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.
64 - Rx queue delay drop.
65 - Support for scattered TX frames.
66 - Advanced support for scattered Rx frames with tunable buffer attributes.
67 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
68 - RSS using different combinations of fields: L3 only, L4 only or both,
69 and source only, destination only or both.
70 - Several RSS hash keys, one for each flow type.
71 - Default RSS operation with no hash key specification.
72 - Configurable RETA table.
73 - Link flow control (pause frame).
74 - Support for multiple MAC addresses.
78 - RX CRC stripping configuration.
79 - TX mbuf fast free offload.
80 - Promiscuous mode on PF and VF.
81 - Multicast promiscuous mode on PF and VF.
82 - Hardware checksum offloads.
83 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
85 - Flow API, including :ref:`flow_isolated_mode`.
87 - KVM and VMware ESX SR-IOV modes are supported.
88 - RSS hash result is supported.
89 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
90 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
92 - Statistics query including Basic, Extended and per queue.
94 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
95 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
96 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
97 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
98 - Flow insertion rate of more then million flows per second, when using Direct Rules.
99 - Support for multiple rte_flow groups.
100 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
103 - Multiple-thread flow insertion.
104 - Matching on IPv4 Internet Header Length (IHL).
105 - Matching on GTP extension header with raw encap/decap action.
106 - Matching on Geneve TLV option header with raw encap/decap action.
107 - RSS support in sample action.
108 - E-Switch mirroring and jump.
109 - E-Switch mirroring and modify.
110 - 21844 flow priorities for ingress or egress flow groups greater than 0 and for any transfer
112 - Flow metering, including meter policy API.
113 - Flow meter hierarchy.
114 - Flow integrity offload API.
115 - Connection tracking.
116 - Sub-Function representors.
125 On Windows, the features are limited:
127 - Promiscuous mode is not supported
128 - The following rules are supported:
130 - IPv4/UDP with CVLAN filtering
131 - Unicast MAC filtering
133 - Additional rules are supported from WinOF2 version 2.70:
135 - IPv4/TCP with CVLAN filtering
136 - L4 steering rules for port RSS of UDP, TCP and IP
138 - For secondary process:
140 - Forked secondary process not supported.
141 - External memory unregistered in EAL memseg list cannot be used for DMA
142 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
143 primary process and remapped to the same virtual address in secondary
144 process. If the external memory is registered by primary process but has
145 different virtual address in secondary process, unexpected error may happen.
149 - Counters of received packets and bytes number of devices in same share group are same.
150 - Counters of received packets and bytes number of queues in same group and queue ID are same.
152 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
153 specific VLAN will match for VLAN packets as well:
155 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
156 Meaning, the flow rule::
158 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
160 Will only match vlan packets with vid=3. and the flow rule::
162 flow create 0 ingress pattern eth / ipv4 / end ...
164 Will match any ipv4 packet (VLAN included).
166 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
168 - 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.
171 flow create 0 ingress pattern eth / ipv4 / end ...
173 Will match any ipv4 packet.
176 flow create 0 ingress pattern eth / vlan / end ...
177 flow create 0 ingress pattern eth has_vlan is 1 / end ...
178 flow create 0 ingress pattern eth type is 0x8100 / end ...
180 Will match single-tagged packets only, with any VLAN ID value.
183 flow create 0 ingress pattern eth type is 0x88A8 / end ...
184 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
186 Will match multi-tagged packets only, with any VLAN ID value.
188 - A flow pattern with 2 sequential VLAN items is not supported.
190 - VLAN pop offload command:
192 - Flow rules having a VLAN pop offload command as one of their actions and
193 are lacking a match on VLAN as one of their items are not supported.
194 - The command is not supported on egress traffic in NIC mode.
196 - VLAN push offload is not supported on ingress traffic in NIC mode.
198 - VLAN set PCP offload is not supported on existing headers.
200 - A multi segment packet must have not more segments than reported by dev_infos_get()
201 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
202 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
203 inline settings) to 58.
205 - Match on VXLAN supports the following fields only:
208 - Last reserved 8-bits
210 Last reserved 8-bits matching is only supported When using DV flow
211 engine (``dv_flow_en`` = 1).
212 For ConnectX-5, the UDP destination port must be the standard one (4789).
213 Group zero's behavior may differ which depends on FW.
214 Matching value equals 0 (value & mask) is not supported.
216 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
218 - Match on Geneve header supports the following fields only:
225 - Match on Geneve TLV option is supported on the following fields:
232 Only one Class/Type/Length Geneve TLV option is supported per shared device.
233 Class/Type/Length fields must be specified as well as masks.
234 Class/Type/Length specified masks must be full.
235 Matching Geneve TLV option without specifying data is not supported.
236 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
238 - VF: flow rules created on VF devices can only match traffic targeted at the
239 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
241 - Match on GTP tunnel header item supports the following fields only:
243 - v_pt_rsv_flags: E flag, S flag, PN flag
247 - Match on GTP extension header only for GTP PDU session container (next
248 extension header type = 0x85).
249 - Match on GTP extension header is not supported in group 0.
251 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
252 The flows within group 0 and set metadata action are rejected by hardware.
256 MAC addresses not already present in the bridge table of the associated
257 kernel network device will be added and cleaned up by the PMD when closing
258 the device. In case of ungraceful program termination, some entries may
259 remain present and should be removed manually by other means.
261 - Buffer split offload is supported with regular Rx burst routine only,
262 no MPRQ feature or vectorized code can be engaged.
264 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
265 externally attached to a user-provided mbuf with having RTE_MBUF_F_EXTERNAL in
266 ol_flags. As the mempool for the external buffer is managed by PMD, all the
267 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
268 the external buffers will be freed by PMD and the application which still
269 holds the external buffers may be corrupted.
271 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
272 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
273 supported. Some Rx packets may not have RTE_MBUF_F_RX_RSS_HASH.
275 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
276 and allmulticast mode are both set to off.
277 To receive IPv6 Multicast messages on VM, explicitly set the relevant
278 MAC address using rte_eth_dev_mac_addr_add() API.
280 - To support a mixed traffic pattern (some buffers from local host memory, some
281 buffers from other devices) with high bandwidth, a mbuf flag is used.
283 An application hints the PMD whether or not it should try to inline the
284 given mbuf data buffer. PMD should do the best effort to act upon this request.
286 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
287 registered by application with rte_mbuf_dynflag_register(). This flag is
288 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
289 which is intended to be used by the application.
291 To query the supported specific flags in runtime,
292 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
293 currently (over present hardware and configuration) supported specific flags.
294 The "not inline hint" feature operating flow is the following one:
297 - probe the devices, ports are created
298 - query the port capabilities
299 - if port supporting the feature is found
300 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
301 - application starts the ports
302 - on ``dev_start()`` PMD checks whether the feature flag is registered and
303 enables the feature support in datapath
304 - application might set the registered flag bit in ``ol_flags`` field
305 of mbuf being sent and PMD will handle ones appropriately.
307 - The amount of descriptors in Tx queue may be limited by data inline settings.
308 Inline data require the more descriptor building blocks and overall block
309 amount may exceed the hardware supported limits. The application should
310 reduce the requested Tx size or adjust data inline settings with
311 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
313 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
314 parameter should be specified.
315 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
316 being sent it tries to synchronize the time of packet appearing on
317 the wire with the specified packet timestamp. It the specified one
318 is in the past it should be ignored, if one is in the distant future
319 it should be capped with some reasonable value (in range of seconds).
320 These specific cases ("too late" and "distant future") can be optionally
321 reported via device xstats to assist applications to detect the
322 time-related problems.
324 The timestamp upper "too-distant-future" limit
325 at the moment of invoking the Tx burst routine
326 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
327 Please note, for the testpmd txonly mode,
328 the limit is deduced from the expression::
330 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
332 There is no any packet reordering according timestamps is supposed,
333 neither within packet burst, nor between packets, it is an entirely
334 application responsibility to generate packets and its timestamps
335 in desired order. The timestamps can be put only in the first packet
336 in the burst providing the entire burst scheduling.
