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 - Support for scattered TX frames.
65 - Advanced support for scattered Rx frames with tunable buffer attributes.
66 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
67 - RSS using different combinations of fields: L3 only, L4 only or both,
68 and source only, destination only or both.
69 - Several RSS hash keys, one for each flow type.
70 - Default RSS operation with no hash key specification.
71 - Configurable RETA table.
72 - Link flow control (pause frame).
73 - Support for multiple MAC addresses.
77 - RX CRC stripping configuration.
78 - TX mbuf fast free offload.
79 - Promiscuous mode on PF and VF.
80 - Multicast promiscuous mode on PF and VF.
81 - Hardware checksum offloads.
82 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
84 - Flow API, including :ref:`flow_isolated_mode`.
86 - KVM and VMware ESX SR-IOV modes are supported.
87 - RSS hash result is supported.
88 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
89 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
91 - Statistics query including Basic, Extended and per queue.
93 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
94 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
95 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
96 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
97 - Flow insertion rate of more then million flows per second, when using Direct Rules.
98 - Support for multiple rte_flow groups.
99 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
102 - Multiple-thread flow insertion.
103 - Matching on IPv4 Internet Header Length (IHL).
104 - Matching on GTP extension header with raw encap/decap action.
105 - Matching on Geneve TLV option header with raw encap/decap action.
106 - RSS support in sample action.
107 - E-Switch mirroring and jump.
108 - E-Switch mirroring and modify.
109 - 21844 flow priorities for ingress or egress flow groups greater than 0 and for any transfer
111 - Flow metering, including meter policy API.
112 - Flow meter hierarchy.
113 - Flow integrity offload API.
114 - Connection tracking.
115 - Sub-Function representors.
124 On Windows, the features are limited:
126 - Promiscuous mode is not supported
127 - The following rules are supported:
129 - IPv4/UDP with CVLAN filtering
130 - Unicast MAC filtering
132 - Additional rules are supported from WinOF2 version 2.70:
134 - IPv4/TCP with CVLAN filtering
135 - L4 steering rules for port RSS of UDP, TCP and IP
137 - For secondary process:
139 - Forked secondary process not supported.
140 - External memory unregistered in EAL memseg list cannot be used for DMA
141 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
142 primary process and remapped to the same virtual address in secondary
143 process. If the external memory is registered by primary process but has
144 different virtual address in secondary process, unexpected error may happen.
148 - Counters of received packets and bytes number of devices in same share group are same.
149 - Counters of received packets and bytes number of queues in same group and queue ID are same.
151 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
152 specific VLAN will match for VLAN packets as well:
154 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
155 Meaning, the flow rule::
157 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
159 Will only match vlan packets with vid=3. and the flow rule::
161 flow create 0 ingress pattern eth / ipv4 / end ...
163 Will match any ipv4 packet (VLAN included).
165 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
167 - 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.
170 flow create 0 ingress pattern eth / ipv4 / end ...
172 Will match any ipv4 packet.
175 flow create 0 ingress pattern eth / vlan / end ...
176 flow create 0 ingress pattern eth has_vlan is 1 / end ...
177 flow create 0 ingress pattern eth type is 0x8100 / end ...
179 Will match single-tagged packets only, with any VLAN ID value.
182 flow create 0 ingress pattern eth type is 0x88A8 / end ...
183 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
185 Will match multi-tagged packets only, with any VLAN ID value.
187 - A flow pattern with 2 sequential VLAN items is not supported.
189 - VLAN pop offload command:
191 - Flow rules having a VLAN pop offload command as one of their actions and
192 are lacking a match on VLAN as one of their items are not supported.
193 - The command is not supported on egress traffic in NIC mode.
195 - VLAN push offload is not supported on ingress traffic in NIC mode.
197 - VLAN set PCP offload is not supported on existing headers.
199 - A multi segment packet must have not more segments than reported by dev_infos_get()
200 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
201 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
202 inline settings) to 58.
204 - Match on VXLAN supports the following fields only:
207 - Last reserved 8-bits
209 Last reserved 8-bits matching is only supported When using DV flow
210 engine (``dv_flow_en`` = 1).
211 For ConnectX-5, the UDP destination port must be the standard one (4789).
212 Group zero's behavior may differ which depends on FW.
213 Matching value equals 0 (value & mask) is not supported.
215 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
217 - Match on Geneve header supports the following fields only:
224 - Match on Geneve TLV option is supported on the following fields:
231 Only one Class/Type/Length Geneve TLV option is supported per shared device.
232 Class/Type/Length fields must be specified as well as masks.
233 Class/Type/Length specified masks must be full.
234 Matching Geneve TLV option without specifying data is not supported.
235 Matching Geneve TLV option with ``data & mask == 0`` is not supported.
237 - VF: flow rules created on VF devices can only match traffic targeted at the
238 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
240 - Match on GTP tunnel header item supports the following fields only:
242 - v_pt_rsv_flags: E flag, S flag, PN flag
246 - Match on GTP extension header only for GTP PDU session container (next
247 extension header type = 0x85).
248 - Match on GTP extension header is not supported in group 0.
250 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
251 The flows within group 0 and set metadata action are rejected by hardware.
255 MAC addresses not already present in the bridge table of the associated
256 kernel network device will be added and cleaned up by the PMD when closing
257 the device. In case of ungraceful program termination, some entries may
258 remain present and should be removed manually by other means.
260 - Buffer split offload is supported with regular Rx burst routine only,
261 no MPRQ feature or vectorized code can be engaged.
263 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
264 externally attached to a user-provided mbuf with having RTE_MBUF_F_EXTERNAL in
265 ol_flags. As the mempool for the external buffer is managed by PMD, all the
266 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
267 the external buffers will be freed by PMD and the application which still
268 holds the external buffers may be corrupted.
270 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
271 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
272 supported. Some Rx packets may not have RTE_MBUF_F_RX_RSS_HASH.
274 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
275 and allmulticast mode are both set to off.
276 To receive IPv6 Multicast messages on VM, explicitly set the relevant
277 MAC address using rte_eth_dev_mac_addr_add() API.
279 - To support a mixed traffic pattern (some buffers from local host memory, some
280 buffers from other devices) with high bandwidth, a mbuf flag is used.
282 An application hints the PMD whether or not it should try to inline the
283 given mbuf data buffer. PMD should do the best effort to act upon this request.
285 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
286 registered by application with rte_mbuf_dynflag_register(). This flag is
287 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
288 which is intended to be used by the application.
290 To query the supported specific flags in runtime,
291 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
292 currently (over present hardware and configuration) supported specific flags.
293 The "not inline hint" feature operating flow is the following one:
296 - probe the devices, ports are created
297 - query the port capabilities
298 - if port supporting the feature is found
299 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
300 - application starts the ports
301 - on ``dev_start()`` PMD checks whether the feature flag is registered and
302 enables the feature support in datapath
303 - application might set the registered flag bit in ``ol_flags`` field
304 of mbuf being sent and PMD will handle ones appropriately.
306 - The amount of descriptors in Tx queue may be limited by data inline settings.
307 Inline data require the more descriptor building blocks and overall block
308 amount may exceed the hardware supported limits. The application should
309 reduce the requested Tx size or adjust data inline settings with
310 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
312 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
313 parameter should be specified.
314 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
315 being sent it tries to synchronize the time of packet appearing on
316 the wire with the specified packet timestamp. It the specified one
317 is in the past it should be ignored, if one is in the distant future
318 it should be capped with some reasonable value (in range of seconds).
319 These specific cases ("too late" and "distant future") can be optionally
320 reported via device xstats to assist applications to detect the
321 time-related problems.
