29. Solarflare libefx-based Poll Mode Driver

29.3.1. Equal stride super-buffer mode

When the receive queue uses equal stride super-buffer DMA mode, one HW Rx descriptor carries many Rx buffers which contiguously follow each other with some stride (equal to total size of rte_mbuf as mempool object). Each Rx buffer is an independent rte_mbuf. However dedicated mempool manager must be used when mempool for the Rx queue is created. The manager must support dequeue of the contiguous block of objects and provide mempool info API to get the block size.

Another limitation of a equal stride super-buffer mode, imposed by the firmware, is that it allows for a single RSS context.

29.5.1. Ethernet destinaton individual/group match

Ethernet item supports I/G matching, if only the corresponding bit is set in the mask of destination address. If destinaton address in the spec is multicast, it matches all multicast (and broadcast) packets, oherwise it matches unicast packets that are not filtered by other flow rules.

29.5.2. Exceptions to flow rules

There is a list of exceptional flow rule patterns which will not be accepted by the PMD. A pattern will be rejected if at least one of the conditions is met:

• Filtering by IPv4 or IPv6 EtherType without pattern items of internet layer and above.
• The last item is IPV4 or IPV6, and it’s empty.
• Filtering by TCP or UDP IP transport protocol without pattern items of transport layer and above.
• The last item is TCP or UDP, and it’s empty.

29.8.1. 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_SFC_EFX_PMD (default y)

  Enable compilation of Solarflare libefx-based poll-mode driver.

• CONFIG_RTE_LIBRTE_SFC_EFX_DEBUG (default n)

  Enable compilation of the extra run-time consistency checks.

29.8.2. Per-Device Parameters

The following per-device parameters can be passed via EAL PCI device whitelist option like “-w 02:00.0,arg1=value1,...”.

Case-insensitive 1/y/yes/on or 0/n/no/off may be used to specify boolean parameters value.

• rx_datapath [auto|efx|ef10|ef10_esps] (default auto)

  Choose receive datapath implementation. auto allows the driver itself to make a choice based on firmware features available and required by the datapath implementation. efx chooses libefx-based datapath which supports Rx scatter. ef10 chooses EF10 (SFN7xxx, SFN8xxx, X2xxx) native datapath which is more efficient than libefx-based and provides richer packet type classification, but lacks Rx scatter support. ef10_esps chooses SFNX2xxx equal stride packed stream datapath which may be used on DPDK firmware variant only (see notes about its limitations above).

• tx_datapath [auto|efx|ef10|ef10_simple] (default auto)

  Choose transmit datapath implementation. auto allows the driver itself to make a choice based on firmware features available and required by the datapath implementation. efx chooses libefx-based datapath which supports VLAN insertion (full-feature firmware variant only), TSO and multi-segment mbufs. Mbuf segments may come from different mempools, and mbuf reference counters are treated responsibly. ef10 chooses EF10 (SFN7xxx, SFN8xxx, X2xxx) native datapath which is more efficient than libefx-based but has no VLAN insertion and TSO support yet. Mbuf segments may come from different mempools, and mbuf reference counters are treated responsibly. ef10_simple chooses EF10 (SFN7xxx, SFN8xxx, X2xxx) native datapath which is even more faster then ef10 but does not support multi-segment mbufs, disallows multiple mempools and neglects mbuf reference counters.

• perf_profile [auto|throughput|low-latency] (default throughput)

  Choose hardware tunning to be optimized for either throughput or low-latency. auto allows NIC firmware to make a choice based on installed licences and firmware variant configured using sfboot.

• stats_update_period_ms [long] (default 1000)

  Adjust period in milliseconds to update port hardware statistics. The accepted range is 0 to 65535. The value of 0 may be used to disable periodic statistics update. One should note that it’s only possible to set an arbitrary value on SFN8xxx and X2xxx provided that firmware version is 6.2.1.1033 or higher, otherwise any positive value will select a fixed update period of 1000 milliseconds

• fw_variant [dont-care|full-feature|ultra-low-latency| capture-packed-stream|dpdk] (default dont-care)

  Choose the preferred firmware variant to use. In order for the selected option to have an effect, the sfboot utility must be configured with the auto firmware-variant option. The preferred firmware variant applies to all ports on the NIC. dont-care ensures that the driver can attach to an unprivileged function. The datapath firmware type to use is controlled by the sfboot utility. full-feature chooses full featured firmware. ultra-low-latency chooses firmware with fewer features but lower latency. capture-packed-stream chooses firmware for SolarCapture packed stream mode. dpdk chooses DPDK firmware with equal stride super-buffer Rx mode for higher Rx packet rate and packet marks support and firmware subvariant without checksumming on transmit for higher Tx packet rate if checksumming is not required.

• rxd_wait_timeout_ns [long] (default 200 us)

  Adjust timeout in nanoseconds to head-of-line block to wait for Rx descriptors. The accepted range is 0 to 400 ms. Flow control should be enabled to make it work. The value of 0 disables it and packets are dropped immediately. When a packet is dropped because of no Rx descriptors, rx_nodesc_drop_cnt counter grows. The feature is supported only by the DPDK firmware variant when equal stride super-buffer Rx mode is used.

29.8.3. Dynamic Logging Parameters

One may leverage EAL option “–log-level” to change default levels for the log types supported by the driver. The option is used with an argument typically consisting of two parts separated by a colon.

Level value is the last part which takes a symbolic name (or integer). Log type is the former part which may shell match syntax. Depending on the choice of the expression, the given log level may be used either for some specific log type or for a subset of types.

SFC EFX PMD provides the following log types available for control:

• pmd.net.sfc.driver (default level is notice)

  Affects driver-wide messages unrelated to any particular devices.

• pmd.net.sfc.main (default level is notice)

  Matches a subset of per-port log types registered during runtime. A full name for a particular type may be obtained by appending a dot and a PCI device identifier (XXXX:XX:XX.X) to the prefix.

• pmd.net.sfc.mcdi (default level is notice)

  Extra logging of the communication with the NIC’s management CPU. The format of the log is consumed by the Solarflare netlogdecode cross-platform tool. May be managed per-port, as explained above.