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_pmd_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>`__.
25 Due to external dependencies, this driver is disabled in default configuration
26 of the "make" build. It can be enabled with ``CONFIG_RTE_LIBRTE_MLX5_PMD=y``
27 or by using "meson" build system which will detect dependencies.
32 Besides its dependency on libibverbs (that implies libmlx5 and associated
33 kernel support), librte_pmd_mlx5 relies heavily on system calls for control
34 operations such as querying/updating the MTU and flow control parameters.
36 For security reasons and robustness, this driver only deals with virtual
37 memory addresses. The way resources allocations are handled by the kernel,
38 combined with hardware specifications that allow to handle virtual memory
39 addresses directly, ensure that DPDK applications cannot access random
40 physical memory (or memory that does not belong to the current process).
42 This capability allows the PMD to coexist with kernel network interfaces
43 which remain functional, although they stop receiving unicast packets as
44 long as they share the same MAC address.
45 This means legacy linux control tools (for example: ethtool, ifconfig and
46 more) can operate on the same network interfaces that owned by the DPDK
49 The PMD can use libibverbs and libmlx5 to access the device firmware
50 or directly the hardware components.
51 There are different levels of objects and bypassing abilities
52 to get the best performances:
54 - Verbs is a complete high-level generic API
55 - Direct Verbs is a device-specific API
56 - DevX allows to access firmware objects
57 - Direct Rules manages flow steering at low-level hardware layer
59 Enabling librte_pmd_mlx5 causes DPDK applications to be linked against
65 - Multi arch support: x86_64, POWER8, ARMv8, i686.
66 - Multiple TX and RX queues.
67 - Support for scattered TX and RX frames.
68 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
69 - RSS using different combinations of fields: L3 only, L4 only or both,
70 and source only, destination only or both.
71 - Several RSS hash keys, one for each flow type.
72 - Default RSS operation with no hash key specification.
73 - Configurable RETA table.
74 - Link flow control (pause frame).
75 - Support for multiple MAC addresses.
79 - RX CRC stripping configuration.
80 - Promiscuous mode on PF and VF.
81 - Multicast promiscuous mode on PF and VF.
82 - Hardware checksum offloads.
83 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
85 - Flow API, including :ref:`flow_isolated_mode`.
87 - KVM and VMware ESX SR-IOV modes are supported.
88 - RSS hash result is supported.
89 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
90 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
92 - Statistics query including Basic, Extended and per queue.
94 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP, Geneve, GTP.
95 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
96 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
97 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
98 - Flow insertion rate of more then million flows per second, when using Direct Rules.
99 - Support for multiple rte_flow groups.
100 - Per packet no-inline hint flag to disable packet data copying into Tx descriptors.
107 - For secondary process:
109 - Forked secondary process not supported.
110 - External memory unregistered in EAL memseg list cannot be used for DMA
111 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
112 primary process and remapped to the same virtual address in secondary
113 process. If the external memory is registered by primary process but has
114 different virtual address in secondary process, unexpected error may happen.
116 - When using Verbs flow engine (``dv_flow_en`` = 0), flow pattern without any
117 specific VLAN will match for VLAN packets as well:
119 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
120 Meaning, the flow rule::
122 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
124 Will only match vlan packets with vid=3. and the flow rule::
126 flow create 0 ingress pattern eth / ipv4 / end ...
128 Will match any ipv4 packet (VLAN included).
130 - When using DV flow engine (``dv_flow_en`` = 1), flow pattern without VLAN item
131 will match untagged packets only.
134 flow create 0 ingress pattern eth / ipv4 / end ...
136 Will match untagged packets only.
139 flow create 0 ingress pattern eth / vlan / ipv4 / end ...
141 Will match tagged packets only, with any VLAN ID value.
144 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
146 Will only match tagged packets with VLAN ID 3.
148 - VLAN pop offload command:
150 - Flow rules having a VLAN pop offload command as one of their actions and
151 are lacking a match on VLAN as one of their items are not supported.
152 - The command is not supported on egress traffic.
154 - VLAN push offload is not supported on ingress traffic.
156 - VLAN set PCP offload is not supported on existing headers.
158 - A multi segment packet must have not more segments than reported by dev_infos_get()
159 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
160 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
161 inline settings) to 58.
163 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
164 to 0 are not supported.
166 - VXLAN TSO and checksum offloads are not supported on VM.
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 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
198 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
199 ol_flags. As the mempool for the external buffer is managed by PMD, all the
200 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
201 the external buffers will be freed by PMD and the application which still
202 holds the external buffers may be corrupted.
204 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
205 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
206 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
208 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
209 and allmulticast mode are both set to off.
210 To receive IPv6 Multicast messages on VM, explicitly set the relevant
211 MAC address using rte_eth_dev_mac_addr_add() API.
213 - To support a mixed traffic pattern (some buffers from local host memory, some
214 buffers from other devices) with high bandwidth, a mbuf flag is used.
216 An application hints the PMD whether or not it should try to inline the
217 given mbuf data buffer. PMD should do the best effort to act upon this request.
219 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
220 registered by application with rte_mbuf_dynflag_register(). This flag is
221 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
222 which is intended to be used by the application.
