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 less than 6 segments in case the Tx burst function
128 is set to multi-packet send or Enhanced multi-packet send. Otherwise it must have
129 less than 50 segments.
131 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
132 to 0 are not supported.
134 - VXLAN TSO and checksum offloads are not supported on VM.
136 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
138 - VF: flow rules created on VF devices can only match traffic targeted at the
139 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
143 MAC addresses not already present in the bridge table of the associated
144 kernel network device will be added and cleaned up by the PMD when closing
145 the device. In case of ungraceful program termination, some entries may
146 remain present and should be removed manually by other means.
148 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
149 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
150 ol_flags. As the mempool for the external buffer is managed by PMD, all the
151 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
152 the external buffers will be freed by PMD and the application which still
153 holds the external buffers may be corrupted.
155 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
156 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
157 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
159 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
160 and allmulticast mode are both set to off.
161 To receive IPv6 Multicast messages on VM, explicitly set the relevant
162 MAC address using rte_eth_dev_mac_addr_add() API.
164 - E-Switch decapsulation Flow:
166 - can be applied to PF port only.
167 - must specify VF port action (packet redirection from PF to VF).
168 - optionally may specify tunnel inner source and destination MAC addresses.
170 - E-Switch encapsulation Flow:
172 - can be applied to VF ports only.
173 - must specify PF port action (packet redirection from VF to PF).
175 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
176 mutually exclusive features which cannot be supported together
177 (see :ref:`mlx5_firmware_config`).
181 - Requires DevX to be enabled.
182 - KEEP_CRC offload cannot be supported with LRO.
183 - The first mbuf length, without head-room, must be big enough to include the
189 MLX5 supports various methods to report statistics:
191 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.
193 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.
195 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.
203 These options can be modified in the ``.config`` file.
205 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
207 Toggle compilation of librte_pmd_mlx5 itself.
209 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
211 Build PMD with additional code to make it loadable without hard
212 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
213 on the target system.
215 In this mode, their presence is still required for it to run properly,
216 however their absence won't prevent a DPDK application from starting (with
217 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
218 missing with ``ldd(1)``.
220 It works by moving these dependencies to a purpose-built rdma-core "glue"
221 plug-in which must either be installed in a directory whose name is based
222 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
223 standard location for the dynamic linker (e.g. ``/lib``) if left to the
224 default empty string (``""``).
226 This option has no performance impact.
228 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
230 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
231 in the PMD shared library or the executable static binary.
233 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
235 Toggle debugging code and stricter compilation flags. Enabling this option
236 adds additional run-time checks and debugging messages at the cost of
241 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
242 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
243 64. Default armv8a configuration of make build and meson build set it to 128
244 then brings performance degradation.
246 Environment variables
247 ~~~~~~~~~~~~~~~~~~~~~
251 A list of directories in which to search for the rdma-core "glue" plug-in,
252 separated by colons or semi-colons.
254 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
255 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
256 since ``LD_LIBRARY_PATH`` has no effect in this case.
258 - ``MLX5_SHUT_UP_BF``
260 Configures HW Tx doorbell register as IO-mapped.
262 By default, the HW Tx doorbell is configured as a write-combining register.
263 The register would be flushed to HW usually when the write-combining buffer
264 becomes full, but it depends on CPU design.
266 Except for vectorized Tx burst routines, a write memory barrier is enforced
267 after updating the register so that the update can be immediately visible to
270 When vectorized Tx burst is called, the barrier is set only if the burst size
271 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
272 variable will bring better latency even though the maximum throughput can
275 Run-time configuration
276 ~~~~~~~~~~~~~~~~~~~~~~
278 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
279 because it is affected by their state. Forcing them down prevents packets
282 - **ethtool** operations on related kernel interfaces also affect the PMD.
284 - ``rxq_cqe_comp_en`` parameter [int]
286 A nonzero value enables the compression of CQE on RX side. This feature
287 allows to save PCI bandwidth and improve performance. Enabled by default.
