1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright 2015 6WIND S.A.
3 Copyright 2015 Mellanox Technologies, Ltd
5 .. include:: <isonum.txt>
10 The MLX5 poll mode driver library (**librte_net_mlx5**) provides support
11 for **Mellanox ConnectX-4**, **Mellanox ConnectX-4 Lx** , **Mellanox
12 ConnectX-5**, **Mellanox ConnectX-6**, **Mellanox ConnectX-6 Dx**, **Mellanox
13 ConnectX-6 Lx**, **Mellanox BlueField** and **Mellanox BlueField-2** families
14 of 10/25/40/50/100/200 Gb/s adapters as well as their virtual functions (VF)
17 Information and documentation about these adapters can be found on the
18 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
19 `Mellanox community <http://community.mellanox.com/welcome>`__.
21 There is also a `section dedicated to this poll mode driver
22 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
28 Besides its dependency on libibverbs (that implies libmlx5 and associated
29 kernel support), librte_net_mlx5 relies heavily on system calls for control
30 operations such as querying/updating the MTU and flow control parameters.
32 For security reasons and robustness, this driver only deals with virtual
33 memory addresses. The way resources allocations are handled by the kernel,
34 combined with hardware specifications that allow to handle virtual memory
35 addresses directly, ensure that DPDK applications cannot access random
36 physical memory (or memory that does not belong to the current process).
38 This capability allows the PMD to coexist with kernel network interfaces
39 which remain functional, although they stop receiving unicast packets as
40 long as they share the same MAC address.
41 This means legacy linux control tools (for example: ethtool, ifconfig and
42 more) can operate on the same network interfaces that owned by the DPDK
45 The PMD can use libibverbs and libmlx5 to access the device firmware
46 or directly the hardware components.
47 There are different levels of objects and bypassing abilities
48 to get the best performances:
50 - Verbs is a complete high-level generic API
51 - Direct Verbs is a device-specific API
52 - DevX allows to access firmware objects
53 - Direct Rules manages flow steering at low-level hardware layer
55 Enabling librte_net_mlx5 causes DPDK applications to be linked against
61 - Multi arch support: x86_64, POWER8, ARMv8, i686.
62 - Multiple TX and RX queues.
63 - Support for scattered TX frames.
64 - Advanced support for scattered Rx frames with tunable buffer attributes.
65 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
66 - RSS using different combinations of fields: L3 only, L4 only or both,
67 and source only, destination only or both.
68 - Several RSS hash keys, one for each flow type.
69 - Default RSS operation with no hash key specification.
70 - Configurable RETA table.
71 - Link flow control (pause frame).
72 - Support for multiple MAC addresses.
76 - RX CRC stripping configuration.
77 - TX mbuf fast free offload.
78 - Promiscuous mode on PF and VF.
79 - Multicast promiscuous mode on PF and VF.
80 - Hardware checksum offloads.
81 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
83 - Flow API, including :ref:`flow_isolated_mode`.
85 - KVM and VMware ESX SR-IOV modes are supported.
86 - RSS hash result is supported.
87 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
88 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
90 - Statistics query including Basic, Extended and per queue.
92 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
93 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
94 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
95 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
96 - Flow insertion rate of more then million flows per second, when using Direct Rules.
97 - Support for multiple rte_flow groups.
98 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
101 - Multiple-thread flow insertion.
102 - Matching on GTP extension header with raw encap/decap action.
103 - Matching on Geneve TLV option header with raw encap/decap action.
104 - RSS support in sample action.
105 - E-Switch mirroring and jump.
106 - E-Switch mirroring and modify.
107 - 21844 flow priorities for ingress or egress flow groups greater than 0 and for any transfer
115 On Windows, the features are limited:
117 - Promiscuous mode is not supported
118 - The following rules are supported:
120 - IPv4/UDP with CVLAN filtering
121 - Unicast MAC filtering
123 - For secondary process:
125 - Forked secondary process not supported.
126 - External memory unregistered in EAL memseg list cannot be used for DMA
127 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
128 primary process and remapped to the same virtual address in secondary
129 process. If the external memory is registered by primary process but has
130 different virtual address in secondary process, unexpected error may happen.
132 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
133 specific VLAN will match for VLAN packets as well:
135 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
136 Meaning, the flow rule::
138 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
140 Will only match vlan packets with vid=3. and the flow rule::
142 flow create 0 ingress pattern eth / ipv4 / end ...
144 Will match any ipv4 packet (VLAN included).
146 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
148 - When using DV flow engine (``dv_flow_en`` = 1), flow pattern with any VLAN specification will match only single-tagged packets unless the ETH item ``type`` field is 0x88A8 or the VLAN item ``has_more_vlan`` field is 1.
151 flow create 0 ingress pattern eth / ipv4 / end ...
153 Will match any ipv4 packet.
156 flow create 0 ingress pattern eth / vlan / end ...
157 flow create 0 ingress pattern eth has_vlan is 1 / end ...
158 flow create 0 ingress pattern eth type is 0x8100 / end ...
160 Will match single-tagged packets only, with any VLAN ID value.
163 flow create 0 ingress pattern eth type is 0x88A8 / end ...
164 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
166 Will match multi-tagged packets only, with any VLAN ID value.
168 - A flow pattern with 2 sequential VLAN items is not supported.
170 - VLAN pop offload command:
172 - Flow rules having a VLAN pop offload command as one of their actions and
173 are lacking a match on VLAN as one of their items are not supported.
174 - The command is not supported on egress traffic.
176 - VLAN push offload is not supported on ingress traffic.
178 - VLAN set PCP offload is not supported on existing headers.
180 - A multi segment packet must have not more segments than reported by dev_infos_get()
181 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
182 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
183 inline settings) to 58.
185 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
186 to 0 are not supported.
188 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
190 - Match on Geneve header supports the following fields only:
197 - Match on Geneve TLV option is supported on the following fields:
204 Only one Class/Type/Length Geneve TLV option is supported per shared device.
205 Class/Type/Length fields must be specified as well as masks.
206 Class/Type/Length specified masks must be full.
207 Matching Geneve TLV option without specifying data is not supported.
208 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
210 - VF: flow rules created on VF devices can only match traffic targeted at the
211 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
213 - Match on GTP tunnel header item supports the following fields only:
215 - v_pt_rsv_flags: E flag, S flag, PN flag
219 - Match on GTP extension header only for GTP PDU session container (next
220 extension header type = 0x85).
221 - Match on GTP extension header is not supported in group 0.