224 To query the supported specific flags in runtime,
225 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
226 currently (over present hardware and configuration) supported specific flags.
227 The "not inline hint" feature operating flow is the following one:
230 - probe the devices, ports are created
231 - query the port capabilities
232 - if port supporting the feature is found
233 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
234 - application starts the ports
235 - on ``dev_start()`` PMD checks whether the feature flag is registered and
236 enables the feature support in datapath
237 - application might set the registered flag bit in ``ol_flags`` field
238 of mbuf being sent and PMD will handle ones appropriately.
240 - The amount of descriptors in Tx queue may be limited by data inline settings.
241 Inline data require the more descriptor building blocks and overall block
242 amount may exceed the hardware supported limits. The application should
243 reduce the requested Tx size or adjust data inline settings with
244 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
246 - E-Switch decapsulation Flow:
248 - can be applied to PF port only.
249 - must specify VF port action (packet redirection from PF to VF).
250 - optionally may specify tunnel inner source and destination MAC addresses.
252 - E-Switch encapsulation Flow:
254 - can be applied to VF ports only.
255 - must specify PF port action (packet redirection from VF to PF).
259 - The input buffer, used as outer header, is not validated.
263 - The decapsulation is always done up to the outermost tunnel detected by the HW.
264 - The input buffer, providing the removal size, is not validated.
265 - The buffer size must match the length of the headers to be removed.
267 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
268 mutually exclusive features which cannot be supported together
269 (see :ref:`mlx5_firmware_config`).
273 - Requires DevX and DV flow to be enabled.
274 - KEEP_CRC offload cannot be supported with LRO.
275 - The first mbuf length, without head-room, must be big enough to include the
277 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
278 it with size limited to max LRO size, not to max RX packet length.
279 - LRO can be used with outer header of TCP packets of the standard format:
280 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
282 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
287 MLX5 supports various methods to report statistics:
289 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.
291 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.
293 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.
301 These options can be modified in the ``.config`` file.
303 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
305 Toggle compilation of librte_pmd_mlx5 itself.
307 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
309 Build PMD with additional code to make it loadable without hard
310 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
311 on the target system.
313 In this mode, their presence is still required for it to run properly,
314 however their absence won't prevent a DPDK application from starting (with
315 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
316 missing with ``ldd(1)``.
318 It works by moving these dependencies to a purpose-built rdma-core "glue"
319 plug-in which must either be installed in a directory whose name is based
320 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
321 standard location for the dynamic linker (e.g. ``/lib``) if left to the
322 default empty string (``""``).
324 This option has no performance impact.
326 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
328 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
329 in the PMD shared library or the executable static binary.
331 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
333 Toggle debugging code and stricter compilation flags. Enabling this option
334 adds additional run-time checks and debugging messages at the cost of
339 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
340 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
341 64. Default armv8a configuration of make build and meson build set it to 128
342 then brings performance degradation.
344 This option is available in meson:
346 - ``ibverbs_link`` can be ``static``, ``shared``, or ``dlopen``.
348 Environment variables
349 ~~~~~~~~~~~~~~~~~~~~~
353 A list of directories in which to search for the rdma-core "glue" plug-in,
354 separated by colons or semi-colons.
356 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
357 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
358 since ``LD_LIBRARY_PATH`` has no effect in this case.
360 - ``MLX5_SHUT_UP_BF``
362 Configures HW Tx doorbell register as IO-mapped.
364 By default, the HW Tx doorbell is configured as a write-combining register.
365 The register would be flushed to HW usually when the write-combining buffer
366 becomes full, but it depends on CPU design.
368 Except for vectorized Tx burst routines, a write memory barrier is enforced
369 after updating the register so that the update can be immediately visible to
372 When vectorized Tx burst is called, the barrier is set only if the burst size
373 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
374 variable will bring better latency even though the maximum throughput can
377 Run-time configuration
378 ~~~~~~~~~~~~~~~~~~~~~~
380 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
381 because it is affected by their state. Forcing them down prevents packets
384 - **ethtool** operations on related kernel interfaces also affect the PMD.
389 In order to run as a non-root user,
390 some capabilities must be granted to the application::
392 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
394 Below are the reasons of the need for each capability:
397 When using physical addresses (PA mode), with Linux >= 4.0,
398 for access to ``/proc/self/pagemap``.
401 For device configuration.
404 For raw ethernet queue allocation through kernel driver.
407 For DMA memory pinning.
412 - ``rxq_cqe_comp_en`` parameter [int]
414 A nonzero value enables the compression of CQE on RX side. This feature
415 allows to save PCI bandwidth and improve performance. Enabled by default.
419 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
421 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
424 - ``rxq_cqe_pad_en`` parameter [int]
426 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
427 is aligned with the size of a cacheline of the core. If cacheline size is
428 128B, the CQE size is configured to be 128B even though the device writes
429 only 64B data on the cacheline. This is to avoid unnecessary cache
430 invalidation by device's two consecutive writes on to one cacheline.
431 However in some architecture, it is more beneficial to update entire
432 cacheline with padding the rest 64B rather than striding because
433 read-modify-write could drop performance a lot. On the other hand,
434 writing extra data will consume more PCIe bandwidth and could also drop
435 the maximum throughput. It is recommended to empirically set this
436 parameter. Disabled by default.
