Re: System freeze on shutdown - bisected to commit 29d08b3efddc

From: Marc Dionne
Date: Wed Oct 06 2010 - 23:46:29 EST


On Wed, Oct 6, 2010 at 11:38 PM, Ben Skeggs <skeggsb@xxxxxxxxx> wrote:
> v       gOn Wed, 2010-10-06 at 23:29 -0400, Marc Dionne wrote:
>> On Tue, Oct 5, 2010 at 8:21 PM, Dave Airlie <airlied@xxxxxxxxxx> wrote:
>> > On Tue, 2010-10-05 at 18:24 -0400, Marc Dionne wrote:
>> >> Hi,
>> >>
>> >> My system has been crashing these past few days when issuing "reboot"
>> >> or hitting the Shut Down button in gnome.  I get a series of stack
>> >> traces scrolling on the console before things freeze up completely.
>> >> The traces that I do see at the end point to kmem_cache_alloc called
>> >> from get_empty_filp, but there are typically various other traces
>> >> above it.
>> >
>> > cc'ing Ben.
>> >
>> > Dave.
>>
>> Some more testing shows that what causes the crash is when
>> drm_gem_object_unreference_unlocked gets called in
>> nouveau_notifier_takedown_channel.  The refcount on the object is
>> always 1 at that point - on entry to nouveau_notifier_takedown I see a
>> handle_count of 0 and a refcount of 1 in most cases, and sometimes a
>> handle_count of 1 and a refcount of 2.  Either way the call to
>> drm_gem_object_unreference_unlocked will bring the refcount to 0.
>> Not sure if that's normal at that point.
>>
>> Commenting out that particular call to
>> drm_gem_object_unreference_unlocked gets rid of the crash.
> How about if you remove the drm_get_object_handle_unreference_unlocked()
> call from that function instead?
>
> Ben.

I had tested that earlier and just retested it now, and yes it also
cures the crash.

Marc
--
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/