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.
103 - For secondary process:
105 - Forked secondary process not supported.
106 - External memory unregistered in EAL memseg list cannot be used for DMA
107 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
108 primary process and remapped to the same virtual address in secondary
109 process. If the external memory is registered by primary process but has
110 different virtual address in secondary process, unexpected error may happen.
112 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
113 specific VLAN will match for VLAN packets as well:
115 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
116 Meaning, the flow rule::
118 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
120 Will only match vlan packets with vid=3. and the flow rule::
122 flow create 0 ingress pattern eth / ipv4 / end ...
124 Will match any ipv4 packet (VLAN included).
126 - When using Verbs flow engine (``dv_flow_en`` = 0), multi-tagged(QinQ) match is not supported.
128 - 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.
131 flow create 0 ingress pattern eth / ipv4 / end ...
133 Will match any ipv4 packet.
136 flow create 0 ingress pattern eth / vlan / end ...
137 flow create 0 ingress pattern eth has_vlan is 1 / end ...
138 flow create 0 ingress pattern eth type is 0x8100 / end ...
140 Will match single-tagged packets only, with any VLAN ID value.
143 flow create 0 ingress pattern eth type is 0x88A8 / end ...
144 flow create 0 ingress pattern eth / vlan has_more_vlan is 1 / end ...
146 Will match multi-tagged packets only, with any VLAN ID value.
148 - A flow pattern with 2 sequential VLAN items is not supported.
150 - VLAN pop offload command:
152 - Flow rules having a VLAN pop offload command as one of their actions and
153 are lacking a match on VLAN as one of their items are not supported.
154 - The command is not supported on egress traffic.
156 - VLAN push offload is not supported on ingress traffic.
158 - VLAN set PCP offload is not supported on existing headers.
160 - A multi segment packet must have not more segments than reported by dev_infos_get()
161 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
162 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
163 inline settings) to 58.
165 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
166 to 0 are not supported.
168 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
170 - Match on Geneve header supports the following fields only:
176 Currently, the only supported options length value is 0.
178 - VF: flow rules created on VF devices can only match traffic targeted at the
179 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
181 - Match on GTP tunnel header item supports the following fields only:
183 - v_pt_rsv_flags: E flag, S flag, PN flag
187 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
188 The flows within group 0 and set metadata action are rejected by hardware.
192 MAC addresses not already present in the bridge table of the associated
193 kernel network device will be added and cleaned up by the PMD when closing
194 the device. In case of ungraceful program termination, some entries may
195 remain present and should be removed manually by other means.
197 - Buffer split offload is supported with regular Rx burst routine only,
198 no MPRQ feature or vectorized code can be engaged.
200 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
201 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
202 ol_flags. As the mempool for the external buffer is managed by PMD, all the
203 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
204 the external buffers will be freed by PMD and the application which still
205 holds the external buffers may be corrupted.
207 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
208 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
209 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
211 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
212 and allmulticast mode are both set to off.
213 To receive IPv6 Multicast messages on VM, explicitly set the relevant
214 MAC address using rte_eth_dev_mac_addr_add() API.
216 - To support a mixed traffic pattern (some buffers from local host memory, some
217 buffers from other devices) with high bandwidth, a mbuf flag is used.
219 An application hints the PMD whether or not it should try to inline the
220 given mbuf data buffer. PMD should do the best effort to act upon this request.
222 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
223 registered by application with rte_mbuf_dynflag_register(). This flag is
224 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
225 which is intended to be used by the application.
227 To query the supported specific flags in runtime,
228 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
229 currently (over present hardware and configuration) supported specific flags.
230 The "not inline hint" feature operating flow is the following one:
233 - probe the devices, ports are created
234 - query the port capabilities
235 - if port supporting the feature is found
236 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
237 - application starts the ports
238 - on ``dev_start()`` PMD checks whether the feature flag is registered and
239 enables the feature support in datapath
240 - application might set the registered flag bit in ``ol_flags`` field
241 of mbuf being sent and PMD will handle ones appropriately.
243 - The amount of descriptors in Tx queue may be limited by data inline settings.
244 Inline data require the more descriptor building blocks and overall block
245 amount may exceed the hardware supported limits. The application should
246 reduce the requested Tx size or adjust data inline settings with
247 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
249 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
250 parameter should be specified.
251 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
252 being sent it tries to synchronize the time of packet appearing on
253 the wire with the specified packet timestamp. It the specified one
254 is in the past it should be ignored, if one is in the distant future
255 it should be capped with some reasonable value (in range of seconds).
256 These specific cases ("too late" and "distant future") can be optionally
257 reported via device xstats to assist applications to detect the
258 time-related problems.
260 The timestamp upper "too-distant-future" limit
261 at the moment of invoking the Tx burst routine
262 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
263 Please note, for the testpmd txonly mode,
264 the limit is deduced from the expression::
266 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
268 There is no any packet reordering according timestamps is supposed,
269 neither within packet burst, nor between packets, it is an entirely
270 application responsibility to generate packets and its timestamps
271 in desired order. The timestamps can be put only in the first packet
272 in the burst providing the entire burst scheduling.
274 - E-Switch decapsulation Flow:
276 - can be applied to PF port only.
277 - must specify VF port action (packet redirection from PF to VF).
278 - optionally may specify tunnel inner source and destination MAC addresses.
280 - E-Switch encapsulation Flow:
282 - can be applied to VF ports only.
283 - must specify PF port action (packet redirection from VF to PF).
287 - The input buffer, used as outer header, is not validated.
