1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright 2015 6WIND S.A.
3 Copyright 2015 Mellanox Technologies, Ltd
5 .. include:: <isonum.txt>
10 The MLX5 poll mode driver library (**librte_net_mlx5**) provides support
11 for **Mellanox ConnectX-4**, **Mellanox ConnectX-4 Lx** , **Mellanox
12 ConnectX-5**, **Mellanox ConnectX-6**, **Mellanox ConnectX-6 Dx**, **Mellanox
13 ConnectX-6 Lx**, **Mellanox BlueField** and **Mellanox BlueField-2** families
14 of 10/25/40/50/100/200 Gb/s adapters as well as their virtual functions (VF)
17 Information and documentation about these adapters can be found on the
18 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
19 `Mellanox community <http://community.mellanox.com/welcome>`__.
21 There is also a `section dedicated to this poll mode driver
22 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
28 Besides its dependency on libibverbs (that implies libmlx5 and associated
29 kernel support), librte_net_mlx5 relies heavily on system calls for control
30 operations such as querying/updating the MTU and flow control parameters.
32 For security reasons and robustness, this driver only deals with virtual
33 memory addresses. The way resources allocations are handled by the kernel,
34 combined with hardware specifications that allow to handle virtual memory
35 addresses directly, ensure that DPDK applications cannot access random
36 physical memory (or memory that does not belong to the current process).
38 This capability allows the PMD to coexist with kernel network interfaces
39 which remain functional, although they stop receiving unicast packets as
40 long as they share the same MAC address.
41 This means legacy linux control tools (for example: ethtool, ifconfig and
42 more) can operate on the same network interfaces that owned by the DPDK
45 The PMD can use libibverbs and libmlx5 to access the device firmware
46 or directly the hardware components.
47 There are different levels of objects and bypassing abilities
48 to get the best performances:
50 - Verbs is a complete high-level generic API
51 - Direct Verbs is a device-specific API
52 - DevX allows to access firmware objects
53 - Direct Rules manages flow steering at low-level hardware layer
55 Enabling librte_net_mlx5 causes DPDK applications to be linked against
61 - Multi arch support: x86_64, POWER8, ARMv8, i686.
62 - Multiple TX and RX queues.
63 - Support for scattered TX frames.
64 - Advanced support for scattered Rx frames with tunable buffer attributes.
65 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
66 - RSS using different combinations of fields: L3 only, L4 only or both,
67 and source only, destination only or both.
68 - Several RSS hash keys, one for each flow type.
69 - Default RSS operation with no hash key specification.
70 - Configurable RETA table.
71 - Link flow control (pause frame).
72 - Support for multiple MAC addresses.
76 - RX CRC stripping configuration.
77 - Promiscuous mode on PF and VF.
78 - Multicast promiscuous mode on PF and VF.
79 - Hardware checksum offloads.
80 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
82 - Flow API, including :ref:`flow_isolated_mode`.
84 - KVM and VMware ESX SR-IOV modes are supported.
85 - RSS hash result is supported.
86 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
87 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
89 - Statistics query including Basic, Extended and per queue.
91 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
92 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
93 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
94 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
95 - Flow insertion rate of more then million flows per second, when using Direct Rules.
96 - Support for multiple rte_flow groups.
97 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
100 - Multiple-thread flow insertion.
105 - For secondary process:
107 - Forked secondary process not supported.
108 - External memory unregistered in EAL memseg list cannot be used for DMA
109 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
110 primary process and remapped to the same virtual address in secondary
111 process. If the external memory is registered by primary process but has
112 different virtual address in secondary process, unexpected error may happen.
114 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
115 specific VLAN will match for VLAN packets as well:
117 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
118 Meaning, the flow rule::
120 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
122 Will only match vlan packets with vid=3. and the flow rule::
124 flow create 0 ingress pattern eth / ipv4 / end ...
126 Will match any ipv4 packet (VLAN included).
128 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
130 - 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.
133 flow create 0 ingress pattern eth / ipv4 / end ...
135 Will match any ipv4 packet.
138 flow create 0 ingress pattern eth / vlan / end ...
139 flow create 0 ingress pattern eth has_vlan is 1 / end ...
140 flow create 0 ingress pattern eth type is 0x8100 / end ...
142 Will match single-tagged packets only, with any VLAN ID value.
145 flow create 0 ingress pattern eth type is 0x88A8 / end ...
146 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
148 Will match multi-tagged packets only, with any VLAN ID value.
150 - A flow pattern with 2 sequential VLAN items is not supported.
152 - VLAN pop offload command:
154 - Flow rules having a VLAN pop offload command as one of their actions and
155 are lacking a match on VLAN as one of their items are not supported.
156 - The command is not supported on egress traffic.
158 - VLAN push offload is not supported on ingress traffic.
160 - VLAN set PCP offload is not supported on existing headers.
162 - A multi segment packet must have not more segments than reported by dev_infos_get()
163 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
164 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
165 inline settings) to 58.
167 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
168 to 0 are not supported.
170 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
172 - Match on Geneve header supports the following fields only:
178 Currently, the only supported options length value is 0.
180 - VF: flow rules created on VF devices can only match traffic targeted at the
181 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
183 - Match on GTP tunnel header item supports the following fields only:
185 - v_pt_rsv_flags: E flag, S flag, PN flag
189 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
190 The flows within group 0 and set metadata action are rejected by hardware.
194 MAC addresses not already present in the bridge table of the associated
195 kernel network device will be added and cleaned up by the PMD when closing
196 the device. In case of ungraceful program termination, some entries may
197 remain present and should be removed manually by other means.
199 - Buffer split offload is supported with regular Rx burst routine only,
200 no MPRQ feature or vectorized code can be engaged.
202 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
203 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
204 ol_flags. As the mempool for the external buffer is managed by PMD, all the
205 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
206 the external buffers will be freed by PMD and the application which still
207 holds the external buffers may be corrupted.
209 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
210 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
211 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
213 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
214 and allmulticast mode are both set to off.
215 To receive IPv6 Multicast messages on VM, explicitly set the relevant
216 MAC address using rte_eth_dev_mac_addr_add() API.
218 - To support a mixed traffic pattern (some buffers from local host memory, some
219 buffers from other devices) with high bandwidth, a mbuf flag is used.
221 An application hints the PMD whether or not it should try to inline the
222 given mbuf data buffer. PMD should do the best effort to act upon this request.
224 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
225 registered by application with rte_mbuf_dynflag_register(). This flag is
226 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
227 which is intended to be used by the application.
229 To query the supported specific flags in runtime,
230 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
231 currently (over present hardware and configuration) supported specific flags.
232 The "not inline hint" feature operating flow is the following one:
235 - probe the devices, ports are created
236 - query the port capabilities
237 - if port supporting the feature is found
238 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
239 - application starts the ports
240 - on ``dev_start()`` PMD checks whether the feature flag is registered and
241 enables the feature support in datapath
242 - application might set the registered flag bit in ``ol_flags`` field
243 of mbuf being sent and PMD will handle ones appropriately.
245 - The amount of descriptors in Tx queue may be limited by data inline settings.
246 Inline data require the more descriptor building blocks and overall block
247 amount may exceed the hardware supported limits. The application should
248 reduce the requested Tx size or adjust data inline settings with
249 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
251 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
252 parameter should be specified.
253 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
254 being sent it tries to synchronize the time of packet appearing on
255 the wire with the specified packet timestamp. It the specified one
256 is in the past it should be ignored, if one is in the distant future
257 it should be capped with some reasonable value (in range of seconds).
258 These specific cases ("too late" and "distant future") can be optionally
259 reported via device xstats to assist applications to detect the
260 time-related problems.
262 The timestamp upper "too-distant-future" limit
263 at the moment of invoking the Tx burst routine
264 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
265 Please note, for the testpmd txonly mode,
266 the limit is deduced from the expression::
268 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
270 There is no any packet reordering according timestamps is supposed,
271 neither within packet burst, nor between packets, it is an entirely
272 application responsibility to generate packets and its timestamps
273 in desired order. The timestamps can be put only in the first packet
274 in the burst providing the entire burst scheduling.
276 - E-Switch decapsulation Flow:
278 - can be applied to PF port only.
279 - must specify VF port action (packet redirection from PF to VF).
280 - optionally may specify tunnel inner source and destination MAC addresses.
282 - E-Switch encapsulation Flow:
284 - can be applied to VF ports only.
285 - must specify PF port action (packet redirection from VF to PF).
289 - The input buffer, used as outer header, is not validated.
293 - The decapsulation is always done up to the outermost tunnel detected by the HW.
294 - The input buffer, providing the removal size, is not validated.
295 - The buffer size must match the length of the headers to be removed.
297 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
298 mutually exclusive features which cannot be supported together
299 (see :ref:`mlx5_firmware_config`).
303 - Requires DevX and DV flow to be enabled.
304 - KEEP_CRC offload cannot be supported with LRO.
305 - The first mbuf length, without head-room, must be big enough to include the
307 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
308 it with size limited to max LRO size, not to max RX packet length.
309 - LRO can be used with outer header of TCP packets of the standard format:
310 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
312 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
313 - LRO packet aggregation is performed by HW only for packet size larger than
314 ``lro_min_mss_size``. This value is reported on device start, when debug
319 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
320 for some NICs (such as ConnectX-6 Dx, ConnectX-6 Lx, and BlueField-2).
321 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
325 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and E-Switch steering domain.
326 - The E-Switch Sample flow must have the eswitch_manager VPORT destination (PF or ECPF) and no additional actions.
327 - For ConnectX-5, the ``RTE_FLOW_ACTION_TYPE_SAMPLE`` is typically used as first action in the E-Switch egress flow if with header modify or encapsulation actions.
329 - IPv6 header item 'proto' field, indicating the next header protocol, should
330 not be set as extension header.
331 In case the next header is an extension header, it should not be specified in
332 IPv6 header item 'proto' field.
333 The last extension header item 'next header' field can specify the following
334 header protocol type.
338 - 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.
339 - Hairpin in switchdev SR-IOV mode is not supported till now.
344 MLX5 supports various methods to report statistics:
346 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.
348 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.
350 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.
358 The ibverbs libraries can be linked with this PMD in a number of ways,
359 configured by the ``ibverbs_link`` build option:
361 - ``shared`` (default): the PMD depends on some .so files.
363 - ``dlopen``: Split the dependencies glue in a separate library
364 loaded when needed by dlopen.
365 It make dependencies on libibverbs and libmlx4 optional,
366 and has no performance impact.
368 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
369 in the PMD shared library or the executable static binary.
371 Environment variables
372 ~~~~~~~~~~~~~~~~~~~~~
376 A list of directories in which to search for the rdma-core "glue" plug-in,
377 separated by colons or semi-colons.
379 - ``MLX5_SHUT_UP_BF``
381 Configures HW Tx doorbell register as IO-mapped.
383 By default, the HW Tx doorbell is configured as a write-combining register.
384 The register would be flushed to HW usually when the write-combining buffer
385 becomes full, but it depends on CPU design.
387 Except for vectorized Tx burst routines, a write memory barrier is enforced
388 after updating the register so that the update can be immediately visible to
391 When vectorized Tx burst is called, the barrier is set only if the burst size
392 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
393 variable will bring better latency even though the maximum throughput can
396 Run-time configuration
397 ~~~~~~~~~~~~~~~~~~~~~~
399 - librte_net_mlx5 brings kernel network interfaces up during initialization
400 because it is affected by their state. Forcing them down prevents packets
403 - **ethtool** operations on related kernel interfaces also affect the PMD.
408 In order to run as a non-root user,
409 some capabilities must be granted to the application::
411 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
413 Below are the reasons of the need for each capability:
416 When using physical addresses (PA mode), with Linux >= 4.0,
417 for access to ``/proc/self/pagemap``.
420 For device configuration.
423 For raw ethernet queue allocation through kernel driver.
426 For DMA memory pinning.
431 - ``rxq_cqe_comp_en`` parameter [int]
433 A nonzero value enables the compression of CQE on RX side. This feature
434 allows to save PCI bandwidth and improve performance. Enabled by default.
435 Different compression formats are supported in order to achieve the best
436 performance for different traffic patterns. Hash RSS format is the default.
438 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
439 better compression rate in case of RTE Flow Mark traffic.
