1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright 2015 6WIND S.A.
3 Copyright 2015 Mellanox Technologies, Ltd
8 The MLX5 poll mode driver library (**librte_pmd_mlx5**) provides support
9 for **Mellanox ConnectX-4**, **Mellanox ConnectX-4 Lx** , **Mellanox
10 ConnectX-5**, **Mellanox ConnectX-6** and **Mellanox BlueField** families
11 of 10/25/40/50/100/200 Gb/s adapters as well as their virtual functions (VF)
14 Information and documentation about these adapters can be found on the
15 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
16 `Mellanox community <http://community.mellanox.com/welcome>`__.
18 There is also a `section dedicated to this poll mode driver
19 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
23 Due to external dependencies, this driver is disabled in default configuration
24 of the "make" build. It can be enabled with ``CONFIG_RTE_LIBRTE_MLX5_PMD=y``
25 or by using "meson" build system which will detect dependencies.
30 Besides its dependency on libibverbs (that implies libmlx5 and associated
31 kernel support), librte_pmd_mlx5 relies heavily on system calls for control
32 operations such as querying/updating the MTU and flow control parameters.
34 For security reasons and robustness, this driver only deals with virtual
35 memory addresses. The way resources allocations are handled by the kernel,
36 combined with hardware specifications that allow to handle virtual memory
37 addresses directly, ensure that DPDK applications cannot access random
38 physical memory (or memory that does not belong to the current process).
40 This capability allows the PMD to coexist with kernel network interfaces
41 which remain functional, although they stop receiving unicast packets as
42 long as they share the same MAC address.
43 This means legacy linux control tools (for example: ethtool, ifconfig and
44 more) can operate on the same network interfaces that owned by the DPDK
47 The PMD can use libibverbs and libmlx5 to access the device firmware
48 or directly the hardware components.
49 There are different levels of objects and bypassing abilities
50 to get the best performances:
52 - Verbs is a complete high-level generic API
53 - Direct Verbs is a device-specific API
54 - DevX allows to access firmware objects
55 - Direct Rules manages flow steering at low-level hardware layer
57 Enabling librte_pmd_mlx5 causes DPDK applications to be linked against
63 - Multi arch support: x86_64, POWER8, ARMv8, i686.
64 - Multiple TX and RX queues.
65 - Support for scattered TX and RX frames.
66 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
67 - Several RSS hash keys, one for each flow type.
68 - Default RSS operation with no hash key specification.
69 - Configurable RETA table.
70 - Link flow control (pause frame).
71 - Support for multiple MAC addresses.
75 - RX CRC stripping configuration.
76 - Promiscuous mode on PF and VF.
77 - Multicast promiscuous mode on PF and VF.
78 - Hardware checksum offloads.
79 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
81 - Flow API, including :ref:`flow_isolated_mode`.
83 - KVM and VMware ESX SR-IOV modes are supported.
84 - RSS hash result is supported.
85 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
86 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
88 - Statistics query including Basic, Extended and per queue.
90 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP.
91 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
92 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
93 increment/decrement, count, drop, mark. For details please see :ref:`mlx5_offloads_support`.
94 - Flow insertion rate of more then million flows per second, when using Direct Rules.
95 - Support for multiple rte_flow groups.
101 - For secondary process:
103 - Forked secondary process not supported.
104 - External memory unregistered in EAL memseg list cannot be used for DMA
105 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
106 primary process and remapped to the same virtual address in secondary
107 process. If the external memory is registered by primary process but has
108 different virtual address in secondary process, unexpected error may happen.
110 - Flow pattern without any specific vlan will match for vlan packets as well:
112 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
113 Meaning, the flow rule::
115 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
117 Will only match vlan packets with vid=3. and the flow rules::
119 flow create 0 ingress pattern eth / ipv4 / end ...
123 flow create 0 ingress pattern eth / vlan / ipv4 / end ...
125 Will match any ipv4 packet (VLAN included).
127 - A multi segment packet must have not more segments than reported by dev_infos_get()
128 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
129 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
130 inline settings) to 58.
