[Intel-gfx] [PATCH v2] drm/i915: Fallback to using unmappable memory for scanout

[Chris Wilson]

On Thu, Mar 19, 2015 at 05:34:09PM +0100, Daniel Vetter wrote:
> On Thu, Mar 19, 2015 at 01:10:13PM +0000, Chris Wilson wrote:
> > On Thu, Mar 19, 2015 at 06:31:04PM +0530, Deepak S wrote:
> > > should we skip put_fence in overlay_do_put_image ?
> > 
> > Ah interesting point you raise there. That is buggy code fullstop.
> > We should not be call put_fence if pin_to_display_plane pins the fence.
> > Techinically the overlay could use a fence, the restriction is more to
> > do with an artificial limitation on the overlay API, and so the actual
> > call to i915_gem_object_put_fence() can be removed without any ill
> > side-effects. Something for the next time someone considers gen2-4 code.
> 
> The put_fence in there is iirc to synchronize with the lazy tiling, just
> in case. Or at least that's the story I remember.

Right, that's before we then did the pin_fence. We should have caught
the conflict then. Fortunately, it should just come out in the wash now.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre