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 cannot be supported togeter with IP-in-IP tunnel.
179 - KEEP_CRC offload cannot be supported with LRO.
180 - The first mbuf length, without head-room, must be big enough to include the
186 MLX5 supports various methods to report statistics:
188 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.
190 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.
192 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.
200 These options can be modified in the ``.config`` file.
202 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
204 Toggle compilation of librte_pmd_mlx5 itself.
206 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
208 Build PMD with additional code to make it loadable without hard
209 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
210 on the target system.
212 In this mode, their presence is still required for it to run properly,
213 however their absence won't prevent a DPDK application from starting (with
214 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
215 missing with ``ldd(1)``.
217 It works by moving these dependencies to a purpose-built rdma-core "glue"
218 plug-in which must either be installed in a directory whose name is based
219 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
220 standard location for the dynamic linker (e.g. ``/lib``) if left to the
221 default empty string (``""``).
223 This option has no performance impact.
225 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
227 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
228 in the PMD shared library or the executable static binary.
230 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
232 Toggle debugging code and stricter compilation flags. Enabling this option
233 adds additional run-time checks and debugging messages at the cost of
238 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
239 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
240 64. Default armv8a configuration of make build and meson build set it to 128
241 then brings performance degradation.
243 Environment variables
244 ~~~~~~~~~~~~~~~~~~~~~
248 A list of directories in which to search for the rdma-core "glue" plug-in,
249 separated by colons or semi-colons.
251 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
252 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
253 since ``LD_LIBRARY_PATH`` has no effect in this case.
255 - ``MLX5_SHUT_UP_BF``
257 Configures HW Tx doorbell register as IO-mapped.
259 By default, the HW Tx doorbell is configured as a write-combining register.
260 The register would be flushed to HW usually when the write-combining buffer
261 becomes full, but it depends on CPU design.
263 Except for vectorized Tx burst routines, a write memory barrier is enforced
264 after updating the register so that the update can be immediately visible to
267 When vectorized Tx burst is called, the barrier is set only if the burst size
268 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
269 variable will bring better latency even though the maximum throughput can
272 Run-time configuration
273 ~~~~~~~~~~~~~~~~~~~~~~
275 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
276 because it is affected by their state. Forcing them down prevents packets
279 - **ethtool** operations on related kernel interfaces also affect the PMD.
281 - ``rxq_cqe_comp_en`` parameter [int]
283 A nonzero value enables the compression of CQE on RX side. This feature
284 allows to save PCI bandwidth and improve performance. Enabled by default.
288 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
289 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
291 - ``rxq_cqe_pad_en`` parameter [int]
293 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
294 is aligned with the size of a cacheline of the core. If cacheline size is
295 128B, the CQE size is configured to be 128B even though the device writes
296 only 64B data on the cacheline. This is to avoid unnecessary cache
297 invalidation by device's two consecutive writes on to one cacheline.
298 However in some architecture, it is more beneficial to update entire
299 cacheline with padding the rest 64B rather than striding because
300 read-modify-write could drop performance a lot. On the other hand,
301 writing extra data will consume more PCIe bandwidth and could also drop
302 the maximum throughput. It is recommended to empirically set this
303 parameter. Disabled by default.
307 - CPU having 128B cacheline with ConnectX-5 and BlueField.
309 - ``rxq_pkt_pad_en`` parameter [int]
311 A nonzero value enables padding Rx packet to the size of cacheline on PCI
312 transaction. This feature would waste PCI bandwidth but could improve
313 performance by avoiding partial cacheline write which may cause costly
314 read-modify-copy in memory transaction on some architectures. Disabled by
319 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
320 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
322 - ``mprq_en`` parameter [int]
324 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
325 configured as Multi-Packet RQ if the total number of Rx queues is
326 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
329 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
330 by posting a single large buffer for multiple packets. Instead of posting a
331 buffers per a packet, one large buffer is posted in order to receive multiple
332 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
333 and each stride receives one packet. MPRQ can improve throughput for
334 small-packet traffic.
336 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
337 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
338 configure large stride size enough to accommodate max_rx_pkt_len as long as
339 device allows. Note that this can waste system memory compared to enabling Rx
340 scatter and multi-segment packet.
342 - ``mprq_log_stride_num`` parameter [int]
344 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
345 strides can reduce PCIe traffic further. If configured value is not in the
346 range of device capability, the default value will be set with a warning
347 message. The default value is 4 which is 16 strides per a buffer, valid only
348 if ``mprq_en`` is set.
350 The size of Rx queue should be bigger than the number of strides.
352 - ``mprq_max_memcpy_len`` parameter [int]
354 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
355 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
356 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
357 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
358 A mempool for external buffers will be allocated and managed by PMD. If Rx
359 packet is externally attached, ol_flags field of the mbuf will have
360 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
361 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
363 - ``rxqs_min_mprq`` parameter [int]
365 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
366 greater or equal to this value. The default value is 12, valid only if
369 - ``txq_inline`` parameter [int]
371 Amount of data to be inlined during TX operations. This parameter is
372 deprecated and converted to the new parameter ``txq_inline_max`` providing
373 partial compatibility.
375 - ``txqs_min_inline`` parameter [int]
377 Enable inline data send only when the number of TX queues is greater or equal
380 This option should be used in combination with ``txq_inline_max`` and
381 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
383 If this option is not specified the default value 16 is used for BlueField
384 and 8 for other platforms
386 The data inlining consumes the CPU cycles, so this option is intended to
387 auto enable inline data if we have enough Tx queues, which means we have
388 enough CPU cores and PCI bandwidth is getting more critical and CPU
389 is not supposed to be bottleneck anymore.
