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 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
168 - Match on Geneve header supports the following fields only:
174 Currently, the only supported options length value is 0.
176 - VF: flow rules created on VF devices can only match traffic targeted at the
177 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
179 - Match on GTP tunnel header item supports the following fields only:
181 - v_pt_rsv_flags: E flag, S flag, PN flag
185 - No Tx metadata go to the E-Switch steering domain for the Flow group 0.
186 The flows within group 0 and set metadata action are rejected by hardware.
190 MAC addresses not already present in the bridge table of the associated
191 kernel network device will be added and cleaned up by the PMD when closing
192 the device. In case of ungraceful program termination, some entries may
193 remain present and should be removed manually by other means.
195 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
196 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
197 ol_flags. As the mempool for the external buffer is managed by PMD, all the
198 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
199 the external buffers will be freed by PMD and the application which still
200 holds the external buffers may be corrupted.
202 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
203 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
204 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
206 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
207 and allmulticast mode are both set to off.
208 To receive IPv6 Multicast messages on VM, explicitly set the relevant
209 MAC address using rte_eth_dev_mac_addr_add() API.
211 - To support a mixed traffic pattern (some buffers from local host memory, some
212 buffers from other devices) with high bandwidth, a mbuf flag is used.
214 An application hints the PMD whether or not it should try to inline the
215 given mbuf data buffer. PMD should do the best effort to act upon this request.
217 The hint flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE`` is dynamic,
218 registered by application with rte_mbuf_dynflag_register(). This flag is
219 purely driver-specific and declared in PMD specific header ``rte_pmd_mlx5.h``,
220 which is intended to be used by the application.
222 To query the supported specific flags in runtime,
223 the function ``rte_pmd_mlx5_get_dyn_flag_names`` returns the array of
224 currently (over present hardware and configuration) supported specific flags.
225 The "not inline hint" feature operating flow is the following one:
228 - probe the devices, ports are created
229 - query the port capabilities
230 - if port supporting the feature is found
231 - register dynamic flag ``RTE_PMD_MLX5_FINE_GRANULARITY_INLINE``
232 - application starts the ports
233 - on ``dev_start()`` PMD checks whether the feature flag is registered and
234 enables the feature support in datapath
235 - application might set the registered flag bit in ``ol_flags`` field
236 of mbuf being sent and PMD will handle ones appropriately.
238 - The amount of descriptors in Tx queue may be limited by data inline settings.
239 Inline data require the more descriptor building blocks and overall block
240 amount may exceed the hardware supported limits. The application should
241 reduce the requested Tx size or adjust data inline settings with
242 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
244 - To provide the packet send scheduling on mbuf timestamps the ``tx_pp``
245 parameter should be specified, RTE_MBUF_DYNFIELD_TIMESTAMP_NAME and
246 RTE_MBUF_DYNFLAG_TIMESTAMP_NAME should be registered by application.
247 When PMD sees the RTE_MBUF_DYNFLAG_TIMESTAMP_NAME set on the packet
248 being sent it tries to synchronize the time of packet appearing on
249 the wire with the specified packet timestamp. It the specified one
250 is in the past it should be ignored, if one is in the distant future
251 it should be capped with some reasonable value (in range of seconds).
252 These specific cases ("too late" and "distant future") can be optionally
253 reported via device xstats to assist applications to detect the
254 time-related problems.
256 The timestamp upper "too-distant-future" limit
257 at the moment of invoking the Tx burst routine
258 can be estimated as ``tx_pp`` option (in nanoseconds) multiplied by 2^23.
259 Please note, for the testpmd txonly mode,
260 the limit is deduced from the expression::
262 (n_tx_descriptors / burst_size + 1) * inter_burst_gap
264 There is no any packet reordering according timestamps is supposed,
265 neither within packet burst, nor between packets, it is an entirely
266 application responsibility to generate packets and its timestamps
267 in desired order. The timestamps can be put only in the first packet
268 in the burst providing the entire burst scheduling.
270 - E-Switch decapsulation Flow:
272 - can be applied to PF port only.
273 - must specify VF port action (packet redirection from PF to VF).
274 - optionally may specify tunnel inner source and destination MAC addresses.
276 - E-Switch encapsulation Flow:
278 - can be applied to VF ports only.
279 - must specify PF port action (packet redirection from VF to PF).
283 - The input buffer, used as outer header, is not validated.
287 - The decapsulation is always done up to the outermost tunnel detected by the HW.
288 - The input buffer, providing the removal size, is not validated.
289 - The buffer size must match the length of the headers to be removed.
291 - ICMP(code/type/identifier/sequence number) / ICMP6(code/type) matching, IP-in-IP and MPLS flow matching are all
292 mutually exclusive features which cannot be supported together
293 (see :ref:`mlx5_firmware_config`).
297 - Requires DevX and DV flow to be enabled.
298 - KEEP_CRC offload cannot be supported with LRO.
299 - The first mbuf length, without head-room, must be big enough to include the
301 - Rx queue with LRO offload enabled, receiving a non-LRO packet, can forward
302 it with size limited to max LRO size, not to max RX packet length.
303 - LRO can be used with outer header of TCP packets of the standard format:
304 eth (with or without vlan) / ipv4 or ipv6 / tcp / payload
306 Other TCP packets (e.g. with MPLS label) received on Rx queue with LRO enabled, will be received with bad checksum.
310 - ``DEV_RX_OFFLOAD_KEEP_CRC`` cannot be supported with decapsulation
311 for some NICs (such as ConnectX-6 Dx and BlueField 2).
312 The capability bit ``scatter_fcs_w_decap_disable`` shows NIC support.
316 - Supports ``RTE_FLOW_ACTION_TYPE_SAMPLE`` action only within NIC Rx and E-Switch steering domain.
317 - The E-Switch Sample flow must have the eswitch_manager VPORT destination (PF or ECPF) and no additional actions.
318 - 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.
323 MLX5 supports various methods to report statistics:
325 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.
327 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.
329 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.
337 These options can be modified in the ``.config`` file.
339 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
341 Toggle compilation of librte_pmd_mlx5 itself.
343 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
345 Build PMD with additional code to make it loadable without hard
346 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
347 on the target system.
349 In this mode, their presence is still required for it to run properly,
350 however their absence won't prevent a DPDK application from starting (with
351 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
352 missing with ``ldd(1)``.
354 It works by moving these dependencies to a purpose-built rdma-core "glue"
355 plug-in which must either be installed in a directory whose name is based
356 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
357 standard location for the dynamic linker (e.g. ``/lib``) if left to the
358 default empty string (``""``).
360 This option has no performance impact.
362 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
364 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
365 in the PMD shared library or the executable static binary.
367 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
369 Toggle debugging code and stricter compilation flags. Enabling this option
370 adds additional run-time checks and debugging messages at the cost of
375 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
376 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
377 64. Default armv8a configuration of make build and meson build set it to 128
378 then brings performance degradation.
380 This option is available in meson:
382 - ``ibverbs_link`` can be ``static``, ``shared``, or ``dlopen``.
384 Environment variables
385 ~~~~~~~~~~~~~~~~~~~~~
389 A list of directories in which to search for the rdma-core "glue" plug-in,
390 separated by colons or semi-colons.
392 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
393 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
394 since ``LD_LIBRARY_PATH`` has no effect in this case.
396 - ``MLX5_SHUT_UP_BF``
398 Configures HW Tx doorbell register as IO-mapped.
400 By default, the HW Tx doorbell is configured as a write-combining register.
401 The register would be flushed to HW usually when the write-combining buffer
402 becomes full, but it depends on CPU design.
404 Except for vectorized Tx burst routines, a write memory barrier is enforced
405 after updating the register so that the update can be immediately visible to
408 When vectorized Tx burst is called, the barrier is set only if the burst size
409 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
410 variable will bring better latency even though the maximum throughput can
413 Run-time configuration
414 ~~~~~~~~~~~~~~~~~~~~~~
416 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
417 because it is affected by their state. Forcing them down prevents packets
420 - **ethtool** operations on related kernel interfaces also affect the PMD.
425 In order to run as a non-root user,
426 some capabilities must be granted to the application::
428 setcap cap_sys_admin,cap_net_admin,cap_net_raw,cap_ipc_lock+ep <dpdk-app>
430 Below are the reasons of the need for each capability:
433 When using physical addresses (PA mode), with Linux >= 4.0,
434 for access to ``/proc/self/pagemap``.
437 For device configuration.
440 For raw ethernet queue allocation through kernel driver.
443 For DMA memory pinning.
448 - ``rxq_cqe_comp_en`` parameter [int]
450 A nonzero value enables the compression of CQE on RX side. This feature
451 allows to save PCI bandwidth and improve performance. Enabled by default.
455 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
457 - POWER9 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
460 - ``rxq_cqe_pad_en`` parameter [int]
462 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
463 is aligned with the size of a cacheline of the core. If cacheline size is
464 128B, the CQE size is configured to be 128B even though the device writes
465 only 64B data on the cacheline. This is to avoid unnecessary cache
466 invalidation by device's two consecutive writes on to one cacheline.
467 However in some architecture, it is more beneficial to update entire
468 cacheline with padding the rest 64B rather than striding because
469 read-modify-write could drop performance a lot. On the other hand,
470 writing extra data will consume more PCIe bandwidth and could also drop
471 the maximum throughput. It is recommended to empirically set this
472 parameter. Disabled by default.
476 - CPU having 128B cacheline with ConnectX-5 and BlueField.
478 - ``rxq_pkt_pad_en`` parameter [int]
480 A nonzero value enables padding Rx packet to the size of cacheline on PCI
481 transaction. This feature would waste PCI bandwidth but could improve
482 performance by avoiding partial cacheline write which may cause costly
483 read-modify-copy in memory transaction on some architectures. Disabled by
488 - x86_64 with ConnectX-4, ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
490 - POWER8 and ARMv8 with ConnectX-4 Lx, ConnectX-5, ConnectX-6, ConnectX-6 Dx
493 - ``mprq_en`` parameter [int]
495 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
496 configured as Multi-Packet RQ if the total number of Rx queues is
497 ``rxqs_min_mprq`` or more. Disabled by default.
499 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
500 by posting a single large buffer for multiple packets. Instead of posting a
501 buffers per a packet, one large buffer is posted in order to receive multiple
502 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
503 and each stride receives one packet. MPRQ can improve throughput for
504 small-packet traffic.
506 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
507 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
508 configure large stride size enough to accommodate max_rx_pkt_len as long as
509 device allows. Note that this can waste system memory compared to enabling Rx
510 scatter and multi-segment packet.
512 - ``mprq_log_stride_num`` parameter [int]
514 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
515 strides can reduce PCIe traffic further. If configured value is not in the
516 range of device capability, the default value will be set with a warning
517 message. The default value is 4 which is 16 strides per a buffer, valid only
518 if ``mprq_en`` is set.
520 The size of Rx queue should be bigger than the number of strides.
522 - ``mprq_log_stride_size`` parameter [int]
524 Log 2 of the size of a stride for Multi-Packet Rx queue. Configuring a smaller
525 stride size can save some memory and reduce probability of a depletion of all
526 available strides due to unreleased packets by an application. If configured
527 value is not in the range of device capability, the default value will be set
528 with a warning message. The default value is 11 which is 2048 bytes per a
529 stride, valid only if ``mprq_en`` is set. With ``mprq_log_stride_size`` set
530 it is possible for a packet to span across multiple strides. This mode allows
531 support of jumbo frames (9K) with MPRQ. The memcopy of some packets (or part
532 of a packet if Rx scatter is configured) may be required in case there is no
533 space left for a head room at the end of a stride which incurs some
536 - ``mprq_max_memcpy_len`` parameter [int]
538 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
539 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
540 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
541 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
542 A mempool for external buffers will be allocated and managed by PMD. If Rx
543 packet is externally attached, ol_flags field of the mbuf will have
544 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
545 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
547 - ``rxqs_min_mprq`` parameter [int]
549 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
550 greater or equal to this value. The default value is 12, valid only if
553 - ``txq_inline`` parameter [int]
555 Amount of data to be inlined during TX operations. This parameter is
556 deprecated and converted to the new parameter ``txq_inline_max`` providing
557 partial compatibility.
559 - ``txqs_min_inline`` parameter [int]
561 Enable inline data send only when the number of TX queues is greater or equal
564 This option should be used in combination with ``txq_inline_max`` and
565 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
567 If this option is not specified the default value 16 is used for BlueField
568 and 8 for other platforms
570 The data inlining consumes the CPU cycles, so this option is intended to
571 auto enable inline data if we have enough Tx queues, which means we have
572 enough CPU cores and PCI bandwidth is getting more critical and CPU
573 is not supposed to be bottleneck anymore.
575 The copying data into WQE improves latency and can improve PPS performance
576 when PCI back pressure is detected and may be useful for scenarios involving
577 heavy traffic on many queues.
579 Because additional software logic is necessary to handle this mode, this
580 option should be used with care, as it may lower performance when back
581 pressure is not expected.
583 If inline data are enabled it may affect the maximal size of Tx queue in
584 descriptors because the inline data increase the descriptor size and
585 queue size limits supported by hardware may be exceeded.
587 - ``txq_inline_min`` parameter [int]
589 Minimal amount of data to be inlined into WQE during Tx operations. NICs
590 may require this minimal data amount to operate correctly. The exact value
591 may depend on NIC operation mode, requested offloads, etc. It is strongly
592 recommended to omit this parameter and use the default values. Anyway,
593 applications using this parameter should take into consideration that
594 specifying an inconsistent value may prevent the NIC from sending packets.
596 If ``txq_inline_min`` key is present the specified value (may be aligned
597 by the driver in order not to exceed the limits and provide better descriptor
598 space utilization) will be used by the driver and it is guaranteed that
599 requested amount of data bytes are inlined into the WQE beside other inline
600 settings. This key also may update ``txq_inline_max`` value (default
601 or specified explicitly in devargs) to reserve the space for inline data.
603 If ``txq_inline_min`` key is not present, the value may be queried by the
604 driver from the NIC via DevX if this feature is available. If there is no DevX
605 enabled/supported the value 18 (supposing L2 header including VLAN) is set
606 for ConnectX-4 and ConnectX-4 Lx, and 0 is set by default for ConnectX-5
607 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
610 For ConnectX-4 NIC, driver does not allow specifying value below 18
611 (minimal L2 header, including VLAN), error will be raised.
613 For ConnectX-4 Lx NIC, it is allowed to specify values below 18, but
614 it is not recommended and may prevent NIC from sending packets over
617 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
618 Multi-Packet Write), because last one does not support partial packet inlining.
619 This is not very critical due to minimal data inlining is mostly required
620 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
622 - ``txq_inline_max`` parameter [int]
624 Specifies the maximal packet length to be completely inlined into WQE
625 Ethernet Segment for ordinary SEND method. If packet is larger than specified
626 value, the packet data won't be copied by the driver at all, data buffer
627 is addressed with a pointer. If packet length is less or equal all packet
628 data will be copied into WQE. This may improve PCI bandwidth utilization for
629 short packets significantly but requires the extra CPU cycles.
631 The data inline feature is controlled by number of Tx queues, if number of Tx
632 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
633 is engaged, if there are not enough Tx queues (which means not enough CPU cores
634 and CPU resources are scarce), data inline is not performed by the driver.
635 Assigning ``txqs_min_inline`` with zero always enables the data inline.
637 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
638 by the driver in order not to exceed the limit (930 bytes) and to provide better
639 WQE space filling without gaps, the adjustment is reflected in the debug log.
640 Also, the default value (290) may be decreased in run-time if the large transmit
641 queue size is requested and hardware does not support enough descriptor
642 amount, in this case warning is emitted. If ``txq_inline_max`` key is
643 specified and requested inline settings can not be satisfied then error
646 - ``txq_inline_mpw`` parameter [int]
648 Specifies the maximal packet length to be completely inlined into WQE for
649 Enhanced MPW method. If packet is large the specified value, the packet data
650 won't be copied, and data buffer is addressed with pointer. If packet length
651 is less or equal, all packet data will be copied into WQE. This may improve PCI
652 bandwidth utilization for short packets significantly but requires the extra
655 The data inline feature is controlled by number of TX queues, if number of Tx
656 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
657 is engaged, if there are not enough Tx queues (which means not enough CPU cores
658 and CPU resources are scarce), data inline is not performed by the driver.
659 Assigning ``txqs_min_inline`` with zero always enables the data inline.
661 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
662 by the driver in order not to exceed the limit (930 bytes) and to provide better
663 WQE space filling without gaps, the adjustment is reflected in the debug log.
664 Due to multiple packets may be included to the same WQE with Enhanced Multi
665 Packet Write Method and overall WQE size is limited it is not recommended to
666 specify large values for the ``txq_inline_mpw``. Also, the default value (268)
667 may be decreased in run-time if the large transmit queue size is requested
668 and hardware does not support enough descriptor amount, in this case warning
669 is emitted. If ``txq_inline_mpw`` key is specified and requested inline
670 settings can not be satisfied then error will be raised.
672 - ``txqs_max_vec`` parameter [int]
674 Enable vectorized Tx only when the number of TX queues is less than or
675 equal to this value. This parameter is deprecated and ignored, kept
676 for compatibility issue to not prevent driver from probing.
678 - ``txq_mpw_hdr_dseg_en`` parameter [int]
680 A nonzero value enables including two pointers in the first block of TX
681 descriptor. The parameter is deprecated and ignored, kept for compatibility
684 - ``txq_max_inline_len`` parameter [int]
686 Maximum size of packet to be inlined. This limits the size of packet to
687 be inlined. If the size of a packet is larger than configured value, the
688 packet isn't inlined even though there's enough space remained in the
689 descriptor. Instead, the packet is included with pointer. This parameter
690 is deprecated and converted directly to ``txq_inline_mpw`` providing full
691 compatibility. Valid only if eMPW feature is engaged.
693 - ``txq_mpw_en`` parameter [int]
695 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
696 ConnectX-6, ConnectX-6 Dx and BlueField. eMPW allows the TX burst function to pack
697 up multiple packets in a single descriptor session in order to save PCI bandwidth
698 and improve performance at the cost of a slightly higher CPU usage. When
699 ``txq_inline_mpw`` is set along with ``txq_mpw_en``, TX burst function copies
700 entire packet data on to TX descriptor instead of including pointer of packet.
702 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
703 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
704 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
705 option or reported by the NIC, the eMPW feature is disengaged.
707 - ``tx_db_nc`` parameter [int]
709 The rdma core library can map doorbell register in two ways, depending on the
710 environment variable "MLX5_SHUT_UP_BF":
712 - As regular cached memory (usually with write combining attribute), if the
713 variable is either missing or set to zero.
714 - As non-cached memory, if the variable is present and set to not "0" value.
716 The type of mapping may slightly affect the Tx performance, the optimal choice
717 is strongly relied on the host architecture and should be deduced practically.
719 If ``tx_db_nc`` is set to zero, the doorbell is forced to be mapped to regular
720 memory (with write combining), the PMD will perform the extra write memory barrier
721 after writing to doorbell, it might increase the needed CPU clocks per packet
722 to send, but latency might be improved.
724 If ``tx_db_nc`` is set to one, the doorbell is forced to be mapped to non
725 cached memory, the PMD will not perform the extra write memory barrier
726 after writing to doorbell, on some architectures it might improve the
729 If ``tx_db_nc`` is set to two, the doorbell is forced to be mapped to regular
730 memory, the PMD will use heuristics to decide whether write memory barrier
731 should be performed. For bursts with size multiple of recommended one (64 pkts)
732 it is supposed the next burst is coming and no need to issue the extra memory
733 barrier (it is supposed to be issued in the next coming burst, at least after
734 descriptor writing). It might increase latency (on some hosts till next
735 packets transmit) and should be used with care.
737 If ``tx_db_nc`` is omitted or set to zero, the preset (if any) environment
738 variable "MLX5_SHUT_UP_BF" value is used. If there is no "MLX5_SHUT_UP_BF",
739 the default ``tx_db_nc`` value is zero for ARM64 hosts and one for others.
741 - ``tx_pp`` parameter [int]
743 If a nonzero value is specified the driver creates all necessary internal
744 objects to provide accurate packet send scheduling on mbuf timestamps.
745 The positive value specifies the scheduling granularity in nanoseconds,
746 the packet send will be accurate up to specified digits. The allowed range is
747 from 500 to 1 million of nanoseconds. The negative value specifies the module
748 of granularity and engages the special test mode the check the schedule rate.
749 By default (if the ``tx_pp`` is not specified) send scheduling on timestamps
752 - ``tx_skew`` parameter [int]
754 The parameter adjusts the send packet scheduling on timestamps and represents
755 the average delay between beginning of the transmitting descriptor processing
756 by the hardware and appearance of actual packet data on the wire. The value
757 should be provided in nanoseconds and is valid only if ``tx_pp`` parameter is
758 specified. The default value is zero.
760 - ``tx_vec_en`` parameter [int]
762 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6, ConnectX-6 Dx
763 and BlueField NICs if the number of global Tx queues on the port is less than
764 ``txqs_max_vec``. The parameter is deprecated and ignored.
766 - ``rx_vec_en`` parameter [int]
768 A nonzero value enables Rx vector if the port is not configured in
769 multi-segment otherwise this parameter is ignored.
773 - ``vf_nl_en`` parameter [int]
775 A nonzero value enables Netlink requests from the VF to add/remove MAC
776 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
777 Otherwise the relevant configuration must be run with Linux iproute2 tools.
778 This is a prerequisite to receive this kind of traffic.
780 Enabled by default, valid only on VF devices ignored otherwise.
782 - ``l3_vxlan_en`` parameter [int]
784 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
785 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
786 parameter. This is a prerequisite to receive this kind of traffic.
790 - ``dv_xmeta_en`` parameter [int]
792 A nonzero value enables extensive flow metadata support if device is
793 capable and driver supports it. This can enable extensive support of
794 ``MARK`` and ``META`` item of ``rte_flow``. The newly introduced
795 ``SET_TAG`` and ``SET_META`` actions do not depend on ``dv_xmeta_en``.
797 There are some possible configurations, depending on parameter value:
799 - 0, this is default value, defines the legacy mode, the ``MARK`` and
800 ``META`` related actions and items operate only within NIC Tx and
801 NIC Rx steering domains, no ``MARK`` and ``META`` information crosses
802 the domain boundaries. The ``MARK`` item is 24 bits wide, the ``META``
803 item is 32 bits wide and match supported on egress only.
805 - 1, this engages extensive metadata mode, the ``MARK`` and ``META``
806 related actions and items operate within all supported steering domains,
807 including FDB, ``MARK`` and ``META`` information may cross the domain
808 boundaries. The ``MARK`` item is 24 bits wide, the ``META`` item width
809 depends on kernel and firmware configurations and might be 0, 16 or
810 32 bits. Within NIC Tx domain ``META`` data width is 32 bits for
811 compatibility, the actual width of data transferred to the FDB domain
812 depends on kernel configuration and may be vary. The actual supported
813 width can be retrieved in runtime by series of rte_flow_validate()
816 - 2, this engages extensive metadata mode, the ``MARK`` and ``META``
817 related actions and items operate within all supported steering domains,
818 including FDB, ``MARK`` and ``META`` information may cross the domain
819 boundaries. The ``META`` item is 32 bits wide, the ``MARK`` item width
820 depends on kernel and firmware configurations and might be 0, 16 or
821 24 bits. The actual supported width can be retrieved in runtime by
822 series of rte_flow_validate() trials.
824 +------+-----------+-----------+-------------+-------------+
825 | Mode | ``MARK`` | ``META`` | ``META`` Tx | FDB/Through |
826 +======+===========+===========+=============+=============+
827 | 0 | 24 bits | 32 bits | 32 bits | no |
828 +------+-----------+-----------+-------------+-------------+
829 | 1 | 24 bits | vary 0-32 | 32 bits | yes |
830 +------+-----------+-----------+-------------+-------------+
831 | 2 | vary 0-32 | 32 bits | 32 bits | yes |
832 +------+-----------+-----------+-------------+-------------+
834 If there is no E-Switch configuration the ``dv_xmeta_en`` parameter is
835 ignored and the device is configured to operate in legacy mode (0).