291 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
292 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
294 - ``rxq_cqe_pad_en`` parameter [int]
296 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
297 is aligned with the size of a cacheline of the core. If cacheline size is
298 128B, the CQE size is configured to be 128B even though the device writes
299 only 64B data on the cacheline. This is to avoid unnecessary cache
300 invalidation by device's two consecutive writes on to one cacheline.
301 However in some architecture, it is more beneficial to update entire
302 cacheline with padding the rest 64B rather than striding because
303 read-modify-write could drop performance a lot. On the other hand,
304 writing extra data will consume more PCIe bandwidth and could also drop
305 the maximum throughput. It is recommended to empirically set this
306 parameter. Disabled by default.
310 - CPU having 128B cacheline with ConnectX-5 and BlueField.
312 - ``rxq_pkt_pad_en`` parameter [int]
314 A nonzero value enables padding Rx packet to the size of cacheline on PCI
315 transaction. This feature would waste PCI bandwidth but could improve
316 performance by avoiding partial cacheline write which may cause costly
317 read-modify-copy in memory transaction on some architectures. Disabled by
322 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
323 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
325 - ``mprq_en`` parameter [int]
327 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
328 configured as Multi-Packet RQ if the total number of Rx queues is
329 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
332 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
333 by posting a single large buffer for multiple packets. Instead of posting a
334 buffers per a packet, one large buffer is posted in order to receive multiple
335 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
336 and each stride receives one packet. MPRQ can improve throughput for
337 small-packet traffic.
339 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
340 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
341 configure large stride size enough to accommodate max_rx_pkt_len as long as
342 device allows. Note that this can waste system memory compared to enabling Rx
343 scatter and multi-segment packet.
345 - ``mprq_log_stride_num`` parameter [int]
347 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
348 strides can reduce PCIe traffic further. If configured value is not in the
349 range of device capability, the default value will be set with a warning
350 message. The default value is 4 which is 16 strides per a buffer, valid only
351 if ``mprq_en`` is set.
353 The size of Rx queue should be bigger than the number of strides.
355 - ``mprq_max_memcpy_len`` parameter [int]
357 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
358 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
359 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
360 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
361 A mempool for external buffers will be allocated and managed by PMD. If Rx
362 packet is externally attached, ol_flags field of the mbuf will have
363 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
364 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
366 - ``rxqs_min_mprq`` parameter [int]
368 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
369 greater or equal to this value. The default value is 12, valid only if
372 - ``txq_inline`` parameter [int]
374 Amount of data to be inlined during TX operations. This parameter is
375 deprecated and converted to the new parameter ``txq_inline_max`` providing
376 partial compatibility.
378 - ``txqs_min_inline`` parameter [int]
380 Enable inline data send only when the number of TX queues is greater or equal
383 This option should be used in combination with ``txq_inline_max`` and
384 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
386 If this option is not specified the default value 16 is used for BlueField
387 and 8 for other platforms
389 The data inlining consumes the CPU cycles, so this option is intended to
390 auto enable inline data if we have enough Tx queues, which means we have
391 enough CPU cores and PCI bandwidth is getting more critical and CPU
392 is not supposed to be bottleneck anymore.
394 The copying data into WQE improves latency and can improve PPS performance
395 when PCI back pressure is detected and may be useful for scenarios involving
396 heavy traffic on many queues.
398 Because additional software logic is necessary to handle this mode, this
399 option should be used with care, as it may lower performance when back
400 pressure is not expected.
402 - ``txq_inline_min`` parameter [int]
404 Minimal amount of data to be inlined into WQE during Tx operations. NICs
405 may require this minimal data amount to operate correctly. The exact value
406 may depend on NIC operation mode, requested offloads, etc.
408 If ``txq_inline_min`` key is present the specified value (may be aligned
409 by the driver in order not to exceed the limits and provide better descriptor
410 space utilization) will be used by the driver and it is guaranteed the
411 requested data bytes are inlined into the WQE beside other inline settings.
