GPU Passthrough Not working in Proxmox VE 6

[Jsingh]

I run the following configuration on my new workstation

Threadripper 1950X
x399 Auros Extreme
24 GB ECC Ram
3x512 GB NVME SSD (In Raid Z1)
4X1Tb Sata HDD as Backup in (RAID Z2)
Asus Geforce 1050ti
Asus Geforce 750ti.

GPU Passthrough was working in Proxmox 5.4. I updated to Proxmox 6 and the Windows crashes with nvidia GPUs.
I again started with a clean installation still, the GPU passthrough isn't working. Windows crashes on GPU driver installation.

 

[dcsapak]

did you already see this thread: https://forum.proxmox.com/threads/vm-w-pcie-passthrough-not-working-after-upgrading-to-6-0.56021/
maybe one of the workarounds there works for you

 

[Jsingh]

I tried the workarounds and I am still getting the same issues. I am using the enterprise repository for all the updates

 

[dcsapak]

I tried the workarounds and I am still getting the same issues. I am using the enterprise repository for all the updates

what exactly did you try?

 

[Jsingh]

what exactly did you try?

I tried adding
qm set ID -args '-machine type=q35,kernel_irqchip=on'

and setting the machine type to 3.1 using:

qm set ID -machine pc-q35-3.1

 

[dcsapak]

and you did this while the vm was off ?

anything special in dmesg/journal?

 

[Jsingh]

Yes , I turned off the VM and did the change. I also restarted the system itself.

This is last lines in dmesg
[ 58.635254] vfio-pci 0000:0a:00.0: enabling device (0000 -> 0003)
[ 58.635526] vfio_ecap_init: 0000:0a:00.0 hiding ecap 0x19@0x900
These are in red:
[ 65.426115] kvm [3700]: vcpu0, guest rIP: 0xfffff80514d3ae32 ignored rdmsr: 0xc0011023
[ 65.427889] kvm [3700]: vcpu0, guest rIP: 0xfffff80514d3ae47 ignored wrmsr: 0xc0011023 data 0x100
[ 65.441128] kvm [3700]: vcpu1, guest rIP: 0xfffff80514d3ae32 ignored rdmsr: 0xc0011023
[ 65.442906] kvm [3700]: vcpu1, guest rIP: 0xfffff80514d3ae47 ignored wrmsr: 0xc0011023 data 0x100
[ 65.445620] kvm [3700]: vcpu2, guest rIP: 0xfffff80514d3ae32 ignored rdmsr: 0xc0011023
[ 65.447392] kvm [3700]: vcpu2, guest rIP: 0xfffff80514d3ae47 ignored wrmsr: 0xc0011023 data 0x100
[ 65.450143] kvm [3700]: vcpu3, guest rIP: 0xfffff80514d3ae32 ignored rdmsr: 0xc0011023
[ 65.451910] kvm [3700]: vcpu3, guest rIP: 0xfffff80514d3ae47 ignored wrmsr: 0xc0011023 data 0x100
[ 65.454690] kvm [3700]: vcpu4, guest rIP: 0xfffff80514d3ae32 ignored rdmsr: 0xc0011023
[ 65.456453] kvm [3700]: vcpu4, guest rIP: 0xfffff80514d3ae47 ignored wrmsr: 0xc0011023 data 0x100
[ 65.459203] kvm [3700]: vcpu5, guest rIP: 0xfffff80514d3ae32 ignored rdmsr: 0xc0011023
[ 65.460970] kvm [3700]: vcpu5, guest rIP: 0xfffff80514d3ae47 ignored wrmsr: 0xc0011023 data 0x100

 

[dcsapak]

These are in red:

those are normal on amd based systems, i have those here too on an EPYC cpu (even without passthrough)

looks normal... any logs from inside the vm that might be relevant?

 

[Integral]

I have a Windows 10 machine with GTX 970 passthrough, the pc-q35-3.1 solved this for me as well

 

[JimmyZ]

I have two VMs, each with a 1070 pass-through, upgrading from 5.4 to 6 0 broke it, I've tried both workarounds, kernel_ireqchip=on or pc-q35-3.1, none of them works, it crashes as soon as the GPU driver gets loaded.

update: I upgraded to 6.0-7 and pc-q35-3.1 works now.