132 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
133 to 0 are not supported.
135 - VXLAN TSO and checksum offloads are not supported on VM.
137 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
139 - VF: flow rules created on VF devices can only match traffic targeted at the
140 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
144 MAC addresses not already present in the bridge table of the associated
145 kernel network device will be added and cleaned up by the PMD when closing
146 the device. In case of ungraceful program termination, some entries may
147 remain present and should be removed manually by other means.
149 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
150 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
151 ol_flags. As the mempool for the external buffer is managed by PMD, all the
152 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
153 the external buffers will be freed by PMD and the application which still
154 holds the external buffers may be corrupted.
156 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
157 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
158 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
160 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
161 and allmulticast mode are both set to off.
162 To receive IPv6 Multicast messages on VM, explicitly set the relevant
163 MAC address using rte_eth_dev_mac_addr_add() API.
165 - E-Switch decapsulation Flow:
167 - can be applied to PF port only.
168 - must specify VF port action (packet redirection from PF to VF).
169 - optionally may specify tunnel inner source and destination MAC addresses.
171 - E-Switch encapsulation Flow:
173 - can be applied to VF ports only.
174 - must specify PF port action (packet redirection from VF to PF).
176 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
177 mutually exclusive features which cannot be supported together
178 (see :ref:`mlx5_firmware_config`).
182 - Requires DevX to be enabled.
183 - KEEP_CRC offload cannot be supported with LRO.
184 - The first mbuf length, without head-room, must be big enough to include the
190 MLX5 supports various methods to report statistics:
192 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.
194 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.
196 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.
204 These options can be modified in the ``.config`` file.
206 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
208 Toggle compilation of librte_pmd_mlx5 itself.
210 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
212 Build PMD with additional code to make it loadable without hard
213 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
214 on the target system.
216 In this mode, their presence is still required for it to run properly,
217 however their absence won't prevent a DPDK application from starting (with
218 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
219 missing with ``ldd(1)``.
221 It works by moving these dependencies to a purpose-built rdma-core "glue"
222 plug-in which must either be installed in a directory whose name is based
223 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
224 standard location for the dynamic linker (e.g. ``/lib``) if left to the
225 default empty string (``""``).
227 This option has no performance impact.
229 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
231 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
232 in the PMD shared library or the executable static binary.
234 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
236 Toggle debugging code and stricter compilation flags. Enabling this option
237 adds additional run-time checks and debugging messages at the cost of
242 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
243 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
244 64. Default armv8a configuration of make build and meson build set it to 128
245 then brings performance degradation.
247 Environment variables
248 ~~~~~~~~~~~~~~~~~~~~~
252 A list of directories in which to search for the rdma-core "glue" plug-in,
253 separated by colons or semi-colons.
255 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
256 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
257 since ``LD_LIBRARY_PATH`` has no effect in this case.
259 - ``MLX5_SHUT_UP_BF``
261 Configures HW Tx doorbell register as IO-mapped.
263 By default, the HW Tx doorbell is configured as a write-combining register.
264 The register would be flushed to HW usually when the write-combining buffer
265 becomes full, but it depends on CPU design.
267 Except for vectorized Tx burst routines, a write memory barrier is enforced
268 after updating the register so that the update can be immediately visible to
271 When vectorized Tx burst is called, the barrier is set only if the burst size
272 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
273 variable will bring better latency even though the maximum throughput can
276 Run-time configuration
277 ~~~~~~~~~~~~~~~~~~~~~~
279 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
280 because it is affected by their state. Forcing them down prevents packets
283 - **ethtool** operations on related kernel interfaces also affect the PMD.
285 - ``rxq_cqe_comp_en`` parameter [int]
287 A nonzero value enables the compression of CQE on RX side. This feature
288 allows to save PCI bandwidth and improve performance. Enabled by default.
