[PATCH 0/8] Add 32 bit VDSO time function support

[PATCH 0/8] Add 32 bit VDSO time function support

[stefani]

From: Stefani Seibold <stefani@seibold.net>

This patch add the functions vdso_gettimeofday(), vdso_clock_gettime()
and vdso_time() to the 32 bit VDSO.

The reason to do this was to get a fast reliable time stamp. Many developers
uses TSC to get a fast time stamp, without knowing the pitfalls. VDSO
time functions a fast and a reliable way, because the kernel knows the
best time source and the P- and C-state of the CPU.

The helper library to use the VDSO functions can be download at
http://http://seibold.net/vdso.c
The libary is very small, only 228 lines of code. Compile it with
gcc -Wall -O3 -fpic vdso.c -lrt -shared -o libvdso.so
and use it with LD_PRELOAD=<path>/libvdso.so

This kind of helper must be integrated into glibc, for x86 64 bit and
PowerPC it is already there.

Some linux 32 bit kernel benchmark results (all measurements are in nano
seconds):

Intel(R) Celeron(TM) CPU 400MHz

Average time kernel call:
 gettimeofday(): 1039
 clock_gettime(): 1578
 time(): 526
Average time VDSO call:
 gettimeofday(): 378
 clock_gettime(): 303
 time(): 60

Celeron(R) Dual-Core CPU T3100 1.90GHz

Average time kernel call:
 gettimeofday(): 209
 clock_gettime(): 406
 time(): 135
Average time VDSO call:
 gettimeofday(): 51
 clock_gettime(): 43
 time(): 10

So you can see a performance increase between 4 and 13, depending on the
CPU and the function.

The address layout of the VDSO has changed, because there is no fixed
address space available on a x86 32 bit kernel, despite the name. Because
someone decided to add an offset to the __FIXADDR_TOP for virtualization.

Also the IA32 Emulation uses the whole 4 GB address space, so there is no
fixed address available.

This was the reason not depend on this kind of address and change the layout
of the VDSO. The VDSO for a 32 bit application has now three pages:

+----------------------------------------+
+ VDSO page (includes code) ro+x         +
+----------------------------------------+
+ VVAR page (export kernel variables) ro +
+----------------------------------------+
+ HPET page (mapped registers) ro 
+----------------------------------------+

The VDSO page for a 32 bit resided now on 0xffffc000, followed by the VVAR and
HPET page.

In the non compat mode the VMA of the VDSO is now 3 pages for a 32 bit kernel.
So this decrease the available logical address room by 2 pages.

The patch is against kernel 3.14 (e7651b819e90da924991d727d3c007200a18670d)

Changelog:
25.11.2012 - first release and proof of concept for linux 3.4
11.12.2012 - Port to linux 3.7 and code cleanup
12.12.2012 - fixes suggested by Andy Lutomirski
           - fixes suggested by John Stultz
           - use call VDSO32_vsyscall instead of int 80
           - code cleanup
17.12.2012 - support for IA32_EMULATION, this includes
             - code cleanup
             - include cleanup to fix compile warnings and errors
	     - move out seqcount from seqlock, enable use in VDSO
	     - map FIXMAP and HPET into the 32 bit address space
18.12.2012 - split into separate patches
30.01.2014 - revamp the code
             - code clean up
	     - VDSO layout changed
	     - no fixed addresses
	     - port to 3.14
01.02.2014 - code cleanup
02.02.2014 - code cleanup
             - split into more patches
	     - use HPET_COUNTER instead of hard coded value
	     - fix changelog to the right year ;-)

[PATCH 1/8] Make vsyscall_gtod_data handling x86 generic

[stefani]

From: Stefani Seibold <stefani@seibold.net>

This patch move the vsyscall_gtod_data handling out of vsyscall_64.c
into an additonal file vsyscall_gtod.c to make the functionality
available for x86 32 bit kernel.

It also adds a new vsyscall_32.c which setup the VVAR page.

Signed-off-by: Stefani Seibold <stefani@seibold.net>
---
 arch/x86/Kconfig                   |  4 +--
 arch/x86/include/asm/clocksource.h |  4 ---
 arch/x86/include/asm/fixmap.h      |  2 ++
 arch/x86/include/asm/vvar.h        |  4 +++
 arch/x86/kernel/Makefile           |  3 +-
 arch/x86/kernel/hpet.c             |  4 ---
 arch/x86/kernel/setup.c            |  2 --
 arch/x86/kernel/tsc.c              |  2 --
 arch/x86/kernel/vmlinux.lds.S      |  3 --
 arch/x86/kernel/vsyscall_32.c      | 24 +++++++++++++++
 arch/x86/kernel/vsyscall_64.c      | 44 ----------------------------
 arch/x86/kernel/vsyscall_gtod.c    | 60 ++++++++++++++++++++++++++++++++++++++
 12 files changed, 94 insertions(+), 62 deletions(-)
 create mode 100644 arch/x86/kernel/vsyscall_32.c
 create mode 100644 arch/x86/kernel/vsyscall_gtod.c

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 940e50e..b556f00 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -107,9 +107,9 @@ config X86
 	select HAVE_ARCH_SOFT_DIRTY
 	select CLOCKSOURCE_WATCHDOG
 	select GENERIC_CLOCKEVENTS
-	select ARCH_CLOCKSOURCE_DATA if X86_64
+	select ARCH_CLOCKSOURCE_DATA
 	select GENERIC_CLOCKEVENTS_BROADCAST if X86_64 || (X86_32 && X86_LOCAL_APIC)
-	select GENERIC_TIME_VSYSCALL if X86_64
+	select GENERIC_TIME_VSYSCALL
 	select KTIME_SCALAR if X86_32
 	select GENERIC_STRNCPY_FROM_USER
 	select GENERIC_STRNLEN_USER
diff --git a/arch/x86/include/asm/clocksource.h b/arch/x86/include/asm/clocksource.h
index 16a57f4..eda81dc 100644
--- a/arch/x86/include/asm/clocksource.h
+++ b/arch/x86/include/asm/clocksource.h
@@ -3,8 +3,6 @@
 #ifndef _ASM_X86_CLOCKSOURCE_H
 #define _ASM_X86_CLOCKSOURCE_H
 
-#ifdef CONFIG_X86_64
-
 #define VCLOCK_NONE 0  /* No vDSO clock available.	*/
 #define VCLOCK_TSC  1  /* vDSO should use vread_tsc.	*/
 #define VCLOCK_HPET 2  /* vDSO should use vread_hpet.	*/
@@ -14,6 +12,4 @@ struct arch_clocksource_data {
 	int vclock_mode;
 };
 
-#endif /* CONFIG_X86_64 */
-
 #endif /* _ASM_X86_CLOCKSOURCE_H */
diff --git a/arch/x86/include/asm/fixmap.h b/arch/x86/include/asm/fixmap.h
index 7252cd3..504c04a 100644
--- a/arch/x86/include/asm/fixmap.h
+++ b/arch/x86/include/asm/fixmap.h
@@ -74,6 +74,8 @@ extern unsigned long __FIXADDR_TOP;
 enum fixed_addresses {
 #ifdef CONFIG_X86_32
 	FIX_HOLE,
+	VSYSCALL_HPET,
+	VVAR_PAGE,
 	FIX_VDSO,
 #else
 	VSYSCALL_LAST_PAGE,
diff --git a/arch/x86/include/asm/vvar.h b/arch/x86/include/asm/vvar.h
index d76ac40..c442782 100644
--- a/arch/x86/include/asm/vvar.h
+++ b/arch/x86/include/asm/vvar.h
@@ -17,7 +17,11 @@
  */
 
 /* Base address of vvars.  This is not ABI. */
+#ifdef CONFIG_X86_64
 #define VVAR_ADDRESS (-10*1024*1024 - 4096)
+#else
+#define VVAR_ADDRESS 0xffffd000
+#endif
 
 #if defined(__VVAR_KERNEL_LDS)
 
diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile
index cb648c8..3282eda 100644
--- a/arch/x86/kernel/Makefile
+++ b/arch/x86/kernel/Makefile
@@ -26,7 +26,8 @@ obj-$(CONFIG_IRQ_WORK)  += irq_work.o
 obj-y			+= probe_roms.o
 obj-$(CONFIG_X86_32)	+= i386_ksyms_32.o
 obj-$(CONFIG_X86_64)	+= sys_x86_64.o x8664_ksyms_64.o
-obj-y			+= syscall_$(BITS).o
+obj-y			+= syscall_$(BITS).o vsyscall_gtod.o
+obj-$(CONFIG_X86_32)	+= vsyscall_32.o
 obj-$(CONFIG_X86_64)	+= vsyscall_64.o
 obj-$(CONFIG_X86_64)	+= vsyscall_emu_64.o
 obj-$(CONFIG_SYSFS)	+= ksysfs.o
diff --git a/arch/x86/kernel/hpet.c b/arch/x86/kernel/hpet.c
index da85a8e..54263f0 100644
--- a/arch/x86/kernel/hpet.c
+++ b/arch/x86/kernel/hpet.c
@@ -74,9 +74,7 @@ static inline void hpet_writel(unsigned int d, unsigned int a)
 static inline void hpet_set_mapping(void)
 {
 	hpet_virt_address = ioremap_nocache(hpet_address, HPET_MMAP_SIZE);
-#ifdef CONFIG_X86_64
 	__set_fixmap(VSYSCALL_HPET, hpet_address, PAGE_KERNEL_VVAR_NOCACHE);
-#endif
 }
 
 static inline void hpet_clear_mapping(void)
@@ -752,9 +750,7 @@ static struct clocksource clocksource_hpet = {
 	.mask		= HPET_MASK,
 	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
 	.resume		= hpet_resume_counter,
-#ifdef CONFIG_X86_64
 	.archdata	= { .vclock_mode = VCLOCK_HPET },
-#endif
 };
 
 static int hpet_clocksource_register(void)
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index 06853e6..56ff330 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -1182,9 +1182,7 @@ void __init setup_arch(char **cmdline_p)
 
 	tboot_probe();
 
-#ifdef CONFIG_X86_64
 	map_vsyscall();
-#endif
 
 	generic_apic_probe();
 
diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
index 19e5adb..f49d1a2 100644
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -988,9 +988,7 @@ static struct clocksource clocksource_tsc = {
 	.mask                   = CLOCKSOURCE_MASK(64),
 	.flags                  = CLOCK_SOURCE_IS_CONTINUOUS |
 				  CLOCK_SOURCE_MUST_VERIFY,
-#ifdef CONFIG_X86_64
 	.archdata               = { .vclock_mode = VCLOCK_TSC },
-#endif
 };
 
 void mark_tsc_unstable(char *reason)
diff --git a/arch/x86/kernel/vmlinux.lds.S b/arch/x86/kernel/vmlinux.lds.S
index da6b35a..1d4897b 100644
--- a/arch/x86/kernel/vmlinux.lds.S
+++ b/arch/x86/kernel/vmlinux.lds.S
@@ -147,7 +147,6 @@ SECTIONS
 		_edata = .;
 	} :data
 
-#ifdef CONFIG_X86_64
 
 	. = ALIGN(PAGE_SIZE);
 	__vvar_page = .;
@@ -169,8 +168,6 @@ SECTIONS
 
        . = ALIGN(__vvar_page + PAGE_SIZE, PAGE_SIZE);
 
-#endif /* CONFIG_X86_64 */
-
 	/* Init code and data - will be freed after init */
 	. = ALIGN(PAGE_SIZE);
 	.init.begin : AT(ADDR(.init.begin) - LOAD_OFFSET) {
diff --git a/arch/x86/kernel/vsyscall_32.c b/arch/x86/kernel/vsyscall_32.c
new file mode 100644
index 0000000..f541dfa
--- /dev/null
+++ b/arch/x86/kernel/vsyscall_32.c
@@ -0,0 +1,24 @@
+/*
+ *  Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
+ *  Copyright 2003 Andi Kleen, SuSE Labs.
+ *
+ *  Modified for x86 32 bit arch by Stefani Seibold <stefani@seibold.net>
+ *
+ *  Thanks to hpa@transmeta.com for some useful hint.
+ *  Special thanks to Ingo Molnar for his early experience with
+ *  a different vsyscall implementation for Linux/IA32 and for the name.
+ *
+ */
+
+#include <asm/vsyscall.h>
+#include <asm/pgtable.h>
+#include <asm/fixmap.h>
+
+void __init map_vsyscall(void)
+{
+	extern char __vvar_page;
+	unsigned long physaddr_vvar_page = __pa_symbol(&__vvar_page);
+
+	__set_fixmap(VVAR_PAGE, physaddr_vvar_page, PAGE_KERNEL_VVAR);
+}
+
diff --git a/arch/x86/kernel/vsyscall_64.c b/arch/x86/kernel/vsyscall_64.c
index 1f96f93..50622ed 100644
--- a/arch/x86/kernel/vsyscall_64.c
+++ b/arch/x86/kernel/vsyscall_64.c
@@ -47,14 +47,12 @@
 #include <asm/segment.h>
 #include <asm/desc.h>
 #include <asm/topology.h>
-#include <asm/vgtod.h>
 #include <asm/traps.h>
 
