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 IPv4 Internet Header Length (IHL).
103 - Matching on GTP extension header with raw encap/decap action.
104 - Matching on Geneve TLV option header with raw encap/decap action.
105 - RSS support in sample action.
106 - E-Switch mirroring and jump.
107 - E-Switch mirroring and modify.
108 - 21844 flow priorities for ingress or egress flow groups greater than 0 and for any transfer
110 - Flow metering, including meter policy API.
111 - Flow meter hierarchy.
112 - Flow integrity offload API.
113 - Connection tracking.
114 - Sub-Function representors.
123 On Windows, the features are limited:
125 - Promiscuous mode is not supported
126 - The following rules are supported:
128 - IPv4/UDP with CVLAN filtering
129 - Unicast MAC filtering
131 - Additional rules are supported from WinOF2 version 2.70:
133 - IPv4/TCP with CVLAN filtering
134 - L4 steering rules for port RSS of UDP, TCP and IP
136 - For secondary process:
138 - Forked secondary process not supported.
139 - External memory unregistered in EAL memseg list cannot be used for DMA
140 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
141 primary process and remapped to the same virtual address in secondary
142 process. If the external memory is registered by primary process but has
143 different virtual address in secondary process, unexpected error may happen.
145 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
146 specific VLAN will match for VLAN packets as well:
148 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
149 Meaning, the flow rule::
151 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
153 Will only match vlan packets with vid=3. and the flow rule::
155 flow create 0 ingress pattern eth / ipv4 / end ...
157 Will match any ipv4 packet (VLAN included).
159 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
161 - 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.
164 flow create 0 ingress pattern eth / ipv4 / end ...
166 Will match any ipv4 packet.
169 flow create 0 ingress pattern eth / vlan / end ...
170 flow create 0 ingress pattern eth has_vlan is 1 / end ...
171 flow create 0 ingress pattern eth type is 0x8100 / end ...
173 Will match single-tagged packets only, with any VLAN ID value.
176 flow create 0 ingress pattern eth type is 0x88A8 / end ...
177 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
179 Will match multi-tagged packets only, with any VLAN ID value.
181 - A flow pattern with 2 sequential VLAN items is not supported.
183 - VLAN pop offload command:
185 - Flow rules having a VLAN pop offload command as one of their actions and
186 are lacking a match on VLAN as one of their items are not supported.
187 - The command is not supported on egress traffic in NIC mode.
189 - VLAN push offload is not supported on ingress traffic in NIC mode.
191 - VLAN set PCP offload is not supported on existing headers.
193 - A multi segment packet must have not more segments than reported by dev_infos_get()
194 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
195 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
196 inline settings) to 58.
198 - Match on VXLAN supports the following fields only:
201 - Last reserved 8-bits
203 Last reserved 8-bits matching is only supported When using DV flow
204 engine (``dv_flow_en`` = 1).
205 For ConnectX-5, the UDP destination port must be the standard one (4789).
206 Group zero's behavior may differ which depends on FW.
207 Matching value equals 0 (value & mask) is not supported.
209 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
211 - Match on Geneve header supports the following fields only:
218 - Match on Geneve TLV option is supported on the following fields:
225 Only one Class/Type/Length Geneve TLV option is supported per shared device.
226 Class/Type/Length fields must be specified as well as masks.
227 Class/Type/Length specified masks must be full.
228 Matching Geneve TLV option without specifying data is not supported.
229 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
231 - VF: flow rules created on VF devices can only match traffic targeted at the
232 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
234 - Match on GTP tunnel header item supports the following fields only:
236 - v_pt_rsv_flags: E flag, S flag, PN flag
240 - Match on GTP extension header only for GTP PDU session container (next
241 extension header type = 0x85).
242 - Match on GTP extension header is not supported in group 0.
244 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
245 The flows within group 0 and set metadata action are rejected by hardware.
249 MAC addresses not already present in the bridge table of the associated
250 kernel network device will be added and cleaned up by the PMD when closing
251 the device. In case of ungraceful program termination, some entries may
252 remain present and should be removed manually by other means.
254 - Buffer split offload is supported with regular Rx burst routine only,
255 no MPRQ feature or vectorized code can be engaged.
257 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
258 externally attached to a user-provided mbuf with having RTE_MBUF_F_EXTERNAL in
259 ol_flags. As the mempool for the external buffer is managed by PMD, all the
260 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
261 the external buffers will be freed by PMD and the application which still
262 holds the external buffers may be corrupted.
264 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
265 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
266 supported. Some Rx packets may not have RTE_MBUF_F_RX_RSS_HASH.
268 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
269 and allmulticast mode are both set to off.
270 To receive IPv6 Multicast messages on VM, explicitly set the relevant
271 MAC address using rte_eth_dev_mac_addr_add() API.
273 - To support a mixed traffic pattern (some buffers from local host memory, some
274 buffers from other devices) with high bandwidth, a mbuf flag is used.
276 An application hints the PMD whether or not it should try to inline the
277 given mbuf data buffer. PMD should do the best effort to act upon this request.
279 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
280 registered by application with rte_mbuf_dynflag_register(). This flag is
281 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
282 which is intended to be used by the application.
284 To query the supported specific flags in runtime,
285 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
286 currently (over present hardware and configuration) supported specific flags.
287 The "not inline hint" feature operating flow is the following one:
290 - probe the devices, ports are created
291 - query the port capabilities
292 - if port supporting the feature is found
293 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
294 - application starts the ports
295 - on ``dev_start()`` PMD checks whether the feature flag is registered and
296 enables the feature support in datapath
297 - application might set the registered flag bit in ``ol_flags`` field
298 of mbuf being sent and PMD will handle ones appropriately.
300 - The amount of descriptors in Tx queue may be limited by data inline settings.
301 Inline data require the more descriptor building blocks and overall block
302 amount may exceed the hardware supported limits. The application should
303 reduce the requested Tx size or adjust data inline settings with
304 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
306 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
307 parameter should be specified.
308 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
309 being sent it tries to synchronize the time of packet appearing on
310 the wire with the specified packet timestamp. It the specified one
311 is in the past it should be ignored, if one is in the distant future
312 it should be capped with some reasonable value (in range of seconds).
313 These specific cases ("too late" and "distant future") can be optionally
314 reported via device xstats to assist applications to detect the
315 time-related problems.
317 The timestamp upper "too-distant-future" limit
318 at the moment of invoking the Tx burst routine
319 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
320 Please note, for the testpmd txonly mode,
321 the limit is deduced from the expression::
323 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
325 There is no any packet reordering according timestamps is supposed,
326 neither within packet burst, nor between packets, it is an entirely
327 application responsibility to generate packets and its timestamps
328 in desired order. The timestamps can be put only in the first packet
329 in the burst providing the entire burst scheduling.
331 - E-Switch decapsulation Flow:
333 - can be applied to PF port only.
334 - must specify VF port action (packet redirection from PF to VF).
335 - optionally may specify tunnel inner source and destination MAC addresses.
337 - E-Switch encapsulation Flow:
339 - can be applied to VF ports only.
340 - must specify PF port action (packet redirection from VF to PF).
344 - The input buffer, used as outer header, is not validated.
348 - The decapsulation is always done up to the outermost tunnel detected by the HW.
349 - The input buffer, providing the removal size, is not validated.
350 - The buffer size must match the length of the headers to be removed.
352 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
353 mutually exclusive features which cannot be supported together
354 (see :ref:`mlx5_firmware_config`).
358 - Requires DevX and DV flow to be enabled.