440 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
441 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
442 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
446 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
447 ConnectX-6 Lx, BlueField and BlueField-2.
448 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
449 ConnectX-6 Lx, BlueField and BlueField-2.
451 - ``rxq_cqe_pad_en`` parameter [int]
453 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
454 is aligned with the size of a cacheline of the core. If cacheline size is
455 128B, the CQE size is configured to be 128B even though the device writes
456 only 64B data on the cacheline. This is to avoid unnecessary cache
457 invalidation by device's two consecutive writes on to one cacheline.
458 However in some architecture, it is more beneficial to update entire
459 cacheline with padding the rest 64B rather than striding because
460 read-modify-write could drop performance a lot. On the other hand,
461 writing extra data will consume more PCIe bandwidth and could also drop
462 the maximum throughput. It is recommended to empirically set this
463 parameter. Disabled by default.
467 - CPU having 128B cacheline with ConnectX-5 and BlueField.
469 - ``rxq_pkt_pad_en`` parameter [int]
471 A nonzero value enables padding Rx packet to the size of cacheline on PCI
472 transaction. This feature would waste PCI bandwidth but could improve
473 performance by avoiding partial cacheline write which may cause costly
474 read-modify-copy in memory transaction on some architectures. Disabled by
479 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
480 ConnectX-6 Lx, BlueField and BlueField-2.
481 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx,
482 ConnectX-6 Lx, BlueField and BlueField-2.
484 - ``mprq_en`` parameter [int]
486 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
487 configured as Multi-Packet RQ if the total number of Rx queues is
488 ``rxqs_min_mprq`` or more. Disabled by default.
490 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
491 by posting a single large buffer for multiple packets. Instead of posting a
492 buffers per a packet, one large buffer is posted in order to receive multiple
493 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
494 and each stride receives one packet. MPRQ can improve throughput for
495 small-packet traffic.
497 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
498 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
499 configure large stride size enough to accommodate max_rx_pkt_len as long as
500 device allows. Note that this can waste system memory compared to enabling Rx
501 scatter and multi-segment packet.
503 - ``mprq_log_stride_num`` parameter [int]
505 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
506 strides can reduce PCIe traffic further. If configured value is not in the
507 range of device capability, the default value will be set with a warning
508 message. The default value is 4 which is 16 strides per a buffer, valid only
509 if ``mprq_en`` is set.
511 The size of Rx queue should be bigger than the number of strides.
513 - ``mprq_log_stride_size`` parameter [int]
515 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
516 stride size can save some memory and reduce probability of a depletion of all
517 available strides due to unreleased packets by an application. If configured
518 value is not in the range of device capability, the default value will be set
519 with a warning message. The default value is 11 which is 2048 bytes per a
520 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
521 it is possible for a packet to span across multiple strides. This mode allows
522 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
523 of a packet if Rx scatter is configured) may be required in case there is no
524 space left for a head room at the end of a stride which incurs some
527 - ``mprq_max_memcpy_len`` parameter [int]
529 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
530 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
531 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
532 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
533 A mempool for external buffers will be allocated and managed by PMD. If Rx
534 packet is externally attached, ol_flags field of the mbuf will have
535 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
536 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
538 - ``rxqs_min_mprq`` parameter [int]
540 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
541 greater or equal to this value. The default value is 12, valid only if
544 - ``txq_inline`` parameter [int]
546 Amount of data to be inlined during TX operations. This parameter is
547 deprecated and converted to the new parameter ``txq_inline_max`` providing
548 partial compatibility.
550 - ``txqs_min_inline`` parameter [int]
552 Enable inline data send only when the number of TX queues is greater or equal
555 This option should be used in combination with ``txq_inline_max`` and
556 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
558 If this option is not specified the default value 16 is used for BlueField
559 and 8 for other platforms
561 The data inlining consumes the CPU cycles, so this option is intended to
562 auto enable inline data if we have enough Tx queues, which means we have
563 enough CPU cores and PCI bandwidth is getting more critical and CPU
564 is not supposed to be bottleneck anymore.
566 The copying data into WQE improves latency and can improve PPS performance
567 when PCI back pressure is detected and may be useful for scenarios involving
568 heavy traffic on many queues.
570 Because additional software logic is necessary to handle this mode, this
571 option should be used with care, as it may lower performance when back
572 pressure is not expected.
574 If inline data are enabled it may affect the maximal size of Tx queue in
575 descriptors because the inline data increase the descriptor size and
576 queue size limits supported by hardware may be exceeded.
578 - ``txq_inline_min`` parameter [int]
580 Minimal amount of data to be inlined into WQE during Tx operations. NICs
581 may require this minimal data amount to operate correctly. The exact value
582 may depend on NIC operation mode, requested offloads, etc. It is strongly
583 recommended to omit this parameter and use the default values. Anyway,
584 applications using this parameter should take into consideration that
585 specifying an inconsistent value may prevent the NIC from sending packets.
587 If ``txq_inline_min`` key is present the specified value (may be aligned
588 by the driver in order not to exceed the limits and provide better descriptor
589 space utilization) will be used by the driver and it is guaranteed that
590 requested amount of data bytes are inlined into the WQE beside other inline
591 settings. This key also may update ``txq_inline_max`` value (default
592 or specified explicitly in devargs) to reserve the space for inline data.
594 If ``txq_inline_min`` key is not present, the value may be queried by the
595 driver from the NIC via DevX if this feature is available. If there is no DevX
596 enabled/supported the value 18 (supposing L2 header including VLAN) is set
597 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
598 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
601 For ConnectX-4 NIC, driver does not allow specifying value below 18
602 (minimal L2 header, including VLAN), error will be raised.
604 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
605 it is not recommended and may prevent NIC from sending packets over
608 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
609 Multi-Packet Write), because last one does not support partial packet inlining.
610 This is not very critical due to minimal data inlining is mostly required
611 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
613 - ``txq_inline_max`` parameter [int]
615 Specifies the maximal packet length to be completely inlined into WQE
616 Ethernet Segment for ordinary SEND method. If packet is larger than specified
617 value, the packet data won't be copied by the driver at all, data buffer
618 is addressed with a pointer. If packet length is less or equal all packet
619 data will be copied into WQE. This may improve PCI bandwidth utilization for
620 short packets significantly but requires the extra CPU cycles.
622 The data inline feature is controlled by number of Tx queues, if number of Tx
623 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
624 is engaged, if there are not enough Tx queues (which means not enough CPU cores
625 and CPU resources are scarce), data inline is not performed by the driver.
