I’ve run into a small(ish) issue while testing LXD VMs under ArchLinux.
When I start a VM with multiple CPU cores (e.g. limits.cpu=8) the guest kernel just crashes with a weird “SMP NOPTI” message which is just impossible to google for.
I’ve manually validated the qemu config and found that by simply disabling the virtio-vga device the VM was working fine.
Does anybody have an idea why this could be happening? For my local testing I now just commented out the lines in LXDs config generator concerning virtio-vga, but I’m hoping to find the actual cause…
Some infos:
LXD 4.19 (from git master)
QEMU 6.1.0
Linux 5.14.11
Guest Image: Tested various images, e.g. ubuntu:20.04
I tested on a temporary setup on Fedora 35 with LXD 4.19 as snap and kernel 5.14.11 (CPU: Ryzen 7 3700X), no issue here
Maybe you could try with a different QEMU configuration by using Libvirt instead of LXD (with similar guest configuration) and see if the problem still occurs to identify if the problem come from LXD ?
I’ve done some further testing and this seems to be an issue with a combination of that device, probably somehow my CPU, and the kernel in use by the Ubuntu images.
Trying to simply load e.g. the “vmlinuz-5.11.0-1017-kvm” delivered by the Ubuntu 21.04 image results in said kernel panic, while downloading one of their generic images and loading the included “vmlinuz-5.11.0-37-generic” works without any issues.