292 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
293 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
295 - ``rxq_cqe_pad_en`` parameter [int]
297 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
298 is aligned with the size of a cacheline of the core. If cacheline size is
299 128B, the CQE size is configured to be 128B even though the device writes
300 only 64B data on the cacheline. This is to avoid unnecessary cache
301 invalidation by device's two consecutive writes on to one cacheline.
302 However in some architecture, it is more beneficial to update entire
303 cacheline with padding the rest 64B rather than striding because
304 read-modify-write could drop performance a lot. On the other hand,
305 writing extra data will consume more PCIe bandwidth and could also drop
306 the maximum throughput. It is recommended to empirically set this
307 parameter. Disabled by default.
311 - CPU having 128B cacheline with ConnectX-5 and BlueField.
313 - ``rxq_pkt_pad_en`` parameter [int]
315 A nonzero value enables padding Rx packet to the size of cacheline on PCI
316 transaction. This feature would waste PCI bandwidth but could improve
317 performance by avoiding partial cacheline write which may cause costly
318 read-modify-copy in memory transaction on some architectures. Disabled by
323 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
324 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
326 - ``mprq_en`` parameter [int]
328 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
329 configured as Multi-Packet RQ if the total number of Rx queues is
330 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
333 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
334 by posting a single large buffer for multiple packets. Instead of posting a
335 buffers per a packet, one large buffer is posted in order to receive multiple
336 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
337 and each stride receives one packet. MPRQ can improve throughput for
338 small-packet traffic.
340 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
341 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
342 configure large stride size enough to accommodate max_rx_pkt_len as long as
343 device allows. Note that this can waste system memory compared to enabling Rx
344 scatter and multi-segment packet.
346 - ``mprq_log_stride_num`` parameter [int]
348 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
349 strides can reduce PCIe traffic further. If configured value is not in the
350 range of device capability, the default value will be set with a warning
351 message. The default value is 4 which is 16 strides per a buffer, valid only
352 if ``mprq_en`` is set.
354 The size of Rx queue should be bigger than the number of strides.
356 - ``mprq_max_memcpy_len`` parameter [int]
358 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
359 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
360 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
361 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
362 A mempool for external buffers will be allocated and managed by PMD. If Rx
363 packet is externally attached, ol_flags field of the mbuf will have
364 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
365 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
367 - ``rxqs_min_mprq`` parameter [int]
369 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
370 greater or equal to this value. The default value is 12, valid only if
373 - ``txq_inline`` parameter [int]
375 Amount of data to be inlined during TX operations. This parameter is
376 deprecated and converted to the new parameter ``txq_inline_max`` providing
377 partial compatibility.
379 - ``txqs_min_inline`` parameter [int]
381 Enable inline data send only when the number of TX queues is greater or equal
384 This option should be used in combination with ``txq_inline_max`` and
385 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
387 If this option is not specified the default value 16 is used for BlueField
388 and 8 for other platforms
390 The data inlining consumes the CPU cycles, so this option is intended to
391 auto enable inline data if we have enough Tx queues, which means we have
392 enough CPU cores and PCI bandwidth is getting more critical and CPU
393 is not supposed to be bottleneck anymore.
395 The copying data into WQE improves latency and can improve PPS performance
396 when PCI back pressure is detected and may be useful for scenarios involving
397 heavy traffic on many queues.
399 Because additional software logic is necessary to handle this mode, this
400 option should be used with care, as it may lower performance when back
401 pressure is not expected.
403 - ``txq_inline_min`` parameter [int]
405 Minimal amount of data to be inlined into WQE during Tx operations. NICs
406 may require this minimal data amount to operate correctly. The exact value
407 may depend on NIC operation mode, requested offloads, etc. It is strongly
408 recommended to omit this parameter and use the default values. Anyway,
409 applications using this parameter should take into consideration that
410 specifying an inconsistent value may prevent the NIC from sending packets.
