1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright 2015 6WIND S.A.
3 Copyright 2015 Mellanox Technologies, Ltd
5 .. include:: <isonum.txt>
10 The MLX5 poll mode driver library (**librte_net_mlx5**) provides support
11 for **Mellanox ConnectX-4**, **Mellanox ConnectX-4 Lx** , **Mellanox
12 ConnectX-5**, **Mellanox ConnectX-6**, **Mellanox ConnectX-6 Dx**, **Mellanox
13 ConnectX-6 Lx**, **Mellanox BlueField** and **Mellanox BlueField-2** families
14 of 10/25/40/50/100/200 Gb/s adapters as well as their virtual functions (VF)
17 Information and documentation about these adapters can be found on the
18 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
19 `Mellanox community <http://community.mellanox.com/welcome>`__.
21 There is also a `section dedicated to this poll mode driver
22 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
28 Besides its dependency on libibverbs (that implies libmlx5 and associated
29 kernel support), librte_net_mlx5 relies heavily on system calls for control
30 operations such as querying/updating the MTU and flow control parameters.
32 For security reasons and robustness, this driver only deals with virtual
33 memory addresses. The way resources allocations are handled by the kernel,
34 combined with hardware specifications that allow to handle virtual memory
35 addresses directly, ensure that DPDK applications cannot access random
36 physical memory (or memory that does not belong to the current process).
38 This capability allows the PMD to coexist with kernel network interfaces
39 which remain functional, although they stop receiving unicast packets as
40 long as they share the same MAC address.
41 This means legacy linux control tools (for example: ethtool, ifconfig and
42 more) can operate on the same network interfaces that owned by the DPDK
45 The PMD can use libibverbs and libmlx5 to access the device firmware
46 or directly the hardware components.
47 There are different levels of objects and bypassing abilities
48 to get the best performances:
50 - Verbs is a complete high-level generic API
51 - Direct Verbs is a device-specific API
52 - DevX allows to access firmware objects
53 - Direct Rules manages flow steering at low-level hardware layer
55 Enabling librte_net_mlx5 causes DPDK applications to be linked against
61 - Multi arch support: x86_64, POWER8, ARMv8, i686.
62 - Multiple TX and RX queues.
63 - Support for scattered TX frames.
64 - Advanced support for scattered Rx frames with tunable buffer attributes.
65 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
66 - RSS using different combinations of fields: L3 only, L4 only or both,
67 and source only, destination only or both.
68 - Several RSS hash keys, one for each flow type.
69 - Default RSS operation with no hash key specification.
70 - Configurable RETA table.
71 - Link flow control (pause frame).
72 - Support for multiple MAC addresses.
76 - RX CRC stripping configuration.
77 - TX mbuf fast free offload.
78 - Promiscuous mode on PF and VF.
79 - Multicast promiscuous mode on PF and VF.
80 - Hardware checksum offloads.
81 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
83 - Flow API, including :ref:`flow_isolated_mode`.
85 - KVM and VMware ESX SR-IOV modes are supported.
86 - RSS hash result is supported.
87 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
88 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
90 - Statistics query including Basic, Extended and per queue.
92 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
93 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
94 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
95 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
96 - Flow insertion rate of more then million flows per second, when using Direct Rules.
97 - Support for multiple rte_flow groups.
98 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
101 - Multiple-thread flow insertion.
102 - Matching on GTP extension header with raw encap/decap action.
103 - Matching on Geneve TLV option header with raw encap/decap action.
104 - RSS support in sample action.
105 - E-Switch mirroring and jump.
106 - E-Switch mirroring and modify.
107 - 21844 flow priorities for ingress or egress flow groups greater than 0 and for any transfer
109 - Flow metering, including meter policy API.
116 On Windows, the features are limited:
118 - Promiscuous mode is not supported
119 - The following rules are supported:
121 - IPv4/UDP with CVLAN filtering
122 - Unicast MAC filtering
124 - For secondary process:
126 - Forked secondary process not supported.
127 - External memory unregistered in EAL memseg list cannot be used for DMA
128 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
129 primary process and remapped to the same virtual address in secondary
130 process. If the external memory is registered by primary process but has
131 different virtual address in secondary process, unexpected error may happen.
133 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
134 specific VLAN will match for VLAN packets as well:
136 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
137 Meaning, the flow rule::
139 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
141 Will only match vlan packets with vid=3. and the flow rule::
143 flow create 0 ingress pattern eth / ipv4 / end ...
145 Will match any ipv4 packet (VLAN included).
147 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
149 - When using DV flow engine (``dv_flow_en`` = 1), flow pattern with any VLAN specification will match only single-tagged packets unless the ETH item ``type`` field is 0x88A8 or the VLAN item ``has_more_vlan`` field is 1.
152 flow create 0 ingress pattern eth / ipv4 / end ...
154 Will match any ipv4 packet.
157 flow create 0 ingress pattern eth / vlan / end ...
158 flow create 0 ingress pattern eth has_vlan is 1 / end ...
159 flow create 0 ingress pattern eth type is 0x8100 / end ...
161 Will match single-tagged packets only, with any VLAN ID value.
164 flow create 0 ingress pattern eth type is 0x88A8 / end ...
165 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
167 Will match multi-tagged packets only, with any VLAN ID value.
169 - A flow pattern with 2 sequential VLAN items is not supported.
171 - VLAN pop offload command:
173 - Flow rules having a VLAN pop offload command as one of their actions and
174 are lacking a match on VLAN as one of their items are not supported.
175 - The command is not supported on egress traffic in NIC mode.
177 - VLAN push offload is not supported on ingress traffic in NIC mode.
179 - VLAN set PCP offload is not supported on existing headers.
181 - A multi segment packet must have not more segments than reported by dev_infos_get()
182 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
183 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
184 inline settings) to 58.
186 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
187 to 0 are not supported.
189 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
191 - Match on Geneve header supports the following fields only:
198 - Match on Geneve TLV option is supported on the following fields:
205 Only one Class/Type/Length Geneve TLV option is supported per shared device.
206 Class/Type/Length fields must be specified as well as masks.
207 Class/Type/Length specified masks must be full.
208 Matching Geneve TLV option without specifying data is not supported.
209 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
211 - VF: flow rules created on VF devices can only match traffic targeted at the
212 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
214 - Match on GTP tunnel header item supports the following fields only:
216 - v_pt_rsv_flags: E flag, S flag, PN flag
220 - Match on GTP extension header only for GTP PDU session container (next
221 extension header type = 0x85).
222 - Match on GTP extension header is not supported in group 0.
224 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
225 The flows within group 0 and set metadata action are rejected by hardware.
229 MAC addresses not already present in the bridge table of the associated
230 kernel network device will be added and cleaned up by the PMD when closing
231 the device. In case of ungraceful program termination, some entries may
232 remain present and should be removed manually by other means.
234 - Buffer split offload is supported with regular Rx burst routine only,
235 no MPRQ feature or vectorized code can be engaged.
237 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
238 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
239 ol_flags. As the mempool for the external buffer is managed by PMD, all the
240 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
241 the external buffers will be freed by PMD and the application which still
242 holds the external buffers may be corrupted.
244 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
245 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
246 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
248 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
249 and allmulticast mode are both set to off.
