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.
118 On Windows, the features are limited:
120 - Promiscuous mode is not supported
121 - The following rules are supported:
123 - IPv4/UDP with CVLAN filtering
124 - Unicast MAC filtering
126 - For secondary process:
128 - Forked secondary process not supported.
129 - External memory unregistered in EAL memseg list cannot be used for DMA
130 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
131 primary process and remapped to the same virtual address in secondary
132 process. If the external memory is registered by primary process but has
133 different virtual address in secondary process, unexpected error may happen.
135 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
136 specific VLAN will match for VLAN packets as well:
138 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
139 Meaning, the flow rule::
141 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
143 Will only match vlan packets with vid=3. and the flow rule::
145 flow create 0 ingress pattern eth / ipv4 / end ...
147 Will match any ipv4 packet (VLAN included).
149 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
151 - 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.
154 flow create 0 ingress pattern eth / ipv4 / end ...
156 Will match any ipv4 packet.
159 flow create 0 ingress pattern eth / vlan / end ...
160 flow create 0 ingress pattern eth has_vlan is 1 / end ...
161 flow create 0 ingress pattern eth type is 0x8100 / end ...
163 Will match single-tagged packets only, with any VLAN ID value.
166 flow create 0 ingress pattern eth type is 0x88A8 / end ...
167 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
169 Will match multi-tagged packets only, with any VLAN ID value.
171 - A flow pattern with 2 sequential VLAN items is not supported.
173 - VLAN pop offload command:
175 - Flow rules having a VLAN pop offload command as one of their actions and
176 are lacking a match on VLAN as one of their items are not supported.
177 - The command is not supported on egress traffic in NIC mode.
179 - VLAN push offload is not supported on ingress traffic in NIC mode.
181 - VLAN set PCP offload is not supported on existing headers.
183 - A multi segment packet must have not more segments than reported by dev_infos_get()
184 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
185 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
186 inline settings) to 58.
188 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
189 to 0 are not supported.
191 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
193 - Match on Geneve header supports the following fields only:
200 - Match on Geneve TLV option is supported on the following fields:
207 Only one Class/Type/Length Geneve TLV option is supported per shared device.
208 Class/Type/Length fields must be specified as well as masks.
209 Class/Type/Length specified masks must be full.
210 Matching Geneve TLV option without specifying data is not supported.
211 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
213 - VF: flow rules created on VF devices can only match traffic targeted at the
214 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
216 - Match on GTP tunnel header item supports the following fields only:
218 - v_pt_rsv_flags: E flag, S flag, PN flag
222 - Match on GTP extension header only for GTP PDU session container (next
223 extension header type = 0x85).
224 - Match on GTP extension header is not supported in group 0.
226 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
227 The flows within group 0 and set metadata action are rejected by hardware.
231 MAC addresses not already present in the bridge table of the associated
232 kernel network device will be added and cleaned up by the PMD when closing
233 the device. In case of ungraceful program termination, some entries may
234 remain present and should be removed manually by other means.
236 - Buffer split offload is supported with regular Rx burst routine only,
237 no MPRQ feature or vectorized code can be engaged.
239 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
240 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
241 ol_flags. As the mempool for the external buffer is managed by PMD, all the
242 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
243 the external buffers will be freed by PMD and the application which still
244 holds the external buffers may be corrupted.
246 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
247 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
248 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
250 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
251 and allmulticast mode are both set to off.
252 To receive IPv6 Multicast messages on VM, explicitly set the relevant
253 MAC address using rte_eth_dev_mac_addr_add() API.
255 - To support a mixed traffic pattern (some buffers from local host memory, some
256 buffers from other devices) with high bandwidth, a mbuf flag is used.
258 An application hints the PMD whether or not it should try to inline the
259 given mbuf data buffer. PMD should do the best effort to act upon this request.
261 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
262 registered by application with rte_mbuf_dynflag_register(). This flag is
263 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
264 which is intended to be used by the application.
266 To query the supported specific flags in runtime,
267 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
268 currently (over present hardware and configuration) supported specific flags.
269 The "not inline hint" feature operating flow is the following one:
272 - probe the devices, ports are created
273 - query the port capabilities
274 - if port supporting the feature is found
275 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
276 - application starts the ports
277 - on ``dev_start()`` PMD checks whether the feature flag is registered and
278 enables the feature support in datapath
279 - application might set the registered flag bit in ``ol_flags`` field
280 of mbuf being sent and PMD will handle ones appropriately.
282 - The amount of descriptors in Tx queue may be limited by data inline settings.
283 Inline data require the more descriptor building blocks and overall block
284 amount may exceed the hardware supported limits. The application should
285 reduce the requested Tx size or adjust data inline settings with
286 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
288 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
289 parameter should be specified.
290 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
291 being sent it tries to synchronize the time of packet appearing on
292 the wire with the specified packet timestamp. It the specified one
293 is in the past it should be ignored, if one is in the distant future
294 it should be capped with some reasonable value (in range of seconds).
295 These specific cases ("too late" and "distant future") can be optionally
296 reported via device xstats to assist applications to detect the
297 time-related problems.
299 The timestamp upper "too-distant-future" limit
300 at the moment of invoking the Tx burst routine
301 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
302 Please note, for the testpmd txonly mode,
303 the limit is deduced from the expression::
305 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
307 There is no any packet reordering according timestamps is supposed,
308 neither within packet burst, nor between packets, it is an entirely
309 application responsibility to generate packets and its timestamps
310 in desired order. The timestamps can be put only in the first packet
311 in the burst providing the entire burst scheduling.
313 - E-Switch decapsulation Flow:
315 - can be applied to PF port only.
316 - must specify VF port action (packet redirection from PF to VF).
317 - optionally may specify tunnel inner source and destination MAC addresses.
319 - E-Switch encapsulation Flow:
321 - can be applied to VF ports only.
322 - must specify PF port action (packet redirection from VF to PF).
326 - The input buffer, used as outer header, is not validated.
330 - The decapsulation is always done up to the outermost tunnel detected by the HW.
331 - The input buffer, providing the removal size, is not validated.
332 - The buffer size must match the length of the headers to be removed.
334 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
335 mutually exclusive features which cannot be supported together
336 (see :ref:`mlx5_firmware_config`).
340 - Requires DevX and DV flow to be enabled.
341 - KEEP_CRC offload cannot be supported with LRO.
342 - The first mbuf length, without head-room, must be big enough to include the
344 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
345 it with size limited to max LRO size, not to max RX packet length.
