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 - Promiscuous mode on PF and VF.
78 - Multicast promiscuous mode on PF and VF.
79 - Hardware checksum offloads.
80 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
82 - Flow API, including :ref:`flow_isolated_mode`.
84 - KVM and VMware ESX SR-IOV modes are supported.
85 - RSS hash result is supported.
86 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
87 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
89 - Statistics query including Basic, Extended and per queue.
91 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
92 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
93 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
94 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
95 - Flow insertion rate of more then million flows per second, when using Direct Rules.
96 - Support for multiple rte_flow groups.
97 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
100 - Multiple-thread flow insertion.
101 - Matching on GTP extension header with raw encap/decap action.
102 - Matching on Geneve TLV option header with raw encap/decap action.
103 - RSS support in sample action.
104 - E-Switch mirroring and jump.
105 - E-Switch mirroring and modify.
106 - 21844 flow priorities for ingress or egress flow groups greater than 0 and for any transfer
114 On Windows, the features are limited:
116 - Promiscuous mode is not supported
117 - The following rules are supported:
119 - IPv4/UDP with CVLAN filtering
120 - Unicast MAC filtering
122 - For secondary process:
124 - Forked secondary process not supported.
125 - External memory unregistered in EAL memseg list cannot be used for DMA
126 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
127 primary process and remapped to the same virtual address in secondary
128 process. If the external memory is registered by primary process but has
129 different virtual address in secondary process, unexpected error may happen.
131 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
132 specific VLAN will match for VLAN packets as well:
134 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
135 Meaning, the flow rule::
137 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
139 Will only match vlan packets with vid=3. and the flow rule::
141 flow create 0 ingress pattern eth / ipv4 / end ...
143 Will match any ipv4 packet (VLAN included).
145 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
147 - 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.
150 flow create 0 ingress pattern eth / ipv4 / end ...
152 Will match any ipv4 packet.
155 flow create 0 ingress pattern eth / vlan / end ...
156 flow create 0 ingress pattern eth has_vlan is 1 / end ...
157 flow create 0 ingress pattern eth type is 0x8100 / end ...
159 Will match single-tagged packets only, with any VLAN ID value.
162 flow create 0 ingress pattern eth type is 0x88A8 / end ...
163 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
165 Will match multi-tagged packets only, with any VLAN ID value.
167 - A flow pattern with 2 sequential VLAN items is not supported.
169 - VLAN pop offload command:
171 - Flow rules having a VLAN pop offload command as one of their actions and
172 are lacking a match on VLAN as one of their items are not supported.
173 - The command is not supported on egress traffic.
175 - VLAN push offload is not supported on ingress traffic.
177 - VLAN set PCP offload is not supported on existing headers.
179 - A multi segment packet must have not more segments than reported by dev_infos_get()
180 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
181 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
182 inline settings) to 58.
184 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
185 to 0 are not supported.
187 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
189 - Match on Geneve header supports the following fields only:
196 - Match on Geneve TLV option is supported on the following fields:
203 Only one Class/Type/Length Geneve TLV option is supported per shared device.
204 Class/Type/Length fields must be specified as well as masks.
205 Class/Type/Length specified masks must be full.
206 Matching Geneve TLV option without specifying data is not supported.
207 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
209 - VF: flow rules created on VF devices can only match traffic targeted at the
210 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
212 - Match on GTP tunnel header item supports the following fields only:
214 - v_pt_rsv_flags: E flag, S flag, PN flag
218 - Match on GTP extension header only for GTP PDU session container (next
219 extension header type = 0x85).
220 - Match on GTP extension header is not supported in group 0.
222 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
223 The flows within group 0 and set metadata action are rejected by hardware.
227 MAC addresses not already present in the bridge table of the associated
228 kernel network device will be added and cleaned up by the PMD when closing
229 the device. In case of ungraceful program termination, some entries may
230 remain present and should be removed manually by other means.
232 - Buffer split offload is supported with regular Rx burst routine only,
233 no MPRQ feature or vectorized code can be engaged.
235 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
236 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
237 ol_flags. As the mempool for the external buffer is managed by PMD, all the
238 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
239 the external buffers will be freed by PMD and the application which still
240 holds the external buffers may be corrupted.
242 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
243 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
244 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
246 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
247 and allmulticast mode are both set to off.
248 To receive IPv6 Multicast messages on VM, explicitly set the relevant
249 MAC address using rte_eth_dev_mac_addr_add() API.
251 - To support a mixed traffic pattern (some buffers from local host memory, some
252 buffers from other devices) with high bandwidth, a mbuf flag is used.
254 An application hints the PMD whether or not it should try to inline the
255 given mbuf data buffer. PMD should do the best effort to act upon this request.
257 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
258 registered by application with rte_mbuf_dynflag_register(). This flag is
259 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
260 which is intended to be used by the application.
262 To query the supported specific flags in runtime,
263 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
264 currently (over present hardware and configuration) supported specific flags.
265 The "not inline hint" feature operating flow is the following one:
268 - probe the devices, ports are created
269 - query the port capabilities
270 - if port supporting the feature is found
271 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
272 - application starts the ports
273 - on ``dev_start()`` PMD checks whether the feature flag is registered and
274 enables the feature support in datapath
275 - application might set the registered flag bit in ``ol_flags`` field
276 of mbuf being sent and PMD will handle ones appropriately.
278 - The amount of descriptors in Tx queue may be limited by data inline settings.
279 Inline data require the more descriptor building blocks and overall block
280 amount may exceed the hardware supported limits. The application should
281 reduce the requested Tx size or adjust data inline settings with
282 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
284 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
285 parameter should be specified.
286 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
287 being sent it tries to synchronize the time of packet appearing on
288 the wire with the specified packet timestamp. It the specified one
289 is in the past it should be ignored, if one is in the distant future
290 it should be capped with some reasonable value (in range of seconds).
291 These specific cases ("too late" and "distant future") can be optionally
292 reported via device xstats to assist applications to detect the
293 time-related problems.
