1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright 2015 6WIND S.A.
3 Copyright 2015 Mellanox Technologies, Ltd
5 .. include:: <isonum.txt>
10 The MLX5 poll mode driver library (**librte_net_mlx5**) provides support
11 for **Mellanox ConnectX-4**, **Mellanox ConnectX-4 Lx** , **Mellanox
12 ConnectX-5**, **Mellanox ConnectX-6**, **Mellanox ConnectX-6 Dx**, **Mellanox
13 ConnectX-6 Lx**, **Mellanox BlueField** and **Mellanox BlueField-2** families
14 of 10/25/40/50/100/200 Gb/s adapters as well as their virtual functions (VF)
17 Information and documentation about these adapters can be found on the
18 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
19 `Mellanox community <http://community.mellanox.com/welcome>`__.
21 There is also a `section dedicated to this poll mode driver
22 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
28 Besides its dependency on libibverbs (that implies libmlx5 and associated
29 kernel support), librte_net_mlx5 relies heavily on system calls for control
30 operations such as querying/updating the MTU and flow control parameters.
32 For security reasons and robustness, this driver only deals with virtual
33 memory addresses. The way resources allocations are handled by the kernel,
34 combined with hardware specifications that allow to handle virtual memory
35 addresses directly, ensure that DPDK applications cannot access random
36 physical memory (or memory that does not belong to the current process).
38 This capability allows the PMD to coexist with kernel network interfaces
39 which remain functional, although they stop receiving unicast packets as
40 long as they share the same MAC address.
41 This means legacy linux control tools (for example: ethtool, ifconfig and
42 more) can operate on the same network interfaces that owned by the DPDK
45 The PMD can use libibverbs and libmlx5 to access the device firmware
46 or directly the hardware components.
47 There are different levels of objects and bypassing abilities
48 to get the best performances:
50 - Verbs is a complete high-level generic API
51 - Direct Verbs is a device-specific API
52 - DevX allows to access firmware objects
53 - Direct Rules manages flow steering at low-level hardware layer
55 Enabling librte_net_mlx5 causes DPDK applications to be linked against
61 - Multi arch support: x86_64, POWER8, ARMv8, i686.
62 - Multiple TX and RX queues.
63 - Support for scattered TX frames.
64 - Advanced support for scattered Rx frames with tunable buffer attributes.
65 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
66 - RSS using different combinations of fields: L3 only, L4 only or both,
67 and source only, destination only or both.
68 - Several RSS hash keys, one for each flow type.
69 - Default RSS operation with no hash key specification.
70 - Configurable RETA table.
71 - Link flow control (pause frame).
72 - Support for multiple MAC addresses.
76 - RX CRC stripping configuration.
77 - TX mbuf fast free offload.
78 - Promiscuous mode on PF and VF.
79 - Multicast promiscuous mode on PF and VF.
80 - Hardware checksum offloads.
81 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
83 - Flow API, including :ref:`flow_isolated_mode`.
85 - KVM and VMware ESX SR-IOV modes are supported.
86 - RSS hash result is supported.
87 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
88 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
90 - Statistics query including Basic, Extended and per queue.
92 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
93 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
94 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
95 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
96 - Flow insertion rate of more then million flows per second, when using Direct Rules.
97 - Support for multiple rte_flow groups.
98 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
101 - Multiple-thread flow insertion.
102 - Matching on GTP extension header with raw encap/decap action.
103 - Matching on Geneve TLV option header with raw encap/decap action.
104 - RSS support in sample action.
105 - E-Switch mirroring and jump.
106 - E-Switch mirroring and modify.
107 - 21844 flow priorities for ingress or egress flow groups greater than 0 and for any transfer
109 - Flow metering, including meter policy API.
110 - Flow integrity offload API.
111 - Connection tracking.
112 - Sub-Function representors.
119 On Windows, the features are limited:
121 - Promiscuous mode is not supported
122 - The following rules are supported:
124 - IPv4/UDP with CVLAN filtering
125 - Unicast MAC filtering
127 - For secondary process:
129 - Forked secondary process not supported.
130 - External memory unregistered in EAL memseg list cannot be used for DMA
131 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
132 primary process and remapped to the same virtual address in secondary
133 process. If the external memory is registered by primary process but has
134 different virtual address in secondary process, unexpected error may happen.
136 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
137 specific VLAN will match for VLAN packets as well:
139 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
140 Meaning, the flow rule::
142 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
144 Will only match vlan packets with vid=3. and the flow rule::
146 flow create 0 ingress pattern eth / ipv4 / end ...
148 Will match any ipv4 packet (VLAN included).
150 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
152 - 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.
155 flow create 0 ingress pattern eth / ipv4 / end ...
157 Will match any ipv4 packet.
160 flow create 0 ingress pattern eth / vlan / end ...
161 flow create 0 ingress pattern eth has_vlan is 1 / end ...
162 flow create 0 ingress pattern eth type is 0x8100 / end ...
164 Will match single-tagged packets only, with any VLAN ID value.
167 flow create 0 ingress pattern eth type is 0x88A8 / end ...
168 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
170 Will match multi-tagged packets only, with any VLAN ID value.
172 - A flow pattern with 2 sequential VLAN items is not supported.
174 - VLAN pop offload command:
176 - Flow rules having a VLAN pop offload command as one of their actions and
177 are lacking a match on VLAN as one of their items are not supported.
178 - The command is not supported on egress traffic in NIC mode.
180 - VLAN push offload is not supported on ingress traffic in NIC mode.
182 - VLAN set PCP offload is not supported on existing headers.
184 - A multi segment packet must have not more segments than reported by dev_infos_get()
185 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
186 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
187 inline settings) to 58.
189 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
190 to 0 are not supported.
192 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
194 - Match on Geneve header supports the following fields only:
201 - Match on Geneve TLV option is supported on the following fields:
208 Only one Class/Type/Length Geneve TLV option is supported per shared device.
209 Class/Type/Length fields must be specified as well as masks.
210 Class/Type/Length specified masks must be full.
211 Matching Geneve TLV option without specifying data is not supported.
212 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
214 - VF: flow rules created on VF devices can only match traffic targeted at the
215 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
217 - Match on GTP tunnel header item supports the following fields only:
219 - v_pt_rsv_flags: E flag, S flag, PN flag
223 - Match on GTP extension header only for GTP PDU session container (next
224 extension header type = 0x85).
225 - Match on GTP extension header is not supported in group 0.
227 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
228 The flows within group 0 and set metadata action are rejected by hardware.
232 MAC addresses not already present in the bridge table of the associated
233 kernel network device will be added and cleaned up by the PMD when closing
234 the device. In case of ungraceful program termination, some entries may
235 remain present and should be removed manually by other means.
237 - Buffer split offload is supported with regular Rx burst routine only,
238 no MPRQ feature or vectorized code can be engaged.
240 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
241 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
242 ol_flags. As the mempool for the external buffer is managed by PMD, all the
243 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
244 the external buffers will be freed by PMD and the application which still
245 holds the external buffers may be corrupted.
247 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
248 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
249 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
251 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
252 and allmulticast mode are both set to off.
253 To receive IPv6 Multicast messages on VM, explicitly set the relevant
254 MAC address using rte_eth_dev_mac_addr_add() API.
256 - To support a mixed traffic pattern (some buffers from local host memory, some
257 buffers from other devices) with high bandwidth, a mbuf flag is used.