346 - LRO can be used with outer header of TCP packets of the standard format:
347 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
349 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
350 - LRO packet aggregation is performed by HW only for packet size larger than
351 ``lro_min_mss_size``. This value is reported on device start, when debug
356 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
357 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
358 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
362 - fast free offload assumes the all mbufs being sent are originated from the
363 same memory pool and there is no any extra references to the mbufs (the
364 reference counter for each mbuf is equal 1 on tx_burst call). The latter
365 means there should be no any externally attached buffers in mbufs. It is
366 an application responsibility to provide the correct mbufs if the fast
367 free offload is engaged. The mlx5 PMD implicitly produces the mbufs with
368 externally attached buffers if MPRQ option is enabled, hence, the fast
369 free offload is neither supported nor advertised if there is MPRQ enabled.
373 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and
374 E-Switch steering domain.
375 - For E-Switch Sampling flow with sample ratio > 1, additional actions are not
376 supported in the sample actions list.
377 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as
378 first action in the E-Switch egress flow if with header modify or
379 encapsulation actions.
380 - For NIC Rx flow, supports ``MARK``, ``COUNT``, ``QUEUE``, ``RSS`` in the
382 - For E-Switch mirroring flow, supports ``RAW ENCAP``, ``Port ID``,
383 ``VXLAN ENCAP``, ``NVGRE ENCAP`` in the sample actions list.
387 - Supports the 'set' operation only for ``RTE_FLOW_ACTION_TYPE_MODIFY_FIELD`` action.
388 - Modification of an arbitrary place in a packet via the special ``RTE_FLOW_FIELD_START`` Field ID is not supported.
389 - Modification of the 802.1Q Tag, VXLAN Network or GENEVE Network ID's is not supported.
390 - Encapsulation levels are not supported, can modify outermost header fields only.
391 - Offsets must be 32-bits aligned, cannot skip past the boundary of a field.
393 - IPv6 header item 'proto' field, indicating the next header protocol, should
394 not be set as extension header.
395 In case the next header is an extension header, it should not be specified in
396 IPv6 header item 'proto' field.
397 The last extension header item 'next header' field can specify the following
398 header protocol type.
402 - 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.
403 - Hairpin in switchdev SR-IOV mode is not supported till now.
407 - All the meter colors with drop action will be counted only by the global drop statistics.
408 - Green color is not supported with drop action.
409 - Yellow detection is not supported.
410 - Red color must be with drop action.
411 - Meter statistics are supported only for drop case.
412 - Meter yellow color detection is not supported.
413 - A meter action created with pre-defined policy must be the last action in the flow except single case where the policy actions are:
414 - green: NULL or END.
415 - yellow: NULL or END.
417 - The only supported meter policy actions:
418 - green: QUEUE, RSS, PORT_ID, JUMP, MARK and SET_TAG.
419 - yellow: must be empty.
421 - meter profile packet mode is supported.
425 - Integrity offload is enabled for **ConnectX-6** family.
426 - Verification bits provided by the hardware are ``l3_ok``, ``ipv4_csum_ok``, ``l4_ok``, ``l4_csum_ok``.
427 - ``level`` value 0 references outer headers.
428 - Multiple integrity items not supported in a single flow rule.
429 - Flow rule items supplied by application must explicitly specify network headers referred by integrity item.
430 For example, if integrity item mask sets ``l4_ok`` or ``l4_csum_ok`` bits, reference to L4 network header,
431 TCP or UDP, must be in the rule pattern as well::
433 flow create 0 ingress pattern integrity level is 0 value mask l3_ok value spec l3_ok / eth / ipv6 / end …
435 flow create 0 ingress pattern integrity level is 0 value mask l4_ok value spec 0 / eth / ipv4 proto is udp / end …
437 - Connection tracking:
439 - Cannot co-exist with ASO meter, ASO age action in a single flow rule.
440 - Flow rules insertion rate and memory consumption need more optimization.
442 - 4M connections maximum.
447 MLX5 supports various methods to report statistics:
449 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.
451 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.
453 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.
461 The ibverbs libraries can be linked with this PMD in a number of ways,
462 configured by the ``ibverbs_link`` build option:
464 - ``shared`` (default): the PMD depends on some .so files.
466 - ``dlopen``: Split the dependencies glue in a separate library
467 loaded when needed by dlopen.
468 It make dependencies on libibverbs and libmlx4 optional,
469 and has no performance impact.
471 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
472 in the PMD shared library or the executable static binary.
474 Environment variables
475 ~~~~~~~~~~~~~~~~~~~~~
479 A list of directories in which to search for the rdma-core "glue" plug-in,
480 separated by colons or semi-colons.
482 - ``MLX5_SHUT_UP_BF``
484 Configures HW Tx doorbell register as IO-mapped.
486 By default, the HW Tx doorbell is configured as a write-combining register.
487 The register would be flushed to HW usually when the write-combining buffer
488 becomes full, but it depends on CPU design.
490 Except for vectorized Tx burst routines, a write memory barrier is enforced
491 after updating the register so that the update can be immediately visible to
494 When vectorized Tx burst is called, the barrier is set only if the burst size
495 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
496 variable will bring better latency even though the maximum throughput can
499 Run-time configuration
500 ~~~~~~~~~~~~~~~~~~~~~~
502 - librte_net_mlx5 brings kernel network interfaces up during initialization
503 because it is affected by their state. Forcing them down prevents packets
506 - **ethtool** operations on related kernel interfaces also affect the PMD.
511 In order to run as a non-root user,
512 some capabilities must be granted to the application::
514 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
516 Below are the reasons of the need for each capability:
519 When using physical addresses (PA mode), with Linux >= 4.0,
520 for access to ``/proc/self/pagemap``.
523 For device configuration.
526 For raw ethernet queue allocation through kernel driver.
529 For DMA memory pinning.
534 - ``rxq_cqe_comp_en`` parameter [int]
536 A nonzero value enables the compression of CQE on RX side. This feature
537 allows to save PCI bandwidth and improve performance. Enabled by default.
538 Different compression formats are supported in order to achieve the best
539 performance for different traffic patterns. Default format depends on
540 Multi-Packet Rx queue configuration: Hash RSS format is used in case
541 MPRQ is disabled, Checksum format is used in case MPRQ is enabled.
543 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
544 better compression rate in case of RTE Flow Mark traffic.
545 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
546 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
547 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
548 CQE compression format selection requires DevX to be enabled. If there is
549 no DevX enabled/supported the value is reset to 1 by default.
553 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
554 ConnectX-6 Lx, BlueField and BlueField-2.
555 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
556 ConnectX-6 Lx, BlueField and BlueField-2.
558 - ``rxq_pkt_pad_en`` parameter [int]
560 A nonzero value enables padding Rx packet to the size of cacheline on PCI
561 transaction. This feature would waste PCI bandwidth but could improve
562 performance by avoiding partial cacheline write which may cause costly
563 read-modify-copy in memory transaction on some architectures. Disabled by
568 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
569 ConnectX-6 Lx, BlueField and BlueField-2.
570 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
571 ConnectX-6 Lx, BlueField and BlueField-2.
573 - ``mprq_en`` parameter [int]
575 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
576 configured as Multi-Packet RQ if the total number of Rx queues is
577 ``rxqs_min_mprq`` or more. Disabled by default.