412 This keys also may update ``txq_inline_max`` value (default of specified
413 explicitly in devargs) to reserve the space for inline data.
415 If ``txq_inline_min`` key is not present, the value may be queried by the
416 driver from the NIC via DevX if this feature is available. If there is no DevX
417 enabled/supported the value 18 (supposing L2 header including VLAN) is set
418 for ConnectX-4, value 58 (supposing L2-L4 headers, required by configurations
419 over E-Switch) is set for ConnectX-4 Lx, and 0 is set by default for ConnectX-5
420 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
423 For the ConnectX-4 and ConnectX-4 Lx NICs driver does not allow to set
424 this value below 18 (minimal L2 header, including VLAN).
426 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
427 Multi-Packet Write), because last one does not support partial packet inlining.
428 This is not very critical due to minimal data inlining is mostly required
429 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
431 - ``txq_inline_max`` parameter [int]
433 Specifies the maximal packet length to be completely inlined into WQE
434 Ethernet Segment for ordinary SEND method. If packet is larger than specified
435 value, the packet data won't be copied by the driver at all, data buffer
436 is addressed with a pointer. If packet length is less or equal all packet
437 data will be copied into WQE. This may improve PCI bandwidth utilization for
438 short packets significantly but requires the extra CPU cycles.
440 The data inline feature is controlled by number of Tx queues, if number of Tx
441 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
442 is engaged, if there are not enough Tx queues (which means not enough CPU cores
443 and CPU resources are scarce), data inline is not performed by the driver.
444 Assigning ``txqs_min_inline`` with zero always enables the data inline.
446 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
447 by the driver in order not to exceed the limit (930 bytes) and to provide better
448 WQE space filling without gaps, the adjustment is reflected in the debug log.
450 - ``txq_inline_mpw`` parameter [int]
452 Specifies the maximal packet length to be completely inlined into WQE for
453 Enhanced MPW method. If packet is large the specified value, the packet data
454 won't be copied, and data buffer is addressed with pointer. If packet length
455 is less or equal, all packet data will be copied into WQE. This may improve PCI
456 bandwidth utilization for short packets significantly but requires the extra
459 The data inline feature is controlled by number of TX queues, if number of Tx
460 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
461 is engaged, if there are not enough Tx queues (which means not enough CPU cores
462 and CPU resources are scarce), data inline is not performed by the driver.
463 Assigning ``txqs_min_inline`` with zero always enables the data inline.
465 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
466 by the driver in order not to exceed the limit (930 bytes) and to provide better
467 WQE space filling without gaps, the adjustment is reflected in the debug log.
468 Due to multiple packets may be included to the same WQE with Enhanced Multi
469 Packet Write Method and overall WQE size is limited it is not recommended to
470 specify large values for the ``txq_inline_mpw``.
472 - ``txqs_max_vec`` parameter [int]
474 Enable vectorized Tx only when the number of TX queues is less than or
475 equal to this value. This parameter is deprecated and ignored, kept
476 for compatibility issue to not prevent driver from probing.
478 - ``txq_mpw_hdr_dseg_en`` parameter [int]
480 A nonzero value enables including two pointers in the first block of TX
481 descriptor. The parameter is deprecated and ignored, kept for compatibility
484 - ``txq_max_inline_len`` parameter [int]
486 Maximum size of packet to be inlined. This limits the size of packet to
487 be inlined. If the size of a packet is larger than configured value, the
488 packet isn't inlined even though there's enough space remained in the
489 descriptor. Instead, the packet is included with pointer. This parameter
490 is deprecated and converted directly to ``txq_inline_mpw`` providing full
491 compatibility. Valid only if eMPW feature is engaged.
493 - ``txq_mpw_en`` parameter [int]
495 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
496 ConnectX-6 and BlueField. eMPW allows the TX burst function to pack up multiple
497 packets in a single descriptor session in order to save PCI bandwidth and improve
498 performance at the cost of a slightly higher CPU usage. When ``txq_inline_mpw``
499 is set along with ``txq_mpw_en``, TX burst function copies entire packet
500 data on to TX descriptor instead of including pointer of packet.
