cpu_relax(), rep: nop, and PAUSE

cpu_relax(), rep: nop, and PAUSE

[David Shwatrz]

Hi, kernel newbies,

We have:
#define cpu_relax()	asm volatile("rep; nop")
in arch/x86/boot/boot.h.

Why don't we use the PAUSE assembler instruction here ?

According to Intel manuals,
Intel? 64 and IA-32 Architectures
Software Developer?s Manual
Volume 2B:

"This instruction was introduced in the Pentium 4 processors, but is
backward compatible with all IA-32 processors.
In earlier IA-32 processors, the PAUSE instruction operates like a NOP
instruction. The Pentium 4 and Intel Xeon
processors implement the PAUSE instruction as a delay. The delay is
finite and can be zero for some processors.
This instruction does not change the architectural state of the
processor (that is, it performs essentially a delaying
no-op operation).
This instruction?s operation is the same in non-64-bit modes and 64-bit mode."

And AFAIK, in spinlocks , PAUSE indeed replaced the rep;nop.

Any ideas?

Best,
DS

cpu_relax(), rep: nop, and PAUSE

[Mulyadi Santosa]

On Tue, Feb 19, 2013 at 7:20 PM, David Shwatrz <dshwatrz@gmail.com> wrote:
> Hi, kernel newbies,
>
> We have:
> #define cpu_relax()     asm volatile("rep; nop")
> in arch/x86/boot/boot.h.
>
> Why don't we use the PAUSE assembler instruction here ?

Just guessing, maybe rep+nop could do better power saving because
processor is considered as idle.



-- 
regards,

Mulyadi Santosa
Freelance Linux trainer and consultant

blog: the-hydra.blogspot.com
training: mulyaditraining.blogspot.com

cpu_relax(), rep: nop, and PAUSE

[Valdis.Kletnieks at vt.edu]

On Wed, 20 Feb 2013 01:58:17 +0700, Mulyadi Santosa said:
> On Tue, Feb 19, 2013 at 7:20 PM, David Shwatrz <dshwatrz@gmail.com> wrote:
> > Hi, kernel newbies,
> >
> > We have:
> > #define cpu_relax()     asm volatile("rep; nop")
> > in arch/x86/boot/boot.h.
> >
> > Why don't we use the PAUSE assembler instruction here ?
>
> Just guessing, maybe rep+nop could do better power saving because
> processor is considered as idle.

The 'rep; nop' is actually a placeholder - for some CPUs, a different opcode
gets filled in during boot time.  See arch/x86/kernel/alternative.c and
arch/x86/include/asm/alternative.h for the gory details.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 865 bytes
Desc: not available
Url : http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20130219/a0c43ea8/attachment.bin 

cpu_relax(), rep: nop, and PAUSE

[Adam Lee]

On Tue, Feb 19, 2013 at 02:20:11PM +0200, David Shwatrz wrote:
> Hi, kernel newbies,
> 
> We have:
> #define cpu_relax()	asm volatile("rep; nop")
> in arch/x86/boot/boot.h.
> 
> Why don't we use the PAUSE assembler instruction here ?

But rep_nop and pause ought to be the same, why we change it? "If it
ain't broke, don't fix it."

You can see that by disassembling, they both are "f3 90", "f3" is rep,
"90" is nop.

> And AFAIK, in spinlocks , PAUSE indeed replaced the rep;nop.

Where? I haven't found that.

-- 
Regards,
Adam Lee
http://adam8157.info

cpu_relax(), rep: nop, and PAUSE

[Rainer Müller]

On 2013-02-19 13:20, David Shwatrz wrote:
> We have:
> #define cpu_relax()	asm volatile("rep; nop")
> in arch/x86/boot/boot.h.
> 
> Why don't we use the PAUSE assembler instruction here ?

If you dig further into the Intel x86 manual, the machine instructions
'pause' and 'rep;nop' actually use the exact same encoding. As some
older assembler might not know the 'pause' mnemonic, using 'rep;nop'
here is most probably for backwards compatibility.

Rainer