 #define CREATE_TRACE_POINTS
 #include "vsyscall_trace.h"
 
 DEFINE_VVAR(int, vgetcpu_mode);
-DEFINE_VVAR(struct vsyscall_gtod_data, vsyscall_gtod_data);
 
 static enum { EMULATE, NATIVE, NONE } vsyscall_mode = EMULATE;
 
@@ -77,48 +75,6 @@ static int __init vsyscall_setup(char *str)
 }
 early_param("vsyscall", vsyscall_setup);
 
-void update_vsyscall_tz(void)
-{
-	vsyscall_gtod_data.sys_tz = sys_tz;
-}
-
-void update_vsyscall(struct timekeeper *tk)
-{
-	struct vsyscall_gtod_data *vdata = &vsyscall_gtod_data;
-
-	write_seqcount_begin(&vdata->seq);
-
-	/* copy vsyscall data */
-	vdata->clock.vclock_mode	= tk->clock->archdata.vclock_mode;
-	vdata->clock.cycle_last		= tk->clock->cycle_last;
-	vdata->clock.mask		= tk->clock->mask;
-	vdata->clock.mult		= tk->mult;
-	vdata->clock.shift		= tk->shift;
-
-	vdata->wall_time_sec		= tk->xtime_sec;
-	vdata->wall_time_snsec		= tk->xtime_nsec;
-
-	vdata->monotonic_time_sec	= tk->xtime_sec
-					+ tk->wall_to_monotonic.tv_sec;
-	vdata->monotonic_time_snsec	= tk->xtime_nsec
-					+ (tk->wall_to_monotonic.tv_nsec
-						<< tk->shift);
-	while (vdata->monotonic_time_snsec >=
-					(((u64)NSEC_PER_SEC) << tk->shift)) {
-		vdata->monotonic_time_snsec -=
-					((u64)NSEC_PER_SEC) << tk->shift;
-		vdata->monotonic_time_sec++;
-	}
-
-	vdata->wall_time_coarse.tv_sec	= tk->xtime_sec;
-	vdata->wall_time_coarse.tv_nsec	= (long)(tk->xtime_nsec >> tk->shift);
-
-	vdata->monotonic_time_coarse	= timespec_add(vdata->wall_time_coarse,
-							tk->wall_to_monotonic);
-
-	write_seqcount_end(&vdata->seq);
-}
-
 static void warn_bad_vsyscall(const char *level, struct pt_regs *regs,
 			      const char *message)
 {
diff --git a/arch/x86/kernel/vsyscall_gtod.c b/arch/x86/kernel/vsyscall_gtod.c
new file mode 100644
index 0000000..91862a4
--- /dev/null
+++ b/arch/x86/kernel/vsyscall_gtod.c
@@ -0,0 +1,60 @@
+/*
+ *  Copyright (C) 2001 Andrea Arcangeli <andrea@suse.de> SuSE
+ *  Copyright 2003 Andi Kleen, SuSE Labs.
+ *
+ *  Modified for x86 32 bit architecture by
+ *  Stefani Seibold <stefani@seibold.net>
+ *
+ *  Thanks to hpa@transmeta.com for some useful hint.
+ *  Special thanks to Ingo Molnar for his early experience with
+ *  a different vsyscall implementation for Linux/IA32 and for the name.
+ *
+ */
+
+#include <linux/timekeeper_internal.h>
+#include <asm/vgtod.h>
+
+DEFINE_VVAR(struct vsyscall_gtod_data, vsyscall_gtod_data);
+
+void update_vsyscall_tz(void)
+{
+	vsyscall_gtod_data.sys_tz = sys_tz;
+}
+
+void update_vsyscall(struct timekeeper *tk)
+{
+	struct vsyscall_gtod_data *vdata = &vsyscall_gtod_data;
+
+	write_seqcount_begin(&vdata->seq);
+
+	/* copy vsyscall data */
+	vdata->clock.vclock_mode	= tk->clock->archdata.vclock_mode;
+	vdata->clock.cycle_last		= tk->clock->cycle_last;
+	vdata->clock.mask		= tk->clock->mask;
+	vdata->clock.mult		= tk->mult;
+	vdata->clock.shift		= tk->shift;
+
+	vdata->wall_time_sec		= tk->xtime_sec;
+	vdata->wall_time_snsec		= tk->xtime_nsec;
+
+	vdata->monotonic_time_sec	= tk->xtime_sec
+					+ tk->wall_to_monotonic.tv_sec;
+	vdata->monotonic_time_snsec	= tk->xtime_nsec
+					+ (tk->wall_to_monotonic.tv_nsec
+						<< tk->shift);
+	while (vdata->monotonic_time_snsec >=
+					(((u64)NSEC_PER_SEC) << tk->shift)) {
+		vdata->monotonic_time_snsec -=
+					((u64)NSEC_PER_SEC) << tk->shift;
+		vdata->monotonic_time_sec++;
+	}
+
+	vdata->wall_time_coarse.tv_sec	= tk->xtime_sec;
+	vdata->wall_time_coarse.tv_nsec	= (long)(tk->xtime_nsec >> tk->shift);
+
+	vdata->monotonic_time_coarse	= timespec_add(vdata->wall_time_coarse,
+							tk->wall_to_monotonic);
+
+	write_seqcount_end(&vdata->seq);
+}
+
-- 
1.8.5.3

[PATCH 2/8] Add new func _install_special_mapping() to mmap.c

[stefani]

From: Stefani Seibold <stefani@seibold.net>

The _install_special_mapping() is the new base function for
install_special_mapping(). This function will return a pointer of the
created VMA or a error code in an ERR_PTR()

This new function will be needed by the for the vdso 32 bit support to map the
additonal vvar and hpet pages into the 32 bit address space. This will be done
with io_remap_pfn_range() and remap_pfn_range, which requieres a vm_area_struct.

Signed-off-by: Stefani Seibold <stefani@seibold.net>
---
 include/linux/mm.h |  3 +++
 mm/mmap.c          | 20 ++++++++++++++++----
 2 files changed, 19 insertions(+), 4 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index f28f46e..55342aa 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -1740,6 +1740,9 @@ extern void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file);
 extern struct file *get_mm_exe_file(struct mm_struct *mm);
 
 extern int may_expand_vm(struct mm_struct *mm, unsigned long npages);
+extern struct vm_area_struct *_install_special_mapping(struct mm_struct *mm,
+				   unsigned long addr, unsigned long len,
+				   unsigned long flags, struct page **pages);
 extern int install_special_mapping(struct mm_struct *mm,
 				   unsigned long addr, unsigned long len,
 				   unsigned long flags, struct page **pages);
diff --git a/mm/mmap.c b/mm/mmap.c
index 20ff0c3..81ba54f 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -2918,7 +2918,7 @@ static const struct vm_operations_struct special_mapping_vmops = {
  * The array pointer and the pages it points to are assumed to stay alive
  * for as long as this mapping might exist.
  */
-int install_special_mapping(struct mm_struct *mm,
+struct vm_area_struct *_install_special_mapping(struct mm_struct *mm,
 			    unsigned long addr, unsigned long len,
 			    unsigned long vm_flags, struct page **pages)
 {
@@ -2927,7 +2927,7 @@ int install_special_mapping(struct mm_struct *mm,
 
 	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
 	if (unlikely(vma == NULL))
-		return -ENOMEM;
+		return ERR_PTR(-ENOMEM);
 
 	INIT_LIST_HEAD(&vma->anon_vma_chain);
 	vma->vm_mm = mm;
@@ -2948,11 +2948,23 @@ int install_special_mapping(struct mm_struct *mm,
 
 	perf_event_mmap(vma);
 
-	return 0;
+	return vma;
 
 out:
 	kmem_cache_free(vm_area_cachep, vma);
-	return ret;
+	return ERR_PTR(ret);
+}
+
+int install_special_mapping(struct mm_struct *mm,
+			    unsigned long addr, unsigned long len,
+			    unsigned long vm_flags, struct page **pages)
+{
+	struct vm_area_struct *vma = _install_special_mapping(mm,
+			    addr, len, vm_flags, pages);
+
+	if (IS_ERR(vma))
+		return PTR_ERR(vma);
+	return 0;
 }
 
 static DEFINE_MUTEX(mm_all_locks_mutex);
-- 
1.8.5.3

[PATCH 3/8] revamp vclock_gettime.c

[stefani]

From: Stefani Seibold <stefani@seibold.net>

This intermediate patch revamps the vclock_gettime.c by moving some functions
around. It is only for spliting purpose, to make whole the 32 bit vdso timer
patch easier to review.

Signed-off-by: Stefani Seibold <stefani@seibold.net>
---
 arch/x86/vdso/vclock_gettime.c | 85 +++++++++++++++++++++---------------------
 1 file changed, 42 insertions(+), 43 deletions(-)

diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
index eb5d7a5..bbc8065 100644
--- a/arch/x86/vdso/vclock_gettime.c
+++ b/arch/x86/vdso/vclock_gettime.c
@@ -26,41 +26,26 @@
 
 #define gtod (&VVAR(vsyscall_gtod_data))
 
-notrace static cycle_t vread_tsc(void)
+static notrace cycle_t vread_hpet(void)
 {
-	cycle_t ret;
-	u64 last;
-
-	/*
-	 * Empirically, a fence (of type that depends on the CPU)
-	 * before rdtsc is enough to ensure that rdtsc is ordered
-	 * with respect to loads.  The various CPU manuals are unclear
-	 * as to whether rdtsc can be reordered with later loads,
-	 * but no one has ever seen it happen.
-	 */
-	rdtsc_barrier();
-	ret = (cycle_t)vget_cycles();
-
-	last = VVAR(vsyscall_gtod_data).clock.cycle_last;
-
-	if (likely(ret >= last))
-		return ret;
+	return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + HPET_COUNTER);
+}
 
-	/*
-	 * GCC likes to generate cmov here, but this branch is extremely
-	 * predictable (it's just a funciton of time and the likely is
-	 * very likely) and there's a data dependence, so force GCC
-	 * to generate a branch instead.  I don't barrier() because
-	 * we don't actually need a barrier, and if this function
-	 * ever gets inlined it will generate worse code.
-	 */
-	asm volatile ("");
-	return last;
+notrace static long vdso_fallback_gettime(long clock, struct timespec *ts)
+{
+	long ret;
+	asm("syscall" : "=a" (ret) :
+	    "0" (__NR_clock_gettime), "D" (clock), "S" (ts) : "memory");
+	return ret;
 }
 
-static notrace cycle_t vread_hpet(void)
+notrace static long vdso_fallback_gtod(struct timeval *tv, struct timezone *tz)
 {
-	return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + HPET_COUNTER);
+	long ret;
+
+	asm("syscall" : "=a" (ret) :
+	    "0" (__NR_gettimeofday), "D" (tv), "S" (tz) : "memory");
+	return ret;
 }
 
 #ifdef CONFIG_PARAVIRT_CLOCK
@@ -133,23 +118,37 @@ static notrace cycle_t vread_pvclock(int *mode)
 }
 #endif
 
-notrace static long vdso_fallback_gettime(long clock, struct timespec *ts)
+notrace static cycle_t vread_tsc(void)
 {
-	long ret;
-	asm("syscall" : "=a" (ret) :
-	    "0" (__NR_clock_gettime),"D" (clock), "S" (ts) : "memory");
-	return ret;
-}
+	cycle_t ret;
+	u64 last;
 
-notrace static long vdso_fallback_gtod(struct timeval *tv, struct timezone *tz)
-{
-	long ret;
+	/*
+	 * Empirically, a fence (of type that depends on the CPU)
+	 * before rdtsc is enough to ensure that rdtsc is ordered
+	 * with respect to loads.  The various CPU manuals are unclear
+	 * as to whether rdtsc can be reordered with later loads,
+	 * but no one has ever seen it happen.
+	 */
+	rdtsc_barrier();
+	ret = (cycle_t)vget_cycles();
 