626 Assigning ``txqs_min_inline`` with zero always enables the data inline.
628 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
629 by the driver in order not to exceed the limit (930 bytes) and to provide better
630 WQE space filling without gaps, the adjustment is reflected in the debug log.
631 Also, the default value (290) may be decreased in run-time if the large transmit
632 queue size is requested and hardware does not support enough descriptor
633 amount, in this case warning is emitted. If ``txq_inline_max`` key is
634 specified and requested inline settings can not be satisfied then error
637 - ``txq_inline_mpw`` parameter [int]
639 Specifies the maximal packet length to be completely inlined into WQE for
640 Enhanced MPW method. If packet is large the specified value, the packet data
641 won't be copied, and data buffer is addressed with pointer. If packet length
642 is less or equal, all packet data will be copied into WQE. This may improve PCI
643 bandwidth utilization for short packets significantly but requires the extra
646 The data inline feature is controlled by number of TX queues, if number of Tx
647 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
648 is engaged, if there are not enough Tx queues (which means not enough CPU cores
649 and CPU resources are scarce), data inline is not performed by the driver.
650 Assigning ``txqs_min_inline`` with zero always enables the data inline.
652 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
653 by the driver in order not to exceed the limit (930 bytes) and to provide better
654 WQE space filling without gaps, the adjustment is reflected in the debug log.
655 Due to multiple packets may be included to the same WQE with Enhanced Multi
656 Packet Write Method and overall WQE size is limited it is not recommended to
657 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
658 may be decreased in run-time if the large transmit queue size is requested
659 and hardware does not support enough descriptor amount, in this case warning
660 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
661 settings can not be satisfied then error will be raised.
663 - ``txqs_max_vec`` parameter [int]
665 Enable vectorized Tx only when the number of TX queues is less than or
666 equal to this value. This parameter is deprecated and ignored, kept
667 for compatibility issue to not prevent driver from probing.
669 - ``txq_mpw_hdr_dseg_en`` parameter [int]
671 A nonzero value enables including two pointers in the first block of TX
672 descriptor. The parameter is deprecated and ignored, kept for compatibility
675 - ``txq_max_inline_len`` parameter [int]
677 Maximum size of packet to be inlined. This limits the size of packet to
678 be inlined. If the size of a packet is larger than configured value, the
679 packet isn't inlined even though there's enough space remained in the
680 descriptor. Instead, the packet is included with pointer. This parameter
681 is deprecated and converted directly to ``txq_inline_mpw`` providing full
682 compatibility. Valid only if eMPW feature is engaged.
684 - ``txq_mpw_en`` parameter [int]
686 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
687 ConnectX-6, ConnectX-6 Dx, ConnectX-6 Lx, BlueField, BlueField-2.
688 eMPW allows the Tx burst function to pack up multiple packets
689 in a single descriptor session in order to save PCI bandwidth
690 and improve performance at the cost of a slightly higher CPU usage.
691 When ``txq_inline_mpw`` is set along with ``txq_mpw_en``,
692 Tx burst function copies entire packet data on to Tx descriptor
693 instead of including pointer of packet.
695 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
696 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
697 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
698 option or reported by the NIC, the eMPW feature is disengaged.
700 - ``tx_db_nc`` parameter [int]
702 The rdma core library can map doorbell register in two ways, depending on the
703 environment variable "MLX5_SHUT_UP_BF":
705 - As regular cached memory (usually with write combining attribute), if the
706 variable is either missing or set to zero.
707 - As non-cached memory, if the variable is present and set to not "0" value.
709 The type of mapping may slightly affect the Tx performance, the optimal choice
710 is strongly relied on the host architecture and should be deduced practically.
712 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
713 memory (with write combining), the PMD will perform the extra write memory barrier
714 after writing to doorbell, it might increase the needed CPU clocks per packet
715 to send, but latency might be improved.
717 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
718 cached memory, the PMD will not perform the extra write memory barrier
719 after writing to doorbell, on some architectures it might improve the
722 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
723 memory, the PMD will use heuristics to decide whether write memory barrier
724 should be performed. For bursts with size multiple of recommended one (64 pkts)
725 it is supposed the next burst is coming and no need to issue the extra memory
726 barrier (it is supposed to be issued in the next coming burst, at least after
727 descriptor writing). It might increase latency (on some hosts till next
728 packets transmit) and should be used with care.
730 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
731 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
732 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
734 - ``tx_pp`` parameter [int]
736 If a nonzero value is specified the driver creates all necessary internal
737 objects to provide accurate packet send scheduling on mbuf timestamps.
738 The positive value specifies the scheduling granularity in nanoseconds,
739 the packet send will be accurate up to specified digits. The allowed range is
740 from 500 to 1 million of nanoseconds. The negative value specifies the module
741 of granularity and engages the special test mode the check the schedule rate.
742 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
745 - ``tx_skew`` parameter [int]
747 The parameter adjusts the send packet scheduling on timestamps and represents
748 the average delay between beginning of the transmitting descriptor processing
749 by the hardware and appearance of actual packet data on the wire. The value
750 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
751 specified. The default value is zero.
753 - ``tx_vec_en`` parameter [int]
755 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx,
756 ConnectX-6 Lx, BlueField and BlueField-2 NICs
757 if the number of global Tx queues on the port is less than ``txqs_max_vec``.
758 The parameter is deprecated and ignored.
760 - ``rx_vec_en`` parameter [int]
762 A nonzero value enables Rx vector if the port is not configured in
763 multi-segment otherwise this parameter is ignored.
767 - ``vf_nl_en`` parameter [int]
769 A nonzero value enables Netlink requests from the VF to add/remove MAC
770 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
771 Otherwise the relevant configuration must be run with Linux iproute2 tools.
772 This is a prerequisite to receive this kind of traffic.
774 Enabled by default, valid only on VF devices ignored otherwise.
776 - ``l3_vxlan_en`` parameter [int]
778 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
779 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
780 parameter. This is a prerequisite to receive this kind of traffic.
784 - ``dv_xmeta_en`` parameter [int]
786 A nonzero value enables extensive flow metadata support if device is
787 capable and driver supports it. This can enable extensive support of
788 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
789 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
791 There are some possible configurations, depending on parameter value:
793 - 0, this is default value, defines the legacy mode, the ``MARK`` and
794 ``META`` related actions and items operate only within NIC Tx and
795 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
796 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
797 item is 32 bits wide and match supported on egress only.