250 To receive IPv6 Multicast messages on VM, explicitly set the relevant
251 MAC address using rte_eth_dev_mac_addr_add() API.
253 - To support a mixed traffic pattern (some buffers from local host memory, some
254 buffers from other devices) with high bandwidth, a mbuf flag is used.
256 An application hints the PMD whether or not it should try to inline the
257 given mbuf data buffer. PMD should do the best effort to act upon this request.
259 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
260 registered by application with rte_mbuf_dynflag_register(). This flag is
261 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
262 which is intended to be used by the application.
264 To query the supported specific flags in runtime,
265 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
266 currently (over present hardware and configuration) supported specific flags.
267 The "not inline hint" feature operating flow is the following one:
270 - probe the devices, ports are created
271 - query the port capabilities
272 - if port supporting the feature is found
273 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
274 - application starts the ports
275 - on ``dev_start()`` PMD checks whether the feature flag is registered and
276 enables the feature support in datapath
277 - application might set the registered flag bit in ``ol_flags`` field
278 of mbuf being sent and PMD will handle ones appropriately.
280 - The amount of descriptors in Tx queue may be limited by data inline settings.
281 Inline data require the more descriptor building blocks and overall block
282 amount may exceed the hardware supported limits. The application should
283 reduce the requested Tx size or adjust data inline settings with
284 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
286 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
287 parameter should be specified.
288 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
289 being sent it tries to synchronize the time of packet appearing on
290 the wire with the specified packet timestamp. It the specified one
291 is in the past it should be ignored, if one is in the distant future
292 it should be capped with some reasonable value (in range of seconds).
293 These specific cases ("too late" and "distant future") can be optionally
294 reported via device xstats to assist applications to detect the
295 time-related problems.
297 The timestamp upper "too-distant-future" limit
298 at the moment of invoking the Tx burst routine
299 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
300 Please note, for the testpmd txonly mode,
301 the limit is deduced from the expression::
303 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
305 There is no any packet reordering according timestamps is supposed,
306 neither within packet burst, nor between packets, it is an entirely
307 application responsibility to generate packets and its timestamps
308 in desired order. The timestamps can be put only in the first packet
309 in the burst providing the entire burst scheduling.
311 - E-Switch decapsulation Flow:
313 - can be applied to PF port only.
314 - must specify VF port action (packet redirection from PF to VF).
315 - optionally may specify tunnel inner source and destination MAC addresses.
317 - E-Switch encapsulation Flow:
319 - can be applied to VF ports only.
320 - must specify PF port action (packet redirection from VF to PF).
324 - The input buffer, used as outer header, is not validated.
328 - The decapsulation is always done up to the outermost tunnel detected by the HW.
329 - The input buffer, providing the removal size, is not validated.
330 - The buffer size must match the length of the headers to be removed.
332 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
333 mutually exclusive features which cannot be supported together
334 (see :ref:`mlx5_firmware_config`).
338 - Requires DevX and DV flow to be enabled.
339 - KEEP_CRC offload cannot be supported with LRO.
340 - The first mbuf length, without head-room, must be big enough to include the
342 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
343 it with size limited to max LRO size, not to max RX packet length.
344 - LRO can be used with outer header of TCP packets of the standard format:
345 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
347 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
348 - LRO packet aggregation is performed by HW only for packet size larger than
349 ``lro_min_mss_size``. This value is reported on device start, when debug
354 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
355 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
356 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
360 - fast free offload assumes the all mbufs being sent are originated from the
361 same memory pool and there is no any extra references to the mbufs (the
362 reference counter for each mbuf is equal 1 on tx_burst call). The latter
363 means there should be no any externally attached buffers in mbufs. It is
364 an application responsibility to provide the correct mbufs if the fast
365 free offload is engaged. The mlx5 PMD implicitly produces the mbufs with
366 externally attached buffers if MPRQ option is enabled, hence, the fast
367 free offload is neither supported nor advertised if there is MPRQ enabled.
371 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and
372 E-Switch steering domain.
373 - For E-Switch Sampling flow with sample ratio > 1, additional actions are not
374 supported in the sample actions list.
375 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as
376 first action in the E-Switch egress flow if with header modify or
377 encapsulation actions.
378 - For NIC Rx flow, supports ``MARK``, ``COUNT``, ``QUEUE``, ``RSS`` in the
380 - For E-Switch mirroring flow, supports ``RAW ENCAP``, ``Port ID``,
381 ``VXLAN ENCAP``, ``NVGRE ENCAP`` in the sample actions list.
385 - Supports the 'set' operation only for ``RTE_FLOW_ACTION_TYPE_MODIFY_FIELD`` action.
386 - Modification of an arbitrary place in a packet via the special ``RTE_FLOW_FIELD_START`` Field ID is not supported.
387 - Modification of the 802.1Q Tag, VXLAN Network or GENEVE Network ID's is not supported.
388 - Encapsulation levels are not supported, can modify outermost header fields only.
389 - Offsets must be 32-bits aligned, cannot skip past the boundary of a field.
391 - IPv6 header item 'proto' field, indicating the next header protocol, should
392 not be set as extension header.
393 In case the next header is an extension header, it should not be specified in
394 IPv6 header item 'proto' field.
395 The last extension header item 'next header' field can specify the following
396 header protocol type.
400 - Hairpin between two ports could only manual binding and explicit Tx flow mode. For single port hairpin, all the combinations of auto/manual binding and explicit/implicit Tx flow mode could be supported.
401 - Hairpin in switchdev SR-IOV mode is not supported till now.
405 - All the meter colors with drop action will be counted only by the global drop statistics.
406 - Green color is not supported with drop action.
407 - Yellow detection is not supported.
408 - Red color must be with drop action.
409 - Meter statistics are supported only for drop case.
410 - Meter yellow color detection is not supported.
411 - A meter action created with pre-defined policy must be the last action in the flow except single case where the policy actions are:
412 - green: NULL or END.
413 - yellow: NULL or END.
415 - The only supported meter policy actions:
416 - green: QUEUE, RSS, PORT_ID, JUMP, MARK and SET_TAG.
417 - yellow: must be empty.
423 MLX5 supports various methods to report statistics:
425 Port statistics can be queried using ``rte_eth_stats_get()``. The received and sent statistics are through SW only and counts the number of packets received or sent successfully by the PMD. The imissed counter is the amount of packets that could not be delivered to SW because a queue was full. Packets not received due to congestion in the bus or on the NIC can be queried via the rx_discards_phy xstats counter.
427 Extended statistics can be queried using ``rte_eth_xstats_get()``. The extended statistics expose a wider set of counters counted by the device. The extended port statistics counts the number of packets received or sent successfully by the port. As Mellanox NICs are using the :ref:`Bifurcated Linux Driver <linux_gsg_linux_drivers>` those counters counts also packet received or sent by the Linux kernel. The counters with ``_phy`` suffix counts the total events on the physical port, therefore not valid for VF.
429 Finally per-flow statistics can by queried using ``rte_flow_query`` when attaching a count action for specific flow. The flow counter counts the number of packets received successfully by the port and match the specific flow.
437 The ibverbs libraries can be linked with this PMD in a number of ways,
438 configured by the ``ibverbs_link`` build option:
440 - ``shared`` (default): the PMD depends on some .so files.