295 The timestamp upper "too-distant-future" limit
296 at the moment of invoking the Tx burst routine
297 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
298 Please note, for the testpmd txonly mode,
299 the limit is deduced from the expression::
301 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
303 There is no any packet reordering according timestamps is supposed,
304 neither within packet burst, nor between packets, it is an entirely
305 application responsibility to generate packets and its timestamps
306 in desired order. The timestamps can be put only in the first packet
307 in the burst providing the entire burst scheduling.
309 - E-Switch decapsulation Flow:
311 - can be applied to PF port only.
312 - must specify VF port action (packet redirection from PF to VF).
313 - optionally may specify tunnel inner source and destination MAC addresses.
315 - E-Switch encapsulation Flow:
317 - can be applied to VF ports only.
318 - must specify PF port action (packet redirection from VF to PF).
322 - The input buffer, used as outer header, is not validated.
326 - The decapsulation is always done up to the outermost tunnel detected by the HW.
327 - The input buffer, providing the removal size, is not validated.
328 - The buffer size must match the length of the headers to be removed.
330 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
331 mutually exclusive features which cannot be supported together
332 (see :ref:`mlx5_firmware_config`).
336 - Requires DevX and DV flow to be enabled.
337 - KEEP_CRC offload cannot be supported with LRO.
338 - The first mbuf length, without head-room, must be big enough to include the
340 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
341 it with size limited to max LRO size, not to max RX packet length.
342 - LRO can be used with outer header of TCP packets of the standard format:
343 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
345 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
346 - LRO packet aggregation is performed by HW only for packet size larger than
347 ``lro_min_mss_size``. This value is reported on device start, when debug
352 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
353 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
354 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
358 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and E-Switch steering domain.
359 - The E-Switch Sample flow must have the eswitch_manager VPORT destination (PF or ECPF) and no additional actions.
360 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as first action in the E-Switch egress flow if with header modify or encapsulation actions.
362 - IPv6 header item 'proto' field, indicating the next header protocol, should
363 not be set as extension header.
364 In case the next header is an extension header, it should not be specified in
365 IPv6 header item 'proto' field.
366 The last extension header item 'next header' field can specify the following
367 header protocol type.
371 - 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.
372 - Hairpin in switchdev SR-IOV mode is not supported till now.
377 MLX5 supports various methods to report statistics:
379 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.
381 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.
383 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.
391 The ibverbs libraries can be linked with this PMD in a number of ways,
392 configured by the ``ibverbs_link`` build option:
394 - ``shared`` (default): the PMD depends on some .so files.
396 - ``dlopen``: Split the dependencies glue in a separate library
397 loaded when needed by dlopen.
398 It make dependencies on libibverbs and libmlx4 optional,
399 and has no performance impact.
401 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
402 in the PMD shared library or the executable static binary.
404 Environment variables
405 ~~~~~~~~~~~~~~~~~~~~~
409 A list of directories in which to search for the rdma-core "glue" plug-in,
410 separated by colons or semi-colons.
412 - ``MLX5_SHUT_UP_BF``
414 Configures HW Tx doorbell register as IO-mapped.
416 By default, the HW Tx doorbell is configured as a write-combining register.
417 The register would be flushed to HW usually when the write-combining buffer
418 becomes full, but it depends on CPU design.
420 Except for vectorized Tx burst routines, a write memory barrier is enforced
421 after updating the register so that the update can be immediately visible to
424 When vectorized Tx burst is called, the barrier is set only if the burst size
425 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
426 variable will bring better latency even though the maximum throughput can
429 Run-time configuration
430 ~~~~~~~~~~~~~~~~~~~~~~
432 - librte_net_mlx5 brings kernel network interfaces up during initialization
433 because it is affected by their state. Forcing them down prevents packets
436 - **ethtool** operations on related kernel interfaces also affect the PMD.
441 In order to run as a non-root user,
442 some capabilities must be granted to the application::
444 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
446 Below are the reasons of the need for each capability:
449 When using physical addresses (PA mode), with Linux >= 4.0,
450 for access to ``/proc/self/pagemap``.
453 For device configuration.
456 For raw ethernet queue allocation through kernel driver.
459 For DMA memory pinning.
464 - ``rxq_cqe_comp_en`` parameter [int]
466 A nonzero value enables the compression of CQE on RX side. This feature
467 allows to save PCI bandwidth and improve performance. Enabled by default.
468 Different compression formats are supported in order to achieve the best
469 performance for different traffic patterns. Hash RSS format is the default.
471 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
472 better compression rate in case of RTE Flow Mark traffic.
473 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
474 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
475 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
479 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
480 ConnectX-6 Lx, BlueField and BlueField-2.
481 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
482 ConnectX-6 Lx, BlueField and BlueField-2.
484 - ``rxq_pkt_pad_en`` parameter [int]
486 A nonzero value enables padding Rx packet to the size of cacheline on PCI
487 transaction. This feature would waste PCI bandwidth but could improve
488 performance by avoiding partial cacheline write which may cause costly
489 read-modify-copy in memory transaction on some architectures. Disabled by
494 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
495 ConnectX-6 Lx, BlueField and BlueField-2.
496 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
497 ConnectX-6 Lx, BlueField and BlueField-2.
499 - ``mprq_en`` parameter [int]
501 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
502 configured as Multi-Packet RQ if the total number of Rx queues is
503 ``rxqs_min_mprq`` or more. Disabled by default.
505 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
506 by posting a single large buffer for multiple packets. Instead of posting a
507 buffers per a packet, one large buffer is posted in order to receive multiple
508 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
509 and each stride receives one packet. MPRQ can improve throughput for
510 small-packet traffic.
512 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
513 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
514 configure large stride size enough to accommodate max_rx_pkt_len as long as
515 device allows. Note that this can waste system memory compared to enabling Rx
516 scatter and multi-segment packet.
518 - ``mprq_log_stride_num`` parameter [int]
520 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
521 strides can reduce PCIe traffic further. If configured value is not in the
522 range of device capability, the default value will be set with a warning
523 message. The default value is 4 which is 16 strides per a buffer, valid only
524 if ``mprq_en`` is set.