412 If ``txq_inline_min`` key is present the specified value (may be aligned
413 by the driver in order not to exceed the limits and provide better descriptor
414 space utilization) will be used by the driver and it is guaranteed that
415 requested amount of data bytes are inlined into the WQE beside other inline
416 settings. This key also may update ``txq_inline_max`` value (default
417 or specified explicitly in devargs) to reserve the space for inline data.
419 If ``txq_inline_min`` key is not present, the value may be queried by the
420 driver from the NIC via DevX if this feature is available. If there is no DevX
421 enabled/supported the value 18 (supposing L2 header including VLAN) is set
422 for ConnectX-4 and ConnectX-4LX, and 0 is set by default for ConnectX-5
423 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
426 For ConnectX-4 NIC, driver does not allow specifying value below 18
427 (minimal L2 header, including VLAN), error will be raised.
429 For ConnectX-4LX NIC, it is allowed to specify values below 18, but
430 it is not recommended and may prevent NIC from sending packets over
433 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
434 Multi-Packet Write), because last one does not support partial packet inlining.
435 This is not very critical due to minimal data inlining is mostly required
436 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
438 - ``txq_inline_max`` parameter [int]
440 Specifies the maximal packet length to be completely inlined into WQE
441 Ethernet Segment for ordinary SEND method. If packet is larger than specified
442 value, the packet data won't be copied by the driver at all, data buffer
443 is addressed with a pointer. If packet length is less or equal all packet
444 data will be copied into WQE. This may improve PCI bandwidth utilization for
445 short packets significantly but requires the extra CPU cycles.
447 The data inline feature is controlled by number of Tx queues, if number of Tx
448 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
449 is engaged, if there are not enough Tx queues (which means not enough CPU cores
450 and CPU resources are scarce), data inline is not performed by the driver.
451 Assigning ``txqs_min_inline`` with zero always enables the data inline.
453 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
454 by the driver in order not to exceed the limit (930 bytes) and to provide better
455 WQE space filling without gaps, the adjustment is reflected in the debug log.
457 - ``txq_inline_mpw`` parameter [int]
459 Specifies the maximal packet length to be completely inlined into WQE for
460 Enhanced MPW method. If packet is large the specified value, the packet data
461 won't be copied, and data buffer is addressed with pointer. If packet length
462 is less or equal, all packet data will be copied into WQE. This may improve PCI
463 bandwidth utilization for short packets significantly but requires the extra
466 The data inline feature is controlled by number of TX queues, if number of Tx
467 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
468 is engaged, if there are not enough Tx queues (which means not enough CPU cores
469 and CPU resources are scarce), data inline is not performed by the driver.
470 Assigning ``txqs_min_inline`` with zero always enables the data inline.
472 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
473 by the driver in order not to exceed the limit (930 bytes) and to provide better
474 WQE space filling without gaps, the adjustment is reflected in the debug log.
475 Due to multiple packets may be included to the same WQE with Enhanced Multi
476 Packet Write Method and overall WQE size is limited it is not recommended to
477 specify large values for the ``txq_inline_mpw``.
479 - ``txqs_max_vec`` parameter [int]
481 Enable vectorized Tx only when the number of TX queues is less than or
482 equal to this value. This parameter is deprecated and ignored, kept
483 for compatibility issue to not prevent driver from probing.
485 - ``txq_mpw_hdr_dseg_en`` parameter [int]
487 A nonzero value enables including two pointers in the first block of TX
488 descriptor. The parameter is deprecated and ignored, kept for compatibility
491 - ``txq_max_inline_len`` parameter [int]
493 Maximum size of packet to be inlined. This limits the size of packet to
494 be inlined. If the size of a packet is larger than configured value, the
495 packet isn't inlined even though there's enough space remained in the
496 descriptor. Instead, the packet is included with pointer. This parameter
497 is deprecated and converted directly to ``txq_inline_mpw`` providing full
498 compatibility. Valid only if eMPW feature is engaged.