291 - The decapsulation is always done up to the outermost tunnel detected by the HW.
292 - The input buffer, providing the removal size, is not validated.
293 - The buffer size must match the length of the headers to be removed.
295 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
296 mutually exclusive features which cannot be supported together
297 (see :ref:`mlx5_firmware_config`).
301 - Requires DevX and DV flow to be enabled.
302 - KEEP_CRC offload cannot be supported with LRO.
303 - The first mbuf length, without head-room, must be big enough to include the
305 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
306 it with size limited to max LRO size, not to max RX packet length.
307 - LRO can be used with outer header of TCP packets of the standard format:
308 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
310 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
311 - LRO packet aggregation is performed by HW only for packet size larger than
312 ``lro_min_mss_size``. This value is reported on device start, when debug
317 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
318 for some NICs (such as ConnectX-6 Dx and BlueField 2).
319 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
323 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and E-Switch steering domain.
324 - The E-Switch Sample flow must have the eswitch_manager VPORT destination (PF or ECPF) and no additional actions.
325 - 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.
327 - IPv6 header item 'proto' field, indicating the next header protocol, should
328 not be set as extension header.
329 In case the next header is an extension header, it should not be specified in
330 IPv6 header item 'proto' field.
331 The last extension header item 'next header' field can specify the following
332 header protocol type.
336 - 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.
337 - Hairpin in switchdev SR-IOV mode is not supported till now.
342 MLX5 supports various methods to report statistics:
344 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.
346 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.
348 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.
356 The ibverbs libraries can be linked with this PMD in a number of ways,
357 configured by the ``ibverbs_link`` build option:
359 - ``shared`` (default): the PMD depends on some .so files.
361 - ``dlopen``: Split the dependencies glue in a separate library
362 loaded when needed by dlopen.
363 It make dependencies on libibverbs and libmlx4 optional,
364 and has no performance impact.
366 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
367 in the PMD shared library or the executable static binary.
369 Environment variables
370 ~~~~~~~~~~~~~~~~~~~~~
374 A list of directories in which to search for the rdma-core "glue" plug-in,
375 separated by colons or semi-colons.
377 - ``MLX5_SHUT_UP_BF``
379 Configures HW Tx doorbell register as IO-mapped.
381 By default, the HW Tx doorbell is configured as a write-combining register.
382 The register would be flushed to HW usually when the write-combining buffer
383 becomes full, but it depends on CPU design.
385 Except for vectorized Tx burst routines, a write memory barrier is enforced
386 after updating the register so that the update can be immediately visible to
389 When vectorized Tx burst is called, the barrier is set only if the burst size
390 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
391 variable will bring better latency even though the maximum throughput can
394 Run-time configuration
395 ~~~~~~~~~~~~~~~~~~~~~~
397 - librte_net_mlx5 brings kernel network interfaces up during initialization
398 because it is affected by their state. Forcing them down prevents packets
401 - **ethtool** operations on related kernel interfaces also affect the PMD.
406 In order to run as a non-root user,
407 some capabilities must be granted to the application::
409 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
411 Below are the reasons of the need for each capability:
414 When using physical addresses (PA mode), with Linux >= 4.0,
415 for access to ``/proc/self/pagemap``.
418 For device configuration.
421 For raw ethernet queue allocation through kernel driver.
424 For DMA memory pinning.
429 - ``rxq_cqe_comp_en`` parameter [int]
431 A nonzero value enables the compression of CQE on RX side. This feature
432 allows to save PCI bandwidth and improve performance. Enabled by default.
436 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
438 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
441 - ``rxq_cqe_pad_en`` parameter [int]
443 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
444 is aligned with the size of a cacheline of the core. If cacheline size is
445 128B, the CQE size is configured to be 128B even though the device writes
446 only 64B data on the cacheline. This is to avoid unnecessary cache
447 invalidation by device's two consecutive writes on to one cacheline.
448 However in some architecture, it is more beneficial to update entire
449 cacheline with padding the rest 64B rather than striding because
450 read-modify-write could drop performance a lot. On the other hand,
451 writing extra data will consume more PCIe bandwidth and could also drop
452 the maximum throughput. It is recommended to empirically set this
453 parameter. Disabled by default.
457 - CPU having 128B cacheline with ConnectX-5 and BlueField.
459 - ``rxq_pkt_pad_en`` parameter [int]
461 A nonzero value enables padding Rx packet to the size of cacheline on PCI
462 transaction. This feature would waste PCI bandwidth but could improve
463 performance by avoiding partial cacheline write which may cause costly
464 read-modify-copy in memory transaction on some architectures. Disabled by
469 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
471 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
474 - ``mprq_en`` parameter [int]
476 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
477 configured as Multi-Packet RQ if the total number of Rx queues is
478 ``rxqs_min_mprq`` or more. Disabled by default.
480 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
481 by posting a single large buffer for multiple packets. Instead of posting a
482 buffers per a packet, one large buffer is posted in order to receive multiple
483 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
484 and each stride receives one packet. MPRQ can improve throughput for
485 small-packet traffic.
487 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
488 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
489 configure large stride size enough to accommodate max_rx_pkt_len as long as
490 device allows. Note that this can waste system memory compared to enabling Rx
491 scatter and multi-segment packet.
493 - ``mprq_log_stride_num`` parameter [int]
495 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
496 strides can reduce PCIe traffic further. If configured value is not in the
497 range of device capability, the default value will be set with a warning
498 message. The default value is 4 which is 16 strides per a buffer, valid only
499 if ``mprq_en`` is set.