579 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
580 by posting a single large buffer for multiple packets. Instead of posting a
581 buffers per a packet, one large buffer is posted in order to receive multiple
582 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
583 and each stride receives one packet. MPRQ can improve throughput for
584 small-packet traffic.
586 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
587 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
588 configure large stride size enough to accommodate max_rx_pkt_len as long as
589 device allows. Note that this can waste system memory compared to enabling Rx
590 scatter and multi-segment packet.
592 - ``mprq_log_stride_num`` parameter [int]
594 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
595 strides can reduce PCIe traffic further. If configured value is not in the
596 range of device capability, the default value will be set with a warning
597 message. The default value is 4 which is 16 strides per a buffer, valid only
598 if ``mprq_en`` is set.
600 The size of Rx queue should be bigger than the number of strides.
602 - ``mprq_log_stride_size`` parameter [int]
604 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
605 stride size can save some memory and reduce probability of a depletion of all
606 available strides due to unreleased packets by an application. If configured
607 value is not in the range of device capability, the default value will be set
608 with a warning message. The default value is 11 which is 2048 bytes per a
609 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
610 it is possible for a packet to span across multiple strides. This mode allows
611 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
612 of a packet if Rx scatter is configured) may be required in case there is no
613 space left for a head room at the end of a stride which incurs some
616 - ``mprq_max_memcpy_len`` parameter [int]
618 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
619 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
620 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
621 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
622 A mempool for external buffers will be allocated and managed by PMD. If Rx
623 packet is externally attached, ol_flags field of the mbuf will have
624 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
625 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
627 - ``rxqs_min_mprq`` parameter [int]
629 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
630 greater or equal to this value. The default value is 12, valid only if
633 - ``txq_inline`` parameter [int]
635 Amount of data to be inlined during TX operations. This parameter is
636 deprecated and converted to the new parameter ``txq_inline_max`` providing
637 partial compatibility.
639 - ``txqs_min_inline`` parameter [int]
641 Enable inline data send only when the number of TX queues is greater or equal
644 This option should be used in combination with ``txq_inline_max`` and
645 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
647 If this option is not specified the default value 16 is used for BlueField
648 and 8 for other platforms
650 The data inlining consumes the CPU cycles, so this option is intended to
651 auto enable inline data if we have enough Tx queues, which means we have
652 enough CPU cores and PCI bandwidth is getting more critical and CPU
653 is not supposed to be bottleneck anymore.
655 The copying data into WQE improves latency and can improve PPS performance
656 when PCI back pressure is detected and may be useful for scenarios involving
657 heavy traffic on many queues.
659 Because additional software logic is necessary to handle this mode, this
660 option should be used with care, as it may lower performance when back
661 pressure is not expected.
663 If inline data are enabled it may affect the maximal size of Tx queue in
664 descriptors because the inline data increase the descriptor size and
665 queue size limits supported by hardware may be exceeded.
667 - ``txq_inline_min`` parameter [int]
669 Minimal amount of data to be inlined into WQE during Tx operations. NICs
670 may require this minimal data amount to operate correctly. The exact value
671 may depend on NIC operation mode, requested offloads, etc. It is strongly
672 recommended to omit this parameter and use the default values. Anyway,
673 applications using this parameter should take into consideration that
674 specifying an inconsistent value may prevent the NIC from sending packets.
676 If ``txq_inline_min`` key is present the specified value (may be aligned
677 by the driver in order not to exceed the limits and provide better descriptor
678 space utilization) will be used by the driver and it is guaranteed that
679 requested amount of data bytes are inlined into the WQE beside other inline
680 settings. This key also may update ``txq_inline_max`` value (default
681 or specified explicitly in devargs) to reserve the space for inline data.
683 If ``txq_inline_min`` key is not present, the value may be queried by the
684 driver from the NIC via DevX if this feature is available. If there is no DevX
685 enabled/supported the value 18 (supposing L2 header including VLAN) is set
686 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
687 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
690 For ConnectX-4 NIC, driver does not allow specifying value below 18
691 (minimal L2 header, including VLAN), error will be raised.
693 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
694 it is not recommended and may prevent NIC from sending packets over
697 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
698 Multi-Packet Write), because last one does not support partial packet inlining.
699 This is not very critical due to minimal data inlining is mostly required
700 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
702 - ``txq_inline_max`` parameter [int]
704 Specifies the maximal packet length to be completely inlined into WQE
705 Ethernet Segment for ordinary SEND method. If packet is larger than specified
706 value, the packet data won't be copied by the driver at all, data buffer
707 is addressed with a pointer. If packet length is less or equal all packet
708 data will be copied into WQE. This may improve PCI bandwidth utilization for
709 short packets significantly but requires the extra CPU cycles.
711 The data inline feature is controlled by number of Tx queues, if number of Tx
712 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
713 is engaged, if there are not enough Tx queues (which means not enough CPU cores
714 and CPU resources are scarce), data inline is not performed by the driver.
715 Assigning ``txqs_min_inline`` with zero always enables the data inline.
717 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
718 by the driver in order not to exceed the limit (930 bytes) and to provide better
719 WQE space filling without gaps, the adjustment is reflected in the debug log.
720 Also, the default value (290) may be decreased in run-time if the large transmit
721 queue size is requested and hardware does not support enough descriptor
722 amount, in this case warning is emitted. If ``txq_inline_max`` key is
723 specified and requested inline settings can not be satisfied then error
726 - ``txq_inline_mpw`` parameter [int]
728 Specifies the maximal packet length to be completely inlined into WQE for
729 Enhanced MPW method. If packet is large the specified value, the packet data
730 won't be copied, and data buffer is addressed with pointer. If packet length
731 is less or equal, all packet data will be copied into WQE. This may improve PCI
732 bandwidth utilization for short packets significantly but requires the extra
735 The data inline feature is controlled by number of TX queues, if number of Tx
736 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
737 is engaged, if there are not enough Tx queues (which means not enough CPU cores
738 and CPU resources are scarce), data inline is not performed by the driver.
739 Assigning ``txqs_min_inline`` with zero always enables the data inline.
741 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
742 by the driver in order not to exceed the limit (930 bytes) and to provide better
743 WQE space filling without gaps, the adjustment is reflected in the debug log.
744 Due to multiple packets may be included to the same WQE with Enhanced Multi
745 Packet Write Method and overall WQE size is limited it is not recommended to
746 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
747 may be decreased in run-time if the large transmit queue size is requested
748 and hardware does not support enough descriptor amount, in this case warning
749 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
750 settings can not be satisfied then error will be raised.
