1953677 - Memory leak for VAAPI on CI

Status: RESOLVED FIXED

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

Description

[Alastor Wu [:alwu]]

In bug 1861759, we're going to enable VAAPI testing on the CI. However, there are some memory leakage need to be addressed. See
https://treeherder.mozilla.org/jobs?repo=try&revision=8a28334c20aafa29c7427661de16fcfd050b9856&selectedTaskRun=cb9gF06MTxyn1ag0UO_9Lw.0
https://treeherder.mozilla.org/jobs?repo=try&revision=8a28334c20aafa29c7427661de16fcfd050b9856&selectedTaskRun=WNkWSZC-QneO7Uw0ksEp9Q.0

Martin, do you have a free cycle to take a look on this? Thanks!

Comment 1

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

Yes, will look at it, can reproduce it locally.

Comment 2

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

This comes from DMABufSurface snapshot GL context used to copy dmabuf textures.

Comment 3

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

Attachment: Bug 1953677 [Linux] Delete snapshot GL context when VideoFramePool is deleted r?alwu

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/e61f26594043
[Linux] Delete snapshot GL context when VideoFramePool is deleted r=alwu

Comment 5

[Atila Butkovits]

https://hg.mozilla.org/mozilla-central/rev/e61f26594043

Comment 6

[Alastor Wu [:alwu]]

Hi, Martin,

Even with you patch, I still see the same leaking and a data race for DMABufDevice. Would you mind help me taking a look on this again? If you want to debug it, simply applying patches in bug 1861759 and select mochitest-media-mda-gpu.

Thanks!

Comment 7

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

Sure. I used local testing for it so it may be different from the code paths used on testsuite.

Comment 8

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

I see even worse thread races with the explicit GL context delete from the pool:
https://treeherder.mozilla.org/logviewer?job_id=508140962&repo=try&lineNumber=1786

Comment 9

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

I wonder why we see it on decoding RDD process only and not in main/parent one where dmabuf is used too.

Comment 10

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

The best result I've got so far:
https://treeherder.mozilla.org/jobs?repo=try&revision=20560605ce0387e06e713fcc8f2dbd6e20fd87d3&selectedTaskRun=fIUdonpXSR6stz7nJAKHZg.0

Looks like there are false-positive from gfx drivers so we may suppress thread warning or leak dmabuf device.

Comment 11

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

When Bug 1959645 lands we need to update ignore lists for tsan to suppress warnings from intel drivers.
I'll look at the leaks but looks like we need to suppress them too.

Comment 12

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

Try with Bug 1959645 - https://treeherder.mozilla.org/jobs?repo=try&landoCommitID=134852

Comment 13

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

Attachment: Bug 1953677 [Linux] Allow mem leaks from Intel/MESA drivers r?emilio

Comment 14

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

Attachment: Bug 1953677 [Linux] Allow data race at Intel/MESA drivers r?emilio

Comment 15

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

Attachment: Bug 1953677 [Linux] Release DMABuf snapshot GL content on RDD process shutdown r?emilio

Comment 16

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

try: https://treeherder.mozilla.org/jobs?repo=try&landoCommitID=135163

Comment 17

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

Alwu, with the latest patches I see ' dom/media/test/test_temporary_file_blob_video_plays.html | Test timed out. ' only - not sure what it is.

Comment 18

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/220fadaae86e
https://hg.mozilla.org/integration/autoland/rev/6f78856b1c42
[Linux] Allow mem leaks from Intel/MESA drivers r=decoder
https://github.com/mozilla-firefox/firefox/commit/587071f4a773
https://hg.mozilla.org/integration/autoland/rev/f3577eac27a2
[Linux] Allow data race at Intel/MESA drivers r=decoder
https://github.com/mozilla-firefox/firefox/commit/06b0b40dd111
https://hg.mozilla.org/integration/autoland/rev/01a77d5fb3b5
[Linux] Release DMABuf snapshot GL content on RDD process shutdown r=emilio

Comment 19

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/b6e3d711424e
https://hg.mozilla.org/integration/autoland/rev/f063eff6b11f
Revert "Bug 1953677 [Linux] Release DMABuf snapshot GL content on RDD process shutdown r=emilio" for causing Android build bustages in RDDParent.cpp.

Comment 20

[Serban Stanca [:SerbanS]]

Reverted this because it was causing Android build bustages in RDDParent.cpp.

• Backout link
• Push with failures
• Failure Log
• Failure line: /builds/worker/checkouts/gecko/dom/media/ipc/RDDParent.cpp:62:12: fatal error: 'mozilla/widget/DMABufSurface.h' file not found

Comment 21

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

Updated, try: https://treeherder.mozilla.org/jobs?repo=try&landoCommitID=135645

Comment 22

[Pulsebot]

Pushed by [e-mail]:
https://github.com/mozilla-firefox/firefox/commit/544f08b2c528
https://hg.mozilla.org/integration/autoland/rev/c8f0d42ac3e9
[Linux] Allow mem leaks from Intel/MESA drivers r=decoder
https://github.com/mozilla-firefox/firefox/commit/dd34fa19d336
https://hg.mozilla.org/integration/autoland/rev/756d3a7b91e6
[Linux] Allow data race at Intel/MESA drivers r=decoder
https://github.com/mozilla-firefox/firefox/commit/e470c48949d2
https://hg.mozilla.org/integration/autoland/rev/9f3b13f2b2b2
[Linux] Release DMABuf snapshot GL content on RDD process shutdown r=emilio

Comment 23

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

https://github.com/mozilla-firefox/firefox/commit/544f08b2c528d078cd717a5d3d594828c6519845
https://hg.mozilla.org/mozilla-central/rev/c8f0d42ac3e9
https://github.com/mozilla-firefox/firefox/commit/dd34fa19d336a3b1e0f030e5be835349dd55ae03
https://hg.mozilla.org/mozilla-central/rev/756d3a7b91e6
https://github.com/mozilla-firefox/firefox/commit/e470c48949d27eeb5f42fd09bdf3c946c5190060
https://hg.mozilla.org/mozilla-central/rev/9f3b13f2b2b2