501 The size of Rx queue should be bigger than the number of strides.
503 - ``mprq_log_stride_size`` parameter [int]
505 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
506 stride size can save some memory and reduce probability of a depletion of all
507 available strides due to unreleased packets by an application. If configured
508 value is not in the range of device capability, the default value will be set
509 with a warning message. The default value is 11 which is 2048 bytes per a
510 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
511 it is possible for a packet to span across multiple strides. This mode allows
512 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
513 of a packet if Rx scatter is configured) may be required in case there is no
514 space left for a head room at the end of a stride which incurs some
517 - ``mprq_max_memcpy_len`` parameter [int]
519 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
520 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
521 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
522 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
523 A mempool for external buffers will be allocated and managed by PMD. If Rx
524 packet is externally attached, ol_flags field of the mbuf will have
525 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
526 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
528 - ``rxqs_min_mprq`` parameter [int]
530 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
531 greater or equal to this value. The default value is 12, valid only if
534 - ``txq_inline`` parameter [int]
536 Amount of data to be inlined during TX operations. This parameter is
537 deprecated and converted to the new parameter ``txq_inline_max`` providing
538 partial compatibility.
540 - ``txqs_min_inline`` parameter [int]
542 Enable inline data send only when the number of TX queues is greater or equal
545 This option should be used in combination with ``txq_inline_max`` and
546 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
548 If this option is not specified the default value 16 is used for BlueField
549 and 8 for other platforms
551 The data inlining consumes the CPU cycles, so this option is intended to
552 auto enable inline data if we have enough Tx queues, which means we have
553 enough CPU cores and PCI bandwidth is getting more critical and CPU
554 is not supposed to be bottleneck anymore.
556 The copying data into WQE improves latency and can improve PPS performance
557 when PCI back pressure is detected and may be useful for scenarios involving
558 heavy traffic on many queues.
560 Because additional software logic is necessary to handle this mode, this
561 option should be used with care, as it may lower performance when back
562 pressure is not expected.
564 If inline data are enabled it may affect the maximal size of Tx queue in
565 descriptors because the inline data increase the descriptor size and
566 queue size limits supported by hardware may be exceeded.
568 - ``txq_inline_min`` parameter [int]
570 Minimal amount of data to be inlined into WQE during Tx operations. NICs
571 may require this minimal data amount to operate correctly. The exact value
572 may depend on NIC operation mode, requested offloads, etc. It is strongly
573 recommended to omit this parameter and use the default values. Anyway,
574 applications using this parameter should take into consideration that
575 specifying an inconsistent value may prevent the NIC from sending packets.
577 If ``txq_inline_min`` key is present the specified value (may be aligned
578 by the driver in order not to exceed the limits and provide better descriptor
579 space utilization) will be used by the driver and it is guaranteed that
580 requested amount of data bytes are inlined into the WQE beside other inline
581 settings. This key also may update ``txq_inline_max`` value (default
582 or specified explicitly in devargs) to reserve the space for inline data.
584 If ``txq_inline_min`` key is not present, the value may be queried by the
585 driver from the NIC via DevX if this feature is available. If there is no DevX
586 enabled/supported the value 18 (supposing L2 header including VLAN) is set
587 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
588 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
591 For ConnectX-4 NIC, driver does not allow specifying value below 18
592 (minimal L2 header, including VLAN), error will be raised.
594 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
595 it is not recommended and may prevent NIC from sending packets over
598 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
599 Multi-Packet Write), because last one does not support partial packet inlining.
600 This is not very critical due to minimal data inlining is mostly required
601 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
603 - ``txq_inline_max`` parameter [int]
605 Specifies the maximal packet length to be completely inlined into WQE
606 Ethernet Segment for ordinary SEND method. If packet is larger than specified
607 value, the packet data won't be copied by the driver at all, data buffer
608 is addressed with a pointer. If packet length is less or equal all packet
609 data will be copied into WQE. This may improve PCI bandwidth utilization for
610 short packets significantly but requires the extra CPU cycles.
612 The data inline feature is controlled by number of Tx queues, if number of Tx
613 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
614 is engaged, if there are not enough Tx queues (which means not enough CPU cores
615 and CPU resources are scarce), data inline is not performed by the driver.
616 Assigning ``txqs_min_inline`` with zero always enables the data inline.
618 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
619 by the driver in order not to exceed the limit (930 bytes) and to provide better
620 WQE space filling without gaps, the adjustment is reflected in the debug log.
621 Also, the default value (290) may be decreased in run-time if the large transmit
622 queue size is requested and hardware does not support enough descriptor
623 amount, in this case warning is emitted. If ``txq_inline_max`` key is
624 specified and requested inline settings can not be satisfied then error
627 - ``txq_inline_mpw`` parameter [int]
629 Specifies the maximal packet length to be completely inlined into WQE for
630 Enhanced MPW method. If packet is large the specified value, the packet data
631 won't be copied, and data buffer is addressed with pointer. If packet length
632 is less or equal, all packet data will be copied into WQE. This may improve PCI
633 bandwidth utilization for short packets significantly but requires the extra
636 The data inline feature is controlled by number of TX queues, if number of Tx
637 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
638 is engaged, if there are not enough Tx queues (which means not enough CPU cores
639 and CPU resources are scarce), data inline is not performed by the driver.
