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.
108 On Windows, the features are limited:
110 - Promiscuous mode is not supported
111 - The following rules are supported:
113 - IPv4/UDP with CVLAN filtering
114 - Unicast MAC filtering
116 - For secondary process:
118 - Forked secondary process not supported.
119 - External memory unregistered in EAL memseg list cannot be used for DMA
120 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
121 primary process and remapped to the same virtual address in secondary
122 process. If the external memory is registered by primary process but has
123 different virtual address in secondary process, unexpected error may happen.
125 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
126 specific VLAN will match for VLAN packets as well:
128 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
129 Meaning, the flow rule::
131 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
133 Will only match vlan packets with vid=3. and the flow rule::
135 flow create 0 ingress pattern eth / ipv4 / end ...
137 Will match any ipv4 packet (VLAN included).
139 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
141 - 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.
144 flow create 0 ingress pattern eth / ipv4 / end ...
146 Will match any ipv4 packet.
149 flow create 0 ingress pattern eth / vlan / end ...
150 flow create 0 ingress pattern eth has_vlan is 1 / end ...
151 flow create 0 ingress pattern eth type is 0x8100 / end ...
153 Will match single-tagged packets only, with any VLAN ID value.
156 flow create 0 ingress pattern eth type is 0x88A8 / end ...
157 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
159 Will match multi-tagged packets only, with any VLAN ID value.
161 - A flow pattern with 2 sequential VLAN items is not supported.
163 - VLAN pop offload command:
165 - Flow rules having a VLAN pop offload command as one of their actions and
166 are lacking a match on VLAN as one of their items are not supported.
167 - The command is not supported on egress traffic.
169 - VLAN push offload is not supported on ingress traffic.
171 - VLAN set PCP offload is not supported on existing headers.
173 - A multi segment packet must have not more segments than reported by dev_infos_get()
174 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
175 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
176 inline settings) to 58.
178 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
179 to 0 are not supported.
181 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
183 - Match on Geneve header supports the following fields only:
189 Currently, the only supported options length value is 0.
191 - VF: flow rules created on VF devices can only match traffic targeted at the
192 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
194 - Match on GTP tunnel header item supports the following fields only:
196 - v_pt_rsv_flags: E flag, S flag, PN flag
200 - Match on GTP extension header only for GTP PDU session container (next
201 extension header type = 0x85).
202 - Match on GTP extension header is not supported in group 0.
204 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
205 The flows within group 0 and set metadata action are rejected by hardware.
209 MAC addresses not already present in the bridge table of the associated
210 kernel network device will be added and cleaned up by the PMD when closing
211 the device. In case of ungraceful program termination, some entries may
212 remain present and should be removed manually by other means.
214 - Buffer split offload is supported with regular Rx burst routine only,
215 no MPRQ feature or vectorized code can be engaged.
217 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
218 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
219 ol_flags. As the mempool for the external buffer is managed by PMD, all the
220 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
221 the external buffers will be freed by PMD and the application which still
222 holds the external buffers may be corrupted.
224 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
225 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
226 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
228 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
229 and allmulticast mode are both set to off.
230 To receive IPv6 Multicast messages on VM, explicitly set the relevant
231 MAC address using rte_eth_dev_mac_addr_add() API.
233 - To support a mixed traffic pattern (some buffers from local host memory, some
234 buffers from other devices) with high bandwidth, a mbuf flag is used.
236 An application hints the PMD whether or not it should try to inline the
237 given mbuf data buffer. PMD should do the best effort to act upon this request.
239 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
240 registered by application with rte_mbuf_dynflag_register(). This flag is
241 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
242 which is intended to be used by the application.
244 To query the supported specific flags in runtime,
245 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
246 currently (over present hardware and configuration) supported specific flags.
247 The "not inline hint" feature operating flow is the following one:
250 - probe the devices, ports are created
251 - query the port capabilities
252 - if port supporting the feature is found
253 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
254 - application starts the ports
255 - on ``dev_start()`` PMD checks whether the feature flag is registered and
256 enables the feature support in datapath
257 - application might set the registered flag bit in ``ol_flags`` field
258 of mbuf being sent and PMD will handle ones appropriately.
260 - The amount of descriptors in Tx queue may be limited by data inline settings.
261 Inline data require the more descriptor building blocks and overall block
262 amount may exceed the hardware supported limits. The application should
263 reduce the requested Tx size or adjust data inline settings with
264 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
266 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
267 parameter should be specified.
268 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
269 being sent it tries to synchronize the time of packet appearing on
270 the wire with the specified packet timestamp. It the specified one
271 is in the past it should be ignored, if one is in the distant future
272 it should be capped with some reasonable value (in range of seconds).
273 These specific cases ("too late" and "distant future") can be optionally
274 reported via device xstats to assist applications to detect the
275 time-related problems.
277 The timestamp upper "too-distant-future" limit
278 at the moment of invoking the Tx burst routine
279 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
280 Please note, for the testpmd txonly mode,
281 the limit is deduced from the expression::
283 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
285 There is no any packet reordering according timestamps is supposed,
286 neither within packet burst, nor between packets, it is an entirely
287 application responsibility to generate packets and its timestamps
288 in desired order. The timestamps can be put only in the first packet
289 in the burst providing the entire burst scheduling.
291 - E-Switch decapsulation Flow:
293 - can be applied to PF port only.
294 - must specify VF port action (packet redirection from PF to VF).
295 - optionally may specify tunnel inner source and destination MAC addresses.
297 - E-Switch encapsulation Flow:
299 - can be applied to VF ports only.
300 - must specify PF port action (packet redirection from VF to PF).
304 - The input buffer, used as outer header, is not validated.
308 - The decapsulation is always done up to the outermost tunnel detected by the HW.
309 - The input buffer, providing the removal size, is not validated.
310 - The buffer size must match the length of the headers to be removed.
