1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright(c) 2019-2021 Xilinx, Inc.
3 Copyright(c) 2016-2019 Solarflare Communications Inc.
5 This software was jointly developed between OKTET Labs (under contract
6 for Solarflare) and Solarflare Communications, Inc.
8 Solarflare libefx-based Poll Mode Driver
9 ========================================
11 The SFC EFX PMD (**librte_net_sfc_efx**) provides poll mode driver support
12 for **Solarflare SFN7xxx and SFN8xxx** family of 10/40 Gbps adapters,
13 **Solarflare XtremeScale X2xxx** family of 10/25/40/50/100 Gbps adapters and
14 **Alveo SN1000 SmartNICs** family of 10/25/40/50/100 Gbps adapters.
15 SFC EFX PMD has support for the latest Linux and FreeBSD operating systems.
17 More information can be found at `Solarflare Communications website
18 <http://solarflare.com>`_.
24 SFC EFX PMD has support for:
26 - Multiple transmit and receive queues
28 - Link state information including link status change interrupt
30 - IPv4/IPv6 TCP/UDP transmit checksum offload
32 - Inner IPv4/IPv6 TCP/UDP transmit checksum offload
34 - Port hardware statistics
36 - Extended statistics (see Solarflare Server Adapter User's Guide for
37 the statistics description)
43 - Jumbo frames up to 9K
49 - TCP segmentation offload (TSO) including VXLAN and GENEVE encapsulated
51 - Multicast MAC filter
53 - IPv4/IPv6 TCP/UDP receive checksum offload
55 - Inner IPv4/IPv6 TCP/UDP receive checksum offload
57 - Received packet type information
59 - Receive side scaling (RSS)
63 - Scattered Rx DMA for packet that are larger that a single Rx descriptor
65 - Receive queue interrupts
67 - Deferred receive and transmit queue start
69 - Transmit VLAN insertion (if running firmware variant supports it)
77 - Port representors (see :ref: switch_representation)
80 Non-supported Features
81 ----------------------
83 The features not yet supported include:
85 - Priority-based flow control
87 - Configurable RX CRC stripping (always stripped)
89 - Header split on receive
101 Due to requirements on receive buffer alignment and usage of the receive
102 buffer for the auxiliary packet information provided by the NIC up to
103 extra 269 (14 bytes prefix plus up to 255 bytes for end padding) bytes may be
104 required in the receive buffer.
105 It should be taken into account when mbuf pool for receive is created.
108 Equal stride super-buffer mode
109 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
111 When the receive queue uses equal stride super-buffer DMA mode, one HW Rx
112 descriptor carries many Rx buffers which contiguously follow each other
113 with some stride (equal to total size of rte_mbuf as mempool object).
114 Each Rx buffer is an independent rte_mbuf.
115 However dedicated mempool manager must be used when mempool for the Rx
116 queue is created. The manager must support dequeue of the contiguous
117 block of objects and provide mempool info API to get the block size.
119 Another limitation of a equal stride super-buffer mode, imposed by the
120 firmware, is that it allows for a single RSS context.
126 NVGRE, VXLAN and GENEVE tunnels are supported on SFN8xxx and X2xxx family
127 adapters with full-feature firmware variant running.
128 **sfboot** should be used to configure NIC to run full-feature firmware variant.
129 See Solarflare Server Adapter User's Guide for details.
131 SFN8xxx and X2xxx family adapters provide either inner or outer packet classes.
132 If adapter firmware advertises support for tunnels then the PMD
133 configures the hardware to report inner classes, and outer classes are
134 not reported in received packets.
135 However, for VXLAN and GENEVE tunnels the PMD does report UDP as the
136 outer layer 4 packet type.
138 SFN8xxx and X2xxx family adapters report GENEVE packets as VXLAN.
139 If UDP ports are configured for only one tunnel type then it is safe to
140 treat VXLAN packet type indication as the corresponding UDP tunnel type.
