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 - Promiscuous mode on PF and VF.
78 - Multicast promiscuous mode on PF and VF.
79 - Hardware checksum offloads.
80 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
82 - Flow API, including :ref:`flow_isolated_mode`.
84 - KVM and VMware ESX SR-IOV modes are supported.
85 - RSS hash result is supported.
86 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
87 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
89 - Statistics query including Basic, Extended and per queue.
91 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
92 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
93 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
94 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
95 - Flow insertion rate of more then million flows per second, when using Direct Rules.
96 - Support for multiple rte_flow groups.
97 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
100 - Multiple-thread flow insertion.
101 - Matching on GTP extension header with raw encap/decap action.
102 - Matching on Geneve TLV option header with raw encap/decap action.
103 - RSS support in sample action.
110 On Windows, the features are limited:
112 - Promiscuous mode is not supported
113 - The following rules are supported:
115 - IPv4/UDP with CVLAN filtering
116 - Unicast MAC filtering
118 - For secondary process:
120 - Forked secondary process not supported.
121 - External memory unregistered in EAL memseg list cannot be used for DMA
122 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
123 primary process and remapped to the same virtual address in secondary
124 process. If the external memory is registered by primary process but has
125 different virtual address in secondary process, unexpected error may happen.
127 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
128 specific VLAN will match for VLAN packets as well:
130 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
131 Meaning, the flow rule::
133 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
135 Will only match vlan packets with vid=3. and the flow rule::
137 flow create 0 ingress pattern eth / ipv4 / end ...
139 Will match any ipv4 packet (VLAN included).
141 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
143 - 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.
146 flow create 0 ingress pattern eth / ipv4 / end ...
148 Will match any ipv4 packet.
151 flow create 0 ingress pattern eth / vlan / end ...
152 flow create 0 ingress pattern eth has_vlan is 1 / end ...
153 flow create 0 ingress pattern eth type is 0x8100 / end ...
155 Will match single-tagged packets only, with any VLAN ID value.
158 flow create 0 ingress pattern eth type is 0x88A8 / end ...
159 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
161 Will match multi-tagged packets only, with any VLAN ID value.
163 - A flow pattern with 2 sequential VLAN items is not supported.
165 - VLAN pop offload command:
167 - Flow rules having a VLAN pop offload command as one of their actions and
168 are lacking a match on VLAN as one of their items are not supported.
169 - The command is not supported on egress traffic.
171 - VLAN push offload is not supported on ingress traffic.
173 - VLAN set PCP offload is not supported on existing headers.
175 - A multi segment packet must have not more segments than reported by dev_infos_get()
176 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
177 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
178 inline settings) to 58.
180 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
181 to 0 are not supported.
183 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
185 - Match on Geneve header supports the following fields only:
192 - Match on Geneve TLV option is supported on the following fields:
199 Only one Class/Type/Length Geneve TLV option is supported per shared device.
200 Class/Type/Length fields must be specified as well as masks.
201 Class/Type/Length specified masks must be full.
202 Matching Geneve TLV option without specifying data is not supported.
203 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
205 - VF: flow rules created on VF devices can only match traffic targeted at the
206 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
208 - Match on GTP tunnel header item supports the following fields only:
210 - v_pt_rsv_flags: E flag, S flag, PN flag
214 - Match on GTP extension header only for GTP PDU session container (next
215 extension header type = 0x85).
216 - Match on GTP extension header is not supported in group 0.
218 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
219 The flows within group 0 and set metadata action are rejected by hardware.
223 MAC addresses not already present in the bridge table of the associated
224 kernel network device will be added and cleaned up by the PMD when closing
225 the device. In case of ungraceful program termination, some entries may
226 remain present and should be removed manually by other means.
228 - Buffer split offload is supported with regular Rx burst routine only,
229 no MPRQ feature or vectorized code can be engaged.
231 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
232 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
233 ol_flags. As the mempool for the external buffer is managed by PMD, all the
234 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
235 the external buffers will be freed by PMD and the application which still
236 holds the external buffers may be corrupted.
238 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
239 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
240 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
242 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
243 and allmulticast mode are both set to off.
244 To receive IPv6 Multicast messages on VM, explicitly set the relevant
245 MAC address using rte_eth_dev_mac_addr_add() API.
247 - To support a mixed traffic pattern (some buffers from local host memory, some
248 buffers from other devices) with high bandwidth, a mbuf flag is used.
250 An application hints the PMD whether or not it should try to inline the
251 given mbuf data buffer. PMD should do the best effort to act upon this request.
253 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
254 registered by application with rte_mbuf_dynflag_register(). This flag is
255 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
256 which is intended to be used by the application.
258 To query the supported specific flags in runtime,
259 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
260 currently (over present hardware and configuration) supported specific flags.
261 The "not inline hint" feature operating flow is the following one:
264 - probe the devices, ports are created
265 - query the port capabilities
266 - if port supporting the feature is found
267 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
268 - application starts the ports
269 - on ``dev_start()`` PMD checks whether the feature flag is registered and
270 enables the feature support in datapath
271 - application might set the registered flag bit in ``ol_flags`` field
272 of mbuf being sent and PMD will handle ones appropriately.
274 - The amount of descriptors in Tx queue may be limited by data inline settings.
275 Inline data require the more descriptor building blocks and overall block
276 amount may exceed the hardware supported limits. The application should
277 reduce the requested Tx size or adjust data inline settings with
278 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
280 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
281 parameter should be specified.
282 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
283 being sent it tries to synchronize the time of packet appearing on
284 the wire with the specified packet timestamp. It the specified one
285 is in the past it should be ignored, if one is in the distant future
286 it should be capped with some reasonable value (in range of seconds).
287 These specific cases ("too late" and "distant future") can be optionally
288 reported via device xstats to assist applications to detect the
289 time-related problems.
