GPU Passthrough

[BenW]

So, I started with proxmox. My first project is passing my GPU through to a windows VM. The windows VM itself is installed with UEfi, qm35 and the cpu=cpu: host,hidden=1,flags=+pcid. This went without any problems (well, at least after a few tries).

Added GRUB_CMDLINE_LINUX_DEFAULT="quiet amd_iommu=on" to /etc/default/grub

I then added the vfio, vfio_pci, vfio_virqfd modules to /etc/modules.

Blocked nvidiafb, nvidia, radeon, nouveau in /etc/modprobe.d/pve-blacklist.conf

Added options vfio-pci ids=10de:1e87,10de:10f8,10de:1ad8,10de:1ad9 to /etc/modprobe.d/vfio.conf. Now this was a first problem. It seems like my vfio modules aren't loaded to my guest GPU (it also didn't load for the host GPU when I switched them for testing purposes). So I solved(?) this by adding GRUB_CMDLINE_LINUX_DEFAULT="quiet amd_iommu=on vfio pci.ids=10de:1e87,10de:10f8,10de:1ad8,10de:1ad9" to /etc/default/grub. Now it seemed that the vfio modules were loaded to the guest GPU. I verified this with the lspci -v command.

dmesg | grep 'remapping' --> Remapping is enabled

I've been trying to solve this for a few days now but I can't get it to work. Whenever I run the Windows VM, the screen freezes and after 15 seconds-ish the proxmox server reboots without any error logs. Any help is appreciated.

Kind regards

BIOs settings:

CSM support: enabled --> because my GPU for Proxmox itself uses a dual link connector. (Can this be a problem?)

SVM mode: enabled

Initial display output: PCIe 3 slot --> the host card

ACS enable : enabled



Hardware:

Nvidea RTX 2080 for the guest is in PCIe x16 first slot running at x16

Nvdiea GeForce 6700 for the host is in PCIex4 slot running at x4

Ryzen 1700x

Ax370 gaming k7

 

[Stefan_R]

CSM support: enabled --> because my GPU for Proxmox itself uses a dual link connector. (Can this be a problem?)

I'm unsure what dual link has to do with the Compatibility Support Module, but CSM has been known to cause all kinds of BIOS errors in the past...

Otherwise, with no logs it is really difficult to say anything, but potentially check your IOMMU groups and your isolation therein?

 

[BenW]

I'm unsure what dual link has to do with the Compatibility Support Module, but CSM has been known to cause all kinds of BIOS errors in the past...

Otherwise, with no logs it is really difficult to say anything, but potentially check your IOMMU groups and your isolation therein?

Well, if I turn CSM off, proxmox won't select my other monitor.. Dmesg doesn't give any hints. Where should I look? This is what I get now.

Thanks in advance.