526 The size of Rx queue should be bigger than the number of strides.
528 - ``mprq_log_stride_size`` parameter [int]
530 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
531 stride size can save some memory and reduce probability of a depletion of all
532 available strides due to unreleased packets by an application. If configured
533 value is not in the range of device capability, the default value will be set
534 with a warning message. The default value is 11 which is 2048 bytes per a
535 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
536 it is possible for a packet to span across multiple strides. This mode allows
537 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
538 of a packet if Rx scatter is configured) may be required in case there is no
539 space left for a head room at the end of a stride which incurs some
542 - ``mprq_max_memcpy_len`` parameter [int]
544 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
545 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
546 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
547 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
548 A mempool for external buffers will be allocated and managed by PMD. If Rx
549 packet is externally attached, ol_flags field of the mbuf will have
550 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
551 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
553 - ``rxqs_min_mprq`` parameter [int]
555 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
556 greater or equal to this value. The default value is 12, valid only if
559 - ``txq_inline`` parameter [int]
561 Amount of data to be inlined during TX operations. This parameter is
562 deprecated and converted to the new parameter ``txq_inline_max`` providing
563 partial compatibility.
565 - ``txqs_min_inline`` parameter [int]
567 Enable inline data send only when the number of TX queues is greater or equal
570 This option should be used in combination with ``txq_inline_max`` and
571 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
573 If this option is not specified the default value 16 is used for BlueField
574 and 8 for other platforms
576 The data inlining consumes the CPU cycles, so this option is intended to
577 auto enable inline data if we have enough Tx queues, which means we have
578 enough CPU cores and PCI bandwidth is getting more critical and CPU
579 is not supposed to be bottleneck anymore.
581 The copying data into WQE improves latency and can improve PPS performance
582 when PCI back pressure is detected and may be useful for scenarios involving
583 heavy traffic on many queues.
585 Because additional software logic is necessary to handle this mode, this
586 option should be used with care, as it may lower performance when back
587 pressure is not expected.
589 If inline data are enabled it may affect the maximal size of Tx queue in
590 descriptors because the inline data increase the descriptor size and
591 queue size limits supported by hardware may be exceeded.
593 - ``txq_inline_min`` parameter [int]
595 Minimal amount of data to be inlined into WQE during Tx operations. NICs
596 may require this minimal data amount to operate correctly. The exact value
597 may depend on NIC operation mode, requested offloads, etc. It is strongly
598 recommended to omit this parameter and use the default values. Anyway,
599 applications using this parameter should take into consideration that
600 specifying an inconsistent value may prevent the NIC from sending packets.
602 If ``txq_inline_min`` key is present the specified value (may be aligned
603 by the driver in order not to exceed the limits and provide better descriptor
604 space utilization) will be used by the driver and it is guaranteed that
605 requested amount of data bytes are inlined into the WQE beside other inline
606 settings. This key also may update ``txq_inline_max`` value (default
607 or specified explicitly in devargs) to reserve the space for inline data.
609 If ``txq_inline_min`` key is not present, the value may be queried by the
610 driver from the NIC via DevX if this feature is available. If there is no DevX
611 enabled/supported the value 18 (supposing L2 header including VLAN) is set
612 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
613 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
616 For ConnectX-4 NIC, driver does not allow specifying value below 18
617 (minimal L2 header, including VLAN), error will be raised.
619 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
620 it is not recommended and may prevent NIC from sending packets over
623 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
624 Multi-Packet Write), because last one does not support partial packet inlining.
625 This is not very critical due to minimal data inlining is mostly required
626 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
628 - ``txq_inline_max`` parameter [int]
630 Specifies the maximal packet length to be completely inlined into WQE
631 Ethernet Segment for ordinary SEND method. If packet is larger than specified
632 value, the packet data won't be copied by the driver at all, data buffer
633 is addressed with a pointer. If packet length is less or equal all packet
634 data will be copied into WQE. This may improve PCI bandwidth utilization for
635 short packets significantly but requires the extra CPU cycles.
637 The data inline feature is controlled by number of Tx queues, if number of Tx
638 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
639 is engaged, if there are not enough Tx queues (which means not enough CPU cores
640 and CPU resources are scarce), data inline is not performed by the driver.
641 Assigning ``txqs_min_inline`` with zero always enables the data inline.
643 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
644 by the driver in order not to exceed the limit (930 bytes) and to provide better
645 WQE space filling without gaps, the adjustment is reflected in the debug log.
646 Also, the default value (290) may be decreased in run-time if the large transmit
647 queue size is requested and hardware does not support enough descriptor
648 amount, in this case warning is emitted. If ``txq_inline_max`` key is
649 specified and requested inline settings can not be satisfied then error
652 - ``txq_inline_mpw`` parameter [int]
654 Specifies the maximal packet length to be completely inlined into WQE for
655 Enhanced MPW method. If packet is large the specified value, the packet data
656 won't be copied, and data buffer is addressed with pointer. If packet length
657 is less or equal, all packet data will be copied into WQE. This may improve PCI
658 bandwidth utilization for short packets significantly but requires the extra
661 The data inline feature is controlled by number of TX queues, if number of Tx
662 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
663 is engaged, if there are not enough Tx queues (which means not enough CPU cores
664 and CPU resources are scarce), data inline is not performed by the driver.
665 Assigning ``txqs_min_inline`` with zero always enables the data inline.
667 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
668 by the driver in order not to exceed the limit (930 bytes) and to provide better
669 WQE space filling without gaps, the adjustment is reflected in the debug log.
670 Due to multiple packets may be included to the same WQE with Enhanced Multi
671 Packet Write Method and overall WQE size is limited it is not recommended to
672 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
673 may be decreased in run-time if the large transmit queue size is requested
674 and hardware does not support enough descriptor amount, in this case warning
675 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
676 settings can not be satisfied then error will be raised.
678 - ``txqs_max_vec`` parameter [int]
680 Enable vectorized Tx only when the number of TX queues is less than or
681 equal to this value. This parameter is deprecated and ignored, kept
682 for compatibility issue to not prevent driver from probing.
