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.
104 - E-Switch mirroring and jump.
111 On Windows, the features are limited:
113 - Promiscuous mode is not supported
114 - The following rules are supported:
116 - IPv4/UDP with CVLAN filtering
117 - Unicast MAC filtering
119 - For secondary process:
121 - Forked secondary process not supported.
122 - External memory unregistered in EAL memseg list cannot be used for DMA
123 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
124 primary process and remapped to the same virtual address in secondary
125 process. If the external memory is registered by primary process but has
126 different virtual address in secondary process, unexpected error may happen.
128 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
129 specific VLAN will match for VLAN packets as well:
131 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
132 Meaning, the flow rule::
134 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
136 Will only match vlan packets with vid=3. and the flow rule::
138 flow create 0 ingress pattern eth / ipv4 / end ...
140 Will match any ipv4 packet (VLAN included).
142 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
144 - 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.
147 flow create 0 ingress pattern eth / ipv4 / end ...
149 Will match any ipv4 packet.
152 flow create 0 ingress pattern eth / vlan / end ...
153 flow create 0 ingress pattern eth has_vlan is 1 / end ...
154 flow create 0 ingress pattern eth type is 0x8100 / end ...
156 Will match single-tagged packets only, with any VLAN ID value.
159 flow create 0 ingress pattern eth type is 0x88A8 / end ...
160 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
162 Will match multi-tagged packets only, with any VLAN ID value.
164 - A flow pattern with 2 sequential VLAN items is not supported.
166 - VLAN pop offload command:
168 - Flow rules having a VLAN pop offload command as one of their actions and
169 are lacking a match on VLAN as one of their items are not supported.
170 - The command is not supported on egress traffic.
172 - VLAN push offload is not supported on ingress traffic.
174 - VLAN set PCP offload is not supported on existing headers.
176 - A multi segment packet must have not more segments than reported by dev_infos_get()
177 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
178 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
179 inline settings) to 58.
181 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
182 to 0 are not supported.
184 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
186 - Match on Geneve header supports the following fields only:
193 - Match on Geneve TLV option is supported on the following fields:
200 Only one Class/Type/Length Geneve TLV option is supported per shared device.
201 Class/Type/Length fields must be specified as well as masks.
202 Class/Type/Length specified masks must be full.
203 Matching Geneve TLV option without specifying data is not supported.
204 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
206 - VF: flow rules created on VF devices can only match traffic targeted at the
207 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
209 - Match on GTP tunnel header item supports the following fields only:
211 - v_pt_rsv_flags: E flag, S flag, PN flag
215 - Match on GTP extension header only for GTP PDU session container (next
216 extension header type = 0x85).
217 - Match on GTP extension header is not supported in group 0.
219 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
220 The flows within group 0 and set metadata action are rejected by hardware.
224 MAC addresses not already present in the bridge table of the associated
225 kernel network device will be added and cleaned up by the PMD when closing
226 the device. In case of ungraceful program termination, some entries may
227 remain present and should be removed manually by other means.
229 - Buffer split offload is supported with regular Rx burst routine only,
230 no MPRQ feature or vectorized code can be engaged.
232 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
233 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
234 ol_flags. As the mempool for the external buffer is managed by PMD, all the
235 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
236 the external buffers will be freed by PMD and the application which still
237 holds the external buffers may be corrupted.
239 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
240 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
241 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
243 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
244 and allmulticast mode are both set to off.
245 To receive IPv6 Multicast messages on VM, explicitly set the relevant
246 MAC address using rte_eth_dev_mac_addr_add() API.
248 - To support a mixed traffic pattern (some buffers from local host memory, some
249 buffers from other devices) with high bandwidth, a mbuf flag is used.
251 An application hints the PMD whether or not it should try to inline the
252 given mbuf data buffer. PMD should do the best effort to act upon this request.
254 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
255 registered by application with rte_mbuf_dynflag_register(). This flag is
256 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
257 which is intended to be used by the application.
259 To query the supported specific flags in runtime,
260 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
261 currently (over present hardware and configuration) supported specific flags.
262 The "not inline hint" feature operating flow is the following one:
265 - probe the devices, ports are created
266 - query the port capabilities
267 - if port supporting the feature is found
268 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
269 - application starts the ports
270 - on ``dev_start()`` PMD checks whether the feature flag is registered and
271 enables the feature support in datapath
272 - application might set the registered flag bit in ``ol_flags`` field
273 of mbuf being sent and PMD will handle ones appropriately.
275 - The amount of descriptors in Tx queue may be limited by data inline settings.
276 Inline data require the more descriptor building blocks and overall block
277 amount may exceed the hardware supported limits. The application should
278 reduce the requested Tx size or adjust data inline settings with
279 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
281 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
282 parameter should be specified.
283 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
284 being sent it tries to synchronize the time of packet appearing on
285 the wire with the specified packet timestamp. It the specified one
286 is in the past it should be ignored, if one is in the distant future
287 it should be capped with some reasonable value (in range of seconds).
288 These specific cases ("too late" and "distant future") can be optionally
289 reported via device xstats to assist applications to detect the
290 time-related problems.
292 The timestamp upper "too-distant-future" limit
293 at the moment of invoking the Tx burst routine
294 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
295 Please note, for the testpmd txonly mode,
296 the limit is deduced from the expression::
298 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
300 There is no any packet reordering according timestamps is supposed,
301 neither within packet burst, nor between packets, it is an entirely
302 application responsibility to generate packets and its timestamps
303 in desired order. The timestamps can be put only in the first packet
304 in the burst providing the entire burst scheduling.
306 - E-Switch decapsulation Flow:
308 - can be applied to PF port only.
309 - must specify VF port action (packet redirection from PF to VF).
310 - optionally may specify tunnel inner source and destination MAC addresses.
312 - E-Switch encapsulation Flow:
314 - can be applied to VF ports only.
315 - must specify PF port action (packet redirection from VF to PF).
