[Intel-gfx] [PATCH 1/6] drm/i915: make CRTC enable/disable asynchronous

[Imre Deak]

On Wed, 2014-03-05 at 15:39 -0800, Jesse Barnes wrote:
> On Thu, 06 Mar 2014 01:29:14 +0200
> Imre Deak <imre.deak at intel.com> wrote:
> 
> > On Wed, 2014-03-05 at 14:48 -0800, Jesse Barnes wrote:
> > > This lets us return to userspace more quickly and should improve init
> > > and suspend/resume times as well, allowing us to return to userspace
> > > sooner.
> > 
> > IMHO this is a good move towards a full command queue based solution for
> > kms commands, where eventually we have to think less of concurrency.
> > That is if we can queue all the other kms commands too (flip,
> > set_plane). But I don't see why that wouldn't be possible.
> > 
> > Btw, why do you have a separate disable and enable queue?
> 
> As opposed to a dedicated work queue for both combined?  I had a
> separate queue in an earlier patch, but dropped it while debugging some
> other stuff.  We should bring it back to ensure ordering.  That would
> remove the need for a few of the syncs, and would also let us queue a
> check at the appropriate time on the same queue.

Ah, sorry I confused myself. I didn't notice you used the global
workqueue. I'm not sure about the ordering guarantees of the global
workqueue, but if that's not given then yea a dedicated ordered
workqueue would seem better.

In any case if this can move towards a more generic command queue
solution then it's great.

--Imre