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** and
13 **Mellanox BlueField** families of 10/25/40/50/100/200 Gb/s adapters
14 as well as their virtual functions (VF) in SR-IOV context.
16 Information and documentation about these adapters can be found on the
17 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
18 `Mellanox community <http://community.mellanox.com/welcome>`__.
20 There is also a `section dedicated to this poll mode driver
21 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
27 Besides its dependency on libibverbs (that implies libmlx5 and associated
28 kernel support), librte_net_mlx5 relies heavily on system calls for control
29 operations such as querying/updating the MTU and flow control parameters.
31 For security reasons and robustness, this driver only deals with virtual
32 memory addresses. The way resources allocations are handled by the kernel,
33 combined with hardware specifications that allow to handle virtual memory
34 addresses directly, ensure that DPDK applications cannot access random
35 physical memory (or memory that does not belong to the current process).
37 This capability allows the PMD to coexist with kernel network interfaces
38 which remain functional, although they stop receiving unicast packets as
39 long as they share the same MAC address.
40 This means legacy linux control tools (for example: ethtool, ifconfig and
41 more) can operate on the same network interfaces that owned by the DPDK
44 The PMD can use libibverbs and libmlx5 to access the device firmware
45 or directly the hardware components.
46 There are different levels of objects and bypassing abilities
47 to get the best performances:
49 - Verbs is a complete high-level generic API
50 - Direct Verbs is a device-specific API
51 - DevX allows to access firmware objects
52 - Direct Rules manages flow steering at low-level hardware layer
54 Enabling librte_net_mlx5 causes DPDK applications to be linked against
60 - Multi arch support: x86_64, POWER8, ARMv8, i686.
61 - Multiple TX and RX queues.
62 - Support for scattered TX frames.
63 - Advanced support for scattered Rx frames with tunable buffer attributes.
64 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
65 - RSS using different combinations of fields: L3 only, L4 only or both,
66 and source only, destination only or both.
67 - Several RSS hash keys, one for each flow type.
68 - Default RSS operation with no hash key specification.
69 - Configurable RETA table.
70 - Link flow control (pause frame).
71 - Support for multiple MAC addresses.
75 - RX CRC stripping configuration.
76 - Promiscuous mode on PF and VF.
77 - Multicast promiscuous mode on PF and VF.
78 - Hardware checksum offloads.
79 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
81 - Flow API, including :ref:`flow_isolated_mode`.
83 - KVM and VMware ESX SR-IOV modes are supported.
84 - RSS hash result is supported.
85 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
86 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
88 - Statistics query including Basic, Extended and per queue.
90 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
91 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
92 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
93 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
94 - Flow insertion rate of more then million flows per second, when using Direct Rules.
95 - Support for multiple rte_flow groups.
96 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
99 - Multiple-thread flow insertion.
104 - For secondary process:
106 - Forked secondary process not supported.
107 - External memory unregistered in EAL memseg list cannot be used for DMA
108 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
109 primary process and remapped to the same virtual address in secondary
110 process. If the external memory is registered by primary process but has
111 different virtual address in secondary process, unexpected error may happen.
113 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
114 specific VLAN will match for VLAN packets as well:
116 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
117 Meaning, the flow rule::
119 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
121 Will only match vlan packets with vid=3. and the flow rule::
123 flow create 0 ingress pattern eth / ipv4 / end ...
125 Will match any ipv4 packet (VLAN included).
127 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
129 - 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.
132 flow create 0 ingress pattern eth / ipv4 / end ...
134 Will match any ipv4 packet.
137 flow create 0 ingress pattern eth / vlan / end ...
138 flow create 0 ingress pattern eth has_vlan is 1 / end ...
139 flow create 0 ingress pattern eth type is 0x8100 / end ...
141 Will match single-tagged packets only, with any VLAN ID value.
144 flow create 0 ingress pattern eth type is 0x88A8 / end ...
145 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
147 Will match multi-tagged packets only, with any VLAN ID value.
149 - A flow pattern with 2 sequential VLAN items is not supported.
151 - VLAN pop offload command:
153 - Flow rules having a VLAN pop offload command as one of their actions and
154 are lacking a match on VLAN as one of their items are not supported.
155 - The command is not supported on egress traffic.
157 - VLAN push offload is not supported on ingress traffic.
159 - VLAN set PCP offload is not supported on existing headers.
161 - A multi segment packet must have not more segments than reported by dev_infos_get()
162 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
163 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
164 inline settings) to 58.
166 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
167 to 0 are not supported.
169 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
171 - Match on Geneve header supports the following fields only:
177 Currently, the only supported options length value is 0.
179 - VF: flow rules created on VF devices can only match traffic targeted at the
180 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
182 - Match on GTP tunnel header item supports the following fields only:
184 - v_pt_rsv_flags: E flag, S flag, PN flag
188 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
189 The flows within group 0 and set metadata action are rejected by hardware.
193 MAC addresses not already present in the bridge table of the associated
194 kernel network device will be added and cleaned up by the PMD when closing
195 the device. In case of ungraceful program termination, some entries may
196 remain present and should be removed manually by other means.
198 - Buffer split offload is supported with regular Rx burst routine only,
199 no MPRQ feature or vectorized code can be engaged.
201 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
202 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
203 ol_flags. As the mempool for the external buffer is managed by PMD, all the
204 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
205 the external buffers will be freed by PMD and the application which still
206 holds the external buffers may be corrupted.
208 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
209 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
210 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
212 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
213 and allmulticast mode are both set to off.
214 To receive IPv6 Multicast messages on VM, explicitly set the relevant
215 MAC address using rte_eth_dev_mac_addr_add() API.
217 - To support a mixed traffic pattern (some buffers from local host memory, some
218 buffers from other devices) with high bandwidth, a mbuf flag is used.
220 An application hints the PMD whether or not it should try to inline the
221 given mbuf data buffer. PMD should do the best effort to act upon this request.
