Re: [PATCH RT 00/02] SLOB optimizations

[Zwane Mwaikambo <zwane@arm.linux.org.uk>]

On Wed, 21 Dec 2005, Eric Dumazet wrote:

> Ingo Molnar a �crit :
> > * Eric Dumazet <dada1@cosmosbay.com> wrote:
> > 
> > in any case, on sane platforms (i386, x86_64) an irq-disable is
> > well-optimized in hardware, and is just as fast as a preempt_disable().
> > 
> 
> I'm afraid its not the case on current hardware.
> 
> The irq enable/disable pair count for more than 50% the cpu time spent in
> kmem_cache_alloc()/kmem_cache_free()/kfree()
> 
> oprofile results on a dual Opteron 246 :
> 
> You can see the high profile numbers right after cli and popf(sti)
> instructions, popf being VERY expensive.
> 
> CPU: Hammer, speed 1993.39 MHz (estimated)
> Counted CPU_CLK_UNHALTED events (Cycles outside of halt state) with a unit
> mask of 0x00 (No unit mask) count 50000
> 
> 29993     1.9317  kfree
> 18654     1.2014  kmem_cache_alloc
> 12962     0.8348  kmem_cache_free
> 
> ffffffff8015c370 <kfree>: /* kfree total:  30334  1.9335 */
>    770  0.0491 :ffffffff8015c370:       push   %rbp
>   2477  0.1579 :ffffffff8015c371:       mov    %rdi,%rbp
>                :ffffffff8015c374:       push   %rbx
>                :ffffffff8015c375:       sub    $0x8,%rsp
>   1792  0.1142 :ffffffff8015c379:       test   %rdi,%rdi
>                :ffffffff8015c37c:       je     ffffffff8015c452 <kfree+0xe2>
>    122  0.0078 :ffffffff8015c382:       pushfq
>   1001  0.0638 :ffffffff8015c383:       popq   (%rsp)
>   1456  0.0928 :ffffffff8015c386:       cli
>   2489  0.1586 :ffffffff8015c387:       mov    $0xffffffff7fffffff,%rax    <<
> 
> ...
>     72  0.0046 :ffffffff8015c44e:       pushq  (%rsp)
>   1080  0.0688 :ffffffff8015c451:       popfq
>  13934  0.8882 :ffffffff8015c452:       add    $0x8,%rsp      << HERE >>

Isn't that due to taking an interrupt?