Green artifacts on YouTube videos in Firefox

[br0kenpixel]

I’m using Fedora 40 on a Legion Pro 7 with an AMD CPU. I get green squares on YouTube videos. I also get a few dropped frames.

I can fix the issue by changing my MUX switch to only allow my NVIDIA dGPU. Furthermore I was able to significantly lower the amount of dropped frames by running Firefox with --hwdec=vaapi-copy. I also set up this flag in about:config. Not sure if that’s how I was supposed to do it but it works.

However, the artifacts still remain while using the iGPU.

I also managed to freeze the system while writing this post and trying to play the video. Previously I couldn’t even play videos, because the system would freeze after a few seconds. I managed to fix that with these kernel parameters.

Also, this happens with ALL videos, not just high-quality ones like the one above.

[dokdave (David Tansley)]

I get exactly the same thing with an AMD Ryzen 7 with integrated GPU.

1. Green artifacts on YouTube videos which seem to happen more frequently when I move the mouse over the video
2. Hard crash to black screen when playing YouTube videos (intermitent, managed to get two crashes today requiring power button shutdown)

[yuntaz (Y.Z.)]

It is an upstream regression:

AMD APU regression (full halt) on kernel 6.10 - how to best report?

A new regression for AMD APU’s is present in kernel 6.10 that wil cause intermittent full system crashes in combination with Mesa 24.1.5. The only option is to power down and restart the machine. Kernel 6.9 is unaffected. There are patches for both the kernel and Mesa available upstream, how would one go about to best report this on our local bugtrackers? Source discussion:

Either kernel or Mesa needs to be downgraded.

[dokdave (David Tansley)]

Confirmed. Booted with 6.9.12 kernel and the problem goes away.

[yuntaz (Y.Z.)]

The new 6.10.4 kernel is available. I updated it and played a video for 30 minutes, but the system froze eventually. The issue has not been resolved.

[br0kenpixel]

I don’t know if the crashes are fixed. I don’t have much time to test it right now, but I still get green artifacts.

[litemotiv (Olivier)]

Mesa 24.1.6 is probably the earliest option for a fix. The mainline 6.11 kernel apparently still contains the bug at the moment, so it’s likely that the 6.10 line is also not patched yet.

[br0kenpixel]

That’s not a bad option. Sorry if this is a dumb question, but how can I downgrade/upgrade the Mesa driver to that version? Is there a dnf command for that?

[litemotiv (Olivier)]

Mesa is a collection of packages, it depends on your system which of them are installed. This is on my system:

$ rpm -qa | grep mesa

mesa-libGLU-9.0.3-4.fc40.x86_64
mesa-vulkan-drivers-24.1.5-2.fc40.x86_64
mesa-filesystem-24.1.5-2.fc40.x86_64
mesa-libglapi-24.1.5-2.fc40.x86_64
mesa-dri-drivers-24.1.5-2.fc40.x86_64
mesa-libgbm-24.1.5-2.fc40.x86_64
mesa-libEGL-24.1.5-2.fc40.x86_64
mesa-libGL-24.1.5-2.fc40.x86_64
mesa-va-drivers-freeworld-24.1.5-1.fc40.x86_64

For the current situation, it is probably the easiest approach to keep using the 6.9 kernel until mesa 24.1.6 becomes available. With that mesa update it should be okay to start using the 6.10 kernel again. The proper fix for the kernel will then probably come some time in 6.11, but the fixes in mesa should make sure that the bug in the amdgpu kernel driver will not be triggered anymore until that time.

[br0kenpixel]

Luckily I still had an option in the GRUB menu for the 6.9 kernel. I can confirm that the video plays flawlessly.

[br0kenpixel]

I tried using the laptop today with the AMD graphics (6.10.4 kernel). It crashed twice already, and the “Problem Reporting” app didn’t catch any crashes, but I found this docs page and checked the kernel logs with journalctl.

I coldn’t find any kernel panic, but I do see a lot of weird errors and warnings, I don’t know which ones are relevant, but I’ll try to put most of them here. Maybe it could help.

amd_pstate: failed to enable amd pstate during resume, return -6
powercap intel-rapl:1: PM: parent intel-rapl should not be sleeping
AVC avc:  denied  { execheap } for  pid=93168 comm="code" scontext=unconfined_u:unconfined_r:unconfined_t:s0-s0:c0.c1023 tcontext=unconfined_u:unconfined_r:unconfined_t:s0-s0:c0.c1023 tclass=process permissive=0
kernel: perf: interrupt took too long (2520 > 2500), lowering kernel.perf_event_max_sample_rate to 79000
amdgpu 0000:05:00.0: [drm] *ERROR* dc_dmub_srv_log_diagnostic_data: DMCUB error - collecting diagnostic data

The screen just randomly went black, and the system rebooted.