[libvirt-users] Debugging Windows 10 Guest Flakiness

[David Morrison]

Hello,

I'm not sure if I'm asking this question in the right place, so please 
redirect me if there's a better list.

I recently set up a Windows 10 guest on an Arch Linux host with libvirt 
and KVM/Qemu, using IOMMU to enable GPU passthrough.  I followed this 
guide to set up the VM:

https://wiki.archlinux.org/index.php/PCI_passthrough_via_OVMF

The problem that I'm having is that the guest is very flaky: sometimes 
the guest will just hang indefinitely, and sometimes the guest loses 
network connectivity.  Usually when either of these behaviors happens, 
restarting the libvirtd daemon on the host resolves the issue, but 
that's not always true, and sometimes when I restart libvirtd it causes 
my entire (host) machine to lock up!

I have enabled log_output level 1 in /etc/libvirt/libvirtd.conf and have 
monitored the log file during one of these hangs, but I don't see any 
obvious errors or problems in the log file itself (and there's a lot of 
information to sort through, so I may have missed something).

Some specifics about my setup:

Motherboard: ASUS Sabertooth X79
CPU: Intel i7 4930K
Host GPU: GeForce GT 730
Guest GPU: GeForce GTX 780
Virtual network interface: NAT with the virtio driver

I'm *reasonably* confident that the problem is not with the GPU 
passthrough setup, as I'm able to run FurMark (a GPU burn-in tool) on 
the guest with no problems.  However, I haven't yet been able to 
reliably reproduce the error at all, so I'm pretty much at a loss as to 
where to go from here.

Can anyone on this list point me in a good direction to look, or have 
you seen anything like this before?  I'm happy to provide debug logs or 
any other information that is helpful, as well.

Thanks for your time,
David