linuxppc-dev.lists.ozlabs.org archive mirror
 help / color / mirror / Atom feed
* [RFC PATCH v1] powerpc/accounting: do not account system time on transition to user.
@ 2019-02-08 14:40 Christophe Leroy
  2019-02-19  5:41 ` Nicholas Piggin
  0 siblings, 1 reply; 2+ messages in thread
From: Christophe Leroy @ 2019-02-08 14:40 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Paul Mackerras, Michael Ellerman,
	Nicholas Piggin
  Cc: linuxppc-dev, linux-kernel

Time spent in kernel mode don't need to be accounted on transition
to user space. As far as the time spent in user is known, it
is possible to calculate the time spent in kernel by substracting
the time spent in user.

To do so, this patch modifies vtime_delta() to substract the
time spent in user since the last call to vtime_delta().

This patch gives a 2% improvment of null_syscall() selftest on a 83xx.

Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
But surprisingly, this patch degrades the null_syscall selftest by 20% on the 8xx. Any idea of the reason ?

 arch/powerpc/include/asm/accounting.h | 1 +
 arch/powerpc/include/asm/ppc_asm.h    | 8 +-------
 arch/powerpc/kernel/asm-offsets.c     | 8 ++------
 arch/powerpc/kernel/time.c            | 4 +++-
 4 files changed, 7 insertions(+), 14 deletions(-)

diff --git a/arch/powerpc/include/asm/accounting.h b/arch/powerpc/include/asm/accounting.h
index c607c5d835cc..2f1ff5f9fd7a 100644
--- a/arch/powerpc/include/asm/accounting.h
+++ b/arch/powerpc/include/asm/accounting.h
@@ -27,6 +27,7 @@ struct cpu_accounting_data {
 	/* Internal counters */
 	unsigned long starttime;	/* TB value snapshot */
 	unsigned long starttime_user;	/* TB value on exit to usermode */
+	unsigned long utime_asm;
 #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
 	unsigned long startspurr;	/* SPURR value snapshot */
 	unsigned long utime_sspurr;	/* ->user_time when ->startspurr set */
diff --git a/arch/powerpc/include/asm/ppc_asm.h b/arch/powerpc/include/asm/ppc_asm.h
index e0637730a8e7..be17d570d484 100644
--- a/arch/powerpc/include/asm/ppc_asm.h
+++ b/arch/powerpc/include/asm/ppc_asm.h
@@ -28,9 +28,8 @@
 #define ACCOUNT_STOLEN_TIME
 #else
 #define ACCOUNT_CPU_USER_ENTRY(ptr, ra, rb)				\
-	MFTB(ra);			/* get timebase */		\
 	PPC_LL	rb, ACCOUNT_STARTTIME_USER(ptr);			\
-	PPC_STL	ra, ACCOUNT_STARTTIME(ptr);				\
+	MFTB(ra);			/* get timebase */		\
 	subf	rb,rb,ra;		/* subtract start value */	\
 	PPC_LL	ra, ACCOUNT_USER_TIME(ptr);				\
 	add	ra,ra,rb;		/* add on to user time */	\
@@ -38,12 +37,7 @@
 
 #define ACCOUNT_CPU_USER_EXIT(ptr, ra, rb)				\
 	MFTB(ra);			/* get timebase */		\
-	PPC_LL	rb, ACCOUNT_STARTTIME(ptr);				\
 	PPC_STL	ra, ACCOUNT_STARTTIME_USER(ptr);			\
-	subf	rb,rb,ra;		/* subtract start value */	\
-	PPC_LL	ra, ACCOUNT_SYSTEM_TIME(ptr);				\
-	add	ra,ra,rb;		/* add on to system time */	\
-	PPC_STL	ra, ACCOUNT_SYSTEM_TIME(ptr)
 
