powerpc: Add VDSO version of time
RFC PATCH 1/1 powerpc: Add VDSO version of time
Hi all,
This patch implement the time syscall as vDSO. I have a glibc patch to use it as IFUNC (as latest gettimeofday patch). Below the perf numbers:
Baseline PPC32: 380 nsec Baseline PPC64: 352 nsec vdso PPC32: 20 nsec vdso PPC64: 20 nsec
I focused on 64 bit kernel, do I need to provide a scheme for 32 bits as well?
Any tips, advices, comments?
--
arch/powerpc/kernel/vdso.c +4 -0
@@ -113,6 +113,10 @@ static struct vdso_patch_def vdso_patches[] = { CPU_FTR_USE_TB, 0, "__kernel_get_tbfreq", NULL }, + { +CPU_FTR_USE_TB ,0 , +"__kernel_time" ,NULL + }, };/*
arch/powerpc/kernel/vdso32/gettimeofday.S +29 -0
@@ -181,6 +181,35 @@ V_FUNCTION_END(__kernel_clock_getres) /* + * Exact prototype of time() + * + * time_t time(time *t); + * + */ +V_FUNCTION_BEGIN(__kernel_time) + .cfi_startproc + mflr r12 + .cfi_register lr,r12 + + mr r11,r3 /* r11 holds t */ + bl __get_datapage@local + mr r9, r3 /* datapage ptr in r9 */ + +1: lwz r8,(CFG_TB_UPDATE_COUNT+LOPART)(r9) + lwz r3,STAMP_XTIME+TSPEC_TV_SEC(r9) + andi. r0,r8,1 /* pending update ? loop */ + bne- 1b + + cmplwi r11,0 /* check if t is NULL */ + beq 2f + stw r3,0(r11) /* store result at *t */ +2: mtlr r12 + crclr cr0*4+so + blr + .cfi_endproc +V_FUNCTION_END(__kernel_time) + +/* * This is the core of clock_gettime() and gettimeofday(), * it returns the current time in r3 (seconds) and r4. * On entry, r7 gives the resolution of r4, either USEC_PER_SEC
arch/powerpc/kernel/vdso32/vdso32.lds.S +1 -0
@@ -150,6 +150,7 @@ VERSION #ifdef CONFIG_PPC64 __kernel_getcpu; #endif + __kernel_time; local: *; };
arch/powerpc/kernel/vdso64/gettimeofday.S +29 -0
@@ -164,6 +164,35 @@ V_FUNCTION_BEGIN(__kernel_clock_getres) .cfi_endproc V_FUNCTION_END(__kernel_clock_getres) +/* + * Exact prototype of time() + * + * time_t time(time *t); + * + */ +V_FUNCTION_BEGIN(__kernel_time) + .cfi_startproc + mflr r12 + .cfi_register lr,r12 + + mr r11,r3 /* r11 holds t */ + bl V_LOCAL_FUNC(__get_datapage) + +1: ld r8,CFG_TB_UPDATE_COUNT(r3) + ld r4,STAMP_XTIME+TSPC64_TV_SEC(r3) + andi. r0,r8,1 /* pending update ? loop */ + bne- 1b + + cmpldi r11,0 /* check if t is NULL */ + beq 2f + std r4,0(r11) /* store result at *t */ +2: mtlr r12 + crclr cr0*4+so + mr r3,r4 + blr + .cfi_endproc +V_FUNCTION_END(__kernel_time) + /* * This is the core of clock_gettime() and gettimeofday(),
arch/powerpc/kernel/vdso64/vdso64.lds.S +1 -0
@@ -147,6 +147,7 @@ VERSION __kernel_sync_dicache_p5; __kernel_sigtramp_rt64; __kernel_getcpu; + __kernel_time; local: *; };
Quoted 3 lines · Adhemerval Zanella
I focused on 64 bit kernel, do I need to provide a scheme for 32 bits as well?
You did provide both 32 and 64-bit VDSO implementations so 32-bit kernels should be covered.
Cheers,
Ben.
Quoted 5 lines · Benjamin Herrenschmidt
Quoted 2 lines · Adhemerval Zanella
I focused on 64 bit kernel, do I need to provide a scheme for 32 bits as well?
You did provide both 32 and 64-bit VDSO implementations so 32-bit kernels should be covered.
Indeed and thanks for the reply. Any objection or request about including it?
Thanks.
Hi all,
Just sending a ping about this patch.
Quoted 15 lines · Adhemerval Zanella
Quoted 5 lines · Benjamin Herrenschmidt
Quoted 2 lines · Adhemerval Zanella
I focused on 64 bit kernel, do I need to provide a scheme for 32 bits as well?
