[PATCH weston] libweston/compositor-drm: Reset repaint scheduled status when setting DPMS level to off

[Marius-cristian Vlad]

On Fri, 2018-03-09 at 09:35 +0000, Daniel Stone wrote:
> Hi Marius,
> 
> On 7 March 2018 at 17:36, Marius Vlad <marius-cristian.vlad at nxp.com>
> wrote:
> > Otherwise when setting dpms level DPMS_ON,
> > weston_output_schedule_repaint()
> > will bail out early and never get a chance to wake up the output.
> > 
> > Arguably this could also be done in drm_set_dpms() when setting
> > dpms_off_pending
> > but I figure it better to do it when deferred.
> 
> Thanks, that's a good catch, but I do wonder how it wasn't getting
> tripped up before. In previous releases, we'd call drm_set_dpms()
> from
> anywhere, which would block on the update completing and then set the
> DPMS level. I wonder if it's because we would receive a flip-done
> event anyway and call weston_output_finish_frame() after the DPMS had
> completed, which would drive us into repaint-idle.

Indeed, I've submitted a previous patch because for timeline
instrumentation because of this. weston_output_finish_frame() is being
called with a NULL timestamp.

> 
> I suspect we should be calling finish_frame() to match that
> behaviour,
> to indicate that the previous update has completed, and then
> synchronously applying DPMS state. Pretending that
> disable/destroy/DPMS-off can be called at any time is really biting
> though. :(

I don't know enough about the repaint machinery, should this be done
from drm_set_dpms() or from drm_output_update_complete()?. I'm hitting
some asserts in both cases, maybe I'm doing it wrong :/.



> Cheers,
> Daniel