1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright 2015 6WIND S.A.
3 Copyright 2015 Mellanox Technologies, Ltd
5 .. include:: <isonum.txt>
10 The MLX5 poll mode driver library (**librte_net_mlx5**) provides support
11 for **Mellanox ConnectX-4**, **Mellanox ConnectX-4 Lx** , **Mellanox
12 ConnectX-5**, **Mellanox ConnectX-6**, **Mellanox ConnectX-6 Dx**, **Mellanox
13 ConnectX-6 Lx**, **Mellanox BlueField** and **Mellanox BlueField-2** families
14 of 10/25/40/50/100/200 Gb/s adapters as well as their virtual functions (VF)
17 Information and documentation about these adapters can be found on the
18 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
19 `Mellanox community <http://community.mellanox.com/welcome>`__.
21 There is also a `section dedicated to this poll mode driver
22 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
28 Besides its dependency on libibverbs (that implies libmlx5 and associated
29 kernel support), librte_net_mlx5 relies heavily on system calls for control
30 operations such as querying/updating the MTU and flow control parameters.
32 For security reasons and robustness, this driver only deals with virtual
33 memory addresses. The way resources allocations are handled by the kernel,
34 combined with hardware specifications that allow to handle virtual memory
35 addresses directly, ensure that DPDK applications cannot access random
36 physical memory (or memory that does not belong to the current process).
38 This capability allows the PMD to coexist with kernel network interfaces
39 which remain functional, although they stop receiving unicast packets as
40 long as they share the same MAC address.
41 This means legacy linux control tools (for example: ethtool, ifconfig and
42 more) can operate on the same network interfaces that owned by the DPDK
45 The PMD can use libibverbs and libmlx5 to access the device firmware
46 or directly the hardware components.
47 There are different levels of objects and bypassing abilities
48 to get the best performances:
50 - Verbs is a complete high-level generic API
51 - Direct Verbs is a device-specific API
52 - DevX allows to access firmware objects
53 - Direct Rules manages flow steering at low-level hardware layer
55 Enabling librte_net_mlx5 causes DPDK applications to be linked against
61 - Multi arch support: x86_64, POWER8, ARMv8, i686.
62 - Multiple TX and RX queues.
63 - Support for scattered TX frames.
64 - Advanced support for scattered Rx frames with tunable buffer attributes.
65 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
66 - RSS using different combinations of fields: L3 only, L4 only or both,
67 and source only, destination only or both.
68 - Several RSS hash keys, one for each flow type.
69 - Default RSS operation with no hash key specification.
70 - Configurable RETA table.
71 - Link flow control (pause frame).
72 - Support for multiple MAC addresses.
76 - RX CRC stripping configuration.
77 - TX mbuf fast free offload.
78 - Promiscuous mode on PF and VF.
79 - Multicast promiscuous mode on PF and VF.
80 - Hardware checksum offloads.
81 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
83 - Flow API, including :ref:`flow_isolated_mode`.
85 - KVM and VMware ESX SR-IOV modes are supported.
86 - RSS hash result is supported.
87 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
88 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
90 - Statistics query including Basic, Extended and per queue.
92 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
93 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
94 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
95 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
96 - Flow insertion rate of more then million flows per second, when using Direct Rules.
97 - Support for multiple rte_flow groups.
98 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
101 - Multiple-thread flow insertion.
102 - Matching on GTP extension header with raw encap/decap action.
103 - Matching on Geneve TLV option header with raw encap/decap action.
104 - RSS support in sample action.
105 - E-Switch mirroring and jump.
106 - E-Switch mirroring and modify.
107 - 21844 flow priorities for ingress or egress flow groups greater than 0 and for any transfer
109 - Flow metering, including meter policy API.
110 - Flow integrity offload API.
111 - Connection tracking.
112 - Sub-Function representors.
119 On Windows, the features are limited:
121 - Promiscuous mode is not supported
122 - The following rules are supported:
124 - IPv4/UDP with CVLAN filtering
125 - Unicast MAC filtering
127 - Additional rules are supported from WinOF2 version 2.70:
129 - IPv4/TCP with CVLAN filtering
130 - L4 steering rules for port RSS of UDP, TCP and IP
132 - For secondary process:
134 - Forked secondary process not supported.
135 - External memory unregistered in EAL memseg list cannot be used for DMA
136 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
137 primary process and remapped to the same virtual address in secondary
138 process. If the external memory is registered by primary process but has
139 different virtual address in secondary process, unexpected error may happen.
141 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
142 specific VLAN will match for VLAN packets as well:
144 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
145 Meaning, the flow rule::
147 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
149 Will only match vlan packets with vid=3. and the flow rule::
151 flow create 0 ingress pattern eth / ipv4 / end ...
153 Will match any ipv4 packet (VLAN included).
155 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
157 - 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.
160 flow create 0 ingress pattern eth / ipv4 / end ...
162 Will match any ipv4 packet.
165 flow create 0 ingress pattern eth / vlan / end ...
166 flow create 0 ingress pattern eth has_vlan is 1 / end ...
167 flow create 0 ingress pattern eth type is 0x8100 / end ...
169 Will match single-tagged packets only, with any VLAN ID value.
172 flow create 0 ingress pattern eth type is 0x88A8 / end ...
173 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
175 Will match multi-tagged packets only, with any VLAN ID value.
177 - A flow pattern with 2 sequential VLAN items is not supported.
179 - VLAN pop offload command:
181 - Flow rules having a VLAN pop offload command as one of their actions and
182 are lacking a match on VLAN as one of their items are not supported.
183 - The command is not supported on egress traffic in NIC mode.
185 - VLAN push offload is not supported on ingress traffic in NIC mode.
187 - VLAN set PCP offload is not supported on existing headers.
189 - A multi segment packet must have not more segments than reported by dev_infos_get()
190 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
191 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
192 inline settings) to 58.
194 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
195 to 0 are not supported.
197 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
199 - Match on Geneve header supports the following fields only:
206 - Match on Geneve TLV option is supported on the following fields:
213 Only one Class/Type/Length Geneve TLV option is supported per shared device.
214 Class/Type/Length fields must be specified as well as masks.
215 Class/Type/Length specified masks must be full.
216 Matching Geneve TLV option without specifying data is not supported.
217 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
219 - VF: flow rules created on VF devices can only match traffic targeted at the
220 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
222 - Match on GTP tunnel header item supports the following fields only:
224 - v_pt_rsv_flags: E flag, S flag, PN flag
228 - Match on GTP extension header only for GTP PDU session container (next
229 extension header type = 0x85).
230 - Match on GTP extension header is not supported in group 0.
232 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
233 The flows within group 0 and set metadata action are rejected by hardware.
237 MAC addresses not already present in the bridge table of the associated
238 kernel network device will be added and cleaned up by the PMD when closing
239 the device. In case of ungraceful program termination, some entries may
240 remain present and should be removed manually by other means.
242 - Buffer split offload is supported with regular Rx burst routine only,
243 no MPRQ feature or vectorized code can be engaged.
245 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
246 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
247 ol_flags. As the mempool for the external buffer is managed by PMD, all the
248 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
249 the external buffers will be freed by PMD and the application which still
250 holds the external buffers may be corrupted.
252 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
253 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
254 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
256 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
257 and allmulticast mode are both set to off.
