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 meter hierarchy.
111 - Flow integrity offload API.
112 - Connection tracking.
113 - Sub-Function representors.
120 On Windows, the features are limited:
122 - Promiscuous mode is not supported
123 - The following rules are supported:
125 - IPv4/UDP with CVLAN filtering
126 - Unicast MAC filtering
128 - Additional rules are supported from WinOF2 version 2.70:
130 - IPv4/TCP with CVLAN filtering
131 - L4 steering rules for port RSS of UDP, TCP and IP
133 - For secondary process:
135 - Forked secondary process not supported.
136 - External memory unregistered in EAL memseg list cannot be used for DMA
137 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
138 primary process and remapped to the same virtual address in secondary
139 process. If the external memory is registered by primary process but has
140 different virtual address in secondary process, unexpected error may happen.
142 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
143 specific VLAN will match for VLAN packets as well:
145 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
146 Meaning, the flow rule::
148 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
150 Will only match vlan packets with vid=3. and the flow rule::
152 flow create 0 ingress pattern eth / ipv4 / end ...
154 Will match any ipv4 packet (VLAN included).
156 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
158 - 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.
161 flow create 0 ingress pattern eth / ipv4 / end ...
163 Will match any ipv4 packet.
166 flow create 0 ingress pattern eth / vlan / end ...
167 flow create 0 ingress pattern eth has_vlan is 1 / end ...
168 flow create 0 ingress pattern eth type is 0x8100 / end ...
170 Will match single-tagged packets only, with any VLAN ID value.
173 flow create 0 ingress pattern eth type is 0x88A8 / end ...
174 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
176 Will match multi-tagged packets only, with any VLAN ID value.
178 - A flow pattern with 2 sequential VLAN items is not supported.
180 - VLAN pop offload command:
182 - Flow rules having a VLAN pop offload command as one of their actions and
183 are lacking a match on VLAN as one of their items are not supported.
184 - The command is not supported on egress traffic in NIC mode.
186 - VLAN push offload is not supported on ingress traffic in NIC mode.
188 - VLAN set PCP offload is not supported on existing headers.
190 - A multi segment packet must have not more segments than reported by dev_infos_get()
191 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
192 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
193 inline settings) to 58.
195 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
196 to 0 are not supported.
198 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
200 - Match on Geneve header supports the following fields only:
207 - Match on Geneve TLV option is supported on the following fields:
214 Only one Class/Type/Length Geneve TLV option is supported per shared device.
215 Class/Type/Length fields must be specified as well as masks.
216 Class/Type/Length specified masks must be full.
217 Matching Geneve TLV option without specifying data is not supported.
218 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
220 - VF: flow rules created on VF devices can only match traffic targeted at the
221 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
223 - Match on GTP tunnel header item supports the following fields only:
225 - v_pt_rsv_flags: E flag, S flag, PN flag
229 - Match on GTP extension header only for GTP PDU session container (next
230 extension header type = 0x85).
231 - Match on GTP extension header is not supported in group 0.
233 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
234 The flows within group 0 and set metadata action are rejected by hardware.
238 MAC addresses not already present in the bridge table of the associated
239 kernel network device will be added and cleaned up by the PMD when closing
240 the device. In case of ungraceful program termination, some entries may
241 remain present and should be removed manually by other means.
243 - Buffer split offload is supported with regular Rx burst routine only,
244 no MPRQ feature or vectorized code can be engaged.
246 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
247 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
248 ol_flags. As the mempool for the external buffer is managed by PMD, all the
249 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
250 the external buffers will be freed by PMD and the application which still
251 holds the external buffers may be corrupted.
253 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
254 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
255 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
257 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
258 and allmulticast mode are both set to off.
259 To receive IPv6 Multicast messages on VM, explicitly set the relevant
260 MAC address using rte_eth_dev_mac_addr_add() API.
262 - To support a mixed traffic pattern (some buffers from local host memory, some
263 buffers from other devices) with high bandwidth, a mbuf flag is used.
265 An application hints the PMD whether or not it should try to inline the
266 given mbuf data buffer. PMD should do the best effort to act upon this request.
268 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
269 registered by application with rte_mbuf_dynflag_register(). This flag is
270 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
271 which is intended to be used by the application.
273 To query the supported specific flags in runtime,
274 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
275 currently (over present hardware and configuration) supported specific flags.
276 The "not inline hint" feature operating flow is the following one:
279 - probe the devices, ports are created
280 - query the port capabilities
281 - if port supporting the feature is found
282 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
283 - application starts the ports
284 - on ``dev_start()`` PMD checks whether the feature flag is registered and
285 enables the feature support in datapath
286 - application might set the registered flag bit in ``ol_flags`` field
287 of mbuf being sent and PMD will handle ones appropriately.
289 - The amount of descriptors in Tx queue may be limited by data inline settings.
290 Inline data require the more descriptor building blocks and overall block
291 amount may exceed the hardware supported limits. The application should
292 reduce the requested Tx size or adjust data inline settings with
293 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
295 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
296 parameter should be specified.
297 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
298 being sent it tries to synchronize the time of packet appearing on
299 the wire with the specified packet timestamp. It the specified one
300 is in the past it should be ignored, if one is in the distant future
301 it should be capped with some reasonable value (in range of seconds).
302 These specific cases ("too late" and "distant future") can be optionally
303 reported via device xstats to assist applications to detect the
304 time-related problems.
306 The timestamp upper "too-distant-future" limit
307 at the moment of invoking the Tx burst routine
308 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
309 Please note, for the testpmd txonly mode,
310 the limit is deduced from the expression::
312 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
314 There is no any packet reordering according timestamps is supposed,
315 neither within packet burst, nor between packets, it is an entirely
316 application responsibility to generate packets and its timestamps
317 in desired order. The timestamps can be put only in the first packet
318 in the burst providing the entire burst scheduling.
320 - E-Switch decapsulation Flow:
322 - can be applied to PF port only.
323 - must specify VF port action (packet redirection from PF to VF).
324 - optionally may specify tunnel inner source and destination MAC addresses.
326 - E-Switch encapsulation Flow:
328 - can be applied to VF ports only.
329 - must specify PF port action (packet redirection from VF to PF).
333 - The input buffer, used as outer header, is not validated.
337 - The decapsulation is always done up to the outermost tunnel detected by the HW.
338 - The input buffer, providing the removal size, is not validated.
339 - The buffer size must match the length of the headers to be removed.
341 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
342 mutually exclusive features which cannot be supported together
343 (see :ref:`mlx5_firmware_config`).
347 - Requires DevX and DV flow to be enabled.
348 - KEEP_CRC offload cannot be supported with LRO.
349 - The first mbuf length, without head-room, must be big enough to include the
351 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
352 it with size limited to max LRO size, not to max RX packet length.
353 - LRO can be used with outer header of TCP packets of the standard format:
354 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
356 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
357 - LRO packet aggregation is performed by HW only for packet size larger than
358 ``lro_min_mss_size``. This value is reported on device start, when debug
363 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
364 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
365 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
369 - fast free offload assumes the all mbufs being sent are originated from the
370 same memory pool and there is no any extra references to the mbufs (the
371 reference counter for each mbuf is equal 1 on tx_burst call). The latter
372 means there should be no any externally attached buffers in mbufs. It is
373 an application responsibility to provide the correct mbufs if the fast
374 free offload is engaged. The mlx5 PMD implicitly produces the mbufs with
375 externally attached buffers if MPRQ option is enabled, hence, the fast
376 free offload is neither supported nor advertised if there is MPRQ enabled.
380 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and
381 E-Switch steering domain.
382 - For E-Switch Sampling flow with sample ratio > 1, additional actions are not
383 supported in the sample actions list.
384 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as
385 first action in the E-Switch egress flow if with header modify or
386 encapsulation actions.
387 - For NIC Rx flow, supports ``MARK``, ``COUNT``, ``QUEUE``, ``RSS`` in the
389 - For E-Switch mirroring flow, supports ``RAW ENCAP``, ``Port ID``,
390 ``VXLAN ENCAP``, ``NVGRE ENCAP`` in the sample actions list.
394 - Supports the 'set' operation only for ``RTE_FLOW_ACTION_TYPE_MODIFY_FIELD`` action.
395 - Modification of an arbitrary place in a packet via the special ``RTE_FLOW_FIELD_START`` Field ID is not supported.
396 - Modification of the 802.1Q Tag, VXLAN Network or GENEVE Network ID's is not supported.
397 - Encapsulation levels are not supported, can modify outermost header fields only.
398 - Offsets must be 32-bits aligned, cannot skip past the boundary of a field.
400 - IPv6 header item 'proto' field, indicating the next header protocol, should
401 not be set as extension header.
402 In case the next header is an extension header, it should not be specified in
403 IPv6 header item 'proto' field.
404 The last extension header item 'next header' field can specify the following
405 header protocol type.
409 - 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.
410 - Hairpin in switchdev SR-IOV mode is not supported till now.
414 - All the meter colors with drop action will be counted only by the global drop statistics.
415 - Green color is not supported with drop action.
416 - Yellow detection is not supported.
417 - Red color must be with drop action.
418 - Meter statistics are supported only for drop case.
419 - Meter yellow color detection is not supported.
420 - A meter action created with pre-defined policy must be the last action in the flow except single case where the policy actions are:
421 - green: NULL or END.
422 - yellow: NULL or END.
424 - The only supported meter policy actions:
425 - green: QUEUE, RSS, PORT_ID, JUMP, MARK and SET_TAG.
426 - yellow: must be empty.
428 - meter profile packet mode is supported.
432 - Integrity offload is enabled for **ConnectX-6** family.
433 - Verification bits provided by the hardware are ``l3_ok``, ``ipv4_csum_ok``, ``l4_ok``, ``l4_csum_ok``.
434 - ``level`` value 0 references outer headers.
435 - Multiple integrity items not supported in a single flow rule.
436 - Flow rule items supplied by application must explicitly specify network headers referred by integrity item.
437 For example, if integrity item mask sets ``l4_ok`` or ``l4_csum_ok`` bits, reference to L4 network header,
438 TCP or UDP, must be in the rule pattern as well::
440 flow create 0 ingress pattern integrity level is 0 value mask l3_ok value spec l3_ok / eth / ipv6 / end …
442 flow create 0 ingress pattern integrity level is 0 value mask l4_ok value spec 0 / eth / ipv4 proto is udp / end …
444 - Connection tracking:
446 - Cannot co-exist with ASO meter, ASO age action in a single flow rule.
447 - Flow rules insertion rate and memory consumption need more optimization.
449 - 4M connections maximum.
454 MLX5 supports various methods to report statistics:
456 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.
458 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.
460 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.
468 The ibverbs libraries can be linked with this PMD in a number of ways,
469 configured by the ``ibverbs_link`` build option:
471 - ``shared`` (default): the PMD depends on some .so files.
473 - ``dlopen``: Split the dependencies glue in a separate library
474 loaded when needed by dlopen.
475 It make dependencies on libibverbs and libmlx4 optional,
476 and has no performance impact.
478 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
479 in the PMD shared library or the executable static binary.
481 Environment variables
482 ~~~~~~~~~~~~~~~~~~~~~
486 A list of directories in which to search for the rdma-core "glue" plug-in,
487 separated by colons or semi-colons.
489 - ``MLX5_SHUT_UP_BF``
491 Configures HW Tx doorbell register as IO-mapped.
493 By default, the HW Tx doorbell is configured as a write-combining register.
494 The register would be flushed to HW usually when the write-combining buffer
495 becomes full, but it depends on CPU design.
497 Except for vectorized Tx burst routines, a write memory barrier is enforced
498 after updating the register so that the update can be immediately visible to
501 When vectorized Tx burst is called, the barrier is set only if the burst size
502 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
503 variable will bring better latency even though the maximum throughput can
506 Run-time configuration
507 ~~~~~~~~~~~~~~~~~~~~~~
509 - librte_net_mlx5 brings kernel network interfaces up during initialization
510 because it is affected by their state. Forcing them down prevents packets
513 - **ethtool** operations on related kernel interfaces also affect the PMD.
518 In order to run as a non-root user,
519 some capabilities must be granted to the application::
521 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
523 Below are the reasons of the need for each capability:
526 When using physical addresses (PA mode), with Linux >= 4.0,
527 for access to ``/proc/self/pagemap``.
530 For device configuration.
533 For raw ethernet queue allocation through kernel driver.
536 For DMA memory pinning.
541 - ``rxq_cqe_comp_en`` parameter [int]
543 A nonzero value enables the compression of CQE on RX side. This feature
544 allows to save PCI bandwidth and improve performance. Enabled by default.
545 Different compression formats are supported in order to achieve the best
546 performance for different traffic patterns. Default format depends on
547 Multi-Packet Rx queue configuration: Hash RSS format is used in case
548 MPRQ is disabled, Checksum format is used in case MPRQ is enabled.
550 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
551 better compression rate in case of RTE Flow Mark traffic.
552 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
553 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
554 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
555 CQE compression format selection requires DevX to be enabled. If there is
556 no DevX enabled/supported the value is reset to 1 by default.
560 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
561 ConnectX-6 Lx, BlueField and BlueField-2.
562 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
563 ConnectX-6 Lx, BlueField and BlueField-2.
565 - ``rxq_pkt_pad_en`` parameter [int]
567 A nonzero value enables padding Rx packet to the size of cacheline on PCI
568 transaction. This feature would waste PCI bandwidth but could improve
569 performance by avoiding partial cacheline write which may cause costly
570 read-modify-copy in memory transaction on some architectures. Disabled by
575 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
576 ConnectX-6 Lx, BlueField and BlueField-2.
577 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
578 ConnectX-6 Lx, BlueField and BlueField-2.
580 - ``mprq_en`` parameter [int]
582 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
583 configured as Multi-Packet RQ if the total number of Rx queues is
584 ``rxqs_min_mprq`` or more. Disabled by default.
586 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
587 by posting a single large buffer for multiple packets. Instead of posting a
588 buffers per a packet, one large buffer is posted in order to receive multiple
589 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
590 and each stride receives one packet. MPRQ can improve throughput for
591 small-packet traffic.
593 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
594 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
595 configure large stride size enough to accommodate max_rx_pkt_len as long as
596 device allows. Note that this can waste system memory compared to enabling Rx
597 scatter and multi-segment packet.
599 - ``mprq_log_stride_num`` parameter [int]
601 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
602 strides can reduce PCIe traffic further. If configured value is not in the
603 range of device capability, the default value will be set with a warning
604 message. The default value is 4 which is 16 strides per a buffer, valid only
605 if ``mprq_en`` is set.
607 The size of Rx queue should be bigger than the number of strides.
609 - ``mprq_log_stride_size`` parameter [int]
611 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
612 stride size can save some memory and reduce probability of a depletion of all
613 available strides due to unreleased packets by an application. If configured
614 value is not in the range of device capability, the default value will be set
615 with a warning message. The default value is 11 which is 2048 bytes per a
616 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
617 it is possible for a packet to span across multiple strides. This mode allows
618 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
619 of a packet if Rx scatter is configured) may be required in case there is no
620 space left for a head room at the end of a stride which incurs some
623 - ``mprq_max_memcpy_len`` parameter [int]
625 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
626 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
627 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
628 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
629 A mempool for external buffers will be allocated and managed by PMD. If Rx
630 packet is externally attached, ol_flags field of the mbuf will have
631 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
632 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
634 - ``rxqs_min_mprq`` parameter [int]
636 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
637 greater or equal to this value. The default value is 12, valid only if
640 - ``txq_inline`` parameter [int]
642 Amount of data to be inlined during TX operations. This parameter is
643 deprecated and converted to the new parameter ``txq_inline_max`` providing
644 partial compatibility.
646 - ``txqs_min_inline`` parameter [int]
648 Enable inline data send only when the number of TX queues is greater or equal
651 This option should be used in combination with ``txq_inline_max`` and
652 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
654 If this option is not specified the default value 16 is used for BlueField
655 and 8 for other platforms
657 The data inlining consumes the CPU cycles, so this option is intended to
658 auto enable inline data if we have enough Tx queues, which means we have
659 enough CPU cores and PCI bandwidth is getting more critical and CPU
660 is not supposed to be bottleneck anymore.
662 The copying data into WQE improves latency and can improve PPS performance
663 when PCI back pressure is detected and may be useful for scenarios involving
664 heavy traffic on many queues.
666 Because additional software logic is necessary to handle this mode, this
667 option should be used with care, as it may lower performance when back
668 pressure is not expected.
670 If inline data are enabled it may affect the maximal size of Tx queue in
671 descriptors because the inline data increase the descriptor size and
672 queue size limits supported by hardware may be exceeded.
674 - ``txq_inline_min`` parameter [int]
676 Minimal amount of data to be inlined into WQE during Tx operations. NICs
677 may require this minimal data amount to operate correctly. The exact value
678 may depend on NIC operation mode, requested offloads, etc. It is strongly
679 recommended to omit this parameter and use the default values. Anyway,
680 applications using this parameter should take into consideration that
681 specifying an inconsistent value may prevent the NIC from sending packets.
683 If ``txq_inline_min`` key is present the specified value (may be aligned
684 by the driver in order not to exceed the limits and provide better descriptor
685 space utilization) will be used by the driver and it is guaranteed that
686 requested amount of data bytes are inlined into the WQE beside other inline
687 settings. This key also may update ``txq_inline_max`` value (default
688 or specified explicitly in devargs) to reserve the space for inline data.
690 If ``txq_inline_min`` key is not present, the value may be queried by the
691 driver from the NIC via DevX if this feature is available. If there is no DevX
692 enabled/supported the value 18 (supposing L2 header including VLAN) is set
693 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
694 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
697 For ConnectX-4 NIC, driver does not allow specifying value below 18
698 (minimal L2 header, including VLAN), error will be raised.
700 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
701 it is not recommended and may prevent NIC from sending packets over
704 For ConnectX-4 and ConnectX-4 Lx NICs, automatically configured value
705 is insufficient for some traffic, because they require at least all L2 headers
706 to be inlined. For example, Q-in-Q adds 4 bytes to default 18 bytes
707 of Ethernet and VLAN, thus ``txq_inline_min`` must be set to 22.
708 MPLS would add 4 bytes per label. Final value must account for all possible
709 L2 encapsulation headers used in particular environment.
711 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
712 Multi-Packet Write), because last one does not support partial packet inlining.
713 This is not very critical due to minimal data inlining is mostly required
714 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
716 - ``txq_inline_max`` parameter [int]
718 Specifies the maximal packet length to be completely inlined into WQE
719 Ethernet Segment for ordinary SEND method. If packet is larger than specified
720 value, the packet data won't be copied by the driver at all, data buffer
721 is addressed with a pointer. If packet length is less or equal all packet
722 data will be copied into WQE. This may improve PCI bandwidth utilization for
723 short packets significantly but requires the extra CPU cycles.
725 The data inline feature is controlled by number of Tx queues, if number of Tx
726 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
727 is engaged, if there are not enough Tx queues (which means not enough CPU cores
728 and CPU resources are scarce), data inline is not performed by the driver.
729 Assigning ``txqs_min_inline`` with zero always enables the data inline.
731 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
732 by the driver in order not to exceed the limit (930 bytes) and to provide better
733 WQE space filling without gaps, the adjustment is reflected in the debug log.
734 Also, the default value (290) may be decreased in run-time if the large transmit
735 queue size is requested and hardware does not support enough descriptor
736 amount, in this case warning is emitted. If ``txq_inline_max`` key is
737 specified and requested inline settings can not be satisfied then error
740 - ``txq_inline_mpw`` parameter [int]
742 Specifies the maximal packet length to be completely inlined into WQE for
743 Enhanced MPW method. If packet is large the specified value, the packet data
744 won't be copied, and data buffer is addressed with pointer. If packet length
745 is less or equal, all packet data will be copied into WQE. This may improve PCI
746 bandwidth utilization for short packets significantly but requires the extra
749 The data inline feature is controlled by number of TX queues, if number of Tx
750 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
751 is engaged, if there are not enough Tx queues (which means not enough CPU cores
752 and CPU resources are scarce), data inline is not performed by the driver.
753 Assigning ``txqs_min_inline`` with zero always enables the data inline.
755 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
756 by the driver in order not to exceed the limit (930 bytes) and to provide better
757 WQE space filling without gaps, the adjustment is reflected in the debug log.
758 Due to multiple packets may be included to the same WQE with Enhanced Multi
759 Packet Write Method and overall WQE size is limited it is not recommended to
760 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
761 may be decreased in run-time if the large transmit queue size is requested
762 and hardware does not support enough descriptor amount, in this case warning
763 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
764 settings can not be satisfied then error will be raised.
766 - ``txqs_max_vec`` parameter [int]
768 Enable vectorized Tx only when the number of TX queues is less than or
769 equal to this value. This parameter is deprecated and ignored, kept
770 for compatibility issue to not prevent driver from probing.