359 - KEEP_CRC offload cannot be supported with LRO.
360 - The first mbuf length, without head-room, must be big enough to include the
362 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
363 it with size limited to max LRO size, not to max RX packet length.
364 - LRO can be used with outer header of TCP packets of the standard format:
365 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
367 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
368 - LRO packet aggregation is performed by HW only for packet size larger than
369 ``lro_min_mss_size``. This value is reported on device start, when debug
374 - ``RTE_ETH_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
375 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
376 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
380 - fast free offload assumes the all mbufs being sent are originated from the
381 same memory pool and there is no any extra references to the mbufs (the
382 reference counter for each mbuf is equal 1 on tx_burst call). The latter
383 means there should be no any externally attached buffers in mbufs. It is
384 an application responsibility to provide the correct mbufs if the fast
385 free offload is engaged. The mlx5 PMD implicitly produces the mbufs with
386 externally attached buffers if MPRQ option is enabled, hence, the fast
387 free offload is neither supported nor advertised if there is MPRQ enabled.
391 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and
392 E-Switch steering domain.
393 - For E-Switch Sampling flow with sample ratio > 1, additional actions are not
394 supported in the sample actions list.
395 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as
396 first action in the E-Switch egress flow if with header modify or
397 encapsulation actions.
398 - For NIC Rx flow, supports ``MARK``, ``COUNT``, ``QUEUE``, ``RSS`` in the
400 - For E-Switch mirroring flow, supports ``RAW ENCAP``, ``Port ID``,
401 ``VXLAN ENCAP``, ``NVGRE ENCAP`` in the sample actions list.
405 - Supports the 'set' operation only for ``RTE_FLOW_ACTION_TYPE_MODIFY_FIELD`` action.
406 - Modification of an arbitrary place in a packet via the special ``RTE_FLOW_FIELD_START`` Field ID is not supported.
407 - Modification of the 802.1Q Tag, VXLAN Network or GENEVE Network ID's is not supported.
408 - Encapsulation levels are not supported, can modify outermost header fields only.
409 - Offsets must be 32-bits aligned, cannot skip past the boundary of a field.
411 - IPv6 header item 'proto' field, indicating the next header protocol, should
412 not be set as extension header.
413 In case the next header is an extension header, it should not be specified in
414 IPv6 header item 'proto' field.
415 The last extension header item 'next header' field can specify the following
416 header protocol type.
420 - 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.
421 - Hairpin in switchdev SR-IOV mode is not supported till now.
425 - All the meter colors with drop action will be counted only by the global drop statistics.
426 - Yellow detection is only supported with ASO metering.
427 - Red color must be with drop action.
428 - Meter statistics are supported only for drop case.
429 - A meter action created with pre-defined policy must be the last action in the flow except single case where the policy actions are:
430 - green: NULL or END.
431 - yellow: NULL or END.
433 - The only supported meter policy actions:
434 - green: QUEUE, RSS, PORT_ID, REPRESENTED_PORT, JUMP, DROP, MARK and SET_TAG.
435 - yellow: QUEUE, RSS, PORT_ID, REPRESENTED_PORT, JUMP, DROP, MARK and SET_TAG.
437 - Policy actions of RSS for green and yellow should have the same configuration except queues.
438 - meter profile packet mode is supported.
439 - meter profiles of RFC2697, RFC2698 and RFC4115 are supported.
443 - Integrity offload is enabled for **ConnectX-6** family.
444 - Verification bits provided by the hardware are ``l3_ok``, ``ipv4_csum_ok``, ``l4_ok``, ``l4_csum_ok``.
445 - ``level`` value 0 references outer headers.
446 - Multiple integrity items not supported in a single flow rule.
447 - Flow rule items supplied by application must explicitly specify network headers referred by integrity item.
448 For example, if integrity item mask sets ``l4_ok`` or ``l4_csum_ok`` bits, reference to L4 network header,
449 TCP or UDP, must be in the rule pattern as well::
451 flow create 0 ingress pattern integrity level is 0 value mask l3_ok value spec l3_ok / eth / ipv6 / end …
453 flow create 0 ingress pattern integrity level is 0 value mask l4_ok value spec 0 / eth / ipv4 proto is udp / end …
455 - Connection tracking:
457 - Cannot co-exist with ASO meter, ASO age action in a single flow rule.
458 - Flow rules insertion rate and memory consumption need more optimization.
460 - 4M connections maximum.
462 - Multi-thread flow insertion:
464 - In order to achieve best insertion rate, application should manage the flows per lcore.
465 - Better to disable memory reclaim by setting ``reclaim_mem_mode`` to 0 to accelerate the flow object allocation and release with cache.
469 - TXQ affinity subjects to HW hash once enabled.
474 MLX5 supports various methods to report statistics:
476 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.
478 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.
480 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.
488 The ibverbs libraries can be linked with this PMD in a number of ways,
489 configured by the ``ibverbs_link`` build option:
491 - ``shared`` (default): the PMD depends on some .so files.
493 - ``dlopen``: Split the dependencies glue in a separate library
494 loaded when needed by dlopen.
495 It make dependencies on libibverbs and libmlx4 optional,
496 and has no performance impact.
498 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
499 in the PMD shared library or the executable static binary.
501 Environment variables
502 ~~~~~~~~~~~~~~~~~~~~~
506 A list of directories in which to search for the rdma-core "glue" plug-in,
507 separated by colons or semi-colons.
509 - ``MLX5_SHUT_UP_BF``
511 Configures HW Tx doorbell register as IO-mapped.
513 By default, the HW Tx doorbell is configured as a write-combining register.
514 The register would be flushed to HW usually when the write-combining buffer
515 becomes full, but it depends on CPU design.
517 Except for vectorized Tx burst routines, a write memory barrier is enforced
518 after updating the register so that the update can be immediately visible to
521 When vectorized Tx burst is called, the barrier is set only if the burst size
522 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
523 variable will bring better latency even though the maximum throughput can
526 Run-time configuration
527 ~~~~~~~~~~~~~~~~~~~~~~
529 - librte_net_mlx5 brings kernel network interfaces up during initialization
530 because it is affected by their state. Forcing them down prevents packets
533 - **ethtool** operations on related kernel interfaces also affect the PMD.
538 In order to run as a non-root user,
539 some capabilities must be granted to the application::
541 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
543 Below are the reasons of the need for each capability:
546 When using physical addresses (PA mode), with Linux >= 4.0,
547 for access to ``/proc/self/pagemap``.
550 For device configuration.
553 For raw ethernet queue allocation through kernel driver.
556 For DMA memory pinning.
561 - ``rxq_cqe_comp_en`` parameter [int]
563 A nonzero value enables the compression of CQE on RX side. This feature
564 allows to save PCI bandwidth and improve performance. Enabled by default.
565 Different compression formats are supported in order to achieve the best
566 performance for different traffic patterns. Default format depends on
567 Multi-Packet Rx queue configuration: Hash RSS format is used in case
568 MPRQ is disabled, Checksum format is used in case MPRQ is enabled.
570 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
571 better compression rate in case of RTE Flow Mark traffic.
572 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
573 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
574 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
575 CQE compression format selection requires DevX to be enabled. If there is
576 no DevX enabled/supported the value is reset to 1 by default.