223 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
224 registered by application with rte_mbuf_dynflag_register(). This flag is
225 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
226 which is intended to be used by the application.
228 To query the supported specific flags in runtime,
229 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
230 currently (over present hardware and configuration) supported specific flags.
231 The "not inline hint" feature operating flow is the following one:
234 - probe the devices, ports are created
235 - query the port capabilities
236 - if port supporting the feature is found
237 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
238 - application starts the ports
239 - on ``dev_start()`` PMD checks whether the feature flag is registered and
240 enables the feature support in datapath
241 - application might set the registered flag bit in ``ol_flags`` field
242 of mbuf being sent and PMD will handle ones appropriately.
244 - The amount of descriptors in Tx queue may be limited by data inline settings.
245 Inline data require the more descriptor building blocks and overall block
246 amount may exceed the hardware supported limits. The application should
247 reduce the requested Tx size or adjust data inline settings with
248 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
250 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
251 parameter should be specified.
252 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
253 being sent it tries to synchronize the time of packet appearing on
254 the wire with the specified packet timestamp. It the specified one
255 is in the past it should be ignored, if one is in the distant future
256 it should be capped with some reasonable value (in range of seconds).
257 These specific cases ("too late" and "distant future") can be optionally
258 reported via device xstats to assist applications to detect the
259 time-related problems.
261 The timestamp upper "too-distant-future" limit
262 at the moment of invoking the Tx burst routine
263 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
264 Please note, for the testpmd txonly mode,
265 the limit is deduced from the expression::
267 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
269 There is no any packet reordering according timestamps is supposed,
270 neither within packet burst, nor between packets, it is an entirely
271 application responsibility to generate packets and its timestamps
272 in desired order. The timestamps can be put only in the first packet
273 in the burst providing the entire burst scheduling.
275 - E-Switch decapsulation Flow:
277 - can be applied to PF port only.
278 - must specify VF port action (packet redirection from PF to VF).
279 - optionally may specify tunnel inner source and destination MAC addresses.
281 - E-Switch encapsulation Flow:
283 - can be applied to VF ports only.
284 - must specify PF port action (packet redirection from VF to PF).
288 - The input buffer, used as outer header, is not validated.
292 - The decapsulation is always done up to the outermost tunnel detected by the HW.
293 - The input buffer, providing the removal size, is not validated.
294 - The buffer size must match the length of the headers to be removed.
296 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
297 mutually exclusive features which cannot be supported together
298 (see :ref:`mlx5_firmware_config`).
302 - Requires DevX and DV flow to be enabled.
303 - KEEP_CRC offload cannot be supported with LRO.
304 - The first mbuf length, without head-room, must be big enough to include the
306 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
307 it with size limited to max LRO size, not to max RX packet length.
308 - LRO can be used with outer header of TCP packets of the standard format:
309 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
311 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
312 - LRO packet aggregation is performed by HW only for packet size larger than
313 ``lro_min_mss_size``. This value is reported on device start, when debug
318 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
319 for some NICs (such as ConnectX-6 Dx and BlueField 2).
320 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
324 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and E-Switch steering domain.
325 - The E-Switch Sample flow must have the eswitch_manager VPORT destination (PF or ECPF) and no additional actions.
326 - 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.
328 - IPv6 header item 'proto' field, indicating the next header protocol, should
329 not be set as extension header.
330 In case the next header is an extension header, it should not be specified in
331 IPv6 header item 'proto' field.
332 The last extension header item 'next header' field can specify the following
333 header protocol type.
337 - 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.
338 - Hairpin in switchdev SR-IOV mode is not supported till now.
343 MLX5 supports various methods to report statistics:
345 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.
347 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.
349 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.
357 The ibverbs libraries can be linked with this PMD in a number of ways,
358 configured by the ``ibverbs_link`` build option:
360 - ``shared`` (default): the PMD depends on some .so files.
362 - ``dlopen``: Split the dependencies glue in a separate library
363 loaded when needed by dlopen.
364 It make dependencies on libibverbs and libmlx4 optional,
365 and has no performance impact.
367 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
368 in the PMD shared library or the executable static binary.
370 Environment variables
371 ~~~~~~~~~~~~~~~~~~~~~
375 A list of directories in which to search for the rdma-core "glue" plug-in,
376 separated by colons or semi-colons.
378 - ``MLX5_SHUT_UP_BF``
380 Configures HW Tx doorbell register as IO-mapped.
382 By default, the HW Tx doorbell is configured as a write-combining register.
383 The register would be flushed to HW usually when the write-combining buffer
384 becomes full, but it depends on CPU design.
386 Except for vectorized Tx burst routines, a write memory barrier is enforced
387 after updating the register so that the update can be immediately visible to
390 When vectorized Tx burst is called, the barrier is set only if the burst size
391 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
392 variable will bring better latency even though the maximum throughput can
395 Run-time configuration
396 ~~~~~~~~~~~~~~~~~~~~~~
398 - librte_net_mlx5 brings kernel network interfaces up during initialization
399 because it is affected by their state. Forcing them down prevents packets
402 - **ethtool** operations on related kernel interfaces also affect the PMD.
407 In order to run as a non-root user,
408 some capabilities must be granted to the application::
410 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
412 Below are the reasons of the need for each capability:
415 When using physical addresses (PA mode), with Linux >= 4.0,
416 for access to ``/proc/self/pagemap``.
419 For device configuration.
422 For raw ethernet queue allocation through kernel driver.
425 For DMA memory pinning.
430 - ``rxq_cqe_comp_en`` parameter [int]
432 A nonzero value enables the compression of CQE on RX side. This feature
433 allows to save PCI bandwidth and improve performance. Enabled by default.
434 Different compression formats are supported in order to achieve the best
435 performance for different traffic patterns. Hash RSS format is the default.
