DPDK Physical Ports¶
The netdev datapath allows attaching of DPDK-backed physical interfaces in order to provide high-performance ingress/egress from the host.
Important
To use any DPDK-backed interface, you must ensure your bridge is configured correctly. For more information, refer to DPDK Bridges.
Changed in version 2.7.0: Before Open vSwitch 2.7.0, it was necessary to prefix port names with a
dpdk
prefix. Starting with 2.7.0, this is no longer necessary.
Quick Example¶
This example demonstrates how to bind two dpdk
ports, bound to physical
interfaces identified by hardware IDs 0000:01:00.0
and 0000:01:00.1
, to
an existing bridge called br0
:
$ ovs-vsctl add-port br0 dpdk-p0 \
-- set Interface dpdk-p0 type=dpdk options:dpdk-devargs=0000:01:00.0
$ ovs-vsctl add-port br0 dpdk-p1 \
-- set Interface dpdk-p1 type=dpdk options:dpdk-devargs=0000:01:00.1
For the above example to work, the two physical interfaces must be bound to the DPDK poll-mode drivers in userspace rather than the traditional kernel drivers. See the binding NIC drivers <dpdk-binding-nics> section for details.
Binding NIC Drivers¶
DPDK operates entirely in userspace and, as a result, requires use of its own poll-mode drivers in user space for physical interfaces and a passthrough-style driver for the devices in kernel space.
There are two different tools for binding drivers: driverctl which
is a generic tool for persistently configuring alternative device drivers, and
dpdk-devbind which is a DPDK-specific tool and whose changes do not
persist across reboots. In addition, there are two options available for this
kernel space driver - VFIO (Virtual Function I/O) and UIO (Userspace I/O) -
along with a number of drivers for each option. We will demonstrate examples of
both tools and will use the vfio-pci
driver, which is the more secure,
robust driver of those available. More information can be found in the
DPDK drivers documentation.
To list devices using driverctl, run:
$ driverctl -v list-devices | grep -i net
0000:07:00.0 igb (I350 Gigabit Network Connection (Ethernet Server Adapter I350-T2))
0000:07:00.1 igb (I350 Gigabit Network Connection (Ethernet Server Adapter I350-T2))
You can then bind one or more of these devices using the same tool:
$ driverctl set-override 0000:07:00.0 vfio-pci
Alternatively, to list devices using dpdk-devbind, run:
$ dpdk-devbind --status
Network devices using DPDK-compatible driver
============================================
<none>
Network devices using kernel driver
===================================
0000:07:00.0 'I350 Gigabit Network Connection 1521' if=enp7s0f0 drv=igb unused=igb_uio
0000:07:00.1 'I350 Gigabit Network Connection 1521' if=enp7s0f1 drv=igb unused=igb_uio
Other Network devices
=====================
...
Once again, you can then bind one or more of these devices using the same tool:
$ dpdk-devbind --bind=vfio-pci 0000:07:00.0
Changed in version 2.6.0: Open vSwitch 2.6.0 added support for DPDK 16.07, which in turn renamed the
former dpdk_nic_bind
tool to dpdk-devbind
.
For more information, refer to the DPDK drivers documentation.
Multiqueue¶
Poll Mode Driver (PMD) threads are the threads that do the heavy lifting for
userspace switching. Correct configuration of PMD threads and the Rx
queues they utilize is a requirement in order to deliver the high-performance
possible with DPDK acceleration. It is possible to configure multiple Rx queues
for dpdk
ports, thus ensuring this is not a bottleneck for performance. For
information on configuring PMD threads, refer to PMD Threads.
Traffic Rx Steering¶
Warning
This feature is experimental.
Some control protocols are used to maintain link status between forwarding engines. In SDN environments, these packets share the same physical network with the user data traffic.
When the system is not sized properly, the PMD threads may not be able to process all incoming traffic from the configured Rx queues. When a signaling packet of such protocols is dropped, it can cause link flapping, worsening the situation.
Some physical NICs can be programmed to put these protocols in a dedicated hardware Rx queue using the rte_flow API.
Warning
This feature is not compatible with all NICs. Refer to the DPDK compatibility matrix and vendor documentation for more details.