291 The timestamp upper "too-distant-future" limit
292 at the moment of invoking the Tx burst routine
293 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
294 Please note, for the testpmd txonly mode,
295 the limit is deduced from the expression::
297 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
299 There is no any packet reordering according timestamps is supposed,
300 neither within packet burst, nor between packets, it is an entirely
301 application responsibility to generate packets and its timestamps
302 in desired order. The timestamps can be put only in the first packet
303 in the burst providing the entire burst scheduling.
305 - E-Switch decapsulation Flow:
307 - can be applied to PF port only.
308 - must specify VF port action (packet redirection from PF to VF).
309 - optionally may specify tunnel inner source and destination MAC addresses.
311 - E-Switch encapsulation Flow:
313 - can be applied to VF ports only.
314 - must specify PF port action (packet redirection from VF to PF).
318 - The input buffer, used as outer header, is not validated.
322 - The decapsulation is always done up to the outermost tunnel detected by the HW.
323 - The input buffer, providing the removal size, is not validated.
324 - The buffer size must match the length of the headers to be removed.
326 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
327 mutually exclusive features which cannot be supported together
328 (see :ref:`mlx5_firmware_config`).
332 - Requires DevX and DV flow to be enabled.
333 - KEEP_CRC offload cannot be supported with LRO.
334 - The first mbuf length, without head-room, must be big enough to include the
336 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
337 it with size limited to max LRO size, not to max RX packet length.
338 - LRO can be used with outer header of TCP packets of the standard format:
339 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
341 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
342 - LRO packet aggregation is performed by HW only for packet size larger than
343 ``lro_min_mss_size``. This value is reported on device start, when debug
348 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
349 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
350 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
354 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and E-Switch steering domain.
355 - The E-Switch Sample flow must have the eswitch_manager VPORT destination (PF or ECPF) and no additional actions.
356 - 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.
358 - IPv6 header item 'proto' field, indicating the next header protocol, should
359 not be set as extension header.
360 In case the next header is an extension header, it should not be specified in
361 IPv6 header item 'proto' field.
362 The last extension header item 'next header' field can specify the following
363 header protocol type.
367 - 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.
368 - Hairpin in switchdev SR-IOV mode is not supported till now.
373 MLX5 supports various methods to report statistics:
375 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.
377 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.
379 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.
387 The ibverbs libraries can be linked with this PMD in a number of ways,
388 configured by the ``ibverbs_link`` build option:
390 - ``shared`` (default): the PMD depends on some .so files.
392 - ``dlopen``: Split the dependencies glue in a separate library
393 loaded when needed by dlopen.
394 It make dependencies on libibverbs and libmlx4 optional,
395 and has no performance impact.
397 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
398 in the PMD shared library or the executable static binary.
400 Environment variables
401 ~~~~~~~~~~~~~~~~~~~~~
405 A list of directories in which to search for the rdma-core "glue" plug-in,
406 separated by colons or semi-colons.
408 - ``MLX5_SHUT_UP_BF``
410 Configures HW Tx doorbell register as IO-mapped.
412 By default, the HW Tx doorbell is configured as a write-combining register.
413 The register would be flushed to HW usually when the write-combining buffer
414 becomes full, but it depends on CPU design.
416 Except for vectorized Tx burst routines, a write memory barrier is enforced
417 after updating the register so that the update can be immediately visible to
420 When vectorized Tx burst is called, the barrier is set only if the burst size
421 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
422 variable will bring better latency even though the maximum throughput can
425 Run-time configuration
426 ~~~~~~~~~~~~~~~~~~~~~~
428 - librte_net_mlx5 brings kernel network interfaces up during initialization
429 because it is affected by their state. Forcing them down prevents packets
432 - **ethtool** operations on related kernel interfaces also affect the PMD.
437 In order to run as a non-root user,
438 some capabilities must be granted to the application::
440 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
442 Below are the reasons of the need for each capability:
445 When using physical addresses (PA mode), with Linux >= 4.0,
446 for access to ``/proc/self/pagemap``.
449 For device configuration.
452 For raw ethernet queue allocation through kernel driver.
455 For DMA memory pinning.
460 - ``rxq_cqe_comp_en`` parameter [int]
462 A nonzero value enables the compression of CQE on RX side. This feature
463 allows to save PCI bandwidth and improve performance. Enabled by default.
464 Different compression formats are supported in order to achieve the best
465 performance for different traffic patterns. Hash RSS format is the default.
467 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
468 better compression rate in case of RTE Flow Mark traffic.
469 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
470 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
471 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
475 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
476 ConnectX-6 Lx, BlueField and BlueField-2.
477 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
478 ConnectX-6 Lx, BlueField and BlueField-2.
480 - ``rxq_pkt_pad_en`` parameter [int]
482 A nonzero value enables padding Rx packet to the size of cacheline on PCI
483 transaction. This feature would waste PCI bandwidth but could improve
484 performance by avoiding partial cacheline write which may cause costly
485 read-modify-copy in memory transaction on some architectures. Disabled by
490 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
491 ConnectX-6 Lx, BlueField and BlueField-2.
492 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
493 ConnectX-6 Lx, BlueField and BlueField-2.
495 - ``mprq_en`` parameter [int]
497 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
498 configured as Multi-Packet RQ if the total number of Rx queues is
499 ``rxqs_min_mprq`` or more. Disabled by default.
501 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
502 by posting a single large buffer for multiple packets. Instead of posting a
503 buffers per a packet, one large buffer is posted in order to receive multiple
504 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
505 and each stride receives one packet. MPRQ can improve throughput for
506 small-packet traffic.
508 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
509 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
510 configure large stride size enough to accommodate max_rx_pkt_len as long as
511 device allows. Note that this can waste system memory compared to enabling Rx
512 scatter and multi-segment packet.
514 - ``mprq_log_stride_num`` parameter [int]
516 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
517 strides can reduce PCIe traffic further. If configured value is not in the
518 range of device capability, the default value will be set with a warning
519 message. The default value is 4 which is 16 strides per a buffer, valid only
520 if ``mprq_en`` is set.
522 The size of Rx queue should be bigger than the number of strides.
524 - ``mprq_log_stride_size`` parameter [int]
526 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
527 stride size can save some memory and reduce probability of a depletion of all
528 available strides due to unreleased packets by an application. If configured
529 value is not in the range of device capability, the default value will be set
530 with a warning message. The default value is 11 which is 2048 bytes per a
531 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
532 it is possible for a packet to span across multiple strides. This mode allows
533 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
534 of a packet if Rx scatter is configured) may be required in case there is no
535 space left for a head room at the end of a stride which incurs some
538 - ``mprq_max_memcpy_len`` parameter [int]
540 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
541 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
542 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
543 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
544 A mempool for external buffers will be allocated and managed by PMD. If Rx
545 packet is externally attached, ol_flags field of the mbuf will have
546 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
547 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
549 - ``rxqs_min_mprq`` parameter [int]
551 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
552 greater or equal to this value. The default value is 12, valid only if
555 - ``txq_inline`` parameter [int]
557 Amount of data to be inlined during TX operations. This parameter is
558 deprecated and converted to the new parameter ``txq_inline_max`` providing
559 partial compatibility.
561 - ``txqs_min_inline`` parameter [int]
563 Enable inline data send only when the number of TX queues is greater or equal
566 This option should be used in combination with ``txq_inline_max`` and
567 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
569 If this option is not specified the default value 16 is used for BlueField
570 and 8 for other platforms
572 The data inlining consumes the CPU cycles, so this option is intended to
573 auto enable inline data if we have enough Tx queues, which means we have
574 enough CPU cores and PCI bandwidth is getting more critical and CPU
575 is not supposed to be bottleneck anymore.
577 The copying data into WQE improves latency and can improve PPS performance
578 when PCI back pressure is detected and may be useful for scenarios involving
579 heavy traffic on many queues.
581 Because additional software logic is necessary to handle this mode, this
582 option should be used with care, as it may lower performance when back
583 pressure is not expected.
585 If inline data are enabled it may affect the maximal size of Tx queue in
586 descriptors because the inline data increase the descriptor size and
587 queue size limits supported by hardware may be exceeded.
589 - ``txq_inline_min`` parameter [int]
591 Minimal amount of data to be inlined into WQE during Tx operations. NICs
592 may require this minimal data amount to operate correctly. The exact value
593 may depend on NIC operation mode, requested offloads, etc. It is strongly
594 recommended to omit this parameter and use the default values. Anyway,
595 applications using this parameter should take into consideration that
596 specifying an inconsistent value may prevent the NIC from sending packets.
598 If ``txq_inline_min`` key is present the specified value (may be aligned
599 by the driver in order not to exceed the limits and provide better descriptor
600 space utilization) will be used by the driver and it is guaranteed that
601 requested amount of data bytes are inlined into the WQE beside other inline
602 settings. This key also may update ``txq_inline_max`` value (default
603 or specified explicitly in devargs) to reserve the space for inline data.
605 If ``txq_inline_min`` key is not present, the value may be queried by the
606 driver from the NIC via DevX if this feature is available. If there is no DevX
607 enabled/supported the value 18 (supposing L2 header including VLAN) is set
608 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
609 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
612 For ConnectX-4 NIC, driver does not allow specifying value below 18
613 (minimal L2 header, including VLAN), error will be raised.
615 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
616 it is not recommended and may prevent NIC from sending packets over
619 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
620 Multi-Packet Write), because last one does not support partial packet inlining.
621 This is not very critical due to minimal data inlining is mostly required
622 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
624 - ``txq_inline_max`` parameter [int]
626 Specifies the maximal packet length to be completely inlined into WQE
627 Ethernet Segment for ordinary SEND method. If packet is larger than specified
628 value, the packet data won't be copied by the driver at all, data buffer
629 is addressed with a pointer. If packet length is less or equal all packet
630 data will be copied into WQE. This may improve PCI bandwidth utilization for
631 short packets significantly but requires the extra CPU cycles.
633 The data inline feature is controlled by number of Tx queues, if number of Tx
634 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
635 is engaged, if there are not enough Tx queues (which means not enough CPU cores
636 and CPU resources are scarce), data inline is not performed by the driver.
637 Assigning ``txqs_min_inline`` with zero always enables the data inline.
639 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
640 by the driver in order not to exceed the limit (930 bytes) and to provide better
641 WQE space filling without gaps, the adjustment is reflected in the debug log.
642 Also, the default value (290) may be decreased in run-time if the large transmit
643 queue size is requested and hardware does not support enough descriptor
644 amount, in this case warning is emitted. If ``txq_inline_max`` key is
645 specified and requested inline settings can not be satisfied then error
648 - ``txq_inline_mpw`` parameter [int]
650 Specifies the maximal packet length to be completely inlined into WQE for
651 Enhanced MPW method. If packet is large the specified value, the packet data
652 won't be copied, and data buffer is addressed with pointer. If packet length
653 is less or equal, all packet data will be copied into WQE. This may improve PCI
654 bandwidth utilization for short packets significantly but requires the extra
657 The data inline feature is controlled by number of TX queues, if number of Tx
658 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
659 is engaged, if there are not enough Tx queues (which means not enough CPU cores
660 and CPU resources are scarce), data inline is not performed by the driver.
661 Assigning ``txqs_min_inline`` with zero always enables the data inline.
663 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
664 by the driver in order not to exceed the limit (930 bytes) and to provide better
665 WQE space filling without gaps, the adjustment is reflected in the debug log.
666 Due to multiple packets may be included to the same WQE with Enhanced Multi
667 Packet Write Method and overall WQE size is limited it is not recommended to
668 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
669 may be decreased in run-time if the large transmit queue size is requested
670 and hardware does not support enough descriptor amount, in this case warning
671 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
672 settings can not be satisfied then error will be raised.
674 - ``txqs_max_vec`` parameter [int]
676 Enable vectorized Tx only when the number of TX queues is less than or
677 equal to this value. This parameter is deprecated and ignored, kept
678 for compatibility issue to not prevent driver from probing.
