On Fri, Mar 27, 2015 at 03:19:20PM +0100, Daniel Vetter wrote: > On Fri, Mar 27, 2015 at 11:01:53AM +0000, Chris Wilson wrote: > > When we submit a request to the GPU, we first take the rpm wakelock, and > > only release it once the GPU has been idle for a small period of time > > after all requests have been complete. This means that we are sure no > > new interrupt can arrive whilst we do not hold the rpm wakelock and so > > can drop the individual get/put around every single request inside > > execlists. > > > > Note: to close one potential issue we should mark the GPU as busy > > earlier in __i915_add_request. > > > > Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> > > With the create/add_request rework from John, should we do the idle->busy > check in the request alloc function, together with latching the worker? > Not perfect if the execbuf doesn't go through, but leaves no races and > userspace better submit valid execbufs anyway if it expects performance. I think it should be done in i915_request_commit(). The argument is that that are quite a few places where we may start building a request to decide that it is a no-op (e.g. fixing i915_gpu_idle() for execlists). -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/intel-gfx
