[Intel-gfx] [PATCH] drm/i915: Remove nested work in gpu error handling

[Mika Kuoppala]

Chris Wilson <chris at chris-wilson.co.uk> writes:

> On Mon, Feb 02, 2015 at 10:38:19AM +0100, Daniel Vetter wrote:
>> On Mon, Feb 02, 2015 at 09:17:14AM +0000, Chris Wilson wrote:
>> > On Wed, Jan 28, 2015 at 05:03:14PM +0200, Mika Kuoppala wrote:
>> > > @@ -2616,6 +2612,9 @@ void i915_handle_error(struct drm_device *dev, bool wedged,
>> > >  	va_list args;
>> > >  	char error_msg[80];
>> > >  
>> > > +	if (WARN_ON(mutex_is_locked(&dev_priv->dev->struct_mutex)))
>> > > +		return;
>> > > +
>> > 
>> > Oops, sorry, I should have realised this was wrong earlier. The mutex
>> > breaking occurs later in i915_handle_error.
>> 
>> Oh well, already merged. Also, prts seems to complain that a bunch of
>> hang stress-tests changed from fail to timeout because of this one here.
>> Is this patch accidentally fix a bug and we just need to tune the tests,
>> or is there some new deadlock now? prts results are really thin, per usual
>> :(
>
> Yes. It will also prevent the gpu reset which those tests depend upon.

Jani pointed me on the dups. There will be more I think. The test
should have been more deep into the reset handling and instead of
bailing out we should have requeued ourselves to the least.

Sorry.
-Mika

> -Chris
>
> -- 
> Chris Wilson, Intel Open Source Technology Centre