437 Specifying 2 as a ``rxq_cqe_comp_en`` value selects Flow Tag format for
438 better compression rate in case of RTE Flow Mark traffic.
439 Specifying 3 as a ``rxq_cqe_comp_en`` value selects Checksum format.
440 Specifying 4 as a ``rxq_cqe_comp_en`` value selects L3/L4 Header format for
441 better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic.
445 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
447 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
450 - ``rxq_cqe_pad_en`` parameter [int]
452 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
453 is aligned with the size of a cacheline of the core. If cacheline size is
454 128B, the CQE size is configured to be 128B even though the device writes
455 only 64B data on the cacheline. This is to avoid unnecessary cache
456 invalidation by device's two consecutive writes on to one cacheline.
457 However in some architecture, it is more beneficial to update entire
458 cacheline with padding the rest 64B rather than striding because
459 read-modify-write could drop performance a lot. On the other hand,
460 writing extra data will consume more PCIe bandwidth and could also drop
461 the maximum throughput. It is recommended to empirically set this
462 parameter. Disabled by default.
466 - CPU having 128B cacheline with ConnectX-5 and BlueField.
468 - ``rxq_pkt_pad_en`` parameter [int]
470 A nonzero value enables padding Rx packet to the size of cacheline on PCI
471 transaction. This feature would waste PCI bandwidth but could improve
472 performance by avoiding partial cacheline write which may cause costly
473 read-modify-copy in memory transaction on some architectures. Disabled by
478 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
480 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
483 - ``mprq_en`` parameter [int]
485 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
486 configured as Multi-Packet RQ if the total number of Rx queues is
487 ``rxqs_min_mprq`` or more. Disabled by default.
489 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
490 by posting a single large buffer for multiple packets. Instead of posting a
491 buffers per a packet, one large buffer is posted in order to receive multiple
492 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
493 and each stride receives one packet. MPRQ can improve throughput for
494 small-packet traffic.
496 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
497 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
498 configure large stride size enough to accommodate max_rx_pkt_len as long as
499 device allows. Note that this can waste system memory compared to enabling Rx
500 scatter and multi-segment packet.
502 - ``mprq_log_stride_num`` parameter [int]
504 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
505 strides can reduce PCIe traffic further. If configured value is not in the
506 range of device capability, the default value will be set with a warning
507 message. The default value is 4 which is 16 strides per a buffer, valid only
508 if ``mprq_en`` is set.
510 The size of Rx queue should be bigger than the number of strides.
512 - ``mprq_log_stride_size`` parameter [int]
514 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
515 stride size can save some memory and reduce probability of a depletion of all
516 available strides due to unreleased packets by an application. If configured
517 value is not in the range of device capability, the default value will be set
518 with a warning message. The default value is 11 which is 2048 bytes per a
519 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
520 it is possible for a packet to span across multiple strides. This mode allows
521 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
522 of a packet if Rx scatter is configured) may be required in case there is no
523 space left for a head room at the end of a stride which incurs some
526 - ``mprq_max_memcpy_len`` parameter [int]
528 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
529 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
530 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
531 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
532 A mempool for external buffers will be allocated and managed by PMD. If Rx
533 packet is externally attached, ol_flags field of the mbuf will have
534 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
535 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
537 - ``rxqs_min_mprq`` parameter [int]
539 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
540 greater or equal to this value. The default value is 12, valid only if
543 - ``txq_inline`` parameter [int]
545 Amount of data to be inlined during TX operations. This parameter is
546 deprecated and converted to the new parameter ``txq_inline_max`` providing
547 partial compatibility.
549 - ``txqs_min_inline`` parameter [int]
551 Enable inline data send only when the number of TX queues is greater or equal
554 This option should be used in combination with ``txq_inline_max`` and
555 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
557 If this option is not specified the default value 16 is used for BlueField
558 and 8 for other platforms
560 The data inlining consumes the CPU cycles, so this option is intended to
561 auto enable inline data if we have enough Tx queues, which means we have
562 enough CPU cores and PCI bandwidth is getting more critical and CPU
563 is not supposed to be bottleneck anymore.
565 The copying data into WQE improves latency and can improve PPS performance
566 when PCI back pressure is detected and may be useful for scenarios involving
567 heavy traffic on many queues.
569 Because additional software logic is necessary to handle this mode, this
570 option should be used with care, as it may lower performance when back
571 pressure is not expected.
573 If inline data are enabled it may affect the maximal size of Tx queue in
574 descriptors because the inline data increase the descriptor size and
575 queue size limits supported by hardware may be exceeded.
577 - ``txq_inline_min`` parameter [int]
579 Minimal amount of data to be inlined into WQE during Tx operations. NICs
580 may require this minimal data amount to operate correctly. The exact value
581 may depend on NIC operation mode, requested offloads, etc. It is strongly
582 recommended to omit this parameter and use the default values. Anyway,
583 applications using this parameter should take into consideration that
584 specifying an inconsistent value may prevent the NIC from sending packets.
586 If ``txq_inline_min`` key is present the specified value (may be aligned
587 by the driver in order not to exceed the limits and provide better descriptor
588 space utilization) will be used by the driver and it is guaranteed that
589 requested amount of data bytes are inlined into the WQE beside other inline
590 settings. This key also may update ``txq_inline_max`` value (default
591 or specified explicitly in devargs) to reserve the space for inline data.
593 If ``txq_inline_min`` key is not present, the value may be queried by the
594 driver from the NIC via DevX if this feature is available. If there is no DevX
595 enabled/supported the value 18 (supposing L2 header including VLAN) is set
596 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
597 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
600 For ConnectX-4 NIC, driver does not allow specifying value below 18
601 (minimal L2 header, including VLAN), error will be raised.