323 The timestamp upper "too-distant-future" limit
324 at the moment of invoking the Tx burst routine
325 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
326 Please note, for the testpmd txonly mode,
327 the limit is deduced from the expression::
329 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
331 There is no any packet reordering according timestamps is supposed,
332 neither within packet burst, nor between packets, it is an entirely
333 application responsibility to generate packets and its timestamps
334 in desired order. The timestamps can be put only in the first packet
335 in the burst providing the entire burst scheduling.
337 - E-Switch decapsulation Flow:
339 - can be applied to PF port only.
340 - must specify VF port action (packet redirection from PF to VF).
341 - optionally may specify tunnel inner source and destination MAC addresses.
343 - E-Switch encapsulation Flow:
345 - can be applied to VF ports only.
346 - must specify PF port action (packet redirection from VF to PF).
350 - The input buffer, used as outer header, is not validated.
354 - The decapsulation is always done up to the outermost tunnel detected by the HW.
355 - The input buffer, providing the removal size, is not validated.
356 - The buffer size must match the length of the headers to be removed.
358 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
359 mutually exclusive features which cannot be supported together
360 (see :ref:`mlx5_firmware_config`).
364 - Requires DevX and DV flow to be enabled.
365 - KEEP_CRC offload cannot be supported with LRO.
366 - The first mbuf length, without head-room, must be big enough to include the
368 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
369 it with size limited to max LRO size, not to max RX packet length.
370 - LRO can be used with outer header of TCP packets of the standard format:
371 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
373 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
374 - LRO packet aggregation is performed by HW only for packet size larger than
375 ``lro_min_mss_size``. This value is reported on device start, when debug
380 - ``RTE_ETH_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
381 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
382 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
386 - fast free offload assumes the all mbufs being sent are originated from the
387 same memory pool and there is no any extra references to the mbufs (the
388 reference counter for each mbuf is equal 1 on tx_burst call). The latter
389 means there should be no any externally attached buffers in mbufs. It is
390 an application responsibility to provide the correct mbufs if the fast
391 free offload is engaged. The mlx5 PMD implicitly produces the mbufs with
392 externally attached buffers if MPRQ option is enabled, hence, the fast
393 free offload is neither supported nor advertised if there is MPRQ enabled.
397 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and
398 E-Switch steering domain.
399 - For E-Switch Sampling flow with sample ratio > 1, additional actions are not
400 supported in the sample actions list.
401 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as
402 first action in the E-Switch egress flow if with header modify or
403 encapsulation actions.
404 - For NIC Rx flow, supports ``MARK``, ``COUNT``, ``QUEUE``, ``RSS`` in the
406 - For E-Switch mirroring flow, supports ``RAW ENCAP``, ``Port ID``,
407 ``VXLAN ENCAP``, ``NVGRE ENCAP`` in the sample actions list.
411 - Supports the 'set' operation only for ``RTE_FLOW_ACTION_TYPE_MODIFY_FIELD`` action.
412 - Modification of an arbitrary place in a packet via the special ``RTE_FLOW_FIELD_START`` Field ID is not supported.
413 - Modification of the 802.1Q Tag, VXLAN Network or GENEVE Network ID's is not supported.
414 - Encapsulation levels are not supported, can modify outermost header fields only.
415 - Offsets must be 32-bits aligned, cannot skip past the boundary of a field.
417 - IPv6 header item 'proto' field, indicating the next header protocol, should
418 not be set as extension header.
419 In case the next header is an extension header, it should not be specified in
420 IPv6 header item 'proto' field.
421 The last extension header item 'next header' field can specify the following
422 header protocol type.
426 - 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.
427 - Hairpin in switchdev SR-IOV mode is not supported till now.
431 - All the meter colors with drop action will be counted only by the global drop statistics.
432 - Yellow detection is only supported with ASO metering.
433 - Red color must be with drop action.
434 - Meter statistics are supported only for drop case.
435 - A meter action created with pre-defined policy must be the last action in the flow except single case where the policy actions are:
436 - green: NULL or END.
437 - yellow: NULL or END.
439 - The only supported meter policy actions:
440 - green: QUEUE, RSS, PORT_ID, REPRESENTED_PORT, JUMP, DROP, MARK and SET_TAG.
441 - yellow: QUEUE, RSS, PORT_ID, REPRESENTED_PORT, JUMP, DROP, MARK and SET_TAG.
443 - Policy actions of RSS for green and yellow should have the same configuration except queues.
444 - Policy with RSS/queue action is not supported when ``dv_xmeta_en`` enabled.
445 - meter profile packet mode is supported.
446 - meter profiles of RFC2697, RFC2698 and RFC4115 are supported.
450 - Integrity offload is enabled for **ConnectX-6** family.
451 - Verification bits provided by the hardware are ``l3_ok``, ``ipv4_csum_ok``, ``l4_ok``, ``l4_csum_ok``.
452 - ``level`` value 0 references outer headers.
453 - Multiple integrity items not supported in a single flow rule.
454 - Flow rule items supplied by application must explicitly specify network headers referred by integrity item.
455 For example, if integrity item mask sets ``l4_ok`` or ``l4_csum_ok`` bits, reference to L4 network header,
456 TCP or UDP, must be in the rule pattern as well::
458 flow create 0 ingress pattern integrity level is 0 value mask l3_ok value spec l3_ok / eth / ipv6 / end …
460 flow create 0 ingress pattern integrity level is 0 value mask l4_ok value spec 0 / eth / ipv4 proto is udp / end …
462 - Connection tracking:
464 - Cannot co-exist with ASO meter, ASO age action in a single flow rule.
465 - Flow rules insertion rate and memory consumption need more optimization.
467 - 4M connections maximum.
469 - Multi-thread flow insertion:
471 - In order to achieve best insertion rate, application should manage the flows per lcore.
472 - Better to disable memory reclaim by setting ``reclaim_mem_mode`` to 0 to accelerate the flow object allocation and release with cache.
476 - TXQ affinity subjects to HW hash once enabled.
478 - Bonding under socket direct mode
485 MLX5 supports various methods to report statistics:
487 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.
489 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.
491 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.
499 The ibverbs libraries can be linked with this PMD in a number of ways,
500 configured by the ``ibverbs_link`` build option:
502 - ``shared`` (default): the PMD depends on some .so files.
504 - ``dlopen``: Split the dependencies glue in a separate library
505 loaded when needed by dlopen.
506 It make dependencies on libibverbs and libmlx4 optional,
507 and has no performance impact.
509 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
510 in the PMD shared library or the executable static binary.
512 Environment variables
513 ~~~~~~~~~~~~~~~~~~~~~
517 A list of directories in which to search for the rdma-core "glue" plug-in,
518 separated by colons or semi-colons.
520 - ``MLX5_SHUT_UP_BF``
522 Configures HW Tx doorbell register as IO-mapped.
524 By default, the HW Tx doorbell is configured as a write-combining register.
525 The register would be flushed to HW usually when the write-combining buffer
526 becomes full, but it depends on CPU design.
528 Except for vectorized Tx burst routines, a write memory barrier is enforced
529 after updating the register so that the update can be immediately visible to
532 When vectorized Tx burst is called, the barrier is set only if the burst size
533 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
534 variable will bring better latency even though the maximum throughput can
537 Run-time configuration
538 ~~~~~~~~~~~~~~~~~~~~~~
540 - librte_net_mlx5 brings kernel network interfaces up during initialization
541 because it is affected by their state. Forcing them down prevents packets
544 - **ethtool** operations on related kernel interfaces also affect the PMD.
549 In order to run as a non-root user,
550 some capabilities must be granted to the application::
552 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
554 Below are the reasons of the need for each capability:
557 When using physical addresses (PA mode), with Linux >= 4.0,
558 for access to ``/proc/self/pagemap``.
561 For device configuration.
564 For raw ethernet queue allocation through kernel driver.
567 For DMA memory pinning.
572 - ``rxq_cqe_comp_en`` parameter [int]
574 A nonzero value enables the compression of CQE on RX side. This feature
575 allows to save PCI bandwidth and improve performance. Enabled by default.
