32. ThunderX NICVF Poll Mode Driver

The ThunderX NICVF PMD (librte_pmd_thunderx_nicvf) provides poll mode driver support for the inbuilt NIC found in the Cavium ThunderX SoC family as well as their virtual functions (VF) in SR-IOV context.

More information can be found at Cavium, Inc Official Website.

32.1. Features

Features of the ThunderX PMD are:

  • Multiple queues for TX and RX
  • Receive Side Scaling (RSS)
  • Packet type information
  • Checksum offload
  • Promiscuous mode
  • Multicast mode
  • Port hardware statistics
  • Jumbo frames
  • Link state information
  • Scattered and gather for TX and RX
  • VLAN stripping
  • SR-IOV VF
  • NUMA support
  • Multi queue set support (up to 96 queues (12 queue sets)) per port

32.2. Supported ThunderX SoCs

  • CN88xx
  • CN81xx
  • CN83xx

32.3. Prerequisites

32.4. Pre-Installation Configuration

32.4.1. Config File Options

The following options can be modified in the config file. Please note that enabling debugging options may affect system performance.

  • CONFIG_RTE_LIBRTE_THUNDERX_NICVF_PMD (default y)

    Toggle compilation of the librte_pmd_thunderx_nicvf driver.

  • CONFIG_RTE_LIBRTE_THUNDERX_NICVF_DEBUG_RX (default n)

    Toggle asserts of receive fast path.

  • CONFIG_RTE_LIBRTE_THUNDERX_NICVF_DEBUG_TX (default n)

    Toggle asserts of transmit fast path.

32.5. Driver compilation and testing

Refer to the document compiling and testing a PMD for a NIC for details.

To compile the ThunderX NICVF PMD for Linux arm64 gcc, use arm64-thunderx-linuxapp-gcc as target.

32.6. Linux

32.6.1. SR-IOV: Prerequisites and sample Application Notes

Current ThunderX NIC PF/VF kernel modules maps each physical Ethernet port automatically to virtual function (VF) and presented them as PCIe-like SR-IOV device. This section provides instructions to configure SR-IOV with Linux OS.

  1. Verify PF devices capabilities using lspci:

    lspci -vvv
    

    Example output:

    0002:01:00.0 Ethernet controller: Cavium Networks Device a01e (rev 01)
    ...
    Capabilities: [100 v1] Alternative Routing-ID Interpretation (ARI)
    ...
    Capabilities: [180 v1] Single Root I/O Virtualization (SR-IOV)
    ...
    Kernel driver in use: thunder-nic
    ...
    

    Note

    Unless thunder-nic driver is in use make sure your kernel config includes CONFIG_THUNDER_NIC_PF setting.

  2. Verify VF devices capabilities and drivers using lspci:

    lspci -vvv
    

    Example output:

    0002:01:00.1 Ethernet controller: Cavium Networks Device 0011 (rev 01)
    ...
    Capabilities: [100 v1] Alternative Routing-ID Interpretation (ARI)
    ...
    Kernel driver in use: thunder-nicvf
    ...
    
    0002:01:00.2 Ethernet controller: Cavium Networks Device 0011 (rev 01)
    ...
    Capabilities: [100 v1] Alternative Routing-ID Interpretation (ARI)
    ...
    Kernel driver in use: thunder-nicvf
    ...
    

    Note

    Unless thunder-nicvf driver is in use make sure your kernel config includes CONFIG_THUNDER_NIC_VF setting.

  3. Pass VF device to VM context (PCIe Passthrough):

    The VF devices may be passed through to the guest VM using qemu or virt-manager or virsh etc.

    Example qemu guest launch command:

    sudo qemu-system-aarch64 -name vm1 \
    -machine virt,gic_version=3,accel=kvm,usb=off \
    -cpu host -m 4096 \
    -smp 4,sockets=1,cores=8,threads=1 \
    -nographic -nodefaults \
    -kernel <kernel image> \
    -append "root=/dev/vda console=ttyAMA0 rw hugepagesz=512M hugepages=3" \
    -device vfio-pci,host=0002:01:00.1 \
    -drive file=<rootfs.ext3>,if=none,id=disk1,format=raw  \
    -device virtio-blk-device,scsi=off,drive=disk1,id=virtio-disk1,bootindex=1 \
    -netdev tap,id=net0,ifname=tap0,script=/etc/qemu-ifup_thunder \
    -device virtio-net-device,netdev=net0 \
    -serial stdio \
    -mem-path /dev/huge
    
  4. Enable VFIO-NOIOMMU mode (optional):

    echo 1 > /sys/module/vfio/parameters/enable_unsafe_noiommu_mode
    

    Note

    VFIO-NOIOMMU is required only when running in VM context and should not be enabled otherwise.

  5. Running testpmd:

    Follow instructions available in the document compiling and testing a PMD for a NIC to run testpmd.

    Example output:

    ./arm64-thunderx-linuxapp-gcc/app/testpmd -l 0-3 -n 4 -w 0002:01:00.2 \
      -- -i --no-flush-rx \
      --port-topology=loop
    
    ...
    
