On Tue, Oct 02, 2018 at 06:19:52PM -0700, Jeykumar Sankaran wrote: > On 2018-10-01 13:30, Sean Paul wrote: > > On Wed, Sep 26, 2018 at 11:56:47AM -0700, Jeykumar Sankaran wrote: > > > On 2018-09-19 11:56, Sean Paul wrote: > > > > From: Sean Paul <seanpaul@xxxxxxxxxxxx> > > > > > > > > This patch sprinkles a few async/legacy_cursor_update checks > > > > through commit to ensure that cursor updates aren't blocked on vsync. > > > > There are 2 main components to this, the first is that we don't want > > to > > > > wait_for_commit_done in msm_atomic before returning from > > > > atomic_complete. > > > > The second is that in dpu we don't want to wait for frame_done events > > > > when > > > > updating the cursor. > > > > > > > > Signed-off-by: Sean Paul <seanpaul@xxxxxxxxxxxx> > > > > --- /snip > > > > diff --git a/drivers/gpu/drm/msm/msm_atomic.c > > > > b/drivers/gpu/drm/msm/msm_atomic.c > > > > index c1f1779c980f..7912130ce5ce 100644 > > > > --- a/drivers/gpu/drm/msm/msm_atomic.c > > > > +++ b/drivers/gpu/drm/msm/msm_atomic.c > > > > @@ -76,7 +76,8 @@ void msm_atomic_commit_tail(struct drm_atomic_state > > > > *state) > > > > kms->funcs->commit(kms, state); > > > > } > > > > > > > > - msm_atomic_wait_for_commit_done(dev, state); > > > > + if (!state->legacy_cursor_update) > > > I see state->async_update is updated after validation checks on the > > async > > > capabilities. Shouldn't we use > > > that var instead of state->legacy_cursor_update? > > > > + msm_atomic_wait_for_commit_done(dev, state); > > > > > > > > kms->funcs->complete_commit(kms, state); > > > > > > Do we need to introduce plane helpers atomic_async_update and > > > atomic_async_check in DPU before supporting > > > these wait skips? or are they irrelevant for this legacy async path? > > > > I was trying to limit the scope of this to just cursor updates. I think > > once/if > > support is added for generic async it makes sense to change over to that > > verbage. > > > Since SDM845 doesnt support dedicated CURSOR stages, Well, it has CURSOR type planes, which unlocks the cursor ioctls, so it kind of does :) > I think the right > way to add the cursor support should be to introduce the atomic async > support > in the driver and let the cursor frame update like regulary async commit. > > I need to explore on the right way to fit that in. It's as easy as implementing the async atomic hooks. If they're present, the cursor path goes through them. Upgrading from legacy cursor to generic async is pretty trivial in light of the other issues we're facing here. Sean > > Thanks, > Jeykumar S. > > Sean > > > > > > > > -- > > > Jeykumar S > > -- > Jeykumar S -- Sean Paul, Software Engineer, Google / Chromium OS
