Configuring PCI passthrough (on the example of a USB controller)¶

Configuring PCI device forwarding for instances using the passthrough function in OpenStack. To set up the system, you need a configured computing node with nova, neutron installed and configured.

To perform manipulations, you must first find out the bus and device number, vendor_id, product_id and the name of the driver with which the device is connected in the system.

For example:
```
lspci -nn | grep -i usb
```
Determining the device ID¶

Next step:
```
lspci -nk|grep -A2 -i 8086:8d31
```
Determining the bus and driver number¶
Enable iommu support in the kernel. To enable the option at OS boot, you can apply the addition to the kernel options in GRUB:

To enable iommu, you need to set GRUB_CMDLINE_LINUX_DEFAULT parameter to intel_iommu=on and vfio_iommu_type1.allow_unsafe_interrupts=1``in the ``/etc/default/grub file set the GRUB_CMDLINE_LINUX_DEFAULT. Options are appended to this parameter separated by a space. After editing the file, run the grub-update command (this example is for Intel based systems).
Adding to the file /etc/modules the loading of the necessary kernel modules for forwarding:
```
pci_stub
vfio
vfio_iommu_type1
vfio-pci
vfio_pci
kvm
kvm_intel
```
You need to restart the system to apply the settings.
Checking if the iommu module is enabled in the kernel:
```
virt-host-validate
```
Adding to the black list of drivers for the device that will be forwarded:
```
echo 'blacklist xhci_hcd' >> /etc/modprobe.d/blacklist-xhci_hcd.conf
```
Confirmation that device forwarding is supported (substitute your bus and device number):
```
ls -ld /sys/kernel/iommu_groups/*/devices/*00:14.?/
```
Adding device binding to the vfio-pci module:
- Adding a vendor_id:product_id controller to the vfio module configuration file. If you have several controllers, then you need to specify each one separated by a comma:
```
/etc/modprobe.d/vfio.conf
options vfio-pci ids=8086:8d31
```
- Setting vfio-pci as driver via kernel options:
```
mcedit /etc/default/grub
GRUB_CMDLINE_LINUX_DEFAULT="intel_iommu=on vfio-pci.ids=8086:8d31
vfio_iommu_type1.allow_unsafe_interrupts=1"
update-grub
```
- Verifying that vfio-pci is assigned as a driver:
```
lspci -s 00:14.0 -nnk
```

Setting in nova config file on compute node:

[pci]
passthrough_whitelist: { "vendor_id": "8086", "product_id": "8d31" }

Restarting the nova-compute service:

systemctl restart nova-compute

Setting up the nova service on the controller:

[pci]
alias: { "vendor_id":"8086", "product_id":"8d31", "device_type":"type-PCI",
"name":"usbcontroller" }
[filter_scheduler]
enabled_filters = PciPassthroughFilter
available_filters = nova.scheduler.filters.all_filters

Restarting nova services:

systemctl restart nova-*

Creating a flavor:

openstack flavor create \
--vcpus 8 \
--ram 16384 \
--disk 100 \
--property "pci_passthrough:alias"="usbcontroller:1" \
8_16384_100_usb

where usbcontroller is the name of the alias on the controller, 1 - how many devices will be transferred to the instance.