1829542 - VAAPI/Intel: `GlobalRefRelease failed: already zero reference! UID` is printed to Dmabuf log

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Darkspirit]

Attachment: aboutsupport.txt

1. Play the video for some seconds.
2. Pause video.
3. Wait some seconds.
4. GlobalRefRelease failed: already zero reference! UID is printed to Dmabuf log.

MOZ_LOG="PlatformDecoderModule:5 Dmabuf:5" mozregression --launch 2023-04-23 -P stdout -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605

0:38.64 INFO: b'[Parent 13209: Compositor]: D/Dmabuf DMABufSurface::ReleaseDMABuf() UID 6'
0:38.64 INFO: b'[Parent 13209: Compositor]: D/Dmabuf DMABufSurface::GlobalRefRelease UID 6'
0:38.64 INFO: b'[Child 13335: MediaPDecoder #1]: D/PlatformDecoderModule RemoteMediaDataDecoder[7f4ebc7706d0] ::~RemoteMediaDataDecoder: 7f4ebc7706d0 is released'
0:38.64 INFO: b'[Child 13335: RemVidChild]: D/PlatformDecoderModule RemoteMediaDataDecoder[7f4ebc770430] ::~RemoteMediaDataDecoder: 7f4ebc770430 is released'
0:38.64 INFO: b'[RDD 13403: MediaPDecoder #2]: D/Dmabuf VideoFrameSurface: deleting dmabuf surface UID 3'
0:38.64 INFO: b'[RDD 13403: MediaPDecoder #2]: D/Dmabuf VideoFrameSurface: VAAPI releasing dmabuf surface UID 3 FFMPEG ID 0xf aForFrameRecycle 0 mLib 7f19d73ca1b8 mAVHWFrameContext 7f19b2619240 mHWAVBuffer 7f19b2619280'
0:38.64 INFO: b'[RDD 13403: MediaPDecoder #2]: D/Dmabuf DMABufSurfaceYUV::ReleaseSurface() UID 3'
0:38.64 INFO: b'[RDD 13403: MediaPDecoder #2]: D/Dmabuf DMABufSurfaceYUV::ReleaseTextures() UID 3'
0:38.64 INFO: b'[RDD 13403: MediaPDecoder #2]: D/Dmabuf DMABufSurface::ReleaseDMABuf() UID 3'
0:38.64 INFO: b'[RDD 13403: MediaPDecoder #2]: D/Dmabuf DMABufSurface::GlobalRefRelease UID 3'
0:38.64 INFO: b'[RDD 13403: MediaPDecoder #2]: D/Dmabuf GlobalRefRelease failed: already zero reference! UID 3'
0:38.64 INFO: b'[RDD 13403: MediaPDecoder #2]: D/Dmabuf VideoFrameSurface: deleting dmabuf surface UID 4'
0:38.64 INFO: b'[RDD 13403: MediaPDecoder #2]: D/Dmabuf VideoFrameSurface: VAAPI releasing dmabuf surface UID 4 FFMPEG ID 0x10 aForFrameRecycle 0 mLib 7f19d73ca1b8 mAVHWFrameCon

MOZ_LOG="PlatformDecoderModule:5 Dmabuf:5" mozregression --good 2023-01-23 --bad 2023-04-23 -P stdout -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605

8:46.03 INFO: Last good revision: ce20888662f791f3c22fd361d56d761642e259ec
8:46.03 INFO: First bad revision: cf328273030f09826f924b6d6300981ecc8cbbdd
8:46.03 INFO: Pushlog:
https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=ce20888662f791f3c22fd361d56d761642e259ec&tochange=cf328273030f09826f924b6d6300981ecc8cbbdd

cf328273030f09826f924b6d6300981ecc8cbbdd stransky — Bug 1809026 [Linux] Set Debug log ffmpeg level for VA-API decoding r=alwu
77034005a42a5c11cb15a551aeab7e097fd23bfd stransky — Bug 1809026 [Linux] Check and log FFMPEG HW surfaces ID so we're sure we don't use the same surface twice r=alwu
602247c99125ff2e328f9aaa1ef28b0a51bd5822 stransky — Bug 1809026 [Linux] Move global DMABufSurface management from VideoFrameSurface to DMABUFSurfaceImage r=alwu
31d23318230a0f6cc2bc699a0ce5dbee4c3078f7 stransky — Bug 1809026 [Linux] Reference DMABufSurface when it's created/imported r=emilio