You did provide both 32 and 64-bit VDSO implementations so 32-bit kernels should be covered.
Indeed and thanks for the reply. Any objection or request about including it?
Thanks.
_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://lists.ozlabs.org/listinfo/linuxppc-dev
Hi Benjamin,
Any objection or request about this patch?
Quoted 15 lines · Benjamin Herrenschmidt
Quoted 2 lines · Adhemerval Zanella
I focused on 64 bit kernel, do I need to provide a scheme for 32 bits as well?
You did provide both 32 and 64-bit VDSO implementations so 32-bit kernels should be covered.
Cheers,
Ben._______________________________________________
Linuxppc-dev mailing list
[email protected]
https://lists.ozlabs.org/listinfo/linuxppc-dev
Quoted 6 lines · Adhemerval Zanella
Hi all,
This patch implement the time syscall as vDSO. I have a glibc patch to use it as IFUNC (as latest gettimeofday patch). Below the perf numbers:
Quoted 7 lines
Baseline PPC32: 380 nsec Baseline PPC64: 352 nsec vdso PPC32: 20 nsec vdso PPC64: 20 nsecI focused on 64 bit kernel, do I need to provide a scheme for 32 bits as well?
You did provide a 32-bit implementation. I take it you haven't tested that though? Can you test it?
What happens if I don't have the glibc patch?
cheers
Quoted 12 lines · Michael Ellerman
Quoted 11 lines · Adhemerval Zanella
Hi all,
This patch implement the time syscall as vDSO. I have a glibc patch to use it as IFUNC (as latest gettimeofday patch). Below the perf numbers:
Baseline PPC32: 380 nsec Baseline PPC64: 352 nsec vdso PPC32: 20 nsec vdso PPC64: 20 nsec
Quoted 4 lines
Quoted 2 lines
I focused on 64 bit kernel, do I need to provide a scheme for 32 bits as well?
You did provide a 32-bit implementation. I take it you haven't tested that though? Can you test it?
I haven't test it yet and I believe it won't be troublesome to do so.
Quoted 2 lines
What happens if I don't have the glibc patch?
GLIBC current behavior is to use the syscall.
Quoted 7 lines
cheers
_______________________________________________
Linuxppc-dev mailing list
[email protected]
https://lists.ozlabs.org/listinfo/linuxppc-dev
Quoted 12 lines · Michael Ellerman
Quoted 11 lines · Adhemerval Zanella
Hi all,
This patch implement the time syscall as vDSO. I have a glibc patch to use it as IFUNC (as latest gettimeofday patch). Below the perf numbers:
Baseline PPC32: 380 nsec Baseline PPC64: 352 nsec vdso PPC32: 20 nsec vdso PPC64: 20 nsec
Quoted 4 lines
Quoted 2 lines
I focused on 64 bit kernel, do I need to provide a scheme for 32 bits as well?
You did provide a 32-bit implementation. I take it you haven't tested that though? Can you test it?
Hi,
I didn't build a 32 bit kernel, but I tested 32 bits binaries (that uses the VDSO32 implantation) without any issue. Performance gains are similar.
Quoted 15 lines · Adhemerval Zanella
Quoted 14 lines · Michael Ellerman
Quoted 13 lines · Adhemerval Zanella
Hi all,
This patch implement the time syscall as vDSO. I have a glibc patch to use it as IFUNC (as latest gettimeofday patch). Below the perf numbers:
Baseline PPC32: 380 nsec Baseline PPC64: 352 nsec vdso PPC32: 20 nsec vdso PPC64: 20 nsecI focused on 64 bit kernel, do I need to provide a scheme for 32 bits as well?
Quoted 7 lines
Quoted 2 lines
You did provide a 32-bit implementation. I take it you haven't tested that though? Can you test it?
Hi,
I didn't build a 32 bit kernel, but I tested 32 bits binaries (that uses the VDSO32 implantation) without any issue. Performance gains are similar.
OK. Please send an updated version of the patch which a changelog describing the testing you've done, and add your Signed-off-by to the patch.
cheers
Hi Adhemerval,
Quoted 8 lines
This patch implement the time syscall as vDSO. I have a glibc patch to use it as IFUNC (as latest gettimeofday patch). Below the perf numbers:
Baseline PPC32: 380 nsec Baseline PPC64: 352 nsec vdso PPC32: 20 nsec vdso PPC64: 20 nsec
Very nice speedup. One small performance improvement:
+1: ld r8,CFG_TB_UPDATE_COUNT(r3)
+ ld r4,STAMP_XTIME+TSPC64_TV_SEC(r3)
+ andi. r0,r8,1 /* pending update ? loop */
+ bne- 1b
Since you are only reading one long you shouldn't need to check the update count and loop, you will always see a consistent value. The system call version of time() just does an unprotected load for example.
