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 - Flow pattern without any specific vlan will match for vlan packets as well:
115 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
116 Meaning, the flow rule::
118 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
120 Will only match vlan packets with vid=3. and the flow rules::
122 flow create 0 ingress pattern eth / ipv4 / end ...
126 flow create 0 ingress pattern eth / vlan / ipv4 / end ...
128 Will match any ipv4 packet (VLAN included).
130 - VLAN pop offload command:
132 - Flow rules having a VLAN pop offload command as one of their actions and
133 are lacking a match on VLAN as one of their items are not supported.
134 - The command is not supported on egress traffic.
136 - VLAN push offload is not supported on ingress traffic.
138 - VLAN set PCP offload is not supported on existing headers.
140 - A multi segment packet must have not more segments than reported by dev_infos_get()
141 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
142 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
143 inline settings) to 58.
145 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
146 to 0 are not supported.
148 - VXLAN TSO and checksum offloads are not supported on VM.
150 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
152 - Match on Geneve header supports the following fields only:
158 Currently, the only supported options length value is 0.
160 - VF: flow rules created on VF devices can only match traffic targeted at the
161 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
163 - Match on GTP tunnel header item supports the following fields only:
170 MAC addresses not already present in the bridge table of the associated
171 kernel network device will be added and cleaned up by the PMD when closing
172 the device. In case of ungraceful program termination, some entries may
173 remain present and should be removed manually by other means.
175 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
176 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
177 ol_flags. As the mempool for the external buffer is managed by PMD, all the
178 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
179 the external buffers will be freed by PMD and the application which still
180 holds the external buffers may be corrupted.
182 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
183 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
184 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
186 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
187 and allmulticast mode are both set to off.
188 To receive IPv6 Multicast messages on VM, explicitly set the relevant
189 MAC address using rte_eth_dev_mac_addr_add() API.
191 - The amount of descriptors in Tx queue may be limited by data inline settings.
192 Inline data require the more descriptor building blocks and overall block
193 amount may exceed the hardware supported limits. The application should
194 reduce the requested Tx size or adjust data inline settings with
195 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
197 - E-Switch decapsulation Flow:
199 - can be applied to PF port only.
200 - must specify VF port action (packet redirection from PF to VF).
201 - optionally may specify tunnel inner source and destination MAC addresses.
203 - E-Switch encapsulation Flow:
205 - can be applied to VF ports only.
206 - must specify PF port action (packet redirection from VF to PF).
208 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
209 mutually exclusive features which cannot be supported together
210 (see :ref:`mlx5_firmware_config`).
214 - Requires DevX and DV flow to be enabled.
215 - KEEP_CRC offload cannot be supported with LRO.
216 - The first mbuf length, without head-room, must be big enough to include the
218 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
219 it with size limited to max LRO size, not to max RX packet length.
224 MLX5 supports various methods to report statistics:
226 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.
228 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.
230 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.
238 These options can be modified in the ``.config`` file.
240 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
242 Toggle compilation of librte_pmd_mlx5 itself.
244 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
246 Build PMD with additional code to make it loadable without hard
247 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
248 on the target system.
250 In this mode, their presence is still required for it to run properly,
251 however their absence won't prevent a DPDK application from starting (with
252 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
253 missing with ``ldd(1)``.
255 It works by moving these dependencies to a purpose-built rdma-core "glue"
256 plug-in which must either be installed in a directory whose name is based
257 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
258 standard location for the dynamic linker (e.g. ``/lib``) if left to the
259 default empty string (``""``).
261 This option has no performance impact.
263 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
265 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
266 in the PMD shared library or the executable static binary.
268 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
270 Toggle debugging code and stricter compilation flags. Enabling this option
271 adds additional run-time checks and debugging messages at the cost of
276 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
277 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
278 64. Default armv8a configuration of make build and meson build set it to 128
279 then brings performance degradation.
281 Environment variables
282 ~~~~~~~~~~~~~~~~~~~~~
286 A list of directories in which to search for the rdma-core "glue" plug-in,
287 separated by colons or semi-colons.
289 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
290 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
291 since ``LD_LIBRARY_PATH`` has no effect in this case.
293 - ``MLX5_SHUT_UP_BF``
295 Configures HW Tx doorbell register as IO-mapped.
297 By default, the HW Tx doorbell is configured as a write-combining register.
298 The register would be flushed to HW usually when the write-combining buffer
299 becomes full, but it depends on CPU design.