752 - ``txqs_max_vec`` parameter [int]
754 Enable vectorized Tx only when the number of TX queues is less than or
755 equal to this value. This parameter is deprecated and ignored, kept
756 for compatibility issue to not prevent driver from probing.
758 - ``txq_mpw_hdr_dseg_en`` parameter [int]
760 A nonzero value enables including two pointers in the first block of TX
761 descriptor. The parameter is deprecated and ignored, kept for compatibility
764 - ``txq_max_inline_len`` parameter [int]
766 Maximum size of packet to be inlined. This limits the size of packet to
767 be inlined. If the size of a packet is larger than configured value, the
768 packet isn't inlined even though there's enough space remained in the
769 descriptor. Instead, the packet is included with pointer. This parameter
770 is deprecated and converted directly to ``txq_inline_mpw`` providing full
771 compatibility. Valid only if eMPW feature is engaged.
773 - ``txq_mpw_en`` parameter [int]
775 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
776 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
777 eMPW allows the Tx burst function to pack up multiple packets
778 in a single descriptor session in order to save PCI bandwidth
779 and improve performance at the cost of a slightly higher CPU usage.
780 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
781 Tx burst function copies entire packet data on to Tx descriptor
782 instead of including pointer of packet.
784 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
785 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
786 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
787 option or reported by the NIC, the eMPW feature is disengaged.
789 - ``tx_db_nc`` parameter [int]
791 The rdma core library can map doorbell register in two ways, depending on the
792 environment variable "MLX5_SHUT_UP_BF":
794 - As regular cached memory (usually with write combining attribute), if the
795 variable is either missing or set to zero.
796 - As non-cached memory, if the variable is present and set to not "0" value.
798 The type of mapping may slightly affect the Tx performance, the optimal choice
799 is strongly relied on the host architecture and should be deduced practically.
801 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
802 memory (with write combining), the PMD will perform the extra write memory barrier
803 after writing to doorbell, it might increase the needed CPU clocks per packet
804 to send, but latency might be improved.
806 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
807 cached memory, the PMD will not perform the extra write memory barrier
808 after writing to doorbell, on some architectures it might improve the
811 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
812 memory, the PMD will use heuristics to decide whether write memory barrier
813 should be performed. For bursts with size multiple of recommended one (64 pkts)
814 it is supposed the next burst is coming and no need to issue the extra memory
815 barrier (it is supposed to be issued in the next coming burst, at least after
816 descriptor writing). It might increase latency (on some hosts till next
817 packets transmit) and should be used with care.
819 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
820 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
821 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
823 - ``tx_pp`` parameter [int]
825 If a nonzero value is specified the driver creates all necessary internal
826 objects to provide accurate packet send scheduling on mbuf timestamps.
827 The positive value specifies the scheduling granularity in nanoseconds,
828 the packet send will be accurate up to specified digits. The allowed range is
829 from 500 to 1 million of nanoseconds. The negative value specifies the module
830 of granularity and engages the special test mode the check the schedule rate.
831 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
834 - ``tx_skew`` parameter [int]
836 The parameter adjusts the send packet scheduling on timestamps and represents
837 the average delay between beginning of the transmitting descriptor processing
838 by the hardware and appearance of actual packet data on the wire. The value
839 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
840 specified. The default value is zero.
842 - ``tx_vec_en`` parameter [int]
844 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
845 ConnectX-6 Lx, BlueField and BlueField-2 NICs
846 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
847 The parameter is deprecated and ignored.
849 - ``rx_vec_en`` parameter [int]
851 A nonzero value enables Rx vector if the port is not configured in
852 multi-segment otherwise this parameter is ignored.
856 - ``vf_nl_en`` parameter [int]
858 A nonzero value enables Netlink requests from the VF to add/remove MAC
859 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
860 Otherwise the relevant configuration must be run with Linux iproute2 tools.
861 This is a prerequisite to receive this kind of traffic.
863 Enabled by default, valid only on VF devices ignored otherwise.
865 - ``l3_vxlan_en`` parameter [int]
867 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
868 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
869 parameter. This is a prerequisite to receive this kind of traffic.
873 - ``dv_xmeta_en`` parameter [int]
875 A nonzero value enables extensive flow metadata support if device is
876 capable and driver supports it. This can enable extensive support of
877 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
878 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
880 There are some possible configurations, depending on parameter value:
882 - 0, this is default value, defines the legacy mode, the ``MARK`` and
883 ``META`` related actions and items operate only within NIC Tx and
884 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
885 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
886 item is 32 bits wide and match supported on egress only.
888 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
889 related actions and items operate within all supported steering domains,
890 including FDB, ``MARK`` and ``META`` information may cross the domain
891 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
892 depends on kernel and firmware configurations and might be 0, 16 or
893 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
894 compatibility, the actual width of data transferred to the FDB domain
895 depends on kernel configuration and may be vary. The actual supported
896 width can be retrieved in runtime by series of rte_flow_validate()
899 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
900 related actions and items operate within all supported steering domains,
901 including FDB, ``MARK`` and ``META`` information may cross the domain
902 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
903 depends on kernel and firmware configurations and might be 0, 16 or
904 24 bits. The actual supported width can be retrieved in runtime by
905 series of rte_flow_validate() trials.
907 - 3, this engages tunnel offload mode. In E-Switch configuration, that
908 mode implicitly activates ``dv_xmeta_en=1``.
910 +------+-----------+-----------+-------------+-------------+
911 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
912 +======+===========+===========+=============+=============+
913 | 0 | 24 bits | 32 bits | 32 bits | no |
914 +------+-----------+-----------+-------------+-------------+
915 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
916 +------+-----------+-----------+-------------+-------------+
917 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
918 +------+-----------+-----------+-------------+-------------+
920 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
921 ignored and the device is configured to operate in legacy mode (0).
923 Disabled by default (set to 0).
925 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
926 of the extensive metadata features. The legacy Verbs supports FLAG and
927 MARK metadata actions over NIC Rx steering domain only.
929 Setting META value to zero in flow action means there is no item provided
930 and receiving datapath will not report in mbufs the metadata are present.
931 Setting MARK value to zero in flow action means the zero FDIR ID value
932 will be reported on packet receiving.
934 For the MARK action the last 16 values in the full range are reserved for
935 internal PMD purposes (to emulate FLAG action). The valid range for the
936 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
937 for the 24-bit mode, the flows with the MARK action value outside
938 the specified range will be rejected.
940 - ``dv_flow_en`` parameter [int]
942 A nonzero value enables the DV flow steering assuming it is supported
943 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
945 Enabled by default if supported.
947 - ``dv_esw_en`` parameter [int]
949 A nonzero value enables E-Switch using Direct Rules.
951 Enabled by default if supported.
953 - ``lacp_by_user`` parameter [int]
955 A nonzero value enables the control of LACP traffic by the user application.
956 When a bond exists in the driver, by default it should be managed by the
957 kernel and therefore LACP traffic should be steered to the kernel.
958 If this devarg is set to 1 it will allow the user to manage the bond by
959 itself and not steer LACP traffic to the kernel.
961 Disabled by default (set to 0).
963 - ``mr_ext_memseg_en`` parameter [int]
965 A nonzero value enables extending memseg when registering DMA memory. If
966 enabled, the number of entries in MR (Memory Region) lookup table on datapath
967 is minimized and it benefits performance. On the other hand, it worsens memory
968 utilization because registered memory is pinned by kernel driver. Even if a
969 page in the extended chunk is freed, that doesn't become reusable until the
970 entire memory is freed.
974 - ``representor`` parameter [list]
976 This parameter can be used to instantiate DPDK Ethernet devices from
977 existing port (PF, VF or SF) representors configured on the device.
979 It is a standard parameter whose format is described in
980 :ref:`ethernet_device_standard_device_arguments`.
982 For instance, to probe VF port representors 0 through 2::
984 <PCI_BDF>,representor=vf[0-2]
986 To probe SF port representors 0 through 2::
988 <PCI_BDF>,representor=sf[0-2]
990 To probe VF port representors 0 through 2 on both PFs of bonding device::
992 <Primary_PCI_BDF>,representor=pf[0,1]vf[0-2]
994 - ``max_dump_files_num`` parameter [int]
996 The maximum number of files per PMD entity that may be created for debug information.
997 The files will be created in /var/log directory or in current directory.
999 set to 128 by default.
1001 - ``lro_timeout_usec`` parameter [int]
1003 The maximum allowed duration of an LRO session, in micro-seconds.
1004 PMD will set the nearest value supported by HW, which is not bigger than
1005 the input ``lro_timeout_usec`` value.
1006 If this parameter is not specified, by default PMD will set
1007 the smallest value supported by HW.
1009 - ``hp_buf_log_sz`` parameter [int]
1011 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
1012 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
1013 The capacity of the value is specified by the firmware and the initialization
1014 will get a failure if it is out of scope.
1015 The range of the value is from 11 to 19 right now, and the supported frame
1016 size of a single packet for hairpin is from 512B to 128KB. It might change if
1017 different firmware release is being used. By using a small value, it could
1018 reduce memory consumption but not work with a large frame. If the value is
1019 too large, the memory consumption will be high and some potential performance
1020 degradation will be introduced.
1021 By default, the PMD will set this value to 16, which means that 9KB jumbo
1022 frames will be supported.
1024 - ``reclaim_mem_mode`` parameter [int]
1026 Cache some resources in flow destroy will help flow recreation more efficient.
1027 While some systems may require the all the resources can be reclaimed after
1029 The parameter ``reclaim_mem_mode`` provides the option for user to configure
1030 if the resource cache is needed or not.
1032 There are three options to choose:
1034 - 0. It means the flow resources will be cached as usual. The resources will
1035 be cached, helpful with flow insertion rate.
1037 - 1. It will only enable the DPDK PMD level resources reclaim.
1039 - 2. Both DPDK PMD level and rdma-core low level will be configured as
1042 By default, the PMD will set this value to 0.
1044 - ``sys_mem_en`` parameter [int]
1046 A non-zero value enables the PMD memory management allocating memory
1047 from system by default, without explicit rte memory flag.
1049 By default, the PMD will set this value to 0.
1051 - ``decap_en`` parameter [int]
1053 Some devices do not support FCS (frame checksum) scattering for
1054 tunnel-decapsulated packets.
1055 If set to 0, this option forces the FCS feature and rejects tunnel
1056 decapsulation in the flow engine for such devices.
1058 By default, the PMD will set this value to 1.
1060 .. _mlx5_firmware_config:
1062 Firmware configuration
1063 ~~~~~~~~~~~~~~~~~~~~~~
1065 Firmware features can be configured as key/value pairs.
1067 The command to set a value is::
1069 mlxconfig -d <device> set <key>=<value>
1071 The command to query a value is::
1073 mlxconfig -d <device> query | grep <key>
1075 The device name for the command ``mlxconfig`` can be either the PCI address,
1076 or the mst device name found with::
1080 Below are some firmware configurations listed.
1086 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1092 - maximum number of SR-IOV virtual functions::
1096 - enable DevX (required by Direct Rules and other features)::
1100 - aggressive CQE zipping::
1104 - L3 VXLAN and VXLAN-GPE destination UDP port::
1107 IP_OVER_VXLAN_PORT=<udp dport>
1109 - enable VXLAN-GPE tunnel flow matching::
1111 FLEX_PARSER_PROFILE_ENABLE=0
1113 FLEX_PARSER_PROFILE_ENABLE=2
1115 - enable IP-in-IP tunnel flow matching::
1117 FLEX_PARSER_PROFILE_ENABLE=0
1119 - enable MPLS flow matching::
1121 FLEX_PARSER_PROFILE_ENABLE=1
1123 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1125 FLEX_PARSER_PROFILE_ENABLE=2
1127 - enable Geneve flow matching::
1129 FLEX_PARSER_PROFILE_ENABLE=0
1131 FLEX_PARSER_PROFILE_ENABLE=1
1133 - enable Geneve TLV option flow matching::
1135 FLEX_PARSER_PROFILE_ENABLE=0
1137 - enable GTP flow matching::
1139 FLEX_PARSER_PROFILE_ENABLE=3
1141 - enable eCPRI flow matching::
1143 FLEX_PARSER_PROFILE_ENABLE=4
1149 This driver relies on external libraries and kernel drivers for resources
1150 allocations and initialization. The following dependencies are not part of
1151 DPDK and must be installed separately:
1155 User space Verbs framework used by librte_net_mlx5. This library provides
1156 a generic interface between the kernel and low-level user space drivers
1159 It allows slow and privileged operations (context initialization, hardware
1160 resources allocations) to be managed by the kernel and fast operations to
1161 never leave user space.
1165 Low-level user space driver library for Mellanox
1166 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1169 This library basically implements send/receive calls to the hardware
1172 - **Kernel modules**
1174 They provide the kernel-side Verbs API and low level device drivers that
1175 manage actual hardware initialization and resources sharing with user
1178 Unlike most other PMDs, these modules must remain loaded and bound to
1181 - mlx5_core: hardware driver managing Mellanox
1182 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1184 - mlx5_ib: InifiniBand device driver.
1185 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1187 - **Firmware update**
1189 Mellanox OFED/EN releases include firmware updates for
1190 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1192 Because each release provides new features, these updates must be applied to
1193 match the kernel modules and libraries they come with.
