15. I40E Poll Mode Driver

The I40E PMD (librte_pmd_i40e) provides poll mode driver support for the Intel X710/XL710/X722 10/40 Gbps family of adapters.

15.1. Features

Features of the I40E PMD are:

  • Multiple queues for TX and RX
  • Receiver Side Scaling (RSS)
  • MAC/VLAN filtering
  • Packet type information
  • Flow director
  • Cloud filter
  • Checksum offload
  • VLAN/QinQ stripping and inserting
  • TSO offload
  • Promiscuous mode
  • Multicast mode
  • Port hardware statistics
  • Jumbo frames
  • Link state information
  • Link flow control
  • Mirror on port, VLAN and VSI
  • Interrupt mode for RX
  • Scattered and gather for TX and RX
  • Vector Poll mode driver
  • DCB
  • VMDQ
  • SR-IOV VF
  • Hot plug
  • IEEE1588/802.1AS timestamping
  • VF Daemon (VFD) - EXPERIMENTAL

15.2. Prerequisites

15.3. Pre-Installation Configuration

15.3.1. 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_I40E_PMD (default y)

    Toggle compilation of the librte_pmd_i40e driver.

  • CONFIG_RTE_LIBRTE_I40E_DEBUG_* (default n)

    Toggle display of generic debugging messages.

  • CONFIG_RTE_LIBRTE_I40E_RX_ALLOW_BULK_ALLOC (default y)

    Toggle bulk allocation for RX.

  • CONFIG_RTE_LIBRTE_I40E_INC_VECTOR (default n)

    Toggle the use of Vector PMD instead of normal RX/TX path. To enable vPMD for RX, bulk allocation for Rx must be allowed.

  • CONFIG_RTE_LIBRTE_I40E_16BYTE_RX_DESC (default n)

    Toggle to use a 16-byte RX descriptor, by default the RX descriptor is 32 byte.

  • CONFIG_RTE_LIBRTE_I40E_QUEUE_NUM_PER_PF (default 64)

    Number of queues reserved for PF.

  • CONFIG_RTE_LIBRTE_I40E_QUEUE_NUM_PER_VF (default 4)

    Number of queues reserved for each SR-IOV VF.

  • CONFIG_RTE_LIBRTE_I40E_QUEUE_NUM_PER_VM (default 4)

    Number of queues reserved for each VMDQ Pool.

  • CONFIG_RTE_LIBRTE_I40E_ITR_INTERVAL (default -1)

    Interrupt Throttling interval.

15.4. Driver compilation and testing

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

15.5. SR-IOV: Prerequisites and sample Application Notes

  1. Load the kernel module:

    modprobe i40e
    

    Check the output in dmesg:

    i40e 0000:83:00.1 ens802f0: renamed from eth0
    
  2. Bring up the PF ports:

    ifconfig ens802f0 up
    
  3. Create VF device(s):

    Echo the number of VFs to be created into the sriov_numvfs sysfs entry of the parent PF.

    Example:

    echo 2 > /sys/devices/pci0000:00/0000:00:03.0/0000:81:00.0/sriov_numvfs
    
  4. Assign VF MAC address:

    Assign MAC address to the VF using iproute2 utility. The syntax is:

    ip link set <PF netdev id> vf <VF id> mac <macaddr>
    

    Example:

    ip link set ens802f0 vf 0 mac a0:b0:c0:d0:e0:f0
    
  5. Assign VF to VM, and bring up the VM. Please see the documentation for the I40E/IXGBE/IGB Virtual Function Driver.

  6. Running testpmd:

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

    Example output:

    ...
    EAL: PCI device 0000:83:00.0 on NUMA socket 1
    EAL: probe driver: 8086:1572 rte_i40e_pmd
    EAL: PCI memory mapped at 0x7f7f80000000
    EAL: PCI memory mapped at 0x7f7f80800000
    PMD: eth_i40e_dev_init(): FW 5.0 API 1.5 NVM 05.00.02 eetrack 8000208a
    Interactive-mode selected
    Configuring Port 0 (socket 0)
    ...
    
    PMD: i40e_dev_rx_queue_setup(): Rx Burst Bulk Alloc Preconditions are
    satisfied.Rx Burst Bulk Alloc function will be used on port=0, queue=0.
    
    ...
    Port 0: 68:05:CA:26:85:84
    Checking link statuses...
    Port 0 Link Up - speed 10000 Mbps - full-duplex
    Done
    
    testpmd>
    

15.6. Sample Application Notes

15.6.1. Vlan filter

Vlan filter only works when Promiscuous mode is off.

