[SOLVED] GPU hard-reset on passthrough

[andi-at]

Hello,

I'm using proxmox for a few years with gpu passthrough (radeon rx480) to a Win10 machine without problems.

But since a few weeks/months, not quite sure, i expected some problems. If the windows vm crashes it won't start up again, windows tries to load but after 2 times run into the bootrepair. Thats definitely due to the graphic card, if i remove the passthrough it starts up normal, also if i deinstall the card in windows device messenger it boots normal. But the gpu will not work after reinstalling the driver. However the problem vanishs if i reboot the proxmox host, until the next freeze.

So seems like the card gets stuck at some point wich needs a hard reset to get it working again.

Now i had an idea, i set up a ubuntu box and attached the 'frozen' card, booted it up to desktop without problems, shut down ubuntu, reattach the card to windows and tatam -> it works again without a reboot of the proxmox host. So i guess its some kind of software bug of the card, i dont get any errors in dmesg.

Long text shirt question, is there a way to reset the card manually? I read online that there is no way like a native pci command, but is it possible to reattach the card to the host while running? Removing the card from the vfio module and reattach it?

Regards,
Andi

 

[leesteken]

For Radeon cards, which do not (always) reset properly, there is vendor-reset. I installed it onto my Proxmox host and it works very well for resetting my RX570. Even to the point that I can use the card on the host in between passing it through to VMs.

 

[andi-at]

Thanks a lot, that was exactly what i´m looking for! I´ve installed it and will try it at the next crash, but the readme says it also supports my card, so looks like that will make it.

Mabe it would be a nice feature to implement in as default in proxmox, or at least write it to the wiki?

Anyway thanks a lot and regards,
Andreas