[vfio-users] System lockup

[Zycorax Tokoroa]

Hello all,

I'm not sure if this is the proper place to write about the problem I'm 
facing as I'm not certain it's related to vfio, but it's something I've 
been experiencing since I started using virtualization more intensely.
I'm using Windows 8.1 with a passtrough'ed GTX 970. Before starting to 
use GPU virtualisation I have had no issues with the VM, although my 
usage was way less intense. Now, instead, at times I experience lockups, 
where both the guest and the host (Kubuntu 16.04) freeze and become 
unresponsive, disk activity goes to zero and any and all peripherals 
(USB, wifi, ethernet) die. There is no log from the host side, dmesg, 
syslog and others don't seem to get any write.
I can only report that this happens while there's a somewhat intense 
disk activity, e.g. defragmentation, steam games verification, transfer 
of several GBs of data. I've tried anything I could think of 
disk-related (changing virtual controllers, checking the disk with 
badblocks, changing the physical disk that backs the storage of the VM), 
I've started clean (both in guest and host) several times, tried 
different OSes, I've disabled any and all passtrough, thinking it could 
be a power-related issue, with the PSU unable to support both my GPUs, 
but I seem not to be able to shake this issue off any more; making a new 
VM with windows and getting a similar disk I/O load on it triggers this 
behavior almost all the times.
At this point I'm wondering if I'm somehow hitting a bug in 
qemu/vfio/virtio, or if my issue comes instead from hardware


Hardware:
ASUS X99-Deluxe
i7-5930K
ASUS GTX 960 for the host
ASUS GTX 970 for the guest