576 Different compression formats are supported in order to achieve the best
577 performance for different traffic patterns. Default format depends on
578 Multi-Packet Rx queue configuration: Hash RSS format is used in case
579 MPRQ is disabled, Checksum format is used in case MPRQ is enabled.
581 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
582 better compression rate in case of RTE Flow Mark traffic.
583 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
584 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
585 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
586 CQE compression format selection requires DevX to be enabled. If there is
587 no DevX enabled/supported the value is reset to 1 by default.
591 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
592 ConnectX-6 Lx, BlueField and BlueField-2.
593 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
594 ConnectX-6 Lx, BlueField and BlueField-2.
596 - ``rxq_pkt_pad_en`` parameter [int]
598 A nonzero value enables padding Rx packet to the size of cacheline on PCI
599 transaction. This feature would waste PCI bandwidth but could improve
600 performance by avoiding partial cacheline write which may cause costly
601 read-modify-copy in memory transaction on some architectures. Disabled by
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 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
609 ConnectX-6 Lx, BlueField and BlueField-2.
611 - ``mprq_en`` parameter [int]
613 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
614 configured as Multi-Packet RQ if the total number of Rx queues is
615 ``rxqs_min_mprq`` or more. Disabled by default.
617 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
618 by posting a single large buffer for multiple packets. Instead of posting a
619 buffers per a packet, one large buffer is posted in order to receive multiple
620 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
621 and each stride receives one packet. MPRQ can improve throughput for
622 small-packet traffic.
624 When MPRQ is enabled, MTU can be larger than the size of
625 user-provided mbuf even if RTE_ETH_RX_OFFLOAD_SCATTER isn't enabled. PMD will
626 configure large stride size enough to accommodate MTU as long as
627 device allows. Note that this can waste system memory compared to enabling Rx
628 scatter and multi-segment packet.
630 - ``mprq_log_stride_num`` parameter [int]
632 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
633 strides can reduce PCIe traffic further. If configured value is not in the
634 range of device capability, the default value will be set with a warning
635 message. The default value is 4 which is 16 strides per a buffer, valid only
636 if ``mprq_en`` is set.
638 The size of Rx queue should be bigger than the number of strides.
640 - ``mprq_log_stride_size`` parameter [int]
642 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
643 stride size can save some memory and reduce probability of a depletion of all
644 available strides due to unreleased packets by an application. If configured
645 value is not in the range of device capability, the default value will be set
646 with a warning message. The default value is 11 which is 2048 bytes per a
647 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
648 it is possible for a packet to span across multiple strides. This mode allows
649 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
650 of a packet if Rx scatter is configured) may be required in case there is no
651 space left for a head room at the end of a stride which incurs some
654 - ``mprq_max_memcpy_len`` parameter [int]
656 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
657 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
658 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
659 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
660 A mempool for external buffers will be allocated and managed by PMD. If Rx
661 packet is externally attached, ol_flags field of the mbuf will have
662 RTE_MBUF_F_EXTERNAL and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
663 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
665 - ``rxqs_min_mprq`` parameter [int]
667 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
668 greater or equal to this value. The default value is 12, valid only if
671 - ``txq_inline`` parameter [int]
673 Amount of data to be inlined during TX operations. This parameter is
674 deprecated and converted to the new parameter ``txq_inline_max`` providing
675 partial compatibility.
677 - ``txqs_min_inline`` parameter [int]
679 Enable inline data send only when the number of TX queues is greater or equal
682 This option should be used in combination with ``txq_inline_max`` and
683 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
685 If this option is not specified the default value 16 is used for BlueField
686 and 8 for other platforms
688 The data inlining consumes the CPU cycles, so this option is intended to
689 auto enable inline data if we have enough Tx queues, which means we have
690 enough CPU cores and PCI bandwidth is getting more critical and CPU
691 is not supposed to be bottleneck anymore.
693 The copying data into WQE improves latency and can improve PPS performance
694 when PCI back pressure is detected and may be useful for scenarios involving
695 heavy traffic on many queues.
697 Because additional software logic is necessary to handle this mode, this
698 option should be used with care, as it may lower performance when back
699 pressure is not expected.
701 If inline data are enabled it may affect the maximal size of Tx queue in
702 descriptors because the inline data increase the descriptor size and
703 queue size limits supported by hardware may be exceeded.
705 - ``txq_inline_min`` parameter [int]
707 Minimal amount of data to be inlined into WQE during Tx operations. NICs
708 may require this minimal data amount to operate correctly. The exact value
709 may depend on NIC operation mode, requested offloads, etc. It is strongly
710 recommended to omit this parameter and use the default values. Anyway,
711 applications using this parameter should take into consideration that
712 specifying an inconsistent value may prevent the NIC from sending packets.
714 If ``txq_inline_min`` key is present the specified value (may be aligned
715 by the driver in order not to exceed the limits and provide better descriptor
716 space utilization) will be used by the driver and it is guaranteed that
717 requested amount of data bytes are inlined into the WQE beside other inline
718 settings. This key also may update ``txq_inline_max`` value (default
719 or specified explicitly in devargs) to reserve the space for inline data.
721 If ``txq_inline_min`` key is not present, the value may be queried by the
722 driver from the NIC via DevX if this feature is available. If there is no DevX
723 enabled/supported the value 18 (supposing L2 header including VLAN) is set
724 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
725 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
728 For ConnectX-4 NIC, driver does not allow specifying value below 18
729 (minimal L2 header, including VLAN), error will be raised.
731 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
732 it is not recommended and may prevent NIC from sending packets over
735 For ConnectX-4 and ConnectX-4 Lx NICs, automatically configured value
736 is insufficient for some traffic, because they require at least all L2 headers
737 to be inlined. For example, Q-in-Q adds 4 bytes to default 18 bytes
738 of Ethernet and VLAN, thus ``txq_inline_min`` must be set to 22.
739 MPLS would add 4 bytes per label. Final value must account for all possible
740 L2 encapsulation headers used in particular environment.
742 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
743 Multi-Packet Write), because last one does not support partial packet inlining.
744 This is not very critical due to minimal data inlining is mostly required
745 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
747 - ``txq_inline_max`` parameter [int]
749 Specifies the maximal packet length to be completely inlined into WQE
750 Ethernet Segment for ordinary SEND method. If packet is larger than specified
751 value, the packet data won't be copied by the driver at all, data buffer
752 is addressed with a pointer. If packet length is less or equal all packet
753 data will be copied into WQE. This may improve PCI bandwidth utilization for
754 short packets significantly but requires the extra CPU cycles.
756 The data inline feature is controlled by number of Tx queues, if number of Tx
757 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
758 is engaged, if there are not enough Tx queues (which means not enough CPU cores
759 and CPU resources are scarce), data inline is not performed by the driver.
760 Assigning ``txqs_min_inline`` with zero always enables the data inline.
762 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
763 by the driver in order not to exceed the limit (930 bytes) and to provide better
764 WQE space filling without gaps, the adjustment is reflected in the debug log.
765 Also, the default value (290) may be decreased in run-time if the large transmit
766 queue size is requested and hardware does not support enough descriptor
767 amount, in this case warning is emitted. If ``txq_inline_max`` key is
768 specified and requested inline settings can not be satisfied then error
771 - ``txq_inline_mpw`` parameter [int]
773 Specifies the maximal packet length to be completely inlined into WQE for
774 Enhanced MPW method. If packet is large the specified value, the packet data
775 won't be copied, and data buffer is addressed with pointer. If packet length
776 is less or equal, all packet data will be copied into WQE. This may improve PCI
777 bandwidth utilization for short packets significantly but requires the extra
780 The data inline feature is controlled by number of TX queues, if number of Tx
781 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
782 is engaged, if there are not enough Tx queues (which means not enough CPU cores
783 and CPU resources are scarce), data inline is not performed by the driver.