684 - ``txq_mpw_hdr_dseg_en`` parameter [int]
686 A nonzero value enables including two pointers in the first block of TX
687 descriptor. The parameter is deprecated and ignored, kept for compatibility
690 - ``txq_max_inline_len`` parameter [int]
692 Maximum size of packet to be inlined. This limits the size of packet to
693 be inlined. If the size of a packet is larger than configured value, the
694 packet isn't inlined even though there's enough space remained in the
695 descriptor. Instead, the packet is included with pointer. This parameter
696 is deprecated and converted directly to ``txq_inline_mpw`` providing full
697 compatibility. Valid only if eMPW feature is engaged.
699 - ``txq_mpw_en`` parameter [int]
701 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
702 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
703 eMPW allows the Tx burst function to pack up multiple packets
704 in a single descriptor session in order to save PCI bandwidth
705 and improve performance at the cost of a slightly higher CPU usage.
706 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
707 Tx burst function copies entire packet data on to Tx descriptor
708 instead of including pointer of packet.
710 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
711 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
712 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
713 option or reported by the NIC, the eMPW feature is disengaged.
715 - ``tx_db_nc`` parameter [int]
717 The rdma core library can map doorbell register in two ways, depending on the
718 environment variable "MLX5_SHUT_UP_BF":
720 - As regular cached memory (usually with write combining attribute), if the
721 variable is either missing or set to zero.
722 - As non-cached memory, if the variable is present and set to not "0" value.
724 The type of mapping may slightly affect the Tx performance, the optimal choice
725 is strongly relied on the host architecture and should be deduced practically.
727 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
728 memory (with write combining), the PMD will perform the extra write memory barrier
729 after writing to doorbell, it might increase the needed CPU clocks per packet
730 to send, but latency might be improved.
732 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
733 cached memory, the PMD will not perform the extra write memory barrier
734 after writing to doorbell, on some architectures it might improve the
737 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
738 memory, the PMD will use heuristics to decide whether write memory barrier
739 should be performed. For bursts with size multiple of recommended one (64 pkts)
740 it is supposed the next burst is coming and no need to issue the extra memory
741 barrier (it is supposed to be issued in the next coming burst, at least after
742 descriptor writing). It might increase latency (on some hosts till next
743 packets transmit) and should be used with care.
745 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
746 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
747 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
749 - ``tx_pp`` parameter [int]
751 If a nonzero value is specified the driver creates all necessary internal
752 objects to provide accurate packet send scheduling on mbuf timestamps.
753 The positive value specifies the scheduling granularity in nanoseconds,
754 the packet send will be accurate up to specified digits. The allowed range is
755 from 500 to 1 million of nanoseconds. The negative value specifies the module
756 of granularity and engages the special test mode the check the schedule rate.
757 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
760 - ``tx_skew`` parameter [int]
762 The parameter adjusts the send packet scheduling on timestamps and represents
763 the average delay between beginning of the transmitting descriptor processing
764 by the hardware and appearance of actual packet data on the wire. The value
765 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
766 specified. The default value is zero.
768 - ``tx_vec_en`` parameter [int]
770 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
771 ConnectX-6 Lx, BlueField and BlueField-2 NICs
772 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
773 The parameter is deprecated and ignored.
775 - ``rx_vec_en`` parameter [int]
777 A nonzero value enables Rx vector if the port is not configured in
778 multi-segment otherwise this parameter is ignored.
782 - ``vf_nl_en`` parameter [int]
784 A nonzero value enables Netlink requests from the VF to add/remove MAC
785 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
786 Otherwise the relevant configuration must be run with Linux iproute2 tools.
787 This is a prerequisite to receive this kind of traffic.
789 Enabled by default, valid only on VF devices ignored otherwise.
791 - ``l3_vxlan_en`` parameter [int]
793 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
794 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
795 parameter. This is a prerequisite to receive this kind of traffic.
799 - ``dv_xmeta_en`` parameter [int]
801 A nonzero value enables extensive flow metadata support if device is
802 capable and driver supports it. This can enable extensive support of
803 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
804 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
806 There are some possible configurations, depending on parameter value:
808 - 0, this is default value, defines the legacy mode, the ``MARK`` and
809 ``META`` related actions and items operate only within NIC Tx and
810 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
811 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
812 item is 32 bits wide and match supported on egress only.
814 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
815 related actions and items operate within all supported steering domains,
816 including FDB, ``MARK`` and ``META`` information may cross the domain
817 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
818 depends on kernel and firmware configurations and might be 0, 16 or
819 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
820 compatibility, the actual width of data transferred to the FDB domain
821 depends on kernel configuration and may be vary. The actual supported
822 width can be retrieved in runtime by series of rte_flow_validate()
825 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
826 related actions and items operate within all supported steering domains,
827 including FDB, ``MARK`` and ``META`` information may cross the domain
828 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
829 depends on kernel and firmware configurations and might be 0, 16 or
830 24 bits. The actual supported width can be retrieved in runtime by
831 series of rte_flow_validate() trials.
833 - 3, this engages tunnel offload mode. In E-Switch configuration, that
834 mode implicitly activates ``dv_xmeta_en=1``.
836 +------+-----------+-----------+-------------+-------------+
837 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
838 +======+===========+===========+=============+=============+
839 | 0 | 24 bits | 32 bits | 32 bits | no |
840 +------+-----------+-----------+-------------+-------------+
841 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
842 +------+-----------+-----------+-------------+-------------+
843 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
844 +------+-----------+-----------+-------------+-------------+
846 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
847 ignored and the device is configured to operate in legacy mode (0).
849 Disabled by default (set to 0).
851 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
852 of the extensive metadata features. The legacy Verbs supports FLAG and
853 MARK metadata actions over NIC Rx steering domain only.
855 The setting MARK or META value to zero means there is no item provided and
856 receiving datapath will not report in mbufs these items are present.
858 For the MARK action the last 16 values in the full range are reserved for
859 internal PMD purposes (to emulate FLAG action). The valid range for the
860 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
861 for the 24-bit mode, the flows with the MARK action value outside
862 the specified range will be rejected.
864 - ``dv_flow_en`` parameter [int]
866 A nonzero value enables the DV flow steering assuming it is supported
867 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
869 Enabled by default if supported.
871 - ``dv_esw_en`` parameter [int]
873 A nonzero value enables E-Switch using Direct Rules.
875 Enabled by default if supported.
877 - ``lacp_by_user`` parameter [int]
879 A nonzero value enables the control of LACP traffic by the user application.
880 When a bond exists in the driver, by default it should be managed by the
881 kernel and therefore LACP traffic should be steered to the kernel.
882 If this devarg is set to 1 it will allow the user to manage the bond by
883 itself and not steer LACP traffic to the kernel.
885 Disabled by default (set to 0).
887 - ``mr_ext_memseg_en`` parameter [int]
889 A nonzero value enables extending memseg when registering DMA memory. If
890 enabled, the number of entries in MR (Memory Region) lookup table on datapath
891 is minimized and it benefits performance. On the other hand, it worsens memory
892 utilization because registered memory is pinned by kernel driver. Even if a
893 page in the extended chunk is freed, that doesn't become reusable until the
894 entire memory is freed.
898 - ``representor`` parameter [list]
900 This parameter can be used to instantiate DPDK Ethernet devices from
901 existing port (or VF) representors configured on the device.
903 It is a standard parameter whose format is described in
904 :ref:`ethernet_device_standard_device_arguments`.
906 For instance, to probe port representors 0 through 2::
910 - ``max_dump_files_num`` parameter [int]
912 The maximum number of files per PMD entity that may be created for debug information.
913 The files will be created in /var/log directory or in current directory.
915 set to 128 by default.
917 - ``lro_timeout_usec`` parameter [int]
919 The maximum allowed duration of an LRO session, in micro-seconds.
920 PMD will set the nearest value supported by HW, which is not bigger than
921 the input ``lro_timeout_usec`` value.
922 If this parameter is not specified, by default PMD will set
923 the smallest value supported by HW.
925 - ``hp_buf_log_sz`` parameter [int]
927 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
928 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
929 The capacity of the value is specified by the firmware and the initialization
930 will get a failure if it is out of scope.
931 The range of the value is from 11 to 19 right now, and the supported frame
932 size of a single packet for hairpin is from 512B to 128KB. It might change if
933 different firmware release is being used. By using a small value, it could
934 reduce memory consumption but not work with a large frame. If the value is
935 too large, the memory consumption will be high and some potential performance
936 degradation will be introduced.
937 By default, the PMD will set this value to 16, which means that 9KB jumbo
938 frames will be supported.
940 - ``reclaim_mem_mode`` parameter [int]
942 Cache some resources in flow destroy will help flow recreation more efficient.
943 While some systems may require the all the resources can be reclaimed after
945 The parameter ``reclaim_mem_mode`` provides the option for user to configure
946 if the resource cache is needed or not.
948 There are three options to choose:
950 - 0. It means the flow resources will be cached as usual. The resources will
951 be cached, helpful with flow insertion rate.
953 - 1. It will only enable the DPDK PMD level resources reclaim.
955 - 2. Both DPDK PMD level and rdma-core low level will be configured as
958 By default, the PMD will set this value to 0.
960 - ``sys_mem_en`` parameter [int]
962 A non-zero value enables the PMD memory management allocating memory
963 from system by default, without explicit rte memory flag.
965 By default, the PMD will set this value to 0.
967 - ``decap_en`` parameter [int]
969 Some devices do not support FCS (frame checksum) scattering for
970 tunnel-decapsulated packets.
971 If set to 0, this option forces the FCS feature and rejects tunnel
972 decapsulation in the flow engine for such devices.
974 By default, the PMD will set this value to 1.
976 .. _mlx5_firmware_config:
978 Firmware configuration
979 ~~~~~~~~~~~~~~~~~~~~~~
981 Firmware features can be configured as key/value pairs.
983 The command to set a value is::
985 mlxconfig -d <device> set <key>=<value>
987 The command to query a value is::
989 mlxconfig -d <device> query | grep <key>
991 The device name for the command ``mlxconfig`` can be either the PCI address,
992 or the mst device name found with::
996 Below are some firmware configurations listed.
1002 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1008 - maximum number of SR-IOV virtual functions::
1012 - enable DevX (required by Direct Rules and other features)::
1016 - aggressive CQE zipping::
1020 - L3 VXLAN and VXLAN-GPE destination UDP port::
1023 IP_OVER_VXLAN_PORT=<udp dport>
1025 - enable VXLAN-GPE tunnel flow matching::
1027 FLEX_PARSER_PROFILE_ENABLE=0
1029 FLEX_PARSER_PROFILE_ENABLE=2
1031 - enable IP-in-IP tunnel flow matching::
1033 FLEX_PARSER_PROFILE_ENABLE=0
1035 - enable MPLS flow matching::
1037 FLEX_PARSER_PROFILE_ENABLE=1
1039 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1041 FLEX_PARSER_PROFILE_ENABLE=2
1043 - enable Geneve flow matching::
1045 FLEX_PARSER_PROFILE_ENABLE=0
1047 FLEX_PARSER_PROFILE_ENABLE=1
1049 - enable Geneve TLV option flow matching::
1051 FLEX_PARSER_PROFILE_ENABLE=0
1053 - enable GTP flow matching::
1055 FLEX_PARSER_PROFILE_ENABLE=3
1057 - enable eCPRI flow matching::
1059 FLEX_PARSER_PROFILE_ENABLE=4
1065 This driver relies on external libraries and kernel drivers for resources
1066 allocations and initialization. The following dependencies are not part of
1067 DPDK and must be installed separately:
1071 User space Verbs framework used by librte_net_mlx5. This library provides
1072 a generic interface between the kernel and low-level user space drivers
1075 It allows slow and privileged operations (context initialization, hardware
1076 resources allocations) to be managed by the kernel and fast operations to
1077 never leave user space.
