FD.io VPP  v19.04.2-12-g66b1689
Vector Packet Processing
Intel AVF device plugin for VPP

Overview

This plugins provides native device support for intel Adaptive Virtual Function (AVF). AVF is driver specification for current and future Intel Virtual Function devices. AVF defines communication channel between Physical Functions (PF) and VF. In essence, today this driver can be used only with Intel XL710 / X710 / XXV710 adapters.

Prerequisites

  • Driver requires newer i40e PF linux driver to be installed on the system, which supports virtualchnl interface. This code is tested with i40e driver version 2.4.6.

Driver requires MSI-X interrupt support, which is not supported by uio_pci_generic driver, so vfio-pci needs to be used. On systems without IOMMU vfio driver can still be used with recent kernels which support no-iommu mode.

Known issues

This driver is still in experimental phase, however it shows very good performance numbers. Following items are not implemented (yet).

  • Jumbo MTU support
  • Adaptive mode
  • NUMA support

Usage

System setup

  1. load VFIO driver
    1 sudo modprobe vfio-pci
  2. (systems without IOMMU only) enable unsafe NOIOMMU mode
    1 echo Y | sudo tee /sys/module/vfio/parameters/enable_unsafe_noiommu_mode
  3. Create and bind SR-IOV virtual function(s)

Following script creates VF, assigns MAC address and binds VF to vfio-pci

1 #!/bin/bash
2 
3 if [ $USER != "root" ] ; then
4  echo "Restarting script with sudo..."
5  sudo $0 ${*}
6  exit
7 fi
8 
9 setup () {
10  cd /sys/bus/pci/devices/${1}
11  driver=$(basename $(readlink driver))
12  if [ "${driver}" != "i40e" ]; then
13  echo ${1} | tee driver/unbind
14  echo ${1} | tee /sys/bus/pci/drivers/i40e/bind
15  fi
16  ifname=$(basename net/*)
17  echo 0 | tee sriov_numvfs > /dev/null
18  echo 1 | tee sriov_numvfs > /dev/null
19  ip link set dev ${ifname} vf 0 mac ${2}
20  ip link show dev ${ifname}
21  vf=$(basename $(readlink virtfn0))
22  echo ${vf} | tee virtfn0/driver/unbind
23  echo vfio-pci | tee virtfn0/driver_override
24  echo ${vf} | sudo tee /sys/bus/pci/drivers/vfio-pci/bind
25  echo | tee virtfn0/driver_override
26 }
27 
28 # Setup one VF on PF 0000:3b:00.0 and assign MAC address
29 setup 0000:3b:00.0 00:11:22:33:44:00
30 # Setup one VF on PF 0000:3b:00.1 and assign MAC address
31 setup 0000:3b:00.1 00:11:22:33:44:01

Promisc mode

In cases when interface is used in the L2 mode or promisc mode is needed for some other reason, trust needs to be set to "on" using the linux "ip link" utility.

1 ip link set dev <PF inteface name> vf <VF id> trust on

Interface Creation

Interfaces can be dynamically created by using following CLI:

1 create interface avf 0000:3b:02.0
2 set int state avf-0/3b/2/0 up

Interface Deletion

Interface can be deleted with following CLI:

1 delete interface avf <interface name>

Interface Statistics

Interface statistics can be displayed with sh hardware-interface <if-name> command.