312 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
313 mutually exclusive features which cannot be supported together
314 (see :ref:`mlx5_firmware_config`).
318 - Requires DevX and DV flow to be enabled.
319 - KEEP_CRC offload cannot be supported with LRO.
320 - The first mbuf length, without head-room, must be big enough to include the
322 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
323 it with size limited to max LRO size, not to max RX packet length.
324 - LRO can be used with outer header of TCP packets of the standard format:
325 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
327 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
328 - LRO packet aggregation is performed by HW only for packet size larger than
329 ``lro_min_mss_size``. This value is reported on device start, when debug
334 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
335 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
336 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
340 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and E-Switch steering domain.
341 - The E-Switch Sample flow must have the eswitch_manager VPORT destination (PF or ECPF) and no additional actions.
342 - 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.
344 - IPv6 header item 'proto' field, indicating the next header protocol, should
345 not be set as extension header.
346 In case the next header is an extension header, it should not be specified in
347 IPv6 header item 'proto' field.
348 The last extension header item 'next header' field can specify the following
349 header protocol type.
353 - 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.
354 - Hairpin in switchdev SR-IOV mode is not supported till now.
359 MLX5 supports various methods to report statistics:
361 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.
363 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.
365 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.
373 The ibverbs libraries can be linked with this PMD in a number of ways,
374 configured by the ``ibverbs_link`` build option:
376 - ``shared`` (default): the PMD depends on some .so files.
378 - ``dlopen``: Split the dependencies glue in a separate library
379 loaded when needed by dlopen.
380 It make dependencies on libibverbs and libmlx4 optional,
381 and has no performance impact.
383 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
384 in the PMD shared library or the executable static binary.
386 Environment variables
387 ~~~~~~~~~~~~~~~~~~~~~
391 A list of directories in which to search for the rdma-core "glue" plug-in,
392 separated by colons or semi-colons.
394 - ``MLX5_SHUT_UP_BF``
396 Configures HW Tx doorbell register as IO-mapped.
398 By default, the HW Tx doorbell is configured as a write-combining register.
399 The register would be flushed to HW usually when the write-combining buffer
400 becomes full, but it depends on CPU design.
402 Except for vectorized Tx burst routines, a write memory barrier is enforced
403 after updating the register so that the update can be immediately visible to
406 When vectorized Tx burst is called, the barrier is set only if the burst size
407 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
408 variable will bring better latency even though the maximum throughput can
411 Run-time configuration
412 ~~~~~~~~~~~~~~~~~~~~~~
414 - librte_net_mlx5 brings kernel network interfaces up during initialization
415 because it is affected by their state. Forcing them down prevents packets
418 - **ethtool** operations on related kernel interfaces also affect the PMD.
423 In order to run as a non-root user,
424 some capabilities must be granted to the application::
426 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
428 Below are the reasons of the need for each capability:
431 When using physical addresses (PA mode), with Linux >= 4.0,
432 for access to ``/proc/self/pagemap``.
435 For device configuration.
438 For raw ethernet queue allocation through kernel driver.
441 For DMA memory pinning.
446 - ``rxq_cqe_comp_en`` parameter [int]
448 A nonzero value enables the compression of CQE on RX side. This feature
449 allows to save PCI bandwidth and improve performance. Enabled by default.
450 Different compression formats are supported in order to achieve the best
451 performance for different traffic patterns. Hash RSS format is the default.
453 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
454 better compression rate in case of RTE Flow Mark traffic.
455 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
456 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
457 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
461 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
462 ConnectX-6 Lx, BlueField and BlueField-2.
463 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
464 ConnectX-6 Lx, BlueField and BlueField-2.
466 - ``rxq_pkt_pad_en`` parameter [int]
468 A nonzero value enables padding Rx packet to the size of cacheline on PCI
469 transaction. This feature would waste PCI bandwidth but could improve
470 performance by avoiding partial cacheline write which may cause costly
471 read-modify-copy in memory transaction on some architectures. Disabled by
476 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
477 ConnectX-6 Lx, BlueField and BlueField-2.
478 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
479 ConnectX-6 Lx, BlueField and BlueField-2.
481 - ``mprq_en`` parameter [int]
483 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
484 configured as Multi-Packet RQ if the total number of Rx queues is
485 ``rxqs_min_mprq`` or more. Disabled by default.
487 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
488 by posting a single large buffer for multiple packets. Instead of posting a
489 buffers per a packet, one large buffer is posted in order to receive multiple
490 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
491 and each stride receives one packet. MPRQ can improve throughput for
492 small-packet traffic.
494 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
495 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
496 configure large stride size enough to accommodate max_rx_pkt_len as long as
497 device allows. Note that this can waste system memory compared to enabling Rx
498 scatter and multi-segment packet.
500 - ``mprq_log_stride_num`` parameter [int]
502 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
503 strides can reduce PCIe traffic further. If configured value is not in the
504 range of device capability, the default value will be set with a warning
505 message. The default value is 4 which is 16 strides per a buffer, valid only
506 if ``mprq_en`` is set.
508 The size of Rx queue should be bigger than the number of strides.
510 - ``mprq_log_stride_size`` parameter [int]
512 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
513 stride size can save some memory and reduce probability of a depletion of all
514 available strides due to unreleased packets by an application. If configured
515 value is not in the range of device capability, the default value will be set
516 with a warning message. The default value is 11 which is 2048 bytes per a
517 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
518 it is possible for a packet to span across multiple strides. This mode allows
519 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
520 of a packet if Rx scatter is configured) may be required in case there is no
521 space left for a head room at the end of a stride which incurs some
524 - ``mprq_max_memcpy_len`` parameter [int]
526 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
527 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
528 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
529 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
530 A mempool for external buffers will be allocated and managed by PMD. If Rx
531 packet is externally attached, ol_flags field of the mbuf will have
532 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
533 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
535 - ``rxqs_min_mprq`` parameter [int]
537 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
538 greater or equal to this value. The default value is 12, valid only if
541 - ``txq_inline`` parameter [int]
543 Amount of data to be inlined during TX operations. This parameter is
544 deprecated and converted to the new parameter ``txq_inline_max`` providing
545 partial compatibility.
