[SOLVED] issues with CPU types when migrating to proxmox 8.1

[damo2929]

on proxmox 7.4 I had my VMs set to use Eypc-Rome to match the CPU in the systems which are EPYC 7452. but when use it on 8.1 am getting the following error

kvm: warning: host doesn't support requested feature: CPUID.0DH:EAX.xsaves [bit 3]
kvm: Host doesn't support requested features
TASK ERROR: start failed: QEMU exited with code 1
any ideas on whats causing this to fail after upgrading.

 

[VictorSTS]

This is explained here [1].

If running a cluster and all your hosts have the same CPU, use type host for best performance and CPU instructions availability. If you have different CPUs in your cluster, use the new x86-64-v2-AES CPU, which provides most CPU instrucctions that are useful for most applications and allows live migration among different CPU generations (but not among different CPU vendors).

[1] https://forum.proxmox.com/threads/cant-use-epyc-rome-cpu-after-update.125336/post-546918

 

[damo2929]

Found the issue, it's because proxmox hasn't enabled the microcode updates on the Debian security repo. I had that enabled on the 7.4 to mitigate zen bleed.

As such adding non-free-firmware to the repo and installing amd64-microcode and rebooting fixed the issue. And also fixed up the processors back to the same microcode level as before.

cat /etc/apt/sources.list
deb http://ftp.uk.debian.org/debian bookworm main contrib

deb http://ftp.uk.debian.org/debian bookworm-updates main contrib non-free non-free-firmware

# security updates
deb http://security.debian.org bookworm-security main contrib non-free non-free-firmware

apt-get install amd64-microcode

You don't need to mess about disabling instructions sets which hurts performance to anything hpc or intense like my cluster runs. It might have been fine for the users in the other thread but not for my use case but really firmware should be enabled out of the box

 

[VictorSTS]

Why not just using CPU type "host"?

 

[damo2929]

I have some Eypc 9000 series GPU servers and using host stop migration from those hosts when the VM is scheduled off the full GPU and all the apps the science team use are complied with -march native as such I need to ensure that native is the lowest system class. That's why removing xsave wasn't a starter as it would have required a mass recompile and revalidation.

 

[VictorSTS]

Very specific use case! Thanks a lot for the details.

 

[damo2929]

I have updated my post to show the exact changes needed to resolve.