[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20080718001640.GA1544@elte.hu>
Date: Fri, 18 Jul 2008 02:16:40 +0200
From: Ingo Molnar <mingo@...e.hu>
To: Vegard Nossum <vegard.nossum@...il.com>
Cc: linux-kernel@...r.kernel.org, netdev@...r.kernel.org,
Pekka Enberg <penberg@...helsinki.fi>,
"Rafael J. Wysocki" <rjw@...k.pl>
Subject: Re: [bug, netconsole, SLUB] BUG skbuff_head_cache: Poison
overwritten
* Vegard Nossum <vegard.nossum@...il.com> wrote:
> On Fri, Jul 18, 2008 at 1:52 AM, Ingo Molnar <mingo@...e.hu> wrote:
> > If only we had some kernel technology that could track and validate
> > memory accesses, and point out the cases where we access uninitialized
> > memory, just like Valgrind?
> >
> > ... something like kmemcheck? ;-)
>
> Cool :)
>
> > So i booted that box with tip/master and kmemcheck enabled. (plus a few
> > fixlets to make networking allocations be properly tracked by
> > kmemcheck.)
> >
> > It was a slow bootup and long wait, but it gave a few hits here:
>
> Hm, if you think it was that slow, I am suspecting you were also using
> SLUB debugging.
nope:
# CONFIG_SLUB_DEBUG is not set
CONFIG_SLUB=y
> This can actually be negative, since now SLUB will access the objects
> (+redzone +padding) and possibly trick kmemcheck into thinking they
> were initialized in the first place.
>
> But what we are really looking for is "read from freed memory"
> messages. So I would actually recommend this: Disable kmemcheck's
> reporting of uninitialized memory, simply to make it easier to spot
> the "freed" messages more easily.
>
> Maybe something like this (warning: whitespace-munged):
ok, applied this too.
> If this only happens during boot, it would also be a good idea to
> simply reboot the machine a lot...
yeah, i've got a script for that. Will try it overnight.
Ingo
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists