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_pmd_mlx5**) provides support
11 for **Mellanox ConnectX-4**, **Mellanox ConnectX-4 Lx** , **Mellanox
12 ConnectX-5**, **Mellanox ConnectX-6**, **Mellanox ConnectX-6 Dx** and
13 **Mellanox BlueField** families of 10/25/40/50/100/200 Gb/s adapters
14 as well as their virtual functions (VF) in SR-IOV context.
16 Information and documentation about these adapters can be found on the
17 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
18 `Mellanox community <http://community.mellanox.com/welcome>`__.
20 There is also a `section dedicated to this poll mode driver
21 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
25 Due to external dependencies, this driver is disabled in default configuration
26 of the "make" build. It can be enabled with ``CONFIG_RTE_LIBRTE_MLX5_PMD=y``
27 or by using "meson" build system which will detect dependencies.
32 Besides its dependency on libibverbs (that implies libmlx5 and associated
33 kernel support), librte_pmd_mlx5 relies heavily on system calls for control
34 operations such as querying/updating the MTU and flow control parameters.
36 For security reasons and robustness, this driver only deals with virtual
37 memory addresses. The way resources allocations are handled by the kernel,
38 combined with hardware specifications that allow to handle virtual memory
39 addresses directly, ensure that DPDK applications cannot access random
40 physical memory (or memory that does not belong to the current process).
42 This capability allows the PMD to coexist with kernel network interfaces
43 which remain functional, although they stop receiving unicast packets as
44 long as they share the same MAC address.
45 This means legacy linux control tools (for example: ethtool, ifconfig and
46 more) can operate on the same network interfaces that owned by the DPDK
49 The PMD can use libibverbs and libmlx5 to access the device firmware
50 or directly the hardware components.
51 There are different levels of objects and bypassing abilities
52 to get the best performances:
54 - Verbs is a complete high-level generic API
55 - Direct Verbs is a device-specific API
56 - DevX allows to access firmware objects
57 - Direct Rules manages flow steering at low-level hardware layer
59 Enabling librte_pmd_mlx5 causes DPDK applications to be linked against
65 - Multi arch support: x86_64, POWER8, ARMv8, i686.
66 - Multiple TX and RX queues.
67 - Support for scattered TX and RX frames.
68 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
69 - RSS using different combinations of fields: L3 only, L4 only or both,
70 and source only, destination only or both.
71 - Several RSS hash keys, one for each flow type.
72 - Default RSS operation with no hash key specification.
73 - Configurable RETA table.
74 - Link flow control (pause frame).
75 - Support for multiple MAC addresses.
79 - RX CRC stripping configuration.
80 - Promiscuous mode on PF and VF.
81 - Multicast promiscuous mode on PF and VF.
82 - Hardware checksum offloads.
83 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
85 - Flow API, including :ref:`flow_isolated_mode`.
87 - KVM and VMware ESX SR-IOV modes are supported.
88 - RSS hash result is supported.
89 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
90 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
92 - Statistics query including Basic, Extended and per queue.
94 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
95 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
96 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
97 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
98 - Flow insertion rate of more then million flows per second, when using Direct Rules.
99 - Support for multiple rte_flow groups.
105 - For secondary process:
107 - Forked secondary process not supported.
108 - External memory unregistered in EAL memseg list cannot be used for DMA
109 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
110 primary process and remapped to the same virtual address in secondary
111 process. If the external memory is registered by primary process but has
112 different virtual address in secondary process, unexpected error may happen.
114 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
115 specific VLAN will match for VLAN packets as well:
117 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
118 Meaning, the flow rule::
120 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
122 Will only match vlan packets with vid=3. and the flow rule::
124 flow create 0 ingress pattern eth / ipv4 / end ...
126 Will match any ipv4 packet (VLAN included).
128 - VLAN pop offload command:
130 - Flow rules having a VLAN pop offload command as one of their actions and
131 are lacking a match on VLAN as one of their items are not supported.
132 - The command is not supported on egress traffic.
134 - VLAN push offload is not supported on ingress traffic.
136 - VLAN set PCP offload is not supported on existing headers.
138 - A multi segment packet must have not more segments than reported by dev_infos_get()
139 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
140 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
141 inline settings) to 58.
143 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
144 to 0 are not supported.
146 - VXLAN TSO and checksum offloads are not supported on VM.
148 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
150 - Match on Geneve header supports the following fields only:
156 Currently, the only supported options length value is 0.
158 - VF: flow rules created on VF devices can only match traffic targeted at the
159 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
161 - Match on GTP tunnel header item supports the following fields only:
166 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
167 The flows within group 0 and set metadata action are rejected by hardware.
171 MAC addresses not already present in the bridge table of the associated
172 kernel network device will be added and cleaned up by the PMD when closing
173 the device. In case of ungraceful program termination, some entries may
174 remain present and should be removed manually by other means.
176 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
177 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
178 ol_flags. As the mempool for the external buffer is managed by PMD, all the
179 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
180 the external buffers will be freed by PMD and the application which still
181 holds the external buffers may be corrupted.
183 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
184 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
185 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
187 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
188 and allmulticast mode are both set to off.
189 To receive IPv6 Multicast messages on VM, explicitly set the relevant
190 MAC address using rte_eth_dev_mac_addr_add() API.
192 - The amount of descriptors in Tx queue may be limited by data inline settings.
193 Inline data require the more descriptor building blocks and overall block
194 amount may exceed the hardware supported limits. The application should
195 reduce the requested Tx size or adjust data inline settings with
196 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
198 - E-Switch decapsulation Flow:
200 - can be applied to PF port only.
201 - must specify VF port action (packet redirection from PF to VF).
202 - optionally may specify tunnel inner source and destination MAC addresses.
204 - E-Switch encapsulation Flow:
206 - can be applied to VF ports only.
207 - must specify PF port action (packet redirection from VF to PF).
211 - The input buffer, used as outer header, is not validated.
215 - The decapsulation is always done up to the outermost tunnel detected by the HW.
216 - The input buffer, providing the removal size, is not validated.
217 - The buffer size must match the length of the headers to be removed.
219 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
220 mutually exclusive features which cannot be supported together
221 (see :ref:`mlx5_firmware_config`).
225 - Requires DevX and DV flow to be enabled.
226 - KEEP_CRC offload cannot be supported with LRO.
227 - The first mbuf length, without head-room, must be big enough to include the
229 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
230 it with size limited to max LRO size, not to max RX packet length.
235 MLX5 supports various methods to report statistics:
237 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.
239 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.
241 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.
249 These options can be modified in the ``.config`` file.
251 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
253 Toggle compilation of librte_pmd_mlx5 itself.
255 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
257 Build PMD with additional code to make it loadable without hard
258 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
259 on the target system.
261 In this mode, their presence is still required for it to run properly,
262 however their absence won't prevent a DPDK application from starting (with
263 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
264 missing with ``ldd(1)``.
266 It works by moving these dependencies to a purpose-built rdma-core "glue"
267 plug-in which must either be installed in a directory whose name is based
268 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
269 standard location for the dynamic linker (e.g. ``/lib``) if left to the
270 default empty string (``""``).
272 This option has no performance impact.
274 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
276 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
277 in the PMD shared library or the executable static binary.
279 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
281 Toggle debugging code and stricter compilation flags. Enabling this option
282 adds additional run-time checks and debugging messages at the cost of
287 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
288 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
289 64. Default armv8a configuration of make build and meson build set it to 128
290 then brings performance degradation.
292 This option is available in meson:
294 - ``ibverbs_link`` can be ``static``, ``shared``, or ``dlopen``.
296 Environment variables
297 ~~~~~~~~~~~~~~~~~~~~~
301 A list of directories in which to search for the rdma-core "glue" plug-in,
302 separated by colons or semi-colons.
304 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
305 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
306 since ``LD_LIBRARY_PATH`` has no effect in this case.
308 - ``MLX5_SHUT_UP_BF``
310 Configures HW Tx doorbell register as IO-mapped.
312 By default, the HW Tx doorbell is configured as a write-combining register.
313 The register would be flushed to HW usually when the write-combining buffer
314 becomes full, but it depends on CPU design.
316 Except for vectorized Tx burst routines, a write memory barrier is enforced
317 after updating the register so that the update can be immediately visible to
320 When vectorized Tx burst is called, the barrier is set only if the burst size
321 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
322 variable will bring better latency even though the maximum throughput can
325 Run-time configuration
326 ~~~~~~~~~~~~~~~~~~~~~~
328 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
329 because it is affected by their state. Forcing them down prevents packets
332 - **ethtool** operations on related kernel interfaces also affect the PMD.
334 - ``rxq_cqe_comp_en`` parameter [int]
336 A nonzero value enables the compression of CQE on RX side. This feature
337 allows to save PCI bandwidth and improve performance. Enabled by default.
341 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
343 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
346 - ``rxq_cqe_pad_en`` parameter [int]
348 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
349 is aligned with the size of a cacheline of the core. If cacheline size is
350 128B, the CQE size is configured to be 128B even though the device writes
351 only 64B data on the cacheline. This is to avoid unnecessary cache
352 invalidation by device's two consecutive writes on to one cacheline.
353 However in some architecture, it is more beneficial to update entire
354 cacheline with padding the rest 64B rather than striding because
355 read-modify-write could drop performance a lot. On the other hand,
356 writing extra data will consume more PCIe bandwidth and could also drop
357 the maximum throughput. It is recommended to empirically set this
358 parameter. Disabled by default.
362 - CPU having 128B cacheline with ConnectX-5 and BlueField.
364 - ``rxq_pkt_pad_en`` parameter [int]
366 A nonzero value enables padding Rx packet to the size of cacheline on PCI
367 transaction. This feature would waste PCI bandwidth but could improve
368 performance by avoiding partial cacheline write which may cause costly
369 read-modify-copy in memory transaction on some architectures. Disabled by
374 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
376 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
379 - ``mprq_en`` parameter [int]
381 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
382 configured as Multi-Packet RQ if the total number of Rx queues is
383 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
386 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
387 by posting a single large buffer for multiple packets. Instead of posting a
388 buffers per a packet, one large buffer is posted in order to receive multiple
389 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
390 and each stride receives one packet. MPRQ can improve throughput for
391 small-packet traffic.
393 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
394 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
395 configure large stride size enough to accommodate max_rx_pkt_len as long as
396 device allows. Note that this can waste system memory compared to enabling Rx
397 scatter and multi-segment packet.
399 - ``mprq_log_stride_num`` parameter [int]
401 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
402 strides can reduce PCIe traffic further. If configured value is not in the
403 range of device capability, the default value will be set with a warning
404 message. The default value is 4 which is 16 strides per a buffer, valid only
405 if ``mprq_en`` is set.
407 The size of Rx queue should be bigger than the number of strides.
409 - ``mprq_max_memcpy_len`` parameter [int]
411 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
412 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
413 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
414 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
415 A mempool for external buffers will be allocated and managed by PMD. If Rx
416 packet is externally attached, ol_flags field of the mbuf will have
417 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
418 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
420 - ``rxqs_min_mprq`` parameter [int]
422 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
423 greater or equal to this value. The default value is 12, valid only if
426 - ``txq_inline`` parameter [int]
428 Amount of data to be inlined during TX operations. This parameter is
429 deprecated and converted to the new parameter ``txq_inline_max`` providing
430 partial compatibility.
432 - ``txqs_min_inline`` parameter [int]
434 Enable inline data send only when the number of TX queues is greater or equal
437 This option should be used in combination with ``txq_inline_max`` and
438 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
440 If this option is not specified the default value 16 is used for BlueField
441 and 8 for other platforms
443 The data inlining consumes the CPU cycles, so this option is intended to
444 auto enable inline data if we have enough Tx queues, which means we have
445 enough CPU cores and PCI bandwidth is getting more critical and CPU
446 is not supposed to be bottleneck anymore.
448 The copying data into WQE improves latency and can improve PPS performance
449 when PCI back pressure is detected and may be useful for scenarios involving
450 heavy traffic on many queues.
452 Because additional software logic is necessary to handle this mode, this
453 option should be used with care, as it may lower performance when back
454 pressure is not expected.
456 If inline data are enabled it may affect the maximal size of Tx queue in
457 descriptors because the inline data increase the descriptor size and
458 queue size limits supported by hardware may be exceeded.
460 - ``txq_inline_min`` parameter [int]
462 Minimal amount of data to be inlined into WQE during Tx operations. NICs
463 may require this minimal data amount to operate correctly. The exact value
464 may depend on NIC operation mode, requested offloads, etc. It is strongly
465 recommended to omit this parameter and use the default values. Anyway,
466 applications using this parameter should take into consideration that
467 specifying an inconsistent value may prevent the NIC from sending packets.
469 If ``txq_inline_min`` key is present the specified value (may be aligned
470 by the driver in order not to exceed the limits and provide better descriptor
471 space utilization) will be used by the driver and it is guaranteed that
472 requested amount of data bytes are inlined into the WQE beside other inline
473 settings. This key also may update ``txq_inline_max`` value (default
474 or specified explicitly in devargs) to reserve the space for inline data.
476 If ``txq_inline_min`` key is not present, the value may be queried by the
477 driver from the NIC via DevX if this feature is available. If there is no DevX
478 enabled/supported the value 18 (supposing L2 header including VLAN) is set
479 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
480 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
483 For ConnectX-4 NIC, driver does not allow specifying value below 18
484 (minimal L2 header, including VLAN), error will be raised.
486 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
487 it is not recommended and may prevent NIC from sending packets over
490 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
491 Multi-Packet Write), because last one does not support partial packet inlining.
492 This is not very critical due to minimal data inlining is mostly required
493 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
495 - ``txq_inline_max`` parameter [int]
497 Specifies the maximal packet length to be completely inlined into WQE
498 Ethernet Segment for ordinary SEND method. If packet is larger than specified
499 value, the packet data won't be copied by the driver at all, data buffer
500 is addressed with a pointer. If packet length is less or equal all packet
501 data will be copied into WQE. This may improve PCI bandwidth utilization for
502 short packets significantly but requires the extra CPU cycles.
504 The data inline feature is controlled by number of Tx queues, if number of Tx
505 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
506 is engaged, if there are not enough Tx queues (which means not enough CPU cores
507 and CPU resources are scarce), data inline is not performed by the driver.
508 Assigning ``txqs_min_inline`` with zero always enables the data inline.
510 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
511 by the driver in order not to exceed the limit (930 bytes) and to provide better
512 WQE space filling without gaps, the adjustment is reflected in the debug log.
513 Also, the default value (290) may be decreased in run-time if the large transmit
514 queue size is requested and hardware does not support enough descriptor
515 amount, in this case warning is emitted. If ``txq_inline_max`` key is
516 specified and requested inline settings can not be satisfied then error
519 - ``txq_inline_mpw`` parameter [int]
521 Specifies the maximal packet length to be completely inlined into WQE for
522 Enhanced MPW method. If packet is large the specified value, the packet data
523 won't be copied, and data buffer is addressed with pointer. If packet length
524 is less or equal, all packet data will be copied into WQE. This may improve PCI
525 bandwidth utilization for short packets significantly but requires the extra
528 The data inline feature is controlled by number of TX queues, if number of Tx
529 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
530 is engaged, if there are not enough Tx queues (which means not enough CPU cores
531 and CPU resources are scarce), data inline is not performed by the driver.
532 Assigning ``txqs_min_inline`` with zero always enables the data inline.
534 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
535 by the driver in order not to exceed the limit (930 bytes) and to provide better
536 WQE space filling without gaps, the adjustment is reflected in the debug log.
537 Due to multiple packets may be included to the same WQE with Enhanced Multi
538 Packet Write Method and overall WQE size is limited it is not recommended to
539 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
540 may be decreased in run-time if the large transmit queue size is requested
541 and hardware does not support enough descriptor amount, in this case warning
542 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
543 settings can not be satisfied then error will be raised.
545 - ``txqs_max_vec`` parameter [int]
547 Enable vectorized Tx only when the number of TX queues is less than or
548 equal to this value. This parameter is deprecated and ignored, kept
549 for compatibility issue to not prevent driver from probing.
551 - ``txq_mpw_hdr_dseg_en`` parameter [int]
553 A nonzero value enables including two pointers in the first block of TX
554 descriptor. The parameter is deprecated and ignored, kept for compatibility
557 - ``txq_max_inline_len`` parameter [int]
559 Maximum size of packet to be inlined. This limits the size of packet to
560 be inlined. If the size of a packet is larger than configured value, the
561 packet isn't inlined even though there's enough space remained in the
562 descriptor. Instead, the packet is included with pointer. This parameter
563 is deprecated and converted directly to ``txq_inline_mpw`` providing full
564 compatibility. Valid only if eMPW feature is engaged.
566 - ``txq_mpw_en`` parameter [int]
568 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
569 ConnectX-6, ConnectX-6 Dx and BlueField. eMPW allows the TX burst function to pack
570 up multiple packets in a single descriptor session in order to save PCI bandwidth
571 and improve performance at the cost of a slightly higher CPU usage. When
572 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
573 entire packet data on to TX descriptor instead of including pointer of packet.
575 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
576 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
577 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
578 option or reported by the NIC, the eMPW feature is disengaged.
580 - ``tx_db_nc`` parameter [int]
582 The rdma core library can map doorbell register in two ways, depending on the
583 environment variable "MLX5_SHUT_UP_BF":
585 - As regular cached memory (usually with write combining attribute), if the
586 variable is either missing or set to zero.
587 - As non-cached memory, if the variable is present and set to not "0" value.
589 The type of mapping may slightly affect the Tx performance, the optimal choice
590 is strongly relied on the host architecture and should be deduced practically.
592 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
593 memory (with write combining), the PMD will perform the extra write memory barrier
594 after writing to doorbell, it might increase the needed CPU clocks per packet
595 to send, but latency might be improved.
597 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
598 cached memory, the PMD will not perform the extra write memory barrier
599 after writing to doorbell, on some architectures it might improve the
602 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
603 memory, the PMD will use heuristics to decide whether write memory barrier
604 should be performed. For bursts with size multiple of recommended one (64 pkts)
605 it is supposed the next burst is coming and no need to issue the extra memory
606 barrier (it is supposed to be issued in the next coming burst, at least after
607 descriptor writing). It might increase latency (on some hosts till next
608 packets transmit) and should be used with care.
610 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
611 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
612 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
614 - ``tx_vec_en`` parameter [int]
616 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx
617 and BlueField NICs if the number of global Tx queues on the port is less than
618 ``txqs_max_vec``. The parameter is deprecated and ignored.
620 - ``rx_vec_en`` parameter [int]
622 A nonzero value enables Rx vector if the port is not configured in
623 multi-segment otherwise this parameter is ignored.
627 - ``vf_nl_en`` parameter [int]
629 A nonzero value enables Netlink requests from the VF to add/remove MAC
630 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
631 Otherwise the relevant configuration must be run with Linux iproute2 tools.
632 This is a prerequisite to receive this kind of traffic.
634 Enabled by default, valid only on VF devices ignored otherwise.
636 - ``l3_vxlan_en`` parameter [int]
638 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
639 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
640 parameter. This is a prerequisite to receive this kind of traffic.
644 - ``dv_xmeta_en`` parameter [int]
646 A nonzero value enables extensive flow metadata support if device is
647 capable and driver supports it. This can enable extensive support of
648 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
649 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
651 There are some possible configurations, depending on parameter value:
653 - 0, this is default value, defines the legacy mode, the ``MARK`` and
654 ``META`` related actions and items operate only within NIC Tx and
655 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
656 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
657 item is 32 bits wide and match supported on egress only.
659 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
660 related actions and items operate within all supported steering domains,
661 including FDB, ``MARK`` and ``META`` information may cross the domain
662 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
663 depends on kernel and firmware configurations and might be 0, 16 or
664 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
665 compatibility, the actual width of data transferred to the FDB domain
666 depends on kernel configuration and may be vary. The actual supported
667 width can be retrieved in runtime by series of rte_flow_validate()
670 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
671 related actions and items operate within all supported steering domains,
672 including FDB, ``MARK`` and ``META`` information may cross the domain
673 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
674 depends on kernel and firmware configurations and might be 0, 16 or
675 24 bits. The actual supported width can be retrieved in runtime by
676 series of rte_flow_validate() trials.
678 +------+-----------+-----------+-------------+-------------+
679 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
680 +======+===========+===========+=============+=============+
681 | 0 | 24 bits | 32 bits | 32 bits | no |
682 +------+-----------+-----------+-------------+-------------+
683 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
684 +------+-----------+-----------+-------------+-------------+
685 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
686 +------+-----------+-----------+-------------+-------------+
688 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
689 ignored and the device is configured to operate in legacy mode (0).
691 Disabled by default (set to 0).
693 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
694 of the extensive metadata features. The legacy Verbs supports FLAG and
695 MARK metadata actions over NIC Rx steering domain only.
697 - ``dv_flow_en`` parameter [int]
699 A nonzero value enables the DV flow steering assuming it is supported
700 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
702 Enabled by default if supported.
704 - ``dv_esw_en`` parameter [int]
706 A nonzero value enables E-Switch using Direct Rules.
708 Enabled by default if supported.
710 - ``mr_ext_memseg_en`` parameter [int]
712 A nonzero value enables extending memseg when registering DMA memory. If
713 enabled, the number of entries in MR (Memory Region) lookup table on datapath
714 is minimized and it benefits performance. On the other hand, it worsens memory
715 utilization because registered memory is pinned by kernel driver. Even if a
716 page in the extended chunk is freed, that doesn't become reusable until the
717 entire memory is freed.
721 - ``representor`` parameter [list]
723 This parameter can be used to instantiate DPDK Ethernet devices from
724 existing port (or VF) representors configured on the device.
726 It is a standard parameter whose format is described in
727 :ref:`ethernet_device_standard_device_arguments`.
729 For instance, to probe port representors 0 through 2::
733 - ``max_dump_files_num`` parameter [int]
735 The maximum number of files per PMD entity that may be created for debug information.
736 The files will be created in /var/log directory or in current directory.
738 set to 128 by default.
740 - ``lro_timeout_usec`` parameter [int]
742 The maximum allowed duration of an LRO session, in micro-seconds.
743 PMD will set the nearest value supported by HW, which is not bigger than
744 the input ``lro_timeout_usec`` value.
745 If this parameter is not specified, by default PMD will set
746 the smallest value supported by HW.
748 .. _mlx5_firmware_config:
750 Firmware configuration
751 ~~~~~~~~~~~~~~~~~~~~~~
753 Firmware features can be configured as key/value pairs.
755 The command to set a value is::
757 mlxconfig -d <device> set <key>=<value>
759 The command to query a value is::
761 mlxconfig -d <device> query | grep <key>
763 The device name for the command ``mlxconfig`` can be either the PCI address,
764 or the mst device name found with::
768 Below are some firmware configurations listed.
774 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
780 - maximum number of SR-IOV virtual functions::
784 - enable DevX (required by Direct Rules and other features)::
788 - aggressive CQE zipping::
792 - L3 VXLAN and VXLAN-GPE destination UDP port::
795 IP_OVER_VXLAN_PORT=<udp dport>
797 - enable IP-in-IP tunnel flow matching::
799 FLEX_PARSER_PROFILE_ENABLE=0
801 - enable MPLS flow matching::
803 FLEX_PARSER_PROFILE_ENABLE=1
805 - enable ICMP/ICMP6 code/type fields matching::
807 FLEX_PARSER_PROFILE_ENABLE=2
809 - enable Geneve flow matching::
811 FLEX_PARSER_PROFILE_ENABLE=0
813 - enable GTP flow matching::
815 FLEX_PARSER_PROFILE_ENABLE=3
820 This driver relies on external libraries and kernel drivers for resources
821 allocations and initialization. The following dependencies are not part of
822 DPDK and must be installed separately:
826 User space Verbs framework used by librte_pmd_mlx5. This library provides
827 a generic interface between the kernel and low-level user space drivers
830 It allows slow and privileged operations (context initialization, hardware
831 resources allocations) to be managed by the kernel and fast operations to
832 never leave user space.
