles-pp._raw_spin_lock
      0.58 +-  6%      -0.2        0.34        perf-profile.self.cycles-pp.__handle_mm_fault
      0.55 +-  6%      -0.1        0.44 +-  2%  perf-profile.self.cycles-pp.find_get_entry
      0.22 +-  5%      -0.1        0.16 +-  2%  perf-profile.self.cycles-pp.alloc_set_pte
      0.10 +-  8%      -0.0        0.08        perf-profile.self.cycles-pp.down_read_trylock
      0.09 +-  5%      -0.0        0.07        perf-profile.self.cycles-pp.unlock_page
      0.06            -0.0        0.05        perf-profile.self.cycles-pp.pmd_devmap_trans_unstable
      0.20 +-  2%      +0.0        0.24 +-  3%  perf-profile.self.cycles-pp.__mod_node_page_state
      0.00            +0.1        0.05        perf-profile.self.cycles-pp.finish_fault
      0.00            +0.1        0.05        perf-profile.self.cycles-pp.get_vma_policy
      0.00            +0.1        0.08 +-  6%  perf-profile.self.cycles-pp.__lru_cache_add_active_or_unevictable
      0.00            +0.2        0.25        perf-profile.self.cycles-pp.handle_pte_fault
      0.00            +0.5        0.46 +-  7%  perf-profile.self.cycles-pp.pte_map_lock
     72.26            +2.3       74.52        perf-profile.self.cycles-pp.native_queued_spin_lock_slowpath

[-- Attachment #3: perf-profile.zip --]
[-- Type: application/zip, Size: 19025 bytes --]

Re: [PATCH v11 00/26] Speculative page faults

[Haiyan Song]

[-- Attachment #1: Type: text/plain, Size: 31691 bytes --]

On Mon, Jun 11, 2018 at 05:15:22PM +0200, Laurent Dufour wrote:

Hi Laurent,

For perf date tested on Intel 4s Skylake platform, here attached the compare result
between base and head commit in attachment, which include the perf-profile comparision information.

And also attached some perf-profile.json captured from test result for page_fault2 and page_fault3 for
checking the regression, thanks.


Best regards,
Haiyan Song



> Hi Haiyan,
> 
> I don't have access to the same hardware you ran the test on, but I give a try
> to those test on a Power8 system (2 sockets, 5 cores/s, 8 threads/c, 80 CPUs 32G).
> I run each will-it-scale test 10 times and compute the average.
> 
> test THP enabled		4.17.0-rc4-mm1	spf		delta
> page_fault3_threads		2697.7		2683.5		-0.53%
> page_fault2_threads		170660.6	169574.1	-0.64%
> context_switch1_threads		6915269.2	6877507.3	-0.55%
> context_switch1_processes	6478076.2	6529493.5	0.79%
> rk1				243391.2	238527.5	-2.00%
> 
> Test were launched with the arguments '-t 80 -s 5', only the average report is
> taken in account. Note that page size is 64K by default on ppc64.
> 
> It would be nice if you could capture some perf data to figure out why the
> page_fault2/3 are showing such a performance regression.
> 
> Thanks,
> Laurent.
> 
> On 11/06/2018 09:49, Song, HaiyanX wrote:
> > Hi Laurent,
> > 
> > Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
> > tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
> > V9 patch serials.
> > 
> > The regression result is sorted by the metric will-it-scale.per_thread_ops.
> > branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
> > commit id:
> >   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
> >   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
> > Benchmark: will-it-scale
> > Download link: https://github.com/antonblanchard/will-it-scale/tree/master
> > 
> > Metrics:
> >   will-it-scale.per_process_ops=processes/nr_cpu
> >   will-it-scale.per_thread_ops=threads/nr_cpu
> >   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
> > THP: enable / disable
> > nr_task:100%
> > 
> > 1. Regressions:
> > 
> > a). Enable THP
> > testcase                          base           change      head           metric
> > page_fault3/enable THP           10519          -20.5%       8368      will-it-scale.per_thread_ops
> > page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
> > brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
> > context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
> > context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
> > 
> > b). Disable THP
> > page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
> > page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
> > brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
> > context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
> > brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
> > page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
> > context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
> > 
> > Notes: for the above  values of test result, the higher is better.
> > 
> > 2. Improvement: not found improvement based on the selected test cases.
> > 
> > 
> > Best regards
> > Haiyan Song
> > ________________________________________
> > From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> > Sent: Monday, May 28, 2018 4:54 PM
> > To: Song, HaiyanX
> > Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> > Subject: Re: [PATCH v11 00/26] Speculative page faults
> > 
> > On 28/05/2018 10:22, Haiyan Song wrote:
> >> Hi Laurent,
> >>
> >> Yes, these tests are done on V9 patch.
> > 
> > Do you plan to give this V11 a run ?
> > 
> >>
> >>
> >> Best regards,
> >> Haiyan Song
> >>
> >> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
> >>> On 28/05/2018 07:23, Song, HaiyanX wrote:
> >>>>
> >>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
> >>>> tested on Intel 4s Skylake platform.
> >>>
> >>> Hi,
> >>>
> >>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
> >>> series" while responding to the v11 header series...
> >>> Were these tests done on v9 or v11 ?
> >>>
> >>> Cheers,
> >>> Laurent.
> >>>
> >>>>
> >>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
> >>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
> >>>> Commit id:
> >>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
> >>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
> >>>> Benchmark suite: will-it-scale
> >>>> Download link:
> >>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
> >>>> Metrics:
> >>>>     will-it-scale.per_process_ops=processes/nr_cpu
> >>>>     will-it-scale.per_thread_ops=threads/nr_cpu
> >>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
> >>>> THP: enable / disable
> >>>> nr_task: 100%
> >>>>
> >>>> 1. Regressions:
> >>>> a) THP enabled:
> >>>> testcase                        base            change          head       metric
> >>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
> >>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
> >>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
> >>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
> >>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
> >>>>
> >>>> b) THP disabled:
> >>>> testcase                        base            change          head       metric
> >>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
> >>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
> >>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
> >>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
> >>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
> >>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
> >>>>
> >>>> 2. Improvements:
> >>>> a) THP enabled:
> >>>> testcase                        base            change          head       metric
> >>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
> >>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
> >>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
> >>>>
> >>>> b) THP disabled:
> >>>> testcase                        base            change          head       metric
> >>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
> >>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
> >>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
> >>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
> >>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
> >>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
> >>>>
> >>>> Notes: for above values in column "change", the higher value means that the related testcase result
> >>>> on head commit is better than that on base commit for this benchmark.
> >>>>
> >>>>
> >>>> Best regards
> >>>> Haiyan Song
> >>>>
> >>>> ________________________________________
> >>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> >>>> Sent: Thursday, May 17, 2018 7:06 PM
> >>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
> >>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> >>>> Subject: [PATCH v11 00/26] Speculative page faults
> >>>>
> >>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
> >>>> page fault without holding the mm semaphore [1].
> >>>>
> >>>> The idea is to try to handle user space page faults without holding the
> >>>> mmap_sem. This should allow better concurrency for massively threaded
> >>>> process since the page fault handler will not wait for other threads memory
> >>>> layout change to be done, assuming that this change is done in another part
> >>>> of the process's memory space. This type page fault is named speculative
> >>>> page fault. If the speculative page fault fails because of a concurrency is
> >>>> detected or because underlying PMD or PTE tables are not yet allocating, it
> >>>> is failing its processing and a classic page fault is then tried.
> >>>>
> >>>> The speculative page fault (SPF) has to look for the VMA matching the fault
> >>>> address without holding the mmap_sem, this is done by introducing a rwlock
> >>>> which protects the access to the mm_rb tree. Previously this was done using
> >>>> SRCU but it was introducing a lot of scheduling to process the VMA's
> >>>> freeing operation which was hitting the performance by 20% as reported by
> >>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
> >>>> limiting the locking contention to these operations which are expected to
> >>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
> >>>> our back a reference count is added and 2 services (get_vma() and
> >>>> put_vma()) are introduced to handle the reference count. Once a VMA is
> >>>> fetched from the RB tree using get_vma(), it must be later freed using
> >>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
> >>>> benchmark anymore.
> >>>>
> >>>> The VMA's attributes checked during the speculative page fault processing
> >>>> have to be protected against parallel changes. This is done by using a per
> >>>> VMA sequence lock. This sequence lock allows the speculative page fault
> >>>> handler to fast check for parallel changes in progress and to abort the
> >>>> speculative page fault in that case.
> >>>>
> >>>> Once the VMA has been found, the speculative page fault handler would check
> >>>> for the VMA's attributes to verify that the page fault has to be handled
> >>>> correctly or not. Thus, the VMA is protected through a sequence lock which
> >>>> allows fast detection of concurrent VMA changes. If such a change is
> >>>> detected, the speculative page fault is aborted and a *classic* page fault
> >>>> is tried.  VMA sequence lockings are added when VMA attributes which are
> >>>> checked during the page fault are modified.
> >>>>
> >>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
> >>>> so once the page table is locked, the VMA is valid, so any other changes
> >>>> leading to touching this PTE will need to lock the page table, so no
> >>>> parallel change is possible at this time.
> >>>>
> >>>> The locking of the PTE is done with interrupts disabled, this allows
> >>>> checking for the PMD to ensure that there is not an ongoing collapsing
> >>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
> >>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
> >>>> valid at the time the PTE is locked, we have the guarantee that the
> >>>> collapsing operation will have to wait on the PTE lock to move forward.
> >>>> This allows the SPF handler to map the PTE safely. If the PMD value is
> >>>> different from the one recorded at the beginning of the SPF operation, the
> >>>> classic page fault handler will be called to handle the operation while
> >>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
> >>>> the lock is done using spin_trylock() to avoid dead lock when handling a
> >>>> page fault while a TLB invalidate is requested by another CPU holding the
> >>>> PTE.
> >>>>
> >>>> In pseudo code, this could be seen as:
> >>>>     speculative_page_fault()
> >>>>     {
> >>>>             vma = get_vma()
> >>>>             check vma sequence count
> >>>>             check vma's support
> >>>>             disable interrupt
> >>>>                   check pgd,p4d,...,pte
> >>>>                   save pmd and pte in vmf
> >>>>                   save vma sequence counter in vmf
> >>>>             enable interrupt
> >>>>             check vma sequence count
> >>>>             handle_pte_fault(vma)
> >>>>                     ..
> >>>>                     page = alloc_page()
> >>>>                     pte_map_lock()
> >>>>                             disable interrupt
> >>>>                                     abort if sequence counter has changed
> >>>>                                     abort if pmd or pte has changed
> >>>>                                     pte map and lock
> >>>>                             enable interrupt
> >>>>                     if abort
> >>>>                        free page
> >>>>                        abort
> >>>>                     ...
> >>>>     }
> >>>>
> >>>>     arch_fault_handler()
> >>>>     {
> >>>>             if (speculative_page_fault(&vma))
> >>>>                goto done
> >>>>     again:
> >>>>             lock(mmap_sem)
> >>>>             vma = find_vma();
> >>>>             handle_pte_fault(vma);
> >>>>             if retry
> >>>>                unlock(mmap_sem)
> >>>>                goto again;
> >>>>     done:
> >>>>             handle fault error
> >>>>     }
> >>>>
> >>>> Support for THP is not done because when checking for the PMD, we can be
> >>>> confused by an in progress collapsing operation done by khugepaged. The
> >>>> issue is that pmd_none() could be true either if the PMD is not already
> >>>> populated or if the underlying PTE are in the way to be collapsed. So we
> >>>> cannot safely allocate a PMD if pmd_none() is true.
> >>>>
> >>>> This series add a new software performance event named 'speculative-faults'
> >>>> or 'spf'. It counts the number of successful page fault event handled
> >>>> speculatively. When recording 'faults,spf' events, the faults one is
> >>>> counting the total number of page fault events while 'spf' is only counting
> >>>> the part of the faults processed speculatively.
> >>>>
> >>>> There are some trace events introduced by this series. They allow
> >>>> identifying why the page faults were not processed speculatively. This
> >>>> doesn't take in account the faults generated by a monothreaded process
> >>>> which directly processed while holding the mmap_sem. This trace events are
> >>>> grouped in a system named 'pagefault', they are:
> >>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
> >>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
> >>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
> >>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
> >>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
> >>>>    back.
> >>>>
> >>>> To record all the related events, the easier is to run perf with the
> >>>> following arguments :
> >>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
> >>>>
> >>>> There is also a dedicated vmstat counter showing the number of successful
> >>>> page fault handled speculatively. I can be seen this way:
> >>>> $ grep speculative_pgfault /proc/vmstat
> >>>>
> >>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
> >>>> on x86, PowerPC and arm64.
> >>>>
> >>>> ---------------------
> >>>> Real Workload results
> >>>>
> >>>> As mentioned in previous email, we did non official runs using a "popular
> >>>> in memory multithreaded database product" on 176 cores SMT8 Power system
> >>>> which showed a 30% improvements in the number of transaction processed per
> >>>> second. This run has been done on the v6 series, but changes introduced in
> >>>> this new version should not impact the performance boost seen.
> >>>>
> >>>> Here are the perf data captured during 2 of these runs on top of the v8
> >>>> series:
> >>>>                 vanilla         spf
> >>>> faults          89.418          101.364         +13%
> >>>> spf                n/a           97.989
> >>>>
> >>>> With the SPF kernel, most of the page fault were processed in a speculative
> >>>> way.
> >>>>
> >>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
> >>>> it a try on an android device. He reported that the application launch time
> >>>> was improved in average by 6%, and for large applications (~100 threads) by
> >>>> 20%.
> >>>>
> >>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
> >>>> MSM845 (8 cores) with 6GB (the less is better):
> >>>>
> >>>> Application                             4.9     4.9+spf delta
> >>>> com.tencent.mm                          416     389     -7%
> >>>> com.eg.android.AlipayGphone             1135    986     -13%
> >>>> com.tencent.mtt                         455     454     0%
> >>>> com.qqgame.hlddz                        1497    1409    -6%
> >>>> com.autonavi.minimap                    711     701     -1%
> >>>> com.tencent.tmgp.sgame                  788     748     -5%
> >>>> com.immomo.momo                         501     487     -3%
> >>>> com.tencent.peng                        2145    2112    -2%
> >>>> com.smile.gifmaker                      491     461     -6%
> >>>> com.baidu.BaiduMap                      479     366     -23%
> >>>> com.taobao.taobao                       1341    1198    -11%
> >>>> com.baidu.searchbox                     333     314     -6%
> >>>> com.tencent.mobileqq                    394     384     -3%
> >>>> com.sina.weibo                          907     906     0%
> >>>> com.youku.phone                         816     731     -11%
> >>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
> >>>> com.UCMobile                            415     411     -1%
> >>>> com.tencent.tmgp.ak                     1464    1431    -2%
> >>>> com.tencent.qqmusic                     336     329     -2%
> >>>> com.sankuai.meituan                     1661    1302    -22%
> >>>> com.netease.cloudmusic                  1193    1200    1%
> >>>> air.tv.douyu.android                    4257    4152    -2%
> >>>>
> >>>> ------------------
> >>>> Benchmarks results
> >>>>
> >>>> Base kernel is v4.17.0-rc4-mm1
> >>>> SPF is BASE + this series
> >>>>
> >>>> Kernbench:
> >>>> ----------
> >>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
> >>>> kernel (kernel is build 5 times):
> >>>>
> >>>> Average Half load -j 8
> >>>>                  Run    (std deviation)
> >>>>                  BASE                   SPF
> >>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
> >>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
> >>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
> >>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
> >>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
> >>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
> >>>>
> >>>> Average Optimal load -j 16
> >>>>                  Run    (std deviation)
> >>>>                  BASE                   SPF
> >>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
> >>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
> >>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
> >>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
> >>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
> >>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
> >>>>
> >>>>
> >>>> During a run on the SPF, perf events were captured:
> >>>>  Performance counter stats for '../kernbench -M':
> >>>>          526743764      faults
> >>>>                210      spf
> >>>>                  3      pagefault:spf_vma_changed
> >>>>                  0      pagefault:spf_vma_noanon
> >>>>               2278      pagefault:spf_vma_notsup
> >>>>                  0      pagefault:spf_vma_access
> >>>>                  0      pagefault:spf_pmd_changed
> >>>>
> >>>> Very few speculative page faults were recorded as most of the processes
> >>>> involved are monothreaded (sounds that on this architecture some threads
> >>>> were created during the kernel build processing).
> >>>>
> >>>> Here are the kerbench results on a 80 CPUs Power8 system:
> >>>>
> >>>> Average Half load -j 40
> >>>>                  Run    (std deviation)
> >>>>                  BASE                   SPF
> >>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
> >>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
> >>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
> >>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
> >>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
> >>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
> >>>>
> >>>> Average Optimal load -j 80
> >>>>                  Run    (std deviation)
> >>>>                  BASE                   SPF
> >>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
> >>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
> >>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
> >>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
> >>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
> >>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
> >>>>
> >>>> During a run on the SPF, perf events were captured:
> >>>>  Performance counter stats for '../kernbench -M':
> >>>>          116730856      faults
> >>>>                  0      spf
> >>>>                  3      pagefault:spf_vma_changed
> >>>>                  0      pagefault:spf_vma_noanon
> >>>>                476      pagefault:spf_vma_notsup
> >>>>                  0      pagefault:spf_vma_access
> >>>>                  0      pagefault:spf_pmd_changed
> >>>>
> >>>> Most of the processes involved are monothreaded so SPF is not activated but
> >>>> there is no impact on the performance.
> >>>>
> >>>> Ebizzy:
> >>>> -------
> >>>> The test is counting the number of records per second it can manage, the
> >>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
> >>>> consistent result I repeated the test 100 times and measure the average
> >>>> result. The number is the record processes per second, the higher is the
> >>>> best.
> >>>>
> >>>>                 BASE            SPF             delta
> >>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
> >>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
> >>>>
> >>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
> >>>>  Performance counter stats for './ebizzy -mTt 16':
> >>>>            1706379      faults
> >>>>            1674599      spf
> >>>>              30588      pagefault:spf_vma_changed
> >>>>                  0      pagefault:spf_vma_noanon
> >>>>                363      pagefault:spf_vma_notsup
> >>>>                  0      pagefault:spf_vma_access
> >>>>                  0      pagefault:spf_pmd_changed
> >>>>
> >>>> And the ones captured during a run on a 80 CPUs Power node:
> >>>>  Performance counter stats for './ebizzy -mTt 80':
> >>>>            1874773      faults
> >>>>            1461153      spf
> >>>>             413293      pagefault:spf_vma_changed
> >>>>                  0      pagefault:spf_vma_noanon
> >>>>                200      pagefault:spf_vma_notsup
> >>>>                  0      pagefault:spf_vma_access
> >>>>                  0      pagefault:spf_pmd_changed
> >>>>
> >>>> In ebizzy's case most of the page fault were handled in a speculative way,
> >>>> leading the ebizzy performance boost.
> >>>>
> >>>> ------------------
> >>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
> >>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
> >>>>    and Minchan Kim, hopefully.
> >>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
> >>>>    __do_page_fault().
> >>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
> >>>>    instead
> >>>>    of aborting the speculative page fault handling. Dropping the now
> >>>> useless
> >>>>    trace event pagefault:spf_pte_lock.
> >>>>  - No more try to reuse the fetched VMA during the speculative page fault
> >>>>    handling when retrying is needed. This adds a lot of complexity and
> >>>>    additional tests done didn't show a significant performance improvement.
> >>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
> >>>>
> >>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
> >>>> [2] https://patchwork.kernel.org/patch/9999687/
> >>>>
> >>>>
> >>>> Laurent Dufour (20):
> >>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
> >>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> >>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> >>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
> >>>>   mm: make pte_unmap_same compatible with SPF
> >>>>   mm: introduce INIT_VMA()
> >>>>   mm: protect VMA modifications using VMA sequence count
> >>>>   mm: protect mremap() against SPF hanlder
> >>>>   mm: protect SPF handler against anon_vma changes
> >>>>   mm: cache some VMA fields in the vm_fault structure
> >>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
> >>>>   mm: introduce __lru_cache_add_active_or_unevictable
> >>>>   mm: introduce __vm_normal_page()
> >>>>   mm: introduce __page_add_new_anon_rmap()
> >>>>   mm: protect mm_rb tree with a rwlock
> >>>>   mm: adding speculative page fault failure trace events
> >>>>   perf: add a speculative page fault sw event
> >>>>   perf tools: add support for the SPF perf event
> >>>>   mm: add speculative page fault vmstats
> >>>>   powerpc/mm: add speculative page fault
> >>>>
> >>>> Mahendran Ganesh (2):
> >>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
> >>>>   arm64/mm: add speculative page fault
> >>>>
> >>>> Peter Zijlstra (4):
> >>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
> >>>>   mm: VMA sequence count
> >>>>   mm: provide speculative fault infrastructure
> >>>>   x86/mm: add speculative pagefault handling
> >>>>
> >>>>  arch/arm64/Kconfig                    |   1 +
> >>>>  arch/arm64/mm/fault.c                 |  12 +
> >>>>  arch/powerpc/Kconfig                  |   1 +
> >>>>  arch/powerpc/mm/fault.c               |  16 +
> >>>>  arch/x86/Kconfig                      |   1 +
> >>>>  arch/x86/mm/fault.c                   |  27 +-
> >>>>  fs/exec.c                             |   2 +-
> >>>>  fs/proc/task_mmu.c                    |   5 +-
> >>>>  fs/userfaultfd.c                      |  17 +-
> >>>>  include/linux/hugetlb_inline.h        |   2 +-
> >>>>  include/linux/migrate.h               |   4 +-
> >>>>  include/linux/mm.h                    | 136 +++++++-
> >>>>  include/linux/mm_types.h              |   7 +
> >>>>  include/linux/pagemap.h               |   4 +-
> >>>>  include/linux/rmap.h                  |  12 +-
> >>>>  include/linux/swap.h                  |  10 +-
> >>>>  include/linux/vm_event_item.h         |   3 +
> >>>>  include/trace/events/pagefault.h      |  80 +++++
> >>>>  include/uapi/linux/perf_event.h       |   1 +
> >>>>  kernel/fork.c                         |   5 +-
> >>>>  mm/Kconfig                            |  22 ++
> >>>>  mm/huge_memory.c                      |   6 +-
> >>>>  mm/hugetlb.c                          |   2 +
> >>>>  mm/init-mm.c                          |   3 +
> >>>>  mm/internal.h                         |  20 ++
> >>>>  mm/khugepaged.c                       |   5 +
> >>>>  mm/madvise.c                          |   6 +-
> >>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
> >>>>  mm/mempolicy.c                        |  51 ++-
> >>>>  mm/migrate.c                          |   6 +-
> >>>>  mm/mlock.c                            |  13 +-
> >>>>  mm/mmap.c                             | 229 ++++++++++---
> >>>>  mm/mprotect.c                         |   4 +-
> >>>>  mm/mremap.c                           |  13 +
> >>>>  mm/nommu.c                            |   2 +-
> >>>>  mm/rmap.c                             |   5 +-
> >>>>  mm/swap.c                             |   6 +-
> >>>>  mm/swap_state.c                       |   8 +-
> >>>>  mm/vmstat.c                           |   5 +-
> >>>>  tools/include/uapi/linux/perf_event.h |   1 +
> >>>>  tools/perf/util/evsel.c               |   1 +
> >>>>  tools/perf/util/parse-events.c        |   4 +
> >>>>  tools/perf/util/parse-events.l        |   1 +
> >>>>  tools/perf/util/python.c              |   1 +
> >>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
> >>>>  create mode 100644 include/trace/events/pagefault.h
> >>>>
> >>>> --
> >>>> 2.7.4
> >>>>
> >>>>
> >>>
> >>
> > 
> 

[-- Attachment #2: compare-result.txt --]
[-- Type: text/plain, Size: 183477 bytes --]

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/always/page_fault3/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
         44:3          -13%          43:3     perf-profile.calltrace.cycles-pp.error_entry
         22:3           -6%          22:3     perf-profile.calltrace.cycles-pp.sync_regs.error_entry
         44:3          -13%          44:3     perf-profile.children.cycles-pp.error_entry
         21:3           -7%          21:3     perf-profile.self.cycles-pp.error_entry
         %stddev     %change         %stddev
             \          |                \  
     10519 ±  3%     -20.5%       8368 ±  6%  will-it-scale.per_thread_ops
    118098           +11.2%     131287 ±  2%  will-it-scale.time.involuntary_context_switches
 6.084e+08 ±  3%     -20.4%  4.845e+08 ±  6%  will-it-scale.time.minor_page_faults
      7467            +5.0%       7841        will-it-scale.time.percent_of_cpu_this_job_got
     44922            +5.0%      47176        will-it-scale.time.system_time
   7126337 ±  3%     -15.4%    6025689 ±  6%  will-it-scale.time.voluntary_context_switches
  91905646            -1.3%   90673935        will-it-scale.workload
     27.15 ±  6%      -8.7%      24.80 ± 10%  boot-time.boot
   2516213 ±  6%      +8.3%    2726303        interrupts.CAL:Function_call_interrupts
    388.00 ±  9%     +60.2%     621.67 ± 20%  irq_exception_noise.softirq_nr
     11.28 ±  2%      -1.9        9.37 ±  4%  mpstat.cpu.idle%
     10065 ±140%    +243.4%      34559 ±  4%  numa-numastat.node0.other_node
     18739           -11.6%      16573 ±  3%  uptime.idle
     29406 ±  2%     -11.8%      25929 ±  5%  vmstat.system.cs
    329614 ±  8%     +17.0%     385618 ± 10%  meminfo.DirectMap4k
    237851           +21.2%     288160 ±  5%  meminfo.Inactive
    237615           +21.2%     287924 ±  5%  meminfo.Inactive(anon)
   7917847           -10.7%    7071860        softirqs.RCU
   4784181 ±  3%     -14.5%    4089039 ±  4%  softirqs.SCHED
  45666107 ±  7%     +12.9%   51535472 ±  3%  softirqs.TIMER
 2.617e+09 ±  2%     -13.9%  2.253e+09 ±  6%  cpuidle.C1E.time
   6688774 ±  2%     -12.8%    5835101 ±  5%  cpuidle.C1E.usage
 1.022e+10 ±  2%     -18.0%  8.376e+09 ±  3%  cpuidle.C6.time
  13440993 ±  2%     -16.3%   11243794 ±  4%  cpuidle.C6.usage
     54781 ± 16%     +37.5%      75347 ± 12%  numa-meminfo.node0.Inactive
     54705 ± 16%     +37.7%      75347 ± 12%  numa-meminfo.node0.Inactive(anon)
     52522           +35.0%      70886 ±  6%  numa-meminfo.node2.Inactive
     52443           +34.7%      70653 ±  6%  numa-meminfo.node2.Inactive(anon)
     31046 ±  6%     +30.3%      40457 ± 11%  numa-meminfo.node2.SReclaimable
     58563           +21.1%      70945 ±  6%  proc-vmstat.nr_inactive_anon
     58564           +21.1%      70947 ±  6%  proc-vmstat.nr_zone_inactive_anon
  69701118            -1.2%   68842151        proc-vmstat.pgalloc_normal
 2.765e+10            -1.3%  2.729e+10        proc-vmstat.pgfault
  69330418            -1.2%   68466824        proc-vmstat.pgfree
    118098           +11.2%     131287 ±  2%  time.involuntary_context_switches
 6.084e+08 ±  3%     -20.4%  4.845e+08 ±  6%  time.minor_page_faults
      7467            +5.0%       7841        time.percent_of_cpu_this_job_got
     44922            +5.0%      47176        time.system_time
   7126337 ±  3%     -15.4%    6025689 ±  6%  time.voluntary_context_switches
     13653 ± 16%     +33.5%      18225 ± 12%  numa-vmstat.node0.nr_inactive_anon
     13651 ± 16%     +33.5%      18224 ± 12%  numa-vmstat.node0.nr_zone_inactive_anon
     13069 ±  3%     +30.1%      17001 ±  4%  numa-vmstat.node2.nr_inactive_anon
    134.67 ± 42%     -49.5%      68.00 ± 31%  numa-vmstat.node2.nr_mlock
      7758 ±  6%     +30.4%      10112 ± 11%  numa-vmstat.node2.nr_slab_reclaimable
     13066 ±  3%     +30.1%      16998 ±  4%  numa-vmstat.node2.nr_zone_inactive_anon
      1039 ± 11%     -17.5%     857.33        slabinfo.Acpi-ParseExt.active_objs
      1039 ± 11%     -17.5%     857.33        slabinfo.Acpi-ParseExt.num_objs
      2566 ±  6%      -8.8%       2340 ±  5%  slabinfo.biovec-64.active_objs
      2566 ±  6%      -8.8%       2340 ±  5%  slabinfo.biovec-64.num_objs
    898.33 ±  3%      -9.5%     813.33 ±  3%  slabinfo.kmem_cache_node.active_objs
      1066 ±  2%      -8.0%     981.33 ±  3%  slabinfo.kmem_cache_node.num_objs
      1940            +2.3%       1984        turbostat.Avg_MHz
   6679037 ±  2%     -12.7%    5830270 ±  5%  turbostat.C1E
      2.25 ±  2%      -0.3        1.94 ±  6%  turbostat.C1E%
  13418115           -16.3%   11234510 ±  4%  turbostat.C6
      8.75 ±  2%      -1.6        7.18 ±  3%  turbostat.C6%
      5.99 ±  2%     -14.4%       5.13 ±  4%  turbostat.CPU%c1
      5.01 ±  3%     -20.1%       4.00 ±  4%  turbostat.CPU%c6
      1.77 ±  3%     -34.7%       1.15        turbostat.Pkg%pc2
 1.378e+13            +1.2%  1.394e+13        perf-stat.branch-instructions
      0.98            -0.0        0.94        perf-stat.branch-miss-rate%
 1.344e+11            -2.3%  1.313e+11        perf-stat.branch-misses
 1.076e+11            -1.8%  1.057e+11        perf-stat.cache-misses
 2.258e+11            -2.1%   2.21e+11        perf-stat.cache-references
  17788064 ±  2%     -11.9%   15674207 ±  6%  perf-stat.context-switches
 2.241e+14            +2.4%  2.294e+14        perf-stat.cpu-cycles
 1.929e+13            +2.2%  1.971e+13        perf-stat.dTLB-loads
      4.01            -0.2        3.83        perf-stat.dTLB-store-miss-rate%
 4.519e+11            -1.3%  4.461e+11        perf-stat.dTLB-store-misses
 1.082e+13            +3.6%  1.121e+13        perf-stat.dTLB-stores
  3.02e+10           +23.2%  3.721e+10 ±  3%  perf-stat.iTLB-load-misses
 2.721e+08 ±  8%      -8.8%  2.481e+08 ±  3%  perf-stat.iTLB-loads
 6.985e+13            +1.8%  7.111e+13        perf-stat.instructions
      2313           -17.2%       1914 ±  3%  perf-stat.instructions-per-iTLB-miss
 2.764e+10            -1.3%  2.729e+10        perf-stat.minor-faults
 1.421e+09 ±  2%     -16.4%  1.188e+09 ±  9%  perf-stat.node-load-misses
 1.538e+10            -9.3%  1.395e+10        perf-stat.node-loads
      9.75            +1.4       11.10        perf-stat.node-store-miss-rate%
 3.012e+09           +14.1%  3.437e+09        perf-stat.node-store-misses
 2.789e+10            -1.3%  2.753e+10        perf-stat.node-stores
 2.764e+10            -1.3%  2.729e+10        perf-stat.page-faults
    760059            +3.2%     784235        perf-stat.path-length
    193545 ± 25%     -57.8%      81757 ± 46%  sched_debug.cfs_rq:/.MIN_vruntime.avg
  26516863 ± 19%     -49.7%   13338070 ± 33%  sched_debug.cfs_rq:/.MIN_vruntime.max
   2202271 ± 21%     -53.2%    1029581 ± 38%  sched_debug.cfs_rq:/.MIN_vruntime.stddev
    193545 ± 25%     -57.8%      81757 ± 46%  sched_debug.cfs_rq:/.max_vruntime.avg
  26516863 ± 19%     -49.7%   13338070 ± 33%  sched_debug.cfs_rq:/.max_vruntime.max
   2202271 ± 21%     -53.2%    1029581 ± 38%  sched_debug.cfs_rq:/.max_vruntime.stddev
      0.32 ± 70%    +253.2%       1.14 ± 54%  sched_debug.cfs_rq:/.removed.load_avg.avg
      4.44 ± 70%    +120.7%       9.80 ± 27%  sched_debug.cfs_rq:/.removed.load_avg.stddev
     14.90 ± 70%    +251.0%      52.31 ± 53%  sched_debug.cfs_rq:/.removed.runnable_sum.avg
    205.71 ± 70%    +119.5%     451.60 ± 27%  sched_debug.cfs_rq:/.removed.runnable_sum.stddev
      0.16 ± 70%    +237.9%       0.54 ± 50%  sched_debug.cfs_rq:/.removed.util_avg.avg
      2.23 ± 70%    +114.2%       4.77 ± 24%  sched_debug.cfs_rq:/.removed.util_avg.stddev
    573.70 ±  5%      -9.7%     518.06 ±  6%  sched_debug.cfs_rq:/.util_avg.min
    114.87 ±  8%     +14.1%     131.04 ± 10%  sched_debug.cfs_rq:/.util_est_enqueued.avg
     64.42 ± 54%     -63.9%      23.27 ± 68%  sched_debug.cpu.cpu_load[1].max
      5.05 ± 48%     -55.2%       2.26 ± 51%  sched_debug.cpu.cpu_load[1].stddev
     57.58 ± 59%     -60.3%      22.88 ± 70%  sched_debug.cpu.cpu_load[2].max
     21019 ±  3%     -15.1%      17841 ±  6%  sched_debug.cpu.nr_switches.min
     20797 ±  3%     -15.0%      17670 ±  6%  sched_debug.cpu.sched_count.min
     10287 ±  3%     -15.1%       8736 ±  6%  sched_debug.cpu.sched_goidle.avg
     13693 ±  2%     -10.7%      12233 ±  5%  sched_debug.cpu.sched_goidle.max
      9976 ±  3%     -16.0%       8381 ±  7%  sched_debug.cpu.sched_goidle.min
      0.00 ± 26%     +98.9%       0.00 ± 28%  sched_debug.rt_rq:/.rt_time.min
      4230 ±141%    -100.0%       0.00        latency_stats.avg.trace_module_notify.notifier_call_chain.blocking_notifier_call_chain.do_init_module.load_module.__do_sys_finit_module.do_syscall_64.entry_SYSCALL_64_after_hwframe
     28498 ±141%    -100.0%       0.00        latency_stats.avg.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4065 ±138%     -92.2%     315.33 ± 91%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
      0.00       +3.6e+105%       3641 ±141%  latency_stats.avg.down.console_lock.console_device.tty_lookup_driver.tty_open.chrdev_open.do_dentry_open.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +2.5e+106%      25040 ±141%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +3.4e+106%      34015 ±141%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_get_acl.get_acl.posix_acl_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open
      0.00       +4.8e+106%      47686 ±141%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
      4230 ±141%    -100.0%       0.00        latency_stats.max.trace_module_notify.notifier_call_chain.blocking_notifier_call_chain.do_init_module.load_module.__do_sys_finit_module.do_syscall_64.entry_SYSCALL_64_after_hwframe
     28498 ±141%    -100.0%       0.00        latency_stats.max.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4065 ±138%     -92.2%     315.33 ± 91%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
      4254 ±134%     -88.0%     511.67 ± 90%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
     43093 ± 35%     +76.6%      76099 ±115%  latency_stats.max.blk_execute_rq.scsi_execute.ioctl_internal_command.scsi_set_medium_removal.cdrom_release.[cdrom].sr_block_release.[sr_mod].__blkdev_put.blkdev_close.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64
     24139 ± 70%    +228.5%      79285 ±105%  latency_stats.max.blk_execute_rq.scsi_execute.scsi_test_unit_ready.sr_check_events.[sr_mod].cdrom_check_events.[cdrom].sr_block_check_events.[sr_mod].disk_check_events.disk_clear_events.check_disk_change.sr_block_open.[sr_mod].__blkdev_get.blkdev_get
      0.00       +3.6e+105%       3641 ±141%  latency_stats.max.down.console_lock.console_device.tty_lookup_driver.tty_open.chrdev_open.do_dentry_open.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +2.5e+106%      25040 ±141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +3.4e+106%      34015 ±141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_get_acl.get_acl.posix_acl_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open
      0.00       +6.5e+106%      64518 ±141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
      4230 ±141%    -100.0%       0.00        latency_stats.sum.trace_module_notify.notifier_call_chain.blocking_notifier_call_chain.do_init_module.load_module.__do_sys_finit_module.do_syscall_64.entry_SYSCALL_64_after_hwframe
     28498 ±141%    -100.0%       0.00        latency_stats.sum.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4065 ±138%     -92.2%     315.33 ± 91%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
     57884 ±  9%     +47.3%      85264 ±118%  latency_stats.sum.blk_execute_rq.scsi_execute.ioctl_internal_command.scsi_set_medium_removal.cdrom_release.[cdrom].sr_block_release.[sr_mod].__blkdev_put.blkdev_close.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64
      0.00       +3.6e+105%       3641 ±141%  latency_stats.sum.down.console_lock.console_device.tty_lookup_driver.tty_open.chrdev_open.do_dentry_open.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +2.5e+106%      25040 ±141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +3.4e+106%      34015 ±141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_get_acl.get_acl.posix_acl_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open
      0.00       +9.5e+106%      95373 ±141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
     11.70           -11.7        0.00        perf-profile.calltrace.cycles-pp.__do_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
     11.52           -11.5        0.00        perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
     10.44           -10.4        0.00        perf-profile.calltrace.cycles-pp.shmem_getpage_gfp.shmem_fault.__do_fault.__handle_mm_fault.handle_mm_fault
      9.83            -9.8        0.00        perf-profile.calltrace.cycles-pp.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault.__handle_mm_fault
      9.55            -9.5        0.00        perf-profile.calltrace.cycles-pp.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      9.35            -9.3        0.00        perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      6.81            -6.8        0.00        perf-profile.calltrace.cycles-pp.page_add_file_rmap.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault
      7.71            -0.3        7.45        perf-profile.calltrace.cycles-pp.find_get_entry.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault
      0.59 ±  7%      -0.2        0.35 ± 70%  perf-profile.calltrace.cycles-pp.smp_apic_timer_interrupt.apic_timer_interrupt.__do_page_fault.do_page_fault.page_fault
      0.59 ±  7%      -0.2        0.35 ± 70%  perf-profile.calltrace.cycles-pp.apic_timer_interrupt.__do_page_fault.do_page_fault.page_fault
     10.41            -0.2       10.24        perf-profile.calltrace.cycles-pp.native_irq_return_iret
      7.68            -0.1        7.60        perf-profile.calltrace.cycles-pp.swapgs_restore_regs_and_return_to_usermode
      0.76            -0.1        0.70        perf-profile.calltrace.cycles-pp.down_read_trylock.__do_page_fault.do_page_fault.page_fault
      1.38            -0.0        1.34        perf-profile.calltrace.cycles-pp.do_page_fault
      1.05            -0.0        1.02        perf-profile.calltrace.cycles-pp.trace_graph_entry.do_page_fault
      0.92            +0.0        0.94        perf-profile.calltrace.cycles-pp.find_vma.__do_page_fault.do_page_fault.page_fault
      0.91            +0.0        0.93        perf-profile.calltrace.cycles-pp.vmacache_find.find_vma.__do_page_fault.do_page_fault.page_fault
      0.65            +0.0        0.67        perf-profile.calltrace.cycles-pp.set_page_dirty.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      0.62            +0.0        0.66        perf-profile.calltrace.cycles-pp.page_mapping.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault
      4.15            +0.1        4.27        perf-profile.calltrace.cycles-pp.page_remove_rmap.unmap_page_range.unmap_vmas.unmap_region.do_munmap
     10.17            +0.2       10.39        perf-profile.calltrace.cycles-pp.munmap
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.munmap
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.munmap
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.unmap_region.do_munmap.vm_munmap.__x64_sys_munmap.do_syscall_64
      9.54            +0.2        9.76        perf-profile.calltrace.cycles-pp.unmap_page_range.unmap_vmas.unmap_region.do_munmap.vm_munmap
      9.54            +0.2        9.76        perf-profile.calltrace.cycles-pp.unmap_vmas.unmap_region.do_munmap.vm_munmap.__x64_sys_munmap
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.do_munmap.vm_munmap.__x64_sys_munmap.do_syscall_64.entry_SYSCALL_64_after_hwframe
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.vm_munmap.__x64_sys_munmap.do_syscall_64.entry_SYSCALL_64_after_hwframe.munmap
      9.56            +0.2        9.78        perf-profile.calltrace.cycles-pp.__x64_sys_munmap.do_syscall_64.entry_SYSCALL_64_after_hwframe.munmap
      0.00            +0.6        0.56 ±  2%  perf-profile.calltrace.cycles-pp.lock_page_memcg.page_add_file_rmap.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +0.6        0.59        perf-profile.calltrace.cycles-pp.page_mapping.set_page_dirty.fault_dirty_shared_page.handle_pte_fault.__handle_mm_fault
      0.00            +0.6        0.60        perf-profile.calltrace.cycles-pp.current_time.file_update_time.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +0.7        0.68        perf-profile.calltrace.cycles-pp.___might_sleep.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault
      0.00            +0.7        0.74        perf-profile.calltrace.cycles-pp.unlock_page.fault_dirty_shared_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +0.8        0.80        perf-profile.calltrace.cycles-pp.set_page_dirty.fault_dirty_shared_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +0.9        0.88        perf-profile.calltrace.cycles-pp._raw_spin_lock.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +0.9        0.91        perf-profile.calltrace.cycles-pp.__set_page_dirty_no_writeback.fault_dirty_shared_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +1.3        1.27        perf-profile.calltrace.cycles-pp.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +1.3        1.30        perf-profile.calltrace.cycles-pp.file_update_time.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +2.8        2.76        perf-profile.calltrace.cycles-pp.fault_dirty_shared_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +6.8        6.81        perf-profile.calltrace.cycles-pp.page_add_file_rmap.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +9.4        9.39        perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +9.6        9.59        perf-profile.calltrace.cycles-pp.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +9.8        9.77        perf-profile.calltrace.cycles-pp.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault
      0.00           +10.4       10.37        perf-profile.calltrace.cycles-pp.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault
      0.00           +11.5       11.46        perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00           +11.6       11.60        perf-profile.calltrace.cycles-pp.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00           +26.6       26.62        perf-profile.calltrace.cycles-pp.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      7.88            -0.3        7.61        perf-profile.children.cycles-pp.find_get_entry
      1.34 ±  8%      -0.2        1.16 ±  2%  perf-profile.children.cycles-pp.hrtimer_interrupt
     10.41            -0.2       10.24        perf-profile.children.cycles-pp.native_irq_return_iret
      0.38 ± 28%      -0.1        0.26 ±  4%  perf-profile.children.cycles-pp.tick_sched_timer
     11.80            -0.1       11.68        perf-profile.children.cycles-pp.__do_fault
      0.55 ± 15%      -0.1        0.43 ±  2%  perf-profile.children.cycles-pp.__hrtimer_run_queues
      0.60            -0.1        0.51        perf-profile.children.cycles-pp.pmd_devmap_trans_unstable
      0.38 ± 13%      -0.1        0.29 ±  4%  perf-profile.children.cycles-pp.ktime_get
      7.68            -0.1        7.60        perf-profile.children.cycles-pp.swapgs_restore_regs_and_return_to_usermode
      5.18            -0.1        5.12        perf-profile.children.cycles-pp.trace_graph_entry
      0.79            -0.1        0.73        perf-profile.children.cycles-pp.down_read_trylock
      7.83            -0.1        7.76        perf-profile.children.cycles-pp.sync_regs
      3.01            -0.1        2.94        perf-profile.children.cycles-pp.fault_dirty_shared_page
      1.02            -0.1        0.96        perf-profile.children.cycles-pp._raw_spin_lock
      4.66            -0.1        4.61        perf-profile.children.cycles-pp.prepare_ftrace_return
      0.37 ±  8%      -0.1        0.32 ±  3%  perf-profile.children.cycles-pp.current_kernel_time64
      5.26            -0.1        5.21        perf-profile.children.cycles-pp.ftrace_graph_caller
      0.66 ±  5%      -0.1        0.61        perf-profile.children.cycles-pp.current_time
      0.18 ±  5%      -0.0        0.15 ±  3%  perf-profile.children.cycles-pp.update_process_times
      0.27            -0.0        0.26        perf-profile.children.cycles-pp._cond_resched
      0.16            -0.0        0.15 ±  3%  perf-profile.children.cycles-pp.rcu_all_qs
      0.94            +0.0        0.95        perf-profile.children.cycles-pp.vmacache_find
      0.48            +0.0        0.50        perf-profile.children.cycles-pp.__mod_node_page_state
      0.17            +0.0        0.19 ±  2%  perf-profile.children.cycles-pp.__unlock_page_memcg
      1.07            +0.0        1.10        perf-profile.children.cycles-pp.find_vma
      0.79 ±  3%      +0.1        0.86 ±  2%  perf-profile.children.cycles-pp.lock_page_memcg
      4.29            +0.1        4.40        perf-profile.children.cycles-pp.page_remove_rmap
      1.39 ±  2%      +0.1        1.52        perf-profile.children.cycles-pp.file_update_time
      0.00            +0.2        0.16        perf-profile.children.cycles-pp.__vm_normal_page
      9.63            +0.2        9.84        perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
      9.63            +0.2        9.84        perf-profile.children.cycles-pp.do_syscall_64
      9.63            +0.2        9.84        perf-profile.children.cycles-pp.unmap_page_range
     10.17            +0.2       10.39        perf-profile.children.cycles-pp.munmap
      9.56            +0.2        9.78        perf-profile.children.cycles-pp.unmap_region
      9.56            +0.2        9.78        perf-profile.children.cycles-pp.do_munmap
      9.56            +0.2        9.78        perf-profile.children.cycles-pp.vm_munmap
      9.56            +0.2        9.78        perf-profile.children.cycles-pp.__x64_sys_munmap
      9.54            +0.2        9.77        perf-profile.children.cycles-pp.unmap_vmas
      1.01            +0.2        1.25        perf-profile.children.cycles-pp.___might_sleep
      0.00            +1.6        1.59        perf-profile.children.cycles-pp.pte_map_lock
      0.00           +26.9       26.89        perf-profile.children.cycles-pp.handle_pte_fault
      4.25            -1.0        3.24        perf-profile.self.cycles-pp.__handle_mm_fault
      1.42            -0.3        1.11        perf-profile.self.cycles-pp.alloc_set_pte
      4.87            -0.3        4.59        perf-profile.self.cycles-pp.find_get_entry
     10.41            -0.2       10.24        perf-profile.self.cycles-pp.native_irq_return_iret
      0.37 ± 13%      -0.1        0.28 ±  4%  perf-profile.self.cycles-pp.ktime_get
      0.60            -0.1        0.51        perf-profile.self.cycles-pp.pmd_devmap_trans_unstable
      7.50            -0.1        7.42        perf-profile.self.cycles-pp.swapgs_restore_regs_and_return_to_usermode
      7.83            -0.1        7.76        perf-profile.self.cycles-pp.sync_regs
      4.85            -0.1        4.79        perf-profile.self.cycles-pp.trace_graph_entry
      1.01            -0.1        0.95        perf-profile.self.cycles-pp._raw_spin_lock
      0.78            -0.1        0.73        perf-profile.self.cycles-pp.down_read_trylock
      0.36 ±  9%      -0.1        0.31 ±  4%  perf-profile.self.cycles-pp.current_kernel_time64
      0.28            -0.0        0.23 ±  2%  perf-profile.self.cycles-pp.__do_fault
      1.04            -0.0        1.00        perf-profile.self.cycles-pp.find_lock_entry
      0.30            -0.0        0.28 ±  3%  perf-profile.self.cycles-pp.fault_dirty_shared_page
      0.70            -0.0        0.67        perf-profile.self.cycles-pp.prepare_ftrace_return
      0.44            -0.0        0.42        perf-profile.self.cycles-pp.do_page_fault
      0.16            -0.0        0.14        perf-profile.self.cycles-pp.rcu_all_qs
      0.78            -0.0        0.77        perf-profile.self.cycles-pp.shmem_getpage_gfp
      0.20            -0.0        0.19        perf-profile.self.cycles-pp._cond_resched
      0.50            +0.0        0.51        perf-profile.self.cycles-pp.set_page_dirty
      0.93            +0.0        0.95        perf-profile.self.cycles-pp.vmacache_find
      0.36 ±  2%      +0.0        0.38        perf-profile.self.cycles-pp.__might_sleep
      0.47            +0.0        0.50        perf-profile.self.cycles-pp.__mod_node_page_state
      0.17            +0.0        0.19 ±  2%  perf-profile.self.cycles-pp.__unlock_page_memcg
      2.34            +0.0        2.38        perf-profile.self.cycles-pp.unmap_page_range
      0.78 ±  3%      +0.1        0.85 ±  2%  perf-profile.self.cycles-pp.lock_page_memcg
      2.17            +0.1        2.24        perf-profile.self.cycles-pp.__do_page_fault
      0.00            +0.2        0.16 ±  3%  perf-profile.self.cycles-pp.__vm_normal_page
      1.00            +0.2        1.24        perf-profile.self.cycles-pp.___might_sleep
      0.00            +0.7        0.70        perf-profile.self.cycles-pp.pte_map_lock
      0.00            +1.4        1.42 ±  2%  perf-profile.self.cycles-pp.handle_pte_fault

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/never/context_switch1/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
           :3           33%           1:3     dmesg.WARNING:at#for_ip_interrupt_entry/0x
          2:3          -67%            :3     kmsg.pstore:crypto_comp_decompress_failed,ret=
          2:3          -67%            :3     kmsg.pstore:decompression_failed
         %stddev     %change         %stddev
             \          |                \  
    224431            -1.3%     221567        will-it-scale.per_process_ops
    237006            -2.2%     231907        will-it-scale.per_thread_ops
 1.601e+09 ± 29%     -46.9%  8.501e+08 ± 12%  will-it-scale.time.involuntary_context_switches
      5429            -1.6%       5344        will-it-scale.time.user_time
  88596221            -1.7%   87067269        will-it-scale.workload
      6863 ±  6%      -9.7%       6200        boot-time.idle
    144908 ± 40%     -66.8%      48173 ± 93%  meminfo.CmaFree
      0.00 ± 70%      +0.0        0.00        mpstat.cpu.iowait%
    448336 ± 14%     -34.8%     292125 ±  3%  turbostat.C1
      7684 ±  6%      -9.5%       6957        uptime.idle
 1.601e+09 ± 29%     -46.9%  8.501e+08 ± 12%  time.involuntary_context_switches
      5429            -1.6%       5344        time.user_time
  44013162            -1.7%   43243125        vmstat.system.cs
    207684            -1.1%     205485        vmstat.system.in
   2217033 ± 15%     -15.8%    1866876 ±  2%  cpuidle.C1.time
    451218 ± 14%     -34.7%     294841 ±  2%  cpuidle.C1.usage
     24839 ± 10%     -19.9%      19896        cpuidle.POLL.time
      7656 ± 11%     -38.9%       4676 ±  8%  cpuidle.POLL.usage
      5.48 ± 49%     -67.3%       1.79 ±100%  irq_exception_noise.__do_page_fault.95th
      9.46 ± 21%     -58.2%       3.95 ± 64%  irq_exception_noise.__do_page_fault.99th
     35.67 ±  8%   +1394.4%     533.00 ± 96%  irq_exception_noise.irq_nr
     52109 ±  3%     -16.0%      43784 ±  4%  irq_exception_noise.softirq_time
     36226 ± 40%     -66.7%      12048 ± 93%  proc-vmstat.nr_free_cma
     25916            -1.0%      25659        proc-vmstat.nr_slab_reclaimable
     16279 ±  8%   +2646.1%     447053 ± 82%  proc-vmstat.pgalloc_movable
   2231117           -18.4%    1820828 ± 20%  proc-vmstat.pgalloc_normal
   1109316 ± 46%     -86.9%     145207 ±109%  numa-numastat.node1.local_node
   1114700 ± 45%     -84.5%     172877 ± 85%  numa-numastat.node1.numa_hit
      5523 ±140%    +402.8%      27768 ± 39%  numa-numastat.node1.other_node
     29013 ± 29%   +3048.1%     913379 ± 73%  numa-numastat.node3.local_node
     65032 ± 13%   +1335.1%     933270 ± 70%  numa-numastat.node3.numa_hit
     36018           -44.8%      19897 ± 75%  numa-numastat.node3.other_node
     12.79 ± 21%   +7739.1%       1002 ±136%  sched_debug.cpu.cpu_load[1].max
      1.82 ± 10%   +3901.1%      72.92 ±135%  sched_debug.cpu.cpu_load[1].stddev
      1.71 ±  4%   +5055.8%      88.08 ±137%  sched_debug.cpu.cpu_load[2].stddev
     12.33 ± 23%   +9061.9%       1129 ±139%  sched_debug.cpu.cpu_load[3].max
      1.78 ± 10%   +4514.8%      82.18 ±137%  sched_debug.cpu.cpu_load[3].stddev
      4692 ± 72%    +154.5%      11945 ± 29%  sched_debug.cpu.max_idle_balance_cost.stddev
     23979            -8.3%      21983        slabinfo.kmalloc-96.active_objs
      1358 ±  6%     -17.9%       1114 ±  3%  slabinfo.nsproxy.active_objs
      1358 ±  6%     -17.9%       1114 ±  3%  slabinfo.nsproxy.num_objs
     15229           +12.4%      17119        slabinfo.pde_opener.active_objs
     15229           +12.4%      17119        slabinfo.pde_opener.num_objs
     59541 ±  8%     -10.1%      53537 ±  8%  slabinfo.vm_area_struct.active_objs
     59612 ±  8%     -10.1%      53604 ±  8%  slabinfo.vm_area_struct.num_objs
 4.163e+13            -1.4%  4.105e+13        perf-stat.branch-instructions
 6.537e+11            -1.2%  6.459e+11        perf-stat.branch-misses
 2.667e+10            -1.7%  2.621e+10        perf-stat.context-switches
      1.21            +1.3%       1.22        perf-stat.cpi
    150508            -9.8%     135825 ±  3%  perf-stat.cpu-migrations
      5.75 ± 33%      +5.4       11.11 ± 26%  perf-stat.iTLB-load-miss-rate%
 3.619e+09 ± 36%    +100.9%  7.272e+09 ± 30%  perf-stat.iTLB-load-misses
 2.089e+14            -1.3%  2.062e+14        perf-stat.instructions
     64607 ± 29%     -50.5%      31964 ± 37%  perf-stat.instructions-per-iTLB-miss
      0.83            -1.3%       0.82        perf-stat.ipc
      3972 ±  4%     -14.7%       3388 ±  8%  numa-meminfo.node0.PageTables
    207919 ± 25%     -57.2%      88989 ± 74%  numa-meminfo.node1.Active
    207715 ± 26%     -57.3%      88785 ± 74%  numa-meminfo.node1.Active(anon)
    356529           -34.3%     234069 ±  2%  numa-meminfo.node1.FilePages
    789129 ±  5%     -19.8%     633161 ± 12%  numa-meminfo.node1.MemUsed
     34777 ±  8%     -48.2%      18010 ± 30%  numa-meminfo.node1.SReclaimable
     69641 ±  4%     -20.7%      55250 ± 12%  numa-meminfo.node1.SUnreclaim
    125526 ±  4%     -96.3%       4602 ± 41%  numa-meminfo.node1.Shmem
    104419           -29.8%      73261 ± 16%  numa-meminfo.node1.Slab
    103661 ± 17%     -72.0%      29029 ± 99%  numa-meminfo.node2.Active
    103661 ± 17%     -72.2%      28829 ±101%  numa-meminfo.node2.Active(anon)
    103564 ± 18%     -72.0%      29007 ±100%  numa-meminfo.node2.AnonPages
    671654 ±  7%     -14.6%     573598 ±  4%  numa-meminfo.node2.MemUsed
     44206 ±127%    +301.4%     177465 ± 42%  numa-meminfo.node3.Active
     44206 ±127%    +301.0%     177263 ± 42%  numa-meminfo.node3.Active(anon)
      8738           +12.2%       9805 ±  8%  numa-meminfo.node3.KernelStack
    603605 ±  9%     +27.8%     771554 ± 14%  numa-meminfo.node3.MemUsed
     14438 ±  6%    +122.9%      32181 ± 42%  numa-meminfo.node3.SReclaimable
      2786 ±137%   +3302.0%      94792 ± 71%  numa-meminfo.node3.Shmem
     71461 ±  7%     +45.2%     103771 ± 29%  numa-meminfo.node3.Slab
    247197 ±  4%      -7.8%     227843        numa-meminfo.node3.Unevictable
    991.67 ±  4%     -14.7%     846.00 ±  8%  numa-vmstat.node0.nr_page_table_pages
     51926 ± 26%     -57.3%      22196 ± 74%  numa-vmstat.node1.nr_active_anon
     89137           -34.4%      58516 ±  2%  numa-vmstat.node1.nr_file_pages
      1679 ±  5%     -10.8%       1498 ±  4%  numa-vmstat.node1.nr_mapped
     31386 ±  4%     -96.3%       1150 ± 41%  numa-vmstat.node1.nr_shmem
      8694 ±  8%     -48.2%       4502 ± 30%  numa-vmstat.node1.nr_slab_reclaimable
     17410 ±  4%     -20.7%      13812 ± 12%  numa-vmstat.node1.nr_slab_unreclaimable
     51926 ± 26%     -57.3%      22196 ± 74%  numa-vmstat.node1.nr_zone_active_anon
   1037174 ± 24%     -57.0%     446205 ± 35%  numa-vmstat.node1.numa_hit
    961611 ± 26%     -65.8%     328687 ± 50%  numa-vmstat.node1.numa_local
     75563 ± 44%     +55.5%     117517 ±  9%  numa-vmstat.node1.numa_other
     25914 ± 17%     -72.2%       7206 ±101%  numa-vmstat.node2.nr_active_anon
     25891 ± 18%     -72.0%       7251 ±100%  numa-vmstat.node2.nr_anon_pages
     25914 ± 17%     -72.2%       7206 ±101%  numa-vmstat.node2.nr_zone_active_anon
     11051 ±127%    +301.0%      44309 ± 42%  numa-vmstat.node3.nr_active_anon
     36227 ± 40%     -66.7%      12049 ± 93%  numa-vmstat.node3.nr_free_cma
      0.33 ±141%  +25000.0%      83.67 ± 81%  numa-vmstat.node3.nr_inactive_file
      8739           +12.2%       9806 ±  8%  numa-vmstat.node3.nr_kernel_stack
    696.67 ±137%   +3299.7%      23684 ± 71%  numa-vmstat.node3.nr_shmem
      3609 ±  6%    +122.9%       8044 ± 42%  numa-vmstat.node3.nr_slab_reclaimable
     61799 ±  4%      -7.8%      56960        numa-vmstat.node3.nr_unevictable
     11053 ±127%    +301.4%      44361 ± 42%  numa-vmstat.node3.nr_zone_active_anon
      0.33 ±141%  +25000.0%      83.67 ± 81%  numa-vmstat.node3.nr_zone_inactive_file
     61799 ±  4%      -7.8%      56960        numa-vmstat.node3.nr_zone_unevictable
    217951 ±  8%    +280.8%     829976 ± 65%  numa-vmstat.node3.numa_hit
     91303 ± 19%    +689.3%     720647 ± 77%  numa-vmstat.node3.numa_local
    126648           -13.7%     109329 ± 13%  numa-vmstat.node3.numa_other
      8.54            -0.1        8.40        perf-profile.calltrace.cycles-pp.dequeue_task_fair.__schedule.schedule.pipe_wait.pipe_read
      5.04            -0.1        4.94        perf-profile.calltrace.cycles-pp.__switch_to.read
      3.43            -0.1        3.35        perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.write
      2.77            -0.1        2.72        perf-profile.calltrace.cycles-pp.reweight_entity.enqueue_task_fair.ttwu_do_activate.try_to_wake_up.autoremove_wake_function
      1.99            -0.0        1.94        perf-profile.calltrace.cycles-pp.copy_page_to_iter.pipe_read.__vfs_read.vfs_read.ksys_read
      0.60 ±  2%      -0.0        0.57 ±  2%  perf-profile.calltrace.cycles-pp.find_next_bit.cpumask_next_wrap.select_idle_sibling.select_task_rq_fair.try_to_wake_up
      0.81            -0.0        0.78        perf-profile.calltrace.cycles-pp.___perf_sw_event.__schedule.schedule.pipe_wait.pipe_read
      0.78            +0.0        0.80        perf-profile.calltrace.cycles-pp.__fdget_pos.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe.write
      0.73            +0.0        0.75        perf-profile.calltrace.cycles-pp.__fget_light.__fdget_pos.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.92            +0.0        0.95        perf-profile.calltrace.cycles-pp.check_preempt_wakeup.check_preempt_curr.ttwu_do_wakeup.try_to_wake_up.autoremove_wake_function
      2.11            +0.0        2.15        perf-profile.calltrace.cycles-pp.security_file_permission.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      7.00            -0.1        6.86        perf-profile.children.cycles-pp.syscall_return_via_sysret
      5.26            -0.1        5.14        perf-profile.children.cycles-pp.__switch_to
      5.65            -0.1        5.56        perf-profile.children.cycles-pp.reweight_entity
      2.17            -0.1        2.12        perf-profile.children.cycles-pp.copy_page_to_iter
      2.94            -0.0        2.90        perf-profile.children.cycles-pp.update_cfs_group
      3.11            -0.0        3.07        perf-profile.children.cycles-pp.pick_next_task_fair
      2.59            -0.0        2.55        perf-profile.children.cycles-pp.load_new_mm_cr3
      1.92            -0.0        1.88        perf-profile.children.cycles-pp._raw_spin_lock_irqsave
      1.11            -0.0        1.08 ±  2%  perf-profile.children.cycles-pp.find_next_bit
      0.59            -0.0        0.56        perf-profile.children.cycles-pp.finish_task_switch
      0.14 ± 15%      -0.0        0.11 ± 16%  perf-profile.children.cycles-pp.write@plt
      1.21            -0.0        1.18        perf-profile.children.cycles-pp.set_next_entity
      0.85            -0.0        0.82        perf-profile.children.cycles-pp.___perf_sw_event
      0.13 ±  3%      -0.0        0.11 ±  4%  perf-profile.children.cycles-pp.timespec_trunc
      0.47 ±  2%      -0.0        0.45        perf-profile.children.cycles-pp.anon_pipe_buf_release
      0.38 ±  2%      -0.0        0.36        perf-profile.children.cycles-pp.file_update_time
      0.74            -0.0        0.73        perf-profile.children.cycles-pp.copyout
      0.41 ±  2%      -0.0        0.39        perf-profile.children.cycles-pp.copy_user_enhanced_fast_string
      0.32            -0.0        0.30        perf-profile.children.cycles-pp.__x64_sys_read
      0.14            -0.0        0.12 ±  3%  perf-profile.children.cycles-pp.current_kernel_time64
      0.91            +0.0        0.92        perf-profile.children.cycles-pp.touch_atime
      0.40            +0.0        0.41        perf-profile.children.cycles-pp._cond_resched
      0.18 ±  2%      +0.0        0.20        perf-profile.children.cycles-pp.activate_task
      0.05            +0.0        0.07 ±  6%  perf-profile.children.cycles-pp.default_wake_function
      0.24            +0.0        0.27 ±  3%  perf-profile.children.cycles-pp.rcu_all_qs
      0.60 ±  2%      +0.0        0.64 ±  2%  perf-profile.children.cycles-pp.update_min_vruntime
      0.42 ±  4%      +0.0        0.46 ±  4%  perf-profile.children.cycles-pp.probe_sched_switch
      1.33            +0.0        1.38        perf-profile.children.cycles-pp.__fget_light
      0.53 ±  2%      +0.1        0.58        perf-profile.children.cycles-pp.entry_SYSCALL_64_stage2
      0.31            +0.1        0.36 ±  2%  perf-profile.children.cycles-pp.generic_pipe_buf_confirm
      4.35            +0.1        4.41        perf-profile.children.cycles-pp.switch_mm_irqs_off
      2.52            +0.1        2.58        perf-profile.children.cycles-pp.selinux_file_permission
      0.00            +0.1        0.07 ± 11%  perf-profile.children.cycles-pp.hrtick_update
      7.00            -0.1        6.86        perf-profile.self.cycles-pp.syscall_return_via_sysret
      5.26            -0.1        5.14        perf-profile.self.cycles-pp.__switch_to
      0.29            -0.1        0.19 ±  2%  perf-profile.self.cycles-pp.ksys_read
      1.49            -0.1        1.43        perf-profile.self.cycles-pp.dequeue_task_fair
      2.41            -0.1        2.35        perf-profile.self.cycles-pp.__schedule
      1.46            -0.0        1.41        perf-profile.self.cycles-pp.select_task_rq_fair
      2.94            -0.0        2.90        perf-profile.self.cycles-pp.update_cfs_group
      0.44            -0.0        0.40        perf-profile.self.cycles-pp.dequeue_entity
      0.48            -0.0        0.44        perf-profile.self.cycles-pp.finish_task_switch
      2.59            -0.0        2.55        perf-profile.self.cycles-pp.load_new_mm_cr3
      1.11            -0.0        1.08 ±  2%  perf-profile.self.cycles-pp.find_next_bit
      1.91            -0.0        1.88        perf-profile.self.cycles-pp._raw_spin_lock_irqsave
      0.78            -0.0        0.75        perf-profile.self.cycles-pp.___perf_sw_event
      0.14 ± 15%      -0.0        0.11 ± 16%  perf-profile.self.cycles-pp.write@plt
      0.37            -0.0        0.35 ±  2%  perf-profile.self.cycles-pp.__wake_up_common_lock
      0.20 ±  2%      -0.0        0.17 ±  2%  perf-profile.self.cycles-pp.__fdget_pos
      0.47 ±  2%      -0.0        0.44        perf-profile.self.cycles-pp.anon_pipe_buf_release
      0.87            -0.0        0.85        perf-profile.self.cycles-pp.copy_user_generic_unrolled
      0.13 ±  3%      -0.0        0.11 ±  4%  perf-profile.self.cycles-pp.timespec_trunc
      0.41 ±  2%      -0.0        0.39        perf-profile.self.cycles-pp.copy_user_enhanced_fast_string
      0.38            -0.0        0.36        perf-profile.self.cycles-pp.__wake_up_common
      0.32            -0.0        0.30        perf-profile.self.cycles-pp.__x64_sys_read
      0.14 ±  3%      -0.0        0.12 ±  3%  perf-profile.self.cycles-pp.current_kernel_time64
      0.30            -0.0        0.28        perf-profile.self.cycles-pp.set_next_entity
      0.28 ±  3%      +0.0        0.30        perf-profile.self.cycles-pp._cond_resched
      0.18 ±  2%      +0.0        0.20        perf-profile.self.cycles-pp.activate_task
      0.17 ±  2%      +0.0        0.19        perf-profile.self.cycles-pp.__might_fault
      0.05            +0.0        0.07 ±  6%  perf-profile.self.cycles-pp.default_wake_function
      0.17 ±  2%      +0.0        0.20        perf-profile.self.cycles-pp.ttwu_do_activate
      0.66            +0.0        0.69        perf-profile.self.cycles-pp.write
      0.24            +0.0        0.27 ±  3%  perf-profile.self.cycles-pp.rcu_all_qs
      0.67            +0.0        0.70        perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
      0.60 ±  2%      +0.0        0.64 ±  2%  perf-profile.self.cycles-pp.update_min_vruntime
      0.42 ±  4%      +0.0        0.46 ±  4%  perf-profile.self.cycles-pp.probe_sched_switch
      1.33            +0.0        1.37        perf-profile.self.cycles-pp.__fget_light
      1.61            +0.0        1.66        perf-profile.self.cycles-pp.pipe_read
      0.53 ±  2%      +0.1        0.58        perf-profile.self.cycles-pp.entry_SYSCALL_64_stage2
      0.31            +0.1        0.36 ±  2%  perf-profile.self.cycles-pp.generic_pipe_buf_confirm
      1.04            +0.1        1.11        perf-profile.self.cycles-pp.pipe_write
      0.00            +0.1        0.07 ± 11%  perf-profile.self.cycles-pp.hrtick_update
      2.00            +0.1        2.08        perf-profile.self.cycles-pp.switch_mm_irqs_off

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/never/page_fault3/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
          1:3          -33%            :3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=file_update_time/0x
           :3           33%           1:3     stderr.mount.nfs:Connection_timed_out
         34:3         -401%          22:3     perf-profile.calltrace.cycles-pp.error_entry.testcase
         17:3         -207%          11:3     perf-profile.calltrace.cycles-pp.sync_regs.error_entry.testcase
         34:3         -404%          22:3     perf-profile.children.cycles-pp.error_entry
          0:3           -2%           0:3     perf-profile.children.cycles-pp.error_exit
         16:3         -196%          11:3     perf-profile.self.cycles-pp.error_entry
          0:3           -2%           0:3     perf-profile.self.cycles-pp.error_exit
         %stddev     %change         %stddev
             \          |                \  
    467454            -1.8%     459251        will-it-scale.per_process_ops
     10856 ±  4%     -23.1%       8344 ±  7%  will-it-scale.per_thread_ops
    118134 ±  2%     +11.7%     131943        will-it-scale.time.involuntary_context_switches
 6.277e+08 ±  4%     -23.1%  4.827e+08 ±  7%  will-it-scale.time.minor_page_faults
      7406            +5.8%       7839        will-it-scale.time.percent_of_cpu_this_job_got
     44526            +5.8%      47106        will-it-scale.time.system_time
   7351468 ±  5%     -18.3%    6009014 ±  7%  will-it-scale.time.voluntary_context_switches
  91835846            -2.2%   89778599        will-it-scale.workload
   2534640            +4.3%    2643005 ±  2%  interrupts.CAL:Function_call_interrupts
      2819 ±  5%     +22.9%       3464 ± 18%  kthread_noise.total_time
     30273 ±  4%     -12.7%      26415 ±  5%  vmstat.system.cs
      1.52 ±  2%     +15.2%       1.75 ±  2%  irq_exception_noise.__do_page_fault.99th
    296.67 ± 12%     -36.7%     187.67 ± 12%  irq_exception_noise.softirq_time
    230900 ±  3%     +30.3%     300925 ±  3%  meminfo.Inactive
    230184 ±  3%     +30.4%     300180 ±  3%  meminfo.Inactive(anon)
     11.62 ±  3%      -2.2        9.40 ±  5%  mpstat.cpu.idle%
      0.00 ± 14%      -0.0        0.00 ±  4%  mpstat.cpu.iowait%
   7992174           -11.1%    7101976 ±  3%  softirqs.RCU
   4973624 ±  2%     -12.9%    4333370 ±  2%  softirqs.SCHED
    118134 ±  2%     +11.7%     131943        time.involuntary_context_switches
 6.277e+08 ±  4%     -23.1%  4.827e+08 ±  7%  time.minor_page_faults
      7406            +5.8%       7839        time.percent_of_cpu_this_job_got
     44526            +5.8%      47106        time.system_time
   7351468 ±  5%     -18.3%    6009014 ±  7%  time.voluntary_context_switches
 2.702e+09 ±  5%     -16.7%  2.251e+09 ±  7%  cpuidle.C1E.time
   6834329 ±  5%     -15.8%    5756243 ±  7%  cpuidle.C1E.usage
 1.046e+10 ±  3%     -19.8%  8.389e+09 ±  4%  cpuidle.C6.time
  13961845 ±  3%     -19.3%   11265555 ±  4%  cpuidle.C6.usage
   1309307 ±  7%     -14.8%    1116168 ±  8%  cpuidle.POLL.time
     19774 ±  6%     -13.7%      17063 ±  7%  cpuidle.POLL.usage
      2523 ±  4%     -11.1%       2243 ±  4%  slabinfo.biovec-64.active_objs
      2523 ±  4%     -11.1%       2243 ±  4%  slabinfo.biovec-64.num_objs
      2610 ±  8%     -33.7%       1731 ± 22%  slabinfo.dmaengine-unmap-16.active_objs
      2610 ±  8%     -33.7%       1731 ± 22%  slabinfo.dmaengine-unmap-16.num_objs
      5118 ± 17%     -22.6%       3962 ±  9%  slabinfo.eventpoll_pwq.active_objs
      5118 ± 17%     -22.6%       3962 ±  9%  slabinfo.eventpoll_pwq.num_objs
      4583 ±  3%     -14.0%       3941 ±  4%  slabinfo.sock_inode_cache.active_objs
      4583 ±  3%     -14.0%       3941 ±  4%  slabinfo.sock_inode_cache.num_objs
      1933            +2.6%       1984        turbostat.Avg_MHz
   6832021 ±  5%     -15.8%    5754156 ±  7%  turbostat.C1E
      2.32 ±  5%      -0.4        1.94 ±  7%  turbostat.C1E%
  13954211 ±  3%     -19.3%   11259436 ±  4%  turbostat.C6
      8.97 ±  3%      -1.8        7.20 ±  4%  turbostat.C6%
      6.18 ±  4%     -17.1%       5.13 ±  5%  turbostat.CPU%c1
      5.12 ±  3%     -21.7%       4.01 ±  4%  turbostat.CPU%c6
      1.76 ±  2%     -34.7%       1.15 ±  2%  turbostat.Pkg%pc2
     57314 ±  4%     +30.4%      74717 ±  4%  proc-vmstat.nr_inactive_anon
     57319 ±  4%     +30.4%      74719 ±  4%  proc-vmstat.nr_zone_inactive_anon
     24415 ± 19%     -62.2%       9236 ±  7%  proc-vmstat.numa_hint_faults
  69661453            -1.8%   68405712        proc-vmstat.numa_hit
  69553390            -1.8%   68297790        proc-vmstat.numa_local
      8792 ± 29%     -92.6%     654.33 ± 23%  proc-vmstat.numa_pages_migrated
     40251 ± 32%     -76.5%       9474 ±  3%  proc-vmstat.numa_pte_updates
  69522532            -1.6%   68383074        proc-vmstat.pgalloc_normal
 2.762e+10            -2.2%  2.701e+10        proc-vmstat.pgfault
  68825100            -1.5%   67772256        proc-vmstat.pgfree
      8792 ± 29%     -92.6%     654.33 ± 23%  proc-vmstat.pgmigrate_success
     57992 ±  6%     +56.2%      90591 ±  3%  numa-meminfo.node0.Inactive
     57916 ±  6%     +56.3%      90513 ±  3%  numa-meminfo.node0.Inactive(anon)
     37285 ± 12%     +36.0%      50709 ±  5%  numa-meminfo.node0.SReclaimable
    110971 ±  8%     +22.7%     136209 ±  8%  numa-meminfo.node0.Slab
     23601 ± 55%    +559.5%     155651 ± 36%  numa-meminfo.node1.AnonPages
     62484 ± 12%     +17.5%      73417 ±  3%  numa-meminfo.node1.Inactive
     62323 ± 12%     +17.2%      73023 ±  4%  numa-meminfo.node1.Inactive(anon)
    109714 ± 63%     -85.6%      15832 ± 96%  numa-meminfo.node2.AnonPages
     52236 ± 13%     +22.7%      64074 ±  3%  numa-meminfo.node2.Inactive
     51922 ± 12%     +23.2%      63963 ±  3%  numa-meminfo.node2.Inactive(anon)
     60241 ± 11%     +21.9%      73442 ±  8%  numa-meminfo.node3.Inactive
     60077 ± 12%     +22.0%      73279 ±  8%  numa-meminfo.node3.Inactive(anon)
     14093 ±  6%     +55.9%      21977 ±  3%  numa-vmstat.node0.nr_inactive_anon
      9321 ± 12%     +36.0%      12675 ±  5%  numa-vmstat.node0.nr_slab_reclaimable
     14090 ±  6%     +56.0%      21977 ±  3%  numa-vmstat.node0.nr_zone_inactive_anon
      5900 ± 55%    +559.4%      38909 ± 36%  numa-vmstat.node1.nr_anon_pages
     15413 ± 12%     +14.8%      17688 ±  4%  numa-vmstat.node1.nr_inactive_anon
     15413 ± 12%     +14.8%      17688 ±  4%  numa-vmstat.node1.nr_zone_inactive_anon
     27430 ± 63%     -85.6%       3960 ± 96%  numa-vmstat.node2.nr_anon_pages
     12928 ± 12%     +20.0%      15508 ±  3%  numa-vmstat.node2.nr_inactive_anon
     12927 ± 12%     +20.0%      15507 ±  3%  numa-vmstat.node2.nr_zone_inactive_anon
      6229 ± 10%    +117.5%      13547 ± 44%  numa-vmstat.node3
     14669 ± 11%     +19.6%      17537 ±  7%  numa-vmstat.node3.nr_inactive_anon
     14674 ± 11%     +19.5%      17541 ±  7%  numa-vmstat.node3.nr_zone_inactive_anon
     24617 ±141%    -100.0%       0.00        latency_stats.avg.io_schedule.nfs_lock_and_join_requests.nfs_updatepage.nfs_write_end.generic_perform_write.nfs_file_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      5049 ±105%     -99.4%      28.33 ± 82%  latency_stats.avg.call_rwsem_down_write_failed.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
    152457 ± 27%    +233.6%     508656 ± 92%  latency_stats.avg.max
      0.00       +3.9e+107%     390767 ±141%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_openat
     24617 ±141%    -100.0%       0.00        latency_stats.max.io_schedule.nfs_lock_and_join_requests.nfs_updatepage.nfs_write_end.generic_perform_write.nfs_file_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4240 ±141%    -100.0%       0.00        latency_stats.max.call_rwsem_down_write_failed.do_unlinkat.do_syscall_64.entry_SYSCALL_64_after_hwframe
      8565 ± 70%     -99.1%      80.33 ±115%  latency_stats.max.call_rwsem_down_write_failed.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
    204835 ±  6%    +457.6%    1142244 ±114%  latency_stats.max.max
      0.00       +5.1e+105%       5057 ±141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_access.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_openat.do_filp_open
      0.00         +1e+108%     995083 ±141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_openat
     13175 ±  4%    -100.0%       0.00        latency_stats.sum.io_schedule.__lock_page_or_retry.filemap_fault.__do_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault.page_fault
     24617 ±141%    -100.0%       0.00        latency_stats.sum.io_schedule.nfs_lock_and_join_requests.nfs_updatepage.nfs_write_end.generic_perform_write.nfs_file_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4260 ±141%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_write_failed.do_unlinkat.do_syscall_64.entry_SYSCALL_64_after_hwframe
      8640 ± 70%     -97.5%     216.33 ±108%  latency_stats.sum.call_rwsem_down_write_failed.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      6673 ± 89%     -92.8%     477.67 ± 74%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
      0.00       +4.2e+105%       4228 ±130%  latency_stats.sum.io_schedule.__lock_page_killable.__lock_page_or_retry.filemap_fault.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault.page_fault
      0.00       +7.5e+105%       7450 ± 98%  latency_stats.sum.io_schedule.__lock_page_or_retry.filemap_fault.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault.page_fault
      0.00       +1.3e+106%      13050 ±141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_access.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_openat.do_filp_open
      0.00       +1.5e+110%  1.508e+08 ±141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_openat
      0.97            -0.0        0.94        perf-stat.branch-miss-rate%
 1.329e+11            -2.6%  1.294e+11        perf-stat.branch-misses
 2.254e+11            -1.9%   2.21e+11        perf-stat.cache-references
  18308779 ±  4%     -12.8%   15969618 ±  5%  perf-stat.context-switches
      3.20            +1.8%       3.26        perf-stat.cpi
 2.233e+14            +2.7%  2.293e+14        perf-stat.cpu-cycles
      4.01            -0.2        3.83        perf-stat.dTLB-store-miss-rate%
  4.51e+11            -2.2%   4.41e+11        perf-stat.dTLB-store-misses
  1.08e+13            +2.6%  1.109e+13        perf-stat.dTLB-stores
 3.158e+10 ±  5%     +16.8%  3.689e+10 ±  2%  perf-stat.iTLB-load-misses
      2214 ±  5%     -13.8%       1907 ±  2%  perf-stat.instructions-per-iTLB-miss
      0.31            -1.8%       0.31        perf-stat.ipc
 2.762e+10            -2.2%  2.701e+10        perf-stat.minor-faults
 1.535e+10           -11.2%  1.362e+10        perf-stat.node-loads
      9.75            +1.1       10.89        perf-stat.node-store-miss-rate%
 3.012e+09           +10.6%  3.332e+09 ±  2%  perf-stat.node-store-misses
 2.787e+10            -2.2%  2.725e+10        perf-stat.node-stores
 2.762e+10            -2.2%  2.701e+10        perf-stat.page-faults
    759458            +3.2%     783404        perf-stat.path-length
    246.39 ± 15%     -20.4%     196.12 ±  6%  sched_debug.cfs_rq:/.load_avg.max
      0.21 ±  3%      +9.0%       0.23 ±  4%  sched_debug.cfs_rq:/.nr_running.stddev
     16.64 ± 27%     +61.0%      26.79 ± 17%  sched_debug.cfs_rq:/.nr_spread_over.max
     75.15           -14.4%      64.30 ±  4%  sched_debug.cfs_rq:/.util_avg.stddev
    178.80 ±  3%     +25.4%     224.12 ±  7%  sched_debug.cfs_rq:/.util_est_enqueued.avg
      1075 ±  5%     -12.3%     943.36 ±  2%  sched_debug.cfs_rq:/.util_est_enqueued.max
   2093630 ± 27%     -36.1%    1337941 ± 16%  sched_debug.cpu.avg_idle.max
    297057 ± 11%     +37.8%     409294 ± 14%  sched_debug.cpu.avg_idle.min
    293240 ± 55%     -62.3%     110571 ± 13%  sched_debug.cpu.avg_idle.stddev
    770075 ±  9%     -19.3%     621136 ± 12%  sched_debug.cpu.max_idle_balance_cost.max
     48919 ± 46%     -66.9%      16190 ± 81%  sched_debug.cpu.max_idle_balance_cost.stddev
     21716 ±  5%     -16.8%      18061 ±  7%  sched_debug.cpu.nr_switches.min
     21519 ±  5%     -17.7%      17700 ±  7%  sched_debug.cpu.sched_count.min
     10586 ±  5%     -18.1%       8669 ±  7%  sched_debug.cpu.sched_goidle.avg
     14183 ±  3%     -17.6%      11693 ±  5%  sched_debug.cpu.sched_goidle.max
     10322 ±  5%     -18.6%       8407 ±  7%  sched_debug.cpu.sched_goidle.min
    400.99 ±  8%     -13.0%     348.75 ±  3%  sched_debug.cpu.sched_goidle.stddev
      5459 ±  8%     +10.0%       6006 ±  3%  sched_debug.cpu.ttwu_local.avg
      8.47 ± 42%    +345.8%      37.73 ± 77%  sched_debug.rt_rq:/.rt_time.max
      0.61 ± 42%    +343.0%       2.72 ± 77%  sched_debug.rt_rq:/.rt_time.stddev
     91.98           -30.9       61.11 ± 70%  perf-profile.calltrace.cycles-pp.testcase
      9.05            -9.1        0.00        perf-profile.calltrace.cycles-pp.__do_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      8.91            -8.9        0.00        perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      8.06            -8.1        0.00        perf-profile.calltrace.cycles-pp.shmem_getpage_gfp.shmem_fault.__do_fault.__handle_mm_fault.handle_mm_fault
      7.59            -7.6        0.00        perf-profile.calltrace.cycles-pp.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault.__handle_mm_fault
      7.44            -7.4        0.00        perf-profile.calltrace.cycles-pp.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      7.28            -7.3        0.00        perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      5.31            -5.3        0.00        perf-profile.calltrace.cycles-pp.page_add_file_rmap.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault
      8.08            -2.8        5.30 ± 70%  perf-profile.calltrace.cycles-pp.native_irq_return_iret.testcase
      5.95            -2.1        3.83 ± 70%  perf-profile.calltrace.cycles-pp.find_get_entry.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault
      5.95            -2.0        3.93 ± 70%  perf-profile.calltrace.cycles-pp.swapgs_restore_regs_and_return_to_usermode.testcase
      3.10            -1.1        2.01 ± 70%  perf-profile.calltrace.cycles-pp.__perf_sw_event.__do_page_fault.do_page_fault.page_fault.testcase
      2.36            -0.8        1.55 ± 70%  perf-profile.calltrace.cycles-pp.___perf_sw_event.__perf_sw_event.__do_page_fault.do_page_fault.page_fault
      1.08            -0.4        0.70 ± 70%  perf-profile.calltrace.cycles-pp.do_page_fault.testcase
      0.82            -0.3        0.54 ± 70%  perf-profile.calltrace.cycles-pp.trace_graph_entry.do_page_fault.testcase
      0.77            -0.3        0.50 ± 70%  perf-profile.calltrace.cycles-pp.ftrace_graph_caller.__do_page_fault.do_page_fault.page_fault.testcase
      0.59            -0.2        0.37 ± 70%  perf-profile.calltrace.cycles-pp.down_read_trylock.__do_page_fault.do_page_fault.page_fault.testcase
     91.98           -30.9       61.11 ± 70%  perf-profile.children.cycles-pp.testcase
      9.14            -3.2        5.99 ± 70%  perf-profile.children.cycles-pp.__do_fault
      8.20            -2.8        5.40 ± 70%  perf-profile.children.cycles-pp.shmem_getpage_gfp
      8.08            -2.8        5.31 ± 70%  perf-profile.children.cycles-pp.native_irq_return_iret
      6.08            -2.2        3.92 ± 70%  perf-profile.children.cycles-pp.find_get_entry
      6.08            -2.1        3.96 ± 70%  perf-profile.children.cycles-pp.sync_regs
      5.95            -2.0        3.93 ± 70%  perf-profile.children.cycles-pp.swapgs_restore_regs_and_return_to_usermode
      4.12            -1.4        2.73 ± 70%  perf-profile.children.cycles-pp.ftrace_graph_caller
      3.65            -1.2        2.42 ± 70%  perf-profile.children.cycles-pp.prepare_ftrace_return
      3.18            -1.1        2.07 ± 70%  perf-profile.children.cycles-pp.__perf_sw_event
      2.34            -0.8        1.52 ± 70%  perf-profile.children.cycles-pp.fault_dirty_shared_page
      0.80            -0.3        0.50 ± 70%  perf-profile.children.cycles-pp._raw_spin_lock
      0.76            -0.3        0.50 ± 70%  perf-profile.children.cycles-pp.tlb_flush_mmu_free
      0.61            -0.2        0.39 ± 70%  perf-profile.children.cycles-pp.down_read_trylock
      0.48 ±  2%      -0.2        0.28 ± 70%  perf-profile.children.cycles-pp.pmd_devmap_trans_unstable
      0.26 ±  6%      -0.1        0.15 ± 71%  perf-profile.children.cycles-pp.ktime_get
      0.20 ±  2%      -0.1        0.12 ± 70%  perf-profile.children.cycles-pp.perf_exclude_event
      0.22 ±  2%      -0.1        0.13 ± 70%  perf-profile.children.cycles-pp._cond_resched
      0.17            -0.1        0.11 ± 70%  perf-profile.children.cycles-pp.page_rmapping
      0.13            -0.1        0.07 ± 70%  perf-profile.children.cycles-pp.rcu_all_qs
      0.07            -0.0        0.04 ± 70%  perf-profile.children.cycles-pp.ftrace_lookup_ip
     22.36            -7.8       14.59 ± 70%  perf-profile.self.cycles-pp.testcase
      8.08            -2.8        5.31 ± 70%  perf-profile.self.cycles-pp.native_irq_return_iret
      6.08            -2.1        3.96 ± 70%  perf-profile.self.cycles-pp.sync_regs
      5.81            -2.0        3.84 ± 70%  perf-profile.self.cycles-pp.swapgs_restore_regs_and_return_to_usermode
      3.27            -1.6        1.65 ± 70%  perf-profile.self.cycles-pp.__handle_mm_fault
      3.79            -1.4        2.36 ± 70%  perf-profile.self.cycles-pp.find_get_entry
      3.80            -1.3        2.53 ± 70%  perf-profile.self.cycles-pp.trace_graph_entry
      1.10            -0.5        0.57 ± 70%  perf-profile.self.cycles-pp.alloc_set_pte
      1.24            -0.4        0.81 ± 70%  perf-profile.self.cycles-pp.shmem_fault
      0.80            -0.3        0.50 ± 70%  perf-profile.self.cycles-pp._raw_spin_lock
      0.81            -0.3        0.51 ± 70%  perf-profile.self.cycles-pp.find_lock_entry
      0.80 ±  2%      -0.3        0.51 ± 70%  perf-profile.self.cycles-pp.__perf_sw_event
      0.61            -0.2        0.38 ± 70%  perf-profile.self.cycles-pp.down_read_trylock
      0.60            -0.2        0.39 ± 70%  perf-profile.self.cycles-pp.shmem_getpage_gfp
      0.48            -0.2        0.27 ± 70%  perf-profile.self.cycles-pp.pmd_devmap_trans_unstable
      0.47            -0.2        0.30 ± 70%  perf-profile.self.cycles-pp.file_update_time
      0.34            -0.1        0.22 ± 70%  perf-profile.self.cycles-pp.do_page_fault
      0.22 ±  4%      -0.1        0.11 ± 70%  perf-profile.self.cycles-pp.__do_fault
      0.25 ±  5%      -0.1        0.14 ± 71%  perf-profile.self.cycles-pp.ktime_get
      0.21 ±  2%      -0.1        0.12 ± 70%  perf-profile.self.cycles-pp.finish_fault
      0.23 ±  2%      -0.1        0.14 ± 70%  perf-profile.self.cycles-pp.fault_dirty_shared_page
      0.22 ±  2%      -0.1        0.14 ± 70%  perf-profile.self.cycles-pp.prepare_exit_to_usermode
      0.20 ±  2%      -0.1        0.12 ± 70%  perf-profile.self.cycles-pp.perf_exclude_event
      0.16            -0.1        0.10 ± 70%  perf-profile.self.cycles-pp._cond_resched
      0.13            -0.1        0.07 ± 70%  perf-profile.self.cycles-pp.rcu_all_qs
      0.07            -0.0        0.04 ± 70%  perf-profile.self.cycles-pp.ftrace_lookup_ip

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/always/context_switch1/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
           :3           33%           1:3     dmesg.WARNING:at#for_ip_interrupt_entry/0x
           :3           33%           1:3     dmesg.WARNING:at#for_ip_ret_from_intr/0x
           :3           67%           2:3     kmsg.pstore:crypto_comp_decompress_failed,ret=
           :3           67%           2:3     kmsg.pstore:decompression_failed
         %stddev     %change         %stddev
             \          |                \  
    223910            -1.3%     220930        will-it-scale.per_process_ops
    233722            -1.0%     231288        will-it-scale.per_thread_ops
 6.001e+08 ± 13%     +31.4%  7.887e+08 ±  4%  will-it-scale.time.involuntary_context_switches
     18003 ±  4%     +10.9%      19956        will-it-scale.time.minor_page_faults
  1.29e+10            -2.5%  1.258e+10        will-it-scale.time.voluntary_context_switches
  87865617            -1.2%   86826277        will-it-scale.workload
   2880329 ±  2%      +5.4%    3034904        interrupts.CAL:Function_call_interrupts
   7695018           -23.3%    5905066 ±  8%  meminfo.DirectMap2M
      0.00 ± 39%      -0.0        0.00 ± 78%  mpstat.cpu.iowait%
      4621 ± 12%     +13.4%       5241        proc-vmstat.numa_hint_faults_local
    715714           +27.6%     913142 ± 13%  softirqs.SCHED
    515653 ±  6%     -20.0%     412650 ± 15%  turbostat.C1
  43643516            -1.2%   43127031        vmstat.system.cs
   2893393 ±  4%     -23.6%    2210524 ± 10%  cpuidle.C1.time
    518051 ±  6%     -19.9%     415081 ± 15%  cpuidle.C1.usage
     23.10           +22.9%      28.38 ±  9%  boot-time.boot
     18.38           +23.2%      22.64 ± 12%  boot-time.dhcp
      5216            +5.0%       5478 ±  2%  boot-time.idle
    963.76 ± 44%    +109.7%       2021 ± 34%  irq_exception_noise.__do_page_fault.sum
      6.33 ± 14%    +726.3%      52.33 ± 62%  irq_exception_noise.irq_time
     56524 ±  7%     -18.8%      45915 ±  4%  irq_exception_noise.softirq_time
 6.001e+08 ± 13%     +31.4%  7.887e+08 ±  4%  time.involuntary_context_switches
     18003 ±  4%     +10.9%      19956        time.minor_page_faults
  1.29e+10            -2.5%  1.258e+10        time.voluntary_context_switches
      1386 ±  7%     +15.4%       1600 ± 11%  slabinfo.scsi_sense_cache.active_objs
      1386 ±  7%     +15.4%       1600 ± 11%  slabinfo.scsi_sense_cache.num_objs
      1427 ±  5%      -8.9%       1299 ±  2%  slabinfo.task_group.active_objs
      1427 ±  5%      -8.9%       1299 ±  2%  slabinfo.task_group.num_objs
     65519 ± 12%     +20.6%      79014 ± 16%  numa-meminfo.node0.SUnreclaim
      8484           -11.9%       7475 ±  7%  numa-meminfo.node1.KernelStack
      9264 ± 26%     -33.7%       6146 ±  7%  numa-meminfo.node1.Mapped
      2138 ± 61%    +373.5%      10127 ± 92%  numa-meminfo.node3.Inactive
      2059 ± 61%    +387.8%      10046 ± 93%  numa-meminfo.node3.Inactive(anon)
     16379 ± 12%     +20.6%      19752 ± 16%  numa-vmstat.node0.nr_slab_unreclaimable
      8483           -11.9%       7474 ±  7%  numa-vmstat.node1.nr_kernel_stack
      6250 ± 29%     -42.8%       3575 ± 24%  numa-vmstat.node2
      3798 ± 17%     +63.7%       6218 ±  5%  numa-vmstat.node3
    543.00 ± 61%    +368.1%       2541 ± 91%  numa-vmstat.node3.nr_inactive_anon
    543.33 ± 61%    +367.8%       2541 ± 91%  numa-vmstat.node3.nr_zone_inactive_anon
 4.138e+13            -1.1%   4.09e+13        perf-stat.branch-instructions
 6.569e+11            -2.0%  6.441e+11        perf-stat.branch-misses
 2.645e+10            -1.2%  2.613e+10        perf-stat.context-switches
      1.21            +1.2%       1.23        perf-stat.cpi
    153343 ±  2%     -12.1%     134776        perf-stat.cpu-migrations
 5.966e+13            -1.3%  5.889e+13        perf-stat.dTLB-loads
 3.736e+13            -1.2%   3.69e+13        perf-stat.dTLB-stores
      5.85 ± 15%      +8.8       14.67 ±  9%  perf-stat.iTLB-load-miss-rate%
 3.736e+09 ± 17%    +161.3%   9.76e+09 ± 11%  perf-stat.iTLB-load-misses
 5.987e+10            -5.4%  5.667e+10        perf-stat.iTLB-loads
 2.079e+14            -1.2%  2.054e+14        perf-stat.instructions
     57547 ± 18%     -62.9%      21340 ± 11%  perf-stat.instructions-per-iTLB-miss
      0.82            -1.2%       0.81        perf-stat.ipc
  27502531 ±  8%      +9.5%   30122136 ±  3%  perf-stat.node-store-misses
      1449 ± 27%     -34.6%     948.85        sched_debug.cfs_rq:/.load.min
    319416 ±115%    -188.5%    -282549        sched_debug.cfs_rq:/.spread0.avg
    657044 ± 55%     -88.3%      76887 ± 23%  sched_debug.cfs_rq:/.spread0.max
  -1525243           +54.6%   -2357898        sched_debug.cfs_rq:/.spread0.min
    101614 ±  6%     +30.6%     132713 ± 19%  sched_debug.cpu.avg_idle.stddev
     11.54 ± 41%     -61.2%       4.48        sched_debug.cpu.cpu_load[1].avg
      1369 ± 67%     -98.5%      20.67 ± 48%  sched_debug.cpu.cpu_load[1].max
     99.29 ± 67%     -97.6%       2.35 ± 26%  sched_debug.cpu.cpu_load[1].stddev
      9.58 ± 38%     -55.2%       4.29        sched_debug.cpu.cpu_load[2].avg
      1024 ± 68%     -98.5%      15.27 ± 36%  sched_debug.cpu.cpu_load[2].max
     74.51 ± 67%     -97.3%       1.99 ± 15%  sched_debug.cpu.cpu_load[2].stddev
      7.37 ± 29%     -42.0%       4.28        sched_debug.cpu.cpu_load[3].avg
    600.58 ± 68%     -97.9%      12.48 ± 20%  sched_debug.cpu.cpu_load[3].max
     43.98 ± 66%     -95.8%       1.83 ±  5%  sched_debug.cpu.cpu_load[3].stddev
      5.95 ± 19%     -28.1%       4.28        sched_debug.cpu.cpu_load[4].avg
    325.39 ± 67%     -96.4%      11.67 ± 10%  sched_debug.cpu.cpu_load[4].max
     24.19 ± 65%     -92.5%       1.81 ±  3%  sched_debug.cpu.cpu_load[4].stddev
    907.23 ±  4%     -14.1%     779.70 ± 10%  sched_debug.cpu.nr_load_updates.stddev
      0.00 ± 83%    +122.5%       0.00        sched_debug.rt_rq:/.rt_time.min
      8.49 ±  2%      -0.3        8.21 ±  2%  perf-profile.calltrace.cycles-pp.dequeue_task_fair.__schedule.schedule.pipe_wait.pipe_read
     57.28            -0.3       57.01        perf-profile.calltrace.cycles-pp.read
      5.06            -0.2        4.85        perf-profile.calltrace.cycles-pp.select_task_rq_fair.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_common_lock
      4.98            -0.2        4.78        perf-profile.calltrace.cycles-pp.__switch_to.read
      3.55            -0.2        3.39 ±  2%  perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.read
      2.72            -0.1        2.60        perf-profile.calltrace.cycles-pp.reweight_entity.enqueue_task_fair.ttwu_do_activate.try_to_wake_up.autoremove_wake_function
      2.67            -0.1        2.57 ±  2%  perf-profile.calltrace.cycles-pp.reweight_entity.dequeue_task_fair.__schedule.schedule.pipe_wait
      3.40            -0.1        3.31        perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.write
      3.77            -0.1        3.68        perf-profile.calltrace.cycles-pp.select_idle_sibling.select_task_rq_fair.try_to_wake_up.autoremove_wake_function.__wake_up_common
      1.95            -0.1        1.88        perf-profile.calltrace.cycles-pp.copy_page_to_iter.pipe_read.__vfs_read.vfs_read.ksys_read
      2.19            -0.1        2.13        perf-profile.calltrace.cycles-pp.__switch_to_asm.read
      1.30            -0.1        1.25        perf-profile.calltrace.cycles-pp.update_curr.reweight_entity.enqueue_task_fair.ttwu_do_activate.try_to_wake_up
      1.27            -0.1        1.22 ±  2%  perf-profile.calltrace.cycles-pp.update_curr.reweight_entity.dequeue_task_fair.__schedule.schedule
      2.29            -0.0        2.24        perf-profile.calltrace.cycles-pp.load_new_mm_cr3.switch_mm_irqs_off.__schedule.schedule.pipe_wait
      0.96            -0.0        0.92        perf-profile.calltrace.cycles-pp.__calc_delta.update_curr.reweight_entity.dequeue_task_fair.__schedule
      0.85            -0.0        0.81 ±  3%  perf-profile.calltrace.cycles-pp.cpumask_next_wrap.select_idle_sibling.select_task_rq_fair.try_to_wake_up.autoremove_wake_function
      1.63            -0.0        1.59        perf-profile.calltrace.cycles-pp.native_write_msr.read
      0.72            -0.0        0.69        perf-profile.calltrace.cycles-pp.copyout.copy_page_to_iter.pipe_read.__vfs_read.vfs_read
      0.65 ±  2%      -0.0        0.62        perf-profile.calltrace.cycles-pp._raw_spin_lock_irqsave.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_common_lock
      0.61            -0.0        0.58 ±  2%  perf-profile.calltrace.cycles-pp.find_next_bit.cpumask_next_wrap.select_idle_sibling.select_task_rq_fair.try_to_wake_up
      0.88            -0.0        0.85        perf-profile.calltrace.cycles-pp.touch_atime.pipe_read.__vfs_read.vfs_read.ksys_read
      0.80            -0.0        0.77 ±  2%  perf-profile.calltrace.cycles-pp.___perf_sw_event.__schedule.schedule.pipe_wait.pipe_read
      0.82            -0.0        0.79        perf-profile.calltrace.cycles-pp.prepare_to_wait.pipe_wait.pipe_read.__vfs_read.vfs_read
      0.72            -0.0        0.70        perf-profile.calltrace.cycles-pp.mutex_lock.pipe_write.__vfs_write.vfs_write.ksys_write
      0.56 ±  2%      -0.0        0.53        perf-profile.calltrace.cycles-pp.update_rq_clock.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_common_lock
      0.83            -0.0        0.81        perf-profile.calltrace.cycles-pp.__wake_up_common_lock.pipe_read.__vfs_read.vfs_read.ksys_read
     42.40            +0.3       42.69        perf-profile.calltrace.cycles-pp.write
     31.80            +0.4       32.18        perf-profile.calltrace.cycles-pp.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
     24.35            +0.5       24.84        perf-profile.calltrace.cycles-pp.pipe_wait.pipe_read.__vfs_read.vfs_read.ksys_read
     20.36            +0.6       20.92 ±  2%  perf-profile.calltrace.cycles-pp.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_common_lock.pipe_write
     22.01            +0.6       22.58        perf-profile.calltrace.cycles-pp.schedule.pipe_wait.pipe_read.__vfs_read.vfs_read
     21.87            +0.6       22.46        perf-profile.calltrace.cycles-pp.__schedule.schedule.pipe_wait.pipe_read.__vfs_read
      3.15 ± 11%      +1.0        4.12 ± 14%  perf-profile.calltrace.cycles-pp.ttwu_do_wakeup.try_to_wake_up.autoremove_wake_function.__wake_up_common.__wake_up_common_lock
      1.07 ± 34%      +1.1        2.12 ± 31%  perf-profile.calltrace.cycles-pp.tracing_record_taskinfo_sched_switch.__schedule.schedule.pipe_wait.pipe_read
      0.66 ± 75%      +1.1        1.72 ± 37%  perf-profile.calltrace.cycles-pp.trace_save_cmdline.tracing_record_taskinfo.ttwu_do_wakeup.try_to_wake_up.autoremove_wake_function
      0.75 ± 74%      +1.1        1.88 ± 34%  perf-profile.calltrace.cycles-pp.tracing_record_taskinfo.ttwu_do_wakeup.try_to_wake_up.autoremove_wake_function.__wake_up_common
      0.69 ± 76%      +1.2        1.85 ± 36%  perf-profile.calltrace.cycles-pp.trace_save_cmdline.tracing_record_taskinfo_sched_switch.__schedule.schedule.pipe_wait
      8.73 ±  2%      -0.3        8.45        perf-profile.children.cycles-pp.dequeue_task_fair
     57.28            -0.3       57.01        perf-profile.children.cycles-pp.read
      6.95            -0.2        6.70        perf-profile.children.cycles-pp.syscall_return_via_sysret
      5.57            -0.2        5.35        perf-profile.children.cycles-pp.reweight_entity
      5.26            -0.2        5.05        perf-profile.children.cycles-pp.select_task_rq_fair
      5.19            -0.2        4.99        perf-profile.children.cycles-pp.__switch_to
      4.90            -0.2        4.73 ±  2%  perf-profile.children.cycles-pp.update_curr
      1.27            -0.1        1.13 ±  8%  perf-profile.children.cycles-pp.fsnotify
      3.92            -0.1        3.83        perf-profile.children.cycles-pp.select_idle_sibling
      2.01            -0.1        1.93        perf-profile.children.cycles-pp.__calc_delta
      2.14            -0.1        2.06        perf-profile.children.cycles-pp.copy_page_to_iter
      1.58            -0.1        1.51        perf-profile.children.cycles-pp._raw_spin_unlock_irqrestore
      2.90            -0.1        2.84        perf-profile.children.cycles-pp.update_cfs_group
      1.93            -0.1        1.87        perf-profile.children.cycles-pp._raw_spin_lock_irqsave
      2.35            -0.1        2.29        perf-profile.children.cycles-pp.__switch_to_asm
      1.33            -0.1        1.27 ±  3%  perf-profile.children.cycles-pp.cpumask_next_wrap
      2.57            -0.1        2.52        perf-profile.children.cycles-pp.load_new_mm_cr3
      1.53            -0.1        1.47 ±  2%  perf-profile.children.cycles-pp.__fdget_pos
      1.11            -0.0        1.07 ±  2%  perf-profile.children.cycles-pp.find_next_bit
      1.18            -0.0        1.14        perf-profile.children.cycles-pp.update_rq_clock
      0.88            -0.0        0.83        perf-profile.children.cycles-pp.copy_user_generic_unrolled
      1.70            -0.0        1.65        perf-profile.children.cycles-pp.native_write_msr
      0.97            -0.0        0.93 ±  2%  perf-profile.children.cycles-pp.account_entity_dequeue
      0.59            -0.0        0.56        perf-profile.children.cycles-pp.finish_task_switch
      0.91            -0.0        0.88        perf-profile.children.cycles-pp.touch_atime
      0.69            -0.0        0.65        perf-profile.children.cycles-pp.account_entity_enqueue
      2.13            -0.0        2.09        perf-profile.children.cycles-pp.mutex_lock
      0.32 ±  3%      -0.0        0.29 ±  4%  perf-profile.children.cycles-pp.__sb_start_write
      0.84            -0.0        0.81 ±  2%  perf-profile.children.cycles-pp.___perf_sw_event
      0.89            -0.0        0.87        perf-profile.children.cycles-pp.prepare_to_wait
      0.73            -0.0        0.71        perf-profile.children.cycles-pp.copyout
      0.31 ±  2%      -0.0        0.28 ±  3%  perf-profile.children.cycles-pp.__list_del_entry_valid
      0.46 ±  2%      -0.0        0.44        perf-profile.children.cycles-pp.anon_pipe_buf_release
      0.38            -0.0        0.36 ±  3%  perf-profile.children.cycles-pp.idle_cpu
      0.32            -0.0        0.30 ±  2%  perf-profile.children.cycles-pp.__x64_sys_read
      0.21 ±  2%      -0.0        0.20 ±  2%  perf-profile.children.cycles-pp.deactivate_task
      0.13            -0.0        0.12 ±  4%  perf-profile.children.cycles-pp.timespec_trunc
      0.09            -0.0        0.08        perf-profile.children.cycles-pp.iov_iter_init
      0.08            -0.0        0.07        perf-profile.children.cycles-pp.native_load_tls
      0.11 ±  4%      +0.0        0.12        perf-profile.children.cycles-pp.tick_sched_timer
      0.08 ±  5%      +0.0        0.10 ±  4%  perf-profile.children.cycles-pp.finish_wait
      0.38 ±  2%      +0.0        0.40 ±  2%  perf-profile.children.cycles-pp.file_update_time
      0.31            +0.0        0.33 ±  2%  perf-profile.children.cycles-pp.smp_apic_timer_interrupt
      0.24 ±  3%      +0.0        0.26 ±  3%  perf-profile.children.cycles-pp.rcu_all_qs
      0.39            +0.0        0.41        perf-profile.children.cycles-pp._cond_resched
      0.05            +0.0        0.07 ±  6%  perf-profile.children.cycles-pp.default_wake_function
      0.23 ±  2%      +0.0        0.26 ±  3%  perf-profile.children.cycles-pp.current_time
      0.30            +0.0        0.35 ±  2%  perf-profile.children.cycles-pp.generic_pipe_buf_confirm
      0.52            +0.1        0.58        perf-profile.children.cycles-pp.entry_SYSCALL_64_stage2
      0.00            +0.1        0.08 ±  5%  perf-profile.children.cycles-pp.hrtick_update
     42.40            +0.3       42.69        perf-profile.children.cycles-pp.write
     31.86            +0.4       32.26        perf-profile.children.cycles-pp.__vfs_read
     24.40            +0.5       24.89        perf-profile.children.cycles-pp.pipe_wait
     20.40            +0.6       20.96 ±  2%  perf-profile.children.cycles-pp.try_to_wake_up
     22.30            +0.6       22.89        perf-profile.children.cycles-pp.schedule
     22.22            +0.6       22.84        perf-profile.children.cycles-pp.__schedule
      0.99 ± 36%      +0.9        1.94 ± 32%  perf-profile.children.cycles-pp.tracing_record_taskinfo
      3.30 ± 10%      +1.0        4.27 ± 13%  perf-profile.children.cycles-pp.ttwu_do_wakeup
      1.14 ± 31%      +1.1        2.24 ± 29%  perf-profile.children.cycles-pp.tracing_record_taskinfo_sched_switch
      1.59 ± 46%      +2.0        3.60 ± 36%  perf-profile.children.cycles-pp.trace_save_cmdline
      6.95            -0.2        6.70        perf-profile.self.cycles-pp.syscall_return_via_sysret
      5.19            -0.2        4.99        perf-profile.self.cycles-pp.__switch_to
      1.27            -0.1        1.12 ±  8%  perf-profile.self.cycles-pp.fsnotify
      1.49            -0.1        1.36        perf-profile.self.cycles-pp.select_task_rq_fair
      2.47            -0.1        2.37 ±  2%  perf-profile.self.cycles-pp.reweight_entity
      0.29            -0.1        0.19 ±  2%  perf-profile.self.cycles-pp.ksys_read
      1.50            -0.1        1.42        perf-profile.self.cycles-pp._raw_spin_unlock_irqrestore
      2.01            -0.1        1.93        perf-profile.self.cycles-pp.__calc_delta
      1.93            -0.1        1.86        perf-profile.self.cycles-pp._raw_spin_lock_irqsave
      1.47            -0.1        1.40        perf-profile.self.cycles-pp.dequeue_task_fair
      2.90            -0.1        2.84        perf-profile.self.cycles-pp.update_cfs_group
      1.29            -0.1        1.23        perf-profile.self.cycles-pp.do_syscall_64
      2.57            -0.1        2.52        perf-profile.self.cycles-pp.load_new_mm_cr3
      2.28            -0.1        2.23        perf-profile.self.cycles-pp.__switch_to_asm
      1.80            -0.1        1.75        perf-profile.self.cycles-pp.select_idle_sibling
      1.11            -0.0        1.07 ±  2%  perf-profile.self.cycles-pp.find_next_bit
      0.87            -0.0        0.83        perf-profile.self.cycles-pp.copy_user_generic_unrolled
      0.43            -0.0        0.39 ±  2%  perf-profile.self.cycles-pp.dequeue_entity
      1.70            -0.0        1.65        perf-profile.self.cycles-pp.native_write_msr
      0.92            -0.0        0.88 ±  2%  perf-profile.self.cycles-pp.account_entity_dequeue
      0.48            -0.0        0.44        perf-profile.self.cycles-pp.finish_task_switch
      0.77            -0.0        0.74        perf-profile.self.cycles-pp.___perf_sw_event
      0.66            -0.0        0.63        perf-profile.self.cycles-pp.account_entity_enqueue
      0.46 ±  2%      -0.0        0.43 ±  2%  perf-profile.self.cycles-pp.anon_pipe_buf_release
      0.32 ±  3%      -0.0        0.29 ±  4%  perf-profile.self.cycles-pp.__sb_start_write
      0.31 ±  2%      -0.0        0.28 ±  3%  perf-profile.self.cycles-pp.__list_del_entry_valid
      0.38            -0.0        0.36 ±  3%  perf-profile.self.cycles-pp.idle_cpu
      0.19 ±  4%      -0.0        0.17 ±  2%  perf-profile.self.cycles-pp.__fdget_pos
      0.50            -0.0        0.48        perf-profile.self.cycles-pp.__atime_needs_update
      0.23 ±  2%      -0.0        0.21 ±  3%  perf-profile.self.cycles-pp.touch_atime
      0.31            -0.0        0.30        perf-profile.self.cycles-pp.__x64_sys_read
      0.21 ±  2%      -0.0        0.20 ±  2%  perf-profile.self.cycles-pp.deactivate_task
      0.21 ±  2%      -0.0        0.19        perf-profile.self.cycles-pp.check_preempt_curr
      0.40            -0.0        0.39        perf-profile.self.cycles-pp.autoremove_wake_function
      0.40            -0.0        0.38        perf-profile.self.cycles-pp.copy_user_enhanced_fast_string
      0.27            -0.0        0.26        perf-profile.self.cycles-pp.pipe_wait
      0.13            -0.0        0.12 ±  4%  perf-profile.self.cycles-pp.timespec_trunc
      0.22 ±  2%      -0.0        0.20 ±  2%  perf-profile.self.cycles-pp.put_prev_entity
      0.09            -0.0        0.08        perf-profile.self.cycles-pp.iov_iter_init
      0.08            -0.0        0.07        perf-profile.self.cycles-pp.native_load_tls
      0.11            -0.0        0.10        perf-profile.self.cycles-pp.schedule
      0.12 ±  4%      +0.0        0.13        perf-profile.self.cycles-pp.copyin
      0.08 ±  5%      +0.0        0.10 ±  4%  perf-profile.self.cycles-pp.finish_wait
      0.18            +0.0        0.20 ±  2%  perf-profile.self.cycles-pp.ttwu_do_activate
      0.28 ±  2%      +0.0        0.30 ±  2%  perf-profile.self.cycles-pp._cond_resched
      0.24 ±  3%      +0.0        0.26 ±  3%  perf-profile.self.cycles-pp.rcu_all_qs
      0.05            +0.0        0.07 ±  6%  perf-profile.self.cycles-pp.default_wake_function
      0.08 ± 14%      +0.0        0.11 ± 14%  perf-profile.self.cycles-pp.tracing_record_taskinfo_sched_switch
      0.51            +0.0        0.55 ±  4%  perf-profile.self.cycles-pp.vfs_write
      0.30            +0.0        0.35 ±  2%  perf-profile.self.cycles-pp.generic_pipe_buf_confirm
      0.52            +0.1        0.58        perf-profile.self.cycles-pp.entry_SYSCALL_64_stage2
      0.00            +0.1        0.08 ±  5%  perf-profile.self.cycles-pp.hrtick_update
      1.97            +0.1        2.07 ±  2%  perf-profile.self.cycles-pp.switch_mm_irqs_off
      1.59 ± 46%      +2.0        3.60 ± 36%  perf-profile.self.cycles-pp.trace_save_cmdline