 #ifdef CONFIG_PPC_SPLPAR
 #define ACCOUNT_STOLEN_TIME						\
diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
index 7a1b93c5af63..f2ba7735f56f 100644
--- a/arch/powerpc/kernel/asm-offsets.c
+++ b/arch/powerpc/kernel/asm-offsets.c
@@ -260,19 +260,15 @@ int main(void)
 	OFFSET(PACAHWCPUID, paca_struct, hw_cpu_id);
 	OFFSET(PACAKEXECSTATE, paca_struct, kexec_state);
 	OFFSET(PACA_DSCR_DEFAULT, paca_struct, dscr_default);
-	OFFSET(ACCOUNT_STARTTIME, paca_struct, accounting.starttime);
 	OFFSET(ACCOUNT_STARTTIME_USER, paca_struct, accounting.starttime_user);
-	OFFSET(ACCOUNT_USER_TIME, paca_struct, accounting.utime);
-	OFFSET(ACCOUNT_SYSTEM_TIME, paca_struct, accounting.stime);
+	OFFSET(ACCOUNT_USER_TIME, paca_struct, accounting.utime_asm);
 	OFFSET(PACA_TRAP_SAVE, paca_struct, trap_save);
 	OFFSET(PACA_NAPSTATELOST, paca_struct, nap_state_lost);
 	OFFSET(PACA_SPRG_VDSO, paca_struct, sprg_vdso);
 #else /* CONFIG_PPC64 */
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
-	OFFSET(ACCOUNT_STARTTIME, thread_info, accounting.starttime);
 	OFFSET(ACCOUNT_STARTTIME_USER, thread_info, accounting.starttime_user);
-	OFFSET(ACCOUNT_USER_TIME, thread_info, accounting.utime);
-	OFFSET(ACCOUNT_SYSTEM_TIME, thread_info, accounting.stime);
+	OFFSET(ACCOUNT_USER_TIME, thread_info, accounting.utime_asm);
 #endif
 #endif /* CONFIG_PPC64 */
 
diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
index bc0503ef9c9c..79420643b45f 100644
--- a/arch/powerpc/kernel/time.c
+++ b/arch/powerpc/kernel/time.c
@@ -331,8 +331,10 @@ static unsigned long vtime_delta(struct task_struct *tsk,
 	WARN_ON_ONCE(!irqs_disabled());
 
 	now = mftb();
-	stime = now - acct->starttime;
+	stime = now - acct->starttime - acct->utime_asm;
 	acct->starttime = now;
+	acct->utime += acct->utime_asm;
+	acct->utime_asm = 0;
 
 	*stime_scaled = vtime_delta_scaled(acct, now, stime);
 
-- 
2.13.3


^ permalink raw reply related	[flat|nested] 2+ messages in thread
* Re: [RFC PATCH v1] powerpc/accounting: do not account system time on transition to user.
  2019-02-08 14:40 [RFC PATCH v1] powerpc/accounting: do not account system time on transition to user Christophe Leroy
@ 2019-02-19  5:41 ` Nicholas Piggin
  0 siblings, 0 replies; 2+ messages in thread
From: Nicholas Piggin @ 2019-02-19  5:41 UTC (permalink / raw)
  To: Benjamin Herrenschmidt, Christophe Leroy, Michael Ellerman,
	Paul Mackerras
  Cc: linuxppc-dev, linux-kernel

Christophe Leroy's on February 9, 2019 12:40 am:
> Time spent in kernel mode don't need to be accounted on transition
> to user space. As far as the time spent in user is known, it
> is possible to calculate the time spent in kernel by substracting
> the time spent in user.
> 
> To do so, this patch modifies vtime_delta() to substract the
> time spent in user since the last call to vtime_delta().
> 
> This patch gives a 2% improvment of null_syscall() selftest on a 83xx.
> 
> Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>

This looks like a clever little optimization, although I don't know
this time accounting code very well.

> ---
> But surprisingly, this patch degrades the null_syscall selftest by 20% on the 8xx. Any idea of the reason ?

I don't know microarchitecture of any of those CPUs I'm afraid.
On the 64s CPUs, mftb is what hurts.