640 Assigning ``txqs_min_inline`` with zero always enables the data inline.
642 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
643 by the driver in order not to exceed the limit (930 bytes) and to provide better
644 WQE space filling without gaps, the adjustment is reflected in the debug log.
645 Due to multiple packets may be included to the same WQE with Enhanced Multi
646 Packet Write Method and overall WQE size is limited it is not recommended to
647 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
648 may be decreased in run-time if the large transmit queue size is requested
649 and hardware does not support enough descriptor amount, in this case warning
650 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
651 settings can not be satisfied then error will be raised.
653 - ``txqs_max_vec`` parameter [int]
655 Enable vectorized Tx only when the number of TX queues is less than or
656 equal to this value. This parameter is deprecated and ignored, kept
657 for compatibility issue to not prevent driver from probing.
659 - ``txq_mpw_hdr_dseg_en`` parameter [int]
661 A nonzero value enables including two pointers in the first block of TX
662 descriptor. The parameter is deprecated and ignored, kept for compatibility
665 - ``txq_max_inline_len`` parameter [int]
667 Maximum size of packet to be inlined. This limits the size of packet to
668 be inlined. If the size of a packet is larger than configured value, the
669 packet isn't inlined even though there's enough space remained in the
670 descriptor. Instead, the packet is included with pointer. This parameter
671 is deprecated and converted directly to ``txq_inline_mpw`` providing full
672 compatibility. Valid only if eMPW feature is engaged.
674 - ``txq_mpw_en`` parameter [int]
676 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
677 ConnectX-6, ConnectX-6 Dx and BlueField. eMPW allows the TX burst function to pack
678 up multiple packets in a single descriptor session in order to save PCI bandwidth
679 and improve performance at the cost of a slightly higher CPU usage. When
680 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
681 entire packet data on to TX descriptor instead of including pointer of packet.
683 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
684 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
685 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
686 option or reported by the NIC, the eMPW feature is disengaged.
688 - ``tx_db_nc`` parameter [int]
690 The rdma core library can map doorbell register in two ways, depending on the
691 environment variable "MLX5_SHUT_UP_BF":
693 - As regular cached memory (usually with write combining attribute), if the
694 variable is either missing or set to zero.
695 - As non-cached memory, if the variable is present and set to not "0" value.
697 The type of mapping may slightly affect the Tx performance, the optimal choice
698 is strongly relied on the host architecture and should be deduced practically.
700 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
701 memory (with write combining), the PMD will perform the extra write memory barrier
702 after writing to doorbell, it might increase the needed CPU clocks per packet
703 to send, but latency might be improved.
705 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
706 cached memory, the PMD will not perform the extra write memory barrier
707 after writing to doorbell, on some architectures it might improve the
710 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
711 memory, the PMD will use heuristics to decide whether write memory barrier
712 should be performed. For bursts with size multiple of recommended one (64 pkts)
713 it is supposed the next burst is coming and no need to issue the extra memory
714 barrier (it is supposed to be issued in the next coming burst, at least after
715 descriptor writing). It might increase latency (on some hosts till next
716 packets transmit) and should be used with care.
718 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
719 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
720 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
722 - ``tx_pp`` parameter [int]
724 If a nonzero value is specified the driver creates all necessary internal
725 objects to provide accurate packet send scheduling on mbuf timestamps.
726 The positive value specifies the scheduling granularity in nanoseconds,
727 the packet send will be accurate up to specified digits. The allowed range is
728 from 500 to 1 million of nanoseconds. The negative value specifies the module
729 of granularity and engages the special test mode the check the schedule rate.
730 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
733 - ``tx_skew`` parameter [int]
735 The parameter adjusts the send packet scheduling on timestamps and represents
736 the average delay between beginning of the transmitting descriptor processing
737 by the hardware and appearance of actual packet data on the wire. The value
738 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
739 specified. The default value is zero.
741 - ``tx_vec_en`` parameter [int]
743 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx
744 and BlueField NICs if the number of global Tx queues on the port is less than
745 ``txqs_max_vec``. The parameter is deprecated and ignored.
747 - ``rx_vec_en`` parameter [int]
749 A nonzero value enables Rx vector if the port is not configured in
750 multi-segment otherwise this parameter is ignored.
754 - ``vf_nl_en`` parameter [int]
756 A nonzero value enables Netlink requests from the VF to add/remove MAC
757 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
758 Otherwise the relevant configuration must be run with Linux iproute2 tools.
759 This is a prerequisite to receive this kind of traffic.
761 Enabled by default, valid only on VF devices ignored otherwise.
763 - ``l3_vxlan_en`` parameter [int]
765 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
766 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
767 parameter. This is a prerequisite to receive this kind of traffic.
771 - ``dv_xmeta_en`` parameter [int]
773 A nonzero value enables extensive flow metadata support if device is
774 capable and driver supports it. This can enable extensive support of
775 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
776 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
778 There are some possible configurations, depending on parameter value:
780 - 0, this is default value, defines the legacy mode, the ``MARK`` and
781 ``META`` related actions and items operate only within NIC Tx and
782 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
783 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
784 item is 32 bits wide and match supported on egress only.
786 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
787 related actions and items operate within all supported steering domains,
788 including FDB, ``MARK`` and ``META`` information may cross the domain
789 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
790 depends on kernel and firmware configurations and might be 0, 16 or
791 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
792 compatibility, the actual width of data transferred to the FDB domain
793 depends on kernel configuration and may be vary. The actual supported
794 width can be retrieved in runtime by series of rte_flow_validate()
797 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
798 related actions and items operate within all supported steering domains,
799 including FDB, ``MARK`` and ``META`` information may cross the domain
800 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
801 depends on kernel and firmware configurations and might be 0, 16 or
802 24 bits. The actual supported width can be retrieved in runtime by
803 series of rte_flow_validate() trials.
805 - 3, this engages tunnel offload mode. In E-Switch configuration, that
806 mode implicitly activates ``dv_xmeta_en=1``.
808 +------+-----------+-----------+-------------+-------------+
809 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
810 +======+===========+===========+=============+=============+
811 | 0 | 24 bits | 32 bits | 32 bits | no |
812 +------+-----------+-----------+-------------+-------------+
813 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
814 +------+-----------+-----------+-------------+-------------+
815 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
816 +------+-----------+-----------+-------------+-------------+
818 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
819 ignored and the device is configured to operate in legacy mode (0).
821 Disabled by default (set to 0).
823 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
824 of the extensive metadata features. The legacy Verbs supports FLAG and
825 MARK metadata actions over NIC Rx steering domain only.
827 - ``dv_flow_en`` parameter [int]
829 A nonzero value enables the DV flow steering assuming it is supported
830 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
832 Enabled by default if supported.
834 - ``dv_esw_en`` parameter [int]
836 A nonzero value enables E-Switch using Direct Rules.
838 Enabled by default if supported.
840 - ``lacp_by_user`` parameter [int]
842 A nonzero value enables the control of LACP traffic by the user application.
843 When a bond exists in the driver, by default it should be managed by the
844 kernel and therefore LACP traffic should be steered to the kernel.
845 If this devarg is set to 1 it will allow the user to manage the bond by
846 itself and not steer LACP traffic to the kernel.
848 Disabled by default (set to 0).
850 - ``mr_ext_memseg_en`` parameter [int]
852 A nonzero value enables extending memseg when registering DMA memory. If
853 enabled, the number of entries in MR (Memory Region) lookup table on datapath
854 is minimized and it benefits performance. On the other hand, it worsens memory
855 utilization because registered memory is pinned by kernel driver. Even if a
856 page in the extended chunk is freed, that doesn't become reusable until the
857 entire memory is freed.
861 - ``representor`` parameter [list]
863 This parameter can be used to instantiate DPDK Ethernet devices from
864 existing port (or VF) representors configured on the device.
866 It is a standard parameter whose format is described in
867 :ref:`ethernet_device_standard_device_arguments`.
869 For instance, to probe port representors 0 through 2::
873 - ``max_dump_files_num`` parameter [int]
875 The maximum number of files per PMD entity that may be created for debug information.
876 The files will be created in /var/log directory or in current directory.
878 set to 128 by default.
880 - ``lro_timeout_usec`` parameter [int]
882 The maximum allowed duration of an LRO session, in micro-seconds.
883 PMD will set the nearest value supported by HW, which is not bigger than
884 the input ``lro_timeout_usec`` value.
885 If this parameter is not specified, by default PMD will set
886 the smallest value supported by HW.
888 - ``hp_buf_log_sz`` parameter [int]
890 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
891 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
892 The capacity of the value is specified by the firmware and the initialization
893 will get a failure if it is out of scope.
894 The range of the value is from 11 to 19 right now, and the supported frame
895 size of a single packet for hairpin is from 512B to 128KB. It might change if
896 different firmware release is being used. By using a small value, it could
897 reduce memory consumption but not work with a large frame. If the value is
898 too large, the memory consumption will be high and some potential performance
899 degradation will be introduced.
900 By default, the PMD will set this value to 16, which means that 9KB jumbo
901 frames will be supported.
903 - ``reclaim_mem_mode`` parameter [int]
905 Cache some resources in flow destroy will help flow recreation more efficient.
906 While some systems may require the all the resources can be reclaimed after
908 The parameter ``reclaim_mem_mode`` provides the option for user to configure
909 if the resource cache is needed or not.
911 There are three options to choose:
913 - 0. It means the flow resources will be cached as usual. The resources will
914 be cached, helpful with flow insertion rate.
916 - 1. It will only enable the DPDK PMD level resources reclaim.
918 - 2. Both DPDK PMD level and rdma-core low level will be configured as
921 By default, the PMD will set this value to 0.
923 - ``sys_mem_en`` parameter [int]
925 A non-zero value enables the PMD memory management allocating memory
926 from system by default, without explicit rte memory flag.
928 By default, the PMD will set this value to 0.
930 - ``decap_en`` parameter [int]
932 Some devices do not support FCS (frame checksum) scattering for
933 tunnel-decapsulated packets.
934 If set to 0, this option forces the FCS feature and rejects tunnel
935 decapsulation in the flow engine for such devices.
937 By default, the PMD will set this value to 1.
939 .. _mlx5_firmware_config:
941 Firmware configuration
942 ~~~~~~~~~~~~~~~~~~~~~~
944 Firmware features can be configured as key/value pairs.
946 The command to set a value is::
948 mlxconfig -d <device> set <key>=<value>
950 The command to query a value is::
952 mlxconfig -d <device> query | grep <key>
954 The device name for the command ``mlxconfig`` can be either the PCI address,
955 or the mst device name found with::
959 Below are some firmware configurations listed.
965 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
971 - maximum number of SR-IOV virtual functions::
975 - enable DevX (required by Direct Rules and other features)::
979 - aggressive CQE zipping::
983 - L3 VXLAN and VXLAN-GPE destination UDP port::
986 IP_OVER_VXLAN_PORT=<udp dport>
988 - enable VXLAN-GPE tunnel flow matching::
990 FLEX_PARSER_PROFILE_ENABLE=0
992 FLEX_PARSER_PROFILE_ENABLE=2
994 - enable IP-in-IP tunnel flow matching::
996 FLEX_PARSER_PROFILE_ENABLE=0
998 - enable MPLS flow matching::
1000 FLEX_PARSER_PROFILE_ENABLE=1
1002 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1004 FLEX_PARSER_PROFILE_ENABLE=2
1006 - enable Geneve flow matching::
1008 FLEX_PARSER_PROFILE_ENABLE=0
1010 FLEX_PARSER_PROFILE_ENABLE=1
1012 - enable GTP flow matching::
1014 FLEX_PARSER_PROFILE_ENABLE=3
1016 - enable eCPRI flow matching::
1018 FLEX_PARSER_PROFILE_ENABLE=4
1024 This driver relies on external libraries and kernel drivers for resources
1025 allocations and initialization. The following dependencies are not part of
1026 DPDK and must be installed separately:
1030 User space Verbs framework used by librte_net_mlx5. This library provides
1031 a generic interface between the kernel and low-level user space drivers
1034 It allows slow and privileged operations (context initialization, hardware
1035 resources allocations) to be managed by the kernel and fast operations to
1036 never leave user space.
1040 Low-level user space driver library for Mellanox
1041 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1044 This library basically implements send/receive calls to the hardware
1047 - **Kernel modules**
1049 They provide the kernel-side Verbs API and low level device drivers that
1050 manage actual hardware initialization and resources sharing with user
1053 Unlike most other PMDs, these modules must remain loaded and bound to
1056 - mlx5_core: hardware driver managing Mellanox
1057 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1059 - mlx5_ib: InifiniBand device driver.
1060 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1062 - **Firmware update**
1064 Mellanox OFED/EN releases include firmware updates for
1065 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1067 Because each release provides new features, these updates must be applied to
1068 match the kernel modules and libraries they come with.
1072 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1078 Either RDMA Core library with a recent enough Linux kernel release
1079 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1082 RDMA Core with Linux Kernel
1083 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1085 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1086 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1087 (see `RDMA Core installation documentation`_)
1088 - When building for i686 use:
1090 - rdma-core version 18.0 or above built with 32bit support.
1091 - Kernel version 4.14.41 or above.
1093 - Starting with rdma-core v21, static libraries can be built::
1096 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1099 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1100 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1106 - Mellanox OFED version: **4.5** and above /
1107 Mellanox EN version: **4.5** and above
1110 - ConnectX-4: **12.21.1000** and above.
1111 - ConnectX-4 Lx: **14.21.1000** and above.
1112 - ConnectX-5: **16.21.1000** and above.
1113 - ConnectX-5 Ex: **16.21.1000** and above.
1114 - ConnectX-6: **20.27.0090** and above.
1115 - ConnectX-6 Dx: **22.27.0090** and above.
1116 - BlueField: **18.25.1010** and above.
1118 While these libraries and kernel modules are available on OpenFabrics
1119 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1120 managers on most distributions, this PMD requires Ethernet extensions that
1121 may not be supported at the moment (this is a work in progress).
1124 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1126 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1127 include the necessary support and should be used in the meantime. For DPDK,
1128 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1129 required from that distribution.
1133 Several versions of Mellanox OFED/EN are available. Installing the version
1134 this DPDK release was developed and tested against is strongly
1135 recommended. Please check the `prerequisites`_.
1140 The following Mellanox device families are supported by the same mlx5 driver:
1150 Below are detailed device names:
1152 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1153 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1154 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1155 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1156 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1157 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1158 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1159 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1160 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1161 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1162 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1163 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1164 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1165 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1166 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1167 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1168 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1169 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1170 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1171 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1172 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1173 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1174 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1175 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1176 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1178 Quick Start Guide on OFED/EN
1179 ----------------------------
1181 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
1184 2. Install the required libraries and kernel modules either by installing
1185 only the required set, or by installing the entire Mellanox OFED/EN::
1187 ./mlnxofedinstall --upstream-libs --dpdk
1189 3. Verify the firmware is the correct one::
1193 4. Verify all ports links are set to Ethernet::
1195 mlxconfig -d <mst device> query | grep LINK_TYPE
1199 Link types may have to be configured to Ethernet::
1201 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1203 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1205 For hypervisors, verify SR-IOV is enabled on the NIC::
1207 mlxconfig -d <mst device> query | grep SRIOV_EN
1210 If needed, configure SR-IOV::
1212 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1213 mlxfwreset -d <mst device> reset
1215 5. Restart the driver::
1217 /etc/init.d/openibd restart
1221 service openibd restart
1223 If link type was changed, firmware must be reset as well::
1225 mlxfwreset -d <mst device> reset
1227 For hypervisors, after reset write the sysfs number of virtual functions
1230 To dynamically instantiate a given number of virtual functions (VFs)::
1232 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1234 6. Install DPDK and you are ready to go.
1235 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1237 Enable switchdev mode
1238 ---------------------
1240 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1241 Representor is a port in DPDK that is connected to a VF in such a way
1242 that assuming there are no offload flows, each packet that is sent from the VF
1243 will be received by the corresponding representor. While each packet that is
1244 sent to a representor will be received by the VF.
1245 This is very useful in case of SRIOV mode, where the first packet that is sent
1246 by the VF will be received by the DPDK application which will decide if this
1247 flow should be offloaded to the E-Switch. After offloading the flow packet
1248 that the VF that are matching the flow will not be received any more by
1249 the DPDK application.