603 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
604 it is not recommended and may prevent NIC from sending packets over
607 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
608 Multi-Packet Write), because last one does not support partial packet inlining.
609 This is not very critical due to minimal data inlining is mostly required
610 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
612 - ``txq_inline_max`` parameter [int]
614 Specifies the maximal packet length to be completely inlined into WQE
615 Ethernet Segment for ordinary SEND method. If packet is larger than specified
616 value, the packet data won't be copied by the driver at all, data buffer
617 is addressed with a pointer. If packet length is less or equal all packet
618 data will be copied into WQE. This may improve PCI bandwidth utilization for
619 short packets significantly but requires the extra CPU cycles.
621 The data inline feature is controlled by number of Tx queues, if number of Tx
622 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
623 is engaged, if there are not enough Tx queues (which means not enough CPU cores
624 and CPU resources are scarce), data inline is not performed by the driver.
625 Assigning ``txqs_min_inline`` with zero always enables the data inline.
627 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
628 by the driver in order not to exceed the limit (930 bytes) and to provide better
629 WQE space filling without gaps, the adjustment is reflected in the debug log.
630 Also, the default value (290) may be decreased in run-time if the large transmit
631 queue size is requested and hardware does not support enough descriptor
632 amount, in this case warning is emitted. If ``txq_inline_max`` key is
633 specified and requested inline settings can not be satisfied then error
636 - ``txq_inline_mpw`` parameter [int]
638 Specifies the maximal packet length to be completely inlined into WQE for
639 Enhanced MPW method. If packet is large the specified value, the packet data
640 won't be copied, and data buffer is addressed with pointer. If packet length
641 is less or equal, all packet data will be copied into WQE. This may improve PCI
642 bandwidth utilization for short packets significantly but requires the extra
645 The data inline feature is controlled by number of TX queues, if number of Tx
646 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
647 is engaged, if there are not enough Tx queues (which means not enough CPU cores
648 and CPU resources are scarce), data inline is not performed by the driver.
649 Assigning ``txqs_min_inline`` with zero always enables the data inline.
651 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
652 by the driver in order not to exceed the limit (930 bytes) and to provide better
653 WQE space filling without gaps, the adjustment is reflected in the debug log.
654 Due to multiple packets may be included to the same WQE with Enhanced Multi
655 Packet Write Method and overall WQE size is limited it is not recommended to
656 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
657 may be decreased in run-time if the large transmit queue size is requested
658 and hardware does not support enough descriptor amount, in this case warning
659 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
660 settings can not be satisfied then error will be raised.
662 - ``txqs_max_vec`` parameter [int]
664 Enable vectorized Tx only when the number of TX queues is less than or
665 equal to this value. This parameter is deprecated and ignored, kept
666 for compatibility issue to not prevent driver from probing.
668 - ``txq_mpw_hdr_dseg_en`` parameter [int]
670 A nonzero value enables including two pointers in the first block of TX
671 descriptor. The parameter is deprecated and ignored, kept for compatibility
674 - ``txq_max_inline_len`` parameter [int]
676 Maximum size of packet to be inlined. This limits the size of packet to
677 be inlined. If the size of a packet is larger than configured value, the
678 packet isn't inlined even though there's enough space remained in the
679 descriptor. Instead, the packet is included with pointer. This parameter
680 is deprecated and converted directly to ``txq_inline_mpw`` providing full
681 compatibility. Valid only if eMPW feature is engaged.
683 - ``txq_mpw_en`` parameter [int]
685 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
686 ConnectX-6, ConnectX-6 Dx and BlueField. eMPW allows the TX burst function to pack
687 up multiple packets in a single descriptor session in order to save PCI bandwidth
688 and improve performance at the cost of a slightly higher CPU usage. When
689 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
690 entire packet data on to TX descriptor instead of including pointer of packet.
692 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
693 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
694 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
695 option or reported by the NIC, the eMPW feature is disengaged.
697 - ``tx_db_nc`` parameter [int]
699 The rdma core library can map doorbell register in two ways, depending on the
700 environment variable "MLX5_SHUT_UP_BF":
702 - As regular cached memory (usually with write combining attribute), if the
703 variable is either missing or set to zero.
704 - As non-cached memory, if the variable is present and set to not "0" value.
706 The type of mapping may slightly affect the Tx performance, the optimal choice
707 is strongly relied on the host architecture and should be deduced practically.
709 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
710 memory (with write combining), the PMD will perform the extra write memory barrier
711 after writing to doorbell, it might increase the needed CPU clocks per packet
712 to send, but latency might be improved.
714 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
715 cached memory, the PMD will not perform the extra write memory barrier
716 after writing to doorbell, on some architectures it might improve the
719 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
720 memory, the PMD will use heuristics to decide whether write memory barrier
721 should be performed. For bursts with size multiple of recommended one (64 pkts)
722 it is supposed the next burst is coming and no need to issue the extra memory
723 barrier (it is supposed to be issued in the next coming burst, at least after
724 descriptor writing). It might increase latency (on some hosts till next
725 packets transmit) and should be used with care.
727 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
728 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
729 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
731 - ``tx_pp`` parameter [int]
733 If a nonzero value is specified the driver creates all necessary internal
734 objects to provide accurate packet send scheduling on mbuf timestamps.
735 The positive value specifies the scheduling granularity in nanoseconds,
736 the packet send will be accurate up to specified digits. The allowed range is
737 from 500 to 1 million of nanoseconds. The negative value specifies the module
738 of granularity and engages the special test mode the check the schedule rate.
739 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
742 - ``tx_skew`` parameter [int]
744 The parameter adjusts the send packet scheduling on timestamps and represents
745 the average delay between beginning of the transmitting descriptor processing
746 by the hardware and appearance of actual packet data on the wire. The value
747 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
748 specified. The default value is zero.
750 - ``tx_vec_en`` parameter [int]
752 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx
753 and BlueField NICs if the number of global Tx queues on the port is less than
754 ``txqs_max_vec``. The parameter is deprecated and ignored.
756 - ``rx_vec_en`` parameter [int]
758 A nonzero value enables Rx vector if the port is not configured in
759 multi-segment otherwise this parameter is ignored.