547 - ``txqs_min_inline`` parameter [int]
549 Enable inline data send only when the number of TX queues is greater or equal
552 This option should be used in combination with ``txq_inline_max`` and
553 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
555 If this option is not specified the default value 16 is used for BlueField
556 and 8 for other platforms
558 The data inlining consumes the CPU cycles, so this option is intended to
559 auto enable inline data if we have enough Tx queues, which means we have
560 enough CPU cores and PCI bandwidth is getting more critical and CPU
561 is not supposed to be bottleneck anymore.
563 The copying data into WQE improves latency and can improve PPS performance
564 when PCI back pressure is detected and may be useful for scenarios involving
565 heavy traffic on many queues.
567 Because additional software logic is necessary to handle this mode, this
568 option should be used with care, as it may lower performance when back
569 pressure is not expected.
571 If inline data are enabled it may affect the maximal size of Tx queue in
572 descriptors because the inline data increase the descriptor size and
573 queue size limits supported by hardware may be exceeded.
575 - ``txq_inline_min`` parameter [int]
577 Minimal amount of data to be inlined into WQE during Tx operations. NICs
578 may require this minimal data amount to operate correctly. The exact value
579 may depend on NIC operation mode, requested offloads, etc. It is strongly
580 recommended to omit this parameter and use the default values. Anyway,
581 applications using this parameter should take into consideration that
582 specifying an inconsistent value may prevent the NIC from sending packets.
584 If ``txq_inline_min`` key is present the specified value (may be aligned
585 by the driver in order not to exceed the limits and provide better descriptor
586 space utilization) will be used by the driver and it is guaranteed that
587 requested amount of data bytes are inlined into the WQE beside other inline
588 settings. This key also may update ``txq_inline_max`` value (default
589 or specified explicitly in devargs) to reserve the space for inline data.
591 If ``txq_inline_min`` key is not present, the value may be queried by the
592 driver from the NIC via DevX if this feature is available. If there is no DevX
593 enabled/supported the value 18 (supposing L2 header including VLAN) is set
594 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
595 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
598 For ConnectX-4 NIC, driver does not allow specifying value below 18
599 (minimal L2 header, including VLAN), error will be raised.
601 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
602 it is not recommended and may prevent NIC from sending packets over
605 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
606 Multi-Packet Write), because last one does not support partial packet inlining.
607 This is not very critical due to minimal data inlining is mostly required
608 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
610 - ``txq_inline_max`` parameter [int]
612 Specifies the maximal packet length to be completely inlined into WQE
613 Ethernet Segment for ordinary SEND method. If packet is larger than specified
614 value, the packet data won't be copied by the driver at all, data buffer
615 is addressed with a pointer. If packet length is less or equal all packet
616 data will be copied into WQE. This may improve PCI bandwidth utilization for
617 short packets significantly but requires the extra CPU cycles.
619 The data inline feature is controlled by number of Tx queues, if number of Tx
620 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
621 is engaged, if there are not enough Tx queues (which means not enough CPU cores
622 and CPU resources are scarce), data inline is not performed by the driver.
623 Assigning ``txqs_min_inline`` with zero always enables the data inline.
625 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
626 by the driver in order not to exceed the limit (930 bytes) and to provide better
627 WQE space filling without gaps, the adjustment is reflected in the debug log.
628 Also, the default value (290) may be decreased in run-time if the large transmit
629 queue size is requested and hardware does not support enough descriptor
630 amount, in this case warning is emitted. If ``txq_inline_max`` key is
631 specified and requested inline settings can not be satisfied then error
634 - ``txq_inline_mpw`` parameter [int]
636 Specifies the maximal packet length to be completely inlined into WQE for
637 Enhanced MPW method. If packet is large the specified value, the packet data
638 won't be copied, and data buffer is addressed with pointer. If packet length
639 is less or equal, all packet data will be copied into WQE. This may improve PCI
640 bandwidth utilization for short packets significantly but requires the extra
643 The data inline feature is controlled by number of TX queues, if number of Tx
644 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
645 is engaged, if there are not enough Tx queues (which means not enough CPU cores
646 and CPU resources are scarce), data inline is not performed by the driver.
647 Assigning ``txqs_min_inline`` with zero always enables the data inline.
649 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
650 by the driver in order not to exceed the limit (930 bytes) and to provide better
651 WQE space filling without gaps, the adjustment is reflected in the debug log.
652 Due to multiple packets may be included to the same WQE with Enhanced Multi
653 Packet Write Method and overall WQE size is limited it is not recommended to
654 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
655 may be decreased in run-time if the large transmit queue size is requested
656 and hardware does not support enough descriptor amount, in this case warning
657 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
658 settings can not be satisfied then error will be raised.
660 - ``txqs_max_vec`` parameter [int]
662 Enable vectorized Tx only when the number of TX queues is less than or
663 equal to this value. This parameter is deprecated and ignored, kept
664 for compatibility issue to not prevent driver from probing.
666 - ``txq_mpw_hdr_dseg_en`` parameter [int]
668 A nonzero value enables including two pointers in the first block of TX
669 descriptor. The parameter is deprecated and ignored, kept for compatibility
672 - ``txq_max_inline_len`` parameter [int]
674 Maximum size of packet to be inlined. This limits the size of packet to
675 be inlined. If the size of a packet is larger than configured value, the
676 packet isn't inlined even though there's enough space remained in the
677 descriptor. Instead, the packet is included with pointer. This parameter
678 is deprecated and converted directly to ``txq_inline_mpw`` providing full
679 compatibility. Valid only if eMPW feature is engaged.
