Re: [PATCH] drm/i915: Check visibility in icl_build_plane_wm

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Tue, Jun 18, 2019 at 02:44:00PM +0300, Ville Syrjälä wrote:
> On Tue, Jun 18, 2019 at 11:16:41AM +0200, Maarten Lankhorst wrote:
> > Op 17-06-2019 om 14:34 schreef Ville Syrjälä:
> > > On Fri, Jun 14, 2019 at 12:39:41PM +0200, Maarten Lankhorst wrote:
> > >> When a planar YUV plane is configured, but the crtc is
> > >> marked inactive, we can end up with a linked plane without
> > >> visibility.
> > > How is that possible? I don't think we should be adding the slave plane
> > > if the master is not visible.
> > 
> > 
> > DPMS off, we calculate the various fields as if the CRTC is on, then disable visibility.
> > 
> > crtc_state->nv12_planes etc still get set, so it works as if the crtc is on.
> > 
> > It's a way of not allowing an invalid result when dpms is off, then breaking on crtc enable.
> 
> Hmm. I wonder when we started to do that. If we're already doing this
> much then I wonder how far we are from just dealing with the FIXME in
> intel_wm_plane_visible() instead?

Still far I guess. Would potentially need to do some surgery on
the pipe ddb allocation as well.

This whole thing is a bit borked. We clear active_planes but still
use it when allocating the Y plane. Hence dpms on could just fail
anyway due to not having a free Y plane (as well as due to
insufficient watermarks).

So if we want to make the Y plane allocation robust I guess we would
also need to move clearing the plane visibility to happen after the 
crtc .atomic_check().

-- 
Ville Syrjälä
Intel
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx