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 Group zero's behavior may differ which depends on FW.
206 Matching value equals 0 (value & mask) is not supported.
208 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
210 - Match on Geneve header supports the following fields only:
217 - Match on Geneve TLV option is supported on the following fields:
224 Only one Class/Type/Length Geneve TLV option is supported per shared device.
225 Class/Type/Length fields must be specified as well as masks.
226 Class/Type/Length specified masks must be full.
227 Matching Geneve TLV option without specifying data is not supported.
228 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
230 - VF: flow rules created on VF devices can only match traffic targeted at the
231 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
233 - Match on GTP tunnel header item supports the following fields only:
235 - v_pt_rsv_flags: E flag, S flag, PN flag
239 - Match on GTP extension header only for GTP PDU session container (next
240 extension header type = 0x85).
241 - Match on GTP extension header is not supported in group 0.
243 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
244 The flows within group 0 and set metadata action are rejected by hardware.
248 MAC addresses not already present in the bridge table of the associated
249 kernel network device will be added and cleaned up by the PMD when closing
250 the device. In case of ungraceful program termination, some entries may
251 remain present and should be removed manually by other means.
253 - Buffer split offload is supported with regular Rx burst routine only,
254 no MPRQ feature or vectorized code can be engaged.
256 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
257 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
258 ol_flags. As the mempool for the external buffer is managed by PMD, all the
259 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
260 the external buffers will be freed by PMD and the application which still
261 holds the external buffers may be corrupted.
263 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
264 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
265 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
267 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
268 and allmulticast mode are both set to off.
269 To receive IPv6 Multicast messages on VM, explicitly set the relevant
270 MAC address using rte_eth_dev_mac_addr_add() API.
272 - To support a mixed traffic pattern (some buffers from local host memory, some
273 buffers from other devices) with high bandwidth, a mbuf flag is used.
275 An application hints the PMD whether or not it should try to inline the
276 given mbuf data buffer. PMD should do the best effort to act upon this request.
278 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
279 registered by application with rte_mbuf_dynflag_register(). This flag is
280 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
281 which is intended to be used by the application.
283 To query the supported specific flags in runtime,
284 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
285 currently (over present hardware and configuration) supported specific flags.
286 The "not inline hint" feature operating flow is the following one:
289 - probe the devices, ports are created
290 - query the port capabilities
291 - if port supporting the feature is found
292 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
293 - application starts the ports
294 - on ``dev_start()`` PMD checks whether the feature flag is registered and
295 enables the feature support in datapath
296 - application might set the registered flag bit in ``ol_flags`` field
297 of mbuf being sent and PMD will handle ones appropriately.
299 - The amount of descriptors in Tx queue may be limited by data inline settings.
300 Inline data require the more descriptor building blocks and overall block
301 amount may exceed the hardware supported limits. The application should
302 reduce the requested Tx size or adjust data inline settings with
303 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
305 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
306 parameter should be specified.
307 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
308 being sent it tries to synchronize the time of packet appearing on
309 the wire with the specified packet timestamp. It the specified one
310 is in the past it should be ignored, if one is in the distant future
311 it should be capped with some reasonable value (in range of seconds).
312 These specific cases ("too late" and "distant future") can be optionally
313 reported via device xstats to assist applications to detect the
314 time-related problems.
316 The timestamp upper "too-distant-future" limit
317 at the moment of invoking the Tx burst routine
318 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
319 Please note, for the testpmd txonly mode,
320 the limit is deduced from the expression::
322 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
324 There is no any packet reordering according timestamps is supposed,
325 neither within packet burst, nor between packets, it is an entirely
326 application responsibility to generate packets and its timestamps
327 in desired order. The timestamps can be put only in the first packet
328 in the burst providing the entire burst scheduling.
330 - E-Switch decapsulation Flow:
332 - can be applied to PF port only.
333 - must specify VF port action (packet redirection from PF to VF).
334 - optionally may specify tunnel inner source and destination MAC addresses.
336 - E-Switch encapsulation Flow:
338 - can be applied to VF ports only.
339 - must specify PF port action (packet redirection from VF to PF).
343 - The input buffer, used as outer header, is not validated.
347 - The decapsulation is always done up to the outermost tunnel detected by the HW.
348 - The input buffer, providing the removal size, is not validated.
349 - The buffer size must match the length of the headers to be removed.
351 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
352 mutually exclusive features which cannot be supported together
353 (see :ref:`mlx5_firmware_config`).
357 - Requires DevX and DV flow to be enabled.
358 - KEEP_CRC offload cannot be supported with LRO.
359 - The first mbuf length, without head-room, must be big enough to include the
361 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
362 it with size limited to max LRO size, not to max RX packet length.
363 - LRO can be used with outer header of TCP packets of the standard format:
364 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
366 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
367 - LRO packet aggregation is performed by HW only for packet size larger than
368 ``lro_min_mss_size``. This value is reported on device start, when debug
373 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
374 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
375 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
379 - fast free offload assumes the all mbufs being sent are originated from the
380 same memory pool and there is no any extra references to the mbufs (the
381 reference counter for each mbuf is equal 1 on tx_burst call). The latter
382 means there should be no any externally attached buffers in mbufs. It is
383 an application responsibility to provide the correct mbufs if the fast
384 free offload is engaged. The mlx5 PMD implicitly produces the mbufs with
385 externally attached buffers if MPRQ option is enabled, hence, the fast
386 free offload is neither supported nor advertised if there is MPRQ enabled.
390 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and
391 E-Switch steering domain.
392 - For E-Switch Sampling flow with sample ratio > 1, additional actions are not
393 supported in the sample actions list.
394 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as
395 first action in the E-Switch egress flow if with header modify or
396 encapsulation actions.
397 - For NIC Rx flow, supports ``MARK``, ``COUNT``, ``QUEUE``, ``RSS`` in the
399 - For E-Switch mirroring flow, supports ``RAW ENCAP``, ``Port ID``,
400 ``VXLAN ENCAP``, ``NVGRE ENCAP`` in the sample actions list.
404 - Supports the 'set' operation only for ``RTE_FLOW_ACTION_TYPE_MODIFY_FIELD`` action.
405 - Modification of an arbitrary place in a packet via the special ``RTE_FLOW_FIELD_START`` Field ID is not supported.
406 - Modification of the 802.1Q Tag, VXLAN Network or GENEVE Network ID's is not supported.
407 - Encapsulation levels are not supported, can modify outermost header fields only.
408 - Offsets must be 32-bits aligned, cannot skip past the boundary of a field.
410 - IPv6 header item 'proto' field, indicating the next header protocol, should
411 not be set as extension header.
412 In case the next header is an extension header, it should not be specified in
413 IPv6 header item 'proto' field.
414 The last extension header item 'next header' field can specify the following
415 header protocol type.
419 - 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.
420 - Hairpin in switchdev SR-IOV mode is not supported till now.
424 - All the meter colors with drop action will be counted only by the global drop statistics.