223 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
224 The flows within group 0 and set metadata action are rejected by hardware.
228 MAC addresses not already present in the bridge table of the associated
229 kernel network device will be added and cleaned up by the PMD when closing
230 the device. In case of ungraceful program termination, some entries may
231 remain present and should be removed manually by other means.
233 - Buffer split offload is supported with regular Rx burst routine only,
234 no MPRQ feature or vectorized code can be engaged.
236 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
237 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
238 ol_flags. As the mempool for the external buffer is managed by PMD, all the
239 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
240 the external buffers will be freed by PMD and the application which still
241 holds the external buffers may be corrupted.
243 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
244 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
245 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
247 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
248 and allmulticast mode are both set to off.
249 To receive IPv6 Multicast messages on VM, explicitly set the relevant
250 MAC address using rte_eth_dev_mac_addr_add() API.
252 - To support a mixed traffic pattern (some buffers from local host memory, some
253 buffers from other devices) with high bandwidth, a mbuf flag is used.
255 An application hints the PMD whether or not it should try to inline the
256 given mbuf data buffer. PMD should do the best effort to act upon this request.
258 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
259 registered by application with rte_mbuf_dynflag_register(). This flag is
260 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
261 which is intended to be used by the application.
263 To query the supported specific flags in runtime,
264 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
265 currently (over present hardware and configuration) supported specific flags.
266 The "not inline hint" feature operating flow is the following one:
269 - probe the devices, ports are created
270 - query the port capabilities
271 - if port supporting the feature is found
272 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
273 - application starts the ports
274 - on ``dev_start()`` PMD checks whether the feature flag is registered and
275 enables the feature support in datapath
276 - application might set the registered flag bit in ``ol_flags`` field
277 of mbuf being sent and PMD will handle ones appropriately.
279 - The amount of descriptors in Tx queue may be limited by data inline settings.
280 Inline data require the more descriptor building blocks and overall block
281 amount may exceed the hardware supported limits. The application should
282 reduce the requested Tx size or adjust data inline settings with
283 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
285 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
286 parameter should be specified.
287 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
288 being sent it tries to synchronize the time of packet appearing on
289 the wire with the specified packet timestamp. It the specified one
290 is in the past it should be ignored, if one is in the distant future
291 it should be capped with some reasonable value (in range of seconds).
292 These specific cases ("too late" and "distant future") can be optionally
293 reported via device xstats to assist applications to detect the
294 time-related problems.
296 The timestamp upper "too-distant-future" limit
297 at the moment of invoking the Tx burst routine
298 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
299 Please note, for the testpmd txonly mode,
300 the limit is deduced from the expression::
302 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
304 There is no any packet reordering according timestamps is supposed,
305 neither within packet burst, nor between packets, it is an entirely
306 application responsibility to generate packets and its timestamps
307 in desired order. The timestamps can be put only in the first packet
308 in the burst providing the entire burst scheduling.
310 - E-Switch decapsulation Flow:
312 - can be applied to PF port only.
313 - must specify VF port action (packet redirection from PF to VF).
314 - optionally may specify tunnel inner source and destination MAC addresses.
316 - E-Switch encapsulation Flow:
318 - can be applied to VF ports only.
319 - must specify PF port action (packet redirection from VF to PF).
323 - The input buffer, used as outer header, is not validated.
327 - The decapsulation is always done up to the outermost tunnel detected by the HW.
328 - The input buffer, providing the removal size, is not validated.
329 - The buffer size must match the length of the headers to be removed.
331 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
332 mutually exclusive features which cannot be supported together
333 (see :ref:`mlx5_firmware_config`).
337 - Requires DevX and DV flow to be enabled.
338 - KEEP_CRC offload cannot be supported with LRO.
339 - The first mbuf length, without head-room, must be big enough to include the
341 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
342 it with size limited to max LRO size, not to max RX packet length.
343 - LRO can be used with outer header of TCP packets of the standard format:
344 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
346 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
347 - LRO packet aggregation is performed by HW only for packet size larger than
348 ``lro_min_mss_size``. This value is reported on device start, when debug
353 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
354 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
355 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
359 - fast free offload assumes the all mbufs being sent are originated from the
360 same memory pool and there is no any extra references to the mbufs (the
361 reference counter for each mbuf is equal 1 on tx_burst call). The latter
362 means there should be no any externally attached buffers in mbufs. It is
363 an application responsibility to provide the correct mbufs if the fast
364 free offload is engaged. The mlx5 PMD implicitly produces the mbufs with
365 externally attached buffers if MPRQ option is enabled, hence, the fast
366 free offload is neither supported nor advertised if there is MPRQ enabled.
370 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and E-Switch steering domain.
371 - The E-Switch Sample flow must have the eswitch_manager VPORT destination (PF or ECPF) and no additional actions.
372 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as first action in the E-Switch egress flow if with header modify or encapsulation actions.
374 - IPv6 header item 'proto' field, indicating the next header protocol, should
375 not be set as extension header.
376 In case the next header is an extension header, it should not be specified in
377 IPv6 header item 'proto' field.
378 The last extension header item 'next header' field can specify the following
379 header protocol type.
383 - Hairpin between two ports could only manual binding and explicit Tx flow mode. For single port hairpin, all the combinations of auto/manual binding and explicit/implicit Tx flow mode could be supported.
384 - Hairpin in switchdev SR-IOV mode is not supported till now.
389 MLX5 supports various methods to report statistics:
391 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.
393 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.
395 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.
403 The ibverbs libraries can be linked with this PMD in a number of ways,
404 configured by the ``ibverbs_link`` build option:
406 - ``shared`` (default): the PMD depends on some .so files.
408 - ``dlopen``: Split the dependencies glue in a separate library
409 loaded when needed by dlopen.
410 It make dependencies on libibverbs and libmlx4 optional,
411 and has no performance impact.
413 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
414 in the PMD shared library or the executable static binary.
416 Environment variables
417 ~~~~~~~~~~~~~~~~~~~~~
421 A list of directories in which to search for the rdma-core "glue" plug-in,
422 separated by colons or semi-colons.
424 - ``MLX5_SHUT_UP_BF``
426 Configures HW Tx doorbell register as IO-mapped.
428 By default, the HW Tx doorbell is configured as a write-combining register.
429 The register would be flushed to HW usually when the write-combining buffer
430 becomes full, but it depends on CPU design.
