drm/atomic: Wait for vblank whenever a plane is added to state.

There's 2 reasons for doing a vblank wait:
- To fulfill uabi expectations, but the legacy ioctls are ill-defined
  enough that we really only need this when we do send out an event.
- To make sure we don't tear down mappings before the scanout engine
  stops accessing it.

The later is problematic with the current code since e.g. rotation
might need a different mapping than normal orientation. And rotation
is a plane property, and not on the fb. Hence we need to remove this
optimization.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
[danvet: Completely new commit message.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/1481204729-9058-5-git-send-email-maarten.lankhorst@linux.intel.com

diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c
index d1d678d63b84dad1f9692af77df187a018e79c39..f5b2a688eb0d0da57bd8d40a9e075f9c76b9be7c 100644 (file)
--- a/drivers/gpu/drm/drm_atomic_helper.c
+++ b/drivers/gpu/drm/drm_atomic_helper.c
@@ -1129,11 +1129,7 @@ drm_atomic_helper_wait_for_vblanks(struct drm_device *dev,
        for_each_crtc_in_state(old_state, crtc, old_crtc_state, i) {
                struct drm_crtc_state *new_crtc_state = crtc->state;
 
-               if (!new_crtc_state->active)
-                       continue;
-
-               if (!drm_atomic_helper_framebuffer_changed(dev,
-                               old_state, crtc))
+               if (!new_crtc_state->active || !new_crtc_state->planes_changed)
                        continue;
 
                ret = drm_crtc_vblank_get(crtc);