1081 Low-level user space driver library for Mellanox
1082 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1085 This library basically implements send/receive calls to the hardware
1088 - **Kernel modules**
1090 They provide the kernel-side Verbs API and low level device drivers that
1091 manage actual hardware initialization and resources sharing with user
1094 Unlike most other PMDs, these modules must remain loaded and bound to
1097 - mlx5_core: hardware driver managing Mellanox
1098 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1100 - mlx5_ib: InifiniBand device driver.
1101 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1103 - **Firmware update**
1105 Mellanox OFED/EN releases include firmware updates for
1106 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1108 Because each release provides new features, these updates must be applied to
1109 match the kernel modules and libraries they come with.
1113 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1119 Either RDMA Core library with a recent enough Linux kernel release
1120 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1123 RDMA Core with Linux Kernel
1124 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1126 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1127 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1128 (see `RDMA Core installation documentation`_)
1129 - When building for i686 use:
1131 - rdma-core version 18.0 or above built with 32bit support.
1132 - Kernel version 4.14.41 or above.
1134 - Starting with rdma-core v21, static libraries can be built::
1137 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1140 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1141 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1147 - Mellanox OFED version: **4.5** and above /
1148 Mellanox EN version: **4.5** and above
1151 - ConnectX-4: **12.21.1000** and above.
1152 - ConnectX-4 Lx: **14.21.1000** and above.
1153 - ConnectX-5: **16.21.1000** and above.
1154 - ConnectX-5 Ex: **16.21.1000** and above.
1155 - ConnectX-6: **20.27.0090** and above.
1156 - ConnectX-6 Dx: **22.27.0090** and above.
1157 - BlueField: **18.25.1010** and above.
1159 While these libraries and kernel modules are available on OpenFabrics
1160 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1161 managers on most distributions, this PMD requires Ethernet extensions that
1162 may not be supported at the moment (this is a work in progress).
1165 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1167 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1168 include the necessary support and should be used in the meantime. For DPDK,
1169 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1170 required from that distribution.
1174 Several versions of Mellanox OFED/EN are available. Installing the version
1175 this DPDK release was developed and tested against is strongly
1176 recommended. Please check the `linux prerequisites`_.
1178 Windows Prerequisites
1179 ---------------------
1181 This driver relies on external libraries and kernel drivers for resources
1182 allocations and initialization. The dependencies in the following sub-sections
1183 are not part of DPDK, and must be installed separately.
1185 Compilation Prerequisites
1186 ~~~~~~~~~~~~~~~~~~~~~~~~~
1188 DevX SDK installation
1189 ^^^^^^^^^^^^^^^^^^^^^
1191 The DevX SDK must be installed on the machine building the Windows PMD.
1192 Additional information can be found at
1193 `How to Integrate Windows DevX in Your Development Environment
1194 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1196 Runtime Prerequisites
1197 ~~~~~~~~~~~~~~~~~~~~~
1199 WinOF2 version 2.60 or higher must be installed on the machine.
1204 The driver can be downloaded from the following site:
1206 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1211 DevX for Windows must be enabled in the Windows registry.
1212 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1213 Additional information can be found in the WinOF2 user manual.
1218 The following Mellanox device families are supported by the same mlx5 driver:
1230 Below are detailed device names:
1232 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1233 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1234 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1235 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1236 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1237 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1238 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1239 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1240 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1241 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1242 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1243 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1244 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1245 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1246 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1247 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1248 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1249 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1250 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1251 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1252 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1253 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1254 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1255 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1256 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1257 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1259 Quick Start Guide on OFED/EN
1260 ----------------------------
1262 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1265 2. Install the required libraries and kernel modules either by installing
1266 only the required set, or by installing the entire Mellanox OFED/EN::
1268 ./mlnxofedinstall --upstream-libs --dpdk
1270 3. Verify the firmware is the correct one::
1274 4. Verify all ports links are set to Ethernet::
1276 mlxconfig -d <mst device> query | grep LINK_TYPE
1280 Link types may have to be configured to Ethernet::
1282 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1284 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1286 For hypervisors, verify SR-IOV is enabled on the NIC::
1288 mlxconfig -d <mst device> query | grep SRIOV_EN
1291 If needed, configure SR-IOV::
1293 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1294 mlxfwreset -d <mst device> reset
1296 5. Restart the driver::
1298 /etc/init.d/openibd restart
1302 service openibd restart
1304 If link type was changed, firmware must be reset as well::
1306 mlxfwreset -d <mst device> reset
1308 For hypervisors, after reset write the sysfs number of virtual functions
1311 To dynamically instantiate a given number of virtual functions (VFs)::
1313 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1315 6. Install DPDK and you are ready to go.
1316 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1318 Enable switchdev mode
1319 ---------------------
1321 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1322 Representor is a port in DPDK that is connected to a VF in such a way
1323 that assuming there are no offload flows, each packet that is sent from the VF
1324 will be received by the corresponding representor. While each packet that is
1325 sent to a representor will be received by the VF.
1326 This is very useful in case of SRIOV mode, where the first packet that is sent
1327 by the VF will be received by the DPDK application which will decide if this
1328 flow should be offloaded to the E-Switch. After offloading the flow packet
1329 that the VF that are matching the flow will not be received any more by
1330 the DPDK application.
1332 1. Enable SRIOV mode::
1334 mlxconfig -d <mst device> set SRIOV_EN=true
1336 2. Configure the max number of VFs::
1338 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1342 mlxfwreset -d <mst device> reset
1344 3. Configure the actual number of VFs::
1346 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1348 4. Unbind the device (can be rebind after the switchdev mode)::
1350 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1352 5. Enbale switchdev mode::
1354 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1359 1. Configure aggressive CQE Zipping for maximum performance::
1361 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1363 To set it back to the default CQE Zipping mode use::
1365 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1367 2. In case of virtualization:
1369 - Make sure that hypervisor kernel is 3.16 or newer.
1370 - Configure boot with ``iommu=pt``.
1371 - Use 1G huge pages.
1372 - Make sure to allocate a VM on huge pages.
1373 - Make sure to set CPU pinning.
