[PATCH] x86/processor.h: Force inlining of cpu_relax()

[PATCH] x86/processor.h: Force inlining of cpu_relax()

[Denys Vlasenko]

On x86, cpu_relax() simply calls rep_nop(), which generates one
instruction, PAUSE (aka REP NOP).

With this config:
http://busybox.net/~vda/kernel_config_OPTIMIZE_INLINING_and_Os
gcc-4.7.2 does not always inline rep_nop(): it generates
several copies of this:

<rep_nop> (16 copies, 194 calls):
       55                      push   %rbp
       48 89 e5                mov    %rsp,%rbp
       f3 90                   pause
       5d                      pop    %rbp
       c3                      retq

See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66122

This patch fixes this via s/inline/__always_inline/
on rep_nop() and cpu_relax().
(Forcing inlining only on rep_nop() causes gcc to
deinline cpu_relax(), with almost no change in generated code).

    text     data      bss       dec     hex filename
88118971 19905208 36421632 144445811 89c1173 vmlinux.before
88118139 19905208 36421632 144444979 89c0e33 vmlinux

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
CC: Ingo Molnar <mingo@kernel.org>
CC: Andy Lutomirski <luto@amacapital.net>
CC: "H. Peter Anvin" <hpa@zytor.com>
CC: Borislav Petkov <bp@alien8.de>
CC: Brian Gerst <brgerst@gmail.com>
CC: x86@kernel.org
CC: linux-kernel@vger.kernel.org
---
 arch/x86/include/asm/processor.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index 19577dd..b55f309 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -556,12 +556,12 @@ static inline unsigned int cpuid_edx(unsigned int op)
 }
 
 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
-static inline void rep_nop(void)
+static __always_inline void rep_nop(void)
 {
 	asm volatile("rep; nop" ::: "memory");
 }
 
-static inline void cpu_relax(void)
+static __always_inline void cpu_relax(void)
 {
 	rep_nop();
 }
-- 
1.8.1.4

Re: [PATCH] x86/processor.h: Force inlining of cpu_relax()

[Borislav Petkov]

On Thu, Sep 24, 2015 at 02:02:29PM +0200, Denys Vlasenko wrote:
> On x86, cpu_relax() simply calls rep_nop(), which generates one
> instruction, PAUSE (aka REP NOP).
> 
> With this config:
> http://busybox.net/~vda/kernel_config_OPTIMIZE_INLINING_and_Os
> gcc-4.7.2 does not always inline rep_nop(): it generates
> several copies of this:
> 
> <rep_nop> (16 copies, 194 calls):
>        55                      push   %rbp
>        48 89 e5                mov    %rsp,%rbp
>        f3 90                   pause
>        5d                      pop    %rbp
>        c3                      retq
> 
> See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66122
> 
> This patch fixes this via s/inline/__always_inline/
> on rep_nop() and cpu_relax().
> (Forcing inlining only on rep_nop() causes gcc to
> deinline cpu_relax(), with almost no change in generated code).
> 
>     text     data      bss       dec     hex filename
> 88118971 19905208 36421632 144445811 89c1173 vmlinux.before
> 88118139 19905208 36421632 144444979 89c0e33 vmlinux

Looks ok to me, text even grows smaller.

Acked-by: Borislav Petkov <bp@suse.de>

-- 
Regards/Gruss,
    Boris.

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

[tip:x86/asm] x86/asm: Force inlining of cpu_relax()

[tip-bot for Denys Vlasenko]

Commit-ID:  0b101e62afe626ecae60173f92f1e0ec72151653
Gitweb:     http://git.kernel.org/tip/0b101e62afe626ecae60173f92f1e0ec72151653
Author:     Denys Vlasenko <dvlasenk@redhat.com>
AuthorDate: Thu, 24 Sep 2015 14:02:29 +0200
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Fri, 25 Sep 2015 09:44:34 +0200

x86/asm: Force inlining of cpu_relax()

On x86, cpu_relax() simply calls rep_nop(), which generates one
instruction, PAUSE (aka REP NOP).

With this config:

  http://busybox.net/~vda/kernel_config_OPTIMIZE_INLINING_and_Os

gcc-4.7.2 does not always inline rep_nop(): it generates several
copies of this:

  <rep_nop> (16 copies, 194 calls):
       55                      push   %rbp
       48 89 e5                mov    %rsp,%rbp
       f3 90                   pause
       5d                      pop    %rbp
       c3                      retq

See: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66122

This patch fixes this via s/inline/__always_inline/
on rep_nop() and cpu_relax().

( Forcing inlining only on rep_nop() causes GCC to
  deinline cpu_relax(), with almost no change in generated code).

      text     data      bss       dec     hex filename
  88118971 19905208 36421632 144445811 89c1173 vmlinux.before
  88118139 19905208 36421632 144444979 89c0e33 vmlinux

Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Link: http://lkml.kernel.org/r/1443096149-27291-1-git-send-email-dvlasenk@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/include/asm/processor.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index 19577dd..b55f309 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -556,12 +556,12 @@ static inline unsigned int cpuid_edx(unsigned int op)
 }
 
 /* REP NOP (PAUSE) is a good thing to insert into busy-wait loops. */
-static inline void rep_nop(void)
+static __always_inline void rep_nop(void)
 {
 	asm volatile("rep; nop" ::: "memory");
 }
 
-static inline void cpu_relax(void)
+static __always_inline void cpu_relax(void)
 {
 	rep_nop();
 }