[Intel-gfx] [PATCH] drm/i915: Cancel pending execlist tasklet upon wedging

[Tvrtko Ursulin]

On 21/06/2017 14:53, Chris Wilson wrote:
> Quoting Tvrtko Ursulin (2017-06-21 14:33:59)
>>
>> On 21/06/2017 13:48, Chris Wilson wrote:
>>> Highly unlikely, but if the stop_machine() did suspend the tasklet, we
>>> want to make sure that when it wakes it finds there is nothing to do.
>>> Otherwise, it will loudly complain that the ELSP port tracking no longer
>>> matches the hardware, and we will be mightly confused.
>>>
>>> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
>>> Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>> ---
>>>    drivers/gpu/drm/i915/i915_gem.c | 7 +++++++
>>>    1 file changed, 7 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
>>> index 1353491c1010..ae3ce1314bd1 100644
>>> --- a/drivers/gpu/drm/i915/i915_gem.c
>>> +++ b/drivers/gpu/drm/i915/i915_gem.c
>>> @@ -3092,6 +3092,13 @@ static void engine_set_wedged(struct intel_engine_cs *engine)
>>>                engine->execlist_first = NULL;
>>>    
>>>                spin_unlock_irqrestore(&engine->timeline->lock, flags);
>>> +
>>> +             /* The port is checked prior to scheduling a tasklet, but
>>> +              * just in case we have suspended the tasklet to do the
>>> +              * wedging make sure that when it wakes, it decides there
>>> +              * is no work to do by clearing the irq_posted bit.
>>> +              */
>>> +             clear_bit(ENGINE_IRQ_EXECLIST, &engine->irq_posted);
>>>        }
>>>    }
>>>    
>>>
>>
>> Or tasklet_kill? Actually I am not sure now how the wedging transition
>> works for request which have pasted the submit callback, but are not in
>> the hardware yet (waiting in the priotree). Can't find at the moment
>> that we have anything dealing with those.
> 
> We kill the rbtree and the tree itself doesn't hold a reference to the
> request (just a reference is held for the ports, as prior to submission
> there is no way we complete the request and so cancel the in-flight
> reference count). That's the
> 
> 	engine->execlist_queue = RB_ROOT;
> 	engine->execlist_first = NULL;
> 
> above.

Yes of course, don't know where I was looking.

Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin at intel.com>

Regards,

Tvrtko