259 An application hints the PMD whether or not it should try to inline the
260 given mbuf data buffer. PMD should do the best effort to act upon this request.
262 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
263 registered by application with rte_mbuf_dynflag_register(). This flag is
264 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
265 which is intended to be used by the application.
267 To query the supported specific flags in runtime,
268 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
269 currently (over present hardware and configuration) supported specific flags.
270 The "not inline hint" feature operating flow is the following one:
273 - probe the devices, ports are created
274 - query the port capabilities
275 - if port supporting the feature is found
276 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
277 - application starts the ports
278 - on ``dev_start()`` PMD checks whether the feature flag is registered and
279 enables the feature support in datapath
280 - application might set the registered flag bit in ``ol_flags`` field
281 of mbuf being sent and PMD will handle ones appropriately.
283 - The amount of descriptors in Tx queue may be limited by data inline settings.
284 Inline data require the more descriptor building blocks and overall block
285 amount may exceed the hardware supported limits. The application should
286 reduce the requested Tx size or adjust data inline settings with
287 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
289 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
290 parameter should be specified.
291 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
292 being sent it tries to synchronize the time of packet appearing on
293 the wire with the specified packet timestamp. It the specified one
294 is in the past it should be ignored, if one is in the distant future
295 it should be capped with some reasonable value (in range of seconds).
296 These specific cases ("too late" and "distant future") can be optionally
297 reported via device xstats to assist applications to detect the
298 time-related problems.
300 The timestamp upper "too-distant-future" limit
301 at the moment of invoking the Tx burst routine
302 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
303 Please note, for the testpmd txonly mode,
304 the limit is deduced from the expression::
306 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
308 There is no any packet reordering according timestamps is supposed,
309 neither within packet burst, nor between packets, it is an entirely
310 application responsibility to generate packets and its timestamps
311 in desired order. The timestamps can be put only in the first packet
312 in the burst providing the entire burst scheduling.
314 - E-Switch decapsulation Flow:
316 - can be applied to PF port only.
317 - must specify VF port action (packet redirection from PF to VF).
318 - optionally may specify tunnel inner source and destination MAC addresses.
320 - E-Switch encapsulation Flow:
322 - can be applied to VF ports only.
323 - must specify PF port action (packet redirection from VF to PF).
327 - The input buffer, used as outer header, is not validated.
331 - The decapsulation is always done up to the outermost tunnel detected by the HW.
332 - The input buffer, providing the removal size, is not validated.
333 - The buffer size must match the length of the headers to be removed.
335 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
336 mutually exclusive features which cannot be supported together
337 (see :ref:`mlx5_firmware_config`).
341 - Requires DevX and DV flow to be enabled.
342 - KEEP_CRC offload cannot be supported with LRO.
343 - The first mbuf length, without head-room, must be big enough to include the
345 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
346 it with size limited to max LRO size, not to max RX packet length.
347 - LRO can be used with outer header of TCP packets of the standard format:
348 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
350 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
351 - LRO packet aggregation is performed by HW only for packet size larger than
352 ``lro_min_mss_size``. This value is reported on device start, when debug
357 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
358 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
359 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
363 - fast free offload assumes the all mbufs being sent are originated from the
364 same memory pool and there is no any extra references to the mbufs (the
365 reference counter for each mbuf is equal 1 on tx_burst call). The latter
366 means there should be no any externally attached buffers in mbufs. It is
367 an application responsibility to provide the correct mbufs if the fast
368 free offload is engaged. The mlx5 PMD implicitly produces the mbufs with
369 externally attached buffers if MPRQ option is enabled, hence, the fast
370 free offload is neither supported nor advertised if there is MPRQ enabled.
374 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and
375 E-Switch steering domain.
376 - For E-Switch Sampling flow with sample ratio > 1, additional actions are not
377 supported in the sample actions list.
378 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as
379 first action in the E-Switch egress flow if with header modify or
380 encapsulation actions.
381 - For NIC Rx flow, supports ``MARK``, ``COUNT``, ``QUEUE``, ``RSS`` in the
383 - For E-Switch mirroring flow, supports ``RAW ENCAP``, ``Port ID``,
384 ``VXLAN ENCAP``, ``NVGRE ENCAP`` in the sample actions list.
388 - Supports the 'set' operation only for ``RTE_FLOW_ACTION_TYPE_MODIFY_FIELD`` action.
389 - Modification of an arbitrary place in a packet via the special ``RTE_FLOW_FIELD_START`` Field ID is not supported.
390 - Modification of the 802.1Q Tag, VXLAN Network or GENEVE Network ID's is not supported.
391 - Encapsulation levels are not supported, can modify outermost header fields only.
392 - Offsets must be 32-bits aligned, cannot skip past the boundary of a field.
394 - IPv6 header item 'proto' field, indicating the next header protocol, should
395 not be set as extension header.
396 In case the next header is an extension header, it should not be specified in
397 IPv6 header item 'proto' field.
398 The last extension header item 'next header' field can specify the following
399 header protocol type.
403 - 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.
404 - Hairpin in switchdev SR-IOV mode is not supported till now.
408 - All the meter colors with drop action will be counted only by the global drop statistics.
409 - Green color is not supported with drop action.
410 - Yellow detection is not supported.
411 - Red color must be with drop action.
412 - Meter statistics are supported only for drop case.
413 - Meter yellow color detection is not supported.
414 - A meter action created with pre-defined policy must be the last action in the flow except single case where the policy actions are:
415 - green: NULL or END.
416 - yellow: NULL or END.
418 - The only supported meter policy actions:
419 - green: QUEUE, RSS, PORT_ID, JUMP, MARK and SET_TAG.
420 - yellow: must be empty.
422 - meter profile packet mode is supported.
426 - Integrity offload is enabled for **ConnectX-6** family.
427 - Verification bits provided by the hardware are ``l3_ok``, ``ipv4_csum_ok``, ``l4_ok``, ``l4_csum_ok``.
428 - ``level`` value 0 references outer headers.
429 - Multiple integrity items not supported in a single flow rule.
430 - Flow rule items supplied by application must explicitly specify network headers referred by integrity item.
431 For example, if integrity item mask sets ``l4_ok`` or ``l4_csum_ok`` bits, reference to L4 network header,
432 TCP or UDP, must be in the rule pattern as well::
434 flow create 0 ingress pattern integrity level is 0 value mask l3_ok value spec l3_ok / eth / ipv6 / end …
436 flow create 0 ingress pattern integrity level is 0 value mask l4_ok value spec 0 / eth / ipv4 proto is udp / end …
438 - Connection tracking:
440 - Cannot co-exist with ASO meter, ASO age action in a single flow rule.
441 - Flow rules insertion rate and memory consumption need more optimization.
443 - 4M connections maximum.
448 MLX5 supports various methods to report statistics:
450 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.
452 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.
454 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.
462 The ibverbs libraries can be linked with this PMD in a number of ways,
463 configured by the ``ibverbs_link`` build option:
465 - ``shared`` (default): the PMD depends on some .so files.
467 - ``dlopen``: Split the dependencies glue in a separate library
468 loaded when needed by dlopen.