681 - ``txq_mpw_en`` parameter [int]
683 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
684 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
685 eMPW allows the Tx burst function to pack up multiple packets
686 in a single descriptor session in order to save PCI bandwidth
687 and improve performance at the cost of a slightly higher CPU usage.
688 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
689 Tx burst function copies entire packet data on to Tx descriptor
690 instead of including pointer of packet.
692 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
693 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
694 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
695 option or reported by the NIC, the eMPW feature is disengaged.
697 - ``tx_db_nc`` parameter [int]
699 The rdma core library can map doorbell register in two ways, depending on the
700 environment variable "MLX5_SHUT_UP_BF":
702 - As regular cached memory (usually with write combining attribute), if the
703 variable is either missing or set to zero.
704 - As non-cached memory, if the variable is present and set to not "0" value.
706 The type of mapping may slightly affect the Tx performance, the optimal choice
707 is strongly relied on the host architecture and should be deduced practically.
709 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
710 memory (with write combining), the PMD will perform the extra write memory barrier
711 after writing to doorbell, it might increase the needed CPU clocks per packet
712 to send, but latency might be improved.
714 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
715 cached memory, the PMD will not perform the extra write memory barrier
716 after writing to doorbell, on some architectures it might improve the
719 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
720 memory, the PMD will use heuristics to decide whether write memory barrier
721 should be performed. For bursts with size multiple of recommended one (64 pkts)
722 it is supposed the next burst is coming and no need to issue the extra memory
723 barrier (it is supposed to be issued in the next coming burst, at least after
724 descriptor writing). It might increase latency (on some hosts till next
725 packets transmit) and should be used with care.
727 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
728 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
729 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
731 - ``tx_pp`` parameter [int]
733 If a nonzero value is specified the driver creates all necessary internal
734 objects to provide accurate packet send scheduling on mbuf timestamps.
735 The positive value specifies the scheduling granularity in nanoseconds,
736 the packet send will be accurate up to specified digits. The allowed range is
737 from 500 to 1 million of nanoseconds. The negative value specifies the module
738 of granularity and engages the special test mode the check the schedule rate.
739 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
742 - ``tx_skew`` parameter [int]
744 The parameter adjusts the send packet scheduling on timestamps and represents
745 the average delay between beginning of the transmitting descriptor processing
746 by the hardware and appearance of actual packet data on the wire. The value
747 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
748 specified. The default value is zero.
750 - ``tx_vec_en`` parameter [int]
752 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
753 ConnectX-6 Lx, BlueField and BlueField-2 NICs
754 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
755 The parameter is deprecated and ignored.
757 - ``rx_vec_en`` parameter [int]
759 A nonzero value enables Rx vector if the port is not configured in
760 multi-segment otherwise this parameter is ignored.
764 - ``vf_nl_en`` parameter [int]
766 A nonzero value enables Netlink requests from the VF to add/remove MAC
767 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
768 Otherwise the relevant configuration must be run with Linux iproute2 tools.
769 This is a prerequisite to receive this kind of traffic.
771 Enabled by default, valid only on VF devices ignored otherwise.
773 - ``l3_vxlan_en`` parameter [int]
775 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
776 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
777 parameter. This is a prerequisite to receive this kind of traffic.
781 - ``dv_xmeta_en`` parameter [int]
783 A nonzero value enables extensive flow metadata support if device is
784 capable and driver supports it. This can enable extensive support of
785 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
786 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
788 There are some possible configurations, depending on parameter value:
790 - 0, this is default value, defines the legacy mode, the ``MARK`` and
791 ``META`` related actions and items operate only within NIC Tx and
792 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
793 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
794 item is 32 bits wide and match supported on egress only.
796 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
797 related actions and items operate within all supported steering domains,
798 including FDB, ``MARK`` and ``META`` information may cross the domain
799 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
800 depends on kernel and firmware configurations and might be 0, 16 or
801 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
802 compatibility, the actual width of data transferred to the FDB domain
803 depends on kernel configuration and may be vary. The actual supported
804 width can be retrieved in runtime by series of rte_flow_validate()
807 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
808 related actions and items operate within all supported steering domains,
809 including FDB, ``MARK`` and ``META`` information may cross the domain
810 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
811 depends on kernel and firmware configurations and might be 0, 16 or
812 24 bits. The actual supported width can be retrieved in runtime by
813 series of rte_flow_validate() trials.
815 - 3, this engages tunnel offload mode. In E-Switch configuration, that
816 mode implicitly activates ``dv_xmeta_en=1``.
818 +------+-----------+-----------+-------------+-------------+
819 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
820 +======+===========+===========+=============+=============+
821 | 0 | 24 bits | 32 bits | 32 bits | no |
822 +------+-----------+-----------+-------------+-------------+
823 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
824 +------+-----------+-----------+-------------+-------------+
825 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
826 +------+-----------+-----------+-------------+-------------+
828 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
829 ignored and the device is configured to operate in legacy mode (0).
831 Disabled by default (set to 0).
833 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
834 of the extensive metadata features. The legacy Verbs supports FLAG and
835 MARK metadata actions over NIC Rx steering domain only.
837 - ``dv_flow_en`` parameter [int]
839 A nonzero value enables the DV flow steering assuming it is supported
840 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
842 Enabled by default if supported.
844 - ``dv_esw_en`` parameter [int]
846 A nonzero value enables E-Switch using Direct Rules.
848 Enabled by default if supported.
850 - ``lacp_by_user`` parameter [int]
852 A nonzero value enables the control of LACP traffic by the user application.
853 When a bond exists in the driver, by default it should be managed by the
854 kernel and therefore LACP traffic should be steered to the kernel.
