1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright 2015 6WIND S.A.
3 Copyright 2015 Mellanox Technologies, Ltd
8 The MLX5 poll mode driver library (**librte_pmd_mlx5**) provides support
9 for **Mellanox ConnectX-4**, **Mellanox ConnectX-4 Lx** , **Mellanox
10 ConnectX-5**, **Mellanox ConnectX-6**, **Mellanox ConnectX-6DX** and
11 **Mellanox BlueField** families of 10/25/40/50/100/200 Gb/s adapters
12 as well as their virtual functions (VF) in SR-IOV context.
14 Information and documentation about these adapters can be found on the
15 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
16 `Mellanox community <http://community.mellanox.com/welcome>`__.
18 There is also a `section dedicated to this poll mode driver
19 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
23 Due to external dependencies, this driver is disabled in default configuration
24 of the "make" build. It can be enabled with ``CONFIG_RTE_LIBRTE_MLX5_PMD=y``
25 or by using "meson" build system which will detect dependencies.
30 Besides its dependency on libibverbs (that implies libmlx5 and associated
31 kernel support), librte_pmd_mlx5 relies heavily on system calls for control
32 operations such as querying/updating the MTU and flow control parameters.
34 For security reasons and robustness, this driver only deals with virtual
35 memory addresses. The way resources allocations are handled by the kernel,
36 combined with hardware specifications that allow to handle virtual memory
37 addresses directly, ensure that DPDK applications cannot access random
38 physical memory (or memory that does not belong to the current process).
40 This capability allows the PMD to coexist with kernel network interfaces
41 which remain functional, although they stop receiving unicast packets as
42 long as they share the same MAC address.
43 This means legacy linux control tools (for example: ethtool, ifconfig and
44 more) can operate on the same network interfaces that owned by the DPDK
47 The PMD can use libibverbs and libmlx5 to access the device firmware
48 or directly the hardware components.
49 There are different levels of objects and bypassing abilities
50 to get the best performances:
52 - Verbs is a complete high-level generic API
53 - Direct Verbs is a device-specific API
54 - DevX allows to access firmware objects
55 - Direct Rules manages flow steering at low-level hardware layer
57 Enabling librte_pmd_mlx5 causes DPDK applications to be linked against
63 - Multi arch support: x86_64, POWER8, ARMv8, i686.
64 - Multiple TX and RX queues.
65 - Support for scattered TX and RX frames.
66 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
67 - RSS using different combinations of fields: L3 only, L4 only or both,
68 and source only, destination only or both.
69 - Several RSS hash keys, one for each flow type.
70 - Default RSS operation with no hash key specification.
71 - Configurable RETA table.
72 - Link flow control (pause frame).
73 - Support for multiple MAC addresses.
77 - RX CRC stripping configuration.
78 - Promiscuous mode on PF and VF.
79 - Multicast promiscuous mode on PF and VF.
80 - Hardware checksum offloads.
81 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
83 - Flow API, including :ref:`flow_isolated_mode`.
85 - KVM and VMware ESX SR-IOV modes are supported.
86 - RSS hash result is supported.
87 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
88 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
90 - Statistics query including Basic, Extended and per queue.
92 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
93 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
94 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
95 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
96 - Flow insertion rate of more then million flows per second, when using Direct Rules.
97 - Support for multiple rte_flow groups.
103 - For secondary process:
105 - Forked secondary process not supported.
106 - External memory unregistered in EAL memseg list cannot be used for DMA
107 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
108 primary process and remapped to the same virtual address in secondary
109 process. If the external memory is registered by primary process but has
110 different virtual address in secondary process, unexpected error may happen.
112 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
113 specific VLAN will match for VLAN packets as well:
115 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
116 Meaning, the flow rule::
118 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
120 Will only match vlan packets with vid=3. and the flow rule::
122 flow create 0 ingress pattern eth / ipv4 / end ...
124 Will match any ipv4 packet (VLAN included).
126 - VLAN pop offload command:
128 - Flow rules having a VLAN pop offload command as one of their actions and
129 are lacking a match on VLAN as one of their items are not supported.
130 - The command is not supported on egress traffic.
132 - VLAN push offload is not supported on ingress traffic.
134 - VLAN set PCP offload is not supported on existing headers.
136 - A multi segment packet must have not more segments than reported by dev_infos_get()
137 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
138 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
139 inline settings) to 58.
141 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
142 to 0 are not supported.
144 - VXLAN TSO and checksum offloads are not supported on VM.
146 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
148 - Match on Geneve header supports the following fields only:
154 Currently, the only supported options length value is 0.
156 - VF: flow rules created on VF devices can only match traffic targeted at the
157 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
159 - Match on GTP tunnel header item supports the following fields only:
164 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
165 The flows within group 0 and set metadata action are rejected by hardware.
169 MAC addresses not already present in the bridge table of the associated
170 kernel network device will be added and cleaned up by the PMD when closing
171 the device. In case of ungraceful program termination, some entries may
172 remain present and should be removed manually by other means.
174 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
175 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
176 ol_flags. As the mempool for the external buffer is managed by PMD, all the
177 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
178 the external buffers will be freed by PMD and the application which still
179 holds the external buffers may be corrupted.
181 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
182 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
183 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
185 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
186 and allmulticast mode are both set to off.
187 To receive IPv6 Multicast messages on VM, explicitly set the relevant
188 MAC address using rte_eth_dev_mac_addr_add() API.
190 - The amount of descriptors in Tx queue may be limited by data inline settings.
191 Inline data require the more descriptor building blocks and overall block
192 amount may exceed the hardware supported limits. The application should
193 reduce the requested Tx size or adjust data inline settings with
194 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
196 - E-Switch decapsulation Flow:
198 - can be applied to PF port only.
199 - must specify VF port action (packet redirection from PF to VF).
200 - optionally may specify tunnel inner source and destination MAC addresses.
202 - E-Switch encapsulation Flow:
204 - can be applied to VF ports only.
205 - must specify PF port action (packet redirection from VF to PF).
209 - The input buffer, used as outer header, is not validated.
213 - The decapsulation is always done up to the outermost tunnel detected by the HW.
214 - The input buffer, providing the removal size, is not validated.
215 - The buffer size must match the length of the headers to be removed.
217 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
218 mutually exclusive features which cannot be supported together
219 (see :ref:`mlx5_firmware_config`).
223 - Requires DevX and DV flow to be enabled.
224 - KEEP_CRC offload cannot be supported with LRO.
225 - The first mbuf length, without head-room, must be big enough to include the
227 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
228 it with size limited to max LRO size, not to max RX packet length.
233 MLX5 supports various methods to report statistics:
235 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.
237 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.
239 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.
247 These options can be modified in the ``.config`` file.
249 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
251 Toggle compilation of librte_pmd_mlx5 itself.
253 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
255 Build PMD with additional code to make it loadable without hard
256 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
257 on the target system.
259 In this mode, their presence is still required for it to run properly,
260 however their absence won't prevent a DPDK application from starting (with
261 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
262 missing with ``ldd(1)``.
264 It works by moving these dependencies to a purpose-built rdma-core "glue"
265 plug-in which must either be installed in a directory whose name is based
266 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
267 standard location for the dynamic linker (e.g. ``/lib``) if left to the
268 default empty string (``""``).
270 This option has no performance impact.
272 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
274 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
275 in the PMD shared library or the executable static binary.
277 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
279 Toggle debugging code and stricter compilation flags. Enabling this option
280 adds additional run-time checks and debugging messages at the cost of
285 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
286 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
287 64. Default armv8a configuration of make build and meson build set it to 128
288 then brings performance degradation.
290 This option is available in meson:
292 - ``ibverbs_link`` can be ``static``, ``shared``, or ``dlopen``.
294 Environment variables
295 ~~~~~~~~~~~~~~~~~~~~~
299 A list of directories in which to search for the rdma-core "glue" plug-in,
300 separated by colons or semi-colons.
302 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
303 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
304 since ``LD_LIBRARY_PATH`` has no effect in this case.
306 - ``MLX5_SHUT_UP_BF``
308 Configures HW Tx doorbell register as IO-mapped.
310 By default, the HW Tx doorbell is configured as a write-combining register.
311 The register would be flushed to HW usually when the write-combining buffer
312 becomes full, but it depends on CPU design.
314 Except for vectorized Tx burst routines, a write memory barrier is enforced
315 after updating the register so that the update can be immediately visible to
318 When vectorized Tx burst is called, the barrier is set only if the burst size
319 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
320 variable will bring better latency even though the maximum throughput can
323 Run-time configuration
324 ~~~~~~~~~~~~~~~~~~~~~~
326 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
327 because it is affected by their state. Forcing them down prevents packets
330 - **ethtool** operations on related kernel interfaces also affect the PMD.
332 - ``rxq_cqe_comp_en`` parameter [int]
334 A nonzero value enables the compression of CQE on RX side. This feature
335 allows to save PCI bandwidth and improve performance. Enabled by default.
339 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
341 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
344 - ``rxq_cqe_pad_en`` parameter [int]
346 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
347 is aligned with the size of a cacheline of the core. If cacheline size is
348 128B, the CQE size is configured to be 128B even though the device writes
349 only 64B data on the cacheline. This is to avoid unnecessary cache
350 invalidation by device's two consecutive writes on to one cacheline.
351 However in some architecture, it is more beneficial to update entire
352 cacheline with padding the rest 64B rather than striding because
353 read-modify-write could drop performance a lot. On the other hand,
354 writing extra data will consume more PCIe bandwidth and could also drop
355 the maximum throughput. It is recommended to empirically set this
356 parameter. Disabled by default.