To start testpmd, and add vlan 10 to port 0:

./app/testpmd -l 0-15 -n 4 -- -i --forward-mode=mac
...

testpmd> set promisc 0 off
testpmd> rx_vlan add 10 0

15.6.2. Flow Director

The Flow Director works in receive mode to identify specific flows or sets of flows and route them to specific queues. The Flow Director filters can match the different fields for different type of packet: flow type, specific input set per flow type and the flexible payload.

The default input set of each flow type is:

ipv4-other : src_ip_address, dst_ip_address
ipv4-frag  : src_ip_address, dst_ip_address
ipv4-tcp   : src_ip_address, dst_ip_address, src_port, dst_port
ipv4-udp   : src_ip_address, dst_ip_address, src_port, dst_port
ipv4-sctp  : src_ip_address, dst_ip_address, src_port, dst_port,
             verification_tag
ipv6-other : src_ip_address, dst_ip_address
ipv6-frag  : src_ip_address, dst_ip_address
ipv6-tcp   : src_ip_address, dst_ip_address, src_port, dst_port
ipv6-udp   : src_ip_address, dst_ip_address, src_port, dst_port
ipv6-sctp  : src_ip_address, dst_ip_address, src_port, dst_port,
             verification_tag
l2_payload : ether_type

The flex payload is selected from offset 0 to 15 of packet’s payload by default, while it is masked out from matching.

Start testpmd with --disable-rss and --pkt-filter-mode=perfect:

./app/testpmd -l 0-15 -n 4 -- -i --disable-rss --pkt-filter-mode=perfect \
              --rxq=8 --txq=8 --nb-cores=8 --nb-ports=1

Add a rule to direct ipv4-udp packet whose dst_ip=2.2.2.5, src_ip=2.2.2.3, src_port=32, dst_port=32 to queue 1:

testpmd> flow_director_filter 0 mode IP add flow ipv4-udp  \
         src 2.2.2.3 32 dst 2.2.2.5 32 vlan 0 flexbytes () \
         fwd pf queue 1 fd_id 1

Check the flow director status:

testpmd> show port fdir 0

######################## FDIR infos for port 0      ####################
  MODE:   PERFECT
  SUPPORTED FLOW TYPE:  ipv4-frag ipv4-tcp ipv4-udp ipv4-sctp ipv4-other
                        ipv6-frag ipv6-tcp ipv6-udp ipv6-sctp ipv6-other
                        l2_payload
  FLEX PAYLOAD INFO:
  max_len:       16          payload_limit: 480
  payload_unit:  2           payload_seg:   3
  bitmask_unit:  2           bitmask_num:   2
  MASK:
    vlan_tci: 0x0000,
    src_ipv4: 0x00000000,
    dst_ipv4: 0x00000000,
    src_port: 0x0000,
    dst_port: 0x0000
    src_ipv6: 0x00000000,0x00000000,0x00000000,0x00000000,
    dst_ipv6: 0x00000000,0x00000000,0x00000000,0x00000000
  FLEX PAYLOAD SRC OFFSET:
    L2_PAYLOAD:    0      1      2      3      4      5      6  ...
    L3_PAYLOAD:    0      1      2      3      4      5      6  ...
    L4_PAYLOAD:    0      1      2      3      4      5      6  ...
  FLEX MASK CFG:
    ipv4-udp:    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ipv4-tcp:    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ipv4-sctp:   00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ipv4-other:  00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ipv4-frag:   00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ipv6-udp:    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ipv6-tcp:    00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ipv6-sctp:   00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ipv6-other:  00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    ipv6-frag:   00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
    l2_payload:  00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
  guarant_count: 1           best_count:    0
  guarant_space: 512         best_space:    7168
  collision:     0           free:          0
  maxhash:       0           maxlen:        0
  add:           0           remove:        0
  f_add:         0           f_remove:      0

Delete all flow director rules on a port:

testpmd> flush_flow_director 0

15.6.3. Floating VEB

The IntelĀ® Ethernet Controller X710 and XL710 Family support a feature called “Floating VEB”.

A Virtual Ethernet Bridge (VEB) is an IEEE Edge Virtual Bridging (EVB) term for functionality that allows local switching between virtual endpoints within a physical endpoint and also with an external bridge/network.

