Skip to content
Switch branches/tags
Go to file

SR-IOV Network device plugin for Kubernetes

Travis CI Go Report Card Weekly minutes

Table of Contents

SR-IOV Network Device Plugin

The SR-IOV network device plugin is Kubernetes device plugin for discovering and advertising SR-IOV virtual functions (VFs) available on a Kubernetes host.


  • Handles SR-IOV capable/not-capable devices (NICs and Accelerators alike)
  • Supports devices with both Kernel and userspace (UIO and VFIO) drivers
  • Allows resource grouping using "Selector"
  • User configurable resourceName
  • Detects Kubelet restarts and auto-re-register
  • Detects Link status (for Linux network devices) and updates associated VFs health accordingly
  • Extensible to support new device types with minimal effort if not already supported
  • Works within virtual deployments of Kubernetes that do not have virtualized-iommu support (VFIO No-IOMMU support)

To deploy workloads with SR-IOV VF this plugin needs to work together with the following two CNI components:

  • Any CNI meta plugin supporting Device Plugin based network provisioning (Multus CNI, or DANM)

    • Retrieves allocated network device information of a Pod

    • During Pod creation, plumbs allocated SR-IOV VF to a Pods network namespace using VF information given by the meta plugin

    • On Pod deletion, reset and release the VF from the Pod

Please follow the Quick Start for multi network interface support in Kubernetes.

Supported SR-IOV NICs

The following NICs were tested with this implementation. However, other SR-IOV capable NICs should work as well.

  • Intel® Ethernet Controller X710 Series 4x10G - PF driver : v2.4.6 - VF driver: v3.5.6

please refer to Intel download center for installing latest Intel Ethernet Controller-X710-Series drivers

  • Intel® 82599ES 10 Gigabit Ethernet Controller
    • PF driver : v4.4.0-k
    • VF driver: v3.2.2-k

please refer to Intel download center for installing latest Intel-® 82599ES 10 Gigabit Ethernet drivers

  • Mellanox ConnectX®-4 Lx EN Adapter
  • Mellanox ConnectX®-5 Adapter

Network card drivers are available as a part of the various linux distributions and upstream. To download the latest Mellanox NIC drivers, click here.

Quick Start

Creating SR-IOV Virtual Functions

Before starting the SR-IOV device plugin you will need to create SR-IOV Virtual Functions on your system. The VF Setup doc will guide you through that process.


  1. Compile SR-IOV-CNI (supported from release 2.0+):
$ git clone
$ cd sriov-cni
$ make
$ cp build/sriov /opt/cni/bin

Build and run SR-IOV network device plugin

You can either build the docker image locally or pull it from docker hub.

If you want to build the docker image locally then follow the following steps:

  1. Clone the sriov-network-device-plugin
$ git clone
$ cd sriov-network-device-plugin
  1. Build docker image binary using make
$ make image

On a successful build, a docker image with tag nfvpe/sriov-device-plugin:latest will be created. You will need to build this image on each node. Alternatively, you could use a local docker registry to host this image.

  1. Create a ConfigMap that defines SR-IOV resrouce pool configuration

Make sure to update the 'config.json' entry in the configMap data to reflect your resource configuration for the device plugin. See Configurations section for supported configuration parameters.

$ kubectl create -f deployments/configMap.yaml
  1. Deploy SR-IOV network device plugin Daemonset
$ kubectl create -f deployments/k8s-v1.16/sriovdp-daemonset.yaml

For K8s version v1.15 or older use deployments/k8s-v1.10-v1.15/sriovdp-daemonset.yaml instead.

Install one compatible CNI meta plugin

A compatible CNI meta-plugin installation is required for SR-IOV CNI plugin to be able to get allocated VF's deviceID in order to configure it.

Option 1 - Multus

Install Multus

Please refer to Multus Quickstart Installation Guide to install Multus.

Network Object CRDs

Multus uses Custom Resource Definitions(CRDs) for defining additional network attachements. These network attachment CRDs follow the standards defined by K8s Network Plumbing Working Group(NPWG). Please refer to Multus documentation for more information.

  1. Create the SR-IOV Network CRD
$ kubectl create -f deployments/sriov-crd.yaml

Option 2 - DANM

This section explains an example deployment of SR-IOV Network device plugin in Kubernetes if you choose DANM as your meta plugin.

Install DANM

Refer to DANM documentation for detailed instructions.

Create SR-IOV type networks

DANM supports the Device Plugin based SR-IOV provisioning with the dynamic level. This means that all DANM API features seamlessly work together with the SR-IOV setup described above, whether you use the lightweight, or the production grade network management APIs. For example manifest objects refer to SR-IOV demo

See following sections on how to configure and run SR-IOV device plugin.


Config parameters

This plugin creates device plugin endpoints based on the configurations given in the config map associated with the SR-IOV device plugin. In json format this file appears as shown below:

    "resourceList": [{
            "resourceName": "intel_sriov_netdevice",
            "selectors": {
                "vendors": ["8086"],
                "devices": ["154c", "10ed"],
                "drivers": ["i40evf", "ixgbevf"]
            "resourceName": "intel_sriov_dpdk",
            "resourcePrefix": "",
            "selectors": {
                "vendors": ["8086"],
                "devices": ["154c", "10ed"],
                "drivers": ["vfio-pci"],
                "pfNames": ["enp0s0f0","enp2s2f1"],
                "needVhostNet": true
            "resourceName": "mlnx_sriov_rdma",
            "resourcePrefix": "",
            "selectors": {
                "vendors": ["15b3"],
                "devices": ["1018"],
                "drivers": ["mlx5_ib"],
                "isRdma": true
            "resourceName": "infiniband_rdma_netdevs",
            "selectors": {
                "linkTypes": ["infiniband"],
                "isRdma": true
            "resourceName": "intel_fpga",
            "deviceType": "accelerator",
            "selectors": {
                    "vendors": ["8086"],
                    "devices": ["0d90"]


"resourceList" should contain a list of config objects. Each config object may consist of following fields:

Field Required Description Type/Defaults Example/Accepted values
"resourceName" Y Endpoint resource name. Should not contain special characters including hyphens and must be unique in the scope of the resource prefix string "sriov_net_A"
"resourcePrefix" N Endpoint resource prefix name override. Should not contain special characters string Default : "" ""
"deviceType" N Device Type for a resource pool. string value of supported types. Default: "netDevice" Currently supported values: "accelerator", "netDevice"
"selectors" N A map of device selectors. The "deviceType" value determines the "selectors" options. json object as string Default: null Example: "selectors": {"vendors": ["8086"],"devices": ["154c"]}

Note: "resourceName" must be unique only in the scope of a given prefix, including the one specified globally in the CLI params, e.g. "", "" and "" are perfectly valid names.

Device selectors

The "deviceType" value determines which selectors are supported for that device. Each selector evaluated in order as listed in selector tables below.

Common selectors

All device types support following common device selectors.

Field Required Description Type/Defaults Example/Accepted values
"vendors" N Target device's vendor Hex code as string string list Default: null "vendors": ["8086"]
"devices" N Target Devices' device Hex code as string string list Default: null "devices": ["154c", "10ed"]
"drivers" N Target device driver names as string string list Default: null "drivers": ["vfio-pci"]
"pciAddresses" N Target device's pci address as string string list Default: null "pciAddresses": ["0000:03:02.0"]

Extended selectors for device type "netDevice"

This selector is applicable when "deviceType" is "netDevice"(note: this is default). In addition to the common selectors from above table, the "netDevice" also supports following selectors.

Field Required Description Type/Defaults Example/Accepted values
"pfNames" N VFs from PF matches list of PF names string list Default: null "pfNames": ["enp2s2f0"] (See follow-up sections for some advance usage of "pfNames")
"rootDevices" N VFs from PF matches list of PF PCI addresses string list Default: null "rootDevices": ["0000:86:00.0"] (See follow-up sections for some advance usage of "rootDevices")
"linkTypes" N The link type of the net device associated with the PCI device string list Default: null "linkTypes": ["ether"]
"ddpProfiles" N A map of device selectors string list Default: null "ddpProfiles": ["GTPv1-C/U IPv4/IPv6 payload"]
"isRdma" N Mount RDMA resources bool values true or false Default: false "isRdma": true
"needVhostNet" N Share /dev/vhost-net bool values true or false Default: false "needVhostNet": true

Extended selectors for device type "accelerator"

This selector is applicable when "deviceType" is "accelerator". The "accelerator" device type currently supports only the common selectors.

Command line arguments

This plugin accepts the following optional run-time command line arguments:

./sriovdp --help

Usage of ./sriovdp:
        log to standard error as well as files
  -config-file string
        JSON device pool config file location (default "/etc/pcidp/config.json")
  -log_backtrace_at value
        when logging hits line file:N, emit a stack trace
  -log_dir string
        If non-empty, write log files in this directory
        log to standard error instead of files
  -resource-prefix string
        resource name prefix used for K8s extended resource (default "")
  -stderrthreshold value
        logs at or above this threshold go to stderr
  -v value
        log level for V logs
  -vmodule value
        comma-separated list of pattern=N settings for file-filtered logging


This plugin does not bind or unbind any driver to any device whether it's PFs or VFs. It also doesn't create Virtual functions either. Usually, the virtual functions are created at boot time when kernel module for the device is loaded. Required device drivers could be loaded on system boot-up time by allow-listing/deny-listing the right modules. But plugin needs to be aware of the driver type of the resources (i.e. devices) that it is registering as K8s extended resource so that it's able to create appropriate Device Specs for the requested resource.

