On Fri, Jan 15, 2016 at 04:05:24PM +0000, Tvrtko Ursulin wrote:
>
> On 15/01/16 16:01, Chris Wilson wrote:
> >On Fri, Jan 15, 2016 at 03:10:29PM +0000, Tvrtko Ursulin wrote:
> >>@@ -1118,15 +1126,22 @@ void intel_lr_context_unpin(struct drm_i915_gem_request *rq)
> >>+ if (--rq->ctx->engine[ring->id].pin_count == 0) {
> >>+ lrc_state_page = i915_gem_object_get_dirty_page(ctx_obj,
> >>+ LRC_STATE_PN);
> >
> >Interesting choice. We called set_page_dirty() when we took the mapping.
> >Should that page flag be preserved whilst we hold the kmap - I think so,
> >i.e. the mm cannot flush the page whilst it has an elevated mapcount. So
> >calling set_page_dirty() again is redundant, right?
>
> If you call mindless copy & paste interesting. :D
>
> Any other concerns or I can respin with that only?
No. I was quibbling over the excess clearing of state on unpinning :)
Pity we have to respin even for innoculous changes just to get a CI
tick.
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
