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 DV flow engine (``dv_flow_en`` = 1), flow pattern without VLAN item
127 will match untagged packets only.
130 flow create 0 ingress pattern eth / ipv4 / end ...
132 Will match untagged packets only.
135 flow create 0 ingress pattern eth / vlan / ipv4 / end ...
137 Will match tagged packets only, with any VLAN ID value.
140 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
142 Will only match tagged packets with VLAN ID 3.
144 - VLAN pop offload command:
146 - Flow rules having a VLAN pop offload command as one of their actions and
147 are lacking a match on VLAN as one of their items are not supported.
148 - The command is not supported on egress traffic.
150 - VLAN push offload is not supported on ingress traffic.
152 - VLAN set PCP offload is not supported on existing headers.
154 - A multi segment packet must have not more segments than reported by dev_infos_get()
155 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
156 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
157 inline settings) to 58.
159 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
160 to 0 are not supported.
162 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
164 - Match on Geneve header supports the following fields only:
170 Currently, the only supported options length value is 0.
172 - VF: flow rules created on VF devices can only match traffic targeted at the
173 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
175 - Match on GTP tunnel header item supports the following fields only:
177 - v_pt_rsv_flags: E flag, S flag, PN flag
181 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
182 The flows within group 0 and set metadata action are rejected by hardware.
186 MAC addresses not already present in the bridge table of the associated
187 kernel network device will be added and cleaned up by the PMD when closing
188 the device. In case of ungraceful program termination, some entries may
189 remain present and should be removed manually by other means.
191 - Buffer split offload is supported with regular Rx burst routine only,
192 no MPRQ feature or vectorized code can be engaged.
194 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
195 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
196 ol_flags. As the mempool for the external buffer is managed by PMD, all the
197 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
198 the external buffers will be freed by PMD and the application which still
199 holds the external buffers may be corrupted.
201 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
202 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
203 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
205 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
206 and allmulticast mode are both set to off.
207 To receive IPv6 Multicast messages on VM, explicitly set the relevant
208 MAC address using rte_eth_dev_mac_addr_add() API.
210 - To support a mixed traffic pattern (some buffers from local host memory, some
211 buffers from other devices) with high bandwidth, a mbuf flag is used.
213 An application hints the PMD whether or not it should try to inline the
214 given mbuf data buffer. PMD should do the best effort to act upon this request.
216 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
217 registered by application with rte_mbuf_dynflag_register(). This flag is
218 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
219 which is intended to be used by the application.
221 To query the supported specific flags in runtime,
222 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
223 currently (over present hardware and configuration) supported specific flags.
224 The "not inline hint" feature operating flow is the following one:
227 - probe the devices, ports are created
228 - query the port capabilities
229 - if port supporting the feature is found
230 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
231 - application starts the ports
232 - on ``dev_start()`` PMD checks whether the feature flag is registered and
233 enables the feature support in datapath
234 - application might set the registered flag bit in ``ol_flags`` field
235 of mbuf being sent and PMD will handle ones appropriately.
237 - The amount of descriptors in Tx queue may be limited by data inline settings.
238 Inline data require the more descriptor building blocks and overall block
239 amount may exceed the hardware supported limits. The application should
240 reduce the requested Tx size or adjust data inline settings with
241 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
243 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
244 parameter should be specified.
245 When PMD sees the RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME set on the packet
246 being sent it tries to synchronize the time of packet appearing on
247 the wire with the specified packet timestamp. It the specified one
248 is in the past it should be ignored, if one is in the distant future
249 it should be capped with some reasonable value (in range of seconds).
250 These specific cases ("too late" and "distant future") can be optionally
251 reported via device xstats to assist applications to detect the
252 time-related problems.
254 The timestamp upper "too-distant-future" limit
255 at the moment of invoking the Tx burst routine
256 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
257 Please note, for the testpmd txonly mode,
258 the limit is deduced from the expression::
260 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
262 There is no any packet reordering according timestamps is supposed,
263 neither within packet burst, nor between packets, it is an entirely
264 application responsibility to generate packets and its timestamps
265 in desired order. The timestamps can be put only in the first packet
266 in the burst providing the entire burst scheduling.
268 - E-Switch decapsulation Flow:
270 - can be applied to PF port only.
271 - must specify VF port action (packet redirection from PF to VF).
272 - optionally may specify tunnel inner source and destination MAC addresses.
274 - E-Switch encapsulation Flow:
276 - can be applied to VF ports only.
277 - must specify PF port action (packet redirection from VF to PF).
281 - The input buffer, used as outer header, is not validated.
285 - The decapsulation is always done up to the outermost tunnel detected by the HW.
286 - The input buffer, providing the removal size, is not validated.
287 - The buffer size must match the length of the headers to be removed.
289 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
290 mutually exclusive features which cannot be supported together
291 (see :ref:`mlx5_firmware_config`).
295 - Requires DevX and DV flow to be enabled.
296 - KEEP_CRC offload cannot be supported with LRO.
297 - The first mbuf length, without head-room, must be big enough to include the
299 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
300 it with size limited to max LRO size, not to max RX packet length.
301 - LRO can be used with outer header of TCP packets of the standard format:
302 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
304 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
305 - LRO packet aggregation is performed by HW only for packet size larger than
306 ``lro_min_mss_size``. This value is reported on device start, when debug
311 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
312 for some NICs (such as ConnectX-6 Dx and BlueField 2).
313 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
317 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and E-Switch steering domain.
318 - The E-Switch Sample flow must have the eswitch_manager VPORT destination (PF or ECPF) and no additional actions.
319 - 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.
321 - IPv6 header item 'proto' field, indicating the next header protocol, should
322 not be set as extension header.
323 In case the next header is an extension header, it should not be specified in
324 IPv6 header item 'proto' field.
325 The last extension header item 'next header' field can specify the following
326 header protocol type.
330 - 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.
331 - Hairpin in switchdev SR-IOV mode is not supported till now.
336 MLX5 supports various methods to report statistics:
338 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.
340 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.
342 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.
350 The ibverbs libraries can be linked with this PMD in a number of ways,
351 configured by the ``ibverbs_link`` build option:
353 - ``shared`` (default): the PMD depends on some .so files.
355 - ``dlopen``: Split the dependencies glue in a separate library
356 loaded when needed by dlopen.
357 It make dependencies on libibverbs and libmlx4 optional,
358 and has no performance impact.
360 - ``static``: Embed static flavor of the dependencies libibverbs and libmlx4
361 in the PMD shared library or the executable static binary.
363 Environment variables
364 ~~~~~~~~~~~~~~~~~~~~~
368 A list of directories in which to search for the rdma-core "glue" plug-in,
369 separated by colons or semi-colons.
371 - ``MLX5_SHUT_UP_BF``
373 Configures HW Tx doorbell register as IO-mapped.
375 By default, the HW Tx doorbell is configured as a write-combining register.
376 The register would be flushed to HW usually when the write-combining buffer
377 becomes full, but it depends on CPU design.
379 Except for vectorized Tx burst routines, a write memory barrier is enforced
380 after updating the register so that the update can be immediately visible to
383 When vectorized Tx burst is called, the barrier is set only if the burst size
384 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
385 variable will bring better latency even though the maximum throughput can
388 Run-time configuration
389 ~~~~~~~~~~~~~~~~~~~~~~
391 - librte_net_mlx5 brings kernel network interfaces up during initialization
392 because it is affected by their state. Forcing them down prevents packets
395 - **ethtool** operations on related kernel interfaces also affect the PMD.
400 In order to run as a non-root user,
401 some capabilities must be granted to the application::
403 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
405 Below are the reasons of the need for each capability:
408 When using physical addresses (PA mode), with Linux >= 4.0,
409 for access to ``/proc/self/pagemap``.
412 For device configuration.
415 For raw ethernet queue allocation through kernel driver.
418 For DMA memory pinning.
423 - ``rxq_cqe_comp_en`` parameter [int]
425 A nonzero value enables the compression of CQE on RX side. This feature
426 allows to save PCI bandwidth and improve performance. Enabled by default.
430 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
432 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
435 - ``rxq_cqe_pad_en`` parameter [int]
437 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
438 is aligned with the size of a cacheline of the core. If cacheline size is
439 128B, the CQE size is configured to be 128B even though the device writes
440 only 64B data on the cacheline. This is to avoid unnecessary cache
441 invalidation by device's two consecutive writes on to one cacheline.
442 However in some architecture, it is more beneficial to update entire
443 cacheline with padding the rest 64B rather than striding because
444 read-modify-write could drop performance a lot. On the other hand,
445 writing extra data will consume more PCIe bandwidth and could also drop
446 the maximum throughput. It is recommended to empirically set this
447 parameter. Disabled by default.
451 - CPU having 128B cacheline with ConnectX-5 and BlueField.
453 - ``rxq_pkt_pad_en`` parameter [int]
455 A nonzero value enables padding Rx packet to the size of cacheline on PCI
456 transaction. This feature would waste PCI bandwidth but could improve
457 performance by avoiding partial cacheline write which may cause costly
458 read-modify-copy in memory transaction on some architectures. Disabled by
463 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
465 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
468 - ``mprq_en`` parameter [int]
470 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
471 configured as Multi-Packet RQ if the total number of Rx queues is
472 ``rxqs_min_mprq`` or more. Disabled by default.
474 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
475 by posting a single large buffer for multiple packets. Instead of posting a
476 buffers per a packet, one large buffer is posted in order to receive multiple
477 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
478 and each stride receives one packet. MPRQ can improve throughput for
479 small-packet traffic.
481 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
482 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
483 configure large stride size enough to accommodate max_rx_pkt_len as long as
484 device allows. Note that this can waste system memory compared to enabling Rx
485 scatter and multi-segment packet.
487 - ``mprq_log_stride_num`` parameter [int]
489 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
490 strides can reduce PCIe traffic further. If configured value is not in the
491 range of device capability, the default value will be set with a warning
492 message. The default value is 4 which is 16 strides per a buffer, valid only
493 if ``mprq_en`` is set.
495 The size of Rx queue should be bigger than the number of strides.
497 - ``mprq_log_stride_size`` parameter [int]
499 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
500 stride size can save some memory and reduce probability of a depletion of all
501 available strides due to unreleased packets by an application. If configured
502 value is not in the range of device capability, the default value will be set
503 with a warning message. The default value is 11 which is 2048 bytes per a
504 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
505 it is possible for a packet to span across multiple strides. This mode allows
506 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
507 of a packet if Rx scatter is configured) may be required in case there is no
508 space left for a head room at the end of a stride which incurs some
511 - ``mprq_max_memcpy_len`` parameter [int]
513 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
514 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
515 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
516 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
517 A mempool for external buffers will be allocated and managed by PMD. If Rx
518 packet is externally attached, ol_flags field of the mbuf will have
519 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
520 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
522 - ``rxqs_min_mprq`` parameter [int]
524 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
525 greater or equal to this value. The default value is 12, valid only if
528 - ``txq_inline`` parameter [int]
530 Amount of data to be inlined during TX operations. This parameter is
531 deprecated and converted to the new parameter ``txq_inline_max`` providing
532 partial compatibility.
534 - ``txqs_min_inline`` parameter [int]
536 Enable inline data send only when the number of TX queues is greater or equal
539 This option should be used in combination with ``txq_inline_max`` and
540 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
542 If this option is not specified the default value 16 is used for BlueField
543 and 8 for other platforms
545 The data inlining consumes the CPU cycles, so this option is intended to
546 auto enable inline data if we have enough Tx queues, which means we have
547 enough CPU cores and PCI bandwidth is getting more critical and CPU
548 is not supposed to be bottleneck anymore.
550 The copying data into WQE improves latency and can improve PPS performance
551 when PCI back pressure is detected and may be useful for scenarios involving
552 heavy traffic on many queues.
554 Because additional software logic is necessary to handle this mode, this
555 option should be used with care, as it may lower performance when back
556 pressure is not expected.
558 If inline data are enabled it may affect the maximal size of Tx queue in
559 descriptors because the inline data increase the descriptor size and
560 queue size limits supported by hardware may be exceeded.
562 - ``txq_inline_min`` parameter [int]
564 Minimal amount of data to be inlined into WQE during Tx operations. NICs
565 may require this minimal data amount to operate correctly. The exact value
566 may depend on NIC operation mode, requested offloads, etc. It is strongly
567 recommended to omit this parameter and use the default values. Anyway,
568 applications using this parameter should take into consideration that
569 specifying an inconsistent value may prevent the NIC from sending packets.
571 If ``txq_inline_min`` key is present the specified value (may be aligned
572 by the driver in order not to exceed the limits and provide better descriptor
573 space utilization) will be used by the driver and it is guaranteed that
574 requested amount of data bytes are inlined into the WQE beside other inline
575 settings. This key also may update ``txq_inline_max`` value (default
576 or specified explicitly in devargs) to reserve the space for inline data.
578 If ``txq_inline_min`` key is not present, the value may be queried by the
579 driver from the NIC via DevX if this feature is available. If there is no DevX
580 enabled/supported the value 18 (supposing L2 header including VLAN) is set
581 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
582 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
585 For ConnectX-4 NIC, driver does not allow specifying value below 18
586 (minimal L2 header, including VLAN), error will be raised.
588 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
589 it is not recommended and may prevent NIC from sending packets over
592 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
593 Multi-Packet Write), because last one does not support partial packet inlining.
594 This is not very critical due to minimal data inlining is mostly required
595 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
597 - ``txq_inline_max`` parameter [int]
599 Specifies the maximal packet length to be completely inlined into WQE
600 Ethernet Segment for ordinary SEND method. If packet is larger than specified
601 value, the packet data won't be copied by the driver at all, data buffer
602 is addressed with a pointer. If packet length is less or equal all packet
603 data will be copied into WQE. This may improve PCI bandwidth utilization for
604 short packets significantly but requires the extra CPU cycles.
606 The data inline feature is controlled by number of Tx queues, if number of Tx
607 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
608 is engaged, if there are not enough Tx queues (which means not enough CPU cores
609 and CPU resources are scarce), data inline is not performed by the driver.
610 Assigning ``txqs_min_inline`` with zero always enables the data inline.
612 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
613 by the driver in order not to exceed the limit (930 bytes) and to provide better
614 WQE space filling without gaps, the adjustment is reflected in the debug log.
615 Also, the default value (290) may be decreased in run-time if the large transmit
616 queue size is requested and hardware does not support enough descriptor
617 amount, in this case warning is emitted. If ``txq_inline_max`` key is
618 specified and requested inline settings can not be satisfied then error
621 - ``txq_inline_mpw`` parameter [int]
623 Specifies the maximal packet length to be completely inlined into WQE for
624 Enhanced MPW method. If packet is large the specified value, the packet data
625 won't be copied, and data buffer is addressed with pointer. If packet length
626 is less or equal, all packet data will be copied into WQE. This may improve PCI
627 bandwidth utilization for short packets significantly but requires the extra
630 The data inline feature is controlled by number of TX queues, if number of Tx
631 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
632 is engaged, if there are not enough Tx queues (which means not enough CPU cores
633 and CPU resources are scarce), data inline is not performed by the driver.
634 Assigning ``txqs_min_inline`` with zero always enables the data inline.
636 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
637 by the driver in order not to exceed the limit (930 bytes) and to provide better
638 WQE space filling without gaps, the adjustment is reflected in the debug log.
639 Due to multiple packets may be included to the same WQE with Enhanced Multi
640 Packet Write Method and overall WQE size is limited it is not recommended to
641 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
642 may be decreased in run-time if the large transmit queue size is requested
643 and hardware does not support enough descriptor amount, in this case warning
644 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
645 settings can not be satisfied then error will be raised.
647 - ``txqs_max_vec`` parameter [int]
649 Enable vectorized Tx only when the number of TX queues is less than or
650 equal to this value. This parameter is deprecated and ignored, kept
651 for compatibility issue to not prevent driver from probing.
653 - ``txq_mpw_hdr_dseg_en`` parameter [int]
655 A nonzero value enables including two pointers in the first block of TX
656 descriptor. The parameter is deprecated and ignored, kept for compatibility
659 - ``txq_max_inline_len`` parameter [int]
661 Maximum size of packet to be inlined. This limits the size of packet to
662 be inlined. If the size of a packet is larger than configured value, the
663 packet isn't inlined even though there's enough space remained in the
664 descriptor. Instead, the packet is included with pointer. This parameter
665 is deprecated and converted directly to ``txq_inline_mpw`` providing full
666 compatibility. Valid only if eMPW feature is engaged.
668 - ``txq_mpw_en`` parameter [int]
670 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
671 ConnectX-6, ConnectX-6 Dx and BlueField. eMPW allows the TX burst function to pack
672 up multiple packets in a single descriptor session in order to save PCI bandwidth
673 and improve performance at the cost of a slightly higher CPU usage. When
674 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
675 entire packet data on to TX descriptor instead of including pointer of packet.
677 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
678 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
679 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
680 option or reported by the NIC, the eMPW feature is disengaged.
682 - ``tx_db_nc`` parameter [int]
684 The rdma core library can map doorbell register in two ways, depending on the
685 environment variable "MLX5_SHUT_UP_BF":
687 - As regular cached memory (usually with write combining attribute), if the
688 variable is either missing or set to zero.
689 - As non-cached memory, if the variable is present and set to not "0" value.
691 The type of mapping may slightly affect the Tx performance, the optimal choice
692 is strongly relied on the host architecture and should be deduced practically.
694 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
695 memory (with write combining), the PMD will perform the extra write memory barrier
696 after writing to doorbell, it might increase the needed CPU clocks per packet
697 to send, but latency might be improved.
699 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
700 cached memory, the PMD will not perform the extra write memory barrier
701 after writing to doorbell, on some architectures it might improve the
704 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
705 memory, the PMD will use heuristics to decide whether write memory barrier
706 should be performed. For bursts with size multiple of recommended one (64 pkts)
707 it is supposed the next burst is coming and no need to issue the extra memory
708 barrier (it is supposed to be issued in the next coming burst, at least after
709 descriptor writing). It might increase latency (on some hosts till next
710 packets transmit) and should be used with care.
712 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
713 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
714 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
716 - ``tx_pp`` parameter [int]
718 If a nonzero value is specified the driver creates all necessary internal
719 objects to provide accurate packet send scheduling on mbuf timestamps.
720 The positive value specifies the scheduling granularity in nanoseconds,
721 the packet send will be accurate up to specified digits. The allowed range is
722 from 500 to 1 million of nanoseconds. The negative value specifies the module
723 of granularity and engages the special test mode the check the schedule rate.
724 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
727 - ``tx_skew`` parameter [int]
729 The parameter adjusts the send packet scheduling on timestamps and represents
730 the average delay between beginning of the transmitting descriptor processing
731 by the hardware and appearance of actual packet data on the wire. The value
732 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
733 specified. The default value is zero.
735 - ``tx_vec_en`` parameter [int]
737 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx
738 and BlueField NICs if the number of global Tx queues on the port is less than
739 ``txqs_max_vec``. The parameter is deprecated and ignored.
741 - ``rx_vec_en`` parameter [int]
743 A nonzero value enables Rx vector if the port is not configured in
744 multi-segment otherwise this parameter is ignored.
748 - ``vf_nl_en`` parameter [int]
750 A nonzero value enables Netlink requests from the VF to add/remove MAC
751 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
752 Otherwise the relevant configuration must be run with Linux iproute2 tools.
753 This is a prerequisite to receive this kind of traffic.
755 Enabled by default, valid only on VF devices ignored otherwise.
757 - ``l3_vxlan_en`` parameter [int]
759 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
760 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
761 parameter. This is a prerequisite to receive this kind of traffic.
765 - ``dv_xmeta_en`` parameter [int]
767 A nonzero value enables extensive flow metadata support if device is
768 capable and driver supports it. This can enable extensive support of
769 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
770 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
772 There are some possible configurations, depending on parameter value:
774 - 0, this is default value, defines the legacy mode, the ``MARK`` and
775 ``META`` related actions and items operate only within NIC Tx and
776 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
777 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
778 item is 32 bits wide and match supported on egress only.
780 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
781 related actions and items operate within all supported steering domains,
782 including FDB, ``MARK`` and ``META`` information may cross the domain
783 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
784 depends on kernel and firmware configurations and might be 0, 16 or
785 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
786 compatibility, the actual width of data transferred to the FDB domain
787 depends on kernel configuration and may be vary. The actual supported
788 width can be retrieved in runtime by series of rte_flow_validate()
791 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
792 related actions and items operate within all supported steering domains,
793 including FDB, ``MARK`` and ``META`` information may cross the domain
794 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
795 depends on kernel and firmware configurations and might be 0, 16 or
796 24 bits. The actual supported width can be retrieved in runtime by
797 series of rte_flow_validate() trials.
799 - 3, this engages tunnel offload mode. In E-Switch configuration, that
800 mode implicitly activates ``dv_xmeta_en=1``.