360 - CPU having 128B cacheline with ConnectX-5 and BlueField.
362 - ``rxq_pkt_pad_en`` parameter [int]
364 A nonzero value enables padding Rx packet to the size of cacheline on PCI
365 transaction. This feature would waste PCI bandwidth but could improve
366 performance by avoiding partial cacheline write which may cause costly
367 read-modify-copy in memory transaction on some architectures. Disabled by
372 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
374 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
377 - ``mprq_en`` parameter [int]
379 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
380 configured as Multi-Packet RQ if the total number of Rx queues is
381 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
384 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
385 by posting a single large buffer for multiple packets. Instead of posting a
386 buffers per a packet, one large buffer is posted in order to receive multiple
387 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
388 and each stride receives one packet. MPRQ can improve throughput for
389 small-packet traffic.
391 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
392 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
393 configure large stride size enough to accommodate max_rx_pkt_len as long as
394 device allows. Note that this can waste system memory compared to enabling Rx
395 scatter and multi-segment packet.
397 - ``mprq_log_stride_num`` parameter [int]
399 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
400 strides can reduce PCIe traffic further. If configured value is not in the
401 range of device capability, the default value will be set with a warning
402 message. The default value is 4 which is 16 strides per a buffer, valid only
403 if ``mprq_en`` is set.
405 The size of Rx queue should be bigger than the number of strides.
407 - ``mprq_max_memcpy_len`` parameter [int]
409 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
410 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
411 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
412 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
413 A mempool for external buffers will be allocated and managed by PMD. If Rx
414 packet is externally attached, ol_flags field of the mbuf will have
415 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
416 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
418 - ``rxqs_min_mprq`` parameter [int]
420 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
421 greater or equal to this value. The default value is 12, valid only if
424 - ``txq_inline`` parameter [int]
426 Amount of data to be inlined during TX operations. This parameter is
427 deprecated and converted to the new parameter ``txq_inline_max`` providing
428 partial compatibility.
430 - ``txqs_min_inline`` parameter [int]
432 Enable inline data send only when the number of TX queues is greater or equal
435 This option should be used in combination with ``txq_inline_max`` and
436 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
438 If this option is not specified the default value 16 is used for BlueField
439 and 8 for other platforms
441 The data inlining consumes the CPU cycles, so this option is intended to
442 auto enable inline data if we have enough Tx queues, which means we have
443 enough CPU cores and PCI bandwidth is getting more critical and CPU
444 is not supposed to be bottleneck anymore.
446 The copying data into WQE improves latency and can improve PPS performance
447 when PCI back pressure is detected and may be useful for scenarios involving
448 heavy traffic on many queues.
450 Because additional software logic is necessary to handle this mode, this
451 option should be used with care, as it may lower performance when back
452 pressure is not expected.
454 If inline data are enabled it may affect the maximal size of Tx queue in
455 descriptors because the inline data increase the descriptor size and
456 queue size limits supported by hardware may be exceeded.
458 - ``txq_inline_min`` parameter [int]
460 Minimal amount of data to be inlined into WQE during Tx operations. NICs
461 may require this minimal data amount to operate correctly. The exact value
462 may depend on NIC operation mode, requested offloads, etc. It is strongly
463 recommended to omit this parameter and use the default values. Anyway,
464 applications using this parameter should take into consideration that
465 specifying an inconsistent value may prevent the NIC from sending packets.
467 If ``txq_inline_min`` key is present the specified value (may be aligned
468 by the driver in order not to exceed the limits and provide better descriptor
469 space utilization) will be used by the driver and it is guaranteed that
470 requested amount of data bytes are inlined into the WQE beside other inline
471 settings. This key also may update ``txq_inline_max`` value (default
472 or specified explicitly in devargs) to reserve the space for inline data.
474 If ``txq_inline_min`` key is not present, the value may be queried by the
475 driver from the NIC via DevX if this feature is available. If there is no DevX
476 enabled/supported the value 18 (supposing L2 header including VLAN) is set
477 for ConnectX-4 and ConnectX-4LX, and 0 is set by default for ConnectX-5
478 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
481 For ConnectX-4 NIC, driver does not allow specifying value below 18
482 (minimal L2 header, including VLAN), error will be raised.
484 For ConnectX-4LX NIC, it is allowed to specify values below 18, but
485 it is not recommended and may prevent NIC from sending packets over
488 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
489 Multi-Packet Write), because last one does not support partial packet inlining.
490 This is not very critical due to minimal data inlining is mostly required
491 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
493 - ``txq_inline_max`` parameter [int]
495 Specifies the maximal packet length to be completely inlined into WQE
496 Ethernet Segment for ordinary SEND method. If packet is larger than specified
497 value, the packet data won't be copied by the driver at all, data buffer
498 is addressed with a pointer. If packet length is less or equal all packet
499 data will be copied into WQE. This may improve PCI bandwidth utilization for
500 short packets significantly but requires the extra CPU cycles.