440 - CPU having 128B cacheline with ConnectX-5 and BlueField.
442 - ``rxq_pkt_pad_en`` parameter [int]
444 A nonzero value enables padding Rx packet to the size of cacheline on PCI
445 transaction. This feature would waste PCI bandwidth but could improve
446 performance by avoiding partial cacheline write which may cause costly
447 read-modify-copy in memory transaction on some architectures. Disabled by
452 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
454 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
457 - ``mprq_en`` parameter [int]
459 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
460 configured as Multi-Packet RQ if the total number of Rx queues is
461 ``rxqs_min_mprq`` or more. Disabled by default.
463 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
464 by posting a single large buffer for multiple packets. Instead of posting a
465 buffers per a packet, one large buffer is posted in order to receive multiple
466 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
467 and each stride receives one packet. MPRQ can improve throughput for
468 small-packet traffic.
470 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
471 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
472 configure large stride size enough to accommodate max_rx_pkt_len as long as
473 device allows. Note that this can waste system memory compared to enabling Rx
474 scatter and multi-segment packet.
476 - ``mprq_log_stride_num`` parameter [int]
478 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
479 strides can reduce PCIe traffic further. If configured value is not in the
480 range of device capability, the default value will be set with a warning
481 message. The default value is 4 which is 16 strides per a buffer, valid only
482 if ``mprq_en`` is set.
484 The size of Rx queue should be bigger than the number of strides.
486 - ``mprq_log_stride_size`` parameter [int]
488 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
489 stride size can save some memory and reduce probability of a depletion of all
490 available strides due to unreleased packets by an application. If configured
491 value is not in the range of device capability, the default value will be set
492 with a warning message. The default value is 11 which is 2048 bytes per a
493 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
494 it is possible for a pcaket to span across multiple strides. This mode allows
495 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
496 of a packet if Rx scatter is configured) may be required in case there is no
497 space left for a head room at the end of a stride which incurs some
500 - ``mprq_max_memcpy_len`` parameter [int]
502 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
503 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
504 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
505 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
506 A mempool for external buffers will be allocated and managed by PMD. If Rx
507 packet is externally attached, ol_flags field of the mbuf will have
508 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
509 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
511 - ``rxqs_min_mprq`` parameter [int]
513 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
514 greater or equal to this value. The default value is 12, valid only if
517 - ``txq_inline`` parameter [int]
519 Amount of data to be inlined during TX operations. This parameter is
520 deprecated and converted to the new parameter ``txq_inline_max`` providing
521 partial compatibility.
523 - ``txqs_min_inline`` parameter [int]
525 Enable inline data send only when the number of TX queues is greater or equal
528 This option should be used in combination with ``txq_inline_max`` and
529 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
531 If this option is not specified the default value 16 is used for BlueField
532 and 8 for other platforms
534 The data inlining consumes the CPU cycles, so this option is intended to
535 auto enable inline data if we have enough Tx queues, which means we have
536 enough CPU cores and PCI bandwidth is getting more critical and CPU
537 is not supposed to be bottleneck anymore.
539 The copying data into WQE improves latency and can improve PPS performance
540 when PCI back pressure is detected and may be useful for scenarios involving
541 heavy traffic on many queues.
543 Because additional software logic is necessary to handle this mode, this
544 option should be used with care, as it may lower performance when back
545 pressure is not expected.
547 If inline data are enabled it may affect the maximal size of Tx queue in
548 descriptors because the inline data increase the descriptor size and
549 queue size limits supported by hardware may be exceeded.
551 - ``txq_inline_min`` parameter [int]
553 Minimal amount of data to be inlined into WQE during Tx operations. NICs
554 may require this minimal data amount to operate correctly. The exact value
555 may depend on NIC operation mode, requested offloads, etc. It is strongly
556 recommended to omit this parameter and use the default values. Anyway,
557 applications using this parameter should take into consideration that
558 specifying an inconsistent value may prevent the NIC from sending packets.
560 If ``txq_inline_min`` key is present the specified value (may be aligned
561 by the driver in order not to exceed the limits and provide better descriptor
562 space utilization) will be used by the driver and it is guaranteed that
563 requested amount of data bytes are inlined into the WQE beside other inline
564 settings. This key also may update ``txq_inline_max`` value (default
565 or specified explicitly in devargs) to reserve the space for inline data.
567 If ``txq_inline_min`` key is not present, the value may be queried by the
568 driver from the NIC via DevX if this feature is available. If there is no DevX
569 enabled/supported the value 18 (supposing L2 header including VLAN) is set
570 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
571 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
574 For ConnectX-4 NIC, driver does not allow specifying value below 18
575 (minimal L2 header, including VLAN), error will be raised.
577 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
578 it is not recommended and may prevent NIC from sending packets over
581 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
582 Multi-Packet Write), because last one does not support partial packet inlining.
583 This is not very critical due to minimal data inlining is mostly required
584 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
586 - ``txq_inline_max`` parameter [int]
588 Specifies the maximal packet length to be completely inlined into WQE
589 Ethernet Segment for ordinary SEND method. If packet is larger than specified
590 value, the packet data won't be copied by the driver at all, data buffer
591 is addressed with a pointer. If packet length is less or equal all packet
592 data will be copied into WQE. This may improve PCI bandwidth utilization for
593 short packets significantly but requires the extra CPU cycles.
