Re: [PATCH 3/4] x86_64 ia32 syscall audit fast-path

[Linus Torvalds <torvalds@linux-foundation.org>]

On Sun, 6 Jul 2008, Roland McGrath wrote:
>
> This adds fast paths for 32-bit syscall entry and exit when
> TIF_SYSCALL_AUDIT is set, but no other kind of syscall tracing.
> These paths does not need to save and restore all registers as
> the general case of tracing does.  Avoiding the iret return path
> when syscall audit is enabled helps performance a lot.

Please don't do it like this.

I realize that Fedora enables audit by default, and that you care about 
the speed of the audit path, but I care about the speed of the *non*audit 
path. And you just slowed it down unnecessarily.


>  	TRACE_IRQS_OFF
> -	testl	$_TIF_ALLWORK_MASK,threadinfo_flags(%r10)
> +	testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),threadinfo_flags(%r10)
>  	jnz	int_ret_from_sys_call
> +	bt	$TIF_SYSCALL_AUDIT,threadinfo_flags(%r10)
> +	jc	sysexit_audit

Please don't add more instructions to the fastpath. Especially not 
relatively slow ones like 'bt'. 

So keep the

	testl   $_TIF_ALLWORK_MASK,threadinfo_flags(%r10)
	jnz     audit_or_other_work

and then in that *slow*path in int_ret_from_sys_call, do

  audit_or_other_work:
	testl $(_TIF_ALLWORK_MASK & ~_TIF_SYSCALL_AUDIT),threadinfo_flags(%r10)
	jnz  int_ret_from_sys_call
  sysexit_audit:
	....

so that we don't have any extra costs at all for the non-audit case.

I realize that maybe you don't care too much about the ia32 emulation, but 
it's going to be the common case for some distros (I think several distros 
end up with a 32-bit user space even when they use a 64-bit kernel).

		Linus