[PATCH] x86_64: use -mskip-rax-setup when available

[PATCH] x86_64: use -mskip-rax-setup when available

[Rasmus Villemoes]

gcc 5 supports the option -mskip-rax-setup to avoid emitting the
two-byte instruction xor %eax,%eax before a vararg function
call. Clearing %eax is redundant in the kernel, since no callee
expects any floating point arguments, and hence never use the incoming
value of %eax for anything.

For a defconfig kernel, the .text saving is around 26 kB:

$ size /tmp/vmlinux.{old,new}
   text    data     bss     dec     hex filename
12221850        1735352 1077248 15034450         e56852 /tmp/vmlinux.old
12195955        1735352 1077248 15008555         e5032b /tmp/vmlinux.new

Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
---

Nothing seems to have happened on the kernel side since H.J. Lu
implemented this in December.  Since gcc 5 is officially released
soon, I think it would be nice if this could soak in -next until the
next merge window.

 arch/x86/Makefile | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/arch/x86/Makefile b/arch/x86/Makefile
index 2fda005bb334..00bc42c4cbd1 100644
--- a/arch/x86/Makefile
+++ b/arch/x86/Makefile
@@ -81,6 +81,8 @@ else
         KBUILD_CFLAGS += $(call cc-option,-mno-80387)
         KBUILD_CFLAGS += $(call cc-option,-mno-fp-ret-in-387)
 
+        KBUILD_CFLAGS += $(call cc-option,-mskip-rax-setup)
+
 	# Use -mpreferred-stack-boundary=3 if supported.
 	KBUILD_CFLAGS += $(call cc-option,-mpreferred-stack-boundary=3)
 
-- 
2.1.3

Re: [PATCH] x86_64: use -mskip-rax-setup when available

[Borislav Petkov]

On Mon, Apr 20, 2015 at 02:28:16PM +0200, Rasmus Villemoes wrote:
> gcc 5 supports the option -mskip-rax-setup to avoid emitting the
> two-byte instruction xor %eax,%eax before a vararg function
> call. Clearing %eax is redundant in the kernel, since no callee
> expects any floating point arguments, and hence never use the incoming
> value of %eax for anything.
> 
> For a defconfig kernel, the .text saving is around 26 kB:
> 
> $ size /tmp/vmlinux.{old,new}
>    text    data     bss     dec     hex filename
> 12221850        1735352 1077248 15034450         e56852 /tmp/vmlinux.old
> 12195955        1735352 1077248 15008555         e5032b /tmp/vmlinux.new
> 
> Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
> ---
> 
> Nothing seems to have happened on the kernel side since H.J. Lu
> implemented this in December.  Since gcc 5 is officially released

I see the same patch from him from Apr. 16th:

https://lkml.kernel.org/r/CAMe9rOrfDo3MGaQc_s5crw3TrYhS8NtqeyRcUFiA0yAV8e9Nig@mail.gmail.com

Why are you sending it too?

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--

Re: [PATCH] x86_64: use -mskip-rax-setup when available

[Rasmus Villemoes]

On Mon, Apr 20 2015, Borislav Petkov <bp@alien8.de> wrote:

> On Mon, Apr 20, 2015 at 02:28:16PM +0200, Rasmus Villemoes wrote:
>> 
>> Nothing seems to have happened on the kernel side since H.J. Lu
>> implemented this in December.  Since gcc 5 is officially released
>
> I see the same patch from him from Apr. 16th:
>
> https://lkml.kernel.org/r/CAMe9rOrfDo3MGaQc_s5crw3TrYhS8NtqeyRcUFiA0yAV8e9Nig@mail.gmail.com
>
> Why are you sending it too?

Hadn't seen that, so I wrongly assumed everybody had forgotten about
it. I don't care about the S-o-b on the trivial Makefile patch, but my
request for a feature-suggested-by or similar still stands.

Rasmus

Re: [PATCH] x86_64: use -mskip-rax-setup when available

[Borislav Petkov]

On Mon, Apr 20, 2015 at 10:24:44PM +0200, Rasmus Villemoes wrote:
> Hadn't seen that, so I wrongly assumed everybody had forgotten about
> it. I don't care about the S-o-b on the trivial Makefile patch, but my
> request for a feature-suggested-by or similar still stands.

