drm: add some comments to drm_wait_vblank and drm_queue_vblank_event
authorIlija Hadzic <ihadzic@research.bell-labs.com>
Mon, 31 Oct 2011 17:11:57 +0000 (13:11 -0400)
committerDave Airlie <airlied@redhat.com>
Fri, 11 Nov 2011 11:16:30 +0000 (11:16 +0000)
during the review of the fix for locks problems in drm_wait_vblank,
a couple of false concerns were raised about how the drm_vblank_get
and drm_vblank_put are used in this function; it turned out that the
code is correct and that it cannot be simplified

add a few comments to explain non-obvious flows in the code,
to prevent "false alarms" in the future

v2: incorporate comments received from Daniel Vetter

Signed-off-by: Ilija Hadzic <ihadzic@research.bell-labs.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
drivers/gpu/drm/drm_irq.c

index cb3794a00f98c541264404164f0eba269736126b..aa3ac0125d9047110b077e84d90a50c47152b3cb 100644 (file)
@@ -1125,6 +1125,7 @@ static int drm_queue_vblank_event(struct drm_device *dev, int pipe,
                trace_drm_vblank_event_delivered(current->pid, pipe,
                                                 vblwait->request.sequence);
        } else {
+               /* drm_handle_vblank_events will call drm_vblank_put */
                list_add_tail(&e->base.link, &dev->vblank_event_list);
                vblwait->reply.sequence = vblwait->request.sequence;
        }
@@ -1205,8 +1206,12 @@ int drm_wait_vblank(struct drm_device *dev, void *data,
                goto done;
        }
 
-       if (flags & _DRM_VBLANK_EVENT)
+       if (flags & _DRM_VBLANK_EVENT) {
+               /* must hold on to the vblank ref until the event fires
+                * drm_vblank_put will be called asynchronously
+                */
                return drm_queue_vblank_event(dev, crtc, vblwait, file_priv);
+       }
 
        if ((flags & _DRM_VBLANK_NEXTONMISS) &&
            (seq - vblwait->request.sequence) <= (1<<23)) {