1693065 - D/PlatformDecoderModule VA-API FFmpeg is disabled by platform when MOZ_WAYLAND_DRM_DEVICE is undefined

Status: RESOLVED MOVED

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

Description

[Aidan Harris]

+++ This bug was initially created as a clone of Bug #1683853 +++

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:87.0) Gecko/20100101 Firefox/87.0

Steps to reproduce:

Start Firefox Nightly 87.0a1 with

MOZ_LOG="PlatformDecoderModule:5" MOZ_DISABLE_RDD_SANDBOX=1 MOZ_ENABLE_WAYLAND=1 firefox

and start video playback .

Actual results:

The logs show:

[Child 3869625: Main Thread]: D/PlatformDecoderModule PDMInitializer, Init PDMs in Content process
[Child 3869625: Main Thread]: D/PlatformDecoderModule VA-API FFmpeg is disabled by platform
[Child 3869625: Main Thread]: D/PlatformDecoderModule VA-API FFmpeg is disabled by platform

Expected results:

The message

D/PlatformDecoderModule VA-API FFmpeg is disabled by platform

should not appear as the configuration options are set correctly. The auto-detection of the DRM device does not appear to be working correctly. Specifying MOZ_WAYLAND_DRM_DEVICE=/dev/dri/renderD128 forces firefox to use the render node specified and correctly enables VA-API and it works without issue. This is on a system with only a single GPU so it should not be necessary to manually specify the DRM device and it used to work fine without specifying this so I think this is a recent regression. Below is the contents of /dev/dri:

$ ls -al /dev/dri
total 0
drwxr-xr-x   3 root root        100 Feb 14 18:10 .
drwxr-xr-x  22 root root       4660 Feb 16 15:04 ..
drwxr-xr-x   2 root root         80 Feb 14 18:10 by-path
crw-rw----+  1 root video  226,   0 Feb 16 07:55 card0
crw-rw-rw-   1 root render 226, 128 Feb 14 18:10 renderD128

and /dev/dri/by-path:

$ ls -al /dev/dri/by-path
total 0
drwxr-xr-x 2 root root  80 Feb 14 18:10 .
drwxr-xr-x 3 root root 100 Feb 14 18:10 ..
lrwxrwxrwx 1 root root   8 Feb 16 07:55 pci-0000:0d:00.0-card -> ../card0
lrwxrwxrwx 1 root root  13 Feb 14 18:10 pci-0000:0d:00.0-render -> ../renderD128
``

Comment 1

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

Robert, any idea here?
Thanks.

Comment 2

[Robert Mader [:rmader]]

Aidan, mind sharing your about:support?

Comment 3

[Aidan Harris]

Attachment: about:support

```

```

Comment 4

[Aidan Harris]

I think this might be relevant but I'm not sure why the test is failing:

Failure Log
(#0) Error: glxtest: Cannot find DRM device
(#1) Error: glxtest: Can't find render node name for DRM device

Comment 5

[Robert Mader [:rmader]]

Thanks, will look into it.

Comment 6

[Robert Mader [:rmader]]

Aidan: no luck reproducing that here - could you run the following try build and paste the output here? It contains some extra debug prints.

https://treeherder.mozilla.org/jobs?repo=try&revision=8dc8dc3dae160a3aeb98c9bd6465932b5d1e00fc

MOZ_ENABLE_WAYLAND=1 mozregression --repo try --launch 8dc8dc3dae160a3aeb98c9bd6465932b5d1e00fc

Comment 7

[Aidan Harris]

(In reply to Robert Mader [:rmader] from comment #6)

Aidan: no luck reproducing that here - could you run the following try build and paste the output here? It contains some extra debug prints.

https://treeherder.mozilla.org/jobs?repo=try&revision=8dc8dc3dae160a3aeb98c9bd6465932b5d1e00fc

MOZ_ENABLE_WAYLAND=1 mozregression --repo try --launch 8dc8dc3dae160a3aeb98c9bd6465932b5d1e00fc

I traced this to an issue in libdrm, sorry for wasting your time:

https://gitlab.freedesktop.org/mesa/drm/-/issues/56

https://gitlab.freedesktop.org/mesa/drm/-/commit/8cb12a2528d795c45bba5f03b3486b4040fb0f45

Comment 8

[Robert Mader [:rmader]]

Hehe funny, so Simon already run into that aswell :) Our code is mostly copied from his (wlroots).

Thanks for tracking it down and reporting back, improving confidence in the solution!