338 - E-Switch decapsulation Flow:
340 - can be applied to PF port only.
341 - must specify VF port action (packet redirection from PF to VF).
342 - optionally may specify tunnel inner source and destination MAC addresses.
344 - E-Switch encapsulation Flow:
346 - can be applied to VF ports only.
347 - must specify PF port action (packet redirection from VF to PF).
351 - The input buffer, used as outer header, is not validated.
355 - The decapsulation is always done up to the outermost tunnel detected by the HW.
356 - The input buffer, providing the removal size, is not validated.
357 - The buffer size must match the length of the headers to be removed.
359 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
360 mutually exclusive features which cannot be supported together
361 (see :ref:`mlx5_firmware_config`).
365 - Requires DevX and DV flow to be enabled.
366 - KEEP_CRC offload cannot be supported with LRO.
367 - The first mbuf length, without head-room, must be big enough to include the
369 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
370 it with size limited to max LRO size, not to max RX packet length.
371 - LRO can be used with outer header of TCP packets of the standard format:
372 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
374 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
375 - LRO packet aggregation is performed by HW only for packet size larger than
376 ``lro_min_mss_size``. This value is reported on device start, when debug
381 - ``RTE_ETH_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
382 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
383 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
387 - fast free offload assumes the all mbufs being sent are originated from the
388 same memory pool and there is no any extra references to the mbufs (the
389 reference counter for each mbuf is equal 1 on tx_burst call). The latter
390 means there should be no any externally attached buffers in mbufs. It is
391 an application responsibility to provide the correct mbufs if the fast
392 free offload is engaged. The mlx5 PMD implicitly produces the mbufs with
393 externally attached buffers if MPRQ option is enabled, hence, the fast
394 free offload is neither supported nor advertised if there is MPRQ enabled.
398 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and
399 E-Switch steering domain.
400 - For E-Switch Sampling flow with sample ratio > 1, additional actions are not
401 supported in the sample actions list.
402 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as
403 first action in the E-Switch egress flow if with header modify or
404 encapsulation actions.
405 - For NIC Rx flow, supports ``MARK``, ``COUNT``, ``QUEUE``, ``RSS`` in the
407 - For E-Switch mirroring flow, supports ``RAW ENCAP``, ``Port ID``,
408 ``VXLAN ENCAP``, ``NVGRE ENCAP`` in the sample actions list.
412 - Supports the 'set' operation only for ``RTE_FLOW_ACTION_TYPE_MODIFY_FIELD`` action.
413 - Modification of an arbitrary place in a packet via the special ``RTE_FLOW_FIELD_START`` Field ID is not supported.
414 - Modification of the 802.1Q Tag, VXLAN Network or GENEVE Network ID's is not supported.
415 - Encapsulation levels are not supported, can modify outermost header fields only.
416 - Offsets must be 32-bits aligned, cannot skip past the boundary of a field.
418 - IPv6 header item 'proto' field, indicating the next header protocol, should
419 not be set as extension header.
420 In case the next header is an extension header, it should not be specified in
421 IPv6 header item 'proto' field.
422 The last extension header item 'next header' field can specify the following
423 header protocol type.
427 - 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.
428 - Hairpin in switchdev SR-IOV mode is not supported till now.
432 - All the meter colors with drop action will be counted only by the global drop statistics.
433 - Yellow detection is only supported with ASO metering.
434 - Red color must be with drop action.
435 - Meter statistics are supported only for drop case.
436 - A meter action created with pre-defined policy must be the last action in the flow except single case where the policy actions are:
437 - green: NULL or END.
438 - yellow: NULL or END.
440 - The only supported meter policy actions:
441 - green: QUEUE, RSS, PORT_ID, REPRESENTED_PORT, JUMP, DROP, MARK and SET_TAG.
442 - yellow: QUEUE, RSS, PORT_ID, REPRESENTED_PORT, JUMP, DROP, MARK and SET_TAG.
444 - Policy actions of RSS for green and yellow should have the same configuration except queues.
445 - Policy with RSS/queue action is not supported when ``dv_xmeta_en`` enabled.
446 - meter profile packet mode is supported.
447 - meter profiles of RFC2697, RFC2698 and RFC4115 are supported.
451 - Integrity offload is enabled for **ConnectX-6** family.
452 - Verification bits provided by the hardware are ``l3_ok``, ``ipv4_csum_ok``, ``l4_ok``, ``l4_csum_ok``.
453 - ``level`` value 0 references outer headers.
454 - Multiple integrity items not supported in a single flow rule.
455 - Flow rule items supplied by application must explicitly specify network headers referred by integrity item.
456 For example, if integrity item mask sets ``l4_ok`` or ``l4_csum_ok`` bits, reference to L4 network header,
457 TCP or UDP, must be in the rule pattern as well::
459 flow create 0 ingress pattern integrity level is 0 value mask l3_ok value spec l3_ok / eth / ipv6 / end …
461 flow create 0 ingress pattern integrity level is 0 value mask l4_ok value spec 0 / eth / ipv4 proto is udp / end …
463 - Connection tracking:
465 - Cannot co-exist with ASO meter, ASO age action in a single flow rule.
466 - Flow rules insertion rate and memory consumption need more optimization.
468 - 4M connections maximum.
470 - Multi-thread flow insertion:
472 - In order to achieve best insertion rate, application should manage the flows per lcore.
473 - Better to disable memory reclaim by setting ``reclaim_mem_mode`` to 0 to accelerate the flow object allocation and release with cache.
477 - TXQ affinity subjects to HW hash once enabled.
479 - Bonding under socket direct mode
485 - CQE timestamp field width is limited by hardware to 63 bits, MSB is zero.
486 - In the free-running mode the timestamp counter is reset on power on
487 and 63-bit value provides over 1800 years of uptime till overflow.
488 - In the real-time mode
489 (configurable with ``REAL_TIME_CLOCK_ENABLE`` firmware settings),
490 the timestamp presents the nanoseconds elapsed since 01-Jan-1970,
491 hardware timestamp overflow will happen on 19-Jan-2038
492 (0x80000000 seconds since 01-Jan-1970).
493 - The send scheduling is based on timestamps
494 from the reference "Clock Queue" completions,
495 the scheduled send timestamps should not be specified with non-zero MSB.
500 MLX5 supports various methods to report statistics:
502 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.
504 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.
506 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.
514 The ibverbs libraries can be linked with this PMD in a number of ways,
515 configured by the ``ibverbs_link`` build option:
517 - ``shared`` (default): the PMD depends on some .so files.
519 - ``dlopen``: Split the dependencies glue in a separate library
520 loaded when needed by dlopen.
521 It make dependencies on libibverbs and libmlx4 optional,
522 and has no performance impact.
524 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
525 in the PMD shared library or the executable static binary.
527 Environment variables
528 ~~~~~~~~~~~~~~~~~~~~~
532 A list of directories in which to search for the rdma-core "glue" plug-in,
533 separated by colons or semi-colons.
535 - ``MLX5_SHUT_UP_BF``
537 Configures HW Tx doorbell register as IO-mapped.
539 By default, the HW Tx doorbell is configured as a write-combining register.
540 The register would be flushed to HW usually when the write-combining buffer
541 becomes full, but it depends on CPU design.
543 Except for vectorized Tx burst routines, a write memory barrier is enforced
544 after updating the register so that the update can be immediately visible to
547 When vectorized Tx burst is called, the barrier is set only if the burst size
548 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
549 variable will bring better latency even though the maximum throughput can
552 Run-time configuration
553 ~~~~~~~~~~~~~~~~~~~~~~
555 - librte_net_mlx5 brings kernel network interfaces up during initialization
556 because it is affected by their state. Forcing them down prevents packets
559 - **ethtool** operations on related kernel interfaces also affect the PMD.