319 - The input buffer, used as outer header, is not validated.
323 - The decapsulation is always done up to the outermost tunnel detected by the HW.
324 - The input buffer, providing the removal size, is not validated.
325 - The buffer size must match the length of the headers to be removed.
327 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
328 mutually exclusive features which cannot be supported together
329 (see :ref:`mlx5_firmware_config`).
333 - Requires DevX and DV flow to be enabled.
334 - KEEP_CRC offload cannot be supported with LRO.
335 - The first mbuf length, without head-room, must be big enough to include the
337 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
338 it with size limited to max LRO size, not to max RX packet length.
339 - LRO can be used with outer header of TCP packets of the standard format:
340 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
342 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
343 - LRO packet aggregation is performed by HW only for packet size larger than
344 ``lro_min_mss_size``. This value is reported on device start, when debug
349 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
350 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
351 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
355 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and E-Switch steering domain.
356 - The E-Switch Sample flow must have the eswitch_manager VPORT destination (PF or ECPF) and no additional actions.
357 - 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.
359 - IPv6 header item 'proto' field, indicating the next header protocol, should
360 not be set as extension header.
361 In case the next header is an extension header, it should not be specified in
362 IPv6 header item 'proto' field.
363 The last extension header item 'next header' field can specify the following
364 header protocol type.
368 - 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.
369 - Hairpin in switchdev SR-IOV mode is not supported till now.
374 MLX5 supports various methods to report statistics:
376 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.
378 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.
380 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.
388 The ibverbs libraries can be linked with this PMD in a number of ways,
389 configured by the ``ibverbs_link`` build option:
391 - ``shared`` (default): the PMD depends on some .so files.
393 - ``dlopen``: Split the dependencies glue in a separate library
394 loaded when needed by dlopen.
395 It make dependencies on libibverbs and libmlx4 optional,
396 and has no performance impact.
398 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
399 in the PMD shared library or the executable static binary.
401 Environment variables
402 ~~~~~~~~~~~~~~~~~~~~~
406 A list of directories in which to search for the rdma-core "glue" plug-in,
407 separated by colons or semi-colons.
409 - ``MLX5_SHUT_UP_BF``
411 Configures HW Tx doorbell register as IO-mapped.
413 By default, the HW Tx doorbell is configured as a write-combining register.
414 The register would be flushed to HW usually when the write-combining buffer
415 becomes full, but it depends on CPU design.
417 Except for vectorized Tx burst routines, a write memory barrier is enforced
418 after updating the register so that the update can be immediately visible to
421 When vectorized Tx burst is called, the barrier is set only if the burst size
422 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
423 variable will bring better latency even though the maximum throughput can
426 Run-time configuration
427 ~~~~~~~~~~~~~~~~~~~~~~
429 - librte_net_mlx5 brings kernel network interfaces up during initialization
430 because it is affected by their state. Forcing them down prevents packets
433 - **ethtool** operations on related kernel interfaces also affect the PMD.
438 In order to run as a non-root user,
439 some capabilities must be granted to the application::
441 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
443 Below are the reasons of the need for each capability:
446 When using physical addresses (PA mode), with Linux >= 4.0,
447 for access to ``/proc/self/pagemap``.
450 For device configuration.
453 For raw ethernet queue allocation through kernel driver.
456 For DMA memory pinning.
461 - ``rxq_cqe_comp_en`` parameter [int]
463 A nonzero value enables the compression of CQE on RX side. This feature
464 allows to save PCI bandwidth and improve performance. Enabled by default.
465 Different compression formats are supported in order to achieve the best
466 performance for different traffic patterns. Hash RSS format is the default.
468 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
469 better compression rate in case of RTE Flow Mark traffic.
470 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
471 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
472 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
476 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
477 ConnectX-6 Lx, BlueField and BlueField-2.
478 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
479 ConnectX-6 Lx, BlueField and BlueField-2.
481 - ``rxq_pkt_pad_en`` parameter [int]
483 A nonzero value enables padding Rx packet to the size of cacheline on PCI
484 transaction. This feature would waste PCI bandwidth but could improve
485 performance by avoiding partial cacheline write which may cause costly
486 read-modify-copy in memory transaction on some architectures. Disabled by
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 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
494 ConnectX-6 Lx, BlueField and BlueField-2.
496 - ``mprq_en`` parameter [int]
498 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
499 configured as Multi-Packet RQ if the total number of Rx queues is
500 ``rxqs_min_mprq`` or more. Disabled by default.
502 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
503 by posting a single large buffer for multiple packets. Instead of posting a
504 buffers per a packet, one large buffer is posted in order to receive multiple
505 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
506 and each stride receives one packet. MPRQ can improve throughput for
507 small-packet traffic.
509 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
510 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
511 configure large stride size enough to accommodate max_rx_pkt_len as long as
512 device allows. Note that this can waste system memory compared to enabling Rx
513 scatter and multi-segment packet.
515 - ``mprq_log_stride_num`` parameter [int]
517 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
518 strides can reduce PCIe traffic further. If configured value is not in the
519 range of device capability, the default value will be set with a warning
520 message. The default value is 4 which is 16 strides per a buffer, valid only
521 if ``mprq_en`` is set.
523 The size of Rx queue should be bigger than the number of strides.
525 - ``mprq_log_stride_size`` parameter [int]
527 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
528 stride size can save some memory and reduce probability of a depletion of all
529 available strides due to unreleased packets by an application. If configured
530 value is not in the range of device capability, the default value will be set
531 with a warning message. The default value is 11 which is 2048 bytes per a
532 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
533 it is possible for a packet to span across multiple strides. This mode allows
534 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
535 of a packet if Rx scatter is configured) may be required in case there is no
536 space left for a head room at the end of a stride which incurs some
539 - ``mprq_max_memcpy_len`` parameter [int]
541 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
542 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
543 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
544 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
545 A mempool for external buffers will be allocated and managed by PMD. If Rx
546 packet is externally attached, ol_flags field of the mbuf will have
547 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
548 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
550 - ``rxqs_min_mprq`` parameter [int]
552 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
553 greater or equal to this value. The default value is 12, valid only if
556 - ``txq_inline`` parameter [int]
558 Amount of data to be inlined during TX operations. This parameter is
559 deprecated and converted to the new parameter ``txq_inline_max`` providing
560 partial compatibility.
