better to always send video model=none when there's no managed video device. Otherwise libvirt adds cirrus which we no longer support
Verified: ovirt-engine-4.4.0-0.0.master.20190910103859.git07162a4.el7 vdsm-4.40.0-907.gitdb55522.el7.x86_64 libvirt-daemon-4.5.0-23.el7_7.1.x86_64 qemu-kvm-ev-2.12.0-33.1.el7.x86_64 sanlock-3.7.3-1.el7.x86_64 2 X Nvidia Tesla M60 under the same host with Nvidia Grid 9.1 GA drivers. Verification scenario: 1. Prepare the next VMs: RHEL 7 VM with mdev-type nvidia-18 Windows 10 VM with 3 mdev_type of nvidia-22 2. Run VMs. For each of the next steps: Verify VMs are running with Video model type="none" (observe vdsm.log VM XML or using virsh dumpxml command) Verify VNC console is showing VMs display. Verify Nvidia instances are assigned to VMs properly. 3. Power off -> run VMs 10 times. 4. Shutdown -> run VMs 10 times. 5. Reboot VMs 10 times. 6. Remove mdev_type instances from the VM and run VMs. Verify QXL device is now listed in VM XML video section. Verify VNC console is showing VM display. 7. Power off VMs, add mdev_type to VMs, run VMs and verify VMs Video model type is "none".
This bugzilla is included in oVirt 4.4.1 release, published on July 8th 2020. Since the problem described in this bug report should be resolved in oVirt 4.4.1 release, it has been closed with a resolution of CURRENT RELEASE. If the solution does not work for you, please open a new bug report.