425 - Yellow detection is only supported with ASO metering.
426 - Red color must be with drop action.
427 - Meter statistics are supported only for drop case.
428 - A meter action created with pre-defined policy must be the last action in the flow except single case where the policy actions are:
429 - green: NULL or END.
430 - yellow: NULL or END.
432 - The only supported meter policy actions:
433 - green: QUEUE, RSS, PORT_ID, JUMP, DROP, MARK and SET_TAG.
434 - yellow: QUEUE, RSS, PORT_ID, JUMP, DROP, MARK and SET_TAG.
436 - Policy actions of RSS for green and yellow should have the same configuration except queues.
437 - meter profile packet mode is supported.
441 - Integrity offload is enabled for **ConnectX-6** family.
442 - Verification bits provided by the hardware are ``l3_ok``, ``ipv4_csum_ok``, ``l4_ok``, ``l4_csum_ok``.
443 - ``level`` value 0 references outer headers.
444 - Multiple integrity items not supported in a single flow rule.
445 - Flow rule items supplied by application must explicitly specify network headers referred by integrity item.
446 For example, if integrity item mask sets ``l4_ok`` or ``l4_csum_ok`` bits, reference to L4 network header,
447 TCP or UDP, must be in the rule pattern as well::
449 flow create 0 ingress pattern integrity level is 0 value mask l3_ok value spec l3_ok / eth / ipv6 / end …
451 flow create 0 ingress pattern integrity level is 0 value mask l4_ok value spec 0 / eth / ipv4 proto is udp / end …
453 - Connection tracking:
455 - Cannot co-exist with ASO meter, ASO age action in a single flow rule.
456 - Flow rules insertion rate and memory consumption need more optimization.
458 - 4M connections maximum.
460 - Multi-thread flow insertion:
462 - In order to achieve best insertion rate, application should manage the flows per lcore.
463 - Better to disable memory reclaim by setting ``reclaim_mem_mode`` to 0 to accelerate the flow object allocation and release with cache.
468 MLX5 supports various methods to report statistics:
470 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.
472 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.
474 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.
482 The ibverbs libraries can be linked with this PMD in a number of ways,
483 configured by the ``ibverbs_link`` build option:
485 - ``shared`` (default): the PMD depends on some .so files.
487 - ``dlopen``: Split the dependencies glue in a separate library
488 loaded when needed by dlopen.
489 It make dependencies on libibverbs and libmlx4 optional,
490 and has no performance impact.
492 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
493 in the PMD shared library or the executable static binary.
495 Environment variables
496 ~~~~~~~~~~~~~~~~~~~~~
500 A list of directories in which to search for the rdma-core "glue" plug-in,
501 separated by colons or semi-colons.
503 - ``MLX5_SHUT_UP_BF``
505 Configures HW Tx doorbell register as IO-mapped.
507 By default, the HW Tx doorbell is configured as a write-combining register.
508 The register would be flushed to HW usually when the write-combining buffer
509 becomes full, but it depends on CPU design.
511 Except for vectorized Tx burst routines, a write memory barrier is enforced
512 after updating the register so that the update can be immediately visible to
515 When vectorized Tx burst is called, the barrier is set only if the burst size
516 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
517 variable will bring better latency even though the maximum throughput can
520 Run-time configuration
521 ~~~~~~~~~~~~~~~~~~~~~~
523 - librte_net_mlx5 brings kernel network interfaces up during initialization
524 because it is affected by their state. Forcing them down prevents packets
527 - **ethtool** operations on related kernel interfaces also affect the PMD.
532 In order to run as a non-root user,
533 some capabilities must be granted to the application::
535 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
537 Below are the reasons of the need for each capability:
540 When using physical addresses (PA mode), with Linux >= 4.0,
541 for access to ``/proc/self/pagemap``.
544 For device configuration.
547 For raw ethernet queue allocation through kernel driver.
550 For DMA memory pinning.
555 - ``rxq_cqe_comp_en`` parameter [int]
557 A nonzero value enables the compression of CQE on RX side. This feature
558 allows to save PCI bandwidth and improve performance. Enabled by default.
559 Different compression formats are supported in order to achieve the best
560 performance for different traffic patterns. Default format depends on
561 Multi-Packet Rx queue configuration: Hash RSS format is used in case
562 MPRQ is disabled, Checksum format is used in case MPRQ is enabled.
564 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
565 better compression rate in case of RTE Flow Mark traffic.
566 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
567 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
568 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
569 CQE compression format selection requires DevX to be enabled. If there is
570 no DevX enabled/supported the value is reset to 1 by default.
574 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
575 ConnectX-6 Lx, BlueField and BlueField-2.
576 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
577 ConnectX-6 Lx, BlueField and BlueField-2.
579 - ``rxq_pkt_pad_en`` parameter [int]
581 A nonzero value enables padding Rx packet to the size of cacheline on PCI
582 transaction. This feature would waste PCI bandwidth but could improve
583 performance by avoiding partial cacheline write which may cause costly
584 read-modify-copy in memory transaction on some architectures. Disabled by
589 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
590 ConnectX-6 Lx, BlueField and BlueField-2.
591 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
592 ConnectX-6 Lx, BlueField and BlueField-2.
594 - ``mprq_en`` parameter [int]
596 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
597 configured as Multi-Packet RQ if the total number of Rx queues is
598 ``rxqs_min_mprq`` or more. Disabled by default.
600 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
601 by posting a single large buffer for multiple packets. Instead of posting a
602 buffers per a packet, one large buffer is posted in order to receive multiple
603 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
604 and each stride receives one packet. MPRQ can improve throughput for
605 small-packet traffic.
607 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
608 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
609 configure large stride size enough to accommodate max_rx_pkt_len as long as
610 device allows. Note that this can waste system memory compared to enabling Rx
611 scatter and multi-segment packet.
613 - ``mprq_log_stride_num`` parameter [int]
615 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
616 strides can reduce PCIe traffic further. If configured value is not in the
617 range of device capability, the default value will be set with a warning
618 message. The default value is 4 which is 16 strides per a buffer, valid only
619 if ``mprq_en`` is set.
621 The size of Rx queue should be bigger than the number of strides.
623 - ``mprq_log_stride_size`` parameter [int]
625 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
626 stride size can save some memory and reduce probability of a depletion of all
627 available strides due to unreleased packets by an application. If configured
628 value is not in the range of device capability, the default value will be set
629 with a warning message. The default value is 11 which is 2048 bytes per a
630 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
631 it is possible for a packet to span across multiple strides. This mode allows
632 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
633 of a packet if Rx scatter is configured) may be required in case there is no
634 space left for a head room at the end of a stride which incurs some
637 - ``mprq_max_memcpy_len`` parameter [int]
639 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
640 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
641 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
642 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
643 A mempool for external buffers will be allocated and managed by PMD. If Rx
644 packet is externally attached, ol_flags field of the mbuf will have
645 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
646 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
648 - ``rxqs_min_mprq`` parameter [int]
650 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
651 greater or equal to this value. The default value is 12, valid only if
654 - ``txq_inline`` parameter [int]
656 Amount of data to be inlined during TX operations. This parameter is
657 deprecated and converted to the new parameter ``txq_inline_max`` providing
658 partial compatibility.
