2002-12-03 15:34:42

by Calin A. Culianu

[permalink] [raw]
Subject: SMP Pentium4 -- PAUSE Instruction


I as wondering -- according to Intel's docs they recommend that on a P4
processor to use the PAUSE instruction (aka rep followed by a nop) inside
any spin loop (such as one used in SMP spinlock code) in order to both
improve processor performance and reduce power consumption.

Is this instruction being used in spin-wait loops? For some reason, I am
having a hard time figuring out whether or not it is being used. There is
a rep_nop() in processor.h.. but I can't determine if that is being called
for spin lock lock/unlock code.


-Calin



2002-12-03 15:49:44

by Nakajima, Jun

[permalink] [raw]
Subject: RE: SMP Pentium4 -- PAUSE Instruction

That one is for tight loops. Spin locks are inlined, and PAUSE is used like:

#define spin_lock_string \
"\n1:\t" \
"lock ; decb %0\n\t" \
"js 2f\n" \
LOCK_SECTION_START("") \
"2:\t" \
"cmpb $0,%0\n\t" \
"rep;nop\n\t" \ <---
"jle 2b\n\t" \
"jmp 1b\n" \
LOCK_SECTION_END

Also take a look at arch/i386/kernel/semaphore.c for read/write_locks.

Jun

> -----Original Message-----
> From: Calin A. Culianu [mailto:[email protected]]
> Sent: Tuesday, December 03, 2002 7:42 AM
> To: Linux Kernel Mailing List
> Subject: SMP Pentium4 -- PAUSE Instruction
>
>
> I as wondering -- according to Intel's docs they recommend that on a P4
> processor to use the PAUSE instruction (aka rep followed by a nop) inside
> any spin loop (such as one used in SMP spinlock code) in order to both
> improve processor performance and reduce power consumption.
>
> Is this instruction being used in spin-wait loops? For some reason, I am
> having a hard time figuring out whether or not it is being used. There is
> a rep_nop() in processor.h.. but I can't determine if that is being called
> for spin lock lock/unlock code.
>
>
> -Calin
>
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/

2002-12-03 15:50:35

by Dave Jones

[permalink] [raw]
Subject: Re: SMP Pentium4 -- PAUSE Instruction

On Tue, Dec 03, 2002 at 10:42:13AM -0500, Calin A. Culianu wrote:
>
> I as wondering -- according to Intel's docs they recommend that on a P4
> processor to use the PAUSE instruction (aka rep followed by a nop) inside
> any spin loop (such as one used in SMP spinlock code) in order to both
> improve processor performance and reduce power consumption.
> Is this instruction being used in spin-wait loops? For some reason, I am
> having a hard time figuring out whether or not it is being used. There is
> a rep_nop() in processor.h.. but I can't determine if that is being called
> for spin lock lock/unlock code.

there's also rep;nop in asm-i386/spinlock.h
See spin_lock_string()

Dave

--
| Dave Jones. http://www.codemonkey.org.uk
| SuSE Labs

2002-12-03 15:58:52

by Arjan van de Ven

[permalink] [raw]
Subject: Re: SMP Pentium4 -- PAUSE Instruction

On Tue, 2002-12-03 at 16:42, Calin A. Culianu wrote:

> Is this instruction being used in spin-wait loops? For some reason, I am
> having a hard time figuring out whether or not it is being used. There is
> a rep_nop() in processor.h.. but I can't determine if that is being called
> for spin lock lock/unlock code.

check cpu_relax() all over the kernel :)
and the spinlock code uses it inside it's own asm directly, not via
rep_nop()


2002-12-03 17:13:53

by Calin A. Culianu

[permalink] [raw]
Subject: Re: SMP Pentium4 -- PAUSE Instruction


:) Oh. Don't I feel like I hit the panic button too early... :/

-Calin

On 3 Dec 2002, Arjan van de Ven wrote:

> On Tue, 2002-12-03 at 16:42, Calin A. Culianu wrote:
>
> > Is this instruction being used in spin-wait loops? For some reason, I am
> > having a hard time figuring out whether or not it is being used. There is
> > a rep_nop() in processor.h.. but I can't determine if that is being called
> > for spin lock lock/unlock code.
>
> check cpu_relax() all over the kernel :)
> and the spinlock code uses it inside it's own asm directly, not via
> rep_nop()
>
>