Hi,
 

I'm running out of ideas. It feels like I have watched every video multiple times without any success.
I try to pass through an Nvidia GT710 to a Windows 10 VM.
My System is an AsRock B450M Pro4-F (BIOS 1.50) running a Ryzen 5 2600 (so no iGPU).

Unfortunately I only have a PCIe x1 left and no second GPU to test if it would work with two GPUs.

IOMMU is enabled and the GPU is in its own IOMMU group.

2020-09-23T11:39:43.986646Z qemu-system-x86_64: -device vfio-pci,host=0000:07:00.0,id=hostdev0,bus=pci.5,addr=0x1,romfile=/mnt/cache/isos/Asus.GT710.2048.170525_edit.rom: Failed to mmap 0000:07:00.0 BAR 3. Performance may be slow
2020-09-23T11:39:45.171024Z qemu-system-x86_64: vfio: Unable to power on device, stuck in D3

What I tried so far:

• I tried it without a BIOS, with the BIOS downloaded from here as wall as the modified BIOS according to the video from Spaceinvader One.
• I also set PCIe ACS override and VFIO allow unsafe interrupts
• I used the VFIO-PCI Config plugin

 

If you need any additional info I will provide it as fast as possible.
Right now I'm just lost...


Edit:
after cleaning up several changes and just using "video-efifb:off vfio-pci.ids=10de:128b,10de:0e0f" in syslinux.cfg I was able to get rid of of the mmap error.
And the "Unable to power on device, stuck in D3" (sometimes) gone.
The connected monitor has some boot output on it and as soon as I start the VM it goes dark and the VM is not booting (static IP remains unpingable)

Edit2:
The solution was to update to BIOS 2.0 with AGESA 1.0.0.4 B even if AsRock writes on their page that "this version is NOT recommended with the Ryzen 5 2600".
AGESA 1.0.0.4 seems to fix the D3 error for multiple vendors.

If I find time the next days i try to clean up all steps to come up with the minimal set of modifications it took me to get my setup up and running.

#### Final words ####
After figuring out that the BIOS/AGESA update did the job for the D3 error it was time to clean up the config mess created during trial and error.

Here is the "minimal" config that made it finally work for me:
Since the GPU I want to pass trough is the only one in the system Unraid shouldn't use it to display it's boot output. This was achieved by adding "video=efifb:off" to syslinux config. Having this parameter set made it even unnecessary to explicitly specify vfio-pci.ids=xxx

 

I tried several vBIOS files the two options which worked for me:

• Techpowerup DL with modification (Spaceinvader video)
• GPU-Z dump with modification
• Dumping from linux directly didn't work and I got a different filesize - maybe the card was in the wrong slot and this only works if the card is not the primary one.

 

At some point a already working config got me a wrong screen resolution and "error 43" after trying a wrong config. This was solved by reinstalling the Nvidia driver.
The card is also properly released an can be shared between multiple VMs (of course not at the same time)

Edited September 25, 20205 yr by TEckl

My setup is a little different, but I am attempting to get a second gpu passthrough to work using a pcie x1 port. I have an powered adapter to fit a x16 amd radeon rx 570. I have successfully (though that was difficult at the time too) gotten a gpu to pass through on the main x16 slot and alos verified this new 570 gpu works in that slot as well. 

 

The problem I am stuck on while trying to pass through to an Ubuntu VM on x1 slot is the boot process starts and Ubuntu trys to boot, but then I get errors which seem to be a result of not getting all the pcie information needed. I end up stalling the boot with an error:

 

busybox v1.30.1 (Ubuntu 1:1.30.1-4ubuntu6.3) built-in shell

(initramfs)

 

Obviously the gpu is somewhat being passed though since I see some activity on my monitor. But did you come across any errors in your Windows VM while using the x1 slot? It almost as if my Ubuntu system has an expectation of a x16 gpu when booting and when it discovers that it only has x1 connection it times out and boot fails.