442 - ``dlopen``: Split the dependencies glue in a separate library
443 loaded when needed by dlopen.
444 It make dependencies on libibverbs and libmlx4 optional,
445 and has no performance impact.
447 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
448 in the PMD shared library or the executable static binary.
450 Environment variables
451 ~~~~~~~~~~~~~~~~~~~~~
455 A list of directories in which to search for the rdma-core "glue" plug-in,
456 separated by colons or semi-colons.
458 - ``MLX5_SHUT_UP_BF``
460 Configures HW Tx doorbell register as IO-mapped.
462 By default, the HW Tx doorbell is configured as a write-combining register.
463 The register would be flushed to HW usually when the write-combining buffer
464 becomes full, but it depends on CPU design.
466 Except for vectorized Tx burst routines, a write memory barrier is enforced
467 after updating the register so that the update can be immediately visible to
470 When vectorized Tx burst is called, the barrier is set only if the burst size
471 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
472 variable will bring better latency even though the maximum throughput can
475 Run-time configuration
476 ~~~~~~~~~~~~~~~~~~~~~~
478 - librte_net_mlx5 brings kernel network interfaces up during initialization
479 because it is affected by their state. Forcing them down prevents packets
482 - **ethtool** operations on related kernel interfaces also affect the PMD.
487 In order to run as a non-root user,
488 some capabilities must be granted to the application::
490 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
492 Below are the reasons of the need for each capability:
495 When using physical addresses (PA mode), with Linux >= 4.0,
496 for access to ``/proc/self/pagemap``.
499 For device configuration.
502 For raw ethernet queue allocation through kernel driver.
505 For DMA memory pinning.
510 - ``rxq_cqe_comp_en`` parameter [int]
512 A nonzero value enables the compression of CQE on RX side. This feature
513 allows to save PCI bandwidth and improve performance. Enabled by default.
514 Different compression formats are supported in order to achieve the best
515 performance for different traffic patterns. Default format depends on
516 Multi-Packet Rx queue configuration: Hash RSS format is used in case
517 MPRQ is disabled, Checksum format is used in case MPRQ is enabled.
519 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
520 better compression rate in case of RTE Flow Mark traffic.
521 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
522 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
523 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
524 CQE compression format selection requires DevX to be enabled. If there is
525 no DevX enabled/supported the value is reset to 1 by default.
529 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
530 ConnectX-6 Lx, BlueField and BlueField-2.
531 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
532 ConnectX-6 Lx, BlueField and BlueField-2.
534 - ``rxq_pkt_pad_en`` parameter [int]
536 A nonzero value enables padding Rx packet to the size of cacheline on PCI
537 transaction. This feature would waste PCI bandwidth but could improve
538 performance by avoiding partial cacheline write which may cause costly
539 read-modify-copy in memory transaction on some architectures. Disabled by
544 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
545 ConnectX-6 Lx, BlueField and BlueField-2.
546 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
547 ConnectX-6 Lx, BlueField and BlueField-2.
549 - ``mprq_en`` parameter [int]
551 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
552 configured as Multi-Packet RQ if the total number of Rx queues is
553 ``rxqs_min_mprq`` or more. Disabled by default.
555 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
556 by posting a single large buffer for multiple packets. Instead of posting a
557 buffers per a packet, one large buffer is posted in order to receive multiple
558 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
559 and each stride receives one packet. MPRQ can improve throughput for
560 small-packet traffic.
562 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
563 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
564 configure large stride size enough to accommodate max_rx_pkt_len as long as
565 device allows. Note that this can waste system memory compared to enabling Rx
566 scatter and multi-segment packet.
568 - ``mprq_log_stride_num`` parameter [int]
570 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
571 strides can reduce PCIe traffic further. If configured value is not in the
572 range of device capability, the default value will be set with a warning
573 message. The default value is 4 which is 16 strides per a buffer, valid only
574 if ``mprq_en`` is set.
576 The size of Rx queue should be bigger than the number of strides.
578 - ``mprq_log_stride_size`` parameter [int]
580 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
581 stride size can save some memory and reduce probability of a depletion of all
582 available strides due to unreleased packets by an application. If configured
583 value is not in the range of device capability, the default value will be set
584 with a warning message. The default value is 11 which is 2048 bytes per a
585 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
586 it is possible for a packet to span across multiple strides. This mode allows
587 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
588 of a packet if Rx scatter is configured) may be required in case there is no
589 space left for a head room at the end of a stride which incurs some
592 - ``mprq_max_memcpy_len`` parameter [int]
594 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
595 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
596 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
597 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
598 A mempool for external buffers will be allocated and managed by PMD. If Rx
599 packet is externally attached, ol_flags field of the mbuf will have
600 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
601 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
603 - ``rxqs_min_mprq`` parameter [int]
605 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
606 greater or equal to this value. The default value is 12, valid only if
609 - ``txq_inline`` parameter [int]
611 Amount of data to be inlined during TX operations. This parameter is
612 deprecated and converted to the new parameter ``txq_inline_max`` providing
613 partial compatibility.
615 - ``txqs_min_inline`` parameter [int]
617 Enable inline data send only when the number of TX queues is greater or equal
620 This option should be used in combination with ``txq_inline_max`` and
621 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
623 If this option is not specified the default value 16 is used for BlueField
624 and 8 for other platforms
626 The data inlining consumes the CPU cycles, so this option is intended to
627 auto enable inline data if we have enough Tx queues, which means we have
628 enough CPU cores and PCI bandwidth is getting more critical and CPU
629 is not supposed to be bottleneck anymore.
631 The copying data into WQE improves latency and can improve PPS performance
632 when PCI back pressure is detected and may be useful for scenarios involving
633 heavy traffic on many queues.
635 Because additional software logic is necessary to handle this mode, this
636 option should be used with care, as it may lower performance when back
637 pressure is not expected.
639 If inline data are enabled it may affect the maximal size of Tx queue in
640 descriptors because the inline data increase the descriptor size and
641 queue size limits supported by hardware may be exceeded.
643 - ``txq_inline_min`` parameter [int]
645 Minimal amount of data to be inlined into WQE during Tx operations. NICs
646 may require this minimal data amount to operate correctly. The exact value
647 may depend on NIC operation mode, requested offloads, etc. It is strongly
648 recommended to omit this parameter and use the default values. Anyway,
649 applications using this parameter should take into consideration that
650 specifying an inconsistent value may prevent the NIC from sending packets.
652 If ``txq_inline_min`` key is present the specified value (may be aligned
653 by the driver in order not to exceed the limits and provide better descriptor
654 space utilization) will be used by the driver and it is guaranteed that
655 requested amount of data bytes are inlined into the WQE beside other inline
656 settings. This key also may update ``txq_inline_max`` value (default
657 or specified explicitly in devargs) to reserve the space for inline data.
659 If ``txq_inline_min`` key is not present, the value may be queried by the
660 driver from the NIC via DevX if this feature is available. If there is no DevX
661 enabled/supported the value 18 (supposing L2 header including VLAN) is set
662 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
663 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
666 For ConnectX-4 NIC, driver does not allow specifying value below 18
667 (minimal L2 header, including VLAN), error will be raised.