500 - ``txq_mpw_en`` parameter [int]
502 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
503 ConnectX-6 and BlueField. eMPW allows the TX burst function to pack up multiple
504 packets in a single descriptor session in order to save PCI bandwidth and improve
505 performance at the cost of a slightly higher CPU usage. When ``txq_inline_mpw``
506 is set along with ``txq_mpw_en``, TX burst function copies entire packet
507 data on to TX descriptor instead of including pointer of packet.
509 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
510 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
511 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
512 option or reported by the NIC, the eMPW feature is disengaged.
514 - ``tx_vec_en`` parameter [int]
516 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6 and BlueField
517 NICs if the number of global Tx queues on the port is less than
518 ``txqs_max_vec``. The parameter is deprecated and ignored.
520 - ``rx_vec_en`` parameter [int]
522 A nonzero value enables Rx vector if the port is not configured in
523 multi-segment otherwise this parameter is ignored.
527 - ``vf_nl_en`` parameter [int]
529 A nonzero value enables Netlink requests from the VF to add/remove MAC
530 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
531 Otherwise the relevant configuration must be run with Linux iproute2 tools.
532 This is a prerequisite to receive this kind of traffic.
534 Enabled by default, valid only on VF devices ignored otherwise.
536 - ``l3_vxlan_en`` parameter [int]
538 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
539 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
540 parameter. This is a prerequisite to receive this kind of traffic.
544 - ``dv_flow_en`` parameter [int]
546 A nonzero value enables the DV flow steering assuming it is supported
551 - ``dv_esw_en`` parameter [int]
553 A nonzero value enables E-Switch using Direct Rules.
555 Enabled by default if supported.
557 - ``mr_ext_memseg_en`` parameter [int]
559 A nonzero value enables extending memseg when registering DMA memory. If
560 enabled, the number of entries in MR (Memory Region) lookup table on datapath
561 is minimized and it benefits performance. On the other hand, it worsens memory
562 utilization because registered memory is pinned by kernel driver. Even if a
563 page in the extended chunk is freed, that doesn't become reusable until the
564 entire memory is freed.
568 - ``representor`` parameter [list]
570 This parameter can be used to instantiate DPDK Ethernet devices from
571 existing port (or VF) representors configured on the device.
573 It is a standard parameter whose format is described in
574 :ref:`ethernet_device_standard_device_arguments`.
576 For instance, to probe port representors 0 through 2::
580 - ``max_dump_files_num`` parameter [int]
582 The maximum number of files per PMD entity that may be created for debug information.
583 The files will be created in /var/log directory or in current directory.
585 set to 128 by default.
587 - ``lro_timeout_usec`` parameter [int]
589 The maximum allowed duration of an LRO session, in micro-seconds.
590 PMD will set the nearest value supported by HW, which is not bigger than
591 the input ``lro_timeout_usec`` value.
592 If this parameter is not specified, by default PMD will set
593 the smallest value supported by HW.
595 .. _mlx5_firmware_config:
597 Firmware configuration
598 ~~~~~~~~~~~~~~~~~~~~~~
600 Firmware features can be configured as key/value pairs.
602 The command to set a value is::
604 mlxconfig -d <device> set <key>=<value>
606 The command to query a value is::
608 mlxconfig -d <device> query | grep <key>
610 The device name for the command ``mlxconfig`` can be either the PCI address,
611 or the mst device name found with::
615 Below are some firmware configurations listed.
621 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
627 - maximum number of SR-IOV virtual functions::
631 - enable DevX (required by Direct Rules and other features)::
635 - aggressive CQE zipping::
639 - L3 VXLAN and VXLAN-GPE destination UDP port::
642 IP_OVER_VXLAN_PORT=<udp dport>
644 - enable IP-in-IP tunnel flow matching::
646 FLEX_PARSER_PROFILE_ENABLE=0
648 - enable MPLS flow matching::
650 FLEX_PARSER_PROFILE_ENABLE=1
652 - enable ICMP/ICMP6 code/type fields matching::
654 FLEX_PARSER_PROFILE_ENABLE=2
659 This driver relies on external libraries and kernel drivers for resources
660 allocations and initialization. The following dependencies are not part of
661 DPDK and must be installed separately:
665 User space Verbs framework used by librte_pmd_mlx5. This library provides
666 a generic interface between the kernel and low-level user space drivers
669 It allows slow and privileged operations (context initialization, hardware
670 resources allocations) to be managed by the kernel and fast operations to
671 never leave user space.