595 The data inline feature is controlled by number of Tx queues, if number of Tx
596 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
597 is engaged, if there are not enough Tx queues (which means not enough CPU cores
598 and CPU resources are scarce), data inline is not performed by the driver.
599 Assigning ``txqs_min_inline`` with zero always enables the data inline.
601 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
602 by the driver in order not to exceed the limit (930 bytes) and to provide better
603 WQE space filling without gaps, the adjustment is reflected in the debug log.
604 Also, the default value (290) may be decreased in run-time if the large transmit
605 queue size is requested and hardware does not support enough descriptor
606 amount, in this case warning is emitted. If ``txq_inline_max`` key is
607 specified and requested inline settings can not be satisfied then error
610 - ``txq_inline_mpw`` parameter [int]
612 Specifies the maximal packet length to be completely inlined into WQE for
613 Enhanced MPW method. If packet is large the specified value, the packet data
614 won't be copied, and data buffer is addressed with pointer. If packet length
615 is less or equal, all packet data will be copied into WQE. This may improve PCI
616 bandwidth utilization for short packets significantly but requires the extra
619 The data inline feature is controlled by number of TX queues, if number of Tx
620 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
621 is engaged, if there are not enough Tx queues (which means not enough CPU cores
622 and CPU resources are scarce), data inline is not performed by the driver.
623 Assigning ``txqs_min_inline`` with zero always enables the data inline.
625 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
626 by the driver in order not to exceed the limit (930 bytes) and to provide better
627 WQE space filling without gaps, the adjustment is reflected in the debug log.
628 Due to multiple packets may be included to the same WQE with Enhanced Multi
629 Packet Write Method and overall WQE size is limited it is not recommended to
630 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
631 may be decreased in run-time if the large transmit queue size is requested
632 and hardware does not support enough descriptor amount, in this case warning
633 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
634 settings can not be satisfied then error will be raised.
636 - ``txqs_max_vec`` parameter [int]
638 Enable vectorized Tx only when the number of TX queues is less than or
639 equal to this value. This parameter is deprecated and ignored, kept
640 for compatibility issue to not prevent driver from probing.
642 - ``txq_mpw_hdr_dseg_en`` parameter [int]
644 A nonzero value enables including two pointers in the first block of TX
645 descriptor. The parameter is deprecated and ignored, kept for compatibility
648 - ``txq_max_inline_len`` parameter [int]
650 Maximum size of packet to be inlined. This limits the size of packet to
651 be inlined. If the size of a packet is larger than configured value, the
652 packet isn't inlined even though there's enough space remained in the
653 descriptor. Instead, the packet is included with pointer. This parameter
654 is deprecated and converted directly to ``txq_inline_mpw`` providing full
655 compatibility. Valid only if eMPW feature is engaged.
657 - ``txq_mpw_en`` parameter [int]
659 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
660 ConnectX-6, ConnectX-6 Dx and BlueField. eMPW allows the TX burst function to pack
661 up multiple packets in a single descriptor session in order to save PCI bandwidth
662 and improve performance at the cost of a slightly higher CPU usage. When
663 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
664 entire packet data on to TX descriptor instead of including pointer of packet.
666 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
667 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
668 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
669 option or reported by the NIC, the eMPW feature is disengaged.
671 - ``tx_db_nc`` parameter [int]
673 The rdma core library can map doorbell register in two ways, depending on the
674 environment variable "MLX5_SHUT_UP_BF":
676 - As regular cached memory (usually with write combining attribute), if the
677 variable is either missing or set to zero.
678 - As non-cached memory, if the variable is present and set to not "0" value.
680 The type of mapping may slightly affect the Tx performance, the optimal choice
681 is strongly relied on the host architecture and should be deduced practically.
683 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
684 memory (with write combining), the PMD will perform the extra write memory barrier
685 after writing to doorbell, it might increase the needed CPU clocks per packet
686 to send, but latency might be improved.
688 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
689 cached memory, the PMD will not perform the extra write memory barrier
690 after writing to doorbell, on some architectures it might improve the
693 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
694 memory, the PMD will use heuristics to decide whether write memory barrier
695 should be performed. For bursts with size multiple of recommended one (64 pkts)
696 it is supposed the next burst is coming and no need to issue the extra memory
697 barrier (it is supposed to be issued in the next coming burst, at least after
698 descriptor writing). It might increase latency (on some hosts till next
699 packets transmit) and should be used with care.
701 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
702 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
703 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
705 - ``tx_vec_en`` parameter [int]
707 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx
708 and BlueField NICs if the number of global Tx queues on the port is less than
709 ``txqs_max_vec``. The parameter is deprecated and ignored.
711 - ``rx_vec_en`` parameter [int]
713 A nonzero value enables Rx vector if the port is not configured in
714 multi-segment otherwise this parameter is ignored.
718 - ``vf_nl_en`` parameter [int]
720 A nonzero value enables Netlink requests from the VF to add/remove MAC
721 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
722 Otherwise the relevant configuration must be run with Linux iproute2 tools.
723 This is a prerequisite to receive this kind of traffic.
725 Enabled by default, valid only on VF devices ignored otherwise.
727 - ``l3_vxlan_en`` parameter [int]
729 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
730 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
731 parameter. This is a prerequisite to receive this kind of traffic.
735 - ``dv_xmeta_en`` parameter [int]
737 A nonzero value enables extensive flow metadata support if device is
738 capable and driver supports it. This can enable extensive support of
739 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
740 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
742 There are some possible configurations, depending on parameter value:
744 - 0, this is default value, defines the legacy mode, the ``MARK`` and
745 ``META`` related actions and items operate only within NIC Tx and
746 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
747 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
748 item is 32 bits wide and match supported on egress only.
750 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
751 related actions and items operate within all supported steering domains,
752 including FDB, ``MARK`` and ``META`` information may cross the domain
753 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
754 depends on kernel and firmware configurations and might be 0, 16 or
755 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
756 compatibility, the actual width of data transferred to the FDB domain
757 depends on kernel configuration and may be vary. The actual supported
758 width can be retrieved in runtime by series of rte_flow_validate()
761 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
762 related actions and items operate within all supported steering domains,
763 including FDB, ``MARK`` and ``META`` information may cross the domain
764 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
765 depends on kernel and firmware configurations and might be 0, 16 or
766 24 bits. The actual supported width can be retrieved in runtime by
767 series of rte_flow_validate() trials.
769 +------+-----------+-----------+-------------+-------------+
770 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
771 +======+===========+===========+=============+=============+
772 | 0 | 24 bits | 32 bits | 32 bits | no |
773 +------+-----------+-----------+-------------+-------------+
774 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
775 +------+-----------+-----------+-------------+-------------+
776 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
777 +------+-----------+-----------+-------------+-------------+
779 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
780 ignored and the device is configured to operate in legacy mode (0).
782 Disabled by default (set to 0).
784 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
785 of the extensive metadata features. The legacy Verbs supports FLAG and
786 MARK metadata actions over NIC Rx steering domain only.
788 - ``dv_flow_en`` parameter [int]
790 A nonzero value enables the DV flow steering assuming it is supported
791 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
793 Enabled by default if supported.
795 - ``dv_esw_en`` parameter [int]
797 A nonzero value enables E-Switch using Direct Rules.
799 Enabled by default if supported.
801 - ``mr_ext_memseg_en`` parameter [int]
803 A nonzero value enables extending memseg when registering DMA memory. If
804 enabled, the number of entries in MR (Memory Region) lookup table on datapath
805 is minimized and it benefits performance. On the other hand, it worsens memory
806 utilization because registered memory is pinned by kernel driver. Even if a
807 page in the extended chunk is freed, that doesn't become reusable until the
808 entire memory is freed.
812 - ``representor`` parameter [list]
814 This parameter can be used to instantiate DPDK Ethernet devices from
815 existing port (or VF) representors configured on the device.
817 It is a standard parameter whose format is described in
818 :ref:`ethernet_device_standard_device_arguments`.
820 For instance, to probe port representors 0 through 2::
824 - ``max_dump_files_num`` parameter [int]
826 The maximum number of files per PMD entity that may be created for debug information.
827 The files will be created in /var/log directory or in current directory.
829 set to 128 by default.
831 - ``lro_timeout_usec`` parameter [int]
833 The maximum allowed duration of an LRO session, in micro-seconds.
834 PMD will set the nearest value supported by HW, which is not bigger than
835 the input ``lro_timeout_usec`` value.
836 If this parameter is not specified, by default PMD will set
837 the smallest value supported by HW.
839 - ``hp_buf_log_sz`` parameter [int]
841 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
842 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
843 The capacity of the value is specified by the firmware and the initialization
844 will get a failure if it is out of scope.
845 The range of the value is from 11 to 19 right now, and the supported frame
846 size of a single packet for hairpin is from 512B to 128KB. It might change if
847 different firmware release is being used. By using a small value, it could
848 reduce memory consumption but not work with a large frame. If the value is
849 too large, the memory consumption will be high and some potential performance
850 degradation will be introduced.
851 By default, the PMD will set this value to 16, which means that 9KB jumbo
852 frames will be supported.
854 .. _mlx5_firmware_config:
856 Firmware configuration
857 ~~~~~~~~~~~~~~~~~~~~~~
859 Firmware features can be configured as key/value pairs.
861 The command to set a value is::
863 mlxconfig -d <device> set <key>=<value>
865 The command to query a value is::
867 mlxconfig -d <device> query | grep <key>
869 The device name for the command ``mlxconfig`` can be either the PCI address,
870 or the mst device name found with::