1375 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1376 for better performance. For VMs, verify that the right CPU
1377 and NUMA node are pinned according to the above. Run::
1381 to identify the NUMA node to which the PCIe adapter is connected.
1383 4. If more than one adapter is used, and root complex capabilities allow
1384 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1385 it is recommended to locate both adapters on the same NUMA node.
1386 This in order to forward packets from one to the other without
1387 NUMA performance penalty.
1389 5. Disable pause frames::
1391 ethtool -A <netdev> rx off tx off
1393 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1394 via the BIOS configuration. Please contact you server provider for more
1395 information about the settings.
1399 On some machines, depends on the machine integrator, it is beneficial
1400 to set the PCI max read request parameter to 1K. This can be
1401 done in the following way:
1403 To query the read request size use::
1405 setpci -s <NIC PCI address> 68.w
1407 If the output is different than 3XXX, set it by::
1409 setpci -s <NIC PCI address> 68.w=3XXX
1411 The XXX can be different on different systems. Make sure to configure
1412 according to the setpci output.
1414 7. To minimize overhead of searching Memory Regions:
1416 - '--socket-mem' is recommended to pin memory by predictable amount.
1417 - Configure per-lcore cache when creating Mempools for packet buffer.
1418 - Refrain from dynamically allocating/freeing memory in run-time.
1423 There are multiple Rx burst functions with different advantages and limitations.
1425 .. table:: Rx burst functions
1427 +-------------------+------------------------+---------+-----------------+------+-------+
1428 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1429 | | | | || comp|| MTU |
1430 +===================+========================+=========+=================+======+=======+
1431 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1432 +-------------------+------------------------+---------+-----------------+------+-------+
1433 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1434 +-------------------+------------------------+---------+-----------------+------+-------+
1435 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1436 | || RxQs >= rxqs_min_mprq | | | | |
1437 +-------------------+------------------------+---------+-----------------+------+-------+
1438 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1439 | || mprq_en=1 | | | | |
1440 | || RxQs >= rxqs_min_mprq | | | | |
1441 +-------------------+------------------------+---------+-----------------+------+-------+
1443 .. _mlx5_offloads_support:
1445 Supported hardware offloads
1446 ---------------------------
1448 .. table:: Minimal SW/HW versions for queue offloads
1450 ============== ===== ===== ========= ===== ========== =============
1451 Offload DPDK Linux rdma-core OFED firmware hardware
1452 ============== ===== ===== ========= ===== ========== =============
1453 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1454 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1455 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1456 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1457 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1458 Buffer Split 20.11 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1459 ============== ===== ===== ========= ===== ========== =============
1461 .. table:: Minimal SW/HW versions for rte_flow offloads
1463 +-----------------------+-----------------+-----------------+
1464 | Offload | with E-Switch | with NIC |
1465 +=======================+=================+=================+
1466 | Count | | DPDK 19.05 | | DPDK 19.02 |
1467 | | | OFED 4.6 | | OFED 4.6 |
1468 | | | rdma-core 24 | | rdma-core 23 |
1469 | | | ConnectX-5 | | ConnectX-5 |
1470 +-----------------------+-----------------+-----------------+
1471 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1472 | | | OFED 4.6 | | OFED 4.5 |
1473 | | | rdma-core 24 | | rdma-core 23 |
1474 | | | ConnectX-5 | | ConnectX-4 |
1475 +-----------------------+-----------------+-----------------+
1476 | Queue / RSS | | | | DPDK 18.11 |
1477 | | | N/A | | OFED 4.5 |
1478 | | | | | rdma-core 23 |
1479 | | | | | ConnectX-4 |
1480 +-----------------------+-----------------+-----------------+
1481 | RSS shared action | | | | DPDK 20.11 |
1482 | | | N/A | | OFED 5.2 |
1483 | | | | | rdma-core 33 |
1484 | | | | | ConnectX-5 |
1485 +-----------------------+-----------------+-----------------+
1486 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1487 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1488 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1489 | | of_set_vlan_pcp / | | | | |
1490 | | of_set_vlan_vid) | | | | |
1491 +-----------------------+-----------------+-----------------+
1492 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1493 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1494 | | | rdma-core 24 | | rdma-core 23 |
1495 | | | ConnectX-5 | | ConnectX-5 |
1496 +-----------------------+-----------------+-----------------+
1497 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1498 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1499 | | | rdma-core 27 | | rdma-core 27 |
1500 | | | ConnectX-5 | | ConnectX-5 |
1501 +-----------------------+-----------------+-----------------+
1502 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1503 | | | OFED 5.1-2 | | OFED 5.1-2 |
1504 | | | rdma-core 32 | | N/A |
1505 | | | ConnectX-5 | | ConnectX-5 |
1506 +-----------------------+-----------------+-----------------+
1507 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1508 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1509 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1510 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1511 | | set_ipv6_dst / | | | | |
1512 | | set_tp_src / | | | | |
1513 | | set_tp_dst / | | | | |
1514 | | dec_ttl / | | | | |
1515 | | set_ttl / | | | | |
1516 | | set_mac_src / | | | | |
1517 | | set_mac_dst) | | | | |
1518 +-----------------------+-----------------+-----------------+
1519 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1520 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1521 | | | | rdma-core 24 | | rdma-core 24 |
1522 | | | | ConnectX-5 | | ConnectX-5 |
1523 +-----------------------+-----------------+-----------------+
1524 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1525 | | | OFED 4.7-1 | | OFED 4.7-1 |
1526 | | | rdma-core 24 | | N/A |
1527 | | | ConnectX-5 | | ConnectX-5 |
1528 +-----------------------+-----------------+-----------------+
1529 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1530 | | | OFED 4.6 | | OFED 4.5 |
1531 | | | rdma-core 24 | | rdma-core 23 |
1532 | | | ConnectX-5 | | ConnectX-4 |
1533 +-----------------------+-----------------+-----------------+
1534 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1535 | | | OFED 4.7-3 | | OFED 4.7-3 |
1536 | | | rdma-core 26 | | rdma-core 26 |
1537 | | | ConnectX-5 | | ConnectX-5 |