That change also caused bug 1820827.

mozregression --repo autoland --launch cf328273030f09826f924b6d6300981ecc8cbbdd -a about:support

Comment 1

[Darkspirit]

Attachment: bug1829542_firstbad.txt

MOZ_LOG="PlatformDecoderModule:5 Dmabuf:5" mozregression --repo autoland --launch cf328273030f09826f924b6d6300981ecc8cbbdd -P stdout -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 > bug1829542_firstbad.txt

Comment 2

[Darkspirit]

Attachment: bug1829542_lastgood.txt

MOZ_LOG="PlatformDecoderModule:5 Dmabuf:5" mozregression --repo autoland --launch ce20888662f791f3c22fd361d56d761642e259ec -P stdout -a https://bug1619882.bmoattachments.org/attachment.cgi?id=9149605 > bug1829542_lastgood.txt

Comment 3

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

That happens when video frames are copied on Intel. Need to check that.

Comment 4

[Ashley Hale [:ahale]]

This bug seems like it belongs more in Media rather than Graphics if I am reading correctly? Or is the video frame copy distinctly on the graphics side of this?

Comment 5

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

Yes, this one involves linux dmabuf implementation.

Comment 6

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1809026

Comment 8

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

Attachment: Bug 1829542 [Linux] Disable copy of DMABufSurface class r?emilio

Comment 9

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

Attachment: Bug 1829542 [Linux] Remove GlobalRefRelease() from GlobalRefCountDelete() r?emilio,alwu

• Remove GlobalRefRelease() from GlobalRefCountDelete(). It matches current scenario where GlobalRefCountCreate() is created an unreferenced counter
  so we also should delete it without unref.
• Call explicitly GlobalRefCountDelete() from ~VideoFrameSurface() to avoid bogus error messages. DMABufSurface may be used by other parts of gecko
  even when video decoder is deleted so we don't care about DMABufSurface ref state here.
• Explicitly call GlobalRefRelease()/GlobalRefCountDelete() from ~DMABufSurface(). That matches DMABufSurface() life time scenario:
  • GlobalRefAdd/GlobalRefRelease in DMABUFSurfaceImage class / RDD process.
  • GlobalRefCountImport()/~DMABufSurface() if DMABufSurface is transfered via. IPC bridge to content/render/compositor process.
• Add DmabufRef MOZ_LOG

Depends on D178282

Comment 10

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

Attachment: Bug 1829542 [Linux] Track used VideoFrameSurface in VideoFramePool r?alwu

• Explicitly track used VideoFrameSurface by mUsed.
• Don't crash if we recycle used dmabuf surface but throw a warning instead. Such scenario produces only visual glitches so no need to hard fail.
• Set mFFMPEGSurfaceID for copied dmabuf surfaces too.

Depends on D178283

Comment 11

[BugBot [:suhaib / :marco/ :calixte]]

Copying crash signatures from duplicate bugs.

Comment 12

[BugBot [:suhaib / :marco/ :calixte]]

The severity field for this bug is set to S3. However, the following bug duplicate has higher severity:

• Bug 1820827: S2

:stransky, could you consider increasing the severity of this bug to S2?

For more information, please visit BugBot documentation.

Comment 13

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/5be595679856
[Linux] Disable copy of DMABufSurface class r=emilio
https://hg.mozilla.org/integration/autoland/rev/a9c9ea4457af
[Linux] Remove GlobalRefRelease() from GlobalRefCountDelete() r=emilio
https://hg.mozilla.org/integration/autoland/rev/865ab9a041f5
[Linux] Track used VideoFrameSurface in VideoFramePool r=alwu

Comment 14

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

Attachment: Bug 1829542 [Linux] Track VideoFrameSurface by mFFMPEGSurfaceID and use Maybe class instead of integer r?alwu

Comment 15

[Narcis Beleuzu [:NarcisB]]

https://hg.mozilla.org/mozilla-central/rev/5be595679856
https://hg.mozilla.org/mozilla-central/rev/a9c9ea4457af
https://hg.mozilla.org/mozilla-central/rev/865ab9a041f5

Comment 16

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/fe5b634984da
[Linux] Track VideoFrameSurface by mFFMPEGSurfaceID and use Maybe class instead of integer r=alwu

Comment 17

[Cosmin Sabou [:CosminS]]

https://hg.mozilla.org/mozilla-central/rev/fe5b634984da