-	asm("syscall" : "=a" (ret) :
-	    "0" (__NR_gettimeofday), "D" (tv), "S" (tz) : "memory");
-	return ret;
-}
+	last = VVAR(vsyscall_gtod_data).clock.cycle_last;
 
+	if (likely(ret >= last))
+		return ret;
+
+	/*
+	 * GCC likes to generate cmov here, but this branch is extremely
+	 * predictable (it's just a funciton of time and the likely is
+	 * very likely) and there's a data dependence, so force GCC
+	 * to generate a branch instead.  I don't barrier() because
+	 * we don't actually need a barrier, and if this function
+	 * ever gets inlined it will generate worse code.
+	 */
+	asm volatile ("");
+	return last;
+}
 
 notrace static inline u64 vgetsns(int *mode)
 {
-- 
1.8.5.3

Re: [PATCH 3/8] revamp vclock_gettime.c

[Andy Lutomirski]

On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
> From: Stefani Seibold <stefani@seibold.net>
>
> This intermediate patch revamps the vclock_gettime.c by moving some functions
> around. It is only for spliting purpose, to make whole the 32 bit vdso timer
> patch easier to review.
>
> Signed-off-by: Stefani Seibold <stefani@seibold.net>

Acked-by: Andy Lutomirski <luto@amacapital.net>

> ---
>  arch/x86/vdso/vclock_gettime.c | 85 +++++++++++++++++++++---------------------
>  1 file changed, 42 insertions(+), 43 deletions(-)
>
> diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
> index eb5d7a5..bbc8065 100644
> --- a/arch/x86/vdso/vclock_gettime.c
> +++ b/arch/x86/vdso/vclock_gettime.c
> @@ -26,41 +26,26 @@
>
>  #define gtod (&VVAR(vsyscall_gtod_data))
>
> -notrace static cycle_t vread_tsc(void)
> +static notrace cycle_t vread_hpet(void)
>  {
> -       cycle_t ret;
> -       u64 last;
> -
> -       /*
> -        * Empirically, a fence (of type that depends on the CPU)
> -        * before rdtsc is enough to ensure that rdtsc is ordered
> -        * with respect to loads.  The various CPU manuals are unclear
> -        * as to whether rdtsc can be reordered with later loads,
> -        * but no one has ever seen it happen.
> -        */
> -       rdtsc_barrier();
> -       ret = (cycle_t)vget_cycles();
> -
> -       last = VVAR(vsyscall_gtod_data).clock.cycle_last;
> -
> -       if (likely(ret >= last))
> -               return ret;
> +       return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + HPET_COUNTER);
> +}
>
> -       /*
> -        * GCC likes to generate cmov here, but this branch is extremely
> -        * predictable (it's just a funciton of time and the likely is
> -        * very likely) and there's a data dependence, so force GCC
> -        * to generate a branch instead.  I don't barrier() because
> -        * we don't actually need a barrier, and if this function
> -        * ever gets inlined it will generate worse code.
> -        */
> -       asm volatile ("");
> -       return last;
> +notrace static long vdso_fallback_gettime(long clock, struct timespec *ts)
> +{
> +       long ret;
> +       asm("syscall" : "=a" (ret) :
> +           "0" (__NR_clock_gettime), "D" (clock), "S" (ts) : "memory");
> +       return ret;
>  }
>
> -static notrace cycle_t vread_hpet(void)
> +notrace static long vdso_fallback_gtod(struct timeval *tv, struct timezone *tz)
>  {
> -       return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + HPET_COUNTER);
> +       long ret;
> +
> +       asm("syscall" : "=a" (ret) :
> +           "0" (__NR_gettimeofday), "D" (tv), "S" (tz) : "memory");
> +       return ret;
>  }
>
>  #ifdef CONFIG_PARAVIRT_CLOCK
> @@ -133,23 +118,37 @@ static notrace cycle_t vread_pvclock(int *mode)
>  }
>  #endif
>
> -notrace static long vdso_fallback_gettime(long clock, struct timespec *ts)
> +notrace static cycle_t vread_tsc(void)
>  {
> -       long ret;
> -       asm("syscall" : "=a" (ret) :
> -           "0" (__NR_clock_gettime),"D" (clock), "S" (ts) : "memory");
> -       return ret;
> -}
> +       cycle_t ret;
> +       u64 last;
>
> -notrace static long vdso_fallback_gtod(struct timeval *tv, struct timezone *tz)
> -{
> -       long ret;
> +       /*
> +        * Empirically, a fence (of type that depends on the CPU)
> +        * before rdtsc is enough to ensure that rdtsc is ordered
> +        * with respect to loads.  The various CPU manuals are unclear
> +        * as to whether rdtsc can be reordered with later loads,
> +        * but no one has ever seen it happen.
> +        */
> +       rdtsc_barrier();
> +       ret = (cycle_t)vget_cycles();
>
> -       asm("syscall" : "=a" (ret) :
> -           "0" (__NR_gettimeofday), "D" (tv), "S" (tz) : "memory");
> -       return ret;
> -}
> +       last = VVAR(vsyscall_gtod_data).clock.cycle_last;
>
> +       if (likely(ret >= last))
> +               return ret;
> +
> +       /*
> +        * GCC likes to generate cmov here, but this branch is extremely
> +        * predictable (it's just a funciton of time and the likely is
> +        * very likely) and there's a data dependence, so force GCC
> +        * to generate a branch instead.  I don't barrier() because
> +        * we don't actually need a barrier, and if this function
> +        * ever gets inlined it will generate worse code.
> +        */
> +       asm volatile ("");
> +       return last;
> +}
>
>  notrace static inline u64 vgetsns(int *mode)
>  {
> --
> 1.8.5.3
>



-- 
Andy Lutomirski
AMA Capital Management, LLC

[PATCH 4/8] vclock_gettime.c __vdso_clock_gettime cleanup

[stefani]

From: Stefani Seibold <stefani@seibold.net>

This patch is a small code cleanup for the __vdso_clock_gettime() function.

It removes the unneeded return values from do_monotonic_coarse() and
do_realtime_coarse() and add a fallback label for doing the kernel
gettimeofday() system call.

Signed-off-by: Stefani Seibold <stefani@seibold.net>
---
 arch/x86/vdso/vclock_gettime.c | 27 ++++++++++++++-------------
 1 file changed, 14 insertions(+), 13 deletions(-)

diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
index bbc8065..fd074dd 100644
--- a/arch/x86/vdso/vclock_gettime.c
+++ b/arch/x86/vdso/vclock_gettime.c
@@ -209,7 +209,7 @@ notrace static int do_monotonic(struct timespec *ts)
 	return mode;
 }
 
-notrace static int do_realtime_coarse(struct timespec *ts)
+notrace static void do_realtime_coarse(struct timespec *ts)
 {
 	unsigned long seq;
 	do {
@@ -217,10 +217,9 @@ notrace static int do_realtime_coarse(struct timespec *ts)
 		ts->tv_sec = gtod->wall_time_coarse.tv_sec;
 		ts->tv_nsec = gtod->wall_time_coarse.tv_nsec;
 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
-	return 0;
 }
 
-notrace static int do_monotonic_coarse(struct timespec *ts)
+notrace static void do_monotonic_coarse(struct timespec *ts)
 {
 	unsigned long seq;
 	do {
@@ -228,30 +227,32 @@ notrace static int do_monotonic_coarse(struct timespec *ts)
 		ts->tv_sec = gtod->monotonic_time_coarse.tv_sec;
 		ts->tv_nsec = gtod->monotonic_time_coarse.tv_nsec;
 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
-
-	return 0;
 }
 
 notrace int __vdso_clock_gettime(clockid_t clock, struct timespec *ts)
 {
-	int ret = VCLOCK_NONE;
-
 	switch (clock) {
 	case CLOCK_REALTIME:
-		ret = do_realtime(ts);
+		if (do_realtime(ts) == VCLOCK_NONE)
+			goto fallback;
 		break;
 	case CLOCK_MONOTONIC:
-		ret = do_monotonic(ts);
+		if (do_monotonic(ts) == VCLOCK_NONE)
+			goto fallback;
 		break;
 	case CLOCK_REALTIME_COARSE:
-		return do_realtime_coarse(ts);
+		do_realtime_coarse(ts);
+		break;
 	case CLOCK_MONOTONIC_COARSE:
-		return do_monotonic_coarse(ts);
+		do_monotonic_coarse(ts);
+		break;
+	default:
+		goto fallback;
 	}
 
-	if (ret == VCLOCK_NONE)
-		return vdso_fallback_gettime(clock, ts);
 	return 0;
+fallback:
+	return vdso_fallback_gettime(clock, ts);
 }
 int clock_gettime(clockid_t, struct timespec *)
 	__attribute__((weak, alias("__vdso_clock_gettime")));
-- 
1.8.5.3

Re: [PATCH 4/8] vclock_gettime.c __vdso_clock_gettime cleanup

[Andy Lutomirski]

On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
> From: Stefani Seibold <stefani@seibold.net>
>
> This patch is a small code cleanup for the __vdso_clock_gettime() function.
>
> It removes the unneeded return values from do_monotonic_coarse() and
> do_realtime_coarse() and add a fallback label for doing the kernel
> gettimeofday() system call.
>

This is possibly worth benchmarking, but it looks harmless, and
possibly even beneficial, to me.  (If the optimizer is in a good mood,
it might do a nicer job with do_monotonic now.)

--Andy

> Signed-off-by: Stefani Seibold <stefani@seibold.net>
> ---
>  arch/x86/vdso/vclock_gettime.c | 27 ++++++++++++++-------------
>  1 file changed, 14 insertions(+), 13 deletions(-)
>
> diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
> index bbc8065..fd074dd 100644
> --- a/arch/x86/vdso/vclock_gettime.c
> +++ b/arch/x86/vdso/vclock_gettime.c
> @@ -209,7 +209,7 @@ notrace static int do_monotonic(struct timespec *ts)
>         return mode;
>  }
>
> -notrace static int do_realtime_coarse(struct timespec *ts)
> +notrace static void do_realtime_coarse(struct timespec *ts)
>  {
>         unsigned long seq;
>         do {
> @@ -217,10 +217,9 @@ notrace static int do_realtime_coarse(struct timespec *ts)
>                 ts->tv_sec = gtod->wall_time_coarse.tv_sec;
>                 ts->tv_nsec = gtod->wall_time_coarse.tv_nsec;
>         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
> -       return 0;
>  }
>
> -notrace static int do_monotonic_coarse(struct timespec *ts)
> +notrace static void do_monotonic_coarse(struct timespec *ts)
>  {
>         unsigned long seq;
>         do {
> @@ -228,30 +227,32 @@ notrace static int do_monotonic_coarse(struct timespec *ts)
>                 ts->tv_sec = gtod->monotonic_time_coarse.tv_sec;
>                 ts->tv_nsec = gtod->monotonic_time_coarse.tv_nsec;
>         } while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
> -
> -       return 0;
>  }
>
>  notrace int __vdso_clock_gettime(clockid_t clock, struct timespec *ts)
>  {
> -       int ret = VCLOCK_NONE;
> -
>         switch (clock) {
>         case CLOCK_REALTIME:
> -               ret = do_realtime(ts);
> +               if (do_realtime(ts) == VCLOCK_NONE)
> +                       goto fallback;
>                 break;
>         case CLOCK_MONOTONIC:
> -               ret = do_monotonic(ts);
> +               if (do_monotonic(ts) == VCLOCK_NONE)
> +                       goto fallback;
>                 break;
>         case CLOCK_REALTIME_COARSE:
> -               return do_realtime_coarse(ts);
> +               do_realtime_coarse(ts);
> +               break;
>         case CLOCK_MONOTONIC_COARSE:
> -               return do_monotonic_coarse(ts);
> +               do_monotonic_coarse(ts);
> +               break;
> +       default:
> +               goto fallback;
>         }
>
> -       if (ret == VCLOCK_NONE)
> -               return vdso_fallback_gettime(clock, ts);
>         return 0;
> +fallback:
> +       return vdso_fallback_gettime(clock, ts);
>  }
>  int clock_gettime(clockid_t, struct timespec *)
>         __attribute__((weak, alias("__vdso_clock_gettime")));
> --
> 1.8.5.3
>



-- 
Andy Lutomirski
AMA Capital Management, LLC

[PATCH 5/8] replace VVAR(vsyscall_gtod_data) by gtod macro

[stefani]

From: Stefani Seibold <stefani@seibold.net>

There a currently more than 30 users of the gtod macro, so replace the
last VVAR(vsyscall_gtod_data) by gtod macro.