763 - ``vf_nl_en`` parameter [int]
765 A nonzero value enables Netlink requests from the VF to add/remove MAC
766 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
767 Otherwise the relevant configuration must be run with Linux iproute2 tools.
768 This is a prerequisite to receive this kind of traffic.
770 Enabled by default, valid only on VF devices ignored otherwise.
772 - ``l3_vxlan_en`` parameter [int]
774 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
775 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
776 parameter. This is a prerequisite to receive this kind of traffic.
780 - ``dv_xmeta_en`` parameter [int]
782 A nonzero value enables extensive flow metadata support if device is
783 capable and driver supports it. This can enable extensive support of
784 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
785 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
787 There are some possible configurations, depending on parameter value:
789 - 0, this is default value, defines the legacy mode, the ``MARK`` and
790 ``META`` related actions and items operate only within NIC Tx and
791 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
792 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
793 item is 32 bits wide and match supported on egress only.
795 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
796 related actions and items operate within all supported steering domains,
797 including FDB, ``MARK`` and ``META`` information may cross the domain
798 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
799 depends on kernel and firmware configurations and might be 0, 16 or
800 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
801 compatibility, the actual width of data transferred to the FDB domain
802 depends on kernel configuration and may be vary. The actual supported
803 width can be retrieved in runtime by series of rte_flow_validate()
806 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
807 related actions and items operate within all supported steering domains,
808 including FDB, ``MARK`` and ``META`` information may cross the domain
809 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
810 depends on kernel and firmware configurations and might be 0, 16 or
811 24 bits. The actual supported width can be retrieved in runtime by
812 series of rte_flow_validate() trials.
814 - 3, this engages tunnel offload mode. In E-Switch configuration, that
815 mode implicitly activates ``dv_xmeta_en=1``.
817 +------+-----------+-----------+-------------+-------------+
818 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
819 +======+===========+===========+=============+=============+
820 | 0 | 24 bits | 32 bits | 32 bits | no |
821 +------+-----------+-----------+-------------+-------------+
822 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
823 +------+-----------+-----------+-------------+-------------+
824 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
825 +------+-----------+-----------+-------------+-------------+
827 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
828 ignored and the device is configured to operate in legacy mode (0).
830 Disabled by default (set to 0).
832 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
833 of the extensive metadata features. The legacy Verbs supports FLAG and
834 MARK metadata actions over NIC Rx steering domain only.
836 - ``dv_flow_en`` parameter [int]
838 A nonzero value enables the DV flow steering assuming it is supported
839 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
841 Enabled by default if supported.
843 - ``dv_esw_en`` parameter [int]
845 A nonzero value enables E-Switch using Direct Rules.
847 Enabled by default if supported.
849 - ``lacp_by_user`` parameter [int]
851 A nonzero value enables the control of LACP traffic by the user application.
852 When a bond exists in the driver, by default it should be managed by the
853 kernel and therefore LACP traffic should be steered to the kernel.
854 If this devarg is set to 1 it will allow the user to manage the bond by
855 itself and not steer LACP traffic to the kernel.
857 Disabled by default (set to 0).
859 - ``mr_ext_memseg_en`` parameter [int]
861 A nonzero value enables extending memseg when registering DMA memory. If
862 enabled, the number of entries in MR (Memory Region) lookup table on datapath
863 is minimized and it benefits performance. On the other hand, it worsens memory
864 utilization because registered memory is pinned by kernel driver. Even if a
865 page in the extended chunk is freed, that doesn't become reusable until the
866 entire memory is freed.
870 - ``representor`` parameter [list]
872 This parameter can be used to instantiate DPDK Ethernet devices from
873 existing port (or VF) representors configured on the device.
875 It is a standard parameter whose format is described in
876 :ref:`ethernet_device_standard_device_arguments`.
878 For instance, to probe port representors 0 through 2::
882 - ``max_dump_files_num`` parameter [int]
884 The maximum number of files per PMD entity that may be created for debug information.
885 The files will be created in /var/log directory or in current directory.
887 set to 128 by default.
889 - ``lro_timeout_usec`` parameter [int]
891 The maximum allowed duration of an LRO session, in micro-seconds.
892 PMD will set the nearest value supported by HW, which is not bigger than
893 the input ``lro_timeout_usec`` value.
894 If this parameter is not specified, by default PMD will set
895 the smallest value supported by HW.
897 - ``hp_buf_log_sz`` parameter [int]
899 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
900 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
901 The capacity of the value is specified by the firmware and the initialization
902 will get a failure if it is out of scope.
903 The range of the value is from 11 to 19 right now, and the supported frame
904 size of a single packet for hairpin is from 512B to 128KB. It might change if
905 different firmware release is being used. By using a small value, it could
906 reduce memory consumption but not work with a large frame. If the value is
907 too large, the memory consumption will be high and some potential performance
908 degradation will be introduced.
909 By default, the PMD will set this value to 16, which means that 9KB jumbo
910 frames will be supported.
912 - ``reclaim_mem_mode`` parameter [int]
914 Cache some resources in flow destroy will help flow recreation more efficient.
915 While some systems may require the all the resources can be reclaimed after
917 The parameter ``reclaim_mem_mode`` provides the option for user to configure
918 if the resource cache is needed or not.
920 There are three options to choose:
922 - 0. It means the flow resources will be cached as usual. The resources will
923 be cached, helpful with flow insertion rate.
925 - 1. It will only enable the DPDK PMD level resources reclaim.
927 - 2. Both DPDK PMD level and rdma-core low level will be configured as
930 By default, the PMD will set this value to 0.
932 - ``sys_mem_en`` parameter [int]
934 A non-zero value enables the PMD memory management allocating memory
935 from system by default, without explicit rte memory flag.
937 By default, the PMD will set this value to 0.
939 - ``decap_en`` parameter [int]
941 Some devices do not support FCS (frame checksum) scattering for
942 tunnel-decapsulated packets.
