The workaround to disable hardware acceleration from the description of the issue kind of works.
I added
Section "Device"
Identifier "card0"
Driver "amdgpu"
Option "AccelMethod" "none"
EndSection
to /etc/X11/xorg.conf.d/50-video.conf.
I also added preempt=none to GRUB_CMDLINE_LINUX in /etc/default/grub as described in Where to add kernel parameter? and verified it’s in use by checking proc/cmdline contains the string.
The discussion is led by isodude (I envy your patience and persistence) trying out many different kernel versions, xen versions & kernel patches. Other people are weighing here and there. I for one failed to extract anything that would improve my situation.
The gist seems to be that amd does not include xen in the qa process resulting in a cubersome experience for qubes users: They experience lots of small errors the first time because few/no other people run xen on their desktops.
This leaves me sad because it seems I’m left with two options:
a) buy an intel machine again and use qubes
b) keep the machine and use linux
Considering time and money required to stay with qubes I’ll probably have to postpone my migration to qubes another few years.