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).
207 - The input buffer, used as outer header, is not validated.
211 - The decapsulation is always done up to the outermost tunnel detected by the HW.
212 - The input buffer, providing the removal size, is not validated.
213 - The buffer size must match the length of the headers to be removed.
215 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
216 mutually exclusive features which cannot be supported together
217 (see :ref:`mlx5_firmware_config`).
221 - Requires DevX and DV flow to be enabled.
222 - KEEP_CRC offload cannot be supported with LRO.
223 - The first mbuf length, without head-room, must be big enough to include the
225 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
226 it with size limited to max LRO size, not to max RX packet length.
231 MLX5 supports various methods to report statistics:
233 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.
235 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.
237 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.
245 These options can be modified in the ``.config`` file.
247 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
249 Toggle compilation of librte_pmd_mlx5 itself.
251 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
253 Build PMD with additional code to make it loadable without hard
254 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
255 on the target system.
257 In this mode, their presence is still required for it to run properly,
258 however their absence won't prevent a DPDK application from starting (with
259 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
260 missing with ``ldd(1)``.
262 It works by moving these dependencies to a purpose-built rdma-core "glue"
263 plug-in which must either be installed in a directory whose name is based
264 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
265 standard location for the dynamic linker (e.g. ``/lib``) if left to the
266 default empty string (``""``).
268 This option has no performance impact.
270 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
272 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
273 in the PMD shared library or the executable static binary.
275 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
277 Toggle debugging code and stricter compilation flags. Enabling this option
278 adds additional run-time checks and debugging messages at the cost of
283 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
284 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
285 64. Default armv8a configuration of make build and meson build set it to 128
286 then brings performance degradation.
288 This option is available in meson:
290 - ``ibverbs_link`` can be ``static``, ``shared``, or ``dlopen``.
292 Environment variables
293 ~~~~~~~~~~~~~~~~~~~~~
297 A list of directories in which to search for the rdma-core "glue" plug-in,
298 separated by colons or semi-colons.
300 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
301 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
302 since ``LD_LIBRARY_PATH`` has no effect in this case.
304 - ``MLX5_SHUT_UP_BF``
306 Configures HW Tx doorbell register as IO-mapped.
308 By default, the HW Tx doorbell is configured as a write-combining register.
309 The register would be flushed to HW usually when the write-combining buffer
310 becomes full, but it depends on CPU design.
312 Except for vectorized Tx burst routines, a write memory barrier is enforced
313 after updating the register so that the update can be immediately visible to
316 When vectorized Tx burst is called, the barrier is set only if the burst size
317 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
318 variable will bring better latency even though the maximum throughput can
321 Run-time configuration
322 ~~~~~~~~~~~~~~~~~~~~~~
324 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
325 because it is affected by their state. Forcing them down prevents packets
328 - **ethtool** operations on related kernel interfaces also affect the PMD.
330 - ``rxq_cqe_comp_en`` parameter [int]
332 A nonzero value enables the compression of CQE on RX side. This feature
333 allows to save PCI bandwidth and improve performance. Enabled by default.
337 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
339 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
342 - ``rxq_cqe_pad_en`` parameter [int]
344 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
345 is aligned with the size of a cacheline of the core. If cacheline size is
346 128B, the CQE size is configured to be 128B even though the device writes
347 only 64B data on the cacheline. This is to avoid unnecessary cache
348 invalidation by device's two consecutive writes on to one cacheline.
349 However in some architecture, it is more beneficial to update entire
350 cacheline with padding the rest 64B rather than striding because
351 read-modify-write could drop performance a lot. On the other hand,
352 writing extra data will consume more PCIe bandwidth and could also drop
353 the maximum throughput. It is recommended to empirically set this
354 parameter. Disabled by default.
358 - CPU having 128B cacheline with ConnectX-5 and BlueField.
360 - ``rxq_pkt_pad_en`` parameter [int]
362 A nonzero value enables padding Rx packet to the size of cacheline on PCI
363 transaction. This feature would waste PCI bandwidth but could improve
364 performance by avoiding partial cacheline write which may cause costly
365 read-modify-copy in memory transaction on some architectures. Disabled by
370 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
372 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
375 - ``mprq_en`` parameter [int]
377 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
378 configured as Multi-Packet RQ if the total number of Rx queues is
379 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
382 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
383 by posting a single large buffer for multiple packets. Instead of posting a
384 buffers per a packet, one large buffer is posted in order to receive multiple
385 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
386 and each stride receives one packet. MPRQ can improve throughput for
387 small-packet traffic.
389 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
390 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
391 configure large stride size enough to accommodate max_rx_pkt_len as long as
392 device allows. Note that this can waste system memory compared to enabling Rx
393 scatter and multi-segment packet.
395 - ``mprq_log_stride_num`` parameter [int]
397 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
398 strides can reduce PCIe traffic further. If configured value is not in the
399 range of device capability, the default value will be set with a warning
400 message. The default value is 4 which is 16 strides per a buffer, valid only
401 if ``mprq_en`` is set.
403 The size of Rx queue should be bigger than the number of strides.
405 - ``mprq_max_memcpy_len`` parameter [int]
407 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
408 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
409 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
410 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
411 A mempool for external buffers will be allocated and managed by PMD. If Rx
412 packet is externally attached, ol_flags field of the mbuf will have
413 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
414 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
416 - ``rxqs_min_mprq`` parameter [int]
418 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
419 greater or equal to this value. The default value is 12, valid only if
422 - ``txq_inline`` parameter [int]
424 Amount of data to be inlined during TX operations. This parameter is
425 deprecated and converted to the new parameter ``txq_inline_max`` providing
426 partial compatibility.
428 - ``txqs_min_inline`` parameter [int]
430 Enable inline data send only when the number of TX queues is greater or equal
433 This option should be used in combination with ``txq_inline_max`` and
434 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
436 If this option is not specified the default value 16 is used for BlueField
437 and 8 for other platforms
439 The data inlining consumes the CPU cycles, so this option is intended to
440 auto enable inline data if we have enough Tx queues, which means we have
441 enough CPU cores and PCI bandwidth is getting more critical and CPU
442 is not supposed to be bottleneck anymore.
444 The copying data into WQE improves latency and can improve PPS performance
445 when PCI back pressure is detected and may be useful for scenarios involving
446 heavy traffic on many queues.
448 Because additional software logic is necessary to handle this mode, this
449 option should be used with care, as it may lower performance when back
450 pressure is not expected.
452 If inline data are enabled it may affect the maximal size of Tx queue in
453 descriptors because the inline data increase the descriptor size and
454 queue size limits supported by hardware may be exceeded.
456 - ``txq_inline_min`` parameter [int]
458 Minimal amount of data to be inlined into WQE during Tx operations. NICs
459 may require this minimal data amount to operate correctly. The exact value
460 may depend on NIC operation mode, requested offloads, etc. It is strongly
461 recommended to omit this parameter and use the default values. Anyway,
462 applications using this parameter should take into consideration that
463 specifying an inconsistent value may prevent the NIC from sending packets.
465 If ``txq_inline_min`` key is present the specified value (may be aligned
466 by the driver in order not to exceed the limits and provide better descriptor
467 space utilization) will be used by the driver and it is guaranteed that
468 requested amount of data bytes are inlined into the WQE beside other inline
469 settings. This key also may update ``txq_inline_max`` value (default
470 or specified explicitly in devargs) to reserve the space for inline data.
472 If ``txq_inline_min`` key is not present, the value may be queried by the
473 driver from the NIC via DevX if this feature is available. If there is no DevX
474 enabled/supported the value 18 (supposing L2 header including VLAN) is set
475 for ConnectX-4 and ConnectX-4LX, and 0 is set by default for ConnectX-5
476 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
479 For ConnectX-4 NIC, driver does not allow specifying value below 18
480 (minimal L2 header, including VLAN), error will be raised.
482 For ConnectX-4LX NIC, it is allowed to specify values below 18, but
483 it is not recommended and may prevent NIC from sending packets over
486 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
487 Multi-Packet Write), because last one does not support partial packet inlining.
488 This is not very critical due to minimal data inlining is mostly required
489 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
491 - ``txq_inline_max`` parameter [int]
493 Specifies the maximal packet length to be completely inlined into WQE
494 Ethernet Segment for ordinary SEND method. If packet is larger than specified
495 value, the packet data won't be copied by the driver at all, data buffer
496 is addressed with a pointer. If packet length is less or equal all packet
497 data will be copied into WQE. This may improve PCI bandwidth utilization for
498 short packets significantly but requires the extra CPU cycles.
500 The data inline feature is controlled by number of Tx queues, if number of Tx
501 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
502 is engaged, if there are not enough Tx queues (which means not enough CPU cores
503 and CPU resources are scarce), data inline is not performed by the driver.
504 Assigning ``txqs_min_inline`` with zero always enables the data inline.
506 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
507 by the driver in order not to exceed the limit (930 bytes) and to provide better
508 WQE space filling without gaps, the adjustment is reflected in the debug log.
509 Also, the default value (290) may be decreased in run-time if the large transmit
510 queue size is requested and hardware does not support enough descriptor
511 amount, in this case warning is emitted. If ``txq_inline_max`` key is
512 specified and requested inline settings can not be satisfied then error
515 - ``txq_inline_mpw`` parameter [int]
517 Specifies the maximal packet length to be completely inlined into WQE for
518 Enhanced MPW method. If packet is large the specified value, the packet data
519 won't be copied, and data buffer is addressed with pointer. If packet length
520 is less or equal, all packet data will be copied into WQE. This may improve PCI
521 bandwidth utilization for short packets significantly but requires the extra
524 The data inline feature is controlled by number of TX queues, if number of Tx
525 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
526 is engaged, if there are not enough Tx queues (which means not enough CPU cores
527 and CPU resources are scarce), data inline is not performed by the driver.
528 Assigning ``txqs_min_inline`` with zero always enables the data inline.
530 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
531 by the driver in order not to exceed the limit (930 bytes) and to provide better
532 WQE space filling without gaps, the adjustment is reflected in the debug log.
533 Due to multiple packets may be included to the same WQE with Enhanced Multi
534 Packet Write Method and overall WQE size is limited it is not recommended to
535 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
536 may be decreased in run-time if the large transmit queue size is requested
537 and hardware does not support enough descriptor amount, in this case warning
538 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
539 settings can not be satisfied then error will be raised.
541 - ``txqs_max_vec`` parameter [int]
543 Enable vectorized Tx only when the number of TX queues is less than or
544 equal to this value. This parameter is deprecated and ignored, kept
545 for compatibility issue to not prevent driver from probing.
547 - ``txq_mpw_hdr_dseg_en`` parameter [int]
549 A nonzero value enables including two pointers in the first block of TX
550 descriptor. The parameter is deprecated and ignored, kept for compatibility
553 - ``txq_max_inline_len`` parameter [int]
555 Maximum size of packet to be inlined. This limits the size of packet to
556 be inlined. If the size of a packet is larger than configured value, the
557 packet isn't inlined even though there's enough space remained in the
558 descriptor. Instead, the packet is included with pointer. This parameter
559 is deprecated and converted directly to ``txq_inline_mpw`` providing full
560 compatibility. Valid only if eMPW feature is engaged.
562 - ``txq_mpw_en`` parameter [int]
564 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
565 ConnectX-6, ConnectX-6 DX and BlueField. eMPW allows the TX burst function to pack
566 up multiple packets in a single descriptor session in order to save PCI bandwidth
567 and improve performance at the cost of a slightly higher CPU usage. When
568 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
569 entire packet data on to TX descriptor instead of including pointer of packet.
571 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
572 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
573 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
574 option or reported by the NIC, the eMPW feature is disengaged.
576 - ``tx_db_nc`` parameter [int]
578 The rdma core library can map doorbell register in two ways, depending on the
579 environment variable "MLX5_SHUT_UP_BF":
581 - As regular cached memory (usually with write combining attribute), if the
582 variable is either missing or set to zero.
583 - As non-cached memory, if the variable is present and set to not "0" value.
585 The type of mapping may slightly affect the Tx performance, the optimal choice
586 is strongly relied on the host architecture and should be deduced practically.
588 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
589 memory (with write combining), the PMD will perform the extra write memory barrier
590 after writing to doorbell, it might increase the needed CPU clocks per packet
591 to send, but latency might be improved.
593 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
594 cached memory, the PMD will not perform the extra write memory barrier
595 after writing to doorbell, on some architectures it might improve the
598 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
599 memory, the PMD will use heuristics to decide whether write memory barrier
600 should be performed. For bursts with size multiple of recommended one (64 pkts)
601 it is supposed the next burst is coming and no need to issue the extra memory
602 barrier (it is supposed to be issued in the next coming burst, at least after
603 descriptor writing). It might increase latency (on some hosts till next
604 packets transmit) and should be used with care.
606 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
607 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
608 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
610 - ``tx_vec_en`` parameter [int]
612 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 DX
613 and BlueField NICs if the number of global Tx queues on the port is less than
614 ``txqs_max_vec``. The parameter is deprecated and ignored.
616 - ``rx_vec_en`` parameter [int]
618 A nonzero value enables Rx vector if the port is not configured in
619 multi-segment otherwise this parameter is ignored.
623 - ``vf_nl_en`` parameter [int]
625 A nonzero value enables Netlink requests from the VF to add/remove MAC
626 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
627 Otherwise the relevant configuration must be run with Linux iproute2 tools.
628 This is a prerequisite to receive this kind of traffic.
630 Enabled by default, valid only on VF devices ignored otherwise.
632 - ``l3_vxlan_en`` parameter [int]
634 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
635 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
636 parameter. This is a prerequisite to receive this kind of traffic.
640 - ``dv_xmeta_en`` parameter [int]
642 A nonzero value enables extensive flow metadata support if device is
643 capable and driver supports it. This can enable extensive support of
644 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
645 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
647 There are some possible configurations, depending on parameter value:
649 - 0, this is default value, defines the legacy mode, the ``MARK`` and
650 ``META`` related actions and items operate only within NIC Tx and
651 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
652 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
653 item is 32 bits wide and match supported on egress only.
655 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
656 related actions and items operate within all supported steering domains,
657 including FDB, ``MARK`` and ``META`` information may cross the domain
658 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
659 depends on kernel and firmware configurations and might be 0, 16 or
660 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
661 compatibility, the actual width of data transferred to the FDB domain
662 depends on kernel configuration and may be vary. The actual supported
663 width can be retrieved in runtime by series of rte_flow_validate()
666 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
667 related actions and items operate within all supported steering domains,
668 including FDB, ``MARK`` and ``META`` information may cross the domain
669 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
670 depends on kernel and firmware configurations and might be 0, 16 or
671 24 bits. The actual supported width can be retrieved in runtime by
672 series of rte_flow_validate() trials.
674 +------+-----------+-----------+-------------+-------------+
675 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
676 +======+===========+===========+=============+=============+
677 | 0 | 24 bits | 32 bits | 32 bits | no |
678 +------+-----------+-----------+-------------+-------------+
679 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
680 +------+-----------+-----------+-------------+-------------+
681 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
682 +------+-----------+-----------+-------------+-------------+
684 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
685 ignored and the device is configured to operate in legacy mode (0).
687 Disabled by default (set to 0).
689 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
690 of the extensive metadata features. The legacy Verbs supports FLAG and
691 MARK metadata actions over NIC Rx steering domain only.
693 - ``dv_flow_en`` parameter [int]
695 A nonzero value enables the DV flow steering assuming it is supported
696 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
698 Enabled by default if supported.
700 - ``dv_esw_en`` parameter [int]
702 A nonzero value enables E-Switch using Direct Rules.
704 Enabled by default if supported.
706 - ``mr_ext_memseg_en`` parameter [int]
708 A nonzero value enables extending memseg when registering DMA memory. If
709 enabled, the number of entries in MR (Memory Region) lookup table on datapath
710 is minimized and it benefits performance. On the other hand, it worsens memory
711 utilization because registered memory is pinned by kernel driver. Even if a
712 page in the extended chunk is freed, that doesn't become reusable until the
713 entire memory is freed.
717 - ``representor`` parameter [list]
719 This parameter can be used to instantiate DPDK Ethernet devices from
720 existing port (or VF) representors configured on the device.
722 It is a standard parameter whose format is described in
723 :ref:`ethernet_device_standard_device_arguments`.
725 For instance, to probe port representors 0 through 2::
729 - ``max_dump_files_num`` parameter [int]
731 The maximum number of files per PMD entity that may be created for debug information.
732 The files will be created in /var/log directory or in current directory.
734 set to 128 by default.
736 - ``lro_timeout_usec`` parameter [int]
738 The maximum allowed duration of an LRO session, in micro-seconds.
739 PMD will set the nearest value supported by HW, which is not bigger than
740 the input ``lro_timeout_usec`` value.
741 If this parameter is not specified, by default PMD will set
742 the smallest value supported by HW.
744 .. _mlx5_firmware_config:
746 Firmware configuration
747 ~~~~~~~~~~~~~~~~~~~~~~
749 Firmware features can be configured as key/value pairs.
751 The command to set a value is::
753 mlxconfig -d <device> set <key>=<value>
755 The command to query a value is::
757 mlxconfig -d <device> query | grep <key>
759 The device name for the command ``mlxconfig`` can be either the PCI address,
760 or the mst device name found with::
764 Below are some firmware configurations listed.
770 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
776 - maximum number of SR-IOV virtual functions::
780 - enable DevX (required by Direct Rules and other features)::
784 - aggressive CQE zipping::
788 - L3 VXLAN and VXLAN-GPE destination UDP port::
791 IP_OVER_VXLAN_PORT=<udp dport>
793 - enable IP-in-IP tunnel flow matching::
795 FLEX_PARSER_PROFILE_ENABLE=0
797 - enable MPLS flow matching::
799 FLEX_PARSER_PROFILE_ENABLE=1
801 - enable ICMP/ICMP6 code/type fields matching::
803 FLEX_PARSER_PROFILE_ENABLE=2
805 - enable Geneve flow matching::
807 FLEX_PARSER_PROFILE_ENABLE=0
809 - enable GTP flow matching::
811 FLEX_PARSER_PROFILE_ENABLE=3
816 This driver relies on external libraries and kernel drivers for resources
817 allocations and initialization. The following dependencies are not part of
818 DPDK and must be installed separately:
822 User space Verbs framework used by librte_pmd_mlx5. This library provides
823 a generic interface between the kernel and low-level user space drivers
826 It allows slow and privileged operations (context initialization, hardware
827 resources allocations) to be managed by the kernel and fast operations to
828 never leave user space.
832 Low-level user space driver library for Mellanox
833 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
836 This library basically implements send/receive calls to the hardware
841 They provide the kernel-side Verbs API and low level device drivers that
842 manage actual hardware initialization and resources sharing with user
845 Unlike most other PMDs, these modules must remain loaded and bound to
848 - mlx5_core: hardware driver managing Mellanox
849 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
851 - mlx5_ib: InifiniBand device driver.
852 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
854 - **Firmware update**
856 Mellanox OFED/EN releases include firmware updates for
857 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
859 Because each release provides new features, these updates must be applied to
860 match the kernel modules and libraries they come with.
864 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
870 Either RDMA Core library with a recent enough Linux kernel release
871 (recommended) or Mellanox OFED/EN, which provides compatibility with older
874 RDMA Core with Linux Kernel
875 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
877 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
878 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
879 (see `RDMA Core installation documentation`_)
880 - When building for i686 use:
882 - rdma-core version 18.0 or above built with 32bit support.
883 - Kernel version 4.14.41 or above.
885 - Starting with rdma-core v21, static libraries can be built::
888 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
891 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
892 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
894 If rdma-core libraries are built but not installed, DPDK makefile can link them,
895 thanks to these environment variables:
897 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
898 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
899 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
904 - Mellanox OFED version: ** 4.5, 4.6** /
905 Mellanox EN version: **4.5, 4.6**
908 - ConnectX-4: **12.21.1000** and above.
909 - ConnectX-4 Lx: **14.21.1000** and above.
910 - ConnectX-5: **16.21.1000** and above.
911 - ConnectX-5 Ex: **16.21.1000** and above.
912 - ConnectX-6: **20.99.5374** and above.
913 - ConnectX-6 DX: **22.27.0090** and above.
914 - BlueField: **18.25.1010** and above.
916 While these libraries and kernel modules are available on OpenFabrics
917 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
918 managers on most distributions, this PMD requires Ethernet extensions that
919 may not be supported at the moment (this is a work in progress).
922 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
924 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
925 include the necessary support and should be used in the meantime. For DPDK,
926 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
927 required from that distribution.
931 Several versions of Mellanox OFED/EN are available. Installing the version
932 this DPDK release was developed and tested against is strongly
933 recommended. Please check the `prerequisites`_.
938 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
939 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
940 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
941 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
942 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
943 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
944 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
945 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
946 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
947 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
948 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
949 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
950 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
951 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
952 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
953 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
954 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
955 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
956 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
957 * Mellanox(R) ConnectX(R)-6 200G MCX654106A-HCAT (4x200G)
958 * Mellanox(R) ConnectX(R)-6DX EN 100G MCX623106AN-CDAT (2*100g)
959 * Mellanox(R) ConnectX(R)-6DX EN 200G MCX623105AN-VDAT (1*200g)
961 Quick Start Guide on OFED/EN
962 ----------------------------
964 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
967 2. Install the required libraries and kernel modules either by installing
968 only the required set, or by installing the entire Mellanox OFED/EN::
970 ./mlnxofedinstall --upstream-libs --dpdk
972 3. Verify the firmware is the correct one::
976 4. Verify all ports links are set to Ethernet::
978 mlxconfig -d <mst device> query | grep LINK_TYPE
982 Link types may have to be configured to Ethernet::
984 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
986 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
988 For hypervisors, verify SR-IOV is enabled on the NIC::
990 mlxconfig -d <mst device> query | grep SRIOV_EN
993 If needed, configure SR-IOV::
995 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
996 mlxfwreset -d <mst device> reset
998 5. Restart the driver::
1000 /etc/init.d/openibd restart
1004 service openibd restart
1006 If link type was changed, firmware must be reset as well::
1008 mlxfwreset -d <mst device> reset
1010 For hypervisors, after reset write the sysfs number of virtual functions
1013 To dynamically instantiate a given number of virtual functions (VFs)::
1015 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1017 6. Compile DPDK and you are ready to go. See instructions on
1018 :ref:`Development Kit Build System <Development_Kit_Build_System>`
1020 Enable switchdev mode
1021 ---------------------
1023 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1024 Representor is a port in DPDK that is connected to a VF in such a way
1025 that assuming there are no offload flows, each packet that is sent from the VF
1026 will be received by the corresponding representor. While each packet that is
1027 sent to a representor will be received by the VF.
1028 This is very useful in case of SRIOV mode, where the first packet that is sent
1029 by the VF will be received by the DPDK application which will decide if this
1030 flow should be offloaded to the E-Switch. After offloading the flow packet
1031 that the VF that are matching the flow will not be received any more by
1032 the DPDK application.
1034 1. Enable SRIOV mode::
1036 mlxconfig -d <mst device> set SRIOV_EN=true
1038 2. Configure the max number of VFs::
1040 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1044 mlxfwreset -d <mst device> reset
1046 3. Configure the actual number of VFs::
1048 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1050 4. Unbind the device (can be rebind after the switchdev mode)::
1052 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1054 5. Enbale switchdev mode::
1056 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1061 1. Configure aggressive CQE Zipping for maximum performance::
1063 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1065 To set it back to the default CQE Zipping mode use::
1067 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1069 2. In case of virtualization:
1071 - Make sure that hypervisor kernel is 3.16 or newer.
1072 - Configure boot with ``iommu=pt``.
1073 - Use 1G huge pages.
1074 - Make sure to allocate a VM on huge pages.
1075 - Make sure to set CPU pinning.
1077 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1078 for better performance. For VMs, verify that the right CPU
1079 and NUMA node are pinned according to the above. Run::
1083 to identify the NUMA node to which the PCIe adapter is connected.
1085 4. If more than one adapter is used, and root complex capabilities allow
1086 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1087 it is recommended to locate both adapters on the same NUMA node.
1088 This in order to forward packets from one to the other without
1089 NUMA performance penalty.
1091 5. Disable pause frames::
1093 ethtool -A <netdev> rx off tx off
1095 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1096 via the BIOS configuration. Please contact you server provider for more
1097 information about the settings.
1101 On some machines, depends on the machine integrator, it is beneficial
1102 to set the PCI max read request parameter to 1K. This can be
1103 done in the following way:
1105 To query the read request size use::
1107 setpci -s <NIC PCI address> 68.w
1109 If the output is different than 3XXX, set it by::
1111 setpci -s <NIC PCI address> 68.w=3XXX
1113 The XXX can be different on different systems. Make sure to configure
1114 according to the setpci output.
1116 7. To minimize overhead of searching Memory Regions:
1118 - '--socket-mem' is recommended to pin memory by predictable amount.
1119 - Configure per-lcore cache when creating Mempools for packet buffer.
1120 - Refrain from dynamically allocating/freeing memory in run-time.
1122 .. _mlx5_offloads_support:
1124 Supported hardware offloads
1125 ---------------------------
1127 .. table:: Minimal SW/HW versions for queue offloads
1129 ============== ===== ===== ========= ===== ========== ==========
1130 Offload DPDK Linux rdma-core OFED firmware hardware
1131 ============== ===== ===== ========= ===== ========== ==========
1132 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1133 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1134 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1135 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1136 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1137 ============== ===== ===== ========= ===== ========== ==========
1139 .. table:: Minimal SW/HW versions for rte_flow offloads
1141 +-----------------------+-----------------+-----------------+
1142 | Offload | with E-Switch | with NIC |
1143 +=======================+=================+=================+
1144 | Count | | DPDK 19.05 | | DPDK 19.02 |
1145 | | | OFED 4.6 | | OFED 4.6 |
1146 | | | rdma-core 24 | | rdma-core 23 |
1147 | | | ConnectX-5 | | ConnectX-5 |
1148 +-----------------------+-----------------+-----------------+
1149 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1150 | | | OFED 4.6 | | OFED 4.5 |
1151 | | | rdma-core 24 | | rdma-core 23 |
1152 | | | ConnectX-5 | | ConnectX-4 |
1153 +-----------------------+-----------------+-----------------+
1154 | Queue / RSS | | | | DPDK 18.11 |
1155 | | | N/A | | OFED 4.5 |
1156 | | | | | rdma-core 23 |
1157 | | | | | ConnectX-4 |
1158 +-----------------------+-----------------+-----------------+
1159 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1160 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1161 | | | rdma-core 24 | | rdma-core 23 |
1162 | | | ConnectX-5 | | ConnectX-5 |
1163 +-----------------------+-----------------+-----------------+
1164 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1165 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1166 | | | rdma-core 27 | | rdma-core 27 |
1167 | | | ConnectX-5 | | ConnectX-5 |
1168 +-----------------------+-----------------+-----------------+
1169 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1170 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1171 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1172 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1173 | | set_ipv6_dst / | | | | |
1174 | | set_tp_src / | | | | |
1175 | | set_tp_dst / | | | | |
1176 | | dec_ttl / | | | | |
1177 | | set_ttl / | | | | |
1178 | | set_mac_src / | | | | |
1179 | | set_mac_dst) | | | | |
1180 +-----------------------+-----------------+-----------------+
1181 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1182 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1183 | | | | rdma-core 24 | | rdma-core 24 |
1184 | | | | ConnectX-5 | | ConnectX-5 |
1185 +-----------------------+-----------------+-----------------+
1186 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1187 | | | OFED 4.7-1 | | OFED 4.7-1 |
1188 | | | rdma-core 24 | | N/A |
1189 | | | ConnectX-5 | | ConnectX-5 |
1190 +-----------------------+-----------------+-----------------+
1191 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1192 | | | OFED 4.6 | | OFED 4.5 |
1193 | | | rdma-core 24 | | rdma-core 23 |
1194 | | | ConnectX-5 | | ConnectX-4 |
1195 +-----------------------+-----------------+-----------------+
1196 | Port ID | | DPDK 19.05 | | N/A |
1197 | | | OFED 4.7-1 | | N/A |
1198 | | | rdma-core 24 | | N/A |
1199 | | | ConnectX-5 | | N/A |
1200 +-----------------------+-----------------+-----------------+
1201 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1202 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1203 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1204 | | of_set_vlan_pcp / | | | | |
1205 | | of_set_vlan_vid) | | | | |
1206 +-----------------------+-----------------+-----------------+
1207 | Hairpin | | | | DPDK 19.11 |
1208 | | | N/A | | OFED 4.7-3 |
1209 | | | | | rdma-core 26 |
1210 | | | | | ConnectX-5 |
1211 +-----------------------+-----------------+-----------------+
1212 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1213 | | | OFED 4.7-3 | | OFED 4.7-3 |
1214 | | | rdma-core 26 | | rdma-core 26 |
1215 | | | ConnectX-5 | | ConnectX-5 |
1216 +-----------------------+-----------------+-----------------+
1217 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1218 | | | OFED 4.7-3 | | OFED 4.7-3 |
1219 | | | rdma-core 26 | | rdma-core 26 |
1220 | | | ConnectX-5 | | ConnectX-5 |
1221 +-----------------------+-----------------+-----------------+
1226 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1227 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1229 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1230 command-line parameter to enable additional protocols (UDP and TCP as well
1231 as IP), the following commands must be entered from its CLI to get the same
1232 behavior as librte_pmd_mlx4::
1235 > port config all rss all
1241 This section demonstrates how to launch **testpmd** with Mellanox
1242 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1244 #. Load the kernel modules::
1246 modprobe -a ib_uverbs mlx5_core mlx5_ib
1248 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1251 /etc/init.d/openibd restart
1255 User space I/O kernel modules (uio and igb_uio) are not used and do
1256 not have to be loaded.
1258 #. Make sure Ethernet interfaces are in working order and linked to kernel
1259 verbs. Related sysfs entries should be present::
1261 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1270 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1273 for intf in eth2 eth3 eth4 eth5;
1275 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1278 sed -n 's,.*/\(.*\),-w \1,p'
1287 #. Request huge pages::
1289 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1291 #. Start testpmd with basic parameters::
1293 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
1298 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1299 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1300 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1301 PMD: librte_pmd_mlx5: 1 port(s) detected
1302 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1303 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1304 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1305 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1306 PMD: librte_pmd_mlx5: 1 port(s) detected
1307 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1308 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1309 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1310 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1311 PMD: librte_pmd_mlx5: 1 port(s) detected
1312 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1313 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1314 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1315 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1316 PMD: librte_pmd_mlx5: 1 port(s) detected
1317 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1318 Interactive-mode selected
1319 Configuring Port 0 (socket 0)
1320 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1321 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1322 Port 0: E4:1D:2D:E7:0C:FE
1323 Configuring Port 1 (socket 0)
1324 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1325 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1326 Port 1: E4:1D:2D:E7:0C:FF
1327 Configuring Port 2 (socket 0)
1328 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1329 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1330 Port 2: E4:1D:2D:E7:0C:FA
1331 Configuring Port 3 (socket 0)
1332 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1333 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1334 Port 3: E4:1D:2D:E7:0C:FB
1335 Checking link statuses...
1336 Port 0 Link Up - speed 40000 Mbps - full-duplex
1337 Port 1 Link Up - speed 40000 Mbps - full-duplex
1338 Port 2 Link Up - speed 10000 Mbps - full-duplex
1339 Port 3 Link Up - speed 10000 Mbps - full-duplex
1346 This section demonstrates how to dump flows. Currently, it's possible to dump
1347 all flows with assistance of external tools.
1349 #. 2 ways to get flow raw file:
1351 - Using testpmd CLI:
1353 .. code-block:: console
1355 testpmd> flow dump <port> <output_file>
1357 - call rte_flow_dev_dump api:
1359 .. code-block:: console
1361 rte_flow_dev_dump(port, file, NULL);
1363 #. Dump human-readable flows from raw file:
1365 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1367 .. code-block:: console
1369 mlx_steering_dump.py -f <output_file>