836 Low-level user space driver library for Mellanox
837 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
840 This library basically implements send/receive calls to the hardware
845 They provide the kernel-side Verbs API and low level device drivers that
846 manage actual hardware initialization and resources sharing with user
849 Unlike most other PMDs, these modules must remain loaded and bound to
852 - mlx5_core: hardware driver managing Mellanox
853 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
855 - mlx5_ib: InifiniBand device driver.
856 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
858 - **Firmware update**
860 Mellanox OFED/EN releases include firmware updates for
861 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
863 Because each release provides new features, these updates must be applied to
864 match the kernel modules and libraries they come with.
868 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
874 Either RDMA Core library with a recent enough Linux kernel release
875 (recommended) or Mellanox OFED/EN, which provides compatibility with older
878 RDMA Core with Linux Kernel
879 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
881 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
882 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
883 (see `RDMA Core installation documentation`_)
884 - When building for i686 use:
886 - rdma-core version 18.0 or above built with 32bit support.
887 - Kernel version 4.14.41 or above.
889 - Starting with rdma-core v21, static libraries can be built::
892 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
895 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
896 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
898 If rdma-core libraries are built but not installed, DPDK makefile can link them,
899 thanks to these environment variables:
901 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
902 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
903 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
908 - Mellanox OFED version: ** 4.5, 4.6** /
909 Mellanox EN version: **4.5, 4.6**
912 - ConnectX-4: **12.21.1000** and above.
913 - ConnectX-4 Lx: **14.21.1000** and above.
914 - ConnectX-5: **16.21.1000** and above.
915 - ConnectX-5 Ex: **16.21.1000** and above.
916 - ConnectX-6: **20.99.5374** and above.
917 - ConnectX-6 Dx: **22.27.0090** and above.
918 - BlueField: **18.25.1010** and above.
920 While these libraries and kernel modules are available on OpenFabrics
921 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
922 managers on most distributions, this PMD requires Ethernet extensions that
923 may not be supported at the moment (this is a work in progress).
926 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
928 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
929 include the necessary support and should be used in the meantime. For DPDK,
930 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
931 required from that distribution.
935 Several versions of Mellanox OFED/EN are available. Installing the version
936 this DPDK release was developed and tested against is strongly
937 recommended. Please check the `prerequisites`_.
942 The following Mellanox device families are supported by the same mlx5 driver:
952 Below are detailed device names:
954 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
955 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
956 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
957 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
958 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
959 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
960 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
961 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
962 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
963 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
964 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
965 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
966 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
967 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
968 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
969 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
970 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
971 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
972 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
973 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
974 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
975 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
976 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
977 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
978 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
980 Quick Start Guide on OFED/EN
981 ----------------------------
983 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
986 2. Install the required libraries and kernel modules either by installing
987 only the required set, or by installing the entire Mellanox OFED/EN::
989 ./mlnxofedinstall --upstream-libs --dpdk
991 3. Verify the firmware is the correct one::
995 4. Verify all ports links are set to Ethernet::
997 mlxconfig -d <mst device> query | grep LINK_TYPE
1001 Link types may have to be configured to Ethernet::
1003 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1005 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1007 For hypervisors, verify SR-IOV is enabled on the NIC::
1009 mlxconfig -d <mst device> query | grep SRIOV_EN
1012 If needed, configure SR-IOV::
1014 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1015 mlxfwreset -d <mst device> reset
1017 5. Restart the driver::
1019 /etc/init.d/openibd restart
1023 service openibd restart
1025 If link type was changed, firmware must be reset as well::
1027 mlxfwreset -d <mst device> reset
1029 For hypervisors, after reset write the sysfs number of virtual functions
1032 To dynamically instantiate a given number of virtual functions (VFs)::
1034 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1036 6. Compile DPDK and you are ready to go. See instructions on
1037 :ref:`Development Kit Build System <Development_Kit_Build_System>`
1039 Enable switchdev mode
1040 ---------------------
1042 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1043 Representor is a port in DPDK that is connected to a VF in such a way
1044 that assuming there are no offload flows, each packet that is sent from the VF
1045 will be received by the corresponding representor. While each packet that is
1046 sent to a representor will be received by the VF.
1047 This is very useful in case of SRIOV mode, where the first packet that is sent
1048 by the VF will be received by the DPDK application which will decide if this
1049 flow should be offloaded to the E-Switch. After offloading the flow packet
1050 that the VF that are matching the flow will not be received any more by
1051 the DPDK application.
