bcm2835-codec: SOURCE_CHANGE queued after last_buffer_dequeued, decoder stops without error
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 32/100
- Issue type
- Bug
- Clarity
- Clearly specified
- Activity status
- Active
- Tech stack
- c
- Domain
- operating-systems
Research direction
Start from handle_fmt_changed() in the bcm2835-codec driver: it sets vq->last_buffer_dequeued before queue_res_chg_event(), so a v4l2_m2m_poll landing between the two writes reports EPOLLIN without EPOLLPRI and the decoder exits via GST_V4L2_FLOW_LAST_BUFFER. The proposed fix reorders the event before the flag and adds wake_up(&vq->done_wq); validate it against v4l2_m2m_poll and gst_v4l2_object_poll. Done means no silent stop across many looped starts on bcm2835 hardware with debug=1. Note the reporter is already testing this and will open a PR, so coordinate before starting.
Written by the indexing model from the issue text.
Description
Pi 4 Model B Rev 1.5 4 GB, kernel 6.18.39-v8 (raspberrypi/linux 8307cc8d), GStreamer 1.26.11, playbin3 → v4l2h264dec → glimagesink under sway. H.264 High 4.1 1920x1080 25 fps + AAC, MP4 over HTTPS.
Our player loops a fixed playlist: the same images and the same two MP4 files, hundreds of times a day. At each start of the H.264 decoder, the firmware reports Format changed to 1920x1088 with an unchanged format (Format was 1920x1088). Most starts play fine. 1 start in 40 to 300 stops and never resumes: no bus error, the pipeline never prerolls, the screen stays on the last frame until a restart. The file and the firmware event stay the same, so we looked for a timing race. We see the freeze on 6 display units and reproduce it on a test unit with a different screen.
Sequence (bcm2835_codec debug=1, GST_DEBUG v4l2*:5):
- capture STREAMON (
bcm2835_codec_start_streaming: type: 9) - 18 to 28 ms later,
handle_fmt_changed: Format changed to 1920x1088and, in the same ms,gst_v4l2_video_dec_loop: Leaving output thread: custom-success - no
Sending EOS event; the decoder loop never dequeues the SOURCE_CHANGE event - on one more start, a flush dequeued the event 1 s later and playback recovered: 2 hits in 86 starts
Cause
handle_fmt_changed()setsvq->last_buffer_dequeued = true, then callsqueue_res_chg_event().v4l2_m2m_pollreports EPOLLIN from the flag underdone_lock, and EPOLLPRI from the event underfh_lock. A poll between the two writes returns EPOLLIN without EPOLLPRI.gst_v4l2_object_pollchecks POLLPRI first. Without it, the pool calls DQBUF, gets EPIPE and returnsGST_V4L2_FLOW_LAST_BUFFER.v4l2videodecthen leaves its output thread and posts no error.- During start-up, each returned input buffer wakes the capture poller, every millisecond or so, which keeps the window busy.
Proposed fix: queue the event first, then set the flag and wake the capture queue pollers.
+ /* A poll that sees the LAST state must also see the event */
+ queue_res_chg_event(ctx);
+
vq = v4l2_m2m_get_vq(ctx->fh.m2m_ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE);
- if (vq->streaming)
+ if (vq->streaming) {
vq->last_buffer_dequeued = true;
-
- queue_res_chg_event(ctx);
+ wake_up(&vq->done_wq);
+ }
We're testing it now and will open a PR once it holds.
Related: #5059 shows the same start-up Format changed with an unchanged format. It's the same trigger and a different bug; neither fix covers the other.
- Dominant language
- C
- Stars
- 13.3k
- Forks
- 5.5k
- Avg merge
- 4d 10h
- Merged PRs (30d)
- 31
Getting set up
This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from raspberrypi/linux
-
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
raspberrypi/linux#7415 · 2 comments · 1 reaction ·
Maintainers usually reply within 1 day
-
rp1-cfe doesn't forward V4L2_EVENT_SOURCE_CHANGE event from csi-2 sensor driver to userspace appOpen
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
raspberrypi/linux#7399 · 1 comment ·
Maintainers usually reply within 1 day
-
Difficulty 1/5 Under an hour Newbie friendliness 82/100
raspberrypi/linux#7357 · 1 comment ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
raspberrypi/linux#7054 · 2 comments ·
Maintainers usually reply within 1 day
-
Difficulty 4/5 3-5 days Newbie friendliness 48/100
raspberrypi/linux#7770 ·
Maintainers usually reply within 1 day
All issues in raspberrypi/linux
Similar issues
-
Difficulty 1/5 Under an hour Newbie friendliness 85/100
Maintainers usually reply within 1 day
-
slice
Difficulty 2/5 Under an hour Newbie friendliness 72/100
Life-Framework/Life-Framework#3465 ·
Maintainers usually reply within 1 day