580 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
581 ConnectX-6 Lx, BlueField and BlueField-2.
582 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
583 ConnectX-6 Lx, BlueField and BlueField-2.
585 - ``rxq_pkt_pad_en`` parameter [int]
587 A nonzero value enables padding Rx packet to the size of cacheline on PCI
588 transaction. This feature would waste PCI bandwidth but could improve
589 performance by avoiding partial cacheline write which may cause costly
590 read-modify-copy in memory transaction on some architectures. Disabled by
595 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
596 ConnectX-6 Lx, BlueField and BlueField-2.
597 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
598 ConnectX-6 Lx, BlueField and BlueField-2.
600 - ``mprq_en`` parameter [int]
602 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
603 configured as Multi-Packet RQ if the total number of Rx queues is
604 ``rxqs_min_mprq`` or more. Disabled by default.
606 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
607 by posting a single large buffer for multiple packets. Instead of posting a
608 buffers per a packet, one large buffer is posted in order to receive multiple
609 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
610 and each stride receives one packet. MPRQ can improve throughput for
611 small-packet traffic.
613 When MPRQ is enabled, MTU can be larger than the size of
614 user-provided mbuf even if RTE_ETH_RX_OFFLOAD_SCATTER isn't enabled. PMD will
615 configure large stride size enough to accommodate MTU as long as
616 device allows. Note that this can waste system memory compared to enabling Rx
617 scatter and multi-segment packet.
619 - ``mprq_log_stride_num`` parameter [int]
621 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
622 strides can reduce PCIe traffic further. If configured value is not in the
623 range of device capability, the default value will be set with a warning
624 message. The default value is 4 which is 16 strides per a buffer, valid only
625 if ``mprq_en`` is set.
627 The size of Rx queue should be bigger than the number of strides.
629 - ``mprq_log_stride_size`` parameter [int]
631 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
632 stride size can save some memory and reduce probability of a depletion of all
633 available strides due to unreleased packets by an application. If configured
634 value is not in the range of device capability, the default value will be set
635 with a warning message. The default value is 11 which is 2048 bytes per a
636 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
637 it is possible for a packet to span across multiple strides. This mode allows
638 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
639 of a packet if Rx scatter is configured) may be required in case there is no
640 space left for a head room at the end of a stride which incurs some
643 - ``mprq_max_memcpy_len`` parameter [int]
645 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
646 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
647 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
648 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
649 A mempool for external buffers will be allocated and managed by PMD. If Rx
650 packet is externally attached, ol_flags field of the mbuf will have
651 RTE_MBUF_F_EXTERNAL and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
652 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
654 - ``rxqs_min_mprq`` parameter [int]
656 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
657 greater or equal to this value. The default value is 12, valid only if
660 - ``txq_inline`` parameter [int]
662 Amount of data to be inlined during TX operations. This parameter is
663 deprecated and converted to the new parameter ``txq_inline_max`` providing
664 partial compatibility.
666 - ``txqs_min_inline`` parameter [int]
668 Enable inline data send only when the number of TX queues is greater or equal
671 This option should be used in combination with ``txq_inline_max`` and
672 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
674 If this option is not specified the default value 16 is used for BlueField
675 and 8 for other platforms
677 The data inlining consumes the CPU cycles, so this option is intended to
678 auto enable inline data if we have enough Tx queues, which means we have
679 enough CPU cores and PCI bandwidth is getting more critical and CPU
680 is not supposed to be bottleneck anymore.
682 The copying data into WQE improves latency and can improve PPS performance
683 when PCI back pressure is detected and may be useful for scenarios involving
684 heavy traffic on many queues.
686 Because additional software logic is necessary to handle this mode, this
687 option should be used with care, as it may lower performance when back
688 pressure is not expected.
690 If inline data are enabled it may affect the maximal size of Tx queue in
691 descriptors because the inline data increase the descriptor size and
692 queue size limits supported by hardware may be exceeded.
694 - ``txq_inline_min`` parameter [int]
696 Minimal amount of data to be inlined into WQE during Tx operations. NICs
697 may require this minimal data amount to operate correctly. The exact value
698 may depend on NIC operation mode, requested offloads, etc. It is strongly
699 recommended to omit this parameter and use the default values. Anyway,
700 applications using this parameter should take into consideration that
701 specifying an inconsistent value may prevent the NIC from sending packets.
703 If ``txq_inline_min`` key is present the specified value (may be aligned
704 by the driver in order not to exceed the limits and provide better descriptor
705 space utilization) will be used by the driver and it is guaranteed that
706 requested amount of data bytes are inlined into the WQE beside other inline
707 settings. This key also may update ``txq_inline_max`` value (default
708 or specified explicitly in devargs) to reserve the space for inline data.
710 If ``txq_inline_min`` key is not present, the value may be queried by the
711 driver from the NIC via DevX if this feature is available. If there is no DevX
712 enabled/supported the value 18 (supposing L2 header including VLAN) is set
713 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
714 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
717 For ConnectX-4 NIC, driver does not allow specifying value below 18
718 (minimal L2 header, including VLAN), error will be raised.
720 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
721 it is not recommended and may prevent NIC from sending packets over
724 For ConnectX-4 and ConnectX-4 Lx NICs, automatically configured value
725 is insufficient for some traffic, because they require at least all L2 headers
726 to be inlined. For example, Q-in-Q adds 4 bytes to default 18 bytes
727 of Ethernet and VLAN, thus ``txq_inline_min`` must be set to 22.
728 MPLS would add 4 bytes per label. Final value must account for all possible
729 L2 encapsulation headers used in particular environment.
731 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
732 Multi-Packet Write), because last one does not support partial packet inlining.
733 This is not very critical due to minimal data inlining is mostly required
734 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
736 - ``txq_inline_max`` parameter [int]
738 Specifies the maximal packet length to be completely inlined into WQE
739 Ethernet Segment for ordinary SEND method. If packet is larger than specified
740 value, the packet data won't be copied by the driver at all, data buffer
741 is addressed with a pointer. If packet length is less or equal all packet
742 data will be copied into WQE. This may improve PCI bandwidth utilization for
743 short packets significantly but requires the extra CPU cycles.
745 The data inline feature is controlled by number of Tx queues, if number of Tx
746 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
747 is engaged, if there are not enough Tx queues (which means not enough CPU cores
748 and CPU resources are scarce), data inline is not performed by the driver.
749 Assigning ``txqs_min_inline`` with zero always enables the data inline.
751 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
752 by the driver in order not to exceed the limit (930 bytes) and to provide better
753 WQE space filling without gaps, the adjustment is reflected in the debug log.
754 Also, the default value (290) may be decreased in run-time if the large transmit
755 queue size is requested and hardware does not support enough descriptor
756 amount, in this case warning is emitted. If ``txq_inline_max`` key is
757 specified and requested inline settings can not be satisfied then error
760 - ``txq_inline_mpw`` parameter [int]
762 Specifies the maximal packet length to be completely inlined into WQE for
763 Enhanced MPW method. If packet is large the specified value, the packet data
764 won't be copied, and data buffer is addressed with pointer. If packet length
765 is less or equal, all packet data will be copied into WQE. This may improve PCI
766 bandwidth utilization for short packets significantly but requires the extra
769 The data inline feature is controlled by number of TX queues, if number of Tx
770 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
771 is engaged, if there are not enough Tx queues (which means not enough CPU cores
772 and CPU resources are scarce), data inline is not performed by the driver.
773 Assigning ``txqs_min_inline`` with zero always enables the data inline.
775 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
776 by the driver in order not to exceed the limit (930 bytes) and to provide better
777 WQE space filling without gaps, the adjustment is reflected in the debug log.
778 Due to multiple packets may be included to the same WQE with Enhanced Multi
779 Packet Write Method and overall WQE size is limited it is not recommended to
780 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
781 may be decreased in run-time if the large transmit queue size is requested
782 and hardware does not support enough descriptor amount, in this case warning
783 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
784 settings can not be satisfied then error will be raised.
786 - ``txqs_max_vec`` parameter [int]
788 Enable vectorized Tx only when the number of TX queues is less than or
789 equal to this value. This parameter is deprecated and ignored, kept
790 for compatibility issue to not prevent driver from probing.
792 - ``txq_mpw_hdr_dseg_en`` parameter [int]
794 A nonzero value enables including two pointers in the first block of TX
795 descriptor. The parameter is deprecated and ignored, kept for compatibility
798 - ``txq_max_inline_len`` parameter [int]
800 Maximum size of packet to be inlined. This limits the size of packet to
801 be inlined. If the size of a packet is larger than configured value, the
802 packet isn't inlined even though there's enough space remained in the
803 descriptor. Instead, the packet is included with pointer. This parameter
804 is deprecated and converted directly to ``txq_inline_mpw`` providing full
805 compatibility. Valid only if eMPW feature is engaged.
807 - ``txq_mpw_en`` parameter [int]
809 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
810 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
811 eMPW allows the Tx burst function to pack up multiple packets
812 in a single descriptor session in order to save PCI bandwidth
813 and improve performance at the cost of a slightly higher CPU usage.
814 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
815 Tx burst function copies entire packet data on to Tx descriptor
816 instead of including pointer of packet.
818 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
819 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
820 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
821 option or reported by the NIC, the eMPW feature is disengaged.
823 - ``tx_db_nc`` parameter [int]
825 The rdma core library can map doorbell register in two ways, depending on the
826 environment variable "MLX5_SHUT_UP_BF":
828 - As regular cached memory (usually with write combining attribute), if the
829 variable is either missing or set to zero.
830 - As non-cached memory, if the variable is present and set to not "0" value.
832 The type of mapping may slightly affect the Tx performance, the optimal choice
833 is strongly relied on the host architecture and should be deduced practically.
835 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
836 memory (with write combining), the PMD will perform the extra write memory barrier
837 after writing to doorbell, it might increase the needed CPU clocks per packet
838 to send, but latency might be improved.
840 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
841 cached memory, the PMD will not perform the extra write memory barrier
842 after writing to doorbell, on some architectures it might improve the
845 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
846 memory, the PMD will use heuristics to decide whether write memory barrier
847 should be performed. For bursts with size multiple of recommended one (64 pkts)
848 it is supposed the next burst is coming and no need to issue the extra memory
849 barrier (it is supposed to be issued in the next coming burst, at least after
850 descriptor writing). It might increase latency (on some hosts till next
851 packets transmit) and should be used with care.
853 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
854 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
855 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
857 - ``tx_pp`` parameter [int]
859 If a nonzero value is specified the driver creates all necessary internal
860 objects to provide accurate packet send scheduling on mbuf timestamps.
861 The positive value specifies the scheduling granularity in nanoseconds,
862 the packet send will be accurate up to specified digits. The allowed range is
863 from 500 to 1 million of nanoseconds. The negative value specifies the module
864 of granularity and engages the special test mode the check the schedule rate.
865 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
868 - ``tx_skew`` parameter [int]
870 The parameter adjusts the send packet scheduling on timestamps and represents
871 the average delay between beginning of the transmitting descriptor processing
872 by the hardware and appearance of actual packet data on the wire. The value
873 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
874 specified. The default value is zero.
876 - ``tx_vec_en`` parameter [int]
878 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
879 ConnectX-6 Lx, BlueField and BlueField-2 NICs
880 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
881 The parameter is deprecated and ignored.
883 - ``rx_vec_en`` parameter [int]
885 A nonzero value enables Rx vector if the port is not configured in
886 multi-segment otherwise this parameter is ignored.
890 - ``vf_nl_en`` parameter [int]
892 A nonzero value enables Netlink requests from the VF to add/remove MAC
893 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
894 Otherwise the relevant configuration must be run with Linux iproute2 tools.
895 This is a prerequisite to receive this kind of traffic.
897 Enabled by default, valid only on VF devices ignored otherwise.
899 - ``l3_vxlan_en`` parameter [int]
901 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
902 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
903 parameter. This is a prerequisite to receive this kind of traffic.
907 - ``dv_xmeta_en`` parameter [int]
909 A nonzero value enables extensive flow metadata support if device is
910 capable and driver supports it. This can enable extensive support of
911 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
912 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
914 There are some possible configurations, depending on parameter value:
916 - 0, this is default value, defines the legacy mode, the ``MARK`` and
917 ``META`` related actions and items operate only within NIC Tx and
918 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
919 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
920 item is 32 bits wide and match supported on egress only.
922 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
923 related actions and items operate within all supported steering domains,
924 including FDB, ``MARK`` and ``META`` information may cross the domain
925 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
926 depends on kernel and firmware configurations and might be 0, 16 or
927 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
928 compatibility, the actual width of data transferred to the FDB domain
929 depends on kernel configuration and may be vary. The actual supported
930 width can be retrieved in runtime by series of rte_flow_validate()
933 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
934 related actions and items operate within all supported steering domains,
935 including FDB, ``MARK`` and ``META`` information may cross the domain
936 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
937 depends on kernel and firmware configurations and might be 0, 16 or
938 24 bits. The actual supported width can be retrieved in runtime by
939 series of rte_flow_validate() trials.
941 - 3, this engages tunnel offload mode. In E-Switch configuration, that
942 mode implicitly activates ``dv_xmeta_en=1``.
944 +------+-----------+-----------+-------------+-------------+
945 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
946 +======+===========+===========+=============+=============+
947 | 0 | 24 bits | 32 bits | 32 bits | no |
948 +------+-----------+-----------+-------------+-------------+
949 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
950 +------+-----------+-----------+-------------+-------------+
951 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
952 +------+-----------+-----------+-------------+-------------+
954 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
955 ignored and the device is configured to operate in legacy mode (0).
957 Disabled by default (set to 0).
959 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
960 of the extensive metadata features. The legacy Verbs supports FLAG and
961 MARK metadata actions over NIC Rx steering domain only.
963 Setting META value to zero in flow action means there is no item provided
964 and receiving datapath will not report in mbufs the metadata are present.
965 Setting MARK value to zero in flow action means the zero FDIR ID value
966 will be reported on packet receiving.
968 For the MARK action the last 16 values in the full range are reserved for
969 internal PMD purposes (to emulate FLAG action). The valid range for the
970 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
971 for the 24-bit mode, the flows with the MARK action value outside
972 the specified range will be rejected.
974 - ``dv_flow_en`` parameter [int]
976 A nonzero value enables the DV flow steering assuming it is supported
977 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
979 Enabled by default if supported.
981 - ``dv_esw_en`` parameter [int]
983 A nonzero value enables E-Switch using Direct Rules.
985 Enabled by default if supported.
987 - ``lacp_by_user`` parameter [int]
989 A nonzero value enables the control of LACP traffic by the user application.
990 When a bond exists in the driver, by default it should be managed by the
991 kernel and therefore LACP traffic should be steered to the kernel.
992 If this devarg is set to 1 it will allow the user to manage the bond by
993 itself and not steer LACP traffic to the kernel.
995 Disabled by default (set to 0).
997 - ``mr_ext_memseg_en`` parameter [int]
999 A nonzero value enables extending memseg when registering DMA memory. If
1000 enabled, the number of entries in MR (Memory Region) lookup table on datapath
1001 is minimized and it benefits performance. On the other hand, it worsens memory
1002 utilization because registered memory is pinned by kernel driver. Even if a
1003 page in the extended chunk is freed, that doesn't become reusable until the
1004 entire memory is freed.