675 Low-level user space driver library for Mellanox
676 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
679 This library basically implements send/receive calls to the hardware
684 They provide the kernel-side Verbs API and low level device drivers that
685 manage actual hardware initialization and resources sharing with user
688 Unlike most other PMDs, these modules must remain loaded and bound to
691 - mlx5_core: hardware driver managing Mellanox
692 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
694 - mlx5_ib: InifiniBand device driver.
695 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
697 - **Firmware update**
699 Mellanox OFED/EN releases include firmware updates for
700 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
702 Because each release provides new features, these updates must be applied to
703 match the kernel modules and libraries they come with.
707 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
713 Either RDMA Core library with a recent enough Linux kernel release
714 (recommended) or Mellanox OFED/EN, which provides compatibility with older
717 RDMA Core with Linux Kernel
718 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
720 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
721 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
722 (see `RDMA Core installation documentation`_)
723 - When building for i686 use:
725 - rdma-core version 18.0 or above built with 32bit support.
726 - Kernel version 4.14.41 or above.
728 - Starting with rdma-core v21, static libraries can be built::
731 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
734 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
735 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
737 If rdma-core libraries are built but not installed, DPDK makefile can link them,
738 thanks to these environment variables:
740 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
741 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
742 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
747 - Mellanox OFED version: ** 4.5, 4.6** /
748 Mellanox EN version: **4.5, 4.6**
751 - ConnectX-4: **12.21.1000** and above.
752 - ConnectX-4 Lx: **14.21.1000** and above.
753 - ConnectX-5: **16.21.1000** and above.
754 - ConnectX-5 Ex: **16.21.1000** and above.
755 - ConnectX-6: **20.99.5374** and above.
756 - BlueField: **18.25.1010** and above.
758 While these libraries and kernel modules are available on OpenFabrics
759 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
760 managers on most distributions, this PMD requires Ethernet extensions that
761 may not be supported at the moment (this is a work in progress).
764 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
766 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
767 include the necessary support and should be used in the meantime. For DPDK,
768 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
769 required from that distribution.
773 Several versions of Mellanox OFED/EN are available. Installing the version
774 this DPDK release was developed and tested against is strongly
775 recommended. Please check the `prerequisites`_.
780 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
781 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
782 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
783 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
784 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
785 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
786 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
787 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
788 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
789 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
790 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
791 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
792 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
793 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
794 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
795 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
796 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
797 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
798 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
800 Quick Start Guide on OFED/EN
801 ----------------------------
803 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
806 2. Install the required libraries and kernel modules either by installing
807 only the required set, or by installing the entire Mellanox OFED/EN::
809 ./mlnxofedinstall --upstream-libs --dpdk
811 3. Verify the firmware is the correct one::
815 4. Verify all ports links are set to Ethernet::
817 mlxconfig -d <mst device> query | grep LINK_TYPE
821 Link types may have to be configured to Ethernet::
823 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
825 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
827 For hypervisors, verify SR-IOV is enabled on the NIC::
829 mlxconfig -d <mst device> query | grep SRIOV_EN
832 If needed, configure SR-IOV::
834 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
835 mlxfwreset -d <mst device> reset
837 5. Restart the driver::
839 /etc/init.d/openibd restart
843 service openibd restart
845 If link type was changed, firmware must be reset as well::
847 mlxfwreset -d <mst device> reset
849 For hypervisors, after reset write the sysfs number of virtual functions
852 To dynamically instantiate a given number of virtual functions (VFs)::
854 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
856 6. Compile DPDK and you are ready to go. See instructions on
857 :ref:`Development Kit Build System <Development_Kit_Build_System>`
859 Enable switchdev mode
860 ---------------------
862 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
863 Representor is a port in DPDK that is connected to a VF in such a way
864 that assuming there are no offload flows, each packet that is sent from the VF
865 will be received by the corresponding representor. While each packet that is
866 sent to a representor will be received by the VF.