Rx steering must be enabled for specific protocols per port. The
rx-steering
option takes one of the following values:
rss
Do regular RSS on all configured Rx queues. This is the default behaviour.
rss+lacp
Do regular RSS on all configured Rx queues. An extra Rx queue is configured for LACP packets (ether type
0x8809
).
Example:
$ ovs-vsctl add-port br0 dpdk-p0 -- set Interface dpdk-p0 type=dpdk \
options:dpdk-devargs=0000:01:00.0 options:n_rxq=2 \
options:rx-steering=rss+lacp
Note
If multiple Rx queues are already configured, regular hash-based RSS (Receive Side Scaling) queue balancing is done on all but the extra Rx queue.
Tip
You can check if Rx steering is supported on a port with the following command:
$ ovs-vsctl get interface dpdk-p0 status
{..., rss_queues="0-1", rx_steering_queue="2"}
This will also show in ovs-vswitchd.log
:
INFO|dpdk-p0: rx-steering: redirecting lacp traffic to queue 2
INFO|dpdk-p0: rx-steering: applying rss on queues 0-1
If the hardware does not support redirecting the specified protocols to a dedicated queue, it will be explicit:
$ ovs-vsctl get interface dpdk-p0 status
{..., rx-steering=unsupported}
More details can often be found in ovs-vswitchd.log
:
WARN|dpdk-p0: rx-steering: failed to add lacp flow: Unsupported pattern
To disable Rx steering on a port, use the following command:
$ ovs-vsctl remove Interface dpdk-p0 options rx-steering
You can see that it has been disabled in ovs-vswitchd.log
:
INFO|dpdk-p0: rx-steering: default rss
Warning
This feature is mutually exclusive with other-config:hw-offload
as it
may conflict with the offloaded flows. If both are enabled, rx-steering
will fall back to default rss
mode.
Flow Control¶
Flow control can be enabled only on DPDK physical ports. To enable flow control support at Tx side while adding a port, run:
$ ovs-vsctl add-port br0 dpdk-p0 -- set Interface dpdk-p0 type=dpdk \
options:dpdk-devargs=0000:01:00.0 options:tx-flow-ctrl=true
Similarly, to enable Rx flow control, run:
$ ovs-vsctl add-port br0 dpdk-p0 -- set Interface dpdk-p0 type=dpdk \
options:dpdk-devargs=0000:01:00.0 options:rx-flow-ctrl=true
To enable flow control auto-negotiation, run:
$ ovs-vsctl add-port br0 dpdk-p0 -- set Interface dpdk-p0 type=dpdk \
options:dpdk-devargs=0000:01:00.0 options:flow-ctrl-autoneg=true
To turn on the Tx flow control at run time for an existing port, run:
$ ovs-vsctl set Interface dpdk-p0 options:tx-flow-ctrl=true
The flow control parameters can be turned off by setting false
to the
respective parameter. To disable the flow control at Tx side, run:
$ ovs-vsctl set Interface dpdk-p0 options:tx-flow-ctrl=false
Rx Checksum Offload¶
By default, DPDK physical ports are enabled with Rx checksum offload.
Rx checksum offload can offer performance improvement only for tunneling traffic in OVS-DPDK because the checksum validation of tunnel packets is offloaded to the NIC. Also enabling Rx checksum may slightly reduce the performance of non-tunnel traffic, specifically for smaller size packet.
Hotplugging¶
OVS supports port hotplugging, allowing the use of physical ports that were not bound to DPDK when ovs-vswitchd was started.
Warning
This feature is not compatible with all NICs. Refer to vendor documentation for more information.
Important
Ports must be bound to DPDK. Refer to Binding NIC Drivers for more information.
To hotplug a port, simply add it like any other port:
$ ovs-vsctl add-port br0 dpdkx -- set Interface dpdkx type=dpdk \
options:dpdk-devargs=0000:01:00.0
Ports can be detached using the del-port
command:
$ ovs-vsctl del-port dpdkx
This should both delete the port and detach the device. If successful, you
should see an INFO
log. For example:
INFO|Device '0000:04:00.1' has been detached
If the log is not seen then the port can be detached like so:
$ ovs-appctl netdev-dpdk/detach 0000:01:00.0
Warning
Detaching should not be done if a device is known to be non-detachable, as this may cause the device to behave improperly when added back with add-port. The Chelsio Terminator adapters which use the cxgbe driver seem to be an example of this behavior; check the driver documentation if this is suspected.