Signed-off-by: Stefani Seibold <stefani@seibold.net>
---
 arch/x86/vdso/vclock_gettime.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
index fd074dd..743f277 100644
--- a/arch/x86/vdso/vclock_gettime.c
+++ b/arch/x86/vdso/vclock_gettime.c
@@ -109,7 +109,7 @@ static notrace cycle_t vread_pvclock(int *mode)
 		*mode = VCLOCK_NONE;
 
 	/* refer to tsc.c read_tsc() comment for rationale */
-	last = VVAR(vsyscall_gtod_data).clock.cycle_last;
+	last = gtod->clock.cycle_last;
 
 	if (likely(ret >= last))
 		return ret;
@@ -133,7 +133,7 @@ notrace static cycle_t vread_tsc(void)
 	rdtsc_barrier();
 	ret = (cycle_t)vget_cycles();
 
-	last = VVAR(vsyscall_gtod_data).clock.cycle_last;
+	last = gtod->clock.cycle_last;
 
 	if (likely(ret >= last))
 		return ret;
@@ -288,7 +288,7 @@ int gettimeofday(struct timeval *, struct timezone *)
 notrace time_t __vdso_time(time_t *t)
 {
 	/* This is atomic on x86_64 so we don't need any locks. */
-	time_t result = ACCESS_ONCE(VVAR(vsyscall_gtod_data).wall_time_sec);
+	time_t result = ACCESS_ONCE(gtod->wall_time_sec);
 
 	if (t)
 		*t = result;
-- 
1.8.5.3

Re: [PATCH 5/8] replace VVAR(vsyscall_gtod_data) by gtod macro

[Andy Lutomirski]

On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
> From: Stefani Seibold <stefani@seibold.net>
>
> There a currently more than 30 users of the gtod macro, so replace the
> last VVAR(vsyscall_gtod_data) by gtod macro.

Acked-by: Andy Lutomirski <luto@amacapital.net>

>
> Signed-off-by: Stefani Seibold <stefani@seibold.net>
> ---
>  arch/x86/vdso/vclock_gettime.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
> index fd074dd..743f277 100644
> --- a/arch/x86/vdso/vclock_gettime.c
> +++ b/arch/x86/vdso/vclock_gettime.c
> @@ -109,7 +109,7 @@ static notrace cycle_t vread_pvclock(int *mode)
>                 *mode = VCLOCK_NONE;
>
>         /* refer to tsc.c read_tsc() comment for rationale */
> -       last = VVAR(vsyscall_gtod_data).clock.cycle_last;
> +       last = gtod->clock.cycle_last;
>
>         if (likely(ret >= last))
>                 return ret;
> @@ -133,7 +133,7 @@ notrace static cycle_t vread_tsc(void)
>         rdtsc_barrier();
>         ret = (cycle_t)vget_cycles();
>
> -       last = VVAR(vsyscall_gtod_data).clock.cycle_last;
> +       last = gtod->clock.cycle_last;
>
>         if (likely(ret >= last))
>                 return ret;
> @@ -288,7 +288,7 @@ int gettimeofday(struct timeval *, struct timezone *)
>  notrace time_t __vdso_time(time_t *t)
>  {
>         /* This is atomic on x86_64 so we don't need any locks. */
> -       time_t result = ACCESS_ONCE(VVAR(vsyscall_gtod_data).wall_time_sec);
> +       time_t result = ACCESS_ONCE(gtod->wall_time_sec);
>
>         if (t)
>                 *t = result;
> --
> 1.8.5.3
>



-- 
Andy Lutomirski
AMA Capital Management, LLC

[PATCH 6/8] cleanup __vdso_gettimeofday

[stefani]

From: Stefani Seibold <stefani@seibold.net>

This patch do a little cleanup for the __vdso_gettimeofday() function.

It kick out an unneeded ret local variable and makes the code faster if
only the timezone is needed.

Signed-off-by: Stefani Seibold <stefani@seibold.net>
---
 arch/x86/vdso/vclock_gettime.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
index 743f277..bf969a0 100644
--- a/arch/x86/vdso/vclock_gettime.c
+++ b/arch/x86/vdso/vclock_gettime.c
@@ -259,13 +259,12 @@ int clock_gettime(clockid_t, struct timespec *)
 
 notrace int __vdso_gettimeofday(struct timeval *tv, struct timezone *tz)
 {
-	long ret = VCLOCK_NONE;
-
 	if (likely(tv != NULL)) {
 		BUILD_BUG_ON(offsetof(struct timeval, tv_usec) !=
 			     offsetof(struct timespec, tv_nsec) ||
 			     sizeof(*tv) != sizeof(struct timespec));
-		ret = do_realtime((struct timespec *)tv);
+		if (do_realtime((struct timespec *)tv) == VCLOCK_NONE)
+			return vdso_fallback_gtod(tv, tz);
 		tv->tv_usec /= 1000;
 	}
 	if (unlikely(tz != NULL)) {
@@ -274,8 +273,6 @@ notrace int __vdso_gettimeofday(struct timeval *tv, struct timezone *tz)
 		tz->tz_dsttime = gtod->sys_tz.tz_dsttime;
 	}
 
-	if (ret == VCLOCK_NONE)
-		return vdso_fallback_gtod(tv, tz);
 	return 0;
 }
 int gettimeofday(struct timeval *, struct timezone *)
-- 
1.8.5.3

Re: [PATCH 6/8] cleanup __vdso_gettimeofday

[H. Peter Anvin]

Even reading the timezone via gettimeofday is the uncommon case... never mind only the timezome.  Whatever you do please don't slow down the common case.

On February 2, 2014 3:27:13 AM PST, stefani@seibold.net wrote:
>From: Stefani Seibold <stefani@seibold.net>
>
>This patch do a little cleanup for the __vdso_gettimeofday() function.
>
>It kick out an unneeded ret local variable and makes the code faster if
>only the timezone is needed.
>
>Signed-off-by: Stefani Seibold <stefani@seibold.net>
>---
> arch/x86/vdso/vclock_gettime.c | 7 ++-----
> 1 file changed, 2 insertions(+), 5 deletions(-)
>
>diff --git a/arch/x86/vdso/vclock_gettime.c
>b/arch/x86/vdso/vclock_gettime.c
>index 743f277..bf969a0 100644
>--- a/arch/x86/vdso/vclock_gettime.c
>+++ b/arch/x86/vdso/vclock_gettime.c
>@@ -259,13 +259,12 @@ int clock_gettime(clockid_t, struct timespec *)
> 
>notrace int __vdso_gettimeofday(struct timeval *tv, struct timezone
>*tz)
> {
>-	long ret = VCLOCK_NONE;
>-
> 	if (likely(tv != NULL)) {
> 		BUILD_BUG_ON(offsetof(struct timeval, tv_usec) !=
> 			     offsetof(struct timespec, tv_nsec) ||
> 			     sizeof(*tv) != sizeof(struct timespec));
>-		ret = do_realtime((struct timespec *)tv);
>+		if (do_realtime((struct timespec *)tv) == VCLOCK_NONE)
>+			return vdso_fallback_gtod(tv, tz);
> 		tv->tv_usec /= 1000;
> 	}
> 	if (unlikely(tz != NULL)) {
>@@ -274,8 +273,6 @@ notrace int __vdso_gettimeofday(struct timeval *tv,
>struct timezone *tz)
> 		tz->tz_dsttime = gtod->sys_tz.tz_dsttime;
> 	}
> 
>-	if (ret == VCLOCK_NONE)
>-		return vdso_fallback_gtod(tv, tz);
> 	return 0;
> }
> int gettimeofday(struct timeval *, struct timezone *)

-- 
Sent from my mobile phone.  Please pardon brevity and lack of formatting.

Re: [PATCH 6/8] cleanup __vdso_gettimeofday

[Stefani Seibold]

On Sun, 2014-02-02 at 07:43 -0800, H. Peter Anvin wrote:
> Even reading the timezone via gettimeofday is the uncommon case... never mind only the timezome.  Whatever you do please don't slow down the common case.
> 
> On February 2, 2014 3:27:13 AM PST, stefani@seibold.net wrote:
> >From: Stefani Seibold <stefani@seibold.net>
> >
> >This patch do a little cleanup for the __vdso_gettimeofday() function.
> >
> >It kick out an unneeded ret local variable and makes the code faster if
> >only the timezone is needed.
> >
> >Signed-off-by: Stefani Seibold <stefani@seibold.net>
> >---
> > arch/x86/vdso/vclock_gettime.c | 7 ++-----
> > 1 file changed, 2 insertions(+), 5 deletions(-)
> >
> >diff --git a/arch/x86/vdso/vclock_gettime.c
> >b/arch/x86/vdso/vclock_gettime.c
> >index 743f277..bf969a0 100644
> >--- a/arch/x86/vdso/vclock_gettime.c
> >+++ b/arch/x86/vdso/vclock_gettime.c
> >@@ -259,13 +259,12 @@ int clock_gettime(clockid_t, struct timespec *)
> > 
> >notrace int __vdso_gettimeofday(struct timeval *tv, struct timezone
> >*tz)
> > {
> >-	long ret = VCLOCK_NONE;
> >-
> > 	if (likely(tv != NULL)) {
> > 		BUILD_BUG_ON(offsetof(struct timeval, tv_usec) !=
> > 			     offsetof(struct timespec, tv_nsec) ||
> > 			     sizeof(*tv) != sizeof(struct timespec));
> >-		ret = do_realtime((struct timespec *)tv);
> >+		if (do_realtime((struct timespec *)tv) == VCLOCK_NONE)
> >+			return vdso_fallback_gtod(tv, tz);
> > 		tv->tv_usec /= 1000;
> > 	}
> > 	if (unlikely(tz != NULL)) {
> >@@ -274,8 +273,6 @@ notrace int __vdso_gettimeofday(struct timeval *tv,
> >struct timezone *tz)
> > 		tz->tz_dsttime = gtod->sys_tz.tz_dsttime;
> > 	}
> > 
> >-	if (ret == VCLOCK_NONE)
> >-		return vdso_fallback_gtod(tv, tz);
> > 	return 0;
> > }
> > int gettimeofday(struct timeval *, struct timezone *)
> 

I don't see where i slow down the common case! It is exactly the same
behaviour as the original code. Only in the uncommon case where the tv
== NULL and tz != NULL is a bit faster.

Re: [PATCH 6/8] cleanup __vdso_gettimeofday

[Andy Lutomirski]

On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
> From: Stefani Seibold <stefani@seibold.net>
>
> This patch do a little cleanup for the __vdso_gettimeofday() function.
>
> It kick out an unneeded ret local variable and makes the code faster if
> only the timezone is needed.
>
> Signed-off-by: Stefani Seibold <stefani@seibold.net>

Acked-by: Andy Lutomirski <luto@amacapital.net>

> ---
>  arch/x86/vdso/vclock_gettime.c | 7 ++-----
>  1 file changed, 2 insertions(+), 5 deletions(-)
>
> diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
> index 743f277..bf969a0 100644
> --- a/arch/x86/vdso/vclock_gettime.c
> +++ b/arch/x86/vdso/vclock_gettime.c
> @@ -259,13 +259,12 @@ int clock_gettime(clockid_t, struct timespec *)
>
>  notrace int __vdso_gettimeofday(struct timeval *tv, struct timezone *tz)
>  {
> -       long ret = VCLOCK_NONE;
> -
>         if (likely(tv != NULL)) {
>                 BUILD_BUG_ON(offsetof(struct timeval, tv_usec) !=
>                              offsetof(struct timespec, tv_nsec) ||
>                              sizeof(*tv) != sizeof(struct timespec));
> -               ret = do_realtime((struct timespec *)tv);
> +               if (do_realtime((struct timespec *)tv) == VCLOCK_NONE)
> +                       return vdso_fallback_gtod(tv, tz);
>                 tv->tv_usec /= 1000;
>         }
>         if (unlikely(tz != NULL)) {
> @@ -274,8 +273,6 @@ notrace int __vdso_gettimeofday(struct timeval *tv, struct timezone *tz)
>                 tz->tz_dsttime = gtod->sys_tz.tz_dsttime;
>         }
>
> -       if (ret == VCLOCK_NONE)
> -               return vdso_fallback_gtod(tv, tz);
>         return 0;
>  }
>  int gettimeofday(struct timeval *, struct timezone *)
> --
> 1.8.5.3
>



-- 
Andy Lutomirski
AMA Capital Management, LLC

[PATCH 7/8] Add 32 bit VDSO time support for 32 bit kernel

[stefani]

From: Stefani Seibold <stefani@seibold.net>

This patch add the time support for 32 bit a VDSO to a 32 bit kernel.

