[Intel-gfx] [PATCH 50/64] drm/i915: Prepare i915_gem_active for annotations

[Tvrtko Ursulin]

On 13/07/16 16:58, Chris Wilson wrote:
> On Wed, Jul 13, 2016 at 04:40:03PM +0100, Tvrtko Ursulin wrote:

[snip]

>>>   	} else {
>>>   		for (i = 0; i < I915_NUM_ENGINES; i++) {
>>>   			struct drm_i915_gem_request *req;
>>>
>>> -			req = obj->last_read[i].request;
>>> +			req = i915_gem_active_peek(&obj->last_read[i]);
>>>   			if (req == NULL)
>>>   				continue;
>>>
>>> -			requests[n++] = i915_gem_request_get(req);
>>> +			requests[n++] = req;
>>>   		}
>>>   	}
>>>
>>> @@ -2383,25 +2386,27 @@ void i915_vma_move_to_active(struct i915_vma *vma,
>>>   static void
>>>   i915_gem_object_retire__write(struct drm_i915_gem_object *obj)
>>>   {
>>> -	GEM_BUG_ON(!obj->last_write.request);
>>> -	GEM_BUG_ON(!(obj->active & intel_engine_flag(obj->last_write.request->engine)));
>>> +	GEM_BUG_ON(!__i915_gem_active_is_busy(&obj->last_write));
>>> +	GEM_BUG_ON(!(obj->active & intel_engine_flag(i915_gem_active_get_engine(&obj->last_write))));
>>>
>>> -	i915_gem_request_assign(&obj->last_write.request, NULL);
>>> +	i915_gem_active_set(&obj->last_write, NULL);
>>
>> Aha!
>
> Drat. Didn't think I did that...
>
> Oh well, no excuses now but to go back in time and make the change
> earlier. It does get removed eventually!

Probably not worth it. You can have a special dispensation since I am 
reviewing all the same lines of code patch after patch anyway. :)

Regards,

Tvrtko