432 Except for vectorized Tx burst routines, a write memory barrier is enforced
433 after updating the register so that the update can be immediately visible to
436 When vectorized Tx burst is called, the barrier is set only if the burst size
437 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
438 variable will bring better latency even though the maximum throughput can
441 Run-time configuration
442 ~~~~~~~~~~~~~~~~~~~~~~
444 - librte_net_mlx5 brings kernel network interfaces up during initialization
445 because it is affected by their state. Forcing them down prevents packets
448 - **ethtool** operations on related kernel interfaces also affect the PMD.
453 In order to run as a non-root user,
454 some capabilities must be granted to the application::
456 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
458 Below are the reasons of the need for each capability:
461 When using physical addresses (PA mode), with Linux >= 4.0,
462 for access to ``/proc/self/pagemap``.
465 For device configuration.
468 For raw ethernet queue allocation through kernel driver.
471 For DMA memory pinning.
476 - ``rxq_cqe_comp_en`` parameter [int]
478 A nonzero value enables the compression of CQE on RX side. This feature
479 allows to save PCI bandwidth and improve performance. Enabled by default.
480 Different compression formats are supported in order to achieve the best
481 performance for different traffic patterns. Hash RSS format is the default.
483 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
484 better compression rate in case of RTE Flow Mark traffic.
485 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
486 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
487 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
491 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
492 ConnectX-6 Lx, BlueField and BlueField-2.
493 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
494 ConnectX-6 Lx, BlueField and BlueField-2.
496 - ``rxq_pkt_pad_en`` parameter [int]
498 A nonzero value enables padding Rx packet to the size of cacheline on PCI
499 transaction. This feature would waste PCI bandwidth but could improve
500 performance by avoiding partial cacheline write which may cause costly
501 read-modify-copy in memory transaction on some architectures. Disabled by
506 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
507 ConnectX-6 Lx, BlueField and BlueField-2.
508 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
509 ConnectX-6 Lx, BlueField and BlueField-2.
511 - ``mprq_en`` parameter [int]
513 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
514 configured as Multi-Packet RQ if the total number of Rx queues is
515 ``rxqs_min_mprq`` or more. Disabled by default.
517 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
518 by posting a single large buffer for multiple packets. Instead of posting a
519 buffers per a packet, one large buffer is posted in order to receive multiple
520 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
521 and each stride receives one packet. MPRQ can improve throughput for
522 small-packet traffic.
524 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
525 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
526 configure large stride size enough to accommodate max_rx_pkt_len as long as
527 device allows. Note that this can waste system memory compared to enabling Rx
528 scatter and multi-segment packet.
530 - ``mprq_log_stride_num`` parameter [int]
532 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
533 strides can reduce PCIe traffic further. If configured value is not in the
534 range of device capability, the default value will be set with a warning
535 message. The default value is 4 which is 16 strides per a buffer, valid only
536 if ``mprq_en`` is set.
538 The size of Rx queue should be bigger than the number of strides.
540 - ``mprq_log_stride_size`` parameter [int]
542 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
543 stride size can save some memory and reduce probability of a depletion of all
544 available strides due to unreleased packets by an application. If configured
545 value is not in the range of device capability, the default value will be set
546 with a warning message. The default value is 11 which is 2048 bytes per a
547 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
548 it is possible for a packet to span across multiple strides. This mode allows
549 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
550 of a packet if Rx scatter is configured) may be required in case there is no
551 space left for a head room at the end of a stride which incurs some
554 - ``mprq_max_memcpy_len`` parameter [int]
556 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
557 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
558 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
559 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
560 A mempool for external buffers will be allocated and managed by PMD. If Rx
561 packet is externally attached, ol_flags field of the mbuf will have
562 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
563 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
565 - ``rxqs_min_mprq`` parameter [int]
567 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
568 greater or equal to this value. The default value is 12, valid only if
571 - ``txq_inline`` parameter [int]
573 Amount of data to be inlined during TX operations. This parameter is
574 deprecated and converted to the new parameter ``txq_inline_max`` providing
575 partial compatibility.
577 - ``txqs_min_inline`` parameter [int]
579 Enable inline data send only when the number of TX queues is greater or equal
582 This option should be used in combination with ``txq_inline_max`` and
583 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
585 If this option is not specified the default value 16 is used for BlueField
586 and 8 for other platforms
588 The data inlining consumes the CPU cycles, so this option is intended to
589 auto enable inline data if we have enough Tx queues, which means we have
590 enough CPU cores and PCI bandwidth is getting more critical and CPU
591 is not supposed to be bottleneck anymore.
593 The copying data into WQE improves latency and can improve PPS performance
594 when PCI back pressure is detected and may be useful for scenarios involving
595 heavy traffic on many queues.
597 Because additional software logic is necessary to handle this mode, this
598 option should be used with care, as it may lower performance when back
599 pressure is not expected.
601 If inline data are enabled it may affect the maximal size of Tx queue in
602 descriptors because the inline data increase the descriptor size and
603 queue size limits supported by hardware may be exceeded.
605 - ``txq_inline_min`` parameter [int]
607 Minimal amount of data to be inlined into WQE during Tx operations. NICs
608 may require this minimal data amount to operate correctly. The exact value
609 may depend on NIC operation mode, requested offloads, etc. It is strongly
610 recommended to omit this parameter and use the default values. Anyway,
611 applications using this parameter should take into consideration that
612 specifying an inconsistent value may prevent the NIC from sending packets.
614 If ``txq_inline_min`` key is present the specified value (may be aligned
615 by the driver in order not to exceed the limits and provide better descriptor
616 space utilization) will be used by the driver and it is guaranteed that
617 requested amount of data bytes are inlined into the WQE beside other inline
618 settings. This key also may update ``txq_inline_max`` value (default
619 or specified explicitly in devargs) to reserve the space for inline data.
621 If ``txq_inline_min`` key is not present, the value may be queried by the
622 driver from the NIC via DevX if this feature is available. If there is no DevX
623 enabled/supported the value 18 (supposing L2 header including VLAN) is set
624 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
625 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
628 For ConnectX-4 NIC, driver does not allow specifying value below 18
629 (minimal L2 header, including VLAN), error will be raised.
631 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
632 it is not recommended and may prevent NIC from sending packets over
635 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
636 Multi-Packet Write), because last one does not support partial packet inlining.
637 This is not very critical due to minimal data inlining is mostly required
638 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
640 - ``txq_inline_max`` parameter [int]
642 Specifies the maximal packet length to be completely inlined into WQE
643 Ethernet Segment for ordinary SEND method. If packet is larger than specified
644 value, the packet data won't be copied by the driver at all, data buffer
645 is addressed with a pointer. If packet length is less or equal all packet
646 data will be copied into WQE. This may improve PCI bandwidth utilization for
647 short packets significantly but requires the extra CPU cycles.