502 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
503 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
504 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
505 option or reported by the NIC, the eMPW feature is disengaged.
507 - ``tx_vec_en`` parameter [int]
509 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6 and BlueField
510 NICs if the number of global Tx queues on the port is less than
511 ``txqs_max_vec``. The parameter is deprecated and ignored.
513 - ``rx_vec_en`` parameter [int]
515 A nonzero value enables Rx vector if the port is not configured in
516 multi-segment otherwise this parameter is ignored.
520 - ``vf_nl_en`` parameter [int]
522 A nonzero value enables Netlink requests from the VF to add/remove MAC
523 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
524 Otherwise the relevant configuration must be run with Linux iproute2 tools.
525 This is a prerequisite to receive this kind of traffic.
527 Enabled by default, valid only on VF devices ignored otherwise.
529 - ``l3_vxlan_en`` parameter [int]
531 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
532 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
533 parameter. This is a prerequisite to receive this kind of traffic.
537 - ``dv_flow_en`` parameter [int]
539 A nonzero value enables the DV flow steering assuming it is supported
544 - ``dv_esw_en`` parameter [int]
546 A nonzero value enables E-Switch using Direct Rules.
548 Enabled by default if supported.
550 - ``mr_ext_memseg_en`` parameter [int]
552 A nonzero value enables extending memseg when registering DMA memory. If
553 enabled, the number of entries in MR (Memory Region) lookup table on datapath
554 is minimized and it benefits performance. On the other hand, it worsens memory
555 utilization because registered memory is pinned by kernel driver. Even if a
556 page in the extended chunk is freed, that doesn't become reusable until the
557 entire memory is freed.
561 - ``representor`` parameter [list]
563 This parameter can be used to instantiate DPDK Ethernet devices from
564 existing port (or VF) representors configured on the device.
566 It is a standard parameter whose format is described in
567 :ref:`ethernet_device_standard_device_arguments`.
569 For instance, to probe port representors 0 through 2::
573 - ``max_dump_files_num`` parameter [int]
575 The maximum number of files per PMD entity that may be created for debug information.
576 The files will be created in /var/log directory or in current directory.
578 set to 128 by default.
580 - ``lro_timeout_usec`` parameter [int]
582 The maximum allowed duration of an LRO session, in micro-seconds.
583 PMD will set the nearest value supported by HW, which is not bigger than
584 the input ``lro_timeout_usec`` value.
585 If this parameter is not specified, by default PMD will set
586 the smallest value supported by HW.
588 .. _mlx5_firmware_config:
590 Firmware configuration
591 ~~~~~~~~~~~~~~~~~~~~~~
593 Firmware features can be configured as key/value pairs.
595 The command to set a value is::
597 mlxconfig -d <device> set <key>=<value>
599 The command to query a value is::
601 mlxconfig -d <device> query | grep <key>
603 The device name for the command ``mlxconfig`` can be either the PCI address,
604 or the mst device name found with::
608 Below are some firmware configurations listed.
614 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
620 - maximum number of SR-IOV virtual functions::
624 - enable DevX (required by Direct Rules and other features)::
628 - aggressive CQE zipping::
632 - L3 VXLAN and VXLAN-GPE destination UDP port::
635 IP_OVER_VXLAN_PORT=<udp dport>
637 - enable IP-in-IP tunnel flow matching::
639 FLEX_PARSER_PROFILE_ENABLE=0
641 - enable MPLS flow matching::
643 FLEX_PARSER_PROFILE_ENABLE=1
645 - enable ICMP/ICMP6 code/type fields matching::
647 FLEX_PARSER_PROFILE_ENABLE=2
652 This driver relies on external libraries and kernel drivers for resources
653 allocations and initialization. The following dependencies are not part of
654 DPDK and must be installed separately:
658 User space Verbs framework used by librte_pmd_mlx5. This library provides
659 a generic interface between the kernel and low-level user space drivers
662 It allows slow and privileged operations (context initialization, hardware
663 resources allocations) to be managed by the kernel and fast operations to
664 never leave user space.