772 - ``txq_mpw_hdr_dseg_en`` parameter [int]
774 A nonzero value enables including two pointers in the first block of TX
775 descriptor. The parameter is deprecated and ignored, kept for compatibility
778 - ``txq_max_inline_len`` parameter [int]
780 Maximum size of packet to be inlined. This limits the size of packet to
781 be inlined. If the size of a packet is larger than configured value, the
782 packet isn't inlined even though there's enough space remained in the
783 descriptor. Instead, the packet is included with pointer. This parameter
784 is deprecated and converted directly to ``txq_inline_mpw`` providing full
785 compatibility. Valid only if eMPW feature is engaged.
787 - ``txq_mpw_en`` parameter [int]
789 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
790 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
791 eMPW allows the Tx burst function to pack up multiple packets
792 in a single descriptor session in order to save PCI bandwidth
793 and improve performance at the cost of a slightly higher CPU usage.
794 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
795 Tx burst function copies entire packet data on to Tx descriptor
796 instead of including pointer of packet.
798 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
799 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
800 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
801 option or reported by the NIC, the eMPW feature is disengaged.
803 - ``tx_db_nc`` parameter [int]
805 The rdma core library can map doorbell register in two ways, depending on the
806 environment variable "MLX5_SHUT_UP_BF":
808 - As regular cached memory (usually with write combining attribute), if the
809 variable is either missing or set to zero.
810 - As non-cached memory, if the variable is present and set to not "0" value.
812 The type of mapping may slightly affect the Tx performance, the optimal choice
813 is strongly relied on the host architecture and should be deduced practically.
815 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
816 memory (with write combining), the PMD will perform the extra write memory barrier
817 after writing to doorbell, it might increase the needed CPU clocks per packet
818 to send, but latency might be improved.
820 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
821 cached memory, the PMD will not perform the extra write memory barrier
822 after writing to doorbell, on some architectures it might improve the
825 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
826 memory, the PMD will use heuristics to decide whether write memory barrier
827 should be performed. For bursts with size multiple of recommended one (64 pkts)
828 it is supposed the next burst is coming and no need to issue the extra memory
829 barrier (it is supposed to be issued in the next coming burst, at least after
830 descriptor writing). It might increase latency (on some hosts till next
831 packets transmit) and should be used with care.
833 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
834 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
835 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
837 - ``tx_pp`` parameter [int]
839 If a nonzero value is specified the driver creates all necessary internal
840 objects to provide accurate packet send scheduling on mbuf timestamps.
841 The positive value specifies the scheduling granularity in nanoseconds,
842 the packet send will be accurate up to specified digits. The allowed range is
843 from 500 to 1 million of nanoseconds. The negative value specifies the module
844 of granularity and engages the special test mode the check the schedule rate.
845 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
848 - ``tx_skew`` parameter [int]
850 The parameter adjusts the send packet scheduling on timestamps and represents
851 the average delay between beginning of the transmitting descriptor processing
852 by the hardware and appearance of actual packet data on the wire. The value
853 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
854 specified. The default value is zero.
856 - ``tx_vec_en`` parameter [int]
858 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
859 ConnectX-6 Lx, BlueField and BlueField-2 NICs
860 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
861 The parameter is deprecated and ignored.
863 - ``rx_vec_en`` parameter [int]
865 A nonzero value enables Rx vector if the port is not configured in
866 multi-segment otherwise this parameter is ignored.
870 - ``vf_nl_en`` parameter [int]
872 A nonzero value enables Netlink requests from the VF to add/remove MAC
873 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
874 Otherwise the relevant configuration must be run with Linux iproute2 tools.
875 This is a prerequisite to receive this kind of traffic.
877 Enabled by default, valid only on VF devices ignored otherwise.
879 - ``l3_vxlan_en`` parameter [int]
881 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
882 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
883 parameter. This is a prerequisite to receive this kind of traffic.
887 - ``dv_xmeta_en`` parameter [int]
889 A nonzero value enables extensive flow metadata support if device is
890 capable and driver supports it. This can enable extensive support of
891 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
892 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
894 There are some possible configurations, depending on parameter value:
896 - 0, this is default value, defines the legacy mode, the ``MARK`` and
897 ``META`` related actions and items operate only within NIC Tx and
898 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
899 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
900 item is 32 bits wide and match supported on egress only.
902 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
903 related actions and items operate within all supported steering domains,
904 including FDB, ``MARK`` and ``META`` information may cross the domain
905 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
906 depends on kernel and firmware configurations and might be 0, 16 or
907 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
908 compatibility, the actual width of data transferred to the FDB domain
909 depends on kernel configuration and may be vary. The actual supported
910 width can be retrieved in runtime by series of rte_flow_validate()
913 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
914 related actions and items operate within all supported steering domains,
915 including FDB, ``MARK`` and ``META`` information may cross the domain
916 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
917 depends on kernel and firmware configurations and might be 0, 16 or
918 24 bits. The actual supported width can be retrieved in runtime by
919 series of rte_flow_validate() trials.
921 - 3, this engages tunnel offload mode. In E-Switch configuration, that
922 mode implicitly activates ``dv_xmeta_en=1``.
924 +------+-----------+-----------+-------------+-------------+
925 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
926 +======+===========+===========+=============+=============+
927 | 0 | 24 bits | 32 bits | 32 bits | no |
928 +------+-----------+-----------+-------------+-------------+
929 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
930 +------+-----------+-----------+-------------+-------------+
931 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
932 +------+-----------+-----------+-------------+-------------+
934 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
935 ignored and the device is configured to operate in legacy mode (0).
937 Disabled by default (set to 0).
939 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
940 of the extensive metadata features. The legacy Verbs supports FLAG and
941 MARK metadata actions over NIC Rx steering domain only.
943 Setting META value to zero in flow action means there is no item provided
944 and receiving datapath will not report in mbufs the metadata are present.
945 Setting MARK value to zero in flow action means the zero FDIR ID value
946 will be reported on packet receiving.
948 For the MARK action the last 16 values in the full range are reserved for
949 internal PMD purposes (to emulate FLAG action). The valid range for the
950 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
951 for the 24-bit mode, the flows with the MARK action value outside
952 the specified range will be rejected.
954 - ``dv_flow_en`` parameter [int]
956 A nonzero value enables the DV flow steering assuming it is supported
957 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
959 Enabled by default if supported.
961 - ``dv_esw_en`` parameter [int]
963 A nonzero value enables E-Switch using Direct Rules.
965 Enabled by default if supported.
967 - ``lacp_by_user`` parameter [int]
969 A nonzero value enables the control of LACP traffic by the user application.
970 When a bond exists in the driver, by default it should be managed by the
971 kernel and therefore LACP traffic should be steered to the kernel.
972 If this devarg is set to 1 it will allow the user to manage the bond by
973 itself and not steer LACP traffic to the kernel.
975 Disabled by default (set to 0).
977 - ``mr_ext_memseg_en`` parameter [int]
979 A nonzero value enables extending memseg when registering DMA memory. If
980 enabled, the number of entries in MR (Memory Region) lookup table on datapath
981 is minimized and it benefits performance. On the other hand, it worsens memory
982 utilization because registered memory is pinned by kernel driver. Even if a
983 page in the extended chunk is freed, that doesn't become reusable until the
984 entire memory is freed.
988 - ``representor`` parameter [list]
990 This parameter can be used to instantiate DPDK Ethernet devices from
991 existing port (PF, VF or SF) representors configured on the device.
993 It is a standard parameter whose format is described in
994 :ref:`ethernet_device_standard_device_arguments`.
996 For instance, to probe VF port representors 0 through 2::
998 <PCI_BDF>,representor=vf[0-2]
1000 To probe SF port representors 0 through 2::
1002 <PCI_BDF>,representor=sf[0-2]
1004 To probe VF port representors 0 through 2 on both PFs of bonding device::
1006 <Primary_PCI_BDF>,representor=pf[0,1]vf[0-2]
1008 - ``max_dump_files_num`` parameter [int]
1010 The maximum number of files per PMD entity that may be created for debug information.
1011 The files will be created in /var/log directory or in current directory.
1013 set to 128 by default.
1015 - ``lro_timeout_usec`` parameter [int]
1017 The maximum allowed duration of an LRO session, in micro-seconds.
1018 PMD will set the nearest value supported by HW, which is not bigger than
1019 the input ``lro_timeout_usec`` value.
1020 If this parameter is not specified, by default PMD will set
1021 the smallest value supported by HW.
1023 - ``hp_buf_log_sz`` parameter [int]
1025 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
1026 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
1027 The capacity of the value is specified by the firmware and the initialization
1028 will get a failure if it is out of scope.
1029 The range of the value is from 11 to 19 right now, and the supported frame
1030 size of a single packet for hairpin is from 512B to 128KB. It might change if
1031 different firmware release is being used. By using a small value, it could
1032 reduce memory consumption but not work with a large frame. If the value is
1033 too large, the memory consumption will be high and some potential performance
1034 degradation will be introduced.
1035 By default, the PMD will set this value to 16, which means that 9KB jumbo
1036 frames will be supported.
1038 - ``reclaim_mem_mode`` parameter [int]
1040 Cache some resources in flow destroy will help flow recreation more efficient.
1041 While some systems may require the all the resources can be reclaimed after
1043 The parameter ``reclaim_mem_mode`` provides the option for user to configure
1044 if the resource cache is needed or not.
1046 There are three options to choose:
1048 - 0. It means the flow resources will be cached as usual. The resources will
1049 be cached, helpful with flow insertion rate.
1051 - 1. It will only enable the DPDK PMD level resources reclaim.
1053 - 2. Both DPDK PMD level and rdma-core low level will be configured as
1056 By default, the PMD will set this value to 0.
1058 - ``sys_mem_en`` parameter [int]
1060 A non-zero value enables the PMD memory management allocating memory
1061 from system by default, without explicit rte memory flag.
1063 By default, the PMD will set this value to 0.
1065 - ``decap_en`` parameter [int]
1067 Some devices do not support FCS (frame checksum) scattering for
1068 tunnel-decapsulated packets.
1069 If set to 0, this option forces the FCS feature and rejects tunnel
1070 decapsulation in the flow engine for such devices.
1072 By default, the PMD will set this value to 1.
1074 - ``allow_duplicate_pattern`` parameter [int]
1076 There are two options to choose:
1078 - 0. Prevent insertion of rules with the same pattern items on non-root table.
1079 In this case, only the first rule is inserted and the following rules are
1080 rejected and error code EEXIST is returned.
1082 - 1. Allow insertion of rules with the same pattern items.
1083 In this case, all rules are inserted but only the first rule takes effect,
1084 the next rule takes effect only if the previous rules are deleted.
1086 By default, the PMD will set this value to 1.
1088 .. _mlx5_firmware_config:
1090 Firmware configuration
1091 ~~~~~~~~~~~~~~~~~~~~~~
1093 Firmware features can be configured as key/value pairs.
1095 The command to set a value is::
1097 mlxconfig -d <device> set <key>=<value>
1099 The command to query a value is::
1101 mlxconfig -d <device> query | grep <key>
1103 The device name for the command ``mlxconfig`` can be either the PCI address,
1104 or the mst device name found with::
1108 Below are some firmware configurations listed.
1114 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1120 - maximum number of SR-IOV virtual functions::
1124 - enable DevX (required by Direct Rules and other features)::
1128 - aggressive CQE zipping::
1132 - L3 VXLAN and VXLAN-GPE destination UDP port::
1135 IP_OVER_VXLAN_PORT=<udp dport>
1137 - enable VXLAN-GPE tunnel flow matching::
1139 FLEX_PARSER_PROFILE_ENABLE=0
1141 FLEX_PARSER_PROFILE_ENABLE=2
1143 - enable IP-in-IP tunnel flow matching::
1145 FLEX_PARSER_PROFILE_ENABLE=0
1147 - enable MPLS flow matching::
1149 FLEX_PARSER_PROFILE_ENABLE=1
1151 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1153 FLEX_PARSER_PROFILE_ENABLE=2
1155 - enable Geneve flow matching::
1157 FLEX_PARSER_PROFILE_ENABLE=0
1159 FLEX_PARSER_PROFILE_ENABLE=1
1161 - enable Geneve TLV option flow matching::
1163 FLEX_PARSER_PROFILE_ENABLE=0
1165 - enable GTP flow matching::
1167 FLEX_PARSER_PROFILE_ENABLE=3
1169 - enable eCPRI flow matching::
1171 FLEX_PARSER_PROFILE_ENABLE=4
1177 This driver relies on external libraries and kernel drivers for resources
1178 allocations and initialization. The following dependencies are not part of
1179 DPDK and must be installed separately:
1183 User space Verbs framework used by librte_net_mlx5. This library provides
1184 a generic interface between the kernel and low-level user space drivers
1187 It allows slow and privileged operations (context initialization, hardware
1188 resources allocations) to be managed by the kernel and fast operations to
1189 never leave user space.
1193 Low-level user space driver library for Mellanox
1194 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1197 This library basically implements send/receive calls to the hardware
1200 - **Kernel modules**
1202 They provide the kernel-side Verbs API and low level device drivers that
1203 manage actual hardware initialization and resources sharing with user
1206 Unlike most other PMDs, these modules must remain loaded and bound to
1209 - mlx5_core: hardware driver managing Mellanox
1210 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1212 - mlx5_ib: InifiniBand device driver.
1213 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1215 - **Firmware update**
1217 Mellanox OFED/EN releases include firmware updates for
1218 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1220 Because each release provides new features, these updates must be applied to
1221 match the kernel modules and libraries they come with.
1225 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1231 Either RDMA Core library with a recent enough Linux kernel release
1232 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1235 RDMA Core with Linux Kernel
1236 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1238 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1239 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1240 (see `RDMA Core installation documentation`_)
1241 - When building for i686 use:
1243 - rdma-core version 18.0 or above built with 32bit support.
1244 - Kernel version 4.14.41 or above.