562 - ``txqs_min_inline`` parameter [int]
564 Enable inline data send only when the number of TX queues is greater or equal
567 This option should be used in combination with ``txq_inline_max`` and
568 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
570 If this option is not specified the default value 16 is used for BlueField
571 and 8 for other platforms
573 The data inlining consumes the CPU cycles, so this option is intended to
574 auto enable inline data if we have enough Tx queues, which means we have
575 enough CPU cores and PCI bandwidth is getting more critical and CPU
576 is not supposed to be bottleneck anymore.
578 The copying data into WQE improves latency and can improve PPS performance
579 when PCI back pressure is detected and may be useful for scenarios involving
580 heavy traffic on many queues.
582 Because additional software logic is necessary to handle this mode, this
583 option should be used with care, as it may lower performance when back
584 pressure is not expected.
586 If inline data are enabled it may affect the maximal size of Tx queue in
587 descriptors because the inline data increase the descriptor size and
588 queue size limits supported by hardware may be exceeded.
590 - ``txq_inline_min`` parameter [int]
592 Minimal amount of data to be inlined into WQE during Tx operations. NICs
593 may require this minimal data amount to operate correctly. The exact value
594 may depend on NIC operation mode, requested offloads, etc. It is strongly
595 recommended to omit this parameter and use the default values. Anyway,
596 applications using this parameter should take into consideration that
597 specifying an inconsistent value may prevent the NIC from sending packets.
599 If ``txq_inline_min`` key is present the specified value (may be aligned
600 by the driver in order not to exceed the limits and provide better descriptor
601 space utilization) will be used by the driver and it is guaranteed that
602 requested amount of data bytes are inlined into the WQE beside other inline
603 settings. This key also may update ``txq_inline_max`` value (default
604 or specified explicitly in devargs) to reserve the space for inline data.
606 If ``txq_inline_min`` key is not present, the value may be queried by the
607 driver from the NIC via DevX if this feature is available. If there is no DevX
608 enabled/supported the value 18 (supposing L2 header including VLAN) is set
609 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
610 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
613 For ConnectX-4 NIC, driver does not allow specifying value below 18
614 (minimal L2 header, including VLAN), error will be raised.
616 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
617 it is not recommended and may prevent NIC from sending packets over
620 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
621 Multi-Packet Write), because last one does not support partial packet inlining.
622 This is not very critical due to minimal data inlining is mostly required
623 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
625 - ``txq_inline_max`` parameter [int]
627 Specifies the maximal packet length to be completely inlined into WQE
628 Ethernet Segment for ordinary SEND method. If packet is larger than specified
629 value, the packet data won't be copied by the driver at all, data buffer
630 is addressed with a pointer. If packet length is less or equal all packet
631 data will be copied into WQE. This may improve PCI bandwidth utilization for
632 short packets significantly but requires the extra CPU cycles.
634 The data inline feature is controlled by number of Tx queues, if number of Tx
635 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
636 is engaged, if there are not enough Tx queues (which means not enough CPU cores
637 and CPU resources are scarce), data inline is not performed by the driver.
638 Assigning ``txqs_min_inline`` with zero always enables the data inline.
640 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
641 by the driver in order not to exceed the limit (930 bytes) and to provide better
642 WQE space filling without gaps, the adjustment is reflected in the debug log.
643 Also, the default value (290) may be decreased in run-time if the large transmit
644 queue size is requested and hardware does not support enough descriptor
645 amount, in this case warning is emitted. If ``txq_inline_max`` key is
646 specified and requested inline settings can not be satisfied then error
649 - ``txq_inline_mpw`` parameter [int]
651 Specifies the maximal packet length to be completely inlined into WQE for
652 Enhanced MPW method. If packet is large the specified value, the packet data
653 won't be copied, and data buffer is addressed with pointer. If packet length
654 is less or equal, all packet data will be copied into WQE. This may improve PCI
655 bandwidth utilization for short packets significantly but requires the extra
658 The data inline feature is controlled by number of TX queues, if number of Tx
659 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
660 is engaged, if there are not enough Tx queues (which means not enough CPU cores
661 and CPU resources are scarce), data inline is not performed by the driver.
662 Assigning ``txqs_min_inline`` with zero always enables the data inline.
664 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
665 by the driver in order not to exceed the limit (930 bytes) and to provide better
666 WQE space filling without gaps, the adjustment is reflected in the debug log.
667 Due to multiple packets may be included to the same WQE with Enhanced Multi
668 Packet Write Method and overall WQE size is limited it is not recommended to
669 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
670 may be decreased in run-time if the large transmit queue size is requested
671 and hardware does not support enough descriptor amount, in this case warning
672 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
673 settings can not be satisfied then error will be raised.
675 - ``txqs_max_vec`` parameter [int]
677 Enable vectorized Tx only when the number of TX queues is less than or
678 equal to this value. This parameter is deprecated and ignored, kept
679 for compatibility issue to not prevent driver from probing.
681 - ``txq_mpw_hdr_dseg_en`` parameter [int]
683 A nonzero value enables including two pointers in the first block of TX
684 descriptor. The parameter is deprecated and ignored, kept for compatibility
687 - ``txq_max_inline_len`` parameter [int]
689 Maximum size of packet to be inlined. This limits the size of packet to
690 be inlined. If the size of a packet is larger than configured value, the
691 packet isn't inlined even though there's enough space remained in the
692 descriptor. Instead, the packet is included with pointer. This parameter
693 is deprecated and converted directly to ``txq_inline_mpw`` providing full
694 compatibility. Valid only if eMPW feature is engaged.
696 - ``txq_mpw_en`` parameter [int]
698 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
699 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
700 eMPW allows the Tx burst function to pack up multiple packets
701 in a single descriptor session in order to save PCI bandwidth
702 and improve performance at the cost of a slightly higher CPU usage.
703 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
704 Tx burst function copies entire packet data on to Tx descriptor
705 instead of including pointer of packet.
707 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
708 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
709 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
710 option or reported by the NIC, the eMPW feature is disengaged.
712 - ``tx_db_nc`` parameter [int]
714 The rdma core library can map doorbell register in two ways, depending on the
715 environment variable "MLX5_SHUT_UP_BF":
717 - As regular cached memory (usually with write combining attribute), if the
718 variable is either missing or set to zero.
719 - As non-cached memory, if the variable is present and set to not "0" value.
721 The type of mapping may slightly affect the Tx performance, the optimal choice
722 is strongly relied on the host architecture and should be deduced practically.
724 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
725 memory (with write combining), the PMD will perform the extra write memory barrier
726 after writing to doorbell, it might increase the needed CPU clocks per packet
727 to send, but latency might be improved.
729 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
730 cached memory, the PMD will not perform the extra write memory barrier
731 after writing to doorbell, on some architectures it might improve the
734 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
735 memory, the PMD will use heuristics to decide whether write memory barrier
736 should be performed. For bursts with size multiple of recommended one (64 pkts)
737 it is supposed the next burst is coming and no need to issue the extra memory
738 barrier (it is supposed to be issued in the next coming burst, at least after
739 descriptor writing). It might increase latency (on some hosts till next
740 packets transmit) and should be used with care.
742 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
743 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
744 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
746 - ``tx_pp`` parameter [int]
748 If a nonzero value is specified the driver creates all necessary internal
749 objects to provide accurate packet send scheduling on mbuf timestamps.
750 The positive value specifies the scheduling granularity in nanoseconds,
751 the packet send will be accurate up to specified digits. The allowed range is
752 from 500 to 1 million of nanoseconds. The negative value specifies the module
753 of granularity and engages the special test mode the check the schedule rate.
754 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
757 - ``tx_skew`` parameter [int]
759 The parameter adjusts the send packet scheduling on timestamps and represents
760 the average delay between beginning of the transmitting descriptor processing
761 by the hardware and appearance of actual packet data on the wire. The value
762 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
763 specified. The default value is zero.
765 - ``tx_vec_en`` parameter [int]
767 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
768 ConnectX-6 Lx, BlueField and BlueField-2 NICs
769 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
770 The parameter is deprecated and ignored.
772 - ``rx_vec_en`` parameter [int]
774 A nonzero value enables Rx vector if the port is not configured in
775 multi-segment otherwise this parameter is ignored.
779 - ``vf_nl_en`` parameter [int]
781 A nonzero value enables Netlink requests from the VF to add/remove MAC
782 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
783 Otherwise the relevant configuration must be run with Linux iproute2 tools.
784 This is a prerequisite to receive this kind of traffic.
786 Enabled by default, valid only on VF devices ignored otherwise.
788 - ``l3_vxlan_en`` parameter [int]
790 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
791 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
792 parameter. This is a prerequisite to receive this kind of traffic.
796 - ``dv_xmeta_en`` parameter [int]
798 A nonzero value enables extensive flow metadata support if device is
799 capable and driver supports it. This can enable extensive support of
800 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
801 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
803 There are some possible configurations, depending on parameter value:
805 - 0, this is default value, defines the legacy mode, the ``MARK`` and
806 ``META`` related actions and items operate only within NIC Tx and
807 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
808 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
809 item is 32 bits wide and match supported on egress only.
811 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
812 related actions and items operate within all supported steering domains,
813 including FDB, ``MARK`` and ``META`` information may cross the domain
814 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
815 depends on kernel and firmware configurations and might be 0, 16 or
816 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
817 compatibility, the actual width of data transferred to the FDB domain
818 depends on kernel configuration and may be vary. The actual supported
819 width can be retrieved in runtime by series of rte_flow_validate()
822 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
823 related actions and items operate within all supported steering domains,
824 including FDB, ``MARK`` and ``META`` information may cross the domain
825 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
826 depends on kernel and firmware configurations and might be 0, 16 or
827 24 bits. The actual supported width can be retrieved in runtime by
828 series of rte_flow_validate() trials.
830 - 3, this engages tunnel offload mode. In E-Switch configuration, that
831 mode implicitly activates ``dv_xmeta_en=1``.
833 +------+-----------+-----------+-------------+-------------+
834 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
835 +======+===========+===========+=============+=============+
836 | 0 | 24 bits | 32 bits | 32 bits | no |
837 +------+-----------+-----------+-------------+-------------+
838 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
839 +------+-----------+-----------+-------------+-------------+
840 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
841 +------+-----------+-----------+-------------+-------------+
843 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
844 ignored and the device is configured to operate in legacy mode (0).
846 Disabled by default (set to 0).
848 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
849 of the extensive metadata features. The legacy Verbs supports FLAG and
850 MARK metadata actions over NIC Rx steering domain only.
852 - ``dv_flow_en`` parameter [int]
854 A nonzero value enables the DV flow steering assuming it is supported
855 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
857 Enabled by default if supported.
859 - ``dv_esw_en`` parameter [int]
861 A nonzero value enables E-Switch using Direct Rules.
863 Enabled by default if supported.
865 - ``lacp_by_user`` parameter [int]
867 A nonzero value enables the control of LACP traffic by the user application.
868 When a bond exists in the driver, by default it should be managed by the
869 kernel and therefore LACP traffic should be steered to the kernel.
870 If this devarg is set to 1 it will allow the user to manage the bond by
871 itself and not steer LACP traffic to the kernel.