802 +------+-----------+-----------+-------------+-------------+
803 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
804 +======+===========+===========+=============+=============+
805 | 0 | 24 bits | 32 bits | 32 bits | no |
806 +------+-----------+-----------+-------------+-------------+
807 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
808 +------+-----------+-----------+-------------+-------------+
809 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
810 +------+-----------+-----------+-------------+-------------+
812 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
813 ignored and the device is configured to operate in legacy mode (0).
815 Disabled by default (set to 0).
817 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
818 of the extensive metadata features. The legacy Verbs supports FLAG and
819 MARK metadata actions over NIC Rx steering domain only.
821 - ``dv_flow_en`` parameter [int]
823 A nonzero value enables the DV flow steering assuming it is supported
824 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
826 Enabled by default if supported.
828 - ``dv_esw_en`` parameter [int]
830 A nonzero value enables E-Switch using Direct Rules.
832 Enabled by default if supported.
834 - ``lacp_by_user`` parameter [int]
836 A nonzero value enables the control of LACP traffic by the user application.
837 When a bond exists in the driver, by default it should be managed by the
838 kernel and therefore LACP traffic should be steered to the kernel.
839 If this devarg is set to 1 it will allow the user to manage the bond by
840 itself and not steer LACP traffic to the kernel.
842 Disabled by default (set to 0).
844 - ``mr_ext_memseg_en`` parameter [int]
846 A nonzero value enables extending memseg when registering DMA memory. If
847 enabled, the number of entries in MR (Memory Region) lookup table on datapath
848 is minimized and it benefits performance. On the other hand, it worsens memory
849 utilization because registered memory is pinned by kernel driver. Even if a
850 page in the extended chunk is freed, that doesn't become reusable until the
851 entire memory is freed.
855 - ``representor`` parameter [list]
857 This parameter can be used to instantiate DPDK Ethernet devices from
858 existing port (or VF) representors configured on the device.
860 It is a standard parameter whose format is described in
861 :ref:`ethernet_device_standard_device_arguments`.
863 For instance, to probe port representors 0 through 2::
867 - ``max_dump_files_num`` parameter [int]
869 The maximum number of files per PMD entity that may be created for debug information.
870 The files will be created in /var/log directory or in current directory.
872 set to 128 by default.
874 - ``lro_timeout_usec`` parameter [int]
876 The maximum allowed duration of an LRO session, in micro-seconds.
877 PMD will set the nearest value supported by HW, which is not bigger than
878 the input ``lro_timeout_usec`` value.
879 If this parameter is not specified, by default PMD will set
880 the smallest value supported by HW.
882 - ``hp_buf_log_sz`` parameter [int]
884 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
885 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
886 The capacity of the value is specified by the firmware and the initialization
887 will get a failure if it is out of scope.
888 The range of the value is from 11 to 19 right now, and the supported frame
889 size of a single packet for hairpin is from 512B to 128KB. It might change if
890 different firmware release is being used. By using a small value, it could
891 reduce memory consumption but not work with a large frame. If the value is
892 too large, the memory consumption will be high and some potential performance
893 degradation will be introduced.
894 By default, the PMD will set this value to 16, which means that 9KB jumbo
895 frames will be supported.
897 - ``reclaim_mem_mode`` parameter [int]
899 Cache some resources in flow destroy will help flow recreation more efficient.
900 While some systems may require the all the resources can be reclaimed after
902 The parameter ``reclaim_mem_mode`` provides the option for user to configure
903 if the resource cache is needed or not.
905 There are three options to choose:
907 - 0. It means the flow resources will be cached as usual. The resources will
908 be cached, helpful with flow insertion rate.
910 - 1. It will only enable the DPDK PMD level resources reclaim.
912 - 2. Both DPDK PMD level and rdma-core low level will be configured as
915 By default, the PMD will set this value to 0.
917 - ``sys_mem_en`` parameter [int]
919 A non-zero value enables the PMD memory management allocating memory
920 from system by default, without explicit rte memory flag.
922 By default, the PMD will set this value to 0.
924 - ``decap_en`` parameter [int]
926 Some devices do not support FCS (frame checksum) scattering for
927 tunnel-decapsulated packets.
928 If set to 0, this option forces the FCS feature and rejects tunnel
929 decapsulation in the flow engine for such devices.
931 By default, the PMD will set this value to 1.
933 .. _mlx5_firmware_config:
935 Firmware configuration
936 ~~~~~~~~~~~~~~~~~~~~~~
938 Firmware features can be configured as key/value pairs.
940 The command to set a value is::
942 mlxconfig -d <device> set <key>=<value>
944 The command to query a value is::
946 mlxconfig -d <device> query | grep <key>
948 The device name for the command ``mlxconfig`` can be either the PCI address,
949 or the mst device name found with::