855 If this devarg is set to 1 it will allow the user to manage the bond by
856 itself and not steer LACP traffic to the kernel.
858 Disabled by default (set to 0).
860 - ``mr_ext_memseg_en`` parameter [int]
862 A nonzero value enables extending memseg when registering DMA memory. If
863 enabled, the number of entries in MR (Memory Region) lookup table on datapath
864 is minimized and it benefits performance. On the other hand, it worsens memory
865 utilization because registered memory is pinned by kernel driver. Even if a
866 page in the extended chunk is freed, that doesn't become reusable until the
867 entire memory is freed.
871 - ``representor`` parameter [list]
873 This parameter can be used to instantiate DPDK Ethernet devices from
874 existing port (or VF) representors configured on the device.
876 It is a standard parameter whose format is described in
877 :ref:`ethernet_device_standard_device_arguments`.
879 For instance, to probe port representors 0 through 2::
883 - ``max_dump_files_num`` parameter [int]
885 The maximum number of files per PMD entity that may be created for debug information.
886 The files will be created in /var/log directory or in current directory.
888 set to 128 by default.
890 - ``lro_timeout_usec`` parameter [int]
892 The maximum allowed duration of an LRO session, in micro-seconds.
893 PMD will set the nearest value supported by HW, which is not bigger than
894 the input ``lro_timeout_usec`` value.
895 If this parameter is not specified, by default PMD will set
896 the smallest value supported by HW.
898 - ``hp_buf_log_sz`` parameter [int]
900 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
901 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
902 The capacity of the value is specified by the firmware and the initialization
903 will get a failure if it is out of scope.
904 The range of the value is from 11 to 19 right now, and the supported frame
905 size of a single packet for hairpin is from 512B to 128KB. It might change if
906 different firmware release is being used. By using a small value, it could
907 reduce memory consumption but not work with a large frame. If the value is
908 too large, the memory consumption will be high and some potential performance
909 degradation will be introduced.
910 By default, the PMD will set this value to 16, which means that 9KB jumbo
911 frames will be supported.
913 - ``reclaim_mem_mode`` parameter [int]
915 Cache some resources in flow destroy will help flow recreation more efficient.
916 While some systems may require the all the resources can be reclaimed after
918 The parameter ``reclaim_mem_mode`` provides the option for user to configure
919 if the resource cache is needed or not.
921 There are three options to choose:
923 - 0. It means the flow resources will be cached as usual. The resources will
924 be cached, helpful with flow insertion rate.
926 - 1. It will only enable the DPDK PMD level resources reclaim.
928 - 2. Both DPDK PMD level and rdma-core low level will be configured as
931 By default, the PMD will set this value to 0.
933 - ``sys_mem_en`` parameter [int]
935 A non-zero value enables the PMD memory management allocating memory
936 from system by default, without explicit rte memory flag.
938 By default, the PMD will set this value to 0.
940 - ``decap_en`` parameter [int]
942 Some devices do not support FCS (frame checksum) scattering for
943 tunnel-decapsulated packets.
944 If set to 0, this option forces the FCS feature and rejects tunnel
945 decapsulation in the flow engine for such devices.
947 By default, the PMD will set this value to 1.
949 .. _mlx5_firmware_config:
951 Firmware configuration
952 ~~~~~~~~~~~~~~~~~~~~~~
954 Firmware features can be configured as key/value pairs.
956 The command to set a value is::
958 mlxconfig -d <device> set <key>=<value>
960 The command to query a value is::
962 mlxconfig -d <device> query | grep <key>
964 The device name for the command ``mlxconfig`` can be either the PCI address,
965 or the mst device name found with::
969 Below are some firmware configurations listed.
975 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
981 - maximum number of SR-IOV virtual functions::
985 - enable DevX (required by Direct Rules and other features)::
989 - aggressive CQE zipping::
993 - L3 VXLAN and VXLAN-GPE destination UDP port::
996 IP_OVER_VXLAN_PORT=<udp dport>
998 - enable VXLAN-GPE tunnel flow matching::
1000 FLEX_PARSER_PROFILE_ENABLE=0
1002 FLEX_PARSER_PROFILE_ENABLE=2
1004 - enable IP-in-IP tunnel flow matching::
1006 FLEX_PARSER_PROFILE_ENABLE=0
1008 - enable MPLS flow matching::
1010 FLEX_PARSER_PROFILE_ENABLE=1
1012 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1014 FLEX_PARSER_PROFILE_ENABLE=2
1016 - enable Geneve flow matching::
1018 FLEX_PARSER_PROFILE_ENABLE=0
1020 FLEX_PARSER_PROFILE_ENABLE=1
1022 - enable GTP flow matching::
1024 FLEX_PARSER_PROFILE_ENABLE=3
1026 - enable eCPRI flow matching::
1028 FLEX_PARSER_PROFILE_ENABLE=4
1034 This driver relies on external libraries and kernel drivers for resources
1035 allocations and initialization. The following dependencies are not part of
1036 DPDK and must be installed separately:
1040 User space Verbs framework used by librte_net_mlx5. This library provides
1041 a generic interface between the kernel and low-level user space drivers
1044 It allows slow and privileged operations (context initialization, hardware
1045 resources allocations) to be managed by the kernel and fast operations to
1046 never leave user space.
1050 Low-level user space driver library for Mellanox
1051 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1054 This library basically implements send/receive calls to the hardware
1057 - **Kernel modules**
1059 They provide the kernel-side Verbs API and low level device drivers that
1060 manage actual hardware initialization and resources sharing with user
1063 Unlike most other PMDs, these modules must remain loaded and bound to
1066 - mlx5_core: hardware driver managing Mellanox
1067 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1069 - mlx5_ib: InifiniBand device driver.
1070 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1072 - **Firmware update**
1074 Mellanox OFED/EN releases include firmware updates for
1075 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1077 Because each release provides new features, these updates must be applied to
1078 match the kernel modules and libraries they come with.
