[Intel-gfx] [PATCH] drm/i915: Use our own workqueue to avoid wedging the system along with the GPU.

[Chris Wilson]

I've just encountered a situation whereby saturating the GPU leaves the
system barely responsive as the system waits for dev->struct_mutex.
Either using a separate work queue or adjusting the retire_request work
handler to use mutex_trylock() makes the system responsive again (though
obviously not the application that is submitting 4s of rendering between
throttles.
-ickle