GitLab

Currently, in Reset() we only call TryCancelAll() on the client
sequence to cancel tasks on device sequence. However, TryCancelAll()
cancels tasks in a best-effort, meaning a few tasks on device sequence
can still run. The tasks posts new tasks to the client sequence.
Therefore, V4L2ImageProcessor might still return previous frames after
Reset() is called.

In this CL, we have these changes:
1. Post additional task ResetTask() to device sequence after
   cancelling previous tasks.
2. Clear |running_jobs_| at ResetTask(), and check the dequeued buffer
   id at Dequeue(). Only return the buffers at |running_jobs_|.
3. Invalidate WeakPtr of the client sequence to cancel pending tasks
   on the client sequence.

With 1. and 2., we could guarantee that after ResetTask(), we don't
post task to |client_task_runner_| for returning frames. With 3., we
could guarantee the tasks on |client_task_runner_| posted by the tasks
on |device_task_runner_| prior to ResetTask() are cancelled.

Bug: 1004727
Test: Run video_decode_accelerator_tests on Kukui and Hana
Test: Run video_encode_accelerator_unittest on Kukui and Hana

Change-Id: Ic2f8084f7e7658488dac9e7d39b0f22be347795e
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1906858
Commit-Queue: Chih-Yu Huang <akahuang@chromium.org>
Reviewed-by: Alexandre Courbot <acourbot@chromium.org>
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Cr-Commit-Position: refs/heads/master@{#715612}