Quoted 4 lines
I focused on 64 bit kernel, do I need to provide a scheme for 32 bits as well?
Any tips, advices, comments?
With the above change and with Michael's comments covered (decent changelog entry and Signed-off-by):
Anton
Adhemerval Zanella
Acked-by Anton BlanchardQuoted 3 lines · Anton Blanchard
Since you are only reading one long you shouldn't need to check the update count and loop, you will always see a consistent value. The system call version of
time()just does an unprotected load for example.
Fixed.
Quoted 4 lines
With the above change and with Michael's comments covered (decent changelog entry and Signed-off-by):
Acked-by Anton Blanchard <[email protected]>
Thanks for the review, below the updated patch:
From: Adhemerval Zanella <[email protected]>
This patch implement the time syscall as vDSO. The performance speedups are:
Baseline PPC32: 380 nsec Baseline PPC64: 350 nsec vdso PPC32: 20 nsec vsdo PPC64: 20 nsec
Tested on 64 bit build with both 32 bit and 64 bit userland.
--- arch/powerpc/kernel/vdso.c | 4 ++++ arch/powerpc/kernel/vdso32/gettimeofday.S | 26 ++++++++++++++++++++++++++ arch/powerpc/kernel/vdso32/vdso32.lds.S | 1 + arch/powerpc/kernel/vdso64/gettimeofday.S | 26 ++++++++++++++++++++++++++ arch/powerpc/kernel/vdso64/vdso64.lds.S | 1 + 5 files changed, 58 insertions(+), 0 deletions(-)
arch/powerpc/kernel/vdso.c +4 -0
@@ -113,6 +113,10 @@ static struct vdso_patch_def vdso_patches[] = { CPU_FTR_USE_TB, 0, "__kernel_get_tbfreq", NULL }, + { +CPU_FTR_USE_TB ,0 , +"__kernel_time" ,NULL + }, };/*
arch/powerpc/kernel/vdso32/gettimeofday.S +26 -0
@@ -181,6 +181,32 @@ V_FUNCTION_END(__kernel_clock_getres) /* + * Exact prototype of time() + * + * time_t time(time *t); + * + */ +V_FUNCTION_BEGIN(__kernel_time) + .cfi_startproc + mflr r12 + .cfi_register lr,r12 + + mr r11,r3 /* r11 holds t */ + bl __get_datapage@local + mr r9, r3 /* datapage ptr in r9 */ + + lwz r3,STAMP_XTIME+TSPEC_TV_SEC(r9) + + cmplwi r11,0 /* check if t is NULL */ + beq 2f + stw r3,0(r11) /* store result at *t */ +2: mtlr r12 + crclr cr0*4+so + blr + .cfi_endproc +V_FUNCTION_END(__kernel_time) + +/* * This is the core of clock_gettime() and gettimeofday(), * it returns the current time in r3 (seconds) and r4. * On entry, r7 gives the resolution of r4, either USEC_PER_SEC
arch/powerpc/kernel/vdso32/vdso32.lds.S +1 -0
@@ -150,6 +150,7 @@ VERSION #ifdef CONFIG_PPC64 __kernel_getcpu; #endif + __kernel_time; local: *; };
arch/powerpc/kernel/vdso64/gettimeofday.S +26 -0
@@ -164,6 +164,32 @@ V_FUNCTION_BEGIN(__kernel_clock_getres) .cfi_endproc V_FUNCTION_END(__kernel_clock_getres) +/* + * Exact prototype of time() + * + * time_t time(time *t); + * + */ +V_FUNCTION_BEGIN(__kernel_time) + .cfi_startproc + mflr r12 + .cfi_register lr,r12 + + mr r11,r3 /* r11 holds t */ + bl V_LOCAL_FUNC(__get_datapage) + + ld r4,STAMP_XTIME+TSPC64_TV_SEC(r3) + + cmpldi r11,0 /* check if t is NULL */ + beq 2f + std r4,0(r11) /* store result at *t */ +2: mtlr r12 + crclr cr0*4+so + mr r3,r4 + blr + .cfi_endproc +V_FUNCTION_END(__kernel_time) + /* * This is the core of clock_gettime() and gettimeofday(),
arch/powerpc/kernel/vdso64/vdso64.lds.S +1 -0
@@ -147,6 +147,7 @@ VERSION __kernel_sync_dicache_p5; __kernel_sigtramp_rt64; __kernel_getcpu; + __kernel_time; local: *; };