Assigning Devices to VMs

In order to assign a whole PCI(e) device to a VM, one should use qvm-pci tool. E.g.

lspci

Find the BDF address of the device you want to assign, and then:

qvm-pci -a <vmname> <bdf>

E.g. assuming 00:1a.0 is a BDF of the device I want to assign to the “personal” domain:

qvm-pci -a personal 00:1a.0

Note that one can only assign full PCI or PCI Express devices. This means one cannot assign single USB devices – only the whole USB controller with whatever USB devices connected to it. This limit is imposed by PC and VT-d architecture.

Using Qubes Manager

TODO

<screenshot>

Finding the right USB controller

If you want assign certain USB device to a VM (by attaching a whole USB controller), you need to figure out which PCI device is the right controller. First check to which USB bus the device is connected:

lsusb

For example I want assign a broadband modem to the netvm. In lsusb output it can be listed as something like this (in this case device isn’t fully identified):

Bus 003 Device 003: ID 413c:818d Dell Computer Corp.

The device is connected to the USB bus #3. Then check which other devices are connected to the same bus - all of them will be assigned to the same VM. Now is the time to find right USB controller:

readlink /sys/bus/usb/devices/usb3

This should output something like:

../../../devices/pci-0/pci0000:00/0000:00:1a.0/usb3

Now you see BDF address in the path (right before final usb3). Strip leading “0000:” and pass the rest to qvm-pci tool:

qvm-pci -a netvm 00:1a.0

Possible issues

DMA buffer size

VMs with assigned PCI devices in Qubes have allocated a small buffer for DMA operations (called swiotlb). By default it is 2MB, but some devices need a larger buffer. To change this allocation, edit VM’s kernel parameters (this is expressed in 512B chunks):

# qvm-prefs netvm |grep kernelopts
kernelopts       : iommu=soft swiotlb=2048 (default)
# qvm-prefs -s netvm kernelopts "iommu=soft swiotlb=4096"

This is ​known to be needed for Realtek RTL8111DL Gigabit Ethernet Controller.

PCI passthrough issues

Sometimes PCI arbitrator is too strict. There is a way to enable permissive mode for it. Create /etc/systemd/system/qubes-pre-netvm.service:

[Unit]
Description=Netvm fixup
Before=qubes-netvm.service

[Service]
ExecStart=/bin/sh -c 'echo 0000:04:00.0 > /sys/bus/pci/drivers/pciback/permissive'
Type=oneshot
RemainAfterExit=yes

[Install]
WantedBy=multi-user.target

Then enable it with systemctl enable qubes-pre-netvm.service

See also: ​https://groups.google.com/forum/#!topic/qubes-users/Fs94QAc3vQI, ​http://wiki.xen.org/wiki/Xen_PCI_Passthrough

NOTE: By setting the permissive flag for the PCI device, you’re potentially weakening the device isolation, especially if your system is not equipped with VT-d Interrupt Remapping unit – see ​this paper, page 7 for more details.

Bringing PCI device back to dom0

By default device detached from some VM (or when VM with PCI device attached get shut down) isn’t attached back to dom0. This is an intended feature. A device which was previously assigned to a less trusted AppVM could attack dom0 if it were automatically reassigned there. In order to re-enable the device in dom0, either:

  1. Reboot the physical machine.

or

  1. Go to the sysfs (/sys/bus/pci), find the right device, detach it from the pciback driver and attach back to the original driver. Replace <BDF> with your device, for example 00:1c.2:
echo 0000:<BDF> > /sys/bus/pci/drivers/pciback/unbind
    MODALIAS=`cat /sys/bus/pci/devices/0000:<BDF>/modalias`
    MOD=`modprobe -R $MODALIAS | head -n 1`
    echo <BDF> > /sys/bus/pci/drivers/$MOD/bind