Re: [Bug #15004] i915: *ERROR* Execbuf while wedged

[Chris Mason <chris.mason-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org>]

On Fri, Feb 05, 2010 at 11:17:47AM -0800, Jesse Barnes wrote:
> On Fri, 5 Feb 2010 14:09:16 -0500
> Chris Mason <chris.mason-QHcLZuEGTsvQT0dZR+AlfA@public.gmane.org> wrote:
> 
> > On Fri, Feb 05, 2010 at 11:01:00AM -0800, Jesse Barnes wrote:
> > > On Mon,  1 Feb 2010 01:43:19 +0100 (CET)
> > > "Rafael J. Wysocki" <rjw-KKrjLPT3xs0@public.gmane.org> wrote:
> > > 
> > > > This message has been generated automatically as a part of a
> > > > report of regressions introduced between 2.6.31 and 2.6.32.
> > > > 
> > > > The following bug entry is on the current list of known
> > > > regressions introduced between 2.6.31 and 2.6.32.  Please verify
> > > > if it still should be listed and let me know (either way).
> > > > 
> > > > 
> > > > Bug-Entry	:
> > > > http://bugzilla.kernel.org/show_bug.cgi?id=15004
> > > > Subject		: i915: *ERROR* Execbuf while wedged
> > > > Submitter	: tomas m <tmezzadra-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> > > > Date		: 2010-01-07 18:53 (25 days old)
> > > 
> > > Can you bisect this to a particular kernel commit?  The particular
> > > error message means the kernel detected a GPU hang.  That's usually
> > > a userspace bug, but the kernel should recover from it.
> > 
> > I see these about once a week, which would be a very difficult bisect.
> > The kernel does recover when I reboot, but beyond that the messages
> > loop forever.
> > 
> > I haven't yet seen it on 2.6.33-rc, but I'm still struggling with
> > suspend/resume failures and haven't really had a one week up time yet.

Ok, updating to rc7 and updating my xf86 driver to 2.10 seems to have
fixed up my suspend/resume problems.  So, I should be able to trigger
the execbuf problem again.

> 
> The fdo bug referenced from kernel bugzilla has a small workaround you
> might try.  It forces the driver to try to recover from the hang, so
> you might not need to reboot.

Well, the rebooting isn't a huge deal, but if there's something I can
track/kick or force to core dump, would it help?

-chris