669 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
670 it is not recommended and may prevent NIC from sending packets over
673 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
674 Multi-Packet Write), because last one does not support partial packet inlining.
675 This is not very critical due to minimal data inlining is mostly required
676 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
678 - ``txq_inline_max`` parameter [int]
680 Specifies the maximal packet length to be completely inlined into WQE
681 Ethernet Segment for ordinary SEND method. If packet is larger than specified
682 value, the packet data won't be copied by the driver at all, data buffer
683 is addressed with a pointer. If packet length is less or equal all packet
684 data will be copied into WQE. This may improve PCI bandwidth utilization for
685 short packets significantly but requires the extra CPU cycles.
687 The data inline feature is controlled by number of Tx queues, if number of Tx
688 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
689 is engaged, if there are not enough Tx queues (which means not enough CPU cores
690 and CPU resources are scarce), data inline is not performed by the driver.
691 Assigning ``txqs_min_inline`` with zero always enables the data inline.
693 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
694 by the driver in order not to exceed the limit (930 bytes) and to provide better
695 WQE space filling without gaps, the adjustment is reflected in the debug log.
696 Also, the default value (290) may be decreased in run-time if the large transmit
697 queue size is requested and hardware does not support enough descriptor
698 amount, in this case warning is emitted. If ``txq_inline_max`` key is
699 specified and requested inline settings can not be satisfied then error
702 - ``txq_inline_mpw`` parameter [int]
704 Specifies the maximal packet length to be completely inlined into WQE for
705 Enhanced MPW method. If packet is large the specified value, the packet data
706 won't be copied, and data buffer is addressed with pointer. If packet length
707 is less or equal, all packet data will be copied into WQE. This may improve PCI
708 bandwidth utilization for short packets significantly but requires the extra
711 The data inline feature is controlled by number of TX queues, if number of Tx
712 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
713 is engaged, if there are not enough Tx queues (which means not enough CPU cores
714 and CPU resources are scarce), data inline is not performed by the driver.
715 Assigning ``txqs_min_inline`` with zero always enables the data inline.
717 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
718 by the driver in order not to exceed the limit (930 bytes) and to provide better
719 WQE space filling without gaps, the adjustment is reflected in the debug log.
720 Due to multiple packets may be included to the same WQE with Enhanced Multi
721 Packet Write Method and overall WQE size is limited it is not recommended to
722 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
723 may be decreased in run-time if the large transmit queue size is requested
724 and hardware does not support enough descriptor amount, in this case warning
725 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
726 settings can not be satisfied then error will be raised.
728 - ``txqs_max_vec`` parameter [int]
730 Enable vectorized Tx only when the number of TX queues is less than or
731 equal to this value. This parameter is deprecated and ignored, kept
732 for compatibility issue to not prevent driver from probing.
734 - ``txq_mpw_hdr_dseg_en`` parameter [int]
736 A nonzero value enables including two pointers in the first block of TX
737 descriptor. The parameter is deprecated and ignored, kept for compatibility
740 - ``txq_max_inline_len`` parameter [int]
742 Maximum size of packet to be inlined. This limits the size of packet to
743 be inlined. If the size of a packet is larger than configured value, the
744 packet isn't inlined even though there's enough space remained in the
745 descriptor. Instead, the packet is included with pointer. This parameter
746 is deprecated and converted directly to ``txq_inline_mpw`` providing full
747 compatibility. Valid only if eMPW feature is engaged.
749 - ``txq_mpw_en`` parameter [int]
751 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
752 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
753 eMPW allows the Tx burst function to pack up multiple packets
754 in a single descriptor session in order to save PCI bandwidth
755 and improve performance at the cost of a slightly higher CPU usage.
756 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
757 Tx burst function copies entire packet data on to Tx descriptor
758 instead of including pointer of packet.
760 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
761 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
762 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
763 option or reported by the NIC, the eMPW feature is disengaged.
765 - ``tx_db_nc`` parameter [int]
767 The rdma core library can map doorbell register in two ways, depending on the
768 environment variable "MLX5_SHUT_UP_BF":
770 - As regular cached memory (usually with write combining attribute), if the
771 variable is either missing or set to zero.
772 - As non-cached memory, if the variable is present and set to not "0" value.
774 The type of mapping may slightly affect the Tx performance, the optimal choice
775 is strongly relied on the host architecture and should be deduced practically.
777 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
778 memory (with write combining), the PMD will perform the extra write memory barrier
779 after writing to doorbell, it might increase the needed CPU clocks per packet
780 to send, but latency might be improved.
782 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
783 cached memory, the PMD will not perform the extra write memory barrier
784 after writing to doorbell, on some architectures it might improve the
787 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
788 memory, the PMD will use heuristics to decide whether write memory barrier
789 should be performed. For bursts with size multiple of recommended one (64 pkts)
790 it is supposed the next burst is coming and no need to issue the extra memory
791 barrier (it is supposed to be issued in the next coming burst, at least after
792 descriptor writing). It might increase latency (on some hosts till next
793 packets transmit) and should be used with care.
795 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
796 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
797 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
799 - ``tx_pp`` parameter [int]
801 If a nonzero value is specified the driver creates all necessary internal
802 objects to provide accurate packet send scheduling on mbuf timestamps.
803 The positive value specifies the scheduling granularity in nanoseconds,
804 the packet send will be accurate up to specified digits. The allowed range is
805 from 500 to 1 million of nanoseconds. The negative value specifies the module
806 of granularity and engages the special test mode the check the schedule rate.
807 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
810 - ``tx_skew`` parameter [int]
812 The parameter adjusts the send packet scheduling on timestamps and represents
813 the average delay between beginning of the transmitting descriptor processing
814 by the hardware and appearance of actual packet data on the wire. The value
815 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
816 specified. The default value is zero.
818 - ``tx_vec_en`` parameter [int]
820 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
821 ConnectX-6 Lx, BlueField and BlueField-2 NICs
822 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
823 The parameter is deprecated and ignored.
825 - ``rx_vec_en`` parameter [int]
827 A nonzero value enables Rx vector if the port is not configured in
828 multi-segment otherwise this parameter is ignored.
832 - ``vf_nl_en`` parameter [int]
834 A nonzero value enables Netlink requests from the VF to add/remove MAC
835 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
836 Otherwise the relevant configuration must be run with Linux iproute2 tools.
837 This is a prerequisite to receive this kind of traffic.
839 Enabled by default, valid only on VF devices ignored otherwise.
841 - ``l3_vxlan_en`` parameter [int]
843 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
844 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
845 parameter. This is a prerequisite to receive this kind of traffic.
849 - ``dv_xmeta_en`` parameter [int]
851 A nonzero value enables extensive flow metadata support if device is
852 capable and driver supports it. This can enable extensive support of
853 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
854 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
856 There are some possible configurations, depending on parameter value:
858 - 0, this is default value, defines the legacy mode, the ``MARK`` and
859 ``META`` related actions and items operate only within NIC Tx and
860 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
861 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
862 item is 32 bits wide and match supported on egress only.