873 Disabled by default (set to 0).
875 - ``mr_ext_memseg_en`` parameter [int]
877 A nonzero value enables extending memseg when registering DMA memory. If
878 enabled, the number of entries in MR (Memory Region) lookup table on datapath
879 is minimized and it benefits performance. On the other hand, it worsens memory
880 utilization because registered memory is pinned by kernel driver. Even if a
881 page in the extended chunk is freed, that doesn't become reusable until the
882 entire memory is freed.
886 - ``representor`` parameter [list]
888 This parameter can be used to instantiate DPDK Ethernet devices from
889 existing port (or VF) representors configured on the device.
891 It is a standard parameter whose format is described in
892 :ref:`ethernet_device_standard_device_arguments`.
894 For instance, to probe port representors 0 through 2::
898 - ``max_dump_files_num`` parameter [int]
900 The maximum number of files per PMD entity that may be created for debug information.
901 The files will be created in /var/log directory or in current directory.
903 set to 128 by default.
905 - ``lro_timeout_usec`` parameter [int]
907 The maximum allowed duration of an LRO session, in micro-seconds.
908 PMD will set the nearest value supported by HW, which is not bigger than
909 the input ``lro_timeout_usec`` value.
910 If this parameter is not specified, by default PMD will set
911 the smallest value supported by HW.
913 - ``hp_buf_log_sz`` parameter [int]
915 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
916 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
917 The capacity of the value is specified by the firmware and the initialization
918 will get a failure if it is out of scope.
919 The range of the value is from 11 to 19 right now, and the supported frame
920 size of a single packet for hairpin is from 512B to 128KB. It might change if
921 different firmware release is being used. By using a small value, it could
922 reduce memory consumption but not work with a large frame. If the value is
923 too large, the memory consumption will be high and some potential performance
924 degradation will be introduced.
925 By default, the PMD will set this value to 16, which means that 9KB jumbo
926 frames will be supported.
928 - ``reclaim_mem_mode`` parameter [int]
930 Cache some resources in flow destroy will help flow recreation more efficient.
931 While some systems may require the all the resources can be reclaimed after
933 The parameter ``reclaim_mem_mode`` provides the option for user to configure
934 if the resource cache is needed or not.
936 There are three options to choose:
938 - 0. It means the flow resources will be cached as usual. The resources will
939 be cached, helpful with flow insertion rate.
941 - 1. It will only enable the DPDK PMD level resources reclaim.
943 - 2. Both DPDK PMD level and rdma-core low level will be configured as
946 By default, the PMD will set this value to 0.
948 - ``sys_mem_en`` parameter [int]
950 A non-zero value enables the PMD memory management allocating memory
951 from system by default, without explicit rte memory flag.
953 By default, the PMD will set this value to 0.
955 - ``decap_en`` parameter [int]
957 Some devices do not support FCS (frame checksum) scattering for
958 tunnel-decapsulated packets.
959 If set to 0, this option forces the FCS feature and rejects tunnel
960 decapsulation in the flow engine for such devices.
962 By default, the PMD will set this value to 1.
964 .. _mlx5_firmware_config:
966 Firmware configuration
967 ~~~~~~~~~~~~~~~~~~~~~~
969 Firmware features can be configured as key/value pairs.
971 The command to set a value is::
973 mlxconfig -d <device> set <key>=<value>
975 The command to query a value is::
977 mlxconfig -d <device> query | grep <key>
979 The device name for the command ``mlxconfig`` can be either the PCI address,
980 or the mst device name found with::
984 Below are some firmware configurations listed.
990 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
996 - maximum number of SR-IOV virtual functions::
1000 - enable DevX (required by Direct Rules and other features)::
1004 - aggressive CQE zipping::
1008 - L3 VXLAN and VXLAN-GPE destination UDP port::
1011 IP_OVER_VXLAN_PORT=<udp dport>
1013 - enable VXLAN-GPE tunnel flow matching::
1015 FLEX_PARSER_PROFILE_ENABLE=0
1017 FLEX_PARSER_PROFILE_ENABLE=2
1019 - enable IP-in-IP tunnel flow matching::
1021 FLEX_PARSER_PROFILE_ENABLE=0
1023 - enable MPLS flow matching::
1025 FLEX_PARSER_PROFILE_ENABLE=1
1027 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1029 FLEX_PARSER_PROFILE_ENABLE=2
1031 - enable Geneve flow matching::
1033 FLEX_PARSER_PROFILE_ENABLE=0
1035 FLEX_PARSER_PROFILE_ENABLE=1
1037 - enable Geneve TLV option flow matching::
1039 FLEX_PARSER_PROFILE_ENABLE=0
1041 - enable GTP flow matching::
1043 FLEX_PARSER_PROFILE_ENABLE=3
1045 - enable eCPRI flow matching::
1047 FLEX_PARSER_PROFILE_ENABLE=4
1053 This driver relies on external libraries and kernel drivers for resources
1054 allocations and initialization. The following dependencies are not part of
1055 DPDK and must be installed separately:
1059 User space Verbs framework used by librte_net_mlx5. This library provides
1060 a generic interface between the kernel and low-level user space drivers
1063 It allows slow and privileged operations (context initialization, hardware
1064 resources allocations) to be managed by the kernel and fast operations to
1065 never leave user space.
1069 Low-level user space driver library for Mellanox
1070 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1073 This library basically implements send/receive calls to the hardware
1076 - **Kernel modules**
1078 They provide the kernel-side Verbs API and low level device drivers that
1079 manage actual hardware initialization and resources sharing with user
1082 Unlike most other PMDs, these modules must remain loaded and bound to
1085 - mlx5_core: hardware driver managing Mellanox
1086 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1088 - mlx5_ib: InifiniBand device driver.
1089 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1091 - **Firmware update**
1093 Mellanox OFED/EN releases include firmware updates for
1094 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1096 Because each release provides new features, these updates must be applied to
1097 match the kernel modules and libraries they come with.