469 It make dependencies on libibverbs and libmlx4 optional,
470 and has no performance impact.
472 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
473 in the PMD shared library or the executable static binary.
475 Environment variables
476 ~~~~~~~~~~~~~~~~~~~~~
480 A list of directories in which to search for the rdma-core "glue" plug-in,
481 separated by colons or semi-colons.
483 - ``MLX5_SHUT_UP_BF``
485 Configures HW Tx doorbell register as IO-mapped.
487 By default, the HW Tx doorbell is configured as a write-combining register.
488 The register would be flushed to HW usually when the write-combining buffer
489 becomes full, but it depends on CPU design.
491 Except for vectorized Tx burst routines, a write memory barrier is enforced
492 after updating the register so that the update can be immediately visible to
495 When vectorized Tx burst is called, the barrier is set only if the burst size
496 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
497 variable will bring better latency even though the maximum throughput can
500 Run-time configuration
501 ~~~~~~~~~~~~~~~~~~~~~~
503 - librte_net_mlx5 brings kernel network interfaces up during initialization
504 because it is affected by their state. Forcing them down prevents packets
507 - **ethtool** operations on related kernel interfaces also affect the PMD.
512 In order to run as a non-root user,
513 some capabilities must be granted to the application::
515 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
517 Below are the reasons of the need for each capability:
520 When using physical addresses (PA mode), with Linux >= 4.0,
521 for access to ``/proc/self/pagemap``.
524 For device configuration.
527 For raw ethernet queue allocation through kernel driver.
530 For DMA memory pinning.
535 - ``rxq_cqe_comp_en`` parameter [int]
537 A nonzero value enables the compression of CQE on RX side. This feature
538 allows to save PCI bandwidth and improve performance. Enabled by default.
539 Different compression formats are supported in order to achieve the best
540 performance for different traffic patterns. Default format depends on
541 Multi-Packet Rx queue configuration: Hash RSS format is used in case
542 MPRQ is disabled, Checksum format is used in case MPRQ is enabled.
544 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
545 better compression rate in case of RTE Flow Mark traffic.
546 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
547 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
548 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
549 CQE compression format selection requires DevX to be enabled. If there is
550 no DevX enabled/supported the value is reset to 1 by default.
554 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
555 ConnectX-6 Lx, BlueField and BlueField-2.
556 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
557 ConnectX-6 Lx, BlueField and BlueField-2.
559 - ``rxq_pkt_pad_en`` parameter [int]
561 A nonzero value enables padding Rx packet to the size of cacheline on PCI
562 transaction. This feature would waste PCI bandwidth but could improve
563 performance by avoiding partial cacheline write which may cause costly
564 read-modify-copy in memory transaction on some architectures. Disabled by
569 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
570 ConnectX-6 Lx, BlueField and BlueField-2.
571 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
572 ConnectX-6 Lx, BlueField and BlueField-2.
574 - ``mprq_en`` parameter [int]
576 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
577 configured as Multi-Packet RQ if the total number of Rx queues is
578 ``rxqs_min_mprq`` or more. Disabled by default.
580 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
581 by posting a single large buffer for multiple packets. Instead of posting a
582 buffers per a packet, one large buffer is posted in order to receive multiple
583 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
584 and each stride receives one packet. MPRQ can improve throughput for
585 small-packet traffic.
587 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
588 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
589 configure large stride size enough to accommodate max_rx_pkt_len as long as
590 device allows. Note that this can waste system memory compared to enabling Rx
591 scatter and multi-segment packet.
593 - ``mprq_log_stride_num`` parameter [int]
595 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
596 strides can reduce PCIe traffic further. If configured value is not in the
597 range of device capability, the default value will be set with a warning
598 message. The default value is 4 which is 16 strides per a buffer, valid only
599 if ``mprq_en`` is set.
601 The size of Rx queue should be bigger than the number of strides.
603 - ``mprq_log_stride_size`` parameter [int]
605 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
606 stride size can save some memory and reduce probability of a depletion of all
607 available strides due to unreleased packets by an application. If configured
608 value is not in the range of device capability, the default value will be set
609 with a warning message. The default value is 11 which is 2048 bytes per a
610 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
611 it is possible for a packet to span across multiple strides. This mode allows
612 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
613 of a packet if Rx scatter is configured) may be required in case there is no
614 space left for a head room at the end of a stride which incurs some
617 - ``mprq_max_memcpy_len`` parameter [int]
619 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
620 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
621 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
622 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
623 A mempool for external buffers will be allocated and managed by PMD. If Rx
624 packet is externally attached, ol_flags field of the mbuf will have
625 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
626 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
628 - ``rxqs_min_mprq`` parameter [int]
630 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
631 greater or equal to this value. The default value is 12, valid only if
634 - ``txq_inline`` parameter [int]
636 Amount of data to be inlined during TX operations. This parameter is
637 deprecated and converted to the new parameter ``txq_inline_max`` providing
638 partial compatibility.
640 - ``txqs_min_inline`` parameter [int]
642 Enable inline data send only when the number of TX queues is greater or equal
645 This option should be used in combination with ``txq_inline_max`` and
646 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
648 If this option is not specified the default value 16 is used for BlueField
649 and 8 for other platforms
651 The data inlining consumes the CPU cycles, so this option is intended to
652 auto enable inline data if we have enough Tx queues, which means we have
653 enough CPU cores and PCI bandwidth is getting more critical and CPU
654 is not supposed to be bottleneck anymore.
656 The copying data into WQE improves latency and can improve PPS performance
657 when PCI back pressure is detected and may be useful for scenarios involving
658 heavy traffic on many queues.
660 Because additional software logic is necessary to handle this mode, this
661 option should be used with care, as it may lower performance when back
662 pressure is not expected.
664 If inline data are enabled it may affect the maximal size of Tx queue in
665 descriptors because the inline data increase the descriptor size and
666 queue size limits supported by hardware may be exceeded.
668 - ``txq_inline_min`` parameter [int]
670 Minimal amount of data to be inlined into WQE during Tx operations. NICs
671 may require this minimal data amount to operate correctly. The exact value
672 may depend on NIC operation mode, requested offloads, etc. It is strongly
673 recommended to omit this parameter and use the default values. Anyway,
674 applications using this parameter should take into consideration that
675 specifying an inconsistent value may prevent the NIC from sending packets.
677 If ``txq_inline_min`` key is present the specified value (may be aligned
678 by the driver in order not to exceed the limits and provide better descriptor
679 space utilization) will be used by the driver and it is guaranteed that
680 requested amount of data bytes are inlined into the WQE beside other inline
681 settings. This key also may update ``txq_inline_max`` value (default
682 or specified explicitly in devargs) to reserve the space for inline data.
684 If ``txq_inline_min`` key is not present, the value may be queried by the
685 driver from the NIC via DevX if this feature is available. If there is no DevX
686 enabled/supported the value 18 (supposing L2 header including VLAN) is set
687 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
688 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
691 For ConnectX-4 NIC, driver does not allow specifying value below 18
692 (minimal L2 header, including VLAN), error will be raised.