660 - ``txqs_min_inline`` parameter [int]
662 Enable inline data send only when the number of TX queues is greater or equal
665 This option should be used in combination with ``txq_inline_max`` and
666 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
668 If this option is not specified the default value 16 is used for BlueField
669 and 8 for other platforms
671 The data inlining consumes the CPU cycles, so this option is intended to
672 auto enable inline data if we have enough Tx queues, which means we have
673 enough CPU cores and PCI bandwidth is getting more critical and CPU
674 is not supposed to be bottleneck anymore.
676 The copying data into WQE improves latency and can improve PPS performance
677 when PCI back pressure is detected and may be useful for scenarios involving
678 heavy traffic on many queues.
680 Because additional software logic is necessary to handle this mode, this
681 option should be used with care, as it may lower performance when back
682 pressure is not expected.
684 If inline data are enabled it may affect the maximal size of Tx queue in
685 descriptors because the inline data increase the descriptor size and
686 queue size limits supported by hardware may be exceeded.
688 - ``txq_inline_min`` parameter [int]
690 Minimal amount of data to be inlined into WQE during Tx operations. NICs
691 may require this minimal data amount to operate correctly. The exact value
692 may depend on NIC operation mode, requested offloads, etc. It is strongly
693 recommended to omit this parameter and use the default values. Anyway,
694 applications using this parameter should take into consideration that
695 specifying an inconsistent value may prevent the NIC from sending packets.
697 If ``txq_inline_min`` key is present the specified value (may be aligned
698 by the driver in order not to exceed the limits and provide better descriptor
699 space utilization) will be used by the driver and it is guaranteed that
700 requested amount of data bytes are inlined into the WQE beside other inline
701 settings. This key also may update ``txq_inline_max`` value (default
702 or specified explicitly in devargs) to reserve the space for inline data.
704 If ``txq_inline_min`` key is not present, the value may be queried by the
705 driver from the NIC via DevX if this feature is available. If there is no DevX
706 enabled/supported the value 18 (supposing L2 header including VLAN) is set
707 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
708 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
711 For ConnectX-4 NIC, driver does not allow specifying value below 18
712 (minimal L2 header, including VLAN), error will be raised.
714 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
715 it is not recommended and may prevent NIC from sending packets over
718 For ConnectX-4 and ConnectX-4 Lx NICs, automatically configured value
719 is insufficient for some traffic, because they require at least all L2 headers
720 to be inlined. For example, Q-in-Q adds 4 bytes to default 18 bytes
721 of Ethernet and VLAN, thus ``txq_inline_min`` must be set to 22.
722 MPLS would add 4 bytes per label. Final value must account for all possible
723 L2 encapsulation headers used in particular environment.
725 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
726 Multi-Packet Write), because last one does not support partial packet inlining.
727 This is not very critical due to minimal data inlining is mostly required
728 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
730 - ``txq_inline_max`` parameter [int]
732 Specifies the maximal packet length to be completely inlined into WQE
733 Ethernet Segment for ordinary SEND method. If packet is larger than specified
734 value, the packet data won't be copied by the driver at all, data buffer
735 is addressed with a pointer. If packet length is less or equal all packet
736 data will be copied into WQE. This may improve PCI bandwidth utilization for
737 short packets significantly but requires the extra CPU cycles.
739 The data inline feature is controlled by number of Tx queues, if number of Tx
740 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
741 is engaged, if there are not enough Tx queues (which means not enough CPU cores
742 and CPU resources are scarce), data inline is not performed by the driver.
743 Assigning ``txqs_min_inline`` with zero always enables the data inline.
745 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
746 by the driver in order not to exceed the limit (930 bytes) and to provide better
747 WQE space filling without gaps, the adjustment is reflected in the debug log.
748 Also, the default value (290) may be decreased in run-time if the large transmit
749 queue size is requested and hardware does not support enough descriptor
750 amount, in this case warning is emitted. If ``txq_inline_max`` key is
751 specified and requested inline settings can not be satisfied then error
754 - ``txq_inline_mpw`` parameter [int]
756 Specifies the maximal packet length to be completely inlined into WQE for
757 Enhanced MPW method. If packet is large the specified value, the packet data
758 won't be copied, and data buffer is addressed with pointer. If packet length
759 is less or equal, all packet data will be copied into WQE. This may improve PCI
760 bandwidth utilization for short packets significantly but requires the extra
763 The data inline feature is controlled by number of TX queues, if number of Tx
764 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
765 is engaged, if there are not enough Tx queues (which means not enough CPU cores
766 and CPU resources are scarce), data inline is not performed by the driver.
767 Assigning ``txqs_min_inline`` with zero always enables the data inline.
769 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
770 by the driver in order not to exceed the limit (930 bytes) and to provide better
771 WQE space filling without gaps, the adjustment is reflected in the debug log.
772 Due to multiple packets may be included to the same WQE with Enhanced Multi
773 Packet Write Method and overall WQE size is limited it is not recommended to
774 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
775 may be decreased in run-time if the large transmit queue size is requested
776 and hardware does not support enough descriptor amount, in this case warning
777 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
778 settings can not be satisfied then error will be raised.
780 - ``txqs_max_vec`` parameter [int]
782 Enable vectorized Tx only when the number of TX queues is less than or
783 equal to this value. This parameter is deprecated and ignored, kept
784 for compatibility issue to not prevent driver from probing.
786 - ``txq_mpw_hdr_dseg_en`` parameter [int]
788 A nonzero value enables including two pointers in the first block of TX
789 descriptor. The parameter is deprecated and ignored, kept for compatibility
792 - ``txq_max_inline_len`` parameter [int]
794 Maximum size of packet to be inlined. This limits the size of packet to
795 be inlined. If the size of a packet is larger than configured value, the
796 packet isn't inlined even though there's enough space remained in the
797 descriptor. Instead, the packet is included with pointer. This parameter
798 is deprecated and converted directly to ``txq_inline_mpw`` providing full
799 compatibility. Valid only if eMPW feature is engaged.
801 - ``txq_mpw_en`` parameter [int]
803 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
804 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
805 eMPW allows the Tx burst function to pack up multiple packets
806 in a single descriptor session in order to save PCI bandwidth
807 and improve performance at the cost of a slightly higher CPU usage.
808 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
809 Tx burst function copies entire packet data on to Tx descriptor
810 instead of including pointer of packet.
812 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
813 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
814 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
815 option or reported by the NIC, the eMPW feature is disengaged.
817 - ``tx_db_nc`` parameter [int]
819 The rdma core library can map doorbell register in two ways, depending on the
820 environment variable "MLX5_SHUT_UP_BF":
822 - As regular cached memory (usually with write combining attribute), if the
823 variable is either missing or set to zero.