799 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
800 related actions and items operate within all supported steering domains,
801 including FDB, ``MARK`` and ``META`` information may cross the domain
802 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
803 depends on kernel and firmware configurations and might be 0, 16 or
804 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
805 compatibility, the actual width of data transferred to the FDB domain
806 depends on kernel configuration and may be vary. The actual supported
807 width can be retrieved in runtime by series of rte_flow_validate()
810 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
811 related actions and items operate within all supported steering domains,
812 including FDB, ``MARK`` and ``META`` information may cross the domain
813 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
814 depends on kernel and firmware configurations and might be 0, 16 or
815 24 bits. The actual supported width can be retrieved in runtime by
816 series of rte_flow_validate() trials.
818 - 3, this engages tunnel offload mode. In E-Switch configuration, that
819 mode implicitly activates ``dv_xmeta_en=1``.
821 +------+-----------+-----------+-------------+-------------+
822 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
823 +======+===========+===========+=============+=============+
824 | 0 | 24 bits | 32 bits | 32 bits | no |
825 +------+-----------+-----------+-------------+-------------+
826 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
827 +------+-----------+-----------+-------------+-------------+
828 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
829 +------+-----------+-----------+-------------+-------------+
831 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
832 ignored and the device is configured to operate in legacy mode (0).
834 Disabled by default (set to 0).
836 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
837 of the extensive metadata features. The legacy Verbs supports FLAG and
838 MARK metadata actions over NIC Rx steering domain only.
840 - ``dv_flow_en`` parameter [int]
842 A nonzero value enables the DV flow steering assuming it is supported
843 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
845 Enabled by default if supported.
847 - ``dv_esw_en`` parameter [int]
849 A nonzero value enables E-Switch using Direct Rules.
851 Enabled by default if supported.
853 - ``lacp_by_user`` parameter [int]
855 A nonzero value enables the control of LACP traffic by the user application.
856 When a bond exists in the driver, by default it should be managed by the
857 kernel and therefore LACP traffic should be steered to the kernel.
858 If this devarg is set to 1 it will allow the user to manage the bond by
859 itself and not steer LACP traffic to the kernel.
861 Disabled by default (set to 0).
863 - ``mr_ext_memseg_en`` parameter [int]
865 A nonzero value enables extending memseg when registering DMA memory. If
866 enabled, the number of entries in MR (Memory Region) lookup table on datapath
867 is minimized and it benefits performance. On the other hand, it worsens memory
868 utilization because registered memory is pinned by kernel driver. Even if a
869 page in the extended chunk is freed, that doesn't become reusable until the
870 entire memory is freed.
874 - ``representor`` parameter [list]
876 This parameter can be used to instantiate DPDK Ethernet devices from
877 existing port (or VF) representors configured on the device.
879 It is a standard parameter whose format is described in
880 :ref:`ethernet_device_standard_device_arguments`.
882 For instance, to probe port representors 0 through 2::
886 - ``max_dump_files_num`` parameter [int]
888 The maximum number of files per PMD entity that may be created for debug information.
889 The files will be created in /var/log directory or in current directory.
891 set to 128 by default.
893 - ``lro_timeout_usec`` parameter [int]
895 The maximum allowed duration of an LRO session, in micro-seconds.
896 PMD will set the nearest value supported by HW, which is not bigger than
897 the input ``lro_timeout_usec`` value.
898 If this parameter is not specified, by default PMD will set
899 the smallest value supported by HW.
901 - ``hp_buf_log_sz`` parameter [int]
903 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
904 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
905 The capacity of the value is specified by the firmware and the initialization
906 will get a failure if it is out of scope.
907 The range of the value is from 11 to 19 right now, and the supported frame
908 size of a single packet for hairpin is from 512B to 128KB. It might change if
909 different firmware release is being used. By using a small value, it could
910 reduce memory consumption but not work with a large frame. If the value is
911 too large, the memory consumption will be high and some potential performance
912 degradation will be introduced.
913 By default, the PMD will set this value to 16, which means that 9KB jumbo
914 frames will be supported.
916 - ``reclaim_mem_mode`` parameter [int]
918 Cache some resources in flow destroy will help flow recreation more efficient.
919 While some systems may require the all the resources can be reclaimed after
921 The parameter ``reclaim_mem_mode`` provides the option for user to configure
922 if the resource cache is needed or not.
924 There are three options to choose:
926 - 0. It means the flow resources will be cached as usual. The resources will
927 be cached, helpful with flow insertion rate.
929 - 1. It will only enable the DPDK PMD level resources reclaim.
931 - 2. Both DPDK PMD level and rdma-core low level will be configured as
934 By default, the PMD will set this value to 0.
936 - ``sys_mem_en`` parameter [int]
938 A non-zero value enables the PMD memory management allocating memory
939 from system by default, without explicit rte memory flag.
941 By default, the PMD will set this value to 0.
943 - ``decap_en`` parameter [int]
945 Some devices do not support FCS (frame checksum) scattering for
946 tunnel-decapsulated packets.
947 If set to 0, this option forces the FCS feature and rejects tunnel
948 decapsulation in the flow engine for such devices.
950 By default, the PMD will set this value to 1.
952 .. _mlx5_firmware_config:
954 Firmware configuration
955 ~~~~~~~~~~~~~~~~~~~~~~
957 Firmware features can be configured as key/value pairs.
959 The command to set a value is::
961 mlxconfig -d <device> set <key>=<value>
963 The command to query a value is::
965 mlxconfig -d <device> query | grep <key>
967 The device name for the command ``mlxconfig`` can be either the PCI address,
968 or the mst device name found with::
