RE: [PATCH] Fix WARN_ON / WARN_ON_ONCE regression

[Ananiev, Leonid I]

> That's one cache miss.  One.  For the remainder of the benchmark,
> __warn_once is in cache and there are no more misses.  That's how
caches
> work ;)

Before patch for WARN_ON_ONCE we never had risk to get a cache miss for
__warn_once.
After patch we have.
> __warn_once is in cache and there are no more misses.
There is no insuerence that it is in the cache.

Next:
CPU 1Mb cache cache_alignment : 128  => -16% degradation of tbench
CPU 2Mb cache cache_alignment : 128  => -1%  degradation
CPU 4Mb cache cache_alignment : 64  => -70% degradation

An evictees depends from cache size and line size. last is more
essential in considered case.

Leonid
-----Original Message-----
From: Andrew Morton [mailto:akpm@xxxxxxxx] 
Sent: Wednesday, October 04, 2006 9:28 PM
To: Ananiev, Leonid I
Cc: tim.c.chen@xxxxxxxxxxxxxxx; Jeremy Fitzhardinge;
herbert@xxxxxxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx
Subject: Re: [PATCH] Fix WARN_ON / WARN_ON_ONCE regression

On Wed, 4 Oct 2006 20:57:57 +0400
"Ananiev, Leonid I" <leonid.i.ananiev@xxxxxxxxx> wrote:

> >Guys.  Please.  Help us out here.  None of this makes sense, and it's
> > possible that we have an underlying problem in there which we need
to
> know
> > about.
>  This is explantion:
> 
> The static variable __warn_once was "never" read (until there is no
bug)
> before patch "Let WARN_ON/WARN_ON_ONCE return the condition"
>
http://kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commi
> t;h=684f978347deb42d180373ac4c427f82ef963171
>  in WARN_ON_ONCE's line 
> - if (unlikely((condition) && __warn_once)) { \
> because 'condition' is false. There was no cache miss as a result.
> 
> Cache miss for __warn_once is happened in new lines
> + if (likely(__warn_once)) \
> + if (WARN_ON(__ret_warn_once)) \
> 

That's one cache miss.  One.  For the remainder of the benchmark,
__warn_once is in cache and there are no more misses.  That's how caches
work ;)

But it appears this isn't happening.  Why?
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/