> 
>  arch/powerpc/include/asm/accounting.h | 1 +
>  arch/powerpc/include/asm/ppc_asm.h    | 8 +-------
>  arch/powerpc/kernel/asm-offsets.c     | 8 ++------
>  arch/powerpc/kernel/time.c            | 4 +++-
>  4 files changed, 7 insertions(+), 14 deletions(-)
> 
> diff --git a/arch/powerpc/include/asm/accounting.h b/arch/powerpc/include/asm/accounting.h
> index c607c5d835cc..2f1ff5f9fd7a 100644
> --- a/arch/powerpc/include/asm/accounting.h
> +++ b/arch/powerpc/include/asm/accounting.h
> @@ -27,6 +27,7 @@ struct cpu_accounting_data {
>  	/* Internal counters */
>  	unsigned long starttime;	/* TB value snapshot */
>  	unsigned long starttime_user;	/* TB value on exit to usermode */
> +	unsigned long utime_asm;
>  #ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
>  	unsigned long startspurr;	/* SPURR value snapshot */
>  	unsigned long utime_sspurr;	/* ->user_time when ->startspurr set */
> diff --git a/arch/powerpc/include/asm/ppc_asm.h b/arch/powerpc/include/asm/ppc_asm.h
> index e0637730a8e7..be17d570d484 100644
> --- a/arch/powerpc/include/asm/ppc_asm.h
> +++ b/arch/powerpc/include/asm/ppc_asm.h
> @@ -28,9 +28,8 @@
>  #define ACCOUNT_STOLEN_TIME
>  #else
>  #define ACCOUNT_CPU_USER_ENTRY(ptr, ra, rb)				\
> -	MFTB(ra);			/* get timebase */		\
>  	PPC_LL	rb, ACCOUNT_STARTTIME_USER(ptr);			\
> -	PPC_STL	ra, ACCOUNT_STARTTIME(ptr);				\
> +	MFTB(ra);			/* get timebase */		\
>  	subf	rb,rb,ra;		/* subtract start value */	\
>  	PPC_LL	ra, ACCOUNT_USER_TIME(ptr);				\
>  	add	ra,ra,rb;		/* add on to user time */	\
> @@ -38,12 +37,7 @@
>  
>  #define ACCOUNT_CPU_USER_EXIT(ptr, ra, rb)				\
>  	MFTB(ra);			/* get timebase */		\
> -	PPC_LL	rb, ACCOUNT_STARTTIME(ptr);				\
>  	PPC_STL	ra, ACCOUNT_STARTTIME_USER(ptr);			\
> -	subf	rb,rb,ra;		/* subtract start value */	\
> -	PPC_LL	ra, ACCOUNT_SYSTEM_TIME(ptr);				\
> -	add	ra,ra,rb;		/* add on to system time */	\
> -	PPC_STL	ra, ACCOUNT_SYSTEM_TIME(ptr)
>  
>  #ifdef CONFIG_PPC_SPLPAR
>  #define ACCOUNT_STOLEN_TIME						\
> diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
> index 7a1b93c5af63..f2ba7735f56f 100644
> --- a/arch/powerpc/kernel/asm-offsets.c
> +++ b/arch/powerpc/kernel/asm-offsets.c
> @@ -260,19 +260,15 @@ int main(void)
>  	OFFSET(PACAHWCPUID, paca_struct, hw_cpu_id);
>  	OFFSET(PACAKEXECSTATE, paca_struct, kexec_state);
>  	OFFSET(PACA_DSCR_DEFAULT, paca_struct, dscr_default);
> -	OFFSET(ACCOUNT_STARTTIME, paca_struct, accounting.starttime);
>  	OFFSET(ACCOUNT_STARTTIME_USER, paca_struct, accounting.starttime_user);
> -	OFFSET(ACCOUNT_USER_TIME, paca_struct, accounting.utime);
> -	OFFSET(ACCOUNT_SYSTEM_TIME, paca_struct, accounting.stime);
> +	OFFSET(ACCOUNT_USER_TIME, paca_struct, accounting.utime_asm);
>  	OFFSET(PACA_TRAP_SAVE, paca_struct, trap_save);
>  	OFFSET(PACA_NAPSTATELOST, paca_struct, nap_state_lost);
>  	OFFSET(PACA_SPRG_VDSO, paca_struct, sprg_vdso);
>  #else /* CONFIG_PPC64 */
>  #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
> -	OFFSET(ACCOUNT_STARTTIME, thread_info, accounting.starttime);
>  	OFFSET(ACCOUNT_STARTTIME_USER, thread_info, accounting.starttime_user);
> -	OFFSET(ACCOUNT_USER_TIME, thread_info, accounting.utime);
> -	OFFSET(ACCOUNT_SYSTEM_TIME, thread_info, accounting.stime);
> +	OFFSET(ACCOUNT_USER_TIME, thread_info, accounting.utime_asm);
>  #endif
>  #endif /* CONFIG_PPC64 */
>  
> diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
> index bc0503ef9c9c..79420643b45f 100644
> --- a/arch/powerpc/kernel/time.c
> +++ b/arch/powerpc/kernel/time.c
> @@ -331,8 +331,10 @@ static unsigned long vtime_delta(struct task_struct *tsk,
>  	WARN_ON_ONCE(!irqs_disabled());
>  
>  	now = mftb();
> -	stime = now - acct->starttime;
> +	stime = now - acct->starttime - acct->utime_asm;
>  	acct->starttime = now;
> +	acct->utime += acct->utime_asm;
> +	acct->utime_asm = 0;
>  
>  	*stime_scaled = vtime_delta_scaled(acct, now, stime);
>  
> -- 
> 2.13.3
> 
> 

^ permalink raw reply	[flat|nested] 2+ messages in thread
end of thread, other threads:[~2019-02-19  5:43 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-02-08 14:40 [RFC PATCH v1] powerpc/accounting: do not account system time on transition to user Christophe Leroy
2019-02-19  5:41 ` Nicholas Piggin

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).