680 - ``txq_mpw_hdr_dseg_en`` parameter [int]
682 A nonzero value enables including two pointers in the first block of TX
683 descriptor. The parameter is deprecated and ignored, kept for compatibility
686 - ``txq_max_inline_len`` parameter [int]
688 Maximum size of packet to be inlined. This limits the size of packet to
689 be inlined. If the size of a packet is larger than configured value, the
690 packet isn't inlined even though there's enough space remained in the
691 descriptor. Instead, the packet is included with pointer. This parameter
692 is deprecated and converted directly to ``txq_inline_mpw`` providing full
693 compatibility. Valid only if eMPW feature is engaged.
695 - ``txq_mpw_en`` parameter [int]
697 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
698 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
699 eMPW allows the Tx burst function to pack up multiple packets
700 in a single descriptor session in order to save PCI bandwidth
701 and improve performance at the cost of a slightly higher CPU usage.
702 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
703 Tx burst function copies entire packet data on to Tx descriptor
704 instead of including pointer of packet.
706 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
707 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
708 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
709 option or reported by the NIC, the eMPW feature is disengaged.
711 - ``tx_db_nc`` parameter [int]
713 The rdma core library can map doorbell register in two ways, depending on the
714 environment variable "MLX5_SHUT_UP_BF":
716 - As regular cached memory (usually with write combining attribute), if the
717 variable is either missing or set to zero.
718 - As non-cached memory, if the variable is present and set to not "0" value.
720 The type of mapping may slightly affect the Tx performance, the optimal choice
721 is strongly relied on the host architecture and should be deduced practically.
723 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
724 memory (with write combining), the PMD will perform the extra write memory barrier
725 after writing to doorbell, it might increase the needed CPU clocks per packet
726 to send, but latency might be improved.
728 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
729 cached memory, the PMD will not perform the extra write memory barrier
730 after writing to doorbell, on some architectures it might improve the
733 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
734 memory, the PMD will use heuristics to decide whether write memory barrier
735 should be performed. For bursts with size multiple of recommended one (64 pkts)
736 it is supposed the next burst is coming and no need to issue the extra memory
737 barrier (it is supposed to be issued in the next coming burst, at least after
738 descriptor writing). It might increase latency (on some hosts till next
739 packets transmit) and should be used with care.
741 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
742 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
743 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
745 - ``tx_pp`` parameter [int]
747 If a nonzero value is specified the driver creates all necessary internal
748 objects to provide accurate packet send scheduling on mbuf timestamps.
749 The positive value specifies the scheduling granularity in nanoseconds,
750 the packet send will be accurate up to specified digits. The allowed range is
751 from 500 to 1 million of nanoseconds. The negative value specifies the module
752 of granularity and engages the special test mode the check the schedule rate.
753 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
756 - ``tx_skew`` parameter [int]
758 The parameter adjusts the send packet scheduling on timestamps and represents
759 the average delay between beginning of the transmitting descriptor processing
760 by the hardware and appearance of actual packet data on the wire. The value
761 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
762 specified. The default value is zero.
764 - ``tx_vec_en`` parameter [int]
766 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
767 ConnectX-6 Lx, BlueField and BlueField-2 NICs
768 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
769 The parameter is deprecated and ignored.
771 - ``rx_vec_en`` parameter [int]
773 A nonzero value enables Rx vector if the port is not configured in
774 multi-segment otherwise this parameter is ignored.
778 - ``vf_nl_en`` parameter [int]
780 A nonzero value enables Netlink requests from the VF to add/remove MAC
781 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
782 Otherwise the relevant configuration must be run with Linux iproute2 tools.
783 This is a prerequisite to receive this kind of traffic.
785 Enabled by default, valid only on VF devices ignored otherwise.
787 - ``l3_vxlan_en`` parameter [int]
789 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
790 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
791 parameter. This is a prerequisite to receive this kind of traffic.
795 - ``dv_xmeta_en`` parameter [int]
797 A nonzero value enables extensive flow metadata support if device is
798 capable and driver supports it. This can enable extensive support of
799 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
800 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
802 There are some possible configurations, depending on parameter value:
804 - 0, this is default value, defines the legacy mode, the ``MARK`` and
805 ``META`` related actions and items operate only within NIC Tx and
806 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
807 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
808 item is 32 bits wide and match supported on egress only.
810 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
811 related actions and items operate within all supported steering domains,
812 including FDB, ``MARK`` and ``META`` information may cross the domain
813 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
814 depends on kernel and firmware configurations and might be 0, 16 or
815 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
816 compatibility, the actual width of data transferred to the FDB domain
817 depends on kernel configuration and may be vary. The actual supported
818 width can be retrieved in runtime by series of rte_flow_validate()
821 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
822 related actions and items operate within all supported steering domains,
823 including FDB, ``MARK`` and ``META`` information may cross the domain
824 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
825 depends on kernel and firmware configurations and might be 0, 16 or
826 24 bits. The actual supported width can be retrieved in runtime by
827 series of rte_flow_validate() trials.
829 - 3, this engages tunnel offload mode. In E-Switch configuration, that
830 mode implicitly activates ``dv_xmeta_en=1``.
832 +------+-----------+-----------+-------------+-------------+
833 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
834 +======+===========+===========+=============+=============+
835 | 0 | 24 bits | 32 bits | 32 bits | no |
836 +------+-----------+-----------+-------------+-------------+
837 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
838 +------+-----------+-----------+-------------+-------------+
839 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
840 +------+-----------+-----------+-------------+-------------+
842 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
843 ignored and the device is configured to operate in legacy mode (0).
845 Disabled by default (set to 0).
847 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
848 of the extensive metadata features. The legacy Verbs supports FLAG and
849 MARK metadata actions over NIC Rx steering domain only.
851 - ``dv_flow_en`` parameter [int]
853 A nonzero value enables the DV flow steering assuming it is supported
854 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
856 Enabled by default if supported.
858 - ``dv_esw_en`` parameter [int]
860 A nonzero value enables E-Switch using Direct Rules.
862 Enabled by default if supported.
864 - ``lacp_by_user`` parameter [int]
866 A nonzero value enables the control of LACP traffic by the user application.
867 When a bond exists in the driver, by default it should be managed by the
868 kernel and therefore LACP traffic should be steered to the kernel.
869 If this devarg is set to 1 it will allow the user to manage the bond by
870 itself and not steer LACP traffic to the kernel.
872 Disabled by default (set to 0).
874 - ``mr_ext_memseg_en`` parameter [int]
876 A nonzero value enables extending memseg when registering DMA memory. If
877 enabled, the number of entries in MR (Memory Region) lookup table on datapath
878 is minimized and it benefits performance. On the other hand, it worsens memory
879 utilization because registered memory is pinned by kernel driver. Even if a
880 page in the extended chunk is freed, that doesn't become reusable until the
881 entire memory is freed.
885 - ``representor`` parameter [list]
887 This parameter can be used to instantiate DPDK Ethernet devices from
888 existing port (or VF) representors configured on the device.
890 It is a standard parameter whose format is described in
891 :ref:`ethernet_device_standard_device_arguments`.
893 For instance, to probe port representors 0 through 2::
897 - ``max_dump_files_num`` parameter [int]
899 The maximum number of files per PMD entity that may be created for debug information.
900 The files will be created in /var/log directory or in current directory.
902 set to 128 by default.
904 - ``lro_timeout_usec`` parameter [int]
906 The maximum allowed duration of an LRO session, in micro-seconds.
907 PMD will set the nearest value supported by HW, which is not bigger than
908 the input ``lro_timeout_usec`` value.
909 If this parameter is not specified, by default PMD will set
910 the smallest value supported by HW.
912 - ``hp_buf_log_sz`` parameter [int]
914 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
915 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
916 The capacity of the value is specified by the firmware and the initialization
917 will get a failure if it is out of scope.
918 The range of the value is from 11 to 19 right now, and the supported frame
919 size of a single packet for hairpin is from 512B to 128KB. It might change if
920 different firmware release is being used. By using a small value, it could
921 reduce memory consumption but not work with a large frame. If the value is
922 too large, the memory consumption will be high and some potential performance
923 degradation will be introduced.
924 By default, the PMD will set this value to 16, which means that 9KB jumbo
925 frames will be supported.
927 - ``reclaim_mem_mode`` parameter [int]
929 Cache some resources in flow destroy will help flow recreation more efficient.
930 While some systems may require the all the resources can be reclaimed after
932 The parameter ``reclaim_mem_mode`` provides the option for user to configure
933 if the resource cache is needed or not.
935 There are three options to choose:
937 - 0. It means the flow resources will be cached as usual. The resources will
938 be cached, helpful with flow insertion rate.
940 - 1. It will only enable the DPDK PMD level resources reclaim.
942 - 2. Both DPDK PMD level and rdma-core low level will be configured as
945 By default, the PMD will set this value to 0.
947 - ``sys_mem_en`` parameter [int]
949 A non-zero value enables the PMD memory management allocating memory
950 from system by default, without explicit rte memory flag.
952 By default, the PMD will set this value to 0.
954 - ``decap_en`` parameter [int]
956 Some devices do not support FCS (frame checksum) scattering for
957 tunnel-decapsulated packets.
958 If set to 0, this option forces the FCS feature and rejects tunnel
959 decapsulation in the flow engine for such devices.
961 By default, the PMD will set this value to 1.
963 .. _mlx5_firmware_config:
965 Firmware configuration
966 ~~~~~~~~~~~~~~~~~~~~~~
968 Firmware features can be configured as key/value pairs.
970 The command to set a value is::
972 mlxconfig -d <device> set <key>=<value>
974 The command to query a value is::
976 mlxconfig -d <device> query | grep <key>
978 The device name for the command ``mlxconfig`` can be either the PCI address,
979 or the mst device name found with::
983 Below are some firmware configurations listed.
989 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
995 - maximum number of SR-IOV virtual functions::
999 - enable DevX (required by Direct Rules and other features)::
1003 - aggressive CQE zipping::
1007 - L3 VXLAN and VXLAN-GPE destination UDP port::
1010 IP_OVER_VXLAN_PORT=<udp dport>
1012 - enable VXLAN-GPE tunnel flow matching::
1014 FLEX_PARSER_PROFILE_ENABLE=0
1016 FLEX_PARSER_PROFILE_ENABLE=2
1018 - enable IP-in-IP tunnel flow matching::
1020 FLEX_PARSER_PROFILE_ENABLE=0
1022 - enable MPLS flow matching::
1024 FLEX_PARSER_PROFILE_ENABLE=1
1026 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1028 FLEX_PARSER_PROFILE_ENABLE=2
1030 - enable Geneve flow matching::
1032 FLEX_PARSER_PROFILE_ENABLE=0
1034 FLEX_PARSER_PROFILE_ENABLE=1
1036 - enable Geneve TLV option flow matching::
1038 FLEX_PARSER_PROFILE_ENABLE=0
1040 - enable GTP flow matching::
1042 FLEX_PARSER_PROFILE_ENABLE=3
1044 - enable eCPRI flow matching::
1046 FLEX_PARSER_PROFILE_ENABLE=4
1052 This driver relies on external libraries and kernel drivers for resources
1053 allocations and initialization. The following dependencies are not part of
1054 DPDK and must be installed separately:
1058 User space Verbs framework used by librte_net_mlx5. This library provides
1059 a generic interface between the kernel and low-level user space drivers
1062 It allows slow and privileged operations (context initialization, hardware
1063 resources allocations) to be managed by the kernel and fast operations to
1064 never leave user space.
1068 Low-level user space driver library for Mellanox
1069 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1072 This library basically implements send/receive calls to the hardware
1075 - **Kernel modules**
1077 They provide the kernel-side Verbs API and low level device drivers that
1078 manage actual hardware initialization and resources sharing with user
1081 Unlike most other PMDs, these modules must remain loaded and bound to
1084 - mlx5_core: hardware driver managing Mellanox
1085 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1087 - mlx5_ib: InifiniBand device driver.