564 In order to run as a non-root user,
565 some capabilities must be granted to the application::
567 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
569 Below are the reasons of the need for each capability:
572 When using physical addresses (PA mode), with Linux >= 4.0,
573 for access to ``/proc/self/pagemap``.
576 For device configuration.
579 For raw ethernet queue allocation through kernel driver.
582 For DMA memory pinning.
587 - ``rxq_cqe_comp_en`` parameter [int]
589 A nonzero value enables the compression of CQE on RX side. This feature
590 allows to save PCI bandwidth and improve performance. Enabled by default.
591 Different compression formats are supported in order to achieve the best
592 performance for different traffic patterns. Default format depends on
593 Multi-Packet Rx queue configuration: Hash RSS format is used in case
594 MPRQ is disabled, Checksum format is used in case MPRQ is enabled.
596 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
597 better compression rate in case of RTE Flow Mark traffic.
598 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
599 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
600 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
601 CQE compression format selection requires DevX to be enabled. If there is
602 no DevX enabled/supported the value is reset to 1 by default.
606 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
607 ConnectX-6 Lx, BlueField and BlueField-2.
608 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
609 ConnectX-6 Lx, BlueField and BlueField-2.
611 - ``rxq_pkt_pad_en`` parameter [int]
613 A nonzero value enables padding Rx packet to the size of cacheline on PCI
614 transaction. This feature would waste PCI bandwidth but could improve
615 performance by avoiding partial cacheline write which may cause costly
616 read-modify-copy in memory transaction on some architectures. Disabled by
621 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
622 ConnectX-6 Lx, BlueField and BlueField-2.
623 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
624 ConnectX-6 Lx, BlueField and BlueField-2.
626 - ``delay_drop`` parameter [int]
628 Bitmask value for the Rx queue delay drop attribute. Bit 0 is used for the
629 standard Rx queue and bit 1 is used for the hairpin Rx queue. By default, the
630 delay drop is disabled for all Rx queues. It will be ignored if the port does
631 not support the attribute even if it is enabled explicitly.
633 The packets being received will not be dropped immediately when the WQEs are
634 exhausted in a Rx queue with delay drop enabled.
636 A timeout value is set in the driver to control the waiting time before
637 dropping a packet. Once the timer is expired, the delay drop will be
638 deactivated for all the Rx queues with this feature enable. To re-activeate
639 it, a rearming is needed and it is part of the kernel driver starting from
642 To enable / disable the delay drop rearming, the private flag ``dropless_rq``
643 can be set and queried via ethtool:
645 - ethtool --set-priv-flags <netdev> dropless_rq on (/ off)
646 - ethtool --show-priv-flags <netdev>
648 The configuration flag is global per PF and can only be set on the PF, once
649 it is on, all the VFs', SFs' and representors' Rx queues will share the timer
652 - ``mprq_en`` parameter [int]
654 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
655 configured as Multi-Packet RQ if the total number of Rx queues is
656 ``rxqs_min_mprq`` or more. Disabled by default.
658 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
659 by posting a single large buffer for multiple packets. Instead of posting a
660 buffers per a packet, one large buffer is posted in order to receive multiple
661 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
662 and each stride receives one packet. MPRQ can improve throughput for
663 small-packet traffic.
665 When MPRQ is enabled, MTU can be larger than the size of
666 user-provided mbuf even if RTE_ETH_RX_OFFLOAD_SCATTER isn't enabled. PMD will
667 configure large stride size enough to accommodate MTU as long as
668 device allows. Note that this can waste system memory compared to enabling Rx
669 scatter and multi-segment packet.
671 - ``mprq_log_stride_num`` parameter [int]
673 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
674 strides can reduce PCIe traffic further. If configured value is not in the
675 range of device capability, the default value will be set with a warning
676 message. The default value is 4 which is 16 strides per a buffer, valid only
677 if ``mprq_en`` is set.
679 The size of Rx queue should be bigger than the number of strides.
681 - ``mprq_log_stride_size`` parameter [int]
683 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
684 stride size can save some memory and reduce probability of a depletion of all
685 available strides due to unreleased packets by an application. If configured
686 value is not in the range of device capability, the default value will be set
687 with a warning message. The default value is 11 which is 2048 bytes per a
688 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
689 it is possible for a packet to span across multiple strides. This mode allows
690 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
691 of a packet if Rx scatter is configured) may be required in case there is no
692 space left for a head room at the end of a stride which incurs some
695 - ``mprq_max_memcpy_len`` parameter [int]
697 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
698 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
699 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
700 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
701 A mempool for external buffers will be allocated and managed by PMD. If Rx
702 packet is externally attached, ol_flags field of the mbuf will have
703 RTE_MBUF_F_EXTERNAL and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
704 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
706 - ``rxqs_min_mprq`` parameter [int]
708 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
709 greater or equal to this value. The default value is 12, valid only if
712 - ``txq_inline`` parameter [int]
714 Amount of data to be inlined during TX operations. This parameter is
715 deprecated and converted to the new parameter ``txq_inline_max`` providing
716 partial compatibility.
718 - ``txqs_min_inline`` parameter [int]
720 Enable inline data send only when the number of TX queues is greater or equal
723 This option should be used in combination with ``txq_inline_max`` and
724 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
726 If this option is not specified the default value 16 is used for BlueField
727 and 8 for other platforms
729 The data inlining consumes the CPU cycles, so this option is intended to
730 auto enable inline data if we have enough Tx queues, which means we have
731 enough CPU cores and PCI bandwidth is getting more critical and CPU
732 is not supposed to be bottleneck anymore.
734 The copying data into WQE improves latency and can improve PPS performance
735 when PCI back pressure is detected and may be useful for scenarios involving
736 heavy traffic on many queues.
738 Because additional software logic is necessary to handle this mode, this
739 option should be used with care, as it may lower performance when back
740 pressure is not expected.
742 If inline data are enabled it may affect the maximal size of Tx queue in
743 descriptors because the inline data increase the descriptor size and
744 queue size limits supported by hardware may be exceeded.
746 - ``txq_inline_min`` parameter [int]
748 Minimal amount of data to be inlined into WQE during Tx operations. NICs
749 may require this minimal data amount to operate correctly. The exact value
750 may depend on NIC operation mode, requested offloads, etc. It is strongly
751 recommended to omit this parameter and use the default values. Anyway,
752 applications using this parameter should take into consideration that
753 specifying an inconsistent value may prevent the NIC from sending packets.
755 If ``txq_inline_min`` key is present the specified value (may be aligned
756 by the driver in order not to exceed the limits and provide better descriptor
757 space utilization) will be used by the driver and it is guaranteed that
758 requested amount of data bytes are inlined into the WQE beside other inline
759 settings. This key also may update ``txq_inline_max`` value (default
760 or specified explicitly in devargs) to reserve the space for inline data.
762 If ``txq_inline_min`` key is not present, the value may be queried by the
763 driver from the NIC via DevX if this feature is available. If there is no DevX
764 enabled/supported the value 18 (supposing L2 header including VLAN) is set
765 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
766 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
769 For ConnectX-4 NIC, driver does not allow specifying value below 18
770 (minimal L2 header, including VLAN), error will be raised.
772 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
773 it is not recommended and may prevent NIC from sending packets over
776 For ConnectX-4 and ConnectX-4 Lx NICs, automatically configured value
777 is insufficient for some traffic, because they require at least all L2 headers
778 to be inlined. For example, Q-in-Q adds 4 bytes to default 18 bytes
779 of Ethernet and VLAN, thus ``txq_inline_min`` must be set to 22.
780 MPLS would add 4 bytes per label. Final value must account for all possible
781 L2 encapsulation headers used in particular environment.
783 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
784 Multi-Packet Write), because last one does not support partial packet inlining.
785 This is not very critical due to minimal data inlining is mostly required
786 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
788 - ``txq_inline_max`` parameter [int]
790 Specifies the maximal packet length to be completely inlined into WQE
791 Ethernet Segment for ordinary SEND method. If packet is larger than specified
792 value, the packet data won't be copied by the driver at all, data buffer
793 is addressed with a pointer. If packet length is less or equal all packet
794 data will be copied into WQE. This may improve PCI bandwidth utilization for
795 short packets significantly but requires the extra CPU cycles.
797 The data inline feature is controlled by number of Tx queues, if number of Tx
798 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
799 is engaged, if there are not enough Tx queues (which means not enough CPU cores
800 and CPU resources are scarce), data inline is not performed by the driver.
801 Assigning ``txqs_min_inline`` with zero always enables the data inline.
803 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
804 by the driver in order not to exceed the limit (930 bytes) and to provide better
805 WQE space filling without gaps, the adjustment is reflected in the debug log.
806 Also, the default value (290) may be decreased in run-time if the large transmit
807 queue size is requested and hardware does not support enough descriptor
808 amount, in this case warning is emitted. If ``txq_inline_max`` key is
809 specified and requested inline settings can not be satisfied then error
812 - ``txq_inline_mpw`` parameter [int]
814 Specifies the maximal packet length to be completely inlined into WQE for
815 Enhanced MPW method. If packet is large the specified value, the packet data
816 won't be copied, and data buffer is addressed with pointer. If packet length
817 is less or equal, all packet data will be copied into WQE. This may improve PCI
818 bandwidth utilization for short packets significantly but requires the extra
821 The data inline feature is controlled by number of TX queues, if number of Tx
822 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
823 is engaged, if there are not enough Tx queues (which means not enough CPU cores
824 and CPU resources are scarce), data inline is not performed by the driver.
825 Assigning ``txqs_min_inline`` with zero always enables the data inline.
827 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
828 by the driver in order not to exceed the limit (930 bytes) and to provide better
829 WQE space filling without gaps, the adjustment is reflected in the debug log.
830 Due to multiple packets may be included to the same WQE with Enhanced Multi
831 Packet Write Method and overall WQE size is limited it is not recommended to
832 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
833 may be decreased in run-time if the large transmit queue size is requested
834 and hardware does not support enough descriptor amount, in this case warning
835 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
836 settings can not be satisfied then error will be raised.
838 - ``txqs_max_vec`` parameter [int]
840 Enable vectorized Tx only when the number of TX queues is less than or
841 equal to this value. This parameter is deprecated and ignored, kept
842 for compatibility issue to not prevent driver from probing.
844 - ``txq_mpw_hdr_dseg_en`` parameter [int]
846 A nonzero value enables including two pointers in the first block of TX
847 descriptor. The parameter is deprecated and ignored, kept for compatibility
850 - ``txq_max_inline_len`` parameter [int]
852 Maximum size of packet to be inlined. This limits the size of packet to
853 be inlined. If the size of a packet is larger than configured value, the
854 packet isn't inlined even though there's enough space remained in the
855 descriptor. Instead, the packet is included with pointer. This parameter
856 is deprecated and converted directly to ``txq_inline_mpw`` providing full
857 compatibility. Valid only if eMPW feature is engaged.
859 - ``txq_mpw_en`` parameter [int]
861 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
862 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
863 eMPW allows the Tx burst function to pack up multiple packets
864 in a single descriptor session in order to save PCI bandwidth
865 and improve performance at the cost of a slightly higher CPU usage.
866 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
867 Tx burst function copies entire packet data on to Tx descriptor
868 instead of including pointer of packet.
870 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
871 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
872 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
873 option or reported by the NIC, the eMPW feature is disengaged.
875 - ``tx_db_nc`` parameter [int]
877 The rdma core library can map doorbell register in two ways, depending on the
878 environment variable "MLX5_SHUT_UP_BF":
880 - As regular cached memory (usually with write combining attribute), if the
881 variable is either missing or set to zero.
882 - As non-cached memory, if the variable is present and set to not "0" value.
884 The type of mapping may slightly affect the Tx performance, the optimal choice
885 is strongly relied on the host architecture and should be deduced practically.
887 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
888 memory (with write combining), the PMD will perform the extra write memory barrier
889 after writing to doorbell, it might increase the needed CPU clocks per packet
890 to send, but latency might be improved.
892 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
893 cached memory, the PMD will not perform the extra write memory barrier
894 after writing to doorbell, on some architectures it might improve the
897 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
898 memory, the PMD will use heuristics to decide whether write memory barrier
899 should be performed. For bursts with size multiple of recommended one (64 pkts)
900 it is supposed the next burst is coming and no need to issue the extra memory
901 barrier (it is supposed to be issued in the next coming burst, at least after
902 descriptor writing). It might increase latency (on some hosts till next
903 packets transmit) and should be used with care.
905 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
906 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
907 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
909 - ``tx_pp`` parameter [int]
911 If a nonzero value is specified the driver creates all necessary internal
912 objects to provide accurate packet send scheduling on mbuf timestamps.
913 The positive value specifies the scheduling granularity in nanoseconds,
914 the packet send will be accurate up to specified digits. The allowed range is
915 from 500 to 1 million of nanoseconds. The negative value specifies the module
916 of granularity and engages the special test mode the check the schedule rate.
917 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
920 - ``tx_skew`` parameter [int]
922 The parameter adjusts the send packet scheduling on timestamps and represents
923 the average delay between beginning of the transmitting descriptor processing
924 by the hardware and appearance of actual packet data on the wire. The value
925 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
926 specified. The default value is zero.
928 - ``tx_vec_en`` parameter [int]
930 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
931 ConnectX-6 Lx, BlueField and BlueField-2 NICs
932 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
933 The parameter is deprecated and ignored.
935 - ``rx_vec_en`` parameter [int]
937 A nonzero value enables Rx vector if the port is not configured in
938 multi-segment otherwise this parameter is ignored.
942 - ``vf_nl_en`` parameter [int]
944 A nonzero value enables Netlink requests from the VF to add/remove MAC
945 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
946 Otherwise the relevant configuration must be run with Linux iproute2 tools.
947 This is a prerequisite to receive this kind of traffic.
949 Enabled by default, valid only on VF devices ignored otherwise.
951 - ``l3_vxlan_en`` parameter [int]
953 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
954 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
955 parameter. This is a prerequisite to receive this kind of traffic.
959 - ``dv_xmeta_en`` parameter [int]
961 A nonzero value enables extensive flow metadata support if device is
962 capable and driver supports it. This can enable extensive support of
963 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
964 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
966 There are some possible configurations, depending on parameter value:
968 - 0, this is default value, defines the legacy mode, the ``MARK`` and
969 ``META`` related actions and items operate only within NIC Tx and
970 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
971 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
972 item is 32 bits wide and match supported on egress only.
974 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
975 related actions and items operate within all supported steering domains,
976 including FDB, ``MARK`` and ``META`` information may cross the domain
977 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
978 depends on kernel and firmware configurations and might be 0, 16 or
979 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
980 compatibility, the actual width of data transferred to the FDB domain
981 depends on kernel configuration and may be vary. The actual supported
982 width can be retrieved in runtime by series of rte_flow_validate()
985 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
986 related actions and items operate within all supported steering domains,
987 including FDB, ``MARK`` and ``META`` information may cross the domain
988 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
989 depends on kernel and firmware configurations and might be 0, 16 or
990 24 bits. The actual supported width can be retrieved in runtime by
991 series of rte_flow_validate() trials.
993 - 3, this engages tunnel offload mode. In E-Switch configuration, that
994 mode implicitly activates ``dv_xmeta_en=1``.
996 +------+-----------+-----------+-------------+-------------+
997 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
998 +======+===========+===========+=============+=============+
999 | 0 | 24 bits | 32 bits | 32 bits | no |
1000 +------+-----------+-----------+-------------+-------------+
1001 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
1002 +------+-----------+-----------+-------------+-------------+
1003 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
1004 +------+-----------+-----------+-------------+-------------+
1006 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
1007 ignored and the device is configured to operate in legacy mode (0).
1009 Disabled by default (set to 0).
1011 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
1012 of the extensive metadata features. The legacy Verbs supports FLAG and
1013 MARK metadata actions over NIC Rx steering domain only.
1015 Setting META value to zero in flow action means there is no item provided
1016 and receiving datapath will not report in mbufs the metadata are present.
1017 Setting MARK value to zero in flow action means the zero FDIR ID value
1018 will be reported on packet receiving.
1020 For the MARK action the last 16 values in the full range are reserved for
1021 internal PMD purposes (to emulate FLAG action). The valid range for the
1022 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
1023 for the 24-bit mode, the flows with the MARK action value outside
1024 the specified range will be rejected.
1026 - ``dv_flow_en`` parameter [int]
1028 A nonzero value enables the DV flow steering assuming it is supported
1029 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
1031 Enabled by default if supported.
1033 - ``dv_esw_en`` parameter [int]
1035 A nonzero value enables E-Switch using Direct Rules.
1037 Enabled by default if supported.
1039 - ``lacp_by_user`` parameter [int]
1041 A nonzero value enables the control of LACP traffic by the user application.
1042 When a bond exists in the driver, by default it should be managed by the
1043 kernel and therefore LACP traffic should be steered to the kernel.
1044 If this devarg is set to 1 it will allow the user to manage the bond by
1045 itself and not steer LACP traffic to the kernel.
1047 Disabled by default (set to 0).
1049 - ``mr_ext_memseg_en`` parameter [int]
1051 A nonzero value enables extending memseg when registering DMA memory. If
1052 enabled, the number of entries in MR (Memory Region) lookup table on datapath
1053 is minimized and it benefits performance. On the other hand, it worsens memory
1054 utilization because registered memory is pinned by kernel driver. Even if a
1055 page in the extended chunk is freed, that doesn't become reusable until the
1056 entire memory is freed.
1060 - ``mr_mempool_reg_en`` parameter [int]
1062 A nonzero value enables implicit registration of DMA memory of all mempools
1063 except those having ``RTE_MEMPOOL_F_NON_IO``. This flag is set automatically
1064 for mempools populated with non-contiguous objects or those without IOVA.
1065 The effect is that when a packet from a mempool is transmitted,
1066 its memory is already registered for DMA in the PMD and no registration
1067 will happen on the data path. The tradeoff is extra work on the creation
1068 of each mempool and increased HW resource use if some mempools
1069 are not used with MLX5 devices.
1073 - ``representor`` parameter [list]
1075 This parameter can be used to instantiate DPDK Ethernet devices from
1076 existing port (PF, VF or SF) representors configured on the device.
1078 It is a standard parameter whose format is described in
1079 :ref:`ethernet_device_standard_device_arguments`.
1081 For instance, to probe VF port representors 0 through 2::
1083 <PCI_BDF>,representor=vf[0-2]
1085 To probe SF port representors 0 through 2::
1087 <PCI_BDF>,representor=sf[0-2]
1089 To probe VF port representors 0 through 2 on both PFs of bonding device::
1091 <Primary_PCI_BDF>,representor=pf[0,1]vf[0-2]
1093 - ``max_dump_files_num`` parameter [int]
1095 The maximum number of files per PMD entity that may be created for debug information.
1096 The files will be created in /var/log directory or in current directory.
1098 set to 128 by default.
1100 - ``lro_timeout_usec`` parameter [int]
1102 The maximum allowed duration of an LRO session, in micro-seconds.
1103 PMD will set the nearest value supported by HW, which is not bigger than
1104 the input ``lro_timeout_usec`` value.
1105 If this parameter is not specified, by default PMD will set
1106 the smallest value supported by HW.
1108 - ``hp_buf_log_sz`` parameter [int]
1110 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
1111 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
1112 The capacity of the value is specified by the firmware and the initialization
1113 will get a failure if it is out of scope.
1114 The range of the value is from 11 to 19 right now, and the supported frame
1115 size of a single packet for hairpin is from 512B to 128KB. It might change if
1116 different firmware release is being used. By using a small value, it could
1117 reduce memory consumption but not work with a large frame. If the value is
1118 too large, the memory consumption will be high and some potential performance
1119 degradation will be introduced.
1120 By default, the PMD will set this value to 16, which means that 9KB jumbo
1121 frames will be supported.
1123 - ``reclaim_mem_mode`` parameter [int]
1125 Cache some resources in flow destroy will help flow recreation more efficient.
1126 While some systems may require the all the resources can be reclaimed after
1128 The parameter ``reclaim_mem_mode`` provides the option for user to configure
1129 if the resource cache is needed or not.
1131 There are three options to choose:
1133 - 0. It means the flow resources will be cached as usual. The resources will
1134 be cached, helpful with flow insertion rate.
1136 - 1. It will only enable the DPDK PMD level resources reclaim.
1138 - 2. Both DPDK PMD level and rdma-core low level will be configured as
1141 By default, the PMD will set this value to 0.
1143 - ``sys_mem_en`` parameter [int]
1145 A non-zero value enables the PMD memory management allocating memory
1146 from system by default, without explicit rte memory flag.
1148 By default, the PMD will set this value to 0.
1150 - ``decap_en`` parameter [int]
1152 Some devices do not support FCS (frame checksum) scattering for
1153 tunnel-decapsulated packets.
1154 If set to 0, this option forces the FCS feature and rejects tunnel
1155 decapsulation in the flow engine for such devices.
1157 By default, the PMD will set this value to 1.
1159 - ``allow_duplicate_pattern`` parameter [int]
1161 There are two options to choose:
1163 - 0. Prevent insertion of rules with the same pattern items on non-root table.
1164 In this case, only the first rule is inserted and the following rules are
1165 rejected and error code EEXIST is returned.
1167 - 1. Allow insertion of rules with the same pattern items.
1168 In this case, all rules are inserted but only the first rule takes effect,
1169 the next rule takes effect only if the previous rules are deleted.
1171 By default, the PMD will set this value to 1.
1173 .. _mlx5_firmware_config:
1175 Firmware configuration
1176 ~~~~~~~~~~~~~~~~~~~~~~
1178 Firmware features can be configured as key/value pairs.
1180 The command to set a value is::
1182 mlxconfig -d <device> set <key>=<value>
1184 The command to query a value is::
1186 mlxconfig -d <device> query | grep <key>
1188 The device name for the command ``mlxconfig`` can be either the PCI address,
1189 or the mst device name found with::
1193 Below are some firmware configurations listed.
1199 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1205 - maximum number of SR-IOV virtual functions::
1209 - enable DevX (required by Direct Rules and other features)::
1213 - aggressive CQE zipping::
1217 - L3 VXLAN and VXLAN-GPE destination UDP port::
1220 IP_OVER_VXLAN_PORT=<udp dport>
1222 - enable VXLAN-GPE tunnel flow matching::
1224 FLEX_PARSER_PROFILE_ENABLE=0
1226 FLEX_PARSER_PROFILE_ENABLE=2
1228 - enable IP-in-IP tunnel flow matching::
1230 FLEX_PARSER_PROFILE_ENABLE=0
1232 - enable MPLS flow matching::
1234 FLEX_PARSER_PROFILE_ENABLE=1
1236 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1238 FLEX_PARSER_PROFILE_ENABLE=2
1240 - enable Geneve flow matching::
1242 FLEX_PARSER_PROFILE_ENABLE=0
1244 FLEX_PARSER_PROFILE_ENABLE=1
1246 - enable Geneve TLV option flow matching::
1248 FLEX_PARSER_PROFILE_ENABLE=0
1250 - enable GTP flow matching::
1252 FLEX_PARSER_PROFILE_ENABLE=3
1254 - enable eCPRI flow matching::
1256 FLEX_PARSER_PROFILE_ENABLE=4
1259 - enable realtime timestamp format::
1261 REAL_TIME_CLOCK_ENABLE=1
1266 This driver relies on external libraries and kernel drivers for resources
1267 allocations and initialization. The following dependencies are not part of
1268 DPDK and must be installed separately:
1272 User space Verbs framework used by librte_net_mlx5. This library provides
1273 a generic interface between the kernel and low-level user space drivers
1276 It allows slow and privileged operations (context initialization, hardware
1277 resources allocations) to be managed by the kernel and fast operations to
1278 never leave user space.
1282 Low-level user space driver library for Mellanox
1283 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1286 This library basically implements send/receive calls to the hardware
1289 - **Kernel modules**
1291 They provide the kernel-side Verbs API and low level device drivers that
1292 manage actual hardware initialization and resources sharing with user
1295 Unlike most other PMDs, these modules must remain loaded and bound to
1298 - mlx5_core: hardware driver managing Mellanox
1299 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1301 - mlx5_ib: InifiniBand device driver.
1302 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1304 - **Firmware update**
1306 Mellanox OFED/EN releases include firmware updates for
1307 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1309 Because each release provides new features, these updates must be applied to
1310 match the kernel modules and libraries they come with.
1314 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1320 Either RDMA Core library with a recent enough Linux kernel release
1321 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1324 RDMA Core with Linux Kernel
1325 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1327 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1328 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1329 (see `RDMA Core installation documentation`_)
1330 - When building for i686 use:
1332 - rdma-core version 18.0 or above built with 32bit support.
1333 - Kernel version 4.14.41 or above.
1335 - Starting with rdma-core v21, static libraries can be built::
1338 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1341 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1342 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1348 - Mellanox OFED version: **4.5** and above /
1349 Mellanox EN version: **4.5** and above
1352 - ConnectX-4: **12.21.1000** and above.
1353 - ConnectX-4 Lx: **14.21.1000** and above.
1354 - ConnectX-5: **16.21.1000** and above.
1355 - ConnectX-5 Ex: **16.21.1000** and above.
1356 - ConnectX-6: **20.27.0090** and above.
1357 - ConnectX-6 Dx: **22.27.0090** and above.
1358 - BlueField: **18.25.1010** and above.
1360 While these libraries and kernel modules are available on OpenFabrics
1361 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1362 managers on most distributions, this PMD requires Ethernet extensions that
1363 may not be supported at the moment (this is a work in progress).
1366 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1368 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1369 include the necessary support and should be used in the meantime. For DPDK,
1370 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1371 required from that distribution.
1375 Several versions of Mellanox OFED/EN are available. Installing the version
1376 this DPDK release was developed and tested against is strongly
1377 recommended. Please check the `linux prerequisites`_.
1379 Windows Prerequisites
1380 ---------------------
1382 This driver relies on external libraries and kernel drivers for resources
1383 allocations and initialization. The dependencies in the following sub-sections
1384 are not part of DPDK, and must be installed separately.
1386 Compilation Prerequisites
1387 ~~~~~~~~~~~~~~~~~~~~~~~~~
1389 DevX SDK installation
1390 ^^^^^^^^^^^^^^^^^^^^^
1392 The DevX SDK must be installed on the machine building the Windows PMD.
1393 Additional information can be found at
1394 `How to Integrate Windows DevX in Your Development Environment
1395 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1397 Runtime Prerequisites
1398 ~~~~~~~~~~~~~~~~~~~~~
1400 WinOF2 version 2.60 or higher must be installed on the machine.
1405 The driver can be downloaded from the following site:
1407 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1412 DevX for Windows must be enabled in the Windows registry.
1413 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1414 Additional information can be found in the WinOF2 user manual.
1419 The following Mellanox device families are supported by the same mlx5 driver:
1431 Below are detailed device names:
1433 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1434 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1435 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1436 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1437 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1438 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1439 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1440 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1441 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1442 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1443 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1444 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1445 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1446 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1447 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1448 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1449 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1450 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1451 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1452 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1453 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1454 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1455 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1456 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1457 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1458 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1460 Quick Start Guide on OFED/EN
1461 ----------------------------
1463 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1466 2. Install the required libraries and kernel modules either by installing
1467 only the required set, or by installing the entire Mellanox OFED/EN::
1469 ./mlnxofedinstall --upstream-libs --dpdk
1471 3. Verify the firmware is the correct one::
1475 4. Verify all ports links are set to Ethernet::
1477 mlxconfig -d <mst device> query | grep LINK_TYPE
1481 Link types may have to be configured to Ethernet::
1483 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1485 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1487 For hypervisors, verify SR-IOV is enabled on the NIC::
1489 mlxconfig -d <mst device> query | grep SRIOV_EN
1492 If needed, configure SR-IOV::
1494 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1495 mlxfwreset -d <mst device> reset
1497 5. Restart the driver::
1499 /etc/init.d/openibd restart
1503 service openibd restart
1505 If link type was changed, firmware must be reset as well::
1507 mlxfwreset -d <mst device> reset
1509 For hypervisors, after reset write the sysfs number of virtual functions
1512 To dynamically instantiate a given number of virtual functions (VFs)::
1514 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1516 6. Install DPDK and you are ready to go.
1517 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1519 Enable switchdev mode
1520 ---------------------
1522 Switchdev mode is a mode in E-Switch, that binds between representor and VF or SF.
1523 Representor is a port in DPDK that is connected to a VF or SF in such a way
1524 that assuming there are no offload flows, each packet that is sent from the VF or SF
1525 will be received by the corresponding representor. While each packet that is or SF
1526 sent to a representor will be received by the VF or SF.
1527 This is very useful in case of SRIOV mode, where the first packet that is sent
1528 by the VF or SF will be received by the DPDK application which will decide if this
1529 flow should be offloaded to the E-Switch. After offloading the flow packet
1530 that the VF or SF that are matching the flow will not be received any more by
1531 the DPDK application.
1533 1. Enable SRIOV mode::
1535 mlxconfig -d <mst device> set SRIOV_EN=true
1537 2. Configure the max number of VFs::
1539 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1543 mlxfwreset -d <mst device> reset
1545 3. Configure the actual number of VFs::
1547 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1549 4. Unbind the device (can be rebind after the switchdev mode)::
1551 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1553 5. Enable switchdev mode::
1555 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1557 Sub-Function support
1558 --------------------
1560 Sub-Function is a portion of the PCI device, a SF netdev has its own
1561 dedicated queues (txq, rxq).
1562 A SF shares PCI level resources with other SFs and/or with its parent PCI function.
1566 OFED version >= 5.4-0.3.3.0
1568 1. Configure SF feature::
1570 # Run mlxconfig on both PFs on host and ECPFs on BlueField.
1571 mlxconfig -d <mst device> set PER_PF_NUM_SF=1 PF_TOTAL_SF=252 PF_SF_BAR_SIZE=12
1573 2. Enable switchdev mode::
1575 mlxdevm dev eswitch set pci/<DBDF> mode switchdev
1579 mlxdevm port add pci/<DBDF> flavour pcisf pfnum 0 sfnum <sfnum>
1581 Get SFID from output: pci/<DBDF>/<SFID>
1583 4. Modify MAC address::
1585 mlxdevm port function set pci/<DBDF>/<SFID> hw_addr <MAC>
1587 5. Activate SF port::
1589 mlxdevm port function set pci/<DBDF>/<ID> state active
1591 6. Devargs to probe SF device::
1593 auxiliary:mlx5_core.sf.<num>,dv_flow_en=1
1595 Sub-Function representor support
1596 --------------------------------
1598 A SF netdev supports E-Switch representation offload
1599 similar to PF and VF representors.
1600 Use <sfnum> to probe SF representor::
1602 testpmd> port attach <PCI_BDF>,representor=sf<sfnum>,dv_flow_en=1
1607 1. Configure aggressive CQE Zipping for maximum performance::
1609 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1611 To set it back to the default CQE Zipping mode use::
1613 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1615 2. In case of virtualization:
1617 - Make sure that hypervisor kernel is 3.16 or newer.
1618 - Configure boot with ``iommu=pt``.