A “Floating” VEB doesn’t have an uplink connection to the outside world so all switching is done internally and remains within the host. As such, this feature provides security benefits.

In addition, a Floating VEB overcomes a limitation of normal VEBs where they cannot forward packets when the physical link is down. Floating VEBs don’t need to connect to the NIC port so they can still forward traffic from VF to VF even when the physical link is down.

Therefore, with this feature enabled VFs can be limited to communicating with each other but not an outside network, and they can do so even when there is no physical uplink on the associated NIC port.

To enable this feature, the user should pass a devargs parameter to the EAL, for example:

-w 84:00.0,enable_floating_veb=1

In this configuration the PMD will use the floating VEB feature for all the VFs created by this PF device.

Alternatively, the user can specify which VFs need to connect to this floating VEB using the floating_veb_list argument:

-w 84:00.0,enable_floating_veb=1,floating_veb_list=1;3-4

In this example VF1, VF3 and VF4 connect to the floating VEB, while other VFs connect to the normal VEB.

The current implementation only supports one floating VEB and one regular VEB. VFs can connect to a floating VEB or a regular VEB according to the configuration passed on the EAL command line.

The floating VEB functionality requires a NIC firmware version of 5.0 or greater.

15.7. Limitations or Known issues

15.7.1. MPLS packet classification on X710/XL710

For firmware versions prior to 5.0, MPLS packets are not recognized by the NIC. The L2 Payload flow type in flow director can be used to classify MPLS packet by using a command in testpmd like:

testpmd> flow_director_filter 0 mode IP add flow l2_payload ether
0x8847 flexbytes () fwd pf queue <N> fd_id <M>

With the NIC firmware version 5.0 or greater, some limited MPLS support is added: Native MPLS (MPLS in Ethernet) skip is implemented, while no new packet type, no classification or offload are possible. With this change, L2 Payload flow type in flow director cannot be used to classify MPLS packet as with previous firmware versions. Meanwhile, the Ethertype filter can be used to classify MPLS packet by using a command in testpmd like:

testpmd> ethertype_filter 0 add mac_ignr 00:00:00:00:00:00 ethertype
0x8847 fwd queue <M>

15.7.2. 16 Byte RX Descriptor setting on DPDK VF

Currently the VF’s RX descriptor mode is decided by PF. There’s no PF-VF interface for VF to request the RX descriptor mode, also no interface to notify VF its own RX descriptor mode. For all available versions of the i40e driver, these drivers don’t support 16 byte RX descriptor. If the Linux i40e kernel driver is used as host driver, while DPDK i40e PMD is used as the VF driver, DPDK cannot choose 16 byte receive descriptor. The reason is that the RX descriptor is already set to 32 byte by the i40e kernel driver. That is to say, user should keep CONFIG_RTE_LIBRTE_I40E_16BYTE_RX_DESC=n in config file. In the future, if the Linux i40e driver supports 16 byte RX descriptor, user should make sure the DPDK VF uses the same RX descriptor mode, 16 byte or 32 byte, as the PF driver.

The same rule for DPDK PF + DPDK VF. The PF and VF should use the same RX descriptor mode. Or the VF RX will not work.

15.7.3. Receive packets with Ethertype 0x88A8

Due to the FW limitation, PF can receive packets with Ethertype 0x88A8 only when floating VEB is disabled.

15.7.4. Incorrect Rx statistics when packet is oversize

When a packet is over maximum frame size, the packet is dropped. However the Rx statistics, when calling rte_eth_stats_get incorrectly shows it as received.

15.7.5. VF & TC max bandwidth setting

The per VF max bandwidth and per TC max bandwidth cannot be enabled in parallel. The dehavior is different when handling per VF and per TC max bandwidth setting. When enabling per VF max bandwidth, SW will check if per TC max bandwidth is enabled. If so, return failure. When enabling per TC max bandwidth, SW will check if per VF max bandwidth is enabled. If so, disable per VF max bandwidth and continue with per TC max bandwidth setting.

15.7.6. TC TX scheduling mode setting

There’re 2 TX scheduling modes for TCs, round robin and strict priority mode. If a TC is set to strict priority mode, it can consume unlimited bandwidth. It means if APP has set the max bandwidth for that TC, it comes to no effect. It’s suggested to set the strict priority mode for a TC that is latency sensitive but no consuming much bandwidth.