For example, if the driver type is uio (i.e. igb_uio.ko) then there are specific device files to add in Device Spec. For vfio-pci, device files are different. And if it is Linux kernel network driver then there is no device file to be added.

The idea here is, user creates a resource config for each resource pool as shown in Config parameters by specifying the resource name, a list resource "selectors".

The device plugin will initially discover all PCI network resources in the host and populate an initial "device list". Each "resource pool" then applies its selectors on this list and add devices that satisfies the selector's constraints. Each selector narrows down the list of devices for the resource pool. Currently, the selectors are applied in following order:

  1. "vendors" - The vendor hex code of device
  2. "devices" - The device hex code of device
  3. "drivers" - The driver name the device is registered with
  4. "pciAddresses" - The pci address of the device in BDF notation
  5. "pfNames" - The Physical function name
  6. "rootDevices" - The Physical function PCI address
  7. "linkTypes" - The link type of the net device associated with the PCI device.

The "pfNames" and "rootDevices" selectors can be used to specify a list and/or range of VFs for a pool in the below format:





`<PFName>`     - is the PF interface name
`<RootDevice>` - is the PF PCI address
`<SingleVF>`   - is a single VF index (0-based) that is included into the pool
`<FirstVF>`    - is the first VF index (0-based) that is included into the range
`<LastVF>`     - is the last VF index (0-based) that is included into the range


The selector for interface named netpf0 and VF 0, 2 upto 7 (included 2 and 7) and 9 will look like:

"pfNames": ["netpf0#0,2-7,9"]

The selector for PCI address 0000:86:00.0 and VF 0, 1, 3, 4 will look like:

"rootDevices": ["0000:86:00.0#0-1,3,4"]

If only PF network interface or PF PCI address is specified in the selector, then assuming that all VFs of this interface are going to the pool.


  • Load device's (Physical function if it is SR-IOV capable) kernel module and bind the driver to the PF
  • Create required Virtual functions
  • Bind all VF with right drivers
  • Create a resource config map
  • Run SR-IOV device plugin (as daemonset)

On successful run, the allocatable resource list for the node should be updated with resource discovered by the plugin as shown below. Note that the resource name is appended with the -resource-prefix i.e. "".

$ kubectl get node node1 -o json | jq '.status.allocatable'

  "cpu": "8",
  "ephemeral-storage": "169986638772",
  "hugepages-1Gi": "0",
  "hugepages-2Mi": "8Gi",
  "": "8",
  "": "8",
  "memory": "7880620Ki",
  "pods": "1k"

Example deployments

We assume that you have working K8s cluster configured with one of the supported meta plugins for multi-network support. Please see Features and Quick Start sections for more information on required CNI plugins.

Deploy the Device Plugin

The images directory contains example Dockerfile, sample specs along with build scripts to deploy the SR-IOV device plugin as daemonset. Please see for more information about the Docker images.

# Create ConfigMap
$ kubectl create -f deployments/configMap.yaml
configmap/sriovdp-config created

# Create sriov-device-plugin-daemonset
$ kubectl create -f deployments/k8s-v1.16/sriovdp-daemonset.yaml
serviceaccount/sriov-device-plugin created
daemonset.apps/kube-sriov-device-plugin-amd64 created

$ kubectl -n kube-system get pods
NAMESPACE     NAME                                   READY   STATUS    RESTARTS   AGE
kube-system   kube-sriov-device-plugin-amd64-46wpv   1/1     Running   0          4s

Deploy SR-IOV workloads when Multus is used

There are some example Pod specs and related network CRD yaml files in deployments directory for a sample deployment with Multus.

Leave the SR-IOV device plugin running and open a new terminal session for following steps.

