1861759 - Enable media-gpu on Linux

Status: RESOLVED FIXED

(Core :: Audio/Video: Playback, task, P3)

Comment 1

[Alastor Wu [:alwu]]

Attachment: Bug 1861759 - part1 : enable media-gpu on Linux.

Comment 2

[Alastor Wu [:alwu]]

Attachment: Bug 1861759 - part2 : only run media-gpu on linux 18.04 worker which has acutal hardware.

Comment 3

[Martin Stránský [:stransky] (ni? me)]

After Bug 1953677 fixed I run another try with logging enabled:
https://treeherder.mozilla.org/jobs?repo=try&revision=96a86a87d32a037d68406f78aa13eceb04fcb721&selectedTaskRun=Kjr5dAYbS6GhBD1n5FzLYA.0

There's one issue remaining:
https://treeherder.mozilla.org/jobs?repo=try&revision=96a86a87d32a037d68406f78aa13eceb04fcb721&selectedTaskRun=Kjr5dAYbS6GhBD1n5FzLYA.0

TEST-UNEXPECTED-FAIL | dom/media/test/test_temporary_file_blob_video_plays.html | Test timed out. -

There's the related log:

[task 2025-06-09T08:22:39.200+00:00] 08:22:39     INFO - GECKO(5008) | [AVHWFramesContext @ 0x7f8932d04300] Direct mapping possible.
[task 2025-06-09T08:22:39.201+00:00] 08:22:39     INFO - GECKO(5008) | [vp8 @ 0x7f895e1ee800] Decode context initialised: 0x1000000/0x2000000.
[task 2025-06-09T08:22:39.201+00:00] 08:22:39     INFO - GECKO(5008) | [vp8 @ 0x7f895e1ee800] Param buffer (type 0, 112 bytes) is 0x8000000.
[task 2025-06-09T08:22:39.202+00:00] 08:22:39     INFO - GECKO(5008) | [vp8 @ 0x7f895e1ee800] Param buffer (type 13, 1072 bytes) is 0x8000001.
[task 2025-06-09T08:22:39.202+00:00] 08:22:39     INFO - GECKO(5008) | [vp8 @ 0x7f895e1ee800] Param buffer (type 1, 64 bytes) is 0x8000002.
[task 2025-06-09T08:22:39.203+00:00] 08:22:39     INFO - GECKO(5008) | [vp8 @ 0x7f895e1ee800] Slice 0 param buffer (72 bytes) is 0x8000003.
[task 2025-06-09T08:22:39.203+00:00] 08:22:39     INFO - GECKO(5008) | [vp8 @ 0x7f895e1ee800] Slice 0 data buffer (241 bytes) is 0x8000004.
[task 2025-06-09T08:22:39.204+00:00] 08:22:39     INFO - GECKO(5008) | [vp8 @ 0x7f895e1ee800] Decode to surface 0x4000000.
[task 2025-06-09T08:22:39.205+00:00] 08:22:39     INFO - GECKO(5008) | [RDD 5108: MediaPDecoder #1]: D/FFmpegVideo FFVPX: Incorrect frame duration, skipping decode stats.
[task 2025-06-09T08:22:39.205+00:00] 08:22:39     INFO - GECKO(5008) | [RDD 5108: MediaPDecoder #1]: D/FFmpegVideo FFVPX: VA-API Got one frame output with pts=0 dts=0 duration=0
[task 2025-06-09T08:22:39.206+00:00] 08:22:39     INFO - GECKO(5008) | [RDD 5108: MediaPDecoder #2]: D/FFmpegVideo FFVPX: ProcessFlush()
[task 2025-06-09T08:22:39.208+00:00] 08:22:39     INFO - GECKO(5008) | [RDD 5108: MediaPDecoder #2]: D/FFmpegVideo FFVPX: FFmpegDataDecoder: flushing buffers
[task 2025-06-09T08:22:39.209+00:00] 08:22:39     INFO - GECKO(5008) | [RDD 5108: MediaPDecoder #1]: D/FFmpegVideo FFVPX: FFmpegDataDecoder: shutdown
[task 2025-06-09T08:24:02.700+00:00] 08:24:02     INFO - GECKO(5008) | [Parent 5008, sqldb:places.sqlite #3] WARNING: Suboptimal indexes for the SQL statement 0x7f484c8f7d40 (http://mzl.la/1FuID0j).: file /builds/worker/checkouts/gecko/storage/mozStoragePrivateHelpers.cpp:114
[task 2025-06-09T08:24:02.700+00:00] 08:24:02     INFO - GECKO(5008) | [Parent 5008, sqldb:places.sqlite #3] WARNING: Suboptimal indexes for the SQL statement 0x7f484c8f7d40 (http://mzl.la/1FuID0j).: file /builds/worker/checkouts/gecko/storage/mozStoragePrivateHelpers.cpp:114
[task 2025-06-09T08:28:07.821+00:00] 08:28:07     INFO - TEST-INFO | started process screentopng
[task 2025-06-09T08:28:07.948+00:00] 08:28:07     INFO - TEST-INFO | screentopng: exit 0
[task 2025-06-09T08:28:07.949+00:00] 08:28:07     INFO - Buffered messages logged at 08:22:39
[task 2025-06-09T08:28:07.949+00:00] 08:28:07     INFO - Drawing color 255,0,0,1
[task 2025-06-09T08:28:07.950+00:00] 08:28:07     INFO - TEST-PASS | dom/media/test/test_temporary_file_blob_video_plays.html | Media recorder stream = canvas stream at the start of recording 
[task 2025-06-09T08:28:07.951+00:00] 08:28:07     INFO - TEST-PASS | dom/media/test/test_temporary_file_blob_video_plays.html | Media recorder should be recording 
[task 2025-06-09T08:28:07.951+00:00] 08:28:07     INFO - Got 'start' event
[task 2025-06-09T08:28:07.952+00:00] 08:28:07     INFO - TEST-PASS | dom/media/test/test_temporary_file_blob_video_plays.html | Should only get one dataavailable event 
[task 2025-06-09T08:28:07.952+00:00] 08:28:07     INFO - Got 'stop' event
[task 2025-06-09T08:28:07.953+00:00] 08:28:07     INFO - TEST-PASS | dom/media/test/test_temporary_file_blob_video_plays.html | Should have gotten a data blob 
[task 2025-06-09T08:28:07.954+00:00] 08:28:07     INFO - Waiting for video recorded-video to match [255,0,0,255] - red (Should become red)
[task 2025-06-09T08:28:07.954+00:00] 08:28:07     INFO - Frame: 0 IsPixel ref=255,0,0,255 threshold=128 value=0,0,0,0
[task 2025-06-09T08:28:07.955+00:00] 08:28:07     INFO - Buffered messages finished
[task 2025-06-09T08:28:07.956+00:00] 08:28:07     INFO - TEST-UNEXPECTED-FAIL | dom/media/test/test_temporary_file_blob_video_plays.html | Test timed out. - 

