Trying to pass my R9 380 (UEFI supported) to a windows VM. I get it to work partially.
My Gpu is handed off to the guest VM as the Windows VM detects that the display adapter is an AMD card.*** It is initially using the microsoft basic display adapter driver so I try to install the AMD drivers. However whenever I try to install the amd driver I get maybe halfway with installing the visual drivers and my VM freezes up and then reboots. I checked my /var/log/libvirt/qemu/wintest.log and received a whole bunch of these types of errors:
2016-10-08T22:26:15.940074Z qemu-system-x86_64: vfio_region_write(0000:03:00.0:region0+0xf9fe7f0, 0xdef4def4,4) failed: Device or resource busy
I am not 100% sure what this means but I am assuming that my GPU is not completely released from the host and something from the host is still bound to it/using it that shouldn't be. The card is in it's own IOMMMU group and on boot it is bound to the "pci-stub" driver and then once the VM starts it becomes bound to the "vfio-pci" drivers. I am using the latest OVMF for my firmware.
I have triple checked and all my hardware is (theoretically) compatible with what I am trying to do. I am using a dual socket lga 1366 supermicro board that has VT-d. My GPU has a UEFI compatible ROM flashed on it.
AT this point I just don't know how to trouble shoot as I am not understanding what I should be looking for from the above error message. I could use some help.
*** So initially when I boot the Host system the GPU picks up the ubuntu host and displays the CLI. Once I start my windows guest VM the GPU acts like a new signal is available and no longer shows the CLI and instead moves to a black screen. However once windows starts to boot the monitor acts like it has been disconnected and goes into "standby" mode (aka sleep mode). The GPU won't pick up ubuntu's CLI until I reboot the Host system again (regardless of whether the VM guest is running or not.)