972 Below are some firmware configurations listed.
978 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
984 - maximum number of SR-IOV virtual functions::
988 - enable DevX (required by Direct Rules and other features)::
992 - aggressive CQE zipping::
996 - L3 VXLAN and VXLAN-GPE destination UDP port::
999 IP_OVER_VXLAN_PORT=<udp dport>
1001 - enable VXLAN-GPE tunnel flow matching::
1003 FLEX_PARSER_PROFILE_ENABLE=0
1005 FLEX_PARSER_PROFILE_ENABLE=2
1007 - enable IP-in-IP tunnel flow matching::
1009 FLEX_PARSER_PROFILE_ENABLE=0
1011 - enable MPLS flow matching::
1013 FLEX_PARSER_PROFILE_ENABLE=1
1015 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1017 FLEX_PARSER_PROFILE_ENABLE=2
1019 - enable Geneve flow matching::
1021 FLEX_PARSER_PROFILE_ENABLE=0
1023 FLEX_PARSER_PROFILE_ENABLE=1
1025 - enable GTP flow matching::
1027 FLEX_PARSER_PROFILE_ENABLE=3
1029 - enable eCPRI flow matching::
1031 FLEX_PARSER_PROFILE_ENABLE=4
1037 This driver relies on external libraries and kernel drivers for resources
1038 allocations and initialization. The following dependencies are not part of
1039 DPDK and must be installed separately:
1043 User space Verbs framework used by librte_net_mlx5. This library provides
1044 a generic interface between the kernel and low-level user space drivers
1047 It allows slow and privileged operations (context initialization, hardware
1048 resources allocations) to be managed by the kernel and fast operations to
1049 never leave user space.
1053 Low-level user space driver library for Mellanox
1054 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1057 This library basically implements send/receive calls to the hardware
1060 - **Kernel modules**
1062 They provide the kernel-side Verbs API and low level device drivers that
1063 manage actual hardware initialization and resources sharing with user
1066 Unlike most other PMDs, these modules must remain loaded and bound to
1069 - mlx5_core: hardware driver managing Mellanox
1070 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1072 - mlx5_ib: InifiniBand device driver.
1073 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1075 - **Firmware update**
1077 Mellanox OFED/EN releases include firmware updates for
1078 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1080 Because each release provides new features, these updates must be applied to
1081 match the kernel modules and libraries they come with.
1085 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1091 Either RDMA Core library with a recent enough Linux kernel release
1092 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1095 RDMA Core with Linux Kernel
1096 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1098 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1099 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1100 (see `RDMA Core installation documentation`_)
1101 - When building for i686 use:
1103 - rdma-core version 18.0 or above built with 32bit support.
1104 - Kernel version 4.14.41 or above.
1106 - Starting with rdma-core v21, static libraries can be built::
1109 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1112 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1113 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1119 - Mellanox OFED version: **4.5** and above /
1120 Mellanox EN version: **4.5** and above
1123 - ConnectX-4: **12.21.1000** and above.
1124 - ConnectX-4 Lx: **14.21.1000** and above.
1125 - ConnectX-5: **16.21.1000** and above.
1126 - ConnectX-5 Ex: **16.21.1000** and above.
1127 - ConnectX-6: **20.27.0090** and above.
1128 - ConnectX-6 Dx: **22.27.0090** and above.
1129 - BlueField: **18.25.1010** and above.
1131 While these libraries and kernel modules are available on OpenFabrics
1132 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1133 managers on most distributions, this PMD requires Ethernet extensions that
1134 may not be supported at the moment (this is a work in progress).
1137 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1139 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1140 include the necessary support and should be used in the meantime. For DPDK,
1141 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1142 required from that distribution.
1146 Several versions of Mellanox OFED/EN are available. Installing the version
1147 this DPDK release was developed and tested against is strongly
1148 recommended. Please check the `prerequisites`_.
1153 The following Mellanox device families are supported by the same mlx5 driver:
1165 Below are detailed device names:
1167 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1168 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1169 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1170 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1171 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1172 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1173 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1174 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1175 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1176 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1177 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1178 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1179 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1180 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1181 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1182 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1183 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1184 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1185 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1186 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1187 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1188 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1189 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1190 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1191 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1192 * Mellanox\ |reg| ConnectX\ |reg|-6 Lx EN 25G MCX631102AN-ADAT (2x25G)
1194 Quick Start Guide on OFED/EN
1195 ----------------------------
1197 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
1200 2. Install the required libraries and kernel modules either by installing
1201 only the required set, or by installing the entire Mellanox OFED/EN::
1203 ./mlnxofedinstall --upstream-libs --dpdk
1205 3. Verify the firmware is the correct one::
1209 4. Verify all ports links are set to Ethernet::
1211 mlxconfig -d <mst device> query | grep LINK_TYPE
1215 Link types may have to be configured to Ethernet::
1217 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1219 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1221 For hypervisors, verify SR-IOV is enabled on the NIC::
1223 mlxconfig -d <mst device> query | grep SRIOV_EN
1226 If needed, configure SR-IOV::
1228 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1229 mlxfwreset -d <mst device> reset
1231 5. Restart the driver::
1233 /etc/init.d/openibd restart
1237 service openibd restart
1239 If link type was changed, firmware must be reset as well::
1241 mlxfwreset -d <mst device> reset
1243 For hypervisors, after reset write the sysfs number of virtual functions
1246 To dynamically instantiate a given number of virtual functions (VFs)::
1248 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1250 6. Install DPDK and you are ready to go.
1251 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1253 Enable switchdev mode
1254 ---------------------
1256 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1257 Representor is a port in DPDK that is connected to a VF in such a way
1258 that assuming there are no offload flows, each packet that is sent from the VF
1259 will be received by the corresponding representor. While each packet that is
1260 sent to a representor will be received by the VF.
1261 This is very useful in case of SRIOV mode, where the first packet that is sent
1262 by the VF will be received by the DPDK application which will decide if this
1263 flow should be offloaded to the E-Switch. After offloading the flow packet
1264 that the VF that are matching the flow will not be received any more by
1265 the DPDK application.
1267 1. Enable SRIOV mode::
1269 mlxconfig -d <mst device> set SRIOV_EN=true
1271 2. Configure the max number of VFs::
1273 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1277 mlxfwreset -d <mst device> reset
1279 3. Configure the actual number of VFs::
1281 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1283 4. Unbind the device (can be rebind after the switchdev mode)::
1285 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1287 5. Enbale switchdev mode::
1289 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1294 1. Configure aggressive CQE Zipping for maximum performance::
1296 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1298 To set it back to the default CQE Zipping mode use::
1300 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1302 2. In case of virtualization:
1304 - Make sure that hypervisor kernel is 3.16 or newer.
1305 - Configure boot with ``iommu=pt``.
1306 - Use 1G huge pages.
1307 - Make sure to allocate a VM on huge pages.