784 Assigning ``txqs_min_inline`` with zero always enables the data inline.
786 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
787 by the driver in order not to exceed the limit (930 bytes) and to provide better
788 WQE space filling without gaps, the adjustment is reflected in the debug log.
789 Due to multiple packets may be included to the same WQE with Enhanced Multi
790 Packet Write Method and overall WQE size is limited it is not recommended to
791 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
792 may be decreased in run-time if the large transmit queue size is requested
793 and hardware does not support enough descriptor amount, in this case warning
794 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
795 settings can not be satisfied then error will be raised.
797 - ``txqs_max_vec`` parameter [int]
799 Enable vectorized Tx only when the number of TX queues is less than or
800 equal to this value. This parameter is deprecated and ignored, kept
801 for compatibility issue to not prevent driver from probing.
803 - ``txq_mpw_hdr_dseg_en`` parameter [int]
805 A nonzero value enables including two pointers in the first block of TX
806 descriptor. The parameter is deprecated and ignored, kept for compatibility
809 - ``txq_max_inline_len`` parameter [int]
811 Maximum size of packet to be inlined. This limits the size of packet to
812 be inlined. If the size of a packet is larger than configured value, the
813 packet isn't inlined even though there's enough space remained in the
814 descriptor. Instead, the packet is included with pointer. This parameter
815 is deprecated and converted directly to ``txq_inline_mpw`` providing full
816 compatibility. Valid only if eMPW feature is engaged.
818 - ``txq_mpw_en`` parameter [int]
820 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
821 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
822 eMPW allows the Tx burst function to pack up multiple packets
823 in a single descriptor session in order to save PCI bandwidth
824 and improve performance at the cost of a slightly higher CPU usage.
825 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
826 Tx burst function copies entire packet data on to Tx descriptor
827 instead of including pointer of packet.
829 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
830 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
831 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
832 option or reported by the NIC, the eMPW feature is disengaged.
834 - ``tx_db_nc`` parameter [int]
836 The rdma core library can map doorbell register in two ways, depending on the
837 environment variable "MLX5_SHUT_UP_BF":
839 - As regular cached memory (usually with write combining attribute), if the
840 variable is either missing or set to zero.
841 - As non-cached memory, if the variable is present and set to not "0" value.
843 The type of mapping may slightly affect the Tx performance, the optimal choice
844 is strongly relied on the host architecture and should be deduced practically.
846 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
847 memory (with write combining), the PMD will perform the extra write memory barrier
848 after writing to doorbell, it might increase the needed CPU clocks per packet
849 to send, but latency might be improved.
851 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
852 cached memory, the PMD will not perform the extra write memory barrier
853 after writing to doorbell, on some architectures it might improve the
856 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
857 memory, the PMD will use heuristics to decide whether write memory barrier
858 should be performed. For bursts with size multiple of recommended one (64 pkts)
859 it is supposed the next burst is coming and no need to issue the extra memory
860 barrier (it is supposed to be issued in the next coming burst, at least after
861 descriptor writing). It might increase latency (on some hosts till next
862 packets transmit) and should be used with care.
864 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
865 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
866 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
868 - ``tx_pp`` parameter [int]
870 If a nonzero value is specified the driver creates all necessary internal
871 objects to provide accurate packet send scheduling on mbuf timestamps.
872 The positive value specifies the scheduling granularity in nanoseconds,
873 the packet send will be accurate up to specified digits. The allowed range is
874 from 500 to 1 million of nanoseconds. The negative value specifies the module
875 of granularity and engages the special test mode the check the schedule rate.
876 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
879 - ``tx_skew`` parameter [int]
881 The parameter adjusts the send packet scheduling on timestamps and represents
882 the average delay between beginning of the transmitting descriptor processing
883 by the hardware and appearance of actual packet data on the wire. The value
884 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
885 specified. The default value is zero.
887 - ``tx_vec_en`` parameter [int]
889 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
890 ConnectX-6 Lx, BlueField and BlueField-2 NICs
891 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
892 The parameter is deprecated and ignored.
894 - ``rx_vec_en`` parameter [int]
896 A nonzero value enables Rx vector if the port is not configured in
897 multi-segment otherwise this parameter is ignored.
901 - ``vf_nl_en`` parameter [int]
903 A nonzero value enables Netlink requests from the VF to add/remove MAC
904 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
905 Otherwise the relevant configuration must be run with Linux iproute2 tools.
906 This is a prerequisite to receive this kind of traffic.
908 Enabled by default, valid only on VF devices ignored otherwise.
910 - ``l3_vxlan_en`` parameter [int]
912 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
913 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
914 parameter. This is a prerequisite to receive this kind of traffic.
918 - ``dv_xmeta_en`` parameter [int]
920 A nonzero value enables extensive flow metadata support if device is
921 capable and driver supports it. This can enable extensive support of
922 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
923 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
925 There are some possible configurations, depending on parameter value:
927 - 0, this is default value, defines the legacy mode, the ``MARK`` and
928 ``META`` related actions and items operate only within NIC Tx and
929 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
930 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
931 item is 32 bits wide and match supported on egress only.
933 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
934 related actions and items operate within all supported steering domains,
935 including FDB, ``MARK`` and ``META`` information may cross the domain
936 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
937 depends on kernel and firmware configurations and might be 0, 16 or
938 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
939 compatibility, the actual width of data transferred to the FDB domain
940 depends on kernel configuration and may be vary. The actual supported
941 width can be retrieved in runtime by series of rte_flow_validate()
944 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
945 related actions and items operate within all supported steering domains,
946 including FDB, ``MARK`` and ``META`` information may cross the domain
947 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
948 depends on kernel and firmware configurations and might be 0, 16 or
949 24 bits. The actual supported width can be retrieved in runtime by
950 series of rte_flow_validate() trials.
952 - 3, this engages tunnel offload mode. In E-Switch configuration, that
953 mode implicitly activates ``dv_xmeta_en=1``.
955 +------+-----------+-----------+-------------+-------------+
956 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
957 +======+===========+===========+=============+=============+
958 | 0 | 24 bits | 32 bits | 32 bits | no |
959 +------+-----------+-----------+-------------+-------------+
960 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
961 +------+-----------+-----------+-------------+-------------+
962 | 2 | vary 0-24 | 32 bits | 32 bits | yes |
963 +------+-----------+-----------+-------------+-------------+
965 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
966 ignored and the device is configured to operate in legacy mode (0).
968 Disabled by default (set to 0).
970 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
971 of the extensive metadata features. The legacy Verbs supports FLAG and
972 MARK metadata actions over NIC Rx steering domain only.
974 Setting META value to zero in flow action means there is no item provided
975 and receiving datapath will not report in mbufs the metadata are present.
976 Setting MARK value to zero in flow action means the zero FDIR ID value
977 will be reported on packet receiving.
979 For the MARK action the last 16 values in the full range are reserved for
980 internal PMD purposes (to emulate FLAG action). The valid range for the
981 MARK action values is 0-0xFFEF for the 16-bit mode and 0-xFFFFEF
982 for the 24-bit mode, the flows with the MARK action value outside
983 the specified range will be rejected.
985 - ``dv_flow_en`` parameter [int]
987 A nonzero value enables the DV flow steering assuming it is supported
988 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
990 Enabled by default if supported.
992 - ``dv_esw_en`` parameter [int]
994 A nonzero value enables E-Switch using Direct Rules.
996 Enabled by default if supported.
998 - ``lacp_by_user`` parameter [int]
1000 A nonzero value enables the control of LACP traffic by the user application.
1001 When a bond exists in the driver, by default it should be managed by the
1002 kernel and therefore LACP traffic should be steered to the kernel.
