Help resetting a 3060 Ti after VM shutdown

[MrnMrk]

So, i have been searching around the whole day, and I think the problem is that the gpu does not reset once the VM is either shutdown or rebooted, PVE just get in a very "freezy" state, and dmesg reports as this: https://i.imgur.com/8Mv9dzy.png. Is there a way to fix this problem? vendor-reset only works for AMD gpus.

 

[leesteken]

vendor-reset only works for AMD gpus.

Ask NVidia to support the open source driver or at least provide documentation to the people making the nouveau driver and vendor-reset.

 

[MrnMrk]

Ask NVidia to support the open source driver or at least provide documentation to the people making the nouveau driver and vendor-reset.

So you are saying that everyone who is using a GeForce GPU does have this issue?

 

[leesteken]

So you are saying that everyone who is using a GeForce GPU does have this issue?

No, I'm saying that it's really difficult to fix reset issues on NVidia GPUs without their help.

 

[MrnMrk]

No, I'm saying that it's really difficult to fix reset issues on NVidia GPUs without their help.

so i have another question, is it more of a VM software driver WHQL side or GPU VBIOS?

 

[leesteken]

so i have another question, is it more of a VM software driver WHQL side or GPU VBIOS?

For a lot of AMD GPUs, it is a hardware (or GPU firmware) issue that it reported FLR support but does not (function level) reset properly and PCI-bus reset also does not work. It could be fixed by using (Linux, open source) driver parts (and a lot of effort of volunteers) and put into vendor-reset, as it's too much code (for a quirk fix) to go into the Linux kernel.
Some GPUs models from some manufacturers reset fine, while the same model from others does not and sometimes whole generations/sub-series do or do not.

I cannot comment on NVidia GPUs or drivers, as I have little experience with them (because of them breaking passthrough for consumer GPUs on purpose in the past) and I don't do this (deep level of) Linux kernel/driver development myself.

 

[damarges]

Maybe this can help you (it did for me):
https://forum.proxmox.com/threads/nvidia-gpu-on-host-unusable-after-use-in-vm.137207/#post-628426

 

[MrnMrk]

Maybe this can help you (it did for me):
https://forum.proxmox.com/threads/nvidia-gpu-on-host-unusable-after-use-in-vm.137207/#post-628426

This didnt work for me, i fixed it by disabling ROM Bar and All functions

 

[emunt6]

Hi!

Try the "suspend/resume" - sometimes works.

$> qm start <vmid>
$> qm suspend <vmid>
$> qm resume <vmid>

If you lucky, the GPU will work again.