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 - Several RSS hash keys, one for each flow type.
68 - Default RSS operation with no hash key specification.
69 - Configurable RETA table.
70 - Link flow control (pause frame).
71 - Support for multiple MAC addresses.
75 - RX CRC stripping configuration.
76 - Promiscuous mode on PF and VF.
77 - Multicast promiscuous mode on PF and VF.
78 - Hardware checksum offloads.
79 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
81 - Flow API, including :ref:`flow_isolated_mode`.
83 - KVM and VMware ESX SR-IOV modes are supported.
84 - RSS hash result is supported.
85 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
86 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
88 - Statistics query including Basic, Extended and per queue.
90 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve.
91 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
92 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
93 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
94 - Flow insertion rate of more then million flows per second, when using Direct Rules.
95 - Support for multiple rte_flow groups.
101 - For secondary process:
103 - Forked secondary process not supported.
104 - External memory unregistered in EAL memseg list cannot be used for DMA
105 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
106 primary process and remapped to the same virtual address in secondary
107 process. If the external memory is registered by primary process but has
108 different virtual address in secondary process, unexpected error may happen.
110 - Flow pattern without any specific vlan will match for vlan packets as well:
112 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
113 Meaning, the flow rule::
115 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
117 Will only match vlan packets with vid=3. and the flow rules::
119 flow create 0 ingress pattern eth / ipv4 / end ...
123 flow create 0 ingress pattern eth / vlan / 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()``).
162 MAC addresses not already present in the bridge table of the associated
163 kernel network device will be added and cleaned up by the PMD when closing
164 the device. In case of ungraceful program termination, some entries may
165 remain present and should be removed manually by other means.
167 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
168 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
169 ol_flags. As the mempool for the external buffer is managed by PMD, all the
170 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
171 the external buffers will be freed by PMD and the application which still
172 holds the external buffers may be corrupted.
174 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
175 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
176 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
178 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
179 and allmulticast mode are both set to off.
180 To receive IPv6 Multicast messages on VM, explicitly set the relevant
181 MAC address using rte_eth_dev_mac_addr_add() API.
183 - The amount of descriptors in Tx queue may be limited by data inline settings.
184 Inline data require the more descriptor building blocks and overall block
185 amount may exceed the hardware supported limits. The application should
186 reduce the requested Tx size or adjust data inline settings with
187 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
189 - E-Switch decapsulation Flow:
191 - can be applied to PF port only.
192 - must specify VF port action (packet redirection from PF to VF).
193 - optionally may specify tunnel inner source and destination MAC addresses.
195 - E-Switch encapsulation Flow:
197 - can be applied to VF ports only.
198 - must specify PF port action (packet redirection from VF to PF).
200 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
201 mutually exclusive features which cannot be supported together
202 (see :ref:`mlx5_firmware_config`).
206 - Requires DevX and DV flow to be enabled.
207 - KEEP_CRC offload cannot be supported with LRO.
208 - The first mbuf length, without head-room, must be big enough to include the
214 MLX5 supports various methods to report statistics:
216 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.
218 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.
220 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.
228 These options can be modified in the ``.config`` file.
230 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
232 Toggle compilation of librte_pmd_mlx5 itself.
234 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
236 Build PMD with additional code to make it loadable without hard
237 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
238 on the target system.
240 In this mode, their presence is still required for it to run properly,
241 however their absence won't prevent a DPDK application from starting (with
242 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
243 missing with ``ldd(1)``.
245 It works by moving these dependencies to a purpose-built rdma-core "glue"
246 plug-in which must either be installed in a directory whose name is based
247 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
248 standard location for the dynamic linker (e.g. ``/lib``) if left to the
249 default empty string (``""``).
251 This option has no performance impact.
253 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
255 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
256 in the PMD shared library or the executable static binary.
