Proxmox not booting after working GPU passthrough

[cube10025]

So, I managed to get a VM working with GPU passthrough. Initially, I was just testing it on my server, but I eventually decided to use it as my main server host. During this process, I had to open the computer to install some additional cards, including an Ethernet card. However, while doing this, I accidentally hit the CMOS button. As a result, Proxmox now refuses to start. The system does boot when the PCI graphics card is removed.

The machine I'm using is an HP Elitedesk 800 G1 SFF.

I need to access the BIOS settings and disable the PCI graphics card. Unfortunately, I can't enter the BIOS with the PCI card installed. Moreover, without the card, the VGA settings in BIOS are grayed out.

Strangely, I've managed to boot into Windows. Within Windows, I've noticed a resource conflict with the PCI bus. I didn't anticipate that any of the GRUB settings would impact the motherboard in this way.

I'm wondering if there's a way to reset the PCI address or resolve the resource conflict.

For reference, here are the commands I used to set up GPU passthrough:

1. Opened the GRUB configuration:
   arduinoCopy code
   nano /etc/default/grub
2. Modified the GRUB default command line:
   makefileCopy code
   #GRUB_CMDLINE_LINUX_DEFAULT="quiet"
   GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on"
3. Updated GRUB:
   sqlCopy code
   update-grub
4. Edited the modules configuration:
   bashCopy code
   nano /etc/modules
   vfiovfio_iommu_type1vfio_pcivfio_virqfd
5. Added options for unsafe interrupts:
   bashCopy code
   echo "options vfio_iommu_type1 allow_unsafe_interrupts=1" > /etc/modprobe.d/iommu_unsafe_interrupts.conf
6. Modified the blacklist configuration:
   bashCopy code
   nano /etc/modprobe.d/blacklist.conf
   Added the following lines:
   Copy code
   blacklist radeon
   blacklist nouveau
   blacklist nvidia
7. Unbound the GPU from its current driver:
   rubyCopy code
   echo "0000:01:00.0" > /sys/bus/pci/drivers/vfio-pci/unbind
8. Bound the GPU to the vfio-pci driver:
   rubyCopy code
   echo "0000:01:00.0" > /sys/bus/pci/drivers/vfio-pci/bind
   
   I might have done something more...

I'd greatly appreciate any advice or solutions to the current issue. Thank you in advance for your assistance!

 

[leesteken]

EDIT: Nevermind, I don't have a clue. Maybe look in the manual if some slots are mutually exclusive?

 

[cube10025]

What does "refuses to start" mean? Please be aware that adding (or removing) PCI(e) devices changes the PCI IDs of other devices. The network interface name is derived from the PCI ID. There are lots of threads about people losing connection to Proxmox after adding/removing PCI(e) devices, and the solution is to determine the new network interface name (with ip a) and edditing the network configuration (with nano /etc/network/interfaces) to match again.

EDIT: On second reading, it sounds like a weird motherboard issue. Look in the manual if some slots are mutually exclusive and/or try moving devices to different slots (which can also affect the PCI ID).

The proxmox host refuses to boot, I can see it not even getting past GRUB based on the lights. There is no screen I can not get in to bios and change the vga. IP is not the problem. As said I can boot without PCI graphiccard but then I can´t see iommu addresses to change.

 

[cube10025]

For those encountering a similar issue, after two days of attempts.

After trying various solutions, including different graphics cards, I inserted a PCI Ethernet card into my system, and it was recognized during boot-up.

When I reinserted the original PCI graphics card, the issue persisted. Attempts to address it through the Windows Registry Editor and Device Manager didn't yield immediate results.

After a system reboot for an unrelated reason, the graphics card started functioning correctly. This allowed me to boot into Proxmox and resolved the initial issue.

I remain unsure about the exact cause. I think reinstalling the graphics card drivers in Windows helped resolve the problem. I think windows fixed the issue, probably a different card in the PCI slot.