864 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
865 related actions and items operate within all supported steering domains,
866 including FDB, ``MARK`` and ``META`` information may cross the domain
867 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
868 depends on kernel and firmware configurations and might be 0, 16 or
869 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
870 compatibility, the actual width of data transferred to the FDB domain
871 depends on kernel configuration and may be vary. The actual supported
872 width can be retrieved in runtime by series of rte_flow_validate()
875 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
876 related actions and items operate within all supported steering domains,
877 including FDB, ``MARK`` and ``META`` information may cross the domain
878 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
879 depends on kernel and firmware configurations and might be 0, 16 or
880 24 bits. The actual supported width can be retrieved in runtime by
881 series of rte_flow_validate() trials.
883 - 3, this engages tunnel offload mode. In E-Switch configuration, that
884 mode implicitly activates ``dv_xmeta_en=1``.
886 +------+-----------+-----------+-------------+-------------+
887 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
888 +======+===========+===========+=============+=============+
889 | 0 | 24 bits | 32 bits | 32 bits | no |
890 +------+-----------+-----------+-------------+-------------+
891 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
892 +------+-----------+-----------+-------------+-------------+
893 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
894 +------+-----------+-----------+-------------+-------------+
896 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
897 ignored and the device is configured to operate in legacy mode (0).
899 Disabled by default (set to 0).
901 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
902 of the extensive metadata features. The legacy Verbs supports FLAG and
903 MARK metadata actions over NIC Rx steering domain only.
905 Setting META value to zero in flow action means there is no item provided
906 and receiving datapath will not report in mbufs the metadata are present.
907 Setting MARK value to zero in flow action means the zero FDIR ID value
908 will be reported on packet receiving.
910 For the MARK action the last 16 values in the full range are reserved for
911 internal PMD purposes (to emulate FLAG action). The valid range for the
912 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
913 for the 24-bit mode, the flows with the MARK action value outside
914 the specified range will be rejected.
916 - ``dv_flow_en`` parameter [int]
918 A nonzero value enables the DV flow steering assuming it is supported
919 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
921 Enabled by default if supported.
923 - ``dv_esw_en`` parameter [int]
925 A nonzero value enables E-Switch using Direct Rules.
927 Enabled by default if supported.
929 - ``lacp_by_user`` parameter [int]
931 A nonzero value enables the control of LACP traffic by the user application.
932 When a bond exists in the driver, by default it should be managed by the
933 kernel and therefore LACP traffic should be steered to the kernel.
934 If this devarg is set to 1 it will allow the user to manage the bond by
935 itself and not steer LACP traffic to the kernel.
937 Disabled by default (set to 0).
939 - ``mr_ext_memseg_en`` parameter [int]
941 A nonzero value enables extending memseg when registering DMA memory. If
942 enabled, the number of entries in MR (Memory Region) lookup table on datapath
943 is minimized and it benefits performance. On the other hand, it worsens memory
944 utilization because registered memory is pinned by kernel driver. Even if a
945 page in the extended chunk is freed, that doesn't become reusable until the
946 entire memory is freed.
950 - ``representor`` parameter [list]
952 This parameter can be used to instantiate DPDK Ethernet devices from
953 existing port (PF, VF or SF) representors configured on the device.
955 It is a standard parameter whose format is described in
956 :ref:`ethernet_device_standard_device_arguments`.
958 For instance, to probe VF port representors 0 through 2::
960 <PCI_BDF>,representor=vf[0-2]
962 To probe SF port representors 0 through 2::
964 <PCI_BDF>,representor=sf[0-2]
966 To probe VF port representors 0 through 2 on both PFs of bonding device::
968 <Primary_PCI_BDF>,representor=pf[0,1]vf[0-2]
970 - ``max_dump_files_num`` parameter [int]
972 The maximum number of files per PMD entity that may be created for debug information.
973 The files will be created in /var/log directory or in current directory.
975 set to 128 by default.
977 - ``lro_timeout_usec`` parameter [int]
979 The maximum allowed duration of an LRO session, in micro-seconds.
980 PMD will set the nearest value supported by HW, which is not bigger than
981 the input ``lro_timeout_usec`` value.
982 If this parameter is not specified, by default PMD will set
983 the smallest value supported by HW.
985 - ``hp_buf_log_sz`` parameter [int]
987 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
988 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
989 The capacity of the value is specified by the firmware and the initialization
990 will get a failure if it is out of scope.
991 The range of the value is from 11 to 19 right now, and the supported frame
992 size of a single packet for hairpin is from 512B to 128KB. It might change if
993 different firmware release is being used. By using a small value, it could
994 reduce memory consumption but not work with a large frame. If the value is
995 too large, the memory consumption will be high and some potential performance
996 degradation will be introduced.
997 By default, the PMD will set this value to 16, which means that 9KB jumbo
998 frames will be supported.
1000 - ``reclaim_mem_mode`` parameter [int]
1002 Cache some resources in flow destroy will help flow recreation more efficient.
1003 While some systems may require the all the resources can be reclaimed after
1005 The parameter ``reclaim_mem_mode`` provides the option for user to configure
1006 if the resource cache is needed or not.
1008 There are three options to choose:
1010 - 0. It means the flow resources will be cached as usual. The resources will
1011 be cached, helpful with flow insertion rate.
1013 - 1. It will only enable the DPDK PMD level resources reclaim.
1015 - 2. Both DPDK PMD level and rdma-core low level will be configured as
1018 By default, the PMD will set this value to 0.
1020 - ``sys_mem_en`` parameter [int]
1022 A non-zero value enables the PMD memory management allocating memory
1023 from system by default, without explicit rte memory flag.
1025 By default, the PMD will set this value to 0.
1027 - ``decap_en`` parameter [int]
1029 Some devices do not support FCS (frame checksum) scattering for
1030 tunnel-decapsulated packets.
1031 If set to 0, this option forces the FCS feature and rejects tunnel
1032 decapsulation in the flow engine for such devices.
1034 By default, the PMD will set this value to 1.
1036 .. _mlx5_firmware_config:
1038 Firmware configuration
1039 ~~~~~~~~~~~~~~~~~~~~~~
1041 Firmware features can be configured as key/value pairs.
1043 The command to set a value is::
1045 mlxconfig -d <device> set <key>=<value>
1047 The command to query a value is::
1049 mlxconfig -d <device> query | grep <key>
1051 The device name for the command ``mlxconfig`` can be either the PCI address,
1052 or the mst device name found with::
1056 Below are some firmware configurations listed.
1062 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1068 - maximum number of SR-IOV virtual functions::
1072 - enable DevX (required by Direct Rules and other features)::
1076 - aggressive CQE zipping::
1080 - L3 VXLAN and VXLAN-GPE destination UDP port::
1083 IP_OVER_VXLAN_PORT=<udp dport>
1085 - enable VXLAN-GPE tunnel flow matching::
1087 FLEX_PARSER_PROFILE_ENABLE=0
1089 FLEX_PARSER_PROFILE_ENABLE=2
1091 - enable IP-in-IP tunnel flow matching::
1093 FLEX_PARSER_PROFILE_ENABLE=0
1095 - enable MPLS flow matching::
1097 FLEX_PARSER_PROFILE_ENABLE=1
1099 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1101 FLEX_PARSER_PROFILE_ENABLE=2
1103 - enable Geneve flow matching::
1105 FLEX_PARSER_PROFILE_ENABLE=0
1107 FLEX_PARSER_PROFILE_ENABLE=1
1109 - enable Geneve TLV option flow matching::
1111 FLEX_PARSER_PROFILE_ENABLE=0
1113 - enable GTP flow matching::
1115 FLEX_PARSER_PROFILE_ENABLE=3
1117 - enable eCPRI flow matching::
1119 FLEX_PARSER_PROFILE_ENABLE=4
1125 This driver relies on external libraries and kernel drivers for resources
1126 allocations and initialization. The following dependencies are not part of
1127 DPDK and must be installed separately:
1131 User space Verbs framework used by librte_net_mlx5. This library provides
1132 a generic interface between the kernel and low-level user space drivers
1135 It allows slow and privileged operations (context initialization, hardware
1136 resources allocations) to be managed by the kernel and fast operations to
1137 never leave user space.