For 32 bit programs running on a 32 bit kernel, the same mechanism is
used as for 64 bit programs running on a 64 bit kernel.

Signed-off-by: Stefani Seibold <stefani@seibold.net>
---
 arch/x86/include/asm/elf.h            |  2 +-
 arch/x86/include/asm/vdso.h           |  3 ++
 arch/x86/include/asm/vdso32.h         | 10 ++++++
 arch/x86/vdso/Makefile                |  7 +++++
 arch/x86/vdso/vclock_gettime.c        | 59 +++++++++++++++++++++++++++++++++--
 arch/x86/vdso/vdso-layout.lds.S       | 25 +++++++++++++++
 arch/x86/vdso/vdso32-setup.c          | 47 ++++++++++++++++++++++------
 arch/x86/vdso/vdso32/vclock_gettime.c | 16 ++++++++++
 arch/x86/vdso/vdso32/vdso32.lds.S     | 11 ++++++-
 9 files changed, 167 insertions(+), 13 deletions(-)
 create mode 100644 arch/x86/include/asm/vdso32.h
 create mode 100644 arch/x86/vdso/vdso32/vclock_gettime.c

diff --git a/arch/x86/include/asm/elf.h b/arch/x86/include/asm/elf.h
index 9c999c1..21bae90 100644
--- a/arch/x86/include/asm/elf.h
+++ b/arch/x86/include/asm/elf.h
@@ -289,7 +289,7 @@ do {									\
 
 #else /* CONFIG_X86_32 */
 
-#define VDSO_HIGH_BASE		0xffffe000U /* CONFIG_COMPAT_VDSO address */
+#define VDSO_HIGH_BASE		0xffffc000U /* CONFIG_COMPAT_VDSO address */
 
 /* 1GB for 64bit, 8MB for 32bit */
 #define STACK_RND_MASK (test_thread_flag(TIF_ADDR32) ? 0x7ff : 0x3fffff)
diff --git a/arch/x86/include/asm/vdso.h b/arch/x86/include/asm/vdso.h
index fddb53d..fe3cef9 100644
--- a/arch/x86/include/asm/vdso.h
+++ b/arch/x86/include/asm/vdso.h
@@ -2,6 +2,9 @@
 #define _ASM_X86_VDSO_H
 
 #if defined CONFIG_X86_32 || defined CONFIG_COMPAT
+
+#include <asm/vdso32.h>
+
 extern const char VDSO32_PRELINK[];
 
 /*
diff --git a/arch/x86/include/asm/vdso32.h b/arch/x86/include/asm/vdso32.h
new file mode 100644
index 0000000..7dd2eb8
--- /dev/null
+++ b/arch/x86/include/asm/vdso32.h
@@ -0,0 +1,10 @@
+#ifndef _ASM_X86_VDSO32_H
+#define _ASM_X86_VDSO32_H
+
+#define VDSO_BASE_PAGE	0
+#define VDSO_VVAR_PAGE	1
+#define VDSO_HPET_PAGE	2
+#define	VDSO_PAGES	3
+#define	VDSO_OFFSET(x)	((x) * PAGE_SIZE)
+
+#endif
diff --git a/arch/x86/vdso/Makefile b/arch/x86/vdso/Makefile
index fd14be1..1ff5b0a 100644
--- a/arch/x86/vdso/Makefile
+++ b/arch/x86/vdso/Makefile
@@ -145,8 +145,15 @@ KBUILD_AFLAGS_32 := $(filter-out -m64,$(KBUILD_AFLAGS))
 $(vdso32-images:%=$(obj)/%.dbg): KBUILD_AFLAGS = $(KBUILD_AFLAGS_32)
 $(vdso32-images:%=$(obj)/%.dbg): asflags-$(CONFIG_X86_64) += -m32
 
+KBUILD_CFLAGS_32 := $(filter-out -m64,$(KBUILD_CFLAGS))
+KBUILD_CFLAGS_32 := $(filter-out -mcmodel=kernel,$(KBUILD_CFLAGS_32))
+KBUILD_CFLAGS_32 := $(filter-out -fno-pic,$(KBUILD_CFLAGS_32))
+KBUILD_CFLAGS_32 += -m32 -msoft-float -mregparm=3 -freg-struct-return -fpic
+$(vdso32-images:%=$(obj)/%.dbg): KBUILD_CFLAGS = $(KBUILD_CFLAGS_32)
+
 $(vdso32-images:%=$(obj)/%.dbg): $(obj)/vdso32-%.so.dbg: FORCE \
 				 $(obj)/vdso32/vdso32.lds \
+				 $(obj)/vdso32/vclock_gettime.o \
 				 $(obj)/vdso32/note.o \
 				 $(obj)/vdso32/%.o
 	$(call if_changed,vdso)
diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
index bf969a0..42f641c 100644
--- a/arch/x86/vdso/vclock_gettime.c
+++ b/arch/x86/vdso/vclock_gettime.c
@@ -4,6 +4,9 @@
  *
  * Fast user context implementation of clock_gettime, gettimeofday, and time.
  *
+ * 32 Bit compat layer by Stefani Seibold <stefani@seibold.net>
+ *  sponsored by Rohde & Schwarz GmbH & Co. KG Munich/Germany
+ *
  * The code should have no internal unresolved relocations.
  * Check with readelf after changing.
  */
@@ -24,6 +27,8 @@
 #include <asm/io.h>
 #include <asm/pvclock.h>
 
+#ifndef BUILD_VDSO32
+
 #define gtod (&VVAR(vsyscall_gtod_data))
 
 static notrace cycle_t vread_hpet(void)
@@ -47,6 +52,54 @@ notrace static long vdso_fallback_gtod(struct timeval *tv, struct timezone *tz)
 	    "0" (__NR_gettimeofday), "D" (tv), "S" (tz) : "memory");
 	return ret;
 }
+#else
+
+struct vsyscall_gtod_data vvar_vsyscall_gtod_data
+	__attribute__((visibility("hidden")));
+
+u8 hpet_page
+	__attribute__((visibility("hidden")));
+
+#define gtod (&vvar_vsyscall_gtod_data)
+
+#ifdef CONFIG_HPET_TIMER
+static notrace cycle_t vread_hpet(void)
+{
+	return readl(&hpet_page + HPET_COUNTER);
+}
+#endif
+
+notrace static long vdso_fallback_gettime(long clock, struct timespec *ts)
+{
+	long ret;
+
+	asm(
+		"push %%ebx \n"
+		"mov %2,%%ebx \n"
+		"call VDSO32_vsyscall \n"
+		"pop %%ebx \n"
+		: "=a" (ret)
+		: "0" (__NR_clock_gettime), "d" (clock), "c" (ts)
+		: "memory");
+	return ret;
+}
+
+notrace static long vdso_fallback_gtod(struct timeval *tv,
+		struct timezone *tz)
+{
+	long ret;
+
+	asm(
+		"push %%ebx \n"
+		"mov %2,%%ebx \n"
+		"call VDSO32_vsyscall \n"
+		"pop %%ebx \n"
+		: "=a" (ret)
+		: "0" (__NR_gettimeofday), "d" (tv), "c" (tz)
+		: "memory");
+	return ret;
+}
+#endif
 
 #ifdef CONFIG_PARAVIRT_CLOCK
 
@@ -152,12 +205,14 @@ notrace static cycle_t vread_tsc(void)
 
 notrace static inline u64 vgetsns(int *mode)
 {
-	long v;
+	u64 v;
 	cycles_t cycles;
 	if (gtod->clock.vclock_mode == VCLOCK_TSC)
 		cycles = vread_tsc();
+#ifdef CONFIG_HPET_TIMER
 	else if (gtod->clock.vclock_mode == VCLOCK_HPET)
 		cycles = vread_hpet();
+#endif
 #ifdef CONFIG_PARAVIRT_CLOCK
 	else if (gtod->clock.vclock_mode == VCLOCK_PVCLOCK)
 		cycles = vread_pvclock(mode);
@@ -284,7 +339,7 @@ int gettimeofday(struct timeval *, struct timezone *)
  */
 notrace time_t __vdso_time(time_t *t)
 {
-	/* This is atomic on x86_64 so we don't need any locks. */
+	/* This is atomic on x86 so we don't need any locks. */
 	time_t result = ACCESS_ONCE(gtod->wall_time_sec);
 
 	if (t)
diff --git a/arch/x86/vdso/vdso-layout.lds.S b/arch/x86/vdso/vdso-layout.lds.S
index 634a2cf..086567b 100644
--- a/arch/x86/vdso/vdso-layout.lds.S
+++ b/arch/x86/vdso/vdso-layout.lds.S
@@ -44,6 +44,27 @@ SECTIONS
 	. = ALIGN(0x100);
 
 	.text		: { *(.text*) }			:text	=0x90909090
+
+#ifdef BUILD_VDSO32
+	. = ALIGN(PAGE_SIZE);
+
+	.vvar_sect : {
+		vvar = .;
+
+	/* Place all vvars at the offsets in asm/vvar.h. */
+#define EMIT_VVAR(name, offset) vvar_ ## name = vvar + offset;
+#define __VVAR_KERNEL_LDS
+#include <asm/vvar.h>
+#undef __VVAR_KERNEL_LDS
+#undef EMIT_VVAR
+	} :text :vvar_sect
+
+	. += PAGE_SIZE;
+
+	.hpet_sect : {
+		hpet_page = .;
+	} :text :hpet_sect
+#endif
 }
 
 /*
@@ -61,4 +82,8 @@ PHDRS
 	dynamic		PT_DYNAMIC	FLAGS(4);		/* PF_R */
 	note		PT_NOTE		FLAGS(4);		/* PF_R */
 	eh_frame_hdr	PT_GNU_EH_FRAME;
+#ifdef BUILD_VDSO32
+	vvar_sect	PT_NULL		FLAGS(4);		/* PF_R */
+	hpet_sect	PT_NULL		FLAGS(4);		/* PF_R */
+#endif
 }
diff --git a/arch/x86/vdso/vdso32-setup.c b/arch/x86/vdso/vdso32-setup.c
index d6bfb87..eb2050c 100644
--- a/arch/x86/vdso/vdso32-setup.c
+++ b/arch/x86/vdso/vdso32-setup.c
@@ -25,6 +25,7 @@
 #include <asm/tlbflush.h>
 #include <asm/vdso.h>
 #include <asm/proto.h>
+#include <asm/fixmap.h>
 
 enum {
 	VDSO_DISABLED = 0,
@@ -193,7 +194,7 @@ static __init void relocate_vdso(Elf32_Ehdr *ehdr)
 	}
 }
 
-static struct page *vdso32_pages[1];
+static struct page *vdso32_pages[VDSO_PAGES];
 
 #ifdef CONFIG_X86_64
 
@@ -310,6 +311,11 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 	unsigned long addr;
 	int ret = 0;
 	bool compat;
+	struct vm_area_struct *vma;
+	extern char __vvar_page;
+#ifdef CONFIG_HPET_TIMER
+	extern unsigned long hpet_address;
+#endif
 