258 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
260 Toggle debugging code and stricter compilation flags. Enabling this option
261 adds additional run-time checks and debugging messages at the cost of
266 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
267 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
268 64. Default armv8a configuration of make build and meson build set it to 128
269 then brings performance degradation.
271 Environment variables
272 ~~~~~~~~~~~~~~~~~~~~~
276 A list of directories in which to search for the rdma-core "glue" plug-in,
277 separated by colons or semi-colons.
279 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
280 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
281 since ``LD_LIBRARY_PATH`` has no effect in this case.
283 - ``MLX5_SHUT_UP_BF``
285 Configures HW Tx doorbell register as IO-mapped.
287 By default, the HW Tx doorbell is configured as a write-combining register.
288 The register would be flushed to HW usually when the write-combining buffer
289 becomes full, but it depends on CPU design.
291 Except for vectorized Tx burst routines, a write memory barrier is enforced
292 after updating the register so that the update can be immediately visible to
295 When vectorized Tx burst is called, the barrier is set only if the burst size
296 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
297 variable will bring better latency even though the maximum throughput can
300 Run-time configuration
301 ~~~~~~~~~~~~~~~~~~~~~~
303 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
304 because it is affected by their state. Forcing them down prevents packets
307 - **ethtool** operations on related kernel interfaces also affect the PMD.
309 - ``rxq_cqe_comp_en`` parameter [int]
311 A nonzero value enables the compression of CQE on RX side. This feature
312 allows to save PCI bandwidth and improve performance. Enabled by default.
316 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
318 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
321 - ``rxq_cqe_pad_en`` parameter [int]
323 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
324 is aligned with the size of a cacheline of the core. If cacheline size is
325 128B, the CQE size is configured to be 128B even though the device writes
326 only 64B data on the cacheline. This is to avoid unnecessary cache
327 invalidation by device's two consecutive writes on to one cacheline.
328 However in some architecture, it is more beneficial to update entire
329 cacheline with padding the rest 64B rather than striding because
330 read-modify-write could drop performance a lot. On the other hand,
331 writing extra data will consume more PCIe bandwidth and could also drop
332 the maximum throughput. It is recommended to empirically set this
333 parameter. Disabled by default.
337 - CPU having 128B cacheline with ConnectX-5 and BlueField.
339 - ``rxq_pkt_pad_en`` parameter [int]
341 A nonzero value enables padding Rx packet to the size of cacheline on PCI
342 transaction. This feature would waste PCI bandwidth but could improve
343 performance by avoiding partial cacheline write which may cause costly
344 read-modify-copy in memory transaction on some architectures. Disabled by
349 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
351 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6, ConnectX-6 DX
354 - ``mprq_en`` parameter [int]
356 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
357 configured as Multi-Packet RQ if the total number of Rx queues is
358 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
361 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
362 by posting a single large buffer for multiple packets. Instead of posting a
363 buffers per a packet, one large buffer is posted in order to receive multiple
364 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
365 and each stride receives one packet. MPRQ can improve throughput for
366 small-packet traffic.
368 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
369 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
370 configure large stride size enough to accommodate max_rx_pkt_len as long as
371 device allows. Note that this can waste system memory compared to enabling Rx
372 scatter and multi-segment packet.
374 - ``mprq_log_stride_num`` parameter [int]
376 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
377 strides can reduce PCIe traffic further. If configured value is not in the
378 range of device capability, the default value will be set with a warning
379 message. The default value is 4 which is 16 strides per a buffer, valid only
380 if ``mprq_en`` is set.
382 The size of Rx queue should be bigger than the number of strides.
384 - ``mprq_max_memcpy_len`` parameter [int]
386 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
387 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
388 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
389 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
390 A mempool for external buffers will be allocated and managed by PMD. If Rx
391 packet is externally attached, ol_flags field of the mbuf will have
392 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
393 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
395 - ``rxqs_min_mprq`` parameter [int]
397 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
398 greater or equal to this value. The default value is 12, valid only if
401 - ``txq_inline`` parameter [int]
403 Amount of data to be inlined during TX operations. This parameter is
404 deprecated and converted to the new parameter ``txq_inline_max`` providing
405 partial compatibility.
