There have been two separate issues with AMD APU’s. One could be more or less resolved (or at least avoided) by using a patched Mesa version, but the other one can still pop up and crash your session.
For that second issue a kernel driver update is needed that will likely land in kernel 6.11. This particular bug is a regression and does not occur in kernel 6.9, if you have an older kernel still installed you can boot into that to avoid it. Kernel 6.10 is affected, so for a more definitive fix you would have to wait until kernel 6.11.
Assuming your application uses the GNOME runtime, or freedesktop-sdk runtime, the components used to build the runtime are recorded in /usr/manifest.json. I bet the KDE runtime does this too? (Fedora runtimes probably don’t?)
$ flatpak run --command=bash com.example.AppID
[📦 com.example.AppID ~]$ less /usr/manifest.json
You can use that to see the versions of pretty much everything other than mesa. But it will show the version of mesa used to build the runtime itself, which is not what you want. What you actually care about is the version used by the mesa extension you have installed:
$ flatpak list | grep Mesa
Mesa org.freedesktop.Platform.GL.default 24.2.1 23.08 flathub system
Mesa (Extra) org.freedesktop.Platform.GL.default 24.2.1 23.08-extra flathub system
Mesa org.freedesktop.Platform.GL.default 24.2.1 24.08 gnome-nightly system
Mesa (Extra) org.freedesktop.Platform.GL.default 24.2.1 24.08extra gnome-nightly system
As you see, I have base and extra mesa extensions installed for both 23.08 and 24.08, and both use mesa version 24.2.1. In case you find different mesa versions here, you’ll need some way to tell which is used by your application, but I don’t know how. What I can tell you is that GNOME 45 and 46 runtimes are based on freedesktop-sdk 23.08, and 47 is based on freedesktop-sdk 24.08.
Do you have a reference for this? It seems highly unusual that amdgpu developers would just accept that 6.10 is broken forever instead of backporting the fix or reverting the commit that regressed.
So the bug can be triggered by an interaction between Mesa and the amdgpu driver. Initially when both were unpatched it could cause full system hangs, which is severe. When Mesa was patched the severity of crashes seems to have reduced to mostly visual artifacts and occassional session crashes, which is an an improvement.
I could personally reproduce the behaviour reliably on a system with Mesa 24.1.7 and kernel 6.10.3, running only Firefox (non-flatpak) and no other programs. Other reports on freedesktop.org have indicated similar behaviour with 6.10.X kernels.
An AMD developer asked for a bisect on the kernel mailinglist:
Which was then made and submitted to the ticket below. It does not indicate unfortunately which kernel release is going to incorporate a fix. If it already landed in one of the 6.10 kernels after .3 it would be good to find a commit message for it (i haven’t been able to find one).
Edit: i have tested with kernel 6.10.9 / mesa 24.1.7 and the issue is still present there.