837 Disabled by default (set to 0).
839 The Direct Verbs/Rules (engaged with ``dv_flow_en`` = 1) supports all
840 of the extensive metadata features. The legacy Verbs supports FLAG and
841 MARK metadata actions over NIC Rx steering domain only.
843 - ``dv_flow_en`` parameter [int]
845 A nonzero value enables the DV flow steering assuming it is supported
846 by the driver (RDMA Core library version is rdma-core-24.0 or higher).
848 Enabled by default if supported.
850 - ``dv_esw_en`` parameter [int]
852 A nonzero value enables E-Switch using Direct Rules.
854 Enabled by default if supported.
856 - ``lacp_by_user`` parameter [int]
858 A nonzero value enables the control of LACP traffic by the user application.
859 When a bond exists in the driver, by default it should be managed by the
860 kernel and therefore LACP traffic should be steered to the kernel.
861 If this devarg is set to 1 it will allow the user to manage the bond by
862 itself and not steer LACP traffic to the kernel.
864 Disabled by default (set to 0).
866 - ``mr_ext_memseg_en`` parameter [int]
868 A nonzero value enables extending memseg when registering DMA memory. If
869 enabled, the number of entries in MR (Memory Region) lookup table on datapath
870 is minimized and it benefits performance. On the other hand, it worsens memory
871 utilization because registered memory is pinned by kernel driver. Even if a
872 page in the extended chunk is freed, that doesn't become reusable until the
873 entire memory is freed.
877 - ``representor`` parameter [list]
879 This parameter can be used to instantiate DPDK Ethernet devices from
880 existing port (or VF) representors configured on the device.
882 It is a standard parameter whose format is described in
883 :ref:`ethernet_device_standard_device_arguments`.
885 For instance, to probe port representors 0 through 2::
889 - ``max_dump_files_num`` parameter [int]
891 The maximum number of files per PMD entity that may be created for debug information.
892 The files will be created in /var/log directory or in current directory.
894 set to 128 by default.
896 - ``lro_timeout_usec`` parameter [int]
898 The maximum allowed duration of an LRO session, in micro-seconds.
899 PMD will set the nearest value supported by HW, which is not bigger than
900 the input ``lro_timeout_usec`` value.
901 If this parameter is not specified, by default PMD will set
902 the smallest value supported by HW.
904 - ``hp_buf_log_sz`` parameter [int]
906 The total data buffer size of a hairpin queue (logarithmic form), in bytes.
907 PMD will set the data buffer size to 2 ** ``hp_buf_log_sz``, both for RX & TX.
908 The capacity of the value is specified by the firmware and the initialization
909 will get a failure if it is out of scope.
910 The range of the value is from 11 to 19 right now, and the supported frame
911 size of a single packet for hairpin is from 512B to 128KB. It might change if
912 different firmware release is being used. By using a small value, it could
913 reduce memory consumption but not work with a large frame. If the value is
914 too large, the memory consumption will be high and some potential performance
915 degradation will be introduced.
916 By default, the PMD will set this value to 16, which means that 9KB jumbo
917 frames will be supported.
919 - ``reclaim_mem_mode`` parameter [int]
921 Cache some resources in flow destroy will help flow recreation more efficient.
922 While some systems may require the all the resources can be reclaimed after
924 The parameter ``reclaim_mem_mode`` provides the option for user to configure
925 if the resource cache is needed or not.
927 There are three options to choose:
929 - 0. It means the flow resources will be cached as usual. The resources will
930 be cached, helpful with flow insertion rate.
932 - 1. It will only enable the DPDK PMD level resources reclaim.
934 - 2. Both DPDK PMD level and rdma-core low level will be configured as
937 By default, the PMD will set this value to 0.
939 - ``sys_mem_en`` parameter [int]
941 A non-zero value enables the PMD memory management allocating memory
942 from system by default, without explicit rte memory flag.
944 By default, the PMD will set this value to 0.
946 - ``decap_en`` parameter [int]
948 Some devices do not support FCS (frame checksum) scattering for
949 tunnel-decapsulated packets.
950 If set to 0, this option forces the FCS feature and rejects tunnel
951 decapsulation in the flow engine for such devices.
953 By default, the PMD will set this value to 1.
955 .. _mlx5_firmware_config:
957 Firmware configuration
958 ~~~~~~~~~~~~~~~~~~~~~~
960 Firmware features can be configured as key/value pairs.
962 The command to set a value is::
964 mlxconfig -d <device> set <key>=<value>
966 The command to query a value is::
968 mlxconfig -d <device> query | grep <key>
970 The device name for the command ``mlxconfig`` can be either the PCI address,
971 or the mst device name found with::
975 Below are some firmware configurations listed.
981 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
987 - maximum number of SR-IOV virtual functions::
991 - enable DevX (required by Direct Rules and other features)::
995 - aggressive CQE zipping::
999 - L3 VXLAN and VXLAN-GPE destination UDP port::
1002 IP_OVER_VXLAN_PORT=<udp dport>
1004 - enable VXLAN-GPE tunnel flow matching::
1006 FLEX_PARSER_PROFILE_ENABLE=0
1008 FLEX_PARSER_PROFILE_ENABLE=2
1010 - enable IP-in-IP tunnel flow matching::
1012 FLEX_PARSER_PROFILE_ENABLE=0
1014 - enable MPLS flow matching::
1016 FLEX_PARSER_PROFILE_ENABLE=1
1018 - enable ICMP(code/type/identifier/sequence number) / ICMP6(code/type) fields matching::
1020 FLEX_PARSER_PROFILE_ENABLE=2
1022 - enable Geneve flow matching::
1024 FLEX_PARSER_PROFILE_ENABLE=0
1026 FLEX_PARSER_PROFILE_ENABLE=1
1028 - enable GTP flow matching::
1030 FLEX_PARSER_PROFILE_ENABLE=3
1032 - enable eCPRI flow matching::
1034 FLEX_PARSER_PROFILE_ENABLE=4
1040 This driver relies on external libraries and kernel drivers for resources
1041 allocations and initialization. The following dependencies are not part of
1042 DPDK and must be installed separately:
1046 User space Verbs framework used by librte_pmd_mlx5. This library provides
1047 a generic interface between the kernel and low-level user space drivers
1050 It allows slow and privileged operations (context initialization, hardware
1051 resources allocations) to be managed by the kernel and fast operations to
1052 never leave user space.