258 To receive IPv6 Multicast messages on VM, explicitly set the relevant
259 MAC address using rte_eth_dev_mac_addr_add() API.
261 - To support a mixed traffic pattern (some buffers from local host memory, some
262 buffers from other devices) with high bandwidth, a mbuf flag is used.
264 An application hints the PMD whether or not it should try to inline the
265 given mbuf data buffer. PMD should do the best effort to act upon this request.
267 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
268 registered by application with rte_mbuf_dynflag_register(). This flag is
269 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
270 which is intended to be used by the application.
272 To query the supported specific flags in runtime,
273 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
274 currently (over present hardware and configuration) supported specific flags.
275 The "not inline hint" feature operating flow is the following one:
278 - probe the devices, ports are created
279 - query the port capabilities
280 - if port supporting the feature is found
281 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
282 - application starts the ports
283 - on ``dev_start()`` PMD checks whether the feature flag is registered and
284 enables the feature support in datapath
285 - application might set the registered flag bit in ``ol_flags`` field
286 of mbuf being sent and PMD will handle ones appropriately.
288 - The amount of descriptors in Tx queue may be limited by data inline settings.
289 Inline data require the more descriptor building blocks and overall block
290 amount may exceed the hardware supported limits. The application should
291 reduce the requested Tx size or adjust data inline settings with
292 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
294 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
295 parameter should be specified.
296 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
297 being sent it tries to synchronize the time of packet appearing on
298 the wire with the specified packet timestamp. It the specified one
299 is in the past it should be ignored, if one is in the distant future
300 it should be capped with some reasonable value (in range of seconds).
301 These specific cases ("too late" and "distant future") can be optionally
302 reported via device xstats to assist applications to detect the
303 time-related problems.
305 The timestamp upper "too-distant-future" limit
306 at the moment of invoking the Tx burst routine
307 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
308 Please note, for the testpmd txonly mode,
309 the limit is deduced from the expression::
311 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
313 There is no any packet reordering according timestamps is supposed,
314 neither within packet burst, nor between packets, it is an entirely
315 application responsibility to generate packets and its timestamps
316 in desired order. The timestamps can be put only in the first packet
317 in the burst providing the entire burst scheduling.
319 - E-Switch decapsulation Flow:
321 - can be applied to PF port only.
322 - must specify VF port action (packet redirection from PF to VF).
323 - optionally may specify tunnel inner source and destination MAC addresses.
325 - E-Switch encapsulation Flow:
327 - can be applied to VF ports only.
328 - must specify PF port action (packet redirection from VF to PF).
332 - The input buffer, used as outer header, is not validated.
336 - The decapsulation is always done up to the outermost tunnel detected by the HW.
337 - The input buffer, providing the removal size, is not validated.
338 - The buffer size must match the length of the headers to be removed.
340 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
341 mutually exclusive features which cannot be supported together
342 (see :ref:`mlx5_firmware_config`).
346 - Requires DevX and DV flow to be enabled.
347 - KEEP_CRC offload cannot be supported with LRO.
348 - The first mbuf length, without head-room, must be big enough to include the
350 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
351 it with size limited to max LRO size, not to max RX packet length.
352 - LRO can be used with outer header of TCP packets of the standard format:
353 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
355 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
356 - LRO packet aggregation is performed by HW only for packet size larger than
357 ``lro_min_mss_size``. This value is reported on device start, when debug
362 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
363 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
364 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
368 - fast free offload assumes the all mbufs being sent are originated from the
369 same memory pool and there is no any extra references to the mbufs (the
370 reference counter for each mbuf is equal 1 on tx_burst call). The latter
371 means there should be no any externally attached buffers in mbufs. It is
372 an application responsibility to provide the correct mbufs if the fast
373 free offload is engaged. The mlx5 PMD implicitly produces the mbufs with
374 externally attached buffers if MPRQ option is enabled, hence, the fast
375 free offload is neither supported nor advertised if there is MPRQ enabled.
379 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and
380 E-Switch steering domain.
381 - For E-Switch Sampling flow with sample ratio > 1, additional actions are not
382 supported in the sample actions list.
383 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as
384 first action in the E-Switch egress flow if with header modify or
385 encapsulation actions.
386 - For NIC Rx flow, supports ``MARK``, ``COUNT``, ``QUEUE``, ``RSS`` in the
388 - For E-Switch mirroring flow, supports ``RAW ENCAP``, ``Port ID``,
389 ``VXLAN ENCAP``, ``NVGRE ENCAP`` in the sample actions list.
393 - Supports the 'set' operation only for ``RTE_FLOW_ACTION_TYPE_MODIFY_FIELD`` action.
394 - Modification of an arbitrary place in a packet via the special ``RTE_FLOW_FIELD_START`` Field ID is not supported.
395 - Modification of the 802.1Q Tag, VXLAN Network or GENEVE Network ID's is not supported.
396 - Encapsulation levels are not supported, can modify outermost header fields only.
397 - Offsets must be 32-bits aligned, cannot skip past the boundary of a field.
399 - IPv6 header item 'proto' field, indicating the next header protocol, should
400 not be set as extension header.
401 In case the next header is an extension header, it should not be specified in
402 IPv6 header item 'proto' field.
403 The last extension header item 'next header' field can specify the following
404 header protocol type.
408 - 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.
409 - Hairpin in switchdev SR-IOV mode is not supported till now.
413 - All the meter colors with drop action will be counted only by the global drop statistics.
414 - Green color is not supported with drop action.
415 - Yellow detection is not supported.
416 - Red color must be with drop action.
417 - Meter statistics are supported only for drop case.
418 - Meter yellow color detection is not supported.
419 - A meter action created with pre-defined policy must be the last action in the flow except single case where the policy actions are:
420 - green: NULL or END.
421 - yellow: NULL or END.
423 - The only supported meter policy actions:
424 - green: QUEUE, RSS, PORT_ID, JUMP, MARK and SET_TAG.
425 - yellow: must be empty.
427 - meter profile packet mode is supported.
431 - Integrity offload is enabled for **ConnectX-6** family.
432 - Verification bits provided by the hardware are ``l3_ok``, ``ipv4_csum_ok``, ``l4_ok``, ``l4_csum_ok``.
433 - ``level`` value 0 references outer headers.
434 - Multiple integrity items not supported in a single flow rule.
435 - Flow rule items supplied by application must explicitly specify network headers referred by integrity item.
436 For example, if integrity item mask sets ``l4_ok`` or ``l4_csum_ok`` bits, reference to L4 network header,
437 TCP or UDP, must be in the rule pattern as well::
439 flow create 0 ingress pattern integrity level is 0 value mask l3_ok value spec l3_ok / eth / ipv6 / end …
441 flow create 0 ingress pattern integrity level is 0 value mask l4_ok value spec 0 / eth / ipv4 proto is udp / end …
443 - Connection tracking:
445 - Cannot co-exist with ASO meter, ASO age action in a single flow rule.
446 - Flow rules insertion rate and memory consumption need more optimization.
448 - 4M connections maximum.
453 MLX5 supports various methods to report statistics:
455 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.
457 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.
459 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.
467 The ibverbs libraries can be linked with this PMD in a number of ways,
468 configured by the ``ibverbs_link`` build option:
470 - ``shared`` (default): the PMD depends on some .so files.
472 - ``dlopen``: Split the dependencies glue in a separate library
473 loaded when needed by dlopen.
474 It make dependencies on libibverbs and libmlx4 optional,
475 and has no performance impact.