1003 If this devarg is set to 1 it will allow the user to manage the bond by
1004 itself and not steer LACP traffic to the kernel.
1006 Disabled by default (set to 0).
1008 - ``mr_ext_memseg_en`` parameter [int]
1010 A nonzero value enables extending memseg when registering DMA memory. If
1011 enabled, the number of entries in MR (Memory Region) lookup table on datapath
1012 is minimized and it benefits performance. On the other hand, it worsens memory
1013 utilization because registered memory is pinned by kernel driver. Even if a
1014 page in the extended chunk is freed, that doesn't become reusable until the
1015 entire memory is freed.
1019 - ``mr_mempool_reg_en`` parameter [int]
1021 A nonzero value enables implicit registration of DMA memory of all mempools
1022 except those having ``RTE_MEMPOOL_F_NON_IO``. This flag is set automatically
1023 for mempools populated with non-contiguous objects or those without IOVA.
1024 The effect is that when a packet from a mempool is transmitted,
1025 its memory is already registered for DMA in the PMD and no registration
1026 will happen on the data path. The tradeoff is extra work on the creation
1027 of each mempool and increased HW resource use if some mempools
1028 are not used with MLX5 devices.
1032 - ``representor`` parameter [list]
1034 This parameter can be used to instantiate DPDK Ethernet devices from
1035 existing port (PF, VF or SF) representors configured on the device.
1037 It is a standard parameter whose format is described in
1038 :ref:`ethernet_device_standard_device_arguments`.
1040 For instance, to probe VF port representors 0 through 2::
1042 <PCI_BDF>,representor=vf[0-2]
1044 To probe SF port representors 0 through 2::
1046 <PCI_BDF>,representor=sf[0-2]
1048 To probe VF port representors 0 through 2 on both PFs of bonding device::
1050 <Primary_PCI_BDF>,representor=pf[0,1]vf[0-2]
1052 - ``max_dump_files_num`` parameter [int]
1054 The maximum number of files per PMD entity that may be created for debug information.
1055 The files will be created in /var/log directory or in current directory.
1057 set to 128 by default.
1059 - ``lro_timeout_usec`` parameter [int]
1061 The maximum allowed duration of an LRO session, in micro-seconds.
1062 PMD will set the nearest value supported by HW, which is not bigger than
1063 the input ``lro_timeout_usec`` value.
1064 If this parameter is not specified, by default PMD will set
1065 the smallest value supported by HW.
1067 - ``hp_buf_log_sz`` parameter [int]
1069 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
1070 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
1071 The capacity of the value is specified by the firmware and the initialization
1072 will get a failure if it is out of scope.
1073 The range of the value is from 11 to 19 right now, and the supported frame
1074 size of a single packet for hairpin is from 512B to 128KB. It might change if
1075 different firmware release is being used. By using a small value, it could
1076 reduce memory consumption but not work with a large frame. If the value is
1077 too large, the memory consumption will be high and some potential performance
1078 degradation will be introduced.
1079 By default, the PMD will set this value to 16, which means that 9KB jumbo
1080 frames will be supported.
1082 - ``reclaim_mem_mode`` parameter [int]
1084 Cache some resources in flow destroy will help flow recreation more efficient.
1085 While some systems may require the all the resources can be reclaimed after
1087 The parameter ``reclaim_mem_mode`` provides the option for user to configure
1088 if the resource cache is needed or not.
1090 There are three options to choose:
1092 - 0. It means the flow resources will be cached as usual. The resources will
1093 be cached, helpful with flow insertion rate.
1095 - 1. It will only enable the DPDK PMD level resources reclaim.
1097 - 2. Both DPDK PMD level and rdma-core low level will be configured as
1100 By default, the PMD will set this value to 0.
1102 - ``sys_mem_en`` parameter [int]
1104 A non-zero value enables the PMD memory management allocating memory
1105 from system by default, without explicit rte memory flag.
1107 By default, the PMD will set this value to 0.
1109 - ``decap_en`` parameter [int]
1111 Some devices do not support FCS (frame checksum) scattering for
1112 tunnel-decapsulated packets.
1113 If set to 0, this option forces the FCS feature and rejects tunnel
1114 decapsulation in the flow engine for such devices.
1116 By default, the PMD will set this value to 1.
1118 - ``allow_duplicate_pattern`` parameter [int]
1120 There are two options to choose:
1122 - 0. Prevent insertion of rules with the same pattern items on non-root table.
1123 In this case, only the first rule is inserted and the following rules are
1124 rejected and error code EEXIST is returned.
1126 - 1. Allow insertion of rules with the same pattern items.
1127 In this case, all rules are inserted but only the first rule takes effect,
1128 the next rule takes effect only if the previous rules are deleted.
1130 By default, the PMD will set this value to 1.
1132 .. _mlx5_firmware_config:
1134 Firmware configuration
1135 ~~~~~~~~~~~~~~~~~~~~~~
1137 Firmware features can be configured as key/value pairs.
1139 The command to set a value is::
1141 mlxconfig -d <device> set <key>=<value>
1143 The command to query a value is::
1145 mlxconfig -d <device> query | grep <key>
1147 The device name for the command ``mlxconfig`` can be either the PCI address,
1148 or the mst device name found with::
1152 Below are some firmware configurations listed.
1158 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1164 - maximum number of SR-IOV virtual functions::
1168 - enable DevX (required by Direct Rules and other features)::
1172 - aggressive CQE zipping::
1176 - L3 VXLAN and VXLAN-GPE destination UDP port::
1179 IP_OVER_VXLAN_PORT=<udp dport>
1181 - enable VXLAN-GPE tunnel flow matching::
1183 FLEX_PARSER_PROFILE_ENABLE=0
1185 FLEX_PARSER_PROFILE_ENABLE=2
1187 - enable IP-in-IP tunnel flow matching::
1189 FLEX_PARSER_PROFILE_ENABLE=0
1191 - enable MPLS flow matching::
1193 FLEX_PARSER_PROFILE_ENABLE=1
1195 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1197 FLEX_PARSER_PROFILE_ENABLE=2
1199 - enable Geneve flow matching::
1201 FLEX_PARSER_PROFILE_ENABLE=0
1203 FLEX_PARSER_PROFILE_ENABLE=1
1205 - enable Geneve TLV option flow matching::
1207 FLEX_PARSER_PROFILE_ENABLE=0
1209 - enable GTP flow matching::
1211 FLEX_PARSER_PROFILE_ENABLE=3
1213 - enable eCPRI flow matching::
1215 FLEX_PARSER_PROFILE_ENABLE=4
1221 This driver relies on external libraries and kernel drivers for resources
1222 allocations and initialization. The following dependencies are not part of
1223 DPDK and must be installed separately:
1227 User space Verbs framework used by librte_net_mlx5. This library provides
1228 a generic interface between the kernel and low-level user space drivers
1231 It allows slow and privileged operations (context initialization, hardware
1232 resources allocations) to be managed by the kernel and fast operations to
1233 never leave user space.
1237 Low-level user space driver library for Mellanox
1238 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1241 This library basically implements send/receive calls to the hardware
1244 - **Kernel modules**
1246 They provide the kernel-side Verbs API and low level device drivers that
1247 manage actual hardware initialization and resources sharing with user
1250 Unlike most other PMDs, these modules must remain loaded and bound to
1253 - mlx5_core: hardware driver managing Mellanox
1254 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1256 - mlx5_ib: InifiniBand device driver.
1257 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1259 - **Firmware update**
1261 Mellanox OFED/EN releases include firmware updates for
1262 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1264 Because each release provides new features, these updates must be applied to
1265 match the kernel modules and libraries they come with.