824 - As non-cached memory, if the variable is present and set to not "0" value.
826 The type of mapping may slightly affect the Tx performance, the optimal choice
827 is strongly relied on the host architecture and should be deduced practically.
829 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
830 memory (with write combining), the PMD will perform the extra write memory barrier
831 after writing to doorbell, it might increase the needed CPU clocks per packet
832 to send, but latency might be improved.
834 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
835 cached memory, the PMD will not perform the extra write memory barrier
836 after writing to doorbell, on some architectures it might improve the
839 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
840 memory, the PMD will use heuristics to decide whether write memory barrier
841 should be performed. For bursts with size multiple of recommended one (64 pkts)
842 it is supposed the next burst is coming and no need to issue the extra memory
843 barrier (it is supposed to be issued in the next coming burst, at least after
844 descriptor writing). It might increase latency (on some hosts till next
845 packets transmit) and should be used with care.
847 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
848 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
849 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
851 - ``tx_pp`` parameter [int]
853 If a nonzero value is specified the driver creates all necessary internal
854 objects to provide accurate packet send scheduling on mbuf timestamps.
855 The positive value specifies the scheduling granularity in nanoseconds,
856 the packet send will be accurate up to specified digits. The allowed range is
857 from 500 to 1 million of nanoseconds. The negative value specifies the module
858 of granularity and engages the special test mode the check the schedule rate.
859 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
862 - ``tx_skew`` parameter [int]
864 The parameter adjusts the send packet scheduling on timestamps and represents
865 the average delay between beginning of the transmitting descriptor processing
866 by the hardware and appearance of actual packet data on the wire. The value
867 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
868 specified. The default value is zero.
870 - ``tx_vec_en`` parameter [int]
872 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
873 ConnectX-6 Lx, BlueField and BlueField-2 NICs
874 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
875 The parameter is deprecated and ignored.
877 - ``rx_vec_en`` parameter [int]
879 A nonzero value enables Rx vector if the port is not configured in
880 multi-segment otherwise this parameter is ignored.
884 - ``vf_nl_en`` parameter [int]
886 A nonzero value enables Netlink requests from the VF to add/remove MAC
887 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
888 Otherwise the relevant configuration must be run with Linux iproute2 tools.
889 This is a prerequisite to receive this kind of traffic.
891 Enabled by default, valid only on VF devices ignored otherwise.
893 - ``l3_vxlan_en`` parameter [int]
895 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
896 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
897 parameter. This is a prerequisite to receive this kind of traffic.
901 - ``dv_xmeta_en`` parameter [int]
903 A nonzero value enables extensive flow metadata support if device is
904 capable and driver supports it. This can enable extensive support of
905 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
906 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
908 There are some possible configurations, depending on parameter value:
910 - 0, this is default value, defines the legacy mode, the ``MARK`` and
911 ``META`` related actions and items operate only within NIC Tx and
912 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
913 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
914 item is 32 bits wide and match supported on egress only.
916 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
917 related actions and items operate within all supported steering domains,
918 including FDB, ``MARK`` and ``META`` information may cross the domain
919 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
920 depends on kernel and firmware configurations and might be 0, 16 or
921 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
922 compatibility, the actual width of data transferred to the FDB domain
923 depends on kernel configuration and may be vary. The actual supported
924 width can be retrieved in runtime by series of rte_flow_validate()
927 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
928 related actions and items operate within all supported steering domains,
929 including FDB, ``MARK`` and ``META`` information may cross the domain
930 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
931 depends on kernel and firmware configurations and might be 0, 16 or
932 24 bits. The actual supported width can be retrieved in runtime by
933 series of rte_flow_validate() trials.
935 - 3, this engages tunnel offload mode. In E-Switch configuration, that
936 mode implicitly activates ``dv_xmeta_en=1``.
938 +------+-----------+-----------+-------------+-------------+
939 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
940 +======+===========+===========+=============+=============+
941 | 0 | 24 bits | 32 bits | 32 bits | no |
942 +------+-----------+-----------+-------------+-------------+
943 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
944 +------+-----------+-----------+-------------+-------------+
945 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
946 +------+-----------+-----------+-------------+-------------+
948 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
949 ignored and the device is configured to operate in legacy mode (0).
951 Disabled by default (set to 0).
953 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
954 of the extensive metadata features. The legacy Verbs supports FLAG and
955 MARK metadata actions over NIC Rx steering domain only.
957 Setting META value to zero in flow action means there is no item provided
958 and receiving datapath will not report in mbufs the metadata are present.
959 Setting MARK value to zero in flow action means the zero FDIR ID value
960 will be reported on packet receiving.
962 For the MARK action the last 16 values in the full range are reserved for
963 internal PMD purposes (to emulate FLAG action). The valid range for the
964 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
965 for the 24-bit mode, the flows with the MARK action value outside
966 the specified range will be rejected.
968 - ``dv_flow_en`` parameter [int]
970 A nonzero value enables the DV flow steering assuming it is supported
971 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
973 Enabled by default if supported.
975 - ``dv_esw_en`` parameter [int]
977 A nonzero value enables E-Switch using Direct Rules.
979 Enabled by default if supported.
981 - ``lacp_by_user`` parameter [int]
983 A nonzero value enables the control of LACP traffic by the user application.
984 When a bond exists in the driver, by default it should be managed by the
985 kernel and therefore LACP traffic should be steered to the kernel.
986 If this devarg is set to 1 it will allow the user to manage the bond by
987 itself and not steer LACP traffic to the kernel.
989 Disabled by default (set to 0).
991 - ``mr_ext_memseg_en`` parameter [int]
993 A nonzero value enables extending memseg when registering DMA memory. If
994 enabled, the number of entries in MR (Memory Region) lookup table on datapath
995 is minimized and it benefits performance. On the other hand, it worsens memory
996 utilization because registered memory is pinned by kernel driver. Even if a
997 page in the extended chunk is freed, that doesn't become reusable until the
998 entire memory is freed.
1002 - ``representor`` parameter [list]
1004 This parameter can be used to instantiate DPDK Ethernet devices from
1005 existing port (PF, VF or SF) representors configured on the device.
1007 It is a standard parameter whose format is described in
1008 :ref:`ethernet_device_standard_device_arguments`.
1010 For instance, to probe VF port representors 0 through 2::
1012 <PCI_BDF>,representor=vf[0-2]
1014 To probe SF port representors 0 through 2::
1016 <PCI_BDF>,representor=sf[0-2]
1018 To probe VF port representors 0 through 2 on both PFs of bonding device::
1020 <Primary_PCI_BDF>,representor=pf[0,1]vf[0-2]
1022 - ``max_dump_files_num`` parameter [int]
1024 The maximum number of files per PMD entity that may be created for debug information.
1025 The files will be created in /var/log directory or in current directory.
1027 set to 128 by default.