1056 Low-level user space driver library for Mellanox
1057 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
1060 This library basically implements send/receive calls to the hardware
1063 - **Kernel modules**
1065 They provide the kernel-side Verbs API and low level device drivers that
1066 manage actual hardware initialization and resources sharing with user
1069 Unlike most other PMDs, these modules must remain loaded and bound to
1072 - mlx5_core: hardware driver managing Mellanox
1073 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
1075 - mlx5_ib: InifiniBand device driver.
1076 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
1078 - **Firmware update**
1080 Mellanox OFED/EN releases include firmware updates for
1081 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
1083 Because each release provides new features, these updates must be applied to
1084 match the kernel modules and libraries they come with.
1088 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
1094 Either RDMA Core library with a recent enough Linux kernel release
1095 (recommended) or Mellanox OFED/EN, which provides compatibility with older
1098 RDMA Core with Linux Kernel
1099 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
1101 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
1102 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
1103 (see `RDMA Core installation documentation`_)
1104 - When building for i686 use:
1106 - rdma-core version 18.0 or above built with 32bit support.
1107 - Kernel version 4.14.41 or above.
1109 - Starting with rdma-core v21, static libraries can be built::
1112 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
1115 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
1116 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
1118 If rdma-core libraries are built but not installed, DPDK makefile can link them,
1119 thanks to these environment variables:
1121 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
1122 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
1123 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
1128 - Mellanox OFED version: **4.5** and above /
1129 Mellanox EN version: **4.5** and above
1132 - ConnectX-4: **12.21.1000** and above.
1133 - ConnectX-4 Lx: **14.21.1000** and above.
1134 - ConnectX-5: **16.21.1000** and above.
1135 - ConnectX-5 Ex: **16.21.1000** and above.
1136 - ConnectX-6: **20.27.0090** and above.
1137 - ConnectX-6 Dx: **22.27.0090** and above.
1138 - BlueField: **18.25.1010** and above.
1140 While these libraries and kernel modules are available on OpenFabrics
1141 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
1142 managers on most distributions, this PMD requires Ethernet extensions that
1143 may not be supported at the moment (this is a work in progress).
1146 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
1148 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
1149 include the necessary support and should be used in the meantime. For DPDK,
1150 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
1151 required from that distribution.
1155 Several versions of Mellanox OFED/EN are available. Installing the version
1156 this DPDK release was developed and tested against is strongly
1157 recommended. Please check the `prerequisites`_.
1162 The following Mellanox device families are supported by the same mlx5 driver:
1172 Below are detailed device names:
1174 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX4111A-XCAT (1x10G)
1175 * Mellanox\ |reg| ConnectX\ |reg|-4 10G MCX412A-XCAT (2x10G)
1176 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX4111A-ACAT (1x25G)
1177 * Mellanox\ |reg| ConnectX\ |reg|-4 25G MCX412A-ACAT (2x25G)
1178 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX413A-BCAT (1x40G)
1179 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX4131A-BCAT (1x40G)
1180 * Mellanox\ |reg| ConnectX\ |reg|-4 40G MCX415A-BCAT (1x40G)
1181 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX413A-GCAT (1x50G)
1182 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX4131A-GCAT (1x50G)
1183 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX414A-BCAT (2x50G)
1184 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-GCAT (1x50G)
1185 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-BCAT (2x50G)
1186 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX416A-GCAT (2x50G)
1187 * Mellanox\ |reg| ConnectX\ |reg|-4 50G MCX415A-CCAT (1x100G)
1188 * Mellanox\ |reg| ConnectX\ |reg|-4 100G MCX416A-CCAT (2x100G)
1189 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4111A-XCAT (1x10G)
1190 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 10G MCX4121A-XCAT (2x10G)
1191 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4111A-ACAT (1x25G)
1192 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 25G MCX4121A-ACAT (2x25G)
1193 * Mellanox\ |reg| ConnectX\ |reg|-4 Lx 40G MCX4131A-BCAT (1x40G)
1194 * Mellanox\ |reg| ConnectX\ |reg|-5 100G MCX556A-ECAT (2x100G)
1195 * Mellanox\ |reg| ConnectX\ |reg|-5 Ex EN 100G MCX516A-CDAT (2x100G)
1196 * Mellanox\ |reg| ConnectX\ |reg|-6 200G MCX654106A-HCAT (2x200G)
1197 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 100G MCX623106AN-CDAT (2x100G)
1198 * Mellanox\ |reg| ConnectX\ |reg|-6 Dx EN 200G MCX623105AN-VDAT (1x200G)
1200 Quick Start Guide on OFED/EN
1201 ----------------------------
1203 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
1206 2. Install the required libraries and kernel modules either by installing
1207 only the required set, or by installing the entire Mellanox OFED/EN::
1209 ./mlnxofedinstall --upstream-libs --dpdk
1211 3. Verify the firmware is the correct one::
1215 4. Verify all ports links are set to Ethernet::
1217 mlxconfig -d <mst device> query | grep LINK_TYPE
1221 Link types may have to be configured to Ethernet::
1223 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
1225 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
1227 For hypervisors, verify SR-IOV is enabled on the NIC::
1229 mlxconfig -d <mst device> query | grep SRIOV_EN
1232 If needed, configure SR-IOV::
1234 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
1235 mlxfwreset -d <mst device> reset
1237 5. Restart the driver::
1239 /etc/init.d/openibd restart
1243 service openibd restart
1245 If link type was changed, firmware must be reset as well::
1247 mlxfwreset -d <mst device> reset
1249 For hypervisors, after reset write the sysfs number of virtual functions
1252 To dynamically instantiate a given number of virtual functions (VFs)::
1254 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
1256 6. Install DPDK and you are ready to go.