Hotplugging with IGB_UIO¶
Important
As of DPDK v20.11 IGB_UIO has been deprecated and is no longer built as part of the default DPDK library. Below is intended for those who wish to use IGB_UIO outside of the standard DPDK build from v20.11 onwards.
As of DPDK v19.11, default igb_uio hotplugging behavior changed from previous DPDK versions.
From DPDK v19.11 onwards, if no device is bound to igb_uio when OVS is launched then the IOVA mode may be set to virtual addressing for DPDK. This is incompatible for hotplugging with igb_uio.
To hotplug a port with igb_uio in this case, DPDK must be configured to use physical addressing for IOVA mode. For more information regarding IOVA modes in DPDK please refer to the DPDK IOVA Mode Detection.
To configure OVS DPDK to use physical addressing for IOVA:
$ ovs-vsctl --no-wait set Open_vSwitch . \
other_config:dpdk-extra="--iova-mode=pa"
Note
Changing IOVA mode requires restarting the ovs-vswitchd application.
Representors¶
DPDK representors enable configuring a phy port to a guest (VM) machine.
OVS resides in the hypervisor which has one or more physical interfaces also known as the physical functions (PFs). If a PF supports SR-IOV it can be used to enable communication with the VMs via Virtual Functions (VFs). The VFs are virtual PCIe devices created from the physical Ethernet controller.
DPDK models a physical interface as a rte device on top of which an eth device is created. DPDK (version 18.xx) introduced the representors eth devices. A representor device represents the VF eth device (VM side) on the hypervisor side and operates on top of a PF. Representors are multi devices created on top of one PF.
For more information, refer to the DPDK documentation.
Prior to port representors there was a one-to-one relationship between the PF and the eth device. With port representors the relationship becomes one PF to many eth devices. In case of two representors ports, when one of the ports is closed - the PCI bus cannot be detached until the second representor port is closed as well.
When configuring a PF-based port, OVS traditionally assigns the device PCI
address in devargs. For an existing bridge called br0
and PCI address
0000:08:00.0
an add-port
command is written as:
$ ovs-vsctl add-port br0 dpdk-pf -- set Interface dpdk-pf type=dpdk \
options:dpdk-devargs=0000:08:00.0
When configuring a VF-based port, DPDK uses an extended devargs syntax which has the following format:
BDBF,representor=<representor identifier>
This syntax shows that a representor is an enumerated eth device (with
a representor identifier) which uses the PF PCI address.
The following commands add representors of VF 3 and 5 using PCI device address
0000:08:00.0
:
$ ovs-vsctl add-port br0 dpdk-rep3 -- set Interface dpdk-rep3 type=dpdk \
options:dpdk-devargs=0000:08:00.0,representor=vf3
$ ovs-vsctl add-port br0 dpdk-rep5 -- set Interface dpdk-rep5 type=dpdk \
options:dpdk-devargs=0000:08:00.0,representor=vf5
Important
Representors ports are configured prior to OVS invocation and independently of it, or by other means as well. Please consult a NIC vendor instructions on how to establish representors.
Intel NICs ixgbe and i40e
In the following example we create one representor on PF address
0000:05:00.0
. Once the NIC is bounded to a DPDK compatible PMD the
representor is created:
# echo 1 > /sys/bus/pci/devices/0000\:05\:00.0/max_vfs
Mellanox NICs ConnectX-4, ConnectX-5 and ConnectX-6
In the following example we create two representors on PF address
0000:05:00.0
and net device name enp3s0f0
.
Ensure SR-IOV is enabled on the system.
Enable IOMMU in Linux by adding intel_iommu=on
to kernel parameters, for
example, using GRUB (see /etc/grub/grub.conf).
Verify the PF PCI address prior to representors creation:
# lspci | grep Mellanox 05:00.0 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4] 05:00.1 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4]
Create the two VFs on the compute node:
# echo 2 > /sys/class/net/enp3s0f0/device/sriov_numvfs
Verify the VFs creation:
# lspci | grep Mellanox 05:00.0 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4] 05:00.1 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4] 05:00.2 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4 Virtual Function] 05:00.3 Ethernet controller: Mellanox Technologies MT27700 Family [ConnectX-4 Virtual Function]
Unbind the relevant VFs 0000:05:00.2..0000:05:00.3:
# echo 0000:05:00.2 > /sys/bus/pci/drivers/mlx5_core/unbind # echo 0000:05:00.3 > /sys/bus/pci/drivers/mlx5_core/unbind
Change e-switch mode.