874 Below are some firmware configurations listed.
880 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
886 - maximum number of SR-IOV virtual functions::
890 - enable DevX (required by Direct Rules and other features)::
894 - aggressive CQE zipping::
898 - L3 VXLAN and VXLAN-GPE destination UDP port::
901 IP_OVER_VXLAN_PORT=<udp dport>
903 - enable VXLAN-GPE tunnel flow matching::
905 FLEX_PARSER_PROFILE_ENABLE=0
907 FLEX_PARSER_PROFILE_ENABLE=2
909 - enable IP-in-IP tunnel flow matching::
911 FLEX_PARSER_PROFILE_ENABLE=0
913 - enable MPLS flow matching::
915 FLEX_PARSER_PROFILE_ENABLE=1
917 - enable ICMP/ICMP6 code/type fields matching::
919 FLEX_PARSER_PROFILE_ENABLE=2
921 - enable Geneve flow matching::
923 FLEX_PARSER_PROFILE_ENABLE=0
925 FLEX_PARSER_PROFILE_ENABLE=1
927 - enable GTP flow matching::
929 FLEX_PARSER_PROFILE_ENABLE=3
934 This driver relies on external libraries and kernel drivers for resources
935 allocations and initialization. The following dependencies are not part of
936 DPDK and must be installed separately:
940 User space Verbs framework used by librte_pmd_mlx5. This library provides
941 a generic interface between the kernel and low-level user space drivers
944 It allows slow and privileged operations (context initialization, hardware
945 resources allocations) to be managed by the kernel and fast operations to
946 never leave user space.
950 Low-level user space driver library for Mellanox
951 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
954 This library basically implements send/receive calls to the hardware
959 They provide the kernel-side Verbs API and low level device drivers that
960 manage actual hardware initialization and resources sharing with user
963 Unlike most other PMDs, these modules must remain loaded and bound to
966 - mlx5_core: hardware driver managing Mellanox
967 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
969 - mlx5_ib: InifiniBand device driver.
970 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
972 - **Firmware update**
974 Mellanox OFED/EN releases include firmware updates for
975 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
977 Because each release provides new features, these updates must be applied to
978 match the kernel modules and libraries they come with.
982 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
988 Either RDMA Core library with a recent enough Linux kernel release
989 (recommended) or Mellanox OFED/EN, which provides compatibility with older
992 RDMA Core with Linux Kernel
993 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
995 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
996 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
997 (see `RDMA Core installation documentation`_)
998 - When building for i686 use:
1000 - rdma-core version 18.0 or above built with 32bit support.
1001 - Kernel version 4.14.41 or above.
1003 - Starting with rdma-core v21, static libraries can be built::
1006 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1009 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1010 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1012 If rdma-core libraries are built but not installed, DPDK makefile can link them,
1013 thanks to these environment variables:
1015 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
1016 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
1017 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
1022 - Mellanox OFED version: **4.5** and above /
1023 Mellanox EN version: **4.5** and above
1026 - ConnectX-4: **12.21.1000** and above.
1027 - ConnectX-4 Lx: **14.21.1000** and above.
1028 - ConnectX-5: **16.21.1000** and above.
1029 - ConnectX-5 Ex: **16.21.1000** and above.
1030 - ConnectX-6: **20.27.0090** and above.
1031 - ConnectX-6 Dx: **22.27.0090** and above.
1032 - BlueField: **18.25.1010** and above.
1034 While these libraries and kernel modules are available on OpenFabrics
1035 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1036 managers on most distributions, this PMD requires Ethernet extensions that
1037 may not be supported at the moment (this is a work in progress).
1040 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1042 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1043 include the necessary support and should be used in the meantime. For DPDK,
1044 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1045 required from that distribution.
1049 Several versions of Mellanox OFED/EN are available. Installing the version
1050 this DPDK release was developed and tested against is strongly
1051 recommended. Please check the `prerequisites`_.
1056 The following Mellanox device families are supported by the same mlx5 driver:
1066 Below are detailed device names:
1068 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1069 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1070 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1071 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1072 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1073 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1074 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1075 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1076 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1077 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1078 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1079 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1080 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1081 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1082 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1083 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1084 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1085 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1086 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1087 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1088 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1089 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1090 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1091 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1092 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1094 Quick Start Guide on OFED/EN
1095 ----------------------------
1097 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
1100 2. Install the required libraries and kernel modules either by installing
1101 only the required set, or by installing the entire Mellanox OFED/EN::
1103 ./mlnxofedinstall --upstream-libs --dpdk
1105 3. Verify the firmware is the correct one::
1109 4. Verify all ports links are set to Ethernet::
1111 mlxconfig -d <mst device> query | grep LINK_TYPE
1115 Link types may have to be configured to Ethernet::
1117 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1119 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1121 For hypervisors, verify SR-IOV is enabled on the NIC::
1123 mlxconfig -d <mst device> query | grep SRIOV_EN
1126 If needed, configure SR-IOV::
1128 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1129 mlxfwreset -d <mst device> reset
1131 5. Restart the driver::
1133 /etc/init.d/openibd restart
1137 service openibd restart
1139 If link type was changed, firmware must be reset as well::
1141 mlxfwreset -d <mst device> reset
1143 For hypervisors, after reset write the sysfs number of virtual functions
1146 To dynamically instantiate a given number of virtual functions (VFs)::
1148 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1150 6. Compile DPDK and you are ready to go. See instructions on
1151 :ref:`Development Kit Build System <Development_Kit_Build_System>`
1153 Enable switchdev mode
1154 ---------------------
1156 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1157 Representor is a port in DPDK that is connected to a VF in such a way
1158 that assuming there are no offload flows, each packet that is sent from the VF
1159 will be received by the corresponding representor. While each packet that is
1160 sent to a representor will be received by the VF.
