I’ve been experiencing a recurring issue where plasmashell crashes (the panels and desktop disappear) almost every day. It usually happens the next day after leaving the PC on or waking it from sleep.
Clearing ~/.cache/qmlcache and ~/.cache/plasma* temporarily fixes the issue, but it always comes back the next day. Additionally, image rendering in my terminal (e.g., fastfetch via Kitty) completely breaks during this time, showing raw “PNG” text instead of the actual image.
My Environment:
OS: Arch Linux
Kernel: 6.19.10-tkg-bore-dirty
DE/WM: KDE Plasma (KWin / Wayland)
GPU: AMD Radeon RX 6600
Mesa: 26.0.4-arch1.1
What I’ve checked so far:
pacman -Qk shows all packages are intact (0 missing files).
btrfs device stats / shows 0 errors (hardware/filesystem is healthy).
However, dmesg shows this warning: amdgpu 0000:08:00.0: amdgpu: SMU driver if version not matched
Here is a snippet from coredumpctl info plasmashell, which shows threads involving libgallium and Qt:
It seems like there might be a mismatch between the custom TKG kernel and the bleeding-edge Mesa 26.0.4, causing buffer or memory corruption over time.
Has anyone experienced similar issues with recent Mesa updates on AMD GPUs, or does anyone have suggestions on how to debug this further? Any insights would be greatly appreciated!
I’d suggest first trying the LTS or Stable Kernel, see if that fixes it, maybe the 6.19.10 build has incompatibility in the TKG patches.
Since you’re on Arch, you can be ahead of the stable firmware releases, check if you have the latest Linux Firmware, and check your Firmware/Kernel Sync; SMU Mismatch can happen when the kernel is newer than the firmware expects.
Mesa 26 is still very new, trying downgrading to an earlier version like 25.x see if that helps in any way.
Does journalctl -p 3 -xb show any “GPU Fault” Errors?
❯ journalctl -p 3 -xb
April 20 16:27:37 kernel: Error: Driver ‘efi-framebuffer’ is already registered, aborting…
April 20 16:27:37 kernel: amdgpu: Overdrive is enabled, please disable it before reporting any bugs unrelated to overdrive.
April 20 16:28:40 sddm-helper[793]: gkr-pam: unable to locate daemon control file
Now I’m using zen kernel.
If it still crashes on Zen, I’ll definitely try downgrading Mesa to 25.x as you suggested.
Thank you
Looking at your command output, you could also try another step before downgrading, the kernel of your amdgpu clearly states that Overdrive is enabled;
There’s a few ways you can disable it, since I don’t know the exact specifications of your system:
* A GUI tool like corectrl (if you have it installed).
* Kernel command-line parameters. For example, you might try adding amdgpu.overdrive=0 to your GRUB or systemd-boot configuration.
* Consulting your motherboard's documentation or AMD's support resources for your specific GPU on how to manage Overdrive settings.
I actually tried the Zen kernel two days ago, but ran into a critical issue: lact (the control tool I use) didn’t play nice with it.
On the Zen kernel, I couldn’t get the fan curves or power adjustments to apply correctly. As a result, the default fan speeds were way too conservative, and the system hit a thermal shutdown during high-load gaming (the Junction temp is a beast on this specific setup).
That’s why I’m sticking with a custom linux-tkg build. It’s the only one that consistently allows lact to
Aggressively ramp up the fan curve to prevent those thermal shutdowns.
Downclock/Undervolt the GPU to maintain stability in demanding titles like MonGil: STAR DIVE, ARMORED CORE VI.
The Mesa downgrade was specifically to fix the plasmashell crashes, which seemed independent of the kernel choice but were making the desktop unusable.