1197 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1203 Either RDMA Core library with a recent enough Linux kernel release
1204 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1207 RDMA Core with Linux Kernel
1208 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1210 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1211 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1212 (see `RDMA Core installation documentation`_)
1213 - When building for i686 use:
1215 - rdma-core version 18.0 or above built with 32bit support.
1216 - Kernel version 4.14.41 or above.
1218 - Starting with rdma-core v21, static libraries can be built::
1221 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1224 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1225 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1231 - Mellanox OFED version: **4.5** and above /
1232 Mellanox EN version: **4.5** and above
1235 - ConnectX-4: **12.21.1000** and above.
1236 - ConnectX-4 Lx: **14.21.1000** and above.
1237 - ConnectX-5: **16.21.1000** and above.
1238 - ConnectX-5 Ex: **16.21.1000** and above.
1239 - ConnectX-6: **20.27.0090** and above.
1240 - ConnectX-6 Dx: **22.27.0090** and above.
1241 - BlueField: **18.25.1010** and above.
1243 While these libraries and kernel modules are available on OpenFabrics
1244 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1245 managers on most distributions, this PMD requires Ethernet extensions that
1246 may not be supported at the moment (this is a work in progress).
1249 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1251 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1252 include the necessary support and should be used in the meantime. For DPDK,
1253 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1254 required from that distribution.
1258 Several versions of Mellanox OFED/EN are available. Installing the version
1259 this DPDK release was developed and tested against is strongly
1260 recommended. Please check the `linux prerequisites`_.
1262 Windows Prerequisites
1263 ---------------------
1265 This driver relies on external libraries and kernel drivers for resources
1266 allocations and initialization. The dependencies in the following sub-sections
1267 are not part of DPDK, and must be installed separately.
1269 Compilation Prerequisites
1270 ~~~~~~~~~~~~~~~~~~~~~~~~~
1272 DevX SDK installation
1273 ^^^^^^^^^^^^^^^^^^^^^
1275 The DevX SDK must be installed on the machine building the Windows PMD.
1276 Additional information can be found at
1277 `How to Integrate Windows DevX in Your Development Environment
1278 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1280 Runtime Prerequisites
1281 ~~~~~~~~~~~~~~~~~~~~~
1283 WinOF2 version 2.60 or higher must be installed on the machine.
1288 The driver can be downloaded from the following site:
1290 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1295 DevX for Windows must be enabled in the Windows registry.
1296 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1297 Additional information can be found in the WinOF2 user manual.
1302 The following Mellanox device families are supported by the same mlx5 driver:
1314 Below are detailed device names:
1316 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1317 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1318 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1319 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1320 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1321 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1322 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1323 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1324 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1325 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1326 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1327 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1328 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1329 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1330 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1331 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1332 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1333 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1334 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1335 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1336 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1337 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1338 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1339 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1340 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1341 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1343 Quick Start Guide on OFED/EN
1344 ----------------------------
1346 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1349 2. Install the required libraries and kernel modules either by installing
1350 only the required set, or by installing the entire Mellanox OFED/EN::
1352 ./mlnxofedinstall --upstream-libs --dpdk
1354 3. Verify the firmware is the correct one::
1358 4. Verify all ports links are set to Ethernet::
1360 mlxconfig -d <mst device> query | grep LINK_TYPE
1364 Link types may have to be configured to Ethernet::
1366 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1368 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1370 For hypervisors, verify SR-IOV is enabled on the NIC::
1372 mlxconfig -d <mst device> query | grep SRIOV_EN
1375 If needed, configure SR-IOV::
1377 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1378 mlxfwreset -d <mst device> reset
1380 5. Restart the driver::
1382 /etc/init.d/openibd restart
1386 service openibd restart
1388 If link type was changed, firmware must be reset as well::
1390 mlxfwreset -d <mst device> reset
1392 For hypervisors, after reset write the sysfs number of virtual functions
1395 To dynamically instantiate a given number of virtual functions (VFs)::
1397 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1399 6. Install DPDK and you are ready to go.
1400 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1402 Enable switchdev mode
1403 ---------------------
1405 Switchdev mode is a mode in E-Switch, that binds between representor and VF or SF.
1406 Representor is a port in DPDK that is connected to a VF or SF in such a way
1407 that assuming there are no offload flows, each packet that is sent from the VF or SF
1408 will be received by the corresponding representor. While each packet that is or SF
1409 sent to a representor will be received by the VF or SF.
1410 This is very useful in case of SRIOV mode, where the first packet that is sent
1411 by the VF or SF will be received by the DPDK application which will decide if this
1412 flow should be offloaded to the E-Switch. After offloading the flow packet
1413 that the VF or SF that are matching the flow will not be received any more by
1414 the DPDK application.
1416 1. Enable SRIOV mode::
1418 mlxconfig -d <mst device> set SRIOV_EN=true
1420 2. Configure the max number of VFs::
1422 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1426 mlxfwreset -d <mst device> reset
1428 3. Configure the actual number of VFs::
1430 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1432 4. Unbind the device (can be rebind after the switchdev mode)::
1434 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1436 5. Enbale switchdev mode::
1438 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1440 SubFunction representor support
1441 -------------------------------
1442 SubFunction is a portion of the PCI device, a SF netdev has its own
1443 dedicated queues(txq, rxq). A SF netdev supports E-Switch representation
1444 offload similar to existing PF and VF representors. A SF shares PCI
1445 level resources with other SFs and/or with its parent PCI function.
1447 1. Configure SF feature::
1449 mlxconfig -d <mst device> set PF_BAR2_SIZE=<0/1/2/3> PF_BAR2_ENABLE=1
1451 Value of PF_BAR2_SIZE:
1460 mlxfwreset -d <mst device> reset
1462 3. Enable switchdev mode::
1464 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1468 mlnx-sf -d <PCI_BDF> -a create
1470 5. Probe SF representor::
1472 testpmd> port attach <PCI_BDF>,representor=sf0,dv_flow_en=1
1477 1. Configure aggressive CQE Zipping for maximum performance::
1479 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1481 To set it back to the default CQE Zipping mode use::
1483 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1485 2. In case of virtualization:
1487 - Make sure that hypervisor kernel is 3.16 or newer.
1488 - Configure boot with ``iommu=pt``.
1489 - Use 1G huge pages.
1490 - Make sure to allocate a VM on huge pages.
1491 - Make sure to set CPU pinning.
1493 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1494 for better performance. For VMs, verify that the right CPU
1495 and NUMA node are pinned according to the above. Run::
1499 to identify the NUMA node to which the PCIe adapter is connected.
1501 4. If more than one adapter is used, and root complex capabilities allow
1502 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1503 it is recommended to locate both adapters on the same NUMA node.
1504 This in order to forward packets from one to the other without
1505 NUMA performance penalty.
1507 5. Disable pause frames::
1509 ethtool -A <netdev> rx off tx off
1511 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1512 via the BIOS configuration. Please contact you server provider for more
1513 information about the settings.