694 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
695 it is not recommended and may prevent NIC from sending packets over
698 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
699 Multi-Packet Write), because last one does not support partial packet inlining.
700 This is not very critical due to minimal data inlining is mostly required
701 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
703 - ``txq_inline_max`` parameter [int]
705 Specifies the maximal packet length to be completely inlined into WQE
706 Ethernet Segment for ordinary SEND method. If packet is larger than specified
707 value, the packet data won't be copied by the driver at all, data buffer
708 is addressed with a pointer. If packet length is less or equal all packet
709 data will be copied into WQE. This may improve PCI bandwidth utilization for
710 short packets significantly but requires the extra CPU cycles.
712 The data inline feature is controlled by number of Tx queues, if number of Tx
713 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
714 is engaged, if there are not enough Tx queues (which means not enough CPU cores
715 and CPU resources are scarce), data inline is not performed by the driver.
716 Assigning ``txqs_min_inline`` with zero always enables the data inline.
718 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
719 by the driver in order not to exceed the limit (930 bytes) and to provide better
720 WQE space filling without gaps, the adjustment is reflected in the debug log.
721 Also, the default value (290) may be decreased in run-time if the large transmit
722 queue size is requested and hardware does not support enough descriptor
723 amount, in this case warning is emitted. If ``txq_inline_max`` key is
724 specified and requested inline settings can not be satisfied then error
727 - ``txq_inline_mpw`` parameter [int]
729 Specifies the maximal packet length to be completely inlined into WQE for
730 Enhanced MPW method. If packet is large the specified value, the packet data
731 won't be copied, and data buffer is addressed with pointer. If packet length
732 is less or equal, all packet data will be copied into WQE. This may improve PCI
733 bandwidth utilization for short packets significantly but requires the extra
736 The data inline feature is controlled by number of TX queues, if number of Tx
737 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
738 is engaged, if there are not enough Tx queues (which means not enough CPU cores
739 and CPU resources are scarce), data inline is not performed by the driver.
740 Assigning ``txqs_min_inline`` with zero always enables the data inline.
742 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
743 by the driver in order not to exceed the limit (930 bytes) and to provide better
744 WQE space filling without gaps, the adjustment is reflected in the debug log.
745 Due to multiple packets may be included to the same WQE with Enhanced Multi
746 Packet Write Method and overall WQE size is limited it is not recommended to
747 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
748 may be decreased in run-time if the large transmit queue size is requested
749 and hardware does not support enough descriptor amount, in this case warning
750 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
751 settings can not be satisfied then error will be raised.
753 - ``txqs_max_vec`` parameter [int]
755 Enable vectorized Tx only when the number of TX queues is less than or
756 equal to this value. This parameter is deprecated and ignored, kept
757 for compatibility issue to not prevent driver from probing.
759 - ``txq_mpw_hdr_dseg_en`` parameter [int]
761 A nonzero value enables including two pointers in the first block of TX
762 descriptor. The parameter is deprecated and ignored, kept for compatibility
765 - ``txq_max_inline_len`` parameter [int]
767 Maximum size of packet to be inlined. This limits the size of packet to
768 be inlined. If the size of a packet is larger than configured value, the
769 packet isn't inlined even though there's enough space remained in the
770 descriptor. Instead, the packet is included with pointer. This parameter
771 is deprecated and converted directly to ``txq_inline_mpw`` providing full
772 compatibility. Valid only if eMPW feature is engaged.
774 - ``txq_mpw_en`` parameter [int]
776 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
777 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
778 eMPW allows the Tx burst function to pack up multiple packets
779 in a single descriptor session in order to save PCI bandwidth
780 and improve performance at the cost of a slightly higher CPU usage.
781 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
782 Tx burst function copies entire packet data on to Tx descriptor
783 instead of including pointer of packet.
785 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
786 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
787 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
788 option or reported by the NIC, the eMPW feature is disengaged.
790 - ``tx_db_nc`` parameter [int]
792 The rdma core library can map doorbell register in two ways, depending on the
793 environment variable "MLX5_SHUT_UP_BF":
795 - As regular cached memory (usually with write combining attribute), if the
796 variable is either missing or set to zero.
797 - As non-cached memory, if the variable is present and set to not "0" value.
799 The type of mapping may slightly affect the Tx performance, the optimal choice
800 is strongly relied on the host architecture and should be deduced practically.
802 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
803 memory (with write combining), the PMD will perform the extra write memory barrier
804 after writing to doorbell, it might increase the needed CPU clocks per packet
805 to send, but latency might be improved.
807 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
808 cached memory, the PMD will not perform the extra write memory barrier
809 after writing to doorbell, on some architectures it might improve the
812 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
813 memory, the PMD will use heuristics to decide whether write memory barrier
814 should be performed. For bursts with size multiple of recommended one (64 pkts)
815 it is supposed the next burst is coming and no need to issue the extra memory
816 barrier (it is supposed to be issued in the next coming burst, at least after
817 descriptor writing). It might increase latency (on some hosts till next
818 packets transmit) and should be used with care.
820 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
821 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
822 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
824 - ``tx_pp`` parameter [int]
826 If a nonzero value is specified the driver creates all necessary internal
827 objects to provide accurate packet send scheduling on mbuf timestamps.
828 The positive value specifies the scheduling granularity in nanoseconds,
829 the packet send will be accurate up to specified digits. The allowed range is
830 from 500 to 1 million of nanoseconds. The negative value specifies the module
831 of granularity and engages the special test mode the check the schedule rate.
832 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
835 - ``tx_skew`` parameter [int]
837 The parameter adjusts the send packet scheduling on timestamps and represents
838 the average delay between beginning of the transmitting descriptor processing
839 by the hardware and appearance of actual packet data on the wire. The value
840 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
841 specified. The default value is zero.
843 - ``tx_vec_en`` parameter [int]
845 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
846 ConnectX-6 Lx, BlueField and BlueField-2 NICs
847 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
848 The parameter is deprecated and ignored.
850 - ``rx_vec_en`` parameter [int]
852 A nonzero value enables Rx vector if the port is not configured in
853 multi-segment otherwise this parameter is ignored.
857 - ``vf_nl_en`` parameter [int]
859 A nonzero value enables Netlink requests from the VF to add/remove MAC
860 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
861 Otherwise the relevant configuration must be run with Linux iproute2 tools.
862 This is a prerequisite to receive this kind of traffic.
864 Enabled by default, valid only on VF devices ignored otherwise.
866 - ``l3_vxlan_en`` parameter [int]
868 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
869 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
870 parameter. This is a prerequisite to receive this kind of traffic.
874 - ``dv_xmeta_en`` parameter [int]
876 A nonzero value enables extensive flow metadata support if device is
877 capable and driver supports it. This can enable extensive support of
878 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
879 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
881 There are some possible configurations, depending on parameter value:
883 - 0, this is default value, defines the legacy mode, the ``MARK`` and
884 ``META`` related actions and items operate only within NIC Tx and
885 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
886 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
887 item is 32 bits wide and match supported on egress only.
889 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
890 related actions and items operate within all supported steering domains,
891 including FDB, ``MARK`` and ``META`` information may cross the domain
892 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
893 depends on kernel and firmware configurations and might be 0, 16 or
894 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
895 compatibility, the actual width of data transferred to the FDB domain
896 depends on kernel configuration and may be vary. The actual supported
897 width can be retrieved in runtime by series of rte_flow_validate()
900 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
901 related actions and items operate within all supported steering domains,
902 including FDB, ``MARK`` and ``META`` information may cross the domain
903 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
904 depends on kernel and firmware configurations and might be 0, 16 or
905 24 bits. The actual supported width can be retrieved in runtime by
906 series of rte_flow_validate() trials.