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/never/brk1/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
           :3           33%           1:3     kmsg.pstore:crypto_comp_decompress_failed,ret=
           :3           33%           1:3     kmsg.pstore:decompression_failed
         %stddev     %change         %stddev
             \          |                \  
    997317            -2.0%     977778        will-it-scale.per_process_ops
    957.00            -7.9%     881.00 ±  3%  will-it-scale.per_thread_ops
     18.42 ±  3%      -8.2%      16.90        will-it-scale.time.user_time
 1.917e+08            -2.0%  1.879e+08        will-it-scale.workload
     18.42 ±  3%      -8.2%      16.90        time.user_time
      0.30 ± 11%     -36.7%       0.19 ± 11%  turbostat.Pkg%pc2
     57539 ± 51%    +140.6%     138439 ± 31%  meminfo.CmaFree
    410877 ± 11%     -22.1%     320082 ± 22%  meminfo.DirectMap4k
    343575 ± 27%     +71.3%     588703 ± 31%  numa-numastat.node0.local_node
    374176 ± 24%     +63.3%     611007 ± 27%  numa-numastat.node0.numa_hit
   1056347 ±  4%     -39.9%     634843 ± 38%  numa-numastat.node3.local_node
   1060682 ±  4%     -39.0%     646862 ± 35%  numa-numastat.node3.numa_hit
     14383 ± 51%    +140.6%      34608 ± 31%  proc-vmstat.nr_free_cma
    179.00            +2.4%     183.33        proc-vmstat.nr_inactive_file
    179.00            +2.4%     183.33        proc-vmstat.nr_zone_inactive_file
    564483 ±  3%     -38.0%     350064 ± 36%  proc-vmstat.pgalloc_movable
   1811959           +10.8%    2008488 ±  5%  proc-vmstat.pgalloc_normal
      7153 ± 42%     -94.0%     431.33 ±119%  latency_stats.max.pipe_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      6627 ±141%    +380.5%      31843 ±110%  latency_stats.max.call_rwsem_down_write_failed_killable.do_mprotect_pkey.__x64_sys_mprotect.do_syscall_64.entry_SYSCALL_64_after_hwframe
     15244 ± 31%     -99.9%      15.00 ±141%  latency_stats.sum.call_rwsem_down_read_failed.__do_page_fault.do_page_fault.page_fault.__get_user_8.exit_robust_list.mm_release.do_exit.do_group_exit.get_signal.do_signal.exit_to_usermode_loop
      4301 ±117%     -83.7%     700.33 ±  6%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
     12153 ± 28%     -83.1%       2056 ± 70%  latency_stats.sum.pipe_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      6772 ±141%   +1105.8%      81665 ±127%  latency_stats.sum.call_rwsem_down_write_failed_killable.do_mprotect_pkey.__x64_sys_mprotect.do_syscall_64.entry_SYSCALL_64_after_hwframe
 2.465e+13            -1.3%  2.434e+13        perf-stat.branch-instructions
 2.691e+11            -2.1%  2.635e+11        perf-stat.branch-misses
 3.402e+13            -1.4%  3.355e+13        perf-stat.dTLB-loads
 1.694e+13            +1.4%  1.718e+13        perf-stat.dTLB-stores
      1.75 ± 50%      +4.7        6.45 ± 11%  perf-stat.iTLB-load-miss-rate%
 4.077e+08 ± 48%    +232.3%  1.355e+09 ± 11%  perf-stat.iTLB-load-misses
  2.31e+10 ±  2%     -14.9%  1.965e+10 ±  3%  perf-stat.iTLB-loads
 1.163e+14            -1.6%  1.144e+14        perf-stat.instructions
    346171 ± 36%     -75.3%      85575 ± 11%  perf-stat.instructions-per-iTLB-miss
 6.174e+08 ±  2%      -9.5%  5.589e+08        perf-stat.node-store-misses
    595.00 ± 10%     +31.4%     782.00 ±  3%  slabinfo.Acpi-State.active_objs
    595.00 ± 10%     +31.4%     782.00 ±  3%  slabinfo.Acpi-State.num_objs
      2831 ±  3%     -14.0%       2434 ±  5%  slabinfo.avtab_node.active_objs
      2831 ±  3%     -14.0%       2434 ±  5%  slabinfo.avtab_node.num_objs
    934.00           -10.9%     832.33 ±  5%  slabinfo.inotify_inode_mark.active_objs
    934.00           -10.9%     832.33 ±  5%  slabinfo.inotify_inode_mark.num_objs
      1232 ±  4%     +13.4%       1397 ±  6%  slabinfo.nsproxy.active_objs
      1232 ±  4%     +13.4%       1397 ±  6%  slabinfo.nsproxy.num_objs
    499.67 ± 12%     +24.8%     623.67 ± 10%  slabinfo.secpath_cache.active_objs
    499.67 ± 12%     +24.8%     623.67 ± 10%  slabinfo.secpath_cache.num_objs
     31393 ± 84%    +220.1%     100477 ± 21%  numa-meminfo.node0.Active
     31393 ± 84%    +220.1%     100477 ± 21%  numa-meminfo.node0.Active(anon)
     30013 ± 85%    +232.1%      99661 ± 21%  numa-meminfo.node0.AnonPages
     21603 ± 34%     -85.0%       3237 ±100%  numa-meminfo.node0.Inactive
     21528 ± 34%     -85.0%       3237 ±100%  numa-meminfo.node0.Inactive(anon)
     10247 ± 35%     -46.4%       5495        numa-meminfo.node0.Mapped
     35388 ± 14%     -41.6%      20670 ± 15%  numa-meminfo.node0.SReclaimable
     22911 ± 29%     -82.3%       4057 ± 84%  numa-meminfo.node0.Shmem
    117387 ±  9%     -22.5%      90986 ± 12%  numa-meminfo.node0.Slab
     68863 ± 67%     +77.7%     122351 ± 13%  numa-meminfo.node1.Active
     68863 ± 67%     +77.7%     122351 ± 13%  numa-meminfo.node1.Active(anon)
    228376           +22.3%     279406 ± 17%  numa-meminfo.node1.FilePages
      1481 ±116%   +1062.1%      17218 ± 39%  numa-meminfo.node1.Inactive
      1481 ±116%   +1062.0%      17216 ± 39%  numa-meminfo.node1.Inactive(anon)
      6593 ±  2%     +11.7%       7367 ±  3%  numa-meminfo.node1.KernelStack
    596227 ±  8%     +18.0%     703748 ±  4%  numa-meminfo.node1.MemUsed
     15298 ± 12%     +88.5%      28843 ± 36%  numa-meminfo.node1.SReclaimable
     52718 ±  9%     +21.0%      63810 ± 11%  numa-meminfo.node1.SUnreclaim
      1808 ± 97%   +2723.8%      51054 ± 97%  numa-meminfo.node1.Shmem
     68017 ±  5%     +36.2%      92654 ± 18%  numa-meminfo.node1.Slab
    125541 ± 29%     -64.9%      44024 ± 98%  numa-meminfo.node3.Active
    125137 ± 29%     -65.0%      43823 ± 98%  numa-meminfo.node3.Active(anon)
     93173 ± 25%     -87.8%      11381 ± 20%  numa-meminfo.node3.AnonPages
      9150 ±  5%      -9.3%       8301 ±  8%  numa-meminfo.node3.KernelStack
      7848 ± 84%    +220.0%      25118 ± 21%  numa-vmstat.node0.nr_active_anon
      7503 ± 85%    +232.1%      24914 ± 21%  numa-vmstat.node0.nr_anon_pages
      5381 ± 34%     -85.0%     809.00 ±100%  numa-vmstat.node0.nr_inactive_anon
      2559 ± 35%     -46.4%       1372        numa-vmstat.node0.nr_mapped
      5727 ± 29%     -82.3%       1014 ± 84%  numa-vmstat.node0.nr_shmem
      8846 ± 14%     -41.6%       5167 ± 15%  numa-vmstat.node0.nr_slab_reclaimable
      7848 ± 84%    +220.0%      25118 ± 21%  numa-vmstat.node0.nr_zone_active_anon
      5381 ± 34%     -85.0%     809.00 ±100%  numa-vmstat.node0.nr_zone_inactive_anon
      4821 ±  2%     +30.3%       6283 ± 15%  numa-vmstat.node1
     17215 ± 67%     +77.7%      30591 ± 13%  numa-vmstat.node1.nr_active_anon
     57093           +22.3%      69850 ± 17%  numa-vmstat.node1.nr_file_pages
    370.00 ±116%   +1061.8%       4298 ± 39%  numa-vmstat.node1.nr_inactive_anon
      6593 ±  2%     +11.7%       7366 ±  3%  numa-vmstat.node1.nr_kernel_stack
    451.67 ± 97%   +2725.6%      12762 ± 97%  numa-vmstat.node1.nr_shmem
      3824 ± 12%     +88.6%       7211 ± 36%  numa-vmstat.node1.nr_slab_reclaimable
     13179 ±  9%     +21.0%      15952 ± 11%  numa-vmstat.node1.nr_slab_unreclaimable
     17215 ± 67%     +77.7%      30591 ± 13%  numa-vmstat.node1.nr_zone_active_anon
    370.00 ±116%   +1061.8%       4298 ± 39%  numa-vmstat.node1.nr_zone_inactive_anon
    364789 ± 12%     +62.8%     593926 ± 34%  numa-vmstat.node1.numa_hit
    239539 ± 19%     +95.4%     468113 ± 43%  numa-vmstat.node1.numa_local
     71.00 ± 28%     +42.3%     101.00        numa-vmstat.node2.nr_mlock
     31285 ± 29%     -65.0%      10960 ± 98%  numa-vmstat.node3.nr_active_anon
     23292 ± 25%     -87.8%       2844 ± 19%  numa-vmstat.node3.nr_anon_pages
     14339 ± 52%    +141.1%      34566 ± 32%  numa-vmstat.node3.nr_free_cma
      9151 ±  5%      -9.3%       8299 ±  8%  numa-vmstat.node3.nr_kernel_stack
     31305 ± 29%     -64.9%      10975 ± 98%  numa-vmstat.node3.nr_zone_active_anon
    930131 ±  3%     -35.9%     596006 ± 34%  numa-vmstat.node3.numa_hit
    836455 ±  3%     -40.9%     493947 ± 44%  numa-vmstat.node3.numa_local
     75182 ± 58%     -83.8%      12160 ±  2%  sched_debug.cfs_rq:/.load.max
      6.65 ±  5%     -10.6%       5.94 ±  6%  sched_debug.cfs_rq:/.load_avg.avg
      0.16 ±  7%     +22.6%       0.20 ± 12%  sched_debug.cfs_rq:/.nr_running.stddev
      5.58 ± 24%    +427.7%      29.42 ± 93%  sched_debug.cfs_rq:/.nr_spread_over.max
      0.54 ± 15%    +306.8%       2.19 ± 86%  sched_debug.cfs_rq:/.nr_spread_over.stddev
      1.05 ± 25%     -65.1%       0.37 ± 71%  sched_debug.cfs_rq:/.removed.load_avg.avg
      9.62 ± 11%     -50.7%       4.74 ± 70%  sched_debug.cfs_rq:/.removed.load_avg.stddev
     48.70 ± 25%     -65.1%      17.02 ± 71%  sched_debug.cfs_rq:/.removed.runnable_sum.avg
    444.31 ± 11%     -50.7%     219.26 ± 70%  sched_debug.cfs_rq:/.removed.runnable_sum.stddev
      0.47 ± 13%     -60.9%       0.19 ± 71%  sched_debug.cfs_rq:/.removed.util_avg.avg
      4.47 ±  4%     -46.5%       2.39 ± 70%  sched_debug.cfs_rq:/.removed.util_avg.stddev
      1.64 ±  7%     +22.1%       2.00 ± 13%  sched_debug.cfs_rq:/.runnable_load_avg.stddev
     74653 ± 59%     -84.4%      11676        sched_debug.cfs_rq:/.runnable_weight.max
   -119169          -491.3%     466350 ± 27%  sched_debug.cfs_rq:/.spread0.avg
    517161 ± 30%    +145.8%    1271292 ± 23%  sched_debug.cfs_rq:/.spread0.max
    624.79 ±  5%     -14.2%     535.76 ±  7%  sched_debug.cfs_rq:/.util_est_enqueued.avg
    247.91 ± 32%     -99.8%       0.48 ±  8%  sched_debug.cfs_rq:/.util_est_enqueued.min
    179704 ±  3%     +30.4%     234297 ± 16%  sched_debug.cpu.avg_idle.stddev
      1.56 ±  9%     +24.4%       1.94 ± 14%  sched_debug.cpu.cpu_load[0].stddev
      1.50 ±  6%     +27.7%       1.91 ± 14%  sched_debug.cpu.cpu_load[1].stddev
      1.45 ±  3%     +30.8%       1.90 ± 14%  sched_debug.cpu.cpu_load[2].stddev
      1.43 ±  3%     +36.1%       1.95 ± 11%  sched_debug.cpu.cpu_load[3].stddev
      1.55 ±  7%     +43.5%       2.22 ±  7%  sched_debug.cpu.cpu_load[4].stddev
     10004 ±  3%     -11.6%       8839 ±  3%  sched_debug.cpu.curr->pid.avg
      1146 ± 26%     +52.2%       1745 ±  7%  sched_debug.cpu.curr->pid.min
      3162 ±  6%     +25.4%       3966 ± 11%  sched_debug.cpu.curr->pid.stddev
    403738 ±  3%     -11.7%     356696 ±  7%  sched_debug.cpu.nr_switches.max
      0.08 ± 21%     +78.2%       0.14 ± 14%  sched_debug.cpu.nr_uninterruptible.avg
    404435 ±  3%     -11.8%     356732 ±  7%  sched_debug.cpu.sched_count.max
      4.17            -0.3        3.87        perf-profile.calltrace.cycles-pp.kmem_cache_alloc.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      2.40            -0.2        2.17        perf-profile.calltrace.cycles-pp.vma_compute_subtree_gap.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk
      7.58            -0.2        7.36        perf-profile.calltrace.cycles-pp.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     15.00            -0.2       14.81        perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.brk
      7.83            -0.2        7.66        perf-profile.calltrace.cycles-pp.unmap_vmas.unmap_region.do_munmap.__x64_sys_brk.do_syscall_64
     28.66            -0.1       28.51        perf-profile.calltrace.cycles-pp.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      2.15            -0.1        2.03        perf-profile.calltrace.cycles-pp.vma_compute_subtree_gap.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.07            -0.1        0.99        perf-profile.calltrace.cycles-pp.memcpy_erms.strlcpy.perf_event_mmap.do_brk_flags.__x64_sys_brk
      1.03            -0.1        0.95        perf-profile.calltrace.cycles-pp.kmem_cache_free.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64
      7.33            -0.1        7.25        perf-profile.calltrace.cycles-pp.unmap_page_range.unmap_vmas.unmap_region.do_munmap.__x64_sys_brk
      0.76            -0.1        0.69        perf-profile.calltrace.cycles-pp.__vm_enough_memory.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     11.85            -0.1       11.77        perf-profile.calltrace.cycles-pp.unmap_region.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.64            -0.1        1.57        perf-profile.calltrace.cycles-pp.strlcpy.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64
      1.06            -0.1        0.99        perf-profile.calltrace.cycles-pp.__indirect_thunk_start.brk
      0.73            -0.1        0.67        perf-profile.calltrace.cycles-pp.sync_mm_rss.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      4.59            -0.1        4.52        perf-profile.calltrace.cycles-pp.security_vm_enough_memory_mm.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      2.82            -0.1        2.76        perf-profile.calltrace.cycles-pp.selinux_vm_enough_memory.security_vm_enough_memory_mm.do_brk_flags.__x64_sys_brk.do_syscall_64
      2.89            -0.1        2.84        perf-profile.calltrace.cycles-pp.down_write_killable.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      3.37            -0.1        3.32        perf-profile.calltrace.cycles-pp.get_unmapped_area.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.99            -0.0        1.94        perf-profile.calltrace.cycles-pp.cred_has_capability.selinux_vm_enough_memory.security_vm_enough_memory_mm.do_brk_flags.__x64_sys_brk
      2.32            -0.0        2.27        perf-profile.calltrace.cycles-pp.perf_iterate_sb.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64
      1.88            -0.0        1.84        perf-profile.calltrace.cycles-pp.security_mmap_addr.get_unmapped_area.do_brk_flags.__x64_sys_brk.do_syscall_64
      0.77            -0.0        0.73        perf-profile.calltrace.cycles-pp._raw_spin_lock.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      1.62            -0.0        1.59        perf-profile.calltrace.cycles-pp.memset_erms.kmem_cache_alloc.do_brk_flags.__x64_sys_brk.do_syscall_64
      0.81            -0.0        0.79        perf-profile.calltrace.cycles-pp.___might_sleep.down_write_killable.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.66            -0.0        0.64        perf-profile.calltrace.cycles-pp.arch_get_unmapped_area_topdown.brk
      0.72            +0.0        0.74        perf-profile.calltrace.cycles-pp.do_munmap.brk
      0.90            +0.0        0.93        perf-profile.calltrace.cycles-pp.___might_sleep.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      4.40            +0.1        4.47        perf-profile.calltrace.cycles-pp.find_vma.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.96            +0.1        2.09        perf-profile.calltrace.cycles-pp.vmacache_find.find_vma.do_munmap.__x64_sys_brk.do_syscall_64
      0.52 ±  2%      +0.2        0.68        perf-profile.calltrace.cycles-pp.__vma_link_rb.brk
      0.35 ± 70%      +0.2        0.54 ±  2%  perf-profile.calltrace.cycles-pp.find_vma.brk
      2.20            +0.3        2.50        perf-profile.calltrace.cycles-pp.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     64.62            +0.3       64.94        perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.brk
     60.53            +0.4       60.92        perf-profile.calltrace.cycles-pp.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
     63.20            +0.4       63.60        perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      3.73            +0.5        4.26        perf-profile.calltrace.cycles-pp.vma_link.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00            +0.6        0.56        perf-profile.calltrace.cycles-pp.free_pgtables.unmap_region.do_munmap.__x64_sys_brk.do_syscall_64
     24.54            +0.6       25.14        perf-profile.calltrace.cycles-pp.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      0.00            +0.6        0.64        perf-profile.calltrace.cycles-pp.put_vma.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64
      0.71            +0.6        1.36        perf-profile.calltrace.cycles-pp.__vma_rb_erase.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00            +0.7        0.70        perf-profile.calltrace.cycles-pp._raw_write_lock.__vma_rb_erase.do_munmap.__x64_sys_brk.do_syscall_64
      3.10            +0.7        3.82        perf-profile.calltrace.cycles-pp.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk.do_syscall_64
      0.00            +0.8        0.76        perf-profile.calltrace.cycles-pp._raw_write_lock.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk
      0.00            +0.8        0.85        perf-profile.calltrace.cycles-pp.__vma_merge.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      5.09            -0.5        4.62        perf-profile.children.cycles-pp.vma_compute_subtree_gap
      4.54            -0.3        4.21        perf-profile.children.cycles-pp.kmem_cache_alloc
      8.11            -0.2        7.89        perf-profile.children.cycles-pp.perf_event_mmap
      8.05            -0.2        7.85        perf-profile.children.cycles-pp.unmap_vmas
     15.01            -0.2       14.81        perf-profile.children.cycles-pp.syscall_return_via_sysret
     29.20            -0.1       29.06        perf-profile.children.cycles-pp.do_brk_flags
      1.11            -0.1        1.00        perf-profile.children.cycles-pp.kmem_cache_free
     12.28            -0.1       12.17        perf-profile.children.cycles-pp.unmap_region
      7.83            -0.1        7.74        perf-profile.children.cycles-pp.unmap_page_range
      0.87 ±  3%      -0.1        0.79        perf-profile.children.cycles-pp.__vm_enough_memory
      1.29            -0.1        1.22        perf-profile.children.cycles-pp.__indirect_thunk_start
      1.81            -0.1        1.74        perf-profile.children.cycles-pp.strlcpy
      4.65            -0.1        4.58        perf-profile.children.cycles-pp.security_vm_enough_memory_mm
      3.08            -0.1        3.02        perf-profile.children.cycles-pp.down_write_killable
      2.88            -0.1        2.82        perf-profile.children.cycles-pp.selinux_vm_enough_memory
      0.73            -0.1        0.67        perf-profile.children.cycles-pp.sync_mm_rss
      3.65            -0.1        3.59        perf-profile.children.cycles-pp.get_unmapped_area
      2.26            -0.1        2.20        perf-profile.children.cycles-pp.cred_has_capability
      1.12            -0.1        1.07        perf-profile.children.cycles-pp.memcpy_erms
      0.39            -0.0        0.35        perf-profile.children.cycles-pp.__rb_insert_augmented
      2.52            -0.0        2.48        perf-profile.children.cycles-pp.perf_iterate_sb
      2.13            -0.0        2.09        perf-profile.children.cycles-pp.security_mmap_addr
      0.55 ±  2%      -0.0        0.52        perf-profile.children.cycles-pp.unmap_single_vma
      1.62            -0.0        1.59        perf-profile.children.cycles-pp.memset_erms
      0.13 ±  3%      -0.0        0.11 ±  4%  perf-profile.children.cycles-pp.__vma_link_file
      0.80            -0.0        0.77        perf-profile.children.cycles-pp._raw_spin_lock
      0.43            -0.0        0.41        perf-profile.children.cycles-pp.strlen
      0.07 ±  6%      -0.0        0.06 ±  8%  perf-profile.children.cycles-pp.should_failslab
      0.43            -0.0        0.42        perf-profile.children.cycles-pp.may_expand_vm
      0.15            +0.0        0.16        perf-profile.children.cycles-pp.__vma_link_list
      0.45            +0.0        0.47        perf-profile.children.cycles-pp.rcu_all_qs
      0.81            +0.1        0.89        perf-profile.children.cycles-pp.free_pgtables
      6.35            +0.1        6.49        perf-profile.children.cycles-pp.find_vma
      2.28            +0.2        2.45        perf-profile.children.cycles-pp.vmacache_find
     64.66            +0.3       64.98        perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
      2.42            +0.3        2.76        perf-profile.children.cycles-pp.remove_vma
     61.77            +0.4       62.13        perf-profile.children.cycles-pp.__x64_sys_brk
     63.40            +0.4       63.79        perf-profile.children.cycles-pp.do_syscall_64
      1.27            +0.4        1.72        perf-profile.children.cycles-pp.__vma_rb_erase
      4.02            +0.5        4.53        perf-profile.children.cycles-pp.vma_link
     25.26            +0.6       25.89        perf-profile.children.cycles-pp.do_munmap
      0.00            +0.7        0.70        perf-profile.children.cycles-pp.put_vma
      3.80            +0.7        4.53        perf-profile.children.cycles-pp.__vma_link_rb
      0.00            +1.2        1.24        perf-profile.children.cycles-pp.__vma_merge
      0.00            +1.5        1.51        perf-profile.children.cycles-pp._raw_write_lock
      5.07            -0.5        4.60        perf-profile.self.cycles-pp.vma_compute_subtree_gap
      0.59            -0.2        0.38        perf-profile.self.cycles-pp.remove_vma
     15.01            -0.2       14.81        perf-profile.self.cycles-pp.syscall_return_via_sysret
      3.15            -0.2        2.96        perf-profile.self.cycles-pp.do_munmap
      0.98            -0.1        0.87        perf-profile.self.cycles-pp.__vma_rb_erase
      1.10            -0.1        0.99        perf-profile.self.cycles-pp.kmem_cache_free
      0.68            -0.1        0.58        perf-profile.self.cycles-pp.__vm_enough_memory
      0.42            -0.1        0.33        perf-profile.self.cycles-pp.unmap_vmas
      3.62            -0.1        3.53        perf-profile.self.cycles-pp.perf_event_mmap
      1.41            -0.1        1.34        perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
      1.29            -0.1        1.22        perf-profile.self.cycles-pp.__indirect_thunk_start
      0.73            -0.1        0.66        perf-profile.self.cycles-pp.sync_mm_rss
      2.96            -0.1        2.90        perf-profile.self.cycles-pp.__x64_sys_brk
      3.24            -0.1        3.19        perf-profile.self.cycles-pp.brk
      1.11            -0.0        1.07        perf-profile.self.cycles-pp.memcpy_erms
      0.53 ±  3%      -0.0        0.49 ±  2%  perf-profile.self.cycles-pp.vma_link
      0.73            -0.0        0.69        perf-profile.self.cycles-pp.unmap_region
      1.66            -0.0        1.61        perf-profile.self.cycles-pp.down_write_killable
      0.39            -0.0        0.35        perf-profile.self.cycles-pp.__rb_insert_augmented
      1.74            -0.0        1.71        perf-profile.self.cycles-pp.kmem_cache_alloc
      0.55 ±  2%      -0.0        0.52        perf-profile.self.cycles-pp.unmap_single_vma
      1.61            -0.0        1.59        perf-profile.self.cycles-pp.memset_erms
      0.80            -0.0        0.77        perf-profile.self.cycles-pp._raw_spin_lock
      0.13            -0.0        0.11 ±  4%  perf-profile.self.cycles-pp.__vma_link_file
      0.43            -0.0        0.41        perf-profile.self.cycles-pp.strlen
      0.07 ±  6%      -0.0        0.06 ±  8%  perf-profile.self.cycles-pp.should_failslab
      0.81            -0.0        0.79        perf-profile.self.cycles-pp.tlb_finish_mmu
      0.15            +0.0        0.16        perf-profile.self.cycles-pp.__vma_link_list
      0.45            +0.0        0.47        perf-profile.self.cycles-pp.rcu_all_qs
      0.71            +0.0        0.72        perf-profile.self.cycles-pp.strlcpy
      0.51            +0.1        0.56        perf-profile.self.cycles-pp.free_pgtables
      1.41            +0.1        1.48        perf-profile.self.cycles-pp.__vma_link_rb
      2.27            +0.2        2.44        perf-profile.self.cycles-pp.vmacache_find
      0.00            +0.7        0.69        perf-profile.self.cycles-pp.put_vma
      0.00            +1.2        1.23        perf-profile.self.cycles-pp.__vma_merge
      0.00            +1.5        1.50        perf-profile.self.cycles-pp._raw_write_lock

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/always/brk1/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
           :3           33%           1:3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=schedule_tail/0x
           :3           33%           1:3     kmsg.DHCP/BOOTP:Reply_not_for_us_on_eth#,op[#]xid[#]
         %stddev     %change         %stddev
             \          |                \  
    998475            -2.2%     976893        will-it-scale.per_process_ops
    625.87            -2.3%     611.42        will-it-scale.time.elapsed_time
    625.87            -2.3%     611.42        will-it-scale.time.elapsed_time.max
      8158            -1.9%       8000        will-it-scale.time.maximum_resident_set_size
     18.42 ±  2%     -11.9%      16.24        will-it-scale.time.user_time
  34349225 ± 13%     -14.5%   29371024 ± 17%  will-it-scale.time.voluntary_context_switches
 1.919e+08            -2.2%  1.877e+08        will-it-scale.workload
      1639 ± 23%     -18.4%       1337 ± 30%  meminfo.Mlocked
     17748 ± 82%    +103.1%      36051        numa-numastat.node3.other_node
  33410486 ± 14%     -14.8%   28449258 ± 18%  cpuidle.C1.usage
    698749 ± 15%     -18.0%     573307 ± 20%  cpuidle.POLL.usage
   3013702 ± 14%     -15.1%    2559405 ± 17%  softirqs.SCHED
  54361293 ±  2%     -19.0%   44044816 ±  2%  softirqs.TIMER
  33408303 ± 14%     -14.9%   28447123 ± 18%  turbostat.C1
      0.34 ± 16%     -52.0%       0.16 ± 15%  turbostat.Pkg%pc2
      1310 ± 74%    +412.1%       6710 ± 58%  irq_exception_noise.__do_page_fault.samples
      3209 ± 74%    +281.9%      12258 ± 53%  irq_exception_noise.__do_page_fault.sum
    600.67 ±132%     -96.0%      24.00 ± 23%  irq_exception_noise.irq_nr
     99557 ±  7%     -24.0%      75627 ±  7%  irq_exception_noise.softirq_nr
     41424 ±  9%     -24.6%      31253 ±  6%  irq_exception_noise.softirq_time
    625.87            -2.3%     611.42        time.elapsed_time
    625.87            -2.3%     611.42        time.elapsed_time.max
      8158            -1.9%       8000        time.maximum_resident_set_size
     18.42 ±  2%     -11.9%      16.24        time.user_time
  34349225 ± 13%     -14.5%   29371024 ± 17%  time.voluntary_context_switches
    988.00 ±  8%     +14.5%       1131 ±  2%  slabinfo.Acpi-ParseExt.active_objs
    988.00 ±  8%     +14.5%       1131 ±  2%  slabinfo.Acpi-ParseExt.num_objs
      2384 ±  3%     +21.1%       2888 ± 11%  slabinfo.pool_workqueue.active_objs
      2474 ±  2%     +20.4%       2979 ± 11%  slabinfo.pool_workqueue.num_objs
    490.33 ± 10%     -19.2%     396.00 ± 11%  slabinfo.secpath_cache.active_objs
    490.33 ± 10%     -19.2%     396.00 ± 11%  slabinfo.secpath_cache.num_objs
      1123 ±  7%     +14.2%       1282 ±  3%  slabinfo.skbuff_fclone_cache.active_objs
      1123 ±  7%     +14.2%       1282 ±  3%  slabinfo.skbuff_fclone_cache.num_objs
      1.09            -0.0        1.07        perf-stat.branch-miss-rate%
 2.691e+11            -2.4%  2.628e+11        perf-stat.branch-misses
  71981351 ± 12%     -13.8%   62013509 ± 16%  perf-stat.context-switches
 1.697e+13            +1.1%  1.715e+13        perf-stat.dTLB-stores
      2.36 ± 29%      +4.4        6.76 ± 11%  perf-stat.iTLB-load-miss-rate%
  5.21e+08 ± 28%    +194.8%  1.536e+09 ± 10%  perf-stat.iTLB-load-misses
    239983 ± 24%     -68.4%      75819 ± 11%  perf-stat.instructions-per-iTLB-miss
   3295653 ±  2%      -6.3%    3088753 ±  3%  perf-stat.node-stores
    606239            +1.1%     612799        perf-stat.path-length
      3755 ± 28%     -37.5%       2346 ± 52%  sched_debug.cfs_rq:/.exec_clock.stddev
     10.45 ±  4%     +24.3%      12.98 ± 18%  sched_debug.cfs_rq:/.load_avg.stddev
      6243 ± 46%     -38.6%       3831 ± 78%  sched_debug.cpu.load.stddev
    867.80 ±  7%     +25.3%       1087 ±  6%  sched_debug.cpu.nr_load_updates.stddev
    395898 ±  3%     -11.1%     352071 ±  7%  sched_debug.cpu.nr_switches.max
    -13.33           -21.1%     -10.52        sched_debug.cpu.nr_uninterruptible.min
    395674 ±  3%     -11.1%     351762 ±  7%  sched_debug.cpu.sched_count.max
     33152 ±  4%     -12.8%      28899        sched_debug.cpu.ttwu_count.min
      0.03 ± 20%     +77.7%       0.05 ± 15%  sched_debug.rt_rq:/.rt_time.max
     89523            +1.8%      91099        proc-vmstat.nr_active_anon
    409.67 ± 23%     -18.4%     334.33 ± 30%  proc-vmstat.nr_mlock
     89530            +1.8%      91117        proc-vmstat.nr_zone_active_anon
   2337130            -2.2%    2286775        proc-vmstat.numa_hit
   2229090            -2.3%    2178626        proc-vmstat.numa_local
      8460 ± 39%     -75.5%       2076 ± 53%  proc-vmstat.numa_pages_migrated
     28643 ± 55%     -83.5%       4727 ± 58%  proc-vmstat.numa_pte_updates
   2695806            -1.8%    2646639        proc-vmstat.pgfault
   2330191            -2.1%    2281197        proc-vmstat.pgfree
      8460 ± 39%     -75.5%       2076 ± 53%  proc-vmstat.pgmigrate_success
    237651 ±  2%     +31.3%     312092 ± 16%  numa-meminfo.node0.FilePages
      8059 ±  2%     +10.7%       8925 ±  7%  numa-meminfo.node0.KernelStack
      6830 ± 25%     +48.8%      10164 ± 35%  numa-meminfo.node0.Mapped
      1612 ± 21%     +70.0%       2740 ± 19%  numa-meminfo.node0.PageTables
     10772 ± 65%    +679.4%      83962 ± 59%  numa-meminfo.node0.Shmem
    163195 ± 15%     -36.9%     103036 ± 32%  numa-meminfo.node1.Active
    163195 ± 15%     -36.9%     103036 ± 32%  numa-meminfo.node1.Active(anon)
      1730 ±  4%     +33.9%       2317 ± 14%  numa-meminfo.node1.PageTables
     55778 ± 19%     +32.5%      73910 ±  8%  numa-meminfo.node1.SUnreclaim
      2671 ± 16%     -45.0%       1469 ± 15%  numa-meminfo.node2.PageTables
     61537 ± 13%     -17.7%      50647 ±  3%  numa-meminfo.node2.SUnreclaim
     48644 ± 94%    +149.8%     121499 ± 11%  numa-meminfo.node3.Active
     48440 ± 94%    +150.4%     121295 ± 11%  numa-meminfo.node3.Active(anon)
     11832 ± 79%     -91.5%       1008 ± 67%  numa-meminfo.node3.Inactive
     11597 ± 82%     -93.3%     772.00 ± 82%  numa-meminfo.node3.Inactive(anon)
     10389 ± 32%     -43.0%       5921 ±  6%  numa-meminfo.node3.Mapped
     33704 ± 24%     -44.2%      18792 ± 15%  numa-meminfo.node3.SReclaimable
    104733 ± 14%     -25.3%      78275 ±  8%  numa-meminfo.node3.Slab
    139329 ±133%     -99.8%     241.67 ± 79%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
      5403 ±139%     -97.5%     137.67 ± 71%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
    165968 ±101%     -61.9%      63304 ± 58%  latency_stats.avg.max
     83.00        +12810.4%      10715 ±140%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_access.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat.filename_lookup
    102.67 ±  6%  +18845.5%      19450 ±140%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    136.33 ± 16%  +25043.5%      34279 ±141%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.__lookup_slow.lookup_slow.walk_component.path_lookupat.filename_lookup
     18497 ±141%    -100.0%       0.00        latency_stats.max.call_rwsem_down_write_failed_killable.vm_mmap_pgoff.ksys_mmap_pgoff.do_syscall_64.entry_SYSCALL_64_after_hwframe
    140500 ±131%     -99.8%     247.00 ± 78%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
      5403 ±139%     -97.5%     137.67 ± 71%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
     87.33 ±  5%  +23963.0%      21015 ±140%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_access.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat.filename_lookup
    136.33 ± 16%  +25043.5%      34279 ±141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.__lookup_slow.lookup_slow.walk_component.path_lookupat.filename_lookup
    149.33 ± 14%  +25485.9%      38208 ±141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
     18761 ±141%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_write_failed_killable.vm_mmap_pgoff.ksys_mmap_pgoff.do_syscall_64.entry_SYSCALL_64_after_hwframe
     23363 ±114%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_read_failed.__do_page_fault.do_page_fault.page_fault.__get_user_8.exit_robust_list.mm_release.do_exit.do_group_exit.get_signal.do_signal.exit_to_usermode_loop
    144810 ±125%     -99.8%     326.67 ± 70%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_do_create.nfs3_proc_create.nfs_create.path_openat.do_filp_open.do_sys_open.do_syscall_64
      5403 ±139%     -97.5%     137.67 ± 71%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.path_openat.do_filp_open.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
     59698 ± 98%     -78.0%      13110 ±141%  latency_stats.sum.call_rwsem_down_read_failed.do_exit.do_group_exit.get_signal.do_signal.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
    166.33        +12768.5%      21404 ±140%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_access.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat.filename_lookup
    825.00 ±  6%  +18761.7%     155609 ±140%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    136.33 ± 16%  +25043.5%      34279 ±141%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup.__lookup_slow.lookup_slow.walk_component.path_lookupat.filename_lookup
     59412 ±  2%     +31.3%      78021 ± 16%  numa-vmstat.node0.nr_file_pages
      8059 ±  2%     +10.7%       8923 ±  7%  numa-vmstat.node0.nr_kernel_stack
      1701 ± 25%     +49.1%       2536 ± 35%  numa-vmstat.node0.nr_mapped
    402.33 ± 21%     +70.0%     684.00 ± 19%  numa-vmstat.node0.nr_page_table_pages
      2692 ± 65%    +679.5%      20988 ± 59%  numa-vmstat.node0.nr_shmem
    622587 ± 36%     +37.7%     857545 ± 13%  numa-vmstat.node0.numa_local
     40797 ± 15%     -36.9%      25757 ± 32%  numa-vmstat.node1.nr_active_anon
    432.00 ±  4%     +33.9%     578.33 ± 14%  numa-vmstat.node1.nr_page_table_pages
     13944 ± 19%     +32.5%      18477 ±  8%  numa-vmstat.node1.nr_slab_unreclaimable
     40797 ± 15%     -36.9%      25757 ± 32%  numa-vmstat.node1.nr_zone_active_anon
    625073 ± 26%     +29.4%     808657 ± 18%  numa-vmstat.node1.numa_hit
    503969 ± 34%     +39.2%     701446 ± 23%  numa-vmstat.node1.numa_local
    137.33 ± 40%     -49.0%      70.00 ± 29%  numa-vmstat.node2.nr_mlock
    667.67 ± 17%     -45.1%     366.33 ± 15%  numa-vmstat.node2.nr_page_table_pages
     15384 ± 13%     -17.7%      12662 ±  3%  numa-vmstat.node2.nr_slab_unreclaimable
     12114 ± 94%    +150.3%      30326 ± 11%  numa-vmstat.node3.nr_active_anon
      2887 ± 83%     -93.4%     190.00 ± 82%  numa-vmstat.node3.nr_inactive_anon
      2632 ± 30%     -39.2%       1600 ±  5%  numa-vmstat.node3.nr_mapped
    101.00           -30.0%      70.67 ± 29%  numa-vmstat.node3.nr_mlock
      8425 ± 24%     -44.2%       4697 ± 15%  numa-vmstat.node3.nr_slab_reclaimable
     12122 ± 94%    +150.3%      30346 ± 11%  numa-vmstat.node3.nr_zone_active_anon
      2887 ± 83%     -93.4%     190.00 ± 82%  numa-vmstat.node3.nr_zone_inactive_anon
    106945 ± 13%     +17.4%     125554        numa-vmstat.node3.numa_other
      4.17            -0.3        3.82        perf-profile.calltrace.cycles-pp.kmem_cache_alloc.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     15.02            -0.3       14.77        perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.brk
      2.42            -0.2        2.18        perf-profile.calltrace.cycles-pp.vma_compute_subtree_gap.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk
      7.60            -0.2        7.39        perf-profile.calltrace.cycles-pp.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      7.79            -0.2        7.63        perf-profile.calltrace.cycles-pp.unmap_vmas.unmap_region.do_munmap.__x64_sys_brk.do_syscall_64
      0.82 ±  9%      -0.1        0.68        perf-profile.calltrace.cycles-pp.__vm_enough_memory.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      2.13            -0.1        2.00        perf-profile.calltrace.cycles-pp.vma_compute_subtree_gap.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.05            -0.1        0.95        perf-profile.calltrace.cycles-pp.kmem_cache_free.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64
      7.31            -0.1        7.21        perf-profile.calltrace.cycles-pp.unmap_page_range.unmap_vmas.unmap_region.do_munmap.__x64_sys_brk
      0.74            -0.1        0.67        perf-profile.calltrace.cycles-pp.sync_mm_rss.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      1.06            -0.1        1.00        perf-profile.calltrace.cycles-pp.memcpy_erms.strlcpy.perf_event_mmap.do_brk_flags.__x64_sys_brk
      3.38            -0.1        3.33        perf-profile.calltrace.cycles-pp.get_unmapped_area.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.05            -0.0        1.00 ±  2%  perf-profile.calltrace.cycles-pp.__indirect_thunk_start.brk
      2.34            -0.0        2.29        perf-profile.calltrace.cycles-pp.perf_iterate_sb.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64
      1.64            -0.0        1.59        perf-profile.calltrace.cycles-pp.strlcpy.perf_event_mmap.do_brk_flags.__x64_sys_brk.do_syscall_64
      1.89            -0.0        1.86        perf-profile.calltrace.cycles-pp.security_mmap_addr.get_unmapped_area.do_brk_flags.__x64_sys_brk.do_syscall_64
      0.76            -0.0        0.73        perf-profile.calltrace.cycles-pp._raw_spin_lock.unmap_page_range.unmap_vmas.unmap_region.do_munmap
      0.57 ±  2%      -0.0        0.55        perf-profile.calltrace.cycles-pp.selinux_mmap_addr.security_mmap_addr.get_unmapped_area.do_brk_flags.__x64_sys_brk
      0.54 ±  2%      +0.0        0.56        perf-profile.calltrace.cycles-pp.do_brk_flags.brk
      0.72            +0.0        0.76 ±  2%  perf-profile.calltrace.cycles-pp.do_munmap.brk
      4.38            +0.1        4.43        perf-profile.calltrace.cycles-pp.find_vma.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      1.96            +0.1        2.04        perf-profile.calltrace.cycles-pp.vmacache_find.find_vma.do_munmap.__x64_sys_brk.do_syscall_64
      0.53            +0.2        0.68        perf-profile.calltrace.cycles-pp.__vma_link_rb.brk
      2.21            +0.3        2.51        perf-profile.calltrace.cycles-pp.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     64.44            +0.5       64.90        perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.brk
     63.04            +0.5       63.54        perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
     60.37            +0.5       60.88        perf-profile.calltrace.cycles-pp.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      3.75            +0.5        4.29        perf-profile.calltrace.cycles-pp.vma_link.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00            +0.6        0.57        perf-profile.calltrace.cycles-pp.free_pgtables.unmap_region.do_munmap.__x64_sys_brk.do_syscall_64
      0.00            +0.6        0.64        perf-profile.calltrace.cycles-pp.put_vma.remove_vma.do_munmap.__x64_sys_brk.do_syscall_64
      0.72            +0.7        1.37        perf-profile.calltrace.cycles-pp.__vma_rb_erase.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
     24.42            +0.7       25.08        perf-profile.calltrace.cycles-pp.do_munmap.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe.brk
      0.00            +0.7        0.71        perf-profile.calltrace.cycles-pp._raw_write_lock.__vma_rb_erase.do_munmap.__x64_sys_brk.do_syscall_64
      3.12            +0.7        3.84        perf-profile.calltrace.cycles-pp.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk.do_syscall_64
      0.00            +0.8        0.77        perf-profile.calltrace.cycles-pp._raw_write_lock.__vma_link_rb.vma_link.do_brk_flags.__x64_sys_brk
      0.00            +0.9        0.85        perf-profile.calltrace.cycles-pp.__vma_merge.do_brk_flags.__x64_sys_brk.do_syscall_64.entry_SYSCALL_64_after_hwframe
      5.10            -0.5        4.60        perf-profile.children.cycles-pp.vma_compute_subtree_gap
      4.53            -0.3        4.18        perf-profile.children.cycles-pp.kmem_cache_alloc
     15.03            -0.3       14.77        perf-profile.children.cycles-pp.syscall_return_via_sysret
      8.13            -0.2        7.92        perf-profile.children.cycles-pp.perf_event_mmap
      8.01            -0.2        7.81        perf-profile.children.cycles-pp.unmap_vmas
      0.97 ± 14%      -0.2        0.78        perf-profile.children.cycles-pp.__vm_enough_memory
      1.13            -0.1        1.00        perf-profile.children.cycles-pp.kmem_cache_free
      7.82            -0.1        7.70        perf-profile.children.cycles-pp.unmap_page_range
     12.23            -0.1       12.13        perf-profile.children.cycles-pp.unmap_region
      0.74            -0.1        0.67        perf-profile.children.cycles-pp.sync_mm_rss
      3.06            -0.1        3.00        perf-profile.children.cycles-pp.down_write_killable
      0.40 ±  2%      -0.1        0.34        perf-profile.children.cycles-pp.__rb_insert_augmented
      1.29            -0.1        1.23        perf-profile.children.cycles-pp.__indirect_thunk_start
      2.54            -0.1        2.49        perf-profile.children.cycles-pp.perf_iterate_sb
      3.66            -0.0        3.61        perf-profile.children.cycles-pp.get_unmapped_area
      1.80            -0.0        1.75        perf-profile.children.cycles-pp.strlcpy
      0.53 ±  2%      -0.0        0.49 ±  2%  perf-profile.children.cycles-pp.cap_capable
      1.57            -0.0        1.53        perf-profile.children.cycles-pp.arch_get_unmapped_area_topdown
      1.11            -0.0        1.08        perf-profile.children.cycles-pp.memcpy_erms
      0.13            -0.0        0.10        perf-profile.children.cycles-pp.__vma_link_file
      0.55            -0.0        0.52        perf-profile.children.cycles-pp.unmap_single_vma
      1.47            -0.0        1.44        perf-profile.children.cycles-pp.cap_vm_enough_memory
      2.14            -0.0        2.12        perf-profile.children.cycles-pp.security_mmap_addr
      0.32            -0.0        0.30        perf-profile.children.cycles-pp.userfaultfd_unmap_complete
      1.25            -0.0        1.23        perf-profile.children.cycles-pp.up_write
      0.50            -0.0        0.49        perf-profile.children.cycles-pp.userfaultfd_unmap_prep
      0.27            -0.0        0.26        perf-profile.children.cycles-pp.tlb_flush_mmu_free
      1.14            -0.0        1.12        perf-profile.children.cycles-pp.__might_sleep
      0.07            -0.0        0.06        perf-profile.children.cycles-pp.should_failslab
      0.72            +0.0        0.74        perf-profile.children.cycles-pp._cond_resched
      0.45            +0.0        0.47        perf-profile.children.cycles-pp.rcu_all_qs
      0.15 ±  3%      +0.0        0.17 ±  4%  perf-profile.children.cycles-pp.__vma_link_list
      0.15 ±  5%      +0.0        0.18 ±  5%  perf-profile.children.cycles-pp.tick_sched_timer
      0.05 ±  8%      +0.1        0.12 ± 17%  perf-profile.children.cycles-pp.perf_mux_hrtimer_handler
      0.80            +0.1        0.89        perf-profile.children.cycles-pp.free_pgtables
      0.22 ±  7%      +0.1        0.31 ±  9%  perf-profile.children.cycles-pp.__hrtimer_run_queues
      0.00            +0.1        0.11 ± 15%  perf-profile.children.cycles-pp.clockevents_program_event
      6.34            +0.1        6.47        perf-profile.children.cycles-pp.find_vma
      2.27            +0.1        2.40        perf-profile.children.cycles-pp.vmacache_find
      0.40 ±  4%      +0.2        0.58 ±  5%  perf-profile.children.cycles-pp.apic_timer_interrupt
      0.40 ±  4%      +0.2        0.58 ±  5%  perf-profile.children.cycles-pp.smp_apic_timer_interrupt
      0.37 ±  4%      +0.2        0.54 ±  5%  perf-profile.children.cycles-pp.hrtimer_interrupt
      0.00            +0.2        0.19 ± 12%  perf-profile.children.cycles-pp.ktime_get
      2.42            +0.3        2.77        perf-profile.children.cycles-pp.remove_vma
     64.49            +0.5       64.94        perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
      1.27            +0.5        1.73        perf-profile.children.cycles-pp.__vma_rb_erase
     61.62            +0.5       62.10        perf-profile.children.cycles-pp.__x64_sys_brk
     63.24            +0.5       63.74        perf-profile.children.cycles-pp.do_syscall_64
      4.03            +0.5        4.56        perf-profile.children.cycles-pp.vma_link
      0.00            +0.7        0.69        perf-profile.children.cycles-pp.put_vma
     25.13            +0.7       25.84        perf-profile.children.cycles-pp.do_munmap
      3.83            +0.7        4.56        perf-profile.children.cycles-pp.__vma_link_rb
      0.00            +1.2        1.25        perf-profile.children.cycles-pp.__vma_merge
      0.00            +1.5        1.53        perf-profile.children.cycles-pp._raw_write_lock
      5.08            -0.5        4.58        perf-profile.self.cycles-pp.vma_compute_subtree_gap
     15.03            -0.3       14.77        perf-profile.self.cycles-pp.syscall_return_via_sysret
      0.59            -0.2        0.39        perf-profile.self.cycles-pp.remove_vma
      0.72 ±  7%      -0.1        0.58        perf-profile.self.cycles-pp.__vm_enough_memory
      1.12            -0.1        0.99        perf-profile.self.cycles-pp.kmem_cache_free
      3.11            -0.1        2.99        perf-profile.self.cycles-pp.do_munmap
      0.99            -0.1        0.88        perf-profile.self.cycles-pp.__vma_rb_erase
      3.63            -0.1        3.52        perf-profile.self.cycles-pp.perf_event_mmap
      3.26            -0.1        3.17        perf-profile.self.cycles-pp.brk
      0.41 ±  2%      -0.1        0.33        perf-profile.self.cycles-pp.unmap_vmas
      0.74            -0.1        0.67        perf-profile.self.cycles-pp.sync_mm_rss
      1.75            -0.1        1.68        perf-profile.self.cycles-pp.kmem_cache_alloc
      0.40 ±  2%      -0.1        0.34        perf-profile.self.cycles-pp.__rb_insert_augmented
      1.29 ±  2%      -0.1        1.23        perf-profile.self.cycles-pp.__indirect_thunk_start
      0.73            -0.0        0.68 ±  2%  perf-profile.self.cycles-pp.unmap_region
      0.53            -0.0        0.49        perf-profile.self.cycles-pp.vma_link
      1.40            -0.0        1.35        perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
      5.22            -0.0        5.18        perf-profile.self.cycles-pp.unmap_page_range
      0.53 ±  2%      -0.0        0.49 ±  2%  perf-profile.self.cycles-pp.cap_capable
      1.11            -0.0        1.07        perf-profile.self.cycles-pp.memcpy_erms
      1.86            -0.0        1.82        perf-profile.self.cycles-pp.perf_iterate_sb
      1.30            -0.0        1.27        perf-profile.self.cycles-pp.arch_get_unmapped_area_topdown
      0.13            -0.0        0.10        perf-profile.self.cycles-pp.__vma_link_file
      0.55            -0.0        0.52        perf-profile.self.cycles-pp.unmap_single_vma
      0.74            -0.0        0.72        perf-profile.self.cycles-pp.selinux_mmap_addr
      0.32            -0.0        0.30        perf-profile.self.cycles-pp.userfaultfd_unmap_complete
      1.13            -0.0        1.12        perf-profile.self.cycles-pp.__might_sleep
      1.24            -0.0        1.23        perf-profile.self.cycles-pp.up_write
      0.50            -0.0        0.49        perf-profile.self.cycles-pp.userfaultfd_unmap_prep
      0.27            -0.0        0.26        perf-profile.self.cycles-pp.tlb_flush_mmu_free
      0.07            -0.0        0.06        perf-profile.self.cycles-pp.should_failslab
      0.45            +0.0        0.47        perf-profile.self.cycles-pp.rcu_all_qs
      0.71            +0.0        0.73        perf-profile.self.cycles-pp.strlcpy
      0.15 ±  3%      +0.0        0.17 ±  4%  perf-profile.self.cycles-pp.__vma_link_list
      0.51            +0.1        0.57        perf-profile.self.cycles-pp.free_pgtables
      1.40            +0.1        1.49        perf-profile.self.cycles-pp.__vma_link_rb
      2.27            +0.1        2.39        perf-profile.self.cycles-pp.vmacache_find
      0.00            +0.2        0.18 ± 12%  perf-profile.self.cycles-pp.ktime_get
      0.00            +0.7        0.69        perf-profile.self.cycles-pp.put_vma
      0.00            +1.2        1.24        perf-profile.self.cycles-pp.__vma_merge
      0.00            +1.5        1.52        perf-profile.self.cycles-pp._raw_write_lock

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/always/page_fault2/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
           :3           33%           1:3     dmesg.WARNING:at#for_ip_native_iret/0x
          1:3          -33%            :3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=__schedule/0x
           :3           33%           1:3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=__slab_free/0x
          1:3          -33%            :3     kmsg.DHCP/BOOTP:Reply_not_for_us_on_eth#,op[#]xid[#]
          3:3         -100%            :3     kmsg.pstore:crypto_comp_decompress_failed,ret=
          3:3         -100%            :3     kmsg.pstore:decompression_failed
          2:3            4%           2:3     perf-profile.calltrace.cycles-pp.sync_regs.error_entry
          5:3            7%           5:3     perf-profile.calltrace.cycles-pp.error_entry
          5:3            7%           5:3     perf-profile.children.cycles-pp.error_entry
          2:3            3%           2:3     perf-profile.self.cycles-pp.error_entry
         %stddev     %change         %stddev
             \          |                \  
      8281 ±  2%     -18.8%       6728        will-it-scale.per_thread_ops
     92778 ±  2%     +17.6%     109080        will-it-scale.time.involuntary_context_switches
  21954366 ±  3%      +4.1%   22857988 ±  2%  will-it-scale.time.maximum_resident_set_size
  4.81e+08 ±  2%     -18.9%  3.899e+08        will-it-scale.time.minor_page_faults
      5804           +12.2%       6512        will-it-scale.time.percent_of_cpu_this_job_got
     34918           +12.2%      39193        will-it-scale.time.system_time
   5638528 ±  2%     -15.3%    4778392        will-it-scale.time.voluntary_context_switches
  15846405            -2.0%   15531034        will-it-scale.workload
   2818137            +1.5%    2861500        interrupts.CAL:Function_call_interrupts
      3.33 ± 28%     -60.0%       1.33 ± 93%  irq_exception_noise.irq_time
      2866           +23.9%       3552 ±  2%  kthread_noise.total_time
   5589674 ± 14%     +31.4%    7344810 ±  6%  meminfo.DirectMap2M
     31169           -16.9%      25906        uptime.idle
     25242 ±  4%     -14.2%      21654 ±  6%  vmstat.system.cs
      7055           -11.6%       6237        boot-time.idle
     21.12           +19.3%      25.19 ±  9%  boot-time.kernel_boot
     20.03 ±  2%      -3.7       16.38        mpstat.cpu.idle%
      0.00 ±  8%      -0.0        0.00 ±  4%  mpstat.cpu.iowait%
   7284147 ±  2%     -16.4%    6092495        softirqs.RCU
   5350756 ±  2%     -10.9%    4769417 ±  4%  softirqs.SCHED
     42933 ± 21%     -28.2%      30807 ±  7%  numa-meminfo.node2.SReclaimable
     63219 ± 13%     -16.6%      52717 ±  6%  numa-meminfo.node2.SUnreclaim
    106153 ± 16%     -21.3%      83525 ±  5%  numa-meminfo.node2.Slab
    247154 ±  4%      -7.6%     228415        numa-meminfo.node3.Unevictable
     11904 ±  4%     +17.1%      13945 ±  8%  numa-vmstat.node0
      2239 ± 22%     -26.6%       1644 ±  2%  numa-vmstat.node2.nr_mapped
     10728 ± 21%     -28.2%       7701 ±  7%  numa-vmstat.node2.nr_slab_reclaimable
     15803 ± 13%     -16.6%      13179 ±  6%  numa-vmstat.node2.nr_slab_unreclaimable
     61788 ±  4%      -7.6%      57103        numa-vmstat.node3.nr_unevictable
     61788 ±  4%      -7.6%      57103        numa-vmstat.node3.nr_zone_unevictable
     92778 ±  2%     +17.6%     109080        time.involuntary_context_switches
  21954366 ±  3%      +4.1%   22857988 ±  2%  time.maximum_resident_set_size
  4.81e+08 ±  2%     -18.9%  3.899e+08        time.minor_page_faults
      5804           +12.2%       6512        time.percent_of_cpu_this_job_got
     34918           +12.2%      39193        time.system_time
   5638528 ±  2%     -15.3%    4778392        time.voluntary_context_switches
   3942289 ±  2%     -10.5%    3528902 ±  2%  cpuidle.C1.time
    242290           -14.2%     207992        cpuidle.C1.usage
  1.64e+09 ±  2%     -15.7%  1.381e+09        cpuidle.C1E.time
   4621281 ±  2%     -14.7%    3939757        cpuidle.C1E.usage
 2.115e+10 ±  2%     -18.5%  1.723e+10        cpuidle.C6.time
  24771099 ±  2%     -18.0%   20305766        cpuidle.C6.usage
   1210810 ±  4%     -17.6%     997270 ±  2%  cpuidle.POLL.time
     18742 ±  3%     -17.0%      15559 ±  2%  cpuidle.POLL.usage
      4135 ±141%    -100.0%       0.00        latency_stats.avg.x86_reserve_hardware.x86_pmu_event_init.perf_try_init_event.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
     33249 ±129%    -100.0%       0.00        latency_stats.max.call_rwsem_down_read_failed.m_start.seq_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4135 ±141%    -100.0%       0.00        latency_stats.max.x86_reserve_hardware.x86_pmu_event_init.perf_try_init_event.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
     65839 ±116%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_read_failed.m_start.seq_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4135 ±141%    -100.0%       0.00        latency_stats.sum.x86_reserve_hardware.x86_pmu_event_init.perf_try_init_event.perf_event_alloc.__do_sys_perf_event_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      8387 ±122%     -90.9%     767.00 ± 13%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    263970 ± 10%     -68.6%      82994 ±  3%  latency_stats.sum.do_syslog.kmsg_read.proc_reg_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      6173 ± 77%    +173.3%      16869 ± 98%  latency_stats.sum.pipe_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
    101.33            -4.6%      96.67        proc-vmstat.nr_anon_transparent_hugepages
     39967            -1.8%      39241        proc-vmstat.nr_slab_reclaimable
     67166            -2.4%      65522        proc-vmstat.nr_slab_unreclaimable
    237743            -3.9%     228396        proc-vmstat.nr_unevictable
    237743            -3.9%     228396        proc-vmstat.nr_zone_unevictable
 4.807e+09            -2.0%   4.71e+09        proc-vmstat.numa_hit
 4.807e+09            -2.0%   4.71e+09        proc-vmstat.numa_local
 4.791e+09            -2.1%   4.69e+09        proc-vmstat.pgalloc_normal
 4.783e+09            -2.0%  4.685e+09        proc-vmstat.pgfault
 4.807e+09            -2.0%  4.709e+09        proc-vmstat.pgfree
      1753            +4.6%       1833        turbostat.Avg_MHz
    239445           -14.1%     205783        turbostat.C1
   4617105 ±  2%     -14.8%    3934693        turbostat.C1E
      1.40 ±  2%      -0.2        1.18        turbostat.C1E%
  24764661 ±  2%     -18.0%   20297643        turbostat.C6
     18.09 ±  2%      -3.4       14.74        turbostat.C6%
      7.53 ±  2%     -17.1%       6.24        turbostat.CPU%c1
     11.88 ±  2%     -19.1%       9.61        turbostat.CPU%c6
      7.62 ±  3%     -20.8%       6.04        turbostat.Pkg%pc2
    388.30            +1.5%     393.93        turbostat.PkgWatt
    390974 ±  8%     +35.8%     530867 ± 11%  sched_debug.cfs_rq:/.min_vruntime.stddev
  -1754042           +75.7%   -3081270        sched_debug.cfs_rq:/.spread0.min
    388140 ±  8%     +36.2%     528494 ± 11%  sched_debug.cfs_rq:/.spread0.stddev
    542.30 ±  3%     -10.0%     488.21 ±  3%  sched_debug.cfs_rq:/.util_avg.min
     53.35 ± 16%     +48.7%      79.35 ± 12%  sched_debug.cfs_rq:/.util_est_enqueued.avg
     30520 ±  6%     -15.2%      25883 ± 12%  sched_debug.cpu.nr_switches.avg
    473770 ± 27%     -37.4%     296623 ± 32%  sched_debug.cpu.nr_switches.max
     17077 ±  2%     -15.1%      14493        sched_debug.cpu.nr_switches.min
     30138 ±  6%     -15.0%      25606 ± 12%  sched_debug.cpu.sched_count.avg
    472345 ± 27%     -37.2%     296419 ± 32%  sched_debug.cpu.sched_count.max
     16858 ±  2%     -15.2%      14299        sched_debug.cpu.sched_count.min
      8358 ±  2%     -15.5%       7063        sched_debug.cpu.sched_goidle.avg
     12225           -13.6%      10565        sched_debug.cpu.sched_goidle.max
      8032 ±  2%     -16.0%       6749        sched_debug.cpu.sched_goidle.min
     14839 ±  6%     -15.3%      12568 ± 12%  sched_debug.cpu.ttwu_count.avg
    235115 ± 28%     -38.3%     145175 ± 31%  sched_debug.cpu.ttwu_count.max
      7627 ±  3%     -15.9%       6413 ±  2%  sched_debug.cpu.ttwu_count.min
    226299 ± 29%     -39.5%     136827 ± 32%  sched_debug.cpu.ttwu_local.max
      0.85            -0.0        0.81        perf-stat.branch-miss-rate%
 3.675e+10            -4.1%  3.523e+10        perf-stat.branch-misses
 4.052e+11            -2.3%  3.958e+11        perf-stat.cache-misses
 7.008e+11            -2.5%  6.832e+11        perf-stat.cache-references
  15320995 ±  4%     -14.3%   13136557 ±  6%  perf-stat.context-switches
      9.16            +4.8%       9.59        perf-stat.cpi
  2.03e+14            +4.6%  2.124e+14        perf-stat.cpu-cycles
     44508            -1.7%      43743        perf-stat.cpu-migrations
      1.30            -0.1        1.24        perf-stat.dTLB-store-miss-rate%
 4.064e+10            -3.5%  3.922e+10        perf-stat.dTLB-store-misses
 3.086e+12            +1.1%  3.119e+12        perf-stat.dTLB-stores
 3.611e+08 ±  6%      -8.5%  3.304e+08 ±  5%  perf-stat.iTLB-loads
      0.11            -4.6%       0.10        perf-stat.ipc
 4.783e+09            -2.0%  4.685e+09        perf-stat.minor-faults
      1.53 ±  2%      -0.3        1.22 ±  8%  perf-stat.node-load-miss-rate%
 1.389e+09 ±  3%     -22.1%  1.083e+09 ±  9%  perf-stat.node-load-misses
 8.922e+10            -1.9%   8.75e+10        perf-stat.node-loads
      5.06            +1.7        6.77 ±  3%  perf-stat.node-store-miss-rate%
 1.204e+09           +29.3%  1.556e+09 ±  3%  perf-stat.node-store-misses
 2.256e+10            -5.1%  2.142e+10 ±  2%  perf-stat.node-stores
 4.783e+09            -2.0%  4.685e+09        perf-stat.page-faults
   1399242            +1.9%    1425404        perf-stat.path-length
      1144 ±  8%     -13.6%     988.00 ±  8%  slabinfo.Acpi-ParseExt.active_objs
      1144 ±  8%     -13.6%     988.00 ±  8%  slabinfo.Acpi-ParseExt.num_objs
      1878 ± 17%     +29.0%       2422 ± 16%  slabinfo.dmaengine-unmap-16.active_objs
      1878 ± 17%     +29.0%       2422 ± 16%  slabinfo.dmaengine-unmap-16.num_objs
      1085 ±  5%     -24.1%     823.33 ±  9%  slabinfo.file_lock_cache.active_objs
      1085 ±  5%     -24.1%     823.33 ±  9%  slabinfo.file_lock_cache.num_objs
     61584 ±  4%     -16.6%      51381 ±  5%  slabinfo.filp.active_objs
    967.00 ±  4%     -16.5%     807.67 ±  5%  slabinfo.filp.active_slabs
     61908 ±  4%     -16.5%      51713 ±  5%  slabinfo.filp.num_objs
    967.00 ±  4%     -16.5%     807.67 ±  5%  slabinfo.filp.num_slabs
      1455           -15.4%       1232 ±  4%  slabinfo.nsproxy.active_objs
      1455           -15.4%       1232 ±  4%  slabinfo.nsproxy.num_objs
     84720 ±  6%     -18.3%      69210 ±  4%  slabinfo.pid.active_objs
      1324 ±  6%     -18.2%       1083 ±  4%  slabinfo.pid.active_slabs
     84820 ±  5%     -18.2%      69386 ±  4%  slabinfo.pid.num_objs
      1324 ±  6%     -18.2%       1083 ±  4%  slabinfo.pid.num_slabs
      2112 ± 18%     -26.3%       1557 ±  5%  slabinfo.scsi_sense_cache.active_objs
      2112 ± 18%     -26.3%       1557 ±  5%  slabinfo.scsi_sense_cache.num_objs
      5018 ±  5%      -7.6%       4635 ±  4%  slabinfo.sock_inode_cache.active_objs
      5018 ±  5%      -7.6%       4635 ±  4%  slabinfo.sock_inode_cache.num_objs
      1193 ±  4%     +13.8%       1358 ±  4%  slabinfo.task_group.active_objs
      1193 ±  4%     +13.8%       1358 ±  4%  slabinfo.task_group.num_objs
     62807 ±  3%     -14.4%      53757 ±  3%  slabinfo.vm_area_struct.active_objs
      1571 ±  3%     -12.1%       1381 ±  3%  slabinfo.vm_area_struct.active_slabs
     62877 ±  3%     -14.3%      53880 ±  3%  slabinfo.vm_area_struct.num_objs
      1571 ±  3%     -12.1%       1381 ±  3%  slabinfo.vm_area_struct.num_slabs
     47.45           -47.4        0.00        perf-profile.calltrace.cycles-pp.alloc_pages_vma.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
     47.16           -47.2        0.00        perf-profile.calltrace.cycles-pp.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault.handle_mm_fault.__do_page_fault
     46.99           -47.0        0.00        perf-profile.calltrace.cycles-pp.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault.handle_mm_fault
     44.95           -44.9        0.00        perf-profile.calltrace.cycles-pp._raw_spin_lock.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault
      7.42 ±  2%      -7.4        0.00        perf-profile.calltrace.cycles-pp.copy_page.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      6.32 ± 10%      -6.3        0.00        perf-profile.calltrace.cycles-pp.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      6.28 ± 10%      -6.3        0.00        perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +0.9        0.85 ± 11%  perf-profile.calltrace.cycles-pp._raw_spin_lock.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +0.9        0.92 ±  4%  perf-profile.calltrace.cycles-pp.__list_del_entry_valid.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault
      0.00            +1.1        1.13 ±  7%  perf-profile.calltrace.cycles-pp.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault
      0.00            +1.2        1.19 ±  7%  perf-profile.calltrace.cycles-pp.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault
      0.00            +1.2        1.22 ±  5%  perf-profile.calltrace.cycles-pp.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +1.3        1.34 ±  7%  perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +1.4        1.36 ±  7%  perf-profile.calltrace.cycles-pp.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +4.5        4.54 ± 19%  perf-profile.calltrace.cycles-pp.pagevec_lru_move_fn.__lru_cache_add.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +4.6        4.64 ± 19%  perf-profile.calltrace.cycles-pp.__lru_cache_add.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +6.6        6.64 ± 15%  perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +6.7        6.68 ± 15%  perf-profile.calltrace.cycles-pp.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +7.5        7.54 ±  5%  perf-profile.calltrace.cycles-pp.copy_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00           +44.6       44.55 ±  3%  perf-profile.calltrace.cycles-pp._raw_spin_lock.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault
      0.00           +46.6       46.63 ±  3%  perf-profile.calltrace.cycles-pp.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault.__handle_mm_fault
      0.00           +46.8       46.81 ±  3%  perf-profile.calltrace.cycles-pp.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00           +47.1       47.10 ±  3%  perf-profile.calltrace.cycles-pp.alloc_pages_vma.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00           +63.1       63.15        perf-profile.calltrace.cycles-pp.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      0.39 ±  3%      +0.0        0.42 ±  3%  perf-profile.children.cycles-pp.radix_tree_lookup_slot
      0.21 ±  3%      +0.0        0.25 ±  5%  perf-profile.children.cycles-pp.__mod_node_page_state
      0.00            +0.1        0.06 ±  8%  perf-profile.children.cycles-pp.get_vma_policy
      0.00            +0.1        0.08 ±  5%  perf-profile.children.cycles-pp.__lru_cache_add_active_or_unevictable
      0.00            +0.2        0.18 ±  6%  perf-profile.children.cycles-pp.__page_add_new_anon_rmap
      0.00            +1.4        1.35 ±  5%  perf-profile.children.cycles-pp.pte_map_lock
      0.00           +63.2       63.21        perf-profile.children.cycles-pp.handle_pte_fault
      1.40 ±  2%      -0.4        1.03 ± 10%  perf-profile.self.cycles-pp._raw_spin_lock
      0.56 ±  3%      -0.2        0.35 ±  6%  perf-profile.self.cycles-pp.__handle_mm_fault
      0.22 ±  3%      -0.0        0.18 ±  7%  perf-profile.self.cycles-pp.alloc_set_pte
      0.09            +0.0        0.10 ±  4%  perf-profile.self.cycles-pp.vmacache_find
      0.39 ±  2%      +0.0        0.41 ±  3%  perf-profile.self.cycles-pp.__radix_tree_lookup
      0.18            +0.0        0.20 ±  6%  perf-profile.self.cycles-pp.mem_cgroup_charge_statistics
      0.17 ±  2%      +0.0        0.20 ±  7%  perf-profile.self.cycles-pp.___might_sleep
      0.33 ±  2%      +0.0        0.36 ±  6%  perf-profile.self.cycles-pp.handle_mm_fault
      0.20 ±  2%      +0.0        0.24 ±  3%  perf-profile.self.cycles-pp.__mod_node_page_state
      0.00            +0.1        0.05        perf-profile.self.cycles-pp.finish_fault
      0.00            +0.1        0.05        perf-profile.self.cycles-pp.get_vma_policy
      0.00            +0.1        0.08 ± 10%  perf-profile.self.cycles-pp.__lru_cache_add_active_or_unevictable
      0.00            +0.2        0.25 ±  5%  perf-profile.self.cycles-pp.handle_pte_fault
      0.00            +0.5        0.49 ±  8%  perf-profile.self.cycles-pp.pte_map_lock

=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/thp_enabled/test/cpufreq_governor:
  lkp-skl-4sp1/will-it-scale/debian-x86_64-2018-04-03.cgz/x86_64-rhel-7.2/gcc-7/100%/never/page_fault2/performance

commit: 
  ba98a1cdad71d259a194461b3a61471b49b14df1
  a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12

ba98a1cdad71d259 a7a8993bfe3ccb54ad468b9f17 
---------------- -------------------------- 
       fail:runs  %reproduction    fail:runs
           |             |             |    
          1:3          -33%            :3     kmsg.DHCP/BOOTP:Reply_not_for_us_on_eth#,op[#]xid[#]
           :3           33%           1:3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=sched_slice/0x
          1:3          -33%            :3     dmesg.WARNING:stack_going_in_the_wrong_direction?ip=schedule_tail/0x
          1:3           24%           2:3     perf-profile.calltrace.cycles-pp.sync_regs.error_entry
          3:3           46%           5:3     perf-profile.calltrace.cycles-pp.error_entry
          5:3           -9%           5:3     perf-profile.children.cycles-pp.error_entry
          2:3           -4%           2:3     perf-profile.self.cycles-pp.error_entry
         %stddev     %change         %stddev
             \          |                \  
      8147           -18.8%       6613        will-it-scale.per_thread_ops
     93113           +17.0%     108982        will-it-scale.time.involuntary_context_switches
 4.732e+08           -19.0%  3.833e+08        will-it-scale.time.minor_page_faults
      5854           +12.0%       6555        will-it-scale.time.percent_of_cpu_this_job_got
     35247           +12.1%      39495        will-it-scale.time.system_time
   5546661           -15.5%    4689314        will-it-scale.time.voluntary_context_switches
  15801637            -1.9%   15504487        will-it-scale.workload
      1.43 ± 11%     -59.7%       0.58 ± 28%  irq_exception_noise.__do_page_fault.min
      2811 ±  3%     +23.7%       3477 ±  3%  kthread_noise.total_time
    292776 ±  5%     +39.6%     408829 ± 21%  meminfo.DirectMap4k
     19.80            -3.7       16.12        mpstat.cpu.idle%
     29940           -14.5%      25593        uptime.idle
     24064 ±  3%      -8.5%      22016        vmstat.system.cs
     34.86            -1.9%      34.19        boot-time.boot
     26.95            -2.8%      26.19 ±  2%  boot-time.kernel_boot
   7190569 ±  2%     -15.2%    6100136 ±  3%  softirqs.RCU
   5513663           -13.8%    4751548        softirqs.SCHED
     18064 ±  2%     +24.3%      22461 ±  7%  numa-vmstat.node0.nr_slab_unreclaimable
      8507 ± 12%     -16.8%       7075 ±  4%  numa-vmstat.node2.nr_slab_reclaimable
     18719 ±  9%     -19.6%      15043 ±  4%  numa-vmstat.node3.nr_slab_unreclaimable
     72265 ±  2%     +24.3%      89855 ±  7%  numa-meminfo.node0.SUnreclaim
    115980 ±  4%     +22.6%     142233 ± 12%  numa-meminfo.node0.Slab
     34035 ± 12%     -16.8%      28307 ±  4%  numa-meminfo.node2.SReclaimable
     74888 ±  9%     -19.7%      60162 ±  4%  numa-meminfo.node3.SUnreclaim
     93113           +17.0%     108982        time.involuntary_context_switches
 4.732e+08           -19.0%  3.833e+08        time.minor_page_faults
      5854           +12.0%       6555        time.percent_of_cpu_this_job_got
     35247           +12.1%      39495        time.system_time
   5546661           -15.5%    4689314        time.voluntary_context_switches
 4.792e+09            -1.9%  4.699e+09        proc-vmstat.numa_hit
 4.791e+09            -1.9%  4.699e+09        proc-vmstat.numa_local
     40447 ± 11%     +13.2%      45804 ±  6%  proc-vmstat.pgactivate
 4.778e+09            -1.9%  4.688e+09        proc-vmstat.pgalloc_normal
 4.767e+09            -1.9%  4.675e+09        proc-vmstat.pgfault
 4.791e+09            -1.9%  4.699e+09        proc-vmstat.pgfree
    230178 ±  2%     -10.1%     206883 ±  3%  cpuidle.C1.usage
 1.617e+09           -15.0%  1.375e+09        cpuidle.C1E.time
   4514401           -14.1%    3878206        cpuidle.C1E.usage
 2.087e+10           -18.5%  1.701e+10        cpuidle.C6.time
  24458365           -18.0%   20045336        cpuidle.C6.usage
   1163758           -16.1%     976094 ±  4%  cpuidle.POLL.time
     17907           -14.6%      15294 ±  4%  cpuidle.POLL.usage
      1758            +4.5%       1838        turbostat.Avg_MHz
    227522 ±  2%     -10.2%     204426 ±  3%  turbostat.C1
   4512700           -14.2%    3873264        turbostat.C1E
      1.39            -0.2        1.18        turbostat.C1E%
  24452583           -18.0%   20039031        turbostat.C6
     17.85            -3.3       14.55        turbostat.C6%
      7.44           -16.8%       6.19        turbostat.CPU%c1
     11.72           -19.3%       9.45        turbostat.CPU%c6
      7.51           -21.3%       5.91        turbostat.Pkg%pc2
    389.33            +1.6%     395.59        turbostat.PkgWatt
    559.33 ± 13%     -17.9%     459.33 ± 20%  slabinfo.dmaengine-unmap-128.active_objs
    559.33 ± 13%     -17.9%     459.33 ± 20%  slabinfo.dmaengine-unmap-128.num_objs
     57734 ±  3%      -5.7%      54421 ±  4%  slabinfo.filp.active_objs
    905.67 ±  3%      -5.6%     854.67 ±  4%  slabinfo.filp.active_slabs
     57981 ±  3%      -5.6%      54720 ±  4%  slabinfo.filp.num_objs
    905.67 ±  3%      -5.6%     854.67 ±  4%  slabinfo.filp.num_slabs
      1378           -12.0%       1212 ±  7%  slabinfo.nsproxy.active_objs
      1378           -12.0%       1212 ±  7%  slabinfo.nsproxy.num_objs
    507.33 ±  7%     -26.8%     371.33 ±  2%  slabinfo.secpath_cache.active_objs
    507.33 ±  7%     -26.8%     371.33 ±  2%  slabinfo.secpath_cache.num_objs
      4788 ±  5%      -8.3%       4391 ±  2%  slabinfo.sock_inode_cache.active_objs
      4788 ±  5%      -8.3%       4391 ±  2%  slabinfo.sock_inode_cache.num_objs
      1431 ±  8%     -12.3%       1255 ±  3%  slabinfo.task_group.active_objs
      1431 ±  8%     -12.3%       1255 ±  3%  slabinfo.task_group.num_objs
      4.27 ± 17%     +27.0%       5.42 ±  7%  sched_debug.cfs_rq:/.runnable_load_avg.avg
     13.44 ± 62%     +73.6%      23.33 ± 24%  sched_debug.cfs_rq:/.runnable_load_avg.stddev
    772.55 ± 21%     -32.7%     520.27 ±  4%  sched_debug.cfs_rq:/.util_est_enqueued.max
      4.39 ± 15%     +29.0%       5.66 ± 11%  sched_debug.cpu.cpu_load[0].avg
    152.09 ± 72%     +83.9%     279.67 ± 33%  sched_debug.cpu.cpu_load[0].max
     13.84 ± 58%     +78.7%      24.72 ± 29%  sched_debug.cpu.cpu_load[0].stddev
      4.53 ± 14%     +25.8%       5.70 ± 10%  sched_debug.cpu.cpu_load[1].avg
    156.58 ± 66%     +76.6%     276.58 ± 33%  sched_debug.cpu.cpu_load[1].max
     14.02 ± 55%     +72.4%      24.17 ± 28%  sched_debug.cpu.cpu_load[1].stddev
      4.87 ± 11%     +17.3%       5.72 ±  9%  sched_debug.cpu.cpu_load[2].avg
      1.58 ±  2%     +13.5%       1.79 ±  6%  sched_debug.cpu.nr_running.max
     16694           -14.6%      14259        sched_debug.cpu.nr_switches.min
     31989 ± 13%     +20.6%      38584 ±  6%  sched_debug.cpu.nr_switches.stddev
     16505           -14.8%      14068        sched_debug.cpu.sched_count.min
     32084 ± 13%     +19.9%      38482 ±  6%  sched_debug.cpu.sched_count.stddev
      8185           -15.0%       6957        sched_debug.cpu.sched_goidle.avg
     12151 ±  2%     -13.5%      10507        sched_debug.cpu.sched_goidle.max
      7867           -15.7%       6631        sched_debug.cpu.sched_goidle.min
      7595           -16.1%       6375        sched_debug.cpu.ttwu_count.min
     15873 ± 13%     +21.2%      19239 ±  6%  sched_debug.cpu.ttwu_count.stddev
      5244 ± 17%     +17.0%       6134 ±  5%  sched_debug.cpu.ttwu_local.avg
     15646 ± 12%     +21.5%      19008 ±  6%  sched_debug.cpu.ttwu_local.stddev
      0.85            -0.0        0.81        perf-stat.branch-miss-rate%
 3.689e+10            -4.6%  3.518e+10        perf-stat.branch-misses
     57.39            +0.6       58.00        perf-stat.cache-miss-rate%
 4.014e+11            -1.2%  3.967e+11        perf-stat.cache-misses
 6.994e+11            -2.2%   6.84e+11        perf-stat.cache-references
  14605393 ±  3%      -8.5%   13369913        perf-stat.context-switches
      9.21            +4.5%       9.63        perf-stat.cpi
 2.037e+14            +4.6%   2.13e+14        perf-stat.cpu-cycles
     44424            -2.0%      43541        perf-stat.cpu-migrations
      1.29            -0.1        1.24        perf-stat.dTLB-store-miss-rate%
 4.018e+10            -2.8%  3.905e+10        perf-stat.dTLB-store-misses
 3.071e+12            +1.4%  3.113e+12        perf-stat.dTLB-stores
     93.04            +1.5       94.51        perf-stat.iTLB-load-miss-rate%
 4.946e+09           +19.3%  5.903e+09 ±  5%  perf-stat.iTLB-load-misses
 3.702e+08            -7.5%  3.423e+08 ±  2%  perf-stat.iTLB-loads
      4470           -15.9%       3760 ±  5%  perf-stat.instructions-per-iTLB-miss
      0.11            -4.3%       0.10        perf-stat.ipc
 4.767e+09            -1.9%  4.675e+09        perf-stat.minor-faults
      1.46 ±  4%      -0.1        1.33 ±  9%  perf-stat.node-load-miss-rate%
      4.91            +1.7        6.65 ±  2%  perf-stat.node-store-miss-rate%
 1.195e+09           +32.8%  1.587e+09 ±  2%  perf-stat.node-store-misses
 2.313e+10            -3.7%  2.227e+10        perf-stat.node-stores
 4.767e+09            -1.9%  4.675e+09        perf-stat.page-faults
   1399047            +2.0%    1427115        perf-stat.path-length
      8908 ± 73%    -100.0%       0.00        latency_stats.avg.call_rwsem_down_read_failed.m_start.seq_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      3604 ±141%    -100.0%       0.00        latency_stats.avg.call_rwsem_down_write_failed.do_unlinkat.do_syscall_64.entry_SYSCALL_64_after_hwframe
     61499 ±130%     -92.6%       4534 ± 16%  latency_stats.avg.expand_files.__alloc_fd.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4391 ±138%     -70.9%       1277 ±129%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
     67311 ±112%     -48.5%      34681 ± 36%  latency_stats.avg.max
      3956 ±138%    +320.4%      16635 ±140%  latency_stats.avg.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    164.67 ± 30%   +7264.0%      12126 ±138%  latency_stats.avg.flush_work.fsnotify_destroy_group.inotify_release.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +5.4e+105%       5367 ±141%  latency_stats.avg.call_rwsem_down_write_failed.unlink_file_vma.free_pgtables.exit_mmap.mmput.flush_old_exec.load_elf_binary.search_binary_handler.do_execveat_common.__x64_sys_execve.do_syscall_64.entry_SYSCALL_64_after_hwframe
     36937 ±119%    -100.0%       0.00        latency_stats.max.call_rwsem_down_read_failed.m_start.seq_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      3604 ±141%    -100.0%       0.00        latency_stats.max.call_rwsem_down_write_failed.do_unlinkat.do_syscall_64.entry_SYSCALL_64_after_hwframe
     84146 ±107%     -72.5%      23171 ± 31%  latency_stats.max.expand_files.__alloc_fd.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4391 ±138%     -70.9%       1277 ±129%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
      5817 ± 83%     -69.7%       1760 ± 67%  latency_stats.max.pipe_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
      6720 ±137%   +1628.2%     116147 ±141%  latency_stats.max.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    164.67 ± 30%   +7264.0%      12126 ±138%  latency_stats.max.flush_work.fsnotify_destroy_group.inotify_release.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +1.2e+106%      12153 ±141%  latency_stats.max.call_rwsem_down_write_failed.unlink_file_vma.free_pgtables.exit_mmap.mmput.flush_old_exec.load_elf_binary.search_binary_handler.do_execveat_common.__x64_sys_execve.do_syscall_64.entry_SYSCALL_64_after_hwframe
    110122 ±120%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_read_failed.m_start.seq_read.__vfs_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
      3604 ±141%    -100.0%       0.00        latency_stats.sum.call_rwsem_down_write_failed.do_unlinkat.do_syscall_64.entry_SYSCALL_64_after_hwframe
  12078828 ±139%     -99.3%      89363 ± 29%  latency_stats.sum.expand_files.__alloc_fd.do_sys_open.do_syscall_64.entry_SYSCALL_64_after_hwframe
    144453 ±120%     -80.9%      27650 ± 19%  latency_stats.sum.poll_schedule_timeout.do_sys_poll.__x64_sys_poll.do_syscall_64.entry_SYSCALL_64_after_hwframe
      4391 ±138%     -70.9%       1277 ±129%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_lookup.nfs_lookup_revalidate.lookup_fast.walk_component.link_path_walk.path_lookupat.filename_lookup
      9438 ± 86%     -68.4%       2980 ± 35%  latency_stats.sum.pipe_write.__vfs_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
     31656 ±138%    +320.4%     133084 ±140%  latency_stats.sum.rpc_wait_bit_killable.__rpc_execute.rpc_run_task.rpc_call_sync.nfs3_rpc_wrapper.nfs3_proc_getattr.__nfs_revalidate_inode.nfs_do_access.nfs_permission.inode_permission.link_path_walk.path_lookupat
    164.67 ± 30%   +7264.0%      12126 ±138%  latency_stats.sum.flush_work.fsnotify_destroy_group.inotify_release.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +8.8e+105%       8760 ±141%  latency_stats.sum.msleep_interruptible.uart_wait_until_sent.tty_wait_until_sent.tty_port_close_start.tty_port_close.tty_release.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +1.3e+106%      12897 ±141%  latency_stats.sum.tty_wait_until_sent.tty_port_close_start.tty_port_close.tty_release.__fput.task_work_run.exit_to_usermode_loop.do_syscall_64.entry_SYSCALL_64_after_hwframe
      0.00       +3.2e+106%      32207 ±141%  latency_stats.sum.call_rwsem_down_write_failed.unlink_file_vma.free_pgtables.exit_mmap.mmput.flush_old_exec.load_elf_binary.search_binary_handler.do_execveat_common.__x64_sys_execve.do_syscall_64.entry_SYSCALL_64_after_hwframe
     44.43 ±  3%     -44.4        0.00        perf-profile.calltrace.cycles-pp.alloc_pages_vma.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
     44.13 ±  3%     -44.1        0.00        perf-profile.calltrace.cycles-pp.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault.handle_mm_fault.__do_page_fault
     43.95 ±  3%     -43.9        0.00        perf-profile.calltrace.cycles-pp.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault.handle_mm_fault
     41.85 ±  4%     -41.9        0.00        perf-profile.calltrace.cycles-pp._raw_spin_lock.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.__handle_mm_fault
      7.74 ±  8%      -7.7        0.00        perf-profile.calltrace.cycles-pp.copy_page.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      7.19 ±  4%      -7.2        0.00        perf-profile.calltrace.cycles-pp.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      7.15 ±  4%      -7.2        0.00        perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      5.09 ±  3%      -5.1        0.00        perf-profile.calltrace.cycles-pp.__lru_cache_add.alloc_set_pte.finish_fault.__handle_mm_fault.handle_mm_fault
      4.99 ±  3%      -5.0        0.00        perf-profile.calltrace.cycles-pp.pagevec_lru_move_fn.__lru_cache_add.alloc_set_pte.finish_fault.__handle_mm_fault
      0.93 ±  6%      -0.1        0.81 ±  2%  perf-profile.calltrace.cycles-pp.find_get_entry.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault
      0.00            +0.8        0.84        perf-profile.calltrace.cycles-pp._raw_spin_lock.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +0.9        0.92 ±  3%  perf-profile.calltrace.cycles-pp.__list_del_entry_valid.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault
      0.00            +1.1        1.08        perf-profile.calltrace.cycles-pp.find_lock_entry.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault
      0.00            +1.1        1.14        perf-profile.calltrace.cycles-pp.shmem_getpage_gfp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault
      0.00            +1.2        1.17        perf-profile.calltrace.cycles-pp.pte_map_lock.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +1.3        1.29        perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +1.3        1.31        perf-profile.calltrace.cycles-pp.__do_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
     61.62            +1.7       63.33        perf-profile.calltrace.cycles-pp.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault.page_fault
     41.73 ±  4%      +3.0       44.75        perf-profile.calltrace.cycles-pp.native_queued_spin_lock_slowpath._raw_spin_lock.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma
      0.00            +4.6        4.55 ± 15%  perf-profile.calltrace.cycles-pp.pagevec_lru_move_fn.__lru_cache_add.alloc_set_pte.finish_fault.handle_pte_fault
      0.00            +4.6        4.65 ± 14%  perf-profile.calltrace.cycles-pp.__lru_cache_add.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault
      0.00            +6.6        6.57 ± 10%  perf-profile.calltrace.cycles-pp.alloc_set_pte.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00            +6.6        6.61 ± 10%  perf-profile.calltrace.cycles-pp.finish_fault.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00            +7.2        7.25 ±  2%  perf-profile.calltrace.cycles-pp.copy_page.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
     41.41 ± 70%     +22.3       63.67        perf-profile.calltrace.cycles-pp.handle_mm_fault.__do_page_fault.do_page_fault.page_fault
     42.19 ± 70%     +22.6       64.75        perf-profile.calltrace.cycles-pp.__do_page_fault.do_page_fault.page_fault
     42.20 ± 70%     +22.6       64.76        perf-profile.calltrace.cycles-pp.do_page_fault.page_fault
     42.27 ± 70%     +22.6       64.86        perf-profile.calltrace.cycles-pp.page_fault
      0.00           +44.9       44.88        perf-profile.calltrace.cycles-pp._raw_spin_lock.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault
      0.00           +46.9       46.92        perf-profile.calltrace.cycles-pp.get_page_from_freelist.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault.__handle_mm_fault
      0.00           +47.1       47.10        perf-profile.calltrace.cycles-pp.__alloc_pages_nodemask.alloc_pages_vma.handle_pte_fault.__handle_mm_fault.handle_mm_fault
      0.00           +47.4       47.37        perf-profile.calltrace.cycles-pp.alloc_pages_vma.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault
      0.00           +63.0       63.00        perf-profile.calltrace.cycles-pp.handle_pte_fault.__handle_mm_fault.handle_mm_fault.__do_page_fault.do_page_fault
      0.97 ±  6%      -0.1        0.84 ±  2%  perf-profile.children.cycles-pp.find_get_entry
      1.23 ±  6%      -0.1        1.11        perf-profile.children.cycles-pp.find_lock_entry
      0.09 ± 10%      -0.0        0.07 ±  6%  perf-profile.children.cycles-pp.unlock_page
      0.19 ±  4%      +0.0        0.21 ±  2%  perf-profile.children.cycles-pp.mem_cgroup_charge_statistics
      0.21 ±  2%      +0.0        0.25        perf-profile.children.cycles-pp.__mod_node_page_state
      0.00            +0.1        0.05 ±  8%  perf-profile.children.cycles-pp.get_vma_policy
      0.00            +0.1        0.08        perf-profile.children.cycles-pp.__lru_cache_add_active_or_unevictable
      0.00            +0.2        0.18 ±  2%  perf-profile.children.cycles-pp.__page_add_new_anon_rmap
      0.00            +1.3        1.30        perf-profile.children.cycles-pp.pte_map_lock
     63.40            +1.6       64.97        perf-profile.children.cycles-pp.__do_page_fault
     63.19            +1.6       64.83        perf-profile.children.cycles-pp.do_page_fault
     61.69            +1.7       63.36        perf-profile.children.cycles-pp.__handle_mm_fault
     63.19            +1.7       64.86        perf-profile.children.cycles-pp.page_fault
     61.99            +1.7       63.70        perf-profile.children.cycles-pp.handle_mm_fault
     72.27            +2.2       74.52        perf-profile.children.cycles-pp.native_queued_spin_lock_slowpath
     67.51            +2.4       69.87        perf-profile.children.cycles-pp._raw_spin_lock
     44.49 ±  3%      +3.0       47.45        perf-profile.children.cycles-pp.alloc_pages_vma
     44.28 ±  3%      +3.0       47.26        perf-profile.children.cycles-pp.__alloc_pages_nodemask
     44.13 ±  3%      +3.0       47.12        perf-profile.children.cycles-pp.get_page_from_freelist
      0.00           +63.1       63.06        perf-profile.children.cycles-pp.handle_pte_fault
      1.46 ±  7%      -0.5        1.01        perf-profile.self.cycles-pp._raw_spin_lock
      0.58 ±  6%      -0.2        0.34        perf-profile.self.cycles-pp.__handle_mm_fault
      0.55 ±  6%      -0.1        0.44 ±  2%  perf-profile.self.cycles-pp.find_get_entry
      0.22 ±  5%      -0.1        0.16 ±  2%  perf-profile.self.cycles-pp.alloc_set_pte
      0.10 ±  8%      -0.0        0.08        perf-profile.self.cycles-pp.down_read_trylock
      0.09 ±  5%      -0.0        0.07        perf-profile.self.cycles-pp.unlock_page
      0.06            -0.0        0.05        perf-profile.self.cycles-pp.pmd_devmap_trans_unstable
      0.20 ±  2%      +0.0        0.24 ±  3%  perf-profile.self.cycles-pp.__mod_node_page_state
      0.00            +0.1        0.05        perf-profile.self.cycles-pp.finish_fault
      0.00            +0.1        0.05        perf-profile.self.cycles-pp.get_vma_policy
      0.00            +0.1        0.08 ±  6%  perf-profile.self.cycles-pp.__lru_cache_add_active_or_unevictable
      0.00            +0.2        0.25        perf-profile.self.cycles-pp.handle_pte_fault
      0.00            +0.5        0.46 ±  7%  perf-profile.self.cycles-pp.pte_map_lock
     72.26            +2.3       74.52        perf-profile.self.cycles-pp.native_queued_spin_lock_slowpath

[-- Attachment #3: perf-profile.zip --]
[-- Type: application/zip, Size: 19025 bytes --]

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

On 11/06/2018 09:49, Song, HaiyanX wrote:
> Hi Laurent,
> 
> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
> V9 patch serials.
> 
> The regression result is sorted by the metric will-it-scale.per_thread_ops.
> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
> commit id:
>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
> Benchmark: will-it-scale
> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
> 
> Metrics:
>   will-it-scale.per_process_ops=processes/nr_cpu
>   will-it-scale.per_thread_ops=threads/nr_cpu
>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
> THP: enable / disable
> nr_task:100%
> 
> 1. Regressions:
> 
> a). Enable THP
> testcase                          base           change      head           metric
> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
> 
> b). Disable THP
> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
> 
> Notes: for the above  values of test result, the higher is better.

I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
get reproducible results. The results have huge variation, even on the vanilla
kernel, and I can't state on any changes due to that.

I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
measure any changes between the vanilla and the SPF patched ones:

test THP enabled		4.17.0-rc4-mm1	spf		delta
page_fault3_threads		2697.7		2683.5		-0.53%
page_fault2_threads		170660.6	169574.1	-0.64%
context_switch1_threads		6915269.2	6877507.3	-0.55%
context_switch1_processes	6478076.2	6529493.5	0.79%
brk1				243391.2	238527.5	-2.00%

Tests were run 10 times, no high variation detected.

Did you see high variation on your side ? How many times the test were run to
compute the average values ?

Thanks,
Laurent.


