lkml.org 
[lkml]   [2011]   [Aug]   [5]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: list corruption in the last few days. (block ? crypto ?)
On Fri, 5 Aug 2011, Pekka Enberg wrote:

> On Fri, Aug 5, 2011 at 11:46 AM, Markus Trippelsdorf
> <markus@trippelsdorf.de> wrote:
> > On 2011.08.04 at 21:00 -0400, Dave Jones wrote:
> >> Sometime in the last week, something was merged which causes my laptop
> >> to lock up occasionally.  I can trigger it most of the time just by
> >> doing a kernel build. When it gets to the final linking stage, it locks up hard.
> >>
> >> I finally managed to coax something out of usb console to get the traces below,
> >> which seem to implicate something in the block layer ?
> >>
> >> my root device is an lvm volume on an dmcrypt'd block dev, which might be relevant,
> >> as I don't see this happening on other machines with simpler setups.
> >>
> >> I'm going to try bisecting, but it might take me a few days, because it's
> >> such a pain in the ass to reproduce this reliably.
> >>
> >> [ 5913.233035] ------------[ cut here ]------------
> >> [ 5913.233097] WARNING: at lib/list_debug.c:53 __list_del_entry+0x8d/0x98()
> >> [ 5913.233101] Hardware name: Adamo 13
> >> [ 5913.233105] list_del corruption. prev->next should be ffffea000434fd20, but was ffffea0004199520
> >
> > See also: http://lkml.org/lkml/2011/8/3/37
>
> That's in networking so SLUB lockless patches are almost certainly the
> issue here. Is this with SLUB debugging enabled or not? Christoph, it
> like the partial lists are getting corrupted somehow.

This is occurring in __slab_free when we are freeing the last
object from a slab page that is on the partial list. It is not frozen so
it is not in use by a processor and thus deactivate_slab cannot be run
on it.

The logical race would be with acquire_slab() but both take the node lock
before doing anything with the lists.

Do you have CONFIG_DEBUG_VM on? If not please do so. This will check if
the frozen state is managed correctly.
\
 
 \ /
  Last update: 2011-08-05 18:35    [from the cache]
©2003-2014 Jasper Spaans. Advertise on this site