943 If set to 0, this option forces the FCS feature and rejects tunnel
944 decapsulation in the flow engine for such devices.
946 By default, the PMD will set this value to 1.
948 .. _mlx5_firmware_config:
950 Firmware configuration
951 ~~~~~~~~~~~~~~~~~~~~~~
953 Firmware features can be configured as key/value pairs.
955 The command to set a value is::
957 mlxconfig -d <device> set <key>=<value>
959 The command to query a value is::
961 mlxconfig -d <device> query | grep <key>
963 The device name for the command ``mlxconfig`` can be either the PCI address,
964 or the mst device name found with::
968 Below are some firmware configurations listed.
974 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
980 - maximum number of SR-IOV virtual functions::
984 - enable DevX (required by Direct Rules and other features)::
988 - aggressive CQE zipping::
992 - L3 VXLAN and VXLAN-GPE destination UDP port::
995 IP_OVER_VXLAN_PORT=<udp dport>
997 - enable VXLAN-GPE tunnel flow matching::
999 FLEX_PARSER_PROFILE_ENABLE=0
1001 FLEX_PARSER_PROFILE_ENABLE=2
1003 - enable IP-in-IP tunnel flow matching::
1005 FLEX_PARSER_PROFILE_ENABLE=0
1007 - enable MPLS flow matching::
1009 FLEX_PARSER_PROFILE_ENABLE=1
1011 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1013 FLEX_PARSER_PROFILE_ENABLE=2
1015 - enable Geneve flow matching::
1017 FLEX_PARSER_PROFILE_ENABLE=0
1019 FLEX_PARSER_PROFILE_ENABLE=1
1021 - enable GTP flow matching::
1023 FLEX_PARSER_PROFILE_ENABLE=3
1025 - enable eCPRI flow matching::
1027 FLEX_PARSER_PROFILE_ENABLE=4
1033 This driver relies on external libraries and kernel drivers for resources
1034 allocations and initialization. The following dependencies are not part of
1035 DPDK and must be installed separately:
1039 User space Verbs framework used by librte_net_mlx5. This library provides
1040 a generic interface between the kernel and low-level user space drivers
1043 It allows slow and privileged operations (context initialization, hardware
1044 resources allocations) to be managed by the kernel and fast operations to
1045 never leave user space.
1049 Low-level user space driver library for Mellanox
1050 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1053 This library basically implements send/receive calls to the hardware
1056 - **Kernel modules**
1058 They provide the kernel-side Verbs API and low level device drivers that
1059 manage actual hardware initialization and resources sharing with user
1062 Unlike most other PMDs, these modules must remain loaded and bound to
1065 - mlx5_core: hardware driver managing Mellanox
1066 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1068 - mlx5_ib: InifiniBand device driver.
1069 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1071 - **Firmware update**
1073 Mellanox OFED/EN releases include firmware updates for
1074 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1076 Because each release provides new features, these updates must be applied to
1077 match the kernel modules and libraries they come with.
1081 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1087 Either RDMA Core library with a recent enough Linux kernel release
1088 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1091 RDMA Core with Linux Kernel
1092 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1094 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1095 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1096 (see `RDMA Core installation documentation`_)
1097 - When building for i686 use:
1099 - rdma-core version 18.0 or above built with 32bit support.
1100 - Kernel version 4.14.41 or above.
1102 - Starting with rdma-core v21, static libraries can be built::
1105 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1108 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1109 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1115 - Mellanox OFED version: **4.5** and above /
1116 Mellanox EN version: **4.5** and above
1119 - ConnectX-4: **12.21.1000** and above.
1120 - ConnectX-4 Lx: **14.21.1000** and above.
1121 - ConnectX-5: **16.21.1000** and above.
1122 - ConnectX-5 Ex: **16.21.1000** and above.
1123 - ConnectX-6: **20.27.0090** and above.
1124 - ConnectX-6 Dx: **22.27.0090** and above.
1125 - BlueField: **18.25.1010** and above.
1127 While these libraries and kernel modules are available on OpenFabrics
1128 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1129 managers on most distributions, this PMD requires Ethernet extensions that
1130 may not be supported at the moment (this is a work in progress).
1133 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1135 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1136 include the necessary support and should be used in the meantime. For DPDK,
1137 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1138 required from that distribution.
1142 Several versions of Mellanox OFED/EN are available. Installing the version
1143 this DPDK release was developed and tested against is strongly
1144 recommended. Please check the `prerequisites`_.
1149 The following Mellanox device families are supported by the same mlx5 driver:
1159 Below are detailed device names:
1161 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1162 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1163 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1164 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1165 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1166 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1167 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1168 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1169 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1170 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1171 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1172 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1173 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1174 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1175 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1176 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1177 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1178 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1179 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1180 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1181 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1182 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1183 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1184 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1185 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1187 Quick Start Guide on OFED/EN
1188 ----------------------------
1190 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
1193 2. Install the required libraries and kernel modules either by installing
1194 only the required set, or by installing the entire Mellanox OFED/EN::
1196 ./mlnxofedinstall --upstream-libs --dpdk
1198 3. Verify the firmware is the correct one::
1202 4. Verify all ports links are set to Ethernet::
1204 mlxconfig -d <mst device> query | grep LINK_TYPE
1208 Link types may have to be configured to Ethernet::
1210 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1212 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1214 For hypervisors, verify SR-IOV is enabled on the NIC::
1216 mlxconfig -d <mst device> query | grep SRIOV_EN
1219 If needed, configure SR-IOV::
1221 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1222 mlxfwreset -d <mst device> reset
1224 5. Restart the driver::
1226 /etc/init.d/openibd restart
1230 service openibd restart
1232 If link type was changed, firmware must be reset as well::
1234 mlxfwreset -d <mst device> reset
1236 For hypervisors, after reset write the sysfs number of virtual functions
1239 To dynamically instantiate a given number of virtual functions (VFs)::
1241 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1243 6. Install DPDK and you are ready to go.
