Re: [PATCH 4/7] drm/vc4: Use drm_atomic_helper_wait_for_flip_done()

[Boris Brezillon <boris.brezillon-wi1+55ScJUtKEb57/3fJTNBPR1lH4CV8@public.gmane.org>]

On Tue, 06 Jun 2017 13:24:33 -0700
Eric Anholt <eric-WhKQ6XTQaPysTnJN9+BGXg@public.gmane.org> wrote:

> Boris Brezillon <boris.brezillon-wi1+55ScJUtKEb57/3fJTNBPR1lH4CV8@public.gmane.org> writes:
> 
> > Replace the drm_atomic_helper_wait_for_vblanks() with a call to
> > drm_atomic_helper_wait_for_flip_done(). This allows better detection of
> > page flip done events which what we are really waiting for in
> > vc4_atomic_complete_commit().
> >
> > With this approach, we also addresse the 'missed single vblank event'
> > problem that can arise when the CRTC is configured in oneshot mode
> > (only a single frame is generated and the CRTC is immediately paused
> > after this frame). Note that this oneshot mode will be used for the
> > writeback connector feature.  
> 
> Should we just use drm_atomic_helper_commit_tail() and make this change
> in the core helper, instead?

Hm, not sure changing the default behavior is such a good idea. I don't
want to break other drivers.

> 
> Actually, I'm confused.  drm_atomic_helper_commit_cleanup_done() seems
> to be waiting for the flip_done on the crtc, already.  What's the
> difference?

Actually, drm_atomic_helper_wait_for_flip_done() is called just before
drm_atomic_helper_cleanup_planes() which in turn is called before
drm_atomic_helper_commit_cleanup_done(). My understanding was that it
was unsafe to call plane->cleanup_fb() on FBs that are still in use, and
the only thing guaranteeing that FBs are not used anymore is the
flip_done event.

Maybe I'm wrong, and FB refcounting is enough to make sure resources
are preserved until page flip is actually done.
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html