From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753218AbeE1FX4 convert rfc822-to-8bit (ORCPT ); Mon, 28 May 2018 01:23:56 -0400 Received: from mga04.intel.com ([192.55.52.120]:56546 "EHLO mga04.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750765AbeE1FXy (ORCPT ); Mon, 28 May 2018 01:23:54 -0400 X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.49,451,1520924400"; d="scan'208";a="50998574" From: "Song, HaiyanX" To: Laurent Dufour , "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: RE: [PATCH v11 00/26] Speculative page faults Thread-Topic: [PATCH v11 00/26] Speculative page faults Thread-Index: AQHT7c81YDLQ4A7f1EGHmQWP/pOb4KREdfjK Date: Mon, 28 May 2018 05:23:48 +0000 Message-ID: <9FE19350E8A7EE45B64D8D63D368C8966B834B67@SHSMSX101.ccr.corp.intel.com> References: <1526555193-7242-1-git-send-email-ldufour@linux.vnet.ibm.com> In-Reply-To: <1526555193-7242-1-git-send-email-ldufour@linux.vnet.ibm.com> Accept-Language: en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-originating-ip: [10.3.86.137] Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 8BIT MIME-Version: 1.0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org 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:*' 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 '. 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 From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga07.intel.com (mga07.intel.com [134.134.136.100]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by lists.ozlabs.org (Postfix) with ESMTPS id 40vQJS5GlmzF0gL for ; Mon, 28 May 2018 15:23:59 +1000 (AEST) From: "Song, HaiyanX" To: Laurent Dufour , "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: RE: [PATCH v11 00/26] Speculative page faults Date: Mon, 28 May 2018 05:23:48 +0000 Message-ID: <9FE19350E8A7EE45B64D8D63D368C8966B834B67@SHSMSX101.ccr.corp.intel.com> References: <1526555193-7242-1-git-send-email-ldufour@linux.vnet.ibm.com> In-Reply-To: <1526555193-7242-1-git-send-email-ldufour@linux.vnet.ibm.com> Content-Type: text/plain; charset="us-ascii" MIME-Version: 1.0 List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , =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:*' =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 '. 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=