1053 1. Enable SRIOV mode::
1055 mlxconfig -d <mst device> set SRIOV_EN=true
1057 2. Configure the max number of VFs::
1059 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1063 mlxfwreset -d <mst device> reset
1065 3. Configure the actual number of VFs::
1067 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1069 4. Unbind the device (can be rebind after the switchdev mode)::
1071 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1073 5. Enbale switchdev mode::
1075 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1080 1. Configure aggressive CQE Zipping for maximum performance::
1082 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1084 To set it back to the default CQE Zipping mode use::
1086 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1088 2. In case of virtualization:
1090 - Make sure that hypervisor kernel is 3.16 or newer.
1091 - Configure boot with ``iommu=pt``.
1092 - Use 1G huge pages.
1093 - Make sure to allocate a VM on huge pages.
1094 - Make sure to set CPU pinning.
1096 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1097 for better performance. For VMs, verify that the right CPU
1098 and NUMA node are pinned according to the above. Run::
1102 to identify the NUMA node to which the PCIe adapter is connected.
1104 4. If more than one adapter is used, and root complex capabilities allow
1105 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1106 it is recommended to locate both adapters on the same NUMA node.
1107 This in order to forward packets from one to the other without
1108 NUMA performance penalty.
1110 5. Disable pause frames::
1112 ethtool -A <netdev> rx off tx off
1114 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1115 via the BIOS configuration. Please contact you server provider for more
1116 information about the settings.
1120 On some machines, depends on the machine integrator, it is beneficial
1121 to set the PCI max read request parameter to 1K. This can be
1122 done in the following way:
1124 To query the read request size use::
1126 setpci -s <NIC PCI address> 68.w
1128 If the output is different than 3XXX, set it by::
1130 setpci -s <NIC PCI address> 68.w=3XXX
1132 The XXX can be different on different systems. Make sure to configure
1133 according to the setpci output.
1135 7. To minimize overhead of searching Memory Regions:
1137 - '--socket-mem' is recommended to pin memory by predictable amount.
1138 - Configure per-lcore cache when creating Mempools for packet buffer.
1139 - Refrain from dynamically allocating/freeing memory in run-time.
1141 .. _mlx5_offloads_support:
1143 Supported hardware offloads
1144 ---------------------------
1146 .. table:: Minimal SW/HW versions for queue offloads
1148 ============== ===== ===== ========= ===== ========== ==========
1149 Offload DPDK Linux rdma-core OFED firmware hardware
1150 ============== ===== ===== ========= ===== ========== ==========
1151 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1152 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1153 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1154 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1155 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1156 ============== ===== ===== ========= ===== ========== ==========
1158 .. table:: Minimal SW/HW versions for rte_flow offloads
1160 +-----------------------+-----------------+-----------------+
1161 | Offload | with E-Switch | with NIC |
1162 +=======================+=================+=================+
1163 | Count | | DPDK 19.05 | | DPDK 19.02 |
1164 | | | OFED 4.6 | | OFED 4.6 |
1165 | | | rdma-core 24 | | rdma-core 23 |
1166 | | | ConnectX-5 | | ConnectX-5 |
1167 +-----------------------+-----------------+-----------------+
1168 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1169 | | | OFED 4.6 | | OFED 4.5 |
1170 | | | rdma-core 24 | | rdma-core 23 |
1171 | | | ConnectX-5 | | ConnectX-4 |
1172 +-----------------------+-----------------+-----------------+
1173 | Queue / RSS | | | | DPDK 18.11 |
1174 | | | N/A | | OFED 4.5 |
1175 | | | | | rdma-core 23 |
1176 | | | | | ConnectX-4 |
1177 +-----------------------+-----------------+-----------------+
1178 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1179 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1180 | | | rdma-core 24 | | rdma-core 23 |
1181 | | | ConnectX-5 | | ConnectX-5 |
1182 +-----------------------+-----------------+-----------------+
1183 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1184 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1185 | | | rdma-core 27 | | rdma-core 27 |
1186 | | | ConnectX-5 | | ConnectX-5 |
1187 +-----------------------+-----------------+-----------------+
1188 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1189 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1190 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1191 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1192 | | set_ipv6_dst / | | | | |
1193 | | set_tp_src / | | | | |
1194 | | set_tp_dst / | | | | |
1195 | | dec_ttl / | | | | |
1196 | | set_ttl / | | | | |
1197 | | set_mac_src / | | | | |
1198 | | set_mac_dst) | | | | |
1199 +-----------------------+-----------------+-----------------+
1200 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1201 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1202 | | | | rdma-core 24 | | rdma-core 24 |
1203 | | | | ConnectX-5 | | ConnectX-5 |
1204 +-----------------------+-----------------+-----------------+
1205 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1206 | | | OFED 4.7-1 | | OFED 4.7-1 |
1207 | | | rdma-core 24 | | N/A |