15.7.7. VF performance is impacted by PCI extended tag setting

To reach maximum NIC performance in the VF the PCI extended tag must be enabled. The DPDK I40E PF driver will set this feature during initialization, but the kernel PF driver does not. So when running traffic on a VF which is managed by the kernel PF driver, a significant NIC performance downgrade has been observed (for 64 byte packets, there is about 25% linerate downgrade for a 25G device and about 35% for a 40G device).

For kernel version >= 4.11, the kernel’s PCI driver will enable the extended tag if it detects that the device supports it. So by default, this is not an issue. For kernels <= 4.11 or when the PCI extended tag is disabled it can be enabled using the steps below.

  1. Get the current value of the PCI configure register:

    setpci -s <XX:XX.X> a8.w
    
  2. Set bit 8:

    value = value | 0x100
    
  3. Set the PCI configure register with new value:

    setpci -s <XX:XX.X> a8.w=<value>
    

15.7.8. Vlan strip of VF

The VF vlan strip function is only supported in the i40e kernel driver >= 2.1.26.

15.8. High Performance of Small Packets on 40G NIC

As there might be firmware fixes for performance enhancement in latest version of firmware image, the firmware update might be needed for getting high performance. Check with the local Intel’s Network Division application engineers for firmware updates. Users should consult the release notes specific to a DPDK release to identify the validated firmware version for a NIC using the i40e driver.

15.8.1. Use 16 Bytes RX Descriptor Size

As i40e PMD supports both 16 and 32 bytes RX descriptor sizes, and 16 bytes size can provide helps to high performance of small packets. Configuration of CONFIG_RTE_LIBRTE_I40E_16BYTE_RX_DESC in config files can be changed to use 16 bytes size RX descriptors.

15.8.2. High Performance and per Packet Latency Tradeoff

Due to the hardware design, the interrupt signal inside NIC is needed for per packet descriptor write-back. The minimum interval of interrupts could be set at compile time by CONFIG_RTE_LIBRTE_I40E_ITR_INTERVAL in configuration files. Though there is a default configuration, the interval could be tuned by the users with that configuration item depends on what the user cares about more, performance or per packet latency.

15.9. Example of getting best performance with l3fwd example

The following is an example of running the DPDK l3fwd sample application to get high performance with an Intel server platform and Intel XL710 NICs.

The example scenario is to get best performance with two Intel XL710 40GbE ports. See Fig. 15.1 for the performance test setup.

Fig. 15.1 Performance Test Setup

  1. Add two Intel XL710 NICs to the platform, and use one port per card to get best performance. The reason for using two NICs is to overcome a PCIe Gen3’s limitation since it cannot provide 80G bandwidth for two 40G ports, but two different PCIe Gen3 x8 slot can. Refer to the sample NICs output above, then we can select 82:00.0 and 85:00.0 as test ports:

    82:00.0 Ethernet [0200]: Intel XL710 for 40GbE QSFP+ [8086:1583]
    85:00.0 Ethernet [0200]: Intel XL710 for 40GbE QSFP+ [8086:1583]
    
  2. Connect the ports to the traffic generator. For high speed testing, it’s best to use a hardware traffic generator.

  3. Check the PCI devices numa node (socket id) and get the cores number on the exact socket id. In this case, 82:00.0 and 85:00.0 are both in socket 1, and the cores on socket 1 in the referenced platform are 18-35 and 54-71. Note: Don’t use 2 logical cores on the same core (e.g core18 has 2 logical cores, core18 and core54), instead, use 2 logical cores from different cores (e.g core18 and core19).

  4. Bind these two ports to igb_uio.

  5. As to XL710 40G port, we need at least two queue pairs to achieve best performance, then two queues per port will be required, and each queue pair will need a dedicated CPU core for receiving/transmitting packets.

  6. The DPDK sample application l3fwd will be used for performance testing, with using two ports for bi-directional forwarding. Compile the l3fwd sample with the default lpm mode.

  7. The command line of running l3fwd would be something like the following:

    ./l3fwd -l 18-21 -n 4 -w 82:00.0 -w 85:00.0 \
            -- -p 0x3 --config '(0,0,18),(0,1,19),(1,0,20),(1,1,21)'
    

    This means that the application uses core 18 for port 0, queue pair 0 forwarding, core 19 for port 0, queue pair 1 forwarding, core 20 for port 1, queue pair 0 forwarding, and core 21 for port 1, queue pair 1 forwarding.

  8. Configure the traffic at a traffic generator.

    • Start creating a stream on packet generator.
    • Set the Ethernet II type to 0x0800.