477 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
478 in the PMD shared library or the executable static binary.
480 Environment variables
481 ~~~~~~~~~~~~~~~~~~~~~
485 A list of directories in which to search for the rdma-core "glue" plug-in,
486 separated by colons or semi-colons.
488 - ``MLX5_SHUT_UP_BF``
490 Configures HW Tx doorbell register as IO-mapped.
492 By default, the HW Tx doorbell is configured as a write-combining register.
493 The register would be flushed to HW usually when the write-combining buffer
494 becomes full, but it depends on CPU design.
496 Except for vectorized Tx burst routines, a write memory barrier is enforced
497 after updating the register so that the update can be immediately visible to
500 When vectorized Tx burst is called, the barrier is set only if the burst size
501 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
502 variable will bring better latency even though the maximum throughput can
505 Run-time configuration
506 ~~~~~~~~~~~~~~~~~~~~~~
508 - librte_net_mlx5 brings kernel network interfaces up during initialization
509 because it is affected by their state. Forcing them down prevents packets
512 - **ethtool** operations on related kernel interfaces also affect the PMD.
517 In order to run as a non-root user,
518 some capabilities must be granted to the application::
520 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
522 Below are the reasons of the need for each capability:
525 When using physical addresses (PA mode), with Linux >= 4.0,
526 for access to ``/proc/self/pagemap``.
529 For device configuration.
532 For raw ethernet queue allocation through kernel driver.
535 For DMA memory pinning.
540 - ``rxq_cqe_comp_en`` parameter [int]
542 A nonzero value enables the compression of CQE on RX side. This feature
543 allows to save PCI bandwidth and improve performance. Enabled by default.
544 Different compression formats are supported in order to achieve the best
545 performance for different traffic patterns. Default format depends on
546 Multi-Packet Rx queue configuration: Hash RSS format is used in case
547 MPRQ is disabled, Checksum format is used in case MPRQ is enabled.
549 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
550 better compression rate in case of RTE Flow Mark traffic.
551 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
552 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
553 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
554 CQE compression format selection requires DevX to be enabled. If there is
555 no DevX enabled/supported the value is reset to 1 by default.
559 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
560 ConnectX-6 Lx, BlueField and BlueField-2.
561 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
562 ConnectX-6 Lx, BlueField and BlueField-2.
564 - ``rxq_pkt_pad_en`` parameter [int]
566 A nonzero value enables padding Rx packet to the size of cacheline on PCI
567 transaction. This feature would waste PCI bandwidth but could improve
568 performance by avoiding partial cacheline write which may cause costly
569 read-modify-copy in memory transaction on some architectures. Disabled by
574 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
575 ConnectX-6 Lx, BlueField and BlueField-2.
576 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
577 ConnectX-6 Lx, BlueField and BlueField-2.
579 - ``mprq_en`` parameter [int]
581 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
582 configured as Multi-Packet RQ if the total number of Rx queues is
583 ``rxqs_min_mprq`` or more. Disabled by default.
585 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
586 by posting a single large buffer for multiple packets. Instead of posting a
587 buffers per a packet, one large buffer is posted in order to receive multiple
588 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
589 and each stride receives one packet. MPRQ can improve throughput for
590 small-packet traffic.
592 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
593 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
594 configure large stride size enough to accommodate max_rx_pkt_len as long as
595 device allows. Note that this can waste system memory compared to enabling Rx
596 scatter and multi-segment packet.
598 - ``mprq_log_stride_num`` parameter [int]
600 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
601 strides can reduce PCIe traffic further. If configured value is not in the
602 range of device capability, the default value will be set with a warning
603 message. The default value is 4 which is 16 strides per a buffer, valid only
604 if ``mprq_en`` is set.
606 The size of Rx queue should be bigger than the number of strides.
608 - ``mprq_log_stride_size`` parameter [int]
610 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
611 stride size can save some memory and reduce probability of a depletion of all
612 available strides due to unreleased packets by an application. If configured
613 value is not in the range of device capability, the default value will be set
614 with a warning message. The default value is 11 which is 2048 bytes per a
615 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
616 it is possible for a packet to span across multiple strides. This mode allows
617 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
618 of a packet if Rx scatter is configured) may be required in case there is no
619 space left for a head room at the end of a stride which incurs some
622 - ``mprq_max_memcpy_len`` parameter [int]
624 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
625 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
626 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
627 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
628 A mempool for external buffers will be allocated and managed by PMD. If Rx
629 packet is externally attached, ol_flags field of the mbuf will have
630 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
631 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
633 - ``rxqs_min_mprq`` parameter [int]
635 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
636 greater or equal to this value. The default value is 12, valid only if
639 - ``txq_inline`` parameter [int]
641 Amount of data to be inlined during TX operations. This parameter is
642 deprecated and converted to the new parameter ``txq_inline_max`` providing
643 partial compatibility.
645 - ``txqs_min_inline`` parameter [int]
647 Enable inline data send only when the number of TX queues is greater or equal
650 This option should be used in combination with ``txq_inline_max`` and
651 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
653 If this option is not specified the default value 16 is used for BlueField
654 and 8 for other platforms
656 The data inlining consumes the CPU cycles, so this option is intended to
657 auto enable inline data if we have enough Tx queues, which means we have
658 enough CPU cores and PCI bandwidth is getting more critical and CPU
659 is not supposed to be bottleneck anymore.
661 The copying data into WQE improves latency and can improve PPS performance
662 when PCI back pressure is detected and may be useful for scenarios involving
663 heavy traffic on many queues.
665 Because additional software logic is necessary to handle this mode, this
666 option should be used with care, as it may lower performance when back
667 pressure is not expected.
669 If inline data are enabled it may affect the maximal size of Tx queue in
670 descriptors because the inline data increase the descriptor size and
671 queue size limits supported by hardware may be exceeded.
673 - ``txq_inline_min`` parameter [int]
675 Minimal amount of data to be inlined into WQE during Tx operations. NICs
676 may require this minimal data amount to operate correctly. The exact value
677 may depend on NIC operation mode, requested offloads, etc. It is strongly
678 recommended to omit this parameter and use the default values. Anyway,
679 applications using this parameter should take into consideration that
680 specifying an inconsistent value may prevent the NIC from sending packets.
682 If ``txq_inline_min`` key is present the specified value (may be aligned
683 by the driver in order not to exceed the limits and provide better descriptor
684 space utilization) will be used by the driver and it is guaranteed that
685 requested amount of data bytes are inlined into the WQE beside other inline
686 settings. This key also may update ``txq_inline_max`` value (default
687 or specified explicitly in devargs) to reserve the space for inline data.
689 If ``txq_inline_min`` key is not present, the value may be queried by the
690 driver from the NIC via DevX if this feature is available. If there is no DevX
691 enabled/supported the value 18 (supposing L2 header including VLAN) is set
692 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
693 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
696 For ConnectX-4 NIC, driver does not allow specifying value below 18
697 (minimal L2 header, including VLAN), error will be raised.
699 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
700 it is not recommended and may prevent NIC from sending packets over
703 For ConnectX-4 and ConnectX-4 Lx NICs, automatically configured value
704 is insufficient for some traffic, because they require at least all L2 headers
705 to be inlined. For example, Q-in-Q adds 4 bytes to default 18 bytes
706 of Ethernet and VLAN, thus ``txq_inline_min`` must be set to 22.
