GitLab

Expanding/contracting a notification changes its size, which generates a new
LocalSurfaceId. Creating a new LocalSurfaceId makes viz create a new
viz::Surface upon submitting the compositor frame. Viz's SurfaceManager tracks
the lifetime and performs garbage collection for Surfaces, which reference
TransferableResources, which correspond to e.g. the notification buffers.
A viz::Surface can be destroyed and deleted. Destroyed means it is marked for
deletion. Deletion will happen at some point in the future during
GarbageCollectSurfaces.

Normally, a notification would be displayed+embedded in the message center so it
will have real references to the SurfaceIds. This means that the old viz Surface
can be garbage collected immediately if the notification produces more buffers.
But, if the notification is producing buffers while the message centre is
closed, it's not embedded in anything. In this case, Viz creates temporary
references. If you expand and then contract, two Viz Surfaces are created (one
for each LocalSurfaceId) but neither can be deleted - both only have temporary
references. These Surfaces hold onto the associated TransferableResources, the
buffers, and cause buffer starvation on the Android side. Android HWUI starts
waiting on a fence for a buffer but the ANR timeout occurs before the Viz
temporary reference timeout. This causes the SystemUI ANR.

Because the triggers for this bug - "notifications not being visible on the
Chrome side" + "receiving input on a notification on the Android side" are
completely asynchronous we can't realistically prevent these triggers from
occurring. Instead, this change explicitly tells Viz it can release the
resources by pushing empty compositor frames when the notification becomes
invisible.

Test: Android applications still can produce buffers
Test: Notifications appear correctly when closing/opening the message center.
Test: Following the steps in b/130856284, the ANR no longer reproduces.
Bug: b/130856284
Change-Id: I42529267f75d7c30a8bcb03586e6dd08f658c69b
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1645432Reviewed-by: kylechar <kylechar@chromium.org>
Reviewed-by: Peng Huang <penghuang@chromium.org>
Reviewed-by: Mitsuru Oshima <oshima@chromium.org>
Commit-Queue: Eliot Courtney <edcourtney@chromium.org>
Cr-Commit-Position: refs/heads/master@{#669068}