908 - 3, this engages tunnel offload mode. In E-Switch configuration, that
909 mode implicitly activates ``dv_xmeta_en=1``.
911 +------+-----------+-----------+-------------+-------------+
912 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
913 +======+===========+===========+=============+=============+
914 | 0 | 24 bits | 32 bits | 32 bits | no |
915 +------+-----------+-----------+-------------+-------------+
916 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
917 +------+-----------+-----------+-------------+-------------+
918 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
919 +------+-----------+-----------+-------------+-------------+
921 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
922 ignored and the device is configured to operate in legacy mode (0).
924 Disabled by default (set to 0).
926 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
927 of the extensive metadata features. The legacy Verbs supports FLAG and
928 MARK metadata actions over NIC Rx steering domain only.
930 Setting META value to zero in flow action means there is no item provided
931 and receiving datapath will not report in mbufs the metadata are present.
932 Setting MARK value to zero in flow action means the zero FDIR ID value
933 will be reported on packet receiving.
935 For the MARK action the last 16 values in the full range are reserved for
936 internal PMD purposes (to emulate FLAG action). The valid range for the
937 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
938 for the 24-bit mode, the flows with the MARK action value outside
939 the specified range will be rejected.
941 - ``dv_flow_en`` parameter [int]
943 A nonzero value enables the DV flow steering assuming it is supported
944 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
946 Enabled by default if supported.
948 - ``dv_esw_en`` parameter [int]
950 A nonzero value enables E-Switch using Direct Rules.
952 Enabled by default if supported.
954 - ``lacp_by_user`` parameter [int]
956 A nonzero value enables the control of LACP traffic by the user application.
957 When a bond exists in the driver, by default it should be managed by the
958 kernel and therefore LACP traffic should be steered to the kernel.
959 If this devarg is set to 1 it will allow the user to manage the bond by
960 itself and not steer LACP traffic to the kernel.
962 Disabled by default (set to 0).
964 - ``mr_ext_memseg_en`` parameter [int]
966 A nonzero value enables extending memseg when registering DMA memory. If
967 enabled, the number of entries in MR (Memory Region) lookup table on datapath
968 is minimized and it benefits performance. On the other hand, it worsens memory
969 utilization because registered memory is pinned by kernel driver. Even if a
970 page in the extended chunk is freed, that doesn't become reusable until the
971 entire memory is freed.
975 - ``representor`` parameter [list]
977 This parameter can be used to instantiate DPDK Ethernet devices from
978 existing port (PF, VF or SF) representors configured on the device.
980 It is a standard parameter whose format is described in
981 :ref:`ethernet_device_standard_device_arguments`.
983 For instance, to probe VF port representors 0 through 2::
985 <PCI_BDF>,representor=vf[0-2]
987 To probe SF port representors 0 through 2::
989 <PCI_BDF>,representor=sf[0-2]
991 To probe VF port representors 0 through 2 on both PFs of bonding device::
993 <Primary_PCI_BDF>,representor=pf[0,1]vf[0-2]
995 - ``max_dump_files_num`` parameter [int]
997 The maximum number of files per PMD entity that may be created for debug information.
998 The files will be created in /var/log directory or in current directory.
1000 set to 128 by default.
1002 - ``lro_timeout_usec`` parameter [int]
1004 The maximum allowed duration of an LRO session, in micro-seconds.
1005 PMD will set the nearest value supported by HW, which is not bigger than
1006 the input ``lro_timeout_usec`` value.
1007 If this parameter is not specified, by default PMD will set
1008 the smallest value supported by HW.
1010 - ``hp_buf_log_sz`` parameter [int]
1012 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
1013 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
1014 The capacity of the value is specified by the firmware and the initialization
1015 will get a failure if it is out of scope.
1016 The range of the value is from 11 to 19 right now, and the supported frame
1017 size of a single packet for hairpin is from 512B to 128KB. It might change if
1018 different firmware release is being used. By using a small value, it could
1019 reduce memory consumption but not work with a large frame. If the value is
1020 too large, the memory consumption will be high and some potential performance
1021 degradation will be introduced.
1022 By default, the PMD will set this value to 16, which means that 9KB jumbo
1023 frames will be supported.
1025 - ``reclaim_mem_mode`` parameter [int]
1027 Cache some resources in flow destroy will help flow recreation more efficient.
1028 While some systems may require the all the resources can be reclaimed after
1030 The parameter ``reclaim_mem_mode`` provides the option for user to configure
1031 if the resource cache is needed or not.
1033 There are three options to choose:
1035 - 0. It means the flow resources will be cached as usual. The resources will
1036 be cached, helpful with flow insertion rate.
1038 - 1. It will only enable the DPDK PMD level resources reclaim.
1040 - 2. Both DPDK PMD level and rdma-core low level will be configured as
1043 By default, the PMD will set this value to 0.
1045 - ``sys_mem_en`` parameter [int]
1047 A non-zero value enables the PMD memory management allocating memory
1048 from system by default, without explicit rte memory flag.
1050 By default, the PMD will set this value to 0.
1052 - ``decap_en`` parameter [int]
1054 Some devices do not support FCS (frame checksum) scattering for
1055 tunnel-decapsulated packets.
1056 If set to 0, this option forces the FCS feature and rejects tunnel
1057 decapsulation in the flow engine for such devices.
1059 By default, the PMD will set this value to 1.
1061 .. _mlx5_firmware_config:
1063 Firmware configuration
1064 ~~~~~~~~~~~~~~~~~~~~~~
1066 Firmware features can be configured as key/value pairs.
1068 The command to set a value is::
1070 mlxconfig -d <device> set <key>=<value>
1072 The command to query a value is::
1074 mlxconfig -d <device> query | grep <key>
1076 The device name for the command ``mlxconfig`` can be either the PCI address,
1077 or the mst device name found with::
1081 Below are some firmware configurations listed.
1087 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1093 - maximum number of SR-IOV virtual functions::
1097 - enable DevX (required by Direct Rules and other features)::
1101 - aggressive CQE zipping::
1105 - L3 VXLAN and VXLAN-GPE destination UDP port::
1108 IP_OVER_VXLAN_PORT=<udp dport>
1110 - enable VXLAN-GPE tunnel flow matching::
1112 FLEX_PARSER_PROFILE_ENABLE=0
1114 FLEX_PARSER_PROFILE_ENABLE=2
1116 - enable IP-in-IP tunnel flow matching::
1118 FLEX_PARSER_PROFILE_ENABLE=0
1120 - enable MPLS flow matching::
1122 FLEX_PARSER_PROFILE_ENABLE=1
1124 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1126 FLEX_PARSER_PROFILE_ENABLE=2
1128 - enable Geneve flow matching::
1130 FLEX_PARSER_PROFILE_ENABLE=0
1132 FLEX_PARSER_PROFILE_ENABLE=1
1134 - enable Geneve TLV option flow matching::
1136 FLEX_PARSER_PROFILE_ENABLE=0
1138 - enable GTP flow matching::
1140 FLEX_PARSER_PROFILE_ENABLE=3
1142 - enable eCPRI flow matching::
1144 FLEX_PARSER_PROFILE_ENABLE=4
1150 This driver relies on external libraries and kernel drivers for resources
1151 allocations and initialization. The following dependencies are not part of
1152 DPDK and must be installed separately:
1156 User space Verbs framework used by librte_net_mlx5. This library provides
1157 a generic interface between the kernel and low-level user space drivers
1160 It allows slow and privileged operations (context initialization, hardware
1161 resources allocations) to be managed by the kernel and fast operations to
1162 never leave user space.
1166 Low-level user space driver library for Mellanox
1167 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1170 This library basically implements send/receive calls to the hardware
1173 - **Kernel modules**
1175 They provide the kernel-side Verbs API and low level device drivers that
1176 manage actual hardware initialization and resources sharing with user
1179 Unlike most other PMDs, these modules must remain loaded and bound to
1182 - mlx5_core: hardware driver managing Mellanox
1183 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1185 - mlx5_ib: InifiniBand device driver.
1186 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1188 - **Firmware update**
1190 Mellanox OFED/EN releases include firmware updates for
1191 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1193 Because each release provides new features, these updates must be applied to
1194 match the kernel modules and libraries they come with.
1198 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1204 Either RDMA Core library with a recent enough Linux kernel release
1205 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1208 RDMA Core with Linux Kernel
1209 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1211 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1212 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1213 (see `RDMA Core installation documentation`_)
1214 - When building for i686 use:
1216 - rdma-core version 18.0 or above built with 32bit support.
1217 - Kernel version 4.14.41 or above.
1219 - Starting with rdma-core v21, static libraries can be built::
1222 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1225 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1226 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1232 - Mellanox OFED version: **4.5** and above /
1233 Mellanox EN version: **4.5** and above
1236 - ConnectX-4: **12.21.1000** and above.
1237 - ConnectX-4 Lx: **14.21.1000** and above.
1238 - ConnectX-5: **16.21.1000** and above.
1239 - ConnectX-5 Ex: **16.21.1000** and above.
1240 - ConnectX-6: **20.27.0090** and above.
1241 - ConnectX-6 Dx: **22.27.0090** and above.
1242 - BlueField: **18.25.1010** and above.
1244 While these libraries and kernel modules are available on OpenFabrics
1245 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1246 managers on most distributions, this PMD requires Ethernet extensions that
1247 may not be supported at the moment (this is a work in progress).
1250 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1252 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1253 include the necessary support and should be used in the meantime. For DPDK,
1254 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1255 required from that distribution.
1259 Several versions of Mellanox OFED/EN are available. Installing the version
1260 this DPDK release was developed and tested against is strongly
1261 recommended. Please check the `linux prerequisites`_.
1263 Windows Prerequisites
1264 ---------------------
1266 This driver relies on external libraries and kernel drivers for resources
1267 allocations and initialization. The dependencies in the following sub-sections
1268 are not part of DPDK, and must be installed separately.
1270 Compilation Prerequisites
1271 ~~~~~~~~~~~~~~~~~~~~~~~~~
1273 DevX SDK installation
1274 ^^^^^^^^^^^^^^^^^^^^^
1276 The DevX SDK must be installed on the machine building the Windows PMD.
1277 Additional information can be found at
1278 `How to Integrate Windows DevX in Your Development Environment
1279 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1281 Runtime Prerequisites
1282 ~~~~~~~~~~~~~~~~~~~~~
1284 WinOF2 version 2.60 or higher must be installed on the machine.
1289 The driver can be downloaded from the following site:
1291 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1296 DevX for Windows must be enabled in the Windows registry.
1297 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1298 Additional information can be found in the WinOF2 user manual.
1303 The following Mellanox device families are supported by the same mlx5 driver:
1315 Below are detailed device names:
1317 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1318 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1319 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1320 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1321 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1322 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1323 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1324 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1325 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1326 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1327 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1328 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1329 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1330 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1331 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1332 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1333 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1334 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1335 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1336 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1337 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1338 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1339 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1340 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1341 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1342 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1344 Quick Start Guide on OFED/EN
1345 ----------------------------
1347 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1350 2. Install the required libraries and kernel modules either by installing
1351 only the required set, or by installing the entire Mellanox OFED/EN::
1353 ./mlnxofedinstall --upstream-libs --dpdk
1355 3. Verify the firmware is the correct one::
1359 4. Verify all ports links are set to Ethernet::
1361 mlxconfig -d <mst device> query | grep LINK_TYPE
1365 Link types may have to be configured to Ethernet::
1367 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1369 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1371 For hypervisors, verify SR-IOV is enabled on the NIC::
1373 mlxconfig -d <mst device> query | grep SRIOV_EN
1376 If needed, configure SR-IOV::
1378 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1379 mlxfwreset -d <mst device> reset
1381 5. Restart the driver::
1383 /etc/init.d/openibd restart
1387 service openibd restart
1389 If link type was changed, firmware must be reset as well::
1391 mlxfwreset -d <mst device> reset
1393 For hypervisors, after reset write the sysfs number of virtual functions
1396 To dynamically instantiate a given number of virtual functions (VFs)::
1398 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1400 6. Install DPDK and you are ready to go.
1401 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1403 Enable switchdev mode
1404 ---------------------
1406 Switchdev mode is a mode in E-Switch, that binds between representor and VF or SF.
1407 Representor is a port in DPDK that is connected to a VF or SF in such a way
1408 that assuming there are no offload flows, each packet that is sent from the VF or SF
1409 will be received by the corresponding representor. While each packet that is or SF
1410 sent to a representor will be received by the VF or SF.
1411 This is very useful in case of SRIOV mode, where the first packet that is sent
1412 by the VF or SF will be received by the DPDK application which will decide if this
1413 flow should be offloaded to the E-Switch. After offloading the flow packet
1414 that the VF or SF that are matching the flow will not be received any more by
1415 the DPDK application.
1417 1. Enable SRIOV mode::
1419 mlxconfig -d <mst device> set SRIOV_EN=true
1421 2. Configure the max number of VFs::
1423 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1427 mlxfwreset -d <mst device> reset
1429 3. Configure the actual number of VFs::
1431 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1433 4. Unbind the device (can be rebind after the switchdev mode)::
1435 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1437 5. Enbale switchdev mode::
1439 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1441 Sub-Function representor
1442 ------------------------
1444 Sub-Function is a portion of the PCI device, a SF netdev has its own
1445 dedicated queues(txq, rxq). A SF netdev supports E-Switch representation
1446 offload similar to existing PF and VF representors. A SF shares PCI
1447 level resources with other SFs and/or with its parent PCI function.
1449 1. Configure SF feature::
1451 mlxconfig -d <mst device> set PF_BAR2_SIZE=<0/1/2/3> PF_BAR2_ENABLE=1
1453 Value of PF_BAR2_SIZE:
1462 mlxfwreset -d <mst device> reset
1464 3. Enable switchdev mode::
1466 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1470 mlnx-sf -d <PCI_BDF> -a create
1472 5. Probe SF representor::
1474 testpmd> port attach <PCI_BDF>,representor=sf0,dv_flow_en=1
1479 1. Configure aggressive CQE Zipping for maximum performance::
1481 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1483 To set it back to the default CQE Zipping mode use::
1485 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1487 2. In case of virtualization:
1489 - Make sure that hypervisor kernel is 3.16 or newer.