707 MPLS would add 4 bytes per label. Final value must account for all possible
708 L2 encapsulation headers used in particular environment.
710 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
711 Multi-Packet Write), because last one does not support partial packet inlining.
712 This is not very critical due to minimal data inlining is mostly required
713 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
715 - ``txq_inline_max`` parameter [int]
717 Specifies the maximal packet length to be completely inlined into WQE
718 Ethernet Segment for ordinary SEND method. If packet is larger than specified
719 value, the packet data won't be copied by the driver at all, data buffer
720 is addressed with a pointer. If packet length is less or equal all packet
721 data will be copied into WQE. This may improve PCI bandwidth utilization for
722 short packets significantly but requires the extra CPU cycles.
724 The data inline feature is controlled by number of Tx queues, if number of Tx
725 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
726 is engaged, if there are not enough Tx queues (which means not enough CPU cores
727 and CPU resources are scarce), data inline is not performed by the driver.
728 Assigning ``txqs_min_inline`` with zero always enables the data inline.
730 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
731 by the driver in order not to exceed the limit (930 bytes) and to provide better
732 WQE space filling without gaps, the adjustment is reflected in the debug log.
733 Also, the default value (290) may be decreased in run-time if the large transmit
734 queue size is requested and hardware does not support enough descriptor
735 amount, in this case warning is emitted. If ``txq_inline_max`` key is
736 specified and requested inline settings can not be satisfied then error
739 - ``txq_inline_mpw`` parameter [int]
741 Specifies the maximal packet length to be completely inlined into WQE for
742 Enhanced MPW method. If packet is large the specified value, the packet data
743 won't be copied, and data buffer is addressed with pointer. If packet length
744 is less or equal, all packet data will be copied into WQE. This may improve PCI
745 bandwidth utilization for short packets significantly but requires the extra
748 The data inline feature is controlled by number of TX queues, if number of Tx
749 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
750 is engaged, if there are not enough Tx queues (which means not enough CPU cores
751 and CPU resources are scarce), data inline is not performed by the driver.
752 Assigning ``txqs_min_inline`` with zero always enables the data inline.
754 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
755 by the driver in order not to exceed the limit (930 bytes) and to provide better
756 WQE space filling without gaps, the adjustment is reflected in the debug log.
757 Due to multiple packets may be included to the same WQE with Enhanced Multi
758 Packet Write Method and overall WQE size is limited it is not recommended to
759 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
760 may be decreased in run-time if the large transmit queue size is requested
761 and hardware does not support enough descriptor amount, in this case warning
762 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
763 settings can not be satisfied then error will be raised.
765 - ``txqs_max_vec`` parameter [int]
767 Enable vectorized Tx only when the number of TX queues is less than or
768 equal to this value. This parameter is deprecated and ignored, kept
769 for compatibility issue to not prevent driver from probing.
771 - ``txq_mpw_hdr_dseg_en`` parameter [int]
773 A nonzero value enables including two pointers in the first block of TX
774 descriptor. The parameter is deprecated and ignored, kept for compatibility
777 - ``txq_max_inline_len`` parameter [int]
779 Maximum size of packet to be inlined. This limits the size of packet to
780 be inlined. If the size of a packet is larger than configured value, the
781 packet isn't inlined even though there's enough space remained in the
782 descriptor. Instead, the packet is included with pointer. This parameter
783 is deprecated and converted directly to ``txq_inline_mpw`` providing full
784 compatibility. Valid only if eMPW feature is engaged.
786 - ``txq_mpw_en`` parameter [int]
788 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
789 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
790 eMPW allows the Tx burst function to pack up multiple packets
791 in a single descriptor session in order to save PCI bandwidth
792 and improve performance at the cost of a slightly higher CPU usage.
793 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
794 Tx burst function copies entire packet data on to Tx descriptor
795 instead of including pointer of packet.
797 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
798 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
799 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
800 option or reported by the NIC, the eMPW feature is disengaged.
802 - ``tx_db_nc`` parameter [int]
804 The rdma core library can map doorbell register in two ways, depending on the
805 environment variable "MLX5_SHUT_UP_BF":
807 - As regular cached memory (usually with write combining attribute), if the
808 variable is either missing or set to zero.
809 - As non-cached memory, if the variable is present and set to not "0" value.
811 The type of mapping may slightly affect the Tx performance, the optimal choice
812 is strongly relied on the host architecture and should be deduced practically.
814 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
815 memory (with write combining), the PMD will perform the extra write memory barrier
816 after writing to doorbell, it might increase the needed CPU clocks per packet
817 to send, but latency might be improved.
819 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
820 cached memory, the PMD will not perform the extra write memory barrier
821 after writing to doorbell, on some architectures it might improve the
824 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
825 memory, the PMD will use heuristics to decide whether write memory barrier
826 should be performed. For bursts with size multiple of recommended one (64 pkts)
827 it is supposed the next burst is coming and no need to issue the extra memory
828 barrier (it is supposed to be issued in the next coming burst, at least after
829 descriptor writing). It might increase latency (on some hosts till next
830 packets transmit) and should be used with care.
832 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
833 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
834 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
836 - ``tx_pp`` parameter [int]
838 If a nonzero value is specified the driver creates all necessary internal
839 objects to provide accurate packet send scheduling on mbuf timestamps.
840 The positive value specifies the scheduling granularity in nanoseconds,
841 the packet send will be accurate up to specified digits. The allowed range is
842 from 500 to 1 million of nanoseconds. The negative value specifies the module
843 of granularity and engages the special test mode the check the schedule rate.
844 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
847 - ``tx_skew`` parameter [int]
849 The parameter adjusts the send packet scheduling on timestamps and represents
850 the average delay between beginning of the transmitting descriptor processing
851 by the hardware and appearance of actual packet data on the wire. The value
852 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
853 specified. The default value is zero.
855 - ``tx_vec_en`` parameter [int]
857 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
858 ConnectX-6 Lx, BlueField and BlueField-2 NICs
859 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
860 The parameter is deprecated and ignored.
862 - ``rx_vec_en`` parameter [int]
864 A nonzero value enables Rx vector if the port is not configured in
865 multi-segment otherwise this parameter is ignored.
869 - ``vf_nl_en`` parameter [int]
871 A nonzero value enables Netlink requests from the VF to add/remove MAC
872 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
873 Otherwise the relevant configuration must be run with Linux iproute2 tools.
874 This is a prerequisite to receive this kind of traffic.
876 Enabled by default, valid only on VF devices ignored otherwise.
878 - ``l3_vxlan_en`` parameter [int]
880 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
881 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
882 parameter. This is a prerequisite to receive this kind of traffic.
886 - ``dv_xmeta_en`` parameter [int]
888 A nonzero value enables extensive flow metadata support if device is
889 capable and driver supports it. This can enable extensive support of
890 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
891 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
893 There are some possible configurations, depending on parameter value:
895 - 0, this is default value, defines the legacy mode, the ``MARK`` and
896 ``META`` related actions and items operate only within NIC Tx and
897 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
898 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
899 item is 32 bits wide and match supported on egress only.
