Problems with nested virtualization on Proxmox 8

[jagson]

Hello everyone,

I have some small problems with my Proxmox machine since upgrading to v8. The whole upgrade went smoothly without any problems and everything Linux related seems to be running great, but not my Windows 11 VMs. Whenever I start one, I get following dmesg errors for all assigned CPU cores in dmesg:

SVM: kvm [807004]: vcpu0, guest rIP: 0xfffff83488f6cae9 unimplemented wrmsr: 0xc0010115 data 0x0

VMs start and PCIe passthrough is still working but performance is extremely poor. My host has nested virtualization enabled on my Ryzen 2700X and both VMs have all Windows virtualization features required for WSL2 enabled.

The first VM has 8 host-cores with passed through GTX970 and NVMe SSD and the second one has 4 host-cores with its own 1050ti and a SATA-SSD.

What I tried so far was disabling hugepages, going back to q35-7.2 and trying out kernel 6.1 but without success.

 

[davemcl]

Sometimes I find it better to start VM's up using 'qm' when troubleshooting, the same errors dont always appear in syslog.

 

[jagson]

"qm start 3201" is only giving me an "swtpm_setup: Not overwriting existing state file."

 

[leesteken]

I don't know if you have an Intel or AMD but you can use cat /sys/module/kvm_*/parameters/nested (on the Proxmox host) to check if nested virtualization is enabled.

 

[jagson]

It was working perfectly on the 5.x kernel.

cat /etc/modprobe.d/kvm-amd.conf returns options kvm-amd nested=1

cat /sys/module/kvm_amd/parameters/nested returns 1

 

[jagson]

Kernel 5.15.108-1-pve and 5.19.17-2-pve produce the same errors.

 

[MrChoke]

There are dozens of us. I'm having this problem too, but even worse. I can't even get my VMs to boot using cpu host mode and I get the unimplemented wrmsr error in my logs.