1269 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1275 Either RDMA Core library with a recent enough Linux kernel release
1276 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1279 RDMA Core with Linux Kernel
1280 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1282 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1283 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1284 (see `RDMA Core installation documentation`_)
1285 - When building for i686 use:
1287 - rdma-core version 18.0 or above built with 32bit support.
1288 - Kernel version 4.14.41 or above.
1290 - Starting with rdma-core v21, static libraries can be built::
1293 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1296 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1297 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1303 - Mellanox OFED version: **4.5** and above /
1304 Mellanox EN version: **4.5** and above
1307 - ConnectX-4: **12.21.1000** and above.
1308 - ConnectX-4 Lx: **14.21.1000** and above.
1309 - ConnectX-5: **16.21.1000** and above.
1310 - ConnectX-5 Ex: **16.21.1000** and above.
1311 - ConnectX-6: **20.27.0090** and above.
1312 - ConnectX-6 Dx: **22.27.0090** and above.
1313 - BlueField: **18.25.1010** and above.
1315 While these libraries and kernel modules are available on OpenFabrics
1316 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1317 managers on most distributions, this PMD requires Ethernet extensions that
1318 may not be supported at the moment (this is a work in progress).
1321 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1323 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1324 include the necessary support and should be used in the meantime. For DPDK,
1325 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1326 required from that distribution.
1330 Several versions of Mellanox OFED/EN are available. Installing the version
1331 this DPDK release was developed and tested against is strongly
1332 recommended. Please check the `linux prerequisites`_.
1334 Windows Prerequisites
1335 ---------------------
1337 This driver relies on external libraries and kernel drivers for resources
1338 allocations and initialization. The dependencies in the following sub-sections
1339 are not part of DPDK, and must be installed separately.
1341 Compilation Prerequisites
1342 ~~~~~~~~~~~~~~~~~~~~~~~~~
1344 DevX SDK installation
1345 ^^^^^^^^^^^^^^^^^^^^^
1347 The DevX SDK must be installed on the machine building the Windows PMD.
1348 Additional information can be found at
1349 `How to Integrate Windows DevX in Your Development Environment
1350 <https://docs.mellanox.com/display/winof2v250/RShim+Drivers+and+Usage#RShimDriversandUsage-DevXInterface>`__.
1352 Runtime Prerequisites
1353 ~~~~~~~~~~~~~~~~~~~~~
1355 WinOF2 version 2.60 or higher must be installed on the machine.
1360 The driver can be downloaded from the following site:
1362 <https://www.mellanox.com/products/adapter-software/ethernet/windows/winof-2>`__
1367 DevX for Windows must be enabled in the Windows registry.
1368 The keys ``DevxEnabled`` and ``DevxFsRules`` must be set.
1369 Additional information can be found in the WinOF2 user manual.
1374 The following Mellanox device families are supported by the same mlx5 driver:
1386 Below are detailed device names:
1388 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1389 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1390 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1391 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1392 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1393 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1394 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1395 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1396 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1397 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1398 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1399 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1400 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1401 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1402 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1403 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1404 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1405 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1406 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1407 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1408 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1409 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1410 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1411 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1412 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1413 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1415 Quick Start Guide on OFED/EN
1416 ----------------------------
1418 1. Download latest Mellanox OFED/EN. For more info check the `linux prerequisites`_.
1421 2. Install the required libraries and kernel modules either by installing
1422 only the required set, or by installing the entire Mellanox OFED/EN::
1424 ./mlnxofedinstall --upstream-libs --dpdk
1426 3. Verify the firmware is the correct one::
1430 4. Verify all ports links are set to Ethernet::
1432 mlxconfig -d <mst device> query | grep LINK_TYPE
1436 Link types may have to be configured to Ethernet::
1438 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1440 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1442 For hypervisors, verify SR-IOV is enabled on the NIC::
1444 mlxconfig -d <mst device> query | grep SRIOV_EN
1447 If needed, configure SR-IOV::
1449 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1450 mlxfwreset -d <mst device> reset
1452 5. Restart the driver::
1454 /etc/init.d/openibd restart
1458 service openibd restart
1460 If link type was changed, firmware must be reset as well::
1462 mlxfwreset -d <mst device> reset
1464 For hypervisors, after reset write the sysfs number of virtual functions
1467 To dynamically instantiate a given number of virtual functions (VFs)::
1469 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1471 6. Install DPDK and you are ready to go.
1472 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1474 Enable switchdev mode
1475 ---------------------
1477 Switchdev mode is a mode in E-Switch, that binds between representor and VF or SF.
1478 Representor is a port in DPDK that is connected to a VF or SF in such a way
1479 that assuming there are no offload flows, each packet that is sent from the VF or SF
1480 will be received by the corresponding representor. While each packet that is or SF
1481 sent to a representor will be received by the VF or SF.
1482 This is very useful in case of SRIOV mode, where the first packet that is sent
1483 by the VF or SF will be received by the DPDK application which will decide if this
1484 flow should be offloaded to the E-Switch. After offloading the flow packet
1485 that the VF or SF that are matching the flow will not be received any more by
1486 the DPDK application.
1488 1. Enable SRIOV mode::
1490 mlxconfig -d <mst device> set SRIOV_EN=true
1492 2. Configure the max number of VFs::
1494 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1498 mlxfwreset -d <mst device> reset
1500 3. Configure the actual number of VFs::
1502 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1504 4. Unbind the device (can be rebind after the switchdev mode)::
1506 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1508 5. Enable switchdev mode::
1510 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1512 Sub-Function support
1513 --------------------
1515 Sub-Function is a portion of the PCI device, a SF netdev has its own
1516 dedicated queues (txq, rxq).
1517 A SF shares PCI level resources with other SFs and/or with its parent PCI function.
1521 OFED version >= 5.4-0.3.3.0
1523 1. Configure SF feature::
1525 # Run mlxconfig on both PFs on host and ECPFs on BlueField.
1526 mlxconfig -d <mst device> set PER_PF_NUM_SF=1 PF_TOTAL_SF=252 PF_SF_BAR_SIZE=12
1528 2. Enable switchdev mode::
1530 mlxdevm dev eswitch set pci/<DBDF> mode switchdev
1534 mlxdevm port add pci/<DBDF> flavour pcisf pfnum 0 sfnum <sfnum>
1536 Get SFID from output: pci/<DBDF>/<SFID>
1538 4. Modify MAC address::
1540 mlxdevm port function set pci/<DBDF>/<SFID> hw_addr <MAC>
1542 5. Activate SF port::
1544 mlxdevm port function set pci/<DBDF>/<ID> state active
1546 6. Devargs to probe SF device::
1548 auxiliary:mlx5_core.sf.<num>,dv_flow_en=1
1550 Sub-Function representor support
1551 --------------------------------
1553 A SF netdev supports E-Switch representation offload
1554 similar to PF and VF representors.
1555 Use <sfnum> to probe SF representor::
1557 testpmd> port attach <PCI_BDF>,representor=sf<sfnum>,dv_flow_en=1
1562 1. Configure aggressive CQE Zipping for maximum performance::
1564 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1566 To set it back to the default CQE Zipping mode use::
1568 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1570 2. In case of virtualization:
1572 - Make sure that hypervisor kernel is 3.16 or newer.
1573 - Configure boot with ``iommu=pt``.
1574 - Use 1G huge pages.
1575 - Make sure to allocate a VM on huge pages.
1576 - Make sure to set CPU pinning.
1578 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1579 for better performance. For VMs, verify that the right CPU
1580 and NUMA node are pinned according to the above. Run::
1582 lstopo-no-graphics --merge
1584 to identify the NUMA node to which the PCIe adapter is connected.
1586 4. If more than one adapter is used, and root complex capabilities allow
1587 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1588 it is recommended to locate both adapters on the same NUMA node.
1589 This in order to forward packets from one to the other without
1590 NUMA performance penalty.
1592 5. Disable pause frames::
1594 ethtool -A <netdev> rx off tx off
1596 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1597 via the BIOS configuration. Please contact you server provider for more
1598 information about the settings.
1602 On some machines, depends on the machine integrator, it is beneficial
1603 to set the PCI max read request parameter to 1K. This can be
1604 done in the following way:
1606 To query the read request size use::
1608 setpci -s <NIC PCI address> 68.w
1610 If the output is different than 3XXX, set it by::
1612 setpci -s <NIC PCI address> 68.w=3XXX
1614 The XXX can be different on different systems. Make sure to configure
1615 according to the setpci output.
1617 7. To minimize overhead of searching Memory Regions:
1619 - '--socket-mem' is recommended to pin memory by predictable amount.
1620 - Configure per-lcore cache when creating Mempools for packet buffer.
1621 - Refrain from dynamically allocating/freeing memory in run-time.
1626 There are multiple Rx burst functions with different advantages and limitations.
1628 .. table:: Rx burst functions
1630 +-------------------+------------------------+---------+-----------------+------+-------+
1631 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1632 | | | | || comp|| MTU |
1633 +===================+========================+=========+=================+======+=======+
1634 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1635 +-------------------+------------------------+---------+-----------------+------+-------+
1636 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1637 +-------------------+------------------------+---------+-----------------+------+-------+
1638 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1639 | || RxQs >= rxqs_min_mprq | | | | |
1640 +-------------------+------------------------+---------+-----------------+------+-------+
1641 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1642 | || mprq_en=1 | | | | |
1643 | || RxQs >= rxqs_min_mprq | | | | |
1644 +-------------------+------------------------+---------+-----------------+------+-------+
1646 .. _mlx5_offloads_support:
1648 Supported hardware offloads
1649 ---------------------------
1651 .. table:: Minimal SW/HW versions for queue offloads
1653 ============== ===== ===== ========= ===== ========== =============
1654 Offload DPDK Linux rdma-core OFED firmware hardware
1655 ============== ===== ===== ========= ===== ========== =============
1656 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1657 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1658 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1659 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1660 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1661 Tx scheduling 20.08 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1662 Buffer Split 20.11 N/A N/A 5.1-2 16.28.2006 ConnectX-5
1663 ============== ===== ===== ========= ===== ========== =============
1665 .. table:: Minimal SW/HW versions for rte_flow offloads
1667 +-----------------------+-----------------+-----------------+
1668 | Offload | with E-Switch | with NIC |
1669 +=======================+=================+=================+
1670 | Count | | DPDK 19.05 | | DPDK 19.02 |
1671 | | | OFED 4.6 | | OFED 4.6 |
1672 | | | rdma-core 24 | | rdma-core 23 |
1673 | | | ConnectX-5 | | ConnectX-5 |
1674 +-----------------------+-----------------+-----------------+
1675 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1676 | | | OFED 4.6 | | OFED 4.5 |
1677 | | | rdma-core 24 | | rdma-core 23 |
1678 | | | ConnectX-5 | | ConnectX-4 |
1679 +-----------------------+-----------------+-----------------+
1680 | Queue / RSS | | | | DPDK 18.11 |
1681 | | | N/A | | OFED 4.5 |
1682 | | | | | rdma-core 23 |
1683 | | | | | ConnectX-4 |
1684 +-----------------------+-----------------+-----------------+
1685 | Shared action | | | | |
1686 | | | :numref:`sact`| | :numref:`sact`|
1689 +-----------------------+-----------------+-----------------+
1690 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1691 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1692 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1693 | | of_set_vlan_pcp / | | | | |
1694 | | of_set_vlan_vid) | | | | |
1695 +-----------------------+-----------------+-----------------+
1696 | | VLAN | | DPDK 21.05 | | |
1697 | | ingress and / | | OFED 5.3 | | N/A |
1698 | | of_push_vlan / | | ConnectX-6 Dx | | |
1699 +-----------------------+-----------------+-----------------+
1700 | | VLAN | | DPDK 21.05 | | |
1701 | | egress and / | | OFED 5.3 | | N/A |
1702 | | of_pop_vlan / | | ConnectX-6 Dx | | |
1703 +-----------------------+-----------------+-----------------+
1704 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1705 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1706 | | | rdma-core 24 | | rdma-core 23 |
1707 | | | ConnectX-5 | | ConnectX-5 |
1708 +-----------------------+-----------------+-----------------+
1709 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1710 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1711 | | | rdma-core 27 | | rdma-core 27 |
1712 | | | ConnectX-5 | | ConnectX-5 |
1713 +-----------------------+-----------------+-----------------+
1714 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1715 | | | OFED 5.1-2 | | OFED 5.1-2 |
1716 | | | rdma-core 32 | | N/A |
1717 | | | ConnectX-5 | | ConnectX-5 |
1718 +-----------------------+-----------------+-----------------+
1719 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1720 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1721 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1722 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1723 | | set_ipv6_dst / | | | | |
1724 | | set_tp_src / | | | | |
1725 | | set_tp_dst / | | | | |
1726 | | dec_ttl / | | | | |
1727 | | set_ttl / | | | | |
1728 | | set_mac_src / | | | | |
1729 | | set_mac_dst) | | | | |
1730 +-----------------------+-----------------+-----------------+
1731 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1732 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1733 | | | | rdma-core 24 | | rdma-core 24 |
1734 | | | | ConnectX-5 | | ConnectX-5 |
1735 +-----------------------+-----------------+-----------------+
1736 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1737 | | | OFED 4.7-1 | | OFED 4.7-1 |
1738 | | | rdma-core 24 | | N/A |
1739 | | | ConnectX-5 | | ConnectX-5 |
1740 +-----------------------+-----------------+-----------------+
1741 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1742 | | | OFED 4.6 | | OFED 4.5 |
1743 | | | rdma-core 24 | | rdma-core 23 |
1744 | | | ConnectX-5 | | ConnectX-4 |
1745 +-----------------------+-----------------+-----------------+
1746 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1747 | | | OFED 4.7-3 | | OFED 4.7-3 |
1748 | | | rdma-core 26 | | rdma-core 26 |
1749 | | | ConnectX-5 | | ConnectX-5 |
1750 +-----------------------+-----------------+-----------------+
1751 | Port ID | | DPDK 19.05 | | N/A |
1752 | | | OFED 4.7-1 | | N/A |
1753 | | | rdma-core 24 | | N/A |
1754 | | | ConnectX-5 | | N/A |
1755 +-----------------------+-----------------+-----------------+
1756 | Hairpin | | | | DPDK 19.11 |
1757 | | | N/A | | OFED 4.7-3 |
1758 | | | | | rdma-core 26 |
1759 | | | | | ConnectX-5 |
1760 +-----------------------+-----------------+-----------------+
1761 | 2-port Hairpin | | | | DPDK 20.11 |
1762 | | | N/A | | OFED 5.1-2 |
1764 | | | | | ConnectX-5 |
1765 +-----------------------+-----------------+-----------------+
1766 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1767 | | | OFED 4.7-3 | | OFED 4.7-3 |
1768 | | | rdma-core 26 | | rdma-core 26 |
1769 | | | ConnectX-5 | | ConnectX-5 |
1770 +-----------------------+-----------------+-----------------+
1771 | ASO Metering | | DPDK 21.05 | | DPDK 21.05 |
1772 | | | OFED 5.3 | | OFED 5.3 |
1773 | | | rdma-core 33 | | rdma-core 33 |
1774 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1775 +-----------------------+-----------------+-----------------+
1776 | Metering Hierarchy | | DPDK 21.08 | | DPDK 21.08 |
1777 | | | OFED 5.3 | | OFED 5.3 |
1779 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1780 +-----------------------+-----------------+-----------------+
1781 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1782 | | | OFED 5.1-2 | | OFED 5.1-2 |