867 This is very useful in case of SRIOV mode, where the first packet that is sent
868 by the VF will be received by the DPDK application which will decide if this
869 flow should be offloaded to the E-Switch. After offloading the flow packet
870 that the VF that are matching the flow will not be received any more by
871 the DPDK application.
873 1. Enable SRIOV mode::
875 mlxconfig -d <mst device> set SRIOV_EN=true
877 2. Configure the max number of VFs::
879 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
883 mlxfwreset -d <mst device> reset
885 3. Configure the actual number of VFs::
887 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
889 4. Unbind the device (can be rebind after the switchdev mode)::
891 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
893 5. Enbale switchdev mode::
895 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
900 1. Configure aggressive CQE Zipping for maximum performance::
902 mlxconfig -d <mst device> s CQE_COMPRESSION=1
904 To set it back to the default CQE Zipping mode use::
906 mlxconfig -d <mst device> s CQE_COMPRESSION=0
908 2. In case of virtualization:
910 - Make sure that hypervisor kernel is 3.16 or newer.
911 - Configure boot with ``iommu=pt``.
913 - Make sure to allocate a VM on huge pages.
914 - Make sure to set CPU pinning.
916 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
917 for better performance. For VMs, verify that the right CPU
918 and NUMA node are pinned according to the above. Run::
922 to identify the NUMA node to which the PCIe adapter is connected.
924 4. If more than one adapter is used, and root complex capabilities allow
925 to put both adapters on the same NUMA node without PCI bandwidth degradation,
926 it is recommended to locate both adapters on the same NUMA node.
927 This in order to forward packets from one to the other without
928 NUMA performance penalty.
930 5. Disable pause frames::
932 ethtool -A <netdev> rx off tx off
934 6. Verify IO non-posted prefetch is disabled by default. This can be checked
935 via the BIOS configuration. Please contact you server provider for more
936 information about the settings.
940 On some machines, depends on the machine integrator, it is beneficial
941 to set the PCI max read request parameter to 1K. This can be
942 done in the following way:
944 To query the read request size use::
946 setpci -s <NIC PCI address> 68.w
948 If the output is different than 3XXX, set it by::
950 setpci -s <NIC PCI address> 68.w=3XXX
952 The XXX can be different on different systems. Make sure to configure
953 according to the setpci output.
955 7. To minimize overhead of searching Memory Regions:
957 - '--socket-mem' is recommended to pin memory by predictable amount.
958 - Configure per-lcore cache when creating Mempools for packet buffer.
959 - Refrain from dynamically allocating/freeing memory in run-time.
