I believe it happens for any guest OS (search for "amd reset bug"). Some people are using a workaround that involves removing the device (in the sysfs), suspending the host, and then rescanning, but for me, that's not much better than a host reboot, so when the AMD GPU occasionally does get stuck in D3 (preventing starting the VM after a VM reboot), I just reboot the host. It doesn't happen very often, and when it does, it's only when I'm in the process of rebooting the VM anyway.
Previously, I also attempted using a kernel patch that's been floating around (DECLARE_PCI_FIXUP_HEADER ... quirk_no_bus_reset), but it just gives me corrupt graphics when rebooting the VM.
Previously, I also attempted using a kernel patch that's been floating around (DECLARE_PCI_FIXUP_HEADER ... quirk_no_bus_reset), but it just gives me corrupt graphics when rebooting the VM.