1029 - ``lro_timeout_usec`` parameter [int]
1031 The maximum allowed duration of an LRO session, in micro-seconds.
1032 PMD will set the nearest value supported by HW, which is not bigger than
1033 the input ``lro_timeout_usec`` value.
1034 If this parameter is not specified, by default PMD will set
1035 the smallest value supported by HW.
1037 - ``hp_buf_log_sz`` parameter [int]
1039 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
1040 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
1041 The capacity of the value is specified by the firmware and the initialization
1042 will get a failure if it is out of scope.
1043 The range of the value is from 11 to 19 right now, and the supported frame
1044 size of a single packet for hairpin is from 512B to 128KB. It might change if
1045 different firmware release is being used. By using a small value, it could
1046 reduce memory consumption but not work with a large frame. If the value is
1047 too large, the memory consumption will be high and some potential performance
1048 degradation will be introduced.
1049 By default, the PMD will set this value to 16, which means that 9KB jumbo
1050 frames will be supported.
1052 - ``reclaim_mem_mode`` parameter [int]
1054 Cache some resources in flow destroy will help flow recreation more efficient.
1055 While some systems may require the all the resources can be reclaimed after
1057 The parameter ``reclaim_mem_mode`` provides the option for user to configure
1058 if the resource cache is needed or not.
1060 There are three options to choose:
1062 - 0. It means the flow resources will be cached as usual. The resources will
1063 be cached, helpful with flow insertion rate.
1065 - 1. It will only enable the DPDK PMD level resources reclaim.
1067 - 2. Both DPDK PMD level and rdma-core low level will be configured as
1070 By default, the PMD will set this value to 0.
1072 - ``sys_mem_en`` parameter [int]
1074 A non-zero value enables the PMD memory management allocating memory
1075 from system by default, without explicit rte memory flag.
1077 By default, the PMD will set this value to 0.
1079 - ``decap_en`` parameter [int]
1081 Some devices do not support FCS (frame checksum) scattering for
1082 tunnel-decapsulated packets.
1083 If set to 0, this option forces the FCS feature and rejects tunnel
1084 decapsulation in the flow engine for such devices.
1086 By default, the PMD will set this value to 1.
1088 - ``allow_duplicate_pattern`` parameter [int]
1090 There are two options to choose:
1092 - 0. Prevent insertion of rules with the same pattern items on non-root table.
1093 In this case, only the first rule is inserted and the following rules are
1094 rejected and error code EEXIST is returned.
1096 - 1. Allow insertion of rules with the same pattern items.
1097 In this case, all rules are inserted but only the first rule takes effect,
1098 the next rule takes effect only if the previous rules are deleted.
1100 By default, the PMD will set this value to 1.
1102 .. _mlx5_firmware_config:
1104 Firmware configuration
1105 ~~~~~~~~~~~~~~~~~~~~~~
1107 Firmware features can be configured as key/value pairs.
1109 The command to set a value is::
1111 mlxconfig -d <device> set <key>=<value>
1113 The command to query a value is::
1115 mlxconfig -d <device> query | grep <key>
1117 The device name for the command ``mlxconfig`` can be either the PCI address,
1118 or the mst device name found with::
1122 Below are some firmware configurations listed.
1128 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1134 - maximum number of SR-IOV virtual functions::
1138 - enable DevX (required by Direct Rules and other features)::
1142 - aggressive CQE zipping::
1146 - L3 VXLAN and VXLAN-GPE destination UDP port::
1149 IP_OVER_VXLAN_PORT=<udp dport>
1151 - enable VXLAN-GPE tunnel flow matching::
1153 FLEX_PARSER_PROFILE_ENABLE=0
1155 FLEX_PARSER_PROFILE_ENABLE=2
1157 - enable IP-in-IP tunnel flow matching::
1159 FLEX_PARSER_PROFILE_ENABLE=0
1161 - enable MPLS flow matching::
1163 FLEX_PARSER_PROFILE_ENABLE=1
1165 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1167 FLEX_PARSER_PROFILE_ENABLE=2
1169 - enable Geneve flow matching::
1171 FLEX_PARSER_PROFILE_ENABLE=0
1173 FLEX_PARSER_PROFILE_ENABLE=1
1175 - enable Geneve TLV option flow matching::
1177 FLEX_PARSER_PROFILE_ENABLE=0
1179 - enable GTP flow matching::
1181 FLEX_PARSER_PROFILE_ENABLE=3
1183 - enable eCPRI flow matching::
1185 FLEX_PARSER_PROFILE_ENABLE=4
1191 This driver relies on external libraries and kernel drivers for resources
1192 allocations and initialization. The following dependencies are not part of
1193 DPDK and must be installed separately:
1197 User space Verbs framework used by librte_net_mlx5. This library provides
1198 a generic interface between the kernel and low-level user space drivers
1201 It allows slow and privileged operations (context initialization, hardware
1202 resources allocations) to be managed by the kernel and fast operations to
1203 never leave user space.
1207 Low-level user space driver library for Mellanox
1208 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1211 This library basically implements send/receive calls to the hardware
1214 - **Kernel modules**
1216 They provide the kernel-side Verbs API and low level device drivers that
1217 manage actual hardware initialization and resources sharing with user
1220 Unlike most other PMDs, these modules must remain loaded and bound to
1223 - mlx5_core: hardware driver managing Mellanox
1224 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1226 - mlx5_ib: InifiniBand device driver.
1227 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1229 - **Firmware update**
1231 Mellanox OFED/EN releases include firmware updates for
1232 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1234 Because each release provides new features, these updates must be applied to
1235 match the kernel modules and libraries they come with.
1239 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1245 Either RDMA Core library with a recent enough Linux kernel release
1246 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1249 RDMA Core with Linux Kernel
1250 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1252 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1253 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1254 (see `RDMA Core installation documentation`_)
1255 - When building for i686 use:
1257 - rdma-core version 18.0 or above built with 32bit support.
1258 - Kernel version 4.14.41 or above.
1260 - Starting with rdma-core v21, static libraries can be built::
1263 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1266 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1267 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1273 - Mellanox OFED version: **4.5** and above /
1274 Mellanox EN version: **4.5** and above
1277 - ConnectX-4: **12.21.1000** and above.
1278 - ConnectX-4 Lx: **14.21.1000** and above.
1279 - ConnectX-5: **16.21.1000** and above.
1280 - ConnectX-5 Ex: **16.21.1000** and above.
1281 - ConnectX-6: **20.27.0090** and above.
1282 - ConnectX-6 Dx: **22.27.0090** and above.
1283 - BlueField: **18.25.1010** and above.
1285 While these libraries and kernel modules are available on OpenFabrics
1286 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1287 managers on most distributions, this PMD requires Ethernet extensions that
1288 may not be supported at the moment (this is a work in progress).
1291 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1293 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1294 include the necessary support and should be used in the meantime. For DPDK,
1295 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1296 required from that distribution.
1300 Several versions of Mellanox OFED/EN are available. Installing the version
1301 this DPDK release was developed and tested against is strongly
1302 recommended. Please check the `linux prerequisites`_.