1783 | | | rdma-core 32 | | N/A |
1784 | | | ConnectX-5 | | ConnectX-5 |
1785 +-----------------------+-----------------+-----------------+
1786 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1787 | GTP PSC | | OFED 5.2 | | OFED 5.2 |
1788 | | | rdma-core 35 | | rdma-core 35 |
1789 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1790 +-----------------------+-----------------+-----------------+
1791 | Encapsulation | | DPDK 21.02 | | DPDK 21.02 |
1792 | GENEVE TLV option | | OFED 5.2 | | OFED 5.2 |
1793 | | | rdma-core 34 | | rdma-core 34 |
1794 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1795 +-----------------------+-----------------+-----------------+
1796 | Modify Field | | DPDK 21.02 | | DPDK 21.02 |
1797 | | | OFED 5.2 | | OFED 5.2 |
1798 | | | rdma-core 35 | | rdma-core 35 |
1799 | | | ConnectX-5 | | ConnectX-5 |
1800 +-----------------------+-----------------+-----------------+
1801 | Connection tracking | | | | DPDK 21.05 |
1802 | | | N/A | | OFED 5.3 |
1803 | | | | | rdma-core 35 |
1804 | | | | | ConnectX-6 Dx |
1805 +-----------------------+-----------------+-----------------+
1807 .. table:: Minimal SW/HW versions for shared action offload
1810 +-----------------------+-----------------+-----------------+
1811 | Shared Action | with E-Switch | with NIC |
1812 +=======================+=================+=================+
1813 | RSS | | | | DPDK 20.11 |
1814 | | | N/A | | OFED 5.2 |
1815 | | | | | rdma-core 33 |
1816 | | | | | ConnectX-5 |
1817 +-----------------------+-----------------+-----------------+
1818 | Age | | DPDK 20.11 | | DPDK 20.11 |
1819 | | | OFED 5.2 | | OFED 5.2 |
1820 | | | rdma-core 32 | | rdma-core 32 |
1821 | | | ConnectX-6 Dx | | ConnectX-6 Dx |
1822 +-----------------------+-----------------+-----------------+
1823 | Count | | DPDK 21.05 | | DPDK 21.05 |
1824 | | | OFED 4.6 | | OFED 4.6 |
1825 | | | rdma-core 24 | | rdma-core 23 |
1826 | | | ConnectX-5 | | ConnectX-5 |
1827 +-----------------------+-----------------+-----------------+
1832 MARK and META items are interrelated with datapath - they might move from/to
1833 the applications in mbuf fields. Hence, zero value for these items has the
1834 special meaning - it means "no metadata are provided", not zero values are
1835 treated by applications and PMD as valid ones.
1837 Moreover in the flow engine domain the value zero is acceptable to match and
1838 set, and we should allow to specify zero values as rte_flow parameters for the
1839 META and MARK items and actions. In the same time zero mask has no meaning and
1840 should be rejected on validation stage.
1845 Flows are not cached in the driver.
1846 When stopping a device port, all the flows created on this port from the
1847 application will be flushed automatically in the background.
1848 After stopping the device port, all flows on this port become invalid and
1849 not represented in the system.
1850 All references to these flows held by the application should be discarded
1851 directly but neither destroyed nor flushed.
1853 The application should re-create the flows as required after the port restart.
1858 Compared to librte_net_mlx4 that implements a single RSS configuration per
1859 port, librte_net_mlx5 supports per-protocol RSS configuration.
1861 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1862 command-line parameter to enable additional protocols (UDP and TCP as well
1863 as IP), the following commands must be entered from its CLI to get the same
1864 behavior as librte_net_mlx4::
1867 > port config all rss all
1873 This section demonstrates how to launch **testpmd** with Mellanox
1874 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1876 #. Load the kernel modules::
1878 modprobe -a ib_uverbs mlx5_core mlx5_ib
1880 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1883 /etc/init.d/openibd restart
1887 User space I/O kernel modules (uio and igb_uio) are not used and do
1888 not have to be loaded.
1890 #. Make sure Ethernet interfaces are in working order and linked to kernel
1891 verbs. Related sysfs entries should be present::
1893 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1902 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1905 for intf in eth2 eth3 eth4 eth5;
1907 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1910 sed -n 's,.*/\(.*\),-a \1,p'
1919 #. Request huge pages::
1921 dpdk-hugepages.py --setup 2G
1923 #. Start testpmd with basic parameters::
1925 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
1930 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1931 EAL: probe driver: 15b3:1013 librte_net_mlx5
1932 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1933 PMD: librte_net_mlx5: 1 port(s) detected
1934 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1935 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1936 EAL: probe driver: 15b3:1013 librte_net_mlx5
1937 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1938 PMD: librte_net_mlx5: 1 port(s) detected
1939 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1940 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1941 EAL: probe driver: 15b3:1013 librte_net_mlx5
1942 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1943 PMD: librte_net_mlx5: 1 port(s) detected
1944 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1945 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1946 EAL: probe driver: 15b3:1013 librte_net_mlx5
1947 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1948 PMD: librte_net_mlx5: 1 port(s) detected
1949 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1950 Interactive-mode selected
1951 Configuring Port 0 (socket 0)
1952 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1953 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1954 Port 0: E4:1D:2D:E7:0C:FE
1955 Configuring Port 1 (socket 0)
1956 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1957 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1958 Port 1: E4:1D:2D:E7:0C:FF
1959 Configuring Port 2 (socket 0)
1960 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1961 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1962 Port 2: E4:1D:2D:E7:0C:FA
1963 Configuring Port 3 (socket 0)
1964 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1965 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1966 Port 3: E4:1D:2D:E7:0C:FB
1967 Checking link statuses...
1968 Port 0 Link Up - speed 40000 Mbps - full-duplex
1969 Port 1 Link Up - speed 40000 Mbps - full-duplex
1970 Port 2 Link Up - speed 10000 Mbps - full-duplex
1971 Port 3 Link Up - speed 10000 Mbps - full-duplex
1978 This section demonstrates how to dump flows. Currently, it's possible to dump
1979 all flows with assistance of external tools.
1981 #. 2 ways to get flow raw file:
1983 - Using testpmd CLI:
1985 .. code-block:: console
1988 testpmd> flow dump <port> all <output_file>
1990 testpmd> flow dump <port> rule <rule_id> <output_file>
1992 - call rte_flow_dev_dump api:
1994 .. code-block:: console
1996 rte_flow_dev_dump(port, flow, file, NULL);
1998 #. Dump human-readable flows from raw file:
2000 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
2002 .. code-block:: console
2004 mlx_steering_dump.py -f <output_file> -flowptr <flow_ptr>
2006 How to share a meter between ports in the same switch domain
2007 ------------------------------------------------------------
2009 This section demonstrates how to use the shared meter. A meter M can be created
2010 on port X and to be shared with a port Y on the same switch domain by the next way:
2012 .. code-block:: console
2014 flow create X ingress transfer pattern eth / port_id id is Y / end actions meter mtr_id M / end
2016 How to use meter hierarchy
2017 --------------------------
2019 This section demonstrates how to create and use a meter hierarchy.
2020 A termination meter M can be the policy green action of another termination meter N.
2021 The two meters are chained together as a chain. Using meter N in a flow will apply
2022 both the meters in hierarchy on that flow.
2024 .. code-block:: console
2026 add port meter policy 0 1 g_actions queue index 0 / end y_actions end r_actions drop / end
2027 create port meter 0 M 1 1 yes 0xffff 1 0
2028 add port meter policy 0 2 g_actions meter mtr_id M / end y_actions end r_actions drop / end
2029 create port meter 0 N 2 2 yes 0xffff 1 0
2030 flow create 0 ingress group 1 pattern eth / end actions meter mtr_id N / end