1141 Low-level user space driver library for Mellanox
1142 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1145 This library basically implements send/receive calls to the hardware
1148 - **Kernel modules**
1150 They provide the kernel-side Verbs API and low level device drivers that
1151 manage actual hardware initialization and resources sharing with user
1154 Unlike most other PMDs, these modules must remain loaded and bound to
1157 - mlx5_core: hardware driver managing Mellanox
1158 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1160 - mlx5_ib: InifiniBand device driver.
1161 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1163 - **Firmware update**
1165 Mellanox OFED/EN releases include firmware updates for
1166 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1168 Because each release provides new features, these updates must be applied to
1169 match the kernel modules and libraries they come with.
1173 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1179 Either RDMA Core library with a recent enough Linux kernel release
1180 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1183 RDMA Core with Linux Kernel
1184 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1186 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1187 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1188 (see `RDMA Core installation documentation`_)
1189 - When building for i686 use:
1191 - rdma-core version 18.0 or above built with 32bit support.
1192 - Kernel version 4.14.41 or above.
1194 - Starting with rdma-core v21, static libraries can be built::
1197 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1200 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1201 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1207 - Mellanox OFED version: **4.5** and above /
1208 Mellanox EN version: **4.5** and above
1211 - ConnectX-4: **12.21.1000** and above.
1212 - ConnectX-4 Lx: **14.21.1000** and above.
1213 - ConnectX-5: **16.21.1000** and above.
1214 - ConnectX-5 Ex: **16.21.1000** and above.
1215 - ConnectX-6: **20.27.0090** and above.
1216 - ConnectX-6 Dx: **22.27.0090** and above.
1217 - BlueField: **18.25.1010** and above.
1219 While these libraries and kernel modules are available on OpenFabrics
1220 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1221 managers on most distributions, this PMD requires Ethernet extensions that
1222 may not be supported at the moment (this is a work in progress).
1225 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1227 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1228 include the necessary support and should be used in the meantime. For DPDK,
1229 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1230 required from that distribution.
1234 Several versions of Mellanox OFED/EN are available. Installing the version
1235 this DPDK release was developed and tested against is strongly
1236 recommended. Please check the `linux prerequisites`_.
1238 Windows Prerequisites
1239 ---------------------
1241 This driver relies on external libraries and kernel drivers for resources
1242 allocations and initialization. The dependencies in the following sub-sections
1243 are not part of DPDK, and must be installed separately.
1245 Compilation Prerequisites
1246 ~~~~~~~~~~~~~~~~~~~~~~~~~
1248 DevX SDK installation
1249 ^^^^^^^^^^^^^^^^^^^^^
1251 The DevX SDK must be installed on the machine building the Windows PMD.
1252 Additional information can be found at
1253 `How to Integrate Windows DevX in Your Development Environment
1254 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1256 Runtime Prerequisites
1257 ~~~~~~~~~~~~~~~~~~~~~
1259 WinOF2 version 2.60 or higher must be installed on the machine.
1264 The driver can be downloaded from the following site:
1266 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1271 DevX for Windows must be enabled in the Windows registry.
1272 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1273 Additional information can be found in the WinOF2 user manual.
1278 The following Mellanox device families are supported by the same mlx5 driver:
1290 Below are detailed device names:
1292 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1293 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1294 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1295 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1296 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1297 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1298 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1299 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1300 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1301 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1302 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1303 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1304 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1305 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1306 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1307 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1308 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1309 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1310 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1311 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1312 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1313 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1314 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1315 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1316 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1317 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1319 Quick Start Guide on OFED/EN
1320 ----------------------------
1322 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1325 2. Install the required libraries and kernel modules either by installing
1326 only the required set, or by installing the entire Mellanox OFED/EN::
1328 ./mlnxofedinstall --upstream-libs --dpdk
1330 3. Verify the firmware is the correct one::
1334 4. Verify all ports links are set to Ethernet::
1336 mlxconfig -d <mst device> query | grep LINK_TYPE
1340 Link types may have to be configured to Ethernet::
1342 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1344 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1346 For hypervisors, verify SR-IOV is enabled on the NIC::
1348 mlxconfig -d <mst device> query | grep SRIOV_EN
1351 If needed, configure SR-IOV::
1353 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1354 mlxfwreset -d <mst device> reset
1356 5. Restart the driver::
1358 /etc/init.d/openibd restart
1362 service openibd restart
1364 If link type was changed, firmware must be reset as well::
1366 mlxfwreset -d <mst device> reset
1368 For hypervisors, after reset write the sysfs number of virtual functions
1371 To dynamically instantiate a given number of virtual functions (VFs)::
1373 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1375 6. Install DPDK and you are ready to go.
1376 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1378 Enable switchdev mode
1379 ---------------------
1381 Switchdev mode is a mode in E-Switch, that binds between representor and VF or SF.
1382 Representor is a port in DPDK that is connected to a VF or SF in such a way
1383 that assuming there are no offload flows, each packet that is sent from the VF or SF
1384 will be received by the corresponding representor. While each packet that is or SF
1385 sent to a representor will be received by the VF or SF.
1386 This is very useful in case of SRIOV mode, where the first packet that is sent
1387 by the VF or SF will be received by the DPDK application which will decide if this
1388 flow should be offloaded to the E-Switch. After offloading the flow packet
1389 that the VF or SF that are matching the flow will not be received any more by
1390 the DPDK application.
1392 1. Enable SRIOV mode::
1394 mlxconfig -d <mst device> set SRIOV_EN=true
1396 2. Configure the max number of VFs::
1398 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1402 mlxfwreset -d <mst device> reset
1404 3. Configure the actual number of VFs::
1406 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1408 4. Unbind the device (can be rebind after the switchdev mode)::
1410 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1412 5. Enbale switchdev mode::
1414 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1416 SubFunction representor support
1417 -------------------------------
1418 SubFunction is a portion of the PCI device, a SF netdev has its own
1419 dedicated queues(txq, rxq). A SF netdev supports E-Switch representation
1420 offload similar to existing PF and VF representors. A SF shares PCI
1421 level resources with other SFs and/or with its parent PCI function.