1308 - Make sure to set CPU pinning.
1310 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1311 for better performance. For VMs, verify that the right CPU
1312 and NUMA node are pinned according to the above. Run::
1316 to identify the NUMA node to which the PCIe adapter is connected.
1318 4. If more than one adapter is used, and root complex capabilities allow
1319 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1320 it is recommended to locate both adapters on the same NUMA node.
1321 This in order to forward packets from one to the other without
1322 NUMA performance penalty.
1324 5. Disable pause frames::
1326 ethtool -A <netdev> rx off tx off
1328 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1329 via the BIOS configuration. Please contact you server provider for more
1330 information about the settings.
1334 On some machines, depends on the machine integrator, it is beneficial
1335 to set the PCI max read request parameter to 1K. This can be
1336 done in the following way:
1338 To query the read request size use::
1340 setpci -s <NIC PCI address> 68.w
1342 If the output is different than 3XXX, set it by::
1344 setpci -s <NIC PCI address> 68.w=3XXX
1346 The XXX can be different on different systems. Make sure to configure
1347 according to the setpci output.
1349 7. To minimize overhead of searching Memory Regions:
1351 - '--socket-mem' is recommended to pin memory by predictable amount.
1352 - Configure per-lcore cache when creating Mempools for packet buffer.
1353 - Refrain from dynamically allocating/freeing memory in run-time.
1358 There are multiple Rx burst functions with different advantages and limitations.
1360 .. table:: Rx burst functions
1362 +-------------------+------------------------+---------+-----------------+------+-------+
1363 || Function Name || Enabler || Scatter|| Error Recovery || CQE || Large|
1364 | | | | || comp|| MTU |
1365 +===================+========================+=========+=================+======+=======+
1366 | rx_burst | rx_vec_en=0 | Yes | Yes | Yes | Yes |
1367 +-------------------+------------------------+---------+-----------------+------+-------+
1368 | rx_burst_vec | rx_vec_en=1 (default) | No | if CQE comp off | Yes | No |
1369 +-------------------+------------------------+---------+-----------------+------+-------+
1370 | rx_burst_mprq || mprq_en=1 | No | Yes | Yes | Yes |
1371 | || RxQs >= rxqs_min_mprq | | | | |
1372 +-------------------+------------------------+---------+-----------------+------+-------+
1373 | rx_burst_mprq_vec || rx_vec_en=1 (default) | No | if CQE comp off | Yes | Yes |
1374 | || mprq_en=1 | | | | |
1375 | || RxQs >= rxqs_min_mprq | | | | |
1376 +-------------------+------------------------+---------+-----------------+------+-------+
1378 .. _mlx5_offloads_support:
1380 Supported hardware offloads
1381 ---------------------------
1383 .. table:: Minimal SW/HW versions for queue offloads
1385 ============== ===== ===== ========= ===== ========== =============
1386 Offload DPDK Linux rdma-core OFED firmware hardware
1387 ============== ===== ===== ========= ===== ========== =============
1388 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1389 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1390 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1391 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1392 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1393 Buffer Split 20.11 N/A N/A 5.1-2 22.28.2006 ConnectX-6 Dx
1394 ============== ===== ===== ========= ===== ========== =============
1396 .. table:: Minimal SW/HW versions for rte_flow offloads
1398 +-----------------------+-----------------+-----------------+
1399 | Offload | with E-Switch | with NIC |
1400 +=======================+=================+=================+
1401 | Count | | DPDK 19.05 | | DPDK 19.02 |
1402 | | | OFED 4.6 | | OFED 4.6 |
1403 | | | rdma-core 24 | | rdma-core 23 |
1404 | | | ConnectX-5 | | ConnectX-5 |
1405 +-----------------------+-----------------+-----------------+
1406 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1407 | | | OFED 4.6 | | OFED 4.5 |
1408 | | | rdma-core 24 | | rdma-core 23 |
1409 | | | ConnectX-5 | | ConnectX-4 |
1410 +-----------------------+-----------------+-----------------+
1411 | Queue / RSS | | | | DPDK 18.11 |
1412 | | | N/A | | OFED 4.5 |
1413 | | | | | rdma-core 23 |
1414 | | | | | ConnectX-4 |
1415 +-----------------------+-----------------+-----------------+
1416 | RSS shared action | | | | DPDK 20.11 |
1417 | | | N/A | | OFED 5.2 |
1418 | | | | | rdma-core 33 |
1419 | | | | | ConnectX-5 |
1420 +-----------------------+-----------------+-----------------+
1421 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1422 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1423 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1424 | | of_set_vlan_pcp / | | | | |
1425 | | of_set_vlan_vid) | | | | |
1426 +-----------------------+-----------------+-----------------+
1427 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1428 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1429 | | | rdma-core 24 | | rdma-core 23 |
1430 | | | ConnectX-5 | | ConnectX-5 |
1431 +-----------------------+-----------------+-----------------+
1432 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1433 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1434 | | | rdma-core 27 | | rdma-core 27 |
1435 | | | ConnectX-5 | | ConnectX-5 |
1436 +-----------------------+-----------------+-----------------+
1437 | Tunnel Offload | | DPDK 20.11 | | DPDK 20.11 |
1438 | | | OFED 5.1-2 | | OFED 5.1-2 |
1439 | | | rdma-core 32 | | N/A |
1440 | | | ConnectX-5 | | ConnectX-5 |
1441 +-----------------------+-----------------+-----------------+
1442 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1443 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1444 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1445 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1446 | | set_ipv6_dst / | | | | |
1447 | | set_tp_src / | | | | |
1448 | | set_tp_dst / | | | | |
1449 | | dec_ttl / | | | | |
1450 | | set_ttl / | | | | |
1451 | | set_mac_src / | | | | |
1452 | | set_mac_dst) | | | | |
1453 +-----------------------+-----------------+-----------------+
1454 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1455 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1456 | | | | rdma-core 24 | | rdma-core 24 |
1457 | | | | ConnectX-5 | | ConnectX-5 |
1458 +-----------------------+-----------------+-----------------+
1459 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1460 | | | OFED 4.7-1 | | OFED 4.7-1 |
1461 | | | rdma-core 24 | | N/A |
1462 | | | ConnectX-5 | | ConnectX-5 |
1463 +-----------------------+-----------------+-----------------+
1464 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1465 | | | OFED 4.6 | | OFED 4.5 |
