GPU Passthrough Windows 10 - io_page_fault when starting game

[Hydrox1992]

Hi!

My HW config:

MSI A78M-E45
AMD APU A10-7860k
Gigabyte Radeon RX 570 4GB
16GB RAM

I have a problem with setting up GPU Passthrough on Windows 10 VM. VM is working fine on RX 570 until I start a game. Game is always crashing(sometimes after couple of seconds, sometimes instant after start)with on dmesg:

[ 1729.128595] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x000e address=0x0000006304d89ac0 flags=0x0000]
[ 1729.157439] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x000e address=0x00000081fc2e8000 flags=0x0010]
[ 1729.157489] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x000e address=0x00000081fc2e8300 flags=0x0010]
[ 1729.157522] AMD-Vi: Event logged [IO_PAGE_FAULT device=01:00.0 domain=0x000e address=0x00000081fc2e8140 flags=0x0010]

When I've add iommu=soft to grub and restart host I've got information that IOMMU is not supported on this system, and IOMMU groups disappeared.

Windows 10 VM starts fine with(latest AMD Crimson Drivers on VM):

/etc/default/grub:

GRUB_CMDLINE_LINUX_DEFAULT="quiet video=efifb:off"
GRUB_CMDLINE_LINUX=""

VM CONF:

balloon: 0
bios: ovmf
bootdisk: virtio0
cores: 4
efidisk0: storage:200/vm-200-disk-1.raw,size=128K
hostpci0: 01:00,x-vga=on,romfile=RX570G4.f11
machine: q35
memory: 8192
name: GameMachine
net0: virtio=12:D0:70:95:82:99,bridge=vmbr0
numa: 0
ostype: win10
scsihw: virtio-scsi-pci
smbios1: uuid=bf022de4-d283-4271-b09b-122c7de88aee
sockets: 1
usb0: host=3-2
usb1: host=8-1
virtio0: storage:200/vm-200-disk-2.raw,cache=writethrough,discard=on,size=500G

Why soft IOMMU didn't work? Why machine is crashing only when I'm trying to start graphics intensive software and on Windows work fine without any complaining about IOMMU?

Pls help.

 

[TheFunk]

It sounds like we're having similar issues. The wiki page has a couple different settings. Their host GRUB line says to include amd_iommu=on and that if your system doesn't support iommu yet you can allow unsafe interrupts. Not sure if either of those would be helpful but you could always change them back.

One thing, my Windows guest has an error in device manager after I installed the Crimson drivers. I also had to use a Windows 7 guest because trying to install Crimson on my Windows 10 guest was crashing the VM.

Lastly, I read somewhere about large amounts of RAM crashing VMs whenever newer AMD cards are passed through. I also have 8GB of RAM set for my guest. I'm going to experiment with dropping that down to 2GB. I'll let you know if that gives me any sort of results.

I also noticed you're passing through USB devices as well. Maybe wait to pass those through until after you get the card functioning normally. It could be that one of those passthrough devices is creating a conflict or something. Who knows? I'm just spitballing.

So some troubleshooting steps that couldn't hurt:

-Try changing that GRUB config line and update-grub, then reboot the host node. (you can always change this back)
-Try allowing unsafe interrupts (you can always change this back)
-Try a Windows 7 guest instead of 10.
-Check your guest's device manager for some hints
-Try messing with some other configs like RAM and USB passthrough