> 
> 2. Improvement: not found improvement based on the selected test cases.
> 
> 
> Best regards
> Haiyan Song
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Monday, May 28, 2018 4:54 PM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
> 
> On 28/05/2018 10:22, Haiyan Song wrote:
>> Hi Laurent,
>>
>> Yes, these tests are done on V9 patch.
> 
> Do you plan to give this V11 a run ?
> 
>>
>>
>> Best regards,
>> Haiyan Song
>>
>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>
>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>> tested on Intel 4s Skylake platform.
>>>
>>> Hi,
>>>
>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>> series" while responding to the v11 header series...
>>> Were these tests done on v9 or v11 ?
>>>
>>> Cheers,
>>> Laurent.
>>>
>>>>
>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>> Commit id:
>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>> Benchmark suite: will-it-scale
>>>> Download link:
>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>> Metrics:
>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>> THP: enable / disable
>>>> nr_task: 100%
>>>>
>>>> 1. Regressions:
>>>> a) THP enabled:
>>>> testcase                        base            change          head       metric
>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>
>>>> b) THP disabled:
>>>> testcase                        base            change          head       metric
>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>
>>>> 2. Improvements:
>>>> a) THP enabled:
>>>> testcase                        base            change          head       metric
>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>
>>>> b) THP disabled:
>>>> testcase                        base            change          head       metric
>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>
>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>> on head commit is better than that on base commit for this benchmark.
>>>>
>>>>
>>>> Best regards
>>>> Haiyan Song
>>>>
>>>> ________________________________________
>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>
>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>> page fault without holding the mm semaphore [1].
>>>>
>>>> The idea is to try to handle user space page faults without holding the
>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>> process since the page fault handler will not wait for other threads memory
>>>> layout change to be done, assuming that this change is done in another part
>>>> of the process's memory space. This type page fault is named speculative
>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>> is failing its processing and a classic page fault is then tried.
>>>>
>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>> freeing operation which was hitting the performance by 20% as reported by
>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>> limiting the locking contention to these operations which are expected to
>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>> our back a reference count is added and 2 services (get_vma() and
>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>> benchmark anymore.
>>>>
>>>> The VMA's attributes checked during the speculative page fault processing
>>>> have to be protected against parallel changes. This is done by using a per
>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>> handler to fast check for parallel changes in progress and to abort the
>>>> speculative page fault in that case.
>>>>
>>>> Once the VMA has been found, the speculative page fault handler would check
>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>> checked during the page fault are modified.
>>>>
>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>> leading to touching this PTE will need to lock the page table, so no
>>>> parallel change is possible at this time.
>>>>
>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>> different from the one recorded at the beginning of the SPF operation, the
>>>> classic page fault handler will be called to handle the operation while
>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>> PTE.
>>>>
>>>> In pseudo code, this could be seen as:
>>>>     speculative_page_fault()
>>>>     {
>>>>             vma = get_vma()
>>>>             check vma sequence count
>>>>             check vma's support
>>>>             disable interrupt
>>>>                   check pgd,p4d,...,pte
>>>>                   save pmd and pte in vmf
>>>>                   save vma sequence counter in vmf
>>>>             enable interrupt
>>>>             check vma sequence count
>>>>             handle_pte_fault(vma)
>>>>                     ..
>>>>                     page = alloc_page()
>>>>                     pte_map_lock()
>>>>                             disable interrupt
>>>>                                     abort if sequence counter has changed
>>>>                                     abort if pmd or pte has changed
>>>>                                     pte map and lock
>>>>                             enable interrupt
>>>>                     if abort
>>>>                        free page
>>>>                        abort
>>>>                     ...
>>>>     }
>>>>
>>>>     arch_fault_handler()
>>>>     {
>>>>             if (speculative_page_fault(&vma))
>>>>                goto done
>>>>     again:
>>>>             lock(mmap_sem)
>>>>             vma = find_vma();
>>>>             handle_pte_fault(vma);
>>>>             if retry
>>>>                unlock(mmap_sem)
>>>>                goto again;
>>>>     done:
>>>>             handle fault error
>>>>     }
>>>>
>>>> Support for THP is not done because when checking for the PMD, we can be
>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>
>>>> This series add a new software performance event named 'speculative-faults'
>>>> or 'spf'. It counts the number of successful page fault event handled
>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>> counting the total number of page fault events while 'spf' is only counting
>>>> the part of the faults processed speculatively.
>>>>
>>>> There are some trace events introduced by this series. They allow
>>>> identifying why the page faults were not processed speculatively. This
>>>> doesn't take in account the faults generated by a monothreaded process
>>>> which directly processed while holding the mmap_sem. This trace events are
>>>> grouped in a system named 'pagefault', they are:
>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>    back.
>>>>
>>>> To record all the related events, the easier is to run perf with the
>>>> following arguments :
>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>
>>>> There is also a dedicated vmstat counter showing the number of successful
>>>> page fault handled speculatively. I can be seen this way:
>>>> $ grep speculative_pgfault /proc/vmstat
>>>>
>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>> on x86, PowerPC and arm64.
>>>>
>>>> ---------------------
>>>> Real Workload results
>>>>
>>>> As mentioned in previous email, we did non official runs using a "popular
>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>> which showed a 30% improvements in the number of transaction processed per
>>>> second. This run has been done on the v6 series, but changes introduced in
>>>> this new version should not impact the performance boost seen.
>>>>
>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>> series:
>>>>                 vanilla         spf
>>>> faults          89.418          101.364         +13%
>>>> spf                n/a           97.989
>>>>
>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>> way.
>>>>
>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>> it a try on an android device. He reported that the application launch time
>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>> 20%.
>>>>
>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>
>>>> Application                             4.9     4.9+spf delta
>>>> com.tencent.mm                          416     389     -7%
>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>> com.tencent.mtt                         455     454     0%
>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>> com.autonavi.minimap                    711     701     -1%
>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>> com.immomo.momo                         501     487     -3%
>>>> com.tencent.peng                        2145    2112    -2%
>>>> com.smile.gifmaker                      491     461     -6%
>>>> com.baidu.BaiduMap                      479     366     -23%
>>>> com.taobao.taobao                       1341    1198    -11%
>>>> com.baidu.searchbox                     333     314     -6%
>>>> com.tencent.mobileqq                    394     384     -3%
>>>> com.sina.weibo                          907     906     0%
>>>> com.youku.phone                         816     731     -11%
>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>> com.UCMobile                            415     411     -1%
>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>> com.tencent.qqmusic                     336     329     -2%
>>>> com.sankuai.meituan                     1661    1302    -22%
>>>> com.netease.cloudmusic                  1193    1200    1%
>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>
>>>> ------------------
>>>> Benchmarks results
>>>>
>>>> Base kernel is v4.17.0-rc4-mm1
>>>> SPF is BASE + this series
>>>>
>>>> Kernbench:
>>>> ----------
>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>> kernel (kernel is build 5 times):
>>>>
>>>> Average Half load -j 8
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>
>>>> Average Optimal load -j 16
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>
>>>>
>>>> During a run on the SPF, perf events were captured:
>>>>  Performance counter stats for '../kernbench -M':
>>>>          526743764      faults
>>>>                210      spf
>>>>                  3      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>               2278      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> Very few speculative page faults were recorded as most of the processes
>>>> involved are monothreaded (sounds that on this architecture some threads
>>>> were created during the kernel build processing).
>>>>
>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>
>>>> Average Half load -j 40
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>
>>>> Average Optimal load -j 80
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>
>>>> During a run on the SPF, perf events were captured:
>>>>  Performance counter stats for '../kernbench -M':
>>>>          116730856      faults
>>>>                  0      spf
>>>>                  3      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>                476      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>> there is no impact on the performance.
>>>>
>>>> Ebizzy:
>>>> -------
>>>> The test is counting the number of records per second it can manage, the
>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>> consistent result I repeated the test 100 times and measure the average
>>>> result. The number is the record processes per second, the higher is the
>>>> best.
>>>>
>>>>                 BASE            SPF             delta
>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>
>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>            1706379      faults
>>>>            1674599      spf
>>>>              30588      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>                363      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>            1874773      faults
>>>>            1461153      spf
>>>>             413293      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>                200      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>> leading the ebizzy performance boost.
>>>>
>>>> ------------------
>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>    and Minchan Kim, hopefully.
>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>    __do_page_fault().
>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>    instead
>>>>    of aborting the speculative page fault handling. Dropping the now
>>>> useless
>>>>    trace event pagefault:spf_pte_lock.
>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>    additional tests done didn't show a significant performance improvement.
>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>
>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>
>>>>
>>>> Laurent Dufour (20):
>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>   mm: make pte_unmap_same compatible with SPF
>>>>   mm: introduce INIT_VMA()
>>>>   mm: protect VMA modifications using VMA sequence count
>>>>   mm: protect mremap() against SPF hanlder
>>>>   mm: protect SPF handler against anon_vma changes
>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>   mm: introduce __vm_normal_page()
>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>   mm: protect mm_rb tree with a rwlock
>>>>   mm: adding speculative page fault failure trace events
>>>>   perf: add a speculative page fault sw event
>>>>   perf tools: add support for the SPF perf event
>>>>   mm: add speculative page fault vmstats
>>>>   powerpc/mm: add speculative page fault
>>>>
>>>> Mahendran Ganesh (2):
>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>   arm64/mm: add speculative page fault
>>>>
>>>> Peter Zijlstra (4):
>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>   mm: VMA sequence count
>>>>   mm: provide speculative fault infrastructure
>>>>   x86/mm: add speculative pagefault handling
>>>>
>>>>  arch/arm64/Kconfig                    |   1 +
>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>  arch/x86/Kconfig                      |   1 +
>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>  fs/exec.c                             |   2 +-
>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>  fs/userfaultfd.c                      |  17 +-
>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>  include/linux/migrate.h               |   4 +-
>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>  include/linux/mm_types.h              |   7 +
>>>>  include/linux/pagemap.h               |   4 +-
>>>>  include/linux/rmap.h                  |  12 +-
>>>>  include/linux/swap.h                  |  10 +-
>>>>  include/linux/vm_event_item.h         |   3 +
>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>  kernel/fork.c                         |   5 +-
>>>>  mm/Kconfig                            |  22 ++
>>>>  mm/huge_memory.c                      |   6 +-
>>>>  mm/hugetlb.c                          |   2 +
>>>>  mm/init-mm.c                          |   3 +
>>>>  mm/internal.h                         |  20 ++
>>>>  mm/khugepaged.c                       |   5 +
>>>>  mm/madvise.c                          |   6 +-
>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>  mm/mempolicy.c                        |  51 ++-
>>>>  mm/migrate.c                          |   6 +-
>>>>  mm/mlock.c                            |  13 +-
>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>  mm/mprotect.c                         |   4 +-
>>>>  mm/mremap.c                           |  13 +
>>>>  mm/nommu.c                            |   2 +-
>>>>  mm/rmap.c                             |   5 +-
>>>>  mm/swap.c                             |   6 +-
>>>>  mm/swap_state.c                       |   8 +-
>>>>  mm/vmstat.c                           |   5 +-
>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>  tools/perf/util/evsel.c               |   1 +
>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>  tools/perf/util/python.c              |   1 +
>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>
>>>> --
>>>> 2.7.4
>>>>
>>>>
>>>
>>
> 

RE: [PATCH v11 00/26] Speculative page faults

[Song, HaiyanX]

Hi Laurent,


For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.

And I did not find other high variation on test case result.

a). Enable THP
testcase                          base     stddev       change      head     stddev         metric
page_fault3/enable THP           10519      ± 3%        -20.5%      8368      ±6%          will-it-scale.per_thread_ops
page_fault2/enalbe THP            8281      ± 2%        -18.8%      6728                   will-it-scale.per_thread_ops
brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops

b). Disable THP
page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops


Best regards,
Haiyan Song
________________________________________
From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
Sent: Monday, July 02, 2018 4:59 PM
To: Song, HaiyanX
Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
Subject: Re: [PATCH v11 00/26] Speculative page faults

On 11/06/2018 09:49, Song, HaiyanX wrote:
> Hi Laurent,
>
> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
> V9 patch serials.
>
> The regression result is sorted by the metric will-it-scale.per_thread_ops.
> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
> commit id:
>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
> Benchmark: will-it-scale
> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>
> Metrics:
>   will-it-scale.per_process_ops=processes/nr_cpu
>   will-it-scale.per_thread_ops=threads/nr_cpu
>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
> THP: enable / disable
> nr_task:100%
>
> 1. Regressions:
>
> a). Enable THP
> testcase                          base           change      head           metric
> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>
> b). Disable THP
> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>
> Notes: for the above  values of test result, the higher is better.

I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
get reproducible results. The results have huge variation, even on the vanilla
kernel, and I can't state on any changes due to that.

I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
measure any changes between the vanilla and the SPF patched ones:

test THP enabled                4.17.0-rc4-mm1  spf             delta
page_fault3_threads             2697.7          2683.5          -0.53%
page_fault2_threads             170660.6        169574.1        -0.64%
context_switch1_threads         6915269.2       6877507.3       -0.55%
context_switch1_processes       6478076.2       6529493.5       0.79%
brk1                            243391.2        238527.5        -2.00%

Tests were run 10 times, no high variation detected.

Did you see high variation on your side ? How many times the test were run to
compute the average values ?

Thanks,
Laurent.


>
> 2. Improvement: not found improvement based on the selected test cases.
>
>
> Best regards
> Haiyan Song
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Monday, May 28, 2018 4:54 PM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
>
> On 28/05/2018 10:22, Haiyan Song wrote:
>> Hi Laurent,
>>
>> Yes, these tests are done on V9 patch.
>
> Do you plan to give this V11 a run ?
>
>>
>>
>> Best regards,
>> Haiyan Song
>>
>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>
>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>> tested on Intel 4s Skylake platform.
>>>
>>> Hi,
>>>
>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>> series" while responding to the v11 header series...
>>> Were these tests done on v9 or v11 ?
>>>
>>> Cheers,
>>> Laurent.
>>>
>>>>
>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>> Commit id:
>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>> Benchmark suite: will-it-scale
>>>> Download link:
>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>> Metrics:
>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>> THP: enable / disable
>>>> nr_task: 100%
>>>>
>>>> 1. Regressions:
>>>> a) THP enabled:
>>>> testcase                        base            change          head       metric
>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>
>>>> b) THP disabled:
>>>> testcase                        base            change          head       metric
>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>
>>>> 2. Improvements:
>>>> a) THP enabled:
>>>> testcase                        base            change          head       metric
>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>
>>>> b) THP disabled:
>>>> testcase                        base            change          head       metric
>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>
>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>> on head commit is better than that on base commit for this benchmark.
>>>>
>>>>
>>>> Best regards
>>>> Haiyan Song
>>>>
>>>> ________________________________________
>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>
>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>> page fault without holding the mm semaphore [1].
>>>>
>>>> The idea is to try to handle user space page faults without holding the
>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>> process since the page fault handler will not wait for other threads memory
>>>> layout change to be done, assuming that this change is done in another part
>>>> of the process's memory space. This type page fault is named speculative
>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>> is failing its processing and a classic page fault is then tried.
>>>>
>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>> freeing operation which was hitting the performance by 20% as reported by
>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>> limiting the locking contention to these operations which are expected to
>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>> our back a reference count is added and 2 services (get_vma() and
>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>> benchmark anymore.
>>>>
>>>> The VMA's attributes checked during the speculative page fault processing
>>>> have to be protected against parallel changes. This is done by using a per
>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>> handler to fast check for parallel changes in progress and to abort the
>>>> speculative page fault in that case.
>>>>
>>>> Once the VMA has been found, the speculative page fault handler would check
>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>> checked during the page fault are modified.
>>>>
>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>> leading to touching this PTE will need to lock the page table, so no
>>>> parallel change is possible at this time.
>>>>
>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>> different from the one recorded at the beginning of the SPF operation, the
>>>> classic page fault handler will be called to handle the operation while
>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>> PTE.
>>>>
>>>> In pseudo code, this could be seen as:
>>>>     speculative_page_fault()
>>>>     {
>>>>             vma = get_vma()
>>>>             check vma sequence count
>>>>             check vma's support
>>>>             disable interrupt
>>>>                   check pgd,p4d,...,pte
>>>>                   save pmd and pte in vmf
>>>>                   save vma sequence counter in vmf
>>>>             enable interrupt
>>>>             check vma sequence count
>>>>             handle_pte_fault(vma)
>>>>                     ..
>>>>                     page = alloc_page()
>>>>                     pte_map_lock()
>>>>                             disable interrupt
>>>>                                     abort if sequence counter has changed
>>>>                                     abort if pmd or pte has changed
>>>>                                     pte map and lock
>>>>                             enable interrupt
>>>>                     if abort
>>>>                        free page
>>>>                        abort
>>>>                     ...
>>>>     }
>>>>
>>>>     arch_fault_handler()
>>>>     {
>>>>             if (speculative_page_fault(&vma))
>>>>                goto done
>>>>     again:
>>>>             lock(mmap_sem)
>>>>             vma = find_vma();
>>>>             handle_pte_fault(vma);
>>>>             if retry
>>>>                unlock(mmap_sem)
>>>>                goto again;
>>>>     done:
>>>>             handle fault error
>>>>     }
>>>>
>>>> Support for THP is not done because when checking for the PMD, we can be
>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>
>>>> This series add a new software performance event named 'speculative-faults'
>>>> or 'spf'. It counts the number of successful page fault event handled
>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>> counting the total number of page fault events while 'spf' is only counting
>>>> the part of the faults processed speculatively.
>>>>
>>>> There are some trace events introduced by this series. They allow
>>>> identifying why the page faults were not processed speculatively. This
>>>> doesn't take in account the faults generated by a monothreaded process
>>>> which directly processed while holding the mmap_sem. This trace events are
>>>> grouped in a system named 'pagefault', they are:
>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>    back.
>>>>
>>>> To record all the related events, the easier is to run perf with the
>>>> following arguments :
>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>
>>>> There is also a dedicated vmstat counter showing the number of successful
>>>> page fault handled speculatively. I can be seen this way:
>>>> $ grep speculative_pgfault /proc/vmstat
>>>>
>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>> on x86, PowerPC and arm64.
>>>>
>>>> ---------------------
>>>> Real Workload results
>>>>
>>>> As mentioned in previous email, we did non official runs using a "popular
>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>> which showed a 30% improvements in the number of transaction processed per
>>>> second. This run has been done on the v6 series, but changes introduced in
>>>> this new version should not impact the performance boost seen.
>>>>
>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>> series:
>>>>                 vanilla         spf
>>>> faults          89.418          101.364         +13%
>>>> spf                n/a           97.989
>>>>
>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>> way.
>>>>
>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>> it a try on an android device. He reported that the application launch time
>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>> 20%.
>>>>
>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>
>>>> Application                             4.9     4.9+spf delta
>>>> com.tencent.mm                          416     389     -7%
>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>> com.tencent.mtt                         455     454     0%
>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>> com.autonavi.minimap                    711     701     -1%
>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>> com.immomo.momo                         501     487     -3%
>>>> com.tencent.peng                        2145    2112    -2%
>>>> com.smile.gifmaker                      491     461     -6%
>>>> com.baidu.BaiduMap                      479     366     -23%
>>>> com.taobao.taobao                       1341    1198    -11%
>>>> com.baidu.searchbox                     333     314     -6%
>>>> com.tencent.mobileqq                    394     384     -3%
>>>> com.sina.weibo                          907     906     0%
>>>> com.youku.phone                         816     731     -11%
>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>> com.UCMobile                            415     411     -1%
>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>> com.tencent.qqmusic                     336     329     -2%
>>>> com.sankuai.meituan                     1661    1302    -22%
>>>> com.netease.cloudmusic                  1193    1200    1%
>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>
>>>> ------------------
>>>> Benchmarks results
>>>>
>>>> Base kernel is v4.17.0-rc4-mm1
>>>> SPF is BASE + this series
>>>>
>>>> Kernbench:
>>>> ----------
>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>> kernel (kernel is build 5 times):
>>>>
>>>> Average Half load -j 8
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>
>>>> Average Optimal load -j 16
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>
>>>>
>>>> During a run on the SPF, perf events were captured:
>>>>  Performance counter stats for '../kernbench -M':
>>>>          526743764      faults
>>>>                210      spf
>>>>                  3      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>               2278      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> Very few speculative page faults were recorded as most of the processes
>>>> involved are monothreaded (sounds that on this architecture some threads
>>>> were created during the kernel build processing).
>>>>
>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>
>>>> Average Half load -j 40
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>
>>>> Average Optimal load -j 80
>>>>                  Run    (std deviation)
>>>>                  BASE                   SPF
>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>
>>>> During a run on the SPF, perf events were captured:
>>>>  Performance counter stats for '../kernbench -M':
>>>>          116730856      faults
>>>>                  0      spf
>>>>                  3      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>                476      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>> there is no impact on the performance.
>>>>
>>>> Ebizzy:
>>>> -------
>>>> The test is counting the number of records per second it can manage, the
>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>> consistent result I repeated the test 100 times and measure the average
>>>> result. The number is the record processes per second, the higher is the
>>>> best.
>>>>
>>>>                 BASE            SPF             delta
>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>
>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>            1706379      faults
>>>>            1674599      spf
>>>>              30588      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>                363      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>            1874773      faults
>>>>            1461153      spf
>>>>             413293      pagefault:spf_vma_changed
>>>>                  0      pagefault:spf_vma_noanon
>>>>                200      pagefault:spf_vma_notsup
>>>>                  0      pagefault:spf_vma_access
>>>>                  0      pagefault:spf_pmd_changed
>>>>
>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>> leading the ebizzy performance boost.
>>>>
>>>> ------------------
>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>    and Minchan Kim, hopefully.
>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>    __do_page_fault().
>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>    instead
>>>>    of aborting the speculative page fault handling. Dropping the now
>>>> useless
>>>>    trace event pagefault:spf_pte_lock.
>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>    additional tests done didn't show a significant performance improvement.
>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>
>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>
>>>>
>>>> Laurent Dufour (20):
>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>   mm: make pte_unmap_same compatible with SPF
>>>>   mm: introduce INIT_VMA()
>>>>   mm: protect VMA modifications using VMA sequence count
>>>>   mm: protect mremap() against SPF hanlder
>>>>   mm: protect SPF handler against anon_vma changes
>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>   mm: introduce __vm_normal_page()
>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>   mm: protect mm_rb tree with a rwlock
>>>>   mm: adding speculative page fault failure trace events
>>>>   perf: add a speculative page fault sw event
>>>>   perf tools: add support for the SPF perf event
>>>>   mm: add speculative page fault vmstats
>>>>   powerpc/mm: add speculative page fault
>>>>
>>>> Mahendran Ganesh (2):
>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>   arm64/mm: add speculative page fault
>>>>
>>>> Peter Zijlstra (4):
>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>   mm: VMA sequence count
>>>>   mm: provide speculative fault infrastructure
>>>>   x86/mm: add speculative pagefault handling
>>>>
>>>>  arch/arm64/Kconfig                    |   1 +
>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>  arch/x86/Kconfig                      |   1 +
>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>  fs/exec.c                             |   2 +-
>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>  fs/userfaultfd.c                      |  17 +-
>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>  include/linux/migrate.h               |   4 +-
>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>  include/linux/mm_types.h              |   7 +
>>>>  include/linux/pagemap.h               |   4 +-
>>>>  include/linux/rmap.h                  |  12 +-
>>>>  include/linux/swap.h                  |  10 +-
>>>>  include/linux/vm_event_item.h         |   3 +
>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>  kernel/fork.c                         |   5 +-
>>>>  mm/Kconfig                            |  22 ++
>>>>  mm/huge_memory.c                      |   6 +-
>>>>  mm/hugetlb.c                          |   2 +
>>>>  mm/init-mm.c                          |   3 +
>>>>  mm/internal.h                         |  20 ++
>>>>  mm/khugepaged.c                       |   5 +
>>>>  mm/madvise.c                          |   6 +-
>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>  mm/mempolicy.c                        |  51 ++-
>>>>  mm/migrate.c                          |   6 +-
>>>>  mm/mlock.c                            |  13 +-
>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>  mm/mprotect.c                         |   4 +-
>>>>  mm/mremap.c                           |  13 +
>>>>  mm/nommu.c                            |   2 +-
>>>>  mm/rmap.c                             |   5 +-
>>>>  mm/swap.c                             |   6 +-
>>>>  mm/swap_state.c                       |   8 +-
>>>>  mm/vmstat.c                           |   5 +-
>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>  tools/perf/util/evsel.c               |   1 +
>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>  tools/perf/util/python.c              |   1 +
>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>
>>>> --
>>>> 2.7.4
>>>>
>>>>
>>>
>>
>

RE: [PATCH v11 00/26] Speculative page faults

[Song, HaiyanX]

Hi Laurent,=0A=
=0A=
=0A=
For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), t=
he below test cases all were run 3 times.=0A=
I check the test results, only page_fault3_thread/enable THP have 6% stddev=
 for head commit, other tests have lower stddev.=0A=
=0A=
And I did not find other high variation on test case result.=0A=
=0A=
a). Enable THP=0A=
testcase                          base     stddev       change      head   =
  stddev         metric=0A=
page_fault3/enable THP           10519      =B1 3%        -20.5%      8368 =
     =B16%          will-it-scale.per_thread_ops=0A=
page_fault2/enalbe THP            8281      =B1 2%        -18.8%      6728 =
                  will-it-scale.per_thread_ops=0A=
brk1/eanble THP                 998475                   -2.2%    976893   =
                will-it-scale.per_process_ops=0A=
context_switch1/enable THP      223910                   -1.3%    220930   =
                will-it-scale.per_process_ops=0A=
context_switch1/enable THP      233722                   -1.0%    231288   =
                will-it-scale.per_thread_ops=0A=
=0A=
b). Disable THP=0A=
page_fault3/disable THP          10856                  -23.1%      8344   =
                will-it-scale.per_thread_ops=0A=
page_fault2/disable THP           8147                  -18.8%      6613   =
                will-it-scale.per_thread_ops=0A=
brk1/disable THP                   957                    -7.9%      881   =
                will-it-scale.per_thread_ops=0A=
context_switch1/disable THP     237006                    -2.2%    231907  =
                will-it-scale.per_thread_ops=0A=
brk1/disable THP                997317                    -2.0%    977778  =
                will-it-scale.per_process_ops=0A=
page_fault3/disable THP         467454                    -1.8%    459251  =
                will-it-scale.per_process_ops=0A=
context_switch1/disable THP     224431                    -1.3%    221567  =
                will-it-scale.per_process_ops=0A=
=0A=
=0A=
Best regards,=0A=
Haiyan Song=0A=
________________________________________=0A=
From: Laurent Dufour [ldufour@linux.vnet.ibm.com]=0A=
Sent: Monday, July 02, 2018 4:59 PM=0A=
To: Song, HaiyanX=0A=
Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kir=
ill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Mat=
thew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; =
benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Glei=
xner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.s=
enozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi=
; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan K=
im; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; l=
inux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora=
@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs=
.org; x86@kernel.org=0A=
Subject: Re: [PATCH v11 00/26] Speculative page faults=0A=
=0A=
On 11/06/2018 09:49, Song, HaiyanX wrote:=0A=
> Hi Laurent,=0A=
>=0A=
> Regression test for v11 patch serials have been run, some regression is f=
ound by LKP-tools (linux kernel performance)=0A=
> tested on Intel 4s skylake platform. This time only test the cases which =
have been run and found regressions on=0A=
> V9 patch serials.=0A=
>=0A=
> The regression result is sorted by the metric will-it-scale.per_thread_op=
s.=0A=
> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126=0A=
> commit id:=0A=
>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12=0A=
>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1=0A=
> Benchmark: will-it-scale=0A=
> Download link: https://github.com/antonblanchard/will-it-scale/tree/maste=
r=0A=
>=0A=
> Metrics:=0A=
>   will-it-scale.per_process_ops=3Dprocesses/nr_cpu=0A=
>   will-it-scale.per_thread_ops=3Dthreads/nr_cpu=0A=
>   test box: lkp-skl-4sp1(nr_cpu=3D192,memory=3D768G)=0A=
> THP: enable / disable=0A=
> nr_task:100%=0A=
>=0A=
> 1. Regressions:=0A=
>=0A=
> a). Enable THP=0A=
> testcase                          base           change      head        =
   metric=0A=
> page_fault3/enable THP           10519          -20.5%        836      wi=
ll-it-scale.per_thread_ops=0A=
> page_fault2/enalbe THP            8281          -18.8%       6728      wi=
ll-it-scale.per_thread_ops=0A=
> brk1/eanble THP                 998475           -2.2%     976893      wi=
ll-it-scale.per_process_ops=0A=
> context_switch1/enable THP      223910           -1.3%     220930      wi=
ll-it-scale.per_process_ops=0A=
> context_switch1/enable THP      233722           -1.0%     231288      wi=
ll-it-scale.per_thread_ops=0A=
>=0A=
> b). Disable THP=0A=
> page_fault3/disable THP          10856          -23.1%       8344      wi=
ll-it-scale.per_thread_ops=0A=
> page_fault2/disable THP           8147          -18.8%       6613      wi=
ll-it-scale.per_thread_ops=0A=
> brk1/disable THP                   957           -7.9%        881      wi=
ll-it-scale.per_thread_ops=0A=
> context_switch1/disable THP     237006           -2.2%     231907      wi=
ll-it-scale.per_thread_ops=0A=
> brk1/disable THP                997317           -2.0%     977778      wi=
ll-it-scale.per_process_ops=0A=
> page_fault3/disable THP         467454           -1.8%     459251      wi=
ll-it-scale.per_process_ops=0A=
> context_switch1/disable THP     224431           -1.3%     221567      wi=
ll-it-scale.per_process_ops=0A=
>=0A=
> Notes: for the above  values of test result, the higher is better.=0A=
=0A=
I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can'=
t=0A=
get reproducible results. The results have huge variation, even on the vani=
lla=0A=
kernel, and I can't state on any changes due to that.=0A=
=0A=
I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I did=
n't=0A=
measure any changes between the vanilla and the SPF patched ones:=0A=
=0A=
test THP enabled                4.17.0-rc4-mm1  spf             delta=0A=
page_fault3_threads             2697.7          2683.5          -0.53%=0A=
page_fault2_threads             170660.6        169574.1        -0.64%=0A=
context_switch1_threads         6915269.2       6877507.3       -0.55%=0A=
context_switch1_processes       6478076.2       6529493.5       0.79%=0A=
brk1                            243391.2        238527.5        -2.00%=0A=
=0A=
Tests were run 10 times, no high variation detected.=0A=
=0A=
Did you see high variation on your side ? How many times the test were run =
to=0A=
compute the average values ?=0A=
=0A=
Thanks,=0A=
Laurent.=0A=
=0A=
=0A=
>=0A=
> 2. Improvement: not found improvement based on the selected test cases.=
=0A=
>=0A=
>=0A=
> Best regards=0A=
> Haiyan Song=0A=
> ________________________________________=0A=
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of La=
urent Dufour [ldufour@linux.vnet.ibm.com]=0A=
> Sent: Monday, May 28, 2018 4:54 PM=0A=
> To: Song, HaiyanX=0A=
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; k=
irill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; M=
atthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com=
; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gl=
eixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey=
.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Ke=
mi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan=
 Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org;=
 linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharo=
ra@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozla=
bs.org; x86@kernel.org=0A=
> Subject: Re: [PATCH v11 00/26] Speculative page faults=0A=
>=0A=
> On 28/05/2018 10:22, Haiyan Song wrote:=0A=
>> Hi Laurent,=0A=
>>=0A=
>> Yes, these tests are done on V9 patch.=0A=
>=0A=
> Do you plan to give this V11 a run ?=0A=
>=0A=
>>=0A=
>>=0A=
>> Best regards,=0A=
>> Haiyan Song=0A=
>>=0A=
>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:=0A=
>>> On 28/05/2018 07:23, Song, HaiyanX wrote:=0A=
>>>>=0A=
>>>> Some regression and improvements is found by LKP-tools(linux kernel pe=
rformance) on V9 patch series=0A=
>>>> tested on Intel 4s Skylake platform.=0A=
>>>=0A=
>>> Hi,=0A=
>>>=0A=
>>> Thanks for reporting this benchmark results, but you mentioned the "V9 =
patch=0A=
>>> series" while responding to the v11 header series...=0A=
>>> Were these tests done on v9 or v11 ?=0A=
>>>=0A=
>>> Cheers,=0A=
>>> Laurent.=0A=
>>>=0A=
>>>>=0A=
>>>> The regression result is sorted by the metric will-it-scale.per_thread=
_ops.=0A=
>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 pat=
ch series)=0A=
>>>> Commit id:=0A=
>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f=0A=
>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110=0A=
>>>> Benchmark suite: will-it-scale=0A=
>>>> Download link:=0A=
>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests=0A=
>>>> Metrics:=0A=
>>>>     will-it-scale.per_process_ops=3Dprocesses/nr_cpu=0A=
>>>>     will-it-scale.per_thread_ops=3Dthreads/nr_cpu=0A=
>>>> test box: lkp-skl-4sp1(nr_cpu=3D192,memory=3D768G)=0A=
>>>> THP: enable / disable=0A=
>>>> nr_task: 100%=0A=
>>>>=0A=
>>>> 1. Regressions:=0A=
>>>> a) THP enabled:=0A=
>>>> testcase                        base            change          head  =
     metric=0A=
>>>> page_fault3/ enable THP         10092           -17.5%          8323  =
     will-it-scale.per_thread_ops=0A=
>>>> page_fault2/ enable THP          8300           -17.2%          6869  =
     will-it-scale.per_thread_ops=0A=
>>>> brk1/ enable THP                  957.67         -7.6%           885  =
     will-it-scale.per_thread_ops=0A=
>>>> page_fault3/ enable THP        172821            -5.3%        163692  =
     will-it-scale.per_process_ops=0A=
>>>> signal1/ enable THP              9125            -3.2%          8834  =
     will-it-scale.per_process_ops=0A=
>>>>=0A=
>>>> b) THP disabled:=0A=
>>>> testcase                        base            change          head  =
     metric=0A=
>>>> page_fault3/ disable THP        10107           -19.1%          8180  =
     will-it-scale.per_thread_ops=0A=
>>>> page_fault2/ disable THP         8432           -17.8%          6931  =
     will-it-scale.per_thread_ops=0A=
>>>> context_switch1/ disable THP   215389            -6.8%        200776  =
     will-it-scale.per_thread_ops=0A=
>>>> brk1/ disable THP                 939.67         -6.6%           877.3=
3    will-it-scale.per_thread_ops=0A=
>>>> page_fault3/ disable THP       173145            -4.7%        165064  =
     will-it-scale.per_process_ops=0A=
>>>> signal1/ disable THP             9162            -3.9%          8802  =
     will-it-scale.per_process_ops=0A=
>>>>=0A=
>>>> 2. Improvements:=0A=
>>>> a) THP enabled:=0A=
>>>> testcase                        base            change          head  =
     metric=0A=
>>>> malloc1/ enable THP               66.33        +469.8%           383.6=
7    will-it-scale.per_thread_ops=0A=
>>>> writeseek3/ enable THP          2531             +4.5%          2646  =
     will-it-scale.per_thread_ops=0A=
>>>> signal1/ enable THP              989.33          +2.8%          1016  =
     will-it-scale.per_thread_ops=0A=
>>>>=0A=
>>>> b) THP disabled:=0A=
>>>> testcase                        base            change          head  =
     metric=0A=
>>>> malloc1/ disable THP              90.33        +417.3%           467.3=
3    will-it-scale.per_thread_ops=0A=
>>>> read2/ disable THP             58934            +39.2%         82060  =
     will-it-scale.per_thread_ops=0A=
>>>> page_fault1/ disable THP        8607            +36.4%         11736  =
     will-it-scale.per_thread_ops=0A=
>>>> read1/ disable THP            314063            +12.7%        353934  =
     will-it-scale.per_thread_ops=0A=
>>>> writeseek3/ disable THP         2452            +12.5%          2759  =
     will-it-scale.per_thread_ops=0A=
>>>> signal1/ disable THP             971.33          +5.5%          1024  =
     will-it-scale.per_thread_ops=0A=
>>>>=0A=
>>>> Notes: for above values in column "change", the higher value means tha=
t the related testcase result=0A=
>>>> on head commit is better than that on base commit for this benchmark.=
=0A=
>>>>=0A=
>>>>=0A=
>>>> Best regards=0A=
>>>> Haiyan Song=0A=
>>>>=0A=
>>>> ________________________________________=0A=
>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of=
 Laurent Dufour [ldufour@linux.vnet.ibm.com]=0A=
>>>> Sent: Thursday, May 17, 2018 7:06 PM=0A=
>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org=
; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz=
; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.=
com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas=
 Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; ser=
gey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang,=
 Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minc=
han Kim; Punit Agrawal; vinayak menon; Yang Shi=0A=
>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet=
.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.=
com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org=0A=
>>>> Subject: [PATCH v11 00/26] Speculative page faults=0A=
>>>>=0A=
>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to ha=
ndle=0A=
>>>> page fault without holding the mm semaphore [1].=0A=
>>>>=0A=
>>>> The idea is to try to handle user space page faults without holding th=
e=0A=
>>>> mmap_sem. This should allow better concurrency for massively threaded=
=0A=
>>>> process since the page fault handler will not wait for other threads m=
emory=0A=
>>>> layout change to be done, assuming that this change is done in another=
 part=0A=
>>>> of the process's memory space. This type page fault is named speculati=
ve=0A=
>>>> page fault. If the speculative page fault fails because of a concurren=
cy is=0A=
>>>> detected or because underlying PMD or PTE tables are not yet allocatin=
g, it=0A=
>>>> is failing its processing and a classic page fault is then tried.=0A=
>>>>=0A=
>>>> The speculative page fault (SPF) has to look for the VMA matching the =
fault=0A=
>>>> address without holding the mmap_sem, this is done by introducing a rw=
lock=0A=
>>>> which protects the access to the mm_rb tree. Previously this was done =
using=0A=
>>>> SRCU but it was introducing a lot of scheduling to process the VMA's=
=0A=
>>>> freeing operation which was hitting the performance by 20% as reported=
 by=0A=
>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is=
=0A=
>>>> limiting the locking contention to these operations which are expected=
 to=0A=
>>>> be in a O(log n) order. In addition to ensure that the VMA is not free=
d in=0A=
>>>> our back a reference count is added and 2 services (get_vma() and=0A=
>>>> put_vma()) are introduced to handle the reference count. Once a VMA is=
=0A=
>>>> fetched from the RB tree using get_vma(), it must be later freed using=
=0A=
>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale=
=0A=
>>>> benchmark anymore.=0A=
>>>>=0A=
>>>> The VMA's attributes checked during the speculative page fault process=
ing=0A=
>>>> have to be protected against parallel changes. This is done by using a=
 per=0A=
>>>> VMA sequence lock. This sequence lock allows the speculative page faul=
t=0A=
>>>> handler to fast check for parallel changes in progress and to abort th=
e=0A=
>>>> speculative page fault in that case.=0A=
>>>>=0A=
>>>> Once the VMA has been found, the speculative page fault handler would =
check=0A=
>>>> for the VMA's attributes to verify that the page fault has to be handl=
ed=0A=
>>>> correctly or not. Thus, the VMA is protected through a sequence lock w=
hich=0A=
>>>> allows fast detection of concurrent VMA changes. If such a change is=
=0A=
>>>> detected, the speculative page fault is aborted and a *classic* page f=
ault=0A=
>>>> is tried.  VMA sequence lockings are added when VMA attributes which a=
re=0A=
>>>> checked during the page fault are modified.=0A=
>>>>=0A=
>>>> When the PTE is fetched, the VMA is checked to see if it has been chan=
ged,=0A=
>>>> so once the page table is locked, the VMA is valid, so any other chang=
es=0A=
>>>> leading to touching this PTE will need to lock the page table, so no=
=0A=
>>>> parallel change is possible at this time.=0A=
>>>>=0A=
>>>> The locking of the PTE is done with interrupts disabled, this allows=
=0A=
>>>> checking for the PMD to ensure that there is not an ongoing collapsing=
=0A=
>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and the=
n is=0A=
>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd=
 is=0A=
>>>> valid at the time the PTE is locked, we have the guarantee that the=0A=
>>>> collapsing operation will have to wait on the PTE lock to move forward=
.=0A=
>>>> This allows the SPF handler to map the PTE safely. If the PMD value is=
=0A=
>>>> different from the one recorded at the beginning of the SPF operation,=
 the=0A=
>>>> classic page fault handler will be called to handle the operation whil=
e=0A=
>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disa=
bled,=0A=
>>>> the lock is done using spin_trylock() to avoid dead lock when handling=
 a=0A=
>>>> page fault while a TLB invalidate is requested by another CPU holding =
the=0A=
>>>> PTE.=0A=
>>>>=0A=
>>>> In pseudo code, this could be seen as:=0A=
>>>>     speculative_page_fault()=0A=
>>>>     {=0A=
>>>>             vma =3D get_vma()=0A=
>>>>             check vma sequence count=0A=
>>>>             check vma's support=0A=
>>>>             disable interrupt=0A=
>>>>                   check pgd,p4d,...,pte=0A=
>>>>                   save pmd and pte in vmf=0A=
>>>>                   save vma sequence counter in vmf=0A=
>>>>             enable interrupt=0A=
>>>>             check vma sequence count=0A=
>>>>             handle_pte_fault(vma)=0A=
>>>>                     ..=0A=
>>>>                     page =3D alloc_page()=0A=
>>>>                     pte_map_lock()=0A=
>>>>                             disable interrupt=0A=
>>>>                                     abort if sequence counter has chan=
ged=0A=
>>>>                                     abort if pmd or pte has changed=0A=
>>>>                                     pte map and lock=0A=
>>>>                             enable interrupt=0A=
>>>>                     if abort=0A=
>>>>                        free page=0A=
>>>>                        abort=0A=
>>>>                     ...=0A=
>>>>     }=0A=
>>>>=0A=
>>>>     arch_fault_handler()=0A=
>>>>     {=0A=
>>>>             if (speculative_page_fault(&vma))=0A=
>>>>                goto done=0A=
>>>>     again:=0A=
>>>>             lock(mmap_sem)=0A=
>>>>             vma =3D find_vma();=0A=
>>>>             handle_pte_fault(vma);=0A=
>>>>             if retry=0A=
>>>>                unlock(mmap_sem)=0A=
>>>>                goto again;=0A=
>>>>     done:=0A=
>>>>             handle fault error=0A=
>>>>     }=0A=
>>>>=0A=
>>>> Support for THP is not done because when checking for the PMD, we can =
be=0A=
>>>> confused by an in progress collapsing operation done by khugepaged. Th=
e=0A=
>>>> issue is that pmd_none() could be true either if the PMD is not alread=
y=0A=
>>>> populated or if the underlying PTE are in the way to be collapsed. So =
we=0A=
>>>> cannot safely allocate a PMD if pmd_none() is true.=0A=
>>>>=0A=
>>>> This series add a new software performance event named 'speculative-fa=
ults'=0A=
>>>> or 'spf'. It counts the number of successful page fault event handled=
=0A=
>>>> speculatively. When recording 'faults,spf' events, the faults one is=
=0A=
>>>> counting the total number of page fault events while 'spf' is only cou=
nting=0A=
>>>> the part of the faults processed speculatively.=0A=
>>>>=0A=
>>>> There are some trace events introduced by this series. They allow=0A=
>>>> identifying why the page faults were not processed speculatively. This=
=0A=
>>>> doesn't take in account the faults generated by a monothreaded process=
=0A=
>>>> which directly processed while holding the mmap_sem. This trace events=
 are=0A=
>>>> grouped in a system named 'pagefault', they are:=0A=
>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back=
=0A=
>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.=
=0A=
>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported=0A=
>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected=
=0A=
>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in ou=
r=0A=
>>>>    back.=0A=
>>>>=0A=
>>>> To record all the related events, the easier is to run perf with the=
=0A=
>>>> following arguments :=0A=
>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>=0A=
>>>>=0A=
>>>> There is also a dedicated vmstat counter showing the number of success=
ful=0A=
>>>> page fault handled speculatively. I can be seen this way:=0A=
>>>> $ grep speculative_pgfault /proc/vmstat=0A=
>>>>=0A=
>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is funct=
ional=0A=
>>>> on x86, PowerPC and arm64.=0A=
>>>>=0A=
>>>> ---------------------=0A=
>>>> Real Workload results=0A=
>>>>=0A=
>>>> As mentioned in previous email, we did non official runs using a "popu=
lar=0A=
>>>> in memory multithreaded database product" on 176 cores SMT8 Power syst=
em=0A=
>>>> which showed a 30% improvements in the number of transaction processed=
 per=0A=
>>>> second. This run has been done on the v6 series, but changes introduce=
d in=0A=
>>>> this new version should not impact the performance boost seen.=0A=
>>>>=0A=
>>>> Here are the perf data captured during 2 of these runs on top of the v=
8=0A=
>>>> series:=0A=
>>>>                 vanilla         spf=0A=
>>>> faults          89.418          101.364         +13%=0A=
>>>> spf                n/a           97.989=0A=
>>>>=0A=
>>>> With the SPF kernel, most of the page fault were processed in a specul=
ative=0A=
>>>> way.=0A=
>>>>=0A=
>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and =
gave=0A=
>>>> it a try on an android device. He reported that the application launch=
 time=0A=
>>>> was improved in average by 6%, and for large applications (~100 thread=
s) by=0A=
>>>> 20%.=0A=
>>>>=0A=
>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qco=
m=0A=
>>>> MSM845 (8 cores) with 6GB (the less is better):=0A=
>>>>=0A=
>>>> Application                             4.9     4.9+spf delta=0A=
>>>> com.tencent.mm                          416     389     -7%=0A=
>>>> com.eg.android.AlipayGphone             1135    986     -13%=0A=
>>>> com.tencent.mtt                         455     454     0%=0A=
>>>> com.qqgame.hlddz                        1497    1409    -6%=0A=
>>>> com.autonavi.minimap                    711     701     -1%=0A=
>>>> com.tencent.tmgp.sgame                  788     748     -5%=0A=
>>>> com.immomo.momo                         501     487     -3%=0A=
>>>> com.tencent.peng                        2145    2112    -2%=0A=
>>>> com.smile.gifmaker                      491     461     -6%=0A=
>>>> com.baidu.BaiduMap                      479     366     -23%=0A=
>>>> com.taobao.taobao                       1341    1198    -11%=0A=
>>>> com.baidu.searchbox                     333     314     -6%=0A=
>>>> com.tencent.mobileqq                    394     384     -3%=0A=
>>>> com.sina.weibo                          907     906     0%=0A=
>>>> com.youku.phone                         816     731     -11%=0A=
>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%=0A=
>>>> com.UCMobile                            415     411     -1%=0A=
>>>> com.tencent.tmgp.ak                     1464    1431    -2%=0A=
>>>> com.tencent.qqmusic                     336     329     -2%=0A=
>>>> com.sankuai.meituan                     1661    1302    -22%=0A=
>>>> com.netease.cloudmusic                  1193    1200    1%=0A=
>>>> air.tv.douyu.android                    4257    4152    -2%=0A=
>>>>=0A=
>>>> ------------------=0A=
>>>> Benchmarks results=0A=
>>>>=0A=
>>>> Base kernel is v4.17.0-rc4-mm1=0A=
>>>> SPF is BASE + this series=0A=
>>>>=0A=
>>>> Kernbench:=0A=
>>>> ----------=0A=
>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15=
=0A=
>>>> kernel (kernel is build 5 times):=0A=
>>>>=0A=
>>>> Average Half load -j 8=0A=
>>>>                  Run    (std deviation)=0A=
>>>>                  BASE                   SPF=0A=
>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%=
=0A=
>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%=
=0A=
>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%=
=0A=
>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%=
=0A=
>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%=
=0A=
>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%=
=0A=
>>>>=0A=
>>>> Average Optimal load -j 16=0A=
>>>>                  Run    (std deviation)=0A=
>>>>                  BASE                   SPF=0A=
>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%=
=0A=
>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%=
=0A=
>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%=
=0A=
>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%=
=0A=
>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%=
=0A=
>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%=
=0A=
>>>>=0A=
>>>>=0A=
>>>> During a run on the SPF, perf events were captured:=0A=
>>>>  Performance counter stats for '../kernbench -M':=0A=
>>>>          526743764      faults=0A=
>>>>                210      spf=0A=
>>>>                  3      pagefault:spf_vma_changed=0A=
>>>>                  0      pagefault:spf_vma_noanon=0A=
>>>>               2278      pagefault:spf_vma_notsup=0A=
>>>>                  0      pagefault:spf_vma_access=0A=
>>>>                  0      pagefault:spf_pmd_changed=0A=
>>>>=0A=
>>>> Very few speculative page faults were recorded as most of the processe=
s=0A=
>>>> involved are monothreaded (sounds that on this architecture some threa=
ds=0A=
>>>> were created during the kernel build processing).=0A=
>>>>=0A=
>>>> Here are the kerbench results on a 80 CPUs Power8 system:=0A=
>>>>=0A=
>>>> Average Half load -j 40=0A=
>>>>                  Run    (std deviation)=0A=
>>>>                  BASE                   SPF=0A=
>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%=
=0A=
>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%=
=0A=
>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%=
=0A=
>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%=
=0A=
>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%=
=0A=
>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%=
=0A=
>>>>=0A=
>>>> Average Optimal load -j 80=0A=
>>>>                  Run    (std deviation)=0A=
>>>>                  BASE                   SPF=0A=
>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%=
=0A=
>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%=
=0A=
>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%=
=0A=
>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%=
=0A=
>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%=
=0A=
>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%=
=0A=
>>>>=0A=
>>>> During a run on the SPF, perf events were captured:=0A=
>>>>  Performance counter stats for '../kernbench -M':=0A=
>>>>          116730856      faults=0A=
>>>>                  0      spf=0A=
>>>>                  3      pagefault:spf_vma_changed=0A=
>>>>                  0      pagefault:spf_vma_noanon=0A=
>>>>                476      pagefault:spf_vma_notsup=0A=
>>>>                  0      pagefault:spf_vma_access=0A=
>>>>                  0      pagefault:spf_pmd_changed=0A=
>>>>=0A=
>>>> Most of the processes involved are monothreaded so SPF is not activate=
d but=0A=
>>>> there is no impact on the performance.=0A=
>>>>=0A=
>>>> Ebizzy:=0A=
>>>> -------=0A=
>>>> The test is counting the number of records per second it can manage, t=
he=0A=
>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get=
=0A=
>>>> consistent result I repeated the test 100 times and measure the averag=
e=0A=
>>>> result. The number is the record processes per second, the higher is t=
he=0A=
>>>> best.=0A=
>>>>=0A=
>>>>                 BASE            SPF             delta=0A=
>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%=0A=
>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%=0A=
>>>>=0A=
>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM=
:=0A=
>>>>  Performance counter stats for './ebizzy -mTt 16':=0A=
>>>>            1706379      faults=0A=
>>>>            1674599      spf=0A=
>>>>              30588      pagefault:spf_vma_changed=0A=
>>>>                  0      pagefault:spf_vma_noanon=0A=
>>>>                363      pagefault:spf_vma_notsup=0A=
>>>>                  0      pagefault:spf_vma_access=0A=
>>>>                  0      pagefault:spf_pmd_changed=0A=
>>>>=0A=
>>>> And the ones captured during a run on a 80 CPUs Power node:=0A=
>>>>  Performance counter stats for './ebizzy -mTt 80':=0A=
>>>>            1874773      faults=0A=
>>>>            1461153      spf=0A=
>>>>             413293      pagefault:spf_vma_changed=0A=
>>>>                  0      pagefault:spf_vma_noanon=0A=
>>>>                200      pagefault:spf_vma_notsup=0A=
>>>>                  0      pagefault:spf_vma_access=0A=
>>>>                  0      pagefault:spf_pmd_changed=0A=
>>>>=0A=
>>>> In ebizzy's case most of the page fault were handled in a speculative =
way,=0A=
>>>> leading the ebizzy performance boost.=0A=
>>>>=0A=
>>>> ------------------=0A=
>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):=0A=
>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahen=
dran=0A=
>>>>    and Minchan Kim, hopefully.=0A=
>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in=0A=
>>>>    __do_page_fault().=0A=
>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails=
=0A=
>>>>    instead=0A=
>>>>    of aborting the speculative page fault handling. Dropping the now=
=0A=
>>>> useless=0A=
>>>>    trace event pagefault:spf_pte_lock.=0A=
>>>>  - No more try to reuse the fetched VMA during the speculative page fa=
ult=0A=
>>>>    handling when retrying is needed. This adds a lot of complexity and=
=0A=
>>>>    additional tests done didn't show a significant performance improve=
ment.=0A=
>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.=
=0A=
>>>>=0A=
>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at=
-speculative-page-faults-tt965642.html#none=0A=
>>>> [2] https://patchwork.kernel.org/patch/9999687/=0A=
>>>>=0A=
>>>>=0A=
>>>> Laurent Dufour (20):=0A=
>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT=0A=
>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT=0A=
>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT=0A=
>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE=0A=
>>>>   mm: make pte_unmap_same compatible with SPF=0A=
>>>>   mm: introduce INIT_VMA()=0A=
>>>>   mm: protect VMA modifications using VMA sequence count=0A=
>>>>   mm: protect mremap() against SPF hanlder=0A=
>>>>   mm: protect SPF handler against anon_vma changes=0A=
>>>>   mm: cache some VMA fields in the vm_fault structure=0A=
>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()=0A=
>>>>   mm: introduce __lru_cache_add_active_or_unevictable=0A=
>>>>   mm: introduce __vm_normal_page()=0A=
>>>>   mm: introduce __page_add_new_anon_rmap()=0A=
>>>>   mm: protect mm_rb tree with a rwlock=0A=
>>>>   mm: adding speculative page fault failure trace events=0A=
>>>>   perf: add a speculative page fault sw event=0A=
>>>>   perf tools: add support for the SPF perf event=0A=
>>>>   mm: add speculative page fault vmstats=0A=
>>>>   powerpc/mm: add speculative page fault=0A=
>>>>=0A=
>>>> Mahendran Ganesh (2):=0A=
>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT=0A=
>>>>   arm64/mm: add speculative page fault=0A=
>>>>=0A=
>>>> Peter Zijlstra (4):=0A=
>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE=0A=
>>>>   mm: VMA sequence count=0A=
>>>>   mm: provide speculative fault infrastructure=0A=
>>>>   x86/mm: add speculative pagefault handling=0A=
>>>>=0A=
>>>>  arch/arm64/Kconfig                    |   1 +=0A=
>>>>  arch/arm64/mm/fault.c                 |  12 +=0A=
>>>>  arch/powerpc/Kconfig                  |   1 +=0A=
>>>>  arch/powerpc/mm/fault.c               |  16 +=0A=
>>>>  arch/x86/Kconfig                      |   1 +=0A=
>>>>  arch/x86/mm/fault.c                   |  27 +-=0A=
>>>>  fs/exec.c                             |   2 +-=0A=
>>>>  fs/proc/task_mmu.c                    |   5 +-=0A=
>>>>  fs/userfaultfd.c                      |  17 +-=0A=
>>>>  include/linux/hugetlb_inline.h        |   2 +-=0A=
>>>>  include/linux/migrate.h               |   4 +-=0A=
>>>>  include/linux/mm.h                    | 136 +++++++-=0A=
>>>>  include/linux/mm_types.h              |   7 +=0A=
>>>>  include/linux/pagemap.h               |   4 +-=0A=
>>>>  include/linux/rmap.h                  |  12 +-=0A=
>>>>  include/linux/swap.h                  |  10 +-=0A=
>>>>  include/linux/vm_event_item.h         |   3 +=0A=
>>>>  include/trace/events/pagefault.h      |  80 +++++=0A=
>>>>  include/uapi/linux/perf_event.h       |   1 +=0A=
>>>>  kernel/fork.c                         |   5 +-=0A=
>>>>  mm/Kconfig                            |  22 ++=0A=
>>>>  mm/huge_memory.c                      |   6 +-=0A=
>>>>  mm/hugetlb.c                          |   2 +=0A=
>>>>  mm/init-mm.c                          |   3 +=0A=
>>>>  mm/internal.h                         |  20 ++=0A=
>>>>  mm/khugepaged.c                       |   5 +=0A=
>>>>  mm/madvise.c                          |   6 +-=0A=
>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++=
++++-----=0A=
>>>>  mm/mempolicy.c                        |  51 ++-=0A=
>>>>  mm/migrate.c                          |   6 +-=0A=
>>>>  mm/mlock.c                            |  13 +-=0A=
>>>>  mm/mmap.c                             | 229 ++++++++++---=0A=
>>>>  mm/mprotect.c                         |   4 +-=0A=
>>>>  mm/mremap.c                           |  13 +=0A=
>>>>  mm/nommu.c                            |   2 +-=0A=
>>>>  mm/rmap.c                             |   5 +-=0A=
>>>>  mm/swap.c                             |   6 +-=0A=
>>>>  mm/swap_state.c                       |   8 +-=0A=
>>>>  mm/vmstat.c                           |   5 +-=0A=
>>>>  tools/include/uapi/linux/perf_event.h |   1 +=0A=
>>>>  tools/perf/util/evsel.c               |   1 +=0A=
>>>>  tools/perf/util/parse-events.c        |   4 +=0A=
>>>>  tools/perf/util/parse-events.l        |   1 +=0A=
>>>>  tools/perf/util/python.c              |   1 +=0A=
>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)=0A=
>>>>  create mode 100644 include/trace/events/pagefault.h=0A=
>>>>=0A=
>>>> --=0A=
>>>> 2.7.4=0A=
>>>>=0A=
>>>>=0A=
>>>=0A=
>>=0A=
>=0A=
=0A=

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

On 04/07/2018 05:23, Song, HaiyanX wrote:
> Hi Laurent,
> 
> 
> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.

Repeating the test only 3 times seems a bit too low to me.

I'll focus on the higher change for the moment, but I don't have access to such
a hardware.

Is possible to provide a diff between base and SPF of the performance cycles
measured when running page_fault3 and page_fault2 when the 20% change is detected.

Please stay focus on the test case process to see exactly where the series is
impacting.

Thanks,
Laurent.

> 
> And I did not find other high variation on test case result.
> 
> a). Enable THP
> testcase                          base     stddev       change      head     stddev         metric
> page_fault3/enable THP           10519      A+- 3%        -20.5%      8368      A+-6%          will-it-scale.per_thread_ops
> page_fault2/enalbe THP            8281      A+- 2%        -18.8%      6728                   will-it-scale.per_thread_ops
> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
> 
> b). Disable THP
> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
> 
> 
> Best regards,
> Haiyan Song
> ________________________________________
> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Monday, July 02, 2018 4:59 PM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
> 
> On 11/06/2018 09:49, Song, HaiyanX wrote:
>> Hi Laurent,
>>
>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>> V9 patch serials.
>>
>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>> commit id:
>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>> Benchmark: will-it-scale
>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>
>> Metrics:
>>   will-it-scale.per_process_ops=processes/nr_cpu
>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>> THP: enable / disable
>> nr_task:100%
>>
>> 1. Regressions:
>>
>> a). Enable THP
>> testcase                          base           change      head           metric
>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>
>> b). Disable THP
>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>
>> Notes: for the above  values of test result, the higher is better.
> 
> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
> get reproducible results. The results have huge variation, even on the vanilla
> kernel, and I can't state on any changes due to that.
> 
> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
> measure any changes between the vanilla and the SPF patched ones:
> 
> test THP enabled                4.17.0-rc4-mm1  spf             delta
> page_fault3_threads             2697.7          2683.5          -0.53%
> page_fault2_threads             170660.6        169574.1        -0.64%
> context_switch1_threads         6915269.2       6877507.3       -0.55%
> context_switch1_processes       6478076.2       6529493.5       0.79%
> brk1                            243391.2        238527.5        -2.00%
> 
> Tests were run 10 times, no high variation detected.
> 
> Did you see high variation on your side ? How many times the test were run to
> compute the average values ?
> 
> Thanks,
> Laurent.
> 
> 
>>
>> 2. Improvement: not found improvement based on the selected test cases.
>>
>>
>> Best regards
>> Haiyan Song
>> ________________________________________
>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>> Sent: Monday, May 28, 2018 4:54 PM
>> To: Song, HaiyanX
>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>
>> On 28/05/2018 10:22, Haiyan Song wrote:
>>> Hi Laurent,
>>>
>>> Yes, these tests are done on V9 patch.
>>
>> Do you plan to give this V11 a run ?
>>
>>>
>>>
>>> Best regards,
>>> Haiyan Song
>>>
>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>
>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>> tested on Intel 4s Skylake platform.
>>>>
>>>> Hi,
>>>>
>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>> series" while responding to the v11 header series...
>>>> Were these tests done on v9 or v11 ?
>>>>
>>>> Cheers,
>>>> Laurent.
>>>>
>>>>>
>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>> Commit id:
>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>> Benchmark suite: will-it-scale
>>>>> Download link:
>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>> Metrics:
>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>> THP: enable / disable
>>>>> nr_task: 100%
>>>>>
>>>>> 1. Regressions:
>>>>> a) THP enabled:
>>>>> testcase                        base            change          head       metric
>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>
>>>>> b) THP disabled:
>>>>> testcase                        base            change          head       metric
>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>
>>>>> 2. Improvements:
>>>>> a) THP enabled:
>>>>> testcase                        base            change          head       metric
>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>
>>>>> b) THP disabled:
>>>>> testcase                        base            change          head       metric
>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>
>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>
>>>>>
>>>>> Best regards
>>>>> Haiyan Song
>>>>>
>>>>> ________________________________________
>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>
>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>> page fault without holding the mm semaphore [1].
>>>>>
>>>>> The idea is to try to handle user space page faults without holding the
>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>> process since the page fault handler will not wait for other threads memory
>>>>> layout change to be done, assuming that this change is done in another part
>>>>> of the process's memory space. This type page fault is named speculative
>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>> is failing its processing and a classic page fault is then tried.
>>>>>
>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>> limiting the locking contention to these operations which are expected to
>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>> benchmark anymore.
>>>>>
>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>> have to be protected against parallel changes. This is done by using a per
>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>> speculative page fault in that case.
>>>>>
>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>> checked during the page fault are modified.
>>>>>
>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>> parallel change is possible at this time.
>>>>>
>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>> classic page fault handler will be called to handle the operation while
>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>> PTE.
>>>>>
>>>>> In pseudo code, this could be seen as:
>>>>>     speculative_page_fault()
>>>>>     {
>>>>>             vma = get_vma()
>>>>>             check vma sequence count
>>>>>             check vma's support
>>>>>             disable interrupt
>>>>>                   check pgd,p4d,...,pte
>>>>>                   save pmd and pte in vmf
>>>>>                   save vma sequence counter in vmf
>>>>>             enable interrupt
>>>>>             check vma sequence count
>>>>>             handle_pte_fault(vma)
>>>>>                     ..
>>>>>                     page = alloc_page()
>>>>>                     pte_map_lock()
>>>>>                             disable interrupt
>>>>>                                     abort if sequence counter has changed
>>>>>                                     abort if pmd or pte has changed
>>>>>                                     pte map and lock
>>>>>                             enable interrupt
>>>>>                     if abort
>>>>>                        free page
>>>>>                        abort
>>>>>                     ...
>>>>>     }
>>>>>
>>>>>     arch_fault_handler()
>>>>>     {
>>>>>             if (speculative_page_fault(&vma))
>>>>>                goto done
>>>>>     again:
>>>>>             lock(mmap_sem)
>>>>>             vma = find_vma();
>>>>>             handle_pte_fault(vma);
>>>>>             if retry
>>>>>                unlock(mmap_sem)
>>>>>                goto again;
>>>>>     done:
>>>>>             handle fault error
>>>>>     }
>>>>>
>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>
>>>>> This series add a new software performance event named 'speculative-faults'
>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>> the part of the faults processed speculatively.
>>>>>
>>>>> There are some trace events introduced by this series. They allow
>>>>> identifying why the page faults were not processed speculatively. This
>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>> grouped in a system named 'pagefault', they are:
>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>    back.
>>>>>
>>>>> To record all the related events, the easier is to run perf with the
>>>>> following arguments :
>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>
>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>> page fault handled speculatively. I can be seen this way:
>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>
>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>> on x86, PowerPC and arm64.
>>>>>
>>>>> ---------------------
>>>>> Real Workload results
>>>>>
>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>> this new version should not impact the performance boost seen.
>>>>>
>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>> series:
>>>>>                 vanilla         spf
>>>>> faults          89.418          101.364         +13%
>>>>> spf                n/a           97.989
>>>>>
>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>> way.
>>>>>
>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>> it a try on an android device. He reported that the application launch time
>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>> 20%.
>>>>>
>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>
>>>>> Application                             4.9     4.9+spf delta
>>>>> com.tencent.mm                          416     389     -7%
>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>> com.tencent.mtt                         455     454     0%
>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>> com.autonavi.minimap                    711     701     -1%
>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>> com.immomo.momo                         501     487     -3%
>>>>> com.tencent.peng                        2145    2112    -2%
>>>>> com.smile.gifmaker                      491     461     -6%
>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>> com.baidu.searchbox                     333     314     -6%
>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>> com.sina.weibo                          907     906     0%
>>>>> com.youku.phone                         816     731     -11%
>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>> com.UCMobile                            415     411     -1%
>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>
>>>>> ------------------
>>>>> Benchmarks results
>>>>>
>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>> SPF is BASE + this series
>>>>>
>>>>> Kernbench:
>>>>> ----------
>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>> kernel (kernel is build 5 times):
>>>>>
>>>>> Average Half load -j 8
>>>>>                  Run    (std deviation)
>>>>>                  BASE                   SPF
>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>
>>>>> Average Optimal load -j 16
>>>>>                  Run    (std deviation)
>>>>>                  BASE                   SPF
>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>
>>>>>
>>>>> During a run on the SPF, perf events were captured:
>>>>>  Performance counter stats for '../kernbench -M':
>>>>>          526743764      faults
>>>>>                210      spf
>>>>>                  3      pagefault:spf_vma_changed
>>>>>                  0      pagefault:spf_vma_noanon
>>>>>               2278      pagefault:spf_vma_notsup
>>>>>                  0      pagefault:spf_vma_access
>>>>>                  0      pagefault:spf_pmd_changed
>>>>>
>>>>> Very few speculative page faults were recorded as most of the processes
>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>> were created during the kernel build processing).
>>>>>
>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>
>>>>> Average Half load -j 40
>>>>>                  Run    (std deviation)
>>>>>                  BASE                   SPF
>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>
>>>>> Average Optimal load -j 80
>>>>>                  Run    (std deviation)
>>>>>                  BASE                   SPF
>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>
>>>>> During a run on the SPF, perf events were captured:
>>>>>  Performance counter stats for '../kernbench -M':
>>>>>          116730856      faults
>>>>>                  0      spf
>>>>>                  3      pagefault:spf_vma_changed
>>>>>                  0      pagefault:spf_vma_noanon
>>>>>                476      pagefault:spf_vma_notsup
>>>>>                  0      pagefault:spf_vma_access
>>>>>                  0      pagefault:spf_pmd_changed
>>>>>
>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>> there is no impact on the performance.
>>>>>
>>>>> Ebizzy:
>>>>> -------
>>>>> The test is counting the number of records per second it can manage, the
>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>> consistent result I repeated the test 100 times and measure the average
>>>>> result. The number is the record processes per second, the higher is the
>>>>> best.
>>>>>
>>>>>                 BASE            SPF             delta
>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>
>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>            1706379      faults
>>>>>            1674599      spf
>>>>>              30588      pagefault:spf_vma_changed
>>>>>                  0      pagefault:spf_vma_noanon
>>>>>                363      pagefault:spf_vma_notsup
>>>>>                  0      pagefault:spf_vma_access
>>>>>                  0      pagefault:spf_pmd_changed
>>>>>
>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>            1874773      faults
>>>>>            1461153      spf
>>>>>             413293      pagefault:spf_vma_changed
>>>>>                  0      pagefault:spf_vma_noanon
>>>>>                200      pagefault:spf_vma_notsup
>>>>>                  0      pagefault:spf_vma_access
>>>>>                  0      pagefault:spf_pmd_changed
>>>>>
>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>> leading the ebizzy performance boost.
>>>>>
>>>>> ------------------
>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>    and Minchan Kim, hopefully.
>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>    __do_page_fault().
>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>    instead
>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>> useless
>>>>>    trace event pagefault:spf_pte_lock.
>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>
>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>
>>>>>
>>>>> Laurent Dufour (20):
>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>   mm: introduce INIT_VMA()
>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>   mm: protect mremap() against SPF hanlder
>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>   mm: introduce __vm_normal_page()
>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>   mm: adding speculative page fault failure trace events
>>>>>   perf: add a speculative page fault sw event
>>>>>   perf tools: add support for the SPF perf event
>>>>>   mm: add speculative page fault vmstats
>>>>>   powerpc/mm: add speculative page fault
>>>>>
>>>>> Mahendran Ganesh (2):
>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>   arm64/mm: add speculative page fault
>>>>>
>>>>> Peter Zijlstra (4):
>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>   mm: VMA sequence count
>>>>>   mm: provide speculative fault infrastructure
>>>>>   x86/mm: add speculative pagefault handling
>>>>>
>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>  fs/exec.c                             |   2 +-
>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>  include/linux/migrate.h               |   4 +-
>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>  include/linux/mm_types.h              |   7 +
>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>  include/linux/swap.h                  |  10 +-
>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>  kernel/fork.c                         |   5 +-
>>>>>  mm/Kconfig                            |  22 ++
>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>  mm/hugetlb.c                          |   2 +
>>>>>  mm/init-mm.c                          |   3 +
>>>>>  mm/internal.h                         |  20 ++
>>>>>  mm/khugepaged.c                       |   5 +
>>>>>  mm/madvise.c                          |   6 +-
>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>  mm/migrate.c                          |   6 +-
>>>>>  mm/mlock.c                            |  13 +-
>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>  mm/mprotect.c                         |   4 +-
>>>>>  mm/mremap.c                           |  13 +
>>>>>  mm/nommu.c                            |   2 +-
>>>>>  mm/rmap.c                             |   5 +-
>>>>>  mm/swap.c                             |   6 +-
>>>>>  mm/swap_state.c                       |   8 +-
>>>>>  mm/vmstat.c                           |   5 +-
>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>  tools/perf/util/python.c              |   1 +
>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>
>>>>> --
>>>>> 2.7.4
>>>>>
>>>>>
>>>>
>>>
>>
> 
> 

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

On 04/07/2018 05:23, Song, HaiyanX wrote:
> Hi Laurent,
> 
> 
> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.

Repeating the test only 3 times seems a bit too low to me.

I'll focus on the higher change for the moment, but I don't have access to such
a hardware.

Is possible to provide a diff between base and SPF of the performance cycles
measured when running page_fault3 and page_fault2 when the 20% change is detected.

Please stay focus on the test case process to see exactly where the series is
impacting.

Thanks,
Laurent.

> 
> And I did not find other high variation on test case result.
> 
> a). Enable THP
> testcase                          base     stddev       change      head     stddev         metric
> page_fault3/enable THP           10519      ± 3%        -20.5%      8368      ±6%          will-it-scale.per_thread_ops
> page_fault2/enalbe THP            8281      ± 2%        -18.8%      6728                   will-it-scale.per_thread_ops
> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
> 
> b). Disable THP
> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
> 
> 
> Best regards,
> Haiyan Song
> ________________________________________
> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Monday, July 02, 2018 4:59 PM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
> 
> On 11/06/2018 09:49, Song, HaiyanX wrote:
>> Hi Laurent,
>>
>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>> V9 patch serials.
>>
>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>> commit id:
>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>> Benchmark: will-it-scale
>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>
>> Metrics:
>>   will-it-scale.per_process_ops=processes/nr_cpu
>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>> THP: enable / disable
>> nr_task:100%
>>
>> 1. Regressions:
>>
>> a). Enable THP
>> testcase                          base           change      head           metric
>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>
>> b). Disable THP
>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>
>> Notes: for the above  values of test result, the higher is better.
> 
> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
> get reproducible results. The results have huge variation, even on the vanilla
> kernel, and I can't state on any changes due to that.
> 
> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
> measure any changes between the vanilla and the SPF patched ones:
> 
> test THP enabled                4.17.0-rc4-mm1  spf             delta
> page_fault3_threads             2697.7          2683.5          -0.53%
> page_fault2_threads             170660.6        169574.1        -0.64%
> context_switch1_threads         6915269.2       6877507.3       -0.55%
> context_switch1_processes       6478076.2       6529493.5       0.79%
> brk1                            243391.2        238527.5        -2.00%
> 
> Tests were run 10 times, no high variation detected.
> 
> Did you see high variation on your side ? How many times the test were run to
> compute the average values ?
> 
> Thanks,
> Laurent.
> 
> 
>>
>> 2. Improvement: not found improvement based on the selected test cases.
>>
>>
>> Best regards
>> Haiyan Song
>> ________________________________________
>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>> Sent: Monday, May 28, 2018 4:54 PM
>> To: Song, HaiyanX
>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>
>> On 28/05/2018 10:22, Haiyan Song wrote:
>>> Hi Laurent,
>>>
>>> Yes, these tests are done on V9 patch.
>>
>> Do you plan to give this V11 a run ?
>>
>>>
>>>
>>> Best regards,
>>> Haiyan Song
>>>
>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>
>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>> tested on Intel 4s Skylake platform.
>>>>
>>>> Hi,
>>>>
>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>> series" while responding to the v11 header series...
>>>> Were these tests done on v9 or v11 ?
>>>>
>>>> Cheers,
>>>> Laurent.
>>>>
>>>>>
>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>> Commit id:
>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>> Benchmark suite: will-it-scale
>>>>> Download link:
>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>> Metrics:
>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>> THP: enable / disable
>>>>> nr_task: 100%
>>>>>
>>>>> 1. Regressions:
>>>>> a) THP enabled:
>>>>> testcase                        base            change          head       metric
>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>
>>>>> b) THP disabled:
>>>>> testcase                        base            change          head       metric
>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>
>>>>> 2. Improvements:
>>>>> a) THP enabled:
>>>>> testcase                        base            change          head       metric
>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>
>>>>> b) THP disabled:
>>>>> testcase                        base            change          head       metric
>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>
>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>
>>>>>
>>>>> Best regards
>>>>> Haiyan Song
>>>>>
>>>>> ________________________________________
>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>
>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>> page fault without holding the mm semaphore [1].
>>>>>
>>>>> The idea is to try to handle user space page faults without holding the
>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>> process since the page fault handler will not wait for other threads memory
>>>>> layout change to be done, assuming that this change is done in another part
>>>>> of the process's memory space. This type page fault is named speculative
>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>> is failing its processing and a classic page fault is then tried.
>>>>>
>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>> limiting the locking contention to these operations which are expected to
>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>> benchmark anymore.
>>>>>
>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>> have to be protected against parallel changes. This is done by using a per
>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>> speculative page fault in that case.
>>>>>
>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>> checked during the page fault are modified.
>>>>>
>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>> parallel change is possible at this time.
>>>>>
>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>> classic page fault handler will be called to handle the operation while
>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>> PTE.
>>>>>
>>>>> In pseudo code, this could be seen as:
>>>>>     speculative_page_fault()
>>>>>     {
>>>>>             vma = get_vma()
>>>>>             check vma sequence count
>>>>>             check vma's support
>>>>>             disable interrupt
>>>>>                   check pgd,p4d,...,pte
>>>>>                   save pmd and pte in vmf
>>>>>                   save vma sequence counter in vmf
>>>>>             enable interrupt
>>>>>             check vma sequence count
>>>>>             handle_pte_fault(vma)
>>>>>                     ..
>>>>>                     page = alloc_page()
>>>>>                     pte_map_lock()
>>>>>                             disable interrupt
>>>>>                                     abort if sequence counter has changed
>>>>>                                     abort if pmd or pte has changed
>>>>>                                     pte map and lock
>>>>>                             enable interrupt
>>>>>                     if abort
>>>>>                        free page
>>>>>                        abort
>>>>>                     ...
>>>>>     }
>>>>>
>>>>>     arch_fault_handler()
>>>>>     {
>>>>>             if (speculative_page_fault(&vma))
>>>>>                goto done
>>>>>     again:
>>>>>             lock(mmap_sem)
>>>>>             vma = find_vma();
>>>>>             handle_pte_fault(vma);
>>>>>             if retry
>>>>>                unlock(mmap_sem)
>>>>>                goto again;
>>>>>     done:
>>>>>             handle fault error
>>>>>     }
>>>>>
>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>
>>>>> This series add a new software performance event named 'speculative-faults'
>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>> the part of the faults processed speculatively.
>>>>>
>>>>> There are some trace events introduced by this series. They allow
>>>>> identifying why the page faults were not processed speculatively. This
>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>> grouped in a system named 'pagefault', they are:
>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>    back.
>>>>>
>>>>> To record all the related events, the easier is to run perf with the
>>>>> following arguments :
>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>
>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>> page fault handled speculatively. I can be seen this way:
>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>
>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>> on x86, PowerPC and arm64.
>>>>>
>>>>> ---------------------
>>>>> Real Workload results
>>>>>
>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>> this new version should not impact the performance boost seen.
>>>>>
>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>> series:
>>>>>                 vanilla         spf
>>>>> faults          89.418          101.364         +13%
>>>>> spf                n/a           97.989
>>>>>
>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>> way.
>>>>>
>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>> it a try on an android device. He reported that the application launch time
>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>> 20%.
>>>>>
>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>
>>>>> Application                             4.9     4.9+spf delta
>>>>> com.tencent.mm                          416     389     -7%
>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>> com.tencent.mtt                         455     454     0%
>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>> com.autonavi.minimap                    711     701     -1%
>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>> com.immomo.momo                         501     487     -3%
>>>>> com.tencent.peng                        2145    2112    -2%
>>>>> com.smile.gifmaker                      491     461     -6%
>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>> com.baidu.searchbox                     333     314     -6%
>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>> com.sina.weibo                          907     906     0%
>>>>> com.youku.phone                         816     731     -11%
>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>> com.UCMobile                            415     411     -1%
>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>
>>>>> ------------------
>>>>> Benchmarks results
>>>>>
>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>> SPF is BASE + this series
>>>>>
>>>>> Kernbench:
>>>>> ----------
>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>> kernel (kernel is build 5 times):
>>>>>
>>>>> Average Half load -j 8
>>>>>                  Run    (std deviation)
>>>>>                  BASE                   SPF
>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>
>>>>> Average Optimal load -j 16
>>>>>                  Run    (std deviation)
>>>>>                  BASE                   SPF
>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>
>>>>>
>>>>> During a run on the SPF, perf events were captured:
>>>>>  Performance counter stats for '../kernbench -M':
>>>>>          526743764      faults
>>>>>                210      spf
>>>>>                  3      pagefault:spf_vma_changed
>>>>>                  0      pagefault:spf_vma_noanon
>>>>>               2278      pagefault:spf_vma_notsup
>>>>>                  0      pagefault:spf_vma_access
>>>>>                  0      pagefault:spf_pmd_changed
>>>>>
>>>>> Very few speculative page faults were recorded as most of the processes
>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>> were created during the kernel build processing).
>>>>>
>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>
>>>>> Average Half load -j 40
>>>>>                  Run    (std deviation)
>>>>>                  BASE                   SPF
>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>
>>>>> Average Optimal load -j 80
>>>>>                  Run    (std deviation)
>>>>>                  BASE                   SPF
>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>
>>>>> During a run on the SPF, perf events were captured:
>>>>>  Performance counter stats for '../kernbench -M':
>>>>>          116730856      faults
>>>>>                  0      spf
>>>>>                  3      pagefault:spf_vma_changed
>>>>>                  0      pagefault:spf_vma_noanon
>>>>>                476      pagefault:spf_vma_notsup
>>>>>                  0      pagefault:spf_vma_access
>>>>>                  0      pagefault:spf_pmd_changed
>>>>>
>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>> there is no impact on the performance.
>>>>>
>>>>> Ebizzy:
>>>>> -------
>>>>> The test is counting the number of records per second it can manage, the
>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>> consistent result I repeated the test 100 times and measure the average
>>>>> result. The number is the record processes per second, the higher is the
>>>>> best.
>>>>>
>>>>>                 BASE            SPF             delta
>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>
>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>            1706379      faults
>>>>>            1674599      spf
>>>>>              30588      pagefault:spf_vma_changed
>>>>>                  0      pagefault:spf_vma_noanon
>>>>>                363      pagefault:spf_vma_notsup
>>>>>                  0      pagefault:spf_vma_access
>>>>>                  0      pagefault:spf_pmd_changed
>>>>>
>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>            1874773      faults
>>>>>            1461153      spf
>>>>>             413293      pagefault:spf_vma_changed
>>>>>                  0      pagefault:spf_vma_noanon
>>>>>                200      pagefault:spf_vma_notsup
>>>>>                  0      pagefault:spf_vma_access
>>>>>                  0      pagefault:spf_pmd_changed
>>>>>
>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>> leading the ebizzy performance boost.
>>>>>
>>>>> ------------------
>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>    and Minchan Kim, hopefully.
>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>    __do_page_fault().
>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>    instead
>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>> useless
>>>>>    trace event pagefault:spf_pte_lock.
>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>
>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>
>>>>>
>>>>> Laurent Dufour (20):
>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>   mm: introduce INIT_VMA()
>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>   mm: protect mremap() against SPF hanlder
>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>   mm: introduce __vm_normal_page()
>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>   mm: adding speculative page fault failure trace events
>>>>>   perf: add a speculative page fault sw event
>>>>>   perf tools: add support for the SPF perf event
>>>>>   mm: add speculative page fault vmstats
>>>>>   powerpc/mm: add speculative page fault
>>>>>
>>>>> Mahendran Ganesh (2):
>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>   arm64/mm: add speculative page fault
>>>>>
>>>>> Peter Zijlstra (4):
>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>   mm: VMA sequence count
>>>>>   mm: provide speculative fault infrastructure
>>>>>   x86/mm: add speculative pagefault handling
>>>>>
>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>  fs/exec.c                             |   2 +-
>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>  include/linux/migrate.h               |   4 +-
>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>  include/linux/mm_types.h              |   7 +
>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>  include/linux/swap.h                  |  10 +-
>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>  kernel/fork.c                         |   5 +-
>>>>>  mm/Kconfig                            |  22 ++
>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>  mm/hugetlb.c                          |   2 +
>>>>>  mm/init-mm.c                          |   3 +
>>>>>  mm/internal.h                         |  20 ++
>>>>>  mm/khugepaged.c                       |   5 +
>>>>>  mm/madvise.c                          |   6 +-
>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>  mm/migrate.c                          |   6 +-
>>>>>  mm/mlock.c                            |  13 +-
>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>  mm/mprotect.c                         |   4 +-
>>>>>  mm/mremap.c                           |  13 +
>>>>>  mm/nommu.c                            |   2 +-
>>>>>  mm/rmap.c                             |   5 +-
>>>>>  mm/swap.c                             |   6 +-
>>>>>  mm/swap_state.c                       |   8 +-
>>>>>  mm/vmstat.c                           |   5 +-
>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>  tools/perf/util/python.c              |   1 +
>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>
>>>>> --
>>>>> 2.7.4
>>>>>
>>>>>
>>>>
>>>
>>
> 
> 

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Hi Haiyan,

Do you get a chance to capture some performance cycles on your system ?
I still can't get these numbers on my hardware.

Thanks,
Laurent.

On 04/07/2018 09:51, Laurent Dufour wrote:
> On 04/07/2018 05:23, Song, HaiyanX wrote:
>> Hi Laurent,
>>
>>
>> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
>> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.
> 
> Repeating the test only 3 times seems a bit too low to me.
> 
> I'll focus on the higher change for the moment, but I don't have access to such
> a hardware.
> 
> Is possible to provide a diff between base and SPF of the performance cycles
> measured when running page_fault3 and page_fault2 when the 20% change is detected.
> 
> Please stay focus on the test case process to see exactly where the series is
> impacting.
> 
> Thanks,
> Laurent.
> 
>>
>> And I did not find other high variation on test case result.
>>
>> a). Enable THP
>> testcase                          base     stddev       change      head     stddev         metric
>> page_fault3/enable THP           10519      A+- 3%        -20.5%      8368      A+-6%          will-it-scale.per_thread_ops
>> page_fault2/enalbe THP            8281      A+- 2%        -18.8%      6728                   will-it-scale.per_thread_ops
>> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
>> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
>> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
>>
>> b). Disable THP
>> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
>> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
>> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
>> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
>> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
>> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
>> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
>>
>>
>> Best regards,
>> Haiyan Song
>> ________________________________________
>> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
>> Sent: Monday, July 02, 2018 4:59 PM
>> To: Song, HaiyanX
>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>
>> On 11/06/2018 09:49, Song, HaiyanX wrote:
>>> Hi Laurent,
>>>
>>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>>> V9 patch serials.
>>>
>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>>> commit id:
>>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>>> Benchmark: will-it-scale
>>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>>
>>> Metrics:
>>>   will-it-scale.per_process_ops=processes/nr_cpu
>>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>> THP: enable / disable
>>> nr_task:100%
>>>
>>> 1. Regressions:
>>>
>>> a). Enable THP
>>> testcase                          base           change      head           metric
>>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>>
>>> b). Disable THP
>>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>>
>>> Notes: for the above  values of test result, the higher is better.
>>
>> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
>> get reproducible results. The results have huge variation, even on the vanilla
>> kernel, and I can't state on any changes due to that.
>>
>> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
>> measure any changes between the vanilla and the SPF patched ones:
>>
>> test THP enabled                4.17.0-rc4-mm1  spf             delta
>> page_fault3_threads             2697.7          2683.5          -0.53%
>> page_fault2_threads             170660.6        169574.1        -0.64%
>> context_switch1_threads         6915269.2       6877507.3       -0.55%
>> context_switch1_processes       6478076.2       6529493.5       0.79%
>> brk1                            243391.2        238527.5        -2.00%
>>
>> Tests were run 10 times, no high variation detected.
>>
>> Did you see high variation on your side ? How many times the test were run to
>> compute the average values ?
>>
>> Thanks,
>> Laurent.
>>
>>
>>>
>>> 2. Improvement: not found improvement based on the selected test cases.
>>>
>>>
>>> Best regards
>>> Haiyan Song
>>> ________________________________________
>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Monday, May 28, 2018 4:54 PM
>>> To: Song, HaiyanX
>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>
>>> On 28/05/2018 10:22, Haiyan Song wrote:
>>>> Hi Laurent,
>>>>
>>>> Yes, these tests are done on V9 patch.
>>>
>>> Do you plan to give this V11 a run ?
>>>
>>>>
>>>>
>>>> Best regards,
>>>> Haiyan Song
>>>>
>>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>>
>>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>>> tested on Intel 4s Skylake platform.
>>>>>
>>>>> Hi,
>>>>>
>>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>>> series" while responding to the v11 header series...
>>>>> Were these tests done on v9 or v11 ?
>>>>>
>>>>> Cheers,
>>>>> Laurent.
>>>>>
>>>>>>
>>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>>> Commit id:
>>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>>> Benchmark suite: will-it-scale
>>>>>> Download link:
>>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>>> Metrics:
>>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>>> THP: enable / disable
>>>>>> nr_task: 100%
>>>>>>
>>>>>> 1. Regressions:
>>>>>> a) THP enabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>>
>>>>>> b) THP disabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>>
>>>>>> 2. Improvements:
>>>>>> a) THP enabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>>
>>>>>> b) THP disabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>>
>>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>>
>>>>>>
>>>>>> Best regards
>>>>>> Haiyan Song
>>>>>>
>>>>>> ________________________________________
>>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>>
>>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>>> page fault without holding the mm semaphore [1].
>>>>>>
>>>>>> The idea is to try to handle user space page faults without holding the
>>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>>> process since the page fault handler will not wait for other threads memory
>>>>>> layout change to be done, assuming that this change is done in another part
>>>>>> of the process's memory space. This type page fault is named speculative
>>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>>> is failing its processing and a classic page fault is then tried.
>>>>>>
>>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>>> limiting the locking contention to these operations which are expected to
>>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>>> benchmark anymore.
>>>>>>
>>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>>> have to be protected against parallel changes. This is done by using a per
>>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>>> speculative page fault in that case.
>>>>>>
>>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>>> checked during the page fault are modified.
>>>>>>
>>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>>> parallel change is possible at this time.
>>>>>>
>>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>>> classic page fault handler will be called to handle the operation while
>>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>>> PTE.
>>>>>>
>>>>>> In pseudo code, this could be seen as:
>>>>>>     speculative_page_fault()
>>>>>>     {
>>>>>>             vma = get_vma()
>>>>>>             check vma sequence count
>>>>>>             check vma's support
>>>>>>             disable interrupt
>>>>>>                   check pgd,p4d,...,pte
>>>>>>                   save pmd and pte in vmf
>>>>>>                   save vma sequence counter in vmf
>>>>>>             enable interrupt
>>>>>>             check vma sequence count
>>>>>>             handle_pte_fault(vma)
>>>>>>                     ..
>>>>>>                     page = alloc_page()
>>>>>>                     pte_map_lock()
>>>>>>                             disable interrupt
>>>>>>                                     abort if sequence counter has changed
>>>>>>                                     abort if pmd or pte has changed
>>>>>>                                     pte map and lock
>>>>>>                             enable interrupt
>>>>>>                     if abort
>>>>>>                        free page
>>>>>>                        abort
>>>>>>                     ...
>>>>>>     }
>>>>>>
>>>>>>     arch_fault_handler()
>>>>>>     {
>>>>>>             if (speculative_page_fault(&vma))
>>>>>>                goto done
>>>>>>     again:
>>>>>>             lock(mmap_sem)
>>>>>>             vma = find_vma();
>>>>>>             handle_pte_fault(vma);
>>>>>>             if retry
>>>>>>                unlock(mmap_sem)
>>>>>>                goto again;
>>>>>>     done:
>>>>>>             handle fault error
>>>>>>     }
>>>>>>
>>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>>
>>>>>> This series add a new software performance event named 'speculative-faults'
>>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>>> the part of the faults processed speculatively.
>>>>>>
>>>>>> There are some trace events introduced by this series. They allow
>>>>>> identifying why the page faults were not processed speculatively. This
>>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>>> grouped in a system named 'pagefault', they are:
>>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>>    back.
>>>>>>
>>>>>> To record all the related events, the easier is to run perf with the
>>>>>> following arguments :
>>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>>
>>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>>> page fault handled speculatively. I can be seen this way:
>>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>>
>>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>>> on x86, PowerPC and arm64.
>>>>>>
>>>>>> ---------------------
>>>>>> Real Workload results
>>>>>>
>>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>>> this new version should not impact the performance boost seen.
>>>>>>
>>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>>> series:
>>>>>>                 vanilla         spf
>>>>>> faults          89.418          101.364         +13%
>>>>>> spf                n/a           97.989
>>>>>>
>>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>>> way.
>>>>>>
>>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>>> it a try on an android device. He reported that the application launch time
>>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>>> 20%.
>>>>>>
>>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>>
>>>>>> Application                             4.9     4.9+spf delta
>>>>>> com.tencent.mm                          416     389     -7%
>>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>>> com.tencent.mtt                         455     454     0%
>>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>>> com.autonavi.minimap                    711     701     -1%
>>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>>> com.immomo.momo                         501     487     -3%
>>>>>> com.tencent.peng                        2145    2112    -2%
>>>>>> com.smile.gifmaker                      491     461     -6%
>>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>>> com.baidu.searchbox                     333     314     -6%
>>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>>> com.sina.weibo                          907     906     0%
>>>>>> com.youku.phone                         816     731     -11%
>>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>>> com.UCMobile                            415     411     -1%
>>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>>
>>>>>> ------------------
>>>>>> Benchmarks results
>>>>>>
>>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>>> SPF is BASE + this series
>>>>>>
>>>>>> Kernbench:
>>>>>> ----------
>>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>>> kernel (kernel is build 5 times):
>>>>>>
>>>>>> Average Half load -j 8
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>>
>>>>>> Average Optimal load -j 16
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>>
>>>>>>
>>>>>> During a run on the SPF, perf events were captured:
>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>          526743764      faults
>>>>>>                210      spf
>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>               2278      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> Very few speculative page faults were recorded as most of the processes
>>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>>> were created during the kernel build processing).
>>>>>>
>>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>>
>>>>>> Average Half load -j 40
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>>
>>>>>> Average Optimal load -j 80
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>>
>>>>>> During a run on the SPF, perf events were captured:
>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>          116730856      faults
>>>>>>                  0      spf
>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>                476      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>>> there is no impact on the performance.
>>>>>>
>>>>>> Ebizzy:
>>>>>> -------
>>>>>> The test is counting the number of records per second it can manage, the
>>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>>> consistent result I repeated the test 100 times and measure the average
>>>>>> result. The number is the record processes per second, the higher is the
>>>>>> best.
>>>>>>
>>>>>>                 BASE            SPF             delta
>>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>>
>>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>>            1706379      faults
>>>>>>            1674599      spf
>>>>>>              30588      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>                363      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>>            1874773      faults
>>>>>>            1461153      spf
>>>>>>             413293      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>                200      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>>> leading the ebizzy performance boost.
>>>>>>
>>>>>> ------------------
>>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>>    and Minchan Kim, hopefully.
>>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>>    __do_page_fault().
>>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>>    instead
>>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>>> useless
>>>>>>    trace event pagefault:spf_pte_lock.
>>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>>
>>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>>
>>>>>>
>>>>>> Laurent Dufour (20):
>>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>>   mm: introduce INIT_VMA()
>>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>>   mm: protect mremap() against SPF hanlder
>>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>>   mm: introduce __vm_normal_page()
>>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>>   mm: adding speculative page fault failure trace events
>>>>>>   perf: add a speculative page fault sw event
>>>>>>   perf tools: add support for the SPF perf event
>>>>>>   mm: add speculative page fault vmstats
>>>>>>   powerpc/mm: add speculative page fault
>>>>>>
>>>>>> Mahendran Ganesh (2):
>>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>   arm64/mm: add speculative page fault
>>>>>>
>>>>>> Peter Zijlstra (4):
>>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>>   mm: VMA sequence count
>>>>>>   mm: provide speculative fault infrastructure
>>>>>>   x86/mm: add speculative pagefault handling
>>>>>>
>>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>>  fs/exec.c                             |   2 +-
>>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>>  include/linux/migrate.h               |   4 +-
>>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>>  include/linux/mm_types.h              |   7 +
>>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>>  include/linux/swap.h                  |  10 +-
>>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>>  kernel/fork.c                         |   5 +-
>>>>>>  mm/Kconfig                            |  22 ++
>>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>>  mm/hugetlb.c                          |   2 +
>>>>>>  mm/init-mm.c                          |   3 +
>>>>>>  mm/internal.h                         |  20 ++
>>>>>>  mm/khugepaged.c                       |   5 +
>>>>>>  mm/madvise.c                          |   6 +-
>>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>>  mm/migrate.c                          |   6 +-
>>>>>>  mm/mlock.c                            |  13 +-
>>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>>  mm/mprotect.c                         |   4 +-
>>>>>>  mm/mremap.c                           |  13 +
>>>>>>  mm/nommu.c                            |   2 +-
>>>>>>  mm/rmap.c                             |   5 +-
>>>>>>  mm/swap.c                             |   6 +-
>>>>>>  mm/swap_state.c                       |   8 +-
>>>>>>  mm/vmstat.c                           |   5 +-
>>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>>  tools/perf/util/python.c              |   1 +
>>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>>
>>>>>> --
>>>>>> 2.7.4
>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>>
> 

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Hi Haiyan,

Do you get a chance to capture some performance cycles on your system ?
I still can't get these numbers on my hardware.

Thanks,
Laurent.

On 04/07/2018 09:51, Laurent Dufour wrote:
> On 04/07/2018 05:23, Song, HaiyanX wrote:
>> Hi Laurent,
>>
>>
>> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
>> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.
> 
> Repeating the test only 3 times seems a bit too low to me.
> 
> I'll focus on the higher change for the moment, but I don't have access to such
> a hardware.
> 
> Is possible to provide a diff between base and SPF of the performance cycles
> measured when running page_fault3 and page_fault2 when the 20% change is detected.
> 
> Please stay focus on the test case process to see exactly where the series is
> impacting.
> 
> Thanks,
> Laurent.
> 
>>
>> And I did not find other high variation on test case result.
>>
>> a). Enable THP
>> testcase                          base     stddev       change      head     stddev         metric
>> page_fault3/enable THP           10519      ± 3%        -20.5%      8368      ±6%          will-it-scale.per_thread_ops
>> page_fault2/enalbe THP            8281      ± 2%        -18.8%      6728                   will-it-scale.per_thread_ops
>> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
>> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
>> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
>>
>> b). Disable THP
>> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
>> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
>> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
>> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
>> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
>> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
>> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
>>
>>
>> Best regards,
>> Haiyan Song
>> ________________________________________
>> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
>> Sent: Monday, July 02, 2018 4:59 PM
>> To: Song, HaiyanX
>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>
>> On 11/06/2018 09:49, Song, HaiyanX wrote:
>>> Hi Laurent,
>>>
>>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>>> V9 patch serials.
>>>
>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>>> commit id:
>>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>>> Benchmark: will-it-scale
>>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>>
>>> Metrics:
>>>   will-it-scale.per_process_ops=processes/nr_cpu
>>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>> THP: enable / disable
>>> nr_task:100%
>>>
>>> 1. Regressions:
>>>
>>> a). Enable THP
>>> testcase                          base           change      head           metric
>>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>>
>>> b). Disable THP
>>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>>
>>> Notes: for the above  values of test result, the higher is better.
>>
>> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
>> get reproducible results. The results have huge variation, even on the vanilla
>> kernel, and I can't state on any changes due to that.
>>
>> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
>> measure any changes between the vanilla and the SPF patched ones:
>>
>> test THP enabled                4.17.0-rc4-mm1  spf             delta
>> page_fault3_threads             2697.7          2683.5          -0.53%
>> page_fault2_threads             170660.6        169574.1        -0.64%
>> context_switch1_threads         6915269.2       6877507.3       -0.55%
>> context_switch1_processes       6478076.2       6529493.5       0.79%
>> brk1                            243391.2        238527.5        -2.00%
>>
>> Tests were run 10 times, no high variation detected.
>>
>> Did you see high variation on your side ? How many times the test were run to
>> compute the average values ?
>>
>> Thanks,
>> Laurent.
>>
>>
>>>
>>> 2. Improvement: not found improvement based on the selected test cases.
>>>
>>>
>>> Best regards
>>> Haiyan Song
>>> ________________________________________
>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Monday, May 28, 2018 4:54 PM
>>> To: Song, HaiyanX
>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>
>>> On 28/05/2018 10:22, Haiyan Song wrote:
>>>> Hi Laurent,
>>>>
>>>> Yes, these tests are done on V9 patch.
>>>
>>> Do you plan to give this V11 a run ?
>>>
>>>>
>>>>
>>>> Best regards,
>>>> Haiyan Song
>>>>
>>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>>
>>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>>> tested on Intel 4s Skylake platform.
>>>>>
>>>>> Hi,
>>>>>
>>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>>> series" while responding to the v11 header series...
>>>>> Were these tests done on v9 or v11 ?
>>>>>
>>>>> Cheers,
>>>>> Laurent.
>>>>>
>>>>>>
>>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>>> Commit id:
>>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>>> Benchmark suite: will-it-scale
>>>>>> Download link:
>>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>>> Metrics:
>>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>>> THP: enable / disable
>>>>>> nr_task: 100%
>>>>>>
>>>>>> 1. Regressions:
>>>>>> a) THP enabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>>
>>>>>> b) THP disabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>>
>>>>>> 2. Improvements:
>>>>>> a) THP enabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>>
>>>>>> b) THP disabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>>
>>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>>
>>>>>>
>>>>>> Best regards
>>>>>> Haiyan Song
>>>>>>
>>>>>> ________________________________________
>>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>>
>>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>>> page fault without holding the mm semaphore [1].
>>>>>>
>>>>>> The idea is to try to handle user space page faults without holding the
>>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>>> process since the page fault handler will not wait for other threads memory
>>>>>> layout change to be done, assuming that this change is done in another part
>>>>>> of the process's memory space. This type page fault is named speculative
>>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>>> is failing its processing and a classic page fault is then tried.
>>>>>>
>>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>>> limiting the locking contention to these operations which are expected to
>>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>>> benchmark anymore.
>>>>>>
>>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>>> have to be protected against parallel changes. This is done by using a per
>>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>>> speculative page fault in that case.
>>>>>>
>>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>>> checked during the page fault are modified.
>>>>>>
>>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>>> parallel change is possible at this time.
>>>>>>
>>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>>> classic page fault handler will be called to handle the operation while
>>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>>> PTE.
>>>>>>
>>>>>> In pseudo code, this could be seen as:
>>>>>>     speculative_page_fault()
>>>>>>     {
>>>>>>             vma = get_vma()
>>>>>>             check vma sequence count
>>>>>>             check vma's support
>>>>>>             disable interrupt
>>>>>>                   check pgd,p4d,...,pte
>>>>>>                   save pmd and pte in vmf
>>>>>>                   save vma sequence counter in vmf
>>>>>>             enable interrupt
>>>>>>             check vma sequence count
>>>>>>             handle_pte_fault(vma)
>>>>>>                     ..
>>>>>>                     page = alloc_page()
>>>>>>                     pte_map_lock()
>>>>>>                             disable interrupt
>>>>>>                                     abort if sequence counter has changed
>>>>>>                                     abort if pmd or pte has changed
>>>>>>                                     pte map and lock
>>>>>>                             enable interrupt
>>>>>>                     if abort
>>>>>>                        free page
>>>>>>                        abort
>>>>>>                     ...
>>>>>>     }
>>>>>>
>>>>>>     arch_fault_handler()
>>>>>>     {
>>>>>>             if (speculative_page_fault(&vma))
>>>>>>                goto done
>>>>>>     again:
>>>>>>             lock(mmap_sem)
>>>>>>             vma = find_vma();
>>>>>>             handle_pte_fault(vma);
>>>>>>             if retry
>>>>>>                unlock(mmap_sem)
>>>>>>                goto again;
>>>>>>     done:
>>>>>>             handle fault error
>>>>>>     }
>>>>>>
>>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>>
>>>>>> This series add a new software performance event named 'speculative-faults'
>>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>>> the part of the faults processed speculatively.
>>>>>>
>>>>>> There are some trace events introduced by this series. They allow
>>>>>> identifying why the page faults were not processed speculatively. This
>>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>>> grouped in a system named 'pagefault', they are:
>>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>>    back.
>>>>>>
>>>>>> To record all the related events, the easier is to run perf with the
>>>>>> following arguments :
>>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>>
>>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>>> page fault handled speculatively. I can be seen this way:
>>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>>
>>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>>> on x86, PowerPC and arm64.
>>>>>>
>>>>>> ---------------------
>>>>>> Real Workload results
>>>>>>
>>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>>> this new version should not impact the performance boost seen.
>>>>>>
>>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>>> series:
>>>>>>                 vanilla         spf
>>>>>> faults          89.418          101.364         +13%
>>>>>> spf                n/a           97.989
>>>>>>
>>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>>> way.
>>>>>>
>>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>>> it a try on an android device. He reported that the application launch time
>>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>>> 20%.
>>>>>>
>>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>>
>>>>>> Application                             4.9     4.9+spf delta
>>>>>> com.tencent.mm                          416     389     -7%
>>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>>> com.tencent.mtt                         455     454     0%
>>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>>> com.autonavi.minimap                    711     701     -1%
>>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>>> com.immomo.momo                         501     487     -3%
>>>>>> com.tencent.peng                        2145    2112    -2%
>>>>>> com.smile.gifmaker                      491     461     -6%
>>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>>> com.baidu.searchbox                     333     314     -6%
>>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>>> com.sina.weibo                          907     906     0%
>>>>>> com.youku.phone                         816     731     -11%
>>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>>> com.UCMobile                            415     411     -1%
>>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>>
>>>>>> ------------------
>>>>>> Benchmarks results
>>>>>>
>>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>>> SPF is BASE + this series
>>>>>>
>>>>>> Kernbench:
>>>>>> ----------
>>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>>> kernel (kernel is build 5 times):
>>>>>>
>>>>>> Average Half load -j 8
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>>
>>>>>> Average Optimal load -j 16
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>>
>>>>>>
>>>>>> During a run on the SPF, perf events were captured:
>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>          526743764      faults
>>>>>>                210      spf
>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>               2278      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> Very few speculative page faults were recorded as most of the processes
>>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>>> were created during the kernel build processing).
>>>>>>
>>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>>
>>>>>> Average Half load -j 40
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>>
>>>>>> Average Optimal load -j 80
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>>
>>>>>> During a run on the SPF, perf events were captured:
>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>          116730856      faults
>>>>>>                  0      spf
>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>                476      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>>> there is no impact on the performance.
>>>>>>
>>>>>> Ebizzy:
>>>>>> -------
>>>>>> The test is counting the number of records per second it can manage, the
>>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>>> consistent result I repeated the test 100 times and measure the average
>>>>>> result. The number is the record processes per second, the higher is the
>>>>>> best.
>>>>>>
>>>>>>                 BASE            SPF             delta
>>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>>
>>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>>            1706379      faults
>>>>>>            1674599      spf
>>>>>>              30588      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>                363      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>>            1874773      faults
>>>>>>            1461153      spf
>>>>>>             413293      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>                200      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>>> leading the ebizzy performance boost.
>>>>>>
>>>>>> ------------------
>>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>>    and Minchan Kim, hopefully.
>>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>>    __do_page_fault().
>>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>>    instead
>>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>>> useless
>>>>>>    trace event pagefault:spf_pte_lock.
>>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>>
>>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>>
>>>>>>
>>>>>> Laurent Dufour (20):
>>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>>   mm: introduce INIT_VMA()
>>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>>   mm: protect mremap() against SPF hanlder
>>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>>   mm: introduce __vm_normal_page()
>>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>>   mm: adding speculative page fault failure trace events
>>>>>>   perf: add a speculative page fault sw event
>>>>>>   perf tools: add support for the SPF perf event
>>>>>>   mm: add speculative page fault vmstats
>>>>>>   powerpc/mm: add speculative page fault
>>>>>>
>>>>>> Mahendran Ganesh (2):
>>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>   arm64/mm: add speculative page fault
>>>>>>
>>>>>> Peter Zijlstra (4):
>>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>>   mm: VMA sequence count
>>>>>>   mm: provide speculative fault infrastructure
>>>>>>   x86/mm: add speculative pagefault handling
>>>>>>
>>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>>  fs/exec.c                             |   2 +-
>>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>>  include/linux/migrate.h               |   4 +-
>>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>>  include/linux/mm_types.h              |   7 +
>>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>>  include/linux/swap.h                  |  10 +-
>>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>>  kernel/fork.c                         |   5 +-
>>>>>>  mm/Kconfig                            |  22 ++
>>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>>  mm/hugetlb.c                          |   2 +
>>>>>>  mm/init-mm.c                          |   3 +
>>>>>>  mm/internal.h                         |  20 ++
>>>>>>  mm/khugepaged.c                       |   5 +
>>>>>>  mm/madvise.c                          |   6 +-
>>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>>  mm/migrate.c                          |   6 +-
>>>>>>  mm/mlock.c                            |  13 +-
>>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>>  mm/mprotect.c                         |   4 +-
>>>>>>  mm/mremap.c                           |  13 +
>>>>>>  mm/nommu.c                            |   2 +-
>>>>>>  mm/rmap.c                             |   5 +-
>>>>>>  mm/swap.c                             |   6 +-
>>>>>>  mm/swap_state.c                       |   8 +-
>>>>>>  mm/vmstat.c                           |   5 +-
>>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>>  tools/perf/util/python.c              |   1 +
>>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>>
>>>>>> --
>>>>>> 2.7.4
>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>>
> 

RE: [PATCH v11 00/26] Speculative page faults

[Song, HaiyanX]

[-- Attachment #1: Type: text/plain, Size: 37281 bytes --]

Hi Laurent,

I attached the perf-profile.gz file for case page_fault2 and page_fault3. These files were captured during test the related test case. 
Please help to check on these data if it can help you to find the higher change. Thanks.