146 Supported attributes:
151 Supported pattern items (***non-transfer*** rules):
155 - ETH (exact match of source/destination addresses, individual/group match
156 of destination address, EtherType in the outer frame and exact match of
157 destination addresses, individual/group match of destination address in
160 - VLAN (exact match of VID, double-tagging is supported)
162 - IPV4 (exact match of source/destination addresses,
163 IP transport protocol)
165 - IPV6 (exact match of source/destination addresses,
166 IP transport protocol)
168 - TCP (exact match of source/destination ports)
170 - UDP (exact match of source/destination ports)
172 - VXLAN (exact match of VXLAN network identifier)
174 - GENEVE (exact match of virtual network identifier, only Ethernet (0x6558)
175 protocol type is supported)
177 - NVGRE (exact match of virtual subnet ID)
179 Supported actions (***non-transfer*** rules):
189 - FLAG (supported only with ef10_essb Rx datapath)
191 - MARK (supported only with ef10_essb Rx datapath)
193 Supported pattern items (***transfer*** rules):
195 - PORT_REPRESENTOR (cannot repeat; conflicts with other traffic source items)
197 - PORT_ID (cannot repeat; conflicts with other traffic source items)
199 - PHY_PORT (cannot repeat; conflicts with other traffic source items)
201 - PF (cannot repeat; conflicts with other traffic source items)
203 - VF (cannot repeat; conflicts with other traffic source items)
207 - VLAN (double-tagging is supported)
209 - IPV4 (source/destination addresses, IP transport protocol,
210 type of service, time to live)
212 - IPV6 (source/destination addresses, IP transport protocol,
213 traffic class, hop limit)
215 - TCP (source/destination ports, TCP header length + TCP flags)
217 - UDP (source/destination ports)
219 - VXLAN (exact match of VXLAN network identifier)
221 - GENEVE (exact match of virtual network identifier)
223 - NVGRE (exact match of virtual subnet ID)
225 Supported actions (***transfer*** rules):
255 Validating flow rules depends on the firmware variant.
257 The :ref:`flow_isolated_mode` is supported.
259 The implementation is natively thread-safe.
261 Ethernet destination individual/group match
262 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
264 Ethernet item supports I/G matching, if only the corresponding bit is set
265 in the mask of destination address. If destination address in the spec is
266 multicast, it matches all multicast (and broadcast) packets, otherwise it
267 matches unicast packets that are not filtered by other flow rules.
269 Exceptions to flow rules
270 ~~~~~~~~~~~~~~~~~~~~~~~~
272 There is a list of exceptional flow rule patterns which will not be
273 accepted by the PMD. A pattern will be rejected if at least one of the
276 - Filtering by IPv4 or IPv6 EtherType without pattern items of internet
279 - The last item is IPV4 or IPV6, and it's empty.
281 - Filtering by TCP or UDP IP transport protocol without pattern items of
282 transport layer and above.
284 - The last item is TCP or UDP, and it's empty.
292 - Alveo SN1022 SmartNIC
294 - Solarflare XtremeScale Adapters:
296 - Solarflare X2522 Dual Port SFP28 10/25GbE Adapter
298 - Solarflare X2541 Single Port QSFP28 10/25G/100G Adapter
300 - Solarflare X2542 Dual Port QSFP28 10/25G/100G Adapter
302 - Solarflare Flareon [Ultra] Server Adapters:
304 - Solarflare SFN8522 Dual Port SFP+ Server Adapter
306 - Solarflare SFN8522M Dual Port SFP+ Server Adapter
308 - Solarflare SFN8042 Dual Port QSFP+ Server Adapter
310 - Solarflare SFN8542 Dual Port QSFP+ Server Adapter
312 - Solarflare SFN8722 Dual Port SFP+ OCP Server Adapter
314 - Solarflare SFN7002F Dual Port SFP+ Server Adapter
316 - Solarflare SFN7004F Quad Port SFP+ Server Adapter
318 - Solarflare SFN7042Q Dual Port QSFP+ Server Adapter
320 - Solarflare SFN7122F Dual Port SFP+ Server Adapter
322 - Solarflare SFN7124F Quad Port SFP+ Server Adapter
324 - Solarflare SFN7142Q Dual Port QSFP+ Server Adapter
326 - Solarflare SFN7322F Precision Time Synchronization Server Adapter
332 - Requires firmware version:
334 - SFN7xxx: **4.7.1.1001** or higher
336 - SFN8xxx: **6.0.2.1004** or higher
338 Visit `Solarflare Support Downloads <https://support.solarflare.com>`_ to get
339 Solarflare Utilities (either Linux or FreeBSD) with the latest firmware.
340 Follow instructions from Solarflare Server Adapter User's Guide to
341 update firmware and configure the adapter.