1082 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1088 Either RDMA Core library with a recent enough Linux kernel release
1089 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1092 RDMA Core with Linux Kernel
1093 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1095 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1096 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1097 (see `RDMA Core installation documentation`_)
1098 - When building for i686 use:
1100 - rdma-core version 18.0 or above built with 32bit support.
1101 - Kernel version 4.14.41 or above.
1103 - Starting with rdma-core v21, static libraries can be built::
1106 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1109 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1110 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1116 - Mellanox OFED version: **4.5** and above /
1117 Mellanox EN version: **4.5** and above
1120 - ConnectX-4: **12.21.1000** and above.
1121 - ConnectX-4 Lx: **14.21.1000** and above.
1122 - ConnectX-5: **16.21.1000** and above.
1123 - ConnectX-5 Ex: **16.21.1000** and above.
1124 - ConnectX-6: **20.27.0090** and above.
1125 - ConnectX-6 Dx: **22.27.0090** and above.
1126 - BlueField: **18.25.1010** and above.
1128 While these libraries and kernel modules are available on OpenFabrics
1129 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1130 managers on most distributions, this PMD requires Ethernet extensions that
1131 may not be supported at the moment (this is a work in progress).
1134 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1136 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1137 include the necessary support and should be used in the meantime. For DPDK,
1138 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1139 required from that distribution.
1143 Several versions of Mellanox OFED/EN are available. Installing the version
1144 this DPDK release was developed and tested against is strongly
1145 recommended. Please check the `linux prerequisites`_.
1147 Windows Prerequisites
1148 ---------------------
1150 This driver relies on external libraries and kernel drivers for resources
1151 allocations and initialization. The dependencies in the following sub-sections
1152 are not part of DPDK, and must be installed separately.
1154 Compilation Prerequisites
1155 ~~~~~~~~~~~~~~~~~~~~~~~~~
1157 DevX SDK installation
1158 ^^^^^^^^^^^^^^^^^^^^^
1160 The DevX SDK must be installed on the machine building the Windows PMD.
1161 Additional information can be found at
1162 `How to Integrate Windows DevX in Your Development Environment
1163 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1165 Runtime Prerequisites
1166 ~~~~~~~~~~~~~~~~~~~~~
1168 WinOF2 version 2.60 or higher must be installed on the machine.
1173 The driver can be downloaded from the following site:
1175 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1180 DevX for Windows must be enabled in the Windows registry.
1181 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1182 Additional information can be found in the WinOF2 user manual.
1187 The following Mellanox device families are supported by the same mlx5 driver:
1199 Below are detailed device names:
1201 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1202 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1203 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1204 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1205 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1206 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1207 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1208 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1209 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1210 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1211 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1212 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1213 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1214 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1215 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1216 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1217 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1218 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1219 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1220 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1221 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1222 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1223 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1224 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1225 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1226 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1228 Quick Start Guide on OFED/EN
1229 ----------------------------
1231 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1234 2. Install the required libraries and kernel modules either by installing
1235 only the required set, or by installing the entire Mellanox OFED/EN::
1237 ./mlnxofedinstall --upstream-libs --dpdk
1239 3. Verify the firmware is the correct one::
1243 4. Verify all ports links are set to Ethernet::
1245 mlxconfig -d <mst device> query | grep LINK_TYPE
1249 Link types may have to be configured to Ethernet::
1251 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1253 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1255 For hypervisors, verify SR-IOV is enabled on the NIC::
1257 mlxconfig -d <mst device> query | grep SRIOV_EN
1260 If needed, configure SR-IOV::
1262 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1263 mlxfwreset -d <mst device> reset
1265 5. Restart the driver::
1267 /etc/init.d/openibd restart
1271 service openibd restart
1273 If link type was changed, firmware must be reset as well::
1275 mlxfwreset -d <mst device> reset
1277 For hypervisors, after reset write the sysfs number of virtual functions
1280 To dynamically instantiate a given number of virtual functions (VFs)::
1282 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1284 6. Install DPDK and you are ready to go.
1285 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1287 Enable switchdev mode
1288 ---------------------
1290 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1291 Representor is a port in DPDK that is connected to a VF in such a way
1292 that assuming there are no offload flows, each packet that is sent from the VF
1293 will be received by the corresponding representor. While each packet that is
1294 sent to a representor will be received by the VF.
1295 This is very useful in case of SRIOV mode, where the first packet that is sent
1296 by the VF will be received by the DPDK application which will decide if this
1297 flow should be offloaded to the E-Switch. After offloading the flow packet
1298 that the VF that are matching the flow will not be received any more by
1299 the DPDK application.
1301 1. Enable SRIOV mode::
1303 mlxconfig -d <mst device> set SRIOV_EN=true
1305 2. Configure the max number of VFs::
1307 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1311 mlxfwreset -d <mst device> reset
1313 3. Configure the actual number of VFs::
1315 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1317 4. Unbind the device (can be rebind after the switchdev mode)::
1319 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1321 5. Enbale switchdev mode::
1323 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1328 1. Configure aggressive CQE Zipping for maximum performance::
1330 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1332 To set it back to the default CQE Zipping mode use::
1334 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1336 2. In case of virtualization:
1338 - Make sure that hypervisor kernel is 3.16 or newer.
1339 - Configure boot with ``iommu=pt``.
1340 - Use 1G huge pages.
1341 - Make sure to allocate a VM on huge pages.
1342 - Make sure to set CPU pinning.
1344 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1345 for better performance. For VMs, verify that the right CPU
1346 and NUMA node are pinned according to the above. Run::
1350 to identify the NUMA node to which the PCIe adapter is connected.
1352 4. If more than one adapter is used, and root complex capabilities allow
1353 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1354 it is recommended to locate both adapters on the same NUMA node.