File name perf-profile_page_fault2_head_THP-Always.gz, means the perf-profile result get from page_fault2 
    tested for head commit (a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12) with THP_always configuration.

Best regards,
Haiyan Song

________________________________________
From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
Sent: Thursday, July 12, 2018 1:05 AM
To: Song, HaiyanX
Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
Subject: Re: [PATCH v11 00/26] Speculative page faults

Hi Haiyan,

Do you get a chance to capture some performance cycles on your system ?
I still can't get these numbers on my hardware.

Thanks,
Laurent.

On 04/07/2018 09:51, Laurent Dufour wrote:
> On 04/07/2018 05:23, Song, HaiyanX wrote:
>> Hi Laurent,
>>
>>
>> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
>> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.
>
> Repeating the test only 3 times seems a bit too low to me.
>
> I'll focus on the higher change for the moment, but I don't have access to such
> a hardware.
>
> Is possible to provide a diff between base and SPF of the performance cycles
> measured when running page_fault3 and page_fault2 when the 20% change is detected.
>
> Please stay focus on the test case process to see exactly where the series is
> impacting.
>
> Thanks,
> Laurent.
>
>>
>> And I did not find other high variation on test case result.
>>
>> a). Enable THP
>> testcase                          base     stddev       change      head     stddev         metric
>> page_fault3/enable THP           10519      ± 3%        -20.5%      8368      ±6%          will-it-scale.per_thread_ops
>> page_fault2/enalbe THP            8281      ± 2%        -18.8%      6728                   will-it-scale.per_thread_ops
>> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
>> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
>> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
>>
>> b). Disable THP
>> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
>> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
>> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
>> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
>> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
>> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
>> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
>>
>>
>> Best regards,
>> Haiyan Song
>> ________________________________________
>> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
>> Sent: Monday, July 02, 2018 4:59 PM
>> To: Song, HaiyanX
>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>
>> On 11/06/2018 09:49, Song, HaiyanX wrote:
>>> Hi Laurent,
>>>
>>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>>> V9 patch serials.
>>>
>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>>> commit id:
>>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>>> Benchmark: will-it-scale
>>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>>
>>> Metrics:
>>>   will-it-scale.per_process_ops=processes/nr_cpu
>>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>> THP: enable / disable
>>> nr_task:100%
>>>
>>> 1. Regressions:
>>>
>>> a). Enable THP
>>> testcase                          base           change      head           metric
>>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>>
>>> b). Disable THP
>>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>>
>>> Notes: for the above  values of test result, the higher is better.
>>
>> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
>> get reproducible results. The results have huge variation, even on the vanilla
>> kernel, and I can't state on any changes due to that.
>>
>> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
>> measure any changes between the vanilla and the SPF patched ones:
>>
>> test THP enabled                4.17.0-rc4-mm1  spf             delta
>> page_fault3_threads             2697.7          2683.5          -0.53%
>> page_fault2_threads             170660.6        169574.1        -0.64%
>> context_switch1_threads         6915269.2       6877507.3       -0.55%
>> context_switch1_processes       6478076.2       6529493.5       0.79%
>> brk1                            243391.2        238527.5        -2.00%
>>
>> Tests were run 10 times, no high variation detected.
>>
>> Did you see high variation on your side ? How many times the test were run to
>> compute the average values ?
>>
>> Thanks,
>> Laurent.
>>
>>
>>>
>>> 2. Improvement: not found improvement based on the selected test cases.
>>>
>>>
>>> Best regards
>>> Haiyan Song
>>> ________________________________________
>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Monday, May 28, 2018 4:54 PM
>>> To: Song, HaiyanX
>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>
>>> On 28/05/2018 10:22, Haiyan Song wrote:
>>>> Hi Laurent,
>>>>
>>>> Yes, these tests are done on V9 patch.
>>>
>>> Do you plan to give this V11 a run ?
>>>
>>>>
>>>>
>>>> Best regards,
>>>> Haiyan Song
>>>>
>>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>>
>>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>>> tested on Intel 4s Skylake platform.
>>>>>
>>>>> Hi,
>>>>>
>>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>>> series" while responding to the v11 header series...
>>>>> Were these tests done on v9 or v11 ?
>>>>>
>>>>> Cheers,
>>>>> Laurent.
>>>>>
>>>>>>
>>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>>> Commit id:
>>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>>> Benchmark suite: will-it-scale
>>>>>> Download link:
>>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>>> Metrics:
>>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>>> THP: enable / disable
>>>>>> nr_task: 100%
>>>>>>
>>>>>> 1. Regressions:
>>>>>> a) THP enabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>>
>>>>>> b) THP disabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>>
>>>>>> 2. Improvements:
>>>>>> a) THP enabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>>
>>>>>> b) THP disabled:
>>>>>> testcase                        base            change          head       metric
>>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>>
>>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>>
>>>>>>
>>>>>> Best regards
>>>>>> Haiyan Song
>>>>>>
>>>>>> ________________________________________
>>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>>
>>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>>> page fault without holding the mm semaphore [1].
>>>>>>
>>>>>> The idea is to try to handle user space page faults without holding the
>>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>>> process since the page fault handler will not wait for other threads memory
>>>>>> layout change to be done, assuming that this change is done in another part
>>>>>> of the process's memory space. This type page fault is named speculative
>>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>>> is failing its processing and a classic page fault is then tried.
>>>>>>
>>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>>> limiting the locking contention to these operations which are expected to
>>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>>> benchmark anymore.
>>>>>>
>>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>>> have to be protected against parallel changes. This is done by using a per
>>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>>> speculative page fault in that case.
>>>>>>
>>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>>> checked during the page fault are modified.
>>>>>>
>>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>>> parallel change is possible at this time.
>>>>>>
>>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>>> classic page fault handler will be called to handle the operation while
>>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>>> PTE.
>>>>>>
>>>>>> In pseudo code, this could be seen as:
>>>>>>     speculative_page_fault()
>>>>>>     {
>>>>>>             vma = get_vma()
>>>>>>             check vma sequence count
>>>>>>             check vma's support
>>>>>>             disable interrupt
>>>>>>                   check pgd,p4d,...,pte
>>>>>>                   save pmd and pte in vmf
>>>>>>                   save vma sequence counter in vmf
>>>>>>             enable interrupt
>>>>>>             check vma sequence count
>>>>>>             handle_pte_fault(vma)
>>>>>>                     ..
>>>>>>                     page = alloc_page()
>>>>>>                     pte_map_lock()
>>>>>>                             disable interrupt
>>>>>>                                     abort if sequence counter has changed
>>>>>>                                     abort if pmd or pte has changed
>>>>>>                                     pte map and lock
>>>>>>                             enable interrupt
>>>>>>                     if abort
>>>>>>                        free page
>>>>>>                        abort
>>>>>>                     ...
>>>>>>     }
>>>>>>
>>>>>>     arch_fault_handler()
>>>>>>     {
>>>>>>             if (speculative_page_fault(&vma))
>>>>>>                goto done
>>>>>>     again:
>>>>>>             lock(mmap_sem)
>>>>>>             vma = find_vma();
>>>>>>             handle_pte_fault(vma);
>>>>>>             if retry
>>>>>>                unlock(mmap_sem)
>>>>>>                goto again;
>>>>>>     done:
>>>>>>             handle fault error
>>>>>>     }
>>>>>>
>>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>>
>>>>>> This series add a new software performance event named 'speculative-faults'
>>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>>> the part of the faults processed speculatively.
>>>>>>
>>>>>> There are some trace events introduced by this series. They allow
>>>>>> identifying why the page faults were not processed speculatively. This
>>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>>> grouped in a system named 'pagefault', they are:
>>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>>    back.
>>>>>>
>>>>>> To record all the related events, the easier is to run perf with the
>>>>>> following arguments :
>>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>>
>>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>>> page fault handled speculatively. I can be seen this way:
>>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>>
>>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>>> on x86, PowerPC and arm64.
>>>>>>
>>>>>> ---------------------
>>>>>> Real Workload results
>>>>>>
>>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>>> this new version should not impact the performance boost seen.
>>>>>>
>>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>>> series:
>>>>>>                 vanilla         spf
>>>>>> faults          89.418          101.364         +13%
>>>>>> spf                n/a           97.989
>>>>>>
>>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>>> way.
>>>>>>
>>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>>> it a try on an android device. He reported that the application launch time
>>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>>> 20%.
>>>>>>
>>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>>
>>>>>> Application                             4.9     4.9+spf delta
>>>>>> com.tencent.mm                          416     389     -7%
>>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>>> com.tencent.mtt                         455     454     0%
>>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>>> com.autonavi.minimap                    711     701     -1%
>>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>>> com.immomo.momo                         501     487     -3%
>>>>>> com.tencent.peng                        2145    2112    -2%
>>>>>> com.smile.gifmaker                      491     461     -6%
>>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>>> com.baidu.searchbox                     333     314     -6%
>>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>>> com.sina.weibo                          907     906     0%
>>>>>> com.youku.phone                         816     731     -11%
>>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>>> com.UCMobile                            415     411     -1%
>>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>>
>>>>>> ------------------
>>>>>> Benchmarks results
>>>>>>
>>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>>> SPF is BASE + this series
>>>>>>
>>>>>> Kernbench:
>>>>>> ----------
>>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>>> kernel (kernel is build 5 times):
>>>>>>
>>>>>> Average Half load -j 8
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>>
>>>>>> Average Optimal load -j 16
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>>
>>>>>>
>>>>>> During a run on the SPF, perf events were captured:
>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>          526743764      faults
>>>>>>                210      spf
>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>               2278      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> Very few speculative page faults were recorded as most of the processes
>>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>>> were created during the kernel build processing).
>>>>>>
>>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>>
>>>>>> Average Half load -j 40
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>>
>>>>>> Average Optimal load -j 80
>>>>>>                  Run    (std deviation)
>>>>>>                  BASE                   SPF
>>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>>
>>>>>> During a run on the SPF, perf events were captured:
>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>          116730856      faults
>>>>>>                  0      spf
>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>                476      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>>> there is no impact on the performance.
>>>>>>
>>>>>> Ebizzy:
>>>>>> -------
>>>>>> The test is counting the number of records per second it can manage, the
>>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>>> consistent result I repeated the test 100 times and measure the average
>>>>>> result. The number is the record processes per second, the higher is the
>>>>>> best.
>>>>>>
>>>>>>                 BASE            SPF             delta
>>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>>
>>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>>            1706379      faults
>>>>>>            1674599      spf
>>>>>>              30588      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>                363      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>>            1874773      faults
>>>>>>            1461153      spf
>>>>>>             413293      pagefault:spf_vma_changed
>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>                200      pagefault:spf_vma_notsup
>>>>>>                  0      pagefault:spf_vma_access
>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>
>>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>>> leading the ebizzy performance boost.
>>>>>>
>>>>>> ------------------
>>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>>    and Minchan Kim, hopefully.
>>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>>    __do_page_fault().
>>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>>    instead
>>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>>> useless
>>>>>>    trace event pagefault:spf_pte_lock.
>>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>>
>>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>>
>>>>>>
>>>>>> Laurent Dufour (20):
>>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>>   mm: introduce INIT_VMA()
>>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>>   mm: protect mremap() against SPF hanlder
>>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>>   mm: introduce __vm_normal_page()
>>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>>   mm: adding speculative page fault failure trace events
>>>>>>   perf: add a speculative page fault sw event
>>>>>>   perf tools: add support for the SPF perf event
>>>>>>   mm: add speculative page fault vmstats
>>>>>>   powerpc/mm: add speculative page fault
>>>>>>
>>>>>> Mahendran Ganesh (2):
>>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>   arm64/mm: add speculative page fault
>>>>>>
>>>>>> Peter Zijlstra (4):
>>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>>   mm: VMA sequence count
>>>>>>   mm: provide speculative fault infrastructure
>>>>>>   x86/mm: add speculative pagefault handling
>>>>>>
>>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>>  fs/exec.c                             |   2 +-
>>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>>  include/linux/migrate.h               |   4 +-
>>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>>  include/linux/mm_types.h              |   7 +
>>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>>  include/linux/swap.h                  |  10 +-
>>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>>  kernel/fork.c                         |   5 +-
>>>>>>  mm/Kconfig                            |  22 ++
>>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>>  mm/hugetlb.c                          |   2 +
>>>>>>  mm/init-mm.c                          |   3 +
>>>>>>  mm/internal.h                         |  20 ++
>>>>>>  mm/khugepaged.c                       |   5 +
>>>>>>  mm/madvise.c                          |   6 +-
>>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>>  mm/migrate.c                          |   6 +-
>>>>>>  mm/mlock.c                            |  13 +-
>>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>>  mm/mprotect.c                         |   4 +-
>>>>>>  mm/mremap.c                           |  13 +
>>>>>>  mm/nommu.c                            |   2 +-
>>>>>>  mm/rmap.c                             |   5 +-
>>>>>>  mm/swap.c                             |   6 +-
>>>>>>  mm/swap_state.c                       |   8 +-
>>>>>>  mm/vmstat.c                           |   5 +-
>>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>>  tools/perf/util/python.c              |   1 +
>>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>>
>>>>>> --
>>>>>> 2.7.4
>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>
>>
>


[-- Attachment #2: perf-profile_page_fault2_base_THP-Alwasys.gz --]
[-- Type: application/gzip, Size: 10171 bytes --]

[-- Attachment #3: perf-profile_page_fault2_base_thp_never.gz --]
[-- Type: application/gzip, Size: 11474 bytes --]

[-- Attachment #4: perf-profile_page_fault2_head_THP-Always.gz --]
[-- Type: application/gzip, Size: 10374 bytes --]

[-- Attachment #5: perf-profile_page_fault2_head_thp_never.gz --]
[-- Type: application/gzip, Size: 11327 bytes --]

[-- Attachment #6: perf-profile_page_fault3_base_THP-Always.gz --]
[-- Type: application/gzip, Size: 9503 bytes --]

[-- Attachment #7: perf-profile_page_fault3_base_thp_never.gz --]
[-- Type: application/gzip, Size: 9843 bytes --]

[-- Attachment #8: perf-profile_page_fault3_head_THP-Always.gz --]
[-- Type: application/gzip, Size: 9596 bytes --]

[-- Attachment #9: perf-profile_page_fault3_head_thp_never.gz --]
[-- Type: application/gzip, Size: 10137 bytes --]

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

On 13/07/2018 05:56, Song, HaiyanX wrote:
> Hi Laurent,

Hi Haiyan,

Thanks a lot for sharing this perf reports.

I looked at them closely, and I've to admit that I was not able to found a
major difference between the base and the head report, except that
handle_pte_fault() is no more in-lined in the head one.

As expected, __handle_speculative_fault() is never traced since these tests are
dealing with file mapping, not handled in the speculative way.

When running these test did you seen a major differences in the test's result
between base and head ?

>From the number of cycles counted, the biggest difference is page_fault3 when
run with the THP enabled:
				BASE		HEAD		Delta
page_fault2_base_thp_never	1142252426747	1065866197589	-6.69%
page_fault2_base_THP-Alwasys	1124844374523	1076312228927	-4.31%
page_fault3_base_thp_never	1099387298152	1134118402345	3.16%
page_fault3_base_THP-Always	1059370178101	853985561949	-19.39%


The very weird thing is the difference of the delta cycles reported between
thp never and thp always, because the speculative way is aborted when checking
for the vma->ops field, which is the same in both case, and the thp is never
checked. So there is no code covering differnce, on the speculative path,
between these 2 cases. This leads me to think that there are other interactions
interfering in the measure.

Looking at the perf-profile_page_fault3_*_THP-Always, the major differences at
the head of the perf report is the 92% testcase which is weirdly not reported
on the head side :
    92.02%    22.33%  page_fault3_processes  [.] testcase
92.02% testcase

Then the base reported 37.67% for __do_page_fault() where the head reported
48.41%, but the only difference in this function, between base and head, is the
call to handle_speculative_fault(). But this is a macro checking for the fault
flags, and mm->users and then calling __handle_speculative_fault() if needed.
So this can't explain this difference, except if __handle_speculative_fault()
is inlined in __do_page_fault().
Is this the case on your build ?

Haiyan, do you still have the output of the test to check those numbers too ?

Cheers,
Laurent

> I attached the perf-profile.gz file for case page_fault2 and page_fault3. These files were captured during test the related test case. 
> Please help to check on these data if it can help you to find the higher change. Thanks.
> 
> File name perf-profile_page_fault2_head_THP-Always.gz, means the perf-profile result get from page_fault2 
>     tested for head commit (a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12) with THP_always configuration.
> 
> Best regards,
> Haiyan Song
> 
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Thursday, July 12, 2018 1:05 AM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
> 
> Hi Haiyan,
> 
> Do you get a chance to capture some performance cycles on your system ?
> I still can't get these numbers on my hardware.
> 
> Thanks,
> Laurent.
> 
> On 04/07/2018 09:51, Laurent Dufour wrote:
>> On 04/07/2018 05:23, Song, HaiyanX wrote:
>>> Hi Laurent,
>>>
>>>
>>> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
>>> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.
>>
>> Repeating the test only 3 times seems a bit too low to me.
>>
>> I'll focus on the higher change for the moment, but I don't have access to such
>> a hardware.
>>
>> Is possible to provide a diff between base and SPF of the performance cycles
>> measured when running page_fault3 and page_fault2 when the 20% change is detected.
>>
>> Please stay focus on the test case process to see exactly where the series is
>> impacting.
>>
>> Thanks,
>> Laurent.
>>
>>>
>>> And I did not find other high variation on test case result.
>>>
>>> a). Enable THP
>>> testcase                          base     stddev       change      head     stddev         metric
>>> page_fault3/enable THP           10519      +- 3%        -20.5%      8368      +-6%          will-it-scale.per_thread_ops
>>> page_fault2/enalbe THP            8281      +- 2%        -18.8%      6728                   will-it-scale.per_thread_ops
>>> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
>>> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
>>> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
>>>
>>> b). Disable THP
>>> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
>>> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
>>> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
>>> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
>>> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
>>> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
>>> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
>>>
>>>
>>> Best regards,
>>> Haiyan Song
>>> ________________________________________
>>> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Monday, July 02, 2018 4:59 PM
>>> To: Song, HaiyanX
>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>
>>> On 11/06/2018 09:49, Song, HaiyanX wrote:
>>>> Hi Laurent,
>>>>
>>>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>>>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>>>> V9 patch serials.
>>>>
>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>>>> commit id:
>>>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>>>> Benchmark: will-it-scale
>>>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>>>
>>>> Metrics:
>>>>   will-it-scale.per_process_ops=processes/nr_cpu
>>>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>> THP: enable / disable
>>>> nr_task:100%
>>>>
>>>> 1. Regressions:
>>>>
>>>> a). Enable THP
>>>> testcase                          base           change      head           metric
>>>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>>>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>>>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>>>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>>>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>>>
>>>> b). Disable THP
>>>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>>>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>>>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>>>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>>>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>>>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>>>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>>>
>>>> Notes: for the above  values of test result, the higher is better.
>>>
>>> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
>>> get reproducible results. The results have huge variation, even on the vanilla
>>> kernel, and I can't state on any changes due to that.
>>>
>>> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
>>> measure any changes between the vanilla and the SPF patched ones:
>>>
>>> test THP enabled                4.17.0-rc4-mm1  spf             delta
>>> page_fault3_threads             2697.7          2683.5          -0.53%
>>> page_fault2_threads             170660.6        169574.1        -0.64%
>>> context_switch1_threads         6915269.2       6877507.3       -0.55%
>>> context_switch1_processes       6478076.2       6529493.5       0.79%
>>> brk1                            243391.2        238527.5        -2.00%
>>>
>>> Tests were run 10 times, no high variation detected.
>>>
>>> Did you see high variation on your side ? How many times the test were run to
>>> compute the average values ?
>>>
>>> Thanks,
>>> Laurent.
>>>
>>>
>>>>
>>>> 2. Improvement: not found improvement based on the selected test cases.
>>>>
>>>>
>>>> Best regards
>>>> Haiyan Song
>>>> ________________________________________
>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>> Sent: Monday, May 28, 2018 4:54 PM
>>>> To: Song, HaiyanX
>>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>>
>>>> On 28/05/2018 10:22, Haiyan Song wrote:
>>>>> Hi Laurent,
>>>>>
>>>>> Yes, these tests are done on V9 patch.
>>>>
>>>> Do you plan to give this V11 a run ?
>>>>
>>>>>
>>>>>
>>>>> Best regards,
>>>>> Haiyan Song
>>>>>
>>>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>>>
>>>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>>>> tested on Intel 4s Skylake platform.
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>>>> series" while responding to the v11 header series...
>>>>>> Were these tests done on v9 or v11 ?
>>>>>>
>>>>>> Cheers,
>>>>>> Laurent.
>>>>>>
>>>>>>>
>>>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>>>> Commit id:
>>>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>>>> Benchmark suite: will-it-scale
>>>>>>> Download link:
>>>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>>>> Metrics:
>>>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>>>> THP: enable / disable
>>>>>>> nr_task: 100%
>>>>>>>
>>>>>>> 1. Regressions:
>>>>>>> a) THP enabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>>>
>>>>>>> b) THP disabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>>>
>>>>>>> 2. Improvements:
>>>>>>> a) THP enabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>>>
>>>>>>> b) THP disabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>>>
>>>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>>>
>>>>>>>
>>>>>>> Best regards
>>>>>>> Haiyan Song
>>>>>>>
>>>>>>> ________________________________________
>>>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>>>
>>>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>>>> page fault without holding the mm semaphore [1].
>>>>>>>
>>>>>>> The idea is to try to handle user space page faults without holding the
>>>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>>>> process since the page fault handler will not wait for other threads memory
>>>>>>> layout change to be done, assuming that this change is done in another part
>>>>>>> of the process's memory space. This type page fault is named speculative
>>>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>>>> is failing its processing and a classic page fault is then tried.
>>>>>>>
>>>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>>>> limiting the locking contention to these operations which are expected to
>>>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>>>> benchmark anymore.
>>>>>>>
>>>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>>>> have to be protected against parallel changes. This is done by using a per
>>>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>>>> speculative page fault in that case.
>>>>>>>
>>>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>>>> checked during the page fault are modified.
>>>>>>>
>>>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>>>> parallel change is possible at this time.
>>>>>>>
>>>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>>>> classic page fault handler will be called to handle the operation while
>>>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>>>> PTE.
>>>>>>>
>>>>>>> In pseudo code, this could be seen as:
>>>>>>>     speculative_page_fault()
>>>>>>>     {
>>>>>>>             vma = get_vma()
>>>>>>>             check vma sequence count
>>>>>>>             check vma's support
>>>>>>>             disable interrupt
>>>>>>>                   check pgd,p4d,...,pte
>>>>>>>                   save pmd and pte in vmf
>>>>>>>                   save vma sequence counter in vmf
>>>>>>>             enable interrupt
>>>>>>>             check vma sequence count
>>>>>>>             handle_pte_fault(vma)
>>>>>>>                     ..
>>>>>>>                     page = alloc_page()
>>>>>>>                     pte_map_lock()
>>>>>>>                             disable interrupt
>>>>>>>                                     abort if sequence counter has changed
>>>>>>>                                     abort if pmd or pte has changed
>>>>>>>                                     pte map and lock
>>>>>>>                             enable interrupt
>>>>>>>                     if abort
>>>>>>>                        free page
>>>>>>>                        abort
>>>>>>>                     ...
>>>>>>>     }
>>>>>>>
>>>>>>>     arch_fault_handler()
>>>>>>>     {
>>>>>>>             if (speculative_page_fault(&vma))
>>>>>>>                goto done
>>>>>>>     again:
>>>>>>>             lock(mmap_sem)
>>>>>>>             vma = find_vma();
>>>>>>>             handle_pte_fault(vma);
>>>>>>>             if retry
>>>>>>>                unlock(mmap_sem)
>>>>>>>                goto again;
>>>>>>>     done:
>>>>>>>             handle fault error
>>>>>>>     }
>>>>>>>
>>>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>>>
>>>>>>> This series add a new software performance event named 'speculative-faults'
>>>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>>>> the part of the faults processed speculatively.
>>>>>>>
>>>>>>> There are some trace events introduced by this series. They allow
>>>>>>> identifying why the page faults were not processed speculatively. This
>>>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>>>> grouped in a system named 'pagefault', they are:
>>>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>>>    back.
>>>>>>>
>>>>>>> To record all the related events, the easier is to run perf with the
>>>>>>> following arguments :
>>>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>>>
>>>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>>>> page fault handled speculatively. I can be seen this way:
>>>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>>>
>>>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>>>> on x86, PowerPC and arm64.
>>>>>>>
>>>>>>> ---------------------
>>>>>>> Real Workload results
>>>>>>>
>>>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>>>> this new version should not impact the performance boost seen.
>>>>>>>
>>>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>>>> series:
>>>>>>>                 vanilla         spf
>>>>>>> faults          89.418          101.364         +13%
>>>>>>> spf                n/a           97.989
>>>>>>>
>>>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>>>> way.
>>>>>>>
>>>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>>>> it a try on an android device. He reported that the application launch time
>>>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>>>> 20%.
>>>>>>>
>>>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>>>
>>>>>>> Application                             4.9     4.9+spf delta
>>>>>>> com.tencent.mm                          416     389     -7%
>>>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>>>> com.tencent.mtt                         455     454     0%
>>>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>>>> com.autonavi.minimap                    711     701     -1%
>>>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>>>> com.immomo.momo                         501     487     -3%
>>>>>>> com.tencent.peng                        2145    2112    -2%
>>>>>>> com.smile.gifmaker                      491     461     -6%
>>>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>>>> com.baidu.searchbox                     333     314     -6%
>>>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>>>> com.sina.weibo                          907     906     0%
>>>>>>> com.youku.phone                         816     731     -11%
>>>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>>>> com.UCMobile                            415     411     -1%
>>>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>>>
>>>>>>> ------------------
>>>>>>> Benchmarks results
>>>>>>>
>>>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>>>> SPF is BASE + this series
>>>>>>>
>>>>>>> Kernbench:
>>>>>>> ----------
>>>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>>>> kernel (kernel is build 5 times):
>>>>>>>
>>>>>>> Average Half load -j 8
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>>>
>>>>>>> Average Optimal load -j 16
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>>>
>>>>>>>
>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>          526743764      faults
>>>>>>>                210      spf
>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>               2278      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> Very few speculative page faults were recorded as most of the processes
>>>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>>>> were created during the kernel build processing).
>>>>>>>
>>>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>>>
>>>>>>> Average Half load -j 40
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>>>
>>>>>>> Average Optimal load -j 80
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>>>
>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>          116730856      faults
>>>>>>>                  0      spf
>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                476      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>>>> there is no impact on the performance.
>>>>>>>
>>>>>>> Ebizzy:
>>>>>>> -------
>>>>>>> The test is counting the number of records per second it can manage, the
>>>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>>>> consistent result I repeated the test 100 times and measure the average
>>>>>>> result. The number is the record processes per second, the higher is the
>>>>>>> best.
>>>>>>>
>>>>>>>                 BASE            SPF             delta
>>>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>>>
>>>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>>>            1706379      faults
>>>>>>>            1674599      spf
>>>>>>>              30588      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                363      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>>>            1874773      faults
>>>>>>>            1461153      spf
>>>>>>>             413293      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                200      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>>>> leading the ebizzy performance boost.
>>>>>>>
>>>>>>> ------------------
>>>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>>>    and Minchan Kim, hopefully.
>>>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>>>    __do_page_fault().
>>>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>>>    instead
>>>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>>>> useless
>>>>>>>    trace event pagefault:spf_pte_lock.
>>>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>>>
>>>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>>>
>>>>>>>
>>>>>>> Laurent Dufour (20):
>>>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>>>   mm: introduce INIT_VMA()
>>>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>>>   mm: protect mremap() against SPF hanlder
>>>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>>>   mm: introduce __vm_normal_page()
>>>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>>>   mm: adding speculative page fault failure trace events
>>>>>>>   perf: add a speculative page fault sw event
>>>>>>>   perf tools: add support for the SPF perf event
>>>>>>>   mm: add speculative page fault vmstats
>>>>>>>   powerpc/mm: add speculative page fault
>>>>>>>
>>>>>>> Mahendran Ganesh (2):
>>>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   arm64/mm: add speculative page fault
>>>>>>>
>>>>>>> Peter Zijlstra (4):
>>>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>>>   mm: VMA sequence count
>>>>>>>   mm: provide speculative fault infrastructure
>>>>>>>   x86/mm: add speculative pagefault handling
>>>>>>>
>>>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>>>  fs/exec.c                             |   2 +-
>>>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>>>  include/linux/migrate.h               |   4 +-
>>>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>>>  include/linux/mm_types.h              |   7 +
>>>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>>>  include/linux/swap.h                  |  10 +-
>>>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>>>  kernel/fork.c                         |   5 +-
>>>>>>>  mm/Kconfig                            |  22 ++
>>>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>>>  mm/hugetlb.c                          |   2 +
>>>>>>>  mm/init-mm.c                          |   3 +
>>>>>>>  mm/internal.h                         |  20 ++
>>>>>>>  mm/khugepaged.c                       |   5 +
>>>>>>>  mm/madvise.c                          |   6 +-
>>>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>>>  mm/migrate.c                          |   6 +-
>>>>>>>  mm/mlock.c                            |  13 +-
>>>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>>>  mm/mprotect.c                         |   4 +-
>>>>>>>  mm/mremap.c                           |  13 +
>>>>>>>  mm/nommu.c                            |   2 +-
>>>>>>>  mm/rmap.c                             |   5 +-
>>>>>>>  mm/swap.c                             |   6 +-
>>>>>>>  mm/swap_state.c                       |   8 +-
>>>>>>>  mm/vmstat.c                           |   5 +-
>>>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>>>  tools/perf/util/python.c              |   1 +
>>>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>>>
>>>>>>> --
>>>>>>> 2.7.4
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>>
>>
> 

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

On 13/07/2018 05:56, Song, HaiyanX wrote:
> Hi Laurent,

Hi Haiyan,

Thanks a lot for sharing this perf reports.

I looked at them closely, and I've to admit that I was not able to found a
major difference between the base and the head report, except that
handle_pte_fault() is no more in-lined in the head one.

As expected, __handle_speculative_fault() is never traced since these tests are
dealing with file mapping, not handled in the speculative way.

When running these test did you seen a major differences in the test's result
between base and head ?

>From the number of cycles counted, the biggest difference is page_fault3 when
run with the THP enabled:
				BASE		HEAD		Delta
page_fault2_base_thp_never	1142252426747	1065866197589	-6.69%
page_fault2_base_THP-Alwasys	1124844374523	1076312228927	-4.31%
page_fault3_base_thp_never	1099387298152	1134118402345	3.16%
page_fault3_base_THP-Always	1059370178101	853985561949	-19.39%


The very weird thing is the difference of the delta cycles reported between
thp never and thp always, because the speculative way is aborted when checking
for the vma->ops field, which is the same in both case, and the thp is never
checked. So there is no code covering differnce, on the speculative path,
between these 2 cases. This leads me to think that there are other interactions
interfering in the measure.

Looking at the perf-profile_page_fault3_*_THP-Always, the major differences at
the head of the perf report is the 92% testcase which is weirdly not reported
on the head side :
    92.02%    22.33%  page_fault3_processes  [.] testcase
92.02% testcase

Then the base reported 37.67% for __do_page_fault() where the head reported
48.41%, but the only difference in this function, between base and head, is the
call to handle_speculative_fault(). But this is a macro checking for the fault
flags, and mm->users and then calling __handle_speculative_fault() if needed.
So this can't explain this difference, except if __handle_speculative_fault()
is inlined in __do_page_fault().
Is this the case on your build ?

Haiyan, do you still have the output of the test to check those numbers too ?

Cheers,
Laurent

> I attached the perf-profile.gz file for case page_fault2 and page_fault3. These files were captured during test the related test case. 
> Please help to check on these data if it can help you to find the higher change. Thanks.
> 
> File name perf-profile_page_fault2_head_THP-Always.gz, means the perf-profile result get from page_fault2 
>     tested for head commit (a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12) with THP_always configuration.
> 
> Best regards,
> Haiyan Song
> 
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Thursday, July 12, 2018 1:05 AM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
> 
> Hi Haiyan,
> 
> Do you get a chance to capture some performance cycles on your system ?
> I still can't get these numbers on my hardware.
> 
> Thanks,
> Laurent.
> 
> On 04/07/2018 09:51, Laurent Dufour wrote:
>> On 04/07/2018 05:23, Song, HaiyanX wrote:
>>> Hi Laurent,
>>>
>>>
>>> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
>>> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.
>>
>> Repeating the test only 3 times seems a bit too low to me.
>>
>> I'll focus on the higher change for the moment, but I don't have access to such
>> a hardware.
>>
>> Is possible to provide a diff between base and SPF of the performance cycles
>> measured when running page_fault3 and page_fault2 when the 20% change is detected.
>>
>> Please stay focus on the test case process to see exactly where the series is
>> impacting.
>>
>> Thanks,
>> Laurent.
>>
>>>
>>> And I did not find other high variation on test case result.
>>>
>>> a). Enable THP
>>> testcase                          base     stddev       change      head     stddev         metric
>>> page_fault3/enable THP           10519      ± 3%        -20.5%      8368      ±6%          will-it-scale.per_thread_ops
>>> page_fault2/enalbe THP            8281      ± 2%        -18.8%      6728                   will-it-scale.per_thread_ops
>>> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
>>> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
>>> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
>>>
>>> b). Disable THP
>>> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
>>> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
>>> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
>>> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
>>> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
>>> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
>>> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
>>>
>>>
>>> Best regards,
>>> Haiyan Song
>>> ________________________________________
>>> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Monday, July 02, 2018 4:59 PM
>>> To: Song, HaiyanX
>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>
>>> On 11/06/2018 09:49, Song, HaiyanX wrote:
>>>> Hi Laurent,
>>>>
>>>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>>>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>>>> V9 patch serials.
>>>>
>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>>>> commit id:
>>>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>>>> Benchmark: will-it-scale
>>>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>>>
>>>> Metrics:
>>>>   will-it-scale.per_process_ops=processes/nr_cpu
>>>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>> THP: enable / disable
>>>> nr_task:100%
>>>>
>>>> 1. Regressions:
>>>>
>>>> a). Enable THP
>>>> testcase                          base           change      head           metric
>>>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>>>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>>>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>>>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>>>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>>>
>>>> b). Disable THP
>>>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>>>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>>>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>>>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>>>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>>>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>>>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>>>
>>>> Notes: for the above  values of test result, the higher is better.
>>>
>>> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
>>> get reproducible results. The results have huge variation, even on the vanilla
>>> kernel, and I can't state on any changes due to that.
>>>
>>> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
>>> measure any changes between the vanilla and the SPF patched ones:
>>>
>>> test THP enabled                4.17.0-rc4-mm1  spf             delta
>>> page_fault3_threads             2697.7          2683.5          -0.53%
>>> page_fault2_threads             170660.6        169574.1        -0.64%
>>> context_switch1_threads         6915269.2       6877507.3       -0.55%
>>> context_switch1_processes       6478076.2       6529493.5       0.79%
>>> brk1                            243391.2        238527.5        -2.00%
>>>
>>> Tests were run 10 times, no high variation detected.
>>>
>>> Did you see high variation on your side ? How many times the test were run to
>>> compute the average values ?
>>>
>>> Thanks,
>>> Laurent.
>>>
>>>
>>>>
>>>> 2. Improvement: not found improvement based on the selected test cases.
>>>>
>>>>
>>>> Best regards
>>>> Haiyan Song
>>>> ________________________________________
>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>> Sent: Monday, May 28, 2018 4:54 PM
>>>> To: Song, HaiyanX
>>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>>
>>>> On 28/05/2018 10:22, Haiyan Song wrote:
>>>>> Hi Laurent,
>>>>>
>>>>> Yes, these tests are done on V9 patch.
>>>>
>>>> Do you plan to give this V11 a run ?
>>>>
>>>>>
>>>>>
>>>>> Best regards,
>>>>> Haiyan Song
>>>>>
>>>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>>>
>>>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>>>> tested on Intel 4s Skylake platform.
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>>>> series" while responding to the v11 header series...
>>>>>> Were these tests done on v9 or v11 ?
>>>>>>
>>>>>> Cheers,
>>>>>> Laurent.
>>>>>>
>>>>>>>
>>>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>>>> Commit id:
>>>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>>>> Benchmark suite: will-it-scale
>>>>>>> Download link:
>>>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>>>> Metrics:
>>>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>>>> THP: enable / disable
>>>>>>> nr_task: 100%
>>>>>>>
>>>>>>> 1. Regressions:
>>>>>>> a) THP enabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>>>
>>>>>>> b) THP disabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>>>
>>>>>>> 2. Improvements:
>>>>>>> a) THP enabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>>>
>>>>>>> b) THP disabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>>>
>>>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>>>
>>>>>>>
>>>>>>> Best regards
>>>>>>> Haiyan Song
>>>>>>>
>>>>>>> ________________________________________
>>>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>>>
>>>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>>>> page fault without holding the mm semaphore [1].
>>>>>>>
>>>>>>> The idea is to try to handle user space page faults without holding the
>>>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>>>> process since the page fault handler will not wait for other threads memory
>>>>>>> layout change to be done, assuming that this change is done in another part
>>>>>>> of the process's memory space. This type page fault is named speculative
>>>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>>>> is failing its processing and a classic page fault is then tried.
>>>>>>>
>>>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>>>> limiting the locking contention to these operations which are expected to
>>>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>>>> benchmark anymore.
>>>>>>>
>>>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>>>> have to be protected against parallel changes. This is done by using a per
>>>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>>>> speculative page fault in that case.
>>>>>>>
>>>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>>>> checked during the page fault are modified.
>>>>>>>
>>>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>>>> parallel change is possible at this time.
>>>>>>>
>>>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>>>> classic page fault handler will be called to handle the operation while
>>>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>>>> PTE.
>>>>>>>
>>>>>>> In pseudo code, this could be seen as:
>>>>>>>     speculative_page_fault()
>>>>>>>     {
>>>>>>>             vma = get_vma()
>>>>>>>             check vma sequence count
>>>>>>>             check vma's support
>>>>>>>             disable interrupt
>>>>>>>                   check pgd,p4d,...,pte
>>>>>>>                   save pmd and pte in vmf
>>>>>>>                   save vma sequence counter in vmf
>>>>>>>             enable interrupt
>>>>>>>             check vma sequence count
>>>>>>>             handle_pte_fault(vma)
>>>>>>>                     ..
>>>>>>>                     page = alloc_page()
>>>>>>>                     pte_map_lock()
>>>>>>>                             disable interrupt
>>>>>>>                                     abort if sequence counter has changed
>>>>>>>                                     abort if pmd or pte has changed
>>>>>>>                                     pte map and lock
>>>>>>>                             enable interrupt
>>>>>>>                     if abort
>>>>>>>                        free page
>>>>>>>                        abort
>>>>>>>                     ...
>>>>>>>     }
>>>>>>>
>>>>>>>     arch_fault_handler()
>>>>>>>     {
>>>>>>>             if (speculative_page_fault(&vma))
>>>>>>>                goto done
>>>>>>>     again:
>>>>>>>             lock(mmap_sem)
>>>>>>>             vma = find_vma();
>>>>>>>             handle_pte_fault(vma);
>>>>>>>             if retry
>>>>>>>                unlock(mmap_sem)
>>>>>>>                goto again;
>>>>>>>     done:
>>>>>>>             handle fault error
>>>>>>>     }
>>>>>>>
>>>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>>>
>>>>>>> This series add a new software performance event named 'speculative-faults'
>>>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>>>> the part of the faults processed speculatively.
>>>>>>>
>>>>>>> There are some trace events introduced by this series. They allow
>>>>>>> identifying why the page faults were not processed speculatively. This
>>>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>>>> grouped in a system named 'pagefault', they are:
>>>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>>>    back.
>>>>>>>
>>>>>>> To record all the related events, the easier is to run perf with the
>>>>>>> following arguments :
>>>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>>>
>>>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>>>> page fault handled speculatively. I can be seen this way:
>>>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>>>
>>>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>>>> on x86, PowerPC and arm64.
>>>>>>>
>>>>>>> ---------------------
>>>>>>> Real Workload results
>>>>>>>
>>>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>>>> this new version should not impact the performance boost seen.
>>>>>>>
>>>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>>>> series:
>>>>>>>                 vanilla         spf
>>>>>>> faults          89.418          101.364         +13%
>>>>>>> spf                n/a           97.989
>>>>>>>
>>>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>>>> way.
>>>>>>>
>>>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>>>> it a try on an android device. He reported that the application launch time
>>>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>>>> 20%.
>>>>>>>
>>>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>>>
>>>>>>> Application                             4.9     4.9+spf delta
>>>>>>> com.tencent.mm                          416     389     -7%
>>>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>>>> com.tencent.mtt                         455     454     0%
>>>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>>>> com.autonavi.minimap                    711     701     -1%
>>>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>>>> com.immomo.momo                         501     487     -3%
>>>>>>> com.tencent.peng                        2145    2112    -2%
>>>>>>> com.smile.gifmaker                      491     461     -6%
>>>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>>>> com.baidu.searchbox                     333     314     -6%
>>>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>>>> com.sina.weibo                          907     906     0%
>>>>>>> com.youku.phone                         816     731     -11%
>>>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>>>> com.UCMobile                            415     411     -1%
>>>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>>>
>>>>>>> ------------------
>>>>>>> Benchmarks results
>>>>>>>
>>>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>>>> SPF is BASE + this series
>>>>>>>
>>>>>>> Kernbench:
>>>>>>> ----------
>>>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>>>> kernel (kernel is build 5 times):
>>>>>>>
>>>>>>> Average Half load -j 8
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>>>
>>>>>>> Average Optimal load -j 16
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>>>
>>>>>>>
>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>          526743764      faults
>>>>>>>                210      spf
>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>               2278      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> Very few speculative page faults were recorded as most of the processes
>>>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>>>> were created during the kernel build processing).
>>>>>>>
>>>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>>>
>>>>>>> Average Half load -j 40
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>>>
>>>>>>> Average Optimal load -j 80
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>>>
>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>          116730856      faults
>>>>>>>                  0      spf
>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                476      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>>>> there is no impact on the performance.
>>>>>>>
>>>>>>> Ebizzy:
>>>>>>> -------
>>>>>>> The test is counting the number of records per second it can manage, the
>>>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>>>> consistent result I repeated the test 100 times and measure the average
>>>>>>> result. The number is the record processes per second, the higher is the
>>>>>>> best.
>>>>>>>
>>>>>>>                 BASE            SPF             delta
>>>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>>>
>>>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>>>            1706379      faults
>>>>>>>            1674599      spf
>>>>>>>              30588      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                363      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>>>            1874773      faults
>>>>>>>            1461153      spf
>>>>>>>             413293      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                200      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>>>> leading the ebizzy performance boost.
>>>>>>>
>>>>>>> ------------------
>>>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>>>    and Minchan Kim, hopefully.
>>>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>>>    __do_page_fault().
>>>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>>>    instead
>>>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>>>> useless
>>>>>>>    trace event pagefault:spf_pte_lock.
>>>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>>>
>>>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>>>
>>>>>>>
>>>>>>> Laurent Dufour (20):
>>>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>>>   mm: introduce INIT_VMA()
>>>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>>>   mm: protect mremap() against SPF hanlder
>>>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>>>   mm: introduce __vm_normal_page()
>>>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>>>   mm: adding speculative page fault failure trace events
>>>>>>>   perf: add a speculative page fault sw event
>>>>>>>   perf tools: add support for the SPF perf event
>>>>>>>   mm: add speculative page fault vmstats
>>>>>>>   powerpc/mm: add speculative page fault
>>>>>>>
>>>>>>> Mahendran Ganesh (2):
>>>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   arm64/mm: add speculative page fault
>>>>>>>
>>>>>>> Peter Zijlstra (4):
>>>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>>>   mm: VMA sequence count
>>>>>>>   mm: provide speculative fault infrastructure
>>>>>>>   x86/mm: add speculative pagefault handling
>>>>>>>
>>>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>>>  fs/exec.c                             |   2 +-
>>>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>>>  include/linux/migrate.h               |   4 +-
>>>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>>>  include/linux/mm_types.h              |   7 +
>>>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>>>  include/linux/swap.h                  |  10 +-
>>>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>>>  kernel/fork.c                         |   5 +-
>>>>>>>  mm/Kconfig                            |  22 ++
>>>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>>>  mm/hugetlb.c                          |   2 +
>>>>>>>  mm/init-mm.c                          |   3 +
>>>>>>>  mm/internal.h                         |  20 ++
>>>>>>>  mm/khugepaged.c                       |   5 +
>>>>>>>  mm/madvise.c                          |   6 +-
>>>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>>>  mm/migrate.c                          |   6 +-
>>>>>>>  mm/mlock.c                            |  13 +-
>>>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>>>  mm/mprotect.c                         |   4 +-
>>>>>>>  mm/mremap.c                           |  13 +
>>>>>>>  mm/nommu.c                            |   2 +-
>>>>>>>  mm/rmap.c                             |   5 +-
>>>>>>>  mm/swap.c                             |   6 +-
>>>>>>>  mm/swap_state.c                       |   8 +-
>>>>>>>  mm/vmstat.c                           |   5 +-
>>>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>>>  tools/perf/util/python.c              |   1 +
>>>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>>>
>>>>>>> --
>>>>>>> 2.7.4
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>>
>>
> 

Re: [PATCH v11 19/26] mm: provide speculative fault infrastructure

[zhong jiang]

On 2018/5/17 19:06, Laurent Dufour wrote:
> From: Peter Zijlstra <peterz@infradead.org>
>
> Provide infrastructure to do a speculative fault (not holding
> mmap_sem).
>
> The not holding of mmap_sem means we can race against VMA
> change/removal and page-table destruction. We use the SRCU VMA freeing
> to keep the VMA around. We use the VMA seqcount to detect change
> (including umapping / page-table deletion) and we use gup_fast() style
> page-table walking to deal with page-table races.
>
> Once we've obtained the page and are ready to update the PTE, we
> validate if the state we started the fault with is still valid, if
> not, we'll fail the fault with VM_FAULT_RETRY, otherwise we update the
> PTE and we're done.
>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>
> [Manage the newly introduced pte_spinlock() for speculative page
>  fault to fail if the VMA is touched in our back]
> [Rename vma_is_dead() to vma_has_changed() and declare it here]
> [Fetch p4d and pud]
> [Set vmd.sequence in __handle_mm_fault()]
> [Abort speculative path when handle_userfault() has to be called]
> [Add additional VMA's flags checks in handle_speculative_fault()]
> [Clear FAULT_FLAG_ALLOW_RETRY in handle_speculative_fault()]
> [Don't set vmf->pte and vmf->ptl if pte_map_lock() failed]
> [Remove warning comment about waiting for !seq&1 since we don't want
>  to wait]
> [Remove warning about no huge page support, mention it explictly]
> [Don't call do_fault() in the speculative path as __do_fault() calls
>  vma->vm_ops->fault() which may want to release mmap_sem]
> [Only vm_fault pointer argument for vma_has_changed()]
> [Fix check against huge page, calling pmd_trans_huge()]
> [Use READ_ONCE() when reading VMA's fields in the speculative path]
> [Explicitly check for __HAVE_ARCH_PTE_SPECIAL as we can't support for
>  processing done in vm_normal_page()]
> [Check that vma->anon_vma is already set when starting the speculative
>  path]
> [Check for memory policy as we can't support MPOL_INTERLEAVE case due to
>  the processing done in mpol_misplaced()]
> [Don't support VMA growing up or down]
> [Move check on vm_sequence just before calling handle_pte_fault()]
> [Don't build SPF services if !CONFIG_SPECULATIVE_PAGE_FAULT]
> [Add mem cgroup oom check]
> [Use READ_ONCE to access p*d entries]
> [Replace deprecated ACCESS_ONCE() by READ_ONCE() in vma_has_changed()]
> [Don't fetch pte again in handle_pte_fault() when running the speculative
>  path]
> [Check PMD against concurrent collapsing operation]
> [Try spin lock the pte during the speculative path to avoid deadlock with
>  other CPU's invalidating the TLB and requiring this CPU to catch the
>  inter processor's interrupt]
> [Move define of FAULT_FLAG_SPECULATIVE here]
> [Introduce __handle_speculative_fault() and add a check against
>  mm->mm_users in handle_speculative_fault() defined in mm.h]
> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
> ---
>  include/linux/hugetlb_inline.h |   2 +-
>  include/linux/mm.h             |  30 ++++
>  include/linux/pagemap.h        |   4 +-
>  mm/internal.h                  |  16 +-
>  mm/memory.c                    | 340 ++++++++++++++++++++++++++++++++++++++++-
>  5 files changed, 385 insertions(+), 7 deletions(-)
>
> diff --git a/include/linux/hugetlb_inline.h b/include/linux/hugetlb_inline.h
> index 0660a03d37d9..9e25283d6fc9 100644
> --- a/include/linux/hugetlb_inline.h
> +++ b/include/linux/hugetlb_inline.h
> @@ -8,7 +8,7 @@
>  
>  static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
>  {
> -	return !!(vma->vm_flags & VM_HUGETLB);
> +	return !!(READ_ONCE(vma->vm_flags) & VM_HUGETLB);
>  }
>  
>  #else
> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index 05cbba70104b..31acf98a7d92 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -315,6 +315,7 @@ extern pgprot_t protection_map[16];
>  #define FAULT_FLAG_USER		0x40	/* The fault originated in userspace */
>  #define FAULT_FLAG_REMOTE	0x80	/* faulting for non current tsk/mm */
>  #define FAULT_FLAG_INSTRUCTION  0x100	/* The fault was during an instruction fetch */
> +#define FAULT_FLAG_SPECULATIVE	0x200	/* Speculative fault, not holding mmap_sem */
>  
>  #define FAULT_FLAG_TRACE \
>  	{ FAULT_FLAG_WRITE,		"WRITE" }, \
> @@ -343,6 +344,10 @@ struct vm_fault {
>  	gfp_t gfp_mask;			/* gfp mask to be used for allocations */
>  	pgoff_t pgoff;			/* Logical page offset based on vma */
>  	unsigned long address;		/* Faulting virtual address */
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +	unsigned int sequence;
> +	pmd_t orig_pmd;			/* value of PMD at the time of fault */
> +#endif
>  	pmd_t *pmd;			/* Pointer to pmd entry matching
>  					 * the 'address' */
>  	pud_t *pud;			/* Pointer to pud entry matching
> @@ -1415,6 +1420,31 @@ int invalidate_inode_page(struct page *page);
>  #ifdef CONFIG_MMU
>  extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>  		unsigned int flags);
> +
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +extern int __handle_speculative_fault(struct mm_struct *mm,
> +				      unsigned long address,
> +				      unsigned int flags);
> +static inline int handle_speculative_fault(struct mm_struct *mm,
> +					   unsigned long address,
> +					   unsigned int flags)
> +{
> +	/*
> +	 * Try speculative page fault for multithreaded user space task only.
> +	 */
> +	if (!(flags & FAULT_FLAG_USER) || atomic_read(&mm->mm_users) == 1)
> +		return VM_FAULT_RETRY;
> +	return __handle_speculative_fault(mm, address, flags);
> +}
> +#else
> +static inline int handle_speculative_fault(struct mm_struct *mm,
> +					   unsigned long address,
> +					   unsigned int flags)
> +{
> +	return VM_FAULT_RETRY;
> +}
> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
> +
>  extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
>  			    unsigned long address, unsigned int fault_flags,
>  			    bool *unlocked);
> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
> index b1bd2186e6d2..6e2aa4e79af7 100644
> --- a/include/linux/pagemap.h
> +++ b/include/linux/pagemap.h
> @@ -456,8 +456,8 @@ static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
>  	pgoff_t pgoff;
>  	if (unlikely(is_vm_hugetlb_page(vma)))
>  		return linear_hugepage_index(vma, address);
> -	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
> -	pgoff += vma->vm_pgoff;
> +	pgoff = (address - READ_ONCE(vma->vm_start)) >> PAGE_SHIFT;
> +	pgoff += READ_ONCE(vma->vm_pgoff);
>  	return pgoff;
>  }
>  
> diff --git a/mm/internal.h b/mm/internal.h
> index fb2667b20f0a..10b188c87fa4 100644
> --- a/mm/internal.h
> +++ b/mm/internal.h
> @@ -44,7 +44,21 @@ int do_swap_page(struct vm_fault *vmf);
>  extern struct vm_area_struct *get_vma(struct mm_struct *mm,
>  				      unsigned long addr);
>  extern void put_vma(struct vm_area_struct *vma);
> -#endif
> +
> +static inline bool vma_has_changed(struct vm_fault *vmf)
> +{
> +	int ret = RB_EMPTY_NODE(&vmf->vma->vm_rb);
> +	unsigned int seq = READ_ONCE(vmf->vma->vm_sequence.sequence);
> +
> +	/*
> +	 * Matches both the wmb in write_seqlock_{begin,end}() and
> +	 * the wmb in vma_rb_erase().
> +	 */
> +	smp_rmb();
> +
> +	return ret || seq != vmf->sequence;
> +}
> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>  
>  void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
>  		unsigned long floor, unsigned long ceiling);
> diff --git a/mm/memory.c b/mm/memory.c
> index ab32b0b4bd69..7bbbb8c7b9cd 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -769,7 +769,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
>  	if (page)
>  		dump_page(page, "bad pte");
>  	pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
> -		 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
> +		 (void *)addr, READ_ONCE(vma->vm_flags), vma->anon_vma,
> +		 mapping, index);
>  	pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
>  		 vma->vm_file,
>  		 vma->vm_ops ? vma->vm_ops->fault : NULL,
> @@ -2306,6 +2307,118 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
>  }
>  EXPORT_SYMBOL_GPL(apply_to_page_range);
>  
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +static bool pte_spinlock(struct vm_fault *vmf)
> +{
> +	bool ret = false;
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	pmd_t pmdval;
> +#endif
> +
> +	/* Check if vma is still valid */
> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
> +		vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
> +		spin_lock(vmf->ptl);
> +		return true;
> +	}
> +
> +again:
> +	local_irq_disable();
> +	if (vma_has_changed(vmf))
> +		goto out;
> +
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	/*
> +	 * We check if the pmd value is still the same to ensure that there
> +	 * is not a huge collapse operation in progress in our back.
> +	 */
> +	pmdval = READ_ONCE(*vmf->pmd);
> +	if (!pmd_same(pmdval, vmf->orig_pmd))
> +		goto out;
> +#endif
> +
> +	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
> +	if (unlikely(!spin_trylock(vmf->ptl))) {
> +		local_irq_enable();
> +		goto again;
> +	}
> +
> +	if (vma_has_changed(vmf)) {
> +		spin_unlock(vmf->ptl);
> +		goto out;
> +	}
> +
> +	ret = true;
> +out:
> +	local_irq_enable();
> +	return ret;
> +}
> +
> +static bool pte_map_lock(struct vm_fault *vmf)
> +{
> +	bool ret = false;
> +	pte_t *pte;
> +	spinlock_t *ptl;
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	pmd_t pmdval;
> +#endif
> +
> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
> +		vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
> +					       vmf->address, &vmf->ptl);
> +		return true;
> +	}
> +
> +	/*
> +	 * The first vma_has_changed() guarantees the page-tables are still
> +	 * valid, having IRQs disabled ensures they stay around, hence the
> +	 * second vma_has_changed() to make sure they are still valid once
> +	 * we've got the lock. After that a concurrent zap_pte_range() will
> +	 * block on the PTL and thus we're safe.
> +	 */
> +again:
> +	local_irq_disable();
> +	if (vma_has_changed(vmf))
> +		goto out;
> +
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	/*
> +	 * We check if the pmd value is still the same to ensure that there
> +	 * is not a huge collapse operation in progress in our back.
> +	 */
> +	pmdval = READ_ONCE(*vmf->pmd);
> +	if (!pmd_same(pmdval, vmf->orig_pmd))
> +		goto out;
> +#endif
> +
> +	/*
> +	 * Same as pte_offset_map_lock() except that we call
> +	 * spin_trylock() in place of spin_lock() to avoid race with
> +	 * unmap path which may have the lock and wait for this CPU
> +	 * to invalidate TLB but this CPU has irq disabled.
> +	 * Since we are in a speculative patch, accept it could fail
> +	 */
> +	ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
> +	pte = pte_offset_map(vmf->pmd, vmf->address);
> +	if (unlikely(!spin_trylock(ptl))) {
> +		pte_unmap(pte);
> +		local_irq_enable();
> +		goto again;
> +	}
> +
> +	if (vma_has_changed(vmf)) {
> +		pte_unmap_unlock(pte, ptl);
> +		goto out;
> +	}
> +
> +	vmf->pte = pte;
> +	vmf->ptl = ptl;
> +	ret = true;
> +out:
> +	local_irq_enable();
> +	return ret;
> +}
> +#else
>  static inline bool pte_spinlock(struct vm_fault *vmf)
>  {
>  	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
> @@ -2319,6 +2432,7 @@ static inline bool pte_map_lock(struct vm_fault *vmf)
>  				       vmf->address, &vmf->ptl);
>  	return true;
>  }
> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>  
>  /*
>   * handle_pte_fault chooses page fault handler according to an entry which was
> @@ -3208,6 +3322,14 @@ static int do_anonymous_page(struct vm_fault *vmf)
>  		ret = check_stable_address_space(vma->vm_mm);
>  		if (ret)
>  			goto unlock;
> +		/*
> +		 * Don't call the userfaultfd during the speculative path.
> +		 * We already checked for the VMA to not be managed through
> +		 * userfaultfd, but it may be set in our back once we have lock
> +		 * the pte. In such a case we can ignore it this time.
> +		 */
> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
> +			goto setpte;
>  		/* Deliver the page fault to userland, check inside PT lock */
>  		if (userfaultfd_missing(vma)) {
>  			pte_unmap_unlock(vmf->pte, vmf->ptl);
> @@ -3249,7 +3371,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
>  		goto unlock_and_release;
>  
>  	/* Deliver the page fault to userland, check inside PT lock */
> -	if (userfaultfd_missing(vma)) {
> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE) && userfaultfd_missing(vma)) {
>  		pte_unmap_unlock(vmf->pte, vmf->ptl);
>  		mem_cgroup_cancel_charge(page, memcg, false);
>  		put_page(page);
> @@ -3994,13 +4116,22 @@ static int handle_pte_fault(struct vm_fault *vmf)
>  
>  	if (unlikely(pmd_none(*vmf->pmd))) {
>  		/*
> +		 * In the case of the speculative page fault handler we abort
> +		 * the speculative path immediately as the pmd is probably
> +		 * in the way to be converted in a huge one. We will try
> +		 * again holding the mmap_sem (which implies that the collapse
> +		 * operation is done).
> +		 */
> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
> +			return VM_FAULT_RETRY;
> +		/*
>  		 * Leave __pte_alloc() until later: because vm_ops->fault may
>  		 * want to allocate huge page, and if we expose page table
>  		 * for an instant, it will be difficult to retract from
>  		 * concurrent faults and from rmap lookups.
>  		 */
>  		vmf->pte = NULL;
> -	} else {
> +	} else if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>  		/* See comment in pte_alloc_one_map() */
>  		if (pmd_devmap_trans_unstable(vmf->pmd))
>  			return 0;
> @@ -4009,6 +4140,9 @@ static int handle_pte_fault(struct vm_fault *vmf)
>  		 * pmd from under us anymore at this point because we hold the
>  		 * mmap_sem read mode and khugepaged takes it in write mode.
>  		 * So now it's safe to run pte_offset_map().
> +		 * This is not applicable to the speculative page fault handler
> +		 * but in that case, the pte is fetched earlier in
> +		 * handle_speculative_fault().
>  		 */
>  		vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
>  		vmf->orig_pte = *vmf->pte;
> @@ -4031,6 +4165,8 @@ static int handle_pte_fault(struct vm_fault *vmf)
>  	if (!vmf->pte) {
>  		if (vma_is_anonymous(vmf->vma))
>  			return do_anonymous_page(vmf);
> +		else if (vmf->flags & FAULT_FLAG_SPECULATIVE)
> +			return VM_FAULT_RETRY;
>  		else
>  			return do_fault(vmf);
>  	}
> @@ -4128,6 +4264,9 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>  	vmf.pmd = pmd_alloc(mm, vmf.pud, address);
>  	if (!vmf.pmd)
>  		return VM_FAULT_OOM;
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +	vmf.sequence = raw_read_seqcount(&vma->vm_sequence);
> +#endif
>  	if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
>  		ret = create_huge_pmd(&vmf);
>  		if (!(ret & VM_FAULT_FALLBACK))
> @@ -4161,6 +4300,201 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>  	return handle_pte_fault(&vmf);
>  }
>  
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +/*
> + * Tries to handle the page fault in a speculative way, without grabbing the
> + * mmap_sem.
> + */
> +int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
> +			       unsigned int flags)
> +{
> +	struct vm_fault vmf = {
> +		.address = address,
> +	};
> +	pgd_t *pgd, pgdval;
> +	p4d_t *p4d, p4dval;
> +	pud_t pudval;
> +	int seq, ret = VM_FAULT_RETRY;
> +	struct vm_area_struct *vma;
> +#ifdef CONFIG_NUMA
> +	struct mempolicy *pol;
> +#endif
> +
> +	/* Clear flags that may lead to release the mmap_sem to retry */
> +	flags &= ~(FAULT_FLAG_ALLOW_RETRY|FAULT_FLAG_KILLABLE);
> +	flags |= FAULT_FLAG_SPECULATIVE;
> +
> +	vma = get_vma(mm, address);
> +	if (!vma)
> +		return ret;
> +
> +	seq = raw_read_seqcount(&vma->vm_sequence); /* rmb <-> seqlock,vma_rb_erase() */
> +	if (seq & 1)
> +		goto out_put;
> +
> +	/*
> +	 * Can't call vm_ops service has we don't know what they would do
> +	 * with the VMA.
> +	 * This include huge page from hugetlbfs.
> +	 */
> +	if (vma->vm_ops)
> +		goto out_put;
> +
  Hi   Laurent
   
   I think that most of pagefault will leave here.   Is there any case  need to skip ?
  I have tested the following  patch, it work well.

diff --git a/mm/memory.c b/mm/memory.c
index 936128b..9bc1545 100644
 @@ -3893,8 +3898,6 @@ static int handle_pte_fault(struct fault_env *fe)
        if (!fe->pte) {
                if (vma_is_anonymous(fe->vma))
                        return do_anonymous_page(fe);
-               else if (fe->flags & FAULT_FLAG_SPECULATIVE)
-                       return VM_FAULT_RETRY;
                else
                        return do_fault(fe);
        }
@@ -4026,20 +4029,11 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
                goto out_put;
        }
        /*
-        * Can't call vm_ops service has we don't know what they would do
-        * with the VMA.
-        * This include huge page from hugetlbfs.
-        */
-       if (vma->vm_ops) {
-               trace_spf_vma_notsup(_RET_IP_, vma, address);
-               goto out_put;
-       }


Thanks
zhong jiang
> +	/*
> +	 * __anon_vma_prepare() requires the mmap_sem to be held
> +	 * because vm_next and vm_prev must be safe. This can't be guaranteed
> +	 * in the speculative path.
> +	 */
> +	if (unlikely(!vma->anon_vma))
> +		goto out_put;
> +
> +	vmf.vma_flags = READ_ONCE(vma->vm_flags);
> +	vmf.vma_page_prot = READ_ONCE(vma->vm_page_prot);
> +
> +	/* Can't call userland page fault handler in the speculative path */
> +	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING))
> +		goto out_put;
> +
> +	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP)
> +		/*
> +		 * This could be detected by the check address against VMA's
> +		 * boundaries but we want to trace it as not supported instead
> +		 * of changed.
> +		 */
> +		goto out_put;
> +
> +	if (address < READ_ONCE(vma->vm_start)
> +	    || READ_ONCE(vma->vm_end) <= address)
> +		goto out_put;
> +
> +	if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
> +				       flags & FAULT_FLAG_INSTRUCTION,
> +				       flags & FAULT_FLAG_REMOTE)) {
> +		ret = VM_FAULT_SIGSEGV;
> +		goto out_put;
> +	}
> +
> +	/* This is one is required to check that the VMA has write access set */
> +	if (flags & FAULT_FLAG_WRITE) {
> +		if (unlikely(!(vmf.vma_flags & VM_WRITE))) {
> +			ret = VM_FAULT_SIGSEGV;
> +			goto out_put;
> +		}
> +	} else if (unlikely(!(vmf.vma_flags & (VM_READ|VM_EXEC|VM_WRITE)))) {
> +		ret = VM_FAULT_SIGSEGV;
> +		goto out_put;
> +	}
> +
> +#ifdef CONFIG_NUMA
> +	/*
> +	 * MPOL_INTERLEAVE implies additional checks in
> +	 * mpol_misplaced() which are not compatible with the
> +	 *speculative page fault processing.
> +	 */
> +	pol = __get_vma_policy(vma, address);
> +	if (!pol)
> +		pol = get_task_policy(current);
> +	if (pol && pol->mode == MPOL_INTERLEAVE)
> +		goto out_put;
> +#endif
> +
> +	/*
> +	 * Do a speculative lookup of the PTE entry.
> +	 */
> +	local_irq_disable();
> +	pgd = pgd_offset(mm, address);
> +	pgdval = READ_ONCE(*pgd);
> +	if (pgd_none(pgdval) || unlikely(pgd_bad(pgdval)))
> +		goto out_walk;
> +
> +	p4d = p4d_offset(pgd, address);
> +	p4dval = READ_ONCE(*p4d);
> +	if (p4d_none(p4dval) || unlikely(p4d_bad(p4dval)))
> +		goto out_walk;
> +
> +	vmf.pud = pud_offset(p4d, address);
> +	pudval = READ_ONCE(*vmf.pud);
> +	if (pud_none(pudval) || unlikely(pud_bad(pudval)))
> +		goto out_walk;
> +
> +	/* Huge pages at PUD level are not supported. */
> +	if (unlikely(pud_trans_huge(pudval)))
> +		goto out_walk;
> +
> +	vmf.pmd = pmd_offset(vmf.pud, address);
> +	vmf.orig_pmd = READ_ONCE(*vmf.pmd);
> +	/*
> +	 * pmd_none could mean that a hugepage collapse is in progress
> +	 * in our back as collapse_huge_page() mark it before
> +	 * invalidating the pte (which is done once the IPI is catched
> +	 * by all CPU and we have interrupt disabled).
> +	 * For this reason we cannot handle THP in a speculative way since we
> +	 * can't safely indentify an in progress collapse operation done in our
> +	 * back on that PMD.
> +	 * Regarding the order of the following checks, see comment in
> +	 * pmd_devmap_trans_unstable()
> +	 */
> +	if (unlikely(pmd_devmap(vmf.orig_pmd) ||
> +		     pmd_none(vmf.orig_pmd) || pmd_trans_huge(vmf.orig_pmd) ||
> +		     is_swap_pmd(vmf.orig_pmd)))
> +		goto out_walk;
> +
> +	/*
> +	 * The above does not allocate/instantiate page-tables because doing so
> +	 * would lead to the possibility of instantiating page-tables after
> +	 * free_pgtables() -- and consequently leaking them.
> +	 *
> +	 * The result is that we take at least one !speculative fault per PMD
> +	 * in order to instantiate it.
> +	 */
> +
> +	vmf.pte = pte_offset_map(vmf.pmd, address);
> +	vmf.orig_pte = READ_ONCE(*vmf.pte);
> +	barrier(); /* See comment in handle_pte_fault() */
> +	if (pte_none(vmf.orig_pte)) {
> +		pte_unmap(vmf.pte);
> +		vmf.pte = NULL;
> +	}
> +
> +	vmf.vma = vma;
> +	vmf.pgoff = linear_page_index(vma, address);
> +	vmf.gfp_mask = __get_fault_gfp_mask(vma);
> +	vmf.sequence = seq;
> +	vmf.flags = flags;
> +
> +	local_irq_enable();
> +
> +	/*
> +	 * We need to re-validate the VMA after checking the bounds, otherwise
> +	 * we might have a false positive on the bounds.
> +	 */
> +	if (read_seqcount_retry(&vma->vm_sequence, seq))
> +		goto out_put;
> +
> +	mem_cgroup_oom_enable();
> +	ret = handle_pte_fault(&vmf);
> +	mem_cgroup_oom_disable();
> +
> +	put_vma(vma);
> +
> +	/*
> +	 * The task may have entered a memcg OOM situation but
> +	 * if the allocation error was handled gracefully (no
> +	 * VM_FAULT_OOM), there is no need to kill anything.
> +	 * Just clean up the OOM state peacefully.
> +	 */
> +	if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
> +		mem_cgroup_oom_synchronize(false);
> +	return ret;
> +
> +out_walk:
> +	local_irq_enable();
> +out_put:
> +	put_vma(vma);
> +	return ret;
> +}
> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
> +
>  /*
>   * By the time we get here, we already hold the mm semaphore
>   *

Re: [PATCH v11 19/26] mm: provide speculative fault infrastructure

[zhong jiang]

On 2018/5/17 19:06, Laurent Dufour wrote:
> From: Peter Zijlstra <peterz@infradead.org>
>
> Provide infrastructure to do a speculative fault (not holding
> mmap_sem).
>
> The not holding of mmap_sem means we can race against VMA
> change/removal and page-table destruction. We use the SRCU VMA freeing
> to keep the VMA around. We use the VMA seqcount to detect change
> (including umapping / page-table deletion) and we use gup_fast() style
> page-table walking to deal with page-table races.
>
> Once we've obtained the page and are ready to update the PTE, we
> validate if the state we started the fault with is still valid, if
> not, we'll fail the fault with VM_FAULT_RETRY, otherwise we update the
> PTE and we're done.
>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>
> [Manage the newly introduced pte_spinlock() for speculative page
>  fault to fail if the VMA is touched in our back]
> [Rename vma_is_dead() to vma_has_changed() and declare it here]
> [Fetch p4d and pud]
> [Set vmd.sequence in __handle_mm_fault()]
> [Abort speculative path when handle_userfault() has to be called]
> [Add additional VMA's flags checks in handle_speculative_fault()]
> [Clear FAULT_FLAG_ALLOW_RETRY in handle_speculative_fault()]
> [Don't set vmf->pte and vmf->ptl if pte_map_lock() failed]
> [Remove warning comment about waiting for !seq&1 since we don't want
>  to wait]
> [Remove warning about no huge page support, mention it explictly]
> [Don't call do_fault() in the speculative path as __do_fault() calls
>  vma->vm_ops->fault() which may want to release mmap_sem]
> [Only vm_fault pointer argument for vma_has_changed()]
> [Fix check against huge page, calling pmd_trans_huge()]
> [Use READ_ONCE() when reading VMA's fields in the speculative path]
> [Explicitly check for __HAVE_ARCH_PTE_SPECIAL as we can't support for
>  processing done in vm_normal_page()]
> [Check that vma->anon_vma is already set when starting the speculative
>  path]
> [Check for memory policy as we can't support MPOL_INTERLEAVE case due to
>  the processing done in mpol_misplaced()]
> [Don't support VMA growing up or down]
> [Move check on vm_sequence just before calling handle_pte_fault()]
> [Don't build SPF services if !CONFIG_SPECULATIVE_PAGE_FAULT]
> [Add mem cgroup oom check]
> [Use READ_ONCE to access p*d entries]
> [Replace deprecated ACCESS_ONCE() by READ_ONCE() in vma_has_changed()]
> [Don't fetch pte again in handle_pte_fault() when running the speculative
>  path]
> [Check PMD against concurrent collapsing operation]
> [Try spin lock the pte during the speculative path to avoid deadlock with
>  other CPU's invalidating the TLB and requiring this CPU to catch the
>  inter processor's interrupt]
> [Move define of FAULT_FLAG_SPECULATIVE here]
> [Introduce __handle_speculative_fault() and add a check against
>  mm->mm_users in handle_speculative_fault() defined in mm.h]
> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
> ---
>  include/linux/hugetlb_inline.h |   2 +-
>  include/linux/mm.h             |  30 ++++
>  include/linux/pagemap.h        |   4 +-
>  mm/internal.h                  |  16 +-
>  mm/memory.c                    | 340 ++++++++++++++++++++++++++++++++++++++++-
>  5 files changed, 385 insertions(+), 7 deletions(-)
>
> diff --git a/include/linux/hugetlb_inline.h b/include/linux/hugetlb_inline.h
> index 0660a03d37d9..9e25283d6fc9 100644
> --- a/include/linux/hugetlb_inline.h
> +++ b/include/linux/hugetlb_inline.h
> @@ -8,7 +8,7 @@
>  
>  static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
>  {
> -	return !!(vma->vm_flags & VM_HUGETLB);
> +	return !!(READ_ONCE(vma->vm_flags) & VM_HUGETLB);
>  }
>  
>  #else
> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index 05cbba70104b..31acf98a7d92 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -315,6 +315,7 @@ extern pgprot_t protection_map[16];
>  #define FAULT_FLAG_USER		0x40	/* The fault originated in userspace */
>  #define FAULT_FLAG_REMOTE	0x80	/* faulting for non current tsk/mm */
>  #define FAULT_FLAG_INSTRUCTION  0x100	/* The fault was during an instruction fetch */
> +#define FAULT_FLAG_SPECULATIVE	0x200	/* Speculative fault, not holding mmap_sem */
>  
>  #define FAULT_FLAG_TRACE \
>  	{ FAULT_FLAG_WRITE,		"WRITE" }, \
> @@ -343,6 +344,10 @@ struct vm_fault {
>  	gfp_t gfp_mask;			/* gfp mask to be used for allocations */
>  	pgoff_t pgoff;			/* Logical page offset based on vma */
>  	unsigned long address;		/* Faulting virtual address */
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +	unsigned int sequence;
> +	pmd_t orig_pmd;			/* value of PMD at the time of fault */
> +#endif
>  	pmd_t *pmd;			/* Pointer to pmd entry matching
>  					 * the 'address' */
>  	pud_t *pud;			/* Pointer to pud entry matching
> @@ -1415,6 +1420,31 @@ int invalidate_inode_page(struct page *page);
>  #ifdef CONFIG_MMU
>  extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>  		unsigned int flags);
> +
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +extern int __handle_speculative_fault(struct mm_struct *mm,
> +				      unsigned long address,
> +				      unsigned int flags);
> +static inline int handle_speculative_fault(struct mm_struct *mm,
> +					   unsigned long address,
> +					   unsigned int flags)
> +{
> +	/*
> +	 * Try speculative page fault for multithreaded user space task only.
> +	 */
> +	if (!(flags & FAULT_FLAG_USER) || atomic_read(&mm->mm_users) == 1)
> +		return VM_FAULT_RETRY;
> +	return __handle_speculative_fault(mm, address, flags);
> +}
> +#else
> +static inline int handle_speculative_fault(struct mm_struct *mm,
> +					   unsigned long address,
> +					   unsigned int flags)
> +{
> +	return VM_FAULT_RETRY;
> +}
> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
> +
>  extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
>  			    unsigned long address, unsigned int fault_flags,
>  			    bool *unlocked);
> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
> index b1bd2186e6d2..6e2aa4e79af7 100644
> --- a/include/linux/pagemap.h
> +++ b/include/linux/pagemap.h
> @@ -456,8 +456,8 @@ static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
>  	pgoff_t pgoff;
>  	if (unlikely(is_vm_hugetlb_page(vma)))
>  		return linear_hugepage_index(vma, address);
> -	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
> -	pgoff += vma->vm_pgoff;
> +	pgoff = (address - READ_ONCE(vma->vm_start)) >> PAGE_SHIFT;
> +	pgoff += READ_ONCE(vma->vm_pgoff);
>  	return pgoff;
>  }
>  
> diff --git a/mm/internal.h b/mm/internal.h
> index fb2667b20f0a..10b188c87fa4 100644
> --- a/mm/internal.h
> +++ b/mm/internal.h
> @@ -44,7 +44,21 @@ int do_swap_page(struct vm_fault *vmf);
>  extern struct vm_area_struct *get_vma(struct mm_struct *mm,
>  				      unsigned long addr);
>  extern void put_vma(struct vm_area_struct *vma);
> -#endif
> +
> +static inline bool vma_has_changed(struct vm_fault *vmf)
> +{
> +	int ret = RB_EMPTY_NODE(&vmf->vma->vm_rb);
> +	unsigned int seq = READ_ONCE(vmf->vma->vm_sequence.sequence);
> +
> +	/*
> +	 * Matches both the wmb in write_seqlock_{begin,end}() and
> +	 * the wmb in vma_rb_erase().
> +	 */
> +	smp_rmb();
> +
> +	return ret || seq != vmf->sequence;
> +}
> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>  
>  void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
>  		unsigned long floor, unsigned long ceiling);
> diff --git a/mm/memory.c b/mm/memory.c
> index ab32b0b4bd69..7bbbb8c7b9cd 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -769,7 +769,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
>  	if (page)
>  		dump_page(page, "bad pte");
>  	pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
> -		 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
> +		 (void *)addr, READ_ONCE(vma->vm_flags), vma->anon_vma,
> +		 mapping, index);
>  	pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
>  		 vma->vm_file,
>  		 vma->vm_ops ? vma->vm_ops->fault : NULL,
> @@ -2306,6 +2307,118 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
>  }
>  EXPORT_SYMBOL_GPL(apply_to_page_range);
>  
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +static bool pte_spinlock(struct vm_fault *vmf)
> +{
> +	bool ret = false;
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	pmd_t pmdval;
> +#endif
> +
> +	/* Check if vma is still valid */
> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
> +		vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
> +		spin_lock(vmf->ptl);
> +		return true;
> +	}
> +
> +again:
> +	local_irq_disable();
> +	if (vma_has_changed(vmf))
> +		goto out;
> +
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	/*
> +	 * We check if the pmd value is still the same to ensure that there
> +	 * is not a huge collapse operation in progress in our back.
> +	 */
> +	pmdval = READ_ONCE(*vmf->pmd);
> +	if (!pmd_same(pmdval, vmf->orig_pmd))
> +		goto out;
> +#endif
> +
> +	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
> +	if (unlikely(!spin_trylock(vmf->ptl))) {
> +		local_irq_enable();
> +		goto again;
> +	}
> +
> +	if (vma_has_changed(vmf)) {
> +		spin_unlock(vmf->ptl);
> +		goto out;
> +	}
> +
> +	ret = true;
> +out:
> +	local_irq_enable();
> +	return ret;
> +}
> +
> +static bool pte_map_lock(struct vm_fault *vmf)
> +{
> +	bool ret = false;
> +	pte_t *pte;
> +	spinlock_t *ptl;
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	pmd_t pmdval;
> +#endif
> +
> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
> +		vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
> +					       vmf->address, &vmf->ptl);
> +		return true;
> +	}
> +
> +	/*
> +	 * The first vma_has_changed() guarantees the page-tables are still
> +	 * valid, having IRQs disabled ensures they stay around, hence the
> +	 * second vma_has_changed() to make sure they are still valid once
> +	 * we've got the lock. After that a concurrent zap_pte_range() will
> +	 * block on the PTL and thus we're safe.
> +	 */
> +again:
> +	local_irq_disable();
> +	if (vma_has_changed(vmf))
> +		goto out;
> +
> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
> +	/*
> +	 * We check if the pmd value is still the same to ensure that there
> +	 * is not a huge collapse operation in progress in our back.
> +	 */
> +	pmdval = READ_ONCE(*vmf->pmd);
> +	if (!pmd_same(pmdval, vmf->orig_pmd))
> +		goto out;
> +#endif
> +
> +	/*
> +	 * Same as pte_offset_map_lock() except that we call
> +	 * spin_trylock() in place of spin_lock() to avoid race with
> +	 * unmap path which may have the lock and wait for this CPU
> +	 * to invalidate TLB but this CPU has irq disabled.
> +	 * Since we are in a speculative patch, accept it could fail
> +	 */
> +	ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
> +	pte = pte_offset_map(vmf->pmd, vmf->address);
> +	if (unlikely(!spin_trylock(ptl))) {
> +		pte_unmap(pte);
> +		local_irq_enable();
> +		goto again;
> +	}
> +
> +	if (vma_has_changed(vmf)) {
> +		pte_unmap_unlock(pte, ptl);
> +		goto out;
> +	}
> +
> +	vmf->pte = pte;
> +	vmf->ptl = ptl;
> +	ret = true;
> +out:
> +	local_irq_enable();
> +	return ret;
> +}
> +#else
>  static inline bool pte_spinlock(struct vm_fault *vmf)
>  {
>  	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
> @@ -2319,6 +2432,7 @@ static inline bool pte_map_lock(struct vm_fault *vmf)
>  				       vmf->address, &vmf->ptl);
>  	return true;
>  }
> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>  
>  /*
>   * handle_pte_fault chooses page fault handler according to an entry which was
> @@ -3208,6 +3322,14 @@ static int do_anonymous_page(struct vm_fault *vmf)
>  		ret = check_stable_address_space(vma->vm_mm);
>  		if (ret)
>  			goto unlock;
> +		/*
> +		 * Don't call the userfaultfd during the speculative path.
> +		 * We already checked for the VMA to not be managed through
> +		 * userfaultfd, but it may be set in our back once we have lock
> +		 * the pte. In such a case we can ignore it this time.
> +		 */
> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
> +			goto setpte;
>  		/* Deliver the page fault to userland, check inside PT lock */
>  		if (userfaultfd_missing(vma)) {
>  			pte_unmap_unlock(vmf->pte, vmf->ptl);
> @@ -3249,7 +3371,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
>  		goto unlock_and_release;
>  
>  	/* Deliver the page fault to userland, check inside PT lock */
> -	if (userfaultfd_missing(vma)) {
> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE) && userfaultfd_missing(vma)) {
>  		pte_unmap_unlock(vmf->pte, vmf->ptl);
>  		mem_cgroup_cancel_charge(page, memcg, false);
>  		put_page(page);
> @@ -3994,13 +4116,22 @@ static int handle_pte_fault(struct vm_fault *vmf)
>  
>  	if (unlikely(pmd_none(*vmf->pmd))) {
>  		/*
> +		 * In the case of the speculative page fault handler we abort
> +		 * the speculative path immediately as the pmd is probably
> +		 * in the way to be converted in a huge one. We will try
> +		 * again holding the mmap_sem (which implies that the collapse
> +		 * operation is done).
> +		 */
> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
> +			return VM_FAULT_RETRY;
> +		/*
>  		 * Leave __pte_alloc() until later: because vm_ops->fault may
>  		 * want to allocate huge page, and if we expose page table
>  		 * for an instant, it will be difficult to retract from
>  		 * concurrent faults and from rmap lookups.
>  		 */
>  		vmf->pte = NULL;
> -	} else {
> +	} else if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>  		/* See comment in pte_alloc_one_map() */
>  		if (pmd_devmap_trans_unstable(vmf->pmd))
>  			return 0;
> @@ -4009,6 +4140,9 @@ static int handle_pte_fault(struct vm_fault *vmf)
>  		 * pmd from under us anymore at this point because we hold the
>  		 * mmap_sem read mode and khugepaged takes it in write mode.
>  		 * So now it's safe to run pte_offset_map().
> +		 * This is not applicable to the speculative page fault handler
> +		 * but in that case, the pte is fetched earlier in
> +		 * handle_speculative_fault().
>  		 */
>  		vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
>  		vmf->orig_pte = *vmf->pte;
> @@ -4031,6 +4165,8 @@ static int handle_pte_fault(struct vm_fault *vmf)
>  	if (!vmf->pte) {
>  		if (vma_is_anonymous(vmf->vma))
>  			return do_anonymous_page(vmf);
> +		else if (vmf->flags & FAULT_FLAG_SPECULATIVE)
> +			return VM_FAULT_RETRY;
>  		else
>  			return do_fault(vmf);
>  	}
> @@ -4128,6 +4264,9 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>  	vmf.pmd = pmd_alloc(mm, vmf.pud, address);
>  	if (!vmf.pmd)
>  		return VM_FAULT_OOM;
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +	vmf.sequence = raw_read_seqcount(&vma->vm_sequence);
> +#endif
>  	if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
>  		ret = create_huge_pmd(&vmf);
>  		if (!(ret & VM_FAULT_FALLBACK))
> @@ -4161,6 +4300,201 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>  	return handle_pte_fault(&vmf);
>  }
>  
> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
> +/*
> + * Tries to handle the page fault in a speculative way, without grabbing the
> + * mmap_sem.
> + */
> +int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
> +			       unsigned int flags)
> +{
> +	struct vm_fault vmf = {
> +		.address = address,
> +	};
> +	pgd_t *pgd, pgdval;
> +	p4d_t *p4d, p4dval;
> +	pud_t pudval;
> +	int seq, ret = VM_FAULT_RETRY;
> +	struct vm_area_struct *vma;
> +#ifdef CONFIG_NUMA
> +	struct mempolicy *pol;
> +#endif
> +
> +	/* Clear flags that may lead to release the mmap_sem to retry */
> +	flags &= ~(FAULT_FLAG_ALLOW_RETRY|FAULT_FLAG_KILLABLE);
> +	flags |= FAULT_FLAG_SPECULATIVE;
> +
> +	vma = get_vma(mm, address);
> +	if (!vma)
> +		return ret;
> +
> +	seq = raw_read_seqcount(&vma->vm_sequence); /* rmb <-> seqlock,vma_rb_erase() */
> +	if (seq & 1)
> +		goto out_put;
> +
> +	/*
> +	 * Can't call vm_ops service has we don't know what they would do
> +	 * with the VMA.
> +	 * This include huge page from hugetlbfs.
> +	 */
> +	if (vma->vm_ops)
> +		goto out_put;
> +
  Hi   Laurent
   
   I think that most of pagefault will leave here.   Is there any case  need to skip ?
  I have tested the following  patch, it work well.

diff --git a/mm/memory.c b/mm/memory.c
index 936128b..9bc1545 100644
 @@ -3893,8 +3898,6 @@ static int handle_pte_fault(struct fault_env *fe)
        if (!fe->pte) {
                if (vma_is_anonymous(fe->vma))
                        return do_anonymous_page(fe);
-               else if (fe->flags & FAULT_FLAG_SPECULATIVE)
-                       return VM_FAULT_RETRY;
                else
                        return do_fault(fe);
        }
@@ -4026,20 +4029,11 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
                goto out_put;
        }
        /*
-        * Can't call vm_ops service has we don't know what they would do
-        * with the VMA.
-        * This include huge page from hugetlbfs.
-        */
-       if (vma->vm_ops) {
-               trace_spf_vma_notsup(_RET_IP_, vma, address);
-               goto out_put;
-       }


Thanks
zhong jiang
> +	/*
> +	 * __anon_vma_prepare() requires the mmap_sem to be held
> +	 * because vm_next and vm_prev must be safe. This can't be guaranteed
> +	 * in the speculative path.
> +	 */
> +	if (unlikely(!vma->anon_vma))
> +		goto out_put;
> +
> +	vmf.vma_flags = READ_ONCE(vma->vm_flags);
> +	vmf.vma_page_prot = READ_ONCE(vma->vm_page_prot);
> +
> +	/* Can't call userland page fault handler in the speculative path */
> +	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING))
> +		goto out_put;
> +
> +	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP)
> +		/*
> +		 * This could be detected by the check address against VMA's
> +		 * boundaries but we want to trace it as not supported instead
> +		 * of changed.
> +		 */
> +		goto out_put;
> +
> +	if (address < READ_ONCE(vma->vm_start)
> +	    || READ_ONCE(vma->vm_end) <= address)
> +		goto out_put;
> +
> +	if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
> +				       flags & FAULT_FLAG_INSTRUCTION,
> +				       flags & FAULT_FLAG_REMOTE)) {
> +		ret = VM_FAULT_SIGSEGV;
> +		goto out_put;
> +	}
> +
> +	/* This is one is required to check that the VMA has write access set */
> +	if (flags & FAULT_FLAG_WRITE) {
> +		if (unlikely(!(vmf.vma_flags & VM_WRITE))) {
> +			ret = VM_FAULT_SIGSEGV;
> +			goto out_put;
> +		}
> +	} else if (unlikely(!(vmf.vma_flags & (VM_READ|VM_EXEC|VM_WRITE)))) {
> +		ret = VM_FAULT_SIGSEGV;
> +		goto out_put;
> +	}
> +
> +#ifdef CONFIG_NUMA
> +	/*
> +	 * MPOL_INTERLEAVE implies additional checks in
> +	 * mpol_misplaced() which are not compatible with the
> +	 *speculative page fault processing.
> +	 */
> +	pol = __get_vma_policy(vma, address);
> +	if (!pol)
> +		pol = get_task_policy(current);
> +	if (pol && pol->mode == MPOL_INTERLEAVE)
> +		goto out_put;
> +#endif
> +
> +	/*
> +	 * Do a speculative lookup of the PTE entry.
> +	 */
> +	local_irq_disable();
> +	pgd = pgd_offset(mm, address);
> +	pgdval = READ_ONCE(*pgd);
> +	if (pgd_none(pgdval) || unlikely(pgd_bad(pgdval)))
> +		goto out_walk;
> +
> +	p4d = p4d_offset(pgd, address);
> +	p4dval = READ_ONCE(*p4d);
> +	if (p4d_none(p4dval) || unlikely(p4d_bad(p4dval)))
> +		goto out_walk;
> +
> +	vmf.pud = pud_offset(p4d, address);
> +	pudval = READ_ONCE(*vmf.pud);
> +	if (pud_none(pudval) || unlikely(pud_bad(pudval)))
> +		goto out_walk;
> +
> +	/* Huge pages at PUD level are not supported. */
> +	if (unlikely(pud_trans_huge(pudval)))
> +		goto out_walk;
> +
> +	vmf.pmd = pmd_offset(vmf.pud, address);
> +	vmf.orig_pmd = READ_ONCE(*vmf.pmd);
> +	/*
> +	 * pmd_none could mean that a hugepage collapse is in progress
> +	 * in our back as collapse_huge_page() mark it before
> +	 * invalidating the pte (which is done once the IPI is catched
> +	 * by all CPU and we have interrupt disabled).
> +	 * For this reason we cannot handle THP in a speculative way since we
> +	 * can't safely indentify an in progress collapse operation done in our
> +	 * back on that PMD.
> +	 * Regarding the order of the following checks, see comment in
> +	 * pmd_devmap_trans_unstable()
> +	 */
> +	if (unlikely(pmd_devmap(vmf.orig_pmd) ||
> +		     pmd_none(vmf.orig_pmd) || pmd_trans_huge(vmf.orig_pmd) ||
> +		     is_swap_pmd(vmf.orig_pmd)))
> +		goto out_walk;
> +
> +	/*
> +	 * The above does not allocate/instantiate page-tables because doing so
> +	 * would lead to the possibility of instantiating page-tables after
> +	 * free_pgtables() -- and consequently leaking them.
> +	 *
> +	 * The result is that we take at least one !speculative fault per PMD
> +	 * in order to instantiate it.
> +	 */
> +
> +	vmf.pte = pte_offset_map(vmf.pmd, address);
> +	vmf.orig_pte = READ_ONCE(*vmf.pte);
> +	barrier(); /* See comment in handle_pte_fault() */
> +	if (pte_none(vmf.orig_pte)) {
> +		pte_unmap(vmf.pte);
> +		vmf.pte = NULL;
> +	}
> +
> +	vmf.vma = vma;
> +	vmf.pgoff = linear_page_index(vma, address);
> +	vmf.gfp_mask = __get_fault_gfp_mask(vma);
> +	vmf.sequence = seq;
> +	vmf.flags = flags;
> +
> +	local_irq_enable();
> +
> +	/*
> +	 * We need to re-validate the VMA after checking the bounds, otherwise
> +	 * we might have a false positive on the bounds.
> +	 */
> +	if (read_seqcount_retry(&vma->vm_sequence, seq))
> +		goto out_put;
> +
> +	mem_cgroup_oom_enable();
> +	ret = handle_pte_fault(&vmf);
> +	mem_cgroup_oom_disable();
> +
> +	put_vma(vma);
> +
> +	/*
> +	 * The task may have entered a memcg OOM situation but
> +	 * if the allocation error was handled gracefully (no
> +	 * VM_FAULT_OOM), there is no need to kill anything.
> +	 * Just clean up the OOM state peacefully.
> +	 */
> +	if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
> +		mem_cgroup_oom_synchronize(false);
> +	return ret;
> +
> +out_walk:
> +	local_irq_enable();
> +out_put:
> +	put_vma(vma);
> +	return ret;
> +}
> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
> +
>  /*
>   * By the time we get here, we already hold the mm semaphore
>   *

Re: [PATCH v11 19/26] mm: provide speculative fault infrastructure

[Laurent Dufour]

On 24/07/2018 16:26, zhong jiang wrote:
> On 2018/5/17 19:06, Laurent Dufour wrote:
>> From: Peter Zijlstra <peterz@infradead.org>
>>
>> Provide infrastructure to do a speculative fault (not holding
>> mmap_sem).
>>
>> The not holding of mmap_sem means we can race against VMA
>> change/removal and page-table destruction. We use the SRCU VMA freeing
>> to keep the VMA around. We use the VMA seqcount to detect change
>> (including umapping / page-table deletion) and we use gup_fast() style
>> page-table walking to deal with page-table races.
>>
>> Once we've obtained the page and are ready to update the PTE, we
>> validate if the state we started the fault with is still valid, if
>> not, we'll fail the fault with VM_FAULT_RETRY, otherwise we update the
>> PTE and we're done.
>>
>> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>>
>> [Manage the newly introduced pte_spinlock() for speculative page
>>  fault to fail if the VMA is touched in our back]
>> [Rename vma_is_dead() to vma_has_changed() and declare it here]
>> [Fetch p4d and pud]
>> [Set vmd.sequence in __handle_mm_fault()]
>> [Abort speculative path when handle_userfault() has to be called]
>> [Add additional VMA's flags checks in handle_speculative_fault()]
>> [Clear FAULT_FLAG_ALLOW_RETRY in handle_speculative_fault()]
>> [Don't set vmf->pte and vmf->ptl if pte_map_lock() failed]
>> [Remove warning comment about waiting for !seq&1 since we don't want
>>  to wait]
>> [Remove warning about no huge page support, mention it explictly]
>> [Don't call do_fault() in the speculative path as __do_fault() calls
>>  vma->vm_ops->fault() which may want to release mmap_sem]
>> [Only vm_fault pointer argument for vma_has_changed()]
>> [Fix check against huge page, calling pmd_trans_huge()]
>> [Use READ_ONCE() when reading VMA's fields in the speculative path]
>> [Explicitly check for __HAVE_ARCH_PTE_SPECIAL as we can't support for
>>  processing done in vm_normal_page()]
>> [Check that vma->anon_vma is already set when starting the speculative
>>  path]
>> [Check for memory policy as we can't support MPOL_INTERLEAVE case due to
>>  the processing done in mpol_misplaced()]
>> [Don't support VMA growing up or down]
>> [Move check on vm_sequence just before calling handle_pte_fault()]
>> [Don't build SPF services if !CONFIG_SPECULATIVE_PAGE_FAULT]
>> [Add mem cgroup oom check]
>> [Use READ_ONCE to access p*d entries]
>> [Replace deprecated ACCESS_ONCE() by READ_ONCE() in vma_has_changed()]
>> [Don't fetch pte again in handle_pte_fault() when running the speculative
>>  path]
>> [Check PMD against concurrent collapsing operation]
>> [Try spin lock the pte during the speculative path to avoid deadlock with
>>  other CPU's invalidating the TLB and requiring this CPU to catch the
>>  inter processor's interrupt]
>> [Move define of FAULT_FLAG_SPECULATIVE here]
>> [Introduce __handle_speculative_fault() and add a check against
>>  mm->mm_users in handle_speculative_fault() defined in mm.h]
>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>> ---
>>  include/linux/hugetlb_inline.h |   2 +-
>>  include/linux/mm.h             |  30 ++++
>>  include/linux/pagemap.h        |   4 +-
>>  mm/internal.h                  |  16 +-
>>  mm/memory.c                    | 340 ++++++++++++++++++++++++++++++++++++++++-
>>  5 files changed, 385 insertions(+), 7 deletions(-)
>>
>> diff --git a/include/linux/hugetlb_inline.h b/include/linux/hugetlb_inline.h
>> index 0660a03d37d9..9e25283d6fc9 100644
>> --- a/include/linux/hugetlb_inline.h
>> +++ b/include/linux/hugetlb_inline.h
>> @@ -8,7 +8,7 @@
>>  
>>  static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
>>  {
>> -	return !!(vma->vm_flags & VM_HUGETLB);
>> +	return !!(READ_ONCE(vma->vm_flags) & VM_HUGETLB);
>>  }
>>  
>>  #else
>> diff --git a/include/linux/mm.h b/include/linux/mm.h
>> index 05cbba70104b..31acf98a7d92 100644
>> --- a/include/linux/mm.h
>> +++ b/include/linux/mm.h
>> @@ -315,6 +315,7 @@ extern pgprot_t protection_map[16];
>>  #define FAULT_FLAG_USER		0x40	/* The fault originated in userspace */
>>  #define FAULT_FLAG_REMOTE	0x80	/* faulting for non current tsk/mm */
>>  #define FAULT_FLAG_INSTRUCTION  0x100	/* The fault was during an instruction fetch */
>> +#define FAULT_FLAG_SPECULATIVE	0x200	/* Speculative fault, not holding mmap_sem */
>>  
>>  #define FAULT_FLAG_TRACE \
>>  	{ FAULT_FLAG_WRITE,		"WRITE" }, \
>> @@ -343,6 +344,10 @@ struct vm_fault {
>>  	gfp_t gfp_mask;			/* gfp mask to be used for allocations */
>>  	pgoff_t pgoff;			/* Logical page offset based on vma */
>>  	unsigned long address;		/* Faulting virtual address */
>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>> +	unsigned int sequence;
>> +	pmd_t orig_pmd;			/* value of PMD at the time of fault */
>> +#endif
>>  	pmd_t *pmd;			/* Pointer to pmd entry matching
>>  					 * the 'address' */
>>  	pud_t *pud;			/* Pointer to pud entry matching
>> @@ -1415,6 +1420,31 @@ int invalidate_inode_page(struct page *page);
>>  #ifdef CONFIG_MMU
>>  extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>  		unsigned int flags);
>> +
>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>> +extern int __handle_speculative_fault(struct mm_struct *mm,
>> +				      unsigned long address,
>> +				      unsigned int flags);
>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>> +					   unsigned long address,
>> +					   unsigned int flags)
>> +{
>> +	/*
>> +	 * Try speculative page fault for multithreaded user space task only.
>> +	 */
>> +	if (!(flags & FAULT_FLAG_USER) || atomic_read(&mm->mm_users) == 1)
>> +		return VM_FAULT_RETRY;
>> +	return __handle_speculative_fault(mm, address, flags);
>> +}
>> +#else
>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>> +					   unsigned long address,
>> +					   unsigned int flags)
>> +{
>> +	return VM_FAULT_RETRY;
>> +}
>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>> +
>>  extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
>>  			    unsigned long address, unsigned int fault_flags,
>>  			    bool *unlocked);
>> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
>> index b1bd2186e6d2..6e2aa4e79af7 100644
>> --- a/include/linux/pagemap.h
>> +++ b/include/linux/pagemap.h
>> @@ -456,8 +456,8 @@ static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
>>  	pgoff_t pgoff;
>>  	if (unlikely(is_vm_hugetlb_page(vma)))
>>  		return linear_hugepage_index(vma, address);
>> -	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
>> -	pgoff += vma->vm_pgoff;
>> +	pgoff = (address - READ_ONCE(vma->vm_start)) >> PAGE_SHIFT;
>> +	pgoff += READ_ONCE(vma->vm_pgoff);
>>  	return pgoff;
>>  }
>>  
>> diff --git a/mm/internal.h b/mm/internal.h
>> index fb2667b20f0a..10b188c87fa4 100644
>> --- a/mm/internal.h
>> +++ b/mm/internal.h
>> @@ -44,7 +44,21 @@ int do_swap_page(struct vm_fault *vmf);
>>  extern struct vm_area_struct *get_vma(struct mm_struct *mm,
>>  				      unsigned long addr);
>>  extern void put_vma(struct vm_area_struct *vma);
>> -#endif
>> +
>> +static inline bool vma_has_changed(struct vm_fault *vmf)
>> +{
>> +	int ret = RB_EMPTY_NODE(&vmf->vma->vm_rb);
>> +	unsigned int seq = READ_ONCE(vmf->vma->vm_sequence.sequence);
>> +
>> +	/*
>> +	 * Matches both the wmb in write_seqlock_{begin,end}() and
>> +	 * the wmb in vma_rb_erase().
>> +	 */
>> +	smp_rmb();
>> +
>> +	return ret || seq != vmf->sequence;
>> +}
>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>  
>>  void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
>>  		unsigned long floor, unsigned long ceiling);
>> diff --git a/mm/memory.c b/mm/memory.c
>> index ab32b0b4bd69..7bbbb8c7b9cd 100644
>> --- a/mm/memory.c
>> +++ b/mm/memory.c
>> @@ -769,7 +769,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
>>  	if (page)
>>  		dump_page(page, "bad pte");
>>  	pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
>> -		 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
>> +		 (void *)addr, READ_ONCE(vma->vm_flags), vma->anon_vma,
>> +		 mapping, index);
>>  	pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
>>  		 vma->vm_file,
>>  		 vma->vm_ops ? vma->vm_ops->fault : NULL,
>> @@ -2306,6 +2307,118 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
>>  }
>>  EXPORT_SYMBOL_GPL(apply_to_page_range);
>>  
>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>> +static bool pte_spinlock(struct vm_fault *vmf)
>> +{
>> +	bool ret = false;
>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>> +	pmd_t pmdval;
>> +#endif
>> +
>> +	/* Check if vma is still valid */
>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>> +		vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>> +		spin_lock(vmf->ptl);
>> +		return true;
>> +	}
>> +
>> +again:
>> +	local_irq_disable();
>> +	if (vma_has_changed(vmf))
>> +		goto out;
>> +
>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>> +	/*
>> +	 * We check if the pmd value is still the same to ensure that there
>> +	 * is not a huge collapse operation in progress in our back.
>> +	 */
>> +	pmdval = READ_ONCE(*vmf->pmd);
>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>> +		goto out;
>> +#endif
>> +
>> +	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>> +	if (unlikely(!spin_trylock(vmf->ptl))) {
>> +		local_irq_enable();
>> +		goto again;
>> +	}
>> +
>> +	if (vma_has_changed(vmf)) {
>> +		spin_unlock(vmf->ptl);
>> +		goto out;
>> +	}
>> +
>> +	ret = true;
>> +out:
>> +	local_irq_enable();
>> +	return ret;
>> +}
>> +
>> +static bool pte_map_lock(struct vm_fault *vmf)
>> +{
>> +	bool ret = false;
>> +	pte_t *pte;
>> +	spinlock_t *ptl;
>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>> +	pmd_t pmdval;
>> +#endif
>> +
>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>> +		vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
>> +					       vmf->address, &vmf->ptl);
>> +		return true;
>> +	}
>> +
>> +	/*
>> +	 * The first vma_has_changed() guarantees the page-tables are still
>> +	 * valid, having IRQs disabled ensures they stay around, hence the
>> +	 * second vma_has_changed() to make sure they are still valid once
>> +	 * we've got the lock. After that a concurrent zap_pte_range() will
>> +	 * block on the PTL and thus we're safe.
>> +	 */
>> +again:
>> +	local_irq_disable();
>> +	if (vma_has_changed(vmf))
>> +		goto out;
>> +
>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>> +	/*
>> +	 * We check if the pmd value is still the same to ensure that there
>> +	 * is not a huge collapse operation in progress in our back.
>> +	 */
>> +	pmdval = READ_ONCE(*vmf->pmd);
>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>> +		goto out;
>> +#endif
>> +
>> +	/*
>> +	 * Same as pte_offset_map_lock() except that we call
>> +	 * spin_trylock() in place of spin_lock() to avoid race with
>> +	 * unmap path which may have the lock and wait for this CPU
>> +	 * to invalidate TLB but this CPU has irq disabled.
>> +	 * Since we are in a speculative patch, accept it could fail
>> +	 */
>> +	ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>> +	pte = pte_offset_map(vmf->pmd, vmf->address);
>> +	if (unlikely(!spin_trylock(ptl))) {
>> +		pte_unmap(pte);
>> +		local_irq_enable();
>> +		goto again;
>> +	}
>> +
>> +	if (vma_has_changed(vmf)) {
>> +		pte_unmap_unlock(pte, ptl);
>> +		goto out;
>> +	}
>> +
>> +	vmf->pte = pte;
>> +	vmf->ptl = ptl;
>> +	ret = true;
>> +out:
>> +	local_irq_enable();
>> +	return ret;
>> +}
>> +#else
>>  static inline bool pte_spinlock(struct vm_fault *vmf)
>>  {
>>  	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>> @@ -2319,6 +2432,7 @@ static inline bool pte_map_lock(struct vm_fault *vmf)
>>  				       vmf->address, &vmf->ptl);
>>  	return true;
>>  }
>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>  
>>  /*
>>   * handle_pte_fault chooses page fault handler according to an entry which was
>> @@ -3208,6 +3322,14 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>  		ret = check_stable_address_space(vma->vm_mm);
>>  		if (ret)
>>  			goto unlock;
>> +		/*
>> +		 * Don't call the userfaultfd during the speculative path.
>> +		 * We already checked for the VMA to not be managed through
>> +		 * userfaultfd, but it may be set in our back once we have lock
>> +		 * the pte. In such a case we can ignore it this time.
>> +		 */
>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>> +			goto setpte;
>>  		/* Deliver the page fault to userland, check inside PT lock */
>>  		if (userfaultfd_missing(vma)) {
>>  			pte_unmap_unlock(vmf->pte, vmf->ptl);
>> @@ -3249,7 +3371,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>  		goto unlock_and_release;
>>  
>>  	/* Deliver the page fault to userland, check inside PT lock */
>> -	if (userfaultfd_missing(vma)) {
>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE) && userfaultfd_missing(vma)) {
>>  		pte_unmap_unlock(vmf->pte, vmf->ptl);
>>  		mem_cgroup_cancel_charge(page, memcg, false);
>>  		put_page(page);
>> @@ -3994,13 +4116,22 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>  
>>  	if (unlikely(pmd_none(*vmf->pmd))) {
>>  		/*
>> +		 * In the case of the speculative page fault handler we abort
>> +		 * the speculative path immediately as the pmd is probably
>> +		 * in the way to be converted in a huge one. We will try
>> +		 * again holding the mmap_sem (which implies that the collapse
>> +		 * operation is done).
>> +		 */
>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>> +			return VM_FAULT_RETRY;
>> +		/*
>>  		 * Leave __pte_alloc() until later: because vm_ops->fault may
>>  		 * want to allocate huge page, and if we expose page table
>>  		 * for an instant, it will be difficult to retract from
>>  		 * concurrent faults and from rmap lookups.
>>  		 */
>>  		vmf->pte = NULL;
>> -	} else {
>> +	} else if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>  		/* See comment in pte_alloc_one_map() */
>>  		if (pmd_devmap_trans_unstable(vmf->pmd))
>>  			return 0;
>> @@ -4009,6 +4140,9 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>  		 * pmd from under us anymore at this point because we hold the
>>  		 * mmap_sem read mode and khugepaged takes it in write mode.
>>  		 * So now it's safe to run pte_offset_map().
>> +		 * This is not applicable to the speculative page fault handler
>> +		 * but in that case, the pte is fetched earlier in
>> +		 * handle_speculative_fault().
>>  		 */
>>  		vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
>>  		vmf->orig_pte = *vmf->pte;
>> @@ -4031,6 +4165,8 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>  	if (!vmf->pte) {
>>  		if (vma_is_anonymous(vmf->vma))
>>  			return do_anonymous_page(vmf);
>> +		else if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>> +			return VM_FAULT_RETRY;
>>  		else
>>  			return do_fault(vmf);
>>  	}
>> @@ -4128,6 +4264,9 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>  	vmf.pmd = pmd_alloc(mm, vmf.pud, address);
>>  	if (!vmf.pmd)
>>  		return VM_FAULT_OOM;
>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>> +	vmf.sequence = raw_read_seqcount(&vma->vm_sequence);
>> +#endif
>>  	if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
>>  		ret = create_huge_pmd(&vmf);
>>  		if (!(ret & VM_FAULT_FALLBACK))
>> @@ -4161,6 +4300,201 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>  	return handle_pte_fault(&vmf);
>>  }
>>  
>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>> +/*
>> + * Tries to handle the page fault in a speculative way, without grabbing the
>> + * mmap_sem.
>> + */
>> +int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>> +			       unsigned int flags)
>> +{
>> +	struct vm_fault vmf = {
>> +		.address = address,
>> +	};
>> +	pgd_t *pgd, pgdval;
>> +	p4d_t *p4d, p4dval;
>> +	pud_t pudval;
>> +	int seq, ret = VM_FAULT_RETRY;
>> +	struct vm_area_struct *vma;
>> +#ifdef CONFIG_NUMA
>> +	struct mempolicy *pol;
>> +#endif
>> +
>> +	/* Clear flags that may lead to release the mmap_sem to retry */
>> +	flags &= ~(FAULT_FLAG_ALLOW_RETRY|FAULT_FLAG_KILLABLE);
>> +	flags |= FAULT_FLAG_SPECULATIVE;
>> +
>> +	vma = get_vma(mm, address);
>> +	if (!vma)
>> +		return ret;
>> +
>> +	seq = raw_read_seqcount(&vma->vm_sequence); /* rmb <-> seqlock,vma_rb_erase() */
>> +	if (seq & 1)
>> +		goto out_put;
>> +
>> +	/*
>> +	 * Can't call vm_ops service has we don't know what they would do
>> +	 * with the VMA.
>> +	 * This include huge page from hugetlbfs.
>> +	 */
>> +	if (vma->vm_ops)
>> +		goto out_put;
>> +
>   Hi   Laurent
>    
>    I think that most of pagefault will leave here.   Is there any case  need to skip ?
>   I have tested the following  patch, it work well.