1517 On some machines, depends on the machine integrator, it is beneficial
1518 to set the PCI max read request parameter to 1K. This can be
1519 done in the following way:
1521 To query the read request size use::
1523 setpci -s <NIC PCI address> 68.w
1525 If the output is different than 3XXX, set it by::
1527 setpci -s <NIC PCI address> 68.w=3XXX
1529 The XXX can be different on different systems. Make sure to configure
1530 according to the setpci output.
1532 7. To minimize overhead of searching Memory Regions:
1534 - '--socket-mem' is recommended to pin memory by predictable amount.
1535 - Configure per-lcore cache when creating Mempools for packet buffer.
1536 - Refrain from dynamically allocating/freeing memory in run-time.
1541 There are multiple Rx burst functions with different advantages and limitations.
1543 .. table:: Rx burst functions
1545 +-------------------+------------------------+---------+-----------------+------+-------+
1546 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1547 | | | | || comp|| MTU |
1548 +===================+========================+=========+=================+======+=======+
1549 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1550 +-------------------+------------------------+---------+-----------------+------+-------+
1551 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1552 +-------------------+------------------------+---------+-----------------+------+-------+
1553 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1554 | || RxQs >= rxqs_min_mprq | | | | |
1555 +-------------------+------------------------+---------+-----------------+------+-------+
1556 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1557 | || mprq_en=1 | | | | |
1558 | || RxQs >= rxqs_min_mprq | | | | |
1559 +-------------------+------------------------+---------+-----------------+------+-------+
1561 .. _mlx5_offloads_support:
1563 Supported hardware offloads
1564 ---------------------------
1566 .. table:: Minimal SW/HW versions for queue offloads
1568 ============== ===== ===== ========= ===== ========== =============
1569 Offload DPDK Linux rdma-core OFED firmware hardware
1570 ============== ===== ===== ========= ===== ========== =============
1571 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1572 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1573 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1574 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1575 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1576 Tx scheduling 20.08 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1577 Buffer Split 20.11 N/A N/A 5.1-2 16.28.2006 ConnectX-5
1578 ============== ===== ===== ========= ===== ========== =============
1580 .. table:: Minimal SW/HW versions for rte_flow offloads
1582 +-----------------------+-----------------+-----------------+
1583 | Offload | with E-Switch | with NIC |
1584 +=======================+=================+=================+
1585 | Count | | DPDK 19.05 | | DPDK 19.02 |
1586 | | | OFED 4.6 | | OFED 4.6 |
1587 | | | rdma-core 24 | | rdma-core 23 |
1588 | | | ConnectX-5 | | ConnectX-5 |
1589 +-----------------------+-----------------+-----------------+
1590 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1591 | | | OFED 4.6 | | OFED 4.5 |
1592 | | | rdma-core 24 | | rdma-core 23 |
1593 | | | ConnectX-5 | | ConnectX-4 |
1594 +-----------------------+-----------------+-----------------+
1595 | Queue / RSS | | | | DPDK 18.11 |
1596 | | | N/A | | OFED 4.5 |
1597 | | | | | rdma-core 23 |
1598 | | | | | ConnectX-4 |
1599 +-----------------------+-----------------+-----------------+
1600 | Shared action | | | | |
1601 | | | :numref:`sact`| | :numref:`sact`|
1604 +-----------------------+-----------------+-----------------+
1605 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1606 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1607 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1608 | | of_set_vlan_pcp / | | | | |
1609 | | of_set_vlan_vid) | | | | |
1610 +-----------------------+-----------------+-----------------+
1611 | | VLAN | | DPDK 21.05 | | |
1612 | | ingress and / | | OFED 5.3 | | N/A |
1613 | | of_push_vlan / | | ConnectX-6 Dx | | |
1614 +-----------------------+-----------------+-----------------+
1615 | | VLAN | | DPDK 21.05 | | |
1616 | | egress and / | | OFED 5.3 | | N/A |
1617 | | of_pop_vlan / | | ConnectX-6 Dx | | |
1618 +-----------------------+-----------------+-----------------+
1619 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1620 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1621 | | | rdma-core 24 | | rdma-core 23 |
1622 | | | ConnectX-5 | | ConnectX-5 |
1623 +-----------------------+-----------------+-----------------+
1624 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1625 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1626 | | | rdma-core 27 | | rdma-core 27 |
1627 | | | ConnectX-5 | | ConnectX-5 |
1628 +-----------------------+-----------------+-----------------+
1629 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1630 | | | OFED 5.1-2 | | OFED 5.1-2 |
1631 | | | rdma-core 32 | | N/A |
1632 | | | ConnectX-5 | | ConnectX-5 |
1633 +-----------------------+-----------------+-----------------+
1634 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1635 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1636 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1637 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1638 | | set_ipv6_dst / | | | | |
1639 | | set_tp_src / | | | | |
1640 | | set_tp_dst / | | | | |
1641 | | dec_ttl / | | | | |
1642 | | set_ttl / | | | | |
1643 | | set_mac_src / | | | | |
1644 | | set_mac_dst) | | | | |
1645 +-----------------------+-----------------+-----------------+
1646 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1647 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1648 | | | | rdma-core 24 | | rdma-core 24 |
1649 | | | | ConnectX-5 | | ConnectX-5 |
1650 +-----------------------+-----------------+-----------------+
1651 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1652 | | | OFED 4.7-1 | | OFED 4.7-1 |
1653 | | | rdma-core 24 | | N/A |
1654 | | | ConnectX-5 | | ConnectX-5 |
1655 +-----------------------+-----------------+-----------------+
1656 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1657 | | | OFED 4.6 | | OFED 4.5 |
1658 | | | rdma-core 24 | | rdma-core 23 |
1659 | | | ConnectX-5 | | ConnectX-4 |
1660 +-----------------------+-----------------+-----------------+
1661 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1662 | | | OFED 4.7-3 | | OFED 4.7-3 |
1663 | | | rdma-core 26 | | rdma-core 26 |
1664 | | | ConnectX-5 | | ConnectX-5 |
1665 +-----------------------+-----------------+-----------------+
1666 | Port ID | | DPDK 19.05 | | N/A |
1667 | | | OFED 4.7-1 | | N/A |
1668 | | | rdma-core 24 | | N/A |
1669 | | | ConnectX-5 | | N/A |
1670 +-----------------------+-----------------+-----------------+
1671 | Hairpin | | | | DPDK 19.11 |
1672 | | | N/A | | OFED 4.7-3 |
1673 | | | | | rdma-core 26 |
1674 | | | | | ConnectX-5 |
1675 +-----------------------+-----------------+-----------------+
1676 | 2-port Hairpin | | | | DPDK 20.11 |
1677 | | | N/A | | OFED 5.1-2 |
1679 | | | | | ConnectX-5 |
1680 +-----------------------+-----------------+-----------------+
1681 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1682 | | | OFED 4.7-3 | | OFED 4.7-3 |
1683 | | | rdma-core 26 | | rdma-core 26 |