649 The data inline feature is controlled by number of Tx queues, if number of Tx
650 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
651 is engaged, if there are not enough Tx queues (which means not enough CPU cores
652 and CPU resources are scarce), data inline is not performed by the driver.
653 Assigning ``txqs_min_inline`` with zero always enables the data inline.
655 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
656 by the driver in order not to exceed the limit (930 bytes) and to provide better
657 WQE space filling without gaps, the adjustment is reflected in the debug log.
658 Also, the default value (290) may be decreased in run-time if the large transmit
659 queue size is requested and hardware does not support enough descriptor
660 amount, in this case warning is emitted. If ``txq_inline_max`` key is
661 specified and requested inline settings can not be satisfied then error
664 - ``txq_inline_mpw`` parameter [int]
666 Specifies the maximal packet length to be completely inlined into WQE for
667 Enhanced MPW method. If packet is large the specified value, the packet data
668 won't be copied, and data buffer is addressed with pointer. If packet length
669 is less or equal, all packet data will be copied into WQE. This may improve PCI
670 bandwidth utilization for short packets significantly but requires the extra
673 The data inline feature is controlled by number of TX queues, if number of Tx
674 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
675 is engaged, if there are not enough Tx queues (which means not enough CPU cores
676 and CPU resources are scarce), data inline is not performed by the driver.
677 Assigning ``txqs_min_inline`` with zero always enables the data inline.
679 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
680 by the driver in order not to exceed the limit (930 bytes) and to provide better
681 WQE space filling without gaps, the adjustment is reflected in the debug log.
682 Due to multiple packets may be included to the same WQE with Enhanced Multi
683 Packet Write Method and overall WQE size is limited it is not recommended to
684 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
685 may be decreased in run-time if the large transmit queue size is requested
686 and hardware does not support enough descriptor amount, in this case warning
687 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
688 settings can not be satisfied then error will be raised.
690 - ``txqs_max_vec`` parameter [int]
692 Enable vectorized Tx only when the number of TX queues is less than or
693 equal to this value. This parameter is deprecated and ignored, kept
694 for compatibility issue to not prevent driver from probing.
696 - ``txq_mpw_hdr_dseg_en`` parameter [int]
698 A nonzero value enables including two pointers in the first block of TX
699 descriptor. The parameter is deprecated and ignored, kept for compatibility
702 - ``txq_max_inline_len`` parameter [int]
704 Maximum size of packet to be inlined. This limits the size of packet to
705 be inlined. If the size of a packet is larger than configured value, the
706 packet isn't inlined even though there's enough space remained in the
707 descriptor. Instead, the packet is included with pointer. This parameter
708 is deprecated and converted directly to ``txq_inline_mpw`` providing full
709 compatibility. Valid only if eMPW feature is engaged.
711 - ``txq_mpw_en`` parameter [int]
713 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
714 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
715 eMPW allows the Tx burst function to pack up multiple packets
716 in a single descriptor session in order to save PCI bandwidth
717 and improve performance at the cost of a slightly higher CPU usage.
718 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
719 Tx burst function copies entire packet data on to Tx descriptor
720 instead of including pointer of packet.
722 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
723 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
724 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
725 option or reported by the NIC, the eMPW feature is disengaged.
727 - ``tx_db_nc`` parameter [int]
729 The rdma core library can map doorbell register in two ways, depending on the
730 environment variable "MLX5_SHUT_UP_BF":
732 - As regular cached memory (usually with write combining attribute), if the
733 variable is either missing or set to zero.
734 - As non-cached memory, if the variable is present and set to not "0" value.
736 The type of mapping may slightly affect the Tx performance, the optimal choice
737 is strongly relied on the host architecture and should be deduced practically.
739 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
740 memory (with write combining), the PMD will perform the extra write memory barrier
741 after writing to doorbell, it might increase the needed CPU clocks per packet
742 to send, but latency might be improved.
744 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
745 cached memory, the PMD will not perform the extra write memory barrier
746 after writing to doorbell, on some architectures it might improve the
749 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
750 memory, the PMD will use heuristics to decide whether write memory barrier
751 should be performed. For bursts with size multiple of recommended one (64 pkts)
752 it is supposed the next burst is coming and no need to issue the extra memory
753 barrier (it is supposed to be issued in the next coming burst, at least after
754 descriptor writing). It might increase latency (on some hosts till next
755 packets transmit) and should be used with care.
757 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
758 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
759 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
761 - ``tx_pp`` parameter [int]
763 If a nonzero value is specified the driver creates all necessary internal
764 objects to provide accurate packet send scheduling on mbuf timestamps.
765 The positive value specifies the scheduling granularity in nanoseconds,
766 the packet send will be accurate up to specified digits. The allowed range is
767 from 500 to 1 million of nanoseconds. The negative value specifies the module
768 of granularity and engages the special test mode the check the schedule rate.
769 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
772 - ``tx_skew`` parameter [int]
774 The parameter adjusts the send packet scheduling on timestamps and represents
775 the average delay between beginning of the transmitting descriptor processing
776 by the hardware and appearance of actual packet data on the wire. The value
777 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
778 specified. The default value is zero.
780 - ``tx_vec_en`` parameter [int]
782 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
783 ConnectX-6 Lx, BlueField and BlueField-2 NICs
784 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
785 The parameter is deprecated and ignored.
787 - ``rx_vec_en`` parameter [int]
789 A nonzero value enables Rx vector if the port is not configured in
790 multi-segment otherwise this parameter is ignored.
794 - ``vf_nl_en`` parameter [int]
796 A nonzero value enables Netlink requests from the VF to add/remove MAC
797 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
798 Otherwise the relevant configuration must be run with Linux iproute2 tools.
799 This is a prerequisite to receive this kind of traffic.
801 Enabled by default, valid only on VF devices ignored otherwise.
803 - ``l3_vxlan_en`` parameter [int]
805 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
806 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
807 parameter. This is a prerequisite to receive this kind of traffic.
811 - ``dv_xmeta_en`` parameter [int]
813 A nonzero value enables extensive flow metadata support if device is
814 capable and driver supports it. This can enable extensive support of
815 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
816 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
818 There are some possible configurations, depending on parameter value:
820 - 0, this is default value, defines the legacy mode, the ``MARK`` and
821 ``META`` related actions and items operate only within NIC Tx and
822 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
823 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
824 item is 32 bits wide and match supported on egress only.
