[vfio-users] AMD RX 580 - Code 43 device manager error on Win7/Win10 guests

[Gary]

On 25/08/18 06:36, Taiidan at gmx.com wrote:
> Board: KGPE-D16 running coreboot. (these run CPU's similar to the FX series)
> 
> While I receive video output from the guest no matter what I do 2d/3d
> acceleration won't enable.
> 
> I see nothing in host dmesg about this (no BAR issues or what not) - but
> the one thing I noticed is that the device doesn't bind to vfio at boot
> despite the addition of the device ids to the kernel command line.
> 
> Ideas?

You can try adding in /etc/modprobe.d/vfio.conf

# Softdep options may be needed if amdgpu still claims gfx rather than
vfio_pci
#softdep radeon pre: vfio vfio_pci
#softdep amdgpu pre: vfio vfio_pci
#softdep nouveau pre: vfio vfio_pci
#softdep drm pre: vfio vfio_pci
#options kvm_amd avic=1

options vfio-pci ids=1002:67df,1002:aaf0

The commented out options shouldn't be needed but if the loading of vfio
is after the rest it can help switch the order. Don't forget to run
update-grub too.

Some people found they had to have the vfio-pci static bound to
initramfs too, although I didn't need that with the amdgpu (RX580 too)

Also make sure you've specified the ID for both the GPU and the audio
interface of the GPU and that the IDs are correct, the ones you see with
lspci -nn


I originally setup with vfio-pci binding in the above way, but in the
end removed all of this and went with the following setup which I find
more convenient to use.

https://www.mups.co.uk/post/2018/08/seamless-gpu-pass-through-on-linux/

Which allows usage of the card in X session using DRI_PRIME=1 or running
the vm and the gpu will automatically be rebound to vfio-pci and bound
back to amdgpu when you shut it down.

The only issue I've ran into with this setup is the rebinding causes a
kernel panic right before the power off should occur.

Gary