407 - ``txqs_min_inline`` parameter [int]
409 Enable inline data send only when the number of TX queues is greater or equal
412 This option should be used in combination with ``txq_inline_max`` and
413 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
415 If this option is not specified the default value 16 is used for BlueField
416 and 8 for other platforms
418 The data inlining consumes the CPU cycles, so this option is intended to
419 auto enable inline data if we have enough Tx queues, which means we have
420 enough CPU cores and PCI bandwidth is getting more critical and CPU
421 is not supposed to be bottleneck anymore.
423 The copying data into WQE improves latency and can improve PPS performance
424 when PCI back pressure is detected and may be useful for scenarios involving
425 heavy traffic on many queues.
427 Because additional software logic is necessary to handle this mode, this
428 option should be used with care, as it may lower performance when back
429 pressure is not expected.
431 If inline data are enabled it may affect the maximal size of Tx queue in
432 descriptors because the inline data increase the descriptor size and
433 queue size limits supported by hardware may be exceeded.
435 - ``txq_inline_min`` parameter [int]
437 Minimal amount of data to be inlined into WQE during Tx operations. NICs
438 may require this minimal data amount to operate correctly. The exact value
439 may depend on NIC operation mode, requested offloads, etc. It is strongly
440 recommended to omit this parameter and use the default values. Anyway,
441 applications using this parameter should take into consideration that
442 specifying an inconsistent value may prevent the NIC from sending packets.
444 If ``txq_inline_min`` key is present the specified value (may be aligned
445 by the driver in order not to exceed the limits and provide better descriptor
446 space utilization) will be used by the driver and it is guaranteed that
447 requested amount of data bytes are inlined into the WQE beside other inline
448 settings. This key also may update ``txq_inline_max`` value (default
449 or specified explicitly in devargs) to reserve the space for inline data.
451 If ``txq_inline_min`` key is not present, the value may be queried by the
452 driver from the NIC via DevX if this feature is available. If there is no DevX
453 enabled/supported the value 18 (supposing L2 header including VLAN) is set
454 for ConnectX-4 and ConnectX-4LX, and 0 is set by default for ConnectX-5
455 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
458 For ConnectX-4 NIC, driver does not allow specifying value below 18
459 (minimal L2 header, including VLAN), error will be raised.
461 For ConnectX-4LX NIC, it is allowed to specify values below 18, but
462 it is not recommended and may prevent NIC from sending packets over
465 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
466 Multi-Packet Write), because last one does not support partial packet inlining.
467 This is not very critical due to minimal data inlining is mostly required
468 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
470 - ``txq_inline_max`` parameter [int]
472 Specifies the maximal packet length to be completely inlined into WQE
473 Ethernet Segment for ordinary SEND method. If packet is larger than specified
474 value, the packet data won't be copied by the driver at all, data buffer
475 is addressed with a pointer. If packet length is less or equal all packet
476 data will be copied into WQE. This may improve PCI bandwidth utilization for
477 short packets significantly but requires the extra CPU cycles.
479 The data inline feature is controlled by number of Tx queues, if number of Tx
480 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
481 is engaged, if there are not enough Tx queues (which means not enough CPU cores
482 and CPU resources are scarce), data inline is not performed by the driver.