1257 See :doc:`compilation instructions <../linux_gsg/build_dpdk>`.
1259 Enable switchdev mode
1260 ---------------------
1262 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
1263 Representor is a port in DPDK that is connected to a VF in such a way
1264 that assuming there are no offload flows, each packet that is sent from the VF
1265 will be received by the corresponding representor. While each packet that is
1266 sent to a representor will be received by the VF.
1267 This is very useful in case of SRIOV mode, where the first packet that is sent
1268 by the VF will be received by the DPDK application which will decide if this
1269 flow should be offloaded to the E-Switch. After offloading the flow packet
1270 that the VF that are matching the flow will not be received any more by
1271 the DPDK application.
1273 1. Enable SRIOV mode::
1275 mlxconfig -d <mst device> set SRIOV_EN=true
1277 2. Configure the max number of VFs::
1279 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
1283 mlxfwreset -d <mst device> reset
1285 3. Configure the actual number of VFs::
1287 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
1289 4. Unbind the device (can be rebind after the switchdev mode)::
1291 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
1293 5. Enbale switchdev mode::
1295 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
1300 1. Configure aggressive CQE Zipping for maximum performance::
1302 mlxconfig -d <mst device> s CQE_COMPRESSION=1
1304 To set it back to the default CQE Zipping mode use::
1306 mlxconfig -d <mst device> s CQE_COMPRESSION=0
1308 2. In case of virtualization:
1310 - Make sure that hypervisor kernel is 3.16 or newer.
1311 - Configure boot with ``iommu=pt``.
1312 - Use 1G huge pages.
1313 - Make sure to allocate a VM on huge pages.
1314 - Make sure to set CPU pinning.
1316 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
1317 for better performance. For VMs, verify that the right CPU
1318 and NUMA node are pinned according to the above. Run::
1322 to identify the NUMA node to which the PCIe adapter is connected.
1324 4. If more than one adapter is used, and root complex capabilities allow
1325 to put both adapters on the same NUMA node without PCI bandwidth degradation,
1326 it is recommended to locate both adapters on the same NUMA node.
1327 This in order to forward packets from one to the other without
1328 NUMA performance penalty.
1330 5. Disable pause frames::
1332 ethtool -A <netdev> rx off tx off
1334 6. Verify IO non-posted prefetch is disabled by default. This can be checked
1335 via the BIOS configuration. Please contact you server provider for more
1336 information about the settings.
1340 On some machines, depends on the machine integrator, it is beneficial
1341 to set the PCI max read request parameter to 1K. This can be
1342 done in the following way:
1344 To query the read request size use::
1346 setpci -s <NIC PCI address> 68.w
1348 If the output is different than 3XXX, set it by::
1350 setpci -s <NIC PCI address> 68.w=3XXX
1352 The XXX can be different on different systems. Make sure to configure
1353 according to the setpci output.
1355 7. To minimize overhead of searching Memory Regions:
1357 - '--socket-mem' is recommended to pin memory by predictable amount.
1358 - Configure per-lcore cache when creating Mempools for packet buffer.
1359 - Refrain from dynamically allocating/freeing memory in run-time.
1361 .. _mlx5_offloads_support:
1363 Supported hardware offloads
1364 ---------------------------
1366 .. table:: Minimal SW/HW versions for queue offloads
1368 ============== ===== ===== ========= ===== ========== ==========
1369 Offload DPDK Linux rdma-core OFED firmware hardware
1370 ============== ===== ===== ========= ===== ========== ==========
1371 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1372 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1373 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1374 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
1375 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
1376 ============== ===== ===== ========= ===== ========== ==========
1378 .. table:: Minimal SW/HW versions for rte_flow offloads
1380 +-----------------------+-----------------+-----------------+
1381 | Offload | with E-Switch | with NIC |
1382 +=======================+=================+=================+
1383 | Count | | DPDK 19.05 | | DPDK 19.02 |
1384 | | | OFED 4.6 | | OFED 4.6 |
1385 | | | rdma-core 24 | | rdma-core 23 |
1386 | | | ConnectX-5 | | ConnectX-5 |
1387 +-----------------------+-----------------+-----------------+
1388 | Drop | | DPDK 19.05 | | DPDK 18.11 |
1389 | | | OFED 4.6 | | OFED 4.5 |
1390 | | | rdma-core 24 | | rdma-core 23 |
1391 | | | ConnectX-5 | | ConnectX-4 |
1392 +-----------------------+-----------------+-----------------+
1393 | Queue / RSS | | | | DPDK 18.11 |
1394 | | | N/A | | OFED 4.5 |
1395 | | | | | rdma-core 23 |
1396 | | | | | ConnectX-4 |
1397 +-----------------------+-----------------+-----------------+
1398 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1399 | (VXLAN / NVGRE / RAW) | | OFED 4.7-1 | | OFED 4.6 |
1400 | | | rdma-core 24 | | rdma-core 23 |
1401 | | | ConnectX-5 | | ConnectX-5 |
1402 +-----------------------+-----------------+-----------------+
1403 | Encapsulation | | DPDK 19.11 | | DPDK 19.11 |
1404 | GENEVE | | OFED 4.7-3 | | OFED 4.7-3 |
1405 | | | rdma-core 27 | | rdma-core 27 |
1406 | | | ConnectX-5 | | ConnectX-5 |
1407 +-----------------------+-----------------+-----------------+
1408 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1409 | | (set_ipv4_src / | | OFED 4.7-1 | | OFED 4.7-1 |
1410 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 24 |
1411 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1412 | | set_ipv6_dst / | | | | |
1413 | | set_tp_src / | | | | |
1414 | | set_tp_dst / | | | | |
1415 | | dec_ttl / | | | | |
1416 | | set_ttl / | | | | |
1417 | | set_mac_src / | | | | |
1418 | | set_mac_dst) | | | | |
1419 +-----------------------+-----------------+-----------------+
1420 | | Header rewrite | | DPDK 20.02 | | DPDK 20.02 |
1421 | | (set_dscp) | | OFED 5.0 | | OFED 5.0 |
1422 | | | | rdma-core 24 | | rdma-core 24 |
1423 | | | | ConnectX-5 | | ConnectX-5 |
1424 +-----------------------+-----------------+-----------------+
1425 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1426 | | | OFED 4.7-1 | | OFED 4.7-1 |
1427 | | | rdma-core 24 | | N/A |
1428 | | | ConnectX-5 | | ConnectX-5 |
1429 +-----------------------+-----------------+-----------------+
1430 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1431 | | | OFED 4.6 | | OFED 4.5 |
1432 | | | rdma-core 24 | | rdma-core 23 |