1088 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1090 - **Firmware update**
1092 Mellanox OFED/EN releases include firmware updates for
1093 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1095 Because each release provides new features, these updates must be applied to
1096 match the kernel modules and libraries they come with.
1100 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1106 Either RDMA Core library with a recent enough Linux kernel release
1107 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1110 RDMA Core with Linux Kernel
1111 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1113 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1114 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1115 (see `RDMA Core installation documentation`_)
1116 - When building for i686 use:
1118 - rdma-core version 18.0 or above built with 32bit support.
1119 - Kernel version 4.14.41 or above.
1121 - Starting with rdma-core v21, static libraries can be built::
1124 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1127 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1128 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1134 - Mellanox OFED version: **4.5** and above /
1135 Mellanox EN version: **4.5** and above
1138 - ConnectX-4: **12.21.1000** and above.
1139 - ConnectX-4 Lx: **14.21.1000** and above.
1140 - ConnectX-5: **16.21.1000** and above.
1141 - ConnectX-5 Ex: **16.21.1000** and above.
1142 - ConnectX-6: **20.27.0090** and above.
1143 - ConnectX-6 Dx: **22.27.0090** and above.
1144 - BlueField: **18.25.1010** and above.
1146 While these libraries and kernel modules are available on OpenFabrics
1147 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1148 managers on most distributions, this PMD requires Ethernet extensions that
1149 may not be supported at the moment (this is a work in progress).
1152 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1154 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1155 include the necessary support and should be used in the meantime. For DPDK,
1156 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1157 required from that distribution.
1161 Several versions of Mellanox OFED/EN are available. Installing the version
1162 this DPDK release was developed and tested against is strongly
1163 recommended. Please check the `linux prerequisites`_.
1165 Windows Prerequisites
1166 ---------------------
1168 This driver relies on external libraries and kernel drivers for resources
1169 allocations and initialization. The dependencies in the following sub-sections
1170 are not part of DPDK, and must be installed separately.
1172 Compilation Prerequisites
1173 ~~~~~~~~~~~~~~~~~~~~~~~~~
1175 DevX SDK installation
1176 ^^^^^^^^^^^^^^^^^^^^^
1178 The DevX SDK must be installed on the machine building the Windows PMD.
1179 Additional information can be found at
1180 `How to Integrate Windows DevX in Your Development Environment
1181 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1183 Runtime Prerequisites
1184 ~~~~~~~~~~~~~~~~~~~~~
1186 WinOF2 version 2.60 or higher must be installed on the machine.
1191 The driver can be downloaded from the following site:
1193 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1198 DevX for Windows must be enabled in the Windows registry.
1199 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1200 Additional information can be found in the WinOF2 user manual.
1205 The following Mellanox device families are supported by the same mlx5 driver:
1217 Below are detailed device names:
1219 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1220 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1221 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1222 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1223 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1224 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1225 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1226 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1227 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1228 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1229 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1230 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1231 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1232 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1233 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1234 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1235 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1236 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1237 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1238 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1239 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1240 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1241 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1242 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1243 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1244 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1246 Quick Start Guide on OFED/EN
1247 ----------------------------
1249 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1252 2. Install the required libraries and kernel modules either by installing
1253 only the required set, or by installing the entire Mellanox OFED/EN::
1255 ./mlnxofedinstall --upstream-libs --dpdk
1257 3. Verify the firmware is the correct one::
1261 4. Verify all ports links are set to Ethernet::
1263 mlxconfig -d <mst device> query | grep LINK_TYPE
1267 Link types may have to be configured to Ethernet::
1269 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1271 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1273 For hypervisors, verify SR-IOV is enabled on the NIC::
1275 mlxconfig -d <mst device> query | grep SRIOV_EN
1278 If needed, configure SR-IOV::
1280 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1281 mlxfwreset -d <mst device> reset
1283 5. Restart the driver::
1285 /etc/init.d/openibd restart
1289 service openibd restart
1291 If link type was changed, firmware must be reset as well::
1293 mlxfwreset -d <mst device> reset
1295 For hypervisors, after reset write the sysfs number of virtual functions
1298 To dynamically instantiate a given number of virtual functions (VFs)::
1300 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1302 6. Install DPDK and you are ready to go.
1303 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1305 Enable switchdev mode
1306 ---------------------
1308 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1309 Representor is a port in DPDK that is connected to a VF in such a way
1310 that assuming there are no offload flows, each packet that is sent from the VF
1311 will be received by the corresponding representor. While each packet that is
1312 sent to a representor will be received by the VF.
1313 This is very useful in case of SRIOV mode, where the first packet that is sent
1314 by the VF will be received by the DPDK application which will decide if this
1315 flow should be offloaded to the E-Switch. After offloading the flow packet
1316 that the VF that are matching the flow will not be received any more by
1317 the DPDK application.
1319 1. Enable SRIOV mode::
1321 mlxconfig -d <mst device> set SRIOV_EN=true
1323 2. Configure the max number of VFs::
1325 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1329 mlxfwreset -d <mst device> reset
1331 3. Configure the actual number of VFs::
1333 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1335 4. Unbind the device (can be rebind after the switchdev mode)::
1337 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1339 5. Enbale switchdev mode::
1341 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1346 1. Configure aggressive CQE Zipping for maximum performance::
1348 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1350 To set it back to the default CQE Zipping mode use::
1352 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1354 2. In case of virtualization:
1356 - Make sure that hypervisor kernel is 3.16 or newer.
1357 - Configure boot with ``iommu=pt``.
1358 - Use 1G huge pages.
1359 - Make sure to allocate a VM on huge pages.
1360 - Make sure to set CPU pinning.
1362 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1363 for better performance. For VMs, verify that the right CPU
1364 and NUMA node are pinned according to the above. Run::
1368 to identify the NUMA node to which the PCIe adapter is connected.
