Windows XP guest latch up on KVM with recent kernels

Windows XP guest latch up on KVM with recent kernels

[Brad Campbell]

G'day all,

I'm at a bit of a loss. I run a CCTV storage server on a Windows XP 
guest running on an AMD64 (Piledriver) host with 64 bit kernel and 
userspace. This has been running well for a number of years.

I recently upgraded from a 3.15 kernel to 3.17 and now 3.18. The 3.15 
kernel runs fine. Running 3.17 or later causes the VM to latch up after 
between 30 minutes and 24 hours running. I say latch up rather than lock 
up as if I run virt-viewer and wiggle the mouse around it springs back 
to life, although the system clock is now hours behind. I can see when 
it latches up as I have cacti monitoring it's network interface. The 
system clock stops dead as soon as the traffic does.

When it latches up, qemu spins using 100% of however many cores I have 
allocated. It does not respond to the virtio network adapter, and the 
system clock ceases ticking.

I have several XP VM's on this machine, all configured with the same 
hardware (virtio storage & network) and only the CCTV VM exhibits this 
behavior.

Reverting back to 3.15.x makes it go away. I've not had the opportunity 
to try 3.16 yet but it's on my todo list for this weekend.

I've tried several versions of qemu, and I'm currently on git HEAD. The 
problem tracks the kernel.

This is a production box, and given it can take hours to reproduce I 
can't really run a bisect on it. I've also been unable to replicate it 
thus far on an "identical" test machine, although I'll continue to try 
as it'll make bisecting more viable.

It's almost as if the guest stops responding to interrupts, yet console 
activity wakes it back up.

I'm only posting this in the vain hope it pricks someones spidey senses 
and gives me a bit of a leg up in debugging it.

I'm using guest drivers from spice-guest-tools-0.74 on all the guests.

Regards,
Brad