[Intel-gfx] Regular DRM deadlock on SNB

[Ben Widawsky]

On 2012-08-17 12:41, Daniel Stone wrote:
> Hi,
> I've got a MacBookAir4,2 (mid-2011 version with Sandy Bridge), and 
> get
> pretty regular deadlocks in DRM.  I'm not 100% sure they all have the
> same root cause: the most regular one seems to occur under fairly
> heavy I/O pressure, whereas this one just randomly happened whilst
> running the Clutter testsuite.  The I/O one seems to be pretty easy 
> to
> trigger; if the system is under heavy memory pressure and you do a 
> lot
> of I/O, I can hit a deadlock pretty easily.  My usual case for this 
> is
> Chrome running with about a billion tabs open, a large compile just
> having finished, and then copying something between my encrypted 
> /home
> and an external USB drive.  I haven't got the trace for this handy 
> but
> will try to dig it up out of logs, or get it the next time it 
> happens.
>
> I've attached the one I just hit running the Clutter test suite 
> (which
> seemed to start from a BUG and go downhill from there), which is also
> available at http://hastebin.com/detumihuju.xml
>
> If it helps, I'm running under GNOME Shell.
>
> Any advice/help/fixes welcome, please!
>
> Cheers,
> Daniel
>

This doesn't appear to be stock 3.5-rc7 as I couldn't exactly match up 
the BUG_ON to line 2029. Could you tell exactly which BUG_ON you're 
hitting? seqno == 0, or !mutex_is_locked() are the two likely 
candidates.

-- 
Ben Widawsky, Intel Open Source Technology Center