483 Assigning ``txqs_min_inline`` with zero always enables the data inline.
485 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
486 by the driver in order not to exceed the limit (930 bytes) and to provide better
487 WQE space filling without gaps, the adjustment is reflected in the debug log.
488 Also, the default value (290) may be decreased in run-time if the large transmit
489 queue size is requested and hardware does not support enough descriptor
490 amount, in this case warning is emitted. If ``txq_inline_max`` key is
491 specified and requested inline settings can not be satisfied then error
494 - ``txq_inline_mpw`` parameter [int]
496 Specifies the maximal packet length to be completely inlined into WQE for
497 Enhanced MPW method. If packet is large the specified value, the packet data
498 won't be copied, and data buffer is addressed with pointer. If packet length
499 is less or equal, all packet data will be copied into WQE. This may improve PCI
500 bandwidth utilization for short packets significantly but requires the extra
503 The data inline feature is controlled by number of TX queues, if number of Tx
504 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
505 is engaged, if there are not enough Tx queues (which means not enough CPU cores
506 and CPU resources are scarce), data inline is not performed by the driver.
507 Assigning ``txqs_min_inline`` with zero always enables the data inline.
509 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
510 by the driver in order not to exceed the limit (930 bytes) and to provide better
511 WQE space filling without gaps, the adjustment is reflected in the debug log.
512 Due to multiple packets may be included to the same WQE with Enhanced Multi
513 Packet Write Method and overall WQE size is limited it is not recommended to
514 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
515 may be decreased in run-time if the large transmit queue size is requested
516 and hardware does not support enough descriptor amount, in this case warning
517 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
518 settings can not be satisfied then error will be raised.
520 - ``txqs_max_vec`` parameter [int]
522 Enable vectorized Tx only when the number of TX queues is less than or
523 equal to this value. This parameter is deprecated and ignored, kept
524 for compatibility issue to not prevent driver from probing.
526 - ``txq_mpw_hdr_dseg_en`` parameter [int]
528 A nonzero value enables including two pointers in the first block of TX
529 descriptor. The parameter is deprecated and ignored, kept for compatibility
532 - ``txq_max_inline_len`` parameter [int]
534 Maximum size of packet to be inlined. This limits the size of packet to
535 be inlined. If the size of a packet is larger than configured value, the
536 packet isn't inlined even though there's enough space remained in the
537 descriptor. Instead, the packet is included with pointer. This parameter
538 is deprecated and converted directly to ``txq_inline_mpw`` providing full
539 compatibility. Valid only if eMPW feature is engaged.
541 - ``txq_mpw_en`` parameter [int]
543 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
544 ConnectX-6, ConnectX-6 DX and BlueField. eMPW allows the TX burst function to pack
545 up multiple packets in a single descriptor session in order to save PCI bandwidth
546 and improve performance at the cost of a slightly higher CPU usage. When
547 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
548 entire packet data on to TX descriptor instead of including pointer of packet.
550 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
551 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
552 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
553 option or reported by the NIC, the eMPW feature is disengaged.
555 - ``tx_vec_en`` parameter [int]
557 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 DX
558 and BlueField NICs if the number of global Tx queues on the port is less than
559 ``txqs_max_vec``. The parameter is deprecated and ignored.
561 - ``rx_vec_en`` parameter [int]
563 A nonzero value enables Rx vector if the port is not configured in
564 multi-segment otherwise this parameter is ignored.
568 - ``vf_nl_en`` parameter [int]
570 A nonzero value enables Netlink requests from the VF to add/remove MAC
571 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
572 Otherwise the relevant configuration must be run with Linux iproute2 tools.
573 This is a prerequisite to receive this kind of traffic.
575 Enabled by default, valid only on VF devices ignored otherwise.
577 - ``l3_vxlan_en`` parameter [int]
579 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
580 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
581 parameter. This is a prerequisite to receive this kind of traffic.
585 - ``dv_flow_en`` parameter [int]
587 A nonzero value enables the DV flow steering assuming it is supported
592 - ``dv_esw_en`` parameter [int]
594 A nonzero value enables E-Switch using Direct Rules.
596 Enabled by default if supported.
598 - ``mr_ext_memseg_en`` parameter [int]
600 A nonzero value enables extending memseg when registering DMA memory. If
601 enabled, the number of entries in MR (Memory Region) lookup table on datapath
602 is minimized and it benefits performance. On the other hand, it worsens memory
603 utilization because registered memory is pinned by kernel driver. Even if a
604 page in the extended chunk is freed, that doesn't become reusable until the
605 entire memory is freed.
609 - ``representor`` parameter [list]
611 This parameter can be used to instantiate DPDK Ethernet devices from
612 existing port (or VF) representors configured on the device.
614 It is a standard parameter whose format is described in
615 :ref:`ethernet_device_standard_device_arguments`.
617 For instance, to probe port representors 0 through 2::
621 - ``max_dump_files_num`` parameter [int]
623 The maximum number of files per PMD entity that may be created for debug information.
624 The files will be created in /var/log directory or in current directory.
626 set to 128 by default.
628 - ``lro_timeout_usec`` parameter [int]
630 The maximum allowed duration of an LRO session, in micro-seconds.
631 PMD will set the nearest value supported by HW, which is not bigger than
632 the input ``lro_timeout_usec`` value.
633 If this parameter is not specified, by default PMD will set
634 the smallest value supported by HW.
636 .. _mlx5_firmware_config:
638 Firmware configuration
639 ~~~~~~~~~~~~~~~~~~~~~~
641 Firmware features can be configured as key/value pairs.
643 The command to set a value is::
645 mlxconfig -d <device> set <key>=<value>
647 The command to query a value is::
649 mlxconfig -d <device> query | grep <key>
651 The device name for the command ``mlxconfig`` can be either the PCI address,
652 or the mst device name found with::
656 Below are some firmware configurations listed.
662 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
668 - maximum number of SR-IOV virtual functions::
672 - enable DevX (required by Direct Rules and other features)::
676 - aggressive CQE zipping::
680 - L3 VXLAN and VXLAN-GPE destination UDP port::
683 IP_OVER_VXLAN_PORT=<udp dport>
685 - enable IP-in-IP tunnel flow matching::
687 FLEX_PARSER_PROFILE_ENABLE=0
689 - enable MPLS flow matching::
691 FLEX_PARSER_PROFILE_ENABLE=1
693 - enable ICMP/ICMP6 code/type fields matching::
695 FLEX_PARSER_PROFILE_ENABLE=2
697 - enable Geneve flow matching::
699 FLEX_PARSER_PROFILE_ENABLE=0
704 This driver relies on external libraries and kernel drivers for resources
705 allocations and initialization. The following dependencies are not part of
706 DPDK and must be installed separately:
710 User space Verbs framework used by librte_pmd_mlx5. This library provides
711 a generic interface between the kernel and low-level user space drivers
714 It allows slow and privileged operations (context initialization, hardware
715 resources allocations) to be managed by the kernel and fast operations to
716 never leave user space.
720 Low-level user space driver library for Mellanox
721 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
724 This library basically implements send/receive calls to the hardware
729 They provide the kernel-side Verbs API and low level device drivers that
730 manage actual hardware initialization and resources sharing with user
733 Unlike most other PMDs, these modules must remain loaded and bound to
736 - mlx5_core: hardware driver managing Mellanox
737 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
739 - mlx5_ib: InifiniBand device driver.
740 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
742 - **Firmware update**
744 Mellanox OFED/EN releases include firmware updates for
745 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
747 Because each release provides new features, these updates must be applied to
748 match the kernel modules and libraries they come with.
752 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
758 Either RDMA Core library with a recent enough Linux kernel release
759 (recommended) or Mellanox OFED/EN, which provides compatibility with older
762 RDMA Core with Linux Kernel
763 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
765 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
766 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
767 (see `RDMA Core installation documentation`_)
768 - When building for i686 use:
770 - rdma-core version 18.0 or above built with 32bit support.
771 - Kernel version 4.14.41 or above.
773 - Starting with rdma-core v21, static libraries can be built::
776 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
779 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
780 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
782 If rdma-core libraries are built but not installed, DPDK makefile can link them,
783 thanks to these environment variables:
785 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
786 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
787 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
792 - Mellanox OFED version: ** 4.5, 4.6** /
793 Mellanox EN version: **4.5, 4.6**
796 - ConnectX-4: **12.21.1000** and above.
797 - ConnectX-4 Lx: **14.21.1000** and above.
798 - ConnectX-5: **16.21.1000** and above.
799 - ConnectX-5 Ex: **16.21.1000** and above.
800 - ConnectX-6: **20.99.5374** and above.
801 - ConnectX-6 DX: **22.27.0090** and above.
802 - BlueField: **18.25.1010** and above.
804 While these libraries and kernel modules are available on OpenFabrics
805 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
806 managers on most distributions, this PMD requires Ethernet extensions that
807 may not be supported at the moment (this is a work in progress).
810 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
812 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
813 include the necessary support and should be used in the meantime. For DPDK,
814 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
815 required from that distribution.
819 Several versions of Mellanox OFED/EN are available. Installing the version
820 this DPDK release was developed and tested against is strongly
821 recommended. Please check the `prerequisites`_.
826 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
827 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
828 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
829 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
830 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
831 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
832 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
833 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
834 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
835 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
836 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
837 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
838 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
839 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
840 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
841 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
842 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
843 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
844 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
845 * Mellanox(R) ConnectX(R)-6 200G MCX654106A-HCAT (4x200G)
846 * Mellanox(R) ConnectX(R)-6DX EN 100G MCX623106AN-CDAT (2*100g)
847 * Mellanox(R) ConnectX(R)-6DX EN 200G MCX623105AN-VDAT (1*200g)
849 Quick Start Guide on OFED/EN
850 ----------------------------
852 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
855 2. Install the required libraries and kernel modules either by installing
856 only the required set, or by installing the entire Mellanox OFED/EN::
858 ./mlnxofedinstall --upstream-libs --dpdk
860 3. Verify the firmware is the correct one::
864 4. Verify all ports links are set to Ethernet::
866 mlxconfig -d <mst device> query | grep LINK_TYPE
870 Link types may have to be configured to Ethernet::
872 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
874 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
876 For hypervisors, verify SR-IOV is enabled on the NIC::
878 mlxconfig -d <mst device> query | grep SRIOV_EN
881 If needed, configure SR-IOV::
883 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
884 mlxfwreset -d <mst device> reset
886 5. Restart the driver::
888 /etc/init.d/openibd restart
892 service openibd restart
894 If link type was changed, firmware must be reset as well::
896 mlxfwreset -d <mst device> reset
898 For hypervisors, after reset write the sysfs number of virtual functions
901 To dynamically instantiate a given number of virtual functions (VFs)::
903 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
905 6. Compile DPDK and you are ready to go. See instructions on
906 :ref:`Development Kit Build System <Development_Kit_Build_System>`
908 Enable switchdev mode
909 ---------------------
911 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
912 Representor is a port in DPDK that is connected to a VF in such a way
913 that assuming there are no offload flows, each packet that is sent from the VF
914 will be received by the corresponding representor. While each packet that is
915 sent to a representor will be received by the VF.
916 This is very useful in case of SRIOV mode, where the first packet that is sent
917 by the VF will be received by the DPDK application which will decide if this
918 flow should be offloaded to the E-Switch. After offloading the flow packet
919 that the VF that are matching the flow will not be received any more by
920 the DPDK application.
922 1. Enable SRIOV mode::
924 mlxconfig -d <mst device> set SRIOV_EN=true
926 2. Configure the max number of VFs::
928 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
932 mlxfwreset -d <mst device> reset
934 3. Configure the actual number of VFs::
936 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
938 4. Unbind the device (can be rebind after the switchdev mode)::
940 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
942 5. Enbale switchdev mode::
944 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
949 1. Configure aggressive CQE Zipping for maximum performance::
951 mlxconfig -d <mst device> s CQE_COMPRESSION=1
953 To set it back to the default CQE Zipping mode use::
955 mlxconfig -d <mst device> s CQE_COMPRESSION=0
957 2. In case of virtualization:
959 - Make sure that hypervisor kernel is 3.16 or newer.
960 - Configure boot with ``iommu=pt``.
962 - Make sure to allocate a VM on huge pages.
963 - Make sure to set CPU pinning.
965 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
966 for better performance. For VMs, verify that the right CPU
967 and NUMA node are pinned according to the above. Run::
971 to identify the NUMA node to which the PCIe adapter is connected.
973 4. If more than one adapter is used, and root complex capabilities allow
974 to put both adapters on the same NUMA node without PCI bandwidth degradation,
975 it is recommended to locate both adapters on the same NUMA node.
976 This in order to forward packets from one to the other without
977 NUMA performance penalty.
979 5. Disable pause frames::
981 ethtool -A <netdev> rx off tx off
983 6. Verify IO non-posted prefetch is disabled by default. This can be checked
984 via the BIOS configuration. Please contact you server provider for more
985 information about the settings.
989 On some machines, depends on the machine integrator, it is beneficial
990 to set the PCI max read request parameter to 1K. This can be
991 done in the following way:
993 To query the read request size use::
995 setpci -s <NIC PCI address> 68.w
997 If the output is different than 3XXX, set it by::
999 setpci -s <NIC PCI address> 68.w=3XXX
1001 The XXX can be different on different systems. Make sure to configure
1002 according to the setpci output.
1004 7. To minimize overhead of searching Memory Regions:
1006 - '--socket-mem' is recommended to pin memory by predictable amount.
1007 - Configure per-lcore cache when creating Mempools for packet buffer.
1008 - Refrain from dynamically allocating/freeing memory in run-time.
1010 .. _mlx5_offloads_support:
1012 Supported hardware offloads
1013 ---------------------------
1015 .. table:: Minimal SW/HW versions for queue offloads
1017 ============== ===== ===== ========= ===== ========== ==========
1018 Offload DPDK Linux rdma-core OFED firmware hardware
1019 ============== ===== ===== ========= ===== ========== ==========
1020 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1021 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1022 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1023 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1024 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1025 ============== ===== ===== ========= ===== ========== ==========
1027 .. table:: Minimal SW/HW versions for rte_flow offloads
1029 +-----------------------+-----------------+-----------------+
1030 | Offload | with E-Switch | with vNIC |
1031 +=======================+=================+=================+
1032 | Count | | DPDK 19.05 | | DPDK 19.02 |
1033 | | | OFED 4.6 | | OFED 4.6 |
1034 | | | rdma-core 24 | | rdma-core 23 |
1035 | | | ConnectX-5 | | ConnectX-5 |
1036 +-----------------------+-----------------+-----------------+
1037 | Drop / Queue / RSS | | DPDK 19.05 | | DPDK 18.11 |
1038 | | | OFED 4.6 | | OFED 4.5 |
1039 | | | rdma-core 24 | | rdma-core 23 |
1040 | | | ConnectX-5 | | ConnectX-4 |
1041 +-----------------------+-----------------+-----------------+
1042 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1043 | (VXLAN / NVGRE / RAW) | | OFED 4.6-2 | | OFED 4.6 |
1044 | | | rdma-core 24 | | rdma-core 23 |
1045 | | | ConnectX-5 | | ConnectX-5 |
1046 +-----------------------+-----------------+-----------------+
1047 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1048 | | (set_ipv4_src / | | OFED 4.6-2 | | OFED 4.6-2 |
1049 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 23 |
1050 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1051 | | set_ipv6_dst / | | | | |
1052 | | set_tp_src / | | | | |
1053 | | set_tp_dst / | | | | |
1054 | | dec_ttl / | | | | |
1055 | | set_ttl / | | | | |
1056 | | set_mac_src / | | | | |
1057 | | set_mac_dst) | | | | |
1059 | | (of_set_vlan_vid) | | DPDK 19.11 | | DPDK 19.11 |
1060 | | | OFED 4.6-4 | | OFED 4.6-4 |
1061 | | | ConnectX-5 | | ConnectX-5 |
1062 +-----------------------+-----------------+-----------------+
1063 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1064 | | | OFED 4.6-4 | | OFED 4.6-4 |
1065 | | | rdma-core 24 | | N/A |
1066 | | | ConnectX-5 | | ConnectX-5 |
1067 +-----------------------+-----------------+-----------------+