1251 1. Enable SRIOV mode::
1253 mlxconfig -d <mst device> set SRIOV_EN=true
1255 2. Configure the max number of VFs::
1257 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1261 mlxfwreset -d <mst device> reset
1263 3. Configure the actual number of VFs::
1265 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1267 4. Unbind the device (can be rebind after the switchdev mode)::
1269 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1271 5. Enbale switchdev mode::
1273 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1278 1. Configure aggressive CQE Zipping for maximum performance::
1280 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1282 To set it back to the default CQE Zipping mode use::
1284 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1286 2. In case of virtualization:
1288 - Make sure that hypervisor kernel is 3.16 or newer.
1289 - Configure boot with ``iommu=pt``.
1290 - Use 1G huge pages.
1291 - Make sure to allocate a VM on huge pages.
1292 - Make sure to set CPU pinning.
1294 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1295 for better performance. For VMs, verify that the right CPU
1296 and NUMA node are pinned according to the above. Run::
1300 to identify the NUMA node to which the PCIe adapter is connected.
1302 4. If more than one adapter is used, and root complex capabilities allow
1303 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1304 it is recommended to locate both adapters on the same NUMA node.
1305 This in order to forward packets from one to the other without
1306 NUMA performance penalty.
1308 5. Disable pause frames::
1310 ethtool -A <netdev> rx off tx off
1312 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1313 via the BIOS configuration. Please contact you server provider for more
1314 information about the settings.
1318 On some machines, depends on the machine integrator, it is beneficial
1319 to set the PCI max read request parameter to 1K. This can be
1320 done in the following way:
1322 To query the read request size use::
1324 setpci -s <NIC PCI address> 68.w
1326 If the output is different than 3XXX, set it by::
1328 setpci -s <NIC PCI address> 68.w=3XXX
1330 The XXX can be different on different systems. Make sure to configure
1331 according to the setpci output.
1333 7. To minimize overhead of searching Memory Regions:
1335 - '--socket-mem' is recommended to pin memory by predictable amount.
1336 - Configure per-lcore cache when creating Mempools for packet buffer.
1337 - Refrain from dynamically allocating/freeing memory in run-time.
1339 .. _mlx5_offloads_support:
1341 Supported hardware offloads
1342 ---------------------------
1344 .. table:: Minimal SW/HW versions for queue offloads
1346 ============== ===== ===== ========= ===== ========== ==========
1347 Offload DPDK Linux rdma-core OFED firmware hardware
1348 ============== ===== ===== ========= ===== ========== ==========
1349 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1350 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1351 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1352 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1353 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1354 ============== ===== ===== ========= ===== ========== ==========
1356 .. table:: Minimal SW/HW versions for rte_flow offloads
1358 +-----------------------+-----------------+-----------------+
1359 | Offload | with E-Switch | with NIC |
1360 +=======================+=================+=================+
1361 | Count | | DPDK 19.05 | | DPDK 19.02 |
1362 | | | OFED 4.6 | | OFED 4.6 |
1363 | | | rdma-core 24 | | rdma-core 23 |
1364 | | | ConnectX-5 | | ConnectX-5 |
1365 +-----------------------+-----------------+-----------------+
1366 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1367 | | | OFED 4.6 | | OFED 4.5 |
1368 | | | rdma-core 24 | | rdma-core 23 |
1369 | | | ConnectX-5 | | ConnectX-4 |
1370 +-----------------------+-----------------+-----------------+
1371 | Queue / RSS | | | | DPDK 18.11 |
1372 | | | N/A | | OFED 4.5 |
1373 | | | | | rdma-core 23 |
1374 | | | | | ConnectX-4 |
1375 +-----------------------+-----------------+-----------------+
1376 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1377 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1378 | | | rdma-core 24 | | rdma-core 23 |
1379 | | | ConnectX-5 | | ConnectX-5 |
1380 +-----------------------+-----------------+-----------------+
1381 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1382 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1383 | | | rdma-core 27 | | rdma-core 27 |
1384 | | | ConnectX-5 | | ConnectX-5 |
1385 +-----------------------+-----------------+-----------------+
1386 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1387 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1388 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1389 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1390 | | set_ipv6_dst / | | | | |
1391 | | set_tp_src / | | | | |
1392 | | set_tp_dst / | | | | |
1393 | | dec_ttl / | | | | |
1394 | | set_ttl / | | | | |
1395 | | set_mac_src / | | | | |
1396 | | set_mac_dst) | | | | |
1397 +-----------------------+-----------------+-----------------+
1398 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1399 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1400 | | | | rdma-core 24 | | rdma-core 24 |
1401 | | | | ConnectX-5 | | ConnectX-5 |
1402 +-----------------------+-----------------+-----------------+
1403 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1404 | | | OFED 4.7-1 | | OFED 4.7-1 |
1405 | | | rdma-core 24 | | N/A |
1406 | | | ConnectX-5 | | ConnectX-5 |
1407 +-----------------------+-----------------+-----------------+
1408 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1409 | | | OFED 4.6 | | OFED 4.5 |
1410 | | | rdma-core 24 | | rdma-core 23 |
1411 | | | ConnectX-5 | | ConnectX-4 |
1412 +-----------------------+-----------------+-----------------+
1413 | Port ID | | DPDK 19.05 | | N/A |
1414 | | | OFED 4.7-1 | | N/A |
1415 | | | rdma-core 24 | | N/A |
1416 | | | ConnectX-5 | | N/A |
1417 +-----------------------+-----------------+-----------------+
1418 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1419 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1420 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1421 | | of_set_vlan_pcp / | | | | |
1422 | | of_set_vlan_vid) | | | | |
1423 +-----------------------+-----------------+-----------------+
1424 | Hairpin | | | | DPDK 19.11 |
1425 | | | N/A | | OFED 4.7-3 |
1426 | | | | | rdma-core 26 |
1427 | | | | | ConnectX-5 |
1428 +-----------------------+-----------------+-----------------+
1429 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1430 | | | OFED 4.7-3 | | OFED 4.7-3 |
1431 | | | rdma-core 26 | | rdma-core 26 |
1432 | | | ConnectX-5 | | ConnectX-5 |
1433 +-----------------------+-----------------+-----------------+
1434 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1435 | | | OFED 4.7-3 | | OFED 4.7-3 |
1436 | | | rdma-core 26 | | rdma-core 26 |
1437 | | | ConnectX-5 | | ConnectX-5 |
1438 +-----------------------+-----------------+-----------------+
1439 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1440 | | | OFED 5.2 | | OFED 5.2 |
1441 | | | rdma-core 32 | | rdma-core 32 |
1442 | | | ConnectX-5 | | ConnectX-5 |
1443 +-----------------------+-----------------+-----------------+
1448 MARK and META items are interrelated with datapath - they might move from/to
1449 the applications in mbuf fields. Hence, zero value for these items has the
1450 special meaning - it means "no metadata are provided", not zero values are
1451 treated by applications and PMD as valid ones.
