[vfio-users] RX 480 on Windows 10 guest freezes with DX11+ games

[Tomáš M.]

I am using HD 7950 on MB Asus P8Z77-V Pro with i7-3770 on Windows 7 and
Windows 10 for many months now using this script
http://pastebin.com/SeyXWH3E and these boot options:
i915.enable_hd_vgaarb=1 intel_iommu=on intel_iommu=igfx_off

Only issues I had were with few games like Heroes of the Storm, Mad Max,
which caused BSOD (maybe I didn't had "options kvm ignore_msrs=1" in
/etc/modprobe.d/vfio.conf back then)

I decided to try out out RX480, from ASUS
https://www.techpowerup.com/vgabios/185508/asus-rx480-8192-160628

Isolating gpu was simple, just blacklist amdgpu instead of radeon in
/etc/modprobe.d/blacklist.conf and edit the ids in
/etc/modprobe.d/vfio.conf:

options vfio-pci ids=1002:67df,1002:aaf0,8086:1e31
options kvm ignore_msrs=1

and do mkinitcpio -p linux-vfio

I tested that out on linux-vfio 4.8.4 kernel from
https://aur.archlinux.org/pkgbase/linux-vfio/

I tried to run via OVMF, but I only got to UEFI menu on screen connected to
gpu.
I was passing    "-drive if=pflash,format=raw,readonly,file=ovmf_x64.bin" and
tried number of bin files from https://aur.archlinux.org/packages/ovmf-git/
Should I have tried passing different rom for gpu or try to reinstall the
system?

But afaik Alex said there is no real advantage of using OVMF so I just used
the same script for HD7950 and that worked, both Windows upgraded the
drivers without problems.

But only 2d games or older DX9 and lower games like: Unreal Tournament
2004, Freedom Fighters, Original War, Torchlight, Hearthstone worked.
All newer games like Witcher 3, DOOM, Planet Coaster, Overwatch caused
whole PC freeze.
After 4-15s of the game screen went to some weird solid color and
everything froze. I sometimes had to force shutdown, because restart button
was not working.
I watched dmesg -w and vfio process output on host screen and I didn't see
any related errors, just some usb messages after starting the VM.
Menus were okay, but Planet Coaster and Overwatch render something there so
that was quick freeze.
Only newer game that seemed to work for few minutes was ABZU.
MSRS messages were ignored, there were a few of them right after Windows
login. I am not sure if that means anything good or bad.

According to MB manual first PCI 3.0 with 16 links shared some IRQ with USB
controllers, so I tried to use 2nd with 8 links which had less IRQ, which
didn't changed anything.

When it was in there the IOMMU Group was:
IOMMU Group 1 00:01.0 PCI bridge [0604]: Intel Corporation Xeon E3-1200
v2/3rd Gen Core processor PCI Express Root Port [8086:0151] (rev 09)
IOMMU Group 1 00:01.1 PCI bridge [0604]: Intel Corporation Xeon E3-1200
v2/3rd Gen Core processor PCI Express Root Port [8086:0155] (rev 09)
IOMMU Group 1 02:00.0 VGA compatible controller [0300]: Advanced Micro
Devices, Inc. [AMD/ATI] Ellesmere [Radeon RX 470/480] [1002:67df] (rev c7)
IOMMU Group 1 02:00.1 Audio device [0403]: Advanced Micro Devices, Inc.
[AMD/ATI] Device [1002:aaf0]

In 1st PCI-E it was one PCI Bridge less, that should be OK afaik.

Right now I will try the RX480 in other PC with native Arch and Windows to
verify if there is nothing wrong with it, but I doubt it.

I tried turning off the i915 patch, not passing the usb controller (the
8086:1e31) but it was always the same on Windows 7 (16.12 drivers) Windows
10 (17.1 divers).
Any ideas what else to try out?

Thanks
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20170211/cd2c6e46/attachment.htm>