953 Below are some firmware configurations listed.
959 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
965 - maximum number of SR-IOV virtual functions::
969 - enable DevX (required by Direct Rules and other features)::
973 - aggressive CQE zipping::
977 - L3 VXLAN and VXLAN-GPE destination UDP port::
980 IP_OVER_VXLAN_PORT=<udp dport>
982 - enable VXLAN-GPE tunnel flow matching::
984 FLEX_PARSER_PROFILE_ENABLE=0
986 FLEX_PARSER_PROFILE_ENABLE=2
988 - enable IP-in-IP tunnel flow matching::
990 FLEX_PARSER_PROFILE_ENABLE=0
992 - enable MPLS flow matching::
994 FLEX_PARSER_PROFILE_ENABLE=1
996 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
998 FLEX_PARSER_PROFILE_ENABLE=2
1000 - enable Geneve flow matching::
1002 FLEX_PARSER_PROFILE_ENABLE=0
1004 FLEX_PARSER_PROFILE_ENABLE=1
1006 - enable GTP flow matching::
1008 FLEX_PARSER_PROFILE_ENABLE=3
1010 - enable eCPRI flow matching::
1012 FLEX_PARSER_PROFILE_ENABLE=4
1018 This driver relies on external libraries and kernel drivers for resources
1019 allocations and initialization. The following dependencies are not part of
1020 DPDK and must be installed separately:
1024 User space Verbs framework used by librte_net_mlx5. This library provides
1025 a generic interface between the kernel and low-level user space drivers
1028 It allows slow and privileged operations (context initialization, hardware
1029 resources allocations) to be managed by the kernel and fast operations to
1030 never leave user space.
1034 Low-level user space driver library for Mellanox
1035 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1038 This library basically implements send/receive calls to the hardware
1041 - **Kernel modules**
1043 They provide the kernel-side Verbs API and low level device drivers that
1044 manage actual hardware initialization and resources sharing with user
1047 Unlike most other PMDs, these modules must remain loaded and bound to
1050 - mlx5_core: hardware driver managing Mellanox
1051 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1053 - mlx5_ib: InifiniBand device driver.
1054 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1056 - **Firmware update**
1058 Mellanox OFED/EN releases include firmware updates for
1059 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1061 Because each release provides new features, these updates must be applied to
1062 match the kernel modules and libraries they come with.
1066 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1072 Either RDMA Core library with a recent enough Linux kernel release
1073 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1076 RDMA Core with Linux Kernel
1077 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1079 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1080 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1081 (see `RDMA Core installation documentation`_)
1082 - When building for i686 use:
1084 - rdma-core version 18.0 or above built with 32bit support.
1085 - Kernel version 4.14.41 or above.
1087 - Starting with rdma-core v21, static libraries can be built::
1090 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1093 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1094 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1100 - Mellanox OFED version: **4.5** and above /
1101 Mellanox EN version: **4.5** and above
1104 - ConnectX-4: **12.21.1000** and above.
1105 - ConnectX-4 Lx: **14.21.1000** and above.
1106 - ConnectX-5: **16.21.1000** and above.
1107 - ConnectX-5 Ex: **16.21.1000** and above.
1108 - ConnectX-6: **20.27.0090** and above.
1109 - ConnectX-6 Dx: **22.27.0090** and above.
1110 - BlueField: **18.25.1010** and above.
1112 While these libraries and kernel modules are available on OpenFabrics
1113 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1114 managers on most distributions, this PMD requires Ethernet extensions that
1115 may not be supported at the moment (this is a work in progress).
1118 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1120 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1121 include the necessary support and should be used in the meantime. For DPDK,
1122 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1123 required from that distribution.
1127 Several versions of Mellanox OFED/EN are available. Installing the version
1128 this DPDK release was developed and tested against is strongly
1129 recommended. Please check the `prerequisites`_.
1134 The following Mellanox device families are supported by the same mlx5 driver:
1144 Below are detailed device names:
1146 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1147 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1148 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1149 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1150 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1151 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1152 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1153 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1154 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1155 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1156 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1157 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1158 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1159 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1160 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1161 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1162 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1163 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1164 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1165 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1166 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1167 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1168 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1169 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1170 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1172 Quick Start Guide on OFED/EN
1173 ----------------------------
1175 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
1178 2. Install the required libraries and kernel modules either by installing
1179 only the required set, or by installing the entire Mellanox OFED/EN::
1181 ./mlnxofedinstall --upstream-libs --dpdk
1183 3. Verify the firmware is the correct one::
1187 4. Verify all ports links are set to Ethernet::
1189 mlxconfig -d <mst device> query | grep LINK_TYPE
1193 Link types may have to be configured to Ethernet::
1195 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1197 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1199 For hypervisors, verify SR-IOV is enabled on the NIC::
1201 mlxconfig -d <mst device> query | grep SRIOV_EN
1204 If needed, configure SR-IOV::
1206 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1207 mlxfwreset -d <mst device> reset
1209 5. Restart the driver::
1211 /etc/init.d/openibd restart
1215 service openibd restart
1217 If link type was changed, firmware must be reset as well::
1219 mlxfwreset -d <mst device> reset
1221 For hypervisors, after reset write the sysfs number of virtual functions
1224 To dynamically instantiate a given number of virtual functions (VFs)::
1226 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1228 6. Install DPDK and you are ready to go.
1229 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1231 Enable switchdev mode
1232 ---------------------
1234 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1235 Representor is a port in DPDK that is connected to a VF in such a way
1236 that assuming there are no offload flows, each packet that is sent from the VF
1237 will be received by the corresponding representor. While each packet that is
1238 sent to a representor will be received by the VF.
1239 This is very useful in case of SRIOV mode, where the first packet that is sent
1240 by the VF will be received by the DPDK application which will decide if this
1241 flow should be offloaded to the E-Switch. After offloading the flow packet
1242 that the VF that are matching the flow will not be received any more by
1243 the DPDK application.