826 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
827 related actions and items operate within all supported steering domains,
828 including FDB, ``MARK`` and ``META`` information may cross the domain
829 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
830 depends on kernel and firmware configurations and might be 0, 16 or
831 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
832 compatibility, the actual width of data transferred to the FDB domain
833 depends on kernel configuration and may be vary. The actual supported
834 width can be retrieved in runtime by series of rte_flow_validate()
837 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
838 related actions and items operate within all supported steering domains,
839 including FDB, ``MARK`` and ``META`` information may cross the domain
840 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
841 depends on kernel and firmware configurations and might be 0, 16 or
842 24 bits. The actual supported width can be retrieved in runtime by
843 series of rte_flow_validate() trials.
845 - 3, this engages tunnel offload mode. In E-Switch configuration, that
846 mode implicitly activates ``dv_xmeta_en=1``.
848 +------+-----------+-----------+-------------+-------------+
849 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
850 +======+===========+===========+=============+=============+
851 | 0 | 24 bits | 32 bits | 32 bits | no |
852 +------+-----------+-----------+-------------+-------------+
853 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
854 +------+-----------+-----------+-------------+-------------+
855 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
856 +------+-----------+-----------+-------------+-------------+
858 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
859 ignored and the device is configured to operate in legacy mode (0).
861 Disabled by default (set to 0).
863 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
864 of the extensive metadata features. The legacy Verbs supports FLAG and
865 MARK metadata actions over NIC Rx steering domain only.
867 The setting MARK or META value to zero means there is no item provided and
868 receiving datapath will not report in mbufs these items are present.
870 For the MARK action the last 16 values in the full range are reserved for
871 internal PMD purposes (to emulate FLAG action). The valid range for the
872 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
873 for the 24-bit mode, the flows with the MARK action value outside
874 the specified range will be rejected.
876 - ``dv_flow_en`` parameter [int]
878 A nonzero value enables the DV flow steering assuming it is supported
879 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
881 Enabled by default if supported.
883 - ``dv_esw_en`` parameter [int]
885 A nonzero value enables E-Switch using Direct Rules.
887 Enabled by default if supported.
889 - ``lacp_by_user`` parameter [int]
891 A nonzero value enables the control of LACP traffic by the user application.
892 When a bond exists in the driver, by default it should be managed by the
893 kernel and therefore LACP traffic should be steered to the kernel.
894 If this devarg is set to 1 it will allow the user to manage the bond by
895 itself and not steer LACP traffic to the kernel.
897 Disabled by default (set to 0).
899 - ``mr_ext_memseg_en`` parameter [int]
901 A nonzero value enables extending memseg when registering DMA memory. If
902 enabled, the number of entries in MR (Memory Region) lookup table on datapath
903 is minimized and it benefits performance. On the other hand, it worsens memory
904 utilization because registered memory is pinned by kernel driver. Even if a
905 page in the extended chunk is freed, that doesn't become reusable until the
906 entire memory is freed.
910 - ``representor`` parameter [list]
912 This parameter can be used to instantiate DPDK Ethernet devices from
913 existing port (or VF) representors configured on the device.
915 It is a standard parameter whose format is described in
916 :ref:`ethernet_device_standard_device_arguments`.
918 For instance, to probe port representors 0 through 2::
922 - ``max_dump_files_num`` parameter [int]
924 The maximum number of files per PMD entity that may be created for debug information.
925 The files will be created in /var/log directory or in current directory.
927 set to 128 by default.
929 - ``lro_timeout_usec`` parameter [int]
931 The maximum allowed duration of an LRO session, in micro-seconds.
932 PMD will set the nearest value supported by HW, which is not bigger than
933 the input ``lro_timeout_usec`` value.
934 If this parameter is not specified, by default PMD will set
935 the smallest value supported by HW.
937 - ``hp_buf_log_sz`` parameter [int]
939 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
940 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
941 The capacity of the value is specified by the firmware and the initialization
942 will get a failure if it is out of scope.
943 The range of the value is from 11 to 19 right now, and the supported frame
944 size of a single packet for hairpin is from 512B to 128KB. It might change if
945 different firmware release is being used. By using a small value, it could
946 reduce memory consumption but not work with a large frame. If the value is
947 too large, the memory consumption will be high and some potential performance
948 degradation will be introduced.
949 By default, the PMD will set this value to 16, which means that 9KB jumbo
950 frames will be supported.
952 - ``reclaim_mem_mode`` parameter [int]
954 Cache some resources in flow destroy will help flow recreation more efficient.
955 While some systems may require the all the resources can be reclaimed after
957 The parameter ``reclaim_mem_mode`` provides the option for user to configure
958 if the resource cache is needed or not.
960 There are three options to choose:
962 - 0. It means the flow resources will be cached as usual. The resources will
963 be cached, helpful with flow insertion rate.
965 - 1. It will only enable the DPDK PMD level resources reclaim.
967 - 2. Both DPDK PMD level and rdma-core low level will be configured as
970 By default, the PMD will set this value to 0.
972 - ``sys_mem_en`` parameter [int]
974 A non-zero value enables the PMD memory management allocating memory
975 from system by default, without explicit rte memory flag.
977 By default, the PMD will set this value to 0.
979 - ``decap_en`` parameter [int]
981 Some devices do not support FCS (frame checksum) scattering for
982 tunnel-decapsulated packets.
983 If set to 0, this option forces the FCS feature and rejects tunnel
984 decapsulation in the flow engine for such devices.
986 By default, the PMD will set this value to 1.
988 .. _mlx5_firmware_config:
990 Firmware configuration
991 ~~~~~~~~~~~~~~~~~~~~~~
993 Firmware features can be configured as key/value pairs.
995 The command to set a value is::
997 mlxconfig -d <device> set <key>=<value>
999 The command to query a value is::
1001 mlxconfig -d <device> query | grep <key>
1003 The device name for the command ``mlxconfig`` can be either the PCI address,
1004 or the mst device name found with::