668 Low-level user space driver library for Mellanox
669 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
672 This library basically implements send/receive calls to the hardware
677 They provide the kernel-side Verbs API and low level device drivers that
678 manage actual hardware initialization and resources sharing with user
681 Unlike most other PMDs, these modules must remain loaded and bound to
684 - mlx5_core: hardware driver managing Mellanox
685 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
687 - mlx5_ib: InifiniBand device driver.
688 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
690 - **Firmware update**
692 Mellanox OFED/EN releases include firmware updates for
693 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
695 Because each release provides new features, these updates must be applied to
696 match the kernel modules and libraries they come with.
700 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
706 Either RDMA Core library with a recent enough Linux kernel release
707 (recommended) or Mellanox OFED/EN, which provides compatibility with older
710 RDMA Core with Linux Kernel
711 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
713 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
714 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
715 (see `RDMA Core installation documentation`_)
716 - When building for i686 use:
718 - rdma-core version 18.0 or above built with 32bit support.
719 - Kernel version 4.14.41 or above.
721 - Starting with rdma-core v21, static libraries can be built::
724 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
727 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
728 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
730 If rdma-core libraries are built but not installed, DPDK makefile can link them,
731 thanks to these environment variables:
733 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
734 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
735 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
740 - Mellanox OFED version: ** 4.5, 4.6** /
741 Mellanox EN version: **4.5, 4.6**
744 - ConnectX-4: **12.21.1000** and above.
745 - ConnectX-4 Lx: **14.21.1000** and above.
746 - ConnectX-5: **16.21.1000** and above.
747 - ConnectX-5 Ex: **16.21.1000** and above.
748 - ConnectX-6: **20.99.5374** and above.
749 - BlueField: **18.25.1010** and above.
751 While these libraries and kernel modules are available on OpenFabrics
752 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
753 managers on most distributions, this PMD requires Ethernet extensions that
754 may not be supported at the moment (this is a work in progress).
757 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
759 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
760 include the necessary support and should be used in the meantime. For DPDK,
761 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
762 required from that distribution.
766 Several versions of Mellanox OFED/EN are available. Installing the version
767 this DPDK release was developed and tested against is strongly
768 recommended. Please check the `prerequisites`_.
773 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
774 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
775 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
776 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
777 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
778 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
779 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
780 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
781 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
782 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
783 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
784 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
785 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
786 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
787 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
788 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
789 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
790 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
791 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
793 Quick Start Guide on OFED/EN
794 ----------------------------
796 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
799 2. Install the required libraries and kernel modules either by installing
800 only the required set, or by installing the entire Mellanox OFED/EN::
802 ./mlnxofedinstall --upstream-libs --dpdk
804 3. Verify the firmware is the correct one::
808 4. Verify all ports links are set to Ethernet::
810 mlxconfig -d <mst device> query | grep LINK_TYPE
814 Link types may have to be configured to Ethernet::
816 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
818 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
820 For hypervisors, verify SR-IOV is enabled on the NIC::
822 mlxconfig -d <mst device> query | grep SRIOV_EN
825 If needed, configure SR-IOV::
827 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
828 mlxfwreset -d <mst device> reset
830 5. Restart the driver::
832 /etc/init.d/openibd restart
836 service openibd restart
838 If link type was changed, firmware must be reset as well::
840 mlxfwreset -d <mst device> reset
842 For hypervisors, after reset write the sysfs number of virtual functions
845 To dynamically instantiate a given number of virtual functions (VFs)::
847 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
849 6. Compile DPDK and you are ready to go. See instructions on
850 :ref:`Development Kit Build System <Development_Kit_Build_System>`
852 Enable switchdev mode
853 ---------------------
855 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
856 Representor is a port in DPDK that is connected to a VF in such a way
857 that assuming there are no offload flows, each packet that is sent from the VF
858 will be received by the corresponding representor. While each packet that is
859 sent to a representor will be received by the VF.