1245 1. Enable SRIOV mode::
1247 mlxconfig -d <mst device> set SRIOV_EN=true
1249 2. Configure the max number of VFs::
1251 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1255 mlxfwreset -d <mst device> reset
1257 3. Configure the actual number of VFs::
1259 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1261 4. Unbind the device (can be rebind after the switchdev mode)::
1263 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1265 5. Enbale switchdev mode::
1267 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1272 1. Configure aggressive CQE Zipping for maximum performance::
1274 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1276 To set it back to the default CQE Zipping mode use::
1278 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1280 2. In case of virtualization:
1282 - Make sure that hypervisor kernel is 3.16 or newer.
1283 - Configure boot with ``iommu=pt``.
1284 - Use 1G huge pages.
1285 - Make sure to allocate a VM on huge pages.
1286 - Make sure to set CPU pinning.
1288 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1289 for better performance. For VMs, verify that the right CPU
1290 and NUMA node are pinned according to the above. Run::
1294 to identify the NUMA node to which the PCIe adapter is connected.
1296 4. If more than one adapter is used, and root complex capabilities allow
1297 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1298 it is recommended to locate both adapters on the same NUMA node.
1299 This in order to forward packets from one to the other without
1300 NUMA performance penalty.
1302 5. Disable pause frames::
1304 ethtool -A <netdev> rx off tx off
1306 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1307 via the BIOS configuration. Please contact you server provider for more
1308 information about the settings.
1312 On some machines, depends on the machine integrator, it is beneficial
1313 to set the PCI max read request parameter to 1K. This can be
1314 done in the following way:
1316 To query the read request size use::
1318 setpci -s <NIC PCI address> 68.w
1320 If the output is different than 3XXX, set it by::
1322 setpci -s <NIC PCI address> 68.w=3XXX
1324 The XXX can be different on different systems. Make sure to configure
1325 according to the setpci output.
1327 7. To minimize overhead of searching Memory Regions:
1329 - '--socket-mem' is recommended to pin memory by predictable amount.
1330 - Configure per-lcore cache when creating Mempools for packet buffer.
1331 - Refrain from dynamically allocating/freeing memory in run-time.
1333 .. _mlx5_offloads_support:
1335 Supported hardware offloads
1336 ---------------------------
1338 .. table:: Minimal SW/HW versions for queue offloads
1340 ============== ===== ===== ========= ===== ========== ==========
1341 Offload DPDK Linux rdma-core OFED firmware hardware
1342 ============== ===== ===== ========= ===== ========== ==========
1343 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1344 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1345 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1346 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1347 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1348 ============== ===== ===== ========= ===== ========== ==========
1350 .. table:: Minimal SW/HW versions for rte_flow offloads
1352 +-----------------------+-----------------+-----------------+
1353 | Offload | with E-Switch | with NIC |
1354 +=======================+=================+=================+
1355 | Count | | DPDK 19.05 | | DPDK 19.02 |
1356 | | | OFED 4.6 | | OFED 4.6 |
1357 | | | rdma-core 24 | | rdma-core 23 |
1358 | | | ConnectX-5 | | ConnectX-5 |
1359 +-----------------------+-----------------+-----------------+
1360 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1361 | | | OFED 4.6 | | OFED 4.5 |
1362 | | | rdma-core 24 | | rdma-core 23 |
1363 | | | ConnectX-5 | | ConnectX-4 |
1364 +-----------------------+-----------------+-----------------+
1365 | Queue / RSS | | | | DPDK 18.11 |
1366 | | | N/A | | OFED 4.5 |
1367 | | | | | rdma-core 23 |
1368 | | | | | ConnectX-4 |
1369 +-----------------------+-----------------+-----------------+
1370 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1371 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1372 | | | rdma-core 24 | | rdma-core 23 |
1373 | | | ConnectX-5 | | ConnectX-5 |
1374 +-----------------------+-----------------+-----------------+
1375 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1376 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1377 | | | rdma-core 27 | | rdma-core 27 |
1378 | | | ConnectX-5 | | ConnectX-5 |
1379 +-----------------------+-----------------+-----------------+
1380 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1381 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1382 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1383 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1384 | | set_ipv6_dst / | | | | |
1385 | | set_tp_src / | | | | |
1386 | | set_tp_dst / | | | | |
1387 | | dec_ttl / | | | | |
1388 | | set_ttl / | | | | |
1389 | | set_mac_src / | | | | |
1390 | | set_mac_dst) | | | | |
1391 +-----------------------+-----------------+-----------------+
1392 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1393 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1394 | | | | rdma-core 24 | | rdma-core 24 |
1395 | | | | ConnectX-5 | | ConnectX-5 |
1396 +-----------------------+-----------------+-----------------+
1397 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1398 | | | OFED 4.7-1 | | OFED 4.7-1 |
1399 | | | rdma-core 24 | | N/A |
1400 | | | ConnectX-5 | | ConnectX-5 |
1401 +-----------------------+-----------------+-----------------+
1402 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1403 | | | OFED 4.6 | | OFED 4.5 |
1404 | | | rdma-core 24 | | rdma-core 23 |
1405 | | | ConnectX-5 | | ConnectX-4 |
1406 +-----------------------+-----------------+-----------------+
1407 | Port ID | | DPDK 19.05 | | N/A |
1408 | | | OFED 4.7-1 | | N/A |
1409 | | | rdma-core 24 | | N/A |
1410 | | | ConnectX-5 | | N/A |
1411 +-----------------------+-----------------+-----------------+
1412 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1413 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1414 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1415 | | of_set_vlan_pcp / | | | | |
1416 | | of_set_vlan_vid) | | | | |
1417 +-----------------------+-----------------+-----------------+
1418 | Hairpin | | | | DPDK 19.11 |
1419 | | | N/A | | OFED 4.7-3 |
1420 | | | | | rdma-core 26 |
1421 | | | | | ConnectX-5 |
1422 +-----------------------+-----------------+-----------------+
1423 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1424 | | | OFED 4.7-3 | | OFED 4.7-3 |
1425 | | | rdma-core 26 | | rdma-core 26 |
1426 | | | ConnectX-5 | | ConnectX-5 |
1427 +-----------------------+-----------------+-----------------+
1428 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1429 | | | OFED 4.7-3 | | OFED 4.7-3 |
1430 | | | rdma-core 26 | | rdma-core 26 |
1431 | | | ConnectX-5 | | ConnectX-5 |
1432 +-----------------------+-----------------+-----------------+
1433 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1434 | | | OFED 5.2 | | OFED 5.2 |
1435 | | | rdma-core 32 | | rdma-core 32 |
1436 | | | ConnectX-5 | | ConnectX-5 |
1437 +-----------------------+-----------------+-----------------+
1442 MARK and META items are interrelated with datapath - they might move from/to
1443 the applications in mbuf fields. Hence, zero value for these items has the
1444 special meaning - it means "no metadata are provided", not zero values are
1445 treated by applications and PMD as valid ones.