1370 4. If more than one adapter is used, and root complex capabilities allow
1371 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1372 it is recommended to locate both adapters on the same NUMA node.
1373 This in order to forward packets from one to the other without
1374 NUMA performance penalty.
1376 5. Disable pause frames::
1378 ethtool -A <netdev> rx off tx off
1380 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1381 via the BIOS configuration. Please contact you server provider for more
1382 information about the settings.
1386 On some machines, depends on the machine integrator, it is beneficial
1387 to set the PCI max read request parameter to 1K. This can be
1388 done in the following way:
1390 To query the read request size use::
1392 setpci -s <NIC PCI address> 68.w
1394 If the output is different than 3XXX, set it by::
1396 setpci -s <NIC PCI address> 68.w=3XXX
1398 The XXX can be different on different systems. Make sure to configure
1399 according to the setpci output.
1401 7. To minimize overhead of searching Memory Regions:
1403 - '--socket-mem' is recommended to pin memory by predictable amount.
1404 - Configure per-lcore cache when creating Mempools for packet buffer.
1405 - Refrain from dynamically allocating/freeing memory in run-time.
1410 There are multiple Rx burst functions with different advantages and limitations.
1412 .. table:: Rx burst functions
1414 +-------------------+------------------------+---------+-----------------+------+-------+
1415 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1416 | | | | || comp|| MTU |
1417 +===================+========================+=========+=================+======+=======+
1418 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1419 +-------------------+------------------------+---------+-----------------+------+-------+
1420 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1421 +-------------------+------------------------+---------+-----------------+------+-------+
1422 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1423 | || RxQs >= rxqs_min_mprq | | | | |
1424 +-------------------+------------------------+---------+-----------------+------+-------+
1425 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1426 | || mprq_en=1 | | | | |
1427 | || RxQs >= rxqs_min_mprq | | | | |
1428 +-------------------+------------------------+---------+-----------------+------+-------+
1430 .. _mlx5_offloads_support:
1432 Supported hardware offloads
1433 ---------------------------
1435 .. table:: Minimal SW/HW versions for queue offloads
1437 ============== ===== ===== ========= ===== ========== =============
1438 Offload DPDK Linux rdma-core OFED firmware hardware
1439 ============== ===== ===== ========= ===== ========== =============
1440 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1441 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1442 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1443 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1444 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1445 Buffer Split 20.11 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1446 ============== ===== ===== ========= ===== ========== =============
1448 .. table:: Minimal SW/HW versions for rte_flow offloads
1450 +-----------------------+-----------------+-----------------+
1451 | Offload | with E-Switch | with NIC |
1452 +=======================+=================+=================+
1453 | Count | | DPDK 19.05 | | DPDK 19.02 |
1454 | | | OFED 4.6 | | OFED 4.6 |
1455 | | | rdma-core 24 | | rdma-core 23 |
1456 | | | ConnectX-5 | | ConnectX-5 |
1457 +-----------------------+-----------------+-----------------+
1458 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1459 | | | OFED 4.6 | | OFED 4.5 |
1460 | | | rdma-core 24 | | rdma-core 23 |
1461 | | | ConnectX-5 | | ConnectX-4 |
1462 +-----------------------+-----------------+-----------------+
1463 | Queue / RSS | | | | DPDK 18.11 |
1464 | | | N/A | | OFED 4.5 |
1465 | | | | | rdma-core 23 |
1466 | | | | | ConnectX-4 |
1467 +-----------------------+-----------------+-----------------+
1468 | RSS shared action | | | | DPDK 20.11 |
1469 | | | N/A | | OFED 5.2 |
1470 | | | | | rdma-core 33 |
1471 | | | | | ConnectX-5 |
1472 +-----------------------+-----------------+-----------------+
1473 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1474 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1475 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1476 | | of_set_vlan_pcp / | | | | |
1477 | | of_set_vlan_vid) | | | | |
1478 +-----------------------+-----------------+-----------------+
1479 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1480 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1481 | | | rdma-core 24 | | rdma-core 23 |
1482 | | | ConnectX-5 | | ConnectX-5 |
1483 +-----------------------+-----------------+-----------------+
1484 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1485 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1486 | | | rdma-core 27 | | rdma-core 27 |
1487 | | | ConnectX-5 | | ConnectX-5 |
1488 +-----------------------+-----------------+-----------------+
1489 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1490 | | | OFED 5.1-2 | | OFED 5.1-2 |
1491 | | | rdma-core 32 | | N/A |
1492 | | | ConnectX-5 | | ConnectX-5 |
1493 +-----------------------+-----------------+-----------------+
1494 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1495 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1496 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1497 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1498 | | set_ipv6_dst / | | | | |
1499 | | set_tp_src / | | | | |
1500 | | set_tp_dst / | | | | |
1501 | | dec_ttl / | | | | |
1502 | | set_ttl / | | | | |
1503 | | set_mac_src / | | | | |
1504 | | set_mac_dst) | | | | |
1505 +-----------------------+-----------------+-----------------+
1506 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1507 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1508 | | | | rdma-core 24 | | rdma-core 24 |
1509 | | | | ConnectX-5 | | ConnectX-5 |
1510 +-----------------------+-----------------+-----------------+
1511 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1512 | | | OFED 4.7-1 | | OFED 4.7-1 |
1513 | | | rdma-core 24 | | N/A |
1514 | | | ConnectX-5 | | ConnectX-5 |
1515 +-----------------------+-----------------+-----------------+
1516 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1517 | | | OFED 4.6 | | OFED 4.5 |
1518 | | | rdma-core 24 | | rdma-core 23 |
1519 | | | ConnectX-5 | | ConnectX-4 |
1520 +-----------------------+-----------------+-----------------+
1521 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1522 | | | OFED 4.7-3 | | OFED 4.7-3 |
1523 | | | rdma-core 26 | | rdma-core 26 |
1524 | | | ConnectX-5 | | ConnectX-5 |
1525 +-----------------------+-----------------+-----------------+
1526 | Port ID | | DPDK 19.05 | | N/A |