1304 Windows Prerequisites
1305 ---------------------
1307 This driver relies on external libraries and kernel drivers for resources
1308 allocations and initialization. The dependencies in the following sub-sections
1309 are not part of DPDK, and must be installed separately.
1311 Compilation Prerequisites
1312 ~~~~~~~~~~~~~~~~~~~~~~~~~
1314 DevX SDK installation
1315 ^^^^^^^^^^^^^^^^^^^^^
1317 The DevX SDK must be installed on the machine building the Windows PMD.
1318 Additional information can be found at
1319 `How to Integrate Windows DevX in Your Development Environment
1320 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1322 Runtime Prerequisites
1323 ~~~~~~~~~~~~~~~~~~~~~
1325 WinOF2 version 2.60 or higher must be installed on the machine.
1330 The driver can be downloaded from the following site:
1332 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1337 DevX for Windows must be enabled in the Windows registry.
1338 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1339 Additional information can be found in the WinOF2 user manual.
1344 The following Mellanox device families are supported by the same mlx5 driver:
1356 Below are detailed device names:
1358 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1359 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1360 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1361 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1362 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1363 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1364 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1365 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1366 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1367 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1368 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1369 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1370 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1371 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1372 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1373 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1374 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1375 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1376 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1377 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1378 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1379 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1380 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1381 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1382 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1383 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1385 Quick Start Guide on OFED/EN
1386 ----------------------------
1388 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1391 2. Install the required libraries and kernel modules either by installing
1392 only the required set, or by installing the entire Mellanox OFED/EN::
1394 ./mlnxofedinstall --upstream-libs --dpdk
1396 3. Verify the firmware is the correct one::
1400 4. Verify all ports links are set to Ethernet::
1402 mlxconfig -d <mst device> query | grep LINK_TYPE
1406 Link types may have to be configured to Ethernet::
1408 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1410 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1412 For hypervisors, verify SR-IOV is enabled on the NIC::
1414 mlxconfig -d <mst device> query | grep SRIOV_EN
1417 If needed, configure SR-IOV::
1419 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1420 mlxfwreset -d <mst device> reset
1422 5. Restart the driver::
1424 /etc/init.d/openibd restart
1428 service openibd restart
1430 If link type was changed, firmware must be reset as well::
1432 mlxfwreset -d <mst device> reset
1434 For hypervisors, after reset write the sysfs number of virtual functions
1437 To dynamically instantiate a given number of virtual functions (VFs)::
1439 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1441 6. Install DPDK and you are ready to go.
1442 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1444 Enable switchdev mode
1445 ---------------------
1447 Switchdev mode is a mode in E-Switch, that binds between representor and VF or SF.
1448 Representor is a port in DPDK that is connected to a VF or SF in such a way
1449 that assuming there are no offload flows, each packet that is sent from the VF or SF
1450 will be received by the corresponding representor. While each packet that is or SF
1451 sent to a representor will be received by the VF or SF.
1452 This is very useful in case of SRIOV mode, where the first packet that is sent
1453 by the VF or SF will be received by the DPDK application which will decide if this
1454 flow should be offloaded to the E-Switch. After offloading the flow packet
1455 that the VF or SF that are matching the flow will not be received any more by
1456 the DPDK application.
1458 1. Enable SRIOV mode::
1460 mlxconfig -d <mst device> set SRIOV_EN=true
1462 2. Configure the max number of VFs::
1464 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1468 mlxfwreset -d <mst device> reset
1470 3. Configure the actual number of VFs::
1472 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1474 4. Unbind the device (can be rebind after the switchdev mode)::
1476 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1478 5. Enbale switchdev mode::
1480 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1482 Sub-Function support
1483 --------------------
1485 Sub-Function is a portion of the PCI device, a SF netdev has its own
1486 dedicated queues (txq, rxq).
1487 A SF shares PCI level resources with other SFs and/or with its parent PCI function.
1491 OFED version >= 5.4-0.3.3.0
1493 1. Configure SF feature::
1495 # Run mlxconfig on both PFs on host and ECPFs on BlueField.
1496 mlxconfig -d <mst device> set PER_PF_NUM_SF=1 PF_TOTAL_SF=252 PF_SF_BAR_SIZE=12
1498 2. Enable switchdev mode::
1500 mlxdevm dev eswitch set pci/<DBDF> mode switchdev
1504 mlxdevm port add pci/<DBDF> flavour pcisf pfnum 0 sfnum <sfnum>
1506 Get SFID from output: pci/<DBDF>/<SFID>
1508 4. Modify MAC address::
1510 mlxdevm port function set pci/<DBDF>/<SFID> hw_addr <MAC>
1512 5. Activate SF port::
1514 mlxdevm port function set pci/<DBDF>/<ID> state active
1516 6. Devargs to probe SF device::
1518 auxiliary:mlx5_core.sf.<num>,dv_flow_en=1
1520 Sub-Function representor support
1521 --------------------------------
1523 A SF netdev supports E-Switch representation offload
1524 similar to PF and VF representors.
1525 Use <sfnum> to probe SF representor::
1527 testpmd> port attach <PCI_BDF>,representor=sf<sfnum>,dv_flow_en=1
1532 1. Configure aggressive CQE Zipping for maximum performance::
1534 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1536 To set it back to the default CQE Zipping mode use::
1538 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1540 2. In case of virtualization:
1542 - Make sure that hypervisor kernel is 3.16 or newer.
1543 - Configure boot with ``iommu=pt``.
1544 - Use 1G huge pages.
1545 - Make sure to allocate a VM on huge pages.
1546 - Make sure to set CPU pinning.
1548 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1549 for better performance. For VMs, verify that the right CPU
1550 and NUMA node are pinned according to the above. Run::
1552 lstopo-no-graphics --merge
1554 to identify the NUMA node to which the PCIe adapter is connected.
1556 4. If more than one adapter is used, and root complex capabilities allow
1557 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1558 it is recommended to locate both adapters on the same NUMA node.
1559 This in order to forward packets from one to the other without
1560 NUMA performance penalty.
1562 5. Disable pause frames::
1564 ethtool -A <netdev> rx off tx off
1566 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1567 via the BIOS configuration. Please contact you server provider for more
1568 information about the settings.
1572 On some machines, depends on the machine integrator, it is beneficial
1573 to set the PCI max read request parameter to 1K. This can be
1574 done in the following way:
1576 To query the read request size use::
1578 setpci -s <NIC PCI address> 68.w
1580 If the output is different than 3XXX, set it by::
1582 setpci -s <NIC PCI address> 68.w=3XXX
1584 The XXX can be different on different systems. Make sure to configure
1585 according to the setpci output.
1587 7. To minimize overhead of searching Memory Regions:
1589 - '--socket-mem' is recommended to pin memory by predictable amount.
1590 - Configure per-lcore cache when creating Mempools for packet buffer.
1591 - Refrain from dynamically allocating/freeing memory in run-time.