1101 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1107 Either RDMA Core library with a recent enough Linux kernel release
1108 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1111 RDMA Core with Linux Kernel
1112 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1114 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1115 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1116 (see `RDMA Core installation documentation`_)
1117 - When building for i686 use:
1119 - rdma-core version 18.0 or above built with 32bit support.
1120 - Kernel version 4.14.41 or above.
1122 - Starting with rdma-core v21, static libraries can be built::
1125 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1128 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1129 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1135 - Mellanox OFED version: **4.5** and above /
1136 Mellanox EN version: **4.5** and above
1139 - ConnectX-4: **12.21.1000** and above.
1140 - ConnectX-4 Lx: **14.21.1000** and above.
1141 - ConnectX-5: **16.21.1000** and above.
1142 - ConnectX-5 Ex: **16.21.1000** and above.
1143 - ConnectX-6: **20.27.0090** and above.
1144 - ConnectX-6 Dx: **22.27.0090** and above.
1145 - BlueField: **18.25.1010** and above.
1147 While these libraries and kernel modules are available on OpenFabrics
1148 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1149 managers on most distributions, this PMD requires Ethernet extensions that
1150 may not be supported at the moment (this is a work in progress).
1153 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1155 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1156 include the necessary support and should be used in the meantime. For DPDK,
1157 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1158 required from that distribution.
1162 Several versions of Mellanox OFED/EN are available. Installing the version
1163 this DPDK release was developed and tested against is strongly
1164 recommended. Please check the `linux prerequisites`_.
1166 Windows Prerequisites
1167 ---------------------
1169 This driver relies on external libraries and kernel drivers for resources
1170 allocations and initialization. The dependencies in the following sub-sections
1171 are not part of DPDK, and must be installed separately.
1173 Compilation Prerequisites
1174 ~~~~~~~~~~~~~~~~~~~~~~~~~
1176 DevX SDK installation
1177 ^^^^^^^^^^^^^^^^^^^^^
1179 The DevX SDK must be installed on the machine building the Windows PMD.
1180 Additional information can be found at
1181 `How to Integrate Windows DevX in Your Development Environment
1182 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1184 Runtime Prerequisites
1185 ~~~~~~~~~~~~~~~~~~~~~
1187 WinOF2 version 2.60 or higher must be installed on the machine.
1192 The driver can be downloaded from the following site:
1194 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1199 DevX for Windows must be enabled in the Windows registry.
1200 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1201 Additional information can be found in the WinOF2 user manual.
1206 The following Mellanox device families are supported by the same mlx5 driver:
1218 Below are detailed device names:
1220 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1221 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1222 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1223 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1224 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1225 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1226 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1227 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1228 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1229 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1230 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1231 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1232 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1233 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1234 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1235 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1236 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1237 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1238 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1239 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1240 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1241 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1242 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1243 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1244 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1245 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1247 Quick Start Guide on OFED/EN
1248 ----------------------------
1250 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1253 2. Install the required libraries and kernel modules either by installing
1254 only the required set, or by installing the entire Mellanox OFED/EN::
1256 ./mlnxofedinstall --upstream-libs --dpdk
1258 3. Verify the firmware is the correct one::
1262 4. Verify all ports links are set to Ethernet::
1264 mlxconfig -d <mst device> query | grep LINK_TYPE
1268 Link types may have to be configured to Ethernet::
1270 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1272 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1274 For hypervisors, verify SR-IOV is enabled on the NIC::
1276 mlxconfig -d <mst device> query | grep SRIOV_EN
1279 If needed, configure SR-IOV::
1281 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1282 mlxfwreset -d <mst device> reset
1284 5. Restart the driver::
1286 /etc/init.d/openibd restart
1290 service openibd restart
1292 If link type was changed, firmware must be reset as well::
1294 mlxfwreset -d <mst device> reset
1296 For hypervisors, after reset write the sysfs number of virtual functions
1299 To dynamically instantiate a given number of virtual functions (VFs)::
1301 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1303 6. Install DPDK and you are ready to go.
1304 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1306 Enable switchdev mode
1307 ---------------------
1309 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1310 Representor is a port in DPDK that is connected to a VF in such a way
1311 that assuming there are no offload flows, each packet that is sent from the VF
1312 will be received by the corresponding representor. While each packet that is
1313 sent to a representor will be received by the VF.
1314 This is very useful in case of SRIOV mode, where the first packet that is sent
1315 by the VF will be received by the DPDK application which will decide if this
1316 flow should be offloaded to the E-Switch. After offloading the flow packet
1317 that the VF that are matching the flow will not be received any more by
1318 the DPDK application.
1320 1. Enable SRIOV mode::
1322 mlxconfig -d <mst device> set SRIOV_EN=true
1324 2. Configure the max number of VFs::
1326 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1330 mlxfwreset -d <mst device> reset
1332 3. Configure the actual number of VFs::
1334 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1336 4. Unbind the device (can be rebind after the switchdev mode)::
1338 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1340 5. Enbale switchdev mode::
1342 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1347 1. Configure aggressive CQE Zipping for maximum performance::
1349 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1351 To set it back to the default CQE Zipping mode use::
1353 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1355 2. In case of virtualization:
1357 - Make sure that hypervisor kernel is 3.16 or newer.
1358 - Configure boot with ``iommu=pt``.
1359 - Use 1G huge pages.
1360 - Make sure to allocate a VM on huge pages.
1361 - Make sure to set CPU pinning.
1363 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1364 for better performance. For VMs, verify that the right CPU
1365 and NUMA node are pinned according to the above. Run::
1369 to identify the NUMA node to which the PCIe adapter is connected.
1371 4. If more than one adapter is used, and root complex capabilities allow
1372 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1373 it is recommended to locate both adapters on the same NUMA node.