1008 - ``mr_mempool_reg_en`` parameter [int]
1010 A nonzero value enables implicit registration of DMA memory of all mempools
1011 except those having ``RTE_MEMPOOL_F_NON_IO``. This flag is set automatically
1012 for mempools populated with non-contiguous objects or those without IOVA.
1013 The effect is that when a packet from a mempool is transmitted,
1014 its memory is already registered for DMA in the PMD and no registration
1015 will happen on the data path. The tradeoff is extra work on the creation
1016 of each mempool and increased HW resource use if some mempools
1017 are not used with MLX5 devices.
1021 - ``representor`` parameter [list]
1023 This parameter can be used to instantiate DPDK Ethernet devices from
1024 existing port (PF, VF or SF) representors configured on the device.
1026 It is a standard parameter whose format is described in
1027 :ref:`ethernet_device_standard_device_arguments`.
1029 For instance, to probe VF port representors 0 through 2::
1031 <PCI_BDF>,representor=vf[0-2]
1033 To probe SF port representors 0 through 2::
1035 <PCI_BDF>,representor=sf[0-2]
1037 To probe VF port representors 0 through 2 on both PFs of bonding device::
1039 <Primary_PCI_BDF>,representor=pf[0,1]vf[0-2]
1041 - ``max_dump_files_num`` parameter [int]
1043 The maximum number of files per PMD entity that may be created for debug information.
1044 The files will be created in /var/log directory or in current directory.
1046 set to 128 by default.
1048 - ``lro_timeout_usec`` parameter [int]
1050 The maximum allowed duration of an LRO session, in micro-seconds.
1051 PMD will set the nearest value supported by HW, which is not bigger than
1052 the input ``lro_timeout_usec`` value.
1053 If this parameter is not specified, by default PMD will set
1054 the smallest value supported by HW.
1056 - ``hp_buf_log_sz`` parameter [int]
1058 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
1059 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
1060 The capacity of the value is specified by the firmware and the initialization
1061 will get a failure if it is out of scope.
1062 The range of the value is from 11 to 19 right now, and the supported frame
1063 size of a single packet for hairpin is from 512B to 128KB. It might change if
1064 different firmware release is being used. By using a small value, it could
1065 reduce memory consumption but not work with a large frame. If the value is
1066 too large, the memory consumption will be high and some potential performance
1067 degradation will be introduced.
1068 By default, the PMD will set this value to 16, which means that 9KB jumbo
1069 frames will be supported.
1071 - ``reclaim_mem_mode`` parameter [int]
1073 Cache some resources in flow destroy will help flow recreation more efficient.
1074 While some systems may require the all the resources can be reclaimed after
1076 The parameter ``reclaim_mem_mode`` provides the option for user to configure
1077 if the resource cache is needed or not.
1079 There are three options to choose:
1081 - 0. It means the flow resources will be cached as usual. The resources will
1082 be cached, helpful with flow insertion rate.
1084 - 1. It will only enable the DPDK PMD level resources reclaim.
1086 - 2. Both DPDK PMD level and rdma-core low level will be configured as
1089 By default, the PMD will set this value to 0.
1091 - ``sys_mem_en`` parameter [int]
1093 A non-zero value enables the PMD memory management allocating memory
1094 from system by default, without explicit rte memory flag.
1096 By default, the PMD will set this value to 0.
1098 - ``decap_en`` parameter [int]
1100 Some devices do not support FCS (frame checksum) scattering for
1101 tunnel-decapsulated packets.
1102 If set to 0, this option forces the FCS feature and rejects tunnel
1103 decapsulation in the flow engine for such devices.
1105 By default, the PMD will set this value to 1.
1107 - ``allow_duplicate_pattern`` parameter [int]
1109 There are two options to choose:
1111 - 0. Prevent insertion of rules with the same pattern items on non-root table.
1112 In this case, only the first rule is inserted and the following rules are
1113 rejected and error code EEXIST is returned.
1115 - 1. Allow insertion of rules with the same pattern items.
1116 In this case, all rules are inserted but only the first rule takes effect,
1117 the next rule takes effect only if the previous rules are deleted.
1119 By default, the PMD will set this value to 1.
1121 .. _mlx5_firmware_config:
1123 Firmware configuration
1124 ~~~~~~~~~~~~~~~~~~~~~~
1126 Firmware features can be configured as key/value pairs.
1128 The command to set a value is::
1130 mlxconfig -d <device> set <key>=<value>
1132 The command to query a value is::
1134 mlxconfig -d <device> query | grep <key>
1136 The device name for the command ``mlxconfig`` can be either the PCI address,
1137 or the mst device name found with::
1141 Below are some firmware configurations listed.
1147 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1153 - maximum number of SR-IOV virtual functions::
1157 - enable DevX (required by Direct Rules and other features)::
1161 - aggressive CQE zipping::
1165 - L3 VXLAN and VXLAN-GPE destination UDP port::
1168 IP_OVER_VXLAN_PORT=<udp dport>
1170 - enable VXLAN-GPE tunnel flow matching::
1172 FLEX_PARSER_PROFILE_ENABLE=0
1174 FLEX_PARSER_PROFILE_ENABLE=2
1176 - enable IP-in-IP tunnel flow matching::
1178 FLEX_PARSER_PROFILE_ENABLE=0
1180 - enable MPLS flow matching::
1182 FLEX_PARSER_PROFILE_ENABLE=1
1184 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1186 FLEX_PARSER_PROFILE_ENABLE=2
1188 - enable Geneve flow matching::
1190 FLEX_PARSER_PROFILE_ENABLE=0
1192 FLEX_PARSER_PROFILE_ENABLE=1
1194 - enable Geneve TLV option flow matching::
1196 FLEX_PARSER_PROFILE_ENABLE=0
1198 - enable GTP flow matching::
1200 FLEX_PARSER_PROFILE_ENABLE=3
1202 - enable eCPRI flow matching::
1204 FLEX_PARSER_PROFILE_ENABLE=4
1210 This driver relies on external libraries and kernel drivers for resources
1211 allocations and initialization. The following dependencies are not part of
1212 DPDK and must be installed separately:
1216 User space Verbs framework used by librte_net_mlx5. This library provides
1217 a generic interface between the kernel and low-level user space drivers
1220 It allows slow and privileged operations (context initialization, hardware
1221 resources allocations) to be managed by the kernel and fast operations to
1222 never leave user space.
1226 Low-level user space driver library for Mellanox
1227 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1230 This library basically implements send/receive calls to the hardware
1233 - **Kernel modules**
1235 They provide the kernel-side Verbs API and low level device drivers that
1236 manage actual hardware initialization and resources sharing with user
1239 Unlike most other PMDs, these modules must remain loaded and bound to
1242 - mlx5_core: hardware driver managing Mellanox
1243 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1245 - mlx5_ib: InifiniBand device driver.
1246 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1248 - **Firmware update**
1250 Mellanox OFED/EN releases include firmware updates for
1251 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1253 Because each release provides new features, these updates must be applied to
1254 match the kernel modules and libraries they come with.
1258 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1264 Either RDMA Core library with a recent enough Linux kernel release
1265 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1268 RDMA Core with Linux Kernel
1269 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1271 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1272 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1273 (see `RDMA Core installation documentation`_)
1274 - When building for i686 use:
1276 - rdma-core version 18.0 or above built with 32bit support.
