1759464 - VA-API stops being used

Status: RESOLVED WORKSFORME

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

Description

[Paul Menzel]

Attachment: nightly-platformdecodermodule5.txt

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:100.0) Gecko/20100101 Firefox/100.0

Steps to reproduce:

On Debian sid/unstable use BigBlueButton Web configuring system. VP8 is used as codec and supported by the Dell Latitude E7250 with Intel Broadwell, so VA-API can be used.

Actual results:

Since a while, randomly VA-API is not being used or is stopped being used. Today it happened with Nightly 100.0a1, 20220313212642. ccording to intel_gpu_top the video engine was not used, and the fan starts spinning.

In the MOZ_LOG="PlatformDecoderModule:5" logs, I just saw:

[RDD 91746: MediaPDecoder #1]: D/PlatformDecoderModule Created new VA-API DMABufSurface UID = 8
[AVHWDeviceContext @ 0x7f6002cb7e40] Format 0x3132564e -> unknown.
[RDD 91746: MediaPDecoder #1]: D/PlatformDecoderModule VideoFrameSurfaceVAAPI: creating surface UID = 8
[RDD 91746: MediaPDecoder #1]: D/PlatformDecoderModule   failed to create texture over DMABuf memory!

Expected results:

VA-API should have been used, and the video engine be used.

Please find the beginning of the logs attached. The file with all captured messages has a size of 84 MB, I am unable to attach.

Comment 1

[Paul Menzel]

Attachment: Output of `about:support`

Comment 2

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

The Bugbug bot thinks this bug should belong to the 'Core::Graphics: WebRender' component, and is moving the bug to that component. Please correct in case you think the bot is wrong.

Comment 4

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

see https://fedoraproject.org/wiki/Firefox_Hardware_acceleration#Firefox_98_and_later

Comment 5

[Paul Menzel]

Sorry, I though the log contains that information, but this was with MOZ_DISABLE_RDD_SANDBOX=1 – full command line is MOZ_LOG="PlatformDecoderModule:1" MOZ_DISABLE_RDD_SANDBOX=1 nightly |& tee /dev/shm/nightly-decoder.txt.

Comment 6

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

Please attach log of MOZ_LOG="PlatformDecoderModule:5 Dmabuf:5"
Thanks.

Comment 7

[Paul Menzel]

As I haven’t figured out how to reproduce it yet, it might take a while to provide the logs.

Comment 8

[Paul Menzel]

I think it happened today again with Nightly 100.0a1 (20220314214902). First the video engine utilization was only 0.56 % and eight participants (video streams) were shown in BigBlueButton (WebRTC, VP8). After some participants left, checking at the end the video engine utilization was shown as 0.

MOZ_LOG="PlatformDecoderModule:5 Dmabuf:5" MOZ_DISABLE_RDD_SANDBOX=1 nightly >& /dev/shm/nightly-dma.txt

The log is almost half a gigabyte in size, and I didn’t find anything obvious in there. It’d be great if you could extract the relevant lines and attach them here, so I can delete it.

Comment 9

[Akarshan Biswas]

I have the exact same problem which I have reported downstream on Redhat Bugzilla after getting the update in Fedora. If I understand the log correctly, it fails around here:

[Parent 6652: Compositor]: D/Dmabuf DMABufSurfaceYUV::ImportSurfaceDescriptor() UID 1
[Parent 6652: Compositor]: D/Dmabuf     plane 0 fd 172 size 320 x 176 format 20203852
[Parent 6652: Compositor]: D/Dmabuf     plane 1 fd 173 size 160 x 88 format 38385247
[RDD 6979: MediaPDecoder #1]: D/Dmabuf DMABufSurfaceYUV::CreateEGLImage() UID 2 plane 0
[RDD 6979: MediaPDecoder #1]: D/Dmabuf   EGLImageKHR creation failed
[RDD 6979: MediaPDecoder #1]: D/Dmabuf DMABufSurfaceYUV::ReleaseEGLImages() UID 2
[RDD 6979: MediaPDecoder #1]: D/PlatformDecoderModule   failed to create texture over DMABuf memory!

Comment 10

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

(In reply to Paul Menzel from comment #8)

I think it happened today again with Nightly 100.0a1 (20220314214902). First the video engine utilization was only 0.56 % and eight participants (video streams) were shown in BigBlueButton (WebRTC, VP8). After some participants left, checking at the end the video engine utilization was shown as 0.

MOZ_LOG="PlatformDecoderModule:5 Dmabuf:5" MOZ_DISABLE_RDD_SANDBOX=1 nightly >& /dev/shm/nightly-dma.txt

The log is almost half a gigabyte in size, and I didn’t find anything obvious in there. It’d be great if you could extract the relevant lines and attach them here, so I can delete it.

Thanks, you can clear it - it really does not contain any useful info.
The decoder seems to be restarted to SW decode but without any obvious reason.

Do you have any related crash in coredumpctl?
https://fedoraproject.org/wiki/Debugging_guidelines_for_Mozilla_products#Using_coredumpctl_to_get_backtrace
Thanks.

Comment 11

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

Please reopen if you have nay further info.

Comment 12

[Paul Menzel]

Sorry for not replying sooner. I haven’t experienced this in the last months, so something was fixed. Thank you.