1008 Below are some firmware configurations listed.
1014 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1020 - maximum number of SR-IOV virtual functions::
1024 - enable DevX (required by Direct Rules and other features)::
1028 - aggressive CQE zipping::
1032 - L3 VXLAN and VXLAN-GPE destination UDP port::
1035 IP_OVER_VXLAN_PORT=<udp dport>
1037 - enable VXLAN-GPE tunnel flow matching::
1039 FLEX_PARSER_PROFILE_ENABLE=0
1041 FLEX_PARSER_PROFILE_ENABLE=2
1043 - enable IP-in-IP tunnel flow matching::
1045 FLEX_PARSER_PROFILE_ENABLE=0
1047 - enable MPLS flow matching::
1049 FLEX_PARSER_PROFILE_ENABLE=1
1051 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1053 FLEX_PARSER_PROFILE_ENABLE=2
1055 - enable Geneve flow matching::
1057 FLEX_PARSER_PROFILE_ENABLE=0
1059 FLEX_PARSER_PROFILE_ENABLE=1
1061 - enable Geneve TLV option flow matching::
1063 FLEX_PARSER_PROFILE_ENABLE=0
1065 - enable GTP flow matching::
1067 FLEX_PARSER_PROFILE_ENABLE=3
1069 - enable eCPRI flow matching::
1071 FLEX_PARSER_PROFILE_ENABLE=4
1077 This driver relies on external libraries and kernel drivers for resources
1078 allocations and initialization. The following dependencies are not part of
1079 DPDK and must be installed separately:
1083 User space Verbs framework used by librte_net_mlx5. This library provides
1084 a generic interface between the kernel and low-level user space drivers
1087 It allows slow and privileged operations (context initialization, hardware
1088 resources allocations) to be managed by the kernel and fast operations to
1089 never leave user space.
1093 Low-level user space driver library for Mellanox
1094 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1097 This library basically implements send/receive calls to the hardware
1100 - **Kernel modules**
1102 They provide the kernel-side Verbs API and low level device drivers that
1103 manage actual hardware initialization and resources sharing with user
1106 Unlike most other PMDs, these modules must remain loaded and bound to
1109 - mlx5_core: hardware driver managing Mellanox
1110 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1112 - mlx5_ib: InifiniBand device driver.
1113 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1115 - **Firmware update**
1117 Mellanox OFED/EN releases include firmware updates for
1118 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1120 Because each release provides new features, these updates must be applied to
1121 match the kernel modules and libraries they come with.
1125 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1131 Either RDMA Core library with a recent enough Linux kernel release
1132 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1135 RDMA Core with Linux Kernel
1136 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1138 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1139 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1140 (see `RDMA Core installation documentation`_)
1141 - When building for i686 use:
1143 - rdma-core version 18.0 or above built with 32bit support.
1144 - Kernel version 4.14.41 or above.
1146 - Starting with rdma-core v21, static libraries can be built::
1149 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1152 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1153 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1159 - Mellanox OFED version: **4.5** and above /
1160 Mellanox EN version: **4.5** and above
1163 - ConnectX-4: **12.21.1000** and above.
1164 - ConnectX-4 Lx: **14.21.1000** and above.
1165 - ConnectX-5: **16.21.1000** and above.
1166 - ConnectX-5 Ex: **16.21.1000** and above.
1167 - ConnectX-6: **20.27.0090** and above.
1168 - ConnectX-6 Dx: **22.27.0090** and above.
1169 - BlueField: **18.25.1010** and above.
1171 While these libraries and kernel modules are available on OpenFabrics
1172 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1173 managers on most distributions, this PMD requires Ethernet extensions that
1174 may not be supported at the moment (this is a work in progress).
1177 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1179 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1180 include the necessary support and should be used in the meantime. For DPDK,
1181 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1182 required from that distribution.
1186 Several versions of Mellanox OFED/EN are available. Installing the version
1187 this DPDK release was developed and tested against is strongly
1188 recommended. Please check the `linux prerequisites`_.
1190 Windows Prerequisites
1191 ---------------------
1193 This driver relies on external libraries and kernel drivers for resources
1194 allocations and initialization. The dependencies in the following sub-sections
1195 are not part of DPDK, and must be installed separately.
1197 Compilation Prerequisites
1198 ~~~~~~~~~~~~~~~~~~~~~~~~~
1200 DevX SDK installation
1201 ^^^^^^^^^^^^^^^^^^^^^
1203 The DevX SDK must be installed on the machine building the Windows PMD.
1204 Additional information can be found at
1205 `How to Integrate Windows DevX in Your Development Environment
1206 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1208 Runtime Prerequisites
1209 ~~~~~~~~~~~~~~~~~~~~~
1211 WinOF2 version 2.60 or higher must be installed on the machine.
1216 The driver can be downloaded from the following site:
1218 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1223 DevX for Windows must be enabled in the Windows registry.
1224 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1225 Additional information can be found in the WinOF2 user manual.
1230 The following Mellanox device families are supported by the same mlx5 driver:
1242 Below are detailed device names:
1244 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1245 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1246 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1247 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1248 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1249 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1250 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1251 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1252 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1253 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1254 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1255 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1256 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1257 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1258 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1259 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1260 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1261 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1262 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1263 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1264 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1265 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1266 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1267 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1268 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1269 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1271 Quick Start Guide on OFED/EN
1272 ----------------------------
1274 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1277 2. Install the required libraries and kernel modules either by installing
1278 only the required set, or by installing the entire Mellanox OFED/EN::
1280 ./mlnxofedinstall --upstream-libs --dpdk
1282 3. Verify the firmware is the correct one::
1286 4. Verify all ports links are set to Ethernet::
1288 mlxconfig -d <mst device> query | grep LINK_TYPE
1292 Link types may have to be configured to Ethernet::
1294 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1296 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1298 For hypervisors, verify SR-IOV is enabled on the NIC::
1300 mlxconfig -d <mst device> query | grep SRIOV_EN
1303 If needed, configure SR-IOV::
1305 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1306 mlxfwreset -d <mst device> reset
1308 5. Restart the driver::
1310 /etc/init.d/openibd restart
1314 service openibd restart
1316 If link type was changed, firmware must be reset as well::
1318 mlxfwreset -d <mst device> reset
1320 For hypervisors, after reset write the sysfs number of virtual functions
1323 To dynamically instantiate a given number of virtual functions (VFs)::
1325 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1327 6. Install DPDK and you are ready to go.
1328 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1330 Enable switchdev mode
1331 ---------------------
1333 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1334 Representor is a port in DPDK that is connected to a VF in such a way
1335 that assuming there are no offload flows, each packet that is sent from the VF
1336 will be received by the corresponding representor. While each packet that is
1337 sent to a representor will be received by the VF.
1338 This is very useful in case of SRIOV mode, where the first packet that is sent
1339 by the VF will be received by the DPDK application which will decide if this
1340 flow should be offloaded to the E-Switch. After offloading the flow packet
1341 that the VF that are matching the flow will not be received any more by
1342 the DPDK application.