1068 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1069 | | | OFED 4.6 | | OFED 4.5 |
1070 | | | rdma-core 24 | | rdma-core 23 |
1071 | | | ConnectX-5 | | ConnectX-4 |
1072 +-----------------------+-----------------+-----------------+
1073 | Port ID | | DPDK 19.05 | | N/A |
1074 | | | OFED 4.6 | | N/A |
1075 | | | rdma-core 24 | | N/A |
1076 | | | ConnectX-5 | | N/A |
1077 +-----------------------+-----------------+-----------------+
1078 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1079 | | (of_pop_vlan / | | OFED 4.6-4 | | OFED 4.6-4 |
1080 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1081 | | of_set_vlan_pcp / | | |
1082 | | of_set_vlan_vid) | | |
1083 +-----------------------+-----------------+-----------------+
1088 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1089 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1091 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1092 command-line parameter to enable additional protocols (UDP and TCP as well
1093 as IP), the following commands must be entered from its CLI to get the same
1094 behavior as librte_pmd_mlx4::
1097 > port config all rss all
1103 This section demonstrates how to launch **testpmd** with Mellanox
1104 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1106 #. Load the kernel modules::
1108 modprobe -a ib_uverbs mlx5_core mlx5_ib
1110 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1113 /etc/init.d/openibd restart
1117 User space I/O kernel modules (uio and igb_uio) are not used and do
1118 not have to be loaded.
1120 #. Make sure Ethernet interfaces are in working order and linked to kernel
1121 verbs. Related sysfs entries should be present::
1123 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1132 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1135 for intf in eth2 eth3 eth4 eth5;
1137 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1140 sed -n 's,.*/\(.*\),-w \1,p'
1149 #. Request huge pages::
1151 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1153 #. Start testpmd with basic parameters::
1155 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
1160 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1161 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1162 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1163 PMD: librte_pmd_mlx5: 1 port(s) detected
1164 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1165 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1166 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1167 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1168 PMD: librte_pmd_mlx5: 1 port(s) detected
1169 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1170 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1171 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1172 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1173 PMD: librte_pmd_mlx5: 1 port(s) detected
1174 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1175 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1176 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1177 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1178 PMD: librte_pmd_mlx5: 1 port(s) detected
1179 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1180 Interactive-mode selected
1181 Configuring Port 0 (socket 0)
1182 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1183 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1184 Port 0: E4:1D:2D:E7:0C:FE
1185 Configuring Port 1 (socket 0)
1186 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1187 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1188 Port 1: E4:1D:2D:E7:0C:FF
1189 Configuring Port 2 (socket 0)
1190 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1191 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1192 Port 2: E4:1D:2D:E7:0C:FA
1193 Configuring Port 3 (socket 0)
1194 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1195 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1196 Port 3: E4:1D:2D:E7:0C:FB
1197 Checking link statuses...
1198 Port 0 Link Up - speed 40000 Mbps - full-duplex
1199 Port 1 Link Up - speed 40000 Mbps - full-duplex
1200 Port 2 Link Up - speed 10000 Mbps - full-duplex
1201 Port 3 Link Up - speed 10000 Mbps - full-duplex