344 Pre-Installation Configuration
345 ------------------------------
351 The following build-time options may be enabled on build time using
352 ``-Dc_args=`` meson argument (e.g. ``-Dc_args=-DRTE_LIBRTE_SFC_EFX_DEBUG``).
354 Please note that enabling debugging options may affect system performance.
356 - ``RTE_LIBRTE_SFC_EFX_DEBUG`` (undefined by default)
358 Enable compilation of the extra run-time consistency checks.
361 Per-Device Parameters
362 ~~~~~~~~~~~~~~~~~~~~~
364 The following per-device parameters can be passed via EAL PCI device
365 allow option like "-a 02:00.0,arg1=value1,...".
367 Case-insensitive 1/y/yes/on or 0/n/no/off may be used to specify
368 boolean parameters value.
370 - ``class`` [net|vdpa] (default **net**)
372 Choose the mode of operation of ef100 device.
373 **net** device will work as network device and will be probed by net/sfc driver.
374 **vdpa** device will work as vdpa device and will be probed by vdpa/sfc driver.
375 If this parameter is not specified then ef100 device will operate as
378 - ``switch_mode`` [legacy|switchdev] (see below for default)
380 In legacy mode, NIC firmware provides Ethernet virtual bridging (EVB) API
381 to configure switching inside NIC to deliver traffic to physical (PF) and
382 virtual (VF) PCI functions. PF driver is responsible to build the
383 infrastructure for VFs, and traffic goes to/from VF by default in accordance
384 with MAC address assigned, permissions and filters installed by VF drivers.
385 In switchdev mode VF traffic goes via port representor (if any) on PF, and
386 software virtual switch (for example, Open vSwitch) makes the decision.
387 Software virtual switch may install MAE rules to pass established traffic
388 flows via hardware and offload software datapath as the result.
389 Default is legacy, unless representors are specified, in which case switchdev
392 - ``representor`` parameter [list]
394 Instantiate port representor Ethernet devices for specified Virtual
397 It is a standard parameter whose format is described in
398 :ref:`ethernet_device_standard_device_arguments`.
400 - ``rx_datapath`` [auto|efx|ef10|ef10_essb] (default **auto**)
402 Choose receive datapath implementation.
403 **auto** allows the driver itself to make a choice based on firmware
404 features available and required by the datapath implementation.
405 **efx** chooses libefx-based datapath which supports Rx scatter.
406 Supported for SFN7xxx, SFN8xxx and X2xxx family adapters only.
407 **ef10** chooses EF10 (SFN7xxx, SFN8xxx, X2xxx) native datapath which is
408 more efficient than libefx-based and provides richer packet type
410 **ef10_essb** chooses SFNX2xxx equal stride super-buffer datapath
411 which may be used on DPDK firmware variant only
412 (see notes about its limitations above).
413 **ef100** chooses EF100 native datapath which is the only supported
414 Rx datapath for EF100 architecture based NICs.
416 - ``tx_datapath`` [auto|efx|ef10|ef10_simple|ef100] (default **auto**)
418 Choose transmit datapath implementation.
419 **auto** allows the driver itself to make a choice based on firmware
420 features available and required by the datapath implementation.
421 **efx** chooses libefx-based datapath which supports VLAN insertion
422 (full-feature firmware variant only), TSO and multi-segment mbufs.
423 Mbuf segments may come from different mempools, and mbuf reference
424 counters are treated responsibly.
425 Supported for SFN7xxx, SFN8xxx and X2xxx family adapters only.
426 **ef10** chooses EF10 (SFN7xxx, SFN8xxx, X2xxx) native datapath which is
427 more efficient than libefx-based but has no VLAN insertion support yet.