Deploy test Pod connecting to pre-created SR-IOV network

$ kubectl create -f pod-tc1.yaml
pod "testpod1" created

$ kubectl get pods
NAME                  READY     STATUS    RESTARTS   AGE
testpod1        	  1/1       Running   0          3s

Verify Pod network interfaces

$ kubectl exec -it testpod1 -- ip addr show

1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet scope host lo
       valid_lft forever preferred_lft forever
3: eth0@if17511: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1450 qdisc noqueue state UP
    link/ether 0a:58:c0:a8:4a:b1 brd ff:ff:ff:ff:ff:ff link-netnsid 0
    inet scope global eth0
       valid_lft forever preferred_lft forever
17508: net0: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc mq state DOWN qlen 1000
    link/ether ce:d8:06:08:e6:3f brd ff:ff:ff:ff:ff:ff
    inet scope global net0
       valid_lft forever preferred_lft forever

Verify Pod routing table

$ kubectl exec -it testpod1 -- route -n

Kernel IP routing table
Destination     Gateway         Genmask         Flags Metric Ref    Use Iface         UG    0      0        0 eth0   U     0      0        0 net0     UG    0      0        0 eth0   U     0      0        0 eth0

Deploy SR-IOV workloads when DANM is used

Verify the existence of the example SR-IOV networks

[cloudadmin@controller-1 ~]$ kubectl get dnet -n example-sriov
NAME         AGE
management   6s
sriov-a      14m
sriov-b      13m

Connect your networks to existing SR-IOV Device Pools

The Spec.Options.device_pool mandatory parameter denotes the Device Pool used by the network. Make sure this parameter is set to the name(s) of your existing SR-IOV Device Pool(s)!

[cloudadmin@controller-1 ~]$ kubectl describe node | grep -A8 Allocatable
 cpu:                          6
 ephemeral-storage:            50189Mi
 hugepages-1Gi:                0
 hugepages-2Mi:                0
 memory:                       249150992Ki  16     32k    32

[cloudadmin@controller-1 ~]$ kubectl describe dnet sriov-a -n example-sriov | grep device_pool
[cloudadmin@controller-1 ~]$ kubectl describe dnet sriov-b -n example-sriov | grep device_pool

Deploy demo Pod connecting to pre-created SR-IOV networks

First, make sure that your Pod asks appropriate number of Devices from the right Device Pools:

[cloudadmin@controller-1 ~]$ grep -B1 sriov_ sriov_pod.yaml
      requests: '2'
      limits: '2'

Then instantiate the Pod:

[cloudadmin@controller-1 ~]$ kubectl create -f sriov_pod.yaml
pod/sriov-pod created

Verify status and the network connections of the demo Pod

[cloudadmin@controller-1 ~]$ kubectl get pod sriov-pod -n example-sriov
sriov-pod   1/1     Running   0          111s

[cloudadmin@controller-1 ~]$ kubectl exec -n example-sriov -it sriov-pod -- ip addr show
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
3: eth0@if49: <BROADCAST,MULTICAST,UP,LOWER_UP,M-DOWN> mtu 8950 qdisc noqueue
    link/ether 8a:74:fd:e0:ee:fa brd ff:ff:ff:ff:ff:ff
    inet brd scope global eth0
       valid_lft forever preferred_lft forever
9: second_path2: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq qlen 1000
    link/ether e2:19:e0:1b:91:44 brd ff:ff:ff:ff:ff:ff
26: first_path1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq qlen 1000
    link/ether 7e:0d:fa:eb:83:8c brd ff:ff:ff:ff:ff:ff

Pod device information

The allocated device information is exported in Container's environment variable. The variable name is PCIDEVICE_ appended with full extended resource name (e.g. etc.) which is capitailzed and any special characters (".", "/") are replaced with underscore ("_"). In case of multiple devices from same extended resource pool, the device IDs are delimited with commas (",").

For example, if 2 devices are allocated from extended resource then the allocated device information will be found in following env variable: PCIDEVICE_INTEL_COM_SRIOV=0000:03:02.1,0000:03:04.3

Virtual Deployments Support

Configure Device Plugin extended selectors in virtual environments

SR-IOV network device plugin supports running in a virtualized environment. However, not all device selectors are applicable as the VFs are passthrough to the VM without any association to their respective PF, hence any device selector that relies on the association between a VF and its PF will not work and therefore the pfNames and rootDevices extended selectors will not work in a virtual deployment. The common selector pciAddress can be used to select the virtual device.

CNI plugins in virtual environments

SR-IOV CNI plugin doesn't support running in a virtualized environment since it always requires accessing to PF device which usually doesn't exist in VM. The recommended CNI plugin to use in a virtualized environment is the host-device CNI plugin.

Virtual environments with no iommu

SR-IOV network device plugin supports allocating VFIO devices in a virtualized environment without a virtualized iommu. For more information refer to this.

Multi Architecture Support

The supported architectures:

  • AMD64
  • PPC64LE

Buiding image for AMD64:

$ DOCKERFILE=Dockerfile make image 

Buiding image for PPC64LE:

$ DOCKERFILE=images/Dockerfile.ppc64le TAG=nfvpe/sriov-device-plugin:ppc64le make image        

Issues and Contributing

We welcome your feedback and contributions to this project. Please see the for contribution guidelines.

Copyright 2018 © Intel Corporation.