1447 Moreover in the flow engine domain the value zero is acceptable to match and
1448 set, and we should allow to specify zero values as rte_flow parameters for the
1449 META and MARK items and actions. In the same time zero mask has no meaning and
1450 should be rejected on validation stage.
1455 Flows are not cached in the driver.
1456 When stopping a device port, all the flows created on this port from the
1457 application will be flushed automatically in the background.
1458 After stopping the device port, all flows on this port become invalid and
1459 not represented in the system.
1460 All references to these flows held by the application should be discarded
1461 directly but neither destroyed nor flushed.
1463 The application should re-create the flows as required after the port restart.
1468 Compared to librte_net_mlx4 that implements a single RSS configuration per
1469 port, librte_net_mlx5 supports per-protocol RSS configuration.
1471 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1472 command-line parameter to enable additional protocols (UDP and TCP as well
1473 as IP), the following commands must be entered from its CLI to get the same
1474 behavior as librte_net_mlx4::
1477 > port config all rss all
1483 This section demonstrates how to launch **testpmd** with Mellanox
1484 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_net_mlx5.
1486 #. Load the kernel modules::
1488 modprobe -a ib_uverbs mlx5_core mlx5_ib
1490 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1493 /etc/init.d/openibd restart
1497 User space I/O kernel modules (uio and igb_uio) are not used and do
1498 not have to be loaded.
1500 #. Make sure Ethernet interfaces are in working order and linked to kernel
1501 verbs. Related sysfs entries should be present::
1503 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1512 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1515 for intf in eth2 eth3 eth4 eth5;
1517 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1520 sed -n 's,.*/\(.*\),-w \1,p'
1529 #. Request huge pages::
1531 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1533 #. Start testpmd with basic parameters::
1535 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
1540 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1541 EAL: probe driver: 15b3:1013 librte_net_mlx5
1542 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1543 PMD: librte_net_mlx5: 1 port(s) detected
1544 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1545 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1546 EAL: probe driver: 15b3:1013 librte_net_mlx5
1547 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1548 PMD: librte_net_mlx5: 1 port(s) detected
1549 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1550 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1551 EAL: probe driver: 15b3:1013 librte_net_mlx5
1552 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1553 PMD: librte_net_mlx5: 1 port(s) detected
1554 PMD: librte_net_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1555 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1556 EAL: probe driver: 15b3:1013 librte_net_mlx5
1557 PMD: librte_net_mlx5: PCI information matches, using device "mlx5_3" (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:fb
1560 Interactive-mode selected
1561 Configuring Port 0 (socket 0)
1562 PMD: librte_net_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1563 PMD: librte_net_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1564 Port 0: E4:1D:2D:E7:0C:FE
1565 Configuring Port 1 (socket 0)
1566 PMD: librte_net_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1567 PMD: librte_net_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1568 Port 1: E4:1D:2D:E7:0C:FF
1569 Configuring Port 2 (socket 0)
1570 PMD: librte_net_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1571 PMD: librte_net_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1572 Port 2: E4:1D:2D:E7:0C:FA
1573 Configuring Port 3 (socket 0)
1574 PMD: librte_net_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1575 PMD: librte_net_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1576 Port 3: E4:1D:2D:E7:0C:FB
1577 Checking link statuses...
1578 Port 0 Link Up - speed 40000 Mbps - full-duplex
1579 Port 1 Link Up - speed 40000 Mbps - full-duplex
1580 Port 2 Link Up - speed 10000 Mbps - full-duplex
1581 Port 3 Link Up - speed 10000 Mbps - full-duplex
1588 This section demonstrates how to dump flows. Currently, it's possible to dump
1589 all flows with assistance of external tools.
1591 #. 2 ways to get flow raw file:
1593 - Using testpmd CLI:
1595 .. code-block:: console
1597 testpmd> flow dump <port> <output_file>
1599 - call rte_flow_dev_dump api:
1601 .. code-block:: console
1603 rte_flow_dev_dump(port, file, NULL);
1605 #. Dump human-readable flows from raw file:
1607 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1609 .. code-block:: console
1611 mlx_steering_dump.py -f <output_file>