Revert "drm/i915: Switch planes from transitional helpers to full atomic helpers"
authorDaniel Vetter <daniel.vetter@ffwll.ch>
Mon, 2 Mar 2015 15:35:20 +0000 (16:35 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Tue, 17 Mar 2015 21:29:51 +0000 (22:29 +0100)
This reverts commit 3f678c96abb43a977d2ea41aefccdc49e8a3e896.

We've been a bit too optimistic with this one here :(

The trouble is that internally we're still using these plane
update/disable hooks. Which was totally ok pre-atomic since the drm
core did all the book-keeping updating and these just mostly updated
hw state. But with atomic there's lots more going on, and it causes
heaps of trouble with the load detect code.

This one specifically cause a deadlock since both the load detect code
and the nested plane atomic helper functions tried to grab the same
locks. It only blows up because of the evil tricks though we play with
the implicit ww acquire context.

Applying this revert unearths the NULL deref on already freed
framebuffer objects reported as a regression in 4.0 by various people.

Fixing this will be fairly invasive, hence revert even for the
4.1-next queue.

Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Paul Bolle <pebolle@tiscali.nl>
Acked-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
drivers/gpu/drm/i915/intel_display.c

index 213ffc7..089acff 100644 (file)
@@ -12182,8 +12182,8 @@ void intel_plane_destroy(struct drm_plane *plane)
 }
 
 const struct drm_plane_funcs intel_plane_funcs = {
-       .update_plane = drm_atomic_helper_update_plane,
-       .disable_plane = drm_atomic_helper_disable_plane,
+       .update_plane = drm_plane_helper_update,
+       .disable_plane = drm_plane_helper_disable,
        .destroy = intel_plane_destroy,
        .set_property = drm_atomic_helper_plane_set_property,
        .atomic_get_property = intel_plane_atomic_get_property,