 #ifdef CONFIG_X86_X32_ABI
 	if (test_thread_flag(TIF_X32))
@@ -330,7 +336,7 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 	if (compat)
 		addr = VDSO_HIGH_BASE;
 	else {
-		addr = get_unmapped_area(NULL, 0, PAGE_SIZE, 0, 0);
+		addr = get_unmapped_area(NULL, 0, VDSO_OFFSET(VDSO_PAGES), 0, 0);
 		if (IS_ERR_VALUE(addr)) {
 			ret = addr;
 			goto up_fail;
@@ -340,16 +346,39 @@ int arch_setup_additional_pages(struct linux_binprm *bprm, int uses_interp)
 	current->mm->context.vdso = (void *)addr;
 
 	if (compat_uses_vma || !compat) {
-		/*
-		 * MAYWRITE to allow gdb to COW and set breakpoints
-		 */
-		ret = install_special_mapping(mm, addr, PAGE_SIZE,
-					      VM_READ|VM_EXEC|
-					      VM_MAYREAD|VM_MAYWRITE|VM_MAYEXEC,
-					      vdso32_pages);
+
+		vma = _install_special_mapping(mm,
+			addr,
+			VDSO_OFFSET(VDSO_PAGES),
+			VM_READ|VM_EXEC,
+			vdso32_pages);
+
+		if (IS_ERR(vma)) {
+			ret = PTR_ERR(vma);
+			goto up_fail;
+		}
+
+		ret = remap_pfn_range(vma,
+			vma->vm_start + VDSO_OFFSET(VDSO_VVAR_PAGE),
+			__pa_symbol(&__vvar_page) >> PAGE_SHIFT,
+			PAGE_SIZE,
+			PAGE_READONLY);
 
 		if (ret)
 			goto up_fail;
+
+#ifdef CONFIG_HPET_TIMER
+		if (hpet_address) {
+			ret = io_remap_pfn_range(vma,
+				vma->vm_start + VDSO_OFFSET(VDSO_HPET_PAGE),
+				hpet_address >> PAGE_SHIFT,
+				PAGE_SIZE,
+				pgprot_noncached(PAGE_READONLY));
+
+			if (ret)
+				goto up_fail;
+		}
+#endif
 	}
 
 	current_thread_info()->sysenter_return =
diff --git a/arch/x86/vdso/vdso32/vclock_gettime.c b/arch/x86/vdso/vdso32/vclock_gettime.c
new file mode 100644
index 0000000..fab4ec6
--- /dev/null
+++ b/arch/x86/vdso/vdso32/vclock_gettime.c
@@ -0,0 +1,16 @@
+#define BUILD_VDSO32
+
+#ifdef CONFIG_X86_64
+
+#define _ASM_X86_PAGE_H
+
+#define __pa(x)                0
+#define __va(x)                0
+
+#undef CONFIG_ILLEGAL_POINTER_VALUE
+#define CONFIG_ILLEGAL_POINTER_VALUE   0
+
+#endif
+
+#include "../vclock_gettime.c"
+
diff --git a/arch/x86/vdso/vdso32/vdso32.lds.S b/arch/x86/vdso/vdso32/vdso32.lds.S
index 976124b..66e73b2 100644
--- a/arch/x86/vdso/vdso32/vdso32.lds.S
+++ b/arch/x86/vdso/vdso32/vdso32.lds.S
@@ -8,6 +8,9 @@
  * values visible using the asm-x86/vdso.h macros from the kernel proper.
  */
 
+#include <asm/page.h>
+
+#define BUILD_VDSO32
 #define VDSO_PRELINK 0
 #include "../vdso-layout.lds.S"
 
@@ -19,11 +22,14 @@ ENTRY(__kernel_vsyscall);
  */
 VERSION
 {
-	LINUX_2.5 {
+	LINUX_2.6 {
 	global:
 		__kernel_vsyscall;
 		__kernel_sigreturn;
 		__kernel_rt_sigreturn;
+		__vdso_clock_gettime;
+		__vdso_gettimeofday;
+		__vdso_time;
 	local: *;
 	};
 }
@@ -35,3 +41,6 @@ VDSO32_PRELINK		= VDSO_PRELINK;
 VDSO32_vsyscall		= __kernel_vsyscall;
 VDSO32_sigreturn	= __kernel_sigreturn;
 VDSO32_rt_sigreturn	= __kernel_rt_sigreturn;
+VDSO32_clock_gettime	= clock_gettime;
+VDSO32_gettimeofday	= gettimeofday;
+VDSO32_time		= time;
-- 
1.8.5.3

Re: [PATCH 7/8] Add 32 bit VDSO time support for 32 bit kernel

[Andy Lutomirski]

On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
> From: Stefani Seibold <stefani@seibold.net>
>
> This patch add the time support for 32 bit a VDSO to a 32 bit kernel.

[...]

Can you address the review comments from last time around?  For
example, this still seems to have redundant vvar and hpet mappings, it
doesn't use the VVAR macro, it moves the 32-bit compat vDSO, etc.

Also, the usual convention is to use [PATCH v2 3/8], etc, as subjects,
to keep separate versions separate.

--Andy

Re: [PATCH 7/8] Add 32 bit VDSO time support for 32 bit kernel

[Stefani Seibold]

Am Sonntag, den 02.02.2014, 08:46 -0800 schrieb Andy Lutomirski:
> On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
> > From: Stefani Seibold <stefani@seibold.net>
> >
> > This patch add the time support for 32 bit a VDSO to a 32 bit kernel.
> 
> [...]
> 
> Can you address the review comments from last time around?  For
> example, this still seems to have redundant vvar and hpet mappings, it
> doesn't use the VVAR macro, it moves the 32-bit compat vDSO, etc.
> 

I will address the compat VDSO issue.

But the VVAR macro will be not a part of this patch set. If you depend
on this, feel free to create one. From my point of view this is not
feasible without a macro hacking, because the address accessing the vvar
area differs in kernel and VDSO user mode.

I also see no redundant mapping. There are two modes, one is the map of
the kernel area the other maps the VDSO into the user space area. This
is exactly the behaviour of the origin VDSO implementation.
 
> Also, the usual convention is to use [PATCH v2 3/8], etc, as subjects,
> to keep separate versions separate.
> 

Will do this.

- Stefani

Re: [PATCH 7/8] Add 32 bit VDSO time support for 32 bit kernel

[Andy Lutomirski]

On Sun, Feb 2, 2014 at 1:39 PM, Stefani Seibold <stefani@seibold.net> wrote:
> Am Sonntag, den 02.02.2014, 08:46 -0800 schrieb Andy Lutomirski:
>> On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
>> > From: Stefani Seibold <stefani@seibold.net>
>> >
>> > This patch add the time support for 32 bit a VDSO to a 32 bit kernel.
>>
>> [...]
>>
>> Can you address the review comments from last time around?  For
>> example, this still seems to have redundant vvar and hpet mappings, it
>> doesn't use the VVAR macro, it moves the 32-bit compat vDSO, etc.
>>
>
> I will address the compat VDSO issue.
>
> But the VVAR macro will be not a part of this patch set. If you depend
> on this, feel free to create one. From my point of view this is not
> feasible without a macro hacking, because the address accessing the vvar
> area differs in kernel and VDSO user mode.

Sorry, but "I will make the code messier for no apparent reason and I
will not offer to fix it in the same series" gets my NAK.

Hint: I'm talking about two or three lines of code in vvar.h.

>
> I also see no redundant mapping. There are two modes, one is the map of
> the kernel area the other maps the VDSO into the user space area. This
> is exactly the behaviour of the origin VDSO implementation.

No.

In your series there are *three* mappings.  There are:

 - The linear mapping that the kernel loader sets up (the writable
mapping used in the kernel).  This is implicit and, of course, fine.
 - There's the fixmap page, which aliases the normal kernel mapping at
a fixed address with the user, ro, and nx attributes.  The 64-bit vDSO
uses that mapping.  See vdso.h -- it's all arranged pretty clearly.
Your code, for no discernible reason, sets up a fixmap entry on
*32-bit* kernels.
 - The vma that you're setting up adjacent to the actual vdso text.
This is what you are using.

Please choose *one* user-readable mapping for the 32-bit vdso and
stick with it.  If the 64-bit vdso can use it to and userspace doesn't
break, even better.  But a pointless set of extra fixmap entries is
not okay.

hpa's right about the symbol versions, BTW -- those are ABI and will
probably break (at least) every new-enough Go binary.

--Andy

Re: [PATCH 7/8] Add 32 bit VDSO time support for 32 bit kernel

[Stefani Seibold]

Am Sonntag, den 02.02.2014, 16:12 -0800 schrieb Andy Lutomirski:
> On Sun, Feb 2, 2014 at 1:39 PM, Stefani Seibold <stefani@seibold.net> wrote:
> > Am Sonntag, den 02.02.2014, 08:46 -0800 schrieb Andy Lutomirski:
> >> On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
> >> > From: Stefani Seibold <stefani@seibold.net>
> >> >
> >> > This patch add the time support for 32 bit a VDSO to a 32 bit kernel.
> >>
> >> [...]
> >>
> >> Can you address the review comments from last time around?  For
> >> example, this still seems to have redundant vvar and hpet mappings, it
> >> doesn't use the VVAR macro, it moves the 32-bit compat vDSO, etc.
> >>
> >
> > I will address the compat VDSO issue.
> >
> > But the VVAR macro will be not a part of this patch set. If you depend
> > on this, feel free to create one. From my point of view this is not
> > feasible without a macro hacking, because the address accessing the vvar
> > area differs in kernel and VDSO user mode.
> 
> Sorry, but "I will make the code messier for no apparent reason and I
> will not offer to fix it in the same series" gets my NAK.
> 
> Hint: I'm talking about two or three lines of code in vvar.h.
> 

A hint back: if you threat me with a NAK for a requested code sequence
which currently no user, this is far away from professional. I am not
your trainee.

BTW: If it is so easy, send me the two or three lines and i will merge
it ;-)

> >
> > I also see no redundant mapping. There are two modes, one is the map of
> > the kernel area the other maps the VDSO into the user space area. This
> > is exactly the behaviour of the origin VDSO implementation.
> 
> No.
> 
> In your series there are *three* mappings.  There are:
> 
>  - The linear mapping that the kernel loader sets up (the writable
> mapping used in the kernel).  This is implicit and, of course, fine.
>  - There's the fixmap page, which aliases the normal kernel mapping at
> a fixed address with the user, ro, and nx attributes.  The 64-bit vDSO
> uses that mapping.  See vdso.h -- it's all arranged pretty clearly.
> Your code, for no discernible reason, sets up a fixmap entry on
> *32-bit* kernels.
>  - The vma that you're setting up adjacent to the actual vdso text.
> This is what you are using.
> 
> Please choose *one* user-readable mapping for the 32-bit vdso and
> stick with it.  If the 64-bit vdso can use it to and userspace doesn't
> break, even better.  But a pointless set of extra fixmap entries is
> not okay.
> 

Again: I wrote that there are two modes for a 32 bit kernel and
therefore there are two mappings at the same time. Since there are both
ways available in a 32 bit kernel via the vdso32= kernel parameter, both
must be supported.

Due the lack of a real fixmap for a 32 bit kernel (FIXADDR_TOP is a
variable), the HPET and VVAR Page can only relative addressed. So this
pages must located before or after the VDSO. 

This is why i need to setup this pages into the fixmap area, this is the
compat mode "vdso32=2".

For "vdso32=1" i need to map the VDSO Page together with the HPET and
VVAR into the user space.

For compability reasons both mappings are required.

There is only one binary for the VDSO page, regardless of the vdso=
kernel parameter and this code can only do a relative addressing.

A 64 bit kernel can do it in an other way, because there is a real
fixmap area, so this special handling is not needed.

- Stefani

Re: [PATCH 7/8] Add 32 bit VDSO time support for 32 bit kernel

[Andy Lutomirski]

On Sun, Feb 2, 2014 at 11:44 PM, Stefani Seibold <stefani@seibold.net> wrote:
> Am Sonntag, den 02.02.2014, 16:12 -0800 schrieb Andy Lutomirski:
>> On Sun, Feb 2, 2014 at 1:39 PM, Stefani Seibold <stefani@seibold.net> wrote:
>> > Am Sonntag, den 02.02.2014, 08:46 -0800 schrieb Andy Lutomirski:
>> >> On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
>> >> > From: Stefani Seibold <stefani@seibold.net>
>> >> >
>> >> > This patch add the time support for 32 bit a VDSO to a 32 bit kernel.
>> >>
>> >> [...]
>> >>
>> >> Can you address the review comments from last time around?  For
>> >> example, this still seems to have redundant vvar and hpet mappings, it
>> >> doesn't use the VVAR macro, it moves the 32-bit compat vDSO, etc.
>> >>
>> >
>> > I will address the compat VDSO issue.
>> >
>> > But the VVAR macro will be not a part of this patch set. If you depend
>> > on this, feel free to create one. From my point of view this is not
>> > feasible without a macro hacking, because the address accessing the vvar
>> > area differs in kernel and VDSO user mode.
>>
>> Sorry, but "I will make the code messier for no apparent reason and I
>> will not offer to fix it in the same series" gets my NAK.
>>
>> Hint: I'm talking about two or three lines of code in vvar.h.
>>
>
> A hint back: if you threat me with a NAK for a requested code sequence
> which currently no user, this is far away from professional. I am not
> your trainee.
>
> BTW: If it is so easy, send me the two or three lines and i will merge
> it ;-)

Something to the effect of:

#elif defined(BUILD_VDSO32)
#define VVAR(name) (*vvar_ ## name)
#endif

Should do the trick.