1374 This in order to forward packets from one to the other without
1375 NUMA performance penalty.
1377 5. Disable pause frames::
1379 ethtool -A <netdev> rx off tx off
1381 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1382 via the BIOS configuration. Please contact you server provider for more
1383 information about the settings.
1387 On some machines, depends on the machine integrator, it is beneficial
1388 to set the PCI max read request parameter to 1K. This can be
1389 done in the following way:
1391 To query the read request size use::
1393 setpci -s <NIC PCI address> 68.w
1395 If the output is different than 3XXX, set it by::
1397 setpci -s <NIC PCI address> 68.w=3XXX
1399 The XXX can be different on different systems. Make sure to configure
1400 according to the setpci output.
1402 7. To minimize overhead of searching Memory Regions:
1404 - '--socket-mem' is recommended to pin memory by predictable amount.
1405 - Configure per-lcore cache when creating Mempools for packet buffer.
1406 - Refrain from dynamically allocating/freeing memory in run-time.
1411 There are multiple Rx burst functions with different advantages and limitations.
1413 .. table:: Rx burst functions
1415 +-------------------+------------------------+---------+-----------------+------+-------+
1416 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1417 | | | | || comp|| MTU |
1418 +===================+========================+=========+=================+======+=======+
1419 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1420 +-------------------+------------------------+---------+-----------------+------+-------+
1421 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1422 +-------------------+------------------------+---------+-----------------+------+-------+
1423 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1424 | || RxQs >= rxqs_min_mprq | | | | |
1425 +-------------------+------------------------+---------+-----------------+------+-------+
1426 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1427 | || mprq_en=1 | | | | |
1428 | || RxQs >= rxqs_min_mprq | | | | |
1429 +-------------------+------------------------+---------+-----------------+------+-------+
1431 .. _mlx5_offloads_support:
1433 Supported hardware offloads
1434 ---------------------------
1436 .. table:: Minimal SW/HW versions for queue offloads
1438 ============== ===== ===== ========= ===== ========== =============
1439 Offload DPDK Linux rdma-core OFED firmware hardware
1440 ============== ===== ===== ========= ===== ========== =============
1441 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1442 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1443 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1444 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1445 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1446 Buffer Split 20.11 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1447 ============== ===== ===== ========= ===== ========== =============
1449 .. table:: Minimal SW/HW versions for rte_flow offloads
1451 +-----------------------+-----------------+-----------------+
1452 | Offload | with E-Switch | with NIC |
1453 +=======================+=================+=================+
1454 | Count | | DPDK 19.05 | | DPDK 19.02 |
1455 | | | OFED 4.6 | | OFED 4.6 |
1456 | | | rdma-core 24 | | rdma-core 23 |
1457 | | | ConnectX-5 | | ConnectX-5 |
1458 +-----------------------+-----------------+-----------------+
1459 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1460 | | | OFED 4.6 | | OFED 4.5 |
1461 | | | rdma-core 24 | | rdma-core 23 |
1462 | | | ConnectX-5 | | ConnectX-4 |
1463 +-----------------------+-----------------+-----------------+
1464 | Queue / RSS | | | | DPDK 18.11 |
1465 | | | N/A | | OFED 4.5 |
1466 | | | | | rdma-core 23 |
1467 | | | | | ConnectX-4 |
1468 +-----------------------+-----------------+-----------------+
1469 | RSS shared action | | | | DPDK 20.11 |
1470 | | | N/A | | OFED 5.2 |
1471 | | | | | rdma-core 33 |
1472 | | | | | ConnectX-5 |
1473 +-----------------------+-----------------+-----------------+
1474 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1475 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1476 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1477 | | of_set_vlan_pcp / | | | | |
1478 | | of_set_vlan_vid) | | | | |
1479 +-----------------------+-----------------+-----------------+
1480 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1481 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1482 | | | rdma-core 24 | | rdma-core 23 |
1483 | | | ConnectX-5 | | ConnectX-5 |
1484 +-----------------------+-----------------+-----------------+
1485 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1486 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1487 | | | rdma-core 27 | | rdma-core 27 |
1488 | | | ConnectX-5 | | ConnectX-5 |
1489 +-----------------------+-----------------+-----------------+
1490 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1491 | | | OFED 5.1-2 | | OFED 5.1-2 |
1492 | | | rdma-core 32 | | N/A |
1493 | | | ConnectX-5 | | ConnectX-5 |
1494 +-----------------------+-----------------+-----------------+
1495 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1496 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1497 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1498 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1499 | | set_ipv6_dst / | | | | |
1500 | | set_tp_src / | | | | |
1501 | | set_tp_dst / | | | | |
1502 | | dec_ttl / | | | | |
1503 | | set_ttl / | | | | |
1504 | | set_mac_src / | | | | |
1505 | | set_mac_dst) | | | | |
1506 +-----------------------+-----------------+-----------------+
1507 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1508 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1509 | | | | rdma-core 24 | | rdma-core 24 |
1510 | | | | ConnectX-5 | | ConnectX-5 |
1511 +-----------------------+-----------------+-----------------+
1512 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1513 | | | OFED 4.7-1 | | OFED 4.7-1 |
1514 | | | rdma-core 24 | | N/A |
1515 | | | ConnectX-5 | | ConnectX-5 |
1516 +-----------------------+-----------------+-----------------+
1517 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1518 | | | OFED 4.6 | | OFED 4.5 |
1519 | | | rdma-core 24 | | rdma-core 23 |
1520 | | | ConnectX-5 | | ConnectX-4 |
1521 +-----------------------+-----------------+-----------------+
1522 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1523 | | | OFED 4.7-3 | | OFED 4.7-3 |
1524 | | | rdma-core 26 | | rdma-core 26 |
1525 | | | ConnectX-5 | | ConnectX-5 |
1526 +-----------------------+-----------------+-----------------+
1527 | Port ID | | DPDK 19.05 | | N/A |
1528 | | | OFED 4.7-1 | | N/A |
1529 | | | rdma-core 24 | | N/A |
1530 | | | ConnectX-5 | | N/A |