1244 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1246 Enable switchdev mode
1247 ---------------------
1249 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1250 Representor is a port in DPDK that is connected to a VF in such a way
1251 that assuming there are no offload flows, each packet that is sent from the VF
1252 will be received by the corresponding representor. While each packet that is
1253 sent to a representor will be received by the VF.
1254 This is very useful in case of SRIOV mode, where the first packet that is sent
1255 by the VF will be received by the DPDK application which will decide if this
1256 flow should be offloaded to the E-Switch. After offloading the flow packet
1257 that the VF that are matching the flow will not be received any more by
1258 the DPDK application.
1260 1. Enable SRIOV mode::
1262 mlxconfig -d <mst device> set SRIOV_EN=true
1264 2. Configure the max number of VFs::
1266 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1270 mlxfwreset -d <mst device> reset
1272 3. Configure the actual number of VFs::
1274 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1276 4. Unbind the device (can be rebind after the switchdev mode)::
1278 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1280 5. Enbale switchdev mode::
1282 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1287 1. Configure aggressive CQE Zipping for maximum performance::
1289 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1291 To set it back to the default CQE Zipping mode use::
1293 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1295 2. In case of virtualization:
1297 - Make sure that hypervisor kernel is 3.16 or newer.
1298 - Configure boot with ``iommu=pt``.
1299 - Use 1G huge pages.
1300 - Make sure to allocate a VM on huge pages.
1301 - Make sure to set CPU pinning.
1303 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1304 for better performance. For VMs, verify that the right CPU
1305 and NUMA node are pinned according to the above. Run::
1309 to identify the NUMA node to which the PCIe adapter is connected.
1311 4. If more than one adapter is used, and root complex capabilities allow
1312 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1313 it is recommended to locate both adapters on the same NUMA node.
1314 This in order to forward packets from one to the other without
1315 NUMA performance penalty.
1317 5. Disable pause frames::
1319 ethtool -A <netdev> rx off tx off
1321 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1322 via the BIOS configuration. Please contact you server provider for more
1323 information about the settings.
1327 On some machines, depends on the machine integrator, it is beneficial
1328 to set the PCI max read request parameter to 1K. This can be
1329 done in the following way:
1331 To query the read request size use::
1333 setpci -s <NIC PCI address> 68.w
1335 If the output is different than 3XXX, set it by::
1337 setpci -s <NIC PCI address> 68.w=3XXX
1339 The XXX can be different on different systems. Make sure to configure
1340 according to the setpci output.
1342 7. To minimize overhead of searching Memory Regions:
1344 - '--socket-mem' is recommended to pin memory by predictable amount.
1345 - Configure per-lcore cache when creating Mempools for packet buffer.
1346 - Refrain from dynamically allocating/freeing memory in run-time.
1348 .. _mlx5_offloads_support:
1350 Supported hardware offloads
1351 ---------------------------
1353 .. table:: Minimal SW/HW versions for queue offloads
1355 ============== ===== ===== ========= ===== ========== ==========
1356 Offload DPDK Linux rdma-core OFED firmware hardware
1357 ============== ===== ===== ========= ===== ========== ==========
1358 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1359 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1360 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1361 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1362 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1363 ============== ===== ===== ========= ===== ========== ==========
1365 .. table:: Minimal SW/HW versions for rte_flow offloads
1367 +-----------------------+-----------------+-----------------+
1368 | Offload | with E-Switch | with NIC |
1369 +=======================+=================+=================+
1370 | Count | | DPDK 19.05 | | DPDK 19.02 |
1371 | | | OFED 4.6 | | OFED 4.6 |
1372 | | | rdma-core 24 | | rdma-core 23 |
1373 | | | ConnectX-5 | | ConnectX-5 |
1374 +-----------------------+-----------------+-----------------+
1375 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1376 | | | OFED 4.6 | | OFED 4.5 |
1377 | | | rdma-core 24 | | rdma-core 23 |
1378 | | | ConnectX-5 | | ConnectX-4 |
1379 +-----------------------+-----------------+-----------------+
1380 | Queue / RSS | | | | DPDK 18.11 |
1381 | | | N/A | | OFED 4.5 |
1382 | | | | | rdma-core 23 |
1383 | | | | | ConnectX-4 |
1384 +-----------------------+-----------------+-----------------+
1385 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1386 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1387 | | | rdma-core 24 | | rdma-core 23 |
1388 | | | ConnectX-5 | | ConnectX-5 |
1389 +-----------------------+-----------------+-----------------+
1390 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1391 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1392 | | | rdma-core 27 | | rdma-core 27 |
1393 | | | ConnectX-5 | | ConnectX-5 |
1394 +-----------------------+-----------------+-----------------+
1395 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1396 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1397 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1398 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1399 | | set_ipv6_dst / | | | | |
1400 | | set_tp_src / | | | | |
1401 | | set_tp_dst / | | | | |
1402 | | dec_ttl / | | | | |
1403 | | set_ttl / | | | | |
1404 | | set_mac_src / | | | | |
1405 | | set_mac_dst) | | | | |
1406 +-----------------------+-----------------+-----------------+
1407 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1408 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1409 | | | | rdma-core 24 | | rdma-core 24 |
1410 | | | | ConnectX-5 | | ConnectX-5 |
1411 +-----------------------+-----------------+-----------------+
1412 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1413 | | | OFED 4.7-1 | | OFED 4.7-1 |
1414 | | | rdma-core 24 | | N/A |
1415 | | | ConnectX-5 | | ConnectX-5 |
1416 +-----------------------+-----------------+-----------------+
1417 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1418 | | | OFED 4.6 | | OFED 4.5 |
1419 | | | rdma-core 24 | | rdma-core 23 |