1538 +-----------------------+-----------------+-----------------+
1539 | Port ID | | DPDK 19.05 | | N/A |
1540 | | | OFED 4.7-1 | | N/A |
1541 | | | rdma-core 24 | | N/A |
1542 | | | ConnectX-5 | | N/A |
1543 +-----------------------+-----------------+-----------------+
1544 | Hairpin | | | | DPDK 19.11 |
1545 | | | N/A | | OFED 4.7-3 |
1546 | | | | | rdma-core 26 |
1547 | | | | | ConnectX-5 |
1548 +-----------------------+-----------------+-----------------+
1549 | 2-port Hairpin | | | | DPDK 20.11 |
1550 | | | N/A | | OFED 5.1-2 |
1552 | | | | | ConnectX-5 |
1553 +-----------------------+-----------------+-----------------+
1554 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1555 | | | OFED 4.7-3 | | OFED 4.7-3 |
1556 | | | rdma-core 26 | | rdma-core 26 |
1557 | | | ConnectX-5 | | ConnectX-5 |
1558 +-----------------------+-----------------+-----------------+
1559 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1560 | | | OFED 5.1-2 | | OFED 5.1-2 |
1561 | | | rdma-core 32 | | N/A |
1562 | | | ConnectX-5 | | ConnectX-5 |
1563 +-----------------------+-----------------+-----------------+
1564 | Age shared action | | DPDK 20.11 | | DPDK 20.11 |
1565 | | | OFED 5.2 | | OFED 5.2 |
1566 | | | rdma-core 32 | | rdma-core 32 |
1567 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1568 +-----------------------+-----------------+-----------------+
1569 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1570 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1571 | | | rdma-core 35 | | rdma-core 35 |
1572 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1573 +-----------------------+-----------------+-----------------+
1574 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1575 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1576 | | | rdma-core 34 | | rdma-core 34 |
1577 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1578 +-----------------------+-----------------+-----------------+
1583 MARK and META items are interrelated with datapath - they might move from/to
1584 the applications in mbuf fields. Hence, zero value for these items has the
1585 special meaning - it means "no metadata are provided", not zero values are
1586 treated by applications and PMD as valid ones.
1588 Moreover in the flow engine domain the value zero is acceptable to match and
1589 set, and we should allow to specify zero values as rte_flow parameters for the
1590 META and MARK items and actions. In the same time zero mask has no meaning and
1591 should be rejected on validation stage.
1596 Flows are not cached in the driver.
1597 When stopping a device port, all the flows created on this port from the
1598 application will be flushed automatically in the background.
1599 After stopping the device port, all flows on this port become invalid and
1600 not represented in the system.
1601 All references to these flows held by the application should be discarded
1602 directly but neither destroyed nor flushed.
1604 The application should re-create the flows as required after the port restart.
1609 Compared to librte_net_mlx4 that implements a single RSS configuration per
1610 port, librte_net_mlx5 supports per-protocol RSS configuration.
1612 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1613 command-line parameter to enable additional protocols (UDP and TCP as well
1614 as IP), the following commands must be entered from its CLI to get the same
1615 behavior as librte_net_mlx4::
1618 > port config all rss all
1624 This section demonstrates how to launch **testpmd** with Mellanox
1625 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1627 #. Load the kernel modules::
1629 modprobe -a ib_uverbs mlx5_core mlx5_ib
1631 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1634 /etc/init.d/openibd restart
1638 User space I/O kernel modules (uio and igb_uio) are not used and do
1639 not have to be loaded.
1641 #. Make sure Ethernet interfaces are in working order and linked to kernel
1642 verbs. Related sysfs entries should be present::
1644 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1653 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1656 for intf in eth2 eth3 eth4 eth5;
1658 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1661 sed -n 's,.*/\(.*\),-a \1,p'
1670 #. Request huge pages::
1672 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1674 #. Start testpmd with basic parameters::
1676 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
1681 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1682 EAL: probe driver: 15b3:1013 librte_net_mlx5
1683 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1684 PMD: librte_net_mlx5: 1 port(s) detected
1685 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1686 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1687 EAL: probe driver: 15b3:1013 librte_net_mlx5
1688 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1689 PMD: librte_net_mlx5: 1 port(s) detected
1690 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1691 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1692 EAL: probe driver: 15b3:1013 librte_net_mlx5
1693 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1694 PMD: librte_net_mlx5: 1 port(s) detected
1695 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1696 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1697 EAL: probe driver: 15b3:1013 librte_net_mlx5
1698 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1699 PMD: librte_net_mlx5: 1 port(s) detected
1700 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1701 Interactive-mode selected
1702 Configuring Port 0 (socket 0)
1703 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1704 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1705 Port 0: E4:1D:2D:E7:0C:FE
1706 Configuring Port 1 (socket 0)
1707 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1708 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1709 Port 1: E4:1D:2D:E7:0C:FF
1710 Configuring Port 2 (socket 0)
1711 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1712 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1713 Port 2: E4:1D:2D:E7:0C:FA
1714 Configuring Port 3 (socket 0)
1715 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1716 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1717 Port 3: E4:1D:2D:E7:0C:FB
1718 Checking link statuses...
1719 Port 0 Link Up - speed 40000 Mbps - full-duplex
1720 Port 1 Link Up - speed 40000 Mbps - full-duplex
1721 Port 2 Link Up - speed 10000 Mbps - full-duplex
1722 Port 3 Link Up - speed 10000 Mbps - full-duplex
1729 This section demonstrates how to dump flows. Currently, it's possible to dump
1730 all flows with assistance of external tools.
1732 #. 2 ways to get flow raw file:
1734 - Using testpmd CLI:
1736 .. code-block:: console
1738 testpmd> flow dump <port> <output_file>
1740 - call rte_flow_dev_dump api:
1742 .. code-block:: console
1744 rte_flow_dev_dump(port, file, NULL);
1746 #. Dump human-readable flows from raw file:
1748 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1750 .. code-block:: console
1752 mlx_steering_dump.py -f <output_file>