1246 - Starting with rdma-core v21, static libraries can be built::
1249 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1252 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1253 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1259 - Mellanox OFED version: **4.5** and above /
1260 Mellanox EN version: **4.5** and above
1263 - ConnectX-4: **12.21.1000** and above.
1264 - ConnectX-4 Lx: **14.21.1000** and above.
1265 - ConnectX-5: **16.21.1000** and above.
1266 - ConnectX-5 Ex: **16.21.1000** and above.
1267 - ConnectX-6: **20.27.0090** and above.
1268 - ConnectX-6 Dx: **22.27.0090** and above.
1269 - BlueField: **18.25.1010** and above.
1271 While these libraries and kernel modules are available on OpenFabrics
1272 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1273 managers on most distributions, this PMD requires Ethernet extensions that
1274 may not be supported at the moment (this is a work in progress).
1277 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1279 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1280 include the necessary support and should be used in the meantime. For DPDK,
1281 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1282 required from that distribution.
1286 Several versions of Mellanox OFED/EN are available. Installing the version
1287 this DPDK release was developed and tested against is strongly
1288 recommended. Please check the `linux prerequisites`_.
1290 Windows Prerequisites
1291 ---------------------
1293 This driver relies on external libraries and kernel drivers for resources
1294 allocations and initialization. The dependencies in the following sub-sections
1295 are not part of DPDK, and must be installed separately.
1297 Compilation Prerequisites
1298 ~~~~~~~~~~~~~~~~~~~~~~~~~
1300 DevX SDK installation
1301 ^^^^^^^^^^^^^^^^^^^^^
1303 The DevX SDK must be installed on the machine building the Windows PMD.
1304 Additional information can be found at
1305 `How to Integrate Windows DevX in Your Development Environment
1306 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1308 Runtime Prerequisites
1309 ~~~~~~~~~~~~~~~~~~~~~
1311 WinOF2 version 2.60 or higher must be installed on the machine.
1316 The driver can be downloaded from the following site:
1318 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1323 DevX for Windows must be enabled in the Windows registry.
1324 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1325 Additional information can be found in the WinOF2 user manual.
1330 The following Mellanox device families are supported by the same mlx5 driver:
1342 Below are detailed device names:
1344 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1345 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1346 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1347 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1348 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1349 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1350 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1351 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1352 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1353 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1354 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1355 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1356 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1357 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1358 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1359 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1360 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1361 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1362 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1363 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1364 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1365 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1366 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1367 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1368 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1369 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1371 Quick Start Guide on OFED/EN
1372 ----------------------------
1374 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1377 2. Install the required libraries and kernel modules either by installing
1378 only the required set, or by installing the entire Mellanox OFED/EN::
1380 ./mlnxofedinstall --upstream-libs --dpdk
1382 3. Verify the firmware is the correct one::
1386 4. Verify all ports links are set to Ethernet::
1388 mlxconfig -d <mst device> query | grep LINK_TYPE
1392 Link types may have to be configured to Ethernet::
1394 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1396 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1398 For hypervisors, verify SR-IOV is enabled on the NIC::
1400 mlxconfig -d <mst device> query | grep SRIOV_EN
1403 If needed, configure SR-IOV::
1405 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1406 mlxfwreset -d <mst device> reset
1408 5. Restart the driver::
1410 /etc/init.d/openibd restart
1414 service openibd restart
1416 If link type was changed, firmware must be reset as well::
1418 mlxfwreset -d <mst device> reset
1420 For hypervisors, after reset write the sysfs number of virtual functions
1423 To dynamically instantiate a given number of virtual functions (VFs)::
1425 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1427 6. Install DPDK and you are ready to go.
1428 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1430 Enable switchdev mode
1431 ---------------------
1433 Switchdev mode is a mode in E-Switch, that binds between representor and VF or SF.
1434 Representor is a port in DPDK that is connected to a VF or SF in such a way
1435 that assuming there are no offload flows, each packet that is sent from the VF or SF
1436 will be received by the corresponding representor. While each packet that is or SF
1437 sent to a representor will be received by the VF or SF.
1438 This is very useful in case of SRIOV mode, where the first packet that is sent
1439 by the VF or SF will be received by the DPDK application which will decide if this
1440 flow should be offloaded to the E-Switch. After offloading the flow packet
1441 that the VF or SF that are matching the flow will not be received any more by
1442 the DPDK application.
1444 1. Enable SRIOV mode::
1446 mlxconfig -d <mst device> set SRIOV_EN=true
1448 2. Configure the max number of VFs::
1450 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1454 mlxfwreset -d <mst device> reset
1456 3. Configure the actual number of VFs::
1458 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1460 4. Unbind the device (can be rebind after the switchdev mode)::
1462 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1464 5. Enbale switchdev mode::
1466 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1468 Sub-Function representor
1469 ------------------------
1471 Sub-Function is a portion of the PCI device, a SF netdev has its own
1472 dedicated queues(txq, rxq). A SF netdev supports E-Switch representation
1473 offload similar to existing PF and VF representors. A SF shares PCI
1474 level resources with other SFs and/or with its parent PCI function.
1476 1. Configure SF feature::
1478 mlxconfig -d <mst device> set PF_BAR2_SIZE=<0/1/2/3> PF_BAR2_ENABLE=1
1480 Value of PF_BAR2_SIZE:
1489 mlxfwreset -d <mst device> reset
1491 3. Enable switchdev mode::
1493 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1497 mlnx-sf -d <PCI_BDF> -a create
1499 5. Probe SF representor::
1501 testpmd> port attach <PCI_BDF>,representor=sf0,dv_flow_en=1
1506 1. Configure aggressive CQE Zipping for maximum performance::
1508 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1510 To set it back to the default CQE Zipping mode use::
1512 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1514 2. In case of virtualization:
1516 - Make sure that hypervisor kernel is 3.16 or newer.
1517 - Configure boot with ``iommu=pt``.
1518 - Use 1G huge pages.
1519 - Make sure to allocate a VM on huge pages.
1520 - Make sure to set CPU pinning.
1522 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1523 for better performance. For VMs, verify that the right CPU
1524 and NUMA node are pinned according to the above. Run::
1526 lstopo-no-graphics --merge
1528 to identify the NUMA node to which the PCIe adapter is connected.
1530 4. If more than one adapter is used, and root complex capabilities allow
1531 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1532 it is recommended to locate both adapters on the same NUMA node.
1533 This in order to forward packets from one to the other without
1534 NUMA performance penalty.
1536 5. Disable pause frames::
1538 ethtool -A <netdev> rx off tx off
1540 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1541 via the BIOS configuration. Please contact you server provider for more
1542 information about the settings.
1546 On some machines, depends on the machine integrator, it is beneficial
1547 to set the PCI max read request parameter to 1K. This can be
1548 done in the following way:
1550 To query the read request size use::
1552 setpci -s <NIC PCI address> 68.w
1554 If the output is different than 3XXX, set it by::
1556 setpci -s <NIC PCI address> 68.w=3XXX
1558 The XXX can be different on different systems. Make sure to configure
1559 according to the setpci output.