1619 - Use 1G huge pages.
1620 - Make sure to allocate a VM on huge pages.
1621 - Make sure to set CPU pinning.
1623 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1624 for better performance. For VMs, verify that the right CPU
1625 and NUMA node are pinned according to the above. Run::
1627 lstopo-no-graphics --merge
1629 to identify the NUMA node to which the PCIe adapter is connected.
1631 4. If more than one adapter is used, and root complex capabilities allow
1632 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1633 it is recommended to locate both adapters on the same NUMA node.
1634 This in order to forward packets from one to the other without
1635 NUMA performance penalty.
1637 5. Disable pause frames::
1639 ethtool -A <netdev> rx off tx off
1641 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1642 via the BIOS configuration. Please contact you server provider for more
1643 information about the settings.
1647 On some machines, depends on the machine integrator, it is beneficial
1648 to set the PCI max read request parameter to 1K. This can be
1649 done in the following way:
1651 To query the read request size use::
1653 setpci -s <NIC PCI address> 68.w
1655 If the output is different than 3XXX, set it by::
1657 setpci -s <NIC PCI address> 68.w=3XXX
1659 The XXX can be different on different systems. Make sure to configure
1660 according to the setpci output.
1662 7. To minimize overhead of searching Memory Regions:
1664 - '--socket-mem' is recommended to pin memory by predictable amount.
1665 - Configure per-lcore cache when creating Mempools for packet buffer.
1666 - Refrain from dynamically allocating/freeing memory in run-time.
1671 There are multiple Rx burst functions with different advantages and limitations.
1673 .. table:: Rx burst functions
1675 +-------------------+------------------------+---------+-----------------+------+-------+
1676 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1677 | | | | || comp|| MTU |
1678 +===================+========================+=========+=================+======+=======+
1679 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1680 +-------------------+------------------------+---------+-----------------+------+-------+
1681 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1682 +-------------------+------------------------+---------+-----------------+------+-------+
1683 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1684 | || RxQs >= rxqs_min_mprq | | | | |
1685 +-------------------+------------------------+---------+-----------------+------+-------+
1686 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1687 | || mprq_en=1 | | | | |
1688 | || RxQs >= rxqs_min_mprq | | | | |
1689 +-------------------+------------------------+---------+-----------------+------+-------+
1691 .. _mlx5_offloads_support:
1693 Supported hardware offloads
1694 ---------------------------
1696 .. table:: Minimal SW/HW versions for queue offloads
1698 ============== ===== ===== ========= ===== ========== =============
1699 Offload DPDK Linux rdma-core OFED firmware hardware
1700 ============== ===== ===== ========= ===== ========== =============
1701 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1702 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1703 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1704 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1705 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1706 Tx scheduling 20.08 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1707 Buffer Split 20.11 N/A N/A 5.1-2 16.28.2006 ConnectX-5
1708 ============== ===== ===== ========= ===== ========== =============
1710 .. table:: Minimal SW/HW versions for rte_flow offloads
1712 +-----------------------+-----------------+-----------------+
1713 | Offload | with E-Switch | with NIC |
1714 +=======================+=================+=================+
1715 | Count | | DPDK 19.05 | | DPDK 19.02 |
1716 | | | OFED 4.6 | | OFED 4.6 |
1717 | | | rdma-core 24 | | rdma-core 23 |
1718 | | | ConnectX-5 | | ConnectX-5 |
1719 +-----------------------+-----------------+-----------------+
1720 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1721 | | | OFED 4.6 | | OFED 4.5 |
1722 | | | rdma-core 24 | | rdma-core 23 |
1723 | | | ConnectX-5 | | ConnectX-4 |
1724 +-----------------------+-----------------+-----------------+
1725 | Queue / RSS | | | | DPDK 18.11 |
1726 | | | N/A | | OFED 4.5 |
1727 | | | | | rdma-core 23 |
1728 | | | | | ConnectX-4 |
1729 +-----------------------+-----------------+-----------------+
1730 | Shared action | | | | |
1731 | | | :numref:`sact`| | :numref:`sact`|
1734 +-----------------------+-----------------+-----------------+
1735 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1736 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1737 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1738 | | of_set_vlan_pcp / | | | | |
1739 | | of_set_vlan_vid) | | | | |
1740 +-----------------------+-----------------+-----------------+
1741 | | VLAN | | DPDK 21.05 | | |
1742 | | ingress and / | | OFED 5.3 | | N/A |
1743 | | of_push_vlan / | | ConnectX-6 Dx | | |
1744 +-----------------------+-----------------+-----------------+
1745 | | VLAN | | DPDK 21.05 | | |
1746 | | egress and / | | OFED 5.3 | | N/A |
1747 | | of_pop_vlan / | | ConnectX-6 Dx | | |
1748 +-----------------------+-----------------+-----------------+
1749 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1750 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1751 | | | rdma-core 24 | | rdma-core 23 |
1752 | | | ConnectX-5 | | ConnectX-5 |
1753 +-----------------------+-----------------+-----------------+
1754 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1755 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1756 | | | rdma-core 27 | | rdma-core 27 |
1757 | | | ConnectX-5 | | ConnectX-5 |
1758 +-----------------------+-----------------+-----------------+
1759 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1760 | | | OFED 5.1-2 | | OFED 5.1-2 |
1761 | | | rdma-core 32 | | N/A |
1762 | | | ConnectX-5 | | ConnectX-5 |
1763 +-----------------------+-----------------+-----------------+
1764 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1765 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1766 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1767 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1768 | | set_ipv6_dst / | | | | |
1769 | | set_tp_src / | | | | |
1770 | | set_tp_dst / | | | | |
1771 | | dec_ttl / | | | | |
1772 | | set_ttl / | | | | |
1773 | | set_mac_src / | | | | |
1774 | | set_mac_dst) | | | | |
1775 +-----------------------+-----------------+-----------------+
1776 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1777 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1778 | | | | rdma-core 24 | | rdma-core 24 |
1779 | | | | ConnectX-5 | | ConnectX-5 |
1780 +-----------------------+-----------------+-----------------+
1781 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1782 | | | OFED 4.7-1 | | OFED 4.7-1 |
1783 | | | rdma-core 24 | | N/A |
1784 | | | ConnectX-5 | | ConnectX-5 |
1785 +-----------------------+-----------------+-----------------+
1786 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1787 | | | OFED 4.6 | | OFED 4.5 |
1788 | | | rdma-core 24 | | rdma-core 23 |
1789 | | | ConnectX-5 | | ConnectX-4 |
1790 +-----------------------+-----------------+-----------------+
1791 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1792 | | | OFED 4.7-3 | | OFED 4.7-3 |
1793 | | | rdma-core 26 | | rdma-core 26 |
1794 | | | ConnectX-5 | | ConnectX-5 |
1795 +-----------------------+-----------------+-----------------+
1796 | Port ID | | DPDK 19.05 | | N/A |
1797 | | | OFED 4.7-1 | | N/A |
1798 | | | rdma-core 24 | | N/A |
1799 | | | ConnectX-5 | | N/A |
1800 +-----------------------+-----------------+-----------------+
1801 | Hairpin | | | | DPDK 19.11 |
1802 | | | N/A | | OFED 4.7-3 |
1803 | | | | | rdma-core 26 |
1804 | | | | | ConnectX-5 |
1805 +-----------------------+-----------------+-----------------+
1806 | 2-port Hairpin | | | | DPDK 20.11 |
1807 | | | N/A | | OFED 5.1-2 |
1809 | | | | | ConnectX-5 |
1810 +-----------------------+-----------------+-----------------+
1811 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1812 | | | OFED 4.7-3 | | OFED 4.7-3 |
1813 | | | rdma-core 26 | | rdma-core 26 |
1814 | | | ConnectX-5 | | ConnectX-5 |
1815 +-----------------------+-----------------+-----------------+
1816 | ASO Metering | | DPDK 21.05 | | DPDK 21.05 |
1817 | | | OFED 5.3 | | OFED 5.3 |
1818 | | | rdma-core 33 | | rdma-core 33 |
1819 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1820 +-----------------------+-----------------+-----------------+