901 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
902 related actions and items operate within all supported steering domains,
903 including FDB, ``MARK`` and ``META`` information may cross the domain
904 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
905 depends on kernel and firmware configurations and might be 0, 16 or
906 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
907 compatibility, the actual width of data transferred to the FDB domain
908 depends on kernel configuration and may be vary. The actual supported
909 width can be retrieved in runtime by series of rte_flow_validate()
912 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
913 related actions and items operate within all supported steering domains,
914 including FDB, ``MARK`` and ``META`` information may cross the domain
915 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
916 depends on kernel and firmware configurations and might be 0, 16 or
917 24 bits. The actual supported width can be retrieved in runtime by
918 series of rte_flow_validate() trials.
920 - 3, this engages tunnel offload mode. In E-Switch configuration, that
921 mode implicitly activates ``dv_xmeta_en=1``.
923 +------+-----------+-----------+-------------+-------------+
924 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
925 +======+===========+===========+=============+=============+
926 | 0 | 24 bits | 32 bits | 32 bits | no |
927 +------+-----------+-----------+-------------+-------------+
928 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
929 +------+-----------+-----------+-------------+-------------+
930 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
931 +------+-----------+-----------+-------------+-------------+
933 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
934 ignored and the device is configured to operate in legacy mode (0).
936 Disabled by default (set to 0).
938 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
939 of the extensive metadata features. The legacy Verbs supports FLAG and
940 MARK metadata actions over NIC Rx steering domain only.
942 Setting META value to zero in flow action means there is no item provided
943 and receiving datapath will not report in mbufs the metadata are present.
944 Setting MARK value to zero in flow action means the zero FDIR ID value
945 will be reported on packet receiving.
947 For the MARK action the last 16 values in the full range are reserved for
948 internal PMD purposes (to emulate FLAG action). The valid range for the
949 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
950 for the 24-bit mode, the flows with the MARK action value outside
951 the specified range will be rejected.
953 - ``dv_flow_en`` parameter [int]
955 A nonzero value enables the DV flow steering assuming it is supported
956 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
958 Enabled by default if supported.
960 - ``dv_esw_en`` parameter [int]
962 A nonzero value enables E-Switch using Direct Rules.
964 Enabled by default if supported.
966 - ``lacp_by_user`` parameter [int]
968 A nonzero value enables the control of LACP traffic by the user application.
969 When a bond exists in the driver, by default it should be managed by the
970 kernel and therefore LACP traffic should be steered to the kernel.
971 If this devarg is set to 1 it will allow the user to manage the bond by
972 itself and not steer LACP traffic to the kernel.
974 Disabled by default (set to 0).
976 - ``mr_ext_memseg_en`` parameter [int]
978 A nonzero value enables extending memseg when registering DMA memory. If
979 enabled, the number of entries in MR (Memory Region) lookup table on datapath
980 is minimized and it benefits performance. On the other hand, it worsens memory
981 utilization because registered memory is pinned by kernel driver. Even if a
982 page in the extended chunk is freed, that doesn't become reusable until the
983 entire memory is freed.
987 - ``representor`` parameter [list]
989 This parameter can be used to instantiate DPDK Ethernet devices from
990 existing port (PF, VF or SF) representors configured on the device.
992 It is a standard parameter whose format is described in
993 :ref:`ethernet_device_standard_device_arguments`.
995 For instance, to probe VF port representors 0 through 2::
997 <PCI_BDF>,representor=vf[0-2]
999 To probe SF port representors 0 through 2::
1001 <PCI_BDF>,representor=sf[0-2]
1003 To probe VF port representors 0 through 2 on both PFs of bonding device::
1005 <Primary_PCI_BDF>,representor=pf[0,1]vf[0-2]
1007 - ``max_dump_files_num`` parameter [int]
1009 The maximum number of files per PMD entity that may be created for debug information.
1010 The files will be created in /var/log directory or in current directory.
1012 set to 128 by default.
1014 - ``lro_timeout_usec`` parameter [int]
1016 The maximum allowed duration of an LRO session, in micro-seconds.
1017 PMD will set the nearest value supported by HW, which is not bigger than
1018 the input ``lro_timeout_usec`` value.
1019 If this parameter is not specified, by default PMD will set
1020 the smallest value supported by HW.
1022 - ``hp_buf_log_sz`` parameter [int]
1024 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
1025 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
1026 The capacity of the value is specified by the firmware and the initialization
1027 will get a failure if it is out of scope.
1028 The range of the value is from 11 to 19 right now, and the supported frame
1029 size of a single packet for hairpin is from 512B to 128KB. It might change if
1030 different firmware release is being used. By using a small value, it could
1031 reduce memory consumption but not work with a large frame. If the value is
1032 too large, the memory consumption will be high and some potential performance
1033 degradation will be introduced.
1034 By default, the PMD will set this value to 16, which means that 9KB jumbo
1035 frames will be supported.
1037 - ``reclaim_mem_mode`` parameter [int]
1039 Cache some resources in flow destroy will help flow recreation more efficient.
1040 While some systems may require the all the resources can be reclaimed after
1042 The parameter ``reclaim_mem_mode`` provides the option for user to configure
1043 if the resource cache is needed or not.
1045 There are three options to choose:
1047 - 0. It means the flow resources will be cached as usual. The resources will
1048 be cached, helpful with flow insertion rate.
1050 - 1. It will only enable the DPDK PMD level resources reclaim.
1052 - 2. Both DPDK PMD level and rdma-core low level will be configured as
1055 By default, the PMD will set this value to 0.
1057 - ``sys_mem_en`` parameter [int]
1059 A non-zero value enables the PMD memory management allocating memory
1060 from system by default, without explicit rte memory flag.
1062 By default, the PMD will set this value to 0.
1064 - ``decap_en`` parameter [int]
1066 Some devices do not support FCS (frame checksum) scattering for
1067 tunnel-decapsulated packets.
1068 If set to 0, this option forces the FCS feature and rejects tunnel
1069 decapsulation in the flow engine for such devices.
1071 By default, the PMD will set this value to 1.
1073 .. _mlx5_firmware_config:
1075 Firmware configuration
1076 ~~~~~~~~~~~~~~~~~~~~~~
1078 Firmware features can be configured as key/value pairs.
1080 The command to set a value is::
1082 mlxconfig -d <device> set <key>=<value>
1084 The command to query a value is::
1086 mlxconfig -d <device> query | grep <key>
1088 The device name for the command ``mlxconfig`` can be either the PCI address,
1089 or the mst device name found with::
1093 Below are some firmware configurations listed.
1099 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1105 - maximum number of SR-IOV virtual functions::
1109 - enable DevX (required by Direct Rules and other features)::
1113 - aggressive CQE zipping::
1117 - L3 VXLAN and VXLAN-GPE destination UDP port::
1120 IP_OVER_VXLAN_PORT=<udp dport>
1122 - enable VXLAN-GPE tunnel flow matching::
1124 FLEX_PARSER_PROFILE_ENABLE=0
1126 FLEX_PARSER_PROFILE_ENABLE=2
1128 - enable IP-in-IP tunnel flow matching::
1130 FLEX_PARSER_PROFILE_ENABLE=0
1132 - enable MPLS flow matching::
1134 FLEX_PARSER_PROFILE_ENABLE=1
1136 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1138 FLEX_PARSER_PROFILE_ENABLE=2
1140 - enable Geneve flow matching::
1142 FLEX_PARSER_PROFILE_ENABLE=0
1144 FLEX_PARSER_PROFILE_ENABLE=1
1146 - enable Geneve TLV option flow matching::
1148 FLEX_PARSER_PROFILE_ENABLE=0
1150 - enable GTP flow matching::
1152 FLEX_PARSER_PROFILE_ENABLE=3
1154 - enable eCPRI flow matching::
1156 FLEX_PARSER_PROFILE_ENABLE=4
1162 This driver relies on external libraries and kernel drivers for resources
1163 allocations and initialization. The following dependencies are not part of
1164 DPDK and must be installed separately:
1168 User space Verbs framework used by librte_net_mlx5. This library provides
1169 a generic interface between the kernel and low-level user space drivers
1172 It allows slow and privileged operations (context initialization, hardware
1173 resources allocations) to be managed by the kernel and fast operations to
1174 never leave user space.