1684 | | | ConnectX-5 | | ConnectX-5 |
1685 +-----------------------+-----------------+-----------------+
1686 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1687 | | | OFED 5.1-2 | | OFED 5.1-2 |
1688 | | | rdma-core 32 | | N/A |
1689 | | | ConnectX-5 | | ConnectX-5 |
1690 +-----------------------+-----------------+-----------------+
1691 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1692 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1693 | | | rdma-core 35 | | rdma-core 35 |
1694 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1695 +-----------------------+-----------------+-----------------+
1696 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1697 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1698 | | | rdma-core 34 | | rdma-core 34 |
1699 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1700 +-----------------------+-----------------+-----------------+
1701 | Modify Field | | DPDK 21.02 | | DPDK 21.02 |
1702 | | | OFED 5.2 | | OFED 5.2 |
1703 | | | rdma-core 35 | | rdma-core 35 |
1704 | | | ConnectX-5 | | ConnectX-5 |
1705 +-----------------------+-----------------+-----------------+
1706 | Connection tracking | | | | DPDK 21.05 |
1707 | | | N/A | | OFED 5.3 |
1708 | | | | | rdma-core 35 |
1709 | | | | | ConnectX-6 Dx |
1710 +-----------------------+-----------------+-----------------+
1712 .. table:: Minimal SW/HW versions for shared action offload
1715 +-----------------------+-----------------+-----------------+
1716 | Shared Action | with E-Switch | with NIC |
1717 +=======================+=================+=================+
1718 | RSS | | | | DPDK 20.11 |
1719 | | | N/A | | OFED 5.2 |
1720 | | | | | rdma-core 33 |
1721 | | | | | ConnectX-5 |
1722 +-----------------------+-----------------+-----------------+
1723 | Age | | DPDK 20.11 | | DPDK 20.11 |
1724 | | | OFED 5.2 | | OFED 5.2 |
1725 | | | rdma-core 32 | | rdma-core 32 |
1726 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1727 +-----------------------+-----------------+-----------------+
1728 | Count | | DPDK 21.05 | | DPDK 21.05 |
1729 | | | OFED 4.6 | | OFED 4.6 |
1730 | | | rdma-core 24 | | rdma-core 23 |
1731 | | | ConnectX-5 | | ConnectX-5 |
1732 +-----------------------+-----------------+-----------------+
1737 MARK and META items are interrelated with datapath - they might move from/to
1738 the applications in mbuf fields. Hence, zero value for these items has the
1739 special meaning - it means "no metadata are provided", not zero values are
1740 treated by applications and PMD as valid ones.
1742 Moreover in the flow engine domain the value zero is acceptable to match and
1743 set, and we should allow to specify zero values as rte_flow parameters for the
1744 META and MARK items and actions. In the same time zero mask has no meaning and
1745 should be rejected on validation stage.
1750 Flows are not cached in the driver.
1751 When stopping a device port, all the flows created on this port from the
1752 application will be flushed automatically in the background.
1753 After stopping the device port, all flows on this port become invalid and
1754 not represented in the system.
1755 All references to these flows held by the application should be discarded
1756 directly but neither destroyed nor flushed.
1758 The application should re-create the flows as required after the port restart.
1763 Compared to librte_net_mlx4 that implements a single RSS configuration per
1764 port, librte_net_mlx5 supports per-protocol RSS configuration.
1766 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1767 command-line parameter to enable additional protocols (UDP and TCP as well
1768 as IP), the following commands must be entered from its CLI to get the same
1769 behavior as librte_net_mlx4::
1772 > port config all rss all
1778 This section demonstrates how to launch **testpmd** with Mellanox
1779 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1781 #. Load the kernel modules::
1783 modprobe -a ib_uverbs mlx5_core mlx5_ib
1785 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1788 /etc/init.d/openibd restart
1792 User space I/O kernel modules (uio and igb_uio) are not used and do
1793 not have to be loaded.
1795 #. Make sure Ethernet interfaces are in working order and linked to kernel
1796 verbs. Related sysfs entries should be present::
1798 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1807 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1810 for intf in eth2 eth3 eth4 eth5;
1812 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1815 sed -n 's,.*/\(.*\),-a \1,p'
1824 #. Request huge pages::
1826 dpdk-hugepages.py --setup 2G
1828 #. Start testpmd with basic parameters::
1830 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
1835 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1836 EAL: probe driver: 15b3:1013 librte_net_mlx5
1837 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1838 PMD: librte_net_mlx5: 1 port(s) detected
1839 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1840 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1841 EAL: probe driver: 15b3:1013 librte_net_mlx5
1842 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1843 PMD: librte_net_mlx5: 1 port(s) detected
1844 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1845 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1846 EAL: probe driver: 15b3:1013 librte_net_mlx5
1847 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1848 PMD: librte_net_mlx5: 1 port(s) detected
1849 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1850 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1851 EAL: probe driver: 15b3:1013 librte_net_mlx5
1852 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1853 PMD: librte_net_mlx5: 1 port(s) detected
1854 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1855 Interactive-mode selected
1856 Configuring Port 0 (socket 0)
1857 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1858 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1859 Port 0: E4:1D:2D:E7:0C:FE
1860 Configuring Port 1 (socket 0)
1861 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1862 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1863 Port 1: E4:1D:2D:E7:0C:FF
1864 Configuring Port 2 (socket 0)
1865 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1866 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1867 Port 2: E4:1D:2D:E7:0C:FA
1868 Configuring Port 3 (socket 0)
1869 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1870 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1871 Port 3: E4:1D:2D:E7:0C:FB
1872 Checking link statuses...
1873 Port 0 Link Up - speed 40000 Mbps - full-duplex
1874 Port 1 Link Up - speed 40000 Mbps - full-duplex
1875 Port 2 Link Up - speed 10000 Mbps - full-duplex
1876 Port 3 Link Up - speed 10000 Mbps - full-duplex
1883 This section demonstrates how to dump flows. Currently, it's possible to dump
1884 all flows with assistance of external tools.
1886 #. 2 ways to get flow raw file:
1888 - Using testpmd CLI:
1890 .. code-block:: console
1893 testpmd> flow dump <port> all <output_file>
1895 testpmd> flow dump <port> rule <rule_id> <output_file>
1897 - call rte_flow_dev_dump api:
1899 .. code-block:: console
1901 rte_flow_dev_dump(port, flow, file, NULL);
1903 #. Dump human-readable flows from raw file:
1905 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1907 .. code-block:: console
1909 mlx_steering_dump.py -f <output_file> -flowptr <flow_ptr>