The Mellanox NIC has an e-switch on it. Change the e-switch mode from legacy to switchdev using the PF PCI address:
# sudo devlink dev eswitch set pci/0000:05:00.0 mode switchdev
This will create the VF representors network devices in the host OS.
After setting the PF to switchdev mode bind back the relevant VFs:
# echo 0000:05:00.2 > /sys/bus/pci/drivers/mlx5_core/bind # echo 0000:05:00.3 > /sys/bus/pci/drivers/mlx5_core/bind
Restart Open vSwitch
To verify representors correct configuration, execute:
$ ovs-vsctl show
and make sure no errors are indicated.
Port representors are an example of multi devices. There are NICs which support multi devices by other methods than representors for which a generic devargs syntax is used. The generic syntax is based on the device mac address:
class=eth,mac=<MAC address>
For example, the following command adds a port to a bridge called br0
using
an eth device whose mac address is 00:11:22:33:44:55
:
$ ovs-vsctl add-port br0 dpdk-mac -- set Interface dpdk-mac type=dpdk \
options:dpdk-devargs="class=eth,mac=00:11:22:33:44:55"
Representor specific configuration¶
In some topologies, a VF must be configured before being assigned to a
guest (VM) machine. This configuration is done through VF-specific fields
in the options
column of the Interface
table.
Important
Some DPDK port use bifurcated drivers, which means that a kernel netdevice remains when Open vSwitch is stopped.
In such case, any configuration applied to a VF would remain set on the kernel netdevice, and be inherited from it when Open vSwitch is restarted, even if the options described in this section are unset from Open vSwitch.
Configure the VF MAC address:
$ ovs-vsctl set Interface dpdk-rep0 options:dpdk-vf-mac=00:11:22:33:44:55
The requested MAC address is assigned to the port and is listed as part of its options:
$ ovs-appctl dpctl/show
[...]
port 3: dpdk-rep0 (dpdk: ..., dpdk-vf-mac=00:11:22:33:44:55, ...)
$ ovs-vsctl show
[...]
Port dpdk-rep0
Interface dpdk-rep0
type: dpdk
options: {dpdk-devargs="<representor devargs>", dpdk-vf-mac="00:11:22:33:44:55"}
$ ovs-vsctl get Interface dpdk-rep0 status
{dpdk-vf-mac="00:11:22:33:44:55", ...}
$ ovs-vsctl list Interface dpdk-rep0 | grep 'mac_in_use\|options'
mac_in_use : "00:11:22:33:44:55"
options : {dpdk-devargs="<representor devargs>", dpdk-vf-mac="00:11:22:33:44:55"}
The value listed as dpdk-vf-mac
is only a request from the user and is
possibly not yet applied.
When the requested configuration is successfully applied to the port,
this MAC address is then also shown in the column mac_in_use
of
the Interface
table. On failure however, mac_in_use
will keep its
previous value, which will thus differ from dpdk-vf-mac
.
Jumbo Frames¶
DPDK physical ports can be configured to use Jumbo Frames. For more information, refer to Jumbo Frames.
Link State Change (LSC) detection configuration¶
There are two methods to get the information when Link State Change (LSC) happens on a network interface: by polling or interrupt.
Configuring the lsc detection mode has no direct effect on OVS itself, instead it configures the NIC how it should handle link state changes. Processing the link state update request triggered by OVS takes less time using interrupt mode, since the NIC updates its link state in the background, while in polling mode the link state has to be fetched from the firmware every time to fulfil this request.
Note that not all PMD drivers support LSC interrupts.
The default configuration is interrupt mode. To set polling mode, option
dpdk-lsc-interrupt
has to be set to false
.
- Command to set interrupt mode for a specific interface::
$ ovs-vsctl set interface <iface_name> options:dpdk-lsc-interrupt=true
- Command to set polling mode for a specific interface::
$ ovs-vsctl set interface <iface_name> options:dpdk-lsc-interrupt=false