1178 Low-level user space driver library for Mellanox
1179 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1182 This library basically implements send/receive calls to the hardware
1185 - **Kernel modules**
1187 They provide the kernel-side Verbs API and low level device drivers that
1188 manage actual hardware initialization and resources sharing with user
1191 Unlike most other PMDs, these modules must remain loaded and bound to
1194 - mlx5_core: hardware driver managing Mellanox
1195 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1197 - mlx5_ib: InifiniBand device driver.
1198 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1200 - **Firmware update**
1202 Mellanox OFED/EN releases include firmware updates for
1203 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1205 Because each release provides new features, these updates must be applied to
1206 match the kernel modules and libraries they come with.
1210 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1216 Either RDMA Core library with a recent enough Linux kernel release
1217 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1220 RDMA Core with Linux Kernel
1221 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1223 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1224 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1225 (see `RDMA Core installation documentation`_)
1226 - When building for i686 use:
1228 - rdma-core version 18.0 or above built with 32bit support.
1229 - Kernel version 4.14.41 or above.
1231 - Starting with rdma-core v21, static libraries can be built::
1234 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1237 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1238 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1244 - Mellanox OFED version: **4.5** and above /
1245 Mellanox EN version: **4.5** and above
1248 - ConnectX-4: **12.21.1000** and above.
1249 - ConnectX-4 Lx: **14.21.1000** and above.
1250 - ConnectX-5: **16.21.1000** and above.
1251 - ConnectX-5 Ex: **16.21.1000** and above.
1252 - ConnectX-6: **20.27.0090** and above.
1253 - ConnectX-6 Dx: **22.27.0090** and above.
1254 - BlueField: **18.25.1010** and above.
1256 While these libraries and kernel modules are available on OpenFabrics
1257 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1258 managers on most distributions, this PMD requires Ethernet extensions that
1259 may not be supported at the moment (this is a work in progress).
1262 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1264 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1265 include the necessary support and should be used in the meantime. For DPDK,
1266 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1267 required from that distribution.
1271 Several versions of Mellanox OFED/EN are available. Installing the version
1272 this DPDK release was developed and tested against is strongly
1273 recommended. Please check the `linux prerequisites`_.
1275 Windows Prerequisites
1276 ---------------------
1278 This driver relies on external libraries and kernel drivers for resources
1279 allocations and initialization. The dependencies in the following sub-sections
1280 are not part of DPDK, and must be installed separately.
1282 Compilation Prerequisites
1283 ~~~~~~~~~~~~~~~~~~~~~~~~~
1285 DevX SDK installation
1286 ^^^^^^^^^^^^^^^^^^^^^
1288 The DevX SDK must be installed on the machine building the Windows PMD.
1289 Additional information can be found at
1290 `How to Integrate Windows DevX in Your Development Environment
1291 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1293 Runtime Prerequisites
1294 ~~~~~~~~~~~~~~~~~~~~~
1296 WinOF2 version 2.60 or higher must be installed on the machine.
1301 The driver can be downloaded from the following site:
1303 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1308 DevX for Windows must be enabled in the Windows registry.
1309 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1310 Additional information can be found in the WinOF2 user manual.
1315 The following Mellanox device families are supported by the same mlx5 driver:
1327 Below are detailed device names:
1329 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1330 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1331 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1332 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1333 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1334 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1335 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1336 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1337 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1338 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1339 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1340 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1341 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1342 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1343 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1344 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1345 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1346 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1347 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1348 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1349 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1350 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1351 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1352 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1353 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1354 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1356 Quick Start Guide on OFED/EN
1357 ----------------------------
1359 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1362 2. Install the required libraries and kernel modules either by installing
1363 only the required set, or by installing the entire Mellanox OFED/EN::
1365 ./mlnxofedinstall --upstream-libs --dpdk
1367 3. Verify the firmware is the correct one::
1371 4. Verify all ports links are set to Ethernet::
1373 mlxconfig -d <mst device> query | grep LINK_TYPE
1377 Link types may have to be configured to Ethernet::
1379 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1381 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1383 For hypervisors, verify SR-IOV is enabled on the NIC::
1385 mlxconfig -d <mst device> query | grep SRIOV_EN
1388 If needed, configure SR-IOV::
1390 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1391 mlxfwreset -d <mst device> reset
1393 5. Restart the driver::
1395 /etc/init.d/openibd restart
1399 service openibd restart
1401 If link type was changed, firmware must be reset as well::
1403 mlxfwreset -d <mst device> reset
1405 For hypervisors, after reset write the sysfs number of virtual functions
1408 To dynamically instantiate a given number of virtual functions (VFs)::
1410 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1412 6. Install DPDK and you are ready to go.
1413 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1415 Enable switchdev mode
1416 ---------------------
1418 Switchdev mode is a mode in E-Switch, that binds between representor and VF or SF.
1419 Representor is a port in DPDK that is connected to a VF or SF in such a way
1420 that assuming there are no offload flows, each packet that is sent from the VF or SF
1421 will be received by the corresponding representor. While each packet that is or SF
1422 sent to a representor will be received by the VF or SF.
1423 This is very useful in case of SRIOV mode, where the first packet that is sent
1424 by the VF or SF will be received by the DPDK application which will decide if this
1425 flow should be offloaded to the E-Switch. After offloading the flow packet
1426 that the VF or SF that are matching the flow will not be received any more by
1427 the DPDK application.
1429 1. Enable SRIOV mode::
1431 mlxconfig -d <mst device> set SRIOV_EN=true
1433 2. Configure the max number of VFs::
1435 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1439 mlxfwreset -d <mst device> reset
1441 3. Configure the actual number of VFs::
1443 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1445 4. Unbind the device (can be rebind after the switchdev mode)::
1447 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1449 5. Enbale switchdev mode::
1451 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1453 Sub-Function representor
1454 ------------------------
1456 Sub-Function is a portion of the PCI device, a SF netdev has its own
1457 dedicated queues(txq, rxq). A SF netdev supports E-Switch representation
1458 offload similar to existing PF and VF representors. A SF shares PCI
1459 level resources with other SFs and/or with its parent PCI function.
1461 1. Configure SF feature::
1463 mlxconfig -d <mst device> set PF_BAR2_SIZE=<0/1/2/3> PF_BAR2_ENABLE=1
1465 Value of PF_BAR2_SIZE:
1474 mlxfwreset -d <mst device> reset
1476 3. Enable switchdev mode::
1478 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1482 mlnx-sf -d <PCI_BDF> -a create
1484 5. Probe SF representor::
1486 testpmd> port attach <PCI_BDF>,representor=sf0,dv_flow_en=1
1491 1. Configure aggressive CQE Zipping for maximum performance::
1493 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1495 To set it back to the default CQE Zipping mode use::
1497 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1499 2. In case of virtualization:
1501 - Make sure that hypervisor kernel is 3.16 or newer.