1561 7. To minimize overhead of searching Memory Regions:
1563 - '--socket-mem' is recommended to pin memory by predictable amount.
1564 - Configure per-lcore cache when creating Mempools for packet buffer.
1565 - Refrain from dynamically allocating/freeing memory in run-time.
1570 There are multiple Rx burst functions with different advantages and limitations.
1572 .. table:: Rx burst functions
1574 +-------------------+------------------------+---------+-----------------+------+-------+
1575 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1576 | | | | || comp|| MTU |
1577 +===================+========================+=========+=================+======+=======+
1578 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1579 +-------------------+------------------------+---------+-----------------+------+-------+
1580 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1581 +-------------------+------------------------+---------+-----------------+------+-------+
1582 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1583 | || RxQs >= rxqs_min_mprq | | | | |
1584 +-------------------+------------------------+---------+-----------------+------+-------+
1585 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1586 | || mprq_en=1 | | | | |
1587 | || RxQs >= rxqs_min_mprq | | | | |
1588 +-------------------+------------------------+---------+-----------------+------+-------+
1590 .. _mlx5_offloads_support:
1592 Supported hardware offloads
1593 ---------------------------
1595 .. table:: Minimal SW/HW versions for queue offloads
1597 ============== ===== ===== ========= ===== ========== =============
1598 Offload DPDK Linux rdma-core OFED firmware hardware
1599 ============== ===== ===== ========= ===== ========== =============
1600 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1601 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1602 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1603 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1604 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1605 Tx scheduling 20.08 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1606 Buffer Split 20.11 N/A N/A 5.1-2 16.28.2006 ConnectX-5
1607 ============== ===== ===== ========= ===== ========== =============
1609 .. table:: Minimal SW/HW versions for rte_flow offloads
1611 +-----------------------+-----------------+-----------------+
1612 | Offload | with E-Switch | with NIC |
1613 +=======================+=================+=================+
1614 | Count | | DPDK 19.05 | | DPDK 19.02 |
1615 | | | OFED 4.6 | | OFED 4.6 |
1616 | | | rdma-core 24 | | rdma-core 23 |
1617 | | | ConnectX-5 | | ConnectX-5 |
1618 +-----------------------+-----------------+-----------------+
1619 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1620 | | | OFED 4.6 | | OFED 4.5 |
1621 | | | rdma-core 24 | | rdma-core 23 |
1622 | | | ConnectX-5 | | ConnectX-4 |
1623 +-----------------------+-----------------+-----------------+
1624 | Queue / RSS | | | | DPDK 18.11 |
1625 | | | N/A | | OFED 4.5 |
1626 | | | | | rdma-core 23 |
1627 | | | | | ConnectX-4 |
1628 +-----------------------+-----------------+-----------------+
1629 | Shared action | | | | |
1630 | | | :numref:`sact`| | :numref:`sact`|
1633 +-----------------------+-----------------+-----------------+
1634 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1635 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1636 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1637 | | of_set_vlan_pcp / | | | | |
1638 | | of_set_vlan_vid) | | | | |
1639 +-----------------------+-----------------+-----------------+
1640 | | VLAN | | DPDK 21.05 | | |
1641 | | ingress and / | | OFED 5.3 | | N/A |
1642 | | of_push_vlan / | | ConnectX-6 Dx | | |
1643 +-----------------------+-----------------+-----------------+
1644 | | VLAN | | DPDK 21.05 | | |
1645 | | egress and / | | OFED 5.3 | | N/A |
1646 | | of_pop_vlan / | | ConnectX-6 Dx | | |
1647 +-----------------------+-----------------+-----------------+
1648 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1649 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1650 | | | rdma-core 24 | | rdma-core 23 |
1651 | | | ConnectX-5 | | ConnectX-5 |
1652 +-----------------------+-----------------+-----------------+
1653 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1654 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1655 | | | rdma-core 27 | | rdma-core 27 |
1656 | | | ConnectX-5 | | ConnectX-5 |
1657 +-----------------------+-----------------+-----------------+
1658 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1659 | | | OFED 5.1-2 | | OFED 5.1-2 |
1660 | | | rdma-core 32 | | N/A |
1661 | | | ConnectX-5 | | ConnectX-5 |
1662 +-----------------------+-----------------+-----------------+
1663 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1664 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1665 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1666 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1667 | | set_ipv6_dst / | | | | |
1668 | | set_tp_src / | | | | |
1669 | | set_tp_dst / | | | | |
1670 | | dec_ttl / | | | | |
1671 | | set_ttl / | | | | |
1672 | | set_mac_src / | | | | |
1673 | | set_mac_dst) | | | | |
1674 +-----------------------+-----------------+-----------------+
1675 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1676 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1677 | | | | rdma-core 24 | | rdma-core 24 |
1678 | | | | ConnectX-5 | | ConnectX-5 |
1679 +-----------------------+-----------------+-----------------+
1680 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1681 | | | OFED 4.7-1 | | OFED 4.7-1 |
1682 | | | rdma-core 24 | | N/A |
1683 | | | ConnectX-5 | | ConnectX-5 |
1684 +-----------------------+-----------------+-----------------+
1685 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1686 | | | OFED 4.6 | | OFED 4.5 |
1687 | | | rdma-core 24 | | rdma-core 23 |
1688 | | | ConnectX-5 | | ConnectX-4 |
1689 +-----------------------+-----------------+-----------------+
1690 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1691 | | | OFED 4.7-3 | | OFED 4.7-3 |
1692 | | | rdma-core 26 | | rdma-core 26 |
1693 | | | ConnectX-5 | | ConnectX-5 |
1694 +-----------------------+-----------------+-----------------+
1695 | Port ID | | DPDK 19.05 | | N/A |
1696 | | | OFED 4.7-1 | | N/A |
1697 | | | rdma-core 24 | | N/A |
1698 | | | ConnectX-5 | | N/A |
1699 +-----------------------+-----------------+-----------------+
1700 | Hairpin | | | | DPDK 19.11 |
1701 | | | N/A | | OFED 4.7-3 |
1702 | | | | | rdma-core 26 |
1703 | | | | | ConnectX-5 |
1704 +-----------------------+-----------------+-----------------+
1705 | 2-port Hairpin | | | | DPDK 20.11 |
1706 | | | N/A | | OFED 5.1-2 |
1708 | | | | | ConnectX-5 |
1709 +-----------------------+-----------------+-----------------+
1710 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1711 | | | OFED 4.7-3 | | OFED 4.7-3 |
1712 | | | rdma-core 26 | | rdma-core 26 |
1713 | | | ConnectX-5 | | ConnectX-5 |
1714 +-----------------------+-----------------+-----------------+
1715 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1716 | | | OFED 5.1-2 | | OFED 5.1-2 |
1717 | | | rdma-core 32 | | N/A |
1718 | | | ConnectX-5 | | ConnectX-5 |
1719 +-----------------------+-----------------+-----------------+
1720 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1721 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1722 | | | rdma-core 35 | | rdma-core 35 |
1723 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1724 +-----------------------+-----------------+-----------------+
1725 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1726 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1727 | | | rdma-core 34 | | rdma-core 34 |
1728 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1729 +-----------------------+-----------------+-----------------+
1730 | Modify Field | | DPDK 21.02 | | DPDK 21.02 |
1731 | | | OFED 5.2 | | OFED 5.2 |
1732 | | | rdma-core 35 | | rdma-core 35 |
1733 | | | ConnectX-5 | | ConnectX-5 |
1734 +-----------------------+-----------------+-----------------+
1735 | Connection tracking | | | | DPDK 21.05 |
1736 | | | N/A | | OFED 5.3 |
1737 | | | | | rdma-core 35 |
1738 | | | | | ConnectX-6 Dx |
1739 +-----------------------+-----------------+-----------------+
1741 .. table:: Minimal SW/HW versions for shared action offload
1744 +-----------------------+-----------------+-----------------+
1745 | Shared Action | with E-Switch | with NIC |
1746 +=======================+=================+=================+
1747 | RSS | | | | DPDK 20.11 |
1748 | | | N/A | | OFED 5.2 |
1749 | | | | | rdma-core 33 |
1750 | | | | | ConnectX-5 |
1751 +-----------------------+-----------------+-----------------+
1752 | Age | | DPDK 20.11 | | DPDK 20.11 |
1753 | | | OFED 5.2 | | OFED 5.2 |
1754 | | | rdma-core 32 | | rdma-core 32 |
1755 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1756 +-----------------------+-----------------+-----------------+
1757 | Count | | DPDK 21.05 | | DPDK 21.05 |
1758 | | | OFED 4.6 | | OFED 4.6 |
1759 | | | rdma-core 24 | | rdma-core 23 |
1760 | | | ConnectX-5 | | ConnectX-5 |
1761 +-----------------------+-----------------+-----------------+
1766 MARK and META items are interrelated with datapath - they might move from/to
1767 the applications in mbuf fields. Hence, zero value for these items has the
1768 special meaning - it means "no metadata are provided", not zero values are
1769 treated by applications and PMD as valid ones.