1355 This in order to forward packets from one to the other without
1356 NUMA performance penalty.
1358 5. Disable pause frames::
1360 ethtool -A <netdev> rx off tx off
1362 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1363 via the BIOS configuration. Please contact you server provider for more
1364 information about the settings.
1368 On some machines, depends on the machine integrator, it is beneficial
1369 to set the PCI max read request parameter to 1K. This can be
1370 done in the following way:
1372 To query the read request size use::
1374 setpci -s <NIC PCI address> 68.w
1376 If the output is different than 3XXX, set it by::
1378 setpci -s <NIC PCI address> 68.w=3XXX
1380 The XXX can be different on different systems. Make sure to configure
1381 according to the setpci output.
1383 7. To minimize overhead of searching Memory Regions:
1385 - '--socket-mem' is recommended to pin memory by predictable amount.
1386 - Configure per-lcore cache when creating Mempools for packet buffer.
1387 - Refrain from dynamically allocating/freeing memory in run-time.
1392 There are multiple Rx burst functions with different advantages and limitations.
1394 .. table:: Rx burst functions
1396 +-------------------+------------------------+---------+-----------------+------+-------+
1397 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1398 | | | | || comp|| MTU |
1399 +===================+========================+=========+=================+======+=======+
1400 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1401 +-------------------+------------------------+---------+-----------------+------+-------+
1402 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1403 +-------------------+------------------------+---------+-----------------+------+-------+
1404 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1405 | || RxQs >= rxqs_min_mprq | | | | |
1406 +-------------------+------------------------+---------+-----------------+------+-------+
1407 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1408 | || mprq_en=1 | | | | |
1409 | || RxQs >= rxqs_min_mprq | | | | |
1410 +-------------------+------------------------+---------+-----------------+------+-------+
1412 .. _mlx5_offloads_support:
1414 Supported hardware offloads
1415 ---------------------------
1417 .. table:: Minimal SW/HW versions for queue offloads
1419 ============== ===== ===== ========= ===== ========== =============
1420 Offload DPDK Linux rdma-core OFED firmware hardware
1421 ============== ===== ===== ========= ===== ========== =============
1422 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1423 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1424 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1425 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1426 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1427 Buffer Split 20.11 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1428 ============== ===== ===== ========= ===== ========== =============
1430 .. table:: Minimal SW/HW versions for rte_flow offloads
1432 +-----------------------+-----------------+-----------------+
1433 | Offload | with E-Switch | with NIC |
1434 +=======================+=================+=================+
1435 | Count | | DPDK 19.05 | | DPDK 19.02 |
1436 | | | OFED 4.6 | | OFED 4.6 |
1437 | | | rdma-core 24 | | rdma-core 23 |
1438 | | | ConnectX-5 | | ConnectX-5 |
1439 +-----------------------+-----------------+-----------------+
1440 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1441 | | | OFED 4.6 | | OFED 4.5 |
1442 | | | rdma-core 24 | | rdma-core 23 |
1443 | | | ConnectX-5 | | ConnectX-4 |
1444 +-----------------------+-----------------+-----------------+
1445 | Queue / RSS | | | | DPDK 18.11 |
1446 | | | N/A | | OFED 4.5 |
1447 | | | | | rdma-core 23 |
1448 | | | | | ConnectX-4 |
1449 +-----------------------+-----------------+-----------------+
1450 | RSS shared action | | | | DPDK 20.11 |
1451 | | | N/A | | OFED 5.2 |
1452 | | | | | rdma-core 33 |
1453 | | | | | ConnectX-5 |
1454 +-----------------------+-----------------+-----------------+
1455 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1456 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1457 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1458 | | of_set_vlan_pcp / | | | | |
1459 | | of_set_vlan_vid) | | | | |
1460 +-----------------------+-----------------+-----------------+
1461 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1462 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1463 | | | rdma-core 24 | | rdma-core 23 |
1464 | | | ConnectX-5 | | ConnectX-5 |
1465 +-----------------------+-----------------+-----------------+
1466 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1467 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1468 | | | rdma-core 27 | | rdma-core 27 |
1469 | | | ConnectX-5 | | ConnectX-5 |
1470 +-----------------------+-----------------+-----------------+
1471 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1472 | | | OFED 5.1-2 | | OFED 5.1-2 |
1473 | | | rdma-core 32 | | N/A |
1474 | | | ConnectX-5 | | ConnectX-5 |
1475 +-----------------------+-----------------+-----------------+
1476 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1477 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1478 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1479 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1480 | | set_ipv6_dst / | | | | |
1481 | | set_tp_src / | | | | |
1482 | | set_tp_dst / | | | | |
1483 | | dec_ttl / | | | | |
1484 | | set_ttl / | | | | |
1485 | | set_mac_src / | | | | |
1486 | | set_mac_dst) | | | | |
1487 +-----------------------+-----------------+-----------------+
1488 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1489 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1490 | | | | rdma-core 24 | | rdma-core 24 |
1491 | | | | ConnectX-5 | | ConnectX-5 |
1492 +-----------------------+-----------------+-----------------+
1493 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1494 | | | OFED 4.7-1 | | OFED 4.7-1 |
1495 | | | rdma-core 24 | | N/A |
1496 | | | ConnectX-5 | | ConnectX-5 |
1497 +-----------------------+-----------------+-----------------+
1498 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1499 | | | OFED 4.6 | | OFED 4.5 |
1500 | | | rdma-core 24 | | rdma-core 23 |
1501 | | | ConnectX-5 | | ConnectX-4 |
1502 +-----------------------+-----------------+-----------------+
1503 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1504 | | | OFED 4.7-3 | | OFED 4.7-3 |
1505 | | | rdma-core 26 | | rdma-core 26 |
1506 | | | ConnectX-5 | | ConnectX-5 |
1507 +-----------------------+-----------------+-----------------+