1821 | Metering Hierarchy | | DPDK 21.08 | | DPDK 21.08 |
1822 | | | OFED 5.3 | | OFED 5.3 |
1824 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1825 +-----------------------+-----------------+-----------------+
1826 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1827 | | | OFED 5.1-2 | | OFED 5.1-2 |
1828 | | | rdma-core 32 | | N/A |
1829 | | | ConnectX-5 | | ConnectX-5 |
1830 +-----------------------+-----------------+-----------------+
1831 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1832 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1833 | | | rdma-core 35 | | rdma-core 35 |
1834 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1835 +-----------------------+-----------------+-----------------+
1836 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1837 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1838 | | | rdma-core 34 | | rdma-core 34 |
1839 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1840 +-----------------------+-----------------+-----------------+
1841 | Modify Field | | DPDK 21.02 | | DPDK 21.02 |
1842 | | | OFED 5.2 | | OFED 5.2 |
1843 | | | rdma-core 35 | | rdma-core 35 |
1844 | | | ConnectX-5 | | ConnectX-5 |
1845 +-----------------------+-----------------+-----------------+
1846 | Connection tracking | | | | DPDK 21.05 |
1847 | | | N/A | | OFED 5.3 |
1848 | | | | | rdma-core 35 |
1849 | | | | | ConnectX-6 Dx |
1850 +-----------------------+-----------------+-----------------+
1852 .. table:: Minimal SW/HW versions for shared action offload
1855 +-----------------------+-----------------+-----------------+
1856 | Shared Action | with E-Switch | with NIC |
1857 +=======================+=================+=================+
1858 | RSS | | | | DPDK 20.11 |
1859 | | | N/A | | OFED 5.2 |
1860 | | | | | rdma-core 33 |
1861 | | | | | ConnectX-5 |
1862 +-----------------------+-----------------+-----------------+
1863 | Age | | DPDK 20.11 | | DPDK 20.11 |
1864 | | | OFED 5.2 | | OFED 5.2 |
1865 | | | rdma-core 32 | | rdma-core 32 |
1866 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1867 +-----------------------+-----------------+-----------------+
1868 | Count | | DPDK 21.05 | | DPDK 21.05 |
1869 | | | OFED 4.6 | | OFED 4.6 |
1870 | | | rdma-core 24 | | rdma-core 23 |
1871 | | | ConnectX-5 | | ConnectX-5 |
1872 +-----------------------+-----------------+-----------------+
1877 MARK and META items are interrelated with datapath - they might move from/to
1878 the applications in mbuf fields. Hence, zero value for these items has the
1879 special meaning - it means "no metadata are provided", not zero values are
1880 treated by applications and PMD as valid ones.
1882 Moreover in the flow engine domain the value zero is acceptable to match and
1883 set, and we should allow to specify zero values as rte_flow parameters for the
1884 META and MARK items and actions. In the same time zero mask has no meaning and
1885 should be rejected on validation stage.
1890 Flows are not cached in the driver.
1891 When stopping a device port, all the flows created on this port from the
1892 application will be flushed automatically in the background.
1893 After stopping the device port, all flows on this port become invalid and
1894 not represented in the system.
1895 All references to these flows held by the application should be discarded
1896 directly but neither destroyed nor flushed.
1898 The application should re-create the flows as required after the port restart.
1903 Compared to librte_net_mlx4 that implements a single RSS configuration per
1904 port, librte_net_mlx5 supports per-protocol RSS configuration.
1906 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1907 command-line parameter to enable additional protocols (UDP and TCP as well
1908 as IP), the following commands must be entered from its CLI to get the same
1909 behavior as librte_net_mlx4::
1912 > port config all rss all
1918 This section demonstrates how to launch **testpmd** with Mellanox
1919 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1921 #. Load the kernel modules::
1923 modprobe -a ib_uverbs mlx5_core mlx5_ib
1925 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1928 /etc/init.d/openibd restart
1932 User space I/O kernel modules (uio and igb_uio) are not used and do
1933 not have to be loaded.
1935 #. Make sure Ethernet interfaces are in working order and linked to kernel
1936 verbs. Related sysfs entries should be present::
1938 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1947 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1950 for intf in eth2 eth3 eth4 eth5;
1952 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1955 sed -n 's,.*/\(.*\),-a \1,p'
1964 #. Request huge pages::
1966 dpdk-hugepages.py --setup 2G
1968 #. Start testpmd with basic parameters::
1970 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
1975 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1976 EAL: probe driver: 15b3:1013 librte_net_mlx5
1977 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1978 PMD: librte_net_mlx5: 1 port(s) detected
1979 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1980 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1981 EAL: probe driver: 15b3:1013 librte_net_mlx5
1982 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1983 PMD: librte_net_mlx5: 1 port(s) detected
1984 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1985 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1986 EAL: probe driver: 15b3:1013 librte_net_mlx5
1987 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1988 PMD: librte_net_mlx5: 1 port(s) detected
1989 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1990 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1991 EAL: probe driver: 15b3:1013 librte_net_mlx5
1992 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1993 PMD: librte_net_mlx5: 1 port(s) detected
1994 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1995 Interactive-mode selected
1996 Configuring Port 0 (socket 0)
1997 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1998 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1999 Port 0: E4:1D:2D:E7:0C:FE
2000 Configuring Port 1 (socket 0)
2001 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
2002 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
2003 Port 1: E4:1D:2D:E7:0C:FF
2004 Configuring Port 2 (socket 0)
2005 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
2006 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
2007 Port 2: E4:1D:2D:E7:0C:FA
2008 Configuring Port 3 (socket 0)
2009 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
2010 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
2011 Port 3: E4:1D:2D:E7:0C:FB
2012 Checking link statuses...
2013 Port 0 Link Up - speed 40000 Mbps - full-duplex
2014 Port 1 Link Up - speed 40000 Mbps - full-duplex
2015 Port 2 Link Up - speed 10000 Mbps - full-duplex
2016 Port 3 Link Up - speed 10000 Mbps - full-duplex
2023 This section demonstrates how to dump flows. Currently, it's possible to dump
2024 all flows with assistance of external tools.
2026 #. 2 ways to get flow raw file:
2028 - Using testpmd CLI:
2030 .. code-block:: console
2033 testpmd> flow dump <port> all <output_file>
2035 testpmd> flow dump <port> rule <rule_id> <output_file>
2037 - call rte_flow_dev_dump api:
2039 .. code-block:: console
2041 rte_flow_dev_dump(port, flow, file, NULL);
2043 #. Dump human-readable flows from raw file:
2045 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
2047 .. code-block:: console
2049 mlx_steering_dump.py -f <output_file> -flowptr <flow_ptr>
2051 How to share a meter between ports in the same switch domain
2052 ------------------------------------------------------------
2054 This section demonstrates how to use the shared meter. A meter M can be created
2055 on port X and to be shared with a port Y on the same switch domain by the next way:
2057 .. code-block:: console
2059 flow create X ingress transfer pattern eth / port_id id is Y / end actions meter mtr_id M / end
2061 How to use meter hierarchy
2062 --------------------------
2064 This section demonstrates how to create and use a meter hierarchy.
2065 A termination meter M can be the policy green action of another termination meter N.
2066 The two meters are chained together as a chain. Using meter N in a flow will apply
2067 both the meters in hierarchy on that flow.
2069 .. code-block:: console
2071 add port meter policy 0 1 g_actions queue index 0 / end y_actions end r_actions drop / end
2072 create port meter 0 M 1 1 yes 0xffff 1 0
2073 add port meter policy 0 2 g_actions meter mtr_id M / end y_actions end r_actions drop / end
2074 create port meter 0 N 2 2 yes 0xffff 1 0
2075 flow create 0 ingress group 1 pattern eth / end actions meter mtr_id N / end