1502 - Configure boot with ``iommu=pt``.
1503 - Use 1G huge pages.
1504 - Make sure to allocate a VM on huge pages.
1505 - Make sure to set CPU pinning.
1507 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1508 for better performance. For VMs, verify that the right CPU
1509 and NUMA node are pinned according to the above. Run::
1511 lstopo-no-graphics --merge
1513 to identify the NUMA node to which the PCIe adapter is connected.
1515 4. If more than one adapter is used, and root complex capabilities allow
1516 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1517 it is recommended to locate both adapters on the same NUMA node.
1518 This in order to forward packets from one to the other without
1519 NUMA performance penalty.
1521 5. Disable pause frames::
1523 ethtool -A <netdev> rx off tx off
1525 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1526 via the BIOS configuration. Please contact you server provider for more
1527 information about the settings.
1531 On some machines, depends on the machine integrator, it is beneficial
1532 to set the PCI max read request parameter to 1K. This can be
1533 done in the following way:
1535 To query the read request size use::
1537 setpci -s <NIC PCI address> 68.w
1539 If the output is different than 3XXX, set it by::
1541 setpci -s <NIC PCI address> 68.w=3XXX
1543 The XXX can be different on different systems. Make sure to configure
1544 according to the setpci output.
1546 7. To minimize overhead of searching Memory Regions:
1548 - '--socket-mem' is recommended to pin memory by predictable amount.
1549 - Configure per-lcore cache when creating Mempools for packet buffer.
1550 - Refrain from dynamically allocating/freeing memory in run-time.
1555 There are multiple Rx burst functions with different advantages and limitations.
1557 .. table:: Rx burst functions
1559 +-------------------+------------------------+---------+-----------------+------+-------+
1560 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1561 | | | | || comp|| MTU |
1562 +===================+========================+=========+=================+======+=======+
1563 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1564 +-------------------+------------------------+---------+-----------------+------+-------+
1565 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1566 +-------------------+------------------------+---------+-----------------+------+-------+
1567 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1568 | || RxQs >= rxqs_min_mprq | | | | |
1569 +-------------------+------------------------+---------+-----------------+------+-------+
1570 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1571 | || mprq_en=1 | | | | |
1572 | || RxQs >= rxqs_min_mprq | | | | |
1573 +-------------------+------------------------+---------+-----------------+------+-------+
1575 .. _mlx5_offloads_support:
1577 Supported hardware offloads
1578 ---------------------------
1580 .. table:: Minimal SW/HW versions for queue offloads
1582 ============== ===== ===== ========= ===== ========== =============
1583 Offload DPDK Linux rdma-core OFED firmware hardware
1584 ============== ===== ===== ========= ===== ========== =============
1585 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1586 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1587 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1588 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1589 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1590 Tx scheduling 20.08 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1591 Buffer Split 20.11 N/A N/A 5.1-2 16.28.2006 ConnectX-5
1592 ============== ===== ===== ========= ===== ========== =============
1594 .. table:: Minimal SW/HW versions for rte_flow offloads
1596 +-----------------------+-----------------+-----------------+
1597 | Offload | with E-Switch | with NIC |
1598 +=======================+=================+=================+
1599 | Count | | DPDK 19.05 | | DPDK 19.02 |
1600 | | | OFED 4.6 | | OFED 4.6 |
1601 | | | rdma-core 24 | | rdma-core 23 |
1602 | | | ConnectX-5 | | ConnectX-5 |
1603 +-----------------------+-----------------+-----------------+
1604 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1605 | | | OFED 4.6 | | OFED 4.5 |
1606 | | | rdma-core 24 | | rdma-core 23 |
1607 | | | ConnectX-5 | | ConnectX-4 |
1608 +-----------------------+-----------------+-----------------+
1609 | Queue / RSS | | | | DPDK 18.11 |
1610 | | | N/A | | OFED 4.5 |
1611 | | | | | rdma-core 23 |
1612 | | | | | ConnectX-4 |
1613 +-----------------------+-----------------+-----------------+
1614 | Shared action | | | | |
1615 | | | :numref:`sact`| | :numref:`sact`|
1618 +-----------------------+-----------------+-----------------+
1619 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1620 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1621 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1622 | | of_set_vlan_pcp / | | | | |
1623 | | of_set_vlan_vid) | | | | |
1624 +-----------------------+-----------------+-----------------+
1625 | | VLAN | | DPDK 21.05 | | |
1626 | | ingress and / | | OFED 5.3 | | N/A |
1627 | | of_push_vlan / | | ConnectX-6 Dx | | |
1628 +-----------------------+-----------------+-----------------+
1629 | | VLAN | | DPDK 21.05 | | |
1630 | | egress and / | | OFED 5.3 | | N/A |
1631 | | of_pop_vlan / | | ConnectX-6 Dx | | |
1632 +-----------------------+-----------------+-----------------+
1633 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1634 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1635 | | | rdma-core 24 | | rdma-core 23 |
1636 | | | ConnectX-5 | | ConnectX-5 |
1637 +-----------------------+-----------------+-----------------+
1638 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1639 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1640 | | | rdma-core 27 | | rdma-core 27 |
1641 | | | ConnectX-5 | | ConnectX-5 |
1642 +-----------------------+-----------------+-----------------+
1643 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1644 | | | OFED 5.1-2 | | OFED 5.1-2 |
1645 | | | rdma-core 32 | | N/A |
1646 | | | ConnectX-5 | | ConnectX-5 |
1647 +-----------------------+-----------------+-----------------+
1648 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1649 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1650 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1651 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1652 | | set_ipv6_dst / | | | | |
1653 | | set_tp_src / | | | | |
1654 | | set_tp_dst / | | | | |
1655 | | dec_ttl / | | | | |
1656 | | set_ttl / | | | | |
1657 | | set_mac_src / | | | | |
1658 | | set_mac_dst) | | | | |
1659 +-----------------------+-----------------+-----------------+
1660 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1661 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1662 | | | | rdma-core 24 | | rdma-core 24 |
1663 | | | | ConnectX-5 | | ConnectX-5 |
1664 +-----------------------+-----------------+-----------------+
1665 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1666 | | | OFED 4.7-1 | | OFED 4.7-1 |
1667 | | | rdma-core 24 | | N/A |
1668 | | | ConnectX-5 | | ConnectX-5 |
1669 +-----------------------+-----------------+-----------------+
1670 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1671 | | | OFED 4.6 | | OFED 4.5 |
1672 | | | rdma-core 24 | | rdma-core 23 |
1673 | | | ConnectX-5 | | ConnectX-4 |
1674 +-----------------------+-----------------+-----------------+
1675 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1676 | | | OFED 4.7-3 | | OFED 4.7-3 |
1677 | | | rdma-core 26 | | rdma-core 26 |
1678 | | | ConnectX-5 | | ConnectX-5 |
1679 +-----------------------+-----------------+-----------------+
1680 | Port ID | | DPDK 19.05 | | N/A |
1681 | | | OFED 4.7-1 | | N/A |
1682 | | | rdma-core 24 | | N/A |
1683 | | | ConnectX-5 | | N/A |
1684 +-----------------------+-----------------+-----------------+
1685 | Hairpin | | | | DPDK 19.11 |
1686 | | | N/A | | OFED 4.7-3 |
1687 | | | | | rdma-core 26 |
1688 | | | | | ConnectX-5 |
1689 +-----------------------+-----------------+-----------------+
1690 | 2-port Hairpin | | | | DPDK 20.11 |