>
>> >
>> > I also see no redundant mapping. There are two modes, one is the map of
>> > the kernel area the other maps the VDSO into the user space area. This
>> > is exactly the behaviour of the origin VDSO implementation.
>>
>> No.
>>
>> In your series there are *three* mappings.  There are:
>>
>>  - The linear mapping that the kernel loader sets up (the writable
>> mapping used in the kernel).  This is implicit and, of course, fine.
>>  - There's the fixmap page, which aliases the normal kernel mapping at
>> a fixed address with the user, ro, and nx attributes.  The 64-bit vDSO
>> uses that mapping.  See vdso.h -- it's all arranged pretty clearly.
>> Your code, for no discernible reason, sets up a fixmap entry on
>> *32-bit* kernels.
>>  - The vma that you're setting up adjacent to the actual vdso text.
>> This is what you are using.
>>
>> Please choose *one* user-readable mapping for the 32-bit vdso and
>> stick with it.  If the 64-bit vdso can use it to and userspace doesn't
>> break, even better.  But a pointless set of extra fixmap entries is
>> not okay.
>>
>
> Again: I wrote that there are two modes for a 32 bit kernel and
> therefore there are two mappings at the same time. Since there are both
> ways available in a 32 bit kernel via the vdso32= kernel parameter, both
> must be supported.
>
> Due the lack of a real fixmap for a 32 bit kernel (FIXADDR_TOP is a
> variable), the HPET and VVAR Page can only relative addressed. So this
> pages must located before or after the VDSO.
>
> This is why i need to setup this pages into the fixmap area, this is the
> compat mode "vdso32=2".
>
> For "vdso32=1" i need to map the VDSO Page together with the HPET and
> VVAR into the user space.
>
> For compability reasons both mappings are required.

Not at the same time, and I don't think you've guarded the
map_vsyscall call with a check for compat mode.

--Andy

Re: [PATCH 7/8] Add 32 bit VDSO time support for 32 bit kernel

[Stefani Seibold]

Am Montag, den 03.02.2014, 08:36 -0800 schrieb Andy Lutomirski:
> On Sun, Feb 2, 2014 at 11:44 PM, Stefani Seibold <stefani@seibold.net> wrote:
> > Am Sonntag, den 02.02.2014, 16:12 -0800 schrieb Andy Lutomirski:
> >> On Sun, Feb 2, 2014 at 1:39 PM, Stefani Seibold <stefani@seibold.net> wrote:
> >> > Am Sonntag, den 02.02.2014, 08:46 -0800 schrieb Andy Lutomirski:
> >> >> On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
> >> >> > From: Stefani Seibold <stefani@seibold.net>
> >> >> >
> >> >> > This patch add the time support for 32 bit a VDSO to a 32 bit kernel.
> >> >>
> >> >> [...]
> >> >>
> >> >> Can you address the review comments from last time around?  For
> >> >> example, this still seems to have redundant vvar and hpet mappings, it
> >> >> doesn't use the VVAR macro, it moves the 32-bit compat vDSO, etc.
> >> >>
> >> >
> >> > I will address the compat VDSO issue.
> >> >
> >> > But the VVAR macro will be not a part of this patch set. If you depend
> >> > on this, feel free to create one. From my point of view this is not
> >> > feasible without a macro hacking, because the address accessing the vvar
> >> > area differs in kernel and VDSO user mode.
> >>
> >> Sorry, but "I will make the code messier for no apparent reason and I
> >> will not offer to fix it in the same series" gets my NAK.
> >>
> >> Hint: I'm talking about two or three lines of code in vvar.h.
> >>
> >
> > A hint back: if you threat me with a NAK for a requested code sequence
> > which currently no user, this is far away from professional. I am not
> > your trainee.
> >
> > BTW: If it is so easy, send me the two or three lines and i will merge
> > it ;-)
> 
> Something to the effect of:
> 
> #elif defined(BUILD_VDSO32)
> #define VVAR(name) (*vvar_ ## name)
> #endif
> 
> Should do the trick.

You are wrong... 

#ifdef BUILD_VDSO32

#define DECLARE_VVAR(offset, type, name) \
	extern type vvar_ ## name __attribute__((visibility("hidden")));

#define VVAR(name) (vvar_ ## name)

#else

/* Base address of vvars.  This is not ABI. */
#ifdef CONFIG_X86_64
#define VVAR_ADDRESS (-10*1024*1024 - 4096)
#else
extern char __vvar_page;

#define VVAR_ADDRESS (&__vvar_page)
#endif

This would do the trick!

But for 64 bit to 32 bit conversation layer in vclock_gettime.c there is
still a

struct arch_vsyscall_gtod_data arch_vvar_vsyscall_gtod_data
	__attribute__((visibility("hidden")));
#define gtod (&arch_vvar_vsyscall_gtod_data)

needed, because vvar_vsyscall_gtod_data is the 32 bit version, which
would result in incorrect access of the struct members. So my code can't
use this VVAR macro.

> 
> >
> >> >
> >> > I also see no redundant mapping. There are two modes, one is the map of
> >> > the kernel area the other maps the VDSO into the user space area. This
> >> > is exactly the behaviour of the origin VDSO implementation.
> >>
> >> No.
> >>
> >> In your series there are *three* mappings.  There are:
> >>
> >>  - The linear mapping that the kernel loader sets up (the writable
> >> mapping used in the kernel).  This is implicit and, of course, fine.
> >>  - There's the fixmap page, which aliases the normal kernel mapping at
> >> a fixed address with the user, ro, and nx attributes.  The 64-bit vDSO
> >> uses that mapping.  See vdso.h -- it's all arranged pretty clearly.
> >> Your code, for no discernible reason, sets up a fixmap entry on
> >> *32-bit* kernels.
> >>  - The vma that you're setting up adjacent to the actual vdso text.
> >> This is what you are using.
> >>
> >> Please choose *one* user-readable mapping for the 32-bit vdso and
> >> stick with it.  If the 64-bit vdso can use it to and userspace doesn't
> >> break, even better.  But a pointless set of extra fixmap entries is
> >> not okay.
> >>
> >
> > Again: I wrote that there are two modes for a 32 bit kernel and
> > therefore there are two mappings at the same time. Since there are both
> > ways available in a 32 bit kernel via the vdso32= kernel parameter, both
> > must be supported.
> >
> > Due the lack of a real fixmap for a 32 bit kernel (FIXADDR_TOP is a
> > variable), the HPET and VVAR Page can only relative addressed. So this
> > pages must located before or after the VDSO.
> >
> > This is why i need to setup this pages into the fixmap area, this is the
> > compat mode "vdso32=2".
> >
> > For "vdso32=1" i need to map the VDSO Page together with the HPET and
> > VVAR into the user space.
> >
> > For compability reasons both mappings are required.
> 
> Not at the same time, and I don't think you've guarded the
> map_vsyscall call with a check for compat mode.
> 

I see no benefit to check the compat mode in map_vsyscall. The best way
is always map the VVAR page, because it is less error prone. In your
case tt would be also necessary not map the HPET page in non compat mode
too. So why bloat the code?

- Stefani

Re: [PATCH 7/8] Add 32 bit VDSO time support for 32 bit kernel

[Andy Lutomirski]

On Mon, Feb 3, 2014 at 2:01 PM, Stefani Seibold <stefani@seibold.net> wrote:
> Am Montag, den 03.02.2014, 08:36 -0800 schrieb Andy Lutomirski:
>> On Sun, Feb 2, 2014 at 11:44 PM, Stefani Seibold <stefani@seibold.net> wrote:
>> > Am Sonntag, den 02.02.2014, 16:12 -0800 schrieb Andy Lutomirski:
>> >> On Sun, Feb 2, 2014 at 1:39 PM, Stefani Seibold <stefani@seibold.net> wrote:
>> >> > Am Sonntag, den 02.02.2014, 08:46 -0800 schrieb Andy Lutomirski:
>> >> >> On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
>> >> >> > From: Stefani Seibold <stefani@seibold.net>
>> >> >> >
>> >> >> > This patch add the time support for 32 bit a VDSO to a 32 bit kernel.
>> >> >>
>> >> >> [...]
>> >> >>
>> >> >> Can you address the review comments from last time around?  For
>> >> >> example, this still seems to have redundant vvar and hpet mappings, it
>> >> >> doesn't use the VVAR macro, it moves the 32-bit compat vDSO, etc.
>> >> >>
>> >> >
>> >> > I will address the compat VDSO issue.
>> >> >
>> >> > But the VVAR macro will be not a part of this patch set. If you depend
>> >> > on this, feel free to create one. From my point of view this is not
>> >> > feasible without a macro hacking, because the address accessing the vvar
>> >> > area differs in kernel and VDSO user mode.
>> >>
>> >> Sorry, but "I will make the code messier for no apparent reason and I
>> >> will not offer to fix it in the same series" gets my NAK.
>> >>
>> >> Hint: I'm talking about two or three lines of code in vvar.h.
>> >>
>> >
>> > A hint back: if you threat me with a NAK for a requested code sequence
>> > which currently no user, this is far away from professional. I am not
>> > your trainee.
>> >
>> > BTW: If it is so easy, send me the two or three lines and i will merge
>> > it ;-)
>>
>> Something to the effect of:
>>
>> #elif defined(BUILD_VDSO32)
>> #define VVAR(name) (*vvar_ ## name)
>> #endif
>>
>> Should do the trick.
>
> You are wrong...
>
> #ifdef BUILD_VDSO32
>
> #define DECLARE_VVAR(offset, type, name) \
>         extern type vvar_ ## name __attribute__((visibility("hidden")));
>
> #define VVAR(name) (vvar_ ## name)
>
> #else
>
> /* Base address of vvars.  This is not ABI. */
> #ifdef CONFIG_X86_64
> #define VVAR_ADDRESS (-10*1024*1024 - 4096)
> #else
> extern char __vvar_page;
>
> #define VVAR_ADDRESS (&__vvar_page)
> #endif
>
> This would do the trick!
>
> But for 64 bit to 32 bit conversation layer in vclock_gettime.c there is
> still a
>
> struct arch_vsyscall_gtod_data arch_vvar_vsyscall_gtod_data
>         __attribute__((visibility("hidden")));
> #define gtod (&arch_vvar_vsyscall_gtod_data)
>
> needed, because vvar_vsyscall_gtod_data is the 32 bit version, which
> would result in incorrect access of the struct members. So my code can't
> use this VVAR macro.

For 32-on-64, I must have read your code wrong.  Are you sticking two
copies of the same struct with different layout into the vvar page?
If so, wouldn't it be better to only have the variant with a common
layout and use it for all versions of the vdso?