1161 This is very useful in case of SRIOV mode, where the first packet that is sent
1162 by the VF will be received by the DPDK application which will decide if this
1163 flow should be offloaded to the E-Switch. After offloading the flow packet
1164 that the VF that are matching the flow will not be received any more by
1165 the DPDK application.
1167 1. Enable SRIOV mode::
1169 mlxconfig -d <mst device> set SRIOV_EN=true
1171 2. Configure the max number of VFs::
1173 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1177 mlxfwreset -d <mst device> reset
1179 3. Configure the actual number of VFs::
1181 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1183 4. Unbind the device (can be rebind after the switchdev mode)::
1185 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1187 5. Enbale switchdev mode::
1189 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1194 1. Configure aggressive CQE Zipping for maximum performance::
1196 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1198 To set it back to the default CQE Zipping mode use::
1200 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1202 2. In case of virtualization:
1204 - Make sure that hypervisor kernel is 3.16 or newer.
1205 - Configure boot with ``iommu=pt``.
1206 - Use 1G huge pages.
1207 - Make sure to allocate a VM on huge pages.
1208 - Make sure to set CPU pinning.
1210 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1211 for better performance. For VMs, verify that the right CPU
1212 and NUMA node are pinned according to the above. Run::
1216 to identify the NUMA node to which the PCIe adapter is connected.
1218 4. If more than one adapter is used, and root complex capabilities allow
1219 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1220 it is recommended to locate both adapters on the same NUMA node.
1221 This in order to forward packets from one to the other without
1222 NUMA performance penalty.
1224 5. Disable pause frames::
1226 ethtool -A <netdev> rx off tx off
1228 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1229 via the BIOS configuration. Please contact you server provider for more
1230 information about the settings.
1234 On some machines, depends on the machine integrator, it is beneficial
1235 to set the PCI max read request parameter to 1K. This can be
1236 done in the following way:
1238 To query the read request size use::
1240 setpci -s <NIC PCI address> 68.w
1242 If the output is different than 3XXX, set it by::
1244 setpci -s <NIC PCI address> 68.w=3XXX
1246 The XXX can be different on different systems. Make sure to configure
1247 according to the setpci output.
1249 7. To minimize overhead of searching Memory Regions:
1251 - '--socket-mem' is recommended to pin memory by predictable amount.
1252 - Configure per-lcore cache when creating Mempools for packet buffer.
1253 - Refrain from dynamically allocating/freeing memory in run-time.
1255 .. _mlx5_offloads_support:
1257 Supported hardware offloads
1258 ---------------------------
1260 .. table:: Minimal SW/HW versions for queue offloads
1262 ============== ===== ===== ========= ===== ========== ==========
1263 Offload DPDK Linux rdma-core OFED firmware hardware
1264 ============== ===== ===== ========= ===== ========== ==========
1265 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1266 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1267 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1268 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1269 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1270 ============== ===== ===== ========= ===== ========== ==========
1272 .. table:: Minimal SW/HW versions for rte_flow offloads
1274 +-----------------------+-----------------+-----------------+
1275 | Offload | with E-Switch | with NIC |
1276 +=======================+=================+=================+
1277 | Count | | DPDK 19.05 | | DPDK 19.02 |
1278 | | | OFED 4.6 | | OFED 4.6 |
1279 | | | rdma-core 24 | | rdma-core 23 |
1280 | | | ConnectX-5 | | ConnectX-5 |
1281 +-----------------------+-----------------+-----------------+
1282 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1283 | | | OFED 4.6 | | OFED 4.5 |
1284 | | | rdma-core 24 | | rdma-core 23 |
1285 | | | ConnectX-5 | | ConnectX-4 |
1286 +-----------------------+-----------------+-----------------+
1287 | Queue / RSS | | | | DPDK 18.11 |
1288 | | | N/A | | OFED 4.5 |
1289 | | | | | rdma-core 23 |
1290 | | | | | ConnectX-4 |
1291 +-----------------------+-----------------+-----------------+
1292 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1293 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1294 | | | rdma-core 24 | | rdma-core 23 |
1295 | | | ConnectX-5 | | ConnectX-5 |
1296 +-----------------------+-----------------+-----------------+
1297 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1298 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1299 | | | rdma-core 27 | | rdma-core 27 |
1300 | | | ConnectX-5 | | ConnectX-5 |
1301 +-----------------------+-----------------+-----------------+
1302 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1303 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1304 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1305 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1306 | | set_ipv6_dst / | | | | |
1307 | | set_tp_src / | | | | |
1308 | | set_tp_dst / | | | | |
1309 | | dec_ttl / | | | | |
1310 | | set_ttl / | | | | |
1311 | | set_mac_src / | | | | |
1312 | | set_mac_dst) | | | | |
1313 +-----------------------+-----------------+-----------------+
1314 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1315 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1316 | | | | rdma-core 24 | | rdma-core 24 |
1317 | | | | ConnectX-5 | | ConnectX-5 |
1318 +-----------------------+-----------------+-----------------+
1319 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1320 | | | OFED 4.7-1 | | OFED 4.7-1 |
1321 | | | rdma-core 24 | | N/A |
1322 | | | ConnectX-5 | | ConnectX-5 |
1323 +-----------------------+-----------------+-----------------+
1324 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1325 | | | OFED 4.6 | | OFED 4.5 |
1326 | | | rdma-core 24 | | rdma-core 23 |
1327 | | | ConnectX-5 | | ConnectX-4 |
1328 +-----------------------+-----------------+-----------------+