502 The data inline feature is controlled by number of Tx queues, if number of Tx
503 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
504 is engaged, if there are not enough Tx queues (which means not enough CPU cores
505 and CPU resources are scarce), data inline is not performed by the driver.
506 Assigning ``txqs_min_inline`` with zero always enables the data inline.
508 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
509 by the driver in order not to exceed the limit (930 bytes) and to provide better
510 WQE space filling without gaps, the adjustment is reflected in the debug log.
511 Also, the default value (290) may be decreased in run-time if the large transmit
512 queue size is requested and hardware does not support enough descriptor
513 amount, in this case warning is emitted. If ``txq_inline_max`` key is
514 specified and requested inline settings can not be satisfied then error
517 - ``txq_inline_mpw`` parameter [int]
519 Specifies the maximal packet length to be completely inlined into WQE for
520 Enhanced MPW method. If packet is large the specified value, the packet data
521 won't be copied, and data buffer is addressed with pointer. If packet length
522 is less or equal, all packet data will be copied into WQE. This may improve PCI
523 bandwidth utilization for short packets significantly but requires the extra
526 The data inline feature is controlled by number of TX queues, if number of Tx
527 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
528 is engaged, if there are not enough Tx queues (which means not enough CPU cores
529 and CPU resources are scarce), data inline is not performed by the driver.
530 Assigning ``txqs_min_inline`` with zero always enables the data inline.
532 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
533 by the driver in order not to exceed the limit (930 bytes) and to provide better
534 WQE space filling without gaps, the adjustment is reflected in the debug log.
535 Due to multiple packets may be included to the same WQE with Enhanced Multi
536 Packet Write Method and overall WQE size is limited it is not recommended to
537 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
538 may be decreased in run-time if the large transmit queue size is requested
539 and hardware does not support enough descriptor amount, in this case warning
540 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
541 settings can not be satisfied then error will be raised.
543 - ``txqs_max_vec`` parameter [int]
545 Enable vectorized Tx only when the number of TX queues is less than or
546 equal to this value. This parameter is deprecated and ignored, kept
547 for compatibility issue to not prevent driver from probing.
549 - ``txq_mpw_hdr_dseg_en`` parameter [int]
551 A nonzero value enables including two pointers in the first block of TX
552 descriptor. The parameter is deprecated and ignored, kept for compatibility
555 - ``txq_max_inline_len`` parameter [int]
557 Maximum size of packet to be inlined. This limits the size of packet to
558 be inlined. If the size of a packet is larger than configured value, the
559 packet isn't inlined even though there's enough space remained in the
560 descriptor. Instead, the packet is included with pointer. This parameter
561 is deprecated and converted directly to ``txq_inline_mpw`` providing full
562 compatibility. Valid only if eMPW feature is engaged.
564 - ``txq_mpw_en`` parameter [int]
566 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
567 ConnectX-6, ConnectX-6 DX and BlueField. eMPW allows the TX burst function to pack
568 up multiple packets in a single descriptor session in order to save PCI bandwidth
569 and improve performance at the cost of a slightly higher CPU usage. When
570 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
571 entire packet data on to TX descriptor instead of including pointer of packet.
573 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
574 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
575 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
576 option or reported by the NIC, the eMPW feature is disengaged.
578 - ``tx_db_nc`` parameter [int]
580 The rdma core library can map doorbell register in two ways, depending on the
581 environment variable "MLX5_SHUT_UP_BF":
583 - As regular cached memory (usually with write combining attribute), if the
584 variable is either missing or set to zero.
585 - As non-cached memory, if the variable is present and set to not "0" value.
587 The type of mapping may slightly affect the Tx performance, the optimal choice
588 is strongly relied on the host architecture and should be deduced practically.
590 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
591 memory (with write combining), the PMD will perform the extra write memory barrier
592 after writing to doorbell, it might increase the needed CPU clocks per packet
593 to send, but latency might be improved.
595 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
596 cached memory, the PMD will not perform the extra write memory barrier
597 after writing to doorbell, on some architectures it might improve the
600 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
601 memory, the PMD will use heuristics to decide whether write memory barrier
602 should be performed. For bursts with size multiple of recommended one (64 pkts)
603 it is supposed the next burst is coming and no need to issue the extra memory
604 barrier (it is supposed to be issued in the next coming burst, at least after
605 descriptor writing). It might increase latency (on some hosts till next
606 packets transmit) and should be used with care.
608 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
609 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
610 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
612 - ``tx_vec_en`` parameter [int]
614 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 DX
615 and BlueField NICs if the number of global Tx queues on the port is less than
616 ``txqs_max_vec``. The parameter is deprecated and ignored.
618 - ``rx_vec_en`` parameter [int]
620 A nonzero value enables Rx vector if the port is not configured in
621 multi-segment otherwise this parameter is ignored.
625 - ``vf_nl_en`` parameter [int]
627 A nonzero value enables Netlink requests from the VF to add/remove MAC
628 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
629 Otherwise the relevant configuration must be run with Linux iproute2 tools.
630 This is a prerequisite to receive this kind of traffic.
632 Enabled by default, valid only on VF devices ignored otherwise.
634 - ``l3_vxlan_en`` parameter [int]
636 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
637 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
638 parameter. This is a prerequisite to receive this kind of traffic.
642 - ``dv_xmeta_en`` parameter [int]
644 A nonzero value enables extensive flow metadata support if device is
645 capable and driver supports it. This can enable extensive support of
646 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
647 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
649 There are some possible configurations, depending on parameter value:
651 - 0, this is default value, defines the legacy mode, the ``MARK`` and
652 ``META`` related actions and items operate only within NIC Tx and
653 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
654 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
655 item is 32 bits wide and match supported on egress only.
657 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
658 related actions and items operate within all supported steering domains,
659 including FDB, ``MARK`` and ``META`` information may cross the domain
660 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
661 depends on kernel and firmware configurations and might be 0, 16 or
662 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
663 compatibility, the actual width of data transferred to the FDB domain
664 depends on kernel configuration and may be vary. The actual supported
665 width can be retrieved in runtime by series of rte_flow_validate()
668 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
669 related actions and items operate within all supported steering domains,
670 including FDB, ``MARK`` and ``META`` information may cross the domain
671 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
672 depends on kernel and firmware configurations and might be 0, 16 or
673 24 bits. The actual supported width can be retrieved in runtime by
674 series of rte_flow_validate() trials.
676 +------+-----------+-----------+-------------+-------------+
677 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
678 +======+===========+===========+=============+=============+
679 | 0 | 24 bits | 32 bits | 32 bits | no |
680 +------+-----------+-----------+-------------+-------------+
681 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
682 +------+-----------+-----------+-------------+-------------+
683 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
684 +------+-----------+-----------+-------------+-------------+
686 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
687 ignored and the device is configured to operate in legacy mode (0).
689 Disabled by default (set to 0).
691 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
692 of the extensive metadata features. The legacy Verbs supports FLAG and
693 MARK metadata actions over NIC Rx steering domain only.
695 - ``dv_flow_en`` parameter [int]
697 A nonzero value enables the DV flow steering assuming it is supported
698 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
700 Enabled by default if supported.
702 - ``dv_esw_en`` parameter [int]
704 A nonzero value enables E-Switch using Direct Rules.
706 Enabled by default if supported.
708 - ``mr_ext_memseg_en`` parameter [int]
710 A nonzero value enables extending memseg when registering DMA memory. If
711 enabled, the number of entries in MR (Memory Region) lookup table on datapath
712 is minimized and it benefits performance. On the other hand, it worsens memory
713 utilization because registered memory is pinned by kernel driver. Even if a
714 page in the extended chunk is freed, that doesn't become reusable until the
715 entire memory is freed.
719 - ``representor`` parameter [list]
721 This parameter can be used to instantiate DPDK Ethernet devices from
722 existing port (or VF) representors configured on the device.
724 It is a standard parameter whose format is described in
725 :ref:`ethernet_device_standard_device_arguments`.
727 For instance, to probe port representors 0 through 2::
731 - ``max_dump_files_num`` parameter [int]
733 The maximum number of files per PMD entity that may be created for debug information.
734 The files will be created in /var/log directory or in current directory.
736 set to 128 by default.
738 - ``lro_timeout_usec`` parameter [int]
740 The maximum allowed duration of an LRO session, in micro-seconds.
741 PMD will set the nearest value supported by HW, which is not bigger than
742 the input ``lro_timeout_usec`` value.
743 If this parameter is not specified, by default PMD will set
744 the smallest value supported by HW.
746 .. _mlx5_firmware_config:
748 Firmware configuration
749 ~~~~~~~~~~~~~~~~~~~~~~
751 Firmware features can be configured as key/value pairs.
753 The command to set a value is::
755 mlxconfig -d <device> set <key>=<value>
757 The command to query a value is::
759 mlxconfig -d <device> query | grep <key>
761 The device name for the command ``mlxconfig`` can be either the PCI address,
762 or the mst device name found with::
766 Below are some firmware configurations listed.
772 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
778 - maximum number of SR-IOV virtual functions::
782 - enable DevX (required by Direct Rules and other features)::
786 - aggressive CQE zipping::
790 - L3 VXLAN and VXLAN-GPE destination UDP port::
793 IP_OVER_VXLAN_PORT=<udp dport>
795 - enable IP-in-IP tunnel flow matching::
797 FLEX_PARSER_PROFILE_ENABLE=0
799 - enable MPLS flow matching::
801 FLEX_PARSER_PROFILE_ENABLE=1
803 - enable ICMP/ICMP6 code/type fields matching::
805 FLEX_PARSER_PROFILE_ENABLE=2
807 - enable Geneve flow matching::
809 FLEX_PARSER_PROFILE_ENABLE=0
811 - enable GTP flow matching::
813 FLEX_PARSER_PROFILE_ENABLE=3
818 This driver relies on external libraries and kernel drivers for resources
819 allocations and initialization. The following dependencies are not part of
820 DPDK and must be installed separately:
824 User space Verbs framework used by librte_pmd_mlx5. This library provides
825 a generic interface between the kernel and low-level user space drivers
828 It allows slow and privileged operations (context initialization, hardware
829 resources allocations) to be managed by the kernel and fast operations to
830 never leave user space.
834 Low-level user space driver library for Mellanox
835 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
838 This library basically implements send/receive calls to the hardware
843 They provide the kernel-side Verbs API and low level device drivers that
844 manage actual hardware initialization and resources sharing with user
847 Unlike most other PMDs, these modules must remain loaded and bound to
850 - mlx5_core: hardware driver managing Mellanox
851 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
853 - mlx5_ib: InifiniBand device driver.
854 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
856 - **Firmware update**
858 Mellanox OFED/EN releases include firmware updates for
859 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
861 Because each release provides new features, these updates must be applied to
862 match the kernel modules and libraries they come with.
866 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
872 Either RDMA Core library with a recent enough Linux kernel release
873 (recommended) or Mellanox OFED/EN, which provides compatibility with older
876 RDMA Core with Linux Kernel
877 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
879 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
880 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
881 (see `RDMA Core installation documentation`_)
882 - When building for i686 use:
884 - rdma-core version 18.0 or above built with 32bit support.
885 - Kernel version 4.14.41 or above.
887 - Starting with rdma-core v21, static libraries can be built::
890 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
893 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
894 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
896 If rdma-core libraries are built but not installed, DPDK makefile can link them,
897 thanks to these environment variables:
899 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
900 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
901 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
906 - Mellanox OFED version: ** 4.5, 4.6** /
907 Mellanox EN version: **4.5, 4.6**
910 - ConnectX-4: **12.21.1000** and above.
911 - ConnectX-4 Lx: **14.21.1000** and above.
912 - ConnectX-5: **16.21.1000** and above.
913 - ConnectX-5 Ex: **16.21.1000** and above.
914 - ConnectX-6: **20.99.5374** and above.
915 - ConnectX-6 DX: **22.27.0090** and above.
916 - BlueField: **18.25.1010** and above.
918 While these libraries and kernel modules are available on OpenFabrics
919 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
920 managers on most distributions, this PMD requires Ethernet extensions that
921 may not be supported at the moment (this is a work in progress).
924 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
926 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
927 include the necessary support and should be used in the meantime. For DPDK,
928 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
929 required from that distribution.
933 Several versions of Mellanox OFED/EN are available. Installing the version
934 this DPDK release was developed and tested against is strongly
935 recommended. Please check the `prerequisites`_.
940 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
941 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
942 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
943 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
944 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
945 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
946 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
947 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
948 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
949 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
950 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
951 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
952 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
953 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
954 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
955 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
956 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
957 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
958 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
959 * Mellanox(R) ConnectX(R)-6 200G MCX654106A-HCAT (4x200G)
960 * Mellanox(R) ConnectX(R)-6DX EN 100G MCX623106AN-CDAT (2*100g)
961 * Mellanox(R) ConnectX(R)-6DX EN 200G MCX623105AN-VDAT (1*200g)
963 Quick Start Guide on OFED/EN
964 ----------------------------
966 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
969 2. Install the required libraries and kernel modules either by installing
970 only the required set, or by installing the entire Mellanox OFED/EN::
972 ./mlnxofedinstall --upstream-libs --dpdk
974 3. Verify the firmware is the correct one::
978 4. Verify all ports links are set to Ethernet::
980 mlxconfig -d <mst device> query | grep LINK_TYPE
984 Link types may have to be configured to Ethernet::
986 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
988 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
990 For hypervisors, verify SR-IOV is enabled on the NIC::
992 mlxconfig -d <mst device> query | grep SRIOV_EN
995 If needed, configure SR-IOV::
997 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
998 mlxfwreset -d <mst device> reset
1000 5. Restart the driver::
1002 /etc/init.d/openibd restart
1006 service openibd restart
1008 If link type was changed, firmware must be reset as well::
1010 mlxfwreset -d <mst device> reset
1012 For hypervisors, after reset write the sysfs number of virtual functions
1015 To dynamically instantiate a given number of virtual functions (VFs)::
1017 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1019 6. Compile DPDK and you are ready to go. See instructions on
1020 :ref:`Development Kit Build System <Development_Kit_Build_System>`
1022 Enable switchdev mode
1023 ---------------------
1025 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1026 Representor is a port in DPDK that is connected to a VF in such a way
1027 that assuming there are no offload flows, each packet that is sent from the VF
1028 will be received by the corresponding representor. While each packet that is
1029 sent to a representor will be received by the VF.
1030 This is very useful in case of SRIOV mode, where the first packet that is sent
1031 by the VF will be received by the DPDK application which will decide if this
1032 flow should be offloaded to the E-Switch. After offloading the flow packet
1033 that the VF that are matching the flow will not be received any more by
1034 the DPDK application.
1036 1. Enable SRIOV mode::
1038 mlxconfig -d <mst device> set SRIOV_EN=true
1040 2. Configure the max number of VFs::
1042 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1046 mlxfwreset -d <mst device> reset
1048 3. Configure the actual number of VFs::
1050 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1052 4. Unbind the device (can be rebind after the switchdev mode)::
1054 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1056 5. Enbale switchdev mode::
1058 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1063 1. Configure aggressive CQE Zipping for maximum performance::
1065 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1067 To set it back to the default CQE Zipping mode use::
1069 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1071 2. In case of virtualization:
1073 - Make sure that hypervisor kernel is 3.16 or newer.
1074 - Configure boot with ``iommu=pt``.
1075 - Use 1G huge pages.
1076 - Make sure to allocate a VM on huge pages.
1077 - Make sure to set CPU pinning.
1079 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1080 for better performance. For VMs, verify that the right CPU
1081 and NUMA node are pinned according to the above. Run::
1085 to identify the NUMA node to which the PCIe adapter is connected.
1087 4. If more than one adapter is used, and root complex capabilities allow
1088 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1089 it is recommended to locate both adapters on the same NUMA node.
1090 This in order to forward packets from one to the other without
1091 NUMA performance penalty.
1093 5. Disable pause frames::
1095 ethtool -A <netdev> rx off tx off
1097 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1098 via the BIOS configuration. Please contact you server provider for more
1099 information about the settings.
1103 On some machines, depends on the machine integrator, it is beneficial
1104 to set the PCI max read request parameter to 1K. This can be
1105 done in the following way:
1107 To query the read request size use::
1109 setpci -s <NIC PCI address> 68.w
1111 If the output is different than 3XXX, set it by::
1113 setpci -s <NIC PCI address> 68.w=3XXX
1115 The XXX can be different on different systems. Make sure to configure
1116 according to the setpci output.
1118 7. To minimize overhead of searching Memory Regions:
1120 - '--socket-mem' is recommended to pin memory by predictable amount.
1121 - Configure per-lcore cache when creating Mempools for packet buffer.
1122 - Refrain from dynamically allocating/freeing memory in run-time.
1124 .. _mlx5_offloads_support:
1126 Supported hardware offloads
1127 ---------------------------
1129 .. table:: Minimal SW/HW versions for queue offloads
1131 ============== ===== ===== ========= ===== ========== ==========
1132 Offload DPDK Linux rdma-core OFED firmware hardware
1133 ============== ===== ===== ========= ===== ========== ==========
1134 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1135 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1136 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1137 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1138 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1139 ============== ===== ===== ========= ===== ========== ==========
1141 .. table:: Minimal SW/HW versions for rte_flow offloads
1143 +-----------------------+-----------------+-----------------+
1144 | Offload | with E-Switch | with NIC |
1145 +=======================+=================+=================+
1146 | Count | | DPDK 19.05 | | DPDK 19.02 |
1147 | | | OFED 4.6 | | OFED 4.6 |
1148 | | | rdma-core 24 | | rdma-core 23 |
1149 | | | ConnectX-5 | | ConnectX-5 |
1150 +-----------------------+-----------------+-----------------+
1151 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1152 | | | OFED 4.6 | | OFED 4.5 |
1153 | | | rdma-core 24 | | rdma-core 23 |
1154 | | | ConnectX-5 | | ConnectX-4 |
1155 +-----------------------+-----------------+-----------------+
1156 | Queue / RSS | | | | DPDK 18.11 |
1157 | | | N/A | | OFED 4.5 |
1158 | | | | | rdma-core 23 |
1159 | | | | | ConnectX-4 |
1160 +-----------------------+-----------------+-----------------+
1161 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1162 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1163 | | | rdma-core 24 | | rdma-core 23 |
1164 | | | ConnectX-5 | | ConnectX-5 |
1165 +-----------------------+-----------------+-----------------+
1166 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1167 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1168 | | | rdma-core 27 | | rdma-core 27 |
1169 | | | ConnectX-5 | | ConnectX-5 |
1170 +-----------------------+-----------------+-----------------+
1171 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1172 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1173 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1174 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1175 | | set_ipv6_dst / | | | | |
1176 | | set_tp_src / | | | | |
1177 | | set_tp_dst / | | | | |
1178 | | dec_ttl / | | | | |
1179 | | set_ttl / | | | | |
1180 | | set_mac_src / | | | | |
1181 | | set_mac_dst) | | | | |
1182 +-----------------------+-----------------+-----------------+
1183 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1184 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1185 | | | | rdma-core 24 | | rdma-core 24 |
1186 | | | | ConnectX-5 | | ConnectX-5 |
1187 +-----------------------+-----------------+-----------------+
1188 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1189 | | | OFED 4.7-1 | | OFED 4.7-1 |
1190 | | | rdma-core 24 | | N/A |
1191 | | | ConnectX-5 | | ConnectX-5 |
1192 +-----------------------+-----------------+-----------------+
1193 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1194 | | | OFED 4.6 | | OFED 4.5 |
1195 | | | rdma-core 24 | | rdma-core 23 |
1196 | | | ConnectX-5 | | ConnectX-4 |
1197 +-----------------------+-----------------+-----------------+
1198 | Port ID | | DPDK 19.05 | | N/A |
1199 | | | OFED 4.7-1 | | N/A |
1200 | | | rdma-core 24 | | N/A |
1201 | | | ConnectX-5 | | N/A |
1202 +-----------------------+-----------------+-----------------+
1203 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1204 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1205 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1206 | | of_set_vlan_pcp / | | | | |
1207 | | of_set_vlan_vid) | | | | |
1208 +-----------------------+-----------------+-----------------+
1209 | Hairpin | | | | DPDK 19.11 |
1210 | | | N/A | | OFED 4.7-3 |
1211 | | | | | rdma-core 26 |
1212 | | | | | ConnectX-5 |
1213 +-----------------------+-----------------+-----------------+
1214 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1215 | | | OFED 4.7-3 | | OFED 4.7-3 |
1216 | | | rdma-core 26 | | rdma-core 26 |
1217 | | | ConnectX-5 | | ConnectX-5 |
1218 +-----------------------+-----------------+-----------------+
1219 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1220 | | | OFED 4.7-3 | | OFED 4.7-3 |
1221 | | | rdma-core 26 | | rdma-core 26 |
1222 | | | ConnectX-5 | | ConnectX-5 |
1223 +-----------------------+-----------------+-----------------+
1228 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1229 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1231 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1232 command-line parameter to enable additional protocols (UDP and TCP as well
1233 as IP), the following commands must be entered from its CLI to get the same
1234 behavior as librte_pmd_mlx4::
1237 > port config all rss all
1243 This section demonstrates how to launch **testpmd** with Mellanox
1244 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1246 #. Load the kernel modules::
1248 modprobe -a ib_uverbs mlx5_core mlx5_ib
1250 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1253 /etc/init.d/openibd restart
1257 User space I/O kernel modules (uio and igb_uio) are not used and do
1258 not have to be loaded.
1260 #. Make sure Ethernet interfaces are in working order and linked to kernel
1261 verbs. Related sysfs entries should be present::
1263 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1272 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1275 for intf in eth2 eth3 eth4 eth5;
1277 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1280 sed -n 's,.*/\(.*\),-w \1,p'
1289 #. Request huge pages::
1291 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1293 #. Start testpmd with basic parameters::
1295 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
1300 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1301 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1302 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1303 PMD: librte_pmd_mlx5: 1 port(s) detected
1304 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1305 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1306 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1307 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1308 PMD: librte_pmd_mlx5: 1 port(s) detected
1309 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1310 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1311 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1312 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1313 PMD: librte_pmd_mlx5: 1 port(s) detected
1314 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1315 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1316 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1317 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1318 PMD: librte_pmd_mlx5: 1 port(s) detected
1319 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1320 Interactive-mode selected
1321 Configuring Port 0 (socket 0)
1322 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1323 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1324 Port 0: E4:1D:2D:E7:0C:FE
1325 Configuring Port 1 (socket 0)
1326 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1327 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1328 Port 1: E4:1D:2D:E7:0C:FF
1329 Configuring Port 2 (socket 0)
1330 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1331 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1332 Port 2: E4:1D:2D:E7:0C:FA
1333 Configuring Port 3 (socket 0)
1334 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1335 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1336 Port 3: E4:1D:2D:E7:0C:FB
1337 Checking link statuses...
1338 Port 0 Link Up - speed 40000 Mbps - full-duplex
1339 Port 1 Link Up - speed 40000 Mbps - full-duplex
1340 Port 2 Link Up - speed 10000 Mbps - full-duplex
1341 Port 3 Link Up - speed 10000 Mbps - full-duplex
1348 This section demonstrates how to dump flows. Currently, it's possible to dump
1349 all flows with assistance of external tools.
1351 #. 2 ways to get flow raw file:
1353 - Using testpmd CLI:
1355 .. code-block:: console
1357 testpmd> flow dump <port> <output_file>
1359 - call rte_flow_dev_dump api:
1361 .. code-block:: console
1363 rte_flow_dev_dump(port, file, NULL);
1365 #. Dump human-readable flows from raw file:
1367 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1369 .. code-block:: console
1371 mlx_steering_dump.py -f <output_file>