--Andy

Re: [PATCH 7/8] Add 32 bit VDSO time support for 32 bit kernel

[Stefani Seibold]

Am Montag, den 03.02.2014, 14:04 -0800 schrieb Andy Lutomirski:
> On Mon, Feb 3, 2014 at 2:01 PM, Stefani Seibold <stefani@seibold.net> wrote:
> > Am Montag, den 03.02.2014, 08:36 -0800 schrieb Andy Lutomirski:
> >> On Sun, Feb 2, 2014 at 11:44 PM, Stefani Seibold <stefani@seibold.net> wrote:
> >> > Am Sonntag, den 02.02.2014, 16:12 -0800 schrieb Andy Lutomirski:
> >> >> On Sun, Feb 2, 2014 at 1:39 PM, Stefani Seibold <stefani@seibold.net> wrote:
> >> >> > Am Sonntag, den 02.02.2014, 08:46 -0800 schrieb Andy Lutomirski:
> >> >> >> On Sun, Feb 2, 2014 at 3:27 AM,  <stefani@seibold.net> wrote:
> >> >> >> > From: Stefani Seibold <stefani@seibold.net>
> >> >> >> >
> >> >> >> > This patch add the time support for 32 bit a VDSO to a 32 bit kernel.
> >> >> >>
> >> >> >> [...]
> >> >> >>
> >> >> >> Can you address the review comments from last time around?  For
> >> >> >> example, this still seems to have redundant vvar and hpet mappings, it
> >> >> >> doesn't use the VVAR macro, it moves the 32-bit compat vDSO, etc.
> >> >> >>
> >> >> >
> >> >> > I will address the compat VDSO issue.
> >> >> >
> >> >> > But the VVAR macro will be not a part of this patch set. If you depend
> >> >> > on this, feel free to create one. From my point of view this is not
> >> >> > feasible without a macro hacking, because the address accessing the vvar
> >> >> > area differs in kernel and VDSO user mode.
> >> >>
> >> >> Sorry, but "I will make the code messier for no apparent reason and I
> >> >> will not offer to fix it in the same series" gets my NAK.
> >> >>
> >> >> Hint: I'm talking about two or three lines of code in vvar.h.
> >> >>
> >> >
> >> > A hint back: if you threat me with a NAK for a requested code sequence
> >> > which currently no user, this is far away from professional. I am not
> >> > your trainee.
> >> >
> >> > BTW: If it is so easy, send me the two or three lines and i will merge
> >> > it ;-)
> >>
> >> Something to the effect of:
> >>
> >> #elif defined(BUILD_VDSO32)
> >> #define VVAR(name) (*vvar_ ## name)
> >> #endif
> >>
> >> Should do the trick.
> >
> > You are wrong...
> >
> > #ifdef BUILD_VDSO32
> >
> > #define DECLARE_VVAR(offset, type, name) \
> >         extern type vvar_ ## name __attribute__((visibility("hidden")));
> >
> > #define VVAR(name) (vvar_ ## name)
> >
> > #else
> >
> > /* Base address of vvars.  This is not ABI. */
> > #ifdef CONFIG_X86_64
> > #define VVAR_ADDRESS (-10*1024*1024 - 4096)
> > #else
> > extern char __vvar_page;
> >
> > #define VVAR_ADDRESS (&__vvar_page)
> > #endif
> >
> > This would do the trick!
> >
> > But for 64 bit to 32 bit conversation layer in vclock_gettime.c there is
> > still a
> >
> > struct arch_vsyscall_gtod_data arch_vvar_vsyscall_gtod_data
> >         __attribute__((visibility("hidden")));
> > #define gtod (&arch_vvar_vsyscall_gtod_data)
> >
> > needed, because vvar_vsyscall_gtod_data is the 32 bit version, which
> > would result in incorrect access of the struct members. So my code can't
> > use this VVAR macro.
> 
> For 32-on-64, I must have read your code wrong.  Are you sticking two
> copies of the same struct with different layout into the vvar page?
> If so, wouldn't it be better to only have the variant with a common
> layout and use it for all versions of the vdso?
> 

No, only one. But for depend on 32/64 bit the layout differs.

We discuss this topic some days before, so please have a look at the
code and the previous posts.

- Stefani

[PATCH 8/8] Add 32 bit VDSO time support for 64 bit kernel

[stefani]

From: Stefani Seibold <stefani@seibold.net>

This patch add the VDSO time support for the IA32 Emulation Layer.

Due the nature of the kernel headers and the LP64 compiler where the
size of a long and a pointer differs against a 32 bit compiler, there
is a lot of type hacking necessary.

Signed-off-by: Stefani Seibold <stefani@seibold.net>
---
 arch/x86/vdso/vclock_gettime.c        | 109 +++++++++++++++++++++++++++-------
 arch/x86/vdso/vdso32/vclock_gettime.c |   7 +++
 2 files changed, 95 insertions(+), 21 deletions(-)

diff --git a/arch/x86/vdso/vclock_gettime.c b/arch/x86/vdso/vclock_gettime.c
index 42f641c..cd33f05 100644
--- a/arch/x86/vdso/vclock_gettime.c
+++ b/arch/x86/vdso/vclock_gettime.c
@@ -31,12 +31,24 @@
 
 #define gtod (&VVAR(vsyscall_gtod_data))
 
+struct api_timeval {
+	long	tv_sec;		/* seconds */
+	long	tv_usec;	/* microseconds */
+};
+
+struct api_timespec {
+	long	tv_sec;		/* seconds */
+	long	tv_nsec;	/* microseconds */
+};
+
+typedef long api_time_t;
+
 static notrace cycle_t vread_hpet(void)
 {
 	return readl((const void __iomem *)fix_to_virt(VSYSCALL_HPET) + HPET_COUNTER);
 }
 
-notrace static long vdso_fallback_gettime(long clock, struct timespec *ts)
+notrace static long vdso_fallback_gettime(long clock, struct api_timespec *ts)
 {
 	long ret;
 	asm("syscall" : "=a" (ret) :
@@ -44,7 +56,8 @@ notrace static long vdso_fallback_gettime(long clock, struct timespec *ts)
 	return ret;
 }
 
-notrace static long vdso_fallback_gtod(struct timeval *tv, struct timezone *tz)
+notrace static long vdso_fallback_gtod(struct api_timeval *tv,
+		struct timezone *tz)
 {
 	long ret;
 
@@ -54,14 +67,62 @@ notrace static long vdso_fallback_gtod(struct timeval *tv, struct timezone *tz)
 }
 #else
 
+#ifdef CONFIG_IA32_EMULATION
+typedef s64	arch_time_t;
+
+struct arch_timespec {
+	s64	tv_sec;
+	s64	tv_nsec;
+};
+
+#define ALIGN8 __attribute__ ((aligned (8)))
+
+struct arch_vsyscall_gtod_data {
+	seqcount_t	seq ALIGN8;
+
+	struct { /* extract of a clocksource struct */
+		int vclock_mode ALIGN8;
+		cycle_t	cycle_last ALIGN8;
+		cycle_t	mask ALIGN8;
+		u32	mult;
+		u32	shift;
+	} clock;
+
+	/* open coded 'struct timespec' */
+	arch_time_t	wall_time_sec;
+	u64		wall_time_snsec;
+	u64		monotonic_time_snsec;
+	arch_time_t	monotonic_time_sec;
+
+	struct timezone sys_tz;
+	struct arch_timespec wall_time_coarse;
+	struct arch_timespec monotonic_time_coarse;
+};
+
+struct arch_vsyscall_gtod_data vvar_vsyscall_gtod_data
+	__attribute__((visibility("hidden")));
+#else
 struct vsyscall_gtod_data vvar_vsyscall_gtod_data
 	__attribute__((visibility("hidden")));
+#endif
 
 u8 hpet_page
 	__attribute__((visibility("hidden")));
 
 #define gtod (&vvar_vsyscall_gtod_data)
 
+struct api_timeval {
+	s32	tv_sec;		/* seconds */
+	s32	tv_usec;	/* microseconds */
+};
+
+struct api_timespec {
+	s32	tv_sec;		/* seconds */
+	s32	tv_nsec;	/* microseconds */
+};
+
+typedef s32 api_time_t;
+
 #ifdef CONFIG_HPET_TIMER
 static notrace cycle_t vread_hpet(void)
 {
@@ -69,7 +130,7 @@ static notrace cycle_t vread_hpet(void)
 }
 #endif
 
-notrace static long vdso_fallback_gettime(long clock, struct timespec *ts)
+notrace static long vdso_fallback_gettime(long clock, struct api_timespec *ts)
 {
 	long ret;
 
@@ -79,12 +140,12 @@ notrace static long vdso_fallback_gettime(long clock, struct timespec *ts)
 		"call VDSO32_vsyscall \n"
 		"pop %%ebx \n"
 		: "=a" (ret)
-		: "0" (__NR_clock_gettime), "d" (clock), "c" (ts)
+		: "0" (__NR_ia32_clock_gettime), "d" (clock), "c" (ts)
 		: "memory");
 	return ret;
 }
 
-notrace static long vdso_fallback_gtod(struct timeval *tv,
+notrace static long vdso_fallback_gtod(struct api_timeval *tv,
 		struct timezone *tz)
 {
 	long ret;
@@ -95,7 +156,7 @@ notrace static long vdso_fallback_gtod(struct timeval *tv,
 		"call VDSO32_vsyscall \n"
 		"pop %%ebx \n"
 		: "=a" (ret)
-		: "0" (__NR_gettimeofday), "d" (tv), "c" (tz)
+		: "0" (__NR_ia32_gettimeofday), "d" (tv), "c" (tz)
 		: "memory");
 	return ret;
 }
@@ -284,42 +345,48 @@ notrace static void do_monotonic_coarse(struct timespec *ts)
 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
 }
 
-notrace int __vdso_clock_gettime(clockid_t clock, struct timespec *ts)
+notrace int __vdso_clock_gettime(clockid_t clock, struct api_timespec *ts)
 {
+	struct timespec tmp;
+
 	switch (clock) {
 	case CLOCK_REALTIME:
-		if (do_realtime(ts) == VCLOCK_NONE)
+		if (do_realtime(&tmp) == VCLOCK_NONE)
 			goto fallback;
 		break;
 	case CLOCK_MONOTONIC:
-		if (do_monotonic(ts) == VCLOCK_NONE)
+		if (do_monotonic(&tmp) == VCLOCK_NONE)
 			goto fallback;
 		break;
 	case CLOCK_REALTIME_COARSE:
-		do_realtime_coarse(ts);
+		do_realtime_coarse(&tmp);
 		break;
 	case CLOCK_MONOTONIC_COARSE:
-		do_monotonic_coarse(ts);
+		do_monotonic_coarse(&tmp);
 		break;
 	default:
 		goto fallback;
 	}
 
+	ts->tv_sec = tmp.tv_sec;
+	ts->tv_nsec = tmp.tv_nsec;
+
 	return 0;
 fallback:
 	return vdso_fallback_gettime(clock, ts);
 }
-int clock_gettime(clockid_t, struct timespec *)
+int clock_gettime(clockid_t, struct api_timespec *)
 	__attribute__((weak, alias("__vdso_clock_gettime")));
 
-notrace int __vdso_gettimeofday(struct timeval *tv, struct timezone *tz)
+notrace int __vdso_gettimeofday(struct api_timeval *tv, struct timezone *tz)
 {
+	struct timespec tmp;
+
 	if (likely(tv != NULL)) {
-		BUILD_BUG_ON(offsetof(struct timeval, tv_usec) !=
-			     offsetof(struct timespec, tv_nsec) ||
-			     sizeof(*tv) != sizeof(struct timespec));
-		if (do_realtime((struct timespec *)tv) == VCLOCK_NONE)
+		if (do_realtime(&tmp) == VCLOCK_NONE)
 			return vdso_fallback_gtod(tv, tz);
+		tv->tv_sec = tmp.tv_sec;
+		tv->tv_usec = tmp.tv_nsec;
 		tv->tv_usec /= 1000;
 	}
 	if (unlikely(tz != NULL)) {
@@ -330,21 +397,21 @@ notrace int __vdso_gettimeofday(struct timeval *tv, struct timezone *tz)
 
 	return 0;
 }
-int gettimeofday(struct timeval *, struct timezone *)
+int gettimeofday(struct api_timeval *, struct timezone *)
 	__attribute__((weak, alias("__vdso_gettimeofday")));
 
 /*
  * This will break when the xtime seconds get inaccurate, but that is
  * unlikely
  */
-notrace time_t __vdso_time(time_t *t)
+notrace api_time_t __vdso_time(api_time_t *t)
 {
 	/* This is atomic on x86 so we don't need any locks. */
-	time_t result = ACCESS_ONCE(gtod->wall_time_sec);
+	api_time_t result = ACCESS_ONCE(gtod->wall_time_sec);
 
 	if (t)
 		*t = result;
 	return result;
 }
-int time(time_t *t)
+int time(api_time_t *t)
 	__attribute__((weak, alias("__vdso_time")));
diff --git a/arch/x86/vdso/vdso32/vclock_gettime.c b/arch/x86/vdso/vdso32/vclock_gettime.c
index fab4ec6..b6df952 100644
--- a/arch/x86/vdso/vdso32/vclock_gettime.c
+++ b/arch/x86/vdso/vdso32/vclock_gettime.c
@@ -2,6 +2,8 @@
 
 #ifdef CONFIG_X86_64
 
+#include <generated/asm/unistd_32_ia32.h>
+
 #define _ASM_X86_PAGE_H
 
 #define __pa(x)                0
@@ -10,6 +12,11 @@
 #undef CONFIG_ILLEGAL_POINTER_VALUE
 #define CONFIG_ILLEGAL_POINTER_VALUE   0
 
+#else
+
+#define __NR_ia32_clock_gettime	__NR_clock_gettime
+#define __NR_ia32_gettimeofday	__NR_gettimeofday
+
 #endif
 
 #include "../vclock_gettime.c"
-- 
1.8.5.3