1329 | Port ID | | DPDK 19.05 | | N/A |
1330 | | | OFED 4.7-1 | | N/A |
1331 | | | rdma-core 24 | | N/A |
1332 | | | ConnectX-5 | | N/A |
1333 +-----------------------+-----------------+-----------------+
1334 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1335 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1336 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1337 | | of_set_vlan_pcp / | | | | |
1338 | | of_set_vlan_vid) | | | | |
1339 +-----------------------+-----------------+-----------------+
1340 | Hairpin | | | | DPDK 19.11 |
1341 | | | N/A | | OFED 4.7-3 |
1342 | | | | | rdma-core 26 |
1343 | | | | | ConnectX-5 |
1344 +-----------------------+-----------------+-----------------+
1345 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1346 | | | OFED 4.7-3 | | OFED 4.7-3 |
1347 | | | rdma-core 26 | | rdma-core 26 |
1348 | | | ConnectX-5 | | ConnectX-5 |
1349 +-----------------------+-----------------+-----------------+
1350 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1351 | | | OFED 4.7-3 | | OFED 4.7-3 |
1352 | | | rdma-core 26 | | rdma-core 26 |
1353 | | | ConnectX-5 | | ConnectX-5 |
1354 +-----------------------+-----------------+-----------------+
1359 MARK and META items are interrelated with datapath - they might move from/to
1360 the applications in mbuf fields. Hence, zero value for these items has the
1361 special meaning - it means "no metadata are provided", not zero values are
1362 treated by applications and PMD as valid ones.
1364 Moreover in the flow engine domain the value zero is acceptable to match and
1365 set, and we should allow to specify zero values as rte_flow parameters for the
1366 META and MARK items and actions. In the same time zero mask has no meaning and
1367 should be rejected on validation stage.
1372 Flows are not cached in the driver.
1373 When stopping a device port, all the flows created on this port from the
1374 application will be flushed automatically in the background.
1375 After stopping the device port, all flows on this port become invalid and
1376 not represented in the system.
1377 All references to these flows held by the application should be discarded
1378 directly but neither destroyed nor flushed.
1380 The application should re-create the flows as required after the port restart.
1385 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1386 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1388 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1389 command-line parameter to enable additional protocols (UDP and TCP as well
1390 as IP), the following commands must be entered from its CLI to get the same
1391 behavior as librte_pmd_mlx4::
1394 > port config all rss all
1400 This section demonstrates how to launch **testpmd** with Mellanox
1401 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1403 #. Load the kernel modules::
1405 modprobe -a ib_uverbs mlx5_core mlx5_ib
1407 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1410 /etc/init.d/openibd restart
1414 User space I/O kernel modules (uio and igb_uio) are not used and do
1415 not have to be loaded.
1417 #. Make sure Ethernet interfaces are in working order and linked to kernel
1418 verbs. Related sysfs entries should be present::
1420 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1429 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1432 for intf in eth2 eth3 eth4 eth5;
1434 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1437 sed -n 's,.*/\(.*\),-w \1,p'
1446 #. Request huge pages::
1448 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1450 #. Start testpmd with basic parameters::
1452 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
1457 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1458 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1459 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1460 PMD: librte_pmd_mlx5: 1 port(s) detected
1461 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1462 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1463 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1464 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1465 PMD: librte_pmd_mlx5: 1 port(s) detected
1466 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1467 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1468 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1469 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1470 PMD: librte_pmd_mlx5: 1 port(s) detected
1471 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1472 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1473 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1474 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1475 PMD: librte_pmd_mlx5: 1 port(s) detected
1476 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1477 Interactive-mode selected
1478 Configuring Port 0 (socket 0)
1479 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1480 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1481 Port 0: E4:1D:2D:E7:0C:FE
1482 Configuring Port 1 (socket 0)
1483 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1484 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1485 Port 1: E4:1D:2D:E7:0C:FF
1486 Configuring Port 2 (socket 0)
1487 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1488 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1489 Port 2: E4:1D:2D:E7:0C:FA
1490 Configuring Port 3 (socket 0)
1491 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1492 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1493 Port 3: E4:1D:2D:E7:0C:FB
1494 Checking link statuses...
1495 Port 0 Link Up - speed 40000 Mbps - full-duplex
1496 Port 1 Link Up - speed 40000 Mbps - full-duplex
1497 Port 2 Link Up - speed 10000 Mbps - full-duplex
1498 Port 3 Link Up - speed 10000 Mbps - full-duplex
1505 This section demonstrates how to dump flows. Currently, it's possible to dump
1506 all flows with assistance of external tools.
1508 #. 2 ways to get flow raw file:
1510 - Using testpmd CLI:
1512 .. code-block:: console
1514 testpmd> flow dump <port> <output_file>
1516 - call rte_flow_dev_dump api:
1518 .. code-block:: console
1520 rte_flow_dev_dump(port, file, NULL);
1522 #. Dump human-readable flows from raw file:
1524 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1526 .. code-block:: console
1528 mlx_steering_dump.py -f <output_file>