961 .. _mlx5_offloads_support:
963 Supported hardware offloads
964 ---------------------------
966 .. table:: Minimal SW/HW versions for queue offloads
968 ============== ===== ===== ========= ===== ========== ==========
969 Offload DPDK Linux rdma-core OFED firmware hardware
970 ============== ===== ===== ========= ===== ========== ==========
971 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
972 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
973 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
974 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
975 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
976 ============== ===== ===== ========= ===== ========== ==========
978 .. table:: Minimal SW/HW versions for rte_flow offloads
980 +-----------------------+-----------------+-----------------+
981 | Offload | with E-Switch | with vNIC |
982 +=======================+=================+=================+
983 | Count | | DPDK 19.05 | | DPDK 19.02 |
984 | | | OFED 4.6 | | OFED 4.6 |
985 | | | rdma-core 24 | | rdma-core 23 |
986 | | | ConnectX-5 | | ConnectX-5 |
987 +-----------------------+-----------------+-----------------+
988 | Drop / Queue / RSS | | DPDK 19.05 | | DPDK 18.11 |
989 | | | OFED 4.6 | | OFED 4.5 |
990 | | | rdma-core 24 | | rdma-core 23 |
991 | | | ConnectX-5 | | ConnectX-4 |
992 +-----------------------+-----------------+-----------------+
993 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
994 | (VXLAN / NVGRE / RAW) | | OFED 4.6-2 | | OFED 4.6 |
995 | | | rdma-core 24 | | rdma-core 23 |
996 | | | ConnectX-5 | | ConnectX-5 |
997 +-----------------------+-----------------+-----------------+
998 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
999 | | (set_ipv4_src / | | OFED 4.6-2 | | OFED 4.6-2 |
1000 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 23 |
1001 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1002 | | set_ipv6_dst / | | |
1003 | | set_tp_src / | | |
1004 | | set_tp_dst / | | |
1007 | | set_mac_src / | | |
1008 | | set_mac_dst) | | |
1009 +-----------------------+-----------------+-----------------+
1010 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1011 | | | OFED 4.6-4 | | OFED 4.6-4 |
1012 | | | rdma-core 24 | | N/A |
1013 | | | ConnectX-5 | | ConnectX-5 |
1014 +-----------------------+-----------------+-----------------+
1015 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1016 | | | OFED 4.6 | | OFED 4.5 |
1017 | | | rdma-core 24 | | rdma-core 23 |
1018 | | | ConnectX-5 | | ConnectX-4 |
1019 +-----------------------+-----------------+-----------------+
1020 | Port ID | | DPDK 19.05 | | N/A |
1021 | | | OFED 4.6 | | N/A |
1022 | | | rdma-core 24 | | N/A |
1023 | | | ConnectX-5 | | N/A |
1024 +-----------------------+-----------------+-----------------+
1029 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1030 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1032 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1033 command-line parameter to enable additional protocols (UDP and TCP as well
1034 as IP), the following commands must be entered from its CLI to get the same
1035 behavior as librte_pmd_mlx4::
1038 > port config all rss all
1044 This section demonstrates how to launch **testpmd** with Mellanox
1045 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1047 #. Load the kernel modules::
1049 modprobe -a ib_uverbs mlx5_core mlx5_ib
1051 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1054 /etc/init.d/openibd restart
1058 User space I/O kernel modules (uio and igb_uio) are not used and do
1059 not have to be loaded.
1061 #. Make sure Ethernet interfaces are in working order and linked to kernel
1062 verbs. Related sysfs entries should be present::
1064 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1073 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1076 for intf in eth2 eth3 eth4 eth5;
1078 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1081 sed -n 's,.*/\(.*\),-w \1,p'
1090 #. Request huge pages::
1092 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1094 #. Start testpmd with basic parameters::
1096 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
1101 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1102 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1103 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1104 PMD: librte_pmd_mlx5: 1 port(s) detected
1105 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1106 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1107 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1108 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1109 PMD: librte_pmd_mlx5: 1 port(s) detected
1110 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1111 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1112 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1113 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1114 PMD: librte_pmd_mlx5: 1 port(s) detected
1115 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1116 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1117 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1118 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1119 PMD: librte_pmd_mlx5: 1 port(s) detected
1120 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1121 Interactive-mode selected
1122 Configuring Port 0 (socket 0)
1123 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1124 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1125 Port 0: E4:1D:2D:E7:0C:FE
1126 Configuring Port 1 (socket 0)
1127 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1128 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1129 Port 1: E4:1D:2D:E7:0C:FF
1130 Configuring Port 2 (socket 0)
1131 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1132 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1133 Port 2: E4:1D:2D:E7:0C:FA
1134 Configuring Port 3 (socket 0)
1135 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1136 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1137 Port 3: E4:1D:2D:E7:0C:FB
1138 Checking link statuses...
1139 Port 0 Link Up - speed 40000 Mbps - full-duplex
1140 Port 1 Link Up - speed 40000 Mbps - full-duplex
1141 Port 2 Link Up - speed 10000 Mbps - full-duplex
1142 Port 3 Link Up - speed 10000 Mbps - full-duplex