Lockups/100% CPU on virtiofsd on host kernel 6.11.2

Hi,

I am experience very odd behaviors with Linux 6.11.2 as host kernel. Usually vms end up locked/frozen and need to use virsh destroy to close them. When this happens the virtiofsd process remains at 100% CPU or one core. It cannot be killed because it's in Zombie state. Luckily it is yet possible by doing echo 1 | tee /proc/<pidof virtiofsd>/task/*/clear_refs

Looking at the task pids, there are lot of fds listed.

I am not sure the issue is with virtiofsd as I've also seen a kworker thread behave the same. Perhaps an issue with KVM on the 6.11 host kernel?

When I go back to Linux 6.10.x all is good and stable. The guest vm have been using 6.6, 6.10 and 6.11. Some guests are using virtiofsd as root, some just as a extra mounted dir.

I've tested qemu 9.0 and 9.1.

Anyways, thought to mention this in case others have a similar problem.

I use a wrapper script to start virtiofsd:

#!/bin/bash

exec /opt/virtiofsd/virtiofsd-rs-git-1.11.1 --log-level=info --inode-file-handles=mandatory --announce-submounts --cache auto --xattr --posix-acl "${@}"