1771 Moreover in the flow engine domain the value zero is acceptable to match and
1772 set, and we should allow to specify zero values as rte_flow parameters for the
1773 META and MARK items and actions. In the same time zero mask has no meaning and
1774 should be rejected on validation stage.
1779 Flows are not cached in the driver.
1780 When stopping a device port, all the flows created on this port from the
1781 application will be flushed automatically in the background.
1782 After stopping the device port, all flows on this port become invalid and
1783 not represented in the system.
1784 All references to these flows held by the application should be discarded
1785 directly but neither destroyed nor flushed.
1787 The application should re-create the flows as required after the port restart.
1792 Compared to librte_net_mlx4 that implements a single RSS configuration per
1793 port, librte_net_mlx5 supports per-protocol RSS configuration.
1795 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1796 command-line parameter to enable additional protocols (UDP and TCP as well
1797 as IP), the following commands must be entered from its CLI to get the same
1798 behavior as librte_net_mlx4::
1801 > port config all rss all
1807 This section demonstrates how to launch **testpmd** with Mellanox
1808 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1810 #. Load the kernel modules::
1812 modprobe -a ib_uverbs mlx5_core mlx5_ib
1814 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1817 /etc/init.d/openibd restart
1821 User space I/O kernel modules (uio and igb_uio) are not used and do
1822 not have to be loaded.
1824 #. Make sure Ethernet interfaces are in working order and linked to kernel
1825 verbs. Related sysfs entries should be present::
1827 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1836 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1839 for intf in eth2 eth3 eth4 eth5;
1841 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1844 sed -n 's,.*/\(.*\),-a \1,p'
1853 #. Request huge pages::
1855 dpdk-hugepages.py --setup 2G
1857 #. Start testpmd with basic parameters::
1859 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
1864 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1865 EAL: probe driver: 15b3:1013 librte_net_mlx5
1866 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (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:fe
1869 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1870 EAL: probe driver: 15b3:1013 librte_net_mlx5
1871 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1872 PMD: librte_net_mlx5: 1 port(s) detected
1873 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1874 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1875 EAL: probe driver: 15b3:1013 librte_net_mlx5
1876 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1877 PMD: librte_net_mlx5: 1 port(s) detected
1878 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1879 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1880 EAL: probe driver: 15b3:1013 librte_net_mlx5
1881 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1882 PMD: librte_net_mlx5: 1 port(s) detected
1883 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1884 Interactive-mode selected
1885 Configuring Port 0 (socket 0)
1886 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1887 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1888 Port 0: E4:1D:2D:E7:0C:FE
1889 Configuring Port 1 (socket 0)
1890 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1891 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1892 Port 1: E4:1D:2D:E7:0C:FF
1893 Configuring Port 2 (socket 0)
1894 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1895 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1896 Port 2: E4:1D:2D:E7:0C:FA
1897 Configuring Port 3 (socket 0)
1898 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1899 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1900 Port 3: E4:1D:2D:E7:0C:FB
1901 Checking link statuses...
1902 Port 0 Link Up - speed 40000 Mbps - full-duplex
1903 Port 1 Link Up - speed 40000 Mbps - full-duplex
1904 Port 2 Link Up - speed 10000 Mbps - full-duplex
1905 Port 3 Link Up - speed 10000 Mbps - full-duplex
1912 This section demonstrates how to dump flows. Currently, it's possible to dump
1913 all flows with assistance of external tools.
1915 #. 2 ways to get flow raw file:
1917 - Using testpmd CLI:
1919 .. code-block:: console
1922 testpmd> flow dump <port> all <output_file>
1924 testpmd> flow dump <port> rule <rule_id> <output_file>
1926 - call rte_flow_dev_dump api:
1928 .. code-block:: console
1930 rte_flow_dev_dump(port, flow, file, NULL);
1932 #. Dump human-readable flows from raw file:
1934 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1936 .. code-block:: console
1938 mlx_steering_dump.py -f <output_file> -flowptr <flow_ptr>
1940 How to share a meter between ports in the same switch domain
1941 ------------------------------------------------------------
1943 This section demonstrates how to use the shared meter. A meter M can be created
1944 on port X and to be shared with a port Y on the same switch domain by the next way:
1946 .. code-block:: console
1948 flow create X ingress transfer pattern eth / port_id id is Y / end actions meter mtr_id M / end
1950 How to use meter hierarchy
1951 --------------------------
1953 This section demonstrates how to create and use a meter hierarchy.
1954 A termination meter M can be the policy green action of another termination meter N.
1955 The two meters are chained together as a chain. Using meter N in a flow will apply
1956 both the meters in hierarchy on that flow.
1958 .. code-block:: console
1960 add port meter policy 0 1 g_actions queue index 0 / end y_actions end r_actions drop / end
1961 create port meter 0 M 1 1 yes 0xffff 1 0
1962 add port meter policy 0 2 g_actions meter mtr_id M / end y_actions end r_actions drop / end
1963 create port meter 0 N 2 2 yes 0xffff 1 0
1964 flow create 0 ingress group 1 pattern eth / end actions meter mtr_id N / end