301 Except for vectorized Tx burst routines, a write memory barrier is enforced
302 after updating the register so that the update can be immediately visible to
305 When vectorized Tx burst is called, the barrier is set only if the burst size
306 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
307 variable will bring better latency even though the maximum throughput can
310 Run-time configuration
311 ~~~~~~~~~~~~~~~~~~~~~~
313 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
314 because it is affected by their state. Forcing them down prevents packets
317 - **ethtool** operations on related kernel interfaces also affect the PMD.
319 - ``rxq_cqe_comp_en`` parameter [int]
321 A nonzero value enables the compression of CQE on RX side. This feature
322 allows to save PCI bandwidth and improve performance. Enabled by default.
326 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
328 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
331 - ``rxq_cqe_pad_en`` parameter [int]
333 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
334 is aligned with the size of a cacheline of the core. If cacheline size is
335 128B, the CQE size is configured to be 128B even though the device writes
336 only 64B data on the cacheline. This is to avoid unnecessary cache
337 invalidation by device's two consecutive writes on to one cacheline.
338 However in some architecture, it is more beneficial to update entire
339 cacheline with padding the rest 64B rather than striding because
340 read-modify-write could drop performance a lot. On the other hand,
341 writing extra data will consume more PCIe bandwidth and could also drop
342 the maximum throughput. It is recommended to empirically set this
343 parameter. Disabled by default.
347 - CPU having 128B cacheline with ConnectX-5 and BlueField.
349 - ``rxq_pkt_pad_en`` parameter [int]
351 A nonzero value enables padding Rx packet to the size of cacheline on PCI
352 transaction. This feature would waste PCI bandwidth but could improve
353 performance by avoiding partial cacheline write which may cause costly
354 read-modify-copy in memory transaction on some architectures. Disabled by
359 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
361 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
364 - ``mprq_en`` parameter [int]
366 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
367 configured as Multi-Packet RQ if the total number of Rx queues is
368 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
371 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
372 by posting a single large buffer for multiple packets. Instead of posting a
373 buffers per a packet, one large buffer is posted in order to receive multiple
374 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
375 and each stride receives one packet. MPRQ can improve throughput for
376 small-packet traffic.
378 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
379 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
380 configure large stride size enough to accommodate max_rx_pkt_len as long as
381 device allows. Note that this can waste system memory compared to enabling Rx
382 scatter and multi-segment packet.
384 - ``mprq_log_stride_num`` parameter [int]
386 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
387 strides can reduce PCIe traffic further. If configured value is not in the
388 range of device capability, the default value will be set with a warning
389 message. The default value is 4 which is 16 strides per a buffer, valid only
390 if ``mprq_en`` is set.
392 The size of Rx queue should be bigger than the number of strides.
394 - ``mprq_max_memcpy_len`` parameter [int]
396 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
397 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
398 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
399 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
400 A mempool for external buffers will be allocated and managed by PMD. If Rx
401 packet is externally attached, ol_flags field of the mbuf will have
402 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
403 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
405 - ``rxqs_min_mprq`` parameter [int]
407 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
408 greater or equal to this value. The default value is 12, valid only if
411 - ``txq_inline`` parameter [int]
413 Amount of data to be inlined during TX operations. This parameter is
414 deprecated and converted to the new parameter ``txq_inline_max`` providing
415 partial compatibility.
417 - ``txqs_min_inline`` parameter [int]
419 Enable inline data send only when the number of TX queues is greater or equal
422 This option should be used in combination with ``txq_inline_max`` and
423 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
425 If this option is not specified the default value 16 is used for BlueField
426 and 8 for other platforms
428 The data inlining consumes the CPU cycles, so this option is intended to
429 auto enable inline data if we have enough Tx queues, which means we have
430 enough CPU cores and PCI bandwidth is getting more critical and CPU
431 is not supposed to be bottleneck anymore.
433 The copying data into WQE improves latency and can improve PPS performance
434 when PCI back pressure is detected and may be useful for scenarios involving
435 heavy traffic on many queues.
437 Because additional software logic is necessary to handle this mode, this
438 option should be used with care, as it may lower performance when back
439 pressure is not expected.
441 If inline data are enabled it may affect the maximal size of Tx queue in
442 descriptors because the inline data increase the descriptor size and
443 queue size limits supported by hardware may be exceeded.
445 - ``txq_inline_min`` parameter [int]
447 Minimal amount of data to be inlined into WQE during Tx operations. NICs
448 may require this minimal data amount to operate correctly. The exact value
449 may depend on NIC operation mode, requested offloads, etc. It is strongly
450 recommended to omit this parameter and use the default values. Anyway,
451 applications using this parameter should take into consideration that
452 specifying an inconsistent value may prevent the NIC from sending packets.
454 If ``txq_inline_min`` key is present the specified value (may be aligned
455 by the driver in order not to exceed the limits and provide better descriptor
456 space utilization) will be used by the driver and it is guaranteed that
457 requested amount of data bytes are inlined into the WQE beside other inline
458 settings. This key also may update ``txq_inline_max`` value (default
459 or specified explicitly in devargs) to reserve the space for inline data.
461 If ``txq_inline_min`` key is not present, the value may be queried by the
462 driver from the NIC via DevX if this feature is available. If there is no DevX
463 enabled/supported the value 18 (supposing L2 header including VLAN) is set
464 for ConnectX-4 and ConnectX-4LX, and 0 is set by default for ConnectX-5
465 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
468 For ConnectX-4 NIC, driver does not allow specifying value below 18
469 (minimal L2 header, including VLAN), error will be raised.
471 For ConnectX-4LX NIC, it is allowed to specify values below 18, but
472 it is not recommended and may prevent NIC from sending packets over
475 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
476 Multi-Packet Write), because last one does not support partial packet inlining.
477 This is not very critical due to minimal data inlining is mostly required
478 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
480 - ``txq_inline_max`` parameter [int]
482 Specifies the maximal packet length to be completely inlined into WQE
483 Ethernet Segment for ordinary SEND method. If packet is larger than specified
484 value, the packet data won't be copied by the driver at all, data buffer
485 is addressed with a pointer. If packet length is less or equal all packet
486 data will be copied into WQE. This may improve PCI bandwidth utilization for
487 short packets significantly but requires the extra CPU cycles.
489 The data inline feature is controlled by number of Tx queues, if number of Tx
490 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
491 is engaged, if there are not enough Tx queues (which means not enough CPU cores
492 and CPU resources are scarce), data inline is not performed by the driver.
493 Assigning ``txqs_min_inline`` with zero always enables the data inline.
495 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
496 by the driver in order not to exceed the limit (930 bytes) and to provide better
497 WQE space filling without gaps, the adjustment is reflected in the debug log.
498 Also, the default value (290) may be decreased in run-time if the large transmit
499 queue size is requested and hardware does not support enough descriptor
500 amount, in this case warning is emitted. If ``txq_inline_max`` key is
501 specified and requested inline settings can not be satisfied then error
504 - ``txq_inline_mpw`` parameter [int]
506 Specifies the maximal packet length to be completely inlined into WQE for
507 Enhanced MPW method. If packet is large the specified value, the packet data
508 won't be copied, and data buffer is addressed with pointer. If packet length
509 is less or equal, all packet data will be copied into WQE. This may improve PCI
510 bandwidth utilization for short packets significantly but requires the extra
513 The data inline feature is controlled by number of TX queues, if number of Tx
514 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
515 is engaged, if there are not enough Tx queues (which means not enough CPU cores
516 and CPU resources are scarce), data inline is not performed by the driver.
517 Assigning ``txqs_min_inline`` with zero always enables the data inline.
519 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
520 by the driver in order not to exceed the limit (930 bytes) and to provide better
521 WQE space filling without gaps, the adjustment is reflected in the debug log.
522 Due to multiple packets may be included to the same WQE with Enhanced Multi
523 Packet Write Method and overall WQE size is limited it is not recommended to
524 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
525 may be decreased in run-time if the large transmit queue size is requested
526 and hardware does not support enough descriptor amount, in this case warning
527 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
528 settings can not be satisfied then error will be raised.
530 - ``txqs_max_vec`` parameter [int]
532 Enable vectorized Tx only when the number of TX queues is less than or
533 equal to this value. This parameter is deprecated and ignored, kept
534 for compatibility issue to not prevent driver from probing.
536 - ``txq_mpw_hdr_dseg_en`` parameter [int]
538 A nonzero value enables including two pointers in the first block of TX
539 descriptor. The parameter is deprecated and ignored, kept for compatibility
542 - ``txq_max_inline_len`` parameter [int]
544 Maximum size of packet to be inlined. This limits the size of packet to
545 be inlined. If the size of a packet is larger than configured value, the
546 packet isn't inlined even though there's enough space remained in the
547 descriptor. Instead, the packet is included with pointer. This parameter
548 is deprecated and converted directly to ``txq_inline_mpw`` providing full
549 compatibility. Valid only if eMPW feature is engaged.
551 - ``txq_mpw_en`` parameter [int]
553 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
554 ConnectX-6, ConnectX-6 DX and BlueField. eMPW allows the TX burst function to pack
555 up multiple packets in a single descriptor session in order to save PCI bandwidth
556 and improve performance at the cost of a slightly higher CPU usage. When
557 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
558 entire packet data on to TX descriptor instead of including pointer of packet.
560 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
561 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
562 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
563 option or reported by the NIC, the eMPW feature is disengaged.
565 - ``tx_db_nc`` parameter [int]
567 The rdma core library can map doorbell register in two ways, depending on the
568 environment variable "MLX5_SHUT_UP_BF":
570 - As regular cached memory, if the variable is either missing or set to zero.
571 - As non-cached memory, if the variable is present and set to not "0" value.
573 The type of mapping may slightly affect the Tx performance, the optimal choice
574 is strongly relied on the host architecture and should be deduced practically.
576 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
577 memory, the PMD will perform the extra write memory barrier after writing to
578 doorbell, it might increase the needed CPU clocks per packet to send, but
579 latency might be improved.
581 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
582 cached memory, the PMD will not perform the extra write memory barrier
583 after writing to doorbell, on some architectures it might improve the
586 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
587 memory, the PMD will use heuristics to decide whether write memory barrier
588 should be performed. For bursts with size multiple of recommended one (64 pkts)
589 it is supposed the next burst is coming and no need to issue the extra memory
590 barrier (it is supposed to be issued in the next coming burst, at least after
591 descriptor writing). It might increase latency (on some hosts till next
592 packets transmit) and should be used with care.
594 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
595 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
596 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
598 - ``tx_vec_en`` parameter [int]
600 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 DX
601 and BlueField NICs if the number of global Tx queues on the port is less than
602 ``txqs_max_vec``. The parameter is deprecated and ignored.
604 - ``rx_vec_en`` parameter [int]
606 A nonzero value enables Rx vector if the port is not configured in
607 multi-segment otherwise this parameter is ignored.
611 - ``vf_nl_en`` parameter [int]
613 A nonzero value enables Netlink requests from the VF to add/remove MAC
614 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
615 Otherwise the relevant configuration must be run with Linux iproute2 tools.
616 This is a prerequisite to receive this kind of traffic.
618 Enabled by default, valid only on VF devices ignored otherwise.
620 - ``l3_vxlan_en`` parameter [int]
622 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
623 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
624 parameter. This is a prerequisite to receive this kind of traffic.
628 - ``dv_xmeta_en`` parameter [int]
630 A nonzero value enables extensive flow metadata support if device is
631 capable and driver supports it. This can enable extensive support of
632 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
633 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
635 There are some possible configurations, depending on parameter value:
637 - 0, this is default value, defines the legacy mode, the ``MARK`` and
638 ``META`` related actions and items operate only within NIC Tx and
639 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
640 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
641 item is 32 bits wide and match supported on egress only.
643 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
644 related actions and items operate within all supported steering domains,
645 including FDB, ``MARK`` and ``META`` information may cross the domain
646 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
647 depends on kernel and firmware configurations and might be 0, 16 or
648 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
649 compatibility, the actual width of data transferred to the FDB domain
650 depends on kernel configuration and may be vary. The actual supported
651 width can be retrieved in runtime by series of rte_flow_validate()
654 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
655 related actions and items operate within all supported steering domains,
656 including FDB, ``MARK`` and ``META`` information may cross the domain
657 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
658 depends on kernel and firmware configurations and might be 0, 16 or
659 24 bits. The actual supported width can be retrieved in runtime by
660 series of rte_flow_validate() trials.
662 +------+-----------+-----------+-------------+-------------+
663 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
664 +======+===========+===========+=============+=============+
665 | 0 | 24 bits | 32 bits | 32 bits | no |
666 +------+-----------+-----------+-------------+-------------+
667 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
668 +------+-----------+-----------+-------------+-------------+
669 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
670 +------+-----------+-----------+-------------+-------------+
672 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
673 ignored and the device is configured to operate in legacy mode (0).
675 Disabled by default (set to 0).
677 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
678 of the extensive metadata features. The legacy Verbs supports FLAG and
679 MARK metadata actions over NIC Rx steering domain only.
681 - ``dv_flow_en`` parameter [int]
683 A nonzero value enables the DV flow steering assuming it is supported
684 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
686 Enabled by default if supported.
688 - ``dv_esw_en`` parameter [int]
690 A nonzero value enables E-Switch using Direct Rules.
692 Enabled by default if supported.
694 - ``mr_ext_memseg_en`` parameter [int]
696 A nonzero value enables extending memseg when registering DMA memory. If
697 enabled, the number of entries in MR (Memory Region) lookup table on datapath
698 is minimized and it benefits performance. On the other hand, it worsens memory
699 utilization because registered memory is pinned by kernel driver. Even if a
700 page in the extended chunk is freed, that doesn't become reusable until the
701 entire memory is freed.
705 - ``representor`` parameter [list]
707 This parameter can be used to instantiate DPDK Ethernet devices from
708 existing port (or VF) representors configured on the device.
710 It is a standard parameter whose format is described in
711 :ref:`ethernet_device_standard_device_arguments`.
713 For instance, to probe port representors 0 through 2::
717 - ``max_dump_files_num`` parameter [int]
719 The maximum number of files per PMD entity that may be created for debug information.
720 The files will be created in /var/log directory or in current directory.
722 set to 128 by default.
724 - ``lro_timeout_usec`` parameter [int]
726 The maximum allowed duration of an LRO session, in micro-seconds.
727 PMD will set the nearest value supported by HW, which is not bigger than
728 the input ``lro_timeout_usec`` value.
729 If this parameter is not specified, by default PMD will set
730 the smallest value supported by HW.
732 .. _mlx5_firmware_config:
734 Firmware configuration
735 ~~~~~~~~~~~~~~~~~~~~~~
737 Firmware features can be configured as key/value pairs.
739 The command to set a value is::
741 mlxconfig -d <device> set <key>=<value>
743 The command to query a value is::
745 mlxconfig -d <device> query | grep <key>
747 The device name for the command ``mlxconfig`` can be either the PCI address,
748 or the mst device name found with::
752 Below are some firmware configurations listed.
758 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
764 - maximum number of SR-IOV virtual functions::
768 - enable DevX (required by Direct Rules and other features)::
772 - aggressive CQE zipping::
776 - L3 VXLAN and VXLAN-GPE destination UDP port::
779 IP_OVER_VXLAN_PORT=<udp dport>
781 - enable IP-in-IP tunnel flow matching::
783 FLEX_PARSER_PROFILE_ENABLE=0
785 - enable MPLS flow matching::
787 FLEX_PARSER_PROFILE_ENABLE=1
789 - enable ICMP/ICMP6 code/type fields matching::
791 FLEX_PARSER_PROFILE_ENABLE=2
793 - enable Geneve flow matching::
795 FLEX_PARSER_PROFILE_ENABLE=0
797 - enable GTP flow matching::
799 FLEX_PARSER_PROFILE_ENABLE=3
804 This driver relies on external libraries and kernel drivers for resources
805 allocations and initialization. The following dependencies are not part of
806 DPDK and must be installed separately:
810 User space Verbs framework used by librte_pmd_mlx5. This library provides
811 a generic interface between the kernel and low-level user space drivers
814 It allows slow and privileged operations (context initialization, hardware
815 resources allocations) to be managed by the kernel and fast operations to
816 never leave user space.
820 Low-level user space driver library for Mellanox
821 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
824 This library basically implements send/receive calls to the hardware
829 They provide the kernel-side Verbs API and low level device drivers that
830 manage actual hardware initialization and resources sharing with user
833 Unlike most other PMDs, these modules must remain loaded and bound to
836 - mlx5_core: hardware driver managing Mellanox
837 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
839 - mlx5_ib: InifiniBand device driver.
840 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
842 - **Firmware update**
844 Mellanox OFED/EN releases include firmware updates for
845 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
847 Because each release provides new features, these updates must be applied to
848 match the kernel modules and libraries they come with.
852 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
858 Either RDMA Core library with a recent enough Linux kernel release
859 (recommended) or Mellanox OFED/EN, which provides compatibility with older
862 RDMA Core with Linux Kernel
863 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
865 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
866 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
867 (see `RDMA Core installation documentation`_)
868 - When building for i686 use:
870 - rdma-core version 18.0 or above built with 32bit support.
871 - Kernel version 4.14.41 or above.
873 - Starting with rdma-core v21, static libraries can be built::
876 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
879 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
880 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
882 If rdma-core libraries are built but not installed, DPDK makefile can link them,
883 thanks to these environment variables:
885 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
886 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
887 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
892 - Mellanox OFED version: ** 4.5, 4.6** /
893 Mellanox EN version: **4.5, 4.6**
896 - ConnectX-4: **12.21.1000** and above.
897 - ConnectX-4 Lx: **14.21.1000** and above.
898 - ConnectX-5: **16.21.1000** and above.
899 - ConnectX-5 Ex: **16.21.1000** and above.
900 - ConnectX-6: **20.99.5374** and above.
901 - ConnectX-6 DX: **22.27.0090** and above.
902 - BlueField: **18.25.1010** and above.
904 While these libraries and kernel modules are available on OpenFabrics
905 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
906 managers on most distributions, this PMD requires Ethernet extensions that
907 may not be supported at the moment (this is a work in progress).
910 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
912 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
913 include the necessary support and should be used in the meantime. For DPDK,
914 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
915 required from that distribution.
919 Several versions of Mellanox OFED/EN are available. Installing the version
920 this DPDK release was developed and tested against is strongly
921 recommended. Please check the `prerequisites`_.
926 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
927 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
928 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
929 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
930 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
931 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
932 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
933 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
934 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
935 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
936 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
937 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
938 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
939 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
940 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
941 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
942 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
943 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
944 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
945 * Mellanox(R) ConnectX(R)-6 200G MCX654106A-HCAT (4x200G)
946 * Mellanox(R) ConnectX(R)-6DX EN 100G MCX623106AN-CDAT (2*100g)
947 * Mellanox(R) ConnectX(R)-6DX EN 200G MCX623105AN-VDAT (1*200g)
949 Quick Start Guide on OFED/EN
950 ----------------------------
952 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
955 2. Install the required libraries and kernel modules either by installing
956 only the required set, or by installing the entire Mellanox OFED/EN::
958 ./mlnxofedinstall --upstream-libs --dpdk
960 3. Verify the firmware is the correct one::
964 4. Verify all ports links are set to Ethernet::
966 mlxconfig -d <mst device> query | grep LINK_TYPE
970 Link types may have to be configured to Ethernet::
972 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
974 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
976 For hypervisors, verify SR-IOV is enabled on the NIC::
978 mlxconfig -d <mst device> query | grep SRIOV_EN
981 If needed, configure SR-IOV::
983 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
984 mlxfwreset -d <mst device> reset
986 5. Restart the driver::
988 /etc/init.d/openibd restart
992 service openibd restart
994 If link type was changed, firmware must be reset as well::
996 mlxfwreset -d <mst device> reset
998 For hypervisors, after reset write the sysfs number of virtual functions
1001 To dynamically instantiate a given number of virtual functions (VFs)::
1003 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1005 6. Compile DPDK and you are ready to go. See instructions on
1006 :ref:`Development Kit Build System <Development_Kit_Build_System>`
1008 Enable switchdev mode
1009 ---------------------
1011 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1012 Representor is a port in DPDK that is connected to a VF in such a way
1013 that assuming there are no offload flows, each packet that is sent from the VF
1014 will be received by the corresponding representor. While each packet that is
1015 sent to a representor will be received by the VF.
1016 This is very useful in case of SRIOV mode, where the first packet that is sent
1017 by the VF will be received by the DPDK application which will decide if this
1018 flow should be offloaded to the E-Switch. After offloading the flow packet
1019 that the VF that are matching the flow will not be received any more by
1020 the DPDK application.
1022 1. Enable SRIOV mode::
1024 mlxconfig -d <mst device> set SRIOV_EN=true
1026 2. Configure the max number of VFs::
1028 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1032 mlxfwreset -d <mst device> reset
1034 3. Configure the actual number of VFs::
1036 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1038 4. Unbind the device (can be rebind after the switchdev mode)::
1040 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1042 5. Enbale switchdev mode::
1044 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1049 1. Configure aggressive CQE Zipping for maximum performance::
1051 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1053 To set it back to the default CQE Zipping mode use::
1055 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1057 2. In case of virtualization:
1059 - Make sure that hypervisor kernel is 3.16 or newer.
1060 - Configure boot with ``iommu=pt``.
1061 - Use 1G huge pages.
1062 - Make sure to allocate a VM on huge pages.
1063 - Make sure to set CPU pinning.
1065 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1066 for better performance. For VMs, verify that the right CPU
1067 and NUMA node are pinned according to the above. Run::
1071 to identify the NUMA node to which the PCIe adapter is connected.
1073 4. If more than one adapter is used, and root complex capabilities allow
1074 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1075 it is recommended to locate both adapters on the same NUMA node.
1076 This in order to forward packets from one to the other without
1077 NUMA performance penalty.
1079 5. Disable pause frames::
1081 ethtool -A <netdev> rx off tx off
1083 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1084 via the BIOS configuration. Please contact you server provider for more
1085 information about the settings.
1089 On some machines, depends on the machine integrator, it is beneficial
1090 to set the PCI max read request parameter to 1K. This can be
1091 done in the following way:
1093 To query the read request size use::
1095 setpci -s <NIC PCI address> 68.w
1097 If the output is different than 3XXX, set it by::
1099 setpci -s <NIC PCI address> 68.w=3XXX
1101 The XXX can be different on different systems. Make sure to configure
1102 according to the setpci output.
1104 7. To minimize overhead of searching Memory Regions:
1106 - '--socket-mem' is recommended to pin memory by predictable amount.
1107 - Configure per-lcore cache when creating Mempools for packet buffer.
1108 - Refrain from dynamically allocating/freeing memory in run-time.
1110 .. _mlx5_offloads_support:
1112 Supported hardware offloads
1113 ---------------------------
1115 .. table:: Minimal SW/HW versions for queue offloads
1117 ============== ===== ===== ========= ===== ========== ==========
1118 Offload DPDK Linux rdma-core OFED firmware hardware
1119 ============== ===== ===== ========= ===== ========== ==========
1120 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1121 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1122 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1123 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1124 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1125 ============== ===== ===== ========= ===== ========== ==========
1127 .. table:: Minimal SW/HW versions for rte_flow offloads
1129 +-----------------------+-----------------+-----------------+
1130 | Offload | with E-Switch | with NIC |
1131 +=======================+=================+=================+
1132 | Count | | DPDK 19.05 | | DPDK 19.02 |
1133 | | | OFED 4.6 | | OFED 4.6 |
1134 | | | rdma-core 24 | | rdma-core 23 |
1135 | | | ConnectX-5 | | ConnectX-5 |
1136 +-----------------------+-----------------+-----------------+
1137 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1138 | | | OFED 4.6 | | OFED 4.5 |
1139 | | | rdma-core 24 | | rdma-core 23 |
1140 | | | ConnectX-5 | | ConnectX-4 |
1141 +-----------------------+-----------------+-----------------+
1142 | Queue / RSS | | | | DPDK 18.11 |
1143 | | | N/A | | OFED 4.5 |
1144 | | | | | rdma-core 23 |
1145 | | | | | ConnectX-4 |
1146 +-----------------------+-----------------+-----------------+
1147 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1148 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1149 | | | rdma-core 24 | | rdma-core 23 |
1150 | | | ConnectX-5 | | ConnectX-5 |
1151 +-----------------------+-----------------+-----------------+
1152 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1153 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1154 | | | rdma-core 27 | | rdma-core 27 |
1155 | | | ConnectX-5 | | ConnectX-5 |
1156 +-----------------------+-----------------+-----------------+
1157 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1158 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1159 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1160 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1161 | | set_ipv6_dst / | | | | |
1162 | | set_tp_src / | | | | |
1163 | | set_tp_dst / | | | | |
1164 | | dec_ttl / | | | | |
1165 | | set_ttl / | | | | |
1166 | | set_mac_src / | | | | |
1167 | | set_mac_dst) | | | | |
1168 +-----------------------+-----------------+-----------------+
1169 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1170 | | | OFED 4.7-1 | | OFED 4.7-1 |
1171 | | | rdma-core 24 | | N/A |
1172 | | | ConnectX-5 | | ConnectX-5 |
1173 +-----------------------+-----------------+-----------------+
1174 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1175 | | | OFED 4.6 | | OFED 4.5 |
1176 | | | rdma-core 24 | | rdma-core 23 |
1177 | | | ConnectX-5 | | ConnectX-4 |
1178 +-----------------------+-----------------+-----------------+
1179 | Port ID | | DPDK 19.05 | | N/A |
1180 | | | OFED 4.7-1 | | N/A |
1181 | | | rdma-core 24 | | N/A |
1182 | | | ConnectX-5 | | N/A |
1183 +-----------------------+-----------------+-----------------+
1184 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1185 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1186 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1187 | | of_set_vlan_pcp / | | | | |
1188 | | of_set_vlan_vid) | | | | |
1189 +-----------------------+-----------------+-----------------+
1190 | Hairpin | | | | DPDK 19.11 |
1191 | | | N/A | | OFED 4.7-3 |
1192 | | | | | rdma-core 26 |
1193 | | | | | ConnectX-5 |
1194 +-----------------------+-----------------+-----------------+
1195 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1196 | | | OFED 4.7-3 | | OFED 4.7-3 |
1197 | | | rdma-core 26 | | rdma-core 26 |
1198 | | | ConnectX-5 | | ConnectX-5 |
1199 +-----------------------+-----------------+-----------------+
1200 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1201 | | | OFED 4.7-3 | | OFED 4.7-3 |
1202 | | | rdma-core 26 | | rdma-core 26 |
1203 | | | ConnectX-5 | | ConnectX-5 |
1204 +-----------------------+-----------------+-----------------+
1209 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1210 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1212 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1213 command-line parameter to enable additional protocols (UDP and TCP as well
1214 as IP), the following commands must be entered from its CLI to get the same
1215 behavior as librte_pmd_mlx4::
1218 > port config all rss all
1224 This section demonstrates how to launch **testpmd** with Mellanox
1225 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1227 #. Load the kernel modules::
1229 modprobe -a ib_uverbs mlx5_core mlx5_ib
1231 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1234 /etc/init.d/openibd restart
1238 User space I/O kernel modules (uio and igb_uio) are not used and do
1239 not have to be loaded.
1241 #. Make sure Ethernet interfaces are in working order and linked to kernel
1242 verbs. Related sysfs entries should be present::
1244 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1253 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1256 for intf in eth2 eth3 eth4 eth5;
1258 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1261 sed -n 's,.*/\(.*\),-w \1,p'
1270 #. Request huge pages::
1272 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1274 #. Start testpmd with basic parameters::
1276 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
1281 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1282 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1283 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1284 PMD: librte_pmd_mlx5: 1 port(s) detected
1285 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1286 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1287 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1288 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1289 PMD: librte_pmd_mlx5: 1 port(s) detected
1290 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1291 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1292 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1293 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1294 PMD: librte_pmd_mlx5: 1 port(s) detected
1295 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1296 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1297 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1298 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1299 PMD: librte_pmd_mlx5: 1 port(s) detected
1300 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1301 Interactive-mode selected
1302 Configuring Port 0 (socket 0)
1303 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1304 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1305 Port 0: E4:1D:2D:E7:0C:FE
1306 Configuring Port 1 (socket 0)
1307 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1308 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1309 Port 1: E4:1D:2D:E7:0C:FF
1310 Configuring Port 2 (socket 0)
1311 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1312 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1313 Port 2: E4:1D:2D:E7:0C:FA
1314 Configuring Port 3 (socket 0)
1315 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1316 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1317 Port 3: E4:1D:2D:E7:0C:FB
1318 Checking link statuses...
1319 Port 0 Link Up - speed 40000 Mbps - full-duplex
1320 Port 1 Link Up - speed 40000 Mbps - full-duplex
1321 Port 2 Link Up - speed 10000 Mbps - full-duplex
1322 Port 3 Link Up - speed 10000 Mbps - full-duplex
1329 This section demonstrates how to dump flows. Currently, it's possible to dump
1330 all flows with assistance of external tools.
1332 #. 2 ways to get flow raw file:
1334 - Using testpmd CLI:
1336 .. code-block:: console
1338 testpmd> flow dump <port> <output_file>
1340 - call rte_flow_dev_dump api:
1342 .. code-block:: console
1344 rte_flow_dev_dump(port, file, NULL);
1346 #. Dump human-readable flows from raw file:
1348 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1350 .. code-block:: console
1352 mlx_steering_dump.py -f <output_file>