1420 | | | ConnectX-5 | | ConnectX-4 |
1421 +-----------------------+-----------------+-----------------+
1422 | Port ID | | DPDK 19.05 | | N/A |
1423 | | | OFED 4.7-1 | | N/A |
1424 | | | rdma-core 24 | | N/A |
1425 | | | ConnectX-5 | | N/A |
1426 +-----------------------+-----------------+-----------------+
1427 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1428 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1429 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1430 | | of_set_vlan_pcp / | | | | |
1431 | | of_set_vlan_vid) | | | | |
1432 +-----------------------+-----------------+-----------------+
1433 | Hairpin | | | | DPDK 19.11 |
1434 | | | N/A | | OFED 4.7-3 |
1435 | | | | | rdma-core 26 |
1436 | | | | | ConnectX-5 |
1437 +-----------------------+-----------------+-----------------+
1438 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1439 | | | OFED 4.7-3 | | OFED 4.7-3 |
1440 | | | rdma-core 26 | | rdma-core 26 |
1441 | | | ConnectX-5 | | ConnectX-5 |
1442 +-----------------------+-----------------+-----------------+
1443 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1444 | | | OFED 4.7-3 | | OFED 4.7-3 |
1445 | | | rdma-core 26 | | rdma-core 26 |
1446 | | | ConnectX-5 | | ConnectX-5 |
1447 +-----------------------+-----------------+-----------------+
1448 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1449 | | | OFED 5.2 | | OFED 5.2 |
1450 | | | rdma-core 32 | | rdma-core 32 |
1451 | | | ConnectX-5 | | ConnectX-5 |
1452 +-----------------------+-----------------+-----------------+
1457 MARK and META items are interrelated with datapath - they might move from/to
1458 the applications in mbuf fields. Hence, zero value for these items has the
1459 special meaning - it means "no metadata are provided", not zero values are
1460 treated by applications and PMD as valid ones.
1462 Moreover in the flow engine domain the value zero is acceptable to match and
1463 set, and we should allow to specify zero values as rte_flow parameters for the
1464 META and MARK items and actions. In the same time zero mask has no meaning and
1465 should be rejected on validation stage.
1470 Flows are not cached in the driver.
1471 When stopping a device port, all the flows created on this port from the
1472 application will be flushed automatically in the background.
1473 After stopping the device port, all flows on this port become invalid and
1474 not represented in the system.
1475 All references to these flows held by the application should be discarded
1476 directly but neither destroyed nor flushed.
1478 The application should re-create the flows as required after the port restart.
1483 Compared to librte_net_mlx4 that implements a single RSS configuration per
1484 port, librte_net_mlx5 supports per-protocol RSS configuration.
1486 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1487 command-line parameter to enable additional protocols (UDP and TCP as well
1488 as IP), the following commands must be entered from its CLI to get the same
1489 behavior as librte_net_mlx4::
1492 > port config all rss all
1498 This section demonstrates how to launch **testpmd** with Mellanox
1499 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1501 #. Load the kernel modules::
1503 modprobe -a ib_uverbs mlx5_core mlx5_ib
1505 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1508 /etc/init.d/openibd restart
1512 User space I/O kernel modules (uio and igb_uio) are not used and do
1513 not have to be loaded.
1515 #. Make sure Ethernet interfaces are in working order and linked to kernel
1516 verbs. Related sysfs entries should be present::
1518 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1527 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1530 for intf in eth2 eth3 eth4 eth5;
1532 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1535 sed -n 's,.*/\(.*\),-w \1,p'
1544 #. Request huge pages::
1546 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1548 #. Start testpmd with basic parameters::
1550 testpmd -l 8-15 -n 4 -w 05:00.0 -w 05:00.1 -w 06:00.0 -w 06:00.1 -- --rxq=2 --txq=2 -i
1555 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1556 EAL: probe driver: 15b3:1013 librte_net_mlx5
1557 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1558 PMD: librte_net_mlx5: 1 port(s) detected
1559 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1560 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1561 EAL: probe driver: 15b3:1013 librte_net_mlx5
1562 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1563 PMD: librte_net_mlx5: 1 port(s) detected
1564 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1565 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1566 EAL: probe driver: 15b3:1013 librte_net_mlx5
1567 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1568 PMD: librte_net_mlx5: 1 port(s) detected
1569 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1570 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1571 EAL: probe driver: 15b3:1013 librte_net_mlx5
1572 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1573 PMD: librte_net_mlx5: 1 port(s) detected
1574 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1575 Interactive-mode selected
1576 Configuring Port 0 (socket 0)
1577 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1578 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1579 Port 0: E4:1D:2D:E7:0C:FE
1580 Configuring Port 1 (socket 0)
1581 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1582 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1583 Port 1: E4:1D:2D:E7:0C:FF
1584 Configuring Port 2 (socket 0)
1585 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1586 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1587 Port 2: E4:1D:2D:E7:0C:FA
1588 Configuring Port 3 (socket 0)
1589 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1590 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1591 Port 3: E4:1D:2D:E7:0C:FB
1592 Checking link statuses...
1593 Port 0 Link Up - speed 40000 Mbps - full-duplex
1594 Port 1 Link Up - speed 40000 Mbps - full-duplex
1595 Port 2 Link Up - speed 10000 Mbps - full-duplex
1596 Port 3 Link Up - speed 10000 Mbps - full-duplex
1603 This section demonstrates how to dump flows. Currently, it's possible to dump
1604 all flows with assistance of external tools.
1606 #. 2 ways to get flow raw file:
1608 - Using testpmd CLI:
1610 .. code-block:: console
1612 testpmd> flow dump <port> <output_file>
1614 - call rte_flow_dev_dump api:
1616 .. code-block:: console
1618 rte_flow_dev_dump(port, file, NULL);
1620 #. Dump human-readable flows from raw file:
1622 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1624 .. code-block:: console
1626 mlx_steering_dump.py -f <output_file>