860 This is very useful in case of SRIOV mode, where the first packet that is sent
861 by the VF will be received by the DPDK application which will decide if this
862 flow should be offloaded to the E-Switch. After offloading the flow packet
863 that the VF that are matching the flow will not be received any more by
864 the DPDK application.
866 1. Enable SRIOV mode::
868 mlxconfig -d <mst device> set SRIOV_EN=true
870 2. Configure the max number of VFs::
872 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
876 mlxfwreset -d <mst device> reset
878 3. Configure the actual number of VFs::
880 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
882 4. Unbind the device (can be rebind after the switchdev mode)::
884 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
886 5. Enbale switchdev mode::
888 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
893 1. Configure aggressive CQE Zipping for maximum performance::
895 mlxconfig -d <mst device> s CQE_COMPRESSION=1
897 To set it back to the default CQE Zipping mode use::
899 mlxconfig -d <mst device> s CQE_COMPRESSION=0
901 2. In case of virtualization:
903 - Make sure that hypervisor kernel is 3.16 or newer.
904 - Configure boot with ``iommu=pt``.
906 - Make sure to allocate a VM on huge pages.
907 - Make sure to set CPU pinning.
909 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
910 for better performance. For VMs, verify that the right CPU
911 and NUMA node are pinned according to the above. Run::
915 to identify the NUMA node to which the PCIe adapter is connected.
917 4. If more than one adapter is used, and root complex capabilities allow
918 to put both adapters on the same NUMA node without PCI bandwidth degradation,
919 it is recommended to locate both adapters on the same NUMA node.
920 This in order to forward packets from one to the other without
921 NUMA performance penalty.
923 5. Disable pause frames::
925 ethtool -A <netdev> rx off tx off
927 6. Verify IO non-posted prefetch is disabled by default. This can be checked
928 via the BIOS configuration. Please contact you server provider for more
929 information about the settings.
933 On some machines, depends on the machine integrator, it is beneficial
934 to set the PCI max read request parameter to 1K. This can be
935 done in the following way:
937 To query the read request size use::
939 setpci -s <NIC PCI address> 68.w
941 If the output is different than 3XXX, set it by::
943 setpci -s <NIC PCI address> 68.w=3XXX
945 The XXX can be different on different systems. Make sure to configure
946 according to the setpci output.
948 7. To minimize overhead of searching Memory Regions:
950 - '--socket-mem' is recommended to pin memory by predictable amount.
951 - Configure per-lcore cache when creating Mempools for packet buffer.
952 - Refrain from dynamically allocating/freeing memory in run-time.
