WindowsVFIO

Configuration files and instructions for setting up a Windows VM with a VFIO GPU

The script and configuration files here are designed for Arch Linux with a Windows 7 VM being managed with libvirt. Feel free to try this with other setups, but please do not report bugs unless you feel it is relevant to this setup. For more detail, feel free to read the discussion in the following thread:

https://bbs.archlinux.org/viewtopic.php?id=162768

Dependencies

• qemu
• libvirt
• A linux kernel at least at version 3.14 with the option CONFIG_VFIO_PCI_VGA=y

If your graphics card that you are passing is not in it's own iommu group, then you may need to use a patch to override acs.

If you are using the i915 drivers on your host, you may need to use a patch to fix the VGA arbiter.

A packaged mainline kernel with both patches applied can be found in forum's thread.

Instructions

You may either run the setup.sh script to automatically install and set up the required configuration, or manually follow the directions below. If you use the script, follow the directions in Creating a VM below.

Setting up user permissions

Note that this may not be necessary if trying to connect to libvirt as root

If you'd prefer file-permissions instead of polkit, refer to the following section of the Arch Wiki's libvirt page.

1. Create the libvirt group and add your user(s) to it.

2. As root, create the file /etc/polkit-1/rules.d/50-org.libvirt.unix.manage.rules with the following:

   polkit.addRule(function(action, subject) {
       if (action.id == "org.libvirt.unix.manage" &&
           subject.isInGroup("libvirt")) {
           return polkit.Result.YES;
       }
   });
   

Setting up QEMU permissions

1. Create the kvm group.
2. Open up /etc/libvirt/qemu.conf and append user=root and group=kvm to the file.
3. Append the line cgroup_device_acl = [ "/dev/vfio/<IOMMU_GROUP>" ] to the file as well, replacing <IOMMU_GROUP> with the iommu group that your graphics card is in.

The kernel parameters

1. Add intel_iommu=on to your kernel's parameters
2. Get your card's ID with lspci -n (it will be in hex form of ####:####) and add pci-stub.ids=<DEVICE ID> to your kernel parameters, replacing <DEVICE ID> with your card's ID.
3. If you're using the ACS override patch, add pcie_acs_override=downstream to your kernel parameters.
4. If you're using i915 drivers on your host, add i915.enable_hd_vgaarb=1 to your kernel parameters.

Kernel modules

1. Add pci-stub to the MODULES section in /etc/mkinitcpio.conf
2. Blacklist the driver of the card you are passing (either nouveau for Nvidia or radeon for AMD/ATI)
3. Run mkinitcpio to generate the proper startup configuration for your system.

Creating the VM

1. Create your Windows VM using whatever method you prefer. Make sure that you are not using the virtual q35 chipset.
2. Open the VM's configuration and modify the following:

• At the beginning, replace the line:

  <domain type='kvm'>

  with:

  <domain type='kvm' xmlns:qemu='http://libvirt.org/schemas/domain/qemu/1.0'>

• Right after </devices>, add the following lines:

  <qemu:arg value='-vga'/>
  <qemu:arg value='none'/>
  <qemu:arg value='-device'/>
  <qemu:arg value='vfio-pci,host=<DEVICE NUMBER>,x-vga=on'/>
  

Replace <DEVICE NUMBER> with your device's.

Your VM should now be all set to use a VFIO VGA passthrough!