drm/atomic-helpers: document how to implement async commit

No helper function to do it all yet provided since no driver has
support for driver core fences yet. Which we'd need to make the
implementation really generic.

v2: Clarify async howto a bit per the discussion With Rob Clark.

Cc: Rob Clark <robdclark@gmail.com>
Reviewed-by: Sean Paul <seanpaul@chromium.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

diff --git a/drivers/gpu/drm/drm_atomic_helper.c b/drivers/gpu/drm/drm_atomic_helper.c
index 4808e71..6668d2c 100644 (file)
--- a/drivers/gpu/drm/drm_atomic_helper.c
+++ b/drivers/gpu/drm/drm_atomic_helper.c
@@ -848,6 +848,44 @@ int drm_atomic_helper_commit(struct drm_device *dev,
 }
 EXPORT_SYMBOL(drm_atomic_helper_commit);
 
+/**
+ * DOC: implementing async commit
+ *
+ * For now the atomic helpers don't support async commit directly. If there is
+ * real need it could be added though, using the dma-buf fence infrastructure
+ * for generic synchronization with outstanding rendering.
+ *
+ * For now drivers have to implement async commit themselves, with the following
+ * sequence being the recommended one:
+ *
+ * 1. Run drm_atomic_helper_prepare_planes() first. This is the only function
+ * which commit needs to call which can fail, so we want to run it first and
+ * synchronously.
+ *
+ * 2. Synchronize with any outstanding asynchronous commit worker threads which
+ * might be affected the new state update. This can be done by either cancelling
+ * or flushing the work items, depending upon whether the driver can deal with
+ * cancelled updates. Note that it is important to ensure that the framebuffer
+ * cleanup is still done when cancelling.
+ *
+ * For sufficient parallelism it is recommended to have a work item per crtc
+ * (for updates which don't touch global state) and a global one. Then we only
+ * need to synchronize with the crtc work items for changed crtcs and the global
+ * work item, which allows nice concurrent updates on disjoint sets of crtcs.
+ *
+ * 3. The software state is updated synchronously with
+ * drm_atomic_helper_swap_state. Doing this under the protection of all modeset
+ * locks means concurrent callers never see inconsistent state. And doing this
+ * while it's guaranteed that no relevant async worker runs means that async
+ * workers do not need grab any locks. Actually they must not grab locks, for
+ * otherwise the work flushing will deadlock.
+ *
+ * 4. Schedule a work item to do all subsequent steps, using the split-out
+ * commit helpers: a) pre-plane commit b) plane commit c) post-plane commit and
+ * then cleaning up the framebuffers after the old framebuffer is no longer
+ * being displayed.
+ */
+
 /**
  * drm_atomic_helper_prepare_planes - prepare plane resources after commit
  * @dev: DRM device