Re: [PATCH v7 1/4] spinlock: A new lockref structure for locklessupdate of refcount

[Sedat Dilek]

On Fri, Aug 30, 2013 at 6:32 PM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Fri, Aug 30, 2013 at 8:38 AM, Sedat Dilek <sedat.dilek@xxxxxxxxx> wrote:
>>
>> A Samsung series-5 ultrabook.
>>
>> $ grep "model name" /proc/cpuinfo | uniq
>> model name      : Intel(R) Core(TM) i5-2467M CPU @ 1.60GHz
>
> Hmm. Do you have debugging options enabled? Because that CPU should
> have the same core count as mine (two+HT), a slightly smaller cache
> (3M vs 4M) and runs at a noticeably lower frequency (1.6GHz vs 3.5).
> It probably also has slower memory etc, but that should still make it
> maybe half speed of mine. Not 1/20th.
>
> As mentioned, I get numbers in the 65M range. Yours are under 2.7M.
> Even with some thermal throttling, I would expect better than that.
>
> My pixel (1.8GHz i5-3427U) should be a *bit* faster that yours. And I
> get 54M iterations on that.
>
> I saw you mentioned CONFIG_CMPXCHG_LOCKREF=y in your .config, so you
> don't have spinlock debugging enabled, but maybe you have some other
> expensive debug option enabled. Like DEBUG_PAGEALLOC etc.
>

Yeah, this was with some debug-options built (see my attached kernel-config).

> If you get "perf" compiled, mind doing a
>
>     perf record -f -e cycles:pp ./a.out
>     perf report
>
> on it and look what that says?
>

Where is this a.out file from or how to generate it?

- Sedat -

Attachment: config-3.11.0-rc7-1-lockref-small
Description: Binary data