1596 There are multiple Rx burst functions with different advantages and limitations.
1598 .. table:: Rx burst functions
1600 +-------------------+------------------------+---------+-----------------+------+-------+
1601 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1602 | | | | || comp|| MTU |
1603 +===================+========================+=========+=================+======+=======+
1604 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1605 +-------------------+------------------------+---------+-----------------+------+-------+
1606 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1607 +-------------------+------------------------+---------+-----------------+------+-------+
1608 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1609 | || RxQs >= rxqs_min_mprq | | | | |
1610 +-------------------+------------------------+---------+-----------------+------+-------+
1611 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1612 | || mprq_en=1 | | | | |
1613 | || RxQs >= rxqs_min_mprq | | | | |
1614 +-------------------+------------------------+---------+-----------------+------+-------+
1616 .. _mlx5_offloads_support:
1618 Supported hardware offloads
1619 ---------------------------
1621 .. table:: Minimal SW/HW versions for queue offloads
1623 ============== ===== ===== ========= ===== ========== =============
1624 Offload DPDK Linux rdma-core OFED firmware hardware
1625 ============== ===== ===== ========= ===== ========== =============
1626 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1627 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1628 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1629 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1630 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1631 Tx scheduling 20.08 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1632 Buffer Split 20.11 N/A N/A 5.1-2 16.28.2006 ConnectX-5
1633 ============== ===== ===== ========= ===== ========== =============
1635 .. table:: Minimal SW/HW versions for rte_flow offloads
1637 +-----------------------+-----------------+-----------------+
1638 | Offload | with E-Switch | with NIC |
1639 +=======================+=================+=================+
1640 | Count | | DPDK 19.05 | | DPDK 19.02 |
1641 | | | OFED 4.6 | | OFED 4.6 |
1642 | | | rdma-core 24 | | rdma-core 23 |
1643 | | | ConnectX-5 | | ConnectX-5 |
1644 +-----------------------+-----------------+-----------------+
1645 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1646 | | | OFED 4.6 | | OFED 4.5 |
1647 | | | rdma-core 24 | | rdma-core 23 |
1648 | | | ConnectX-5 | | ConnectX-4 |
1649 +-----------------------+-----------------+-----------------+
1650 | Queue / RSS | | | | DPDK 18.11 |
1651 | | | N/A | | OFED 4.5 |
1652 | | | | | rdma-core 23 |
1653 | | | | | ConnectX-4 |
1654 +-----------------------+-----------------+-----------------+
1655 | Shared action | | | | |
1656 | | | :numref:`sact`| | :numref:`sact`|
1659 +-----------------------+-----------------+-----------------+
1660 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1661 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1662 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1663 | | of_set_vlan_pcp / | | | | |
1664 | | of_set_vlan_vid) | | | | |
1665 +-----------------------+-----------------+-----------------+
1666 | | VLAN | | DPDK 21.05 | | |
1667 | | ingress and / | | OFED 5.3 | | N/A |
1668 | | of_push_vlan / | | ConnectX-6 Dx | | |
1669 +-----------------------+-----------------+-----------------+
1670 | | VLAN | | DPDK 21.05 | | |
1671 | | egress and / | | OFED 5.3 | | N/A |
1672 | | of_pop_vlan / | | ConnectX-6 Dx | | |
1673 +-----------------------+-----------------+-----------------+
1674 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1675 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1676 | | | rdma-core 24 | | rdma-core 23 |
1677 | | | ConnectX-5 | | ConnectX-5 |
1678 +-----------------------+-----------------+-----------------+
1679 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1680 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1681 | | | rdma-core 27 | | rdma-core 27 |
1682 | | | ConnectX-5 | | ConnectX-5 |
1683 +-----------------------+-----------------+-----------------+
1684 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1685 | | | OFED 5.1-2 | | OFED 5.1-2 |
1686 | | | rdma-core 32 | | N/A |
1687 | | | ConnectX-5 | | ConnectX-5 |
1688 +-----------------------+-----------------+-----------------+
1689 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1690 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1691 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1692 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1693 | | set_ipv6_dst / | | | | |
1694 | | set_tp_src / | | | | |
1695 | | set_tp_dst / | | | | |
1696 | | dec_ttl / | | | | |
1697 | | set_ttl / | | | | |
1698 | | set_mac_src / | | | | |
1699 | | set_mac_dst) | | | | |
1700 +-----------------------+-----------------+-----------------+
1701 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1702 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1703 | | | | rdma-core 24 | | rdma-core 24 |
1704 | | | | ConnectX-5 | | ConnectX-5 |
1705 +-----------------------+-----------------+-----------------+
1706 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1707 | | | OFED 4.7-1 | | OFED 4.7-1 |
1708 | | | rdma-core 24 | | N/A |
1709 | | | ConnectX-5 | | ConnectX-5 |
1710 +-----------------------+-----------------+-----------------+
1711 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1712 | | | OFED 4.6 | | OFED 4.5 |
1713 | | | rdma-core 24 | | rdma-core 23 |
1714 | | | ConnectX-5 | | ConnectX-4 |
1715 +-----------------------+-----------------+-----------------+
1716 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1717 | | | OFED 4.7-3 | | OFED 4.7-3 |
1718 | | | rdma-core 26 | | rdma-core 26 |
1719 | | | ConnectX-5 | | ConnectX-5 |
1720 +-----------------------+-----------------+-----------------+
1721 | Port ID | | DPDK 19.05 | | N/A |
1722 | | | OFED 4.7-1 | | N/A |
1723 | | | rdma-core 24 | | N/A |
1724 | | | ConnectX-5 | | N/A |
1725 +-----------------------+-----------------+-----------------+
1726 | Hairpin | | | | DPDK 19.11 |
1727 | | | N/A | | OFED 4.7-3 |
1728 | | | | | rdma-core 26 |
1729 | | | | | ConnectX-5 |
1730 +-----------------------+-----------------+-----------------+
1731 | 2-port Hairpin | | | | DPDK 20.11 |
1732 | | | N/A | | OFED 5.1-2 |
1734 | | | | | ConnectX-5 |
1735 +-----------------------+-----------------+-----------------+
1736 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1737 | | | OFED 4.7-3 | | OFED 4.7-3 |
1738 | | | rdma-core 26 | | rdma-core 26 |
1739 | | | ConnectX-5 | | ConnectX-5 |
1740 +-----------------------+-----------------+-----------------+
1741 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1742 | | | OFED 5.1-2 | | OFED 5.1-2 |
1743 | | | rdma-core 32 | | N/A |
1744 | | | ConnectX-5 | | ConnectX-5 |
1745 +-----------------------+-----------------+-----------------+
1746 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1747 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1748 | | | rdma-core 35 | | rdma-core 35 |
1749 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1750 +-----------------------+-----------------+-----------------+
1751 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1752 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1753 | | | rdma-core 34 | | rdma-core 34 |
1754 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1755 +-----------------------+-----------------+-----------------+
1756 | Modify Field | | DPDK 21.02 | | DPDK 21.02 |
1757 | | | OFED 5.2 | | OFED 5.2 |
1758 | | | rdma-core 35 | | rdma-core 35 |
1759 | | | ConnectX-5 | | ConnectX-5 |
1760 +-----------------------+-----------------+-----------------+
1761 | Connection tracking | | | | DPDK 21.05 |
1762 | | | N/A | | OFED 5.3 |
1763 | | | | | rdma-core 35 |
1764 | | | | | ConnectX-6 Dx |
1765 +-----------------------+-----------------+-----------------+
1767 .. table:: Minimal SW/HW versions for shared action offload
1770 +-----------------------+-----------------+-----------------+
1771 | Shared Action | with E-Switch | with NIC |
1772 +=======================+=================+=================+
1773 | RSS | | | | DPDK 20.11 |
1774 | | | N/A | | OFED 5.2 |
1775 | | | | | rdma-core 33 |
1776 | | | | | ConnectX-5 |
1777 +-----------------------+-----------------+-----------------+
1778 | Age | | DPDK 20.11 | | DPDK 20.11 |
1779 | | | OFED 5.2 | | OFED 5.2 |
1780 | | | rdma-core 32 | | rdma-core 32 |
1781 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1782 +-----------------------+-----------------+-----------------+
1783 | Count | | DPDK 21.05 | | DPDK 21.05 |
1784 | | | OFED 4.6 | | OFED 4.6 |
1785 | | | rdma-core 24 | | rdma-core 23 |
1786 | | | ConnectX-5 | | ConnectX-5 |
1787 +-----------------------+-----------------+-----------------+
1792 MARK and META items are interrelated with datapath - they might move from/to
1793 the applications in mbuf fields. Hence, zero value for these items has the
1794 special meaning - it means "no metadata are provided", not zero values are
1795 treated by applications and PMD as valid ones.