1691 | | | N/A | | OFED 5.1-2 |
1693 | | | | | ConnectX-5 |
1694 +-----------------------+-----------------+-----------------+
1695 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1696 | | | OFED 4.7-3 | | OFED 4.7-3 |
1697 | | | rdma-core 26 | | rdma-core 26 |
1698 | | | ConnectX-5 | | ConnectX-5 |
1699 +-----------------------+-----------------+-----------------+
1700 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1701 | | | OFED 5.1-2 | | OFED 5.1-2 |
1702 | | | rdma-core 32 | | N/A |
1703 | | | ConnectX-5 | | ConnectX-5 |
1704 +-----------------------+-----------------+-----------------+
1705 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1706 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1707 | | | rdma-core 35 | | rdma-core 35 |
1708 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1709 +-----------------------+-----------------+-----------------+
1710 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1711 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1712 | | | rdma-core 34 | | rdma-core 34 |
1713 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1714 +-----------------------+-----------------+-----------------+
1715 | Modify Field | | DPDK 21.02 | | DPDK 21.02 |
1716 | | | OFED 5.2 | | OFED 5.2 |
1717 | | | rdma-core 35 | | rdma-core 35 |
1718 | | | ConnectX-5 | | ConnectX-5 |
1719 +-----------------------+-----------------+-----------------+
1720 | Connection tracking | | | | DPDK 21.05 |
1721 | | | N/A | | OFED 5.3 |
1722 | | | | | rdma-core 35 |
1723 | | | | | ConnectX-6 Dx |
1724 +-----------------------+-----------------+-----------------+
1726 .. table:: Minimal SW/HW versions for shared action offload
1729 +-----------------------+-----------------+-----------------+
1730 | Shared Action | with E-Switch | with NIC |
1731 +=======================+=================+=================+
1732 | RSS | | | | DPDK 20.11 |
1733 | | | N/A | | OFED 5.2 |
1734 | | | | | rdma-core 33 |
1735 | | | | | ConnectX-5 |
1736 +-----------------------+-----------------+-----------------+
1737 | Age | | DPDK 20.11 | | DPDK 20.11 |
1738 | | | OFED 5.2 | | OFED 5.2 |
1739 | | | rdma-core 32 | | rdma-core 32 |
1740 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1741 +-----------------------+-----------------+-----------------+
1742 | Count | | DPDK 21.05 | | DPDK 21.05 |
1743 | | | OFED 4.6 | | OFED 4.6 |
1744 | | | rdma-core 24 | | rdma-core 23 |
1745 | | | ConnectX-5 | | ConnectX-5 |
1746 +-----------------------+-----------------+-----------------+
1751 MARK and META items are interrelated with datapath - they might move from/to
1752 the applications in mbuf fields. Hence, zero value for these items has the
1753 special meaning - it means "no metadata are provided", not zero values are
1754 treated by applications and PMD as valid ones.
1756 Moreover in the flow engine domain the value zero is acceptable to match and
1757 set, and we should allow to specify zero values as rte_flow parameters for the
1758 META and MARK items and actions. In the same time zero mask has no meaning and
1759 should be rejected on validation stage.
1764 Flows are not cached in the driver.
1765 When stopping a device port, all the flows created on this port from the
1766 application will be flushed automatically in the background.
1767 After stopping the device port, all flows on this port become invalid and
1768 not represented in the system.
1769 All references to these flows held by the application should be discarded
1770 directly but neither destroyed nor flushed.
1772 The application should re-create the flows as required after the port restart.
1777 Compared to librte_net_mlx4 that implements a single RSS configuration per
1778 port, librte_net_mlx5 supports per-protocol RSS configuration.
1780 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1781 command-line parameter to enable additional protocols (UDP and TCP as well
1782 as IP), the following commands must be entered from its CLI to get the same
1783 behavior as librte_net_mlx4::
1786 > port config all rss all
1792 This section demonstrates how to launch **testpmd** with Mellanox
1793 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1795 #. Load the kernel modules::
1797 modprobe -a ib_uverbs mlx5_core mlx5_ib
1799 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1802 /etc/init.d/openibd restart
1806 User space I/O kernel modules (uio and igb_uio) are not used and do
1807 not have to be loaded.
1809 #. Make sure Ethernet interfaces are in working order and linked to kernel
1810 verbs. Related sysfs entries should be present::
1812 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1821 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1824 for intf in eth2 eth3 eth4 eth5;
1826 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1829 sed -n 's,.*/\(.*\),-a \1,p'
1838 #. Request huge pages::
1840 dpdk-hugepages.py --setup 2G
1842 #. Start testpmd with basic parameters::
1844 dpdk-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
1849 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1850 EAL: probe driver: 15b3:1013 librte_net_mlx5
1851 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1852 PMD: librte_net_mlx5: 1 port(s) detected
1853 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1854 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1855 EAL: probe driver: 15b3:1013 librte_net_mlx5
1856 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1857 PMD: librte_net_mlx5: 1 port(s) detected
1858 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1859 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1860 EAL: probe driver: 15b3:1013 librte_net_mlx5
1861 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1862 PMD: librte_net_mlx5: 1 port(s) detected
1863 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1864 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1865 EAL: probe driver: 15b3:1013 librte_net_mlx5
1866 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1867 PMD: librte_net_mlx5: 1 port(s) detected
1868 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1869 Interactive-mode selected
1870 Configuring Port 0 (socket 0)
1871 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1872 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1873 Port 0: E4:1D:2D:E7:0C:FE
1874 Configuring Port 1 (socket 0)
1875 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1876 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1877 Port 1: E4:1D:2D:E7:0C:FF
1878 Configuring Port 2 (socket 0)
1879 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1880 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1881 Port 2: E4:1D:2D:E7:0C:FA
1882 Configuring Port 3 (socket 0)
1883 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1884 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1885 Port 3: E4:1D:2D:E7:0C:FB
1886 Checking link statuses...
1887 Port 0 Link Up - speed 40000 Mbps - full-duplex
1888 Port 1 Link Up - speed 40000 Mbps - full-duplex
1889 Port 2 Link Up - speed 10000 Mbps - full-duplex
1890 Port 3 Link Up - speed 10000 Mbps - full-duplex
1897 This section demonstrates how to dump flows. Currently, it's possible to dump
1898 all flows with assistance of external tools.
1900 #. 2 ways to get flow raw file:
1902 - Using testpmd CLI:
1904 .. code-block:: console
1907 testpmd> flow dump <port> all <output_file>
1909 testpmd> flow dump <port> rule <rule_id> <output_file>
1911 - call rte_flow_dev_dump api:
1913 .. code-block:: console
1915 rte_flow_dev_dump(port, flow, file, NULL);
1917 #. Dump human-readable flows from raw file:
1919 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1921 .. code-block:: console
1923 mlx_steering_dump.py -f <output_file> -flowptr <flow_ptr>
1925 How to share a meter between ports in the same switch domain
1926 ------------------------------------------------------------
1928 This section demonstrates how to use the shared meter. A meter M can be created
1929 on port X and to be shared with a port Y on the same switch domain by the next way:
1931 .. code-block:: console
1933 flow create X ingress transfer pattern eth / port_id id is Y / end actions meter mtr_id M / end