1277 - Kernel version 4.14.41 or above.
1279 - Starting with rdma-core v21, static libraries can be built::
1282 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1285 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1286 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1292 - Mellanox OFED version: **4.5** and above /
1293 Mellanox EN version: **4.5** and above
1296 - ConnectX-4: **12.21.1000** and above.
1297 - ConnectX-4 Lx: **14.21.1000** and above.
1298 - ConnectX-5: **16.21.1000** and above.
1299 - ConnectX-5 Ex: **16.21.1000** and above.
1300 - ConnectX-6: **20.27.0090** and above.
1301 - ConnectX-6 Dx: **22.27.0090** and above.
1302 - BlueField: **18.25.1010** and above.
1304 While these libraries and kernel modules are available on OpenFabrics
1305 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1306 managers on most distributions, this PMD requires Ethernet extensions that
1307 may not be supported at the moment (this is a work in progress).
1310 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1312 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1313 include the necessary support and should be used in the meantime. For DPDK,
1314 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1315 required from that distribution.
1319 Several versions of Mellanox OFED/EN are available. Installing the version
1320 this DPDK release was developed and tested against is strongly
1321 recommended. Please check the `linux prerequisites`_.
1323 Windows Prerequisites
1324 ---------------------
1326 This driver relies on external libraries and kernel drivers for resources
1327 allocations and initialization. The dependencies in the following sub-sections
1328 are not part of DPDK, and must be installed separately.
1330 Compilation Prerequisites
1331 ~~~~~~~~~~~~~~~~~~~~~~~~~
1333 DevX SDK installation
1334 ^^^^^^^^^^^^^^^^^^^^^
1336 The DevX SDK must be installed on the machine building the Windows PMD.
1337 Additional information can be found at
1338 `How to Integrate Windows DevX in Your Development Environment
1339 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1341 Runtime Prerequisites
1342 ~~~~~~~~~~~~~~~~~~~~~
1344 WinOF2 version 2.60 or higher must be installed on the machine.
1349 The driver can be downloaded from the following site:
1351 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1356 DevX for Windows must be enabled in the Windows registry.
1357 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1358 Additional information can be found in the WinOF2 user manual.
1363 The following Mellanox device families are supported by the same mlx5 driver:
1375 Below are detailed device names:
1377 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1378 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1379 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1380 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1381 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1382 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1383 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1384 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1385 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1386 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1387 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1388 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1389 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1390 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1391 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1392 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1393 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1394 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1395 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1396 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1397 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1398 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1399 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1400 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1401 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1402 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1404 Quick Start Guide on OFED/EN
1405 ----------------------------
1407 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1410 2. Install the required libraries and kernel modules either by installing
1411 only the required set, or by installing the entire Mellanox OFED/EN::
1413 ./mlnxofedinstall --upstream-libs --dpdk
1415 3. Verify the firmware is the correct one::
1419 4. Verify all ports links are set to Ethernet::
1421 mlxconfig -d <mst device> query | grep LINK_TYPE
1425 Link types may have to be configured to Ethernet::
1427 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1429 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1431 For hypervisors, verify SR-IOV is enabled on the NIC::
1433 mlxconfig -d <mst device> query | grep SRIOV_EN
1436 If needed, configure SR-IOV::
1438 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1439 mlxfwreset -d <mst device> reset
1441 5. Restart the driver::
1443 /etc/init.d/openibd restart
1447 service openibd restart
1449 If link type was changed, firmware must be reset as well::
1451 mlxfwreset -d <mst device> reset
1453 For hypervisors, after reset write the sysfs number of virtual functions
1456 To dynamically instantiate a given number of virtual functions (VFs)::
1458 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1460 6. Install DPDK and you are ready to go.
1461 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1463 Enable switchdev mode
1464 ---------------------
1466 Switchdev mode is a mode in E-Switch, that binds between representor and VF or SF.
1467 Representor is a port in DPDK that is connected to a VF or SF in such a way
1468 that assuming there are no offload flows, each packet that is sent from the VF or SF
1469 will be received by the corresponding representor. While each packet that is or SF
1470 sent to a representor will be received by the VF or SF.
1471 This is very useful in case of SRIOV mode, where the first packet that is sent
1472 by the VF or SF will be received by the DPDK application which will decide if this
1473 flow should be offloaded to the E-Switch. After offloading the flow packet
1474 that the VF or SF that are matching the flow will not be received any more by
1475 the DPDK application.
1477 1. Enable SRIOV mode::
1479 mlxconfig -d <mst device> set SRIOV_EN=true
1481 2. Configure the max number of VFs::
1483 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1487 mlxfwreset -d <mst device> reset
1489 3. Configure the actual number of VFs::
1491 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1493 4. Unbind the device (can be rebind after the switchdev mode)::
1495 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1497 5. Enable switchdev mode::
1499 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1501 Sub-Function support
1502 --------------------
1504 Sub-Function is a portion of the PCI device, a SF netdev has its own
1505 dedicated queues (txq, rxq).
1506 A SF shares PCI level resources with other SFs and/or with its parent PCI function.
1510 OFED version >= 5.4-0.3.3.0
1512 1. Configure SF feature::
1514 # Run mlxconfig on both PFs on host and ECPFs on BlueField.
1515 mlxconfig -d <mst device> set PER_PF_NUM_SF=1 PF_TOTAL_SF=252 PF_SF_BAR_SIZE=12
1517 2. Enable switchdev mode::
1519 mlxdevm dev eswitch set pci/<DBDF> mode switchdev
1523 mlxdevm port add pci/<DBDF> flavour pcisf pfnum 0 sfnum <sfnum>
1525 Get SFID from output: pci/<DBDF>/<SFID>
1527 4. Modify MAC address::
1529 mlxdevm port function set pci/<DBDF>/<SFID> hw_addr <MAC>
1531 5. Activate SF port::
1533 mlxdevm port function set pci/<DBDF>/<ID> state active
1535 6. Devargs to probe SF device::
1537 auxiliary:mlx5_core.sf.<num>,dv_flow_en=1
1539 Sub-Function representor support
1540 --------------------------------
1542 A SF netdev supports E-Switch representation offload
1543 similar to PF and VF representors.
1544 Use <sfnum> to probe SF representor::
1546 testpmd> port attach <PCI_BDF>,representor=sf<sfnum>,dv_flow_en=1
1551 1. Configure aggressive CQE Zipping for maximum performance::
1553 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1555 To set it back to the default CQE Zipping mode use::
1557 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1559 2. In case of virtualization:
1561 - Make sure that hypervisor kernel is 3.16 or newer.
1562 - Configure boot with ``iommu=pt``.
1563 - Use 1G huge pages.
1564 - Make sure to allocate a VM on huge pages.
1565 - Make sure to set CPU pinning.
1567 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1568 for better performance. For VMs, verify that the right CPU
1569 and NUMA node are pinned according to the above. Run::
1571 lstopo-no-graphics --merge
1573 to identify the NUMA node to which the PCIe adapter is connected.
1575 4. If more than one adapter is used, and root complex capabilities allow
1576 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1577 it is recommended to locate both adapters on the same NUMA node.
1578 This in order to forward packets from one to the other without
1579 NUMA performance penalty.
1581 5. Disable pause frames::
1583 ethtool -A <netdev> rx off tx off
1585 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1586 via the BIOS configuration. Please contact you server provider for more
1587 information about the settings.