1433 | | | ConnectX-5 | | ConnectX-4 |
1434 +-----------------------+-----------------+-----------------+
1435 | Port ID | | DPDK 19.05 | | N/A |
1436 | | | OFED 4.7-1 | | N/A |
1437 | | | rdma-core 24 | | N/A |
1438 | | | ConnectX-5 | | N/A |
1439 +-----------------------+-----------------+-----------------+
1440 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1441 | | (of_pop_vlan / | | OFED 4.7-1 | | OFED 4.7-1 |
1442 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1443 | | of_set_vlan_pcp / | | | | |
1444 | | of_set_vlan_vid) | | | | |
1445 +-----------------------+-----------------+-----------------+
1446 | Hairpin | | | | DPDK 19.11 |
1447 | | | N/A | | OFED 4.7-3 |
1448 | | | | | rdma-core 26 |
1449 | | | | | ConnectX-5 |
1450 +-----------------------+-----------------+-----------------+
1451 | Meta data | | DPDK 19.11 | | DPDK 19.11 |
1452 | | | OFED 4.7-3 | | OFED 4.7-3 |
1453 | | | rdma-core 26 | | rdma-core 26 |
1454 | | | ConnectX-5 | | ConnectX-5 |
1455 +-----------------------+-----------------+-----------------+
1456 | Metering | | DPDK 19.11 | | DPDK 19.11 |
1457 | | | OFED 4.7-3 | | OFED 4.7-3 |
1458 | | | rdma-core 26 | | rdma-core 26 |
1459 | | | ConnectX-5 | | ConnectX-5 |
1460 +-----------------------+-----------------+-----------------+
1461 | Sampling | | DPDK 20.11 | | DPDK 20.11 |
1462 | | | OFED 5.2 | | OFED 5.2 |
1463 | | | rdma-core 32 | | rdma-core 32 |
1464 | | | ConnectX-5 | | ConnectX-5 |
1465 +-----------------------+-----------------+-----------------+
1470 MARK and META items are interrelated with datapath - they might move from/to
1471 the applications in mbuf fields. Hence, zero value for these items has the
1472 special meaning - it means "no metadata are provided", not zero values are
1473 treated by applications and PMD as valid ones.
1475 Moreover in the flow engine domain the value zero is acceptable to match and
1476 set, and we should allow to specify zero values as rte_flow parameters for the
1477 META and MARK items and actions. In the same time zero mask has no meaning and
1478 should be rejected on validation stage.
1483 Flows are not cached in the driver.
1484 When stopping a device port, all the flows created on this port from the
1485 application will be flushed automatically in the background.
1486 After stopping the device port, all flows on this port become invalid and
1487 not represented in the system.
1488 All references to these flows held by the application should be discarded
1489 directly but neither destroyed nor flushed.
1491 The application should re-create the flows as required after the port restart.
1496 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1497 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1499 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1500 command-line parameter to enable additional protocols (UDP and TCP as well
1501 as IP), the following commands must be entered from its CLI to get the same
1502 behavior as librte_pmd_mlx4::
1505 > port config all rss all
1511 This section demonstrates how to launch **testpmd** with Mellanox
1512 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1514 #. Load the kernel modules::
1516 modprobe -a ib_uverbs mlx5_core mlx5_ib
1518 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1521 /etc/init.d/openibd restart
1525 User space I/O kernel modules (uio and igb_uio) are not used and do
1526 not have to be loaded.
1528 #. Make sure Ethernet interfaces are in working order and linked to kernel
1529 verbs. Related sysfs entries should be present::
1531 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1540 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1543 for intf in eth2 eth3 eth4 eth5;
1545 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1548 sed -n 's,.*/\(.*\),-w \1,p'
1557 #. Request huge pages::
1559 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1561 #. Start testpmd with basic parameters::
1563 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
1568 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1569 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1570 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1571 PMD: librte_pmd_mlx5: 1 port(s) detected
1572 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1573 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1574 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1575 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1576 PMD: librte_pmd_mlx5: 1 port(s) detected
1577 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1578 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1579 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1580 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1581 PMD: librte_pmd_mlx5: 1 port(s) detected
1582 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1583 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1584 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1585 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1586 PMD: librte_pmd_mlx5: 1 port(s) detected
1587 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1588 Interactive-mode selected
1589 Configuring Port 0 (socket 0)
1590 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1591 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1592 Port 0: E4:1D:2D:E7:0C:FE
1593 Configuring Port 1 (socket 0)
1594 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1595 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1596 Port 1: E4:1D:2D:E7:0C:FF
1597 Configuring Port 2 (socket 0)
1598 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1599 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1600 Port 2: E4:1D:2D:E7:0C:FA
1601 Configuring Port 3 (socket 0)
1602 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1603 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1604 Port 3: E4:1D:2D:E7:0C:FB
1605 Checking link statuses...
1606 Port 0 Link Up - speed 40000 Mbps - full-duplex
1607 Port 1 Link Up - speed 40000 Mbps - full-duplex
1608 Port 2 Link Up - speed 10000 Mbps - full-duplex
1609 Port 3 Link Up - speed 10000 Mbps - full-duplex
1616 This section demonstrates how to dump flows. Currently, it's possible to dump
1617 all flows with assistance of external tools.
1619 #. 2 ways to get flow raw file:
1621 - Using testpmd CLI:
1623 .. code-block:: console
1625 testpmd> flow dump <port> <output_file>
1627 - call rte_flow_dev_dump api:
1629 .. code-block:: console
1631 rte_flow_dev_dump(port, file, NULL);
1633 #. Dump human-readable flows from raw file:
1635 Get flow parsing tool from: https://github.com/Mellanox/mlx_steering_dump
1637 .. code-block:: console
1639 mlx_steering_dump.py -f <output_file>