Hi Zhong,

Well this will allow file mapping to be handle in a speculative way, but that's
a bit dangerous today as there is no guaranty that the vm_ops.vm_fault()
operation will be fair.

In the case of the anonymous file mapping that's often not a problem, depending
on the underlying file system, but there are so many cases to check and this is
hard to say this can be done in a speculative way as is.

The huge work to do is to double check that all the code called by
vm_ops.fault() is not dealing with the mmap_sem, which could be handled using
FAULT_FLAG_RETRY_NOWAIT, and care is also needed about the resources that code
is managing as it may assume that it is under the protection of the mmap_sem in
read mode, and that can be done implicitly.

Cheers,
Laurent.

> 
> diff --git a/mm/memory.c b/mm/memory.c
> index 936128b..9bc1545 100644
>  @@ -3893,8 +3898,6 @@ static int handle_pte_fault(struct fault_env *fe)
>         if (!fe->pte) {
>                 if (vma_is_anonymous(fe->vma))
>                         return do_anonymous_page(fe);
> -               else if (fe->flags & FAULT_FLAG_SPECULATIVE)
> -                       return VM_FAULT_RETRY;
>                 else
>                         return do_fault(fe);
>         }
> @@ -4026,20 +4029,11 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>                 goto out_put;
>         }
>         /*
> -        * Can't call vm_ops service has we don't know what they would do
> -        * with the VMA.
> -        * This include huge page from hugetlbfs.
> -        */
> -       if (vma->vm_ops) {
> -               trace_spf_vma_notsup(_RET_IP_, vma, address);
> -               goto out_put;
> -       }
> 
> 
> Thanks
> zhong jiang
>> +	/*
>> +	 * __anon_vma_prepare() requires the mmap_sem to be held
>> +	 * because vm_next and vm_prev must be safe. This can't be guaranteed
>> +	 * in the speculative path.
>> +	 */
>> +	if (unlikely(!vma->anon_vma))
>> +		goto out_put;
>> +
>> +	vmf.vma_flags = READ_ONCE(vma->vm_flags);
>> +	vmf.vma_page_prot = READ_ONCE(vma->vm_page_prot);
>> +
>> +	/* Can't call userland page fault handler in the speculative path */
>> +	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING))
>> +		goto out_put;
>> +
>> +	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP)
>> +		/*
>> +		 * This could be detected by the check address against VMA's
>> +		 * boundaries but we want to trace it as not supported instead
>> +		 * of changed.
>> +		 */
>> +		goto out_put;
>> +
>> +	if (address < READ_ONCE(vma->vm_start)
>> +	    || READ_ONCE(vma->vm_end) <= address)
>> +		goto out_put;
>> +
>> +	if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
>> +				       flags & FAULT_FLAG_INSTRUCTION,
>> +				       flags & FAULT_FLAG_REMOTE)) {
>> +		ret = VM_FAULT_SIGSEGV;
>> +		goto out_put;
>> +	}
>> +
>> +	/* This is one is required to check that the VMA has write access set */
>> +	if (flags & FAULT_FLAG_WRITE) {
>> +		if (unlikely(!(vmf.vma_flags & VM_WRITE))) {
>> +			ret = VM_FAULT_SIGSEGV;
>> +			goto out_put;
>> +		}
>> +	} else if (unlikely(!(vmf.vma_flags & (VM_READ|VM_EXEC|VM_WRITE)))) {
>> +		ret = VM_FAULT_SIGSEGV;
>> +		goto out_put;
>> +	}
>> +
>> +#ifdef CONFIG_NUMA
>> +	/*
>> +	 * MPOL_INTERLEAVE implies additional checks in
>> +	 * mpol_misplaced() which are not compatible with the
>> +	 *speculative page fault processing.
>> +	 */
>> +	pol = __get_vma_policy(vma, address);
>> +	if (!pol)
>> +		pol = get_task_policy(current);
>> +	if (pol && pol->mode == MPOL_INTERLEAVE)
>> +		goto out_put;
>> +#endif
>> +
>> +	/*
>> +	 * Do a speculative lookup of the PTE entry.
>> +	 */
>> +	local_irq_disable();
>> +	pgd = pgd_offset(mm, address);
>> +	pgdval = READ_ONCE(*pgd);
>> +	if (pgd_none(pgdval) || unlikely(pgd_bad(pgdval)))
>> +		goto out_walk;
>> +
>> +	p4d = p4d_offset(pgd, address);
>> +	p4dval = READ_ONCE(*p4d);
>> +	if (p4d_none(p4dval) || unlikely(p4d_bad(p4dval)))
>> +		goto out_walk;
>> +
>> +	vmf.pud = pud_offset(p4d, address);
>> +	pudval = READ_ONCE(*vmf.pud);
>> +	if (pud_none(pudval) || unlikely(pud_bad(pudval)))
>> +		goto out_walk;
>> +
>> +	/* Huge pages at PUD level are not supported. */
>> +	if (unlikely(pud_trans_huge(pudval)))
>> +		goto out_walk;
>> +
>> +	vmf.pmd = pmd_offset(vmf.pud, address);
>> +	vmf.orig_pmd = READ_ONCE(*vmf.pmd);
>> +	/*
>> +	 * pmd_none could mean that a hugepage collapse is in progress
>> +	 * in our back as collapse_huge_page() mark it before
>> +	 * invalidating the pte (which is done once the IPI is catched
>> +	 * by all CPU and we have interrupt disabled).
>> +	 * For this reason we cannot handle THP in a speculative way since we
>> +	 * can't safely indentify an in progress collapse operation done in our
>> +	 * back on that PMD.
>> +	 * Regarding the order of the following checks, see comment in
>> +	 * pmd_devmap_trans_unstable()
>> +	 */
>> +	if (unlikely(pmd_devmap(vmf.orig_pmd) ||
>> +		     pmd_none(vmf.orig_pmd) || pmd_trans_huge(vmf.orig_pmd) ||
>> +		     is_swap_pmd(vmf.orig_pmd)))
>> +		goto out_walk;
>> +
>> +	/*
>> +	 * The above does not allocate/instantiate page-tables because doing so
>> +	 * would lead to the possibility of instantiating page-tables after
>> +	 * free_pgtables() -- and consequently leaking them.
>> +	 *
>> +	 * The result is that we take at least one !speculative fault per PMD
>> +	 * in order to instantiate it.
>> +	 */
>> +
>> +	vmf.pte = pte_offset_map(vmf.pmd, address);
>> +	vmf.orig_pte = READ_ONCE(*vmf.pte);
>> +	barrier(); /* See comment in handle_pte_fault() */
>> +	if (pte_none(vmf.orig_pte)) {
>> +		pte_unmap(vmf.pte);
>> +		vmf.pte = NULL;
>> +	}
>> +
>> +	vmf.vma = vma;
>> +	vmf.pgoff = linear_page_index(vma, address);
>> +	vmf.gfp_mask = __get_fault_gfp_mask(vma);
>> +	vmf.sequence = seq;
>> +	vmf.flags = flags;
>> +
>> +	local_irq_enable();
>> +
>> +	/*
>> +	 * We need to re-validate the VMA after checking the bounds, otherwise
>> +	 * we might have a false positive on the bounds.
>> +	 */
>> +	if (read_seqcount_retry(&vma->vm_sequence, seq))
>> +		goto out_put;
>> +
>> +	mem_cgroup_oom_enable();
>> +	ret = handle_pte_fault(&vmf);
>> +	mem_cgroup_oom_disable();
>> +
>> +	put_vma(vma);
>> +
>> +	/*
>> +	 * The task may have entered a memcg OOM situation but
>> +	 * if the allocation error was handled gracefully (no
>> +	 * VM_FAULT_OOM), there is no need to kill anything.
>> +	 * Just clean up the OOM state peacefully.
>> +	 */
>> +	if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
>> +		mem_cgroup_oom_synchronize(false);
>> +	return ret;
>> +
>> +out_walk:
>> +	local_irq_enable();
>> +out_put:
>> +	put_vma(vma);
>> +	return ret;
>> +}
>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>> +
>>  /*
>>   * By the time we get here, we already hold the mm semaphore
>>   *
> 
> 

Re: [PATCH v11 19/26] mm: provide speculative fault infrastructure

[zhong jiang]

On 2018/7/25 0:10, Laurent Dufour wrote:
>
> On 24/07/2018 16:26, zhong jiang wrote:
>> On 2018/5/17 19:06, Laurent Dufour wrote:
>>> From: Peter Zijlstra <peterz@infradead.org>
>>>
>>> Provide infrastructure to do a speculative fault (not holding
>>> mmap_sem).
>>>
>>> The not holding of mmap_sem means we can race against VMA
>>> change/removal and page-table destruction. We use the SRCU VMA freeing
>>> to keep the VMA around. We use the VMA seqcount to detect change
>>> (including umapping / page-table deletion) and we use gup_fast() style
>>> page-table walking to deal with page-table races.
>>>
>>> Once we've obtained the page and are ready to update the PTE, we
>>> validate if the state we started the fault with is still valid, if
>>> not, we'll fail the fault with VM_FAULT_RETRY, otherwise we update the
>>> PTE and we're done.
>>>
>>> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>>>
>>> [Manage the newly introduced pte_spinlock() for speculative page
>>>  fault to fail if the VMA is touched in our back]
>>> [Rename vma_is_dead() to vma_has_changed() and declare it here]
>>> [Fetch p4d and pud]
>>> [Set vmd.sequence in __handle_mm_fault()]
>>> [Abort speculative path when handle_userfault() has to be called]
>>> [Add additional VMA's flags checks in handle_speculative_fault()]
>>> [Clear FAULT_FLAG_ALLOW_RETRY in handle_speculative_fault()]
>>> [Don't set vmf->pte and vmf->ptl if pte_map_lock() failed]
>>> [Remove warning comment about waiting for !seq&1 since we don't want
>>>  to wait]
>>> [Remove warning about no huge page support, mention it explictly]
>>> [Don't call do_fault() in the speculative path as __do_fault() calls
>>>  vma->vm_ops->fault() which may want to release mmap_sem]
>>> [Only vm_fault pointer argument for vma_has_changed()]
>>> [Fix check against huge page, calling pmd_trans_huge()]
>>> [Use READ_ONCE() when reading VMA's fields in the speculative path]
>>> [Explicitly check for __HAVE_ARCH_PTE_SPECIAL as we can't support for
>>>  processing done in vm_normal_page()]
>>> [Check that vma->anon_vma is already set when starting the speculative
>>>  path]
>>> [Check for memory policy as we can't support MPOL_INTERLEAVE case due to
>>>  the processing done in mpol_misplaced()]
>>> [Don't support VMA growing up or down]
>>> [Move check on vm_sequence just before calling handle_pte_fault()]
>>> [Don't build SPF services if !CONFIG_SPECULATIVE_PAGE_FAULT]
>>> [Add mem cgroup oom check]
>>> [Use READ_ONCE to access p*d entries]
>>> [Replace deprecated ACCESS_ONCE() by READ_ONCE() in vma_has_changed()]
>>> [Don't fetch pte again in handle_pte_fault() when running the speculative
>>>  path]
>>> [Check PMD against concurrent collapsing operation]
>>> [Try spin lock the pte during the speculative path to avoid deadlock with
>>>  other CPU's invalidating the TLB and requiring this CPU to catch the
>>>  inter processor's interrupt]
>>> [Move define of FAULT_FLAG_SPECULATIVE here]
>>> [Introduce __handle_speculative_fault() and add a check against
>>>  mm->mm_users in handle_speculative_fault() defined in mm.h]
>>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>>> ---
>>>  include/linux/hugetlb_inline.h |   2 +-
>>>  include/linux/mm.h             |  30 ++++
>>>  include/linux/pagemap.h        |   4 +-
>>>  mm/internal.h                  |  16 +-
>>>  mm/memory.c                    | 340 ++++++++++++++++++++++++++++++++++++++++-
>>>  5 files changed, 385 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/include/linux/hugetlb_inline.h b/include/linux/hugetlb_inline.h
>>> index 0660a03d37d9..9e25283d6fc9 100644
>>> --- a/include/linux/hugetlb_inline.h
>>> +++ b/include/linux/hugetlb_inline.h
>>> @@ -8,7 +8,7 @@
>>>  
>>>  static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
>>>  {
>>> -	return !!(vma->vm_flags & VM_HUGETLB);
>>> +	return !!(READ_ONCE(vma->vm_flags) & VM_HUGETLB);
>>>  }
>>>  
>>>  #else
>>> diff --git a/include/linux/mm.h b/include/linux/mm.h
>>> index 05cbba70104b..31acf98a7d92 100644
>>> --- a/include/linux/mm.h
>>> +++ b/include/linux/mm.h
>>> @@ -315,6 +315,7 @@ extern pgprot_t protection_map[16];
>>>  #define FAULT_FLAG_USER		0x40	/* The fault originated in userspace */
>>>  #define FAULT_FLAG_REMOTE	0x80	/* faulting for non current tsk/mm */
>>>  #define FAULT_FLAG_INSTRUCTION  0x100	/* The fault was during an instruction fetch */
>>> +#define FAULT_FLAG_SPECULATIVE	0x200	/* Speculative fault, not holding mmap_sem */
>>>  
>>>  #define FAULT_FLAG_TRACE \
>>>  	{ FAULT_FLAG_WRITE,		"WRITE" }, \
>>> @@ -343,6 +344,10 @@ struct vm_fault {
>>>  	gfp_t gfp_mask;			/* gfp mask to be used for allocations */
>>>  	pgoff_t pgoff;			/* Logical page offset based on vma */
>>>  	unsigned long address;		/* Faulting virtual address */
>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>> +	unsigned int sequence;
>>> +	pmd_t orig_pmd;			/* value of PMD at the time of fault */
>>> +#endif
>>>  	pmd_t *pmd;			/* Pointer to pmd entry matching
>>>  					 * the 'address' */
>>>  	pud_t *pud;			/* Pointer to pud entry matching
>>> @@ -1415,6 +1420,31 @@ int invalidate_inode_page(struct page *page);
>>>  #ifdef CONFIG_MMU
>>>  extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>  		unsigned int flags);
>>> +
>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>> +extern int __handle_speculative_fault(struct mm_struct *mm,
>>> +				      unsigned long address,
>>> +				      unsigned int flags);
>>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>>> +					   unsigned long address,
>>> +					   unsigned int flags)
>>> +{
>>> +	/*
>>> +	 * Try speculative page fault for multithreaded user space task only.
>>> +	 */
>>> +	if (!(flags & FAULT_FLAG_USER) || atomic_read(&mm->mm_users) == 1)
>>> +		return VM_FAULT_RETRY;
>>> +	return __handle_speculative_fault(mm, address, flags);
>>> +}
>>> +#else
>>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>>> +					   unsigned long address,
>>> +					   unsigned int flags)
>>> +{
>>> +	return VM_FAULT_RETRY;
>>> +}
>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>> +
>>>  extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
>>>  			    unsigned long address, unsigned int fault_flags,
>>>  			    bool *unlocked);
>>> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
>>> index b1bd2186e6d2..6e2aa4e79af7 100644
>>> --- a/include/linux/pagemap.h
>>> +++ b/include/linux/pagemap.h
>>> @@ -456,8 +456,8 @@ static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
>>>  	pgoff_t pgoff;
>>>  	if (unlikely(is_vm_hugetlb_page(vma)))
>>>  		return linear_hugepage_index(vma, address);
>>> -	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
>>> -	pgoff += vma->vm_pgoff;
>>> +	pgoff = (address - READ_ONCE(vma->vm_start)) >> PAGE_SHIFT;
>>> +	pgoff += READ_ONCE(vma->vm_pgoff);
>>>  	return pgoff;
>>>  }
>>>  
>>> diff --git a/mm/internal.h b/mm/internal.h
>>> index fb2667b20f0a..10b188c87fa4 100644
>>> --- a/mm/internal.h
>>> +++ b/mm/internal.h
>>> @@ -44,7 +44,21 @@ int do_swap_page(struct vm_fault *vmf);
>>>  extern struct vm_area_struct *get_vma(struct mm_struct *mm,
>>>  				      unsigned long addr);
>>>  extern void put_vma(struct vm_area_struct *vma);
>>> -#endif
>>> +
>>> +static inline bool vma_has_changed(struct vm_fault *vmf)
>>> +{
>>> +	int ret = RB_EMPTY_NODE(&vmf->vma->vm_rb);
>>> +	unsigned int seq = READ_ONCE(vmf->vma->vm_sequence.sequence);
>>> +
>>> +	/*
>>> +	 * Matches both the wmb in write_seqlock_{begin,end}() and
>>> +	 * the wmb in vma_rb_erase().
>>> +	 */
>>> +	smp_rmb();
>>> +
>>> +	return ret || seq != vmf->sequence;
>>> +}
>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>  
>>>  void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
>>>  		unsigned long floor, unsigned long ceiling);
>>> diff --git a/mm/memory.c b/mm/memory.c
>>> index ab32b0b4bd69..7bbbb8c7b9cd 100644
>>> --- a/mm/memory.c
>>> +++ b/mm/memory.c
>>> @@ -769,7 +769,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
>>>  	if (page)
>>>  		dump_page(page, "bad pte");
>>>  	pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
>>> -		 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
>>> +		 (void *)addr, READ_ONCE(vma->vm_flags), vma->anon_vma,
>>> +		 mapping, index);
>>>  	pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
>>>  		 vma->vm_file,
>>>  		 vma->vm_ops ? vma->vm_ops->fault : NULL,
>>> @@ -2306,6 +2307,118 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
>>>  }
>>>  EXPORT_SYMBOL_GPL(apply_to_page_range);
>>>  
>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>> +static bool pte_spinlock(struct vm_fault *vmf)
>>> +{
>>> +	bool ret = false;
>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>> +	pmd_t pmdval;
>>> +#endif
>>> +
>>> +	/* Check if vma is still valid */
>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>> +		vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>> +		spin_lock(vmf->ptl);
>>> +		return true;
>>> +	}
>>> +
>>> +again:
>>> +	local_irq_disable();
>>> +	if (vma_has_changed(vmf))
>>> +		goto out;
>>> +
>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>> +	/*
>>> +	 * We check if the pmd value is still the same to ensure that there
>>> +	 * is not a huge collapse operation in progress in our back.
>>> +	 */
>>> +	pmdval = READ_ONCE(*vmf->pmd);
>>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>>> +		goto out;
>>> +#endif
>>> +
>>> +	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>> +	if (unlikely(!spin_trylock(vmf->ptl))) {
>>> +		local_irq_enable();
>>> +		goto again;
>>> +	}
>>> +
>>> +	if (vma_has_changed(vmf)) {
>>> +		spin_unlock(vmf->ptl);
>>> +		goto out;
>>> +	}
>>> +
>>> +	ret = true;
>>> +out:
>>> +	local_irq_enable();
>>> +	return ret;
>>> +}
>>> +
>>> +static bool pte_map_lock(struct vm_fault *vmf)
>>> +{
>>> +	bool ret = false;
>>> +	pte_t *pte;
>>> +	spinlock_t *ptl;
>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>> +	pmd_t pmdval;
>>> +#endif
>>> +
>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>> +		vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
>>> +					       vmf->address, &vmf->ptl);
>>> +		return true;
>>> +	}
>>> +
>>> +	/*
>>> +	 * The first vma_has_changed() guarantees the page-tables are still
>>> +	 * valid, having IRQs disabled ensures they stay around, hence the
>>> +	 * second vma_has_changed() to make sure they are still valid once
>>> +	 * we've got the lock. After that a concurrent zap_pte_range() will
>>> +	 * block on the PTL and thus we're safe.
>>> +	 */
>>> +again:
>>> +	local_irq_disable();
>>> +	if (vma_has_changed(vmf))
>>> +		goto out;
>>> +
>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>> +	/*
>>> +	 * We check if the pmd value is still the same to ensure that there
>>> +	 * is not a huge collapse operation in progress in our back.
>>> +	 */
>>> +	pmdval = READ_ONCE(*vmf->pmd);
>>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>>> +		goto out;
>>> +#endif
>>> +
>>> +	/*
>>> +	 * Same as pte_offset_map_lock() except that we call
>>> +	 * spin_trylock() in place of spin_lock() to avoid race with
>>> +	 * unmap path which may have the lock and wait for this CPU
>>> +	 * to invalidate TLB but this CPU has irq disabled.
>>> +	 * Since we are in a speculative patch, accept it could fail
>>> +	 */
>>> +	ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>> +	pte = pte_offset_map(vmf->pmd, vmf->address);
>>> +	if (unlikely(!spin_trylock(ptl))) {
>>> +		pte_unmap(pte);
>>> +		local_irq_enable();
>>> +		goto again;
>>> +	}
>>> +
>>> +	if (vma_has_changed(vmf)) {
>>> +		pte_unmap_unlock(pte, ptl);
>>> +		goto out;
>>> +	}
>>> +
>>> +	vmf->pte = pte;
>>> +	vmf->ptl = ptl;
>>> +	ret = true;
>>> +out:
>>> +	local_irq_enable();
>>> +	return ret;
>>> +}
>>> +#else
>>>  static inline bool pte_spinlock(struct vm_fault *vmf)
>>>  {
>>>  	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>> @@ -2319,6 +2432,7 @@ static inline bool pte_map_lock(struct vm_fault *vmf)
>>>  				       vmf->address, &vmf->ptl);
>>>  	return true;
>>>  }
>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>  
>>>  /*
>>>   * handle_pte_fault chooses page fault handler according to an entry which was
>>> @@ -3208,6 +3322,14 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>>  		ret = check_stable_address_space(vma->vm_mm);
>>>  		if (ret)
>>>  			goto unlock;
>>> +		/*
>>> +		 * Don't call the userfaultfd during the speculative path.
>>> +		 * We already checked for the VMA to not be managed through
>>> +		 * userfaultfd, but it may be set in our back once we have lock
>>> +		 * the pte. In such a case we can ignore it this time.
>>> +		 */
>>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>> +			goto setpte;
>>>  		/* Deliver the page fault to userland, check inside PT lock */
>>>  		if (userfaultfd_missing(vma)) {
>>>  			pte_unmap_unlock(vmf->pte, vmf->ptl);
>>> @@ -3249,7 +3371,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>>  		goto unlock_and_release;
>>>  
>>>  	/* Deliver the page fault to userland, check inside PT lock */
>>> -	if (userfaultfd_missing(vma)) {
>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE) && userfaultfd_missing(vma)) {
>>>  		pte_unmap_unlock(vmf->pte, vmf->ptl);
>>>  		mem_cgroup_cancel_charge(page, memcg, false);
>>>  		put_page(page);
>>> @@ -3994,13 +4116,22 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>  
>>>  	if (unlikely(pmd_none(*vmf->pmd))) {
>>>  		/*
>>> +		 * In the case of the speculative page fault handler we abort
>>> +		 * the speculative path immediately as the pmd is probably
>>> +		 * in the way to be converted in a huge one. We will try
>>> +		 * again holding the mmap_sem (which implies that the collapse
>>> +		 * operation is done).
>>> +		 */
>>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>> +			return VM_FAULT_RETRY;
>>> +		/*
>>>  		 * Leave __pte_alloc() until later: because vm_ops->fault may
>>>  		 * want to allocate huge page, and if we expose page table
>>>  		 * for an instant, it will be difficult to retract from
>>>  		 * concurrent faults and from rmap lookups.
>>>  		 */
>>>  		vmf->pte = NULL;
>>> -	} else {
>>> +	} else if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>  		/* See comment in pte_alloc_one_map() */
>>>  		if (pmd_devmap_trans_unstable(vmf->pmd))
>>>  			return 0;
>>> @@ -4009,6 +4140,9 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>  		 * pmd from under us anymore at this point because we hold the
>>>  		 * mmap_sem read mode and khugepaged takes it in write mode.
>>>  		 * So now it's safe to run pte_offset_map().
>>> +		 * This is not applicable to the speculative page fault handler
>>> +		 * but in that case, the pte is fetched earlier in
>>> +		 * handle_speculative_fault().
>>>  		 */
>>>  		vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
>>>  		vmf->orig_pte = *vmf->pte;
>>> @@ -4031,6 +4165,8 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>  	if (!vmf->pte) {
>>>  		if (vma_is_anonymous(vmf->vma))
>>>  			return do_anonymous_page(vmf);
>>> +		else if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>> +			return VM_FAULT_RETRY;
>>>  		else
>>>  			return do_fault(vmf);
>>>  	}
>>> @@ -4128,6 +4264,9 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>  	vmf.pmd = pmd_alloc(mm, vmf.pud, address);
>>>  	if (!vmf.pmd)
>>>  		return VM_FAULT_OOM;
>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>> +	vmf.sequence = raw_read_seqcount(&vma->vm_sequence);
>>> +#endif
>>>  	if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
>>>  		ret = create_huge_pmd(&vmf);
>>>  		if (!(ret & VM_FAULT_FALLBACK))
>>> @@ -4161,6 +4300,201 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>  	return handle_pte_fault(&vmf);
>>>  }
>>>  
>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>> +/*
>>> + * Tries to handle the page fault in a speculative way, without grabbing the
>>> + * mmap_sem.
>>> + */
>>> +int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>> +			       unsigned int flags)
>>> +{
>>> +	struct vm_fault vmf = {
>>> +		.address = address,
>>> +	};
>>> +	pgd_t *pgd, pgdval;
>>> +	p4d_t *p4d, p4dval;
>>> +	pud_t pudval;
>>> +	int seq, ret = VM_FAULT_RETRY;
>>> +	struct vm_area_struct *vma;
>>> +#ifdef CONFIG_NUMA
>>> +	struct mempolicy *pol;
>>> +#endif
>>> +
>>> +	/* Clear flags that may lead to release the mmap_sem to retry */
>>> +	flags &= ~(FAULT_FLAG_ALLOW_RETRY|FAULT_FLAG_KILLABLE);
>>> +	flags |= FAULT_FLAG_SPECULATIVE;
>>> +
>>> +	vma = get_vma(mm, address);
>>> +	if (!vma)
>>> +		return ret;
>>> +
>>> +	seq = raw_read_seqcount(&vma->vm_sequence); /* rmb <-> seqlock,vma_rb_erase() */
>>> +	if (seq & 1)
>>> +		goto out_put;
>>> +
>>> +	/*
>>> +	 * Can't call vm_ops service has we don't know what they would do
>>> +	 * with the VMA.
>>> +	 * This include huge page from hugetlbfs.
>>> +	 */
>>> +	if (vma->vm_ops)
>>> +		goto out_put;
>>> +
>>   Hi   Laurent
>>    
>>    I think that most of pagefault will leave here.   Is there any case  need to skip ?
>>   I have tested the following  patch, it work well.
> Hi Zhong,
>
> Well this will allow file mapping to be handle in a speculative way, but that's
> a bit dangerous today as there is no guaranty that the vm_ops.vm_fault()
> operation will be fair.
>
> In the case of the anonymous file mapping that's often not a problem, depending
> on the underlying file system, but there are so many cases to check and this is
> hard to say this can be done in a speculative way as is.
 This patch say that spf just handle anonyous page. but I find that do_swap_page
 also maybe release the mmap_sem without FAULT_FLAG_RETRY_NOWAIT. why is it safe
 to handle the case. I think that the case is similar to file page. Maybe I miss
 something else.

 I test the patches and find just only 18% of the pagefault will enter into the
 speculative page fault during a process startup. As I had said. most of pagefault
 will be handled by ops->fault. I do not know the data you had posted is how to get.
 

 Thanks
 zhong jiang
> The huge work to do is to double check that all the code called by
> vm_ops.fault() is not dealing with the mmap_sem, which could be handled using
> FAULT_FLAG_RETRY_NOWAIT, and care is also needed about the resources that code
> is managing as it may assume that it is under the protection of the mmap_sem in
> read mode, and that can be done implicitly.
>
> Cheers,
> Laurent.
>
>> diff --git a/mm/memory.c b/mm/memory.c
>> index 936128b..9bc1545 100644
>>  @@ -3893,8 +3898,6 @@ static int handle_pte_fault(struct fault_env *fe)
>>         if (!fe->pte) {
>>                 if (vma_is_anonymous(fe->vma))
>>                         return do_anonymous_page(fe);
>> -               else if (fe->flags & FAULT_FLAG_SPECULATIVE)
>> -                       return VM_FAULT_RETRY;
>>                 else
>>                         return do_fault(fe);
>>         }
>> @@ -4026,20 +4029,11 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>                 goto out_put;
>>         }
>>         /*
>> -        * Can't call vm_ops service has we don't know what they would do
>> -        * with the VMA.
>> -        * This include huge page from hugetlbfs.
>> -        */
>> -       if (vma->vm_ops) {
>> -               trace_spf_vma_notsup(_RET_IP_, vma, address);
>> -               goto out_put;
>> -       }
>>
>>
>> Thanks
>> zhong jiang
>>> +	/*
>>> +	 * __anon_vma_prepare() requires the mmap_sem to be held
>>> +	 * because vm_next and vm_prev must be safe. This can't be guaranteed
>>> +	 * in the speculative path.
>>> +	 */
>>> +	if (unlikely(!vma->anon_vma))
>>> +		goto out_put;
>>> +
>>> +	vmf.vma_flags = READ_ONCE(vma->vm_flags);
>>> +	vmf.vma_page_prot = READ_ONCE(vma->vm_page_prot);
>>> +
>>> +	/* Can't call userland page fault handler in the speculative path */
>>> +	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING))
>>> +		goto out_put;
>>> +
>>> +	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP)
>>> +		/*
>>> +		 * This could be detected by the check address against VMA's
>>> +		 * boundaries but we want to trace it as not supported instead
>>> +		 * of changed.
>>> +		 */
>>> +		goto out_put;
>>> +
>>> +	if (address < READ_ONCE(vma->vm_start)
>>> +	    || READ_ONCE(vma->vm_end) <= address)
>>> +		goto out_put;
>>> +
>>> +	if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
>>> +				       flags & FAULT_FLAG_INSTRUCTION,
>>> +				       flags & FAULT_FLAG_REMOTE)) {
>>> +		ret = VM_FAULT_SIGSEGV;
>>> +		goto out_put;
>>> +	}
>>> +
>>> +	/* This is one is required to check that the VMA has write access set */
>>> +	if (flags & FAULT_FLAG_WRITE) {
>>> +		if (unlikely(!(vmf.vma_flags & VM_WRITE))) {
>>> +			ret = VM_FAULT_SIGSEGV;
>>> +			goto out_put;
>>> +		}
>>> +	} else if (unlikely(!(vmf.vma_flags & (VM_READ|VM_EXEC|VM_WRITE)))) {
>>> +		ret = VM_FAULT_SIGSEGV;
>>> +		goto out_put;
>>> +	}
>>> +
>>> +#ifdef CONFIG_NUMA
>>> +	/*
>>> +	 * MPOL_INTERLEAVE implies additional checks in
>>> +	 * mpol_misplaced() which are not compatible with the
>>> +	 *speculative page fault processing.
>>> +	 */
>>> +	pol = __get_vma_policy(vma, address);
>>> +	if (!pol)
>>> +		pol = get_task_policy(current);
>>> +	if (pol && pol->mode == MPOL_INTERLEAVE)
>>> +		goto out_put;
>>> +#endif
>>> +
>>> +	/*
>>> +	 * Do a speculative lookup of the PTE entry.
>>> +	 */
>>> +	local_irq_disable();
>>> +	pgd = pgd_offset(mm, address);
>>> +	pgdval = READ_ONCE(*pgd);
>>> +	if (pgd_none(pgdval) || unlikely(pgd_bad(pgdval)))
>>> +		goto out_walk;
>>> +
>>> +	p4d = p4d_offset(pgd, address);
>>> +	p4dval = READ_ONCE(*p4d);
>>> +	if (p4d_none(p4dval) || unlikely(p4d_bad(p4dval)))
>>> +		goto out_walk;
>>> +
>>> +	vmf.pud = pud_offset(p4d, address);
>>> +	pudval = READ_ONCE(*vmf.pud);
>>> +	if (pud_none(pudval) || unlikely(pud_bad(pudval)))
>>> +		goto out_walk;
>>> +
>>> +	/* Huge pages at PUD level are not supported. */
>>> +	if (unlikely(pud_trans_huge(pudval)))
>>> +		goto out_walk;
>>> +
>>> +	vmf.pmd = pmd_offset(vmf.pud, address);
>>> +	vmf.orig_pmd = READ_ONCE(*vmf.pmd);
>>> +	/*
>>> +	 * pmd_none could mean that a hugepage collapse is in progress
>>> +	 * in our back as collapse_huge_page() mark it before
>>> +	 * invalidating the pte (which is done once the IPI is catched
>>> +	 * by all CPU and we have interrupt disabled).
>>> +	 * For this reason we cannot handle THP in a speculative way since we
>>> +	 * can't safely indentify an in progress collapse operation done in our
>>> +	 * back on that PMD.
>>> +	 * Regarding the order of the following checks, see comment in
>>> +	 * pmd_devmap_trans_unstable()
>>> +	 */
>>> +	if (unlikely(pmd_devmap(vmf.orig_pmd) ||
>>> +		     pmd_none(vmf.orig_pmd) || pmd_trans_huge(vmf.orig_pmd) ||
>>> +		     is_swap_pmd(vmf.orig_pmd)))
>>> +		goto out_walk;
>>> +
>>> +	/*
>>> +	 * The above does not allocate/instantiate page-tables because doing so
>>> +	 * would lead to the possibility of instantiating page-tables after
>>> +	 * free_pgtables() -- and consequently leaking them.
>>> +	 *
>>> +	 * The result is that we take at least one !speculative fault per PMD
>>> +	 * in order to instantiate it.
>>> +	 */
>>> +
>>> +	vmf.pte = pte_offset_map(vmf.pmd, address);
>>> +	vmf.orig_pte = READ_ONCE(*vmf.pte);
>>> +	barrier(); /* See comment in handle_pte_fault() */
>>> +	if (pte_none(vmf.orig_pte)) {
>>> +		pte_unmap(vmf.pte);
>>> +		vmf.pte = NULL;
>>> +	}
>>> +
>>> +	vmf.vma = vma;
>>> +	vmf.pgoff = linear_page_index(vma, address);
>>> +	vmf.gfp_mask = __get_fault_gfp_mask(vma);
>>> +	vmf.sequence = seq;
>>> +	vmf.flags = flags;
>>> +
>>> +	local_irq_enable();
>>> +
>>> +	/*
>>> +	 * We need to re-validate the VMA after checking the bounds, otherwise
>>> +	 * we might have a false positive on the bounds.
>>> +	 */
>>> +	if (read_seqcount_retry(&vma->vm_sequence, seq))
>>> +		goto out_put;
>>> +
>>> +	mem_cgroup_oom_enable();
>>> +	ret = handle_pte_fault(&vmf);
>>> +	mem_cgroup_oom_disable();
>>> +
>>> +	put_vma(vma);
>>> +
>>> +	/*
>>> +	 * The task may have entered a memcg OOM situation but
>>> +	 * if the allocation error was handled gracefully (no
>>> +	 * VM_FAULT_OOM), there is no need to kill anything.
>>> +	 * Just clean up the OOM state peacefully.
>>> +	 */
>>> +	if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
>>> +		mem_cgroup_oom_synchronize(false);
>>> +	return ret;
>>> +
>>> +out_walk:
>>> +	local_irq_enable();
>>> +out_put:
>>> +	put_vma(vma);
>>> +	return ret;
>>> +}
>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>> +
>>>  /*
>>>   * By the time we get here, we already hold the mm semaphore
>>>   *
>>
>
> .
>

Re: [PATCH v11 19/26] mm: provide speculative fault infrastructure

[zhong jiang]

On 2018/7/25 0:10, Laurent Dufour wrote:
>
> On 24/07/2018 16:26, zhong jiang wrote:
>> On 2018/5/17 19:06, Laurent Dufour wrote:
>>> From: Peter Zijlstra <peterz@infradead.org>
>>>
>>> Provide infrastructure to do a speculative fault (not holding
>>> mmap_sem).
>>>
>>> The not holding of mmap_sem means we can race against VMA
>>> change/removal and page-table destruction. We use the SRCU VMA freeing
>>> to keep the VMA around. We use the VMA seqcount to detect change
>>> (including umapping / page-table deletion) and we use gup_fast() style
>>> page-table walking to deal with page-table races.
>>>
>>> Once we've obtained the page and are ready to update the PTE, we
>>> validate if the state we started the fault with is still valid, if
>>> not, we'll fail the fault with VM_FAULT_RETRY, otherwise we update the
>>> PTE and we're done.
>>>
>>> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>>>
>>> [Manage the newly introduced pte_spinlock() for speculative page
>>>  fault to fail if the VMA is touched in our back]
>>> [Rename vma_is_dead() to vma_has_changed() and declare it here]
>>> [Fetch p4d and pud]
>>> [Set vmd.sequence in __handle_mm_fault()]
>>> [Abort speculative path when handle_userfault() has to be called]
>>> [Add additional VMA's flags checks in handle_speculative_fault()]
>>> [Clear FAULT_FLAG_ALLOW_RETRY in handle_speculative_fault()]
>>> [Don't set vmf->pte and vmf->ptl if pte_map_lock() failed]
>>> [Remove warning comment about waiting for !seq&1 since we don't want
>>>  to wait]
>>> [Remove warning about no huge page support, mention it explictly]
>>> [Don't call do_fault() in the speculative path as __do_fault() calls
>>>  vma->vm_ops->fault() which may want to release mmap_sem]
>>> [Only vm_fault pointer argument for vma_has_changed()]
>>> [Fix check against huge page, calling pmd_trans_huge()]
>>> [Use READ_ONCE() when reading VMA's fields in the speculative path]
>>> [Explicitly check for __HAVE_ARCH_PTE_SPECIAL as we can't support for
>>>  processing done in vm_normal_page()]
>>> [Check that vma->anon_vma is already set when starting the speculative
>>>  path]
>>> [Check for memory policy as we can't support MPOL_INTERLEAVE case due to
>>>  the processing done in mpol_misplaced()]
>>> [Don't support VMA growing up or down]
>>> [Move check on vm_sequence just before calling handle_pte_fault()]
>>> [Don't build SPF services if !CONFIG_SPECULATIVE_PAGE_FAULT]
>>> [Add mem cgroup oom check]
>>> [Use READ_ONCE to access p*d entries]
>>> [Replace deprecated ACCESS_ONCE() by READ_ONCE() in vma_has_changed()]
>>> [Don't fetch pte again in handle_pte_fault() when running the speculative
>>>  path]
>>> [Check PMD against concurrent collapsing operation]
>>> [Try spin lock the pte during the speculative path to avoid deadlock with
>>>  other CPU's invalidating the TLB and requiring this CPU to catch the
>>>  inter processor's interrupt]
>>> [Move define of FAULT_FLAG_SPECULATIVE here]
>>> [Introduce __handle_speculative_fault() and add a check against
>>>  mm->mm_users in handle_speculative_fault() defined in mm.h]
>>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>>> ---
>>>  include/linux/hugetlb_inline.h |   2 +-
>>>  include/linux/mm.h             |  30 ++++
>>>  include/linux/pagemap.h        |   4 +-
>>>  mm/internal.h                  |  16 +-
>>>  mm/memory.c                    | 340 ++++++++++++++++++++++++++++++++++++++++-
>>>  5 files changed, 385 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/include/linux/hugetlb_inline.h b/include/linux/hugetlb_inline.h
>>> index 0660a03d37d9..9e25283d6fc9 100644
>>> --- a/include/linux/hugetlb_inline.h
>>> +++ b/include/linux/hugetlb_inline.h
>>> @@ -8,7 +8,7 @@
>>>  
>>>  static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
>>>  {
>>> -	return !!(vma->vm_flags & VM_HUGETLB);
>>> +	return !!(READ_ONCE(vma->vm_flags) & VM_HUGETLB);
>>>  }
>>>  
>>>  #else
>>> diff --git a/include/linux/mm.h b/include/linux/mm.h
>>> index 05cbba70104b..31acf98a7d92 100644
>>> --- a/include/linux/mm.h
>>> +++ b/include/linux/mm.h
>>> @@ -315,6 +315,7 @@ extern pgprot_t protection_map[16];
>>>  #define FAULT_FLAG_USER		0x40	/* The fault originated in userspace */
>>>  #define FAULT_FLAG_REMOTE	0x80	/* faulting for non current tsk/mm */
>>>  #define FAULT_FLAG_INSTRUCTION  0x100	/* The fault was during an instruction fetch */
>>> +#define FAULT_FLAG_SPECULATIVE	0x200	/* Speculative fault, not holding mmap_sem */
>>>  
>>>  #define FAULT_FLAG_TRACE \
>>>  	{ FAULT_FLAG_WRITE,		"WRITE" }, \
>>> @@ -343,6 +344,10 @@ struct vm_fault {
>>>  	gfp_t gfp_mask;			/* gfp mask to be used for allocations */
>>>  	pgoff_t pgoff;			/* Logical page offset based on vma */
>>>  	unsigned long address;		/* Faulting virtual address */
>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>> +	unsigned int sequence;
>>> +	pmd_t orig_pmd;			/* value of PMD at the time of fault */
>>> +#endif
>>>  	pmd_t *pmd;			/* Pointer to pmd entry matching
>>>  					 * the 'address' */
>>>  	pud_t *pud;			/* Pointer to pud entry matching
>>> @@ -1415,6 +1420,31 @@ int invalidate_inode_page(struct page *page);
>>>  #ifdef CONFIG_MMU
>>>  extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>  		unsigned int flags);
>>> +
>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>> +extern int __handle_speculative_fault(struct mm_struct *mm,
>>> +				      unsigned long address,
>>> +				      unsigned int flags);
>>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>>> +					   unsigned long address,
>>> +					   unsigned int flags)
>>> +{
>>> +	/*
>>> +	 * Try speculative page fault for multithreaded user space task only.
>>> +	 */
>>> +	if (!(flags & FAULT_FLAG_USER) || atomic_read(&mm->mm_users) == 1)
>>> +		return VM_FAULT_RETRY;
>>> +	return __handle_speculative_fault(mm, address, flags);
>>> +}
>>> +#else
>>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>>> +					   unsigned long address,
>>> +					   unsigned int flags)
>>> +{
>>> +	return VM_FAULT_RETRY;
>>> +}
>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>> +
>>>  extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
>>>  			    unsigned long address, unsigned int fault_flags,
>>>  			    bool *unlocked);
>>> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
>>> index b1bd2186e6d2..6e2aa4e79af7 100644
>>> --- a/include/linux/pagemap.h
>>> +++ b/include/linux/pagemap.h
>>> @@ -456,8 +456,8 @@ static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
>>>  	pgoff_t pgoff;
>>>  	if (unlikely(is_vm_hugetlb_page(vma)))
>>>  		return linear_hugepage_index(vma, address);
>>> -	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
>>> -	pgoff += vma->vm_pgoff;
>>> +	pgoff = (address - READ_ONCE(vma->vm_start)) >> PAGE_SHIFT;
>>> +	pgoff += READ_ONCE(vma->vm_pgoff);
>>>  	return pgoff;
>>>  }
>>>  
>>> diff --git a/mm/internal.h b/mm/internal.h
>>> index fb2667b20f0a..10b188c87fa4 100644
>>> --- a/mm/internal.h
>>> +++ b/mm/internal.h
>>> @@ -44,7 +44,21 @@ int do_swap_page(struct vm_fault *vmf);
>>>  extern struct vm_area_struct *get_vma(struct mm_struct *mm,
>>>  				      unsigned long addr);
>>>  extern void put_vma(struct vm_area_struct *vma);
>>> -#endif
>>> +
>>> +static inline bool vma_has_changed(struct vm_fault *vmf)
>>> +{
>>> +	int ret = RB_EMPTY_NODE(&vmf->vma->vm_rb);
>>> +	unsigned int seq = READ_ONCE(vmf->vma->vm_sequence.sequence);
>>> +
>>> +	/*
>>> +	 * Matches both the wmb in write_seqlock_{begin,end}() and
>>> +	 * the wmb in vma_rb_erase().
>>> +	 */
>>> +	smp_rmb();
>>> +
>>> +	return ret || seq != vmf->sequence;
>>> +}
>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>  
>>>  void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
>>>  		unsigned long floor, unsigned long ceiling);
>>> diff --git a/mm/memory.c b/mm/memory.c
>>> index ab32b0b4bd69..7bbbb8c7b9cd 100644
>>> --- a/mm/memory.c
>>> +++ b/mm/memory.c
>>> @@ -769,7 +769,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
>>>  	if (page)
>>>  		dump_page(page, "bad pte");
>>>  	pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
>>> -		 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
>>> +		 (void *)addr, READ_ONCE(vma->vm_flags), vma->anon_vma,
>>> +		 mapping, index);
>>>  	pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
>>>  		 vma->vm_file,
>>>  		 vma->vm_ops ? vma->vm_ops->fault : NULL,
>>> @@ -2306,6 +2307,118 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
>>>  }
>>>  EXPORT_SYMBOL_GPL(apply_to_page_range);
>>>  
>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>> +static bool pte_spinlock(struct vm_fault *vmf)
>>> +{
>>> +	bool ret = false;
>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>> +	pmd_t pmdval;
>>> +#endif
>>> +
>>> +	/* Check if vma is still valid */
>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>> +		vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>> +		spin_lock(vmf->ptl);
>>> +		return true;
>>> +	}
>>> +
>>> +again:
>>> +	local_irq_disable();
>>> +	if (vma_has_changed(vmf))
>>> +		goto out;
>>> +
>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>> +	/*
>>> +	 * We check if the pmd value is still the same to ensure that there
>>> +	 * is not a huge collapse operation in progress in our back.
>>> +	 */
>>> +	pmdval = READ_ONCE(*vmf->pmd);
>>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>>> +		goto out;
>>> +#endif
>>> +
>>> +	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>> +	if (unlikely(!spin_trylock(vmf->ptl))) {
>>> +		local_irq_enable();
>>> +		goto again;
>>> +	}
>>> +
>>> +	if (vma_has_changed(vmf)) {
>>> +		spin_unlock(vmf->ptl);
>>> +		goto out;
>>> +	}
>>> +
>>> +	ret = true;
>>> +out:
>>> +	local_irq_enable();
>>> +	return ret;
>>> +}
>>> +
>>> +static bool pte_map_lock(struct vm_fault *vmf)
>>> +{
>>> +	bool ret = false;
>>> +	pte_t *pte;
>>> +	spinlock_t *ptl;
>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>> +	pmd_t pmdval;
>>> +#endif
>>> +
>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>> +		vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
>>> +					       vmf->address, &vmf->ptl);
>>> +		return true;
>>> +	}
>>> +
>>> +	/*
>>> +	 * The first vma_has_changed() guarantees the page-tables are still
>>> +	 * valid, having IRQs disabled ensures they stay around, hence the
>>> +	 * second vma_has_changed() to make sure they are still valid once
>>> +	 * we've got the lock. After that a concurrent zap_pte_range() will
>>> +	 * block on the PTL and thus we're safe.
>>> +	 */
>>> +again:
>>> +	local_irq_disable();
>>> +	if (vma_has_changed(vmf))
>>> +		goto out;
>>> +
>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>> +	/*
>>> +	 * We check if the pmd value is still the same to ensure that there
>>> +	 * is not a huge collapse operation in progress in our back.
>>> +	 */
>>> +	pmdval = READ_ONCE(*vmf->pmd);
>>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>>> +		goto out;
>>> +#endif
>>> +
>>> +	/*
>>> +	 * Same as pte_offset_map_lock() except that we call
>>> +	 * spin_trylock() in place of spin_lock() to avoid race with
>>> +	 * unmap path which may have the lock and wait for this CPU
>>> +	 * to invalidate TLB but this CPU has irq disabled.
>>> +	 * Since we are in a speculative patch, accept it could fail
>>> +	 */
>>> +	ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>> +	pte = pte_offset_map(vmf->pmd, vmf->address);
>>> +	if (unlikely(!spin_trylock(ptl))) {
>>> +		pte_unmap(pte);
>>> +		local_irq_enable();
>>> +		goto again;
>>> +	}
>>> +
>>> +	if (vma_has_changed(vmf)) {
>>> +		pte_unmap_unlock(pte, ptl);
>>> +		goto out;
>>> +	}
>>> +
>>> +	vmf->pte = pte;
>>> +	vmf->ptl = ptl;
>>> +	ret = true;
>>> +out:
>>> +	local_irq_enable();
>>> +	return ret;
>>> +}
>>> +#else
>>>  static inline bool pte_spinlock(struct vm_fault *vmf)
>>>  {
>>>  	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>> @@ -2319,6 +2432,7 @@ static inline bool pte_map_lock(struct vm_fault *vmf)
>>>  				       vmf->address, &vmf->ptl);
>>>  	return true;
>>>  }
>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>  
>>>  /*
>>>   * handle_pte_fault chooses page fault handler according to an entry which was
>>> @@ -3208,6 +3322,14 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>>  		ret = check_stable_address_space(vma->vm_mm);
>>>  		if (ret)
>>>  			goto unlock;
>>> +		/*
>>> +		 * Don't call the userfaultfd during the speculative path.
>>> +		 * We already checked for the VMA to not be managed through
>>> +		 * userfaultfd, but it may be set in our back once we have lock
>>> +		 * the pte. In such a case we can ignore it this time.
>>> +		 */
>>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>> +			goto setpte;
>>>  		/* Deliver the page fault to userland, check inside PT lock */
>>>  		if (userfaultfd_missing(vma)) {
>>>  			pte_unmap_unlock(vmf->pte, vmf->ptl);
>>> @@ -3249,7 +3371,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>>  		goto unlock_and_release;
>>>  
>>>  	/* Deliver the page fault to userland, check inside PT lock */
>>> -	if (userfaultfd_missing(vma)) {
>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE) && userfaultfd_missing(vma)) {
>>>  		pte_unmap_unlock(vmf->pte, vmf->ptl);
>>>  		mem_cgroup_cancel_charge(page, memcg, false);
>>>  		put_page(page);
>>> @@ -3994,13 +4116,22 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>  
>>>  	if (unlikely(pmd_none(*vmf->pmd))) {
>>>  		/*
>>> +		 * In the case of the speculative page fault handler we abort
>>> +		 * the speculative path immediately as the pmd is probably
>>> +		 * in the way to be converted in a huge one. We will try
>>> +		 * again holding the mmap_sem (which implies that the collapse
>>> +		 * operation is done).
>>> +		 */
>>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>> +			return VM_FAULT_RETRY;
>>> +		/*
>>>  		 * Leave __pte_alloc() until later: because vm_ops->fault may
>>>  		 * want to allocate huge page, and if we expose page table
>>>  		 * for an instant, it will be difficult to retract from
>>>  		 * concurrent faults and from rmap lookups.
>>>  		 */
>>>  		vmf->pte = NULL;
>>> -	} else {
>>> +	} else if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>  		/* See comment in pte_alloc_one_map() */
>>>  		if (pmd_devmap_trans_unstable(vmf->pmd))
>>>  			return 0;
>>> @@ -4009,6 +4140,9 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>  		 * pmd from under us anymore at this point because we hold the
>>>  		 * mmap_sem read mode and khugepaged takes it in write mode.
>>>  		 * So now it's safe to run pte_offset_map().
>>> +		 * This is not applicable to the speculative page fault handler
>>> +		 * but in that case, the pte is fetched earlier in
>>> +		 * handle_speculative_fault().
>>>  		 */
>>>  		vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
>>>  		vmf->orig_pte = *vmf->pte;
>>> @@ -4031,6 +4165,8 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>  	if (!vmf->pte) {
>>>  		if (vma_is_anonymous(vmf->vma))
>>>  			return do_anonymous_page(vmf);
>>> +		else if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>> +			return VM_FAULT_RETRY;
>>>  		else
>>>  			return do_fault(vmf);
>>>  	}
>>> @@ -4128,6 +4264,9 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>  	vmf.pmd = pmd_alloc(mm, vmf.pud, address);
>>>  	if (!vmf.pmd)
>>>  		return VM_FAULT_OOM;
>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>> +	vmf.sequence = raw_read_seqcount(&vma->vm_sequence);
>>> +#endif
>>>  	if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
>>>  		ret = create_huge_pmd(&vmf);
>>>  		if (!(ret & VM_FAULT_FALLBACK))
>>> @@ -4161,6 +4300,201 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>  	return handle_pte_fault(&vmf);
>>>  }
>>>  
>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>> +/*
>>> + * Tries to handle the page fault in a speculative way, without grabbing the
>>> + * mmap_sem.
>>> + */
>>> +int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>> +			       unsigned int flags)
>>> +{
>>> +	struct vm_fault vmf = {
>>> +		.address = address,
>>> +	};
>>> +	pgd_t *pgd, pgdval;
>>> +	p4d_t *p4d, p4dval;
>>> +	pud_t pudval;
>>> +	int seq, ret = VM_FAULT_RETRY;
>>> +	struct vm_area_struct *vma;
>>> +#ifdef CONFIG_NUMA
>>> +	struct mempolicy *pol;
>>> +#endif
>>> +
>>> +	/* Clear flags that may lead to release the mmap_sem to retry */
>>> +	flags &= ~(FAULT_FLAG_ALLOW_RETRY|FAULT_FLAG_KILLABLE);
>>> +	flags |= FAULT_FLAG_SPECULATIVE;
>>> +
>>> +	vma = get_vma(mm, address);
>>> +	if (!vma)
>>> +		return ret;
>>> +
>>> +	seq = raw_read_seqcount(&vma->vm_sequence); /* rmb <-> seqlock,vma_rb_erase() */
>>> +	if (seq & 1)
>>> +		goto out_put;
>>> +
>>> +	/*
>>> +	 * Can't call vm_ops service has we don't know what they would do
>>> +	 * with the VMA.
>>> +	 * This include huge page from hugetlbfs.
>>> +	 */
>>> +	if (vma->vm_ops)
>>> +		goto out_put;
>>> +
>>   Hi   Laurent
>>    
>>    I think that most of pagefault will leave here.   Is there any case  need to skip ?
>>   I have tested the following  patch, it work well.
> Hi Zhong,
>
> Well this will allow file mapping to be handle in a speculative way, but that's
> a bit dangerous today as there is no guaranty that the vm_ops.vm_fault()
> operation will be fair.
>
> In the case of the anonymous file mapping that's often not a problem, depending
> on the underlying file system, but there are so many cases to check and this is
> hard to say this can be done in a speculative way as is.
 This patch say that spf just handle anonyous page. but I find that do_swap_page
 also maybe release the mmap_sem without FAULT_FLAG_RETRY_NOWAIT. why is it safe
 to handle the case. I think that the case is similar to file page. Maybe I miss
 something else.

 I test the patches and find just only 18% of the pagefault will enter into the
 speculative page fault during a process startup. As I had said. most of pagefault
 will be handled by ops->fault. I do not know the data you had posted is how to get.
 

 Thanks
 zhong jiang
> The huge work to do is to double check that all the code called by
> vm_ops.fault() is not dealing with the mmap_sem, which could be handled using
> FAULT_FLAG_RETRY_NOWAIT, and care is also needed about the resources that code
> is managing as it may assume that it is under the protection of the mmap_sem in
> read mode, and that can be done implicitly.
>
> Cheers,
> Laurent.
>
>> diff --git a/mm/memory.c b/mm/memory.c
>> index 936128b..9bc1545 100644
>>  @@ -3893,8 +3898,6 @@ static int handle_pte_fault(struct fault_env *fe)
>>         if (!fe->pte) {
>>                 if (vma_is_anonymous(fe->vma))
>>                         return do_anonymous_page(fe);
>> -               else if (fe->flags & FAULT_FLAG_SPECULATIVE)
>> -                       return VM_FAULT_RETRY;
>>                 else
>>                         return do_fault(fe);
>>         }
>> @@ -4026,20 +4029,11 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>                 goto out_put;
>>         }
>>         /*
>> -        * Can't call vm_ops service has we don't know what they would do
>> -        * with the VMA.
>> -        * This include huge page from hugetlbfs.
>> -        */
>> -       if (vma->vm_ops) {
>> -               trace_spf_vma_notsup(_RET_IP_, vma, address);
>> -               goto out_put;
>> -       }
>>
>>
>> Thanks
>> zhong jiang
>>> +	/*
>>> +	 * __anon_vma_prepare() requires the mmap_sem to be held
>>> +	 * because vm_next and vm_prev must be safe. This can't be guaranteed
>>> +	 * in the speculative path.
>>> +	 */
>>> +	if (unlikely(!vma->anon_vma))
>>> +		goto out_put;
>>> +
>>> +	vmf.vma_flags = READ_ONCE(vma->vm_flags);
>>> +	vmf.vma_page_prot = READ_ONCE(vma->vm_page_prot);
>>> +
>>> +	/* Can't call userland page fault handler in the speculative path */
>>> +	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING))
>>> +		goto out_put;
>>> +
>>> +	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP)
>>> +		/*
>>> +		 * This could be detected by the check address against VMA's
>>> +		 * boundaries but we want to trace it as not supported instead
>>> +		 * of changed.
>>> +		 */
>>> +		goto out_put;
>>> +
>>> +	if (address < READ_ONCE(vma->vm_start)
>>> +	    || READ_ONCE(vma->vm_end) <= address)
>>> +		goto out_put;
>>> +
>>> +	if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
>>> +				       flags & FAULT_FLAG_INSTRUCTION,
>>> +				       flags & FAULT_FLAG_REMOTE)) {
>>> +		ret = VM_FAULT_SIGSEGV;
>>> +		goto out_put;
>>> +	}
>>> +
>>> +	/* This is one is required to check that the VMA has write access set */
>>> +	if (flags & FAULT_FLAG_WRITE) {
>>> +		if (unlikely(!(vmf.vma_flags & VM_WRITE))) {
>>> +			ret = VM_FAULT_SIGSEGV;
>>> +			goto out_put;
>>> +		}
>>> +	} else if (unlikely(!(vmf.vma_flags & (VM_READ|VM_EXEC|VM_WRITE)))) {
>>> +		ret = VM_FAULT_SIGSEGV;
>>> +		goto out_put;
>>> +	}
>>> +
>>> +#ifdef CONFIG_NUMA
>>> +	/*
>>> +	 * MPOL_INTERLEAVE implies additional checks in
>>> +	 * mpol_misplaced() which are not compatible with the
>>> +	 *speculative page fault processing.
>>> +	 */
>>> +	pol = __get_vma_policy(vma, address);
>>> +	if (!pol)
>>> +		pol = get_task_policy(current);
>>> +	if (pol && pol->mode == MPOL_INTERLEAVE)
>>> +		goto out_put;
>>> +#endif
>>> +
>>> +	/*
>>> +	 * Do a speculative lookup of the PTE entry.
>>> +	 */
>>> +	local_irq_disable();
>>> +	pgd = pgd_offset(mm, address);
>>> +	pgdval = READ_ONCE(*pgd);
>>> +	if (pgd_none(pgdval) || unlikely(pgd_bad(pgdval)))
>>> +		goto out_walk;
>>> +
>>> +	p4d = p4d_offset(pgd, address);
>>> +	p4dval = READ_ONCE(*p4d);
>>> +	if (p4d_none(p4dval) || unlikely(p4d_bad(p4dval)))
>>> +		goto out_walk;
>>> +
>>> +	vmf.pud = pud_offset(p4d, address);
>>> +	pudval = READ_ONCE(*vmf.pud);
>>> +	if (pud_none(pudval) || unlikely(pud_bad(pudval)))
>>> +		goto out_walk;
>>> +
>>> +	/* Huge pages at PUD level are not supported. */
>>> +	if (unlikely(pud_trans_huge(pudval)))
>>> +		goto out_walk;
>>> +
>>> +	vmf.pmd = pmd_offset(vmf.pud, address);
>>> +	vmf.orig_pmd = READ_ONCE(*vmf.pmd);
>>> +	/*
>>> +	 * pmd_none could mean that a hugepage collapse is in progress
>>> +	 * in our back as collapse_huge_page() mark it before
>>> +	 * invalidating the pte (which is done once the IPI is catched
>>> +	 * by all CPU and we have interrupt disabled).
>>> +	 * For this reason we cannot handle THP in a speculative way since we
>>> +	 * can't safely indentify an in progress collapse operation done in our
>>> +	 * back on that PMD.
>>> +	 * Regarding the order of the following checks, see comment in
>>> +	 * pmd_devmap_trans_unstable()
>>> +	 */
>>> +	if (unlikely(pmd_devmap(vmf.orig_pmd) ||
>>> +		     pmd_none(vmf.orig_pmd) || pmd_trans_huge(vmf.orig_pmd) ||
>>> +		     is_swap_pmd(vmf.orig_pmd)))
>>> +		goto out_walk;
>>> +
>>> +	/*
>>> +	 * The above does not allocate/instantiate page-tables because doing so
>>> +	 * would lead to the possibility of instantiating page-tables after
>>> +	 * free_pgtables() -- and consequently leaking them.
>>> +	 *
>>> +	 * The result is that we take at least one !speculative fault per PMD
>>> +	 * in order to instantiate it.
>>> +	 */
>>> +
>>> +	vmf.pte = pte_offset_map(vmf.pmd, address);
>>> +	vmf.orig_pte = READ_ONCE(*vmf.pte);
>>> +	barrier(); /* See comment in handle_pte_fault() */
>>> +	if (pte_none(vmf.orig_pte)) {
>>> +		pte_unmap(vmf.pte);
>>> +		vmf.pte = NULL;
>>> +	}
>>> +
>>> +	vmf.vma = vma;
>>> +	vmf.pgoff = linear_page_index(vma, address);
>>> +	vmf.gfp_mask = __get_fault_gfp_mask(vma);
>>> +	vmf.sequence = seq;
>>> +	vmf.flags = flags;
>>> +
>>> +	local_irq_enable();
>>> +
>>> +	/*
>>> +	 * We need to re-validate the VMA after checking the bounds, otherwise
>>> +	 * we might have a false positive on the bounds.
>>> +	 */
>>> +	if (read_seqcount_retry(&vma->vm_sequence, seq))
>>> +		goto out_put;
>>> +
>>> +	mem_cgroup_oom_enable();
>>> +	ret = handle_pte_fault(&vmf);
>>> +	mem_cgroup_oom_disable();
>>> +
>>> +	put_vma(vma);
>>> +
>>> +	/*
>>> +	 * The task may have entered a memcg OOM situation but
>>> +	 * if the allocation error was handled gracefully (no
>>> +	 * VM_FAULT_OOM), there is no need to kill anything.
>>> +	 * Just clean up the OOM state peacefully.
>>> +	 */
>>> +	if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
>>> +		mem_cgroup_oom_synchronize(false);
>>> +	return ret;
>>> +
>>> +out_walk:
>>> +	local_irq_enable();
>>> +out_put:
>>> +	put_vma(vma);
>>> +	return ret;
>>> +}
>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>> +
>>>  /*
>>>   * By the time we get here, we already hold the mm semaphore
>>>   *
>>
>
> .
>

Re: [PATCH v11 19/26] mm: provide speculative fault infrastructure

[Laurent Dufour]

On 25/07/2018 11:04, zhong jiang wrote:
> On 2018/7/25 0:10, Laurent Dufour wrote:
>>
>> On 24/07/2018 16:26, zhong jiang wrote:
>>> On 2018/5/17 19:06, Laurent Dufour wrote:
>>>> From: Peter Zijlstra <peterz@infradead.org>
>>>>
>>>> Provide infrastructure to do a speculative fault (not holding
>>>> mmap_sem).
>>>>
>>>> The not holding of mmap_sem means we can race against VMA
>>>> change/removal and page-table destruction. We use the SRCU VMA freeing
>>>> to keep the VMA around. We use the VMA seqcount to detect change
>>>> (including umapping / page-table deletion) and we use gup_fast() style
>>>> page-table walking to deal with page-table races.
>>>>
>>>> Once we've obtained the page and are ready to update the PTE, we
>>>> validate if the state we started the fault with is still valid, if
>>>> not, we'll fail the fault with VM_FAULT_RETRY, otherwise we update the
>>>> PTE and we're done.
>>>>
>>>> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>>>>
>>>> [Manage the newly introduced pte_spinlock() for speculative page
>>>>  fault to fail if the VMA is touched in our back]
>>>> [Rename vma_is_dead() to vma_has_changed() and declare it here]
>>>> [Fetch p4d and pud]
>>>> [Set vmd.sequence in __handle_mm_fault()]
>>>> [Abort speculative path when handle_userfault() has to be called]
>>>> [Add additional VMA's flags checks in handle_speculative_fault()]
>>>> [Clear FAULT_FLAG_ALLOW_RETRY in handle_speculative_fault()]
>>>> [Don't set vmf->pte and vmf->ptl if pte_map_lock() failed]
>>>> [Remove warning comment about waiting for !seq&1 since we don't want
>>>>  to wait]
>>>> [Remove warning about no huge page support, mention it explictly]
>>>> [Don't call do_fault() in the speculative path as __do_fault() calls
>>>>  vma->vm_ops->fault() which may want to release mmap_sem]
>>>> [Only vm_fault pointer argument for vma_has_changed()]
>>>> [Fix check against huge page, calling pmd_trans_huge()]
>>>> [Use READ_ONCE() when reading VMA's fields in the speculative path]
>>>> [Explicitly check for __HAVE_ARCH_PTE_SPECIAL as we can't support for
>>>>  processing done in vm_normal_page()]
>>>> [Check that vma->anon_vma is already set when starting the speculative
>>>>  path]
>>>> [Check for memory policy as we can't support MPOL_INTERLEAVE case due to
>>>>  the processing done in mpol_misplaced()]
>>>> [Don't support VMA growing up or down]
>>>> [Move check on vm_sequence just before calling handle_pte_fault()]
>>>> [Don't build SPF services if !CONFIG_SPECULATIVE_PAGE_FAULT]
>>>> [Add mem cgroup oom check]
>>>> [Use READ_ONCE to access p*d entries]
>>>> [Replace deprecated ACCESS_ONCE() by READ_ONCE() in vma_has_changed()]
>>>> [Don't fetch pte again in handle_pte_fault() when running the speculative
>>>>  path]
>>>> [Check PMD against concurrent collapsing operation]
>>>> [Try spin lock the pte during the speculative path to avoid deadlock with
>>>>  other CPU's invalidating the TLB and requiring this CPU to catch the
>>>>  inter processor's interrupt]
>>>> [Move define of FAULT_FLAG_SPECULATIVE here]
>>>> [Introduce __handle_speculative_fault() and add a check against
>>>>  mm->mm_users in handle_speculative_fault() defined in mm.h]
>>>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>>>> ---
>>>>  include/linux/hugetlb_inline.h |   2 +-
>>>>  include/linux/mm.h             |  30 ++++
>>>>  include/linux/pagemap.h        |   4 +-
>>>>  mm/internal.h                  |  16 +-
>>>>  mm/memory.c                    | 340 ++++++++++++++++++++++++++++++++++++++++-
>>>>  5 files changed, 385 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/include/linux/hugetlb_inline.h b/include/linux/hugetlb_inline.h
>>>> index 0660a03d37d9..9e25283d6fc9 100644
>>>> --- a/include/linux/hugetlb_inline.h
>>>> +++ b/include/linux/hugetlb_inline.h
>>>> @@ -8,7 +8,7 @@
>>>>  
>>>>  static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
>>>>  {
>>>> -	return !!(vma->vm_flags & VM_HUGETLB);
>>>> +	return !!(READ_ONCE(vma->vm_flags) & VM_HUGETLB);
>>>>  }
>>>>  
>>>>  #else
>>>> diff --git a/include/linux/mm.h b/include/linux/mm.h
>>>> index 05cbba70104b..31acf98a7d92 100644
>>>> --- a/include/linux/mm.h
>>>> +++ b/include/linux/mm.h
>>>> @@ -315,6 +315,7 @@ extern pgprot_t protection_map[16];
>>>>  #define FAULT_FLAG_USER		0x40	/* The fault originated in userspace */
>>>>  #define FAULT_FLAG_REMOTE	0x80	/* faulting for non current tsk/mm */
>>>>  #define FAULT_FLAG_INSTRUCTION  0x100	/* The fault was during an instruction fetch */
>>>> +#define FAULT_FLAG_SPECULATIVE	0x200	/* Speculative fault, not holding mmap_sem */
>>>>  
>>>>  #define FAULT_FLAG_TRACE \
>>>>  	{ FAULT_FLAG_WRITE,		"WRITE" }, \
>>>> @@ -343,6 +344,10 @@ struct vm_fault {
>>>>  	gfp_t gfp_mask;			/* gfp mask to be used for allocations */
>>>>  	pgoff_t pgoff;			/* Logical page offset based on vma */
>>>>  	unsigned long address;		/* Faulting virtual address */
>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>> +	unsigned int sequence;
>>>> +	pmd_t orig_pmd;			/* value of PMD at the time of fault */
>>>> +#endif
>>>>  	pmd_t *pmd;			/* Pointer to pmd entry matching
>>>>  					 * the 'address' */
>>>>  	pud_t *pud;			/* Pointer to pud entry matching
>>>> @@ -1415,6 +1420,31 @@ int invalidate_inode_page(struct page *page);
>>>>  #ifdef CONFIG_MMU
>>>>  extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>>  		unsigned int flags);
>>>> +
>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>> +extern int __handle_speculative_fault(struct mm_struct *mm,
>>>> +				      unsigned long address,
>>>> +				      unsigned int flags);
>>>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>>>> +					   unsigned long address,
>>>> +					   unsigned int flags)
>>>> +{
>>>> +	/*
>>>> +	 * Try speculative page fault for multithreaded user space task only.
>>>> +	 */
>>>> +	if (!(flags & FAULT_FLAG_USER) || atomic_read(&mm->mm_users) == 1)
>>>> +		return VM_FAULT_RETRY;
>>>> +	return __handle_speculative_fault(mm, address, flags);
>>>> +}
>>>> +#else
>>>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>>>> +					   unsigned long address,
>>>> +					   unsigned int flags)
>>>> +{
>>>> +	return VM_FAULT_RETRY;
>>>> +}
>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>> +
>>>>  extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
>>>>  			    unsigned long address, unsigned int fault_flags,
>>>>  			    bool *unlocked);
>>>> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
>>>> index b1bd2186e6d2..6e2aa4e79af7 100644
>>>> --- a/include/linux/pagemap.h
>>>> +++ b/include/linux/pagemap.h
>>>> @@ -456,8 +456,8 @@ static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
>>>>  	pgoff_t pgoff;
>>>>  	if (unlikely(is_vm_hugetlb_page(vma)))
>>>>  		return linear_hugepage_index(vma, address);
>>>> -	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
>>>> -	pgoff += vma->vm_pgoff;
>>>> +	pgoff = (address - READ_ONCE(vma->vm_start)) >> PAGE_SHIFT;
>>>> +	pgoff += READ_ONCE(vma->vm_pgoff);
>>>>  	return pgoff;
>>>>  }
>>>>  
>>>> diff --git a/mm/internal.h b/mm/internal.h
>>>> index fb2667b20f0a..10b188c87fa4 100644
>>>> --- a/mm/internal.h
>>>> +++ b/mm/internal.h
>>>> @@ -44,7 +44,21 @@ int do_swap_page(struct vm_fault *vmf);
>>>>  extern struct vm_area_struct *get_vma(struct mm_struct *mm,
>>>>  				      unsigned long addr);
>>>>  extern void put_vma(struct vm_area_struct *vma);
>>>> -#endif
>>>> +
>>>> +static inline bool vma_has_changed(struct vm_fault *vmf)
>>>> +{
>>>> +	int ret = RB_EMPTY_NODE(&vmf->vma->vm_rb);
>>>> +	unsigned int seq = READ_ONCE(vmf->vma->vm_sequence.sequence);
>>>> +
>>>> +	/*
>>>> +	 * Matches both the wmb in write_seqlock_{begin,end}() and
>>>> +	 * the wmb in vma_rb_erase().
>>>> +	 */
>>>> +	smp_rmb();
>>>> +
>>>> +	return ret || seq != vmf->sequence;
>>>> +}
>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>>  
>>>>  void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
>>>>  		unsigned long floor, unsigned long ceiling);
>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>> index ab32b0b4bd69..7bbbb8c7b9cd 100644
>>>> --- a/mm/memory.c
>>>> +++ b/mm/memory.c
>>>> @@ -769,7 +769,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
>>>>  	if (page)
>>>>  		dump_page(page, "bad pte");
>>>>  	pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
>>>> -		 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
>>>> +		 (void *)addr, READ_ONCE(vma->vm_flags), vma->anon_vma,
>>>> +		 mapping, index);
>>>>  	pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
>>>>  		 vma->vm_file,
>>>>  		 vma->vm_ops ? vma->vm_ops->fault : NULL,
>>>> @@ -2306,6 +2307,118 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
>>>>  }
>>>>  EXPORT_SYMBOL_GPL(apply_to_page_range);
>>>>  
>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>> +static bool pte_spinlock(struct vm_fault *vmf)
>>>> +{
>>>> +	bool ret = false;
>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>> +	pmd_t pmdval;
>>>> +#endif
>>>> +
>>>> +	/* Check if vma is still valid */
>>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>> +		vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>> +		spin_lock(vmf->ptl);
>>>> +		return true;
>>>> +	}
>>>> +
>>>> +again:
>>>> +	local_irq_disable();
>>>> +	if (vma_has_changed(vmf))
>>>> +		goto out;
>>>> +
>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>> +	/*
>>>> +	 * We check if the pmd value is still the same to ensure that there
>>>> +	 * is not a huge collapse operation in progress in our back.
>>>> +	 */
>>>> +	pmdval = READ_ONCE(*vmf->pmd);
>>>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>>>> +		goto out;
>>>> +#endif
>>>> +
>>>> +	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>> +	if (unlikely(!spin_trylock(vmf->ptl))) {
>>>> +		local_irq_enable();
>>>> +		goto again;
>>>> +	}
>>>> +
>>>> +	if (vma_has_changed(vmf)) {
>>>> +		spin_unlock(vmf->ptl);
>>>> +		goto out;
>>>> +	}
>>>> +
>>>> +	ret = true;
>>>> +out:
>>>> +	local_irq_enable();
>>>> +	return ret;
>>>> +}
>>>> +
>>>> +static bool pte_map_lock(struct vm_fault *vmf)
>>>> +{
>>>> +	bool ret = false;
>>>> +	pte_t *pte;
>>>> +	spinlock_t *ptl;
>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>> +	pmd_t pmdval;
>>>> +#endif
>>>> +
>>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>> +		vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
>>>> +					       vmf->address, &vmf->ptl);
>>>> +		return true;
>>>> +	}
>>>> +
>>>> +	/*
>>>> +	 * The first vma_has_changed() guarantees the page-tables are still
>>>> +	 * valid, having IRQs disabled ensures they stay around, hence the
>>>> +	 * second vma_has_changed() to make sure they are still valid once
>>>> +	 * we've got the lock. After that a concurrent zap_pte_range() will
>>>> +	 * block on the PTL and thus we're safe.
>>>> +	 */
>>>> +again:
>>>> +	local_irq_disable();
>>>> +	if (vma_has_changed(vmf))
>>>> +		goto out;
>>>> +
>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>> +	/*
>>>> +	 * We check if the pmd value is still the same to ensure that there
>>>> +	 * is not a huge collapse operation in progress in our back.
>>>> +	 */
>>>> +	pmdval = READ_ONCE(*vmf->pmd);
>>>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>>>> +		goto out;
>>>> +#endif
>>>> +
>>>> +	/*
>>>> +	 * Same as pte_offset_map_lock() except that we call
>>>> +	 * spin_trylock() in place of spin_lock() to avoid race with
>>>> +	 * unmap path which may have the lock and wait for this CPU
>>>> +	 * to invalidate TLB but this CPU has irq disabled.
>>>> +	 * Since we are in a speculative patch, accept it could fail
>>>> +	 */
>>>> +	ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>> +	pte = pte_offset_map(vmf->pmd, vmf->address);
>>>> +	if (unlikely(!spin_trylock(ptl))) {
>>>> +		pte_unmap(pte);
>>>> +		local_irq_enable();
>>>> +		goto again;
>>>> +	}
>>>> +
>>>> +	if (vma_has_changed(vmf)) {
>>>> +		pte_unmap_unlock(pte, ptl);
>>>> +		goto out;
>>>> +	}
>>>> +
>>>> +	vmf->pte = pte;
>>>> +	vmf->ptl = ptl;
>>>> +	ret = true;
>>>> +out:
>>>> +	local_irq_enable();
>>>> +	return ret;
>>>> +}
>>>> +#else
>>>>  static inline bool pte_spinlock(struct vm_fault *vmf)
>>>>  {
>>>>  	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>> @@ -2319,6 +2432,7 @@ static inline bool pte_map_lock(struct vm_fault *vmf)
>>>>  				       vmf->address, &vmf->ptl);
>>>>  	return true;
>>>>  }
>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>>  
>>>>  /*
>>>>   * handle_pte_fault chooses page fault handler according to an entry which was
>>>> @@ -3208,6 +3322,14 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>>>  		ret = check_stable_address_space(vma->vm_mm);
>>>>  		if (ret)
>>>>  			goto unlock;
>>>> +		/*
>>>> +		 * Don't call the userfaultfd during the speculative path.
>>>> +		 * We already checked for the VMA to not be managed through
>>>> +		 * userfaultfd, but it may be set in our back once we have lock
>>>> +		 * the pte. In such a case we can ignore it this time.
>>>> +		 */
>>>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>>> +			goto setpte;
>>>>  		/* Deliver the page fault to userland, check inside PT lock */
>>>>  		if (userfaultfd_missing(vma)) {
>>>>  			pte_unmap_unlock(vmf->pte, vmf->ptl);
>>>> @@ -3249,7 +3371,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>>>  		goto unlock_and_release;
>>>>  
>>>>  	/* Deliver the page fault to userland, check inside PT lock */
>>>> -	if (userfaultfd_missing(vma)) {
>>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE) && userfaultfd_missing(vma)) {
>>>>  		pte_unmap_unlock(vmf->pte, vmf->ptl);
>>>>  		mem_cgroup_cancel_charge(page, memcg, false);
>>>>  		put_page(page);
>>>> @@ -3994,13 +4116,22 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>>  
>>>>  	if (unlikely(pmd_none(*vmf->pmd))) {
>>>>  		/*
>>>> +		 * In the case of the speculative page fault handler we abort
>>>> +		 * the speculative path immediately as the pmd is probably
>>>> +		 * in the way to be converted in a huge one. We will try
>>>> +		 * again holding the mmap_sem (which implies that the collapse
>>>> +		 * operation is done).
>>>> +		 */
>>>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>>> +			return VM_FAULT_RETRY;
>>>> +		/*
>>>>  		 * Leave __pte_alloc() until later: because vm_ops->fault may
>>>>  		 * want to allocate huge page, and if we expose page table
>>>>  		 * for an instant, it will be difficult to retract from
>>>>  		 * concurrent faults and from rmap lookups.
>>>>  		 */
>>>>  		vmf->pte = NULL;
>>>> -	} else {
>>>> +	} else if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>>  		/* See comment in pte_alloc_one_map() */
>>>>  		if (pmd_devmap_trans_unstable(vmf->pmd))
>>>>  			return 0;
>>>> @@ -4009,6 +4140,9 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>>  		 * pmd from under us anymore at this point because we hold the
>>>>  		 * mmap_sem read mode and khugepaged takes it in write mode.
>>>>  		 * So now it's safe to run pte_offset_map().
>>>> +		 * This is not applicable to the speculative page fault handler
>>>> +		 * but in that case, the pte is fetched earlier in
>>>> +		 * handle_speculative_fault().
>>>>  		 */
>>>>  		vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
>>>>  		vmf->orig_pte = *vmf->pte;
>>>> @@ -4031,6 +4165,8 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>>  	if (!vmf->pte) {
>>>>  		if (vma_is_anonymous(vmf->vma))
>>>>  			return do_anonymous_page(vmf);
>>>> +		else if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>>> +			return VM_FAULT_RETRY;
>>>>  		else
>>>>  			return do_fault(vmf);
>>>>  	}
>>>> @@ -4128,6 +4264,9 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>>  	vmf.pmd = pmd_alloc(mm, vmf.pud, address);
>>>>  	if (!vmf.pmd)
>>>>  		return VM_FAULT_OOM;
>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>> +	vmf.sequence = raw_read_seqcount(&vma->vm_sequence);
>>>> +#endif
>>>>  	if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
>>>>  		ret = create_huge_pmd(&vmf);
>>>>  		if (!(ret & VM_FAULT_FALLBACK))
>>>> @@ -4161,6 +4300,201 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>>  	return handle_pte_fault(&vmf);
>>>>  }
>>>>  
>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>> +/*
>>>> + * Tries to handle the page fault in a speculative way, without grabbing the
>>>> + * mmap_sem.
>>>> + */
>>>> +int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>> +			       unsigned int flags)
>>>> +{
>>>> +	struct vm_fault vmf = {
>>>> +		.address = address,
>>>> +	};
>>>> +	pgd_t *pgd, pgdval;
>>>> +	p4d_t *p4d, p4dval;
>>>> +	pud_t pudval;
>>>> +	int seq, ret = VM_FAULT_RETRY;
>>>> +	struct vm_area_struct *vma;
>>>> +#ifdef CONFIG_NUMA
>>>> +	struct mempolicy *pol;
>>>> +#endif
>>>> +
>>>> +	/* Clear flags that may lead to release the mmap_sem to retry */
>>>> +	flags &= ~(FAULT_FLAG_ALLOW_RETRY|FAULT_FLAG_KILLABLE);
>>>> +	flags |= FAULT_FLAG_SPECULATIVE;
>>>> +
>>>> +	vma = get_vma(mm, address);
>>>> +	if (!vma)
>>>> +		return ret;
>>>> +
>>>> +	seq = raw_read_seqcount(&vma->vm_sequence); /* rmb <-> seqlock,vma_rb_erase() */
>>>> +	if (seq & 1)
>>>> +		goto out_put;
>>>> +
>>>> +	/*
>>>> +	 * Can't call vm_ops service has we don't know what they would do
>>>> +	 * with the VMA.
>>>> +	 * This include huge page from hugetlbfs.
>>>> +	 */
>>>> +	if (vma->vm_ops)
>>>> +		goto out_put;
>>>> +
>>>   Hi   Laurent
>>>    
>>>    I think that most of pagefault will leave here.   Is there any case  need to skip ?
>>>   I have tested the following  patch, it work well.
>> Hi Zhong,
>>
>> Well this will allow file mapping to be handle in a speculative way, but that's
>> a bit dangerous today as there is no guaranty that the vm_ops.vm_fault()
>> operation will be fair.
>>
>> In the case of the anonymous file mapping that's often not a problem, depending
>> on the underlying file system, but there are so many cases to check and this is
>> hard to say this can be done in a speculative way as is.
>  This patch say that spf just handle anonyous page. but I find that do_swap_page
>  also maybe release the mmap_sem without FAULT_FLAG_RETRY_NOWAIT. why is it safe
>  to handle the case. I think that the case is similar to file page. Maybe I miss
>  something else.

do_swap_page() may released the mmap_sem through the call to
__lock_page_or_retry(), but this can only happen if FAULT_FLAG_ALLOW_RETRY or
FAULT_FLAG_KILLABLE are set and they are unset in __handle_speculative_fault().

> 
>  I test the patches and find just only 18% of the pagefault will enter into the
>  speculative page fault during a process startup. As I had said. most of pagefault
>  will be handled by ops->fault. I do not know the data you had posted is how to get.

I do agree that handling file mapping will be required, but this will add more
complexity to this series, since we need a way for drivers to tell they are
compatible with the speculative path.

May be I should give it a try on the next send.

For my information, what was the performance improvement you seen when handling
file page faulting this way ?

Thanks,
Laurent.

