The PCI passthrough feature in OpenStack allows full access and direct control of a physical PCI device in guests. This mechanism is generic for any kind of PCI device, and runs with a Network Interface Card (NIC), Graphics Processing Unit (GPU), or any other devices that can be attached to a PCI bus. Correct driver installation is the only requirement for the guest to properly use the devices.
Some PCI devices provide Single Root I/O Virtualization and Sharing (SR-IOV) capabilities. When SR-IOV is used, a physical device is virtualized and appears as multiple PCI devices. Virtual PCI devices are assigned to the same or different guests. In the case of PCI passthrough, the full physical device is assigned to only one guest and cannot be shared.
Note
For information on creating servers with virtual SR-IOV devices to guests, refer to the Networking Guide. Attaching SR-IOV ports to existing servers is not currently supported, see bug 1708433 for details.
To enable PCI passthrough, follow the steps below:
Note
The PCI device with address 0000:41:00.0
is used as an example. This
will differ between environments.
nova-scheduler
as specified in Configure nova-scheduler.nova-scheduler
service.Specify the PCI alias for the device.
Configure a PCI alias a1
to request a PCI device with a vendor_id
of
0x8086
and a product_id
of 0x154d
. The vendor_id
and
product_id
correspond the PCI device with address 0000:41:00.0
.
Edit /etc/nova/nova.conf
:
[pci]
alias = { "vendor_id":"8086", "product_id":"154d", "device_type":"type-PF", "name":"a1" }
For more information about the syntax of alias
, refer to
Configuration Options.
Restart the nova-api
service.
Configure a flavor to request two PCI devices, each with vendor_id
of
0x8086
and product_id
of 0x154d
:
# openstack flavor set m1.large --property "pci_passthrough:alias"="a1:2"
For more information about the syntax for pci_passthrough:alias
, refer to
Flavors.
Enable VT-d and IOMMU. For more information, refer to steps one and two in Create Virtual Functions.
Configure nova-compute
to allow the PCI device to pass through to
VMs. Edit /etc/nova/nova.conf
:
[pci]
passthrough_whitelist = { "address": "0000:41:00.0" }
Alternatively specify multiple PCI devices using whitelisting:
[pci]
passthrough_whitelist = { "vendor_id": "8086", "product_id": "10fb" }
All PCI devices matching the vendor_id
and product_id
are added to
the pool of PCI devices available for passthrough to VMs.
For more information about the syntax of passthrough_whitelist
,
refer to Configuration Options.
Specify the PCI alias for the device.
From the Newton release, to resize guest with PCI device, configure the PCI alias on the compute node as well.
Configure a PCI alias a1
to request a PCI device with a vendor_id
of
0x8086
and a product_id
of 0x154d
. The vendor_id
and
product_id
correspond the PCI device with address 0000:41:00.0
.
Edit /etc/nova/nova.conf
:
[pci]
alias = { "vendor_id":"8086", "product_id":"154d", "device_type":"type-PF", "name":"a1" }
For more information about the syntax of alias
, refer to Configuration Options.
Restart the nova-compute
service.
The nova-scheduler
selects a destination host that has PCI devices
available with the specified vendor_id
and product_id
that matches the
alias
from the flavor.
# openstack server create --flavor m1.large --image cirros-0.3.5-x86_64-uec --wait test-pci
Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.