1508 | Port ID | | DPDK 19.05 | | N/A |
1509 | | | OFED 4.7-1 | | N/A |
1510 | | | rdma-core 24 | | N/A |
1511 | | | ConnectX-5 | | N/A |
1512 +-----------------------+-----------------+-----------------+
1513 | Hairpin | | | | DPDK 19.11 |
1514 | | | N/A | | OFED 4.7-3 |
1515 | | | | | rdma-core 26 |
1516 | | | | | ConnectX-5 |
1517 +-----------------------+-----------------+-----------------+
1518 | 2-port Hairpin | | | | DPDK 20.11 |
1519 | | | N/A | | OFED 5.1-2 |
1521 | | | | | ConnectX-5 |
1522 +-----------------------+-----------------+-----------------+
1523 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1524 | | | OFED 4.7-3 | | OFED 4.7-3 |
1525 | | | rdma-core 26 | | rdma-core 26 |
1526 | | | ConnectX-5 | | ConnectX-5 |
1527 +-----------------------+-----------------+-----------------+
1528 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1529 | | | OFED 5.1-2 | | OFED 5.1-2 |
1530 | | | rdma-core 32 | | N/A |
1531 | | | ConnectX-5 | | ConnectX-5 |
1532 +-----------------------+-----------------+-----------------+
1533 | Age shared action | | DPDK 20.11 | | DPDK 20.11 |
1534 | | | OFED 5.2 | | OFED 5.2 |
1535 | | | rdma-core 32 | | rdma-core 32 |
1536 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1537 +-----------------------+-----------------+-----------------+
1538 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1539 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1540 | | | rdma-core 35 | | rdma-core 35 |
1541 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1542 +-----------------------+-----------------+-----------------+
1547 MARK and META items are interrelated with datapath - they might move from/to
1548 the applications in mbuf fields. Hence, zero value for these items has the
1549 special meaning - it means "no metadata are provided", not zero values are
1550 treated by applications and PMD as valid ones.
1552 Moreover in the flow engine domain the value zero is acceptable to match and
1553 set, and we should allow to specify zero values as rte_flow parameters for the
1554 META and MARK items and actions. In the same time zero mask has no meaning and
1555 should be rejected on validation stage.
1560 Flows are not cached in the driver.
1561 When stopping a device port, all the flows created on this port from the
1562 application will be flushed automatically in the background.
1563 After stopping the device port, all flows on this port become invalid and
1564 not represented in the system.
1565 All references to these flows held by the application should be discarded
1566 directly but neither destroyed nor flushed.
1568 The application should re-create the flows as required after the port restart.
1573 Compared to librte_net_mlx4 that implements a single RSS configuration per
1574 port, librte_net_mlx5 supports per-protocol RSS configuration.
1576 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1577 command-line parameter to enable additional protocols (UDP and TCP as well
1578 as IP), the following commands must be entered from its CLI to get the same
1579 behavior as librte_net_mlx4::
1582 > port config all rss all
1588 This section demonstrates how to launch **testpmd** with Mellanox
1589 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1591 #. Load the kernel modules::
1593 modprobe -a ib_uverbs mlx5_core mlx5_ib
1595 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1598 /etc/init.d/openibd restart
1602 User space I/O kernel modules (uio and igb_uio) are not used and do
1603 not have to be loaded.
1605 #. Make sure Ethernet interfaces are in working order and linked to kernel
1606 verbs. Related sysfs entries should be present::
1608 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1617 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1620 for intf in eth2 eth3 eth4 eth5;
1622 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1625 sed -n 's,.*/\(.*\),-a \1,p'
1634 #. Request huge pages::
1636 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1638 #. Start testpmd with basic parameters::
1640 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
1645 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1646 EAL: probe driver: 15b3:1013 librte_net_mlx5
1647 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1648 PMD: librte_net_mlx5: 1 port(s) detected
1649 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1650 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1651 EAL: probe driver: 15b3:1013 librte_net_mlx5
1652 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1653 PMD: librte_net_mlx5: 1 port(s) detected
1654 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1655 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1656 EAL: probe driver: 15b3:1013 librte_net_mlx5
1657 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1658 PMD: librte_net_mlx5: 1 port(s) detected
1659 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1660 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1661 EAL: probe driver: 15b3:1013 librte_net_mlx5
1662 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1663 PMD: librte_net_mlx5: 1 port(s) detected
1664 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1665 Interactive-mode selected
1666 Configuring Port 0 (socket 0)
1667 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1668 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1669 Port 0: E4:1D:2D:E7:0C:FE
1670 Configuring Port 1 (socket 0)
1671 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1672 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1673 Port 1: E4:1D:2D:E7:0C:FF
1674 Configuring Port 2 (socket 0)
1675 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1676 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1677 Port 2: E4:1D:2D:E7:0C:FA
1678 Configuring Port 3 (socket 0)
1679 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1680 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1681 Port 3: E4:1D:2D:E7:0C:FB
1682 Checking link statuses...
1683 Port 0 Link Up - speed 40000 Mbps - full-duplex
1684 Port 1 Link Up - speed 40000 Mbps - full-duplex
1685 Port 2 Link Up - speed 10000 Mbps - full-duplex
1686 Port 3 Link Up - speed 10000 Mbps - full-duplex
1693 This section demonstrates how to dump flows. Currently, it's possible to dump
1694 all flows with assistance of external tools.
1696 #. 2 ways to get flow raw file:
1698 - Using testpmd CLI:
1700 .. code-block:: console
1702 testpmd> flow dump <port> <output_file>
1704 - call rte_flow_dev_dump api:
1706 .. code-block:: console
1708 rte_flow_dev_dump(port, file, NULL);
1710 #. Dump human-readable flows from raw file:
1712 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1714 .. code-block:: console
1716 mlx_steering_dump.py -f <output_file>