> 
> 
>  Thanks
>  zhong jiang
>> The huge work to do is to double check that all the code called by
>> vm_ops.fault() is not dealing with the mmap_sem, which could be handled using
>> FAULT_FLAG_RETRY_NOWAIT, and care is also needed about the resources that code
>> is managing as it may assume that it is under the protection of the mmap_sem in
>> read mode, and that can be done implicitly.
>>
>> Cheers,
>> Laurent.
>>
>>> diff --git a/mm/memory.c b/mm/memory.c
>>> index 936128b..9bc1545 100644
>>>  @@ -3893,8 +3898,6 @@ static int handle_pte_fault(struct fault_env *fe)
>>>         if (!fe->pte) {
>>>                 if (vma_is_anonymous(fe->vma))
>>>                         return do_anonymous_page(fe);
>>> -               else if (fe->flags & FAULT_FLAG_SPECULATIVE)
>>> -                       return VM_FAULT_RETRY;
>>>                 else
>>>                         return do_fault(fe);
>>>         }
>>> @@ -4026,20 +4029,11 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>                 goto out_put;
>>>         }
>>>         /*
>>> -        * Can't call vm_ops service has we don't know what they would do
>>> -        * with the VMA.
>>> -        * This include huge page from hugetlbfs.
>>> -        */
>>> -       if (vma->vm_ops) {
>>> -               trace_spf_vma_notsup(_RET_IP_, vma, address);
>>> -               goto out_put;
>>> -       }
>>>
>>>
>>> Thanks
>>> zhong jiang
>>>> +	/*
>>>> +	 * __anon_vma_prepare() requires the mmap_sem to be held
>>>> +	 * because vm_next and vm_prev must be safe. This can't be guaranteed
>>>> +	 * in the speculative path.
>>>> +	 */
>>>> +	if (unlikely(!vma->anon_vma))
>>>> +		goto out_put;
>>>> +
>>>> +	vmf.vma_flags = READ_ONCE(vma->vm_flags);
>>>> +	vmf.vma_page_prot = READ_ONCE(vma->vm_page_prot);
>>>> +
>>>> +	/* Can't call userland page fault handler in the speculative path */
>>>> +	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING))
>>>> +		goto out_put;
>>>> +
>>>> +	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP)
>>>> +		/*
>>>> +		 * This could be detected by the check address against VMA's
>>>> +		 * boundaries but we want to trace it as not supported instead
>>>> +		 * of changed.
>>>> +		 */
>>>> +		goto out_put;
>>>> +
>>>> +	if (address < READ_ONCE(vma->vm_start)
>>>> +	    || READ_ONCE(vma->vm_end) <= address)
>>>> +		goto out_put;
>>>> +
>>>> +	if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
>>>> +				       flags & FAULT_FLAG_INSTRUCTION,
>>>> +				       flags & FAULT_FLAG_REMOTE)) {
>>>> +		ret = VM_FAULT_SIGSEGV;
>>>> +		goto out_put;
>>>> +	}
>>>> +
>>>> +	/* This is one is required to check that the VMA has write access set */
>>>> +	if (flags & FAULT_FLAG_WRITE) {
>>>> +		if (unlikely(!(vmf.vma_flags & VM_WRITE))) {
>>>> +			ret = VM_FAULT_SIGSEGV;
>>>> +			goto out_put;
>>>> +		}
>>>> +	} else if (unlikely(!(vmf.vma_flags & (VM_READ|VM_EXEC|VM_WRITE)))) {
>>>> +		ret = VM_FAULT_SIGSEGV;
>>>> +		goto out_put;
>>>> +	}
>>>> +
>>>> +#ifdef CONFIG_NUMA
>>>> +	/*
>>>> +	 * MPOL_INTERLEAVE implies additional checks in
>>>> +	 * mpol_misplaced() which are not compatible with the
>>>> +	 *speculative page fault processing.
>>>> +	 */
>>>> +	pol = __get_vma_policy(vma, address);
>>>> +	if (!pol)
>>>> +		pol = get_task_policy(current);
>>>> +	if (pol && pol->mode == MPOL_INTERLEAVE)
>>>> +		goto out_put;
>>>> +#endif
>>>> +
>>>> +	/*
>>>> +	 * Do a speculative lookup of the PTE entry.
>>>> +	 */
>>>> +	local_irq_disable();
>>>> +	pgd = pgd_offset(mm, address);
>>>> +	pgdval = READ_ONCE(*pgd);
>>>> +	if (pgd_none(pgdval) || unlikely(pgd_bad(pgdval)))
>>>> +		goto out_walk;
>>>> +
>>>> +	p4d = p4d_offset(pgd, address);
>>>> +	p4dval = READ_ONCE(*p4d);
>>>> +	if (p4d_none(p4dval) || unlikely(p4d_bad(p4dval)))
>>>> +		goto out_walk;
>>>> +
>>>> +	vmf.pud = pud_offset(p4d, address);
>>>> +	pudval = READ_ONCE(*vmf.pud);
>>>> +	if (pud_none(pudval) || unlikely(pud_bad(pudval)))
>>>> +		goto out_walk;
>>>> +
>>>> +	/* Huge pages at PUD level are not supported. */
>>>> +	if (unlikely(pud_trans_huge(pudval)))
>>>> +		goto out_walk;
>>>> +
>>>> +	vmf.pmd = pmd_offset(vmf.pud, address);
>>>> +	vmf.orig_pmd = READ_ONCE(*vmf.pmd);
>>>> +	/*
>>>> +	 * pmd_none could mean that a hugepage collapse is in progress
>>>> +	 * in our back as collapse_huge_page() mark it before
>>>> +	 * invalidating the pte (which is done once the IPI is catched
>>>> +	 * by all CPU and we have interrupt disabled).
>>>> +	 * For this reason we cannot handle THP in a speculative way since we
>>>> +	 * can't safely indentify an in progress collapse operation done in our
>>>> +	 * back on that PMD.
>>>> +	 * Regarding the order of the following checks, see comment in
>>>> +	 * pmd_devmap_trans_unstable()
>>>> +	 */
>>>> +	if (unlikely(pmd_devmap(vmf.orig_pmd) ||
>>>> +		     pmd_none(vmf.orig_pmd) || pmd_trans_huge(vmf.orig_pmd) ||
>>>> +		     is_swap_pmd(vmf.orig_pmd)))
>>>> +		goto out_walk;
>>>> +
>>>> +	/*
>>>> +	 * The above does not allocate/instantiate page-tables because doing so
>>>> +	 * would lead to the possibility of instantiating page-tables after
>>>> +	 * free_pgtables() -- and consequently leaking them.
>>>> +	 *
>>>> +	 * The result is that we take at least one !speculative fault per PMD
>>>> +	 * in order to instantiate it.
>>>> +	 */
>>>> +
>>>> +	vmf.pte = pte_offset_map(vmf.pmd, address);
>>>> +	vmf.orig_pte = READ_ONCE(*vmf.pte);
>>>> +	barrier(); /* See comment in handle_pte_fault() */
>>>> +	if (pte_none(vmf.orig_pte)) {
>>>> +		pte_unmap(vmf.pte);
>>>> +		vmf.pte = NULL;
>>>> +	}
>>>> +
>>>> +	vmf.vma = vma;
>>>> +	vmf.pgoff = linear_page_index(vma, address);
>>>> +	vmf.gfp_mask = __get_fault_gfp_mask(vma);
>>>> +	vmf.sequence = seq;
>>>> +	vmf.flags = flags;
>>>> +
>>>> +	local_irq_enable();
>>>> +
>>>> +	/*
>>>> +	 * We need to re-validate the VMA after checking the bounds, otherwise
>>>> +	 * we might have a false positive on the bounds.
>>>> +	 */
>>>> +	if (read_seqcount_retry(&vma->vm_sequence, seq))
>>>> +		goto out_put;
>>>> +
>>>> +	mem_cgroup_oom_enable();
>>>> +	ret = handle_pte_fault(&vmf);
>>>> +	mem_cgroup_oom_disable();
>>>> +
>>>> +	put_vma(vma);
>>>> +
>>>> +	/*
>>>> +	 * The task may have entered a memcg OOM situation but
>>>> +	 * if the allocation error was handled gracefully (no
>>>> +	 * VM_FAULT_OOM), there is no need to kill anything.
>>>> +	 * Just clean up the OOM state peacefully.
>>>> +	 */
>>>> +	if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
>>>> +		mem_cgroup_oom_synchronize(false);
>>>> +	return ret;
>>>> +
>>>> +out_walk:
>>>> +	local_irq_enable();
>>>> +out_put:
>>>> +	put_vma(vma);
>>>> +	return ret;
>>>> +}
>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>> +
>>>>  /*
>>>>   * By the time we get here, we already hold the mm semaphore
>>>>   *
>>>
>>
>> .
>>
> 
> 

Re: [PATCH v11 19/26] mm: provide speculative fault infrastructure

[zhong jiang]

On 2018/7/25 18:44, Laurent Dufour wrote:
>
> On 25/07/2018 11:04, zhong jiang wrote:
>> On 2018/7/25 0:10, Laurent Dufour wrote:
>>> On 24/07/2018 16:26, zhong jiang wrote:
>>>> On 2018/5/17 19:06, Laurent Dufour wrote:
>>>>> From: Peter Zijlstra <peterz@infradead.org>
>>>>>
>>>>> Provide infrastructure to do a speculative fault (not holding
>>>>> mmap_sem).
>>>>>
>>>>> The not holding of mmap_sem means we can race against VMA
>>>>> change/removal and page-table destruction. We use the SRCU VMA freeing
>>>>> to keep the VMA around. We use the VMA seqcount to detect change
>>>>> (including umapping / page-table deletion) and we use gup_fast() style
>>>>> page-table walking to deal with page-table races.
>>>>>
>>>>> Once we've obtained the page and are ready to update the PTE, we
>>>>> validate if the state we started the fault with is still valid, if
>>>>> not, we'll fail the fault with VM_FAULT_RETRY, otherwise we update the
>>>>> PTE and we're done.
>>>>>
>>>>> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>>>>>
>>>>> [Manage the newly introduced pte_spinlock() for speculative page
>>>>>  fault to fail if the VMA is touched in our back]
>>>>> [Rename vma_is_dead() to vma_has_changed() and declare it here]
>>>>> [Fetch p4d and pud]
>>>>> [Set vmd.sequence in __handle_mm_fault()]
>>>>> [Abort speculative path when handle_userfault() has to be called]
>>>>> [Add additional VMA's flags checks in handle_speculative_fault()]
>>>>> [Clear FAULT_FLAG_ALLOW_RETRY in handle_speculative_fault()]
>>>>> [Don't set vmf->pte and vmf->ptl if pte_map_lock() failed]
>>>>> [Remove warning comment about waiting for !seq&1 since we don't want
>>>>>  to wait]
>>>>> [Remove warning about no huge page support, mention it explictly]
>>>>> [Don't call do_fault() in the speculative path as __do_fault() calls
>>>>>  vma->vm_ops->fault() which may want to release mmap_sem]
>>>>> [Only vm_fault pointer argument for vma_has_changed()]
>>>>> [Fix check against huge page, calling pmd_trans_huge()]
>>>>> [Use READ_ONCE() when reading VMA's fields in the speculative path]
>>>>> [Explicitly check for __HAVE_ARCH_PTE_SPECIAL as we can't support for
>>>>>  processing done in vm_normal_page()]
>>>>> [Check that vma->anon_vma is already set when starting the speculative
>>>>>  path]
>>>>> [Check for memory policy as we can't support MPOL_INTERLEAVE case due to
>>>>>  the processing done in mpol_misplaced()]
>>>>> [Don't support VMA growing up or down]
>>>>> [Move check on vm_sequence just before calling handle_pte_fault()]
>>>>> [Don't build SPF services if !CONFIG_SPECULATIVE_PAGE_FAULT]
>>>>> [Add mem cgroup oom check]
>>>>> [Use READ_ONCE to access p*d entries]
>>>>> [Replace deprecated ACCESS_ONCE() by READ_ONCE() in vma_has_changed()]
>>>>> [Don't fetch pte again in handle_pte_fault() when running the speculative
>>>>>  path]
>>>>> [Check PMD against concurrent collapsing operation]
>>>>> [Try spin lock the pte during the speculative path to avoid deadlock with
>>>>>  other CPU's invalidating the TLB and requiring this CPU to catch the
>>>>>  inter processor's interrupt]
>>>>> [Move define of FAULT_FLAG_SPECULATIVE here]
>>>>> [Introduce __handle_speculative_fault() and add a check against
>>>>>  mm->mm_users in handle_speculative_fault() defined in mm.h]
>>>>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>>>>> ---
>>>>>  include/linux/hugetlb_inline.h |   2 +-
>>>>>  include/linux/mm.h             |  30 ++++
>>>>>  include/linux/pagemap.h        |   4 +-
>>>>>  mm/internal.h                  |  16 +-
>>>>>  mm/memory.c                    | 340 ++++++++++++++++++++++++++++++++++++++++-
>>>>>  5 files changed, 385 insertions(+), 7 deletions(-)
>>>>>
>>>>> diff --git a/include/linux/hugetlb_inline.h b/include/linux/hugetlb_inline.h
>>>>> index 0660a03d37d9..9e25283d6fc9 100644
>>>>> --- a/include/linux/hugetlb_inline.h
>>>>> +++ b/include/linux/hugetlb_inline.h
>>>>> @@ -8,7 +8,7 @@
>>>>>  
>>>>>  static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
>>>>>  {
>>>>> -	return !!(vma->vm_flags & VM_HUGETLB);
>>>>> +	return !!(READ_ONCE(vma->vm_flags) & VM_HUGETLB);
>>>>>  }
>>>>>  
>>>>>  #else
>>>>> diff --git a/include/linux/mm.h b/include/linux/mm.h
>>>>> index 05cbba70104b..31acf98a7d92 100644
>>>>> --- a/include/linux/mm.h
>>>>> +++ b/include/linux/mm.h
>>>>> @@ -315,6 +315,7 @@ extern pgprot_t protection_map[16];
>>>>>  #define FAULT_FLAG_USER		0x40	/* The fault originated in userspace */
>>>>>  #define FAULT_FLAG_REMOTE	0x80	/* faulting for non current tsk/mm */
>>>>>  #define FAULT_FLAG_INSTRUCTION  0x100	/* The fault was during an instruction fetch */
>>>>> +#define FAULT_FLAG_SPECULATIVE	0x200	/* Speculative fault, not holding mmap_sem */
>>>>>  
>>>>>  #define FAULT_FLAG_TRACE \
>>>>>  	{ FAULT_FLAG_WRITE,		"WRITE" }, \
>>>>> @@ -343,6 +344,10 @@ struct vm_fault {
>>>>>  	gfp_t gfp_mask;			/* gfp mask to be used for allocations */
>>>>>  	pgoff_t pgoff;			/* Logical page offset based on vma */
>>>>>  	unsigned long address;		/* Faulting virtual address */
>>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>>> +	unsigned int sequence;
>>>>> +	pmd_t orig_pmd;			/* value of PMD at the time of fault */
>>>>> +#endif
>>>>>  	pmd_t *pmd;			/* Pointer to pmd entry matching
>>>>>  					 * the 'address' */
>>>>>  	pud_t *pud;			/* Pointer to pud entry matching
>>>>> @@ -1415,6 +1420,31 @@ int invalidate_inode_page(struct page *page);
>>>>>  #ifdef CONFIG_MMU
>>>>>  extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>>>  		unsigned int flags);
>>>>> +
>>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>>> +extern int __handle_speculative_fault(struct mm_struct *mm,
>>>>> +				      unsigned long address,
>>>>> +				      unsigned int flags);
>>>>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>>>>> +					   unsigned long address,
>>>>> +					   unsigned int flags)
>>>>> +{
>>>>> +	/*
>>>>> +	 * Try speculative page fault for multithreaded user space task only.
>>>>> +	 */
>>>>> +	if (!(flags & FAULT_FLAG_USER) || atomic_read(&mm->mm_users) == 1)
>>>>> +		return VM_FAULT_RETRY;
>>>>> +	return __handle_speculative_fault(mm, address, flags);
>>>>> +}
>>>>> +#else
>>>>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>>>>> +					   unsigned long address,
>>>>> +					   unsigned int flags)
>>>>> +{
>>>>> +	return VM_FAULT_RETRY;
>>>>> +}
>>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>>> +
>>>>>  extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
>>>>>  			    unsigned long address, unsigned int fault_flags,
>>>>>  			    bool *unlocked);
>>>>> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
>>>>> index b1bd2186e6d2..6e2aa4e79af7 100644
>>>>> --- a/include/linux/pagemap.h
>>>>> +++ b/include/linux/pagemap.h
>>>>> @@ -456,8 +456,8 @@ static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
>>>>>  	pgoff_t pgoff;
>>>>>  	if (unlikely(is_vm_hugetlb_page(vma)))
>>>>>  		return linear_hugepage_index(vma, address);
>>>>> -	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
>>>>> -	pgoff += vma->vm_pgoff;
>>>>> +	pgoff = (address - READ_ONCE(vma->vm_start)) >> PAGE_SHIFT;
>>>>> +	pgoff += READ_ONCE(vma->vm_pgoff);
>>>>>  	return pgoff;
>>>>>  }
>>>>>  
>>>>> diff --git a/mm/internal.h b/mm/internal.h
>>>>> index fb2667b20f0a..10b188c87fa4 100644
>>>>> --- a/mm/internal.h
>>>>> +++ b/mm/internal.h
>>>>> @@ -44,7 +44,21 @@ int do_swap_page(struct vm_fault *vmf);
>>>>>  extern struct vm_area_struct *get_vma(struct mm_struct *mm,
>>>>>  				      unsigned long addr);
>>>>>  extern void put_vma(struct vm_area_struct *vma);
>>>>> -#endif
>>>>> +
>>>>> +static inline bool vma_has_changed(struct vm_fault *vmf)
>>>>> +{
>>>>> +	int ret = RB_EMPTY_NODE(&vmf->vma->vm_rb);
>>>>> +	unsigned int seq = READ_ONCE(vmf->vma->vm_sequence.sequence);
>>>>> +
>>>>> +	/*
>>>>> +	 * Matches both the wmb in write_seqlock_{begin,end}() and
>>>>> +	 * the wmb in vma_rb_erase().
>>>>> +	 */
>>>>> +	smp_rmb();
>>>>> +
>>>>> +	return ret || seq != vmf->sequence;
>>>>> +}
>>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>>>  
>>>>>  void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
>>>>>  		unsigned long floor, unsigned long ceiling);
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index ab32b0b4bd69..7bbbb8c7b9cd 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -769,7 +769,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
>>>>>  	if (page)
>>>>>  		dump_page(page, "bad pte");
>>>>>  	pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
>>>>> -		 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
>>>>> +		 (void *)addr, READ_ONCE(vma->vm_flags), vma->anon_vma,
>>>>> +		 mapping, index);
>>>>>  	pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
>>>>>  		 vma->vm_file,
>>>>>  		 vma->vm_ops ? vma->vm_ops->fault : NULL,
>>>>> @@ -2306,6 +2307,118 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
>>>>>  }
>>>>>  EXPORT_SYMBOL_GPL(apply_to_page_range);
>>>>>  
>>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>>> +static bool pte_spinlock(struct vm_fault *vmf)
>>>>> +{
>>>>> +	bool ret = false;
>>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>> +	pmd_t pmdval;
>>>>> +#endif
>>>>> +
>>>>> +	/* Check if vma is still valid */
>>>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>>> +		vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>>> +		spin_lock(vmf->ptl);
>>>>> +		return true;
>>>>> +	}
>>>>> +
>>>>> +again:
>>>>> +	local_irq_disable();
>>>>> +	if (vma_has_changed(vmf))
>>>>> +		goto out;
>>>>> +
>>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>> +	/*
>>>>> +	 * We check if the pmd value is still the same to ensure that there
>>>>> +	 * is not a huge collapse operation in progress in our back.
>>>>> +	 */
>>>>> +	pmdval = READ_ONCE(*vmf->pmd);
>>>>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>>>>> +		goto out;
>>>>> +#endif
>>>>> +
>>>>> +	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>>> +	if (unlikely(!spin_trylock(vmf->ptl))) {
>>>>> +		local_irq_enable();
>>>>> +		goto again;
>>>>> +	}
>>>>> +
>>>>> +	if (vma_has_changed(vmf)) {
>>>>> +		spin_unlock(vmf->ptl);
>>>>> +		goto out;
>>>>> +	}
>>>>> +
>>>>> +	ret = true;
>>>>> +out:
>>>>> +	local_irq_enable();
>>>>> +	return ret;
>>>>> +}
>>>>> +
>>>>> +static bool pte_map_lock(struct vm_fault *vmf)
>>>>> +{
>>>>> +	bool ret = false;
>>>>> +	pte_t *pte;
>>>>> +	spinlock_t *ptl;
>>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>> +	pmd_t pmdval;
>>>>> +#endif
>>>>> +
>>>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>>> +		vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
>>>>> +					       vmf->address, &vmf->ptl);
>>>>> +		return true;
>>>>> +	}
>>>>> +
>>>>> +	/*
>>>>> +	 * The first vma_has_changed() guarantees the page-tables are still
>>>>> +	 * valid, having IRQs disabled ensures they stay around, hence the
>>>>> +	 * second vma_has_changed() to make sure they are still valid once
>>>>> +	 * we've got the lock. After that a concurrent zap_pte_range() will
>>>>> +	 * block on the PTL and thus we're safe.
>>>>> +	 */
>>>>> +again:
>>>>> +	local_irq_disable();
>>>>> +	if (vma_has_changed(vmf))
>>>>> +		goto out;
>>>>> +
>>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>> +	/*
>>>>> +	 * We check if the pmd value is still the same to ensure that there
>>>>> +	 * is not a huge collapse operation in progress in our back.
>>>>> +	 */
>>>>> +	pmdval = READ_ONCE(*vmf->pmd);
>>>>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>>>>> +		goto out;
>>>>> +#endif
>>>>> +
>>>>> +	/*
>>>>> +	 * Same as pte_offset_map_lock() except that we call
>>>>> +	 * spin_trylock() in place of spin_lock() to avoid race with
>>>>> +	 * unmap path which may have the lock and wait for this CPU
>>>>> +	 * to invalidate TLB but this CPU has irq disabled.
>>>>> +	 * Since we are in a speculative patch, accept it could fail
>>>>> +	 */
>>>>> +	ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>>> +	pte = pte_offset_map(vmf->pmd, vmf->address);
>>>>> +	if (unlikely(!spin_trylock(ptl))) {
>>>>> +		pte_unmap(pte);
>>>>> +		local_irq_enable();
>>>>> +		goto again;
>>>>> +	}
>>>>> +
>>>>> +	if (vma_has_changed(vmf)) {
>>>>> +		pte_unmap_unlock(pte, ptl);
>>>>> +		goto out;
>>>>> +	}
>>>>> +
>>>>> +	vmf->pte = pte;
>>>>> +	vmf->ptl = ptl;
>>>>> +	ret = true;
>>>>> +out:
>>>>> +	local_irq_enable();
>>>>> +	return ret;
>>>>> +}
>>>>> +#else
>>>>>  static inline bool pte_spinlock(struct vm_fault *vmf)
>>>>>  {
>>>>>  	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>>> @@ -2319,6 +2432,7 @@ static inline bool pte_map_lock(struct vm_fault *vmf)
>>>>>  				       vmf->address, &vmf->ptl);
>>>>>  	return true;
>>>>>  }
>>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>>>  
>>>>>  /*
>>>>>   * handle_pte_fault chooses page fault handler according to an entry which was
>>>>> @@ -3208,6 +3322,14 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>>>>  		ret = check_stable_address_space(vma->vm_mm);
>>>>>  		if (ret)
>>>>>  			goto unlock;
>>>>> +		/*
>>>>> +		 * Don't call the userfaultfd during the speculative path.
>>>>> +		 * We already checked for the VMA to not be managed through
>>>>> +		 * userfaultfd, but it may be set in our back once we have lock
>>>>> +		 * the pte. In such a case we can ignore it this time.
>>>>> +		 */
>>>>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>>>> +			goto setpte;
>>>>>  		/* Deliver the page fault to userland, check inside PT lock */
>>>>>  		if (userfaultfd_missing(vma)) {
>>>>>  			pte_unmap_unlock(vmf->pte, vmf->ptl);
>>>>> @@ -3249,7 +3371,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>>>>  		goto unlock_and_release;
>>>>>  
>>>>>  	/* Deliver the page fault to userland, check inside PT lock */
>>>>> -	if (userfaultfd_missing(vma)) {
>>>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE) && userfaultfd_missing(vma)) {
>>>>>  		pte_unmap_unlock(vmf->pte, vmf->ptl);
>>>>>  		mem_cgroup_cancel_charge(page, memcg, false);
>>>>>  		put_page(page);
>>>>> @@ -3994,13 +4116,22 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>>>  
>>>>>  	if (unlikely(pmd_none(*vmf->pmd))) {
>>>>>  		/*
>>>>> +		 * In the case of the speculative page fault handler we abort
>>>>> +		 * the speculative path immediately as the pmd is probably
>>>>> +		 * in the way to be converted in a huge one. We will try
>>>>> +		 * again holding the mmap_sem (which implies that the collapse
>>>>> +		 * operation is done).
>>>>> +		 */
>>>>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>>>> +			return VM_FAULT_RETRY;
>>>>> +		/*
>>>>>  		 * Leave __pte_alloc() until later: because vm_ops->fault may
>>>>>  		 * want to allocate huge page, and if we expose page table
>>>>>  		 * for an instant, it will be difficult to retract from
>>>>>  		 * concurrent faults and from rmap lookups.
>>>>>  		 */
>>>>>  		vmf->pte = NULL;
>>>>> -	} else {
>>>>> +	} else if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>>>  		/* See comment in pte_alloc_one_map() */
>>>>>  		if (pmd_devmap_trans_unstable(vmf->pmd))
>>>>>  			return 0;
>>>>> @@ -4009,6 +4140,9 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>>>  		 * pmd from under us anymore at this point because we hold the
>>>>>  		 * mmap_sem read mode and khugepaged takes it in write mode.
>>>>>  		 * So now it's safe to run pte_offset_map().
>>>>> +		 * This is not applicable to the speculative page fault handler
>>>>> +		 * but in that case, the pte is fetched earlier in
>>>>> +		 * handle_speculative_fault().
>>>>>  		 */
>>>>>  		vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
>>>>>  		vmf->orig_pte = *vmf->pte;
>>>>> @@ -4031,6 +4165,8 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>>>  	if (!vmf->pte) {
>>>>>  		if (vma_is_anonymous(vmf->vma))
>>>>>  			return do_anonymous_page(vmf);
>>>>> +		else if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>>>> +			return VM_FAULT_RETRY;
>>>>>  		else
>>>>>  			return do_fault(vmf);
>>>>>  	}
>>>>> @@ -4128,6 +4264,9 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>>>  	vmf.pmd = pmd_alloc(mm, vmf.pud, address);
>>>>>  	if (!vmf.pmd)
>>>>>  		return VM_FAULT_OOM;
>>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>>> +	vmf.sequence = raw_read_seqcount(&vma->vm_sequence);
>>>>> +#endif
>>>>>  	if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
>>>>>  		ret = create_huge_pmd(&vmf);
>>>>>  		if (!(ret & VM_FAULT_FALLBACK))
>>>>> @@ -4161,6 +4300,201 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>>>  	return handle_pte_fault(&vmf);
>>>>>  }
>>>>>  
>>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>>> +/*
>>>>> + * Tries to handle the page fault in a speculative way, without grabbing the
>>>>> + * mmap_sem.
>>>>> + */
>>>>> +int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>> +			       unsigned int flags)
>>>>> +{
>>>>> +	struct vm_fault vmf = {
>>>>> +		.address = address,
>>>>> +	};
>>>>> +	pgd_t *pgd, pgdval;
>>>>> +	p4d_t *p4d, p4dval;
>>>>> +	pud_t pudval;
>>>>> +	int seq, ret = VM_FAULT_RETRY;
>>>>> +	struct vm_area_struct *vma;
>>>>> +#ifdef CONFIG_NUMA
>>>>> +	struct mempolicy *pol;
>>>>> +#endif
>>>>> +
>>>>> +	/* Clear flags that may lead to release the mmap_sem to retry */
>>>>> +	flags &= ~(FAULT_FLAG_ALLOW_RETRY|FAULT_FLAG_KILLABLE);
>>>>> +	flags |= FAULT_FLAG_SPECULATIVE;
>>>>> +
>>>>> +	vma = get_vma(mm, address);
>>>>> +	if (!vma)
>>>>> +		return ret;
>>>>> +
>>>>> +	seq = raw_read_seqcount(&vma->vm_sequence); /* rmb <-> seqlock,vma_rb_erase() */
>>>>> +	if (seq & 1)
>>>>> +		goto out_put;
>>>>> +
>>>>> +	/*
>>>>> +	 * Can't call vm_ops service has we don't know what they would do
>>>>> +	 * with the VMA.
>>>>> +	 * This include huge page from hugetlbfs.
>>>>> +	 */
>>>>> +	if (vma->vm_ops)
>>>>> +		goto out_put;
>>>>> +
>>>>   Hi   Laurent
>>>>    
>>>>    I think that most of pagefault will leave here.   Is there any case  need to skip ?
>>>>   I have tested the following  patch, it work well.
>>> Hi Zhong,
>>>
>>> Well this will allow file mapping to be handle in a speculative way, but that's
>>> a bit dangerous today as there is no guaranty that the vm_ops.vm_fault()
>>> operation will be fair.
>>>
>>> In the case of the anonymous file mapping that's often not a problem, depending
>>> on the underlying file system, but there are so many cases to check and this is
>>> hard to say this can be done in a speculative way as is.
>>  This patch say that spf just handle anonyous page. but I find that do_swap_page
>>  also maybe release the mmap_sem without FAULT_FLAG_RETRY_NOWAIT. why is it safe
>>  to handle the case. I think that the case is similar to file page. Maybe I miss
>>  something else.
> do_swap_page() may released the mmap_sem through the call to
> __lock_page_or_retry(), but this can only happen if FAULT_FLAG_ALLOW_RETRY or
> FAULT_FLAG_KILLABLE are set and they are unset in __handle_speculative_fault().
 For spf. Indeed. Thank you for clarification.
>>  I test the patches and find just only 18% of the pagefault will enter into the
>>  speculative page fault during a process startup. As I had said. most of pagefault
>>  will be handled by ops->fault. I do not know the data you had posted is how to get.
> I do agree that handling file mapping will be required, but this will add more
> complexity to this series, since we need a way for drivers to tell they are
> compatible with the speculative path.
 As the above mentioned. the specualtive page fault do not pass FAULT_FLAG_ALLOW_RETRY.
 In other words, File page will not refer to release mmap_sem for spf.
 
 but I am still not quite clear that what should drivers do to compatible with speculatve path.
 The speculative path should not refer to the mmap_sem for filemap_fault.

 Thanks,
 zhong jiang
> May be I should give it a try on the next send.
 Ok, I will try.
> For my information, what was the performance improvement you seen when handling
> file page faulting this way ?
 I am sorry that. It is the data that Ganesh test the launch time on Andriod.
> Thanks,
> Laurent.
>
>>
>>  Thanks
>>  zhong jiang
>>> The huge work to do is to double check that all the code called by
>>> vm_ops.fault() is not dealing with the mmap_sem, which could be handled using
>>> FAULT_FLAG_RETRY_NOWAIT, and care is also needed about the resources that code
>>> is managing as it may assume that it is under the protection of the mmap_sem in
>>> read mode, and that can be done implicitly.
>>>
>>> Cheers,
>>> Laurent.
>>>
>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>> index 936128b..9bc1545 100644
>>>>  @@ -3893,8 +3898,6 @@ static int handle_pte_fault(struct fault_env *fe)
>>>>         if (!fe->pte) {
>>>>                 if (vma_is_anonymous(fe->vma))
>>>>                         return do_anonymous_page(fe);
>>>> -               else if (fe->flags & FAULT_FLAG_SPECULATIVE)
>>>> -                       return VM_FAULT_RETRY;
>>>>                 else
>>>>                         return do_fault(fe);
>>>>         }
>>>> @@ -4026,20 +4029,11 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>                 goto out_put;
>>>>         }
>>>>         /*
>>>> -        * Can't call vm_ops service has we don't know what they would do
>>>> -        * with the VMA.
>>>> -        * This include huge page from hugetlbfs.
>>>> -        */
>>>> -       if (vma->vm_ops) {
>>>> -               trace_spf_vma_notsup(_RET_IP_, vma, address);
>>>> -               goto out_put;
>>>> -       }
>>>>
>>>>
>>>> Thanks
>>>> zhong jiang
>>>>> +	/*
>>>>> +	 * __anon_vma_prepare() requires the mmap_sem to be held
>>>>> +	 * because vm_next and vm_prev must be safe. This can't be guaranteed
>>>>> +	 * in the speculative path.
>>>>> +	 */
>>>>> +	if (unlikely(!vma->anon_vma))
>>>>> +		goto out_put;
>>>>> +
>>>>> +	vmf.vma_flags = READ_ONCE(vma->vm_flags);
>>>>> +	vmf.vma_page_prot = READ_ONCE(vma->vm_page_prot);
>>>>> +
>>>>> +	/* Can't call userland page fault handler in the speculative path */
>>>>> +	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING))
>>>>> +		goto out_put;
>>>>> +
>>>>> +	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP)
>>>>> +		/*
>>>>> +		 * This could be detected by the check address against VMA's
>>>>> +		 * boundaries but we want to trace it as not supported instead
>>>>> +		 * of changed.
>>>>> +		 */
>>>>> +		goto out_put;
>>>>> +
>>>>> +	if (address < READ_ONCE(vma->vm_start)
>>>>> +	    || READ_ONCE(vma->vm_end) <= address)
>>>>> +		goto out_put;
>>>>> +
>>>>> +	if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
>>>>> +				       flags & FAULT_FLAG_INSTRUCTION,
>>>>> +				       flags & FAULT_FLAG_REMOTE)) {
>>>>> +		ret = VM_FAULT_SIGSEGV;
>>>>> +		goto out_put;
>>>>> +	}
>>>>> +
>>>>> +	/* This is one is required to check that the VMA has write access set */
>>>>> +	if (flags & FAULT_FLAG_WRITE) {
>>>>> +		if (unlikely(!(vmf.vma_flags & VM_WRITE))) {
>>>>> +			ret = VM_FAULT_SIGSEGV;
>>>>> +			goto out_put;
>>>>> +		}
>>>>> +	} else if (unlikely(!(vmf.vma_flags & (VM_READ|VM_EXEC|VM_WRITE)))) {
>>>>> +		ret = VM_FAULT_SIGSEGV;
>>>>> +		goto out_put;
>>>>> +	}
>>>>> +
>>>>> +#ifdef CONFIG_NUMA
>>>>> +	/*
>>>>> +	 * MPOL_INTERLEAVE implies additional checks in
>>>>> +	 * mpol_misplaced() which are not compatible with the
>>>>> +	 *speculative page fault processing.
>>>>> +	 */
>>>>> +	pol = __get_vma_policy(vma, address);
>>>>> +	if (!pol)
>>>>> +		pol = get_task_policy(current);
>>>>> +	if (pol && pol->mode == MPOL_INTERLEAVE)
>>>>> +		goto out_put;
>>>>> +#endif
>>>>> +
>>>>> +	/*
>>>>> +	 * Do a speculative lookup of the PTE entry.
>>>>> +	 */
>>>>> +	local_irq_disable();
>>>>> +	pgd = pgd_offset(mm, address);
>>>>> +	pgdval = READ_ONCE(*pgd);
>>>>> +	if (pgd_none(pgdval) || unlikely(pgd_bad(pgdval)))
>>>>> +		goto out_walk;
>>>>> +
>>>>> +	p4d = p4d_offset(pgd, address);
>>>>> +	p4dval = READ_ONCE(*p4d);
>>>>> +	if (p4d_none(p4dval) || unlikely(p4d_bad(p4dval)))
>>>>> +		goto out_walk;
>>>>> +
>>>>> +	vmf.pud = pud_offset(p4d, address);
>>>>> +	pudval = READ_ONCE(*vmf.pud);
>>>>> +	if (pud_none(pudval) || unlikely(pud_bad(pudval)))
>>>>> +		goto out_walk;
>>>>> +
>>>>> +	/* Huge pages at PUD level are not supported. */
>>>>> +	if (unlikely(pud_trans_huge(pudval)))
>>>>> +		goto out_walk;
>>>>> +
>>>>> +	vmf.pmd = pmd_offset(vmf.pud, address);
>>>>> +	vmf.orig_pmd = READ_ONCE(*vmf.pmd);
>>>>> +	/*
>>>>> +	 * pmd_none could mean that a hugepage collapse is in progress
>>>>> +	 * in our back as collapse_huge_page() mark it before
>>>>> +	 * invalidating the pte (which is done once the IPI is catched
>>>>> +	 * by all CPU and we have interrupt disabled).
>>>>> +	 * For this reason we cannot handle THP in a speculative way since we
>>>>> +	 * can't safely indentify an in progress collapse operation done in our
>>>>> +	 * back on that PMD.
>>>>> +	 * Regarding the order of the following checks, see comment in
>>>>> +	 * pmd_devmap_trans_unstable()
>>>>> +	 */
>>>>> +	if (unlikely(pmd_devmap(vmf.orig_pmd) ||
>>>>> +		     pmd_none(vmf.orig_pmd) || pmd_trans_huge(vmf.orig_pmd) ||
>>>>> +		     is_swap_pmd(vmf.orig_pmd)))
>>>>> +		goto out_walk;
>>>>> +
>>>>> +	/*
>>>>> +	 * The above does not allocate/instantiate page-tables because doing so
>>>>> +	 * would lead to the possibility of instantiating page-tables after
>>>>> +	 * free_pgtables() -- and consequently leaking them.
>>>>> +	 *
>>>>> +	 * The result is that we take at least one !speculative fault per PMD
>>>>> +	 * in order to instantiate it.
>>>>> +	 */
>>>>> +
>>>>> +	vmf.pte = pte_offset_map(vmf.pmd, address);
>>>>> +	vmf.orig_pte = READ_ONCE(*vmf.pte);
>>>>> +	barrier(); /* See comment in handle_pte_fault() */
>>>>> +	if (pte_none(vmf.orig_pte)) {
>>>>> +		pte_unmap(vmf.pte);
>>>>> +		vmf.pte = NULL;
>>>>> +	}
>>>>> +
>>>>> +	vmf.vma = vma;
>>>>> +	vmf.pgoff = linear_page_index(vma, address);
>>>>> +	vmf.gfp_mask = __get_fault_gfp_mask(vma);
>>>>> +	vmf.sequence = seq;
>>>>> +	vmf.flags = flags;
>>>>> +
>>>>> +	local_irq_enable();
>>>>> +
>>>>> +	/*
>>>>> +	 * We need to re-validate the VMA after checking the bounds, otherwise
>>>>> +	 * we might have a false positive on the bounds.
>>>>> +	 */
>>>>> +	if (read_seqcount_retry(&vma->vm_sequence, seq))
>>>>> +		goto out_put;
>>>>> +
>>>>> +	mem_cgroup_oom_enable();
>>>>> +	ret = handle_pte_fault(&vmf);
>>>>> +	mem_cgroup_oom_disable();
>>>>> +
>>>>> +	put_vma(vma);
>>>>> +
>>>>> +	/*
>>>>> +	 * The task may have entered a memcg OOM situation but
>>>>> +	 * if the allocation error was handled gracefully (no
>>>>> +	 * VM_FAULT_OOM), there is no need to kill anything.
>>>>> +	 * Just clean up the OOM state peacefully.
>>>>> +	 */
>>>>> +	if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
>>>>> +		mem_cgroup_oom_synchronize(false);
>>>>> +	return ret;
>>>>> +
>>>>> +out_walk:
>>>>> +	local_irq_enable();
>>>>> +out_put:
>>>>> +	put_vma(vma);
>>>>> +	return ret;
>>>>> +}
>>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>>> +
>>>>>  /*
>>>>>   * By the time we get here, we already hold the mm semaphore
>>>>>   *
>>> .
>>>
>>
>
> .
>

Re: [PATCH v11 19/26] mm: provide speculative fault infrastructure

[zhong jiang]

On 2018/7/25 18:44, Laurent Dufour wrote:
>
> On 25/07/2018 11:04, zhong jiang wrote:
>> On 2018/7/25 0:10, Laurent Dufour wrote:
>>> On 24/07/2018 16:26, zhong jiang wrote:
>>>> On 2018/5/17 19:06, Laurent Dufour wrote:
>>>>> From: Peter Zijlstra <peterz@infradead.org>
>>>>>
>>>>> Provide infrastructure to do a speculative fault (not holding
>>>>> mmap_sem).
>>>>>
>>>>> The not holding of mmap_sem means we can race against VMA
>>>>> change/removal and page-table destruction. We use the SRCU VMA freeing
>>>>> to keep the VMA around. We use the VMA seqcount to detect change
>>>>> (including umapping / page-table deletion) and we use gup_fast() style
>>>>> page-table walking to deal with page-table races.
>>>>>
>>>>> Once we've obtained the page and are ready to update the PTE, we
>>>>> validate if the state we started the fault with is still valid, if
>>>>> not, we'll fail the fault with VM_FAULT_RETRY, otherwise we update the
>>>>> PTE and we're done.
>>>>>
>>>>> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>>>>>
>>>>> [Manage the newly introduced pte_spinlock() for speculative page
>>>>>  fault to fail if the VMA is touched in our back]
>>>>> [Rename vma_is_dead() to vma_has_changed() and declare it here]
>>>>> [Fetch p4d and pud]
>>>>> [Set vmd.sequence in __handle_mm_fault()]
>>>>> [Abort speculative path when handle_userfault() has to be called]
>>>>> [Add additional VMA's flags checks in handle_speculative_fault()]
>>>>> [Clear FAULT_FLAG_ALLOW_RETRY in handle_speculative_fault()]
>>>>> [Don't set vmf->pte and vmf->ptl if pte_map_lock() failed]
>>>>> [Remove warning comment about waiting for !seq&1 since we don't want
>>>>>  to wait]
>>>>> [Remove warning about no huge page support, mention it explictly]
>>>>> [Don't call do_fault() in the speculative path as __do_fault() calls
>>>>>  vma->vm_ops->fault() which may want to release mmap_sem]
>>>>> [Only vm_fault pointer argument for vma_has_changed()]
>>>>> [Fix check against huge page, calling pmd_trans_huge()]
>>>>> [Use READ_ONCE() when reading VMA's fields in the speculative path]
>>>>> [Explicitly check for __HAVE_ARCH_PTE_SPECIAL as we can't support for
>>>>>  processing done in vm_normal_page()]
>>>>> [Check that vma->anon_vma is already set when starting the speculative
>>>>>  path]
>>>>> [Check for memory policy as we can't support MPOL_INTERLEAVE case due to
>>>>>  the processing done in mpol_misplaced()]
>>>>> [Don't support VMA growing up or down]
>>>>> [Move check on vm_sequence just before calling handle_pte_fault()]
>>>>> [Don't build SPF services if !CONFIG_SPECULATIVE_PAGE_FAULT]
>>>>> [Add mem cgroup oom check]
>>>>> [Use READ_ONCE to access p*d entries]
>>>>> [Replace deprecated ACCESS_ONCE() by READ_ONCE() in vma_has_changed()]
>>>>> [Don't fetch pte again in handle_pte_fault() when running the speculative
>>>>>  path]
>>>>> [Check PMD against concurrent collapsing operation]
>>>>> [Try spin lock the pte during the speculative path to avoid deadlock with
>>>>>  other CPU's invalidating the TLB and requiring this CPU to catch the
>>>>>  inter processor's interrupt]
>>>>> [Move define of FAULT_FLAG_SPECULATIVE here]
>>>>> [Introduce __handle_speculative_fault() and add a check against
>>>>>  mm->mm_users in handle_speculative_fault() defined in mm.h]
>>>>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>>>>> ---
>>>>>  include/linux/hugetlb_inline.h |   2 +-
>>>>>  include/linux/mm.h             |  30 ++++
>>>>>  include/linux/pagemap.h        |   4 +-
>>>>>  mm/internal.h                  |  16 +-
>>>>>  mm/memory.c                    | 340 ++++++++++++++++++++++++++++++++++++++++-
>>>>>  5 files changed, 385 insertions(+), 7 deletions(-)
>>>>>
>>>>> diff --git a/include/linux/hugetlb_inline.h b/include/linux/hugetlb_inline.h
>>>>> index 0660a03d37d9..9e25283d6fc9 100644
>>>>> --- a/include/linux/hugetlb_inline.h
>>>>> +++ b/include/linux/hugetlb_inline.h
>>>>> @@ -8,7 +8,7 @@
>>>>>  
>>>>>  static inline bool is_vm_hugetlb_page(struct vm_area_struct *vma)
>>>>>  {
>>>>> -	return !!(vma->vm_flags & VM_HUGETLB);
>>>>> +	return !!(READ_ONCE(vma->vm_flags) & VM_HUGETLB);
>>>>>  }
>>>>>  
>>>>>  #else
>>>>> diff --git a/include/linux/mm.h b/include/linux/mm.h
>>>>> index 05cbba70104b..31acf98a7d92 100644
>>>>> --- a/include/linux/mm.h
>>>>> +++ b/include/linux/mm.h
>>>>> @@ -315,6 +315,7 @@ extern pgprot_t protection_map[16];
>>>>>  #define FAULT_FLAG_USER		0x40	/* The fault originated in userspace */
>>>>>  #define FAULT_FLAG_REMOTE	0x80	/* faulting for non current tsk/mm */
>>>>>  #define FAULT_FLAG_INSTRUCTION  0x100	/* The fault was during an instruction fetch */
>>>>> +#define FAULT_FLAG_SPECULATIVE	0x200	/* Speculative fault, not holding mmap_sem */
>>>>>  
>>>>>  #define FAULT_FLAG_TRACE \
>>>>>  	{ FAULT_FLAG_WRITE,		"WRITE" }, \
>>>>> @@ -343,6 +344,10 @@ struct vm_fault {
>>>>>  	gfp_t gfp_mask;			/* gfp mask to be used for allocations */
>>>>>  	pgoff_t pgoff;			/* Logical page offset based on vma */
>>>>>  	unsigned long address;		/* Faulting virtual address */
>>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>>> +	unsigned int sequence;
>>>>> +	pmd_t orig_pmd;			/* value of PMD at the time of fault */
>>>>> +#endif
>>>>>  	pmd_t *pmd;			/* Pointer to pmd entry matching
>>>>>  					 * the 'address' */
>>>>>  	pud_t *pud;			/* Pointer to pud entry matching
>>>>> @@ -1415,6 +1420,31 @@ int invalidate_inode_page(struct page *page);
>>>>>  #ifdef CONFIG_MMU
>>>>>  extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>>>  		unsigned int flags);
>>>>> +
>>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>>> +extern int __handle_speculative_fault(struct mm_struct *mm,
>>>>> +				      unsigned long address,
>>>>> +				      unsigned int flags);
>>>>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>>>>> +					   unsigned long address,
>>>>> +					   unsigned int flags)
>>>>> +{
>>>>> +	/*
>>>>> +	 * Try speculative page fault for multithreaded user space task only.
>>>>> +	 */
>>>>> +	if (!(flags & FAULT_FLAG_USER) || atomic_read(&mm->mm_users) == 1)
>>>>> +		return VM_FAULT_RETRY;
>>>>> +	return __handle_speculative_fault(mm, address, flags);
>>>>> +}
>>>>> +#else
>>>>> +static inline int handle_speculative_fault(struct mm_struct *mm,
>>>>> +					   unsigned long address,
>>>>> +					   unsigned int flags)
>>>>> +{
>>>>> +	return VM_FAULT_RETRY;
>>>>> +}
>>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>>> +
>>>>>  extern int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
>>>>>  			    unsigned long address, unsigned int fault_flags,
>>>>>  			    bool *unlocked);
>>>>> diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
>>>>> index b1bd2186e6d2..6e2aa4e79af7 100644
>>>>> --- a/include/linux/pagemap.h
>>>>> +++ b/include/linux/pagemap.h
>>>>> @@ -456,8 +456,8 @@ static inline pgoff_t linear_page_index(struct vm_area_struct *vma,
>>>>>  	pgoff_t pgoff;
>>>>>  	if (unlikely(is_vm_hugetlb_page(vma)))
>>>>>  		return linear_hugepage_index(vma, address);
>>>>> -	pgoff = (address - vma->vm_start) >> PAGE_SHIFT;
>>>>> -	pgoff += vma->vm_pgoff;
>>>>> +	pgoff = (address - READ_ONCE(vma->vm_start)) >> PAGE_SHIFT;
>>>>> +	pgoff += READ_ONCE(vma->vm_pgoff);
>>>>>  	return pgoff;
>>>>>  }
>>>>>  
>>>>> diff --git a/mm/internal.h b/mm/internal.h
>>>>> index fb2667b20f0a..10b188c87fa4 100644
>>>>> --- a/mm/internal.h
>>>>> +++ b/mm/internal.h
>>>>> @@ -44,7 +44,21 @@ int do_swap_page(struct vm_fault *vmf);
>>>>>  extern struct vm_area_struct *get_vma(struct mm_struct *mm,
>>>>>  				      unsigned long addr);
>>>>>  extern void put_vma(struct vm_area_struct *vma);
>>>>> -#endif
>>>>> +
>>>>> +static inline bool vma_has_changed(struct vm_fault *vmf)
>>>>> +{
>>>>> +	int ret = RB_EMPTY_NODE(&vmf->vma->vm_rb);
>>>>> +	unsigned int seq = READ_ONCE(vmf->vma->vm_sequence.sequence);
>>>>> +
>>>>> +	/*
>>>>> +	 * Matches both the wmb in write_seqlock_{begin,end}() and
>>>>> +	 * the wmb in vma_rb_erase().
>>>>> +	 */
>>>>> +	smp_rmb();
>>>>> +
>>>>> +	return ret || seq != vmf->sequence;
>>>>> +}
>>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>>>  
>>>>>  void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *start_vma,
>>>>>  		unsigned long floor, unsigned long ceiling);
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index ab32b0b4bd69..7bbbb8c7b9cd 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -769,7 +769,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
>>>>>  	if (page)
>>>>>  		dump_page(page, "bad pte");
>>>>>  	pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
>>>>> -		 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
>>>>> +		 (void *)addr, READ_ONCE(vma->vm_flags), vma->anon_vma,
>>>>> +		 mapping, index);
>>>>>  	pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
>>>>>  		 vma->vm_file,
>>>>>  		 vma->vm_ops ? vma->vm_ops->fault : NULL,
>>>>> @@ -2306,6 +2307,118 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
>>>>>  }
>>>>>  EXPORT_SYMBOL_GPL(apply_to_page_range);
>>>>>  
>>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>>> +static bool pte_spinlock(struct vm_fault *vmf)
>>>>> +{
>>>>> +	bool ret = false;
>>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>> +	pmd_t pmdval;
>>>>> +#endif
>>>>> +
>>>>> +	/* Check if vma is still valid */
>>>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>>> +		vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>>> +		spin_lock(vmf->ptl);
>>>>> +		return true;
>>>>> +	}
>>>>> +
>>>>> +again:
>>>>> +	local_irq_disable();
>>>>> +	if (vma_has_changed(vmf))
>>>>> +		goto out;
>>>>> +
>>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>> +	/*
>>>>> +	 * We check if the pmd value is still the same to ensure that there
>>>>> +	 * is not a huge collapse operation in progress in our back.
>>>>> +	 */
>>>>> +	pmdval = READ_ONCE(*vmf->pmd);
>>>>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>>>>> +		goto out;
>>>>> +#endif
>>>>> +
>>>>> +	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>>> +	if (unlikely(!spin_trylock(vmf->ptl))) {
>>>>> +		local_irq_enable();
>>>>> +		goto again;
>>>>> +	}
>>>>> +
>>>>> +	if (vma_has_changed(vmf)) {
>>>>> +		spin_unlock(vmf->ptl);
>>>>> +		goto out;
>>>>> +	}
>>>>> +
>>>>> +	ret = true;
>>>>> +out:
>>>>> +	local_irq_enable();
>>>>> +	return ret;
>>>>> +}
>>>>> +
>>>>> +static bool pte_map_lock(struct vm_fault *vmf)
>>>>> +{
>>>>> +	bool ret = false;
>>>>> +	pte_t *pte;
>>>>> +	spinlock_t *ptl;
>>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>> +	pmd_t pmdval;
>>>>> +#endif
>>>>> +
>>>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>>> +		vmf->pte = pte_offset_map_lock(vmf->vma->vm_mm, vmf->pmd,
>>>>> +					       vmf->address, &vmf->ptl);
>>>>> +		return true;
>>>>> +	}
>>>>> +
>>>>> +	/*
>>>>> +	 * The first vma_has_changed() guarantees the page-tables are still
>>>>> +	 * valid, having IRQs disabled ensures they stay around, hence the
>>>>> +	 * second vma_has_changed() to make sure they are still valid once
>>>>> +	 * we've got the lock. After that a concurrent zap_pte_range() will
>>>>> +	 * block on the PTL and thus we're safe.
>>>>> +	 */
>>>>> +again:
>>>>> +	local_irq_disable();
>>>>> +	if (vma_has_changed(vmf))
>>>>> +		goto out;
>>>>> +
>>>>> +#ifdef CONFIG_TRANSPARENT_HUGEPAGE
>>>>> +	/*
>>>>> +	 * We check if the pmd value is still the same to ensure that there
>>>>> +	 * is not a huge collapse operation in progress in our back.
>>>>> +	 */
>>>>> +	pmdval = READ_ONCE(*vmf->pmd);
>>>>> +	if (!pmd_same(pmdval, vmf->orig_pmd))
>>>>> +		goto out;
>>>>> +#endif
>>>>> +
>>>>> +	/*
>>>>> +	 * Same as pte_offset_map_lock() except that we call
>>>>> +	 * spin_trylock() in place of spin_lock() to avoid race with
>>>>> +	 * unmap path which may have the lock and wait for this CPU
>>>>> +	 * to invalidate TLB but this CPU has irq disabled.
>>>>> +	 * Since we are in a speculative patch, accept it could fail
>>>>> +	 */
>>>>> +	ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>>> +	pte = pte_offset_map(vmf->pmd, vmf->address);
>>>>> +	if (unlikely(!spin_trylock(ptl))) {
>>>>> +		pte_unmap(pte);
>>>>> +		local_irq_enable();
>>>>> +		goto again;
>>>>> +	}
>>>>> +
>>>>> +	if (vma_has_changed(vmf)) {
>>>>> +		pte_unmap_unlock(pte, ptl);
>>>>> +		goto out;
>>>>> +	}
>>>>> +
>>>>> +	vmf->pte = pte;
>>>>> +	vmf->ptl = ptl;
>>>>> +	ret = true;
>>>>> +out:
>>>>> +	local_irq_enable();
>>>>> +	return ret;
>>>>> +}
>>>>> +#else
>>>>>  static inline bool pte_spinlock(struct vm_fault *vmf)
>>>>>  {
>>>>>  	vmf->ptl = pte_lockptr(vmf->vma->vm_mm, vmf->pmd);
>>>>> @@ -2319,6 +2432,7 @@ static inline bool pte_map_lock(struct vm_fault *vmf)
>>>>>  				       vmf->address, &vmf->ptl);
>>>>>  	return true;
>>>>>  }
>>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>>>  
>>>>>  /*
>>>>>   * handle_pte_fault chooses page fault handler according to an entry which was
>>>>> @@ -3208,6 +3322,14 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>>>>  		ret = check_stable_address_space(vma->vm_mm);
>>>>>  		if (ret)
>>>>>  			goto unlock;
>>>>> +		/*
>>>>> +		 * Don't call the userfaultfd during the speculative path.
>>>>> +		 * We already checked for the VMA to not be managed through
>>>>> +		 * userfaultfd, but it may be set in our back once we have lock
>>>>> +		 * the pte. In such a case we can ignore it this time.
>>>>> +		 */
>>>>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>>>> +			goto setpte;
>>>>>  		/* Deliver the page fault to userland, check inside PT lock */
>>>>>  		if (userfaultfd_missing(vma)) {
>>>>>  			pte_unmap_unlock(vmf->pte, vmf->ptl);
>>>>> @@ -3249,7 +3371,7 @@ static int do_anonymous_page(struct vm_fault *vmf)
>>>>>  		goto unlock_and_release;
>>>>>  
>>>>>  	/* Deliver the page fault to userland, check inside PT lock */
>>>>> -	if (userfaultfd_missing(vma)) {
>>>>> +	if (!(vmf->flags & FAULT_FLAG_SPECULATIVE) && userfaultfd_missing(vma)) {
>>>>>  		pte_unmap_unlock(vmf->pte, vmf->ptl);
>>>>>  		mem_cgroup_cancel_charge(page, memcg, false);
>>>>>  		put_page(page);
>>>>> @@ -3994,13 +4116,22 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>>>  
>>>>>  	if (unlikely(pmd_none(*vmf->pmd))) {
>>>>>  		/*
>>>>> +		 * In the case of the speculative page fault handler we abort
>>>>> +		 * the speculative path immediately as the pmd is probably
>>>>> +		 * in the way to be converted in a huge one. We will try
>>>>> +		 * again holding the mmap_sem (which implies that the collapse
>>>>> +		 * operation is done).
>>>>> +		 */
>>>>> +		if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>>>> +			return VM_FAULT_RETRY;
>>>>> +		/*
>>>>>  		 * Leave __pte_alloc() until later: because vm_ops->fault may
>>>>>  		 * want to allocate huge page, and if we expose page table
>>>>>  		 * for an instant, it will be difficult to retract from
>>>>>  		 * concurrent faults and from rmap lookups.
>>>>>  		 */
>>>>>  		vmf->pte = NULL;
>>>>> -	} else {
>>>>> +	} else if (!(vmf->flags & FAULT_FLAG_SPECULATIVE)) {
>>>>>  		/* See comment in pte_alloc_one_map() */
>>>>>  		if (pmd_devmap_trans_unstable(vmf->pmd))
>>>>>  			return 0;
>>>>> @@ -4009,6 +4140,9 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>>>  		 * pmd from under us anymore at this point because we hold the
>>>>>  		 * mmap_sem read mode and khugepaged takes it in write mode.
>>>>>  		 * So now it's safe to run pte_offset_map().
>>>>> +		 * This is not applicable to the speculative page fault handler
>>>>> +		 * but in that case, the pte is fetched earlier in
>>>>> +		 * handle_speculative_fault().
>>>>>  		 */
>>>>>  		vmf->pte = pte_offset_map(vmf->pmd, vmf->address);
>>>>>  		vmf->orig_pte = *vmf->pte;
>>>>> @@ -4031,6 +4165,8 @@ static int handle_pte_fault(struct vm_fault *vmf)
>>>>>  	if (!vmf->pte) {
>>>>>  		if (vma_is_anonymous(vmf->vma))
>>>>>  			return do_anonymous_page(vmf);
>>>>> +		else if (vmf->flags & FAULT_FLAG_SPECULATIVE)
>>>>> +			return VM_FAULT_RETRY;
>>>>>  		else
>>>>>  			return do_fault(vmf);
>>>>>  	}
>>>>> @@ -4128,6 +4264,9 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>>>  	vmf.pmd = pmd_alloc(mm, vmf.pud, address);
>>>>>  	if (!vmf.pmd)
>>>>>  		return VM_FAULT_OOM;
>>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>>> +	vmf.sequence = raw_read_seqcount(&vma->vm_sequence);
>>>>> +#endif
>>>>>  	if (pmd_none(*vmf.pmd) && transparent_hugepage_enabled(vma)) {
>>>>>  		ret = create_huge_pmd(&vmf);
>>>>>  		if (!(ret & VM_FAULT_FALLBACK))
>>>>> @@ -4161,6 +4300,201 @@ static int __handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
>>>>>  	return handle_pte_fault(&vmf);
>>>>>  }
>>>>>  
>>>>> +#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
>>>>> +/*
>>>>> + * Tries to handle the page fault in a speculative way, without grabbing the
>>>>> + * mmap_sem.
>>>>> + */
>>>>> +int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>> +			       unsigned int flags)
>>>>> +{
>>>>> +	struct vm_fault vmf = {
>>>>> +		.address = address,
>>>>> +	};
>>>>> +	pgd_t *pgd, pgdval;
>>>>> +	p4d_t *p4d, p4dval;
>>>>> +	pud_t pudval;
>>>>> +	int seq, ret = VM_FAULT_RETRY;
>>>>> +	struct vm_area_struct *vma;
>>>>> +#ifdef CONFIG_NUMA
>>>>> +	struct mempolicy *pol;
>>>>> +#endif
>>>>> +
>>>>> +	/* Clear flags that may lead to release the mmap_sem to retry */
>>>>> +	flags &= ~(FAULT_FLAG_ALLOW_RETRY|FAULT_FLAG_KILLABLE);
>>>>> +	flags |= FAULT_FLAG_SPECULATIVE;
>>>>> +
>>>>> +	vma = get_vma(mm, address);
>>>>> +	if (!vma)
>>>>> +		return ret;
>>>>> +
>>>>> +	seq = raw_read_seqcount(&vma->vm_sequence); /* rmb <-> seqlock,vma_rb_erase() */
>>>>> +	if (seq & 1)
>>>>> +		goto out_put;
>>>>> +
>>>>> +	/*
>>>>> +	 * Can't call vm_ops service has we don't know what they would do
>>>>> +	 * with the VMA.
>>>>> +	 * This include huge page from hugetlbfs.
>>>>> +	 */
>>>>> +	if (vma->vm_ops)
>>>>> +		goto out_put;
>>>>> +
>>>>   Hi   Laurent
>>>>    
>>>>    I think that most of pagefault will leave here.   Is there any case  need to skip ?
>>>>   I have tested the following  patch, it work well.
>>> Hi Zhong,
>>>
>>> Well this will allow file mapping to be handle in a speculative way, but that's
>>> a bit dangerous today as there is no guaranty that the vm_ops.vm_fault()
>>> operation will be fair.
>>>
>>> In the case of the anonymous file mapping that's often not a problem, depending
>>> on the underlying file system, but there are so many cases to check and this is
>>> hard to say this can be done in a speculative way as is.
>>  This patch say that spf just handle anonyous page. but I find that do_swap_page
>>  also maybe release the mmap_sem without FAULT_FLAG_RETRY_NOWAIT. why is it safe
>>  to handle the case. I think that the case is similar to file page. Maybe I miss
>>  something else.
> do_swap_page() may released the mmap_sem through the call to
> __lock_page_or_retry(), but this can only happen if FAULT_FLAG_ALLOW_RETRY or
> FAULT_FLAG_KILLABLE are set and they are unset in __handle_speculative_fault().
 For spf. Indeed. Thank you for clarification.
>>  I test the patches and find just only 18% of the pagefault will enter into the
>>  speculative page fault during a process startup. As I had said. most of pagefault
>>  will be handled by ops->fault. I do not know the data you had posted is how to get.
> I do agree that handling file mapping will be required, but this will add more
> complexity to this series, since we need a way for drivers to tell they are
> compatible with the speculative path.
 As the above mentioned. the specualtive page fault do not pass FAULT_FLAG_ALLOW_RETRY.
 In other words, File page will not refer to release mmap_sem for spf.
 
 but I am still not quite clear that what should drivers do to compatible with speculatve path.
 The speculative path should not refer to the mmap_sem for filemap_fault.

 Thanks,
 zhong jiang
> May be I should give it a try on the next send.
 Ok, I will try.
> For my information, what was the performance improvement you seen when handling
> file page faulting this way ?
 I am sorry that. It is the data that Ganesh test the launch time on Andriod.
> Thanks,
> Laurent.
>
>>
>>  Thanks
>>  zhong jiang
>>> The huge work to do is to double check that all the code called by
>>> vm_ops.fault() is not dealing with the mmap_sem, which could be handled using
>>> FAULT_FLAG_RETRY_NOWAIT, and care is also needed about the resources that code
>>> is managing as it may assume that it is under the protection of the mmap_sem in
>>> read mode, and that can be done implicitly.
>>>
>>> Cheers,
>>> Laurent.
>>>
>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>> index 936128b..9bc1545 100644
>>>>  @@ -3893,8 +3898,6 @@ static int handle_pte_fault(struct fault_env *fe)
>>>>         if (!fe->pte) {
>>>>                 if (vma_is_anonymous(fe->vma))
>>>>                         return do_anonymous_page(fe);
>>>> -               else if (fe->flags & FAULT_FLAG_SPECULATIVE)
>>>> -                       return VM_FAULT_RETRY;
>>>>                 else
>>>>                         return do_fault(fe);
>>>>         }
>>>> @@ -4026,20 +4029,11 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>                 goto out_put;
>>>>         }
>>>>         /*
>>>> -        * Can't call vm_ops service has we don't know what they would do
>>>> -        * with the VMA.
>>>> -        * This include huge page from hugetlbfs.
>>>> -        */
>>>> -       if (vma->vm_ops) {
>>>> -               trace_spf_vma_notsup(_RET_IP_, vma, address);
>>>> -               goto out_put;
>>>> -       }
>>>>
>>>>
>>>> Thanks
>>>> zhong jiang
>>>>> +	/*
>>>>> +	 * __anon_vma_prepare() requires the mmap_sem to be held
>>>>> +	 * because vm_next and vm_prev must be safe. This can't be guaranteed
>>>>> +	 * in the speculative path.
>>>>> +	 */
>>>>> +	if (unlikely(!vma->anon_vma))
>>>>> +		goto out_put;
>>>>> +
>>>>> +	vmf.vma_flags = READ_ONCE(vma->vm_flags);
>>>>> +	vmf.vma_page_prot = READ_ONCE(vma->vm_page_prot);
>>>>> +
>>>>> +	/* Can't call userland page fault handler in the speculative path */
>>>>> +	if (unlikely(vmf.vma_flags & VM_UFFD_MISSING))
>>>>> +		goto out_put;
>>>>> +
>>>>> +	if (vmf.vma_flags & VM_GROWSDOWN || vmf.vma_flags & VM_GROWSUP)
>>>>> +		/*
>>>>> +		 * This could be detected by the check address against VMA's
>>>>> +		 * boundaries but we want to trace it as not supported instead
>>>>> +		 * of changed.
>>>>> +		 */
>>>>> +		goto out_put;
>>>>> +
>>>>> +	if (address < READ_ONCE(vma->vm_start)
>>>>> +	    || READ_ONCE(vma->vm_end) <= address)
>>>>> +		goto out_put;
>>>>> +
>>>>> +	if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
>>>>> +				       flags & FAULT_FLAG_INSTRUCTION,
>>>>> +				       flags & FAULT_FLAG_REMOTE)) {
>>>>> +		ret = VM_FAULT_SIGSEGV;
>>>>> +		goto out_put;
>>>>> +	}
>>>>> +
>>>>> +	/* This is one is required to check that the VMA has write access set */
>>>>> +	if (flags & FAULT_FLAG_WRITE) {
>>>>> +		if (unlikely(!(vmf.vma_flags & VM_WRITE))) {
>>>>> +			ret = VM_FAULT_SIGSEGV;
>>>>> +			goto out_put;
>>>>> +		}
>>>>> +	} else if (unlikely(!(vmf.vma_flags & (VM_READ|VM_EXEC|VM_WRITE)))) {
>>>>> +		ret = VM_FAULT_SIGSEGV;
>>>>> +		goto out_put;
>>>>> +	}
>>>>> +
>>>>> +#ifdef CONFIG_NUMA
>>>>> +	/*
>>>>> +	 * MPOL_INTERLEAVE implies additional checks in
>>>>> +	 * mpol_misplaced() which are not compatible with the
>>>>> +	 *speculative page fault processing.
>>>>> +	 */
>>>>> +	pol = __get_vma_policy(vma, address);
>>>>> +	if (!pol)
>>>>> +		pol = get_task_policy(current);
>>>>> +	if (pol && pol->mode == MPOL_INTERLEAVE)
>>>>> +		goto out_put;
>>>>> +#endif
>>>>> +
>>>>> +	/*
>>>>> +	 * Do a speculative lookup of the PTE entry.
>>>>> +	 */
>>>>> +	local_irq_disable();
>>>>> +	pgd = pgd_offset(mm, address);
>>>>> +	pgdval = READ_ONCE(*pgd);
>>>>> +	if (pgd_none(pgdval) || unlikely(pgd_bad(pgdval)))
>>>>> +		goto out_walk;
>>>>> +
>>>>> +	p4d = p4d_offset(pgd, address);
>>>>> +	p4dval = READ_ONCE(*p4d);
>>>>> +	if (p4d_none(p4dval) || unlikely(p4d_bad(p4dval)))
>>>>> +		goto out_walk;
>>>>> +
>>>>> +	vmf.pud = pud_offset(p4d, address);
>>>>> +	pudval = READ_ONCE(*vmf.pud);
>>>>> +	if (pud_none(pudval) || unlikely(pud_bad(pudval)))
>>>>> +		goto out_walk;
>>>>> +
>>>>> +	/* Huge pages at PUD level are not supported. */
>>>>> +	if (unlikely(pud_trans_huge(pudval)))
>>>>> +		goto out_walk;
>>>>> +
>>>>> +	vmf.pmd = pmd_offset(vmf.pud, address);
>>>>> +	vmf.orig_pmd = READ_ONCE(*vmf.pmd);
>>>>> +	/*
>>>>> +	 * pmd_none could mean that a hugepage collapse is in progress
>>>>> +	 * in our back as collapse_huge_page() mark it before
>>>>> +	 * invalidating the pte (which is done once the IPI is catched
>>>>> +	 * by all CPU and we have interrupt disabled).
>>>>> +	 * For this reason we cannot handle THP in a speculative way since we
>>>>> +	 * can't safely indentify an in progress collapse operation done in our
>>>>> +	 * back on that PMD.
>>>>> +	 * Regarding the order of the following checks, see comment in
>>>>> +	 * pmd_devmap_trans_unstable()
>>>>> +	 */
>>>>> +	if (unlikely(pmd_devmap(vmf.orig_pmd) ||
>>>>> +		     pmd_none(vmf.orig_pmd) || pmd_trans_huge(vmf.orig_pmd) ||
>>>>> +		     is_swap_pmd(vmf.orig_pmd)))
>>>>> +		goto out_walk;
>>>>> +
>>>>> +	/*
>>>>> +	 * The above does not allocate/instantiate page-tables because doing so
>>>>> +	 * would lead to the possibility of instantiating page-tables after
>>>>> +	 * free_pgtables() -- and consequently leaking them.
>>>>> +	 *
>>>>> +	 * The result is that we take at least one !speculative fault per PMD
>>>>> +	 * in order to instantiate it.
>>>>> +	 */
>>>>> +
>>>>> +	vmf.pte = pte_offset_map(vmf.pmd, address);
>>>>> +	vmf.orig_pte = READ_ONCE(*vmf.pte);
>>>>> +	barrier(); /* See comment in handle_pte_fault() */
>>>>> +	if (pte_none(vmf.orig_pte)) {
>>>>> +		pte_unmap(vmf.pte);
>>>>> +		vmf.pte = NULL;
>>>>> +	}
>>>>> +
>>>>> +	vmf.vma = vma;
>>>>> +	vmf.pgoff = linear_page_index(vma, address);
>>>>> +	vmf.gfp_mask = __get_fault_gfp_mask(vma);
>>>>> +	vmf.sequence = seq;
>>>>> +	vmf.flags = flags;
>>>>> +
>>>>> +	local_irq_enable();
>>>>> +
>>>>> +	/*
>>>>> +	 * We need to re-validate the VMA after checking the bounds, otherwise
>>>>> +	 * we might have a false positive on the bounds.
>>>>> +	 */
>>>>> +	if (read_seqcount_retry(&vma->vm_sequence, seq))
>>>>> +		goto out_put;
>>>>> +
>>>>> +	mem_cgroup_oom_enable();
>>>>> +	ret = handle_pte_fault(&vmf);
>>>>> +	mem_cgroup_oom_disable();
>>>>> +
>>>>> +	put_vma(vma);
>>>>> +
>>>>> +	/*
>>>>> +	 * The task may have entered a memcg OOM situation but
>>>>> +	 * if the allocation error was handled gracefully (no
>>>>> +	 * VM_FAULT_OOM), there is no need to kill anything.
>>>>> +	 * Just clean up the OOM state peacefully.
>>>>> +	 */
>>>>> +	if (task_in_memcg_oom(current) && !(ret & VM_FAULT_OOM))
>>>>> +		mem_cgroup_oom_synchronize(false);
>>>>> +	return ret;
>>>>> +
>>>>> +out_walk:
>>>>> +	local_irq_enable();
>>>>> +out_put:
>>>>> +	put_vma(vma);
>>>>> +	return ret;
>>>>> +}
>>>>> +#endif /* CONFIG_SPECULATIVE_PAGE_FAULT */
>>>>> +
>>>>>  /*
>>>>>   * By the time we get here, we already hold the mm semaphore
>>>>>   *
>>> .
>>>
>>
>
> .
>

RE: [PATCH v11 00/26] Speculative page faults

[Song, HaiyanX]

[-- Attachment #1: Type: text/plain, Size: 42188 bytes --]

Hi Laurent,

Thanks for your analysis for the last perf results.
Your mentioned ," the major differences at the head of the perf report is the 92% testcase which is weirdly not reported
on the head side", which is a bug of 0-day,and it caused the item is not counted in perf. 

I've triggered the test page_fault2 and page_fault3 again only with thread mode of will-it-scale on 0-day (on the same test box,every case tested 3 times).
I checked the perf report have no above mentioned problem.

I have compared them, found some items have difference, such as below case:
       page_fault2-thp-always: handle_mm_fault, base: 45.22%    head: 29.41%
       page_fault3-thp-always: handle_mm_fault, base: 22.95%    head: 14.15%       

So i attached the perf result in mail again, could your have a look again for checking the difference between base and head commit.


Thanks,
Haiyan, Song
 
________________________________________
From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
Sent: Tuesday, July 17, 2018 5:36 PM
To: Song, HaiyanX
Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
Subject: Re: [PATCH v11 00/26] Speculative page faults

On 13/07/2018 05:56, Song, HaiyanX wrote:
> Hi Laurent,

Hi Haiyan,

Thanks a lot for sharing this perf reports.

I looked at them closely, and I've to admit that I was not able to found a
major difference between the base and the head report, except that
handle_pte_fault() is no more in-lined in the head one.

As expected, __handle_speculative_fault() is never traced since these tests are
dealing with file mapping, not handled in the speculative way.

When running these test did you seen a major differences in the test's result
between base and head ?

>From the number of cycles counted, the biggest difference is page_fault3 when
run with the THP enabled:
                                BASE            HEAD            Delta
page_fault2_base_thp_never      1142252426747   1065866197589   -6.69%
page_fault2_base_THP-Alwasys    1124844374523   1076312228927   -4.31%
page_fault3_base_thp_never      1099387298152   1134118402345   3.16%
page_fault3_base_THP-Always     1059370178101   853985561949    -19.39%


The very weird thing is the difference of the delta cycles reported between
thp never and thp always, because the speculative way is aborted when checking
for the vma->ops field, which is the same in both case, and the thp is never
checked. So there is no code covering differnce, on the speculative path,
between these 2 cases. This leads me to think that there are other interactions
interfering in the measure.

Looking at the perf-profile_page_fault3_*_THP-Always, the major differences at
the head of the perf report is the 92% testcase which is weirdly not reported
on the head side :
    92.02%    22.33%  page_fault3_processes  [.] testcase
92.02% testcase

Then the base reported 37.67% for __do_page_fault() where the head reported
48.41%, but the only difference in this function, between base and head, is the
call to handle_speculative_fault(). But this is a macro checking for the fault
flags, and mm->users and then calling __handle_speculative_fault() if needed.
So this can't explain this difference, except if __handle_speculative_fault()
is inlined in __do_page_fault().
Is this the case on your build ?

Haiyan, do you still have the output of the test to check those numbers too ?

Cheers,
Laurent

> I attached the perf-profile.gz file for case page_fault2 and page_fault3. These files were captured during test the related test case.
> Please help to check on these data if it can help you to find the higher change. Thanks.
>
> File name perf-profile_page_fault2_head_THP-Always.gz, means the perf-profile result get from page_fault2
>     tested for head commit (a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12) with THP_always configuration.
>
> Best regards,
> Haiyan Song
>
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Thursday, July 12, 2018 1:05 AM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
>
> Hi Haiyan,
>
> Do you get a chance to capture some performance cycles on your system ?
> I still can't get these numbers on my hardware.
>
> Thanks,
> Laurent.
>
> On 04/07/2018 09:51, Laurent Dufour wrote:
>> On 04/07/2018 05:23, Song, HaiyanX wrote:
>>> Hi Laurent,
>>>
>>>
>>> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
>>> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.
>>
>> Repeating the test only 3 times seems a bit too low to me.
>>
>> I'll focus on the higher change for the moment, but I don't have access to such
>> a hardware.
>>
>> Is possible to provide a diff between base and SPF of the performance cycles
>> measured when running page_fault3 and page_fault2 when the 20% change is detected.
>>
>> Please stay focus on the test case process to see exactly where the series is
>> impacting.
>>
>> Thanks,
>> Laurent.
>>
>>>
>>> And I did not find other high variation on test case result.
>>>
>>> a). Enable THP
>>> testcase                          base     stddev       change      head     stddev         metric
>>> page_fault3/enable THP           10519      ± 3%        -20.5%      8368      ±6%          will-it-scale.per_thread_ops
>>> page_fault2/enalbe THP            8281      ± 2%        -18.8%      6728                   will-it-scale.per_thread_ops
>>> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
>>> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
>>> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
>>>
>>> b). Disable THP
>>> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
>>> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
>>> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
>>> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
>>> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
>>> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
>>> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
>>>
>>>
>>> Best regards,
>>> Haiyan Song
>>> ________________________________________
>>> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Monday, July 02, 2018 4:59 PM
>>> To: Song, HaiyanX
>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>
>>> On 11/06/2018 09:49, Song, HaiyanX wrote:
>>>> Hi Laurent,
>>>>
>>>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>>>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>>>> V9 patch serials.
>>>>
>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>>>> commit id:
>>>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>>>> Benchmark: will-it-scale
>>>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>>>
>>>> Metrics:
>>>>   will-it-scale.per_process_ops=processes/nr_cpu
>>>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>> THP: enable / disable
>>>> nr_task:100%
>>>>
>>>> 1. Regressions:
>>>>
>>>> a). Enable THP
>>>> testcase                          base           change      head           metric
>>>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>>>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>>>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>>>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>>>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>>>
>>>> b). Disable THP
>>>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>>>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>>>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>>>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>>>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>>>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>>>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>>>
>>>> Notes: for the above  values of test result, the higher is better.
>>>
>>> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
>>> get reproducible results. The results have huge variation, even on the vanilla
>>> kernel, and I can't state on any changes due to that.
>>>
>>> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
>>> measure any changes between the vanilla and the SPF patched ones:
>>>
>>> test THP enabled                4.17.0-rc4-mm1  spf             delta
>>> page_fault3_threads             2697.7          2683.5          -0.53%
>>> page_fault2_threads             170660.6        169574.1        -0.64%
>>> context_switch1_threads         6915269.2       6877507.3       -0.55%
>>> context_switch1_processes       6478076.2       6529493.5       0.79%
>>> brk1                            243391.2        238527.5        -2.00%
>>>
>>> Tests were run 10 times, no high variation detected.
>>>
>>> Did you see high variation on your side ? How many times the test were run to
>>> compute the average values ?
>>>
>>> Thanks,
>>> Laurent.
>>>
>>>
>>>>
>>>> 2. Improvement: not found improvement based on the selected test cases.
>>>>
>>>>
>>>> Best regards
>>>> Haiyan Song
>>>> ________________________________________
>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>> Sent: Monday, May 28, 2018 4:54 PM
>>>> To: Song, HaiyanX
>>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>>
>>>> On 28/05/2018 10:22, Haiyan Song wrote:
>>>>> Hi Laurent,
>>>>>
>>>>> Yes, these tests are done on V9 patch.
>>>>
>>>> Do you plan to give this V11 a run ?
>>>>
>>>>>
>>>>>
>>>>> Best regards,
>>>>> Haiyan Song
>>>>>
>>>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>>>
>>>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>>>> tested on Intel 4s Skylake platform.
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>>>> series" while responding to the v11 header series...
>>>>>> Were these tests done on v9 or v11 ?
>>>>>>
>>>>>> Cheers,
>>>>>> Laurent.
>>>>>>
>>>>>>>
>>>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>>>> Commit id:
>>>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>>>> Benchmark suite: will-it-scale
>>>>>>> Download link:
>>>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>>>> Metrics:
>>>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>>>> THP: enable / disable
>>>>>>> nr_task: 100%
>>>>>>>
>>>>>>> 1. Regressions:
>>>>>>> a) THP enabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>>>
>>>>>>> b) THP disabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>>>
>>>>>>> 2. Improvements:
>>>>>>> a) THP enabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>>>
>>>>>>> b) THP disabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>>>
>>>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>>>
>>>>>>>
>>>>>>> Best regards
>>>>>>> Haiyan Song
>>>>>>>
>>>>>>> ________________________________________
>>>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>>>
>>>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>>>> page fault without holding the mm semaphore [1].
>>>>>>>
>>>>>>> The idea is to try to handle user space page faults without holding the
>>>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>>>> process since the page fault handler will not wait for other threads memory
>>>>>>> layout change to be done, assuming that this change is done in another part
>>>>>>> of the process's memory space. This type page fault is named speculative
>>>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>>>> is failing its processing and a classic page fault is then tried.
>>>>>>>
>>>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>>>> limiting the locking contention to these operations which are expected to
>>>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>>>> benchmark anymore.
>>>>>>>
>>>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>>>> have to be protected against parallel changes. This is done by using a per
>>>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>>>> speculative page fault in that case.
>>>>>>>
>>>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>>>> checked during the page fault are modified.
>>>>>>>
>>>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>>>> parallel change is possible at this time.
>>>>>>>
>>>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>>>> classic page fault handler will be called to handle the operation while
>>>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>>>> PTE.
>>>>>>>
>>>>>>> In pseudo code, this could be seen as:
>>>>>>>     speculative_page_fault()
>>>>>>>     {
>>>>>>>             vma = get_vma()
>>>>>>>             check vma sequence count
>>>>>>>             check vma's support
>>>>>>>             disable interrupt
>>>>>>>                   check pgd,p4d,...,pte
>>>>>>>                   save pmd and pte in vmf
>>>>>>>                   save vma sequence counter in vmf
>>>>>>>             enable interrupt
>>>>>>>             check vma sequence count
>>>>>>>             handle_pte_fault(vma)
>>>>>>>                     ..
>>>>>>>                     page = alloc_page()
>>>>>>>                     pte_map_lock()
>>>>>>>                             disable interrupt
>>>>>>>                                     abort if sequence counter has changed
>>>>>>>                                     abort if pmd or pte has changed
>>>>>>>                                     pte map and lock
>>>>>>>                             enable interrupt
>>>>>>>                     if abort
>>>>>>>                        free page
>>>>>>>                        abort
>>>>>>>                     ...
>>>>>>>     }
>>>>>>>
>>>>>>>     arch_fault_handler()
>>>>>>>     {
>>>>>>>             if (speculative_page_fault(&vma))
>>>>>>>                goto done
>>>>>>>     again:
>>>>>>>             lock(mmap_sem)
>>>>>>>             vma = find_vma();
>>>>>>>             handle_pte_fault(vma);
>>>>>>>             if retry
>>>>>>>                unlock(mmap_sem)
>>>>>>>                goto again;
>>>>>>>     done:
>>>>>>>             handle fault error
>>>>>>>     }
>>>>>>>
>>>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>>>
>>>>>>> This series add a new software performance event named 'speculative-faults'
>>>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>>>> the part of the faults processed speculatively.
>>>>>>>
>>>>>>> There are some trace events introduced by this series. They allow
>>>>>>> identifying why the page faults were not processed speculatively. This
>>>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>>>> grouped in a system named 'pagefault', they are:
>>>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>>>    back.
>>>>>>>
>>>>>>> To record all the related events, the easier is to run perf with the
>>>>>>> following arguments :
>>>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>>>
>>>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>>>> page fault handled speculatively. I can be seen this way:
>>>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>>>
>>>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>>>> on x86, PowerPC and arm64.
>>>>>>>
>>>>>>> ---------------------
>>>>>>> Real Workload results
>>>>>>>
>>>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>>>> this new version should not impact the performance boost seen.
>>>>>>>
>>>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>>>> series:
>>>>>>>                 vanilla         spf
>>>>>>> faults          89.418          101.364         +13%
>>>>>>> spf                n/a           97.989
>>>>>>>
>>>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>>>> way.
>>>>>>>
>>>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>>>> it a try on an android device. He reported that the application launch time
>>>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>>>> 20%.
>>>>>>>
>>>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>>>
>>>>>>> Application                             4.9     4.9+spf delta
>>>>>>> com.tencent.mm                          416     389     -7%
>>>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>>>> com.tencent.mtt                         455     454     0%
>>>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>>>> com.autonavi.minimap                    711     701     -1%
>>>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>>>> com.immomo.momo                         501     487     -3%
>>>>>>> com.tencent.peng                        2145    2112    -2%
>>>>>>> com.smile.gifmaker                      491     461     -6%
>>>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>>>> com.baidu.searchbox                     333     314     -6%
>>>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>>>> com.sina.weibo                          907     906     0%
>>>>>>> com.youku.phone                         816     731     -11%
>>>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>>>> com.UCMobile                            415     411     -1%
>>>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>>>
>>>>>>> ------------------
>>>>>>> Benchmarks results
>>>>>>>
>>>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>>>> SPF is BASE + this series
>>>>>>>
>>>>>>> Kernbench:
>>>>>>> ----------
>>>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>>>> kernel (kernel is build 5 times):
>>>>>>>
>>>>>>> Average Half load -j 8
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>>>
>>>>>>> Average Optimal load -j 16
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>>>
>>>>>>>
>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>          526743764      faults
>>>>>>>                210      spf
>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>               2278      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> Very few speculative page faults were recorded as most of the processes
>>>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>>>> were created during the kernel build processing).
>>>>>>>
>>>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>>>
>>>>>>> Average Half load -j 40
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>>>
>>>>>>> Average Optimal load -j 80
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>>>
>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>          116730856      faults
>>>>>>>                  0      spf
>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                476      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>>>> there is no impact on the performance.
>>>>>>>
>>>>>>> Ebizzy:
>>>>>>> -------
>>>>>>> The test is counting the number of records per second it can manage, the
>>>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>>>> consistent result I repeated the test 100 times and measure the average
>>>>>>> result. The number is the record processes per second, the higher is the
>>>>>>> best.
>>>>>>>
>>>>>>>                 BASE            SPF             delta
>>>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>>>
>>>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>>>            1706379      faults
>>>>>>>            1674599      spf
>>>>>>>              30588      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                363      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>>>            1874773      faults
>>>>>>>            1461153      spf
>>>>>>>             413293      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                200      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>>>> leading the ebizzy performance boost.
>>>>>>>
>>>>>>> ------------------
>>>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>>>    and Minchan Kim, hopefully.
>>>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>>>    __do_page_fault().
>>>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>>>    instead
>>>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>>>> useless
>>>>>>>    trace event pagefault:spf_pte_lock.
>>>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>>>
>>>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>>>
>>>>>>>
>>>>>>> Laurent Dufour (20):
>>>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>>>   mm: introduce INIT_VMA()
>>>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>>>   mm: protect mremap() against SPF hanlder
>>>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>>>   mm: introduce __vm_normal_page()
>>>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>>>   mm: adding speculative page fault failure trace events
>>>>>>>   perf: add a speculative page fault sw event
>>>>>>>   perf tools: add support for the SPF perf event
>>>>>>>   mm: add speculative page fault vmstats
>>>>>>>   powerpc/mm: add speculative page fault
>>>>>>>
>>>>>>> Mahendran Ganesh (2):
>>>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   arm64/mm: add speculative page fault
>>>>>>>
>>>>>>> Peter Zijlstra (4):
>>>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>>>   mm: VMA sequence count
>>>>>>>   mm: provide speculative fault infrastructure
>>>>>>>   x86/mm: add speculative pagefault handling
>>>>>>>
>>>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>>>  fs/exec.c                             |   2 +-
>>>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>>>  include/linux/migrate.h               |   4 +-
>>>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>>>  include/linux/mm_types.h              |   7 +
>>>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>>>  include/linux/swap.h                  |  10 +-
>>>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>>>  kernel/fork.c                         |   5 +-
>>>>>>>  mm/Kconfig                            |  22 ++
>>>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>>>  mm/hugetlb.c                          |   2 +
>>>>>>>  mm/init-mm.c                          |   3 +
>>>>>>>  mm/internal.h                         |  20 ++
>>>>>>>  mm/khugepaged.c                       |   5 +
>>>>>>>  mm/madvise.c                          |   6 +-
>>>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>>>  mm/migrate.c                          |   6 +-
>>>>>>>  mm/mlock.c                            |  13 +-
>>>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>>>  mm/mprotect.c                         |   4 +-
>>>>>>>  mm/mremap.c                           |  13 +
>>>>>>>  mm/nommu.c                            |   2 +-
>>>>>>>  mm/rmap.c                             |   5 +-
>>>>>>>  mm/swap.c                             |   6 +-
>>>>>>>  mm/swap_state.c                       |   8 +-
>>>>>>>  mm/vmstat.c                           |   5 +-
>>>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>>>  tools/perf/util/python.c              |   1 +
>>>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>>>
>>>>>>> --
>>>>>>> 2.7.4
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>>
>>
>


[-- Attachment #2: perf-profile_page_fault2_base_thp_always.gz --]
[-- Type: application/gzip, Size: 12167 bytes --]

[-- Attachment #3: perf-profile_page_fault2_base_thp_never.gz --]
[-- Type: application/gzip, Size: 11543 bytes --]

[-- Attachment #4: perf-profile_page_fault2_head_thp_always.gz --]
[-- Type: application/gzip, Size: 12019 bytes --]

[-- Attachment #5: perf-profile_page_fault3_base_thp_always.gz --]
[-- Type: application/gzip, Size: 12701 bytes --]

[-- Attachment #6: perf-profile_page_fault3_base_thp_always.gz --]
[-- Type: application/gzip, Size: 12701 bytes --]

[-- Attachment #7: perf-profile_page_fault3_base_thp_never.gz --]
[-- Type: application/gzip, Size: 12699 bytes --]

RE: [PATCH v11 00/26] Speculative page faults

[Song, HaiyanX]

[-- Attachment #1: Type: text/plain, Size: 43157 bytes --]

Add another 3 perf file.
________________________________________
From: Song, HaiyanX
Sent: Friday, August 03, 2018 2:36 PM
To: Laurent Dufour
Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
Subject: RE: [PATCH v11 00/26] Speculative page faults

Hi Laurent,

Thanks for your analysis for the last perf results.
Your mentioned ," the major differences at the head of the perf report is the 92% testcase which is weirdly not reported
on the head side", which is a bug of 0-day,and it caused the item is not counted in perf.

I've triggered the test page_fault2 and page_fault3 again only with thread mode of will-it-scale on 0-day (on the same test box,every case tested 3 times).
I checked the perf report have no above mentioned problem.

I have compared them, found some items have difference, such as below case:
       page_fault2-thp-always: handle_mm_fault, base: 45.22%    head: 29.41%
       page_fault3-thp-always: handle_mm_fault, base: 22.95%    head: 14.15%

So i attached the perf result in mail again, could your have a look again for checking the difference between base and head commit.


Thanks,
Haiyan, Song

________________________________________
From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
Sent: Tuesday, July 17, 2018 5:36 PM
To: Song, HaiyanX
Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
Subject: Re: [PATCH v11 00/26] Speculative page faults

On 13/07/2018 05:56, Song, HaiyanX wrote:
> Hi Laurent,

Hi Haiyan,

Thanks a lot for sharing this perf reports.

I looked at them closely, and I've to admit that I was not able to found a
major difference between the base and the head report, except that
handle_pte_fault() is no more in-lined in the head one.

As expected, __handle_speculative_fault() is never traced since these tests are
dealing with file mapping, not handled in the speculative way.

When running these test did you seen a major differences in the test's result
between base and head ?

>From the number of cycles counted, the biggest difference is page_fault3 when
run with the THP enabled:
                                BASE            HEAD            Delta
page_fault2_base_thp_never      1142252426747   1065866197589   -6.69%
page_fault2_base_THP-Alwasys    1124844374523   1076312228927   -4.31%
page_fault3_base_thp_never      1099387298152   1134118402345   3.16%
page_fault3_base_THP-Always     1059370178101   853985561949    -19.39%


The very weird thing is the difference of the delta cycles reported between
thp never and thp always, because the speculative way is aborted when checking
for the vma->ops field, which is the same in both case, and the thp is never
checked. So there is no code covering differnce, on the speculative path,
between these 2 cases. This leads me to think that there are other interactions
interfering in the measure.

Looking at the perf-profile_page_fault3_*_THP-Always, the major differences at
the head of the perf report is the 92% testcase which is weirdly not reported
on the head side :
    92.02%    22.33%  page_fault3_processes  [.] testcase
92.02% testcase

Then the base reported 37.67% for __do_page_fault() where the head reported
48.41%, but the only difference in this function, between base and head, is the
call to handle_speculative_fault(). But this is a macro checking for the fault
flags, and mm->users and then calling __handle_speculative_fault() if needed.
So this can't explain this difference, except if __handle_speculative_fault()
is inlined in __do_page_fault().
Is this the case on your build ?

Haiyan, do you still have the output of the test to check those numbers too ?

Cheers,
Laurent

> I attached the perf-profile.gz file for case page_fault2 and page_fault3. These files were captured during test the related test case.
> Please help to check on these data if it can help you to find the higher change. Thanks.
>
> File name perf-profile_page_fault2_head_THP-Always.gz, means the perf-profile result get from page_fault2
>     tested for head commit (a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12) with THP_always configuration.
>
> Best regards,
> Haiyan Song
>
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Thursday, July 12, 2018 1:05 AM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
>
> Hi Haiyan,
>
> Do you get a chance to capture some performance cycles on your system ?
> I still can't get these numbers on my hardware.
>
> Thanks,
> Laurent.
>
> On 04/07/2018 09:51, Laurent Dufour wrote:
>> On 04/07/2018 05:23, Song, HaiyanX wrote:
>>> Hi Laurent,
>>>
>>>
>>> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
>>> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.
>>
>> Repeating the test only 3 times seems a bit too low to me.
>>
>> I'll focus on the higher change for the moment, but I don't have access to such
>> a hardware.
>>
>> Is possible to provide a diff between base and SPF of the performance cycles
>> measured when running page_fault3 and page_fault2 when the 20% change is detected.
>>
>> Please stay focus on the test case process to see exactly where the series is
>> impacting.
>>
>> Thanks,
>> Laurent.
>>
>>>
>>> And I did not find other high variation on test case result.
>>>
>>> a). Enable THP
>>> testcase                          base     stddev       change      head     stddev         metric
>>> page_fault3/enable THP           10519      ± 3%        -20.5%      8368      ±6%          will-it-scale.per_thread_ops
>>> page_fault2/enalbe THP            8281      ± 2%        -18.8%      6728                   will-it-scale.per_thread_ops
>>> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
>>> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
>>> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
>>>
>>> b). Disable THP
>>> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
>>> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
>>> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
>>> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
>>> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
>>> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
>>> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
>>>
>>>
>>> Best regards,
>>> Haiyan Song
>>> ________________________________________
>>> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>> Sent: Monday, July 02, 2018 4:59 PM
>>> To: Song, HaiyanX
>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>
>>> On 11/06/2018 09:49, Song, HaiyanX wrote:
>>>> Hi Laurent,
>>>>
>>>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>>>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>>>> V9 patch serials.
>>>>
>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>>>> commit id:
>>>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>>>> Benchmark: will-it-scale
>>>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>>>
>>>> Metrics:
>>>>   will-it-scale.per_process_ops=processes/nr_cpu
>>>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>> THP: enable / disable
>>>> nr_task:100%
>>>>
>>>> 1. Regressions:
>>>>
>>>> a). Enable THP
>>>> testcase                          base           change      head           metric
>>>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>>>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>>>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>>>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>>>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>>>
>>>> b). Disable THP
>>>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>>>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>>>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>>>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>>>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>>>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>>>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>>>
>>>> Notes: for the above  values of test result, the higher is better.
>>>
>>> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
>>> get reproducible results. The results have huge variation, even on the vanilla
>>> kernel, and I can't state on any changes due to that.
>>>
>>> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
>>> measure any changes between the vanilla and the SPF patched ones:
>>>
>>> test THP enabled                4.17.0-rc4-mm1  spf             delta
>>> page_fault3_threads             2697.7          2683.5          -0.53%
>>> page_fault2_threads             170660.6        169574.1        -0.64%
>>> context_switch1_threads         6915269.2       6877507.3       -0.55%
>>> context_switch1_processes       6478076.2       6529493.5       0.79%
>>> brk1                            243391.2        238527.5        -2.00%
>>>
>>> Tests were run 10 times, no high variation detected.
>>>
>>> Did you see high variation on your side ? How many times the test were run to
>>> compute the average values ?
>>>
>>> Thanks,
>>> Laurent.
>>>
>>>
>>>>
>>>> 2. Improvement: not found improvement based on the selected test cases.
>>>>
>>>>
>>>> Best regards
>>>> Haiyan Song
>>>> ________________________________________
>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>> Sent: Monday, May 28, 2018 4:54 PM
>>>> To: Song, HaiyanX
>>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>>
>>>> On 28/05/2018 10:22, Haiyan Song wrote:
>>>>> Hi Laurent,
>>>>>
>>>>> Yes, these tests are done on V9 patch.
>>>>
>>>> Do you plan to give this V11 a run ?
>>>>
>>>>>
>>>>>
>>>>> Best regards,
>>>>> Haiyan Song
>>>>>
>>>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>>>
>>>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>>>> tested on Intel 4s Skylake platform.
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>>>> series" while responding to the v11 header series...
>>>>>> Were these tests done on v9 or v11 ?
>>>>>>
>>>>>> Cheers,
>>>>>> Laurent.
>>>>>>
>>>>>>>
>>>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>>>> Commit id:
>>>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>>>> Benchmark suite: will-it-scale
>>>>>>> Download link:
>>>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>>>> Metrics:
>>>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>>>> THP: enable / disable
>>>>>>> nr_task: 100%
>>>>>>>
>>>>>>> 1. Regressions:
>>>>>>> a) THP enabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>>>
>>>>>>> b) THP disabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>>>
>>>>>>> 2. Improvements:
>>>>>>> a) THP enabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>>>
>>>>>>> b) THP disabled:
>>>>>>> testcase                        base            change          head       metric
>>>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>>>
>>>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>>>
>>>>>>>
>>>>>>> Best regards
>>>>>>> Haiyan Song
>>>>>>>
>>>>>>> ________________________________________
>>>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>>>
>>>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>>>> page fault without holding the mm semaphore [1].
>>>>>>>
>>>>>>> The idea is to try to handle user space page faults without holding the
>>>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>>>> process since the page fault handler will not wait for other threads memory
>>>>>>> layout change to be done, assuming that this change is done in another part
>>>>>>> of the process's memory space. This type page fault is named speculative
>>>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>>>> is failing its processing and a classic page fault is then tried.
>>>>>>>
>>>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>>>> limiting the locking contention to these operations which are expected to
>>>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>>>> benchmark anymore.
>>>>>>>
>>>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>>>> have to be protected against parallel changes. This is done by using a per
>>>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>>>> speculative page fault in that case.
>>>>>>>
>>>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>>>> checked during the page fault are modified.
>>>>>>>
>>>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>>>> parallel change is possible at this time.
>>>>>>>
>>>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>>>> classic page fault handler will be called to handle the operation while
>>>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>>>> PTE.
>>>>>>>
>>>>>>> In pseudo code, this could be seen as:
>>>>>>>     speculative_page_fault()
>>>>>>>     {
>>>>>>>             vma = get_vma()
>>>>>>>             check vma sequence count
>>>>>>>             check vma's support
>>>>>>>             disable interrupt
>>>>>>>                   check pgd,p4d,...,pte
>>>>>>>                   save pmd and pte in vmf
>>>>>>>                   save vma sequence counter in vmf
>>>>>>>             enable interrupt
>>>>>>>             check vma sequence count
>>>>>>>             handle_pte_fault(vma)
>>>>>>>                     ..
>>>>>>>                     page = alloc_page()
>>>>>>>                     pte_map_lock()
>>>>>>>                             disable interrupt
>>>>>>>                                     abort if sequence counter has changed
>>>>>>>                                     abort if pmd or pte has changed
>>>>>>>                                     pte map and lock
>>>>>>>                             enable interrupt
>>>>>>>                     if abort
>>>>>>>                        free page
>>>>>>>                        abort
>>>>>>>                     ...
>>>>>>>     }
>>>>>>>
>>>>>>>     arch_fault_handler()
>>>>>>>     {
>>>>>>>             if (speculative_page_fault(&vma))
>>>>>>>                goto done
>>>>>>>     again:
>>>>>>>             lock(mmap_sem)
>>>>>>>             vma = find_vma();
>>>>>>>             handle_pte_fault(vma);
>>>>>>>             if retry
>>>>>>>                unlock(mmap_sem)
>>>>>>>                goto again;
>>>>>>>     done:
>>>>>>>             handle fault error
>>>>>>>     }
>>>>>>>
>>>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>>>
>>>>>>> This series add a new software performance event named 'speculative-faults'
>>>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>>>> the part of the faults processed speculatively.
>>>>>>>
>>>>>>> There are some trace events introduced by this series. They allow
>>>>>>> identifying why the page faults were not processed speculatively. This
>>>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>>>> grouped in a system named 'pagefault', they are:
>>>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>>>    back.
>>>>>>>
>>>>>>> To record all the related events, the easier is to run perf with the
>>>>>>> following arguments :
>>>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>>>
>>>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>>>> page fault handled speculatively. I can be seen this way:
>>>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>>>
>>>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>>>> on x86, PowerPC and arm64.
>>>>>>>
>>>>>>> ---------------------
>>>>>>> Real Workload results
>>>>>>>
>>>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>>>> this new version should not impact the performance boost seen.
>>>>>>>
>>>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>>>> series:
>>>>>>>                 vanilla         spf
>>>>>>> faults          89.418          101.364         +13%
>>>>>>> spf                n/a           97.989
>>>>>>>
>>>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>>>> way.
>>>>>>>
>>>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>>>> it a try on an android device. He reported that the application launch time
>>>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>>>> 20%.
>>>>>>>
>>>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>>>
>>>>>>> Application                             4.9     4.9+spf delta
>>>>>>> com.tencent.mm                          416     389     -7%
>>>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>>>> com.tencent.mtt                         455     454     0%
>>>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>>>> com.autonavi.minimap                    711     701     -1%
>>>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>>>> com.immomo.momo                         501     487     -3%
>>>>>>> com.tencent.peng                        2145    2112    -2%
>>>>>>> com.smile.gifmaker                      491     461     -6%
>>>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>>>> com.baidu.searchbox                     333     314     -6%
>>>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>>>> com.sina.weibo                          907     906     0%
>>>>>>> com.youku.phone                         816     731     -11%
>>>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>>>> com.UCMobile                            415     411     -1%
>>>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>>>
>>>>>>> ------------------
>>>>>>> Benchmarks results
>>>>>>>
>>>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>>>> SPF is BASE + this series
>>>>>>>
>>>>>>> Kernbench:
>>>>>>> ----------
>>>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>>>> kernel (kernel is build 5 times):
>>>>>>>
>>>>>>> Average Half load -j 8
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>>>
>>>>>>> Average Optimal load -j 16
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>>>
>>>>>>>
>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>          526743764      faults
>>>>>>>                210      spf
>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>               2278      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> Very few speculative page faults were recorded as most of the processes
>>>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>>>> were created during the kernel build processing).
>>>>>>>
>>>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>>>
>>>>>>> Average Half load -j 40
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>>>
>>>>>>> Average Optimal load -j 80
>>>>>>>                  Run    (std deviation)
>>>>>>>                  BASE                   SPF
>>>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>>>
>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>          116730856      faults
>>>>>>>                  0      spf
>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                476      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>>>> there is no impact on the performance.
>>>>>>>
>>>>>>> Ebizzy:
>>>>>>> -------
>>>>>>> The test is counting the number of records per second it can manage, the
>>>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>>>> consistent result I repeated the test 100 times and measure the average
>>>>>>> result. The number is the record processes per second, the higher is the
>>>>>>> best.
>>>>>>>
>>>>>>>                 BASE            SPF             delta
>>>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>>>
>>>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>>>            1706379      faults
>>>>>>>            1674599      spf
>>>>>>>              30588      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                363      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>>>            1874773      faults
>>>>>>>            1461153      spf
>>>>>>>             413293      pagefault:spf_vma_changed
>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>                200      pagefault:spf_vma_notsup
>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>
>>>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>>>> leading the ebizzy performance boost.
>>>>>>>
>>>>>>> ------------------
>>>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>>>    and Minchan Kim, hopefully.
>>>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>>>    __do_page_fault().
>>>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>>>    instead
>>>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>>>> useless
>>>>>>>    trace event pagefault:spf_pte_lock.
>>>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>>>
>>>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>>>
>>>>>>>
>>>>>>> Laurent Dufour (20):
>>>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>>>   mm: introduce INIT_VMA()
>>>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>>>   mm: protect mremap() against SPF hanlder
>>>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>>>   mm: introduce __vm_normal_page()
>>>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>>>   mm: adding speculative page fault failure trace events
>>>>>>>   perf: add a speculative page fault sw event
>>>>>>>   perf tools: add support for the SPF perf event
>>>>>>>   mm: add speculative page fault vmstats
>>>>>>>   powerpc/mm: add speculative page fault
>>>>>>>
>>>>>>> Mahendran Ganesh (2):
>>>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>   arm64/mm: add speculative page fault
>>>>>>>
>>>>>>> Peter Zijlstra (4):
>>>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>>>   mm: VMA sequence count
>>>>>>>   mm: provide speculative fault infrastructure
>>>>>>>   x86/mm: add speculative pagefault handling
>>>>>>>
>>>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>>>  fs/exec.c                             |   2 +-
>>>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>>>  include/linux/migrate.h               |   4 +-
>>>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>>>  include/linux/mm_types.h              |   7 +
>>>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>>>  include/linux/swap.h                  |  10 +-
>>>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>>>  kernel/fork.c                         |   5 +-
>>>>>>>  mm/Kconfig                            |  22 ++
>>>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>>>  mm/hugetlb.c                          |   2 +
>>>>>>>  mm/init-mm.c                          |   3 +
>>>>>>>  mm/internal.h                         |  20 ++
>>>>>>>  mm/khugepaged.c                       |   5 +
>>>>>>>  mm/madvise.c                          |   6 +-
>>>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>>>  mm/migrate.c                          |   6 +-
>>>>>>>  mm/mlock.c                            |  13 +-
>>>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>>>  mm/mprotect.c                         |   4 +-
>>>>>>>  mm/mremap.c                           |  13 +
>>>>>>>  mm/nommu.c                            |   2 +-
>>>>>>>  mm/rmap.c                             |   5 +-
>>>>>>>  mm/swap.c                             |   6 +-
>>>>>>>  mm/swap_state.c                       |   8 +-
>>>>>>>  mm/vmstat.c                           |   5 +-
>>>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>>>  tools/perf/util/python.c              |   1 +
>>>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>>>
>>>>>>> --
>>>>>>> 2.7.4
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>
>>>
>>
>


[-- Attachment #2: perf-profile_page_fault3_head_thp_always.gz --]
[-- Type: application/gzip, Size: 12909 bytes --]

[-- Attachment #3: perf-profile_page_fault3_head_thp_never.gz --]
[-- Type: application/gzip, Size: 12535 bytes --]

[-- Attachment #4: perf-profile_page_fault2_head_thp_never.gz --]
[-- Type: application/gzip, Size: 11782 bytes --]

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

[-- Attachment #1: Type: text/plain, Size: 43707 bytes --]

On 03/08/2018 08:36, Song, HaiyanX wrote:
> Hi Laurent,

Hi Haiyan,

Sorry for the late answer, I was off a couple of days.

> 
> Thanks for your analysis for the last perf results.
> Your mentioned ," the major differences at the head of the perf report is the 92% testcase which is weirdly not reported
> on the head side", which is a bug of 0-day,and it caused the item is not counted in perf. 
> 
> I've triggered the test page_fault2 and page_fault3 again only with thread mode of will-it-scale on 0-day (on the same test box,every case tested 3 times).
> I checked the perf report have no above mentioned problem.
> 
> I have compared them, found some items have difference, such as below case:
>        page_fault2-thp-always: handle_mm_fault, base: 45.22%    head: 29.41%
>        page_fault3-thp-always: handle_mm_fault, base: 22.95%    head: 14.15%       

These would mean that the system spends lees time running handle_mm_fault()
when SPF is in the picture in this 2 cases which is good. This should lead to
better results with the SPF series, and I can't find any values higher on the
head side.

> 
> So i attached the perf result in mail again, could your have a look again for checking the difference between base and head commit.

I took a close look to all the perf result you sent, but I can't identify any
major difference. But the compiler optimization is getting rid of the
handle_pte_fault() symbol on the base kernel which add complexity to check the
differences.

To get rid of that, I'm proposing that you applied the attached patch to the
spf kernel. This patch is allowing to turn on/off the SPF handler through
/proc/sys/vm/speculative_page_fault.

This should ease the testing by limiting the reboot and avoid kernel's symbols
mismatch. Obviously there is still a small overhead due to the check but it
should not be viewable.

With this patch applied you can simply run
echo 1 > /proc/sys/vm/speculative_page_fault
to run a test with the speculative page fault handler activated. Or run
echo 0 > /proc/sys/vm/speculative_page_fault
to run a test without it.

I'm really sorry to asking that again, but could please run the test
page_fault3_base_THP-Always with and without SPF and capture the perf output.

I think we should focus on that test which showed the biggest regression.

Thanks,
Laurent.