1423 1. Configure SF feature::
1425 mlxconfig -d <mst device> set PF_BAR2_SIZE=<0/1/2/3> PF_BAR2_ENABLE=1
1427 Value of PF_BAR2_SIZE:
1436 mlxfwreset -d <mst device> reset
1438 3. Enable switchdev mode::
1440 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1444 mlnx-sf -d <PCI_BDF> -a create
1446 5. Probe SF representor::
1448 testpmd> port attach <PCI_BDF>,representor=sf0,dv_flow_en=1
1453 1. Configure aggressive CQE Zipping for maximum performance::
1455 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1457 To set it back to the default CQE Zipping mode use::
1459 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1461 2. In case of virtualization:
1463 - Make sure that hypervisor kernel is 3.16 or newer.
1464 - Configure boot with ``iommu=pt``.
1465 - Use 1G huge pages.
1466 - Make sure to allocate a VM on huge pages.
1467 - Make sure to set CPU pinning.
1469 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1470 for better performance. For VMs, verify that the right CPU
1471 and NUMA node are pinned according to the above. Run::
1475 to identify the NUMA node to which the PCIe adapter is connected.
1477 4. If more than one adapter is used, and root complex capabilities allow
1478 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1479 it is recommended to locate both adapters on the same NUMA node.
1480 This in order to forward packets from one to the other without
1481 NUMA performance penalty.
1483 5. Disable pause frames::
1485 ethtool -A <netdev> rx off tx off
1487 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1488 via the BIOS configuration. Please contact you server provider for more
1489 information about the settings.
1493 On some machines, depends on the machine integrator, it is beneficial
1494 to set the PCI max read request parameter to 1K. This can be
1495 done in the following way:
1497 To query the read request size use::
1499 setpci -s <NIC PCI address> 68.w
1501 If the output is different than 3XXX, set it by::
1503 setpci -s <NIC PCI address> 68.w=3XXX
1505 The XXX can be different on different systems. Make sure to configure
1506 according to the setpci output.
1508 7. To minimize overhead of searching Memory Regions:
1510 - '--socket-mem' is recommended to pin memory by predictable amount.
1511 - Configure per-lcore cache when creating Mempools for packet buffer.
1512 - Refrain from dynamically allocating/freeing memory in run-time.
1517 There are multiple Rx burst functions with different advantages and limitations.
1519 .. table:: Rx burst functions
1521 +-------------------+------------------------+---------+-----------------+------+-------+
1522 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1523 | | | | || comp|| MTU |
1524 +===================+========================+=========+=================+======+=======+
1525 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1526 +-------------------+------------------------+---------+-----------------+------+-------+
1527 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1528 +-------------------+------------------------+---------+-----------------+------+-------+
1529 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1530 | || RxQs >= rxqs_min_mprq | | | | |
1531 +-------------------+------------------------+---------+-----------------+------+-------+
1532 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1533 | || mprq_en=1 | | | | |
1534 | || RxQs >= rxqs_min_mprq | | | | |
1535 +-------------------+------------------------+---------+-----------------+------+-------+
1537 .. _mlx5_offloads_support:
1539 Supported hardware offloads
1540 ---------------------------
1542 .. table:: Minimal SW/HW versions for queue offloads
1544 ============== ===== ===== ========= ===== ========== =============
1545 Offload DPDK Linux rdma-core OFED firmware hardware
1546 ============== ===== ===== ========= ===== ========== =============
1547 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1548 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1549 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1550 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1551 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1552 Tx scheduling 20.08 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1553 Buffer Split 20.11 N/A N/A 5.1-2 16.28.2006 ConnectX-5
1554 ============== ===== ===== ========= ===== ========== =============
1556 .. table:: Minimal SW/HW versions for rte_flow offloads
1558 +-----------------------+-----------------+-----------------+
1559 | Offload | with E-Switch | with NIC |
1560 +=======================+=================+=================+
1561 | Count | | DPDK 19.05 | | DPDK 19.02 |
1562 | | | OFED 4.6 | | OFED 4.6 |
1563 | | | rdma-core 24 | | rdma-core 23 |
1564 | | | ConnectX-5 | | ConnectX-5 |
1565 +-----------------------+-----------------+-----------------+
1566 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1567 | | | OFED 4.6 | | OFED 4.5 |
1568 | | | rdma-core 24 | | rdma-core 23 |
1569 | | | ConnectX-5 | | ConnectX-4 |
1570 +-----------------------+-----------------+-----------------+
1571 | Queue / RSS | | | | DPDK 18.11 |
1572 | | | N/A | | OFED 4.5 |
1573 | | | | | rdma-core 23 |
1574 | | | | | ConnectX-4 |
1575 +-----------------------+-----------------+-----------------+
1576 | Shared action | | | | |
1577 | | | :numref:`sact`| | :numref:`sact`|
1580 +-----------------------+-----------------+-----------------+
1581 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1582 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1583 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1584 | | of_set_vlan_pcp / | | | | |
1585 | | of_set_vlan_vid) | | | | |
1586 +-----------------------+-----------------+-----------------+
1587 | | VLAN | | DPDK 21.05 | | |
1588 | | ingress and / | | OFED 5.3 | | N/A |
1589 | | of_push_vlan / | | ConnectX-6 Dx | | |
1590 +-----------------------+-----------------+-----------------+
1591 | | VLAN | | DPDK 21.05 | | |
1592 | | egress and / | | OFED 5.3 | | N/A |
1593 | | of_pop_vlan / | | ConnectX-6 Dx | | |
1594 +-----------------------+-----------------+-----------------+
1595 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1596 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1597 | | | rdma-core 24 | | rdma-core 23 |
1598 | | | ConnectX-5 | | ConnectX-5 |
1599 +-----------------------+-----------------+-----------------+
1600 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1601 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1602 | | | rdma-core 27 | | rdma-core 27 |
1603 | | | ConnectX-5 | | ConnectX-5 |
1604 +-----------------------+-----------------+-----------------+
1605 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1606 | | | OFED 5.1-2 | | OFED 5.1-2 |
1607 | | | rdma-core 32 | | N/A |
1608 | | | ConnectX-5 | | ConnectX-5 |
1609 +-----------------------+-----------------+-----------------+
1610 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1611 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1612 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1613 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1614 | | set_ipv6_dst / | | | | |
1615 | | set_tp_src / | | | | |
1616 | | set_tp_dst / | | | | |
1617 | | dec_ttl / | | | | |
1618 | | set_ttl / | | | | |
1619 | | set_mac_src / | | | | |
1620 | | set_mac_dst) | | | | |
1621 +-----------------------+-----------------+-----------------+
1622 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1623 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1624 | | | | rdma-core 24 | | rdma-core 24 |
1625 | | | | ConnectX-5 | | ConnectX-5 |
1626 +-----------------------+-----------------+-----------------+
1627 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1628 | | | OFED 4.7-1 | | OFED 4.7-1 |
1629 | | | rdma-core 24 | | N/A |
1630 | | | ConnectX-5 | | ConnectX-5 |
1631 +-----------------------+-----------------+-----------------+
1632 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1633 | | | OFED 4.6 | | OFED 4.5 |
1634 | | | rdma-core 24 | | rdma-core 23 |
1635 | | | ConnectX-5 | | ConnectX-4 |