which looks like correct decode to me.

Comment 4

[Alastor Wu [:alwu]]

(In reply to Martin Stránský [:stransky] (ni? me) from comment #3)

After Bug 1953677 fixed I run another try with logging enabled:
https://treeherder.mozilla.org/jobs?repo=try&revision=96a86a87d32a037d68406f78aa13eceb04fcb721&selectedTaskRun=Kjr5dAYbS6GhBD1n5FzLYA.0

There's one issue remaining:
https://treeherder.mozilla.org/jobs?repo=try&revision=96a86a87d32a037d68406f78aa13eceb04fcb721&selectedTaskRun=Kjr5dAYbS6GhBD1n5FzLYA.0

TEST-UNEXPECTED-FAIL | dom/media/test/test_temporary_file_blob_video_plays.html | Test timed out. -

That matches what I previously observed in my patch D241337, which temporarily disables the test. However, after reviewing the test today, it seems this section may be causing the timeout.

I suspect the timeupdate event might not be received because playback has already ended. I’ll run a test to confirm whether this is the case and see if we can fix the timeout. If not, I’ll proceed with disabling the test on Linux media-gpu.

Comment 5

[Alastor Wu [:alwu]]

Okay, the test is still timeout so we need to temporarily disable it. However, I noticed other issues

• Assertions in DMABUFSurfaceImage : 1 2
• Byte buffer leakage : might be related with VAAPI log? See here
• Data race in av_log_default_callback : I found this from the push in comment 3

---

Martin, could you help take a look on them? Thanks!

Comment 6

[Martin Stránský [:stransky] (ni? me)]

Sure, will look at it. The leaks come from logging, will look at the asserts.

Comment 7

[Martin Stránský [:stransky] (ni? me)]

Looks related to VP8 decoding, I'll look at it. The assertion looks related to previous bunch of errors - the bad fd reported above.
We may decode/export VP8 decoded dmabuf surface wrongly and we see the timeout and perhaps empty surface. Will try on hardware with VP8 HW decoding.

Comment 8

[Martin Stránský [:stransky] (ni? me)]

I see the bug locally on HW VP8 decoding (Intel). May be related to dmabuf handling/vp8 decoding.

Comment 9

[Martin Stránský [:stransky] (ni? me)]

Looks like we need to keep VideoFrameSurface present (or at least textures/surface/fd) after FFmpegVideoDecoder::ProcessShutdown(). Right now we delete whole surface pool there.

Comment 10

[Martin Stránský [:stransky] (ni? me)]

With Bug 1976755 we should be fine here (except the memleak caused by logging).

Comment 11

[Alastor Wu [:alwu]]

Thanks! I will check it again, keep my NI.

Comment 12

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/dafa080ee84a
https://hg.mozilla.org/integration/autoland/rev/afd3a964ff38
part1 : enable media-gpu on Linux. r=jmaher
https://github.com/mozilla-firefox/firefox/commit/d2cd95c977df
https://hg.mozilla.org/integration/autoland/rev/596633b82226
part2 : only run media-gpu on linux 18.04 worker which has acutal hardware. r=taskgraph-reviewers,jmaher

Comment 13

[Cosmin Sabou [:CosminS]]

https://hg.mozilla.org/mozilla-central/rev/afd3a964ff38
https://hg.mozilla.org/mozilla-central/rev/596633b82226