1344 1. Enable SRIOV mode::
1346 mlxconfig -d <mst device> set SRIOV_EN=true
1348 2. Configure the max number of VFs::
1350 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1354 mlxfwreset -d <mst device> reset
1356 3. Configure the actual number of VFs::
1358 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1360 4. Unbind the device (can be rebind after the switchdev mode)::
1362 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1364 5. Enbale switchdev mode::
1366 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1371 1. Configure aggressive CQE Zipping for maximum performance::
1373 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1375 To set it back to the default CQE Zipping mode use::
1377 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1379 2. In case of virtualization:
1381 - Make sure that hypervisor kernel is 3.16 or newer.
1382 - Configure boot with ``iommu=pt``.
1383 - Use 1G huge pages.
1384 - Make sure to allocate a VM on huge pages.
1385 - Make sure to set CPU pinning.
1387 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1388 for better performance. For VMs, verify that the right CPU
1389 and NUMA node are pinned according to the above. Run::
1393 to identify the NUMA node to which the PCIe adapter is connected.
1395 4. If more than one adapter is used, and root complex capabilities allow
1396 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1397 it is recommended to locate both adapters on the same NUMA node.
1398 This in order to forward packets from one to the other without
1399 NUMA performance penalty.
1401 5. Disable pause frames::
1403 ethtool -A <netdev> rx off tx off
1405 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1406 via the BIOS configuration. Please contact you server provider for more
1407 information about the settings.
1411 On some machines, depends on the machine integrator, it is beneficial
1412 to set the PCI max read request parameter to 1K. This can be
1413 done in the following way:
1415 To query the read request size use::
1417 setpci -s <NIC PCI address> 68.w
1419 If the output is different than 3XXX, set it by::
1421 setpci -s <NIC PCI address> 68.w=3XXX
1423 The XXX can be different on different systems. Make sure to configure
1424 according to the setpci output.
1426 7. To minimize overhead of searching Memory Regions:
1428 - '--socket-mem' is recommended to pin memory by predictable amount.
1429 - Configure per-lcore cache when creating Mempools for packet buffer.
1430 - Refrain from dynamically allocating/freeing memory in run-time.
1435 There are multiple Rx burst functions with different advantages and limitations.
1437 .. table:: Rx burst functions
1439 +-------------------+------------------------+---------+-----------------+------+-------+
1440 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1441 | | | | || comp|| MTU |
1442 +===================+========================+=========+=================+======+=======+
1443 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1444 +-------------------+------------------------+---------+-----------------+------+-------+
1445 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1446 +-------------------+------------------------+---------+-----------------+------+-------+
1447 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1448 | || RxQs >= rxqs_min_mprq | | | | |
1449 +-------------------+------------------------+---------+-----------------+------+-------+
1450 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1451 | || mprq_en=1 | | | | |
1452 | || RxQs >= rxqs_min_mprq | | | | |
1453 +-------------------+------------------------+---------+-----------------+------+-------+
1455 .. _mlx5_offloads_support:
1457 Supported hardware offloads
1458 ---------------------------
1460 .. table:: Minimal SW/HW versions for queue offloads
1462 ============== ===== ===== ========= ===== ========== =============
1463 Offload DPDK Linux rdma-core OFED firmware hardware
1464 ============== ===== ===== ========= ===== ========== =============
1465 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1466 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1467 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1468 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1469 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1470 Buffer Split 20.11 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1471 ============== ===== ===== ========= ===== ========== =============
1473 .. table:: Minimal SW/HW versions for rte_flow offloads
1475 +-----------------------+-----------------+-----------------+
1476 | Offload | with E-Switch | with NIC |
1477 +=======================+=================+=================+
1478 | Count | | DPDK 19.05 | | DPDK 19.02 |
1479 | | | OFED 4.6 | | OFED 4.6 |
1480 | | | rdma-core 24 | | rdma-core 23 |
1481 | | | ConnectX-5 | | ConnectX-5 |
1482 +-----------------------+-----------------+-----------------+
1483 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1484 | | | OFED 4.6 | | OFED 4.5 |
1485 | | | rdma-core 24 | | rdma-core 23 |
1486 | | | ConnectX-5 | | ConnectX-4 |
1487 +-----------------------+-----------------+-----------------+
1488 | Queue / RSS | | | | DPDK 18.11 |
1489 | | | N/A | | OFED 4.5 |
1490 | | | | | rdma-core 23 |
1491 | | | | | ConnectX-4 |
1492 +-----------------------+-----------------+-----------------+
1493 | RSS shared action | | | | DPDK 20.11 |
1494 | | | N/A | | OFED 5.2 |
1495 | | | | | rdma-core 33 |
1496 | | | | | ConnectX-5 |
1497 +-----------------------+-----------------+-----------------+
1498 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1499 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1500 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1501 | | of_set_vlan_pcp / | | | | |
1502 | | of_set_vlan_vid) | | | | |
1503 +-----------------------+-----------------+-----------------+
1504 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1505 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1506 | | | rdma-core 24 | | rdma-core 23 |
1507 | | | ConnectX-5 | | ConnectX-5 |
1508 +-----------------------+-----------------+-----------------+
1509 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1510 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1511 | | | rdma-core 27 | | rdma-core 27 |
1512 | | | ConnectX-5 | | ConnectX-5 |
1513 +-----------------------+-----------------+-----------------+
1514 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1515 | | | OFED 5.1-2 | | OFED 5.1-2 |
1516 | | | rdma-core 32 | | N/A |
1517 | | | ConnectX-5 | | ConnectX-5 |
1518 +-----------------------+-----------------+-----------------+
1519 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1520 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1521 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1522 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1523 | | set_ipv6_dst / | | | | |
1524 | | set_tp_src / | | | | |
1525 | | set_tp_dst / | | | | |
1526 | | dec_ttl / | | | | |
1527 | | set_ttl / | | | | |
1528 | | set_mac_src / | | | | |
1529 | | set_mac_dst) | | | | |
1530 +-----------------------+-----------------+-----------------+
1531 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1532 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1533 | | | | rdma-core 24 | | rdma-core 24 |
1534 | | | | ConnectX-5 | | ConnectX-5 |
1535 +-----------------------+-----------------+-----------------+
1536 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1537 | | | OFED 4.7-1 | | OFED 4.7-1 |
1538 | | | rdma-core 24 | | N/A |
1539 | | | ConnectX-5 | | ConnectX-5 |