1531 +-----------------------+-----------------+-----------------+
1532 | Hairpin | | | | DPDK 19.11 |
1533 | | | N/A | | OFED 4.7-3 |
1534 | | | | | rdma-core 26 |
1535 | | | | | ConnectX-5 |
1536 +-----------------------+-----------------+-----------------+
1537 | 2-port Hairpin | | | | DPDK 20.11 |
1538 | | | N/A | | OFED 5.1-2 |
1540 | | | | | ConnectX-5 |
1541 +-----------------------+-----------------+-----------------+
1542 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1543 | | | OFED 4.7-3 | | OFED 4.7-3 |
1544 | | | rdma-core 26 | | rdma-core 26 |
1545 | | | ConnectX-5 | | ConnectX-5 |
1546 +-----------------------+-----------------+-----------------+
1547 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1548 | | | OFED 5.1-2 | | OFED 5.1-2 |
1549 | | | rdma-core 32 | | N/A |
1550 | | | ConnectX-5 | | ConnectX-5 |
1551 +-----------------------+-----------------+-----------------+
1552 | Age shared action | | DPDK 20.11 | | DPDK 20.11 |
1553 | | | OFED 5.2 | | OFED 5.2 |
1554 | | | rdma-core 32 | | rdma-core 32 |
1555 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1556 +-----------------------+-----------------+-----------------+
1557 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1558 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1559 | | | rdma-core 35 | | rdma-core 35 |
1560 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1561 +-----------------------+-----------------+-----------------+
1562 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1563 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1564 | | | rdma-core 34 | | rdma-core 34 |
1565 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1566 +-----------------------+-----------------+-----------------+
1571 MARK and META items are interrelated with datapath - they might move from/to
1572 the applications in mbuf fields. Hence, zero value for these items has the
1573 special meaning - it means "no metadata are provided", not zero values are
1574 treated by applications and PMD as valid ones.
1576 Moreover in the flow engine domain the value zero is acceptable to match and
1577 set, and we should allow to specify zero values as rte_flow parameters for the
1578 META and MARK items and actions. In the same time zero mask has no meaning and
1579 should be rejected on validation stage.
1584 Flows are not cached in the driver.
1585 When stopping a device port, all the flows created on this port from the
1586 application will be flushed automatically in the background.
1587 After stopping the device port, all flows on this port become invalid and
1588 not represented in the system.
1589 All references to these flows held by the application should be discarded
1590 directly but neither destroyed nor flushed.
1592 The application should re-create the flows as required after the port restart.
1597 Compared to librte_net_mlx4 that implements a single RSS configuration per
1598 port, librte_net_mlx5 supports per-protocol RSS configuration.
1600 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1601 command-line parameter to enable additional protocols (UDP and TCP as well
1602 as IP), the following commands must be entered from its CLI to get the same
1603 behavior as librte_net_mlx4::
1606 > port config all rss all
1612 This section demonstrates how to launch **testpmd** with Mellanox
1613 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1615 #. Load the kernel modules::
1617 modprobe -a ib_uverbs mlx5_core mlx5_ib
1619 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1622 /etc/init.d/openibd restart
1626 User space I/O kernel modules (uio and igb_uio) are not used and do
1627 not have to be loaded.
1629 #. Make sure Ethernet interfaces are in working order and linked to kernel
1630 verbs. Related sysfs entries should be present::
1632 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1641 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1644 for intf in eth2 eth3 eth4 eth5;
1646 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1649 sed -n 's,.*/\(.*\),-a \1,p'
1658 #. Request huge pages::
1660 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1662 #. Start testpmd with basic parameters::
1664 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
1669 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1670 EAL: probe driver: 15b3:1013 librte_net_mlx5
1671 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1672 PMD: librte_net_mlx5: 1 port(s) detected
1673 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1674 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1675 EAL: probe driver: 15b3:1013 librte_net_mlx5
1676 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1677 PMD: librte_net_mlx5: 1 port(s) detected
1678 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1679 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1680 EAL: probe driver: 15b3:1013 librte_net_mlx5
1681 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1682 PMD: librte_net_mlx5: 1 port(s) detected
1683 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1684 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1685 EAL: probe driver: 15b3:1013 librte_net_mlx5
1686 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1687 PMD: librte_net_mlx5: 1 port(s) detected
1688 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1689 Interactive-mode selected
1690 Configuring Port 0 (socket 0)
1691 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1692 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1693 Port 0: E4:1D:2D:E7:0C:FE
1694 Configuring Port 1 (socket 0)
1695 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1696 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1697 Port 1: E4:1D:2D:E7:0C:FF
1698 Configuring Port 2 (socket 0)
1699 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1700 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1701 Port 2: E4:1D:2D:E7:0C:FA
1702 Configuring Port 3 (socket 0)
1703 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1704 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1705 Port 3: E4:1D:2D:E7:0C:FB
1706 Checking link statuses...
1707 Port 0 Link Up - speed 40000 Mbps - full-duplex
1708 Port 1 Link Up - speed 40000 Mbps - full-duplex
1709 Port 2 Link Up - speed 10000 Mbps - full-duplex
1710 Port 3 Link Up - speed 10000 Mbps - full-duplex
1717 This section demonstrates how to dump flows. Currently, it's possible to dump
1718 all flows with assistance of external tools.
1720 #. 2 ways to get flow raw file:
1722 - Using testpmd CLI:
1724 .. code-block:: console
1726 testpmd> flow dump <port> <output_file>
1728 - call rte_flow_dev_dump api:
1730 .. code-block:: console
1732 rte_flow_dev_dump(port, file, NULL);
1734 #. Dump human-readable flows from raw file:
1736 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1738 .. code-block:: console
1740 mlx_steering_dump.py -f <output_file>