Ok, I think I've got this display card under control - or so it seems. It hasn't crashed since Wednesday.
I guess this is a combo AMD RDNA3 Navi23 and KDE Plasma (+ Wayland?) bug...
I backed out all my Linux kernel command line switches for AMD GPUs and it crashed to green screen in less than a half hour. So, re-adding them stopped that from happening.
... amdgpu.dcdebugmask=0x410 amdgpu.tmz=0 amdgpu.gpu_recovery=1 amdgpu.sg_display=0 ...
And then the cursor thing is fixed <?!> by adding
KWIN_FORCE_SW_CURSOR=1
Currently this is what is set:
cat /proc/$(pidof kwin_wayland)/environ | tr '\0' '\n' | grep KWIN :
KWIN_DRM_NO_AMS=1 KWIN_DRM_NO_DIRECT_SCANOUT=1 KWIN_FORCE_SW_CURSOR=1
From a user experience point of view I don't see any difference with that software cursor setting.
There's another wrinkle to this that might be a factor but I can't really test around it currently. This is a multi-head machine and one screen is connected via HDMI and one is DP.
Perhaps this wouldn't be a problem if both screens were connected via the same type of connector.
Both screens are set to the same refresh frequency. But, in the display control app one reads back at 60.00 Hz while the other is 60.000 Hz. Minor thing right? But in trying to make this work I've seen references to AMD GPU weirdness in multi-head configurations where the refresh rates were different between screens. Red herring perhaps.
Anyway, don't know if there are (or will be) any people trying to make a RX 5800 XT GPU behave under Linux but this seems to be what worked for me and my dual head setup. I'll update if anything changes.
#Linux #AMD #AMDGPU #KDE #KDEPlasma #Plasma #AlpineLinux #kwin #Wayland