1527 | | | OFED 4.7-1 | | N/A |
1528 | | | rdma-core 24 | | N/A |
1529 | | | ConnectX-5 | | N/A |
1530 +-----------------------+-----------------+-----------------+
1531 | Hairpin | | | | DPDK 19.11 |
1532 | | | N/A | | OFED 4.7-3 |
1533 | | | | | rdma-core 26 |
1534 | | | | | ConnectX-5 |
1535 +-----------------------+-----------------+-----------------+
1536 | 2-port Hairpin | | | | DPDK 20.11 |
1537 | | | N/A | | OFED 5.1-2 |
1539 | | | | | ConnectX-5 |
1540 +-----------------------+-----------------+-----------------+
1541 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1542 | | | OFED 4.7-3 | | OFED 4.7-3 |
1543 | | | rdma-core 26 | | rdma-core 26 |
1544 | | | ConnectX-5 | | ConnectX-5 |
1545 +-----------------------+-----------------+-----------------+
1546 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1547 | | | OFED 5.1-2 | | OFED 5.1-2 |
1548 | | | rdma-core 32 | | N/A |
1549 | | | ConnectX-5 | | ConnectX-5 |
1550 +-----------------------+-----------------+-----------------+
1551 | Age shared action | | DPDK 20.11 | | DPDK 20.11 |
1552 | | | OFED 5.2 | | OFED 5.2 |
1553 | | | rdma-core 32 | | rdma-core 32 |
1554 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1555 +-----------------------+-----------------+-----------------+
1556 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1557 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1558 | | | rdma-core 35 | | rdma-core 35 |
1559 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1560 +-----------------------+-----------------+-----------------+
1561 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1562 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1563 | | | rdma-core 34 | | rdma-core 34 |
1564 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1565 +-----------------------+-----------------+-----------------+
1570 MARK and META items are interrelated with datapath - they might move from/to
1571 the applications in mbuf fields. Hence, zero value for these items has the
1572 special meaning - it means "no metadata are provided", not zero values are
1573 treated by applications and PMD as valid ones.
1575 Moreover in the flow engine domain the value zero is acceptable to match and
1576 set, and we should allow to specify zero values as rte_flow parameters for the
1577 META and MARK items and actions. In the same time zero mask has no meaning and
1578 should be rejected on validation stage.
1583 Flows are not cached in the driver.
1584 When stopping a device port, all the flows created on this port from the
1585 application will be flushed automatically in the background.
1586 After stopping the device port, all flows on this port become invalid and
1587 not represented in the system.
1588 All references to these flows held by the application should be discarded
1589 directly but neither destroyed nor flushed.
1591 The application should re-create the flows as required after the port restart.
1596 Compared to librte_net_mlx4 that implements a single RSS configuration per
1597 port, librte_net_mlx5 supports per-protocol RSS configuration.
1599 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1600 command-line parameter to enable additional protocols (UDP and TCP as well
1601 as IP), the following commands must be entered from its CLI to get the same
1602 behavior as librte_net_mlx4::
1605 > port config all rss all
1611 This section demonstrates how to launch **testpmd** with Mellanox
1612 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1614 #. Load the kernel modules::
1616 modprobe -a ib_uverbs mlx5_core mlx5_ib
1618 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1621 /etc/init.d/openibd restart
1625 User space I/O kernel modules (uio and igb_uio) are not used and do
1626 not have to be loaded.
1628 #. Make sure Ethernet interfaces are in working order and linked to kernel
1629 verbs. Related sysfs entries should be present::
1631 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1640 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1643 for intf in eth2 eth3 eth4 eth5;
1645 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1648 sed -n 's,.*/\(.*\),-a \1,p'
1657 #. Request huge pages::
1659 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1661 #. Start testpmd with basic parameters::
1663 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
1668 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1669 EAL: probe driver: 15b3:1013 librte_net_mlx5
1670 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1671 PMD: librte_net_mlx5: 1 port(s) detected
1672 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1673 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1674 EAL: probe driver: 15b3:1013 librte_net_mlx5
1675 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1676 PMD: librte_net_mlx5: 1 port(s) detected
1677 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1678 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1679 EAL: probe driver: 15b3:1013 librte_net_mlx5
1680 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1681 PMD: librte_net_mlx5: 1 port(s) detected
1682 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1683 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1684 EAL: probe driver: 15b3:1013 librte_net_mlx5
1685 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1686 PMD: librte_net_mlx5: 1 port(s) detected
1687 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1688 Interactive-mode selected
1689 Configuring Port 0 (socket 0)
1690 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1691 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1692 Port 0: E4:1D:2D:E7:0C:FE
1693 Configuring Port 1 (socket 0)
1694 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1695 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1696 Port 1: E4:1D:2D:E7:0C:FF
1697 Configuring Port 2 (socket 0)
1698 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1699 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1700 Port 2: E4:1D:2D:E7:0C:FA
1701 Configuring Port 3 (socket 0)
1702 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1703 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1704 Port 3: E4:1D:2D:E7:0C:FB
1705 Checking link statuses...
1706 Port 0 Link Up - speed 40000 Mbps - full-duplex
1707 Port 1 Link Up - speed 40000 Mbps - full-duplex
1708 Port 2 Link Up - speed 10000 Mbps - full-duplex
1709 Port 3 Link Up - speed 10000 Mbps - full-duplex
1716 This section demonstrates how to dump flows. Currently, it's possible to dump
1717 all flows with assistance of external tools.
1719 #. 2 ways to get flow raw file:
1721 - Using testpmd CLI:
1723 .. code-block:: console
1725 testpmd> flow dump <port> <output_file>
1727 - call rte_flow_dev_dump api:
1729 .. code-block:: console
1731 rte_flow_dev_dump(port, file, NULL);
1733 #. Dump human-readable flows from raw file:
1735 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1737 .. code-block:: console
1739 mlx_steering_dump.py -f <output_file>