1208 | | | ConnectX-5 | | ConnectX-5 |
1209 +-----------------------+-----------------+-----------------+
1210 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1211 | | | OFED 4.6 | | OFED 4.5 |
1212 | | | rdma-core 24 | | rdma-core 23 |
1213 | | | ConnectX-5 | | ConnectX-4 |
1214 +-----------------------+-----------------+-----------------+
1215 | Port ID | | DPDK 19.05 | | N/A |
1216 | | | OFED 4.7-1 | | N/A |
1217 | | | rdma-core 24 | | N/A |
1218 | | | ConnectX-5 | | N/A |
1219 +-----------------------+-----------------+-----------------+
1220 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1221 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1222 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1223 | | of_set_vlan_pcp / | | | | |
1224 | | of_set_vlan_vid) | | | | |
1225 +-----------------------+-----------------+-----------------+
1226 | Hairpin | | | | DPDK 19.11 |
1227 | | | N/A | | OFED 4.7-3 |
1228 | | | | | rdma-core 26 |
1229 | | | | | ConnectX-5 |
1230 +-----------------------+-----------------+-----------------+
1231 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1232 | | | OFED 4.7-3 | | OFED 4.7-3 |
1233 | | | rdma-core 26 | | rdma-core 26 |
1234 | | | ConnectX-5 | | ConnectX-5 |
1235 +-----------------------+-----------------+-----------------+
1236 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1237 | | | OFED 4.7-3 | | OFED 4.7-3 |
1238 | | | rdma-core 26 | | rdma-core 26 |
1239 | | | ConnectX-5 | | ConnectX-5 |
1240 +-----------------------+-----------------+-----------------+
1245 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1246 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1248 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1249 command-line parameter to enable additional protocols (UDP and TCP as well
1250 as IP), the following commands must be entered from its CLI to get the same
1251 behavior as librte_pmd_mlx4::
1254 > port config all rss all
1260 This section demonstrates how to launch **testpmd** with Mellanox
1261 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1263 #. Load the kernel modules::
1265 modprobe -a ib_uverbs mlx5_core mlx5_ib
1267 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1270 /etc/init.d/openibd restart
1274 User space I/O kernel modules (uio and igb_uio) are not used and do
1275 not have to be loaded.
1277 #. Make sure Ethernet interfaces are in working order and linked to kernel
1278 verbs. Related sysfs entries should be present::
1280 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1289 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1292 for intf in eth2 eth3 eth4 eth5;
1294 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1297 sed -n 's,.*/\(.*\),-w \1,p'
1306 #. Request huge pages::
1308 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1310 #. Start testpmd with basic parameters::
1312 testpmd -l 8-15 -n 4 -w 05:00.0 -w 05:00.1 -w 06:00.0 -w 06:00.1 -- --rxq=2 --txq=2 -i
1317 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1318 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1319 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1320 PMD: librte_pmd_mlx5: 1 port(s) detected
1321 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1322 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1323 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1324 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1325 PMD: librte_pmd_mlx5: 1 port(s) detected
1326 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1327 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1328 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1329 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1330 PMD: librte_pmd_mlx5: 1 port(s) detected
1331 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1332 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1333 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1334 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1335 PMD: librte_pmd_mlx5: 1 port(s) detected
1336 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1337 Interactive-mode selected
1338 Configuring Port 0 (socket 0)
1339 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1340 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1341 Port 0: E4:1D:2D:E7:0C:FE
1342 Configuring Port 1 (socket 0)
1343 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1344 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1345 Port 1: E4:1D:2D:E7:0C:FF
1346 Configuring Port 2 (socket 0)
1347 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1348 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1349 Port 2: E4:1D:2D:E7:0C:FA
1350 Configuring Port 3 (socket 0)
1351 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1352 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1353 Port 3: E4:1D:2D:E7:0C:FB
1354 Checking link statuses...
1355 Port 0 Link Up - speed 40000 Mbps - full-duplex
1356 Port 1 Link Up - speed 40000 Mbps - full-duplex
1357 Port 2 Link Up - speed 10000 Mbps - full-duplex
1358 Port 3 Link Up - speed 10000 Mbps - full-duplex
1365 This section demonstrates how to dump flows. Currently, it's possible to dump
1366 all flows with assistance of external tools.
1368 #. 2 ways to get flow raw file:
1370 - Using testpmd CLI:
1372 .. code-block:: console
1374 testpmd> flow dump <port> <output_file>
1376 - call rte_flow_dev_dump api:
1378 .. code-block:: console
1380 rte_flow_dev_dump(port, file, NULL);
1382 #. Dump human-readable flows from raw file:
1384 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1386 .. code-block:: console
1388 mlx_steering_dump.py -f <output_file>