> 
> Thanks,
> Haiyan, Song
>  
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Tuesday, July 17, 2018 5:36 PM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
> 
> On 13/07/2018 05:56, Song, HaiyanX wrote:
>> Hi Laurent,
> 
> Hi Haiyan,
> 
> Thanks a lot for sharing this perf reports.
> 
> I looked at them closely, and I've to admit that I was not able to found a
> major difference between the base and the head report, except that
> handle_pte_fault() is no more in-lined in the head one.
> 
> As expected, __handle_speculative_fault() is never traced since these tests are
> dealing with file mapping, not handled in the speculative way.
> 
> When running these test did you seen a major differences in the test's result
> between base and head ?
> 
> From the number of cycles counted, the biggest difference is page_fault3 when
> run with the THP enabled:
>                                 BASE            HEAD            Delta
> page_fault2_base_thp_never      1142252426747   1065866197589   -6.69%
> page_fault2_base_THP-Alwasys    1124844374523   1076312228927   -4.31%
> page_fault3_base_thp_never      1099387298152   1134118402345   3.16%
> page_fault3_base_THP-Always     1059370178101   853985561949    -19.39%
> 
> 
> The very weird thing is the difference of the delta cycles reported between
> thp never and thp always, because the speculative way is aborted when checking
> for the vma->ops field, which is the same in both case, and the thp is never
> checked. So there is no code covering differnce, on the speculative path,
> between these 2 cases. This leads me to think that there are other interactions
> interfering in the measure.
> 
> Looking at the perf-profile_page_fault3_*_THP-Always, the major differences at
> the head of the perf report is the 92% testcase which is weirdly not reported
> on the head side :
>     92.02%    22.33%  page_fault3_processes  [.] testcase
> 92.02% testcase
> 
> Then the base reported 37.67% for __do_page_fault() where the head reported
> 48.41%, but the only difference in this function, between base and head, is the
> call to handle_speculative_fault(). But this is a macro checking for the fault
> flags, and mm->users and then calling __handle_speculative_fault() if needed.
> So this can't explain this difference, except if __handle_speculative_fault()
> is inlined in __do_page_fault().
> Is this the case on your build ?
> 
> Haiyan, do you still have the output of the test to check those numbers too ?
> 
> Cheers,
> Laurent
> 
>> I attached the perf-profile.gz file for case page_fault2 and page_fault3. These files were captured during test the related test case.
>> Please help to check on these data if it can help you to find the higher change. Thanks.
>>
>> File name perf-profile_page_fault2_head_THP-Always.gz, means the perf-profile result get from page_fault2
>>     tested for head commit (a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12) with THP_always configuration.
>>
>> Best regards,
>> Haiyan Song
>>
>> ________________________________________
>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>> Sent: Thursday, July 12, 2018 1:05 AM
>> To: Song, HaiyanX
>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>
>> Hi Haiyan,
>>
>> Do you get a chance to capture some performance cycles on your system ?
>> I still can't get these numbers on my hardware.
>>
>> Thanks,
>> Laurent.
>>
>> On 04/07/2018 09:51, Laurent Dufour wrote:
>>> On 04/07/2018 05:23, Song, HaiyanX wrote:
>>>> Hi Laurent,
>>>>
>>>>
>>>> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
>>>> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.
>>>
>>> Repeating the test only 3 times seems a bit too low to me.
>>>
>>> I'll focus on the higher change for the moment, but I don't have access to such
>>> a hardware.
>>>
>>> Is possible to provide a diff between base and SPF of the performance cycles
>>> measured when running page_fault3 and page_fault2 when the 20% change is detected.
>>>
>>> Please stay focus on the test case process to see exactly where the series is
>>> impacting.
>>>
>>> Thanks,
>>> Laurent.
>>>
>>>>
>>>> And I did not find other high variation on test case result.
>>>>
>>>> a). Enable THP
>>>> testcase                          base     stddev       change      head     stddev         metric
>>>> page_fault3/enable THP           10519      +- 3%        -20.5%      8368      +-6%          will-it-scale.per_thread_ops
>>>> page_fault2/enalbe THP            8281      +- 2%        -18.8%      6728                   will-it-scale.per_thread_ops
>>>> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
>>>> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
>>>> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
>>>>
>>>> b). Disable THP
>>>> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
>>>> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
>>>> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
>>>> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
>>>> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
>>>> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
>>>> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
>>>>
>>>>
>>>> Best regards,
>>>> Haiyan Song
>>>> ________________________________________
>>>> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>> Sent: Monday, July 02, 2018 4:59 PM
>>>> To: Song, HaiyanX
>>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>>
>>>> On 11/06/2018 09:49, Song, HaiyanX wrote:
>>>>> Hi Laurent,
>>>>>
>>>>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>>>>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>>>>> V9 patch serials.
>>>>>
>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>>>>> commit id:
>>>>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>>>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>>>>> Benchmark: will-it-scale
>>>>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>>>>
>>>>> Metrics:
>>>>>   will-it-scale.per_process_ops=processes/nr_cpu
>>>>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>> THP: enable / disable
>>>>> nr_task:100%
>>>>>
>>>>> 1. Regressions:
>>>>>
>>>>> a). Enable THP
>>>>> testcase                          base           change      head           metric
>>>>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>>>>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>>>>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>>>>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>>>>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>>>>
>>>>> b). Disable THP
>>>>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>>>>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>>>>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>>>>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>>>>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>>>>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>>>>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>>>>
>>>>> Notes: for the above  values of test result, the higher is better.
>>>>
>>>> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
>>>> get reproducible results. The results have huge variation, even on the vanilla
>>>> kernel, and I can't state on any changes due to that.
>>>>
>>>> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
>>>> measure any changes between the vanilla and the SPF patched ones:
>>>>
>>>> test THP enabled                4.17.0-rc4-mm1  spf             delta
>>>> page_fault3_threads             2697.7          2683.5          -0.53%
>>>> page_fault2_threads             170660.6        169574.1        -0.64%
>>>> context_switch1_threads         6915269.2       6877507.3       -0.55%
>>>> context_switch1_processes       6478076.2       6529493.5       0.79%
>>>> brk1                            243391.2        238527.5        -2.00%
>>>>
>>>> Tests were run 10 times, no high variation detected.
>>>>
>>>> Did you see high variation on your side ? How many times the test were run to
>>>> compute the average values ?
>>>>
>>>> Thanks,
>>>> Laurent.
>>>>
>>>>
>>>>>
>>>>> 2. Improvement: not found improvement based on the selected test cases.
>>>>>
>>>>>
>>>>> Best regards
>>>>> Haiyan Song
>>>>> ________________________________________
>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>> Sent: Monday, May 28, 2018 4:54 PM
>>>>> To: Song, HaiyanX
>>>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>>>
>>>>> On 28/05/2018 10:22, Haiyan Song wrote:
>>>>>> Hi Laurent,
>>>>>>
>>>>>> Yes, these tests are done on V9 patch.
>>>>>
>>>>> Do you plan to give this V11 a run ?
>>>>>
>>>>>>
>>>>>>
>>>>>> Best regards,
>>>>>> Haiyan Song
>>>>>>
>>>>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>>>>
>>>>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>>>>> tested on Intel 4s Skylake platform.
>>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>>>>> series" while responding to the v11 header series...
>>>>>>> Were these tests done on v9 or v11 ?
>>>>>>>
>>>>>>> Cheers,
>>>>>>> Laurent.
>>>>>>>
>>>>>>>>
>>>>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>>>>> Commit id:
>>>>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>>>>> Benchmark suite: will-it-scale
>>>>>>>> Download link:
>>>>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>>>>> Metrics:
>>>>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>>>>> THP: enable / disable
>>>>>>>> nr_task: 100%
>>>>>>>>
>>>>>>>> 1. Regressions:
>>>>>>>> a) THP enabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>>>>
>>>>>>>> b) THP disabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>>>>
>>>>>>>> 2. Improvements:
>>>>>>>> a) THP enabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>>>>
>>>>>>>> b) THP disabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>>>>
>>>>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>>>>
>>>>>>>>
>>>>>>>> Best regards
>>>>>>>> Haiyan Song
>>>>>>>>
>>>>>>>> ________________________________________
>>>>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>>>>
>>>>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>>>>> page fault without holding the mm semaphore [1].
>>>>>>>>
>>>>>>>> The idea is to try to handle user space page faults without holding the
>>>>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>>>>> process since the page fault handler will not wait for other threads memory
>>>>>>>> layout change to be done, assuming that this change is done in another part
>>>>>>>> of the process's memory space. This type page fault is named speculative
>>>>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>>>>> is failing its processing and a classic page fault is then tried.
>>>>>>>>
>>>>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>>>>> limiting the locking contention to these operations which are expected to
>>>>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>>>>> benchmark anymore.
>>>>>>>>
>>>>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>>>>> have to be protected against parallel changes. This is done by using a per
>>>>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>>>>> speculative page fault in that case.
>>>>>>>>
>>>>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>>>>> checked during the page fault are modified.
>>>>>>>>
>>>>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>>>>> parallel change is possible at this time.
>>>>>>>>
>>>>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>>>>> classic page fault handler will be called to handle the operation while
>>>>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>>>>> PTE.
>>>>>>>>
>>>>>>>> In pseudo code, this could be seen as:
>>>>>>>>     speculative_page_fault()
>>>>>>>>     {
>>>>>>>>             vma = get_vma()
>>>>>>>>             check vma sequence count
>>>>>>>>             check vma's support
>>>>>>>>             disable interrupt
>>>>>>>>                   check pgd,p4d,...,pte
>>>>>>>>                   save pmd and pte in vmf
>>>>>>>>                   save vma sequence counter in vmf
>>>>>>>>             enable interrupt
>>>>>>>>             check vma sequence count
>>>>>>>>             handle_pte_fault(vma)
>>>>>>>>                     ..
>>>>>>>>                     page = alloc_page()
>>>>>>>>                     pte_map_lock()
>>>>>>>>                             disable interrupt
>>>>>>>>                                     abort if sequence counter has changed
>>>>>>>>                                     abort if pmd or pte has changed
>>>>>>>>                                     pte map and lock
>>>>>>>>                             enable interrupt
>>>>>>>>                     if abort
>>>>>>>>                        free page
>>>>>>>>                        abort
>>>>>>>>                     ...
>>>>>>>>     }
>>>>>>>>
>>>>>>>>     arch_fault_handler()
>>>>>>>>     {
>>>>>>>>             if (speculative_page_fault(&vma))
>>>>>>>>                goto done
>>>>>>>>     again:
>>>>>>>>             lock(mmap_sem)
>>>>>>>>             vma = find_vma();
>>>>>>>>             handle_pte_fault(vma);
>>>>>>>>             if retry
>>>>>>>>                unlock(mmap_sem)
>>>>>>>>                goto again;
>>>>>>>>     done:
>>>>>>>>             handle fault error
>>>>>>>>     }
>>>>>>>>
>>>>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>>>>
>>>>>>>> This series add a new software performance event named 'speculative-faults'
>>>>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>>>>> the part of the faults processed speculatively.
>>>>>>>>
>>>>>>>> There are some trace events introduced by this series. They allow
>>>>>>>> identifying why the page faults were not processed speculatively. This
>>>>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>>>>> grouped in a system named 'pagefault', they are:
>>>>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>>>>    back.
>>>>>>>>
>>>>>>>> To record all the related events, the easier is to run perf with the
>>>>>>>> following arguments :
>>>>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>>>>
>>>>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>>>>> page fault handled speculatively. I can be seen this way:
>>>>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>>>>
>>>>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>>>>> on x86, PowerPC and arm64.
>>>>>>>>
>>>>>>>> ---------------------
>>>>>>>> Real Workload results
>>>>>>>>
>>>>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>>>>> this new version should not impact the performance boost seen.
>>>>>>>>
>>>>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>>>>> series:
>>>>>>>>                 vanilla         spf
>>>>>>>> faults          89.418          101.364         +13%
>>>>>>>> spf                n/a           97.989
>>>>>>>>
>>>>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>>>>> way.
>>>>>>>>
>>>>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>>>>> it a try on an android device. He reported that the application launch time
>>>>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>>>>> 20%.
>>>>>>>>
>>>>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>>>>
>>>>>>>> Application                             4.9     4.9+spf delta
>>>>>>>> com.tencent.mm                          416     389     -7%
>>>>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>>>>> com.tencent.mtt                         455     454     0%
>>>>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>>>>> com.autonavi.minimap                    711     701     -1%
>>>>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>>>>> com.immomo.momo                         501     487     -3%
>>>>>>>> com.tencent.peng                        2145    2112    -2%
>>>>>>>> com.smile.gifmaker                      491     461     -6%
>>>>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>>>>> com.baidu.searchbox                     333     314     -6%
>>>>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>>>>> com.sina.weibo                          907     906     0%
>>>>>>>> com.youku.phone                         816     731     -11%
>>>>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>>>>> com.UCMobile                            415     411     -1%
>>>>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>>>>
>>>>>>>> ------------------
>>>>>>>> Benchmarks results
>>>>>>>>
>>>>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>>>>> SPF is BASE + this series
>>>>>>>>
>>>>>>>> Kernbench:
>>>>>>>> ----------
>>>>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>>>>> kernel (kernel is build 5 times):
>>>>>>>>
>>>>>>>> Average Half load -j 8
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>>>>
>>>>>>>> Average Optimal load -j 16
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>>>>
>>>>>>>>
>>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>>          526743764      faults
>>>>>>>>                210      spf
>>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>               2278      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> Very few speculative page faults were recorded as most of the processes
>>>>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>>>>> were created during the kernel build processing).
>>>>>>>>
>>>>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>>>>
>>>>>>>> Average Half load -j 40
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>>>>
>>>>>>>> Average Optimal load -j 80
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>>>>
>>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>>          116730856      faults
>>>>>>>>                  0      spf
>>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>                476      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>>>>> there is no impact on the performance.
>>>>>>>>
>>>>>>>> Ebizzy:
>>>>>>>> -------
>>>>>>>> The test is counting the number of records per second it can manage, the
>>>>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>>>>> consistent result I repeated the test 100 times and measure the average
>>>>>>>> result. The number is the record processes per second, the higher is the
>>>>>>>> best.
>>>>>>>>
>>>>>>>>                 BASE            SPF             delta
>>>>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>>>>
>>>>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>>>>            1706379      faults
>>>>>>>>            1674599      spf
>>>>>>>>              30588      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>                363      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>>>>            1874773      faults
>>>>>>>>            1461153      spf
>>>>>>>>             413293      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>                200      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>>>>> leading the ebizzy performance boost.
>>>>>>>>
>>>>>>>> ------------------
>>>>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>>>>    and Minchan Kim, hopefully.
>>>>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>>>>    __do_page_fault().
>>>>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>>>>    instead
>>>>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>>>>> useless
>>>>>>>>    trace event pagefault:spf_pte_lock.
>>>>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>>>>
>>>>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>>>>
>>>>>>>>
>>>>>>>> Laurent Dufour (20):
>>>>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>>>>   mm: introduce INIT_VMA()
>>>>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>>>>   mm: protect mremap() against SPF hanlder
>>>>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>>>>   mm: introduce __vm_normal_page()
>>>>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>>>>   mm: adding speculative page fault failure trace events
>>>>>>>>   perf: add a speculative page fault sw event
>>>>>>>>   perf tools: add support for the SPF perf event
>>>>>>>>   mm: add speculative page fault vmstats
>>>>>>>>   powerpc/mm: add speculative page fault
>>>>>>>>
>>>>>>>> Mahendran Ganesh (2):
>>>>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>>   arm64/mm: add speculative page fault
>>>>>>>>
>>>>>>>> Peter Zijlstra (4):
>>>>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>>>>   mm: VMA sequence count
>>>>>>>>   mm: provide speculative fault infrastructure
>>>>>>>>   x86/mm: add speculative pagefault handling
>>>>>>>>
>>>>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>>>>  fs/exec.c                             |   2 +-
>>>>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>>>>  include/linux/migrate.h               |   4 +-
>>>>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>>>>  include/linux/mm_types.h              |   7 +
>>>>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>>>>  include/linux/swap.h                  |  10 +-
>>>>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>>>>  kernel/fork.c                         |   5 +-
>>>>>>>>  mm/Kconfig                            |  22 ++
>>>>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>>>>  mm/hugetlb.c                          |   2 +
>>>>>>>>  mm/init-mm.c                          |   3 +
>>>>>>>>  mm/internal.h                         |  20 ++
>>>>>>>>  mm/khugepaged.c                       |   5 +
>>>>>>>>  mm/madvise.c                          |   6 +-
>>>>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>>>>  mm/migrate.c                          |   6 +-
>>>>>>>>  mm/mlock.c                            |  13 +-
>>>>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>>>>  mm/mprotect.c                         |   4 +-
>>>>>>>>  mm/mremap.c                           |  13 +
>>>>>>>>  mm/nommu.c                            |   2 +-
>>>>>>>>  mm/rmap.c                             |   5 +-
>>>>>>>>  mm/swap.c                             |   6 +-
>>>>>>>>  mm/swap_state.c                       |   8 +-
>>>>>>>>  mm/vmstat.c                           |   5 +-
>>>>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>>>>  tools/perf/util/python.c              |   1 +
>>>>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>>>>
>>>>>>>> --
>>>>>>>> 2.7.4
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>>
>>>
>>
> 


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-mm-Add-a-speculative-page-fault-switch-in-sysctl.patch --]
[-- Type: text/x-patch; name="0001-mm-Add-a-speculative-page-fault-switch-in-sysctl.patch", Size: 0 bytes --]

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

[-- Attachment #1: Type: text/plain, Size: 43704 bytes --]

On 03/08/2018 08:36, Song, HaiyanX wrote:
> Hi Laurent,

Hi Haiyan,

Sorry for the late answer, I was off a couple of days.

> 
> Thanks for your analysis for the last perf results.
> Your mentioned ," the major differences at the head of the perf report is the 92% testcase which is weirdly not reported
> on the head side", which is a bug of 0-day,and it caused the item is not counted in perf. 
> 
> I've triggered the test page_fault2 and page_fault3 again only with thread mode of will-it-scale on 0-day (on the same test box,every case tested 3 times).
> I checked the perf report have no above mentioned problem.
> 
> I have compared them, found some items have difference, such as below case:
>        page_fault2-thp-always: handle_mm_fault, base: 45.22%    head: 29.41%
>        page_fault3-thp-always: handle_mm_fault, base: 22.95%    head: 14.15%       

These would mean that the system spends lees time running handle_mm_fault()
when SPF is in the picture in this 2 cases which is good. This should lead to
better results with the SPF series, and I can't find any values higher on the
head side.

> 
> So i attached the perf result in mail again, could your have a look again for checking the difference between base and head commit.

I took a close look to all the perf result you sent, but I can't identify any
major difference. But the compiler optimization is getting rid of the
handle_pte_fault() symbol on the base kernel which add complexity to check the
differences.

To get rid of that, I'm proposing that you applied the attached patch to the
spf kernel. This patch is allowing to turn on/off the SPF handler through
/proc/sys/vm/speculative_page_fault.

This should ease the testing by limiting the reboot and avoid kernel's symbols
mismatch. Obviously there is still a small overhead due to the check but it
should not be viewable.

With this patch applied you can simply run
echo 1 > /proc/sys/vm/speculative_page_fault
to run a test with the speculative page fault handler activated. Or run
echo 0 > /proc/sys/vm/speculative_page_fault
to run a test without it.

I'm really sorry to asking that again, but could please run the test
page_fault3_base_THP-Always with and without SPF and capture the perf output.

I think we should focus on that test which showed the biggest regression.

Thanks,
Laurent.


> 
> Thanks,
> Haiyan, Song
>  
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Tuesday, July 17, 2018 5:36 PM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
> 
> On 13/07/2018 05:56, Song, HaiyanX wrote:
>> Hi Laurent,
> 
> Hi Haiyan,
> 
> Thanks a lot for sharing this perf reports.
> 
> I looked at them closely, and I've to admit that I was not able to found a
> major difference between the base and the head report, except that
> handle_pte_fault() is no more in-lined in the head one.
> 
> As expected, __handle_speculative_fault() is never traced since these tests are
> dealing with file mapping, not handled in the speculative way.
> 
> When running these test did you seen a major differences in the test's result
> between base and head ?
> 
> From the number of cycles counted, the biggest difference is page_fault3 when
> run with the THP enabled:
>                                 BASE            HEAD            Delta
> page_fault2_base_thp_never      1142252426747   1065866197589   -6.69%
> page_fault2_base_THP-Alwasys    1124844374523   1076312228927   -4.31%
> page_fault3_base_thp_never      1099387298152   1134118402345   3.16%
> page_fault3_base_THP-Always     1059370178101   853985561949    -19.39%
> 
> 
> The very weird thing is the difference of the delta cycles reported between
> thp never and thp always, because the speculative way is aborted when checking
> for the vma->ops field, which is the same in both case, and the thp is never
> checked. So there is no code covering differnce, on the speculative path,
> between these 2 cases. This leads me to think that there are other interactions
> interfering in the measure.
> 
> Looking at the perf-profile_page_fault3_*_THP-Always, the major differences at
> the head of the perf report is the 92% testcase which is weirdly not reported
> on the head side :
>     92.02%    22.33%  page_fault3_processes  [.] testcase
> 92.02% testcase
> 
> Then the base reported 37.67% for __do_page_fault() where the head reported
> 48.41%, but the only difference in this function, between base and head, is the
> call to handle_speculative_fault(). But this is a macro checking for the fault
> flags, and mm->users and then calling __handle_speculative_fault() if needed.
> So this can't explain this difference, except if __handle_speculative_fault()
> is inlined in __do_page_fault().
> Is this the case on your build ?
> 
> Haiyan, do you still have the output of the test to check those numbers too ?
> 
> Cheers,
> Laurent
> 
>> I attached the perf-profile.gz file for case page_fault2 and page_fault3. These files were captured during test the related test case.
>> Please help to check on these data if it can help you to find the higher change. Thanks.
>>
>> File name perf-profile_page_fault2_head_THP-Always.gz, means the perf-profile result get from page_fault2
>>     tested for head commit (a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12) with THP_always configuration.
>>
>> Best regards,
>> Haiyan Song
>>
>> ________________________________________
>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>> Sent: Thursday, July 12, 2018 1:05 AM
>> To: Song, HaiyanX
>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>
>> Hi Haiyan,
>>
>> Do you get a chance to capture some performance cycles on your system ?
>> I still can't get these numbers on my hardware.
>>
>> Thanks,
>> Laurent.
>>
>> On 04/07/2018 09:51, Laurent Dufour wrote:
>>> On 04/07/2018 05:23, Song, HaiyanX wrote:
>>>> Hi Laurent,
>>>>
>>>>
>>>> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
>>>> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.
>>>
>>> Repeating the test only 3 times seems a bit too low to me.
>>>
>>> I'll focus on the higher change for the moment, but I don't have access to such
>>> a hardware.
>>>
>>> Is possible to provide a diff between base and SPF of the performance cycles
>>> measured when running page_fault3 and page_fault2 when the 20% change is detected.
>>>
>>> Please stay focus on the test case process to see exactly where the series is
>>> impacting.
>>>
>>> Thanks,
>>> Laurent.
>>>
>>>>
>>>> And I did not find other high variation on test case result.
>>>>
>>>> a). Enable THP
>>>> testcase                          base     stddev       change      head     stddev         metric
>>>> page_fault3/enable THP           10519      ± 3%        -20.5%      8368      ±6%          will-it-scale.per_thread_ops
>>>> page_fault2/enalbe THP            8281      ± 2%        -18.8%      6728                   will-it-scale.per_thread_ops
>>>> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
>>>> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
>>>> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
>>>>
>>>> b). Disable THP
>>>> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
>>>> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
>>>> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
>>>> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
>>>> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
>>>> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
>>>> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
>>>>
>>>>
>>>> Best regards,
>>>> Haiyan Song
>>>> ________________________________________
>>>> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>> Sent: Monday, July 02, 2018 4:59 PM
>>>> To: Song, HaiyanX
>>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>>
>>>> On 11/06/2018 09:49, Song, HaiyanX wrote:
>>>>> Hi Laurent,
>>>>>
>>>>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>>>>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>>>>> V9 patch serials.
>>>>>
>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>>>>> commit id:
>>>>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>>>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>>>>> Benchmark: will-it-scale
>>>>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>>>>
>>>>> Metrics:
>>>>>   will-it-scale.per_process_ops=processes/nr_cpu
>>>>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>> THP: enable / disable
>>>>> nr_task:100%
>>>>>
>>>>> 1. Regressions:
>>>>>
>>>>> a). Enable THP
>>>>> testcase                          base           change      head           metric
>>>>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>>>>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>>>>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>>>>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>>>>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>>>>
>>>>> b). Disable THP
>>>>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>>>>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>>>>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>>>>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>>>>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>>>>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>>>>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>>>>
>>>>> Notes: for the above  values of test result, the higher is better.
>>>>
>>>> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
>>>> get reproducible results. The results have huge variation, even on the vanilla
>>>> kernel, and I can't state on any changes due to that.
>>>>
>>>> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
>>>> measure any changes between the vanilla and the SPF patched ones:
>>>>
>>>> test THP enabled                4.17.0-rc4-mm1  spf             delta
>>>> page_fault3_threads             2697.7          2683.5          -0.53%
>>>> page_fault2_threads             170660.6        169574.1        -0.64%
>>>> context_switch1_threads         6915269.2       6877507.3       -0.55%
>>>> context_switch1_processes       6478076.2       6529493.5       0.79%
>>>> brk1                            243391.2        238527.5        -2.00%
>>>>
>>>> Tests were run 10 times, no high variation detected.
>>>>
>>>> Did you see high variation on your side ? How many times the test were run to
>>>> compute the average values ?
>>>>
>>>> Thanks,
>>>> Laurent.
>>>>
>>>>
>>>>>
>>>>> 2. Improvement: not found improvement based on the selected test cases.
>>>>>
>>>>>
>>>>> Best regards
>>>>> Haiyan Song
>>>>> ________________________________________
>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>> Sent: Monday, May 28, 2018 4:54 PM
>>>>> To: Song, HaiyanX
>>>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>>>
>>>>> On 28/05/2018 10:22, Haiyan Song wrote:
>>>>>> Hi Laurent,
>>>>>>
>>>>>> Yes, these tests are done on V9 patch.
>>>>>
>>>>> Do you plan to give this V11 a run ?
>>>>>
>>>>>>
>>>>>>
>>>>>> Best regards,
>>>>>> Haiyan Song
>>>>>>
>>>>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>>>>
>>>>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>>>>> tested on Intel 4s Skylake platform.
>>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>>>>> series" while responding to the v11 header series...
>>>>>>> Were these tests done on v9 or v11 ?
>>>>>>>
>>>>>>> Cheers,
>>>>>>> Laurent.
>>>>>>>
>>>>>>>>
>>>>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>>>>> Commit id:
>>>>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>>>>> Benchmark suite: will-it-scale
>>>>>>>> Download link:
>>>>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>>>>> Metrics:
>>>>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>>>>> THP: enable / disable
>>>>>>>> nr_task: 100%
>>>>>>>>
>>>>>>>> 1. Regressions:
>>>>>>>> a) THP enabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>>>>
>>>>>>>> b) THP disabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>>>>
>>>>>>>> 2. Improvements:
>>>>>>>> a) THP enabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>>>>
>>>>>>>> b) THP disabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>>>>
>>>>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>>>>
>>>>>>>>
>>>>>>>> Best regards
>>>>>>>> Haiyan Song
>>>>>>>>
>>>>>>>> ________________________________________
>>>>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>>>>
>>>>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>>>>> page fault without holding the mm semaphore [1].
>>>>>>>>
>>>>>>>> The idea is to try to handle user space page faults without holding the
>>>>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>>>>> process since the page fault handler will not wait for other threads memory
>>>>>>>> layout change to be done, assuming that this change is done in another part
>>>>>>>> of the process's memory space. This type page fault is named speculative
>>>>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>>>>> is failing its processing and a classic page fault is then tried.
>>>>>>>>
>>>>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>>>>> limiting the locking contention to these operations which are expected to
>>>>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>>>>> benchmark anymore.
>>>>>>>>
>>>>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>>>>> have to be protected against parallel changes. This is done by using a per
>>>>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>>>>> speculative page fault in that case.
>>>>>>>>
>>>>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>>>>> checked during the page fault are modified.
>>>>>>>>
>>>>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>>>>> parallel change is possible at this time.
>>>>>>>>
>>>>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>>>>> classic page fault handler will be called to handle the operation while
>>>>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>>>>> PTE.
>>>>>>>>
>>>>>>>> In pseudo code, this could be seen as:
>>>>>>>>     speculative_page_fault()
>>>>>>>>     {
>>>>>>>>             vma = get_vma()
>>>>>>>>             check vma sequence count
>>>>>>>>             check vma's support
>>>>>>>>             disable interrupt
>>>>>>>>                   check pgd,p4d,...,pte
>>>>>>>>                   save pmd and pte in vmf
>>>>>>>>                   save vma sequence counter in vmf
>>>>>>>>             enable interrupt
>>>>>>>>             check vma sequence count
>>>>>>>>             handle_pte_fault(vma)
>>>>>>>>                     ..
>>>>>>>>                     page = alloc_page()
>>>>>>>>                     pte_map_lock()
>>>>>>>>                             disable interrupt
>>>>>>>>                                     abort if sequence counter has changed
>>>>>>>>                                     abort if pmd or pte has changed
>>>>>>>>                                     pte map and lock
>>>>>>>>                             enable interrupt
>>>>>>>>                     if abort
>>>>>>>>                        free page
>>>>>>>>                        abort
>>>>>>>>                     ...
>>>>>>>>     }
>>>>>>>>
>>>>>>>>     arch_fault_handler()
>>>>>>>>     {
>>>>>>>>             if (speculative_page_fault(&vma))
>>>>>>>>                goto done
>>>>>>>>     again:
>>>>>>>>             lock(mmap_sem)
>>>>>>>>             vma = find_vma();
>>>>>>>>             handle_pte_fault(vma);
>>>>>>>>             if retry
>>>>>>>>                unlock(mmap_sem)
>>>>>>>>                goto again;
>>>>>>>>     done:
>>>>>>>>             handle fault error
>>>>>>>>     }
>>>>>>>>
>>>>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>>>>
>>>>>>>> This series add a new software performance event named 'speculative-faults'
>>>>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>>>>> the part of the faults processed speculatively.
>>>>>>>>
>>>>>>>> There are some trace events introduced by this series. They allow
>>>>>>>> identifying why the page faults were not processed speculatively. This
>>>>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>>>>> grouped in a system named 'pagefault', they are:
>>>>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>>>>    back.
>>>>>>>>
>>>>>>>> To record all the related events, the easier is to run perf with the
>>>>>>>> following arguments :
>>>>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>>>>
>>>>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>>>>> page fault handled speculatively. I can be seen this way:
>>>>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>>>>
>>>>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>>>>> on x86, PowerPC and arm64.
>>>>>>>>
>>>>>>>> ---------------------
>>>>>>>> Real Workload results
>>>>>>>>
>>>>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>>>>> this new version should not impact the performance boost seen.
>>>>>>>>
>>>>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>>>>> series:
>>>>>>>>                 vanilla         spf
>>>>>>>> faults          89.418          101.364         +13%
>>>>>>>> spf                n/a           97.989
>>>>>>>>
>>>>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>>>>> way.
>>>>>>>>
>>>>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>>>>> it a try on an android device. He reported that the application launch time
>>>>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>>>>> 20%.
>>>>>>>>
>>>>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>>>>
>>>>>>>> Application                             4.9     4.9+spf delta
>>>>>>>> com.tencent.mm                          416     389     -7%
>>>>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>>>>> com.tencent.mtt                         455     454     0%
>>>>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>>>>> com.autonavi.minimap                    711     701     -1%
>>>>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>>>>> com.immomo.momo                         501     487     -3%
>>>>>>>> com.tencent.peng                        2145    2112    -2%
>>>>>>>> com.smile.gifmaker                      491     461     -6%
>>>>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>>>>> com.baidu.searchbox                     333     314     -6%
>>>>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>>>>> com.sina.weibo                          907     906     0%
>>>>>>>> com.youku.phone                         816     731     -11%
>>>>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>>>>> com.UCMobile                            415     411     -1%
>>>>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>>>>
>>>>>>>> ------------------
>>>>>>>> Benchmarks results
>>>>>>>>
>>>>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>>>>> SPF is BASE + this series
>>>>>>>>
>>>>>>>> Kernbench:
>>>>>>>> ----------
>>>>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>>>>> kernel (kernel is build 5 times):
>>>>>>>>
>>>>>>>> Average Half load -j 8
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>>>>
>>>>>>>> Average Optimal load -j 16
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>>>>
>>>>>>>>
>>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>>          526743764      faults
>>>>>>>>                210      spf
>>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>               2278      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> Very few speculative page faults were recorded as most of the processes
>>>>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>>>>> were created during the kernel build processing).
>>>>>>>>
>>>>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>>>>
>>>>>>>> Average Half load -j 40
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>>>>
>>>>>>>> Average Optimal load -j 80
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>>>>
>>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>>          116730856      faults
>>>>>>>>                  0      spf
>>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>                476      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>>>>> there is no impact on the performance.
>>>>>>>>
>>>>>>>> Ebizzy:
>>>>>>>> -------
>>>>>>>> The test is counting the number of records per second it can manage, the
>>>>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>>>>> consistent result I repeated the test 100 times and measure the average
>>>>>>>> result. The number is the record processes per second, the higher is the
>>>>>>>> best.
>>>>>>>>
>>>>>>>>                 BASE            SPF             delta
>>>>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>>>>
>>>>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>>>>            1706379      faults
>>>>>>>>            1674599      spf
>>>>>>>>              30588      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>                363      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>>>>            1874773      faults
>>>>>>>>            1461153      spf
>>>>>>>>             413293      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>                200      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>>>>> leading the ebizzy performance boost.
>>>>>>>>
>>>>>>>> ------------------
>>>>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>>>>    and Minchan Kim, hopefully.
>>>>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>>>>    __do_page_fault().
>>>>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>>>>    instead
>>>>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>>>>> useless
>>>>>>>>    trace event pagefault:spf_pte_lock.
>>>>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>>>>
>>>>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>>>>
>>>>>>>>
>>>>>>>> Laurent Dufour (20):
>>>>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>>>>   mm: introduce INIT_VMA()
>>>>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>>>>   mm: protect mremap() against SPF hanlder
>>>>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>>>>   mm: introduce __vm_normal_page()
>>>>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>>>>   mm: adding speculative page fault failure trace events
>>>>>>>>   perf: add a speculative page fault sw event
>>>>>>>>   perf tools: add support for the SPF perf event
>>>>>>>>   mm: add speculative page fault vmstats
>>>>>>>>   powerpc/mm: add speculative page fault
>>>>>>>>
>>>>>>>> Mahendran Ganesh (2):
>>>>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>>   arm64/mm: add speculative page fault
>>>>>>>>
>>>>>>>> Peter Zijlstra (4):
>>>>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>>>>   mm: VMA sequence count
>>>>>>>>   mm: provide speculative fault infrastructure
>>>>>>>>   x86/mm: add speculative pagefault handling
>>>>>>>>
>>>>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>>>>  fs/exec.c                             |   2 +-
>>>>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>>>>  include/linux/migrate.h               |   4 +-
>>>>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>>>>  include/linux/mm_types.h              |   7 +
>>>>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>>>>  include/linux/swap.h                  |  10 +-
>>>>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>>>>  kernel/fork.c                         |   5 +-
>>>>>>>>  mm/Kconfig                            |  22 ++
>>>>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>>>>  mm/hugetlb.c                          |   2 +
>>>>>>>>  mm/init-mm.c                          |   3 +
>>>>>>>>  mm/internal.h                         |  20 ++
>>>>>>>>  mm/khugepaged.c                       |   5 +
>>>>>>>>  mm/madvise.c                          |   6 +-
>>>>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>>>>  mm/migrate.c                          |   6 +-
>>>>>>>>  mm/mlock.c                            |  13 +-
>>>>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>>>>  mm/mprotect.c                         |   4 +-
>>>>>>>>  mm/mremap.c                           |  13 +
>>>>>>>>  mm/nommu.c                            |   2 +-
>>>>>>>>  mm/rmap.c                             |   5 +-
>>>>>>>>  mm/swap.c                             |   6 +-
>>>>>>>>  mm/swap_state.c                       |   8 +-
>>>>>>>>  mm/vmstat.c                           |   5 +-
>>>>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>>>>  tools/perf/util/python.c              |   1 +
>>>>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>>>>
>>>>>>>> --
>>>>>>>> 2.7.4
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>>
>>>
>>
> 


[-- Attachment #2: 0001-mm-Add-a-speculative-page-fault-switch-in-sysctl.patch --]
[-- Type: text/x-patch, Size: 2326 bytes --]

>From b6c7fa413f25b8574edf8c764b136715c40299c2 Mon Sep 17 00:00:00 2001
From: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Date: Mon, 20 Aug 2018 17:51:26 +0200
Subject: [PATCH] mm: Add a speculative page fault switch in sysctl

This allows to turn on/off the use of the speculative page fault handler.

By default it's turned on.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 include/linux/mm.h | 3 +++
 kernel/sysctl.c    | 9 +++++++++
 mm/memory.c        | 3 +++
 3 files changed, 15 insertions(+)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 31acf98a7d92..ac102efc4c86 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1422,6 +1422,7 @@ extern int handle_mm_fault(struct vm_area_struct *vma, unsigned long address,
 		unsigned int flags);
 
 #ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+extern int sysctl_speculative_page_fault;
 extern int __handle_speculative_fault(struct mm_struct *mm,
 				      unsigned long address,
 				      unsigned int flags);
@@ -1429,6 +1430,8 @@ static inline int handle_speculative_fault(struct mm_struct *mm,
 					   unsigned long address,
 					   unsigned int flags)
 {
+	if (unlikely(!sysctl_speculative_page_fault))
+		return VM_FAULT_RETRY;
 	/*
 	 * Try speculative page fault for multithreaded user space task only.
 	 */
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index f45ed9e696eb..0fb81edd22c1 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -1243,6 +1243,15 @@ static struct ctl_table vm_table[] = {
 		.extra1		= &zero,
 		.extra2		= &two,
 	},
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+	{
+		.procname	= "speculative_page_fault",
+		.data		= &sysctl_speculative_page_fault,
+		.maxlen		= sizeof(sysctl_speculative_page_fault),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+#endif
 	{
 		.procname	= "panic_on_oom",
 		.data		= &sysctl_panic_on_oom,
diff --git a/mm/memory.c b/mm/memory.c
index 48e1cf0a54ef..c3db3bc4347b 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -82,6 +82,9 @@
 
 #define CREATE_TRACE_POINTS
 #include <trace/events/pagefault.h>
+#ifdef CONFIG_SPECULATIVE_PAGE_FAULT
+int sysctl_speculative_page_fault = 1;
+#endif
 
 #if defined(LAST_CPUPID_NOT_IN_PAGE_FLAGS) && !defined(CONFIG_COMPILE_TEST)
 #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
-- 
2.7.4

RE: [PATCH v11 00/26] Speculative page faults

[Song, HaiyanX]

[-- Attachment #1: Type: text/plain, Size: 46630 bytes --]

Hi Laurent,

I am sorry for replying you so late. 
The previous LKP test for this case are running on the same Intel skylake 4s platform, but it need maintain recently. 
So I changed to another test box to run the page_fault3 test case, it is Intel skylake 2s platform (nr_cpu: 104, memory: 64G).

I applied your patch to the SPF kernel (commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12), then triggered below 2 cases test.
a)  Turn on the SPF handler by below command, then run page_fault3-thp-always test.
echo 1 > /proc/sys/vm/speculative_page_fault

b) Turn off the SPF handler by below command, then run page_fault3-thp-always test.
 echo 0 > /proc/sys/vm/speculative_page_fault

Every test run 3 times, and then get test result and capture perf data. 
Here is average result for will-it-scale.per_thread_ops:                                                         
                                                                                          SPF_turn_off       SPF_turn_on
page_fault3-THP-Alwasys.will-it-scale.per_thread_ops    31963                  26285

Best regards,
Haiyan Song

________________________________________
From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
Sent: Wednesday, August 22, 2018 10:23 PM
To: Song, HaiyanX
Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
Subject: Re: [PATCH v11 00/26] Speculative page faults

On 03/08/2018 08:36, Song, HaiyanX wrote:
> Hi Laurent,

Hi Haiyan,

Sorry for the late answer, I was off a couple of days.

>
> Thanks for your analysis for the last perf results.
> Your mentioned ," the major differences at the head of the perf report is the 92% testcase which is weirdly not reported
> on the head side", which is a bug of 0-day,and it caused the item is not counted in perf.
>
> I've triggered the test page_fault2 and page_fault3 again only with thread mode of will-it-scale on 0-day (on the same test box,every case tested 3 times).
> I checked the perf report have no above mentioned problem.
>
> I have compared them, found some items have difference, such as below case:
>        page_fault2-thp-always: handle_mm_fault, base: 45.22%    head: 29.41%
>        page_fault3-thp-always: handle_mm_fault, base: 22.95%    head: 14.15%

These would mean that the system spends lees time running handle_mm_fault()
when SPF is in the picture in this 2 cases which is good. This should lead to
better results with the SPF series, and I can't find any values higher on the
head side.

>
> So i attached the perf result in mail again, could your have a look again for checking the difference between base and head commit.

I took a close look to all the perf result you sent, but I can't identify any
major difference. But the compiler optimization is getting rid of the
handle_pte_fault() symbol on the base kernel which add complexity to check the
differences.

To get rid of that, I'm proposing that you applied the attached patch to the
spf kernel. This patch is allowing to turn on/off the SPF handler through
/proc/sys/vm/speculative_page_fault.

This should ease the testing by limiting the reboot and avoid kernel's symbols
mismatch. Obviously there is still a small overhead due to the check but it
should not be viewable.

With this patch applied you can simply run
echo 1 > /proc/sys/vm/speculative_page_fault
to run a test with the speculative page fault handler activated. Or run
echo 0 > /proc/sys/vm/speculative_page_fault
to run a test without it.

I'm really sorry to asking that again, but could please run the test
page_fault3_base_THP-Always with and without SPF and capture the perf output.

I think we should focus on that test which showed the biggest regression.

Thanks,
Laurent.


>
> Thanks,
> Haiyan, Song
>
> ________________________________________
> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
> Sent: Tuesday, July 17, 2018 5:36 PM
> To: Song, HaiyanX
> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
> Subject: Re: [PATCH v11 00/26] Speculative page faults
>
> On 13/07/2018 05:56, Song, HaiyanX wrote:
>> Hi Laurent,
>
> Hi Haiyan,
>
> Thanks a lot for sharing this perf reports.
>
> I looked at them closely, and I've to admit that I was not able to found a
> major difference between the base and the head report, except that
> handle_pte_fault() is no more in-lined in the head one.
>
> As expected, __handle_speculative_fault() is never traced since these tests are
> dealing with file mapping, not handled in the speculative way.
>
> When running these test did you seen a major differences in the test's result
> between base and head ?
>
> From the number of cycles counted, the biggest difference is page_fault3 when
> run with the THP enabled:
>                                 BASE            HEAD            Delta
> page_fault2_base_thp_never      1142252426747   1065866197589   -6.69%
> page_fault2_base_THP-Alwasys    1124844374523   1076312228927   -4.31%
> page_fault3_base_thp_never      1099387298152   1134118402345   3.16%
> page_fault3_base_THP-Always     1059370178101   853985561949    -19.39%
>
>
> The very weird thing is the difference of the delta cycles reported between
> thp never and thp always, because the speculative way is aborted when checking
> for the vma->ops field, which is the same in both case, and the thp is never
> checked. So there is no code covering differnce, on the speculative path,
> between these 2 cases. This leads me to think that there are other interactions
> interfering in the measure.
>
> Looking at the perf-profile_page_fault3_*_THP-Always, the major differences at
> the head of the perf report is the 92% testcase which is weirdly not reported
> on the head side :
>     92.02%    22.33%  page_fault3_processes  [.] testcase
> 92.02% testcase
>
> Then the base reported 37.67% for __do_page_fault() where the head reported
> 48.41%, but the only difference in this function, between base and head, is the
> call to handle_speculative_fault(). But this is a macro checking for the fault
> flags, and mm->users and then calling __handle_speculative_fault() if needed.
> So this can't explain this difference, except if __handle_speculative_fault()
> is inlined in __do_page_fault().
> Is this the case on your build ?
>
> Haiyan, do you still have the output of the test to check those numbers too ?
>
> Cheers,
> Laurent
>
>> I attached the perf-profile.gz file for case page_fault2 and page_fault3. These files were captured during test the related test case.
>> Please help to check on these data if it can help you to find the higher change. Thanks.
>>
>> File name perf-profile_page_fault2_head_THP-Always.gz, means the perf-profile result get from page_fault2
>>     tested for head commit (a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12) with THP_always configuration.
>>
>> Best regards,
>> Haiyan Song
>>
>> ________________________________________
>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>> Sent: Thursday, July 12, 2018 1:05 AM
>> To: Song, HaiyanX
>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>
>> Hi Haiyan,
>>
>> Do you get a chance to capture some performance cycles on your system ?
>> I still can't get these numbers on my hardware.
>>
>> Thanks,
>> Laurent.
>>
>> On 04/07/2018 09:51, Laurent Dufour wrote:
>>> On 04/07/2018 05:23, Song, HaiyanX wrote:
>>>> Hi Laurent,
>>>>
>>>>
>>>> For the test result on Intel 4s skylake platform (192 CPUs, 768G Memory), the below test cases all were run 3 times.
>>>> I check the test results, only page_fault3_thread/enable THP have 6% stddev for head commit, other tests have lower stddev.
>>>
>>> Repeating the test only 3 times seems a bit too low to me.
>>>
>>> I'll focus on the higher change for the moment, but I don't have access to such
>>> a hardware.
>>>
>>> Is possible to provide a diff between base and SPF of the performance cycles
>>> measured when running page_fault3 and page_fault2 when the 20% change is detected.
>>>
>>> Please stay focus on the test case process to see exactly where the series is
>>> impacting.
>>>
>>> Thanks,
>>> Laurent.
>>>
>>>>
>>>> And I did not find other high variation on test case result.
>>>>
>>>> a). Enable THP
>>>> testcase                          base     stddev       change      head     stddev         metric
>>>> page_fault3/enable THP           10519      ± 3%        -20.5%      8368      ±6%          will-it-scale.per_thread_ops
>>>> page_fault2/enalbe THP            8281      ± 2%        -18.8%      6728                   will-it-scale.per_thread_ops
>>>> brk1/eanble THP                 998475                   -2.2%    976893                   will-it-scale.per_process_ops
>>>> context_switch1/enable THP      223910                   -1.3%    220930                   will-it-scale.per_process_ops
>>>> context_switch1/enable THP      233722                   -1.0%    231288                   will-it-scale.per_thread_ops
>>>>
>>>> b). Disable THP
>>>> page_fault3/disable THP          10856                  -23.1%      8344                   will-it-scale.per_thread_ops
>>>> page_fault2/disable THP           8147                  -18.8%      6613                   will-it-scale.per_thread_ops
>>>> brk1/disable THP                   957                    -7.9%      881                   will-it-scale.per_thread_ops
>>>> context_switch1/disable THP     237006                    -2.2%    231907                  will-it-scale.per_thread_ops
>>>> brk1/disable THP                997317                    -2.0%    977778                  will-it-scale.per_process_ops
>>>> page_fault3/disable THP         467454                    -1.8%    459251                  will-it-scale.per_process_ops
>>>> context_switch1/disable THP     224431                    -1.3%    221567                  will-it-scale.per_process_ops
>>>>
>>>>
>>>> Best regards,
>>>> Haiyan Song
>>>> ________________________________________
>>>> From: Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>> Sent: Monday, July 02, 2018 4:59 PM
>>>> To: Song, HaiyanX
>>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>>
>>>> On 11/06/2018 09:49, Song, HaiyanX wrote:
>>>>> Hi Laurent,
>>>>>
>>>>> Regression test for v11 patch serials have been run, some regression is found by LKP-tools (linux kernel performance)
>>>>> tested on Intel 4s skylake platform. This time only test the cases which have been run and found regressions on
>>>>> V9 patch serials.
>>>>>
>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>> branch: Laurent-Dufour/Speculative-page-faults/20180520-045126
>>>>> commit id:
>>>>>   head commit : a7a8993bfe3ccb54ad468b9f1799649e4ad1ff12
>>>>>   base commit : ba98a1cdad71d259a194461b3a61471b49b14df1
>>>>> Benchmark: will-it-scale
>>>>> Download link: https://github.com/antonblanchard/will-it-scale/tree/master
>>>>>
>>>>> Metrics:
>>>>>   will-it-scale.per_process_ops=processes/nr_cpu
>>>>>   will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>   test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>> THP: enable / disable
>>>>> nr_task:100%
>>>>>
>>>>> 1. Regressions:
>>>>>
>>>>> a). Enable THP
>>>>> testcase                          base           change      head           metric
>>>>> page_fault3/enable THP           10519          -20.5%        836      will-it-scale.per_thread_ops
>>>>> page_fault2/enalbe THP            8281          -18.8%       6728      will-it-scale.per_thread_ops
>>>>> brk1/eanble THP                 998475           -2.2%     976893      will-it-scale.per_process_ops
>>>>> context_switch1/enable THP      223910           -1.3%     220930      will-it-scale.per_process_ops
>>>>> context_switch1/enable THP      233722           -1.0%     231288      will-it-scale.per_thread_ops
>>>>>
>>>>> b). Disable THP
>>>>> page_fault3/disable THP          10856          -23.1%       8344      will-it-scale.per_thread_ops
>>>>> page_fault2/disable THP           8147          -18.8%       6613      will-it-scale.per_thread_ops
>>>>> brk1/disable THP                   957           -7.9%        881      will-it-scale.per_thread_ops
>>>>> context_switch1/disable THP     237006           -2.2%     231907      will-it-scale.per_thread_ops
>>>>> brk1/disable THP                997317           -2.0%     977778      will-it-scale.per_process_ops
>>>>> page_fault3/disable THP         467454           -1.8%     459251      will-it-scale.per_process_ops
>>>>> context_switch1/disable THP     224431           -1.3%     221567      will-it-scale.per_process_ops
>>>>>
>>>>> Notes: for the above  values of test result, the higher is better.
>>>>
>>>> I tried the same tests on my PowerPC victim VM (1024 CPUs, 11TB) and I can't
>>>> get reproducible results. The results have huge variation, even on the vanilla
>>>> kernel, and I can't state on any changes due to that.
>>>>
>>>> I tried on smaller node (80 CPUs, 32G), and the tests ran better, but I didn't
>>>> measure any changes between the vanilla and the SPF patched ones:
>>>>
>>>> test THP enabled                4.17.0-rc4-mm1  spf             delta
>>>> page_fault3_threads             2697.7          2683.5          -0.53%
>>>> page_fault2_threads             170660.6        169574.1        -0.64%
>>>> context_switch1_threads         6915269.2       6877507.3       -0.55%
>>>> context_switch1_processes       6478076.2       6529493.5       0.79%
>>>> brk1                            243391.2        238527.5        -2.00%
>>>>
>>>> Tests were run 10 times, no high variation detected.
>>>>
>>>> Did you see high variation on your side ? How many times the test were run to
>>>> compute the average values ?
>>>>
>>>> Thanks,
>>>> Laurent.
>>>>
>>>>
>>>>>
>>>>> 2. Improvement: not found improvement based on the selected test cases.
>>>>>
>>>>>
>>>>> Best regards
>>>>> Haiyan Song
>>>>> ________________________________________
>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>> Sent: Monday, May 28, 2018 4:54 PM
>>>>> To: Song, HaiyanX
>>>>> Cc: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi; linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>> Subject: Re: [PATCH v11 00/26] Speculative page faults
>>>>>
>>>>> On 28/05/2018 10:22, Haiyan Song wrote:
>>>>>> Hi Laurent,
>>>>>>
>>>>>> Yes, these tests are done on V9 patch.
>>>>>
>>>>> Do you plan to give this V11 a run ?
>>>>>
>>>>>>
>>>>>>
>>>>>> Best regards,
>>>>>> Haiyan Song
>>>>>>
>>>>>> On Mon, May 28, 2018 at 09:51:34AM +0200, Laurent Dufour wrote:
>>>>>>> On 28/05/2018 07:23, Song, HaiyanX wrote:
>>>>>>>>
>>>>>>>> Some regression and improvements is found by LKP-tools(linux kernel performance) on V9 patch series
>>>>>>>> tested on Intel 4s Skylake platform.
>>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>> Thanks for reporting this benchmark results, but you mentioned the "V9 patch
>>>>>>> series" while responding to the v11 header series...
>>>>>>> Were these tests done on v9 or v11 ?
>>>>>>>
>>>>>>> Cheers,
>>>>>>> Laurent.
>>>>>>>
>>>>>>>>
>>>>>>>> The regression result is sorted by the metric will-it-scale.per_thread_ops.
>>>>>>>> Branch: Laurent-Dufour/Speculative-page-faults/20180316-151833 (V9 patch series)
>>>>>>>> Commit id:
>>>>>>>>     base commit: d55f34411b1b126429a823d06c3124c16283231f
>>>>>>>>     head commit: 0355322b3577eeab7669066df42c550a56801110
>>>>>>>> Benchmark suite: will-it-scale
>>>>>>>> Download link:
>>>>>>>> https://github.com/antonblanchard/will-it-scale/tree/master/tests
>>>>>>>> Metrics:
>>>>>>>>     will-it-scale.per_process_ops=processes/nr_cpu
>>>>>>>>     will-it-scale.per_thread_ops=threads/nr_cpu
>>>>>>>> test box: lkp-skl-4sp1(nr_cpu=192,memory=768G)
>>>>>>>> THP: enable / disable
>>>>>>>> nr_task: 100%
>>>>>>>>
>>>>>>>> 1. Regressions:
>>>>>>>> a) THP enabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> page_fault3/ enable THP         10092           -17.5%          8323       will-it-scale.per_thread_ops
>>>>>>>> page_fault2/ enable THP          8300           -17.2%          6869       will-it-scale.per_thread_ops
>>>>>>>> brk1/ enable THP                  957.67         -7.6%           885       will-it-scale.per_thread_ops
>>>>>>>> page_fault3/ enable THP        172821            -5.3%        163692       will-it-scale.per_process_ops
>>>>>>>> signal1/ enable THP              9125            -3.2%          8834       will-it-scale.per_process_ops
>>>>>>>>
>>>>>>>> b) THP disabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> page_fault3/ disable THP        10107           -19.1%          8180       will-it-scale.per_thread_ops
>>>>>>>> page_fault2/ disable THP         8432           -17.8%          6931       will-it-scale.per_thread_ops
>>>>>>>> context_switch1/ disable THP   215389            -6.8%        200776       will-it-scale.per_thread_ops
>>>>>>>> brk1/ disable THP                 939.67         -6.6%           877.33    will-it-scale.per_thread_ops
>>>>>>>> page_fault3/ disable THP       173145            -4.7%        165064       will-it-scale.per_process_ops
>>>>>>>> signal1/ disable THP             9162            -3.9%          8802       will-it-scale.per_process_ops
>>>>>>>>
>>>>>>>> 2. Improvements:
>>>>>>>> a) THP enabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> malloc1/ enable THP               66.33        +469.8%           383.67    will-it-scale.per_thread_ops
>>>>>>>> writeseek3/ enable THP          2531             +4.5%          2646       will-it-scale.per_thread_ops
>>>>>>>> signal1/ enable THP              989.33          +2.8%          1016       will-it-scale.per_thread_ops
>>>>>>>>
>>>>>>>> b) THP disabled:
>>>>>>>> testcase                        base            change          head       metric
>>>>>>>> malloc1/ disable THP              90.33        +417.3%           467.33    will-it-scale.per_thread_ops
>>>>>>>> read2/ disable THP             58934            +39.2%         82060       will-it-scale.per_thread_ops
>>>>>>>> page_fault1/ disable THP        8607            +36.4%         11736       will-it-scale.per_thread_ops
>>>>>>>> read1/ disable THP            314063            +12.7%        353934       will-it-scale.per_thread_ops
>>>>>>>> writeseek3/ disable THP         2452            +12.5%          2759       will-it-scale.per_thread_ops
>>>>>>>> signal1/ disable THP             971.33          +5.5%          1024       will-it-scale.per_thread_ops
>>>>>>>>
>>>>>>>> Notes: for above values in column "change", the higher value means that the related testcase result
>>>>>>>> on head commit is better than that on base commit for this benchmark.
>>>>>>>>
>>>>>>>>
>>>>>>>> Best regards
>>>>>>>> Haiyan Song
>>>>>>>>
>>>>>>>> ________________________________________
>>>>>>>> From: owner-linux-mm@kvack.org [owner-linux-mm@kvack.org] on behalf of Laurent Dufour [ldufour@linux.vnet.ibm.com]
>>>>>>>> Sent: Thursday, May 17, 2018 7:06 PM
>>>>>>>> To: akpm@linux-foundation.org; mhocko@kernel.org; peterz@infradead.org; kirill@shutemov.name; ak@linux.intel.com; dave@stgolabs.net; jack@suse.cz; Matthew Wilcox; khandual@linux.vnet.ibm.com; aneesh.kumar@linux.vnet.ibm.com; benh@kernel.crashing.org; mpe@ellerman.id.au; paulus@samba.org; Thomas Gleixner; Ingo Molnar; hpa@zytor.com; Will Deacon; Sergey Senozhatsky; sergey.senozhatsky.work@gmail.com; Andrea Arcangeli; Alexei Starovoitov; Wang, Kemi; Daniel Jordan; David Rientjes; Jerome Glisse; Ganesh Mahendran; Minchan Kim; Punit Agrawal; vinayak menon; Yang Shi
>>>>>>>> Cc: linux-kernel@vger.kernel.org; linux-mm@kvack.org; haren@linux.vnet.ibm.com; npiggin@gmail.com; bsingharora@gmail.com; paulmck@linux.vnet.ibm.com; Tim Chen; linuxppc-dev@lists.ozlabs.org; x86@kernel.org
>>>>>>>> Subject: [PATCH v11 00/26] Speculative page faults
>>>>>>>>
>>>>>>>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>>>>>>>> page fault without holding the mm semaphore [1].
>>>>>>>>
>>>>>>>> The idea is to try to handle user space page faults without holding the
>>>>>>>> mmap_sem. This should allow better concurrency for massively threaded
>>>>>>>> process since the page fault handler will not wait for other threads memory
>>>>>>>> layout change to be done, assuming that this change is done in another part
>>>>>>>> of the process's memory space. This type page fault is named speculative
>>>>>>>> page fault. If the speculative page fault fails because of a concurrency is
>>>>>>>> detected or because underlying PMD or PTE tables are not yet allocating, it
>>>>>>>> is failing its processing and a classic page fault is then tried.
>>>>>>>>
>>>>>>>> The speculative page fault (SPF) has to look for the VMA matching the fault
>>>>>>>> address without holding the mmap_sem, this is done by introducing a rwlock
>>>>>>>> which protects the access to the mm_rb tree. Previously this was done using
>>>>>>>> SRCU but it was introducing a lot of scheduling to process the VMA's
>>>>>>>> freeing operation which was hitting the performance by 20% as reported by
>>>>>>>> Kemi Wang [2]. Using a rwlock to protect access to the mm_rb tree is
>>>>>>>> limiting the locking contention to these operations which are expected to
>>>>>>>> be in a O(log n) order. In addition to ensure that the VMA is not freed in
>>>>>>>> our back a reference count is added and 2 services (get_vma() and
>>>>>>>> put_vma()) are introduced to handle the reference count. Once a VMA is
>>>>>>>> fetched from the RB tree using get_vma(), it must be later freed using
>>>>>>>> put_vma(). I can't see anymore the overhead I got while will-it-scale
>>>>>>>> benchmark anymore.
>>>>>>>>
>>>>>>>> The VMA's attributes checked during the speculative page fault processing
>>>>>>>> have to be protected against parallel changes. This is done by using a per
>>>>>>>> VMA sequence lock. This sequence lock allows the speculative page fault
>>>>>>>> handler to fast check for parallel changes in progress and to abort the
>>>>>>>> speculative page fault in that case.
>>>>>>>>
>>>>>>>> Once the VMA has been found, the speculative page fault handler would check
>>>>>>>> for the VMA's attributes to verify that the page fault has to be handled
>>>>>>>> correctly or not. Thus, the VMA is protected through a sequence lock which
>>>>>>>> allows fast detection of concurrent VMA changes. If such a change is
>>>>>>>> detected, the speculative page fault is aborted and a *classic* page fault
>>>>>>>> is tried.  VMA sequence lockings are added when VMA attributes which are
>>>>>>>> checked during the page fault are modified.
>>>>>>>>
>>>>>>>> When the PTE is fetched, the VMA is checked to see if it has been changed,
>>>>>>>> so once the page table is locked, the VMA is valid, so any other changes
>>>>>>>> leading to touching this PTE will need to lock the page table, so no
>>>>>>>> parallel change is possible at this time.
>>>>>>>>
>>>>>>>> The locking of the PTE is done with interrupts disabled, this allows
>>>>>>>> checking for the PMD to ensure that there is not an ongoing collapsing
>>>>>>>> operation. Since khugepaged is firstly set the PMD to pmd_none and then is
>>>>>>>> waiting for the other CPU to have caught the IPI interrupt, if the pmd is
>>>>>>>> valid at the time the PTE is locked, we have the guarantee that the
>>>>>>>> collapsing operation will have to wait on the PTE lock to move forward.
>>>>>>>> This allows the SPF handler to map the PTE safely. If the PMD value is
>>>>>>>> different from the one recorded at the beginning of the SPF operation, the
>>>>>>>> classic page fault handler will be called to handle the operation while
>>>>>>>> holding the mmap_sem. As the PTE lock is done with the interrupts disabled,
>>>>>>>> the lock is done using spin_trylock() to avoid dead lock when handling a
>>>>>>>> page fault while a TLB invalidate is requested by another CPU holding the
>>>>>>>> PTE.
>>>>>>>>
>>>>>>>> In pseudo code, this could be seen as:
>>>>>>>>     speculative_page_fault()
>>>>>>>>     {
>>>>>>>>             vma = get_vma()
>>>>>>>>             check vma sequence count
>>>>>>>>             check vma's support
>>>>>>>>             disable interrupt
>>>>>>>>                   check pgd,p4d,...,pte
>>>>>>>>                   save pmd and pte in vmf
>>>>>>>>                   save vma sequence counter in vmf
>>>>>>>>             enable interrupt
>>>>>>>>             check vma sequence count
>>>>>>>>             handle_pte_fault(vma)
>>>>>>>>                     ..
>>>>>>>>                     page = alloc_page()
>>>>>>>>                     pte_map_lock()
>>>>>>>>                             disable interrupt
>>>>>>>>                                     abort if sequence counter has changed
>>>>>>>>                                     abort if pmd or pte has changed
>>>>>>>>                                     pte map and lock
>>>>>>>>                             enable interrupt
>>>>>>>>                     if abort
>>>>>>>>                        free page
>>>>>>>>                        abort
>>>>>>>>                     ...
>>>>>>>>     }
>>>>>>>>
>>>>>>>>     arch_fault_handler()
>>>>>>>>     {
>>>>>>>>             if (speculative_page_fault(&vma))
>>>>>>>>                goto done
>>>>>>>>     again:
>>>>>>>>             lock(mmap_sem)
>>>>>>>>             vma = find_vma();
>>>>>>>>             handle_pte_fault(vma);
>>>>>>>>             if retry
>>>>>>>>                unlock(mmap_sem)
>>>>>>>>                goto again;
>>>>>>>>     done:
>>>>>>>>             handle fault error
>>>>>>>>     }
>>>>>>>>
>>>>>>>> Support for THP is not done because when checking for the PMD, we can be
>>>>>>>> confused by an in progress collapsing operation done by khugepaged. The
>>>>>>>> issue is that pmd_none() could be true either if the PMD is not already
>>>>>>>> populated or if the underlying PTE are in the way to be collapsed. So we
>>>>>>>> cannot safely allocate a PMD if pmd_none() is true.
>>>>>>>>
>>>>>>>> This series add a new software performance event named 'speculative-faults'
>>>>>>>> or 'spf'. It counts the number of successful page fault event handled
>>>>>>>> speculatively. When recording 'faults,spf' events, the faults one is
>>>>>>>> counting the total number of page fault events while 'spf' is only counting
>>>>>>>> the part of the faults processed speculatively.
>>>>>>>>
>>>>>>>> There are some trace events introduced by this series. They allow
>>>>>>>> identifying why the page faults were not processed speculatively. This
>>>>>>>> doesn't take in account the faults generated by a monothreaded process
>>>>>>>> which directly processed while holding the mmap_sem. This trace events are
>>>>>>>> grouped in a system named 'pagefault', they are:
>>>>>>>>  - pagefault:spf_vma_changed : if the VMA has been changed in our back
>>>>>>>>  - pagefault:spf_vma_noanon : the vma->anon_vma field was not yet set.
>>>>>>>>  - pagefault:spf_vma_notsup : the VMA's type is not supported
>>>>>>>>  - pagefault:spf_vma_access : the VMA's access right are not respected
>>>>>>>>  - pagefault:spf_pmd_changed : the upper PMD pointer has changed in our
>>>>>>>>    back.
>>>>>>>>
>>>>>>>> To record all the related events, the easier is to run perf with the
>>>>>>>> following arguments :
>>>>>>>> $ perf stat -e 'faults,spf,pagefault:*' <command>
>>>>>>>>
>>>>>>>> There is also a dedicated vmstat counter showing the number of successful
>>>>>>>> page fault handled speculatively. I can be seen this way:
>>>>>>>> $ grep speculative_pgfault /proc/vmstat
>>>>>>>>
>>>>>>>> This series builds on top of v4.16-mmotm-2018-04-13-17-28 and is functional
>>>>>>>> on x86, PowerPC and arm64.
>>>>>>>>
>>>>>>>> ---------------------
>>>>>>>> Real Workload results
>>>>>>>>
>>>>>>>> As mentioned in previous email, we did non official runs using a "popular
>>>>>>>> in memory multithreaded database product" on 176 cores SMT8 Power system
>>>>>>>> which showed a 30% improvements in the number of transaction processed per
>>>>>>>> second. This run has been done on the v6 series, but changes introduced in
>>>>>>>> this new version should not impact the performance boost seen.
>>>>>>>>
>>>>>>>> Here are the perf data captured during 2 of these runs on top of the v8
>>>>>>>> series:
>>>>>>>>                 vanilla         spf
>>>>>>>> faults          89.418          101.364         +13%
>>>>>>>> spf                n/a           97.989
>>>>>>>>
>>>>>>>> With the SPF kernel, most of the page fault were processed in a speculative
>>>>>>>> way.
>>>>>>>>
>>>>>>>> Ganesh Mahendran had backported the series on top of a 4.9 kernel and gave
>>>>>>>> it a try on an android device. He reported that the application launch time
>>>>>>>> was improved in average by 6%, and for large applications (~100 threads) by
>>>>>>>> 20%.
>>>>>>>>
>>>>>>>> Here are the launch time Ganesh mesured on Android 8.0 on top of a Qcom
>>>>>>>> MSM845 (8 cores) with 6GB (the less is better):
>>>>>>>>
>>>>>>>> Application                             4.9     4.9+spf delta
>>>>>>>> com.tencent.mm                          416     389     -7%
>>>>>>>> com.eg.android.AlipayGphone             1135    986     -13%
>>>>>>>> com.tencent.mtt                         455     454     0%
>>>>>>>> com.qqgame.hlddz                        1497    1409    -6%
>>>>>>>> com.autonavi.minimap                    711     701     -1%
>>>>>>>> com.tencent.tmgp.sgame                  788     748     -5%
>>>>>>>> com.immomo.momo                         501     487     -3%
>>>>>>>> com.tencent.peng                        2145    2112    -2%
>>>>>>>> com.smile.gifmaker                      491     461     -6%
>>>>>>>> com.baidu.BaiduMap                      479     366     -23%
>>>>>>>> com.taobao.taobao                       1341    1198    -11%
>>>>>>>> com.baidu.searchbox                     333     314     -6%
>>>>>>>> com.tencent.mobileqq                    394     384     -3%
>>>>>>>> com.sina.weibo                          907     906     0%
>>>>>>>> com.youku.phone                         816     731     -11%
>>>>>>>> com.happyelements.AndroidAnimal.qq      763     717     -6%
>>>>>>>> com.UCMobile                            415     411     -1%
>>>>>>>> com.tencent.tmgp.ak                     1464    1431    -2%
>>>>>>>> com.tencent.qqmusic                     336     329     -2%
>>>>>>>> com.sankuai.meituan                     1661    1302    -22%
>>>>>>>> com.netease.cloudmusic                  1193    1200    1%
>>>>>>>> air.tv.douyu.android                    4257    4152    -2%
>>>>>>>>
>>>>>>>> ------------------
>>>>>>>> Benchmarks results
>>>>>>>>
>>>>>>>> Base kernel is v4.17.0-rc4-mm1
>>>>>>>> SPF is BASE + this series
>>>>>>>>
>>>>>>>> Kernbench:
>>>>>>>> ----------
>>>>>>>> Here are the results on a 16 CPUs X86 guest using kernbench on a 4.15
>>>>>>>> kernel (kernel is build 5 times):
>>>>>>>>
>>>>>>>> Average Half load -j 8
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     1448.65 (5.72312)      1455.84 (4.84951)       0.50%
>>>>>>>> User    Time     10135.4 (30.3699)      10148.8 (31.1252)       0.13%
>>>>>>>> System  Time     900.47  (2.81131)      923.28  (7.52779)       2.53%
>>>>>>>> Percent CPU      761.4   (1.14018)      760.2   (0.447214)      -0.16%
>>>>>>>> Context Switches 85380   (3419.52)      84748   (1904.44)       -0.74%
>>>>>>>> Sleeps           105064  (1240.96)      105074  (337.612)       0.01%
>>>>>>>>
>>>>>>>> Average Optimal load -j 16
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     920.528 (10.1212)      927.404 (8.91789)       0.75%
>>>>>>>> User    Time     11064.8 (981.142)      11085   (990.897)       0.18%
>>>>>>>> System  Time     979.904 (84.0615)      1001.14 (82.5523)       2.17%
>>>>>>>> Percent CPU      1089.5  (345.894)      1086.1  (343.545)       -0.31%
>>>>>>>> Context Switches 159488  (78156.4)      158223  (77472.1)       -0.79%
>>>>>>>> Sleeps           110566  (5877.49)      110388  (5617.75)       -0.16%
>>>>>>>>
>>>>>>>>
>>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>>          526743764      faults
>>>>>>>>                210      spf
>>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>               2278      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> Very few speculative page faults were recorded as most of the processes
>>>>>>>> involved are monothreaded (sounds that on this architecture some threads
>>>>>>>> were created during the kernel build processing).
>>>>>>>>
>>>>>>>> Here are the kerbench results on a 80 CPUs Power8 system:
>>>>>>>>
>>>>>>>> Average Half load -j 40
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     117.152 (0.774642)     117.166 (0.476057)      0.01%
>>>>>>>> User    Time     4478.52 (24.7688)      4479.76 (9.08555)       0.03%
>>>>>>>> System  Time     131.104 (0.720056)     134.04  (0.708414)      2.24%
>>>>>>>> Percent CPU      3934    (19.7104)      3937.2  (19.0184)       0.08%
>>>>>>>> Context Switches 92125.4 (576.787)      92581.6 (198.622)       0.50%
>>>>>>>> Sleeps           317923  (652.499)      318469  (1255.59)       0.17%
>>>>>>>>
>>>>>>>> Average Optimal load -j 80
>>>>>>>>                  Run    (std deviation)
>>>>>>>>                  BASE                   SPF
>>>>>>>> Elapsed Time     107.73  (0.632416)     107.31  (0.584936)      -0.39%
>>>>>>>> User    Time     5869.86 (1466.72)      5871.71 (1467.27)       0.03%
>>>>>>>> System  Time     153.728 (23.8573)      157.153 (24.3704)       2.23%
>>>>>>>> Percent CPU      5418.6  (1565.17)      5436.7  (1580.91)       0.33%
>>>>>>>> Context Switches 223861  (138865)       225032  (139632)        0.52%
>>>>>>>> Sleeps           330529  (13495.1)      332001  (14746.2)       0.45%
>>>>>>>>
>>>>>>>> During a run on the SPF, perf events were captured:
>>>>>>>>  Performance counter stats for '../kernbench -M':
>>>>>>>>          116730856      faults
>>>>>>>>                  0      spf
>>>>>>>>                  3      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>                476      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> Most of the processes involved are monothreaded so SPF is not activated but
>>>>>>>> there is no impact on the performance.
>>>>>>>>
>>>>>>>> Ebizzy:
>>>>>>>> -------
>>>>>>>> The test is counting the number of records per second it can manage, the
>>>>>>>> higher is the best. I run it like this 'ebizzy -mTt <nrcpus>'. To get
>>>>>>>> consistent result I repeated the test 100 times and measure the average
>>>>>>>> result. The number is the record processes per second, the higher is the
>>>>>>>> best.
>>>>>>>>
>>>>>>>>                 BASE            SPF             delta
>>>>>>>> 16 CPUs x86 VM  742.57          1490.24         100.69%
>>>>>>>> 80 CPUs P8 node 13105.4         24174.23        84.46%
>>>>>>>>
>>>>>>>> Here are the performance counter read during a run on a 16 CPUs x86 VM:
>>>>>>>>  Performance counter stats for './ebizzy -mTt 16':
>>>>>>>>            1706379      faults
>>>>>>>>            1674599      spf
>>>>>>>>              30588      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>                363      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> And the ones captured during a run on a 80 CPUs Power node:
>>>>>>>>  Performance counter stats for './ebizzy -mTt 80':
>>>>>>>>            1874773      faults
>>>>>>>>            1461153      spf
>>>>>>>>             413293      pagefault:spf_vma_changed
>>>>>>>>                  0      pagefault:spf_vma_noanon
>>>>>>>>                200      pagefault:spf_vma_notsup
>>>>>>>>                  0      pagefault:spf_vma_access
>>>>>>>>                  0      pagefault:spf_pmd_changed
>>>>>>>>
>>>>>>>> In ebizzy's case most of the page fault were handled in a speculative way,
>>>>>>>> leading the ebizzy performance boost.
>>>>>>>>
>>>>>>>> ------------------
>>>>>>>> Changes since v10 (https://lkml.org/lkml/2018/4/17/572):
>>>>>>>>  - Accounted for all review feedbacks from Punit Agrawal, Ganesh Mahendran
>>>>>>>>    and Minchan Kim, hopefully.
>>>>>>>>  - Remove unneeded check on CONFIG_SPECULATIVE_PAGE_FAULT in
>>>>>>>>    __do_page_fault().
>>>>>>>>  - Loop in pte_spinlock() and pte_map_lock() when pte try lock fails
>>>>>>>>    instead
>>>>>>>>    of aborting the speculative page fault handling. Dropping the now
>>>>>>>> useless
>>>>>>>>    trace event pagefault:spf_pte_lock.
>>>>>>>>  - No more try to reuse the fetched VMA during the speculative page fault
>>>>>>>>    handling when retrying is needed. This adds a lot of complexity and
>>>>>>>>    additional tests done didn't show a significant performance improvement.
>>>>>>>>  - Convert IS_ENABLED(CONFIG_NUMA) back to #ifdef due to build error.
>>>>>>>>
>>>>>>>> [1] http://linux-kernel.2935.n7.nabble.com/RFC-PATCH-0-6-Another-go-at-speculative-page-faults-tt965642.html#none
>>>>>>>> [2] https://patchwork.kernel.org/patch/9999687/
>>>>>>>>
>>>>>>>>
>>>>>>>> Laurent Dufour (20):
>>>>>>>>   mm: introduce CONFIG_SPECULATIVE_PAGE_FAULT
>>>>>>>>   x86/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>>   powerpc/mm: set ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>>   mm: introduce pte_spinlock for FAULT_FLAG_SPECULATIVE
>>>>>>>>   mm: make pte_unmap_same compatible with SPF
>>>>>>>>   mm: introduce INIT_VMA()
>>>>>>>>   mm: protect VMA modifications using VMA sequence count
>>>>>>>>   mm: protect mremap() against SPF hanlder
>>>>>>>>   mm: protect SPF handler against anon_vma changes
>>>>>>>>   mm: cache some VMA fields in the vm_fault structure
>>>>>>>>   mm/migrate: Pass vm_fault pointer to migrate_misplaced_page()
>>>>>>>>   mm: introduce __lru_cache_add_active_or_unevictable
>>>>>>>>   mm: introduce __vm_normal_page()
>>>>>>>>   mm: introduce __page_add_new_anon_rmap()
>>>>>>>>   mm: protect mm_rb tree with a rwlock
>>>>>>>>   mm: adding speculative page fault failure trace events
>>>>>>>>   perf: add a speculative page fault sw event
>>>>>>>>   perf tools: add support for the SPF perf event
>>>>>>>>   mm: add speculative page fault vmstats
>>>>>>>>   powerpc/mm: add speculative page fault
>>>>>>>>
>>>>>>>> Mahendran Ganesh (2):
>>>>>>>>   arm64/mm: define ARCH_SUPPORTS_SPECULATIVE_PAGE_FAULT
>>>>>>>>   arm64/mm: add speculative page fault
>>>>>>>>
>>>>>>>> Peter Zijlstra (4):
>>>>>>>>   mm: prepare for FAULT_FLAG_SPECULATIVE
>>>>>>>>   mm: VMA sequence count
>>>>>>>>   mm: provide speculative fault infrastructure
>>>>>>>>   x86/mm: add speculative pagefault handling
>>>>>>>>
>>>>>>>>  arch/arm64/Kconfig                    |   1 +
>>>>>>>>  arch/arm64/mm/fault.c                 |  12 +
>>>>>>>>  arch/powerpc/Kconfig                  |   1 +
>>>>>>>>  arch/powerpc/mm/fault.c               |  16 +
>>>>>>>>  arch/x86/Kconfig                      |   1 +
>>>>>>>>  arch/x86/mm/fault.c                   |  27 +-
>>>>>>>>  fs/exec.c                             |   2 +-
>>>>>>>>  fs/proc/task_mmu.c                    |   5 +-
>>>>>>>>  fs/userfaultfd.c                      |  17 +-
>>>>>>>>  include/linux/hugetlb_inline.h        |   2 +-
>>>>>>>>  include/linux/migrate.h               |   4 +-
>>>>>>>>  include/linux/mm.h                    | 136 +++++++-
>>>>>>>>  include/linux/mm_types.h              |   7 +
>>>>>>>>  include/linux/pagemap.h               |   4 +-
>>>>>>>>  include/linux/rmap.h                  |  12 +-
>>>>>>>>  include/linux/swap.h                  |  10 +-
>>>>>>>>  include/linux/vm_event_item.h         |   3 +
>>>>>>>>  include/trace/events/pagefault.h      |  80 +++++
>>>>>>>>  include/uapi/linux/perf_event.h       |   1 +
>>>>>>>>  kernel/fork.c                         |   5 +-
>>>>>>>>  mm/Kconfig                            |  22 ++
>>>>>>>>  mm/huge_memory.c                      |   6 +-
>>>>>>>>  mm/hugetlb.c                          |   2 +
>>>>>>>>  mm/init-mm.c                          |   3 +
>>>>>>>>  mm/internal.h                         |  20 ++
>>>>>>>>  mm/khugepaged.c                       |   5 +
>>>>>>>>  mm/madvise.c                          |   6 +-
>>>>>>>>  mm/memory.c                           | 612 +++++++++++++++++++++++++++++-----
>>>>>>>>  mm/mempolicy.c                        |  51 ++-
>>>>>>>>  mm/migrate.c                          |   6 +-
>>>>>>>>  mm/mlock.c                            |  13 +-
>>>>>>>>  mm/mmap.c                             | 229 ++++++++++---
>>>>>>>>  mm/mprotect.c                         |   4 +-
>>>>>>>>  mm/mremap.c                           |  13 +
>>>>>>>>  mm/nommu.c                            |   2 +-
>>>>>>>>  mm/rmap.c                             |   5 +-
>>>>>>>>  mm/swap.c                             |   6 +-
>>>>>>>>  mm/swap_state.c                       |   8 +-
>>>>>>>>  mm/vmstat.c                           |   5 +-
>>>>>>>>  tools/include/uapi/linux/perf_event.h |   1 +
>>>>>>>>  tools/perf/util/evsel.c               |   1 +
>>>>>>>>  tools/perf/util/parse-events.c        |   4 +
>>>>>>>>  tools/perf/util/parse-events.l        |   1 +
>>>>>>>>  tools/perf/util/python.c              |   1 +
>>>>>>>>  44 files changed, 1161 insertions(+), 211 deletions(-)
>>>>>>>>  create mode 100644 include/trace/events/pagefault.h
>>>>>>>>
>>>>>>>> --
>>>>>>>> 2.7.4
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>>
>>>
>>
>


[-- Attachment #2: perf-profile_page_fault3-head-thp-always-SPF-off.gz --]
[-- Type: application/gzip, Size: 11278 bytes --]

[-- Attachment #3: perf-profile_page_fault3-head-thp-always-SPF-on.gz --]
[-- Type: application/gzip, Size: 11424 bytes --]

Re: [PATCH v11 10/26] mm: protect VMA modifications using VMA sequence count

[vinayak menon]

Hi Laurent,

On Thu, May 17, 2018 at 4:37 PM Laurent Dufour
<ldufour@linux.vnet.ibm.com> wrote:
>
> The VMA sequence count has been introduced to allow fast detection of
> VMA modification when running a page fault handler without holding
> the mmap_sem.
>
> This patch provides protection against the VMA modification done in :
>         - madvise()
>         - mpol_rebind_policy()
>         - vma_replace_policy()
>         - change_prot_numa()
>         - mlock(), munlock()
>         - mprotect()
>         - mmap_region()
>         - collapse_huge_page()
>         - userfaultd registering services
>
> In addition, VMA fields which will be read during the speculative fault
> path needs to be written using WRITE_ONCE to prevent write to be split
> and intermediate values to be pushed to other CPUs.
>
> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
> ---
>  fs/proc/task_mmu.c |  5 ++++-
>  fs/userfaultfd.c   | 17 +++++++++++++----
>  mm/khugepaged.c    |  3 +++
>  mm/madvise.c       |  6 +++++-
>  mm/mempolicy.c     | 51 ++++++++++++++++++++++++++++++++++-----------------
>  mm/mlock.c         | 13 ++++++++-----
>  mm/mmap.c          | 22 +++++++++++++---------
>  mm/mprotect.c      |  4 +++-
>  mm/swap_state.c    |  8 ++++++--
>  9 files changed, 89 insertions(+), 40 deletions(-)
>
>  struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
>                                 struct vm_fault *vmf)
> @@ -665,9 +669,9 @@ static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
>                                      unsigned long *start,
>                                      unsigned long *end)
>  {
> -       *start = max3(lpfn, PFN_DOWN(vma->vm_start),
> +       *start = max3(lpfn, PFN_DOWN(READ_ONCE(vma->vm_start)),
>                       PFN_DOWN(faddr & PMD_MASK));
> -       *end = min3(rpfn, PFN_DOWN(vma->vm_end),
> +       *end = min3(rpfn, PFN_DOWN(READ_ONCE(vma->vm_end)),
>                     PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
>  }
>
> --
> 2.7.4
>

I have got a crash on 4.14 kernel with speculative page faults enabled
and here is my analysis of the problem.
The issue was reported only once.

[23409.303395]  el1_da+0x24/0x84
[23409.303400]  __radix_tree_lookup+0x8/0x90
[23409.303407]  find_get_entry+0x64/0x14c
[23409.303410]  pagecache_get_page+0x5c/0x27c
[23409.303416]  __read_swap_cache_async+0x80/0x260
[23409.303420]  swap_vma_readahead+0x264/0x37c
[23409.303423]  swapin_readahead+0x5c/0x6c
[23409.303428]  do_swap_page+0x128/0x6e4
[23409.303431]  handle_pte_fault+0x230/0xca4
[23409.303435]  __handle_speculative_fault+0x57c/0x7c8
[23409.303438]  do_page_fault+0x228/0x3e8
[23409.303442]  do_translation_fault+0x50/0x6c
[23409.303445]  do_mem_abort+0x5c/0xe0
[23409.303447]  el0_da+0x20/0x24

Process A accesses address ADDR (part of VMA A) and that results in a
translation fault.
Kernel enters __handle_speculative_fault to fix the fault.
Process A enters do_swap_page->swapin_readahead->swap_vma_readahead
from speculative path.
During this time, another process B which shares the same mm, does a
mprotect from another CPU which follows
mprotect_fixup->__split_vma, and it splits VMA A into VMAs A and B.
After the split, ADDR falls into VMA B, but process A is still using
VMA A.
Now ADDR is greater than VMA_A->vm_start and VMA_A->vm_end.
swap_vma_readahead->swap_ra_info uses start and end of vma to
calculate ptes and nr_pte, which goes wrong due to this and finally
resulting in wrong "entry" passed to
swap_vma_readahead->__read_swap_cache_async, and in turn causing
invalid swapper_space
being passed to __read_swap_cache_async->find_get_page, causing an abort.

The fix I have tried is to cache vm_start and vm_end also in vmf and
use it in swap_ra_clamp_pfn. Let me know your thoughts on this. I can
send
the patch I am a using if you feel that is the right thing to do.

Thanks,
Vinayak

Re: [PATCH v11 10/26] mm: protect VMA modifications using VMA sequence count

[vinayak menon]

Hi Laurent,

On Thu, May 17, 2018 at 4:37 PM Laurent Dufour
<ldufour@linux.vnet.ibm.com> wrote:
>
> The VMA sequence count has been introduced to allow fast detection of
> VMA modification when running a page fault handler without holding
> the mmap_sem.
>
> This patch provides protection against the VMA modification done in :
>         - madvise()
>         - mpol_rebind_policy()
>         - vma_replace_policy()
>         - change_prot_numa()
>         - mlock(), munlock()
>         - mprotect()
>         - mmap_region()
>         - collapse_huge_page()
>         - userfaultd registering services
>
> In addition, VMA fields which will be read during the speculative fault
> path needs to be written using WRITE_ONCE to prevent write to be split
> and intermediate values to be pushed to other CPUs.
>
> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
> ---
>  fs/proc/task_mmu.c |  5 ++++-
>  fs/userfaultfd.c   | 17 +++++++++++++----
>  mm/khugepaged.c    |  3 +++
>  mm/madvise.c       |  6 +++++-
>  mm/mempolicy.c     | 51 ++++++++++++++++++++++++++++++++++-----------------
>  mm/mlock.c         | 13 ++++++++-----
>  mm/mmap.c          | 22 +++++++++++++---------
>  mm/mprotect.c      |  4 +++-
>  mm/swap_state.c    |  8 ++++++--
>  9 files changed, 89 insertions(+), 40 deletions(-)
>
>  struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
>                                 struct vm_fault *vmf)
> @@ -665,9 +669,9 @@ static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
>                                      unsigned long *start,
>                                      unsigned long *end)
>  {
> -       *start = max3(lpfn, PFN_DOWN(vma->vm_start),
> +       *start = max3(lpfn, PFN_DOWN(READ_ONCE(vma->vm_start)),
>                       PFN_DOWN(faddr & PMD_MASK));
> -       *end = min3(rpfn, PFN_DOWN(vma->vm_end),
> +       *end = min3(rpfn, PFN_DOWN(READ_ONCE(vma->vm_end)),
>                     PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
>  }
>
> --
> 2.7.4
>

I have got a crash on 4.14 kernel with speculative page faults enabled
and here is my analysis of the problem.
The issue was reported only once.

[23409.303395]  el1_da+0x24/0x84
[23409.303400]  __radix_tree_lookup+0x8/0x90
[23409.303407]  find_get_entry+0x64/0x14c
[23409.303410]  pagecache_get_page+0x5c/0x27c
[23409.303416]  __read_swap_cache_async+0x80/0x260
[23409.303420]  swap_vma_readahead+0x264/0x37c
[23409.303423]  swapin_readahead+0x5c/0x6c
[23409.303428]  do_swap_page+0x128/0x6e4
[23409.303431]  handle_pte_fault+0x230/0xca4
[23409.303435]  __handle_speculative_fault+0x57c/0x7c8
[23409.303438]  do_page_fault+0x228/0x3e8
[23409.303442]  do_translation_fault+0x50/0x6c
[23409.303445]  do_mem_abort+0x5c/0xe0
[23409.303447]  el0_da+0x20/0x24

Process A accesses address ADDR (part of VMA A) and that results in a
translation fault.
Kernel enters __handle_speculative_fault to fix the fault.
Process A enters do_swap_page->swapin_readahead->swap_vma_readahead
from speculative path.
During this time, another process B which shares the same mm, does a
mprotect from another CPU which follows
mprotect_fixup->__split_vma, and it splits VMA A into VMAs A and B.
After the split, ADDR falls into VMA B, but process A is still using
VMA A.
Now ADDR is greater than VMA_A->vm_start and VMA_A->vm_end.
swap_vma_readahead->swap_ra_info uses start and end of vma to
calculate ptes and nr_pte, which goes wrong due to this and finally
resulting in wrong "entry" passed to
swap_vma_readahead->__read_swap_cache_async, and in turn causing
invalid swapper_space
being passed to __read_swap_cache_async->find_get_page, causing an abort.

The fix I have tried is to cache vm_start and vm_end also in vmf and
use it in swap_ra_clamp_pfn. Let me know your thoughts on this. I can
send
the patch I am a using if you feel that is the right thing to do.

Thanks,
Vinayak

Re: [PATCH v11 00/26] Speculative page faults

[Balbir Singh]

On Thu, May 17, 2018 at 01:06:07PM +0200, Laurent Dufour wrote:
> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
> page fault without holding the mm semaphore [1].
> 
> The idea is to try to handle user space page faults without holding the
> mmap_sem. This should allow better concurrency for massively threaded

Question -- I presume mmap_sem (rw_semaphore implementation tested against)
was qrwlock?

Balbir Singh.

Re: [PATCH v11 00/26] Speculative page faults

[Balbir Singh]

On Thu, May 17, 2018 at 01:06:07PM +0200, Laurent Dufour wrote:
> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
> page fault without holding the mm semaphore [1].
> 
> The idea is to try to handle user space page faults without holding the
> mmap_sem. This should allow better concurrency for massively threaded

Question -- I presume mmap_sem (rw_semaphore implementation tested against)
was qrwlock?

Balbir Singh.

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Le 05/11/2018 à 11:42, Balbir Singh a écrit :
> On Thu, May 17, 2018 at 01:06:07PM +0200, Laurent Dufour wrote:
>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>> page fault without holding the mm semaphore [1].
>>
>> The idea is to try to handle user space page faults without holding the
>> mmap_sem. This should allow better concurrency for massively threaded
> 
> Question -- I presume mmap_sem (rw_semaphore implementation tested against)
> was qrwlock?

I don't think so, this series doesn't change the mmap_sem definition so 
it still belongs to the 'struct rw_semaphore'.

Laurent.

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Le 05/11/2018 A  11:42, Balbir Singh a A(C)critA :
> On Thu, May 17, 2018 at 01:06:07PM +0200, Laurent Dufour wrote:
>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>> page fault without holding the mm semaphore [1].
>>
>> The idea is to try to handle user space page faults without holding the
>> mmap_sem. This should allow better concurrency for massively threaded
> 
> Question -- I presume mmap_sem (rw_semaphore implementation tested against)
> was qrwlock?

I don't think so, this series doesn't change the mmap_sem definition so 
it still belongs to the 'struct rw_semaphore'.

Laurent.

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Le 05/11/2018 à 11:42, Balbir Singh a écrit :
> On Thu, May 17, 2018 at 01:06:07PM +0200, Laurent Dufour wrote:
>> This is a port on kernel 4.17 of the work done by Peter Zijlstra to handle
>> page fault without holding the mm semaphore [1].
>>
>> The idea is to try to handle user space page faults without holding the
>> mmap_sem. This should allow better concurrency for massively threaded
> 
> Question -- I presume mmap_sem (rw_semaphore implementation tested against)
> was qrwlock?

I don't think so, this series doesn't change the mmap_sem definition so 
it still belongs to the 'struct rw_semaphore'.

Laurent.

Re: [PATCH v11 10/26] mm: protect VMA modifications using VMA sequence count

[Laurent Dufour]

[-- Attachment #1: Type: text/plain, Size: 4478 bytes --]

Le 05/11/2018 à 08:04, vinayak menon a écrit :
> Hi Laurent,
> 
> On Thu, May 17, 2018 at 4:37 PM Laurent Dufour
> <ldufour@linux.vnet.ibm.com> wrote:
>>
>> The VMA sequence count has been introduced to allow fast detection of
>> VMA modification when running a page fault handler without holding
>> the mmap_sem.
>>
>> This patch provides protection against the VMA modification done in :
>>          - madvise()
>>          - mpol_rebind_policy()
>>          - vma_replace_policy()
>>          - change_prot_numa()
>>          - mlock(), munlock()
>>          - mprotect()
>>          - mmap_region()
>>          - collapse_huge_page()
>>          - userfaultd registering services
>>
>> In addition, VMA fields which will be read during the speculative fault
>> path needs to be written using WRITE_ONCE to prevent write to be split
>> and intermediate values to be pushed to other CPUs.
>>
>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>> ---
>>   fs/proc/task_mmu.c |  5 ++++-
>>   fs/userfaultfd.c   | 17 +++++++++++++----
>>   mm/khugepaged.c    |  3 +++
>>   mm/madvise.c       |  6 +++++-
>>   mm/mempolicy.c     | 51 ++++++++++++++++++++++++++++++++++-----------------
>>   mm/mlock.c         | 13 ++++++++-----
>>   mm/mmap.c          | 22 +++++++++++++---------
>>   mm/mprotect.c      |  4 +++-
>>   mm/swap_state.c    |  8 ++++++--
>>   9 files changed, 89 insertions(+), 40 deletions(-)
>>
>>   struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
>>                                  struct vm_fault *vmf)
>> @@ -665,9 +669,9 @@ static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
>>                                       unsigned long *start,
>>                                       unsigned long *end)
>>   {
>> -       *start = max3(lpfn, PFN_DOWN(vma->vm_start),
>> +       *start = max3(lpfn, PFN_DOWN(READ_ONCE(vma->vm_start)),
>>                        PFN_DOWN(faddr & PMD_MASK));
>> -       *end = min3(rpfn, PFN_DOWN(vma->vm_end),
>> +       *end = min3(rpfn, PFN_DOWN(READ_ONCE(vma->vm_end)),
>>                      PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
>>   }
>>
>> --
>> 2.7.4
>>
> 
> I have got a crash on 4.14 kernel with speculative page faults enabled
> and here is my analysis of the problem.
> The issue was reported only once.

Hi Vinayak,

Thanks for reporting this.

> 
> [23409.303395]  el1_da+0x24/0x84
> [23409.303400]  __radix_tree_lookup+0x8/0x90
> [23409.303407]  find_get_entry+0x64/0x14c
> [23409.303410]  pagecache_get_page+0x5c/0x27c
> [23409.303416]  __read_swap_cache_async+0x80/0x260
> [23409.303420]  swap_vma_readahead+0x264/0x37c
> [23409.303423]  swapin_readahead+0x5c/0x6c
> [23409.303428]  do_swap_page+0x128/0x6e4
> [23409.303431]  handle_pte_fault+0x230/0xca4
> [23409.303435]  __handle_speculative_fault+0x57c/0x7c8
> [23409.303438]  do_page_fault+0x228/0x3e8
> [23409.303442]  do_translation_fault+0x50/0x6c
> [23409.303445]  do_mem_abort+0x5c/0xe0
> [23409.303447]  el0_da+0x20/0x24
> 
> Process A accesses address ADDR (part of VMA A) and that results in a
> translation fault.
> Kernel enters __handle_speculative_fault to fix the fault.
> Process A enters do_swap_page->swapin_readahead->swap_vma_readahead
> from speculative path.
> During this time, another process B which shares the same mm, does a
> mprotect from another CPU which follows
> mprotect_fixup->__split_vma, and it splits VMA A into VMAs A and B.
> After the split, ADDR falls into VMA B, but process A is still using
> VMA A.
> Now ADDR is greater than VMA_A->vm_start and VMA_A->vm_end.
> swap_vma_readahead->swap_ra_info uses start and end of vma to
> calculate ptes and nr_pte, which goes wrong due to this and finally
> resulting in wrong "entry" passed to
> swap_vma_readahead->__read_swap_cache_async, and in turn causing
> invalid swapper_space
> being passed to __read_swap_cache_async->find_get_page, causing an abort.
> 
> The fix I have tried is to cache vm_start and vm_end also in vmf and
> use it in swap_ra_clamp_pfn. Let me know your thoughts on this. I can
> send
> the patch I am a using if you feel that is the right thing to do.

I think the best would be to don't do swap readahead during the 
speculatvive page fault. If the page is found in the swap cache, that's 
fine, but otherwise, we should f	allback to the regular page fault.

The attached -untested- patch is doing this, if you want to give it a 
try. I'll review that for the next series.

Thanks,
Laurent.

[-- Attachment #2: 0001-mm-don-t-do-swap-readahead-during-speculative-page-f.patch --]
[-- Type: text/plain, Size: 1507 bytes --]

From 056afafb0bccea6a356f80f4253ffcd3ef4a1f8d Mon Sep 17 00:00:00 2001
From: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Date: Mon, 5 Nov 2018 18:43:01 +0100
Subject: [PATCH] mm: don't do swap readahead during speculative page fault

Vinayak Menon faced a panic because one thread was page faulting a page in
swap, while another one was mprotecting a part of the VMA leading to a VMA
split.
This raise a panic in swap_vma_readahead() because the VMA's boundaries
were not more matching the faulting address.

To avoid this, if the page is not found in the swap, the speculative page
fault is aborted to retry a regular page fault.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 mm/memory.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/mm/memory.c b/mm/memory.c
index 9dd5ffeb1f7e..720dc9a1b99f 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -3139,6 +3139,16 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 				lru_cache_add_anon(page);
 				swap_readpage(page, true);
 			}
+		} else if (vmf->flags & FAULT_FLAG_SPECULATIVE) {
+			/*
+			 * Don't try readahead during a speculative page fault as
+			 * the VMA's boundaries may change in our back.
+			 * If the page is not in the swap cache and synchronous read
+			 * is disabled, fall back to the regular page fault mechanism.
+			 */
+			delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
+			ret = VM_FAULT_RETRY;
+			goto out;
 		} else {
 			page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
 						vmf);
-- 
2.19.1

Re: [PATCH v11 10/26] mm: protect VMA modifications using VMA sequence count

[Laurent Dufour]

[-- Attachment #1: Type: text/plain, Size: 4480 bytes --]

Le 05/11/2018 A  08:04, vinayak menon a A(C)critA :
> Hi Laurent,
> 
> On Thu, May 17, 2018 at 4:37 PM Laurent Dufour
> <ldufour@linux.vnet.ibm.com> wrote:
>>
>> The VMA sequence count has been introduced to allow fast detection of
>> VMA modification when running a page fault handler without holding
>> the mmap_sem.
>>
>> This patch provides protection against the VMA modification done in :
>>          - madvise()
>>          - mpol_rebind_policy()
>>          - vma_replace_policy()
>>          - change_prot_numa()
>>          - mlock(), munlock()
>>          - mprotect()
>>          - mmap_region()
>>          - collapse_huge_page()
>>          - userfaultd registering services
>>
>> In addition, VMA fields which will be read during the speculative fault
>> path needs to be written using WRITE_ONCE to prevent write to be split
>> and intermediate values to be pushed to other CPUs.
>>
>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>> ---
>>   fs/proc/task_mmu.c |  5 ++++-
>>   fs/userfaultfd.c   | 17 +++++++++++++----
>>   mm/khugepaged.c    |  3 +++
>>   mm/madvise.c       |  6 +++++-
>>   mm/mempolicy.c     | 51 ++++++++++++++++++++++++++++++++++-----------------
>>   mm/mlock.c         | 13 ++++++++-----
>>   mm/mmap.c          | 22 +++++++++++++---------
>>   mm/mprotect.c      |  4 +++-
>>   mm/swap_state.c    |  8 ++++++--
>>   9 files changed, 89 insertions(+), 40 deletions(-)
>>
>>   struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
>>                                  struct vm_fault *vmf)
>> @@ -665,9 +669,9 @@ static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
>>                                       unsigned long *start,
>>                                       unsigned long *end)
>>   {
>> -       *start = max3(lpfn, PFN_DOWN(vma->vm_start),
>> +       *start = max3(lpfn, PFN_DOWN(READ_ONCE(vma->vm_start)),
>>                        PFN_DOWN(faddr & PMD_MASK));
>> -       *end = min3(rpfn, PFN_DOWN(vma->vm_end),
>> +       *end = min3(rpfn, PFN_DOWN(READ_ONCE(vma->vm_end)),
>>                      PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
>>   }
>>
>> --
>> 2.7.4
>>
> 
> I have got a crash on 4.14 kernel with speculative page faults enabled
> and here is my analysis of the problem.
> The issue was reported only once.

Hi Vinayak,

Thanks for reporting this.

> 
> [23409.303395]  el1_da+0x24/0x84
> [23409.303400]  __radix_tree_lookup+0x8/0x90
> [23409.303407]  find_get_entry+0x64/0x14c
> [23409.303410]  pagecache_get_page+0x5c/0x27c
> [23409.303416]  __read_swap_cache_async+0x80/0x260
> [23409.303420]  swap_vma_readahead+0x264/0x37c
> [23409.303423]  swapin_readahead+0x5c/0x6c
> [23409.303428]  do_swap_page+0x128/0x6e4
> [23409.303431]  handle_pte_fault+0x230/0xca4
> [23409.303435]  __handle_speculative_fault+0x57c/0x7c8
> [23409.303438]  do_page_fault+0x228/0x3e8
> [23409.303442]  do_translation_fault+0x50/0x6c
> [23409.303445]  do_mem_abort+0x5c/0xe0
> [23409.303447]  el0_da+0x20/0x24
> 
> Process A accesses address ADDR (part of VMA A) and that results in a
> translation fault.
> Kernel enters __handle_speculative_fault to fix the fault.
> Process A enters do_swap_page->swapin_readahead->swap_vma_readahead
> from speculative path.
> During this time, another process B which shares the same mm, does a
> mprotect from another CPU which follows
> mprotect_fixup->__split_vma, and it splits VMA A into VMAs A and B.
> After the split, ADDR falls into VMA B, but process A is still using
> VMA A.
> Now ADDR is greater than VMA_A->vm_start and VMA_A->vm_end.
> swap_vma_readahead->swap_ra_info uses start and end of vma to
> calculate ptes and nr_pte, which goes wrong due to this and finally
> resulting in wrong "entry" passed to
> swap_vma_readahead->__read_swap_cache_async, and in turn causing
> invalid swapper_space
> being passed to __read_swap_cache_async->find_get_page, causing an abort.
> 
> The fix I have tried is to cache vm_start and vm_end also in vmf and
> use it in swap_ra_clamp_pfn. Let me know your thoughts on this. I can
> send
> the patch I am a using if you feel that is the right thing to do.

I think the best would be to don't do swap readahead during the 
speculatvive page fault. If the page is found in the swap cache, that's 
fine, but otherwise, we should f	allback to the regular page fault.

The attached -untested- patch is doing this, if you want to give it a 
try. I'll review that for the next series.

Thanks,
Laurent.

[-- Attachment #2: 0001-mm-don-t-do-swap-readahead-during-speculative-page-f.patch --]
[-- Type: text/plain, Size: 1507 bytes --]

From 056afafb0bccea6a356f80f4253ffcd3ef4a1f8d Mon Sep 17 00:00:00 2001
From: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Date: Mon, 5 Nov 2018 18:43:01 +0100
Subject: [PATCH] mm: don't do swap readahead during speculative page fault

Vinayak Menon faced a panic because one thread was page faulting a page in
swap, while another one was mprotecting a part of the VMA leading to a VMA
split.
This raise a panic in swap_vma_readahead() because the VMA's boundaries
were not more matching the faulting address.

To avoid this, if the page is not found in the swap, the speculative page
fault is aborted to retry a regular page fault.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 mm/memory.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/mm/memory.c b/mm/memory.c
index 9dd5ffeb1f7e..720dc9a1b99f 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -3139,6 +3139,16 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 				lru_cache_add_anon(page);
 				swap_readpage(page, true);
 			}
+		} else if (vmf->flags & FAULT_FLAG_SPECULATIVE) {
+			/*
+			 * Don't try readahead during a speculative page fault as
+			 * the VMA's boundaries may change in our back.
+			 * If the page is not in the swap cache and synchronous read
+			 * is disabled, fall back to the regular page fault mechanism.
+			 */
+			delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
+			ret = VM_FAULT_RETRY;
+			goto out;
 		} else {
 			page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
 						vmf);
-- 
2.19.1

Re: [PATCH v11 10/26] mm: protect VMA modifications using VMA sequence count

[Laurent Dufour]

[-- Attachment #1: Type: text/plain, Size: 4478 bytes --]

Le 05/11/2018 à 08:04, vinayak menon a écrit :
> Hi Laurent,
> 
> On Thu, May 17, 2018 at 4:37 PM Laurent Dufour
> <ldufour@linux.vnet.ibm.com> wrote:
>>
>> The VMA sequence count has been introduced to allow fast detection of
>> VMA modification when running a page fault handler without holding
>> the mmap_sem.
>>
>> This patch provides protection against the VMA modification done in :
>>          - madvise()
>>          - mpol_rebind_policy()
>>          - vma_replace_policy()
>>          - change_prot_numa()
>>          - mlock(), munlock()
>>          - mprotect()
>>          - mmap_region()
>>          - collapse_huge_page()
>>          - userfaultd registering services
>>
>> In addition, VMA fields which will be read during the speculative fault
>> path needs to be written using WRITE_ONCE to prevent write to be split
>> and intermediate values to be pushed to other CPUs.
>>
>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>> ---
>>   fs/proc/task_mmu.c |  5 ++++-
>>   fs/userfaultfd.c   | 17 +++++++++++++----
>>   mm/khugepaged.c    |  3 +++
>>   mm/madvise.c       |  6 +++++-
>>   mm/mempolicy.c     | 51 ++++++++++++++++++++++++++++++++++-----------------
>>   mm/mlock.c         | 13 ++++++++-----
>>   mm/mmap.c          | 22 +++++++++++++---------
>>   mm/mprotect.c      |  4 +++-
>>   mm/swap_state.c    |  8 ++++++--
>>   9 files changed, 89 insertions(+), 40 deletions(-)
>>
>>   struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
>>                                  struct vm_fault *vmf)
>> @@ -665,9 +669,9 @@ static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
>>                                       unsigned long *start,
>>                                       unsigned long *end)
>>   {
>> -       *start = max3(lpfn, PFN_DOWN(vma->vm_start),
>> +       *start = max3(lpfn, PFN_DOWN(READ_ONCE(vma->vm_start)),
>>                        PFN_DOWN(faddr & PMD_MASK));
>> -       *end = min3(rpfn, PFN_DOWN(vma->vm_end),
>> +       *end = min3(rpfn, PFN_DOWN(READ_ONCE(vma->vm_end)),
>>                      PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
>>   }
>>
>> --
>> 2.7.4
>>
> 
> I have got a crash on 4.14 kernel with speculative page faults enabled
> and here is my analysis of the problem.
> The issue was reported only once.

Hi Vinayak,

Thanks for reporting this.

> 
> [23409.303395]  el1_da+0x24/0x84
> [23409.303400]  __radix_tree_lookup+0x8/0x90
> [23409.303407]  find_get_entry+0x64/0x14c
> [23409.303410]  pagecache_get_page+0x5c/0x27c
> [23409.303416]  __read_swap_cache_async+0x80/0x260
> [23409.303420]  swap_vma_readahead+0x264/0x37c
> [23409.303423]  swapin_readahead+0x5c/0x6c
> [23409.303428]  do_swap_page+0x128/0x6e4
> [23409.303431]  handle_pte_fault+0x230/0xca4
> [23409.303435]  __handle_speculative_fault+0x57c/0x7c8
> [23409.303438]  do_page_fault+0x228/0x3e8
> [23409.303442]  do_translation_fault+0x50/0x6c
> [23409.303445]  do_mem_abort+0x5c/0xe0
> [23409.303447]  el0_da+0x20/0x24
> 
> Process A accesses address ADDR (part of VMA A) and that results in a
> translation fault.
> Kernel enters __handle_speculative_fault to fix the fault.
> Process A enters do_swap_page->swapin_readahead->swap_vma_readahead
> from speculative path.
> During this time, another process B which shares the same mm, does a
> mprotect from another CPU which follows
> mprotect_fixup->__split_vma, and it splits VMA A into VMAs A and B.
> After the split, ADDR falls into VMA B, but process A is still using
> VMA A.
> Now ADDR is greater than VMA_A->vm_start and VMA_A->vm_end.
> swap_vma_readahead->swap_ra_info uses start and end of vma to
> calculate ptes and nr_pte, which goes wrong due to this and finally
> resulting in wrong "entry" passed to
> swap_vma_readahead->__read_swap_cache_async, and in turn causing
> invalid swapper_space
> being passed to __read_swap_cache_async->find_get_page, causing an abort.
> 
> The fix I have tried is to cache vm_start and vm_end also in vmf and
> use it in swap_ra_clamp_pfn. Let me know your thoughts on this. I can
> send
> the patch I am a using if you feel that is the right thing to do.

I think the best would be to don't do swap readahead during the 
speculatvive page fault. If the page is found in the swap cache, that's 
fine, but otherwise, we should f	allback to the regular page fault.

The attached -untested- patch is doing this, if you want to give it a 
try. I'll review that for the next series.

Thanks,
Laurent.

[-- Attachment #2: 0001-mm-don-t-do-swap-readahead-during-speculative-page-f.patch --]
[-- Type: text/plain, Size: 1507 bytes --]

From 056afafb0bccea6a356f80f4253ffcd3ef4a1f8d Mon Sep 17 00:00:00 2001
From: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Date: Mon, 5 Nov 2018 18:43:01 +0100
Subject: [PATCH] mm: don't do swap readahead during speculative page fault

Vinayak Menon faced a panic because one thread was page faulting a page in
swap, while another one was mprotecting a part of the VMA leading to a VMA
split.
This raise a panic in swap_vma_readahead() because the VMA's boundaries
were not more matching the faulting address.

To avoid this, if the page is not found in the swap, the speculative page
fault is aborted to retry a regular page fault.

Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 mm/memory.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

diff --git a/mm/memory.c b/mm/memory.c
index 9dd5ffeb1f7e..720dc9a1b99f 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -3139,6 +3139,16 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 				lru_cache_add_anon(page);
 				swap_readpage(page, true);
 			}
+		} else if (vmf->flags & FAULT_FLAG_SPECULATIVE) {
+			/*
+			 * Don't try readahead during a speculative page fault as
+			 * the VMA's boundaries may change in our back.
+			 * If the page is not in the swap cache and synchronous read
+			 * is disabled, fall back to the regular page fault mechanism.
+			 */
+			delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
+			ret = VM_FAULT_RETRY;
+			goto out;
 		} else {
 			page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE,
 						vmf);
-- 
2.19.1

Re: [PATCH v11 10/26] mm: protect VMA modifications using VMA sequence count

[Vinayak Menon]

On 11/5/2018 11:52 PM, Laurent Dufour wrote:
> Le 05/11/2018 à 08:04, vinayak menon a écrit :
>> Hi Laurent,
>>
>> On Thu, May 17, 2018 at 4:37 PM Laurent Dufour
>> <ldufour@linux.vnet.ibm.com> wrote:
>>>
>>> The VMA sequence count has been introduced to allow fast detection of
>>> VMA modification when running a page fault handler without holding
>>> the mmap_sem.
>>>
>>> This patch provides protection against the VMA modification done in :
>>>          - madvise()
>>>          - mpol_rebind_policy()
>>>          - vma_replace_policy()
>>>          - change_prot_numa()
>>>          - mlock(), munlock()
>>>          - mprotect()
>>>          - mmap_region()
>>>          - collapse_huge_page()
>>>          - userfaultd registering services
>>>
>>> In addition, VMA fields which will be read during the speculative fault
>>> path needs to be written using WRITE_ONCE to prevent write to be split
>>> and intermediate values to be pushed to other CPUs.
>>>
>>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>>> ---
>>>   fs/proc/task_mmu.c |  5 ++++-
>>>   fs/userfaultfd.c   | 17 +++++++++++++----
>>>   mm/khugepaged.c    |  3 +++
>>>   mm/madvise.c       |  6 +++++-
>>>   mm/mempolicy.c     | 51 ++++++++++++++++++++++++++++++++++-----------------
>>>   mm/mlock.c         | 13 ++++++++-----
>>>   mm/mmap.c          | 22 +++++++++++++---------
>>>   mm/mprotect.c      |  4 +++-
>>>   mm/swap_state.c    |  8 ++++++--
>>>   9 files changed, 89 insertions(+), 40 deletions(-)
>>>
>>>   struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
>>>                                  struct vm_fault *vmf)
>>> @@ -665,9 +669,9 @@ static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
>>>                                       unsigned long *start,
>>>                                       unsigned long *end)
>>>   {
>>> -       *start = max3(lpfn, PFN_DOWN(vma->vm_start),
>>> +       *start = max3(lpfn, PFN_DOWN(READ_ONCE(vma->vm_start)),
>>>                        PFN_DOWN(faddr & PMD_MASK));
>>> -       *end = min3(rpfn, PFN_DOWN(vma->vm_end),
>>> +       *end = min3(rpfn, PFN_DOWN(READ_ONCE(vma->vm_end)),
>>>                      PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
>>>   }
>>>
>>> -- 
>>> 2.7.4
>>>
>>
>> I have got a crash on 4.14 kernel with speculative page faults enabled
>> and here is my analysis of the problem.
>> The issue was reported only once.
>
> Hi Vinayak,
>
> Thanks for reporting this.
>
>>
>> [23409.303395]  el1_da+0x24/0x84
>> [23409.303400]  __radix_tree_lookup+0x8/0x90
>> [23409.303407]  find_get_entry+0x64/0x14c
>> [23409.303410]  pagecache_get_page+0x5c/0x27c
>> [23409.303416]  __read_swap_cache_async+0x80/0x260
>> [23409.303420]  swap_vma_readahead+0x264/0x37c
>> [23409.303423]  swapin_readahead+0x5c/0x6c
>> [23409.303428]  do_swap_page+0x128/0x6e4
>> [23409.303431]  handle_pte_fault+0x230/0xca4
>> [23409.303435]  __handle_speculative_fault+0x57c/0x7c8
>> [23409.303438]  do_page_fault+0x228/0x3e8
>> [23409.303442]  do_translation_fault+0x50/0x6c
>> [23409.303445]  do_mem_abort+0x5c/0xe0
>> [23409.303447]  el0_da+0x20/0x24
>>
>> Process A accesses address ADDR (part of VMA A) and that results in a
>> translation fault.
>> Kernel enters __handle_speculative_fault to fix the fault.
>> Process A enters do_swap_page->swapin_readahead->swap_vma_readahead
>> from speculative path.
>> During this time, another process B which shares the same mm, does a
>> mprotect from another CPU which follows
>> mprotect_fixup->__split_vma, and it splits VMA A into VMAs A and B.
>> After the split, ADDR falls into VMA B, but process A is still using
>> VMA A.
>> Now ADDR is greater than VMA_A->vm_start and VMA_A->vm_end.
>> swap_vma_readahead->swap_ra_info uses start and end of vma to
>> calculate ptes and nr_pte, which goes wrong due to this and finally
>> resulting in wrong "entry" passed to
>> swap_vma_readahead->__read_swap_cache_async, and in turn causing
>> invalid swapper_space
>> being passed to __read_swap_cache_async->find_get_page, causing an abort.
>>
>> The fix I have tried is to cache vm_start and vm_end also in vmf and
>> use it in swap_ra_clamp_pfn. Let me know your thoughts on this. I can
>> send
>> the patch I am a using if you feel that is the right thing to do.
>
> I think the best would be to don't do swap readahead during the speculatvive page fault. If the page is found in the swap cache, that's fine, but otherwise, we should f    allback to the regular page fault.
>
> The attached -untested- patch is doing this, if you want to give it a try. I'll review that for the next series.
>

Thanks Laurent. I and going to try this patch.

With this patch, since all non-SWP_SYNCHRONOUS_IO swapins result in non-speculative fault
and a retry, wouldn't this have an impact on some perf numbers ? If so, would caching start
and end be a better option ?

Also, would it make sense to move the FAULT_FLAG_SPECULATIVE check inside swapin_readahead,
in a way that  swap_cluster_readahead can take the speculative path ? swap_cluster_readahead
doesn't seem to use vma values.

Thanks,
Vinayak

Re: [PATCH v11 10/26] mm: protect VMA modifications using VMA sequence count

[Vinayak Menon]

On 11/5/2018 11:52 PM, Laurent Dufour wrote:
> Le 05/11/2018 A  08:04, vinayak menon a A(C)critA :
>> Hi Laurent,
>>
>> On Thu, May 17, 2018 at 4:37 PM Laurent Dufour
>> <ldufour@linux.vnet.ibm.com> wrote:
>>>
>>> The VMA sequence count has been introduced to allow fast detection of
>>> VMA modification when running a page fault handler without holding
>>> the mmap_sem.
>>>
>>> This patch provides protection against the VMA modification done in :
>>> A A A A A A A A  - madvise()
>>> A A A A A A A A  - mpol_rebind_policy()
>>> A A A A A A A A  - vma_replace_policy()
>>> A A A A A A A A  - change_prot_numa()
>>> A A A A A A A A  - mlock(), munlock()
>>> A A A A A A A A  - mprotect()
>>> A A A A A A A A  - mmap_region()
>>> A A A A A A A A  - collapse_huge_page()
>>> A A A A A A A A  - userfaultd registering services
>>>
>>> In addition, VMA fields which will be read during the speculative fault
>>> path needs to be written using WRITE_ONCE to prevent write to be split
>>> and intermediate values to be pushed to other CPUs.
>>>
>>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>>> ---
>>> A  fs/proc/task_mmu.c |A  5 ++++-
>>> A  fs/userfaultfd.cA A  | 17 +++++++++++++----
>>> A  mm/khugepaged.cA A A  |A  3 +++
>>> A  mm/madvise.cA A A A A A  |A  6 +++++-
>>> A  mm/mempolicy.cA A A A  | 51 ++++++++++++++++++++++++++++++++++-----------------
>>> A  mm/mlock.cA A A A A A A A  | 13 ++++++++-----
>>> A  mm/mmap.cA A A A A A A A A  | 22 +++++++++++++---------
>>> A  mm/mprotect.cA A A A A  |A  4 +++-
>>> A  mm/swap_state.cA A A  |A  8 ++++++--
>>> A  9 files changed, 89 insertions(+), 40 deletions(-)
>>>
>>> A  struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
>>> A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A  struct vm_fault *vmf)
>>> @@ -665,9 +669,9 @@ static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
>>> A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A  unsigned long *start,
>>> A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A A  unsigned long *end)
>>> A  {
>>> -A A A A A A  *start = max3(lpfn, PFN_DOWN(vma->vm_start),
>>> +A A A A A A  *start = max3(lpfn, PFN_DOWN(READ_ONCE(vma->vm_start)),
>>> A A A A A A A A A A A A A A A A A A A A A A  PFN_DOWN(faddr & PMD_MASK));
>>> -A A A A A A  *end = min3(rpfn, PFN_DOWN(vma->vm_end),
>>> +A A A A A A  *end = min3(rpfn, PFN_DOWN(READ_ONCE(vma->vm_end)),
>>> A A A A A A A A A A A A A A A A A A A A  PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
>>> A  }
>>>
>>> -- 
>>> 2.7.4
>>>
>>
>> I have got a crash on 4.14 kernel with speculative page faults enabled
>> and here is my analysis of the problem.
>> The issue was reported only once.
>
> Hi Vinayak,
>
> Thanks for reporting this.
>
>>
>> [23409.303395]A  el1_da+0x24/0x84
>> [23409.303400]A  __radix_tree_lookup+0x8/0x90
>> [23409.303407]A  find_get_entry+0x64/0x14c
>> [23409.303410]A  pagecache_get_page+0x5c/0x27c
>> [23409.303416]A  __read_swap_cache_async+0x80/0x260
>> [23409.303420]A  swap_vma_readahead+0x264/0x37c
>> [23409.303423]A  swapin_readahead+0x5c/0x6c
>> [23409.303428]A  do_swap_page+0x128/0x6e4
>> [23409.303431]A  handle_pte_fault+0x230/0xca4
>> [23409.303435]A  __handle_speculative_fault+0x57c/0x7c8
>> [23409.303438]A  do_page_fault+0x228/0x3e8
>> [23409.303442]A  do_translation_fault+0x50/0x6c
>> [23409.303445]A  do_mem_abort+0x5c/0xe0
>> [23409.303447]A  el0_da+0x20/0x24
>>
>> Process A accesses address ADDR (part of VMA A) and that results in a
>> translation fault.
>> Kernel enters __handle_speculative_fault to fix the fault.
>> Process A enters do_swap_page->swapin_readahead->swap_vma_readahead
>> from speculative path.
>> During this time, another process B which shares the same mm, does a
>> mprotect from another CPU which follows
>> mprotect_fixup->__split_vma, and it splits VMA A into VMAs A and B.
>> After the split, ADDR falls into VMA B, but process A is still using
>> VMA A.
>> Now ADDR is greater than VMA_A->vm_start and VMA_A->vm_end.
>> swap_vma_readahead->swap_ra_info uses start and end of vma to
>> calculate ptes and nr_pte, which goes wrong due to this and finally
>> resulting in wrong "entry" passed to
>> swap_vma_readahead->__read_swap_cache_async, and in turn causing
>> invalid swapper_space
>> being passed to __read_swap_cache_async->find_get_page, causing an abort.
>>
>> The fix I have tried is to cache vm_start and vm_end also in vmf and
>> use it in swap_ra_clamp_pfn. Let me know your thoughts on this. I can
>> send
>> the patch I am a using if you feel that is the right thing to do.
>
> I think the best would be to don't do swap readahead during the speculatvive page fault. If the page is found in the swap cache, that's fine, but otherwise, we should fA A A  allback to the regular page fault.
>
> The attached -untested- patch is doing this, if you want to give it a try. I'll review that for the next series.
>

Thanks Laurent. I and going to try this patch.

With this patch, since all non-SWP_SYNCHRONOUS_IO swapins result in non-speculative fault
and a retry, wouldn't this have an impact on some perf numbers ? If so, would caching start
and end be a better option ?

Also, would it make sense to move the FAULT_FLAG_SPECULATIVE check inside swapin_readahead,
in a way thatA  swap_cluster_readahead can take the speculative path ? swap_cluster_readahead
doesn't seem to use vma values.

Thanks,
Vinayak

Re: [PATCH v11 10/26] mm: protect VMA modifications using VMA sequence count

[Vinayak Menon]

On 11/5/2018 11:52 PM, Laurent Dufour wrote:
> Le 05/11/2018 à 08:04, vinayak menon a écrit :
>> Hi Laurent,
>>
>> On Thu, May 17, 2018 at 4:37 PM Laurent Dufour
>> <ldufour@linux.vnet.ibm.com> wrote:
>>>
>>> The VMA sequence count has been introduced to allow fast detection of
>>> VMA modification when running a page fault handler without holding
>>> the mmap_sem.
>>>
>>> This patch provides protection against the VMA modification done in :
>>>          - madvise()
>>>          - mpol_rebind_policy()
>>>          - vma_replace_policy()
>>>          - change_prot_numa()
>>>          - mlock(), munlock()
>>>          - mprotect()
>>>          - mmap_region()
>>>          - collapse_huge_page()
>>>          - userfaultd registering services
>>>
>>> In addition, VMA fields which will be read during the speculative fault
>>> path needs to be written using WRITE_ONCE to prevent write to be split
>>> and intermediate values to be pushed to other CPUs.
>>>
>>> Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
>>> ---
>>>   fs/proc/task_mmu.c |  5 ++++-
>>>   fs/userfaultfd.c   | 17 +++++++++++++----
>>>   mm/khugepaged.c    |  3 +++
>>>   mm/madvise.c       |  6 +++++-
>>>   mm/mempolicy.c     | 51 ++++++++++++++++++++++++++++++++++-----------------
>>>   mm/mlock.c         | 13 ++++++++-----
>>>   mm/mmap.c          | 22 +++++++++++++---------
>>>   mm/mprotect.c      |  4 +++-
>>>   mm/swap_state.c    |  8 ++++++--
>>>   9 files changed, 89 insertions(+), 40 deletions(-)
>>>
>>>   struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
>>>                                  struct vm_fault *vmf)
>>> @@ -665,9 +669,9 @@ static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
>>>                                       unsigned long *start,
>>>                                       unsigned long *end)
>>>   {
>>> -       *start = max3(lpfn, PFN_DOWN(vma->vm_start),
>>> +       *start = max3(lpfn, PFN_DOWN(READ_ONCE(vma->vm_start)),
>>>                        PFN_DOWN(faddr & PMD_MASK));
>>> -       *end = min3(rpfn, PFN_DOWN(vma->vm_end),
>>> +       *end = min3(rpfn, PFN_DOWN(READ_ONCE(vma->vm_end)),
>>>                      PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
>>>   }
>>>
>>> -- 
>>> 2.7.4
>>>
>>
>> I have got a crash on 4.14 kernel with speculative page faults enabled
>> and here is my analysis of the problem.
>> The issue was reported only once.
>
> Hi Vinayak,
>
> Thanks for reporting this.
>
>>
>> [23409.303395]  el1_da+0x24/0x84
>> [23409.303400]  __radix_tree_lookup+0x8/0x90
>> [23409.303407]  find_get_entry+0x64/0x14c
>> [23409.303410]  pagecache_get_page+0x5c/0x27c
>> [23409.303416]  __read_swap_cache_async+0x80/0x260
>> [23409.303420]  swap_vma_readahead+0x264/0x37c
>> [23409.303423]  swapin_readahead+0x5c/0x6c
>> [23409.303428]  do_swap_page+0x128/0x6e4
>> [23409.303431]  handle_pte_fault+0x230/0xca4
>> [23409.303435]  __handle_speculative_fault+0x57c/0x7c8
>> [23409.303438]  do_page_fault+0x228/0x3e8
>> [23409.303442]  do_translation_fault+0x50/0x6c
>> [23409.303445]  do_mem_abort+0x5c/0xe0
>> [23409.303447]  el0_da+0x20/0x24
>>
>> Process A accesses address ADDR (part of VMA A) and that results in a
>> translation fault.
>> Kernel enters __handle_speculative_fault to fix the fault.
>> Process A enters do_swap_page->swapin_readahead->swap_vma_readahead
>> from speculative path.
>> During this time, another process B which shares the same mm, does a
>> mprotect from another CPU which follows
>> mprotect_fixup->__split_vma, and it splits VMA A into VMAs A and B.
>> After the split, ADDR falls into VMA B, but process A is still using
>> VMA A.
>> Now ADDR is greater than VMA_A->vm_start and VMA_A->vm_end.
>> swap_vma_readahead->swap_ra_info uses start and end of vma to
>> calculate ptes and nr_pte, which goes wrong due to this and finally
>> resulting in wrong "entry" passed to
>> swap_vma_readahead->__read_swap_cache_async, and in turn causing
>> invalid swapper_space
>> being passed to __read_swap_cache_async->find_get_page, causing an abort.
>>
>> The fix I have tried is to cache vm_start and vm_end also in vmf and
>> use it in swap_ra_clamp_pfn. Let me know your thoughts on this. I can
>> send
>> the patch I am a using if you feel that is the right thing to do.
>
> I think the best would be to don't do swap readahead during the speculatvive page fault. If the page is found in the swap cache, that's fine, but otherwise, we should f    allback to the regular page fault.
>
> The attached -untested- patch is doing this, if you want to give it a try. I'll review that for the next series.
>

Thanks Laurent. I and going to try this patch.

With this patch, since all non-SWP_SYNCHRONOUS_IO swapins result in non-speculative fault
and a retry, wouldn't this have an impact on some perf numbers ? If so, would caching start
and end be a better option ?

Also, would it make sense to move the FAULT_FLAG_SPECULATIVE check inside swapin_readahead,
in a way that  swap_cluster_readahead can take the speculative path ? swap_cluster_readahead
doesn't seem to use vma values.

Thanks,
Vinayak

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/28 23:45, Laurent Dufour wrote:
> Le 28/01/2019 à 15:09, zhong jiang a écrit :
>> On 2019/1/28 16:59, Laurent Dufour wrote:
>>> Le 25/01/2019 à 13:32, zhong jiang a écrit :
>>>> On 2019/1/24 16:20, Laurent Dufour wrote:
>>>>> Le 22/01/2019 à 17:22, zhong jiang a écrit :
>>>>>> On 2019/1/19 0:24, Laurent Dufour wrote:
>>>>>>> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>>>>>>>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>>>>>>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>>>>>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>>>>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>>>>>>>> Hi Laurent,
>>>>>>>>>>>>
>>>>>>>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>>>>>>>
>>>>>>>>>>> With the patch below, we don't hit the issue.
>>>>>>>>>>>
>>>>>>>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>>>>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>>>>>>>
>>>>>>>>>>> It is observed that the following scenario results in
>>>>>>>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>>>>>>>> forever after few iterations.
>>>>>>>>>>>
>>>>>>>>>>> CPU 1                   CPU 2                    CPU 3
>>>>>>>>>>> Process 1,              Process 1,               Process 1,
>>>>>>>>>>> Thread A                Thread B                 Thread C
>>>>>>>>>>>
>>>>>>>>>>> while (1) {             while (1) {              while(1) {
>>>>>>>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>>>>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>>>>>>>> }                       }
>>>>>>>>>>>
>>>>>>>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>>>>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>>>>>>>> on one of the CPUs at least. This can create a problem if one
>>>>>>>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>>>>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>>>>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>>>>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>>>>>>>> entry can still modify old_page even after it is copied to
>>>>>>>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>>>>>>>> Nice catch and thanks for your investigation!
>>>>>>>>>>
>>>>>>>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>>>>>>>
>>>>>>>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>>>>>> ---
>>>>>>>>>>>       mm/memory.c | 7 +++++++
>>>>>>>>>>>       1 file changed, 7 insertions(+)
>>>>>>>>>>>
>>>>>>>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>>>>>>>> index 52080e4..1ea168ff 100644
>>>>>>>>>>> --- a/mm/memory.c
>>>>>>>>>>> +++ b/mm/memory.c
>>>>>>>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>>>>>>>                      return VM_FAULT_RETRY;
>>>>>>>>>>>              }
>>>>>>>>>>>
>>>>>>>>>>> +       /*
>>>>>>>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>>>>>>>> +        * in copy_one_pte
>>>>>>>>>>> +        */
>>>>>>>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>>>>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>>>>>>>> +
>>>>>>>>>>>              mem_cgroup_oom_enable();
>>>>>>>>>>>              ret = handle_pte_fault(&vmf);
>>>>>>>>>>>              mem_cgroup_oom_disable();
>>>>>>>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>>>>>>>
>>>>>>>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>>>>>>>
>>>>>>>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>>>>>>>
>>>>>>>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>>>>>>>> Hi, Vinayak and Laurent
>>>>>>>>
>>>>>>>> I think the below change will impact the performance significantly. Becuase most of process has many
>>>>>>>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>>>>>>>> call the flush_tlb_mm  later.
>>>>>>>>
>>>>>>>> I think we can try the following way to do.
>>>>>>>>
>>>>>>>> vm_write_begin(vma)
>>>>>>>> copy_pte_range
>>>>>>>> vm_write_end(vma)
>>>>>>>>
>>>>>>>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>>>>>>>> Any thought?
>>>>>>>
>>>>>>> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>>>>>>>
>>>>>>> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>>>>>>>
>>>>>>> Vinayak, Zhong, could you please give it a try ?
>>>>>>>
>>>>>> Hi Laurent
>>>>>>
>>>>>> I apply the patch you had attached and none of any abnormal thing came in two days. It is feasible to fix the issue.
>>>>>
>>>>> Good news !
>>>>>
>>>>>>
>>>>>> but It will better to filter the condition by is_cow_mapping. is it right?
>>>>>>
>>>>>> for example:
>>>>>>
>>>>>> if (is_cow_mapping(mnpt->vm_flags)) {
>>>>>>               ........
>>>>>> }
>>>>>
>>>>> That's doable for sure but I don't think this has to be introduce in dup_mmap().
>>>>> Unless there is a real performance benefit to do so, I don't think dup_mmap() has to mimic underlying checks done in copy_page_range().
>>>>>
>>>>
>>>> Hi, Laurent
>>>>
>>>> I test the performace with microbench after appling the patch. I find
>>>> the page fault latency will increase about 8% than before.  I think we
>>>> should use is_cow_mapping to waken the impact and I will try it out.
>>>
>>> That's interesting,  I would not expect such a higher latency assuming that most of the area not in copied on write are also not managed by the speculative page fault handler (file mapping, etc.). Anyway I'm looking forward to see the result with additional is_cow_mapping() check.
>>>
>> I test the performance again. It is the protect error access latency in lat_sig.c that it will result in a drop of 8% in that testcase.
>
> What is that "protect error access latency in lat_sig.c" ?
>
It is the protect error access that the source code is lat_sig.c in microbench.
>> The page fault latency, In fact, does not impact the performace. It seems to just the fluctuation.
>>
>> Thanks,
>> zhong jiang
>>>> or we can use the following solution to replace as Vinayak has said.
>>>>
>>>> if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>       return VM_FAULT_RETRY;
>>>>
>>>> Even though it will influence the performance of SPF, but at least it does
>>>> not bring in any negative impact. Any thought?
>>>
>>> I don't agree, this checks will completely by pass the SPF handler for all the COW areas, even if there is no race situation.
>>>
>>> Cheers,
>>> Laurent.
>>>>
>>>> Thanks,
>>>>
>>>>
>>>>> Cheers,
>>>>> Laurent.
>>>>>
>>>>>
>>>>> .
>>>>>
>>>>
>>>>
>>>
>>>
>>>
>>
>>
>
>
> .
>

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Le 28/01/2019 à 15:09, zhong jiang a écrit :
> On 2019/1/28 16:59, Laurent Dufour wrote:
>> Le 25/01/2019 à 13:32, zhong jiang a écrit :
>>> On 2019/1/24 16:20, Laurent Dufour wrote:
>>>> Le 22/01/2019 à 17:22, zhong jiang a écrit :
>>>>> On 2019/1/19 0:24, Laurent Dufour wrote:
>>>>>> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>>>>>>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>>>>>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>>>>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>>>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>>>>>>> Hi Laurent,
>>>>>>>>>>>
>>>>>>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>>>>>>
>>>>>>>>>> With the patch below, we don't hit the issue.
>>>>>>>>>>
>>>>>>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>>>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>>>>>>
>>>>>>>>>> It is observed that the following scenario results in
>>>>>>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>>>>>>> forever after few iterations.
>>>>>>>>>>
>>>>>>>>>> CPU 1                   CPU 2                    CPU 3
>>>>>>>>>> Process 1,              Process 1,               Process 1,
>>>>>>>>>> Thread A                Thread B                 Thread C
>>>>>>>>>>
>>>>>>>>>> while (1) {             while (1) {              while(1) {
>>>>>>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>>>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>>>>>>> }                       }
>>>>>>>>>>
>>>>>>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>>>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>>>>>>> on one of the CPUs at least. This can create a problem if one
>>>>>>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>>>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>>>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>>>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>>>>>>> entry can still modify old_page even after it is copied to
>>>>>>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>>>>>>> Nice catch and thanks for your investigation!
>>>>>>>>>
>>>>>>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>>>>>>
>>>>>>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>>>>> ---
>>>>>>>>>>       mm/memory.c | 7 +++++++
>>>>>>>>>>       1 file changed, 7 insertions(+)
>>>>>>>>>>
>>>>>>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>>>>>>> index 52080e4..1ea168ff 100644
>>>>>>>>>> --- a/mm/memory.c
>>>>>>>>>> +++ b/mm/memory.c
>>>>>>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>>>>>>                      return VM_FAULT_RETRY;
>>>>>>>>>>              }
>>>>>>>>>>
>>>>>>>>>> +       /*
>>>>>>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>>>>>>> +        * in copy_one_pte
>>>>>>>>>> +        */
>>>>>>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>>>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>>>>>>> +
>>>>>>>>>>              mem_cgroup_oom_enable();
>>>>>>>>>>              ret = handle_pte_fault(&vmf);
>>>>>>>>>>              mem_cgroup_oom_disable();
>>>>>>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>>>>>>
>>>>>>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>>>>>>
>>>>>>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>>>>>>
>>>>>>>>
>>>>>>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>>>>>>
>>>>>>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>>>>>>> Hi, Vinayak and Laurent
>>>>>>>
>>>>>>> I think the below change will impact the performance significantly. Becuase most of process has many
>>>>>>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>>>>>>> call the flush_tlb_mm  later.
>>>>>>>
>>>>>>> I think we can try the following way to do.
>>>>>>>
>>>>>>> vm_write_begin(vma)
>>>>>>> copy_pte_range
>>>>>>> vm_write_end(vma)
>>>>>>>
>>>>>>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>>>>>>> Any thought?
>>>>>>
>>>>>> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>>>>>>
>>>>>> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>>>>>>
>>>>>> Vinayak, Zhong, could you please give it a try ?
>>>>>>
>>>>> Hi Laurent
>>>>>
>>>>> I apply the patch you had attached and none of any abnormal thing came in two days. It is feasible to fix the issue.
>>>>
>>>> Good news !
>>>>
>>>>>
>>>>> but It will better to filter the condition by is_cow_mapping. is it right?
>>>>>
>>>>> for example:
>>>>>
>>>>> if (is_cow_mapping(mnpt->vm_flags)) {
>>>>>               ........
>>>>> }
>>>>
>>>> That's doable for sure but I don't think this has to be introduce in dup_mmap().
>>>> Unless there is a real performance benefit to do so, I don't think dup_mmap() has to mimic underlying checks done in copy_page_range().
>>>>
>>>
>>> Hi, Laurent
>>>
>>> I test the performace with microbench after appling the patch. I find
>>> the page fault latency will increase about 8% than before.  I think we
>>> should use is_cow_mapping to waken the impact and I will try it out.
>>
>> That's interesting,  I would not expect such a higher latency assuming that most of the area not in copied on write are also not managed by the speculative page fault handler (file mapping, etc.). Anyway I'm looking forward to see the result with additional is_cow_mapping() check.
>>
> I test the performance again. It is the protect error access latency in lat_sig.c that it will result in a drop of 8% in that testcase.

What is that "protect error access latency in lat_sig.c" ?

> The page fault latency, In fact, does not impact the performace. It seems to just the fluctuation.
> 
> Thanks,
> zhong jiang
>>> or we can use the following solution to replace as Vinayak has said.
>>>
>>> if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>       return VM_FAULT_RETRY;
>>>
>>> Even though it will influence the performance of SPF, but at least it does
>>> not bring in any negative impact. Any thought?
>>
>> I don't agree, this checks will completely by pass the SPF handler for all the COW areas, even if there is no race situation.
>>
>> Cheers,
>> Laurent.
>>>
>>> Thanks,
>>>
>>>
>>>> Cheers,
>>>> Laurent.
>>>>
>>>>
>>>> .
>>>>
>>>
>>>
>>
>>
>>
> 
> 

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/28 16:59, Laurent Dufour wrote:
> Le 25/01/2019 à 13:32, zhong jiang a écrit :
>> On 2019/1/24 16:20, Laurent Dufour wrote:
>>> Le 22/01/2019 à 17:22, zhong jiang a écrit :
>>>> On 2019/1/19 0:24, Laurent Dufour wrote:
>>>>> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>>>>>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>>>>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>>>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>>>>>> Hi Laurent,
>>>>>>>>>>
>>>>>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>>>>>
>>>>>>>>> With the patch below, we don't hit the issue.
>>>>>>>>>
>>>>>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>>>>>
>>>>>>>>> It is observed that the following scenario results in
>>>>>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>>>>>> forever after few iterations.
>>>>>>>>>
>>>>>>>>> CPU 1                   CPU 2                    CPU 3
>>>>>>>>> Process 1,              Process 1,               Process 1,
>>>>>>>>> Thread A                Thread B                 Thread C
>>>>>>>>>
>>>>>>>>> while (1) {             while (1) {              while(1) {
>>>>>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>>>>>> }                       }
>>>>>>>>>
>>>>>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>>>>>> on one of the CPUs at least. This can create a problem if one
>>>>>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>>>>>> entry can still modify old_page even after it is copied to
>>>>>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>>>>>> Nice catch and thanks for your investigation!
>>>>>>>>
>>>>>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>>>>>
>>>>>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>>>> ---
>>>>>>>>>      mm/memory.c | 7 +++++++
>>>>>>>>>      1 file changed, 7 insertions(+)
>>>>>>>>>
>>>>>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>>>>>> index 52080e4..1ea168ff 100644
>>>>>>>>> --- a/mm/memory.c
>>>>>>>>> +++ b/mm/memory.c
>>>>>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>>>>>                     return VM_FAULT_RETRY;
>>>>>>>>>             }
>>>>>>>>>
>>>>>>>>> +       /*
>>>>>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>>>>>> +        * in copy_one_pte
>>>>>>>>> +        */
>>>>>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>>>>>> +
>>>>>>>>>             mem_cgroup_oom_enable();
>>>>>>>>>             ret = handle_pte_fault(&vmf);
>>>>>>>>>             mem_cgroup_oom_disable();
>>>>>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>>>>>
>>>>>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>>>>>
>>>>>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>>>>>
>>>>>>>
>>>>>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>>>>>
>>>>>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>>>>>> Hi, Vinayak and Laurent
>>>>>>
>>>>>> I think the below change will impact the performance significantly. Becuase most of process has many
>>>>>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>>>>>> call the flush_tlb_mm  later.
>>>>>>
>>>>>> I think we can try the following way to do.
>>>>>>
>>>>>> vm_write_begin(vma)
>>>>>> copy_pte_range
>>>>>> vm_write_end(vma)
>>>>>>
>>>>>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>>>>>> Any thought?
>>>>>
>>>>> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>>>>>
>>>>> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>>>>>
>>>>> Vinayak, Zhong, could you please give it a try ?
>>>>>
>>>> Hi Laurent
>>>>
>>>> I apply the patch you had attached and none of any abnormal thing came in two days. It is feasible to fix the issue.
>>>
>>> Good news !
>>>
>>>>
>>>> but It will better to filter the condition by is_cow_mapping. is it right?
>>>>
>>>> for example:
>>>>
>>>> if (is_cow_mapping(mnpt->vm_flags)) {
>>>>              ........
>>>> }
>>>
>>> That's doable for sure but I don't think this has to be introduce in dup_mmap().
>>> Unless there is a real performance benefit to do so, I don't think dup_mmap() has to mimic underlying checks done in copy_page_range().
>>>
>>
>> Hi, Laurent
>>
>> I test the performace with microbench after appling the patch. I find
>> the page fault latency will increase about 8% than before.  I think we
>> should use is_cow_mapping to waken the impact and I will try it out.
>
> That's interesting,  I would not expect such a higher latency assuming that most of the area not in copied on write are also not managed by the speculative page fault handler (file mapping, etc.). Anyway I'm looking forward to see the result with additional is_cow_mapping() check.
>
I test the performance again. It is the protect error access latency in lat_sig.c that it will result in a drop of 8% in that testcase.
The page fault latency, In fact, does not impact the performace. It seems to just the fluctuation.

Thanks,
zhong jiang
>> or we can use the following solution to replace as Vinayak has said.
>>
>> if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>      return VM_FAULT_RETRY;
>>
>> Even though it will influence the performance of SPF, but at least it does
>> not bring in any negative impact. Any thought?
>
> I don't agree, this checks will completely by pass the SPF handler for all the COW areas, even if there is no race situation.
>
> Cheers,
> Laurent.
>>
>> Thanks,
>>
>>
>>> Cheers,
>>> Laurent.
>>>
>>>
>>> .
>>>
>>
>>
>
>
>

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/28 16:59, Laurent Dufour wrote:
> Le 25/01/2019 à 13:32, zhong jiang a écrit :
>> On 2019/1/24 16:20, Laurent Dufour wrote:
>>> Le 22/01/2019 à 17:22, zhong jiang a écrit :
>>>> On 2019/1/19 0:24, Laurent Dufour wrote:
>>>>> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>>>>>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>>>>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>>>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>>>>>> Hi Laurent,
>>>>>>>>>>
>>>>>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>>>>>
>>>>>>>>> With the patch below, we don't hit the issue.
>>>>>>>>>
>>>>>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>>>>>
>>>>>>>>> It is observed that the following scenario results in
>>>>>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>>>>>> forever after few iterations.
>>>>>>>>>
>>>>>>>>> CPU 1                   CPU 2                    CPU 3
>>>>>>>>> Process 1,              Process 1,               Process 1,
>>>>>>>>> Thread A                Thread B                 Thread C
>>>>>>>>>
>>>>>>>>> while (1) {             while (1) {              while(1) {
>>>>>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>>>>>> }                       }
>>>>>>>>>
>>>>>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>>>>>> on one of the CPUs at least. This can create a problem if one
>>>>>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>>>>>> entry can still modify old_page even after it is copied to
>>>>>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>>>>>> Nice catch and thanks for your investigation!
>>>>>>>>
>>>>>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>>>>>
>>>>>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>>>> ---
>>>>>>>>>      mm/memory.c | 7 +++++++
>>>>>>>>>      1 file changed, 7 insertions(+)
>>>>>>>>>
>>>>>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>>>>>> index 52080e4..1ea168ff 100644
>>>>>>>>> --- a/mm/memory.c
>>>>>>>>> +++ b/mm/memory.c
>>>>>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>>>>>                     return VM_FAULT_RETRY;
>>>>>>>>>             }
>>>>>>>>>
>>>>>>>>> +       /*
>>>>>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>>>>>> +        * in copy_one_pte
>>>>>>>>> +        */
>>>>>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>>>>>> +
>>>>>>>>>             mem_cgroup_oom_enable();
>>>>>>>>>             ret = handle_pte_fault(&vmf);
>>>>>>>>>             mem_cgroup_oom_disable();
>>>>>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>>>>>
>>>>>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>>>>>
>>>>>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>>>>>
>>>>>>>
>>>>>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>>>>>
>>>>>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>>>>>> Hi, Vinayak and Laurent
>>>>>>
>>>>>> I think the below change will impact the performance significantly. Becuase most of process has many
>>>>>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>>>>>> call the flush_tlb_mm  later.
>>>>>>
>>>>>> I think we can try the following way to do.
>>>>>>
>>>>>> vm_write_begin(vma)
>>>>>> copy_pte_range
>>>>>> vm_write_end(vma)
>>>>>>
>>>>>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>>>>>> Any thought?
>>>>>
>>>>> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>>>>>
>>>>> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>>>>>
>>>>> Vinayak, Zhong, could you please give it a try ?
>>>>>
>>>> Hi Laurent
>>>>
>>>> I apply the patch you had attached and none of any abnormal thing came in two days. It is feasible to fix the issue.
>>>
>>> Good news !
>>>
>>>>
>>>> but It will better to filter the condition by is_cow_mapping. is it right?
>>>>
>>>> for example:
>>>>
>>>> if (is_cow_mapping(mnpt->vm_flags)) {
>>>>              ........
>>>> }
>>>
>>> That's doable for sure but I don't think this has to be introduce in dup_mmap().
>>> Unless there is a real performance benefit to do so, I don't think dup_mmap() has to mimic underlying checks done in copy_page_range().
>>>
>>
>> Hi, Laurent
>>
>> I test the performace with microbench after appling the patch. I find
>> the page fault latency will increase about 8% than before.  I think we
>> should use is_cow_mapping to waken the impact and I will try it out.
>
> That's interesting,  I would not expect such a higher latency assuming that most of the area not in copied on write are also not managed by the speculative page fault handler (file mapping, etc.). Anyway I'm looking forward to see the result with additional is_cow_mapping() check.
>
I test the performance again. It is the protect error access latency in lat_sig.c that it will result in a drop of 8% in that testcase.
The page fault latency, In fact, does not impact the performace. It seems to just the fluctuation.

Thanks,
zhong jiang
>> or we can use the following solution to replace as Vinayak has said.
>>
>> if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>      return VM_FAULT_RETRY;
>>
>> Even though it will influence the performance of SPF, but at least it does
>> not bring in any negative impact. Any thought?
>
> I don't agree, this checks will completely by pass the SPF handler for all the COW areas, even if there is no race situation.
>
> Cheers,
> Laurent.
>>
>> Thanks,
>>
>>
>>> Cheers,
>>> Laurent.
>>>
>>>
>>> .
>>>
>>
>>
>
>
>

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Le 25/01/2019 à 13:32, zhong jiang a écrit :
> On 2019/1/24 16:20, Laurent Dufour wrote:
>> Le 22/01/2019 à 17:22, zhong jiang a écrit :
>>> On 2019/1/19 0:24, Laurent Dufour wrote:
>>>> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>>>>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>>>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>>>>> Hi Laurent,
>>>>>>>>>
>>>>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>>>>
>>>>>>>> With the patch below, we don't hit the issue.
>>>>>>>>
>>>>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>>>>
>>>>>>>> It is observed that the following scenario results in
>>>>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>>>>> forever after few iterations.
>>>>>>>>
>>>>>>>> CPU 1                   CPU 2                    CPU 3
>>>>>>>> Process 1,              Process 1,               Process 1,
>>>>>>>> Thread A                Thread B                 Thread C
>>>>>>>>
>>>>>>>> while (1) {             while (1) {              while(1) {
>>>>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>>>>> }                       }
>>>>>>>>
>>>>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>>>>> on one of the CPUs at least. This can create a problem if one
>>>>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>>>>> entry can still modify old_page even after it is copied to
>>>>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>>>>> Nice catch and thanks for your investigation!
>>>>>>>
>>>>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>>>>
>>>>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>>> ---
>>>>>>>>      mm/memory.c | 7 +++++++
>>>>>>>>      1 file changed, 7 insertions(+)
>>>>>>>>
>>>>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>>>>> index 52080e4..1ea168ff 100644
>>>>>>>> --- a/mm/memory.c
>>>>>>>> +++ b/mm/memory.c
>>>>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>>>>                     return VM_FAULT_RETRY;
>>>>>>>>             }
>>>>>>>>
>>>>>>>> +       /*
>>>>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>>>>> +        * in copy_one_pte
>>>>>>>> +        */
>>>>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>>>>> +
>>>>>>>>             mem_cgroup_oom_enable();
>>>>>>>>             ret = handle_pte_fault(&vmf);
>>>>>>>>             mem_cgroup_oom_disable();
>>>>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>>>>
>>>>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>>>>
>>>>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>>>>
>>>>>>
>>>>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>>>>
>>>>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>>>>> Hi, Vinayak and Laurent
>>>>>
>>>>> I think the below change will impact the performance significantly. Becuase most of process has many
>>>>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>>>>> call the flush_tlb_mm  later.
>>>>>
>>>>> I think we can try the following way to do.
>>>>>
>>>>> vm_write_begin(vma)
>>>>> copy_pte_range
>>>>> vm_write_end(vma)
>>>>>
>>>>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>>>>> Any thought?
>>>>
>>>> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>>>>
>>>> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>>>>
>>>> Vinayak, Zhong, could you please give it a try ?
>>>>
>>> Hi Laurent
>>>
>>> I apply the patch you had attached and none of any abnormal thing came in two days. It is feasible to fix the issue.
>>
>> Good news !
>>
>>>
>>> but It will better to filter the condition by is_cow_mapping. is it right?
>>>
>>> for example:
>>>
>>> if (is_cow_mapping(mnpt->vm_flags)) {
>>>              ........
>>> }
>>
>> That's doable for sure but I don't think this has to be introduce in dup_mmap().
>> Unless there is a real performance benefit to do so, I don't think dup_mmap() has to mimic underlying checks done in copy_page_range().
>>
> 
> Hi, Laurent
> 
> I test the performace with microbench after appling the patch. I find
> the page fault latency will increase about 8% than before.  I think we
> should use is_cow_mapping to waken the impact and I will try it out.

That's interesting,  I would not expect such a higher latency assuming 
that most of the area not in copied on write are also not managed by the 
speculative page fault handler (file mapping, etc.). Anyway I'm looking 
forward to see the result with additional is_cow_mapping() check.

> or we can use the following solution to replace as Vinayak has said.
> 
> if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>      return VM_FAULT_RETRY;
> 
> Even though it will influence the performance of SPF, but at least it does
> not bring in any negative impact. Any thought?

I don't agree, this checks will completely by pass the SPF handler for 
all the COW areas, even if there is no race situation.

Cheers,
Laurent.
> 
> Thanks,
> 
> 
>> Cheers,
>> Laurent.
>>
>>
>> .
>>
> 
> 

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/24 16:20, Laurent Dufour wrote:
> Le 22/01/2019 à 17:22, zhong jiang a écrit :
>> On 2019/1/19 0:24, Laurent Dufour wrote:
>>> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>>>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>>>> Hi Laurent,
>>>>>>>>
>>>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>>>
>>>>>>> With the patch below, we don't hit the issue.
>>>>>>>
>>>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>>>
>>>>>>> It is observed that the following scenario results in
>>>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>>>> forever after few iterations.
>>>>>>>
>>>>>>> CPU 1                   CPU 2                    CPU 3
>>>>>>> Process 1,              Process 1,               Process 1,
>>>>>>> Thread A                Thread B                 Thread C
>>>>>>>
>>>>>>> while (1) {             while (1) {              while(1) {
>>>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>>>> }                       }
>>>>>>>
>>>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>>>> on one of the CPUs at least. This can create a problem if one
>>>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>>>> entry can still modify old_page even after it is copied to
>>>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>>>> Nice catch and thanks for your investigation!
>>>>>>
>>>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>>>
>>>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>> ---
>>>>>>>     mm/memory.c | 7 +++++++
>>>>>>>     1 file changed, 7 insertions(+)
>>>>>>>
>>>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>>>> index 52080e4..1ea168ff 100644
>>>>>>> --- a/mm/memory.c
>>>>>>> +++ b/mm/memory.c
>>>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>>>                    return VM_FAULT_RETRY;
>>>>>>>            }
>>>>>>>
>>>>>>> +       /*
>>>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>>>> +        * in copy_one_pte
>>>>>>> +        */
>>>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>>>> +
>>>>>>>            mem_cgroup_oom_enable();
>>>>>>>            ret = handle_pte_fault(&vmf);
>>>>>>>            mem_cgroup_oom_disable();
>>>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>>>
>>>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>>>
>>>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>>>
>>>>>
>>>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>>>
>>>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>>>> Hi, Vinayak and Laurent
>>>>
>>>> I think the below change will impact the performance significantly. Becuase most of process has many
>>>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>>>> call the flush_tlb_mm  later.
>>>>
>>>> I think we can try the following way to do.
>>>>
>>>> vm_write_begin(vma)
>>>> copy_pte_range
>>>> vm_write_end(vma)
>>>>
>>>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>>>> Any thought?
>>>
>>> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>>>
>>> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>>>
>>> Vinayak, Zhong, could you please give it a try ?
>>>
>> Hi Laurent
>>
>> I apply the patch you had attached and none of any abnormal thing came in two days. It is feasible to fix the issue.
>
> Good news !
>
>>
>> but It will better to filter the condition by is_cow_mapping. is it right?
>>
>> for example:
>>
>> if (is_cow_mapping(mnpt->vm_flags)) {
>>             ........
>> }
>
> That's doable for sure but I don't think this has to be introduce in dup_mmap().
> Unless there is a real performance benefit to do so, I don't think dup_mmap() has to mimic underlying checks done in copy_page_range().
>

Hi, Laurent

I test the performace with microbench after appling the patch. I find 
the page fault latency will increase about 8% than before.  I think we
should use is_cow_mapping to waken the impact and I will try it out.

or we can use the following solution to replace as Vinayak has said.
 
if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
    return VM_FAULT_RETRY;

Even though it will influence the performance of SPF, but at least it does
not bring in any negative impact. Any thought?

Thanks,


> Cheers,
> Laurent.
>
>
> .
>

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/24 16:20, Laurent Dufour wrote:
> Le 22/01/2019 à 17:22, zhong jiang a écrit :
>> On 2019/1/19 0:24, Laurent Dufour wrote:
>>> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>>>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>>>> Hi Laurent,
>>>>>>>>
>>>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>>>
>>>>>>> With the patch below, we don't hit the issue.
>>>>>>>
>>>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>>>
>>>>>>> It is observed that the following scenario results in
>>>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>>>> forever after few iterations.
>>>>>>>
>>>>>>> CPU 1                   CPU 2                    CPU 3
>>>>>>> Process 1,              Process 1,               Process 1,
>>>>>>> Thread A                Thread B                 Thread C
>>>>>>>
>>>>>>> while (1) {             while (1) {              while(1) {
>>>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>>>> }                       }
>>>>>>>
>>>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>>>> on one of the CPUs at least. This can create a problem if one
>>>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>>>> entry can still modify old_page even after it is copied to
>>>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>>>> Nice catch and thanks for your investigation!
>>>>>>
>>>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>>>
>>>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>>> ---
>>>>>>>     mm/memory.c | 7 +++++++
>>>>>>>     1 file changed, 7 insertions(+)
>>>>>>>
>>>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>>>> index 52080e4..1ea168ff 100644
>>>>>>> --- a/mm/memory.c
>>>>>>> +++ b/mm/memory.c
>>>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>>>                    return VM_FAULT_RETRY;
>>>>>>>            }
>>>>>>>
>>>>>>> +       /*
>>>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>>>> +        * in copy_one_pte
>>>>>>> +        */
>>>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>>>> +
>>>>>>>            mem_cgroup_oom_enable();
>>>>>>>            ret = handle_pte_fault(&vmf);
>>>>>>>            mem_cgroup_oom_disable();
>>>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>>>
>>>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>>>
>>>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>>>
>>>>>
>>>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>>>
>>>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>>>> Hi, Vinayak and Laurent
>>>>
>>>> I think the below change will impact the performance significantly. Becuase most of process has many
>>>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>>>> call the flush_tlb_mm  later.
>>>>
>>>> I think we can try the following way to do.
>>>>
>>>> vm_write_begin(vma)
>>>> copy_pte_range
>>>> vm_write_end(vma)
>>>>
>>>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>>>> Any thought?
>>>
>>> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>>>
>>> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>>>
>>> Vinayak, Zhong, could you please give it a try ?
>>>
>> Hi Laurent
>>
>> I apply the patch you had attached and none of any abnormal thing came in two days. It is feasible to fix the issue.
>
> Good news !
>
>>
>> but It will better to filter the condition by is_cow_mapping. is it right?
>>
>> for example:
>>
>> if (is_cow_mapping(mnpt->vm_flags)) {
>>             ........
>> }
>
> That's doable for sure but I don't think this has to be introduce in dup_mmap().
> Unless there is a real performance benefit to do so, I don't think dup_mmap() has to mimic underlying checks done in copy_page_range().
>

Hi, Laurent

I test the performace with microbench after appling the patch. I find 
the page fault latency will increase about 8% than before.  I think we
should use is_cow_mapping to waken the impact and I will try it out.

or we can use the following solution to replace as Vinayak has said.
 
if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
    return VM_FAULT_RETRY;

Even though it will influence the performance of SPF, but at least it does
not bring in any negative impact. Any thought?

Thanks,


> Cheers,
> Laurent.
>
>
> .
>

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Le 22/01/2019 à 17:22, zhong jiang a écrit :
> On 2019/1/19 0:24, Laurent Dufour wrote:
>> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>>> Hi Laurent,
>>>>>>>
>>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>>
>>>>>> With the patch below, we don't hit the issue.
>>>>>>
>>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>>
>>>>>> It is observed that the following scenario results in
>>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>>> forever after few iterations.
>>>>>>
>>>>>> CPU 1                   CPU 2                    CPU 3
>>>>>> Process 1,              Process 1,               Process 1,
>>>>>> Thread A                Thread B                 Thread C
>>>>>>
>>>>>> while (1) {             while (1) {              while(1) {
>>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>>> }                       }
>>>>>>
>>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>>> on one of the CPUs at least. This can create a problem if one
>>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>>> entry can still modify old_page even after it is copied to
>>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>>> Nice catch and thanks for your investigation!
>>>>>
>>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>>
>>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>> ---
>>>>>>     mm/memory.c | 7 +++++++
>>>>>>     1 file changed, 7 insertions(+)
>>>>>>
>>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>>> index 52080e4..1ea168ff 100644
>>>>>> --- a/mm/memory.c
>>>>>> +++ b/mm/memory.c
>>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>>                    return VM_FAULT_RETRY;
>>>>>>            }
>>>>>>
>>>>>> +       /*
>>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>>> +        * in copy_one_pte
>>>>>> +        */
>>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>>> +
>>>>>>            mem_cgroup_oom_enable();
>>>>>>            ret = handle_pte_fault(&vmf);
>>>>>>            mem_cgroup_oom_disable();
>>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>>
>>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>>
>>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>>
>>>>
>>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>>
>>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>>> Hi, Vinayak and Laurent
>>>
>>> I think the below change will impact the performance significantly. Becuase most of process has many
>>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>>> call the flush_tlb_mm  later.
>>>
>>> I think we can try the following way to do.
>>>
>>> vm_write_begin(vma)
>>> copy_pte_range
>>> vm_write_end(vma)
>>>
>>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>>> Any thought?
>>
>> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>>
>> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>>
>> Vinayak, Zhong, could you please give it a try ?
>>
> Hi Laurent
> 
> I apply the patch you had attached and none of any abnormal thing came in two days. It is feasible to fix the issue.

Good news !

> 
> but It will better to filter the condition by is_cow_mapping. is it right?
> 
> for example:
> 
> if (is_cow_mapping(mnpt->vm_flags)) {
>        
>      ........
> }

That's doable for sure but I don't think this has to be introduce in 
dup_mmap().
Unless there is a real performance benefit to do so, I don't think 
dup_mmap() has to mimic underlying checks done in copy_page_range().

Cheers,
Laurent.

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/19 0:24, Laurent Dufour wrote:
> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>> Hi Laurent,
>>>>>>
>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>
>>>>> With the patch below, we don't hit the issue.
>>>>>
>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>
>>>>> It is observed that the following scenario results in
>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>> forever after few iterations.
>>>>>
>>>>> CPU 1                   CPU 2                    CPU 3
>>>>> Process 1,              Process 1,               Process 1,
>>>>> Thread A                Thread B                 Thread C
>>>>>
>>>>> while (1) {             while (1) {              while(1) {
>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>> }                       }
>>>>>
>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>> on one of the CPUs at least. This can create a problem if one
>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>> entry can still modify old_page even after it is copied to
>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>> Nice catch and thanks for your investigation!
>>>>
>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>
>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> ---
>>>>>    mm/memory.c | 7 +++++++
>>>>>    1 file changed, 7 insertions(+)
>>>>>
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index 52080e4..1ea168ff 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>                   return VM_FAULT_RETRY;
>>>>>           }
>>>>>
>>>>> +       /*
>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>> +        * in copy_one_pte
>>>>> +        */
>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>> +
>>>>>           mem_cgroup_oom_enable();
>>>>>           ret = handle_pte_fault(&vmf);
>>>>>           mem_cgroup_oom_disable();
>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>
>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>
>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>
>>>
>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>
>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>> Hi, Vinayak and Laurent
>>
>> I think the below change will impact the performance significantly. Becuase most of process has many
>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>> call the flush_tlb_mm  later.
>>
>> I think we can try the following way to do.
>>
>> vm_write_begin(vma)
>> copy_pte_range
>> vm_write_end(vma)
>>
>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>> Any thought?
>
> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>
> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>
> Vinayak, Zhong, could you please give it a try ?
>
Hi Laurent

I apply the patch you had attached and none of any abnormal thing came in two days. It is feasible to fix the issue.

but It will better to filter the condition by is_cow_mapping. is it right?

for example:

if (is_cow_mapping(mnpt->vm_flags)) {
      
    ........
}
   
Thanks,
zhong jiang
> Thanks,
> Laurent.
>

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/19 0:24, Laurent Dufour wrote:
> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>> Hi Laurent,
>>>>>>
>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>
>>>>> With the patch below, we don't hit the issue.
>>>>>
>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>
>>>>> It is observed that the following scenario results in
>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>> forever after few iterations.
>>>>>
>>>>> CPU 1                   CPU 2                    CPU 3
>>>>> Process 1,              Process 1,               Process 1,
>>>>> Thread A                Thread B                 Thread C
>>>>>
>>>>> while (1) {             while (1) {              while(1) {
>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>> }                       }
>>>>>
>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>> on one of the CPUs at least. This can create a problem if one
>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>> entry can still modify old_page even after it is copied to
>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>> Nice catch and thanks for your investigation!
>>>>
>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>
>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> ---
>>>>>    mm/memory.c | 7 +++++++
>>>>>    1 file changed, 7 insertions(+)
>>>>>
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index 52080e4..1ea168ff 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>                   return VM_FAULT_RETRY;
>>>>>           }
>>>>>
>>>>> +       /*
>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>> +        * in copy_one_pte
>>>>> +        */
>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>> +
>>>>>           mem_cgroup_oom_enable();
>>>>>           ret = handle_pte_fault(&vmf);
>>>>>           mem_cgroup_oom_disable();
>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>
>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>
>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>
>>>
>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>
>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>> Hi, Vinayak and Laurent
>>
>> I think the below change will impact the performance significantly. Becuase most of process has many
>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>> call the flush_tlb_mm  later.
>>
>> I think we can try the following way to do.
>>
>> vm_write_begin(vma)
>> copy_pte_range
>> vm_write_end(vma)
>>
>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>> Any thought?
>
> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>
> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>
> Vinayak, Zhong, could you please give it a try ?
>
Hi Laurent

I apply the patch you had attached and none of any abnormal thing came in two days. It is feasible to fix the issue.

but It will better to filter the condition by is_cow_mapping. is it right?

for example:

if (is_cow_mapping(mnpt->vm_flags)) {
      
    ........
}
   
Thanks,
zhong jiang
> Thanks,
> Laurent.
>

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/19 0:24, Laurent Dufour wrote:
> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>> Hi Laurent,
>>>>>>
>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>
>>>>> With the patch below, we don't hit the issue.
>>>>>
>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>
>>>>> It is observed that the following scenario results in
>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>> forever after few iterations.
>>>>>
>>>>> CPU 1                   CPU 2                    CPU 3
>>>>> Process 1,              Process 1,               Process 1,
>>>>> Thread A                Thread B                 Thread C
>>>>>
>>>>> while (1) {             while (1) {              while(1) {
>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>> }                       }
>>>>>
>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>> on one of the CPUs at least. This can create a problem if one
>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>> entry can still modify old_page even after it is copied to
>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>> Nice catch and thanks for your investigation!
>>>>
>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>
>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> ---
>>>>>    mm/memory.c | 7 +++++++
>>>>>    1 file changed, 7 insertions(+)
>>>>>
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index 52080e4..1ea168ff 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>                   return VM_FAULT_RETRY;
>>>>>           }
>>>>>
>>>>> +       /*
>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>> +        * in copy_one_pte
>>>>> +        */
>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>> +
>>>>>           mem_cgroup_oom_enable();
>>>>>           ret = handle_pte_fault(&vmf);
>>>>>           mem_cgroup_oom_disable();
>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>
>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>
>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>
>>>
>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>
>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>> Hi, Vinayak and Laurent
>>
>> I think the below change will impact the performance significantly. Becuase most of process has many
>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>> call the flush_tlb_mm  later.
>>
>> I think we can try the following way to do.
>>
>> vm_write_begin(vma)
>> copy_pte_range
>> vm_write_end(vma)
>>
>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>> Any thought?
>
> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>
> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>
> Vinayak, Zhong, could you please give it a try ?
>
Thanks, look good to me. I will try it.

Sincerely,
zhong jiang
> Thanks,
> Laurent.
>

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/19 0:24, Laurent Dufour wrote:
> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>> Hi Laurent,
>>>>>>
>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>
>>>>> With the patch below, we don't hit the issue.
>>>>>
>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>
>>>>> It is observed that the following scenario results in
>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>> forever after few iterations.
>>>>>
>>>>> CPU 1                   CPU 2                    CPU 3
>>>>> Process 1,              Process 1,               Process 1,
>>>>> Thread A                Thread B                 Thread C
>>>>>
>>>>> while (1) {             while (1) {              while(1) {
>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>> }                       }
>>>>>
>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>> on one of the CPUs at least. This can create a problem if one
>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>> entry can still modify old_page even after it is copied to
>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>> Nice catch and thanks for your investigation!
>>>>
>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>
>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> ---
>>>>>    mm/memory.c | 7 +++++++
>>>>>    1 file changed, 7 insertions(+)
>>>>>
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index 52080e4..1ea168ff 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>                   return VM_FAULT_RETRY;
>>>>>           }
>>>>>
>>>>> +       /*
>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>> +        * in copy_one_pte
>>>>> +        */
>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>> +
>>>>>           mem_cgroup_oom_enable();
>>>>>           ret = handle_pte_fault(&vmf);
>>>>>           mem_cgroup_oom_disable();
>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>
>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>
>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>
>>>
>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>
>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>> Hi, Vinayak and Laurent
>>
>> I think the below change will impact the performance significantly. Becuase most of process has many
>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>> call the flush_tlb_mm  later.
>>
>> I think we can try the following way to do.
>>
>> vm_write_begin(vma)
>> copy_pte_range
>> vm_write_end(vma)
>>
>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>> Any thought?
>
> Here is a new version of the patch fixing this issue. There is no additional TLB flush, all the fix is belonging on vm_write_{begin,end} calls.
>
> I did some test on x86_64 and PowerPC but that needs to be double check on arm64.
>
> Vinayak, Zhong, could you please give it a try ?
>
Thanks, look good to me. I will try it.

Sincerely,
zhong jiang
> Thanks,
> Laurent.
>

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

[-- Attachment #1: Type: text/plain, Size: 4671 bytes --]

Le 17/01/2019 à 16:51, zhong jiang a écrit :
> On 2019/1/16 19:41, Vinayak Menon wrote:
>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>> Hi Laurent,
>>>>>
>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>
>>>> With the patch below, we don't hit the issue.
>>>>
>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>
>>>> It is observed that the following scenario results in
>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>> forever after few iterations.
>>>>
>>>> CPU 1                   CPU 2                    CPU 3
>>>> Process 1,              Process 1,               Process 1,
>>>> Thread A                Thread B                 Thread C
>>>>
>>>> while (1) {             while (1) {              while(1) {
>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>> }                       }
>>>>
>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>> (of lock l), stale tlb entries can exist with write permissions
>>>> on one of the CPUs at least. This can create a problem if one
>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>> entry can still modify old_page even after it is copied to
>>>> new_page by wp_page_copy, thus causing a corruption.
>>> Nice catch and thanks for your investigation!
>>>
>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>
>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>> ---
>>>>    mm/memory.c | 7 +++++++
>>>>    1 file changed, 7 insertions(+)
>>>>
>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>> index 52080e4..1ea168ff 100644
>>>> --- a/mm/memory.c
>>>> +++ b/mm/memory.c
>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>                   return VM_FAULT_RETRY;
>>>>           }
>>>>
>>>> +       /*
>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>> +        * in copy_one_pte
>>>> +        */
>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>> +               flush_tlb_page(vmf.vma, address);
>>>> +
>>>>           mem_cgroup_oom_enable();
>>>>           ret = handle_pte_fault(&vmf);
>>>>           mem_cgroup_oom_disable();
>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>
>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>
>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>
>>
>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>
>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
> Hi, Vinayak and Laurent
> 
> I think the below change will impact the performance significantly. Becuase most of process has many
> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
> call the flush_tlb_mm  later.
> 
> I think we can try the following way to do.
> 
> vm_write_begin(vma)
> copy_pte_range
> vm_write_end(vma)
> 
> The speculative page fault will return to grap the mmap_sem to run the nromal path.
> Any thought?

Here is a new version of the patch fixing this issue. There is no 
additional TLB flush, all the fix is belonging on vm_write_{begin,end} 
calls.

I did some test on x86_64 and PowerPC but that needs to be double check 
on arm64.

Vinayak, Zhong, could you please give it a try ?

Thanks,
Laurent.


[-- Attachment #2: 0001-mm-protect-against-PTE-changes-done-by-dup_mmap.patch --]
[-- Type: text/plain, Size: 4050 bytes --]

From 3be977febb9ff93d516a2d222cca4b5a52472a9f Mon Sep 17 00:00:00 2001
From: Laurent Dufour <ldufour@linux.ibm.com>
Date: Fri, 18 Jan 2019 16:19:08 +0100
Subject: [PATCH] mm: protect against PTE changes done by dup_mmap()

Vinayak Menon and Ganesh Mahendran reported that the following scenario may
lead to thread being blocked due to data corruption:

    CPU 1                   CPU 2                    CPU 3
    Process 1,              Process 1,               Process 1,
    Thread A                Thread B                 Thread C

    while (1) {             while (1) {              while(1) {
    pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
    pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
    }                       }

In the details this happens because :

    CPU 1                CPU 2                       CPU 3
    fork()
    copy_pte_range()
      set PTE rdonly
    got to next VMA...
     .                   PTE is seen rdonly          PTE still writable
     .                   thread is writing to page
     .                   -> page fault
     .                     copy the page             Thread writes to page
     .                      .                        -> no page fault
     .                     update the PTE
     .                     flush TLB for that PTE
   flush TLB                                        PTE are now rdonly

So the write done by the CPU 3 is interfering with the page copy operation
done by CPU 2, leading to the data corruption.

To avoid this we mark all the VMA involved in the COW mechanism as changing
by calling vm_write_begin(). This ensures that the speculative page fault
handler will not try to handle a fault on these pages.
The marker is set until the TLB is flushed, ensuring that all the CPUs will
now see the PTE as not writable.
Once the TLB is flush, the marker is removed by calling vm_write_end().

The variable last is used to keep tracked of the latest VMA marked to
handle the error path where part of the VMA may have been marked.

Reported-by: Ganesh Mahendran <opensource.ganesh@gmail.com>
Reported-by: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Laurent Dufour <ldufour@linux.ibm.com>
---
 kernel/fork.c | 30 ++++++++++++++++++++++++++++--
 1 file changed, 28 insertions(+), 2 deletions(-)

diff --git a/kernel/fork.c b/kernel/fork.c
index f1258c2ade09..39854b97d06a 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -395,7 +395,7 @@ EXPORT_SYMBOL(free_task);
 static __latent_entropy int dup_mmap(struct mm_struct *mm,
 					struct mm_struct *oldmm)
 {
-	struct vm_area_struct *mpnt, *tmp, *prev, **pprev;
+	struct vm_area_struct *mpnt, *tmp, *prev, **pprev, *last = NULL;
 	struct rb_node **rb_link, *rb_parent;
 	int retval;
 	unsigned long charge;
@@ -515,8 +515,18 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
 		rb_parent = &tmp->vm_rb;
 
 		mm->map_count++;
-		if (!(tmp->vm_flags & VM_WIPEONFORK))
+		if (!(tmp->vm_flags & VM_WIPEONFORK)) {
+			if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT)) {
+				/*
+				 * Mark this VMA as changing to prevent the
+				 * speculative page fault hanlder to process
+				 * it until the TLB are flushed below.
+				 */
+				last = mpnt;
+				vm_write_begin(mpnt);
+			}
 			retval = copy_page_range(mm, oldmm, mpnt);
+		}
 
 		if (tmp->vm_ops && tmp->vm_ops->open)
 			tmp->vm_ops->open(tmp);
@@ -530,6 +540,22 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
 out:
 	up_write(&mm->mmap_sem);
 	flush_tlb_mm(oldmm);
+
+	if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT)) {
+		/*
+		 * Since the TLB has been flush, we can safely unmark the
+		 * copied VMAs and allows the speculative page fault handler to
+		 * process them again.
+		 * Walk back the VMA list from the last marked VMA.
+		 */
+		for (; last; last = last->vm_prev) {
+			if (last->vm_flags & VM_DONTCOPY)
+				continue;
+			if (!(last->vm_flags & VM_WIPEONFORK))
+				vm_write_end(last);
+		}
+	}
+
 	up_write(&oldmm->mmap_sem);
 	dup_userfaultfd_complete(&uf);
 fail_uprobe_end:
-- 
2.20.1

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Le 18/01/2019 à 16:41, zhong jiang a écrit :
> On 2019/1/18 17:29, Laurent Dufour wrote:
>> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>>> Hi Laurent,
>>>>>>>
>>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>>
>>>>>> With the patch below, we don't hit the issue.
>>>>>>
>>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>>
>>>>>> It is observed that the following scenario results in
>>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>>> forever after few iterations.
>>>>>>
>>>>>> CPU 1                   CPU 2                    CPU 3
>>>>>> Process 1,              Process 1,               Process 1,
>>>>>> Thread A                Thread B                 Thread C
>>>>>>
>>>>>> while (1) {             while (1) {              while(1) {
>>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>>> }                       }
>>>>>>
>>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>>> on one of the CPUs at least. This can create a problem if one
>>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>>> entry can still modify old_page even after it is copied to
>>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>>> Nice catch and thanks for your investigation!
>>>>>
>>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>>
>>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>>> ---
>>>>>>     mm/memory.c | 7 +++++++
>>>>>>     1 file changed, 7 insertions(+)
>>>>>>
>>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>>> index 52080e4..1ea168ff 100644
>>>>>> --- a/mm/memory.c
>>>>>> +++ b/mm/memory.c
>>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>>                    return VM_FAULT_RETRY;
>>>>>>            }
>>>>>>
>>>>>> +       /*
>>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>>> +        * in copy_one_pte
>>>>>> +        */
>>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>>> +
>>>>>>            mem_cgroup_oom_enable();
>>>>>>            ret = handle_pte_fault(&vmf);
>>>>>>            mem_cgroup_oom_disable();
>>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>>
>>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>>
>>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>>
>>>>
>>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>>
>>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>>> Hi, Vinayak and Laurent
>>>
>>> I think the below change will impact the performance significantly. Becuase most of process has many
>>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>>> call the flush_tlb_mm  later.
>>>
>>> I think we can try the following way to do.
>>>
>>> vm_write_begin(vma)
>>> copy_pte_range
>>> vm_write_end(vma)
>>>
>>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>>> Any thought?
>>
>> Hi Zhong,
>>
>> I agree that flushing the TLB could have a bad impact on the performance, but tagging the VMA when copy_pte_range() is not fixing the issue as the VMA must be flagged until the PTE are flushed.
>>
>> Here is what happens:
>>
>> CPU A                CPU B                       CPU C
>> fork()
>> copy_pte_range()
>>    set PTE rdonly
>> got to next VMA...
>>   .                   PTE is seen rdonly             PTE still writable
>>   .                   thread is writing to page
>>   .                   -> page fault
>>   .                     copy the page             Thread writes to page
>>   .                      .                        -> no page fault
>>   .                     update the PTE
>>   .                     flush TLB for that PTE
>> flush TLB                                        PTE are now rdonly
>>
>> So the write done by the CPU C is interfering with the page copy operation done by CPU B, leading to the data corruption.
>>
> I want to know the case if the CPU B has finished in front of the CPU C that the data still is vaild ?

If the CPU B has done the flush TLB then the CPU C will write data to 
the right page. If the CPU B has not yet done the flush of the TLB as 
the time the CPU C is writing data, then this roughly the same issue.

Anyway this is fixed with the patch I'm about to sent for testing on arm64.

Cheers,
Laurent.

> 
> This is to say, the old_page will be changed from other cpu because of the access from other cpu.
> 
> Maybe this is a stupid qestion :-)
> 
> Thanks,
> zhong jiang.
>> Flushing the PTE in copy_pte_range() is fixing the issue as the CPU C is seeing the PTE as rdonly earlier. But this impacts performance.
>>
>> Another option, I'll work on is to flag _all the COW eligible_ VMA before starting copying them and until the PTE are flushed on the CPU A.
>> This way when the CPU B will page fault the speculative handler will abort because the VMA is in the way to be touched.
>>
>> But I need to ensure that all the calls to copy_pte_range() are handling this correctly.
>>
>> Laurent.
>>
>>
>> .
>>
> 
> 

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/18 17:29, Laurent Dufour wrote:
> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>> Hi Laurent,
>>>>>>
>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>
>>>>> With the patch below, we don't hit the issue.
>>>>>
>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>
>>>>> It is observed that the following scenario results in
>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>> forever after few iterations.
>>>>>
>>>>> CPU 1                   CPU 2                    CPU 3
>>>>> Process 1,              Process 1,               Process 1,
>>>>> Thread A                Thread B                 Thread C
>>>>>
>>>>> while (1) {             while (1) {              while(1) {
>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>> }                       }
>>>>>
>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>> on one of the CPUs at least. This can create a problem if one
>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>> entry can still modify old_page even after it is copied to
>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>> Nice catch and thanks for your investigation!
>>>>
>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>
>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> ---
>>>>>    mm/memory.c | 7 +++++++
>>>>>    1 file changed, 7 insertions(+)
>>>>>
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index 52080e4..1ea168ff 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>                   return VM_FAULT_RETRY;
>>>>>           }
>>>>>
>>>>> +       /*
>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>> +        * in copy_one_pte
>>>>> +        */
>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>> +
>>>>>           mem_cgroup_oom_enable();
>>>>>           ret = handle_pte_fault(&vmf);
>>>>>           mem_cgroup_oom_disable();
>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>
>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>
>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>
>>>
>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>
>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>> Hi, Vinayak and Laurent
>>
>> I think the below change will impact the performance significantly. Becuase most of process has many
>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>> call the flush_tlb_mm  later.
>>
>> I think we can try the following way to do.
>>
>> vm_write_begin(vma)
>> copy_pte_range
>> vm_write_end(vma)
>>
>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>> Any thought?
>
> Hi Zhong,
>
> I agree that flushing the TLB could have a bad impact on the performance, but tagging the VMA when copy_pte_range() is not fixing the issue as the VMA must be flagged until the PTE are flushed.
>
> Here is what happens:
>
> CPU A                CPU B                       CPU C
> fork()
> copy_pte_range()
>   set PTE rdonly
> got to next VMA...           
>  .                   PTE is seen rdonly             PTE still writable
>  .                   thread is writing to page
>  .                   -> page fault
>  .                     copy the page             Thread writes to page
>  .                      .                        -> no page fault
>  .                     update the PTE
>  .                     flush TLB for that PTE
> flush TLB                                        PTE are now rdonly  
>
> So the write done by the CPU C is interfering with the page copy operation done by CPU B, leading to the data corruption.
>
I want to know the case if the CPU B has finished in front of the CPU C that the data still is vaild ?

This is to say, the old_page will be changed from other cpu because of the access from other cpu.

Maybe this is a stupid qestion :-)

Thanks,
zhong jiang.
> Flushing the PTE in copy_pte_range() is fixing the issue as the CPU C is seeing the PTE as rdonly earlier. But this impacts performance.
>
> Another option, I'll work on is to flag _all the COW eligible_ VMA before starting copying them and until the PTE are flushed on the CPU A.
> This way when the CPU B will page fault the speculative handler will abort because the VMA is in the way to be touched.
>
> But I need to ensure that all the calls to copy_pte_range() are handling this correctly.
>
> Laurent.
>
>
> .
>

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/18 17:29, Laurent Dufour wrote:
> Le 17/01/2019 à 16:51, zhong jiang a écrit :
>> On 2019/1/16 19:41, Vinayak Menon wrote:
>>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>>> Hi Laurent,
>>>>>>
>>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>>
>>>>> With the patch below, we don't hit the issue.
>>>>>
>>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>>
>>>>> It is observed that the following scenario results in
>>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>>> forever after few iterations.
>>>>>
>>>>> CPU 1                   CPU 2                    CPU 3
>>>>> Process 1,              Process 1,               Process 1,
>>>>> Thread A                Thread B                 Thread C
>>>>>
>>>>> while (1) {             while (1) {              while(1) {
>>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>>> }                       }
>>>>>
>>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>>> (of lock l), stale tlb entries can exist with write permissions
>>>>> on one of the CPUs at least. This can create a problem if one
>>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>>> entry can still modify old_page even after it is copied to
>>>>> new_page by wp_page_copy, thus causing a corruption.
>>>> Nice catch and thanks for your investigation!
>>>>
>>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>>
>>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>>> ---
>>>>>    mm/memory.c | 7 +++++++
>>>>>    1 file changed, 7 insertions(+)
>>>>>
>>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>>> index 52080e4..1ea168ff 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>>                   return VM_FAULT_RETRY;
>>>>>           }
>>>>>
>>>>> +       /*
>>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>>> +        * in copy_one_pte
>>>>> +        */
>>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>>> +               flush_tlb_page(vmf.vma, address);
>>>>> +
>>>>>           mem_cgroup_oom_enable();
>>>>>           ret = handle_pte_fault(&vmf);
>>>>>           mem_cgroup_oom_disable();
>>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>>
>>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>>
>>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>>
>>>
>>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>>
>>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
>> Hi, Vinayak and Laurent
>>
>> I think the below change will impact the performance significantly. Becuase most of process has many
>> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
>> call the flush_tlb_mm  later.
>>
>> I think we can try the following way to do.
>>
>> vm_write_begin(vma)
>> copy_pte_range
>> vm_write_end(vma)
>>
>> The speculative page fault will return to grap the mmap_sem to run the nromal path.
>> Any thought?
>
> Hi Zhong,
>
> I agree that flushing the TLB could have a bad impact on the performance, but tagging the VMA when copy_pte_range() is not fixing the issue as the VMA must be flagged until the PTE are flushed.
>
> Here is what happens:
>
> CPU A                CPU B                       CPU C
> fork()
> copy_pte_range()
>   set PTE rdonly
> got to next VMA...           
>  .                   PTE is seen rdonly             PTE still writable
>  .                   thread is writing to page
>  .                   -> page fault
>  .                     copy the page             Thread writes to page
>  .                      .                        -> no page fault
>  .                     update the PTE
>  .                     flush TLB for that PTE
> flush TLB                                        PTE are now rdonly  
>
> So the write done by the CPU C is interfering with the page copy operation done by CPU B, leading to the data corruption.
>
I want to know the case if the CPU B has finished in front of the CPU C that the data still is vaild ?

This is to say, the old_page will be changed from other cpu because of the access from other cpu.

Maybe this is a stupid qestion :-)

Thanks,
zhong jiang.
> Flushing the PTE in copy_pte_range() is fixing the issue as the CPU C is seeing the PTE as rdonly earlier. But this impacts performance.
>
> Another option, I'll work on is to flag _all the COW eligible_ VMA before starting copying them and until the PTE are flushed on the CPU A.
> This way when the CPU B will page fault the speculative handler will abort because the VMA is in the way to be touched.
>
> But I need to ensure that all the calls to copy_pte_range() are handling this correctly.
>
> Laurent.
>
>
> .
>

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Le 17/01/2019 à 16:51, zhong jiang a écrit :
> On 2019/1/16 19:41, Vinayak Menon wrote:
>> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>>> Hi Laurent,
>>>>>
>>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>>
>>>> With the patch below, we don't hit the issue.
>>>>
>>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>>
>>>> It is observed that the following scenario results in
>>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>>> forever after few iterations.
>>>>
>>>> CPU 1                   CPU 2                    CPU 3
>>>> Process 1,              Process 1,               Process 1,
>>>> Thread A                Thread B                 Thread C
>>>>
>>>> while (1) {             while (1) {              while(1) {
>>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>>> }                       }
>>>>
>>>> When from thread C, copy_one_pte write-protects the parent pte
>>>> (of lock l), stale tlb entries can exist with write permissions
>>>> on one of the CPUs at least. This can create a problem if one
>>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>>> does not take mmap_sem it can proceed further fixing a fault soon
>>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>>> entry can still modify old_page even after it is copied to
>>>> new_page by wp_page_copy, thus causing a corruption.
>>> Nice catch and thanks for your investigation!
>>>
>>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>>
>>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>>> ---
>>>>    mm/memory.c | 7 +++++++
>>>>    1 file changed, 7 insertions(+)
>>>>
>>>> diff --git a/mm/memory.c b/mm/memory.c
>>>> index 52080e4..1ea168ff 100644
>>>> --- a/mm/memory.c
>>>> +++ b/mm/memory.c
>>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>>                   return VM_FAULT_RETRY;
>>>>           }
>>>>
>>>> +       /*
>>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>>> +        * in copy_one_pte
>>>> +        */
>>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>>> +               flush_tlb_page(vmf.vma, address);
>>>> +
>>>>           mem_cgroup_oom_enable();
>>>>           ret = handle_pte_fault(&vmf);
>>>>           mem_cgroup_oom_disable();
>>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>>
>>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>>
>> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>>
>>
>>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>>
>> Your patch works fine on arm64 with a minor change. Thanks Laurent.
> Hi, Vinayak and Laurent
> 
> I think the below change will impact the performance significantly. Becuase most of process has many
> vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
> call the flush_tlb_mm  later.
> 
> I think we can try the following way to do.
> 
> vm_write_begin(vma)
> copy_pte_range
> vm_write_end(vma)
> 
> The speculative page fault will return to grap the mmap_sem to run the nromal path.
> Any thought?

Hi Zhong,

I agree that flushing the TLB could have a bad impact on the 
performance, but tagging the VMA when copy_pte_range() is not fixing the 
issue as the VMA must be flagged until the PTE are flushed.

Here is what happens:

CPU A                CPU B                       CPU C
fork()
copy_pte_range()
   set PTE rdonly
got to next VMA...			
  .                   PTE is seen rdonly	         PTE still writable
  .                   thread is writing to page
  .                   -> page fault
  .                     copy the page             Thread writes to page
  .                      .                        -> no page fault
  .                     update the PTE
  .                     flush TLB for that PTE
flush TLB                                        PTE are now rdonly

So the write done by the CPU C is interfering with the page copy 
operation done by CPU B, leading to the data corruption.

Flushing the PTE in copy_pte_range() is fixing the issue as the CPU C is 
seeing the PTE as rdonly earlier. But this impacts performance.

Another option, I'll work on is to flag _all the COW eligible_ VMA 
before starting copying them and until the PTE are flushed on the CPU A.
This way when the CPU B will page fault the speculative handler will 
abort because the VMA is in the way to be touched.

But I need to ensure that all the calls to copy_pte_range() are handling 
this correctly.

Laurent.

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/16 19:41, Vinayak Menon wrote:
> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>> Hi Laurent,
>>>>
>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>
>>> With the patch below, we don't hit the issue.
>>>
>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>
>>> It is observed that the following scenario results in
>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>> forever after few iterations.
>>>
>>> CPU 1                   CPU 2                    CPU 3
>>> Process 1,              Process 1,               Process 1,
>>> Thread A                Thread B                 Thread C
>>>
>>> while (1) {             while (1) {              while(1) {
>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>> }                       }
>>>
>>> When from thread C, copy_one_pte write-protects the parent pte
>>> (of lock l), stale tlb entries can exist with write permissions
>>> on one of the CPUs at least. This can create a problem if one
>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>> does not take mmap_sem it can proceed further fixing a fault soon
>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>> entry can still modify old_page even after it is copied to
>>> new_page by wp_page_copy, thus causing a corruption.
>> Nice catch and thanks for your investigation!
>>
>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>
>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>> ---
>>>   mm/memory.c | 7 +++++++
>>>   1 file changed, 7 insertions(+)
>>>
>>> diff --git a/mm/memory.c b/mm/memory.c
>>> index 52080e4..1ea168ff 100644
>>> --- a/mm/memory.c
>>> +++ b/mm/memory.c
>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>                  return VM_FAULT_RETRY;
>>>          }
>>>
>>> +       /*
>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>> +        * in copy_one_pte
>>> +        */
>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>> +               flush_tlb_page(vmf.vma, address);
>>> +
>>>          mem_cgroup_oom_enable();
>>>          ret = handle_pte_fault(&vmf);
>>>          mem_cgroup_oom_disable();
>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>
>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>
> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>
>
>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>
> Your patch works fine on arm64 with a minor change. Thanks Laurent.
Hi, Vinayak and Laurent

I think the below change will impact the performance significantly. Becuase most of process has many
vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
call the flush_tlb_mm  later.

I think we can try the following way to do.

vm_write_begin(vma)
copy_pte_range
vm_write_end(vma)

The speculative page fault will return to grap the mmap_sem to run the nromal path.
Any thought?

Thanks,
zhong jiang
> diff --git a/mm/memory.c b/mm/memory.c
> index 52080e4..4767095 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1087,6 +1087,7 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
>         spinlock_t *src_ptl, *dst_ptl;
>         int progress = 0;
>         int rss[NR_MM_COUNTERS];
> +       unsigned long orig_addr = addr;
>         swp_entry_t entry = (swp_entry_t){0};
>
>  again:
> @@ -1125,6 +1126,15 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
>         } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
>
>         arch_leave_lazy_mmu_mode();
> +
> +       /*
> +        * Prevent the page fault handler to copy the page while stale tlb entry
> +        * are still not flushed.
> +        */
> +       if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT) &&
> +               is_cow_mapping(vma->vm_flags))
> +                       flush_tlb_range(vma, orig_addr, end);
> +
>         spin_unlock(src_ptl);
>         pte_unmap(orig_src_pte);
>         add_mm_rss_vec(dst_mm, rss);
>
> Thanks,
>
> Vinayak
>
>
>

Re: [PATCH v11 00/26] Speculative page faults

[zhong jiang]

On 2019/1/16 19:41, Vinayak Menon wrote:
> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>> Hi Laurent,
>>>>
>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>
>>> With the patch below, we don't hit the issue.
>>>
>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>
>>> It is observed that the following scenario results in
>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>> forever after few iterations.
>>>
>>> CPU 1                   CPU 2                    CPU 3
>>> Process 1,              Process 1,               Process 1,
>>> Thread A                Thread B                 Thread C
>>>
>>> while (1) {             while (1) {              while(1) {
>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>> }                       }
>>>
>>> When from thread C, copy_one_pte write-protects the parent pte
>>> (of lock l), stale tlb entries can exist with write permissions
>>> on one of the CPUs at least. This can create a problem if one
>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>> does not take mmap_sem it can proceed further fixing a fault soon
>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>> entry can still modify old_page even after it is copied to
>>> new_page by wp_page_copy, thus causing a corruption.
>> Nice catch and thanks for your investigation!
>>
>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>
>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>> ---
>>>   mm/memory.c | 7 +++++++
>>>   1 file changed, 7 insertions(+)
>>>
>>> diff --git a/mm/memory.c b/mm/memory.c
>>> index 52080e4..1ea168ff 100644
>>> --- a/mm/memory.c
>>> +++ b/mm/memory.c
>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>                  return VM_FAULT_RETRY;
>>>          }
>>>
>>> +       /*
>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>> +        * in copy_one_pte
>>> +        */
>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>> +               flush_tlb_page(vmf.vma, address);
>>> +
>>>          mem_cgroup_oom_enable();
>>>          ret = handle_pte_fault(&vmf);
>>>          mem_cgroup_oom_disable();
>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>
>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
>
> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?
>
>
>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>
> Your patch works fine on arm64 with a minor change. Thanks Laurent.
Hi, Vinayak and Laurent

I think the below change will impact the performance significantly. Becuase most of process has many
vmas with cow flags. Flush the tlb in advance is not the better way to avoid the issue and it will
call the flush_tlb_mm  later.

I think we can try the following way to do.

vm_write_begin(vma)
copy_pte_range
vm_write_end(vma)

The speculative page fault will return to grap the mmap_sem to run the nromal path.
Any thought?

Thanks,
zhong jiang
> diff --git a/mm/memory.c b/mm/memory.c
> index 52080e4..4767095 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1087,6 +1087,7 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
>         spinlock_t *src_ptl, *dst_ptl;
>         int progress = 0;
>         int rss[NR_MM_COUNTERS];
> +       unsigned long orig_addr = addr;
>         swp_entry_t entry = (swp_entry_t){0};
>
>  again:
> @@ -1125,6 +1126,15 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
>         } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
>
>         arch_leave_lazy_mmu_mode();
> +
> +       /*
> +        * Prevent the page fault handler to copy the page while stale tlb entry
> +        * are still not flushed.
> +        */
> +       if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT) &&
> +               is_cow_mapping(vma->vm_flags))
> +                       flush_tlb_range(vma, orig_addr, end);
> +
>         spin_unlock(src_ptl);
>         pte_unmap(orig_src_pte);
>         add_mm_rss_vec(dst_mm, rss);
>
> Thanks,
>
> Vinayak
>
>
>

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

Le 16/01/2019 à 12:41, Vinayak Menon a écrit :
> 
> On 1/15/2019 1:54 PM, Laurent Dufour wrote:
>> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>>> Hi Laurent,
>>>>
>>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>>
>>>
>>> With the patch below, we don't hit the issue.
>>>
>>> From: Vinayak Menon <vinmenon@codeaurora.org>
>>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>>
>>> It is observed that the following scenario results in
>>> threads A and B of process 1 blocking on pthread_mutex_lock
>>> forever after few iterations.
>>>
>>> CPU 1                   CPU 2                    CPU 3
>>> Process 1,              Process 1,               Process 1,
>>> Thread A                Thread B                 Thread C
>>>
>>> while (1) {             while (1) {              while(1) {
>>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>>> }                       }
>>>
>>> When from thread C, copy_one_pte write-protects the parent pte
>>> (of lock l), stale tlb entries can exist with write permissions
>>> on one of the CPUs at least. This can create a problem if one
>>> of the threads A or B hits the write fault. Though dup_mmap calls
>>> flush_tlb_mm after copy_page_range, since speculative page fault
>>> does not take mmap_sem it can proceed further fixing a fault soon
>>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>>> entry can still modify old_page even after it is copied to
>>> new_page by wp_page_copy, thus causing a corruption.
>>
>> Nice catch and thanks for your investigation!
>>
>> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>>
>>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>>> ---
>>>    mm/memory.c | 7 +++++++
>>>    1 file changed, 7 insertions(+)
>>>
>>> diff --git a/mm/memory.c b/mm/memory.c
>>> index 52080e4..1ea168ff 100644
>>> --- a/mm/memory.c
>>> +++ b/mm/memory.c
>>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>>                   return VM_FAULT_RETRY;
>>>           }
>>>
>>> +       /*
>>> +        * Discard tlb entries created before ptep_set_wrprotect
>>> +        * in copy_one_pte
>>> +        */
>>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>>> +               flush_tlb_page(vmf.vma, address);
>>> +
>>>           mem_cgroup_oom_enable();
>>>           ret = handle_pte_fault(&vmf);
>>>           mem_cgroup_oom_disable();
>>
>> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>>
>> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.
> 
> 
> But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?

I think this should be less costly to do it per vma at the time of the 
fork instead of per page hit once the fork has been done, since this 
will happen in both the forked task and the forking one (the COW pages 
are concerning the both sides of the fork).

>>
>> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>>
> 
> Your patch works fine on arm64 with a minor change. Thanks Laurent.

Yup my mistake !
I tried to shrink the patch after testing it, sounds that I shrunk it 
far too much...

> 
> diff --git a/mm/memory.c b/mm/memory.c
> index 52080e4..4767095 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -1087,6 +1087,7 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
>          spinlock_t *src_ptl, *dst_ptl;
>          int progress = 0;
>          int rss[NR_MM_COUNTERS];
> +       unsigned long orig_addr = addr;
>          swp_entry_t entry = (swp_entry_t){0};
> 
>   again:
> @@ -1125,6 +1126,15 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
>          } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
> 
>          arch_leave_lazy_mmu_mode();
> +
> +       /*
> +        * Prevent the page fault handler to copy the page while stale tlb entry
> +        * are still not flushed.
> +        */
> +       if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT) &&
> +               is_cow_mapping(vma->vm_flags))
> +                       flush_tlb_range(vma, orig_addr, end);
> +
>          spin_unlock(src_ptl);
>          pte_unmap(orig_src_pte);
>          add_mm_rss_vec(dst_mm, rss);
> 
> Thanks,
> 
> Vinayak
> 

Re: [PATCH v11 00/26] Speculative page faults

[Vinayak Menon]

On 1/15/2019 1:54 PM, Laurent Dufour wrote:
> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>> Hi Laurent,
>>>
>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>
>>
>> With the patch below, we don't hit the issue.
>>
>> From: Vinayak Menon <vinmenon@codeaurora.org>
>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>
>> It is observed that the following scenario results in
>> threads A and B of process 1 blocking on pthread_mutex_lock
>> forever after few iterations.
>>
>> CPU 1                   CPU 2                    CPU 3
>> Process 1,              Process 1,               Process 1,
>> Thread A                Thread B                 Thread C
>>
>> while (1) {             while (1) {              while(1) {
>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>> }                       }
>>
>> When from thread C, copy_one_pte write-protects the parent pte
>> (of lock l), stale tlb entries can exist with write permissions
>> on one of the CPUs at least. This can create a problem if one
>> of the threads A or B hits the write fault. Though dup_mmap calls
>> flush_tlb_mm after copy_page_range, since speculative page fault
>> does not take mmap_sem it can proceed further fixing a fault soon
>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>> entry can still modify old_page even after it is copied to
>> new_page by wp_page_copy, thus causing a corruption.
>
> Nice catch and thanks for your investigation!
>
> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>
>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>> ---
>>   mm/memory.c | 7 +++++++
>>   1 file changed, 7 insertions(+)
>>
>> diff --git a/mm/memory.c b/mm/memory.c
>> index 52080e4..1ea168ff 100644
>> --- a/mm/memory.c
>> +++ b/mm/memory.c
>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>                  return VM_FAULT_RETRY;
>>          }
>>
>> +       /*
>> +        * Discard tlb entries created before ptep_set_wrprotect
>> +        * in copy_one_pte
>> +        */
>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>> +               flush_tlb_page(vmf.vma, address);
>> +
>>          mem_cgroup_oom_enable();
>>          ret = handle_pte_fault(&vmf);
>>          mem_cgroup_oom_disable();
>
> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>
> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.


But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?


>
> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>

Your patch works fine on arm64 with a minor change. Thanks Laurent.

diff --git a/mm/memory.c b/mm/memory.c
index 52080e4..4767095 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1087,6 +1087,7 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
        spinlock_t *src_ptl, *dst_ptl;
        int progress = 0;
        int rss[NR_MM_COUNTERS];
+       unsigned long orig_addr = addr;
        swp_entry_t entry = (swp_entry_t){0};

 again:
@@ -1125,6 +1126,15 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
        } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);

        arch_leave_lazy_mmu_mode();
+
+       /*
+        * Prevent the page fault handler to copy the page while stale tlb entry
+        * are still not flushed.
+        */
+       if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT) &&
+               is_cow_mapping(vma->vm_flags))
+                       flush_tlb_range(vma, orig_addr, end);
+
        spin_unlock(src_ptl);
        pte_unmap(orig_src_pte);
        add_mm_rss_vec(dst_mm, rss);

Thanks,

Vinayak

Re: [PATCH v11 00/26] Speculative page faults

[Vinayak Menon]

On 1/15/2019 1:54 PM, Laurent Dufour wrote:
> Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
>> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>>> Hi Laurent,
>>>
>>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
>>
>>
>> With the patch below, we don't hit the issue.
>>
>> From: Vinayak Menon <vinmenon@codeaurora.org>
>> Date: Mon, 14 Jan 2019 16:06:34 +0530
>> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
>>
>> It is observed that the following scenario results in
>> threads A and B of process 1 blocking on pthread_mutex_lock
>> forever after few iterations.
>>
>> CPU 1                   CPU 2                    CPU 3
>> Process 1,              Process 1,               Process 1,
>> Thread A                Thread B                 Thread C
>>
>> while (1) {             while (1) {              while(1) {
>> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
>> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
>> }                       }
>>
>> When from thread C, copy_one_pte write-protects the parent pte
>> (of lock l), stale tlb entries can exist with write permissions
>> on one of the CPUs at least. This can create a problem if one
>> of the threads A or B hits the write fault. Though dup_mmap calls
>> flush_tlb_mm after copy_page_range, since speculative page fault
>> does not take mmap_sem it can proceed further fixing a fault soon
>> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
>> entry can still modify old_page even after it is copied to
>> new_page by wp_page_copy, thus causing a corruption.
>
> Nice catch and thanks for your investigation!
>
> There is a real synchronization issue here between copy_page_range() and the speculative page fault handler. I didn't get it on PowerVM since the TLB are flushed when arch_exit_lazy_mode() is called in copy_page_range() but now, I can get it when running on x86_64.
>
>> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
>> ---
>>   mm/memory.c | 7 +++++++
>>   1 file changed, 7 insertions(+)
>>
>> diff --git a/mm/memory.c b/mm/memory.c
>> index 52080e4..1ea168ff 100644
>> --- a/mm/memory.c
>> +++ b/mm/memory.c
>> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>>                  return VM_FAULT_RETRY;
>>          }
>>
>> +       /*
>> +        * Discard tlb entries created before ptep_set_wrprotect
>> +        * in copy_one_pte
>> +        */
>> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
>> +               flush_tlb_page(vmf.vma, address);
>> +
>>          mem_cgroup_oom_enable();
>>          ret = handle_pte_fault(&vmf);
>>          mem_cgroup_oom_disable();
>
> Your patch is fixing the race but I'm wondering about the cost of these tlb flushes. Here we are flushing on a per page basis (architecture like x86_64 are smarter and flush more pages) but there is a request to flush a range of tlb entries each time a cow page is newly touched. I think there could be some bad impact here.
>
> Another option would be to flush the range in copy_pte_range() before unlocking the page table lock. This will flush entries flush_tlb_mm() would later handle in dup_mmap() but that will be called once per fork per cow VMA.


But wouldn't this cause an unnecessary impact if most of the COW pages remain untouched (which I assume would be the usual case) and thus do not create a fault ?


>
> I tried the attached patch which seems to fix the issue on x86_64. Could you please give it a try on arm64 ?
>

Your patch works fine on arm64 with a minor change. Thanks Laurent.

diff --git a/mm/memory.c b/mm/memory.c
index 52080e4..4767095 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1087,6 +1087,7 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
        spinlock_t *src_ptl, *dst_ptl;
        int progress = 0;
        int rss[NR_MM_COUNTERS];
+       unsigned long orig_addr = addr;
        swp_entry_t entry = (swp_entry_t){0};

 again:
@@ -1125,6 +1126,15 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
        } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);

        arch_leave_lazy_mmu_mode();
+
+       /*
+        * Prevent the page fault handler to copy the page while stale tlb entry
+        * are still not flushed.
+        */
+       if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT) &&
+               is_cow_mapping(vma->vm_flags))
+                       flush_tlb_range(vma, orig_addr, end);
+
        spin_unlock(src_ptl);
        pte_unmap(orig_src_pte);
        add_mm_rss_vec(dst_mm, rss);

Thanks,

Vinayak

Re: [PATCH v11 00/26] Speculative page faults

[Laurent Dufour]

[-- Attachment #1: Type: text/plain, Size: 3581 bytes --]

Le 14/01/2019 à 14:19, Vinayak Menon a écrit :
> On 1/11/2019 9:13 PM, Vinayak Menon wrote:
>> Hi Laurent,
>>
>> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).
> 
> 
> With the patch below, we don't hit the issue.
> 
> From: Vinayak Menon <vinmenon@codeaurora.org>
> Date: Mon, 14 Jan 2019 16:06:34 +0530
> Subject: [PATCH] mm: flush stale tlb entries on speculative write fault
> 
> It is observed that the following scenario results in
> threads A and B of process 1 blocking on pthread_mutex_lock
> forever after few iterations.
> 
> CPU 1                   CPU 2                    CPU 3
> Process 1,              Process 1,               Process 1,
> Thread A                Thread B                 Thread C
> 
> while (1) {             while (1) {              while(1) {
> pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
> pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
> }                       }
> 
> When from thread C, copy_one_pte write-protects the parent pte
> (of lock l), stale tlb entries can exist with write permissions
> on one of the CPUs at least. This can create a problem if one
> of the threads A or B hits the write fault. Though dup_mmap calls
> flush_tlb_mm after copy_page_range, since speculative page fault
> does not take mmap_sem it can proceed further fixing a fault soon
> after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
> entry can still modify old_page even after it is copied to
> new_page by wp_page_copy, thus causing a corruption.

Nice catch and thanks for your investigation!

There is a real synchronization issue here between copy_page_range() and 
the speculative page fault handler. I didn't get it on PowerVM since the 
TLB are flushed when arch_exit_lazy_mode() is called in 
copy_page_range() but now, I can get it when running on x86_64.

> Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
> ---
>   mm/memory.c | 7 +++++++
>   1 file changed, 7 insertions(+)
> 
> diff --git a/mm/memory.c b/mm/memory.c
> index 52080e4..1ea168ff 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
>                  return VM_FAULT_RETRY;
>          }
> 
> +       /*
> +        * Discard tlb entries created before ptep_set_wrprotect
> +        * in copy_one_pte
> +        */
> +       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
> +               flush_tlb_page(vmf.vma, address);
> +
>          mem_cgroup_oom_enable();
>          ret = handle_pte_fault(&vmf);
>          mem_cgroup_oom_disable();

Your patch is fixing the race but I'm wondering about the cost of these 
tlb flushes. Here we are flushing on a per page basis (architecture like 
x86_64 are smarter and flush more pages) but there is a request to flush 
a range of tlb entries each time a cow page is newly touched. I think 
there could be some bad impact here.

Another option would be to flush the range in copy_pte_range() before 
unlocking the page table lock. This will flush entries flush_tlb_mm() 
would later handle in dup_mmap() but that will be called once per fork 
per cow VMA.

I tried the attached patch which seems to fix the issue on x86_64. Could 
you please give it a try on arm64 ?

Thanks,
Laurent.


[-- Attachment #2: 0001-mm-flush-TLB-once-pages-are-copied-when-SPF-is-on.patch --]
[-- Type: text/plain, Size: 2137 bytes --]

From 9847338187c5c7e2d387d14765452d00fa60981e Mon Sep 17 00:00:00 2001
From: Laurent Dufour <ldufour@linux.vnet.ibm.com>
Date: Mon, 14 Jan 2019 18:35:39 +0100
Subject: [PATCH] mm: flush TLB once pages are copied when SPF is on

Vinayak Menon reported that the following scenario results in
threads A and B of process 1 blocking on pthread_mutex_lock
forever after few iterations.

CPU 1                   CPU 2                    CPU 3
Process 1,              Process 1,               Process 1,
Thread A                Thread B                 Thread C

while (1) {             while (1) {              while(1) {
pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
}                       }

When from thread C, copy_one_pte write-protects the parent pte
(of lock l), stale tlb entries can exist with write permissions
on one of the CPUs at least. This can create a problem if one
of the threads A or B hits the write fault. Though dup_mmap calls
flush_tlb_mm after copy_page_range, since speculative page fault
does not take mmap_sem it can proceed further fixing a fault soon
after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
entry can still modify old_page even after it is copied to
new_page by wp_page_copy, thus causing a corruption.

Reported-by: Vinayak Menon <vinmenon@codeaurora.org>
Signed-off-by: Laurent Dufour <ldufour@linux.vnet.ibm.com>
---
 mm/memory.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/mm/memory.c b/mm/memory.c
index 48e1cf0a54ef..b7501294e0a0 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -1112,6 +1112,15 @@ static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
 	} while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
 
 	arch_leave_lazy_mmu_mode();
+
+	/*
+	 * Prevent the page fault handler to copy the page while stale tlb entry
+	 * are still not flushed.
+	 */
+	if (IS_ENABLED(CONFIG_SPECULATIVE_PAGE_FAULT) &&
+	    is_cow_mapping(vma->vm_flags))
+		flush_tlb_range(vma, addr, end);
+
 	spin_unlock(src_ptl);
 	pte_unmap(orig_src_pte);
 	add_mm_rss_vec(dst_mm, rss);
-- 
2.20.1

Re: [PATCH v11 00/26] Speculative page faults

[Vinayak Menon]

On 1/11/2019 9:13 PM, Vinayak Menon wrote:
> Hi Laurent,
>
> We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).


With the patch below, we don't hit the issue.

From: Vinayak Menon <vinmenon@codeaurora.org>
Date: Mon, 14 Jan 2019 16:06:34 +0530
Subject: [PATCH] mm: flush stale tlb entries on speculative write fault

It is observed that the following scenario results in
threads A and B of process 1 blocking on pthread_mutex_lock
forever after few iterations.

CPU 1                   CPU 2                    CPU 3
Process 1,              Process 1,               Process 1,
Thread A                Thread B                 Thread C

while (1) {             while (1) {              while(1) {
pthread_mutex_lock(l)   pthread_mutex_lock(l)    fork
pthread_mutex_unlock(l) pthread_mutex_unlock(l)  }
}                       }

When from thread C, copy_one_pte write-protects the parent pte
(of lock l), stale tlb entries can exist with write permissions
on one of the CPUs at least. This can create a problem if one
of the threads A or B hits the write fault. Though dup_mmap calls
flush_tlb_mm after copy_page_range, since speculative page fault
does not take mmap_sem it can proceed further fixing a fault soon
after CPU 3 does ptep_set_wrprotect. But the CPU with stale tlb
entry can still modify old_page even after it is copied to
new_page by wp_page_copy, thus causing a corruption.

Signed-off-by: Vinayak Menon <vinmenon@codeaurora.org>
---
 mm/memory.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/mm/memory.c b/mm/memory.c
index 52080e4..1ea168ff 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4507,6 +4507,13 @@ int __handle_speculative_fault(struct mm_struct *mm, unsigned long address,
                return VM_FAULT_RETRY;
        }

+       /*
+        * Discard tlb entries created before ptep_set_wrprotect
+        * in copy_one_pte
+        */
+       if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))
+               flush_tlb_page(vmf.vma, address);
+
        mem_cgroup_oom_enable();
        ret = handle_pte_fault(&vmf);
        mem_cgroup_oom_disable();
--
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of the Code Aurora Forum, hosted by The Linux Foundation

[PATCH v11 00/26] Speculative page faults

[Vinayak Menon]

Hi Laurent,

We are observing an issue with speculative page fault with the following test code on ARM64 (4.14 kernel, 8 cores).

Steps:

1) Run the program

2) The 2 threads will try lock/unlock and prints messages, and main thread does fork. In around 1 minute time, lock/unlock threads will hang on pthread_mutex_lock.


Initially the issue was reported when ptrace was tried on apps. Later it was discovered that the write protect done by fork is causing the issue and the below test code was created.


Observations:

1) We have tried disabling SPF and the issue disappears.

2) Adding this piece of code in __handle_speculative_fault also fixes the problem.

if (flags & FAULT_FLAG_WRITE && !pte_write(vmf.orig_pte))

    return VM_FAULT_RETRY;

3) As an experiment we tried encapsulating handle_speculative_fault with down_read(mmap_sem) and that too fixes the problem.

4) It is observed that while in wp_page_copy, the contents of the old_page changes which should not ideally happen as the pte is !pte_write.

5) To prove that it is a race, we tried affining the threads to single core, and the issue disappears.

Let us know if you want us to try out any experiments.

Thanks,
Vinayak

/**test.c***/

#include <stdio.h>
#include <pthread.h>
#include <unistd.h>
#include <string.h>
#include <stdlib.h>

#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>

#define UNUSED(expr) do { (void)(expr); } while(0)
#define print printf

pthread_mutex_t m;
pthread_cond_t c;

static void* cond_do(void* arg)
{
  UNUSED(arg);
  do {
    print("%s:%d state=%u addr: %lx\n", __func__, __LINE__, *(uint32_t*)(&m), (unsigned long)&m);
    pthread_mutex_lock(&m);
 
    print("%s:%d state=%u addr: %lx\n", __func__, __LINE__, *(uint32_t*)(&m), (unsigned long)&m);
    pthread_mutex_unlock(&m);
    print("%s:%d state=%u addr: %lx\n", __func__, __LINE__, *(uint32_t*)(&m), (unsigned long)&m);
  } while (true);

  return NULL;
}

static void* sig_do(void* arg)
{
  UNUSED(arg);
  do {
    print("%s:%d state=%u addr: %lx\n", __func__, __LINE__, *(uint32_t*)(&m), (unsigned long)&m);
    pthread_mutex_lock(&m);

    print("%s:%d state=%u addr: %lx\n", __func__, __LINE__, *(uint32_t*)(&m), (unsigned long)&m);
    pthread_mutex_unlock(&m);
    print("%s:%d state=%u addr: %lx\n", __func__, __LINE__, *(uint32_t*)(&m), (unsigned long)&m);
  } while (true);
  return NULL;
}

int main()
{
  pthread_t sig;
  pthread_t cond;
  pthread_mutex_init(&m, NULL);
 
  pthread_create(&cond, NULL, cond_do, NULL);
  pthread_create(&sig, NULL, sig_do, NULL);

  while(1) {
      if (!fork()) {
          usleep(500);
          abort();
      }
      usleep(550);
  }

  pthread_join(sig, NULL);
  pthread_join(cond, NULL);

  return 0;
}