linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH] x86-64: adjust ia32entry branches for better static prediction
@ 2009-10-20  7:10 Jan Beulich
  2009-10-20  7:21 ` Eric Dumazet
  0 siblings, 1 reply; 4+ messages in thread
From: Jan Beulich @ 2009-10-20  7:10 UTC (permalink / raw)
  To: mingo, tglx, hpa; +Cc: linux-kernel

Swap branch pairs so that the static branch prediction logic would
consider the syscall-number-in-range case the taken path.

Signed-off-by: Jan Beulich <jbeulich@novell.com>

---
 arch/x86/ia32/ia32entry.S |   18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

--- linux-2.6.32-rc5/arch/x86/ia32/ia32entry.S	2009-10-19 13:12:43.000000000 +0200
+++ 2.6.32-rc5-x86_64-ia32-syscall-trace-branch-dir/arch/x86/ia32/ia32entry.S	2009-10-19 09:36:39.000000000 +0200
@@ -249,9 +249,9 @@ sysenter_tracesys:
 	call	syscall_trace_enter
 	LOAD_ARGS32 ARGOFFSET  /* reload args from stack in case ptrace changed it */
 	RESTORE_REST
-	cmpl	$(IA32_NR_syscalls-1),%eax
-	ja	int_ret_from_sys_call /* sysenter_tracesys has set RAX(%rsp) */
-	jmp	sysenter_do_call
+	cmpl	$IA32_NR_syscalls,%eax
+	jb	sysenter_do_call
+	jmp	int_ret_from_sys_call /* sysenter_tracesys has set RAX(%rsp) */
 	CFI_ENDPROC
 ENDPROC(ia32_sysenter_target)
 
@@ -368,9 +368,9 @@ cstar_tracesys:
 	LOAD_ARGS32 ARGOFFSET, 1  /* reload args from stack in case ptrace changed it */
 	RESTORE_REST
 	xchgl %ebp,%r9d
-	cmpl $(IA32_NR_syscalls-1),%eax
-	ja int_ret_from_sys_call /* cstar_tracesys has set RAX(%rsp) */
-	jmp cstar_do_call
+	cmpl $IA32_NR_syscalls,%eax
+	jb cstar_do_call
+	jmp int_ret_from_sys_call /* cstar_tracesys has set RAX(%rsp) */
 END(ia32_cstar_target)
 				
 ia32_badarg:
@@ -445,9 +445,9 @@ ia32_tracesys:			 
 	call syscall_trace_enter
 	LOAD_ARGS32 ARGOFFSET  /* reload args from stack in case ptrace changed it */
 	RESTORE_REST
-	cmpl $(IA32_NR_syscalls-1),%eax
-	ja  int_ret_from_sys_call	/* ia32_tracesys has set RAX(%rsp) */
-	jmp ia32_do_call
+	cmpl $IA32_NR_syscalls,%eax
+	jb ia32_do_call
+	jmp int_ret_from_sys_call	/* ia32_tracesys has set RAX(%rsp) */
 END(ia32_syscall)
 
 ia32_badsys:




^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH] x86-64: adjust ia32entry branches for better static  prediction
  2009-10-20  7:10 [PATCH] x86-64: adjust ia32entry branches for better static prediction Jan Beulich
@ 2009-10-20  7:21 ` Eric Dumazet
  2009-10-20  7:24   ` H. Peter Anvin
  0 siblings, 1 reply; 4+ messages in thread
From: Eric Dumazet @ 2009-10-20  7:21 UTC (permalink / raw)
  To: Jan Beulich; +Cc: mingo, tglx, hpa, linux-kernel

Jan Beulich a écrit :
> Swap branch pairs so that the static branch prediction logic would
> consider the syscall-number-in-range case the taken path.
> 
> Signed-off-by: Jan Beulich <jbeulich@novell.com>
> 
> ---
>  arch/x86/ia32/ia32entry.S |   18 +++++++++---------
>  1 file changed, 9 insertions(+), 9 deletions(-)
> 
> --- linux-2.6.32-rc5/arch/x86/ia32/ia32entry.S	2009-10-19 13:12:43.000000000 +0200
> +++ 2.6.32-rc5-x86_64-ia32-syscall-trace-branch-dir/arch/x86/ia32/ia32entry.S	2009-10-19 09:36:39.000000000 +0200
> @@ -249,9 +249,9 @@ sysenter_tracesys:
>  	call	syscall_trace_enter
>  	LOAD_ARGS32 ARGOFFSET  /* reload args from stack in case ptrace changed it */
>  	RESTORE_REST
> -	cmpl	$(IA32_NR_syscalls-1),%eax
> -	ja	int_ret_from_sys_call /* sysenter_tracesys has set RAX(%rsp) */
> -	jmp	sysenter_do_call
> +	cmpl	$IA32_NR_syscalls,%eax
> +	jb	sysenter_do_call
> +	jmp	int_ret_from_sys_call /* sysenter_tracesys has set RAX(%rsp) */
>  	CFI_ENDPROC

on cmov enabled build, we could use cmov instruction to load eax with number
that calls enosys_syscall(), to avoid loading -ENOSYS,RAX
and these branches...