1466 | | | rdma-core 24 | | rdma-core 23 |
1467 | | | ConnectX-5 | | ConnectX-4 |
1468 +-----------------------+-----------------+-----------------+
1469 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1470 | | | OFED 4.7-3 | | OFED 4.7-3 |
1471 | | | rdma-core 26 | | rdma-core 26 |
1472 | | | ConnectX-5 | | ConnectX-5 |
1473 +-----------------------+-----------------+-----------------+
1474 | Port ID | | DPDK 19.05 | | N/A |
1475 | | | OFED 4.7-1 | | N/A |
1476 | | | rdma-core 24 | | N/A |
1477 | | | ConnectX-5 | | N/A |
1478 +-----------------------+-----------------+-----------------+
1479 | Hairpin | | | | DPDK 19.11 |
1480 | | | N/A | | OFED 4.7-3 |
1481 | | | | | rdma-core 26 |
1482 | | | | | ConnectX-5 |
1483 +-----------------------+-----------------+-----------------+
1484 | 2-port Hairpin | | | | DPDK 20.11 |
1485 | | | N/A | | OFED 5.1-2 |
1487 | | | | | ConnectX-5 |
1488 +-----------------------+-----------------+-----------------+
1489 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1490 | | | OFED 4.7-3 | | OFED 4.7-3 |
1491 | | | rdma-core 26 | | rdma-core 26 |
1492 | | | ConnectX-5 | | ConnectX-5 |
1493 +-----------------------+-----------------+-----------------+
1494 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1495 | | | OFED 5.1-2 | | OFED 5.1-2 |
1496 | | | rdma-core 32 | | N/A |
1497 | | | ConnectX-5 | | ConnectX-5 |
1498 +-----------------------+-----------------+-----------------+
1499 | Age shared action | | DPDK 20.11 | | DPDK 20.11 |
1500 | | | OFED 5.2 | | OFED 5.2 |
1501 | | | rdma-core 32 | | rdma-core 32 |
1502 | | | ConnectX-6 Dx| | ConnectX-6 Dx |
1503 +-----------------------+-----------------+-----------------+
1508 MARK and META items are interrelated with datapath - they might move from/to
1509 the applications in mbuf fields. Hence, zero value for these items has the
1510 special meaning - it means "no metadata are provided", not zero values are
1511 treated by applications and PMD as valid ones.
1513 Moreover in the flow engine domain the value zero is acceptable to match and
1514 set, and we should allow to specify zero values as rte_flow parameters for the
1515 META and MARK items and actions. In the same time zero mask has no meaning and
1516 should be rejected on validation stage.
1521 Flows are not cached in the driver.
1522 When stopping a device port, all the flows created on this port from the
1523 application will be flushed automatically in the background.
1524 After stopping the device port, all flows on this port become invalid and
1525 not represented in the system.
1526 All references to these flows held by the application should be discarded
1527 directly but neither destroyed nor flushed.
1529 The application should re-create the flows as required after the port restart.
1534 Compared to librte_net_mlx4 that implements a single RSS configuration per
1535 port, librte_net_mlx5 supports per-protocol RSS configuration.
1537 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1538 command-line parameter to enable additional protocols (UDP and TCP as well
1539 as IP), the following commands must be entered from its CLI to get the same
1540 behavior as librte_net_mlx4::
1543 > port config all rss all
1549 This section demonstrates how to launch **testpmd** with Mellanox
1550 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1552 #. Load the kernel modules::
1554 modprobe -a ib_uverbs mlx5_core mlx5_ib
1556 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1559 /etc/init.d/openibd restart
1563 User space I/O kernel modules (uio and igb_uio) are not used and do
1564 not have to be loaded.
1566 #. Make sure Ethernet interfaces are in working order and linked to kernel
1567 verbs. Related sysfs entries should be present::
1569 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1578 #. Optionally, retrieve their PCI bus addresses for to be used with the allow list::
1581 for intf in eth2 eth3 eth4 eth5;
1583 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1586 sed -n 's,.*/\(.*\),-a \1,p'
1595 #. Request huge pages::
1597 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1599 #. Start testpmd with basic parameters::
1601 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
1606 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1607 EAL: probe driver: 15b3:1013 librte_net_mlx5
1608 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1609 PMD: librte_net_mlx5: 1 port(s) detected
1610 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1611 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1612 EAL: probe driver: 15b3:1013 librte_net_mlx5
1613 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1614 PMD: librte_net_mlx5: 1 port(s) detected
1615 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1616 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1617 EAL: probe driver: 15b3:1013 librte_net_mlx5
1618 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1619 PMD: librte_net_mlx5: 1 port(s) detected
1620 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1621 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1622 EAL: probe driver: 15b3:1013 librte_net_mlx5
1623 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1624 PMD: librte_net_mlx5: 1 port(s) detected
1625 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1626 Interactive-mode selected
1627 Configuring Port 0 (socket 0)
1628 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1629 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1630 Port 0: E4:1D:2D:E7:0C:FE
1631 Configuring Port 1 (socket 0)
1632 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1633 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1634 Port 1: E4:1D:2D:E7:0C:FF
1635 Configuring Port 2 (socket 0)
1636 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1637 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1638 Port 2: E4:1D:2D:E7:0C:FA
1639 Configuring Port 3 (socket 0)
1640 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1641 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1642 Port 3: E4:1D:2D:E7:0C:FB
1643 Checking link statuses...
1644 Port 0 Link Up - speed 40000 Mbps - full-duplex
1645 Port 1 Link Up - speed 40000 Mbps - full-duplex
1646 Port 2 Link Up - speed 10000 Mbps - full-duplex
1647 Port 3 Link Up - speed 10000 Mbps - full-duplex
1654 This section demonstrates how to dump flows. Currently, it's possible to dump
1655 all flows with assistance of external tools.
1657 #. 2 ways to get flow raw file:
1659 - Using testpmd CLI:
1661 .. code-block:: console
1663 testpmd> flow dump <port> <output_file>
1665 - call rte_flow_dev_dump api:
1667 .. code-block:: console
1669 rte_flow_dev_dump(port, file, NULL);
1671 #. Dump human-readable flows from raw file:
1673 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1675 .. code-block:: console
1677 mlx_steering_dump.py -f <output_file>