1453 Moreover in the flow engine domain the value zero is acceptable to match and
1454 set, and we should allow to specify zero values as rte_flow parameters for the
1455 META and MARK items and actions. In the same time zero mask has no meaning and
1456 should be rejected on validation stage.
1461 Flows are not cached in the driver.
1462 When stopping a device port, all the flows created on this port from the
1463 application will be flushed automatically in the background.
1464 After stopping the device port, all flows on this port become invalid and
1465 not represented in the system.
1466 All references to these flows held by the application should be discarded
1467 directly but neither destroyed nor flushed.
1469 The application should re-create the flows as required after the port restart.
1474 Compared to librte_net_mlx4 that implements a single RSS configuration per
1475 port, librte_net_mlx5 supports per-protocol RSS configuration.
1477 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1478 command-line parameter to enable additional protocols (UDP and TCP as well
1479 as IP), the following commands must be entered from its CLI to get the same
1480 behavior as librte_net_mlx4::
1483 > port config all rss all
1489 This section demonstrates how to launch **testpmd** with Mellanox
1490 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1492 #. Load the kernel modules::
1494 modprobe -a ib_uverbs mlx5_core mlx5_ib
1496 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1499 /etc/init.d/openibd restart
1503 User space I/O kernel modules (uio and igb_uio) are not used and do
1504 not have to be loaded.
1506 #. Make sure Ethernet interfaces are in working order and linked to kernel
1507 verbs. Related sysfs entries should be present::
1509 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1518 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1521 for intf in eth2 eth3 eth4 eth5;
1523 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1526 sed -n 's,.*/\(.*\),-w \1,p'
1535 #. Request huge pages::
1537 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1539 #. Start testpmd with basic parameters::
1541 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
1546 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1547 EAL: probe driver: 15b3:1013 librte_net_mlx5
1548 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1549 PMD: librte_net_mlx5: 1 port(s) detected
1550 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1551 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1552 EAL: probe driver: 15b3:1013 librte_net_mlx5
1553 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1554 PMD: librte_net_mlx5: 1 port(s) detected
1555 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1556 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1557 EAL: probe driver: 15b3:1013 librte_net_mlx5
1558 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1559 PMD: librte_net_mlx5: 1 port(s) detected
1560 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1561 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1562 EAL: probe driver: 15b3:1013 librte_net_mlx5
1563 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1564 PMD: librte_net_mlx5: 1 port(s) detected
1565 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1566 Interactive-mode selected
1567 Configuring Port 0 (socket 0)
1568 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1569 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1570 Port 0: E4:1D:2D:E7:0C:FE
1571 Configuring Port 1 (socket 0)
1572 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1573 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1574 Port 1: E4:1D:2D:E7:0C:FF
1575 Configuring Port 2 (socket 0)
1576 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1577 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1578 Port 2: E4:1D:2D:E7:0C:FA
1579 Configuring Port 3 (socket 0)
1580 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1581 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1582 Port 3: E4:1D:2D:E7:0C:FB
1583 Checking link statuses...
1584 Port 0 Link Up - speed 40000 Mbps - full-duplex
1585 Port 1 Link Up - speed 40000 Mbps - full-duplex
1586 Port 2 Link Up - speed 10000 Mbps - full-duplex
1587 Port 3 Link Up - speed 10000 Mbps - full-duplex
1594 This section demonstrates how to dump flows. Currently, it's possible to dump
1595 all flows with assistance of external tools.
1597 #. 2 ways to get flow raw file:
1599 - Using testpmd CLI:
1601 .. code-block:: console
1603 testpmd> flow dump <port> <output_file>
1605 - call rte_flow_dev_dump api:
1607 .. code-block:: console
1609 rte_flow_dev_dump(port, file, NULL);
1611 #. Dump human-readable flows from raw file:
1613 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1615 .. code-block:: console
1617 mlx_steering_dump.py -f <output_file>