diff options
author | Laurent Pinchart <laurent.pinchart@ideasonboard.com> | 2022-06-21 22:47:53 +0300 |
---|---|---|
committer | Laurent Pinchart <laurent.pinchart@ideasonboard.com> | 2022-07-04 23:08:12 +0300 |
commit | b3fd8e14b52b5a42aca3c7dadca064b95cf19758 (patch) | |
tree | a206e837a8e3defd2151ff657e5c60b05621a272 /Documentation/guides/pipeline-handler.rst | |
parent | 29ef923877c6eea202ad9f8ab39d219a8a953188 (diff) |
gstreamer: Fix race conditions in task pause/resume
The task run function races with two other threads that want to resume
the task: the requestCompleted() handler and the buffer-notify signal
handler. If the former queues completed requests or the latter queues
back buffers to the pool, and then resume the task, after the task run
handler checks the queues but before it attemps to pause the task, then
the task may be paused without noticing that more work is available.
The most immediate way to fix this is to take the stream_lock in the
requestCompleted() and buffer-notify signal handlers, or cover the whole
task run handler with the GstLibcameraSrcState lock. This could cause
long delays in the requestCompleted() handler, so that's not a good
option.
Instead, pause the task unconditionally at the beginning of its run
function, and track while processing buffers and requests if the task
needs to be resumed. It may also get resumed externally by the
buffer-notify signal handler or the request completion handler, which
are guaranteed not to race due to the lock taken by the gst_task_pause()
and gst_task_resume() functions.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Nicolas Dufresne <nicolas.dufresne@collabora.com>
Diffstat (limited to 'Documentation/guides/pipeline-handler.rst')
0 files changed, 0 insertions, 0 deletions