428 Mbuf segments may come from different mempools, and mbuf reference
429 counters are treated responsibly.
430 **ef10_simple** chooses EF10 (SFN7xxx, SFN8xxx, X2xxx) native datapath which
431 is even more faster then **ef10** but does not support multi-segment
432 mbufs, disallows multiple mempools and neglects mbuf reference counters.
433 **ef100** chooses EF100 native datapath which supports multi-segment
434 mbufs, VLAN insertion, inner/outer IPv4 and TCP/UDP checksum and TCP
435 segmentation offloads including VXLAN and GENEVE IPv4/IPv6 tunnels.
437 - ``perf_profile`` [auto|throughput|low-latency] (default **throughput**)
439 Choose hardware tuning to be optimized for either throughput or
441 **auto** allows NIC firmware to make a choice based on
442 installed licenses and firmware variant configured using **sfboot**.
444 - ``stats_update_period_ms`` [long] (default **1000**)
446 Adjust period in milliseconds to update port hardware statistics.
447 The accepted range is 0 to 65535. The value of **0** may be used
448 to disable periodic statistics update. One should note that it's
449 only possible to set an arbitrary value on SFN8xxx and X2xxx provided that
450 firmware version is 6.2.1.1033 or higher, otherwise any positive
451 value will select a fixed update period of **1000** milliseconds
453 - ``fw_variant`` [dont-care|full-feature|ultra-low-latency|
454 capture-packed-stream|dpdk] (default **dont-care**)
456 Choose the preferred firmware variant to use.
457 The parameter is supported for SFN7xxX, SFN8xxx and X2xxx families only.
458 In order for the selected option to have an effect, the **sfboot** utility
459 must be configured with the **auto** firmware-variant option.
460 The preferred firmware variant applies to all ports on the NIC.
461 **dont-care** ensures that the driver can attach to an unprivileged function.
462 The datapath firmware type to use is controlled by the **sfboot**
464 **full-feature** chooses full featured firmware.
465 **ultra-low-latency** chooses firmware with fewer features but lower latency.
466 **capture-packed-stream** chooses firmware for SolarCapture packed stream
468 **dpdk** chooses DPDK firmware with equal stride super-buffer Rx mode
469 for higher Rx packet rate and packet marks support and firmware subvariant
470 without checksumming on transmit for higher Tx packet rate if
471 checksumming is not required.
473 - ``rxd_wait_timeout_ns`` [long] (default **200 us**)
475 Adjust timeout in nanoseconds to head-of-line block to wait for
477 The accepted range is 0 to 400 ms.
478 Flow control should be enabled to make it work.
479 The value of **0** disables it and packets are dropped immediately.
480 When a packet is dropped because of no Rx descriptors,
481 ``rx_nodesc_drop_cnt`` counter grows.
482 The feature is supported only by the DPDK firmware variant when equal
483 stride super-buffer Rx mode is used.
486 Dynamic Logging Parameters
487 ~~~~~~~~~~~~~~~~~~~~~~~~~~
489 One may leverage EAL option "--log-level" to change default levels
490 for the log types supported by the driver. The option is used with
491 an argument typically consisting of two parts separated by a colon.
493 Level value is the last part which takes a symbolic name (or integer).
494 Log type is the former part which may shell match syntax.
495 Depending on the choice of the expression, the given log level may
496 be used either for some specific log type or for a subset of types.
498 SFC EFX PMD provides the following log types available for control:
500 - ``pmd.net.sfc.driver`` (default level is **notice**)
502 Affects driver-wide messages unrelated to any particular devices.
504 - ``pmd.net.sfc.main`` (default level is **notice**)
506 Matches a subset of per-port log types registered during runtime.
507 A full name for a particular type may be obtained by appending a
508 dot and a PCI device identifier (``XXXX:XX:XX.X``) to the prefix.
510 - ``pmd.net.sfc.mcdi`` (default level is **notice**)
512 Extra logging of the communication with the NIC's management CPU.
513 The format of the log is consumed by the Solarflare netlogdecode
514 cross-platform tool. May be managed per-port, as explained above.