I don't understand what that means. Are you saying, you want the patch
in or you want attribution or ...?

/me is confused.

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--

Re: [PATCH] x86_64: use -mskip-rax-setup when available

[Rasmus Villemoes]

On Mon, Apr 20 2015, Borislav Petkov <bp@alien8.de> wrote:

> On Mon, Apr 20, 2015 at 10:24:44PM +0200, Rasmus Villemoes wrote:
>> Hadn't seen that, so I wrongly assumed everybody had forgotten about
>> it. I don't care about the S-o-b on the trivial Makefile patch, but my
>> request for a feature-suggested-by or similar still stands.
>
> I don't understand what that means. Are you saying, you want the patch
> in

Mostly this, of course, or the whole thing would have been
pointless.

> or you want attribution or ...?

That would be nice. 

Rasmus

Re: [PATCH] x86_64: use -mskip-rax-setup when available

[Borislav Petkov]

On Thu, Apr 23, 2015 at 01:00:26PM +0200, Rasmus Villemoes wrote:
> > or you want attribution or ...?
> 
> That would be nice.

I fail to see for what - I see H.J. Lu's patch from Dec. 2014 and you
repeating the same patch months later.

Don't get me wrong - I'd gladly add your attribution too but it is his
patch which should go because it was the first. It is only fair this way
IMO.

Now, if you go and test his patch on latest linus/master, I'd gladly add
your Tested-by...

Thanks.

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--

Re: [PATCH] x86_64: use -mskip-rax-setup when available

[Rasmus Villemoes]

On Thu, Apr 23 2015, Borislav Petkov <bp@alien8.de> wrote:

> On Thu, Apr 23, 2015 at 01:00:26PM +0200, Rasmus Villemoes wrote:
>> > or you want attribution or ...?
>> 
>> That would be nice.
>
> I fail to see for what - I see H.J. Lu's patch from Dec. 2014 and you
> repeating the same patch months later.

I think we're talking past each other. What I'd like credit for is
coming up with the idea of omitting the xor %eax,%eax in the first place
(the original mail was sent to x86@kernel.org [cc'ed to a few x86
maintainers], but that doesn't seem to be publicly archived).

So it is very much not about (attribution for) the Makefile patch itself.

> [...] it is his patch which should go because it was the first. It is
> only fair this way IMO.

Agreed - all I ask for is being mentioned somewhere in the commit
log. Again, I only sent the April 20 patch because I hadn't seen
anything happen since December.

Rasmus

Re: [PATCH] x86_64: use -mskip-rax-setup when available

[Borislav Petkov]

On Thu, Apr 23, 2015 at 02:55:02PM +0200, Rasmus Villemoes wrote:
> I think we're talking past each other. What I'd like credit for is
> coming up with the idea of omitting the xor %eax,%eax in the first place
> (the original mail was sent to x86@kernel.org [cc'ed to a few x86
> maintainers], but that doesn't seem to be publicly archived).

Ok, now *this* I was missing. I'll add that information to the commit
message.

> Agreed - all I ask for is being mentioned somewhere in the commit
> log. Again, I only sent the April 20 patch because I hadn't seen
> anything happen since December.

Right, and I think situations like that can be much easier handled in
the future by you or the person who's interested in the patch, replying
to the original submission and asking what the status is instead of
doing a new and identical patch.

We're on the same page now :-)

Thanks!

-- 
Regards/Gruss,
    Boris.

ECO tip #101: Trim your mails when you reply.
--

Re: [PATCH] x86_64: use -mskip-rax-setup when available

[H.J. Lu]

On Thu, Apr 23, 2015 at 5:29 AM, Borislav Petkov <bp@alien8.de> wrote:
> On Thu, Apr 23, 2015 at 01:00:26PM +0200, Rasmus Villemoes wrote:
>> > or you want attribution or ...?
>>
>> That would be nice.
>
> I fail to see for what - I see H.J. Lu's patch from Dec. 2014 and you
> repeating the same patch months later.

This optimization I implemented in GCC was suggested by Rasmus.
He should be recognized.

> Don't get me wrong - I'd gladly add your attribution too but it is his
> patch which should go because it was the first. It is only fair this way
> IMO.
>
> Now, if you go and test his patch on latest linus/master, I'd gladly add
> your Tested-by...

Thanks.

-- 
H.J.