^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH] x86-64: adjust ia32entry branches for better static  prediction
  2009-10-20  7:21 ` Eric Dumazet
@ 2009-10-20  7:24   ` H. Peter Anvin
  2009-10-26  8:28     ` Andi Kleen
  0 siblings, 1 reply; 4+ messages in thread
From: H. Peter Anvin @ 2009-10-20  7:24 UTC (permalink / raw)
  To: Eric Dumazet; +Cc: Jan Beulich, mingo, tglx, linux-kernel

On 10/20/2009 04:21 PM, Eric Dumazet wrote:
>>
>> --- linux-2.6.32-rc5/arch/x86/ia32/ia32entry.S	2009-10-19 13:12:43.000000000 +0200
>> +++ 2.6.32-rc5-x86_64-ia32-syscall-trace-branch-dir/arch/x86/ia32/ia32entry.S	2009-10-19 09:36:39.000000000 +0200
>> @@ -249,9 +249,9 @@ sysenter_tracesys:
>>   	call	syscall_trace_enter
>>   	LOAD_ARGS32 ARGOFFSET  /* reload args from stack in case ptrace changed it */
>>   	RESTORE_REST
>> -	cmpl	$(IA32_NR_syscalls-1),%eax
>> -	ja	int_ret_from_sys_call /* sysenter_tracesys has set RAX(%rsp) */
>> -	jmp	sysenter_do_call
>> +	cmpl	$IA32_NR_syscalls,%eax
>> +	jb	sysenter_do_call
>> +	jmp	int_ret_from_sys_call /* sysenter_tracesys has set RAX(%rsp) */
>>   	CFI_ENDPROC
>
> on cmov enabled build, we could use cmov instruction to load eax with number
> that calls enosys_syscall(), to avoid loading -ENOSYS,RAX
> and these branches...
>

It's x86-64, cmov is guaranteed.

	-hpa

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH] x86-64: adjust ia32entry branches for better static  prediction
  2009-10-20  7:24   ` H. Peter Anvin
@ 2009-10-26  8:28     ` Andi Kleen
  0 siblings, 0 replies; 4+ messages in thread
From: Andi Kleen @ 2009-10-26  8:28 UTC (permalink / raw)
  To: H. Peter Anvin; +Cc: Eric Dumazet, Jan Beulich, mingo, tglx, linux-kernel

"H. Peter Anvin" <hpa@zytor.com> writes:

> On 10/20/2009 04:21 PM, Eric Dumazet wrote:
>>>
>>> --- linux-2.6.32-rc5/arch/x86/ia32/ia32entry.S	2009-10-19 13:12:43.000000000 +0200
>>> +++ 2.6.32-rc5-x86_64-ia32-syscall-trace-branch-dir/arch/x86/ia32/ia32entry.S	2009-10-19 09:36:39.000000000 +0200
>>> @@ -249,9 +249,9 @@ sysenter_tracesys:
>>>   	call	syscall_trace_enter
>>>   	LOAD_ARGS32 ARGOFFSET  /* reload args from stack in case ptrace changed it */
>>>   	RESTORE_REST
>>> -	cmpl	$(IA32_NR_syscalls-1),%eax
>>> -	ja	int_ret_from_sys_call /* sysenter_tracesys has set RAX(%rsp) */
>>> -	jmp	sysenter_do_call
>>> +	cmpl	$IA32_NR_syscalls,%eax
>>> +	jb	sysenter_do_call
>>> +	jmp	int_ret_from_sys_call /* sysenter_tracesys has set RAX(%rsp) */
>>>   	CFI_ENDPROC
>>
>> on cmov enabled build, we could use cmov instruction to load eax with number
>> that calls enosys_syscall(), to avoid loading -ENOSYS,RAX
>> and these branches...
>>
>
> It's x86-64, cmov is guaranteed.

But it's also not necessarily faster.

-Andi
-- 
ak@linux.intel.com -- Speaking for myself only.

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2009-10-26  8:28 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2009-10-20  7:10 [PATCH] x86-64: adjust ia32entry branches for better static prediction Jan Beulich
2009-10-20  7:21 ` Eric Dumazet
2009-10-20  7:24   ` H. Peter Anvin
2009-10-26  8:28     ` Andi Kleen

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).