Unable to start VM with PCIe GPU passthrough

[BottledBoa]

Hi, I've been using Proxmox for about a year and this is my first attempt at GPU passthrough (or any passthrough for that matters). I followed several tutorials but nothing seems to work. The VM works just fine up until the moment I add the PCI device to the VM, after which the VM just goes unresponsive and either goes into recovery mode or freezes at boot, as per screenshot attached.

Can anyone help me troubleshoot it? The issue is occurring on Proxmox Virtual Environment 7.4-17, here's a list of files I modified:

on /etc/default/grub
GRUB_CMDLINE_LINUX_DEFAULT="quiet intel_iommu=on pcie_acs_override=downstream,multif>

on /etc/modules
vfio vfio_iommu_type1 vfio_pci vfio_virqfd

on /etc/modprobe.d/iommu_unsafe_interrupts.conf
options vfio_iommu_type1 allow_unsafe_interrupts=1

on /etc/modprobe.d/kvm.conf
options kvm ignore_msrs=1

on /etc/modprobe.d/blacklist.conf
blacklist radeon blacklist amdgpu blacklist nvidia blacklist nouveau

on /etc/modprobe.d/vfio.conf
options vfio-pci ids=1002:682b,1002:aab0 disable_vga=1
(where 1002:682b and 1002:aab0 are the GPU I want to pass through and the associated audio driver IDs respectively)

on /etc/pve/qemu-server/301.conf (where 301 is the VM ID)
cpu: host,hidden=1,flags=+pcid args: -cpu 'host,+kvm_pv_unhalt,+kvm_pv_eoi,hv_vendor_id=NV43FIX,kvm=off'

finally, on /proc/cmdline
initcall_blacklist=sysfb_init

Thanks in advance for any replies!

 

[leesteken]

You are doing a lot of things are or were necessary for NVidia GPUs but you appear to passthrough a AMD GPU for which such work-around are counter productive.
It would help to know which AMD GPU as many need vendor-reset to function with passthrough. It would also help to show the VM configuration file and to know if you can boot your system without using the GPU you want to passthrough. All this has been discussed many times over on this forum, so searching it a bit for your specific GPU might help also.

 

[BottledBoa]

You are doing a lot of things are or were necessary for NVidia GPUs but you appear to passthrough a AMD GPU for which such work-around are counter productive.
It would help to know which AMD GPU as many need vendor-reset to function with passthrough. It would also help to show the VM configuration file and to know if you can boot your system without using the GPU you want to passthrough. All this has been discussed many times over on this forum, so searching it a bit for your specific GPU might help also.

You may be right, I've been banging my head against this wall for nearly a week so I've tried every possible thing I could find. The GPU I'm trying to passthrough is indeed an AMD GPU, and more precisely a Radeon HD 8830M/Cape Verde PRO/ Venus LE/ R7 250 or equivalent, according to Proxmox itself.
I've seen a couple post with similar issues but none of the solutions that worked for others have worked for me so far, that's the main reason for me opening an account just to post this, I apologize if this has already been discussed and solved many times over, I'll have a better look through the forums here.

In the meantime, yes, as stated in the post, the OS boots just fine without the PCI device attached; here's the current VM configuration file:
GNU nano 5.4 /etc/pve/qemu-server/301.conf agent: 1 args: -cpu 'host,+kvm_pv_unhalt,+kvm_pv_eoi,hv_vendor_id=NV43FIX,kvm=off' balloon: 0 bios: ovmf boot: order=ide0;net0 cores: 8 cpu: host,hidden=1,flags=+pcid efidisk0: STG01:301/vm-301-disk-0.qcow2,efitype=4m,pre-enrolled-keys=1,size=528K ide0: STG01:301/vm-301-disk-1.qcow2,size=300G machine: pc-q35-7.2 memory: 16384 meta: creation-qemu=7.2.0,ctime=1700339055 name: WIN11PS01 net0: e1000=FE:2D:DF:0B:27:F4,bridge=vmbr0,firewall=1 numa: 0 ostype: win11 scsihw: virtio-scsi-single smbios1: uuid=ce5e836a-1507-46ee-9fda-74cf32e02aa3 sockets: 1 tpmstate0: STG01:301/vm-301-disk-2.raw,size=4M,version=v2.0 vga: qxl,memory=64 vmgenid: 9dd9107a-6bcc-4442-9cff-f958b499a797

 

[leesteken]

That GPU probably does not reset properly. Can you boot your Proxmox host with another GPU, like and integrated graphics of a different GPU? Can you make sure you have no output on the Radeon Cape Verde PRO/ Venus LE/ R7 250 (which one is it?) when the Proxmox host starts? Otherwise, you'll need to boot your Proxmox host headless, which makes troubleshooting more difficult.

 

[BottledBoa]

That GPU probably does not reset properly. Can you boot your Proxmox host with another GPU, like and integrated graphics of a different GPU? Can you make sure you have no output on the Radeon Cape Verde PRO/ Venus LE/ R7 250 (which one is it?) when the Proxmox host starts? Otherwise, you'll need to boot your Proxmox host headless, which makes troubleshooting more difficult.

I have an Nvidia Tesla on the way, but for now this is the only GPU I have on the server, which does not feature integrated graphics whatsoever; I was using it more as a test to get familiar with passthrough in advance. When the host boots with a monitor plugged in, I do get some display output through the Radeon card, despite all the settings that should prevent it. As for booting it headless, how would I go about it? I can probably manage it via web interface or SSH if need be. Thanks!