1490 - Configure boot with ``iommu=pt``.
1491 - Use 1G huge pages.
1492 - Make sure to allocate a VM on huge pages.
1493 - Make sure to set CPU pinning.
1495 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1496 for better performance. For VMs, verify that the right CPU
1497 and NUMA node are pinned according to the above. Run::
1499 lstopo-no-graphics --merge
1501 to identify the NUMA node to which the PCIe adapter is connected.
1503 4. If more than one adapter is used, and root complex capabilities allow
1504 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1505 it is recommended to locate both adapters on the same NUMA node.
1506 This in order to forward packets from one to the other without
1507 NUMA performance penalty.
1509 5. Disable pause frames::
1511 ethtool -A <netdev> rx off tx off
1513 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1514 via the BIOS configuration. Please contact you server provider for more
1515 information about the settings.
1519 On some machines, depends on the machine integrator, it is beneficial
1520 to set the PCI max read request parameter to 1K. This can be
1521 done in the following way:
1523 To query the read request size use::
1525 setpci -s <NIC PCI address> 68.w
1527 If the output is different than 3XXX, set it by::
1529 setpci -s <NIC PCI address> 68.w=3XXX
1531 The XXX can be different on different systems. Make sure to configure
1532 according to the setpci output.
1534 7. To minimize overhead of searching Memory Regions:
1536 - '--socket-mem' is recommended to pin memory by predictable amount.
1537 - Configure per-lcore cache when creating Mempools for packet buffer.
1538 - Refrain from dynamically allocating/freeing memory in run-time.
1543 There are multiple Rx burst functions with different advantages and limitations.
1545 .. table:: Rx burst functions
1547 +-------------------+------------------------+---------+-----------------+------+-------+
1548 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1549 | | | | || comp|| MTU |
1550 +===================+========================+=========+=================+======+=======+
1551 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1552 +-------------------+------------------------+---------+-----------------+------+-------+
1553 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1554 +-------------------+------------------------+---------+-----------------+------+-------+
1555 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1556 | || RxQs >= rxqs_min_mprq | | | | |
1557 +-------------------+------------------------+---------+-----------------+------+-------+
1558 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1559 | || mprq_en=1 | | | | |
1560 | || RxQs >= rxqs_min_mprq | | | | |
1561 +-------------------+------------------------+---------+-----------------+------+-------+
1563 .. _mlx5_offloads_support:
1565 Supported hardware offloads
1566 ---------------------------
1568 .. table:: Minimal SW/HW versions for queue offloads
1570 ============== ===== ===== ========= ===== ========== =============
1571 Offload DPDK Linux rdma-core OFED firmware hardware
1572 ============== ===== ===== ========= ===== ========== =============
1573 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1574 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1575 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1576 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1577 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1578 Tx scheduling 20.08 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1579 Buffer Split 20.11 N/A N/A 5.1-2 16.28.2006 ConnectX-5
1580 ============== ===== ===== ========= ===== ========== =============
1582 .. table:: Minimal SW/HW versions for rte_flow offloads
1584 +-----------------------+-----------------+-----------------+
1585 | Offload | with E-Switch | with NIC |
1586 +=======================+=================+=================+
1587 | Count | | DPDK 19.05 | | DPDK 19.02 |
1588 | | | OFED 4.6 | | OFED 4.6 |
1589 | | | rdma-core 24 | | rdma-core 23 |
1590 | | | ConnectX-5 | | ConnectX-5 |
1591 +-----------------------+-----------------+-----------------+
1592 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1593 | | | OFED 4.6 | | OFED 4.5 |
1594 | | | rdma-core 24 | | rdma-core 23 |
1595 | | | ConnectX-5 | | ConnectX-4 |
1596 +-----------------------+-----------------+-----------------+
1597 | Queue / RSS | | | | DPDK 18.11 |
1598 | | | N/A | | OFED 4.5 |
1599 | | | | | rdma-core 23 |
1600 | | | | | ConnectX-4 |
1601 +-----------------------+-----------------+-----------------+
1602 | Shared action | | | | |
1603 | | | :numref:`sact`| | :numref:`sact`|
1606 +-----------------------+-----------------+-----------------+
1607 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1608 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1609 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1610 | | of_set_vlan_pcp / | | | | |
1611 | | of_set_vlan_vid) | | | | |
1612 +-----------------------+-----------------+-----------------+
1613 | | VLAN | | DPDK 21.05 | | |
1614 | | ingress and / | | OFED 5.3 | | N/A |
1615 | | of_push_vlan / | | ConnectX-6 Dx | | |
1616 +-----------------------+-----------------+-----------------+
1617 | | VLAN | | DPDK 21.05 | | |
1618 | | egress and / | | OFED 5.3 | | N/A |
1619 | | of_pop_vlan / | | ConnectX-6 Dx | | |
1620 +-----------------------+-----------------+-----------------+
1621 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1622 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1623 | | | rdma-core 24 | | rdma-core 23 |
1624 | | | ConnectX-5 | | ConnectX-5 |
1625 +-----------------------+-----------------+-----------------+
1626 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1627 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1628 | | | rdma-core 27 | | rdma-core 27 |
1629 | | | ConnectX-5 | | ConnectX-5 |
1630 +-----------------------+-----------------+-----------------+
1631 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1632 | | | OFED 5.1-2 | | OFED 5.1-2 |
1633 | | | rdma-core 32 | | N/A |
1634 | | | ConnectX-5 | | ConnectX-5 |
1635 +-----------------------+-----------------+-----------------+
1636 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1637 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1638 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1639 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1640 | | set_ipv6_dst / | | | | |
1641 | | set_tp_src / | | | | |
1642 | | set_tp_dst / | | | | |
1643 | | dec_ttl / | | | | |
1644 | | set_ttl / | | | | |
1645 | | set_mac_src / | | | | |
1646 | | set_mac_dst) | | | | |
1647 +-----------------------+-----------------+-----------------+
1648 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1649 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1650 | | | | rdma-core 24 | | rdma-core 24 |
1651 | | | | ConnectX-5 | | ConnectX-5 |
1652 +-----------------------+-----------------+-----------------+
1653 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1654 | | | OFED 4.7-1 | | OFED 4.7-1 |
1655 | | | rdma-core 24 | | N/A |
1656 | | | ConnectX-5 | | ConnectX-5 |
1657 +-----------------------+-----------------+-----------------+
1658 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1659 | | | OFED 4.6 | | OFED 4.5 |
1660 | | | rdma-core 24 | | rdma-core 23 |
1661 | | | ConnectX-5 | | ConnectX-4 |
1662 +-----------------------+-----------------+-----------------+
1663 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1664 | | | OFED 4.7-3 | | OFED 4.7-3 |
1665 | | | rdma-core 26 | | rdma-core 26 |
1666 | | | ConnectX-5 | | ConnectX-5 |
1667 +-----------------------+-----------------+-----------------+
1668 | Port ID | | DPDK 19.05 | | N/A |
1669 | | | OFED 4.7-1 | | N/A |
1670 | | | rdma-core 24 | | N/A |
1671 | | | ConnectX-5 | | N/A |
1672 +-----------------------+-----------------+-----------------+