1797 Moreover in the flow engine domain the value zero is acceptable to match and
1798 set, and we should allow to specify zero values as rte_flow parameters for the
1799 META and MARK items and actions. In the same time zero mask has no meaning and
1800 should be rejected on validation stage.
1805 Flows are not cached in the driver.
1806 When stopping a device port, all the flows created on this port from the
1807 application will be flushed automatically in the background.
1808 After stopping the device port, all flows on this port become invalid and
1809 not represented in the system.
1810 All references to these flows held by the application should be discarded
1811 directly but neither destroyed nor flushed.
1813 The application should re-create the flows as required after the port restart.
1818 Compared to librte_net_mlx4 that implements a single RSS configuration per
1819 port, librte_net_mlx5 supports per-protocol RSS configuration.
1821 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1822 command-line parameter to enable additional protocols (UDP and TCP as well
1823 as IP), the following commands must be entered from its CLI to get the same
1824 behavior as librte_net_mlx4::
1827 > port config all rss all
1833 This section demonstrates how to launch **testpmd** with Mellanox
1834 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1836 #. Load the kernel modules::
1838 modprobe -a ib_uverbs mlx5_core mlx5_ib
1840 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1843 /etc/init.d/openibd restart
1847 User space I/O kernel modules (uio and igb_uio) are not used and do
1848 not have to be loaded.
1850 #. Make sure Ethernet interfaces are in working order and linked to kernel
1851 verbs. Related sysfs entries should be present::
1853 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1862 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1865 for intf in eth2 eth3 eth4 eth5;
1867 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1870 sed -n 's,.*/\(.*\),-a \1,p'
1879 #. Request huge pages::
1881 dpdk-hugepages.py --setup 2G
1883 #. Start testpmd with basic parameters::
1885 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
1890 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1891 EAL: probe driver: 15b3:1013 librte_net_mlx5
1892 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1893 PMD: librte_net_mlx5: 1 port(s) detected
1894 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1895 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1896 EAL: probe driver: 15b3:1013 librte_net_mlx5
1897 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1898 PMD: librte_net_mlx5: 1 port(s) detected
1899 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1900 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1901 EAL: probe driver: 15b3:1013 librte_net_mlx5
1902 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1903 PMD: librte_net_mlx5: 1 port(s) detected
1904 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1905 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1906 EAL: probe driver: 15b3:1013 librte_net_mlx5
1907 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1908 PMD: librte_net_mlx5: 1 port(s) detected
1909 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1910 Interactive-mode selected
1911 Configuring Port 0 (socket 0)
1912 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1913 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1914 Port 0: E4:1D:2D:E7:0C:FE
1915 Configuring Port 1 (socket 0)
1916 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1917 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1918 Port 1: E4:1D:2D:E7:0C:FF
1919 Configuring Port 2 (socket 0)
1920 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1921 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1922 Port 2: E4:1D:2D:E7:0C:FA
1923 Configuring Port 3 (socket 0)
1924 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1925 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1926 Port 3: E4:1D:2D:E7:0C:FB
1927 Checking link statuses...
1928 Port 0 Link Up - speed 40000 Mbps - full-duplex
1929 Port 1 Link Up - speed 40000 Mbps - full-duplex
1930 Port 2 Link Up - speed 10000 Mbps - full-duplex
1931 Port 3 Link Up - speed 10000 Mbps - full-duplex
1938 This section demonstrates how to dump flows. Currently, it's possible to dump
1939 all flows with assistance of external tools.
1941 #. 2 ways to get flow raw file:
1943 - Using testpmd CLI:
1945 .. code-block:: console
1948 testpmd> flow dump <port> all <output_file>
1950 testpmd> flow dump <port> rule <rule_id> <output_file>
1952 - call rte_flow_dev_dump api:
1954 .. code-block:: console
1956 rte_flow_dev_dump(port, flow, file, NULL);
1958 #. Dump human-readable flows from raw file:
1960 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1962 .. code-block:: console
1964 mlx_steering_dump.py -f <output_file> -flowptr <flow_ptr>
1966 How to share a meter between ports in the same switch domain
1967 ------------------------------------------------------------
1969 This section demonstrates how to use the shared meter. A meter M can be created
1970 on port X and to be shared with a port Y on the same switch domain by the next way:
1972 .. code-block:: console
1974 flow create X ingress transfer pattern eth / port_id id is Y / end actions meter mtr_id M / end
1976 How to use meter hierarchy
1977 --------------------------
1979 This section demonstrates how to create and use a meter hierarchy.
1980 A termination meter M can be the policy green action of another termination meter N.
1981 The two meters are chained together as a chain. Using meter N in a flow will apply
1982 both the meters in hierarchy on that flow.
1984 .. code-block:: console
1986 add port meter policy 0 1 g_actions queue index 0 / end y_actions end r_actions drop / end
1987 create port meter 0 M 1 1 yes 0xffff 1 0
1988 add port meter policy 0 2 g_actions meter mtr_id M / end y_actions end r_actions drop / end
1989 create port meter 0 N 2 2 yes 0xffff 1 0
1990 flow create 0 ingress group 1 pattern eth / end actions meter mtr_id N / end