954 .. _mlx5_offloads_support:
956 Supported hardware offloads
957 ---------------------------
959 .. table:: Minimal SW/HW versions for queue offloads
961 ============== ===== ===== ========= ===== ========== ==========
962 Offload DPDK Linux rdma-core OFED firmware hardware
963 ============== ===== ===== ========= ===== ========== ==========
964 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
965 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
966 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
967 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
968 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
969 ============== ===== ===== ========= ===== ========== ==========
971 .. table:: Minimal SW/HW versions for rte_flow offloads
973 +-----------------------+-----------------+-----------------+
974 | Offload | with E-Switch | with vNIC |
975 +=======================+=================+=================+
976 | Count | | DPDK 19.05 | | DPDK 19.02 |
977 | | | OFED 4.6 | | OFED 4.6 |
978 | | | rdma-core 24 | | rdma-core 23 |
979 | | | ConnectX-5 | | ConnectX-5 |
980 +-----------------------+-----------------+-----------------+
981 | Drop / Queue / RSS | | DPDK 19.05 | | DPDK 18.11 |
982 | | | OFED 4.6 | | OFED 4.5 |
983 | | | rdma-core 24 | | rdma-core 23 |
984 | | | ConnectX-5 | | ConnectX-4 |
985 +-----------------------+-----------------+-----------------+
986 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
987 | (VXLAN / NVGRE / RAW) | | OFED 4.6-2 | | OFED 4.6 |
988 | | | rdma-core 24 | | rdma-core 23 |
989 | | | ConnectX-5 | | ConnectX-5 |
990 +-----------------------+-----------------+-----------------+
991 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
992 | | (set_ipv4_src / | | OFED 4.6-2 | | OFED 4.6-2 |
993 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 23 |
994 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
995 | | set_ipv6_dst / | | |
996 | | set_tp_src / | | |
997 | | set_tp_dst / | | |
1000 | | set_mac_src / | | |
1001 | | set_mac_dst) | | |
1002 +-----------------------+-----------------+-----------------+
1003 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1004 | | | OFED 4.6-4 | | OFED 4.6-4 |
1005 | | | rdma-core 24 | | N/A |
1006 | | | ConnectX-5 | | ConnectX-5 |
1007 +-----------------------+-----------------+-----------------+
1008 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1009 | | | OFED 4.6 | | OFED 4.5 |
1010 | | | rdma-core 24 | | rdma-core 23 |
1011 | | | ConnectX-5 | | ConnectX-4 |
1012 +-----------------------+-----------------+-----------------+
1013 | Port ID | | DPDK 19.05 | | N/A |
1014 | | | OFED 4.6 | | N/A |
1015 | | | rdma-core 24 | | N/A |
1016 | | | ConnectX-5 | | N/A |
1017 +-----------------------+-----------------+-----------------+
1022 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1023 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1025 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1026 command-line parameter to enable additional protocols (UDP and TCP as well
1027 as IP), the following commands must be entered from its CLI to get the same
1028 behavior as librte_pmd_mlx4::
1031 > port config all rss all
1037 This section demonstrates how to launch **testpmd** with Mellanox
1038 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1040 #. Load the kernel modules::
1042 modprobe -a ib_uverbs mlx5_core mlx5_ib
1044 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1047 /etc/init.d/openibd restart
1051 User space I/O kernel modules (uio and igb_uio) are not used and do
1052 not have to be loaded.
1054 #. Make sure Ethernet interfaces are in working order and linked to kernel
1055 verbs. Related sysfs entries should be present::
1057 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1066 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1069 for intf in eth2 eth3 eth4 eth5;
1071 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1074 sed -n 's,.*/\(.*\),-w \1,p'
1083 #. Request huge pages::
1085 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1087 #. Start testpmd with basic parameters::
1089 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
1094 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1095 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1096 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1097 PMD: librte_pmd_mlx5: 1 port(s) detected
1098 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1099 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1100 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1101 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1102 PMD: librte_pmd_mlx5: 1 port(s) detected
1103 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1104 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1105 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1106 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1107 PMD: librte_pmd_mlx5: 1 port(s) detected
1108 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1109 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1110 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1111 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1112 PMD: librte_pmd_mlx5: 1 port(s) detected
1113 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1114 Interactive-mode selected
1115 Configuring Port 0 (socket 0)
1116 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1117 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1118 Port 0: E4:1D:2D:E7:0C:FE
1119 Configuring Port 1 (socket 0)
1120 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1121 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1122 Port 1: E4:1D:2D:E7:0C:FF
1123 Configuring Port 2 (socket 0)
1124 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1125 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1126 Port 2: E4:1D:2D:E7:0C:FA
1127 Configuring Port 3 (socket 0)
1128 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1129 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1130 Port 3: E4:1D:2D:E7:0C:FB
1131 Checking link statuses...
1132 Port 0 Link Up - speed 40000 Mbps - full-duplex
1133 Port 1 Link Up - speed 40000 Mbps - full-duplex
1134 Port 2 Link Up - speed 10000 Mbps - full-duplex
1135 Port 3 Link Up - speed 10000 Mbps - full-duplex