1673 | Hairpin | | | | DPDK 19.11 |
1674 | | | N/A | | OFED 4.7-3 |
1675 | | | | | rdma-core 26 |
1676 | | | | | ConnectX-5 |
1677 +-----------------------+-----------------+-----------------+
1678 | 2-port Hairpin | | | | DPDK 20.11 |
1679 | | | N/A | | OFED 5.1-2 |
1681 | | | | | ConnectX-5 |
1682 +-----------------------+-----------------+-----------------+
1683 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1684 | | | OFED 4.7-3 | | OFED 4.7-3 |
1685 | | | rdma-core 26 | | rdma-core 26 |
1686 | | | ConnectX-5 | | ConnectX-5 |
1687 +-----------------------+-----------------+-----------------+
1688 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1689 | | | OFED 5.1-2 | | OFED 5.1-2 |
1690 | | | rdma-core 32 | | N/A |
1691 | | | ConnectX-5 | | ConnectX-5 |
1692 +-----------------------+-----------------+-----------------+
1693 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1694 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1695 | | | rdma-core 35 | | rdma-core 35 |
1696 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1697 +-----------------------+-----------------+-----------------+
1698 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1699 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1700 | | | rdma-core 34 | | rdma-core 34 |
1701 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1702 +-----------------------+-----------------+-----------------+
1703 | Modify Field | | DPDK 21.02 | | DPDK 21.02 |
1704 | | | OFED 5.2 | | OFED 5.2 |
1705 | | | rdma-core 35 | | rdma-core 35 |
1706 | | | ConnectX-5 | | ConnectX-5 |
1707 +-----------------------+-----------------+-----------------+
1708 | Connection tracking | | | | DPDK 21.05 |
1709 | | | N/A | | OFED 5.3 |
1710 | | | | | rdma-core 35 |
1711 | | | | | ConnectX-6 Dx |
1712 +-----------------------+-----------------+-----------------+
1714 .. table:: Minimal SW/HW versions for shared action offload
1717 +-----------------------+-----------------+-----------------+
1718 | Shared Action | with E-Switch | with NIC |
1719 +=======================+=================+=================+
1720 | RSS | | | | DPDK 20.11 |
1721 | | | N/A | | OFED 5.2 |
1722 | | | | | rdma-core 33 |
1723 | | | | | ConnectX-5 |
1724 +-----------------------+-----------------+-----------------+
1725 | Age | | DPDK 20.11 | | DPDK 20.11 |
1726 | | | OFED 5.2 | | OFED 5.2 |
1727 | | | rdma-core 32 | | rdma-core 32 |
1728 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1729 +-----------------------+-----------------+-----------------+
1730 | Count | | DPDK 21.05 | | DPDK 21.05 |
1731 | | | OFED 4.6 | | OFED 4.6 |
1732 | | | rdma-core 24 | | rdma-core 23 |
1733 | | | ConnectX-5 | | ConnectX-5 |
1734 +-----------------------+-----------------+-----------------+
1739 MARK and META items are interrelated with datapath - they might move from/to
1740 the applications in mbuf fields. Hence, zero value for these items has the
1741 special meaning - it means "no metadata are provided", not zero values are
1742 treated by applications and PMD as valid ones.
1744 Moreover in the flow engine domain the value zero is acceptable to match and
1745 set, and we should allow to specify zero values as rte_flow parameters for the
1746 META and MARK items and actions. In the same time zero mask has no meaning and
1747 should be rejected on validation stage.
1752 Flows are not cached in the driver.
1753 When stopping a device port, all the flows created on this port from the
1754 application will be flushed automatically in the background.
1755 After stopping the device port, all flows on this port become invalid and
1756 not represented in the system.
1757 All references to these flows held by the application should be discarded
1758 directly but neither destroyed nor flushed.
1760 The application should re-create the flows as required after the port restart.
1765 Compared to librte_net_mlx4 that implements a single RSS configuration per
1766 port, librte_net_mlx5 supports per-protocol RSS configuration.
1768 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1769 command-line parameter to enable additional protocols (UDP and TCP as well
1770 as IP), the following commands must be entered from its CLI to get the same
1771 behavior as librte_net_mlx4::
1774 > port config all rss all
1780 This section demonstrates how to launch **testpmd** with Mellanox
1781 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1783 #. Load the kernel modules::
1785 modprobe -a ib_uverbs mlx5_core mlx5_ib
1787 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1790 /etc/init.d/openibd restart
1794 User space I/O kernel modules (uio and igb_uio) are not used and do
1795 not have to be loaded.
1797 #. Make sure Ethernet interfaces are in working order and linked to kernel
1798 verbs. Related sysfs entries should be present::
1800 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1809 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1812 for intf in eth2 eth3 eth4 eth5;
1814 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1817 sed -n 's,.*/\(.*\),-a \1,p'
1826 #. Request huge pages::
1828 dpdk-hugepages.py --setup 2G
1830 #. Start testpmd with basic parameters::
1832 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
1837 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1838 EAL: probe driver: 15b3:1013 librte_net_mlx5
1839 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1840 PMD: librte_net_mlx5: 1 port(s) detected
1841 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1842 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1843 EAL: probe driver: 15b3:1013 librte_net_mlx5
1844 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1845 PMD: librte_net_mlx5: 1 port(s) detected
1846 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1847 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1848 EAL: probe driver: 15b3:1013 librte_net_mlx5
1849 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1850 PMD: librte_net_mlx5: 1 port(s) detected
1851 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1852 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1853 EAL: probe driver: 15b3:1013 librte_net_mlx5
1854 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1855 PMD: librte_net_mlx5: 1 port(s) detected
1856 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1857 Interactive-mode selected
1858 Configuring Port 0 (socket 0)
1859 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1860 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1861 Port 0: E4:1D:2D:E7:0C:FE
1862 Configuring Port 1 (socket 0)
1863 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1864 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1865 Port 1: E4:1D:2D:E7:0C:FF
1866 Configuring Port 2 (socket 0)
1867 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1868 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1869 Port 2: E4:1D:2D:E7:0C:FA
1870 Configuring Port 3 (socket 0)
1871 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1872 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1873 Port 3: E4:1D:2D:E7:0C:FB
1874 Checking link statuses...
1875 Port 0 Link Up - speed 40000 Mbps - full-duplex
1876 Port 1 Link Up - speed 40000 Mbps - full-duplex
1877 Port 2 Link Up - speed 10000 Mbps - full-duplex
1878 Port 3 Link Up - speed 10000 Mbps - full-duplex
1885 This section demonstrates how to dump flows. Currently, it's possible to dump
1886 all flows with assistance of external tools.
1888 #. 2 ways to get flow raw file:
1890 - Using testpmd CLI:
1892 .. code-block:: console
1895 testpmd> flow dump <port> all <output_file>
1897 testpmd> flow dump <port> rule <rule_id> <output_file>
1899 - call rte_flow_dev_dump api:
1901 .. code-block:: console
1903 rte_flow_dev_dump(port, flow, file, NULL);
1905 #. Dump human-readable flows from raw file:
1907 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1909 .. code-block:: console
1911 mlx_steering_dump.py -f <output_file> -flowptr <flow_ptr>