1636 +-----------------------+-----------------+-----------------+
1637 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1638 | | | OFED 4.7-3 | | OFED 4.7-3 |
1639 | | | rdma-core 26 | | rdma-core 26 |
1640 | | | ConnectX-5 | | ConnectX-5 |
1641 +-----------------------+-----------------+-----------------+
1642 | Port ID | | DPDK 19.05 | | N/A |
1643 | | | OFED 4.7-1 | | N/A |
1644 | | | rdma-core 24 | | N/A |
1645 | | | ConnectX-5 | | N/A |
1646 +-----------------------+-----------------+-----------------+
1647 | Hairpin | | | | DPDK 19.11 |
1648 | | | N/A | | OFED 4.7-3 |
1649 | | | | | rdma-core 26 |
1650 | | | | | ConnectX-5 |
1651 +-----------------------+-----------------+-----------------+
1652 | 2-port Hairpin | | | | DPDK 20.11 |
1653 | | | N/A | | OFED 5.1-2 |
1655 | | | | | ConnectX-5 |
1656 +-----------------------+-----------------+-----------------+
1657 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1658 | | | OFED 4.7-3 | | OFED 4.7-3 |
1659 | | | rdma-core 26 | | rdma-core 26 |
1660 | | | ConnectX-5 | | ConnectX-5 |
1661 +-----------------------+-----------------+-----------------+
1662 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1663 | | | OFED 5.1-2 | | OFED 5.1-2 |
1664 | | | rdma-core 32 | | N/A |
1665 | | | ConnectX-5 | | ConnectX-5 |
1666 +-----------------------+-----------------+-----------------+
1667 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1668 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1669 | | | rdma-core 35 | | rdma-core 35 |
1670 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1671 +-----------------------+-----------------+-----------------+
1672 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1673 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1674 | | | rdma-core 34 | | rdma-core 34 |
1675 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1676 +-----------------------+-----------------+-----------------+
1677 | Modify Field | | DPDK 21.02 | | DPDK 21.02 |
1678 | | | OFED 5.2 | | OFED 5.2 |
1679 | | | rdma-core 35 | | rdma-core 35 |
1680 | | | ConnectX-5 | | ConnectX-5 |
1681 +-----------------------+-----------------+-----------------+
1683 .. table:: Minimal SW/HW versions for shared action offload
1686 +-----------------------+-----------------+-----------------+
1687 | Shared Action | with E-Switch | with NIC |
1688 +=======================+=================+=================+
1689 | RSS | | | | DPDK 20.11 |
1690 | | | N/A | | OFED 5.2 |
1691 | | | | | rdma-core 33 |
1692 | | | | | ConnectX-5 |
1693 +-----------------------+-----------------+-----------------+
1694 | Age | | DPDK 20.11 | | DPDK 20.11 |
1695 | | | OFED 5.2 | | OFED 5.2 |
1696 | | | rdma-core 32 | | rdma-core 32 |
1697 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1698 +-----------------------+-----------------+-----------------+
1703 MARK and META items are interrelated with datapath - they might move from/to
1704 the applications in mbuf fields. Hence, zero value for these items has the
1705 special meaning - it means "no metadata are provided", not zero values are
1706 treated by applications and PMD as valid ones.
1708 Moreover in the flow engine domain the value zero is acceptable to match and
1709 set, and we should allow to specify zero values as rte_flow parameters for the
1710 META and MARK items and actions. In the same time zero mask has no meaning and
1711 should be rejected on validation stage.
1716 Flows are not cached in the driver.
1717 When stopping a device port, all the flows created on this port from the
1718 application will be flushed automatically in the background.
1719 After stopping the device port, all flows on this port become invalid and
1720 not represented in the system.
1721 All references to these flows held by the application should be discarded
1722 directly but neither destroyed nor flushed.
1724 The application should re-create the flows as required after the port restart.
1729 Compared to librte_net_mlx4 that implements a single RSS configuration per
1730 port, librte_net_mlx5 supports per-protocol RSS configuration.
1732 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1733 command-line parameter to enable additional protocols (UDP and TCP as well
1734 as IP), the following commands must be entered from its CLI to get the same
1735 behavior as librte_net_mlx4::
1738 > port config all rss all
1744 This section demonstrates how to launch **testpmd** with Mellanox
1745 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1747 #. Load the kernel modules::
1749 modprobe -a ib_uverbs mlx5_core mlx5_ib
1751 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1754 /etc/init.d/openibd restart
1758 User space I/O kernel modules (uio and igb_uio) are not used and do
1759 not have to be loaded.
1761 #. Make sure Ethernet interfaces are in working order and linked to kernel
1762 verbs. Related sysfs entries should be present::
1764 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1773 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1776 for intf in eth2 eth3 eth4 eth5;
1778 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1781 sed -n 's,.*/\(.*\),-a \1,p'
1790 #. Request huge pages::
1792 dpdk-hugepages.py --setup 2G
1794 #. Start testpmd with basic parameters::
1796 dpdk-testpmd -l 8-15 -n 4 -a 05:00.0 -a 05:00.1 -a 06:00.0 -a 06:00.1 -- --rxq=2 --txq=2 -i
1801 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1802 EAL: probe driver: 15b3:1013 librte_net_mlx5
1803 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1804 PMD: librte_net_mlx5: 1 port(s) detected
1805 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1806 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1807 EAL: probe driver: 15b3:1013 librte_net_mlx5
1808 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1809 PMD: librte_net_mlx5: 1 port(s) detected
1810 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1811 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1812 EAL: probe driver: 15b3:1013 librte_net_mlx5
1813 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1814 PMD: librte_net_mlx5: 1 port(s) detected
1815 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1816 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1817 EAL: probe driver: 15b3:1013 librte_net_mlx5
1818 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1819 PMD: librte_net_mlx5: 1 port(s) detected
1820 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1821 Interactive-mode selected
1822 Configuring Port 0 (socket 0)
1823 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1824 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1825 Port 0: E4:1D:2D:E7:0C:FE
1826 Configuring Port 1 (socket 0)
1827 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1828 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1829 Port 1: E4:1D:2D:E7:0C:FF
1830 Configuring Port 2 (socket 0)
1831 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1832 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1833 Port 2: E4:1D:2D:E7:0C:FA
1834 Configuring Port 3 (socket 0)
1835 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1836 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1837 Port 3: E4:1D:2D:E7:0C:FB
1838 Checking link statuses...
1839 Port 0 Link Up - speed 40000 Mbps - full-duplex
1840 Port 1 Link Up - speed 40000 Mbps - full-duplex
1841 Port 2 Link Up - speed 10000 Mbps - full-duplex
1842 Port 3 Link Up - speed 10000 Mbps - full-duplex
1849 This section demonstrates how to dump flows. Currently, it's possible to dump
1850 all flows with assistance of external tools.
1852 #. 2 ways to get flow raw file:
1854 - Using testpmd CLI:
1856 .. code-block:: console
1859 testpmd> flow dump <port> all <output_file>
1861 testpmd> flow dump <port> rule <rule_id> <output_file>
1863 - call rte_flow_dev_dump api:
1865 .. code-block:: console
1867 rte_flow_dev_dump(port, flow, file, NULL);
1869 #. Dump human-readable flows from raw file:
1871 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1873 .. code-block:: console
1875 mlx_steering_dump.py -f <output_file> -flowptr <flow_ptr>