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.
93 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
94 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
95 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
96 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
97 - Flow insertion rate of more then million flows per second, when using Direct Rules.
98 - Support for multiple rte_flow groups.
104 - For secondary process:
106 - Forked secondary process not supported.
107 - External memory unregistered in EAL memseg list cannot be used for DMA
108 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
109 primary process and remapped to the same virtual address in secondary
110 process. If the external memory is registered by primary process but has
111 different virtual address in secondary process, unexpected error may happen.
113 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
114 specific VLAN will match for VLAN packets as well:
116 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
117 Meaning, the flow rule::
119 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
121 Will only match vlan packets with vid=3. and the flow rule::
123 flow create 0 ingress pattern eth / ipv4 / end ...
125 Will match any ipv4 packet (VLAN included).
127 - VLAN pop offload command:
129 - Flow rules having a VLAN pop offload command as one of their actions and
130 are lacking a match on VLAN as one of their items are not supported.
131 - The command is not supported on egress traffic.
133 - VLAN push offload is not supported on ingress traffic.
135 - VLAN set PCP offload is not supported on existing headers.
137 - A multi segment packet must have not more segments than reported by dev_infos_get()
138 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
139 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
140 inline settings) to 58.
142 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
143 to 0 are not supported.
145 - VXLAN TSO and checksum offloads are not supported on VM.
147 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
149 - Match on Geneve header supports the following fields only:
155 Currently, the only supported options length value is 0.
157 - VF: flow rules created on VF devices can only match traffic targeted at the
158 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
160 - Match on GTP tunnel header item supports the following fields only:
167 MAC addresses not already present in the bridge table of the associated
168 kernel network device will be added and cleaned up by the PMD when closing
169 the device. In case of ungraceful program termination, some entries may
170 remain present and should be removed manually by other means.
172 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
173 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
174 ol_flags. As the mempool for the external buffer is managed by PMD, all the
175 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
176 the external buffers will be freed by PMD and the application which still
177 holds the external buffers may be corrupted.
179 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
180 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
181 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
183 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
184 and allmulticast mode are both set to off.
185 To receive IPv6 Multicast messages on VM, explicitly set the relevant
186 MAC address using rte_eth_dev_mac_addr_add() API.
188 - The amount of descriptors in Tx queue may be limited by data inline settings.
189 Inline data require the more descriptor building blocks and overall block
190 amount may exceed the hardware supported limits. The application should
191 reduce the requested Tx size or adjust data inline settings with
192 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
194 - E-Switch decapsulation Flow:
196 - can be applied to PF port only.
197 - must specify VF port action (packet redirection from PF to VF).
198 - optionally may specify tunnel inner source and destination MAC addresses.
200 - E-Switch encapsulation Flow:
202 - can be applied to VF ports only.
203 - must specify PF port action (packet redirection from VF to PF).
205 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
206 mutually exclusive features which cannot be supported together
207 (see :ref:`mlx5_firmware_config`).
211 - Requires DevX and DV flow to be enabled.
212 - KEEP_CRC offload cannot be supported with LRO.
213 - The first mbuf length, without head-room, must be big enough to include the
215 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
216 it with size limited to max LRO size, not to max RX packet length.
221 MLX5 supports various methods to report statistics:
223 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.
225 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.
227 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.
235 These options can be modified in the ``.config`` file.
237 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
239 Toggle compilation of librte_pmd_mlx5 itself.
241 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
243 Build PMD with additional code to make it loadable without hard
244 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
245 on the target system.
247 In this mode, their presence is still required for it to run properly,
248 however their absence won't prevent a DPDK application from starting (with
249 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
250 missing with ``ldd(1)``.
252 It works by moving these dependencies to a purpose-built rdma-core "glue"
253 plug-in which must either be installed in a directory whose name is based
254 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
255 standard location for the dynamic linker (e.g. ``/lib``) if left to the
256 default empty string (``""``).
258 This option has no performance impact.
260 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
262 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
263 in the PMD shared library or the executable static binary.
265 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
267 Toggle debugging code and stricter compilation flags. Enabling this option
268 adds additional run-time checks and debugging messages at the cost of
273 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
274 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
275 64. Default armv8a configuration of make build and meson build set it to 128
276 then brings performance degradation.
278 Environment variables
279 ~~~~~~~~~~~~~~~~~~~~~
283 A list of directories in which to search for the rdma-core "glue" plug-in,
284 separated by colons or semi-colons.
286 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
287 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
288 since ``LD_LIBRARY_PATH`` has no effect in this case.
290 - ``MLX5_SHUT_UP_BF``
292 Configures HW Tx doorbell register as IO-mapped.
294 By default, the HW Tx doorbell is configured as a write-combining register.
295 The register would be flushed to HW usually when the write-combining buffer
296 becomes full, but it depends on CPU design.
298 Except for vectorized Tx burst routines, a write memory barrier is enforced
299 after updating the register so that the update can be immediately visible to
302 When vectorized Tx burst is called, the barrier is set only if the burst size
303 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
304 variable will bring better latency even though the maximum throughput can
307 Run-time configuration
308 ~~~~~~~~~~~~~~~~~~~~~~
310 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
311 because it is affected by their state. Forcing them down prevents packets
314 - **ethtool** operations on related kernel interfaces also affect the PMD.
316 - ``rxq_cqe_comp_en`` parameter [int]
318 A nonzero value enables the compression of CQE on RX side. This feature
319 allows to save PCI bandwidth and improve performance. Enabled by default.
323 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
325 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
328 - ``rxq_cqe_pad_en`` parameter [int]
330 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
331 is aligned with the size of a cacheline of the core. If cacheline size is
332 128B, the CQE size is configured to be 128B even though the device writes
333 only 64B data on the cacheline. This is to avoid unnecessary cache
334 invalidation by device's two consecutive writes on to one cacheline.
335 However in some architecture, it is more beneficial to update entire
336 cacheline with padding the rest 64B rather than striding because
337 read-modify-write could drop performance a lot. On the other hand,
338 writing extra data will consume more PCIe bandwidth and could also drop
339 the maximum throughput. It is recommended to empirically set this
340 parameter. Disabled by default.
344 - CPU having 128B cacheline with ConnectX-5 and BlueField.
346 - ``rxq_pkt_pad_en`` parameter [int]
348 A nonzero value enables padding Rx packet to the size of cacheline on PCI
349 transaction. This feature would waste PCI bandwidth but could improve
350 performance by avoiding partial cacheline write which may cause costly
351 read-modify-copy in memory transaction on some architectures. Disabled by
356 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
358 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
361 - ``mprq_en`` parameter [int]
363 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
364 configured as Multi-Packet RQ if the total number of Rx queues is
365 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
368 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
369 by posting a single large buffer for multiple packets. Instead of posting a
370 buffers per a packet, one large buffer is posted in order to receive multiple
371 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
372 and each stride receives one packet. MPRQ can improve throughput for
373 small-packet traffic.
375 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
376 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
377 configure large stride size enough to accommodate max_rx_pkt_len as long as
378 device allows. Note that this can waste system memory compared to enabling Rx
379 scatter and multi-segment packet.
381 - ``mprq_log_stride_num`` parameter [int]
383 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
384 strides can reduce PCIe traffic further. If configured value is not in the
385 range of device capability, the default value will be set with a warning
386 message. The default value is 4 which is 16 strides per a buffer, valid only
387 if ``mprq_en`` is set.
389 The size of Rx queue should be bigger than the number of strides.
391 - ``mprq_max_memcpy_len`` parameter [int]
393 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
394 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
395 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
396 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
397 A mempool for external buffers will be allocated and managed by PMD. If Rx
398 packet is externally attached, ol_flags field of the mbuf will have
399 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
400 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
402 - ``rxqs_min_mprq`` parameter [int]
404 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
405 greater or equal to this value. The default value is 12, valid only if
408 - ``txq_inline`` parameter [int]
410 Amount of data to be inlined during TX operations. This parameter is
411 deprecated and converted to the new parameter ``txq_inline_max`` providing
412 partial compatibility.
414 - ``txqs_min_inline`` parameter [int]
416 Enable inline data send only when the number of TX queues is greater or equal
419 This option should be used in combination with ``txq_inline_max`` and
420 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
422 If this option is not specified the default value 16 is used for BlueField
423 and 8 for other platforms
425 The data inlining consumes the CPU cycles, so this option is intended to
426 auto enable inline data if we have enough Tx queues, which means we have
427 enough CPU cores and PCI bandwidth is getting more critical and CPU
428 is not supposed to be bottleneck anymore.
430 The copying data into WQE improves latency and can improve PPS performance
431 when PCI back pressure is detected and may be useful for scenarios involving
432 heavy traffic on many queues.
434 Because additional software logic is necessary to handle this mode, this
435 option should be used with care, as it may lower performance when back
436 pressure is not expected.
438 If inline data are enabled it may affect the maximal size of Tx queue in
439 descriptors because the inline data increase the descriptor size and
440 queue size limits supported by hardware may be exceeded.
442 - ``txq_inline_min`` parameter [int]
444 Minimal amount of data to be inlined into WQE during Tx operations. NICs
445 may require this minimal data amount to operate correctly. The exact value
446 may depend on NIC operation mode, requested offloads, etc. It is strongly
447 recommended to omit this parameter and use the default values. Anyway,
448 applications using this parameter should take into consideration that
449 specifying an inconsistent value may prevent the NIC from sending packets.
451 If ``txq_inline_min`` key is present the specified value (may be aligned
452 by the driver in order not to exceed the limits and provide better descriptor
453 space utilization) will be used by the driver and it is guaranteed that
454 requested amount of data bytes are inlined into the WQE beside other inline
455 settings. This key also may update ``txq_inline_max`` value (default
456 or specified explicitly in devargs) to reserve the space for inline data.
458 If ``txq_inline_min`` key is not present, the value may be queried by the
459 driver from the NIC via DevX if this feature is available. If there is no DevX
460 enabled/supported the value 18 (supposing L2 header including VLAN) is set
461 for ConnectX-4 and ConnectX-4LX, and 0 is set by default for ConnectX-5
462 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
465 For ConnectX-4 NIC, driver does not allow specifying value below 18
466 (minimal L2 header, including VLAN), error will be raised.
468 For ConnectX-4LX NIC, it is allowed to specify values below 18, but
469 it is not recommended and may prevent NIC from sending packets over
472 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
473 Multi-Packet Write), because last one does not support partial packet inlining.
474 This is not very critical due to minimal data inlining is mostly required
475 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
477 - ``txq_inline_max`` parameter [int]
479 Specifies the maximal packet length to be completely inlined into WQE
480 Ethernet Segment for ordinary SEND method. If packet is larger than specified
481 value, the packet data won't be copied by the driver at all, data buffer
482 is addressed with a pointer. If packet length is less or equal all packet
483 data will be copied into WQE. This may improve PCI bandwidth utilization for
484 short packets significantly but requires the extra CPU cycles.
486 The data inline feature is controlled by number of Tx queues, if number of Tx
487 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
488 is engaged, if there are not enough Tx queues (which means not enough CPU cores
489 and CPU resources are scarce), data inline is not performed by the driver.
490 Assigning ``txqs_min_inline`` with zero always enables the data inline.
492 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
493 by the driver in order not to exceed the limit (930 bytes) and to provide better
494 WQE space filling without gaps, the adjustment is reflected in the debug log.
495 Also, the default value (290) may be decreased in run-time if the large transmit
496 queue size is requested and hardware does not support enough descriptor
497 amount, in this case warning is emitted. If ``txq_inline_max`` key is
498 specified and requested inline settings can not be satisfied then error
501 - ``txq_inline_mpw`` parameter [int]
503 Specifies the maximal packet length to be completely inlined into WQE for
504 Enhanced MPW method. If packet is large the specified value, the packet data
505 won't be copied, and data buffer is addressed with pointer. If packet length
506 is less or equal, all packet data will be copied into WQE. This may improve PCI
507 bandwidth utilization for short packets significantly but requires the extra
510 The data inline feature is controlled by number of TX queues, if number of Tx
511 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
512 is engaged, if there are not enough Tx queues (which means not enough CPU cores
513 and CPU resources are scarce), data inline is not performed by the driver.
514 Assigning ``txqs_min_inline`` with zero always enables the data inline.
516 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
517 by the driver in order not to exceed the limit (930 bytes) and to provide better
518 WQE space filling without gaps, the adjustment is reflected in the debug log.
519 Due to multiple packets may be included to the same WQE with Enhanced Multi
520 Packet Write Method and overall WQE size is limited it is not recommended to
521 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
522 may be decreased in run-time if the large transmit queue size is requested
523 and hardware does not support enough descriptor amount, in this case warning
524 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
525 settings can not be satisfied then error will be raised.
527 - ``txqs_max_vec`` parameter [int]
529 Enable vectorized Tx only when the number of TX queues is less than or
530 equal to this value. This parameter is deprecated and ignored, kept
531 for compatibility issue to not prevent driver from probing.
533 - ``txq_mpw_hdr_dseg_en`` parameter [int]
535 A nonzero value enables including two pointers in the first block of TX
536 descriptor. The parameter is deprecated and ignored, kept for compatibility
539 - ``txq_max_inline_len`` parameter [int]
541 Maximum size of packet to be inlined. This limits the size of packet to
542 be inlined. If the size of a packet is larger than configured value, the
543 packet isn't inlined even though there's enough space remained in the
544 descriptor. Instead, the packet is included with pointer. This parameter
545 is deprecated and converted directly to ``txq_inline_mpw`` providing full
546 compatibility. Valid only if eMPW feature is engaged.
548 - ``txq_mpw_en`` parameter [int]
550 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
551 ConnectX-6, ConnectX-6 DX and BlueField. eMPW allows the TX burst function to pack
552 up multiple packets in a single descriptor session in order to save PCI bandwidth
553 and improve performance at the cost of a slightly higher CPU usage. When
554 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
555 entire packet data on to TX descriptor instead of including pointer of packet.
557 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
558 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
559 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
560 option or reported by the NIC, the eMPW feature is disengaged.
562 - ``tx_db_nc`` parameter [int]
564 The rdma core library can map doorbell register in two ways, depending on the
565 environment variable "MLX5_SHUT_UP_BF":
567 - As regular cached memory, if the variable is either missing or set to zero.
568 - As non-cached memory, if the variable is present and set to not "0" value.
570 The type of mapping may slightly affect the Tx performance, the optimal choice
571 is strongly relied on the host architecture and should be deduced practically.
573 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
574 memory, the PMD will perform the extra write memory barrier after writing to
575 doorbell, it might increase the needed CPU clocks per packet to send, but
576 latency might be improved.
578 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
579 cached memory, the PMD will not perform the extra write memory barrier
580 after writing to doorbell, on some architectures it might improve the
583 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
584 memory, the PMD will use heuristics to decide whether write memory barrier
585 should be performed. For bursts with size multiple of recommended one (64 pkts)
586 it is supposed the next burst is coming and no need to issue the extra memory
587 barrier (it is supposed to be issued in the next coming burst, at least after
588 descriptor writing). It might increase latency (on some hosts till next
589 packets transmit) and should be used with care.
591 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
592 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
593 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
595 - ``tx_vec_en`` parameter [int]
597 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 DX
598 and BlueField NICs if the number of global Tx queues on the port is less than
599 ``txqs_max_vec``. The parameter is deprecated and ignored.
601 - ``rx_vec_en`` parameter [int]
603 A nonzero value enables Rx vector if the port is not configured in
604 multi-segment otherwise this parameter is ignored.
608 - ``vf_nl_en`` parameter [int]
610 A nonzero value enables Netlink requests from the VF to add/remove MAC
611 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
612 Otherwise the relevant configuration must be run with Linux iproute2 tools.
613 This is a prerequisite to receive this kind of traffic.
615 Enabled by default, valid only on VF devices ignored otherwise.
617 - ``l3_vxlan_en`` parameter [int]
619 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
620 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
621 parameter. This is a prerequisite to receive this kind of traffic.
625 - ``dv_xmeta_en`` parameter [int]
627 A nonzero value enables extensive flow metadata support if device is
628 capable and driver supports it. This can enable extensive support of
629 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
630 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
632 There are some possible configurations, depending on parameter value:
634 - 0, this is default value, defines the legacy mode, the ``MARK`` and
635 ``META`` related actions and items operate only within NIC Tx and
636 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
637 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
638 item is 32 bits wide and match supported on egress only.
640 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
641 related actions and items operate within all supported steering domains,
642 including FDB, ``MARK`` and ``META`` information may cross the domain
643 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
644 depends on kernel and firmware configurations and might be 0, 16 or
645 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
646 compatibility, the actual width of data transferred to the FDB domain
647 depends on kernel configuration and may be vary. The actual supported
648 width can be retrieved in runtime by series of rte_flow_validate()
651 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
652 related actions and items operate within all supported steering domains,
653 including FDB, ``MARK`` and ``META`` information may cross the domain
654 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
655 depends on kernel and firmware configurations and might be 0, 16 or
656 24 bits. The actual supported width can be retrieved in runtime by
657 series of rte_flow_validate() trials.
659 +------+-----------+-----------+-------------+-------------+
660 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
661 +======+===========+===========+=============+=============+
662 | 0 | 24 bits | 32 bits | 32 bits | no |
663 +------+-----------+-----------+-------------+-------------+
664 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
665 +------+-----------+-----------+-------------+-------------+
666 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
667 +------+-----------+-----------+-------------+-------------+
669 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
670 ignored and the device is configured to operate in legacy mode (0).
672 Disabled by default (set to 0).
674 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
675 of the extensive metadata features. The legacy Verbs supports FLAG and
676 MARK metadata actions over NIC Rx steering domain only.
678 - ``dv_flow_en`` parameter [int]
680 A nonzero value enables the DV flow steering assuming it is supported
681 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
683 Enabled by default if supported.
685 - ``dv_esw_en`` parameter [int]
687 A nonzero value enables E-Switch using Direct Rules.
689 Enabled by default if supported.
691 - ``mr_ext_memseg_en`` parameter [int]
693 A nonzero value enables extending memseg when registering DMA memory. If
694 enabled, the number of entries in MR (Memory Region) lookup table on datapath
695 is minimized and it benefits performance. On the other hand, it worsens memory
696 utilization because registered memory is pinned by kernel driver. Even if a
697 page in the extended chunk is freed, that doesn't become reusable until the
698 entire memory is freed.
702 - ``representor`` parameter [list]
704 This parameter can be used to instantiate DPDK Ethernet devices from
705 existing port (or VF) representors configured on the device.
707 It is a standard parameter whose format is described in
708 :ref:`ethernet_device_standard_device_arguments`.
710 For instance, to probe port representors 0 through 2::
714 - ``max_dump_files_num`` parameter [int]
716 The maximum number of files per PMD entity that may be created for debug information.
717 The files will be created in /var/log directory or in current directory.
719 set to 128 by default.
721 - ``lro_timeout_usec`` parameter [int]
723 The maximum allowed duration of an LRO session, in micro-seconds.
724 PMD will set the nearest value supported by HW, which is not bigger than
725 the input ``lro_timeout_usec`` value.
726 If this parameter is not specified, by default PMD will set
727 the smallest value supported by HW.
729 .. _mlx5_firmware_config:
731 Firmware configuration
732 ~~~~~~~~~~~~~~~~~~~~~~
734 Firmware features can be configured as key/value pairs.
736 The command to set a value is::
738 mlxconfig -d <device> set <key>=<value>
740 The command to query a value is::
742 mlxconfig -d <device> query | grep <key>
744 The device name for the command ``mlxconfig`` can be either the PCI address,
745 or the mst device name found with::
749 Below are some firmware configurations listed.
755 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
761 - maximum number of SR-IOV virtual functions::
765 - enable DevX (required by Direct Rules and other features)::
769 - aggressive CQE zipping::
773 - L3 VXLAN and VXLAN-GPE destination UDP port::
776 IP_OVER_VXLAN_PORT=<udp dport>
778 - enable IP-in-IP tunnel flow matching::
780 FLEX_PARSER_PROFILE_ENABLE=0
782 - enable MPLS flow matching::
784 FLEX_PARSER_PROFILE_ENABLE=1
786 - enable ICMP/ICMP6 code/type fields matching::
788 FLEX_PARSER_PROFILE_ENABLE=2
790 - enable Geneve flow matching::
792 FLEX_PARSER_PROFILE_ENABLE=0
794 - enable GTP flow matching::
796 FLEX_PARSER_PROFILE_ENABLE=3
801 This driver relies on external libraries and kernel drivers for resources
802 allocations and initialization. The following dependencies are not part of
803 DPDK and must be installed separately:
807 User space Verbs framework used by librte_pmd_mlx5. This library provides
808 a generic interface between the kernel and low-level user space drivers
811 It allows slow and privileged operations (context initialization, hardware
812 resources allocations) to be managed by the kernel and fast operations to
813 never leave user space.
817 Low-level user space driver library for Mellanox
818 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
821 This library basically implements send/receive calls to the hardware
826 They provide the kernel-side Verbs API and low level device drivers that
827 manage actual hardware initialization and resources sharing with user
830 Unlike most other PMDs, these modules must remain loaded and bound to
833 - mlx5_core: hardware driver managing Mellanox
834 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
836 - mlx5_ib: InifiniBand device driver.
837 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
839 - **Firmware update**
841 Mellanox OFED/EN releases include firmware updates for
842 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
844 Because each release provides new features, these updates must be applied to
845 match the kernel modules and libraries they come with.
849 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
855 Either RDMA Core library with a recent enough Linux kernel release
856 (recommended) or Mellanox OFED/EN, which provides compatibility with older
859 RDMA Core with Linux Kernel
860 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
862 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
863 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
864 (see `RDMA Core installation documentation`_)
865 - When building for i686 use:
867 - rdma-core version 18.0 or above built with 32bit support.
868 - Kernel version 4.14.41 or above.
870 - Starting with rdma-core v21, static libraries can be built::
873 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
876 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
877 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
879 If rdma-core libraries are built but not installed, DPDK makefile can link them,
880 thanks to these environment variables:
882 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
883 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
884 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
889 - Mellanox OFED version: ** 4.5, 4.6** /
890 Mellanox EN version: **4.5, 4.6**
893 - ConnectX-4: **12.21.1000** and above.
894 - ConnectX-4 Lx: **14.21.1000** and above.
895 - ConnectX-5: **16.21.1000** and above.
896 - ConnectX-5 Ex: **16.21.1000** and above.
897 - ConnectX-6: **20.99.5374** and above.
898 - ConnectX-6 DX: **22.27.0090** and above.
899 - BlueField: **18.25.1010** and above.
901 While these libraries and kernel modules are available on OpenFabrics
902 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
903 managers on most distributions, this PMD requires Ethernet extensions that
904 may not be supported at the moment (this is a work in progress).
907 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
909 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
910 include the necessary support and should be used in the meantime. For DPDK,
911 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
912 required from that distribution.
916 Several versions of Mellanox OFED/EN are available. Installing the version
917 this DPDK release was developed and tested against is strongly
918 recommended. Please check the `prerequisites`_.
923 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
924 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
925 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
926 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
927 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
928 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
929 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
930 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
931 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
932 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
933 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
934 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
935 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
936 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
937 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
938 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
939 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
940 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
941 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
942 * Mellanox(R) ConnectX(R)-6 200G MCX654106A-HCAT (4x200G)
943 * Mellanox(R) ConnectX(R)-6DX EN 100G MCX623106AN-CDAT (2*100g)
944 * Mellanox(R) ConnectX(R)-6DX EN 200G MCX623105AN-VDAT (1*200g)
946 Quick Start Guide on OFED/EN
947 ----------------------------
949 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
952 2. Install the required libraries and kernel modules either by installing
953 only the required set, or by installing the entire Mellanox OFED/EN::
955 ./mlnxofedinstall --upstream-libs --dpdk
957 3. Verify the firmware is the correct one::
961 4. Verify all ports links are set to Ethernet::
963 mlxconfig -d <mst device> query | grep LINK_TYPE
967 Link types may have to be configured to Ethernet::
969 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
971 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
973 For hypervisors, verify SR-IOV is enabled on the NIC::
975 mlxconfig -d <mst device> query | grep SRIOV_EN
978 If needed, configure SR-IOV::
980 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
981 mlxfwreset -d <mst device> reset
983 5. Restart the driver::
985 /etc/init.d/openibd restart
989 service openibd restart
991 If link type was changed, firmware must be reset as well::
993 mlxfwreset -d <mst device> reset
995 For hypervisors, after reset write the sysfs number of virtual functions
998 To dynamically instantiate a given number of virtual functions (VFs)::
1000 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1002 6. Compile DPDK and you are ready to go. See instructions on
1003 :ref:`Development Kit Build System <Development_Kit_Build_System>`
1005 Enable switchdev mode
1006 ---------------------
1008 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1009 Representor is a port in DPDK that is connected to a VF in such a way
1010 that assuming there are no offload flows, each packet that is sent from the VF
1011 will be received by the corresponding representor. While each packet that is
1012 sent to a representor will be received by the VF.
1013 This is very useful in case of SRIOV mode, where the first packet that is sent
1014 by the VF will be received by the DPDK application which will decide if this
1015 flow should be offloaded to the E-Switch. After offloading the flow packet
1016 that the VF that are matching the flow will not be received any more by
1017 the DPDK application.
1019 1. Enable SRIOV mode::
1021 mlxconfig -d <mst device> set SRIOV_EN=true
1023 2. Configure the max number of VFs::
1025 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1029 mlxfwreset -d <mst device> reset
1031 3. Configure the actual number of VFs::
1033 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1035 4. Unbind the device (can be rebind after the switchdev mode)::
1037 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1039 5. Enbale switchdev mode::
1041 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1046 1. Configure aggressive CQE Zipping for maximum performance::
1048 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1050 To set it back to the default CQE Zipping mode use::
1052 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1054 2. In case of virtualization:
1056 - Make sure that hypervisor kernel is 3.16 or newer.
1057 - Configure boot with ``iommu=pt``.
1058 - Use 1G huge pages.
1059 - Make sure to allocate a VM on huge pages.
1060 - Make sure to set CPU pinning.
1062 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1063 for better performance. For VMs, verify that the right CPU
1064 and NUMA node are pinned according to the above. Run::
1068 to identify the NUMA node to which the PCIe adapter is connected.
1070 4. If more than one adapter is used, and root complex capabilities allow
1071 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1072 it is recommended to locate both adapters on the same NUMA node.
1073 This in order to forward packets from one to the other without
1074 NUMA performance penalty.
1076 5. Disable pause frames::
1078 ethtool -A <netdev> rx off tx off
1080 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1081 via the BIOS configuration. Please contact you server provider for more
1082 information about the settings.
1086 On some machines, depends on the machine integrator, it is beneficial
1087 to set the PCI max read request parameter to 1K. This can be
1088 done in the following way:
1090 To query the read request size use::
1092 setpci -s <NIC PCI address> 68.w
1094 If the output is different than 3XXX, set it by::
1096 setpci -s <NIC PCI address> 68.w=3XXX
1098 The XXX can be different on different systems. Make sure to configure
1099 according to the setpci output.
1101 7. To minimize overhead of searching Memory Regions:
1103 - '--socket-mem' is recommended to pin memory by predictable amount.
1104 - Configure per-lcore cache when creating Mempools for packet buffer.
1105 - Refrain from dynamically allocating/freeing memory in run-time.
1107 .. _mlx5_offloads_support:
1109 Supported hardware offloads
1110 ---------------------------
1112 .. table:: Minimal SW/HW versions for queue offloads
1114 ============== ===== ===== ========= ===== ========== ==========
1115 Offload DPDK Linux rdma-core OFED firmware hardware
1116 ============== ===== ===== ========= ===== ========== ==========
1117 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1118 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1119 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1120 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1121 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1122 ============== ===== ===== ========= ===== ========== ==========
1124 .. table:: Minimal SW/HW versions for rte_flow offloads
1126 +-----------------------+-----------------+-----------------+
1127 | Offload | with E-Switch | with NIC |
1128 +=======================+=================+=================+
1129 | Count | | DPDK 19.05 | | DPDK 19.02 |
1130 | | | OFED 4.6 | | OFED 4.6 |
1131 | | | rdma-core 24 | | rdma-core 23 |
1132 | | | ConnectX-5 | | ConnectX-5 |
1133 +-----------------------+-----------------+-----------------+
1134 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1135 | | | OFED 4.6 | | OFED 4.5 |
1136 | | | rdma-core 24 | | rdma-core 23 |
1137 | | | ConnectX-5 | | ConnectX-4 |
1138 +-----------------------+-----------------+-----------------+
1139 | Queue / RSS | | | | DPDK 18.11 |
1140 | | | N/A | | OFED 4.5 |
1141 | | | | | rdma-core 23 |
1142 | | | | | ConnectX-4 |
1143 +-----------------------+-----------------+-----------------+
1144 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1145 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1146 | | | rdma-core 24 | | rdma-core 23 |
1147 | | | ConnectX-5 | | ConnectX-5 |
1148 +-----------------------+-----------------+-----------------+
1149 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1150 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1151 | | | rdma-core 27 | | rdma-core 27 |
1152 | | | ConnectX-5 | | ConnectX-5 |
1153 +-----------------------+-----------------+-----------------+
1154 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1155 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1156 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1157 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1158 | | set_ipv6_dst / | | | | |
1159 | | set_tp_src / | | | | |
1160 | | set_tp_dst / | | | | |
1161 | | dec_ttl / | | | | |
1162 | | set_ttl / | | | | |
1163 | | set_mac_src / | | | | |
1164 | | set_mac_dst) | | | | |
1165 +-----------------------+-----------------+-----------------+
1166 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1167 | | | OFED 4.7-1 | | OFED 4.7-1 |
1168 | | | rdma-core 24 | | N/A |
1169 | | | ConnectX-5 | | ConnectX-5 |
1170 +-----------------------+-----------------+-----------------+
1171 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1172 | | | OFED 4.6 | | OFED 4.5 |
1173 | | | rdma-core 24 | | rdma-core 23 |
1174 | | | ConnectX-5 | | ConnectX-4 |
1175 +-----------------------+-----------------+-----------------+
1176 | Port ID | | DPDK 19.05 | | N/A |
1177 | | | OFED 4.7-1 | | N/A |
1178 | | | rdma-core 24 | | N/A |
1179 | | | ConnectX-5 | | N/A |
1180 +-----------------------+-----------------+-----------------+
1181 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1182 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1183 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1184 | | of_set_vlan_pcp / | | | | |
1185 | | of_set_vlan_vid) | | | | |
1186 +-----------------------+-----------------+-----------------+
1187 | Hairpin | | | | DPDK 19.11 |
1188 | | | N/A | | OFED 4.7-3 |
1189 | | | | | rdma-core 26 |
1190 | | | | | ConnectX-5 |
1191 +-----------------------+-----------------+-----------------+
1192 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1193 | | | OFED 4.7-3 | | OFED 4.7-3 |
1194 | | | rdma-core 26 | | rdma-core 26 |
1195 | | | ConnectX-5 | | ConnectX-5 |
1196 +-----------------------+-----------------+-----------------+
1197 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1198 | | | OFED 4.7-3 | | OFED 4.7-3 |
1199 | | | rdma-core 26 | | rdma-core 26 |
1200 | | | ConnectX-5 | | ConnectX-5 |
1201 +-----------------------+-----------------+-----------------+
1206 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1207 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1209 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1210 command-line parameter to enable additional protocols (UDP and TCP as well
1211 as IP), the following commands must be entered from its CLI to get the same
1212 behavior as librte_pmd_mlx4::
1215 > port config all rss all
1221 This section demonstrates how to launch **testpmd** with Mellanox
1222 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1224 #. Load the kernel modules::
1226 modprobe -a ib_uverbs mlx5_core mlx5_ib
1228 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1231 /etc/init.d/openibd restart
1235 User space I/O kernel modules (uio and igb_uio) are not used and do
1236 not have to be loaded.
1238 #. Make sure Ethernet interfaces are in working order and linked to kernel
1239 verbs. Related sysfs entries should be present::
1241 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1250 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1253 for intf in eth2 eth3 eth4 eth5;
1255 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1258 sed -n 's,.*/\(.*\),-w \1,p'
1267 #. Request huge pages::
1269 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1271 #. Start testpmd with basic parameters::
1273 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
1278 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1279 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1280 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1281 PMD: librte_pmd_mlx5: 1 port(s) detected
1282 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1283 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1284 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1285 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1286 PMD: librte_pmd_mlx5: 1 port(s) detected
1287 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1288 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1289 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1290 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1291 PMD: librte_pmd_mlx5: 1 port(s) detected
1292 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1293 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1294 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1295 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1296 PMD: librte_pmd_mlx5: 1 port(s) detected
1297 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1298 Interactive-mode selected
1299 Configuring Port 0 (socket 0)
1300 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1301 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1302 Port 0: E4:1D:2D:E7:0C:FE
1303 Configuring Port 1 (socket 0)
1304 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1305 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1306 Port 1: E4:1D:2D:E7:0C:FF
1307 Configuring Port 2 (socket 0)
1308 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1309 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1310 Port 2: E4:1D:2D:E7:0C:FA
1311 Configuring Port 3 (socket 0)
1312 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1313 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1314 Port 3: E4:1D:2D:E7:0C:FB
1315 Checking link statuses...
1316 Port 0 Link Up - speed 40000 Mbps - full-duplex
1317 Port 1 Link Up - speed 40000 Mbps - full-duplex
1318 Port 2 Link Up - speed 10000 Mbps - full-duplex
1319 Port 3 Link Up - speed 10000 Mbps - full-duplex
1326 This section demonstrates how to dump flows. Currently, it's possible to dump
1327 all flows with assistance of external tools.
1329 #. 2 ways to get flow raw file:
1331 - Using testpmd CLI:
1333 .. code-block:: console
1335 testpmd> flow dump <port> <output_file>
1337 - call rte_flow_dev_dump api:
1339 .. code-block:: console
1341 rte_flow_dev_dump(port, file, NULL);
1343 #. Dump human-readable flows from raw file:
1345 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1347 .. code-block:: console
1349 mlx_steering_dump.py -f <output_file>