On Thu, Apr 28, 2016 at 06:20:30PM +0100, Dave Gordon wrote: > On 28/04/16 17:36, Chris Wilson wrote: > >On Thu, Apr 28, 2016 at 05:26:21PM +0100, Dave Gordon wrote: > >>intel_pin_and_map_ringbuffer_obj() should have been setting the dirty > >>flag, but wasn't; and the LRC code in intel_lr_context_do_pin() is also > >>updated to use the new function. > > > >Technically it doesn't need to. Whilst there is any valid content in the > >ringbuffer, it is pinned for use by requests and the hadware. Unlike the > >context object whose content does need to persist when inactive. > >-Chris > > What about across suspend and/or hibernate? Any chance of there > being valid (new) content in a ringbuffer that gets thrown away? > Though presumably we insist that all pending work is completed > before allowing hibernation/suspend anyway? Correct, before suspend we wait for all GPU work to be completed. We don't want the GPU writing into memory after the suspend/hibernation thinks we are ready. The pages are pinned by our activity as well. -Chris -- Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