1591 On some machines, depends on the machine integrator, it is beneficial
1592 to set the PCI max read request parameter to 1K. This can be
1593 done in the following way:
1595 To query the read request size use::
1597 setpci -s <NIC PCI address> 68.w
1599 If the output is different than 3XXX, set it by::
1601 setpci -s <NIC PCI address> 68.w=3XXX
1603 The XXX can be different on different systems. Make sure to configure
1604 according to the setpci output.
1606 7. To minimize overhead of searching Memory Regions:
1608 - '--socket-mem' is recommended to pin memory by predictable amount.
1609 - Configure per-lcore cache when creating Mempools for packet buffer.
1610 - Refrain from dynamically allocating/freeing memory in run-time.
1615 There are multiple Rx burst functions with different advantages and limitations.
1617 .. table:: Rx burst functions
1619 +-------------------+------------------------+---------+-----------------+------+-------+
1620 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1621 | | | | || comp|| MTU |
1622 +===================+========================+=========+=================+======+=======+
1623 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1624 +-------------------+------------------------+---------+-----------------+------+-------+
1625 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1626 +-------------------+------------------------+---------+-----------------+------+-------+
1627 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1628 | || RxQs >= rxqs_min_mprq | | | | |
1629 +-------------------+------------------------+---------+-----------------+------+-------+
1630 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1631 | || mprq_en=1 | | | | |
1632 | || RxQs >= rxqs_min_mprq | | | | |
1633 +-------------------+------------------------+---------+-----------------+------+-------+
1635 .. _mlx5_offloads_support:
1637 Supported hardware offloads
1638 ---------------------------
1640 .. table:: Minimal SW/HW versions for queue offloads
1642 ============== ===== ===== ========= ===== ========== =============
1643 Offload DPDK Linux rdma-core OFED firmware hardware
1644 ============== ===== ===== ========= ===== ========== =============
1645 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1646 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1647 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1648 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1649 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1650 Tx scheduling 20.08 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1651 Buffer Split 20.11 N/A N/A 5.1-2 16.28.2006 ConnectX-5
1652 ============== ===== ===== ========= ===== ========== =============
1654 .. table:: Minimal SW/HW versions for rte_flow offloads
1656 +-----------------------+-----------------+-----------------+
1657 | Offload | with E-Switch | with NIC |
1658 +=======================+=================+=================+
1659 | Count | | DPDK 19.05 | | DPDK 19.02 |
1660 | | | OFED 4.6 | | OFED 4.6 |
1661 | | | rdma-core 24 | | rdma-core 23 |
1662 | | | ConnectX-5 | | ConnectX-5 |
1663 +-----------------------+-----------------+-----------------+
1664 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1665 | | | OFED 4.6 | | OFED 4.5 |
1666 | | | rdma-core 24 | | rdma-core 23 |
1667 | | | ConnectX-5 | | ConnectX-4 |
1668 +-----------------------+-----------------+-----------------+
1669 | Queue / RSS | | | | DPDK 18.11 |
1670 | | | N/A | | OFED 4.5 |
1671 | | | | | rdma-core 23 |
1672 | | | | | ConnectX-4 |
1673 +-----------------------+-----------------+-----------------+
1674 | Shared action | | | | |
1675 | | | :numref:`sact`| | :numref:`sact`|
1678 +-----------------------+-----------------+-----------------+
1679 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1680 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1681 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1682 | | of_set_vlan_pcp / | | | | |
1683 | | of_set_vlan_vid) | | | | |
1684 +-----------------------+-----------------+-----------------+
1685 | | VLAN | | DPDK 21.05 | | |
1686 | | ingress and / | | OFED 5.3 | | N/A |
1687 | | of_push_vlan / | | ConnectX-6 Dx | | |
1688 +-----------------------+-----------------+-----------------+
1689 | | VLAN | | DPDK 21.05 | | |
1690 | | egress and / | | OFED 5.3 | | N/A |
1691 | | of_pop_vlan / | | ConnectX-6 Dx | | |
1692 +-----------------------+-----------------+-----------------+
1693 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1694 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1695 | | | rdma-core 24 | | rdma-core 23 |
1696 | | | ConnectX-5 | | ConnectX-5 |
1697 +-----------------------+-----------------+-----------------+
1698 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1699 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1700 | | | rdma-core 27 | | rdma-core 27 |
1701 | | | ConnectX-5 | | ConnectX-5 |
1702 +-----------------------+-----------------+-----------------+
1703 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1704 | | | OFED 5.1-2 | | OFED 5.1-2 |
1705 | | | rdma-core 32 | | N/A |
1706 | | | ConnectX-5 | | ConnectX-5 |
1707 +-----------------------+-----------------+-----------------+
1708 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1709 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1710 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1711 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1712 | | set_ipv6_dst / | | | | |
1713 | | set_tp_src / | | | | |
1714 | | set_tp_dst / | | | | |
1715 | | dec_ttl / | | | | |
1716 | | set_ttl / | | | | |
1717 | | set_mac_src / | | | | |
1718 | | set_mac_dst) | | | | |
1719 +-----------------------+-----------------+-----------------+
1720 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1721 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1722 | | | | rdma-core 24 | | rdma-core 24 |
1723 | | | | ConnectX-5 | | ConnectX-5 |
1724 +-----------------------+-----------------+-----------------+
1725 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1726 | | | OFED 4.7-1 | | OFED 4.7-1 |
1727 | | | rdma-core 24 | | N/A |
1728 | | | ConnectX-5 | | ConnectX-5 |
1729 +-----------------------+-----------------+-----------------+
1730 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1731 | | | OFED 4.6 | | OFED 4.5 |
1732 | | | rdma-core 24 | | rdma-core 23 |
1733 | | | ConnectX-5 | | ConnectX-4 |
1734 +-----------------------+-----------------+-----------------+
1735 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1736 | | | OFED 4.7-3 | | OFED 4.7-3 |
1737 | | | rdma-core 26 | | rdma-core 26 |
1738 | | | ConnectX-5 | | ConnectX-5 |
1739 +-----------------------+-----------------+-----------------+
1740 | Port ID | | DPDK 19.05 | | N/A |
1741 | | | OFED 4.7-1 | | N/A |
1742 | | | rdma-core 24 | | N/A |
1743 | | | ConnectX-5 | | N/A |
1744 +-----------------------+-----------------+-----------------+
1745 | Hairpin | | | | DPDK 19.11 |
1746 | | | N/A | | OFED 4.7-3 |
1747 | | | | | rdma-core 26 |
1748 | | | | | ConnectX-5 |
1749 +-----------------------+-----------------+-----------------+
1750 | 2-port Hairpin | | | | DPDK 20.11 |
1751 | | | N/A | | OFED 5.1-2 |
1753 | | | | | ConnectX-5 |
1754 +-----------------------+-----------------+-----------------+
1755 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1756 | | | OFED 4.7-3 | | OFED 4.7-3 |
1757 | | | rdma-core 26 | | rdma-core 26 |
1758 | | | ConnectX-5 | | ConnectX-5 |
1759 +-----------------------+-----------------+-----------------+
1760 | ASO Metering | | DPDK 21.05 | | DPDK 21.05 |
1761 | | | OFED 5.3 | | OFED 5.3 |
1762 | | | rdma-core 33 | | rdma-core 33 |
1763 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1764 +-----------------------+-----------------+-----------------+
1765 | Metering Hierarchy | | DPDK 21.08 | | DPDK 21.08 |
1766 | | | OFED 5.3 | | OFED 5.3 |
1768 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1769 +-----------------------+-----------------+-----------------+
1770 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1771 | | | OFED 5.1-2 | | OFED 5.1-2 |
1772 | | | rdma-core 32 | | N/A |
1773 | | | ConnectX-5 | | ConnectX-5 |
1774 +-----------------------+-----------------+-----------------+