1540 +-----------------------+-----------------+-----------------+
1541 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1542 | | | OFED 4.6 | | OFED 4.5 |
1543 | | | rdma-core 24 | | rdma-core 23 |
1544 | | | ConnectX-5 | | ConnectX-4 |
1545 +-----------------------+-----------------+-----------------+
1546 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1547 | | | OFED 4.7-3 | | OFED 4.7-3 |
1548 | | | rdma-core 26 | | rdma-core 26 |
1549 | | | ConnectX-5 | | ConnectX-5 |
1550 +-----------------------+-----------------+-----------------+
1551 | Port ID | | DPDK 19.05 | | N/A |
1552 | | | OFED 4.7-1 | | N/A |
1553 | | | rdma-core 24 | | N/A |
1554 | | | ConnectX-5 | | N/A |
1555 +-----------------------+-----------------+-----------------+
1556 | Hairpin | | | | DPDK 19.11 |
1557 | | | N/A | | OFED 4.7-3 |
1558 | | | | | rdma-core 26 |
1559 | | | | | ConnectX-5 |
1560 +-----------------------+-----------------+-----------------+
1561 | 2-port Hairpin | | | | DPDK 20.11 |
1562 | | | N/A | | OFED 5.1-2 |
1564 | | | | | ConnectX-5 |
1565 +-----------------------+-----------------+-----------------+
1566 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1567 | | | OFED 4.7-3 | | OFED 4.7-3 |
1568 | | | rdma-core 26 | | rdma-core 26 |
1569 | | | ConnectX-5 | | ConnectX-5 |
1570 +-----------------------+-----------------+-----------------+
1571 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1572 | | | OFED 5.1-2 | | OFED 5.1-2 |
1573 | | | rdma-core 32 | | N/A |
1574 | | | ConnectX-5 | | ConnectX-5 |
1575 +-----------------------+-----------------+-----------------+
1576 | Age shared action | | DPDK 20.11 | | DPDK 20.11 |
1577 | | | OFED 5.2 | | OFED 5.2 |
1578 | | | rdma-core 32 | | rdma-core 32 |
1579 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1580 +-----------------------+-----------------+-----------------+
1581 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1582 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1583 | | | rdma-core 35 | | rdma-core 35 |
1584 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1585 +-----------------------+-----------------+-----------------+
1586 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1587 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1588 | | | rdma-core 34 | | rdma-core 34 |
1589 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1590 +-----------------------+-----------------+-----------------+
1595 MARK and META items are interrelated with datapath - they might move from/to
1596 the applications in mbuf fields. Hence, zero value for these items has the
1597 special meaning - it means "no metadata are provided", not zero values are
1598 treated by applications and PMD as valid ones.
1600 Moreover in the flow engine domain the value zero is acceptable to match and
1601 set, and we should allow to specify zero values as rte_flow parameters for the
1602 META and MARK items and actions. In the same time zero mask has no meaning and
1603 should be rejected on validation stage.
1608 Flows are not cached in the driver.
1609 When stopping a device port, all the flows created on this port from the
1610 application will be flushed automatically in the background.
1611 After stopping the device port, all flows on this port become invalid and
1612 not represented in the system.
1613 All references to these flows held by the application should be discarded
1614 directly but neither destroyed nor flushed.
1616 The application should re-create the flows as required after the port restart.
1621 Compared to librte_net_mlx4 that implements a single RSS configuration per
1622 port, librte_net_mlx5 supports per-protocol RSS configuration.
1624 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1625 command-line parameter to enable additional protocols (UDP and TCP as well
1626 as IP), the following commands must be entered from its CLI to get the same
1627 behavior as librte_net_mlx4::
1630 > port config all rss all
1636 This section demonstrates how to launch **testpmd** with Mellanox
1637 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1639 #. Load the kernel modules::
1641 modprobe -a ib_uverbs mlx5_core mlx5_ib
1643 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1646 /etc/init.d/openibd restart
1650 User space I/O kernel modules (uio and igb_uio) are not used and do
1651 not have to be loaded.
1653 #. Make sure Ethernet interfaces are in working order and linked to kernel
1654 verbs. Related sysfs entries should be present::
1656 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1665 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1668 for intf in eth2 eth3 eth4 eth5;
1670 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1673 sed -n 's,.*/\(.*\),-a \1,p'
1682 #. Request huge pages::
1684 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1686 #. Start testpmd with basic parameters::
1688 testpmd -l 8-15 -n 4 -a 05:00.0 -a 05:00.1 -a 06:00.0 -a 06:00.1 -- --rxq=2 --txq=2 -i
1693 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1694 EAL: probe driver: 15b3:1013 librte_net_mlx5
1695 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1696 PMD: librte_net_mlx5: 1 port(s) detected
1697 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1698 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1699 EAL: probe driver: 15b3:1013 librte_net_mlx5
1700 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1701 PMD: librte_net_mlx5: 1 port(s) detected
1702 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1703 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1704 EAL: probe driver: 15b3:1013 librte_net_mlx5
1705 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1706 PMD: librte_net_mlx5: 1 port(s) detected
1707 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1708 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1709 EAL: probe driver: 15b3:1013 librte_net_mlx5
1710 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1711 PMD: librte_net_mlx5: 1 port(s) detected
1712 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1713 Interactive-mode selected
1714 Configuring Port 0 (socket 0)
1715 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1716 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1717 Port 0: E4:1D:2D:E7:0C:FE
1718 Configuring Port 1 (socket 0)
1719 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1720 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1721 Port 1: E4:1D:2D:E7:0C:FF
1722 Configuring Port 2 (socket 0)
1723 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1724 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1725 Port 2: E4:1D:2D:E7:0C:FA
1726 Configuring Port 3 (socket 0)
1727 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1728 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1729 Port 3: E4:1D:2D:E7:0C:FB
1730 Checking link statuses...
1731 Port 0 Link Up - speed 40000 Mbps - full-duplex
1732 Port 1 Link Up - speed 40000 Mbps - full-duplex
1733 Port 2 Link Up - speed 10000 Mbps - full-duplex
1734 Port 3 Link Up - speed 10000 Mbps - full-duplex
1741 This section demonstrates how to dump flows. Currently, it's possible to dump
1742 all flows with assistance of external tools.
1744 #. 2 ways to get flow raw file:
1746 - Using testpmd CLI:
1748 .. code-block:: console
1750 testpmd> flow dump <port> <output_file>
1752 - call rte_flow_dev_dump api:
1754 .. code-block:: console
1756 rte_flow_dev_dump(port, file, NULL);
1758 #. Dump human-readable flows from raw file:
1760 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1762 .. code-block:: console
1764 mlx_steering_dump.py -f <output_file>