    PMD: rte_nicvf_pmd_init(): librte_pmd_thunderx nicvf version 1.0
    
    ...
    EAL:   probe driver: 177d:11 rte_nicvf_pmd
    EAL:   using IOMMU type 1 (Type 1)
    EAL:   PCI memory mapped at 0x3ffade50000
    EAL: Trying to map BAR 4 that contains the MSI-X table.
         Trying offsets: 0x40000000000:0x0000, 0x10000:0x1f0000
    EAL:   PCI memory mapped at 0x3ffadc60000
    PMD: nicvf_eth_dev_init(): nicvf: device (177d:11) 2:1:0:2
    PMD: nicvf_eth_dev_init(): node=0 vf=1 mode=tns-bypass sqs=false
         loopback_supported=true
    PMD: nicvf_eth_dev_init(): Port 0 (177d:11) mac=a6:c6:d9:17:78:01
    Interactive-mode selected
    Configuring Port 0 (socket 0)
    ...
    
    PMD: nicvf_dev_configure(): Configured ethdev port0 hwcap=0x0
    Port 0: A6:C6:D9:17:78:01
    Checking link statuses...
    Port 0 Link Up - speed 10000 Mbps - full-duplex
    Done
    testpmd>
    

32.6.2. Multiple Queue Set per DPDK port configuration

There are two types of VFs:

  • Primary VF
  • Secondary VF

Each port consists of a primary VF and n secondary VF(s). Each VF provides 8 Tx/Rx queues to a port. When a given port is configured to use more than 8 queues, it requires one (or more) secondary VF. Each secondary VF adds 8 additional queues to the queue set.

During PMD driver initialization, the primary VF’s are enumerated by checking the specific flag (see sqs message in DPDK boot log - sqs indicates secondary queue set). They are at the beginning of VF list (the remain ones are secondary VF’s).

The primary VFs are used as master queue sets. Secondary VFs provide additional queue sets for primary ones. If a port is configured for more then 8 queues than it will request for additional queues from secondary VFs.

Secondary VFs cannot be shared between primary VFs.

Primary VFs are present on the beginning of the ‘Network devices using kernel driver’ list, secondary VFs are on the remaining on the remaining part of the list.

Note

The VNIC driver in the multiqueue setup works differently than other drivers like ixgbe. We need to bind separately each specific queue set device with the usertools/dpdk-devbind.py utility.

Note

Depending on the hardware used, the kernel driver sets a threshold vf_id. VFs that try to attached with an id below or equal to this boundary are considered primary VFs. VFs that try to attach with an id above this boundary are considered secondary VFs.

32.6.3. Example device binding

If a system has three interfaces, a total of 18 VF devices will be created on a non-NUMA machine.

Note

NUMA systems have 12 VFs per port and non-NUMA 6 VFs per port.

# usertools/dpdk-devbind.py --status

Network devices using DPDK-compatible driver
============================================
<none>

Network devices using kernel driver
===================================
0000:01:10.0 'Device a026' if= drv=thunder-BGX unused=vfio-pci,uio_pci_generic
0000:01:10.1 'Device a026' if= drv=thunder-BGX unused=vfio-pci,uio_pci_generic
0002:01:00.0 'Device a01e' if= drv=thunder-nic unused=vfio-pci,uio_pci_generic
0002:01:00.1 'Device 0011' if=eth0 drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:00.2 'Device 0011' if=eth1 drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:00.3 'Device 0011' if=eth2 drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:00.4 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:00.5 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:00.6 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:00.7 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:01.0 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:01.1 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:01.2 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:01.3 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:01.4 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:01.5 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:01.6 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:01.7 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:02.0 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:02.1 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic
0002:01:02.2 'Device 0011' if= drv=thunder-nicvf unused=vfio-pci,uio_pci_generic

Other network devices
=====================
0002:00:03.0 'Device a01f' unused=vfio-pci,uio_pci_generic

We want to bind two physical interfaces with 24 queues each device, we attach two primary VFs and four secondary queues. In our example we choose two 10G interfaces eth1 (0002:01:00.2) and eth2 (0002:01:00.3). We will choose four secondary queue sets from the ending of the list (0002:01:01.7-0002:01:02.2).

  1. Bind two primary VFs to the vfio-pci driver:

    usertools/dpdk-devbind.py -b vfio-pci 0002:01:00.2
    usertools/dpdk-devbind.py -b vfio-pci 0002:01:00.3
    
  2. Bind four primary VFs to the vfio-pci driver:

    usertools/dpdk-devbind.py -b vfio-pci 0002:01:01.7
    usertools/dpdk-devbind.py -b vfio-pci 0002:01:02.0
    usertools/dpdk-devbind.py -b vfio-pci 0002:01:02.1
    usertools/dpdk-devbind.py -b vfio-pci 0002:01:02.2
    

The nicvf thunderx driver will make use of attached secondary VFs automatically during the interface configuration stage.

32.7. Limitations

32.7.1. CRC striping

The ThunderX SoC family NICs strip the CRC for every packets coming into the host interface irrespective of the offload configuration.

32.7.2. Maximum packet length

The ThunderX SoC family NICs support a maximum of a 9K jumbo frame. The value is fixed and cannot be changed. So, even when the rxmode.max_rx_pkt_len member of struct rte_eth_conf is set to a value lower than 9200, frames up to 9200 bytes can still reach the host interface.

32.7.3. Maximum packet segments

The ThunderX SoC family NICs support up to 12 segments per packet when working in scatter/gather mode. So, setting MTU will result with EINVAL when the frame size does not fit in the maximum number of segments.