1775 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1776 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1777 | | | rdma-core 35 | | rdma-core 35 |
1778 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1779 +-----------------------+-----------------+-----------------+
1780 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1781 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1782 | | | rdma-core 34 | | rdma-core 34 |
1783 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1784 +-----------------------+-----------------+-----------------+
1785 | Modify Field | | DPDK 21.02 | | DPDK 21.02 |
1786 | | | OFED 5.2 | | OFED 5.2 |
1787 | | | rdma-core 35 | | rdma-core 35 |
1788 | | | ConnectX-5 | | ConnectX-5 |
1789 +-----------------------+-----------------+-----------------+
1790 | Connection tracking | | | | DPDK 21.05 |
1791 | | | N/A | | OFED 5.3 |
1792 | | | | | rdma-core 35 |
1793 | | | | | ConnectX-6 Dx |
1794 +-----------------------+-----------------+-----------------+
1796 .. table:: Minimal SW/HW versions for shared action offload
1799 +-----------------------+-----------------+-----------------+
1800 | Shared Action | with E-Switch | with NIC |
1801 +=======================+=================+=================+
1802 | RSS | | | | DPDK 20.11 |
1803 | | | N/A | | OFED 5.2 |
1804 | | | | | rdma-core 33 |
1805 | | | | | ConnectX-5 |
1806 +-----------------------+-----------------+-----------------+
1807 | Age | | DPDK 20.11 | | DPDK 20.11 |
1808 | | | OFED 5.2 | | OFED 5.2 |
1809 | | | rdma-core 32 | | rdma-core 32 |
1810 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1811 +-----------------------+-----------------+-----------------+
1812 | Count | | DPDK 21.05 | | DPDK 21.05 |
1813 | | | OFED 4.6 | | OFED 4.6 |
1814 | | | rdma-core 24 | | rdma-core 23 |
1815 | | | ConnectX-5 | | ConnectX-5 |
1816 +-----------------------+-----------------+-----------------+
1821 MARK and META items are interrelated with datapath - they might move from/to
1822 the applications in mbuf fields. Hence, zero value for these items has the
1823 special meaning - it means "no metadata are provided", not zero values are
1824 treated by applications and PMD as valid ones.
1826 Moreover in the flow engine domain the value zero is acceptable to match and
1827 set, and we should allow to specify zero values as rte_flow parameters for the
1828 META and MARK items and actions. In the same time zero mask has no meaning and
1829 should be rejected on validation stage.
1834 Flows are not cached in the driver.
1835 When stopping a device port, all the flows created on this port from the
1836 application will be flushed automatically in the background.
1837 After stopping the device port, all flows on this port become invalid and
1838 not represented in the system.
1839 All references to these flows held by the application should be discarded
1840 directly but neither destroyed nor flushed.
1842 The application should re-create the flows as required after the port restart.
1847 Compared to librte_net_mlx4 that implements a single RSS configuration per
1848 port, librte_net_mlx5 supports per-protocol RSS configuration.
1850 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1851 command-line parameter to enable additional protocols (UDP and TCP as well
1852 as IP), the following commands must be entered from its CLI to get the same
1853 behavior as librte_net_mlx4::
1856 > port config all rss all
1862 This section demonstrates how to launch **testpmd** with Mellanox
1863 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1865 #. Load the kernel modules::
1867 modprobe -a ib_uverbs mlx5_core mlx5_ib
1869 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1872 /etc/init.d/openibd restart
1876 User space I/O kernel modules (uio and igb_uio) are not used and do
1877 not have to be loaded.
1879 #. Make sure Ethernet interfaces are in working order and linked to kernel
1880 verbs. Related sysfs entries should be present::
1882 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1891 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1894 for intf in eth2 eth3 eth4 eth5;
1896 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1899 sed -n 's,.*/\(.*\),-a \1,p'
1908 #. Request huge pages::
1910 dpdk-hugepages.py --setup 2G
1912 #. Start testpmd with basic parameters::
1914 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
1919 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1920 EAL: probe driver: 15b3:1013 librte_net_mlx5
1921 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1922 PMD: librte_net_mlx5: 1 port(s) detected
1923 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1924 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1925 EAL: probe driver: 15b3:1013 librte_net_mlx5
1926 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1927 PMD: librte_net_mlx5: 1 port(s) detected
1928 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1929 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1930 EAL: probe driver: 15b3:1013 librte_net_mlx5
1931 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1932 PMD: librte_net_mlx5: 1 port(s) detected
1933 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1934 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1935 EAL: probe driver: 15b3:1013 librte_net_mlx5
1936 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1937 PMD: librte_net_mlx5: 1 port(s) detected
1938 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1939 Interactive-mode selected
1940 Configuring Port 0 (socket 0)
1941 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1942 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1943 Port 0: E4:1D:2D:E7:0C:FE
1944 Configuring Port 1 (socket 0)
1945 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1946 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1947 Port 1: E4:1D:2D:E7:0C:FF
1948 Configuring Port 2 (socket 0)
1949 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1950 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1951 Port 2: E4:1D:2D:E7:0C:FA
1952 Configuring Port 3 (socket 0)
1953 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1954 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1955 Port 3: E4:1D:2D:E7:0C:FB
1956 Checking link statuses...
1957 Port 0 Link Up - speed 40000 Mbps - full-duplex
1958 Port 1 Link Up - speed 40000 Mbps - full-duplex
1959 Port 2 Link Up - speed 10000 Mbps - full-duplex
1960 Port 3 Link Up - speed 10000 Mbps - full-duplex
1967 This section demonstrates how to dump flows. Currently, it's possible to dump
1968 all flows with assistance of external tools.
1970 #. 2 ways to get flow raw file:
1972 - Using testpmd CLI:
1974 .. code-block:: console
1977 testpmd> flow dump <port> all <output_file>
1979 testpmd> flow dump <port> rule <rule_id> <output_file>
1981 - call rte_flow_dev_dump api:
1983 .. code-block:: console
1985 rte_flow_dev_dump(port, flow, file, NULL);
1987 #. Dump human-readable flows from raw file:
1989 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1991 .. code-block:: console
1993 mlx_steering_dump.py -f <output_file> -flowptr <flow_ptr>
1995 How to share a meter between ports in the same switch domain
1996 ------------------------------------------------------------
1998 This section demonstrates how to use the shared meter. A meter M can be created
1999 on port X and to be shared with a port Y on the same switch domain by the next way:
2001 .. code-block:: console
2003 flow create X ingress transfer pattern eth / port_id id is Y / end actions meter mtr_id M / end
2005 How to use meter hierarchy
2006 --------------------------
2008 This section demonstrates how to create and use a meter hierarchy.
2009 A termination meter M can be the policy green action of another termination meter N.
2010 The two meters are chained together as a chain. Using meter N in a flow will apply
2011 both the meters in hierarchy on that flow.
2013 .. code-block:: console
2015 add port meter policy 0 1 g_actions queue index 0 / end y_actions end r_actions drop / end
2016 create port meter 0 M 1 1 yes 0xffff 1 0
2017 add port meter policy 0 2 g_actions meter mtr_id M / end y_actions end r_actions drop / end
2018 create port meter 0 N 2 2 yes 0xffff 1 0
2019 flow create 0 ingress group 1 pattern eth / end actions meter mtr_id N / end