@DustyArmstrong

Thanks for responding to the GitHub issue. It’s great that more people want this working; it’s difficult to gain traction otherwise.

Regarding your list, it’s correct. A reboot should be on the second place. You need to reboot only to detach your PCI device (video card) from its driver and assign it to the pciback driver instead on the next boot. This effectively creates a reservation for the device and allows you to dynamically assign it to VMs.

Once your card is free from other kernel drivers, the rest doesn’t require a reboot.

@yannsionneau said:

Can you retry with an up-to-date xcp-ng 8.3 please?

FYI on recent XCP-ng 8.3 versions the pci-passthrough will enable the ROM expansion bar. The guest VM will have access to it, so no need to pass it via qemu anymore.
See my comment on GitHub: https://github.com/xcp-ng/xcp/issues/786#issuecomment-4281846490

Regards,

Yann

I tried, yes. I confirmed as well on github that the patch does what it suppose to do - it exposes ROM BAR, but that alone is not sufficient to get our cards working. More in my other topic..

I believe it will still fix a lot of issues, and it could potentially fix amd gpu passthrough for some card, but not Phoenix, Raphael and this generation of Ryzen iGPUs (not sure specifically about Barcelo)..

But it is a good progress anyway as I wouldn't be able to fix this ever on my own, so I am really glad it is done. Now I will probably get back to the topic and try to patch whatever else is needit and give it to people in a form of rpm package for the time being.. we will see, it is all about time

@DustyArmstrong

Thanks for responding to the GitHub issue. It’s great that more people want this working; it’s difficult to gain traction otherwise.

Regarding your list, it’s correct. A reboot should be on the second place. You need to reboot only to detach your PCI device (video card) from its driver and assign it to the pciback driver instead on the next boot. This effectively creates a reservation for the device and allows you to dynamically assign it to VMs.

Once your card is free from other kernel drivers, the rest doesn’t require a reboot.

@DustyArmstrong @teddyastie

Guys, try this https://xcp-ng.org/forum/post/103216

A deep dive is on the linked medium article in that post.

You're right in your findings about the ROMBAR.

It is not a huge problem that is not passed through. What annoys me is that the support for "fake" ROMBAR (config options rombar and romfile) is present in the xen_pci_passthrough QEMU device, but it doesn't work

I found a workaround by using the “loader” QEMU device. This device simply places your vBIOS in the right memory region where the amdgpu driver can find it. It is a quite nice, clean solution, and I hope my pull request for required changes in qemu-wrapper will be integrated soon. In my opinion, it doesn't hurt anything, but I am still waiting.

@Tristis-Oris sharing my google drive link via chat... should be easy to upload it there

@Tristis-Oris thanks a lot, but is says: The transfer you requested has been deleted. sorry for troubles

@msupport said in GPU support and Nvidia Grid vGPU:

@Tristis-Oris
The driver version 17.1 worked for me, 17.2 and 17.3 crashed the Windows drivers
https://we.tl/t-VozEeV8TFB

May I ask for a re-up? For some reason, I don't see version 17 in my Nvidia Licensing Portal The latest is 16