391 The copying data into WQE improves latency and can improve PPS performance
392 when PCI back pressure is detected and may be useful for scenarios involving
393 heavy traffic on many queues.
395 Because additional software logic is necessary to handle this mode, this
396 option should be used with care, as it may lower performance when back
397 pressure is not expected.
399 - ``txq_inline_min`` parameter [int]
401 Minimal amount of data to be inlined into WQE during Tx operations. NICs
402 may require this minimal data amount to operate correctly. The exact value
403 may depend on NIC operation mode, requested offloads, etc.
405 If ``txq_inline_min`` key is present the specified value (may be aligned
406 by the driver in order not to exceed the limits and provide better descriptor
407 space utilization) will be used by the driver and it is guaranteed the
408 requested data bytes are inlined into the WQE beside other inline settings.
409 This keys also may update ``txq_inline_max`` value (default of specified
410 explicitly in devargs) to reserve the space for inline data.
412 If ``txq_inline_min`` key is not present, the value may be queried by the
413 driver from the NIC via DevX if this feature is available. If there is no DevX
414 enabled/supported the value 18 (supposing L2 header including VLAN) is set
415 for ConnectX-4, value 58 (supposing L2-L4 headers, required by configurations
416 over E-Switch) is set for ConnectX-4 Lx, and 0 is set by default for ConnectX-5
417 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
420 For the ConnectX-4 and ConnectX-4 Lx NICs driver does not allow to set
421 this value below 18 (minimal L2 header, including VLAN).
423 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
424 Multi-Packet Write), because last one does not support partial packet inlining.
425 This is not very critical due to minimal data inlining is mostly required
426 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
428 - ``txq_inline_max`` parameter [int]
430 Specifies the maximal packet length to be completely inlined into WQE
431 Ethernet Segment for ordinary SEND method. If packet is larger than specified
432 value, the packet data won't be copied by the driver at all, data buffer
433 is addressed with a pointer. If packet length is less or equal all packet
434 data will be copied into WQE. This may improve PCI bandwidth utilization for
435 short packets significantly but requires the extra CPU cycles.
437 The data inline feature is controlled by number of Tx queues, if number of Tx
438 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
439 is engaged, if there are not enough Tx queues (which means not enough CPU cores
440 and CPU resources are scarce), data inline is not performed by the driver.
441 Assigning ``txqs_min_inline`` with zero always enables the data inline.
443 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
444 by the driver in order not to exceed the limit (930 bytes) and to provide better
445 WQE space filling without gaps, the adjustment is reflected in the debug log.
447 - ``txq_inline_mpw`` parameter [int]
449 Specifies the maximal packet length to be completely inlined into WQE for
450 Enhanced MPW method. If packet is large the specified value, the packet data
451 won't be copied, and data buffer is addressed with pointer. If packet length
452 is less or equal, all packet data will be copied into WQE. This may improve PCI
453 bandwidth utilization for short packets significantly but requires the extra
456 The data inline feature is controlled by number of TX queues, if number of Tx
457 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
458 is engaged, if there are not enough Tx queues (which means not enough CPU cores
459 and CPU resources are scarce), data inline is not performed by the driver.
460 Assigning ``txqs_min_inline`` with zero always enables the data inline.
462 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
463 by the driver in order not to exceed the limit (930 bytes) and to provide better
464 WQE space filling without gaps, the adjustment is reflected in the debug log.
465 Due to multiple packets may be included to the same WQE with Enhanced Multi
466 Packet Write Method and overall WQE size is limited it is not recommended to
467 specify large values for the ``txq_inline_mpw``.
469 - ``txqs_max_vec`` parameter [int]
471 Enable vectorized Tx only when the number of TX queues is less than or
472 equal to this value. This parameter is deprecated and ignored, kept
473 for compatibility issue to not prevent driver from probing.
475 - ``txq_mpw_hdr_dseg_en`` parameter [int]
477 A nonzero value enables including two pointers in the first block of TX
478 descriptor. The parameter is deprecated and ignored, kept for compatibility
481 - ``txq_max_inline_len`` parameter [int]
483 Maximum size of packet to be inlined. This limits the size of packet to
484 be inlined. If the size of a packet is larger than configured value, the
485 packet isn't inlined even though there's enough space remained in the
486 descriptor. Instead, the packet is included with pointer. This parameter
487 is deprecated and converted directly to ``txq_inline_mpw`` providing full
488 compatibility. Valid only if eMPW feature is engaged.
490 - ``txq_mpw_en`` parameter [int]
492 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
493 ConnectX-6 and BlueField. eMPW allows the TX burst function to pack up multiple
494 packets in a single descriptor session in order to save PCI bandwidth and improve
495 performance at the cost of a slightly higher CPU usage. When ``txq_inline_mpw``
496 is set along with ``txq_mpw_en``, TX burst function copies entire packet
497 data on to TX descriptor instead of including pointer of packet.
499 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
500 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
501 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
502 option or reported by the NIC, the eMPW feature is disengaged.
504 - ``tx_vec_en`` parameter [int]
506 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6 and BlueField
507 NICs if the number of global Tx queues on the port is less than
508 ``txqs_max_vec``. The parameter is deprecated and ignored.
510 - ``rx_vec_en`` parameter [int]
512 A nonzero value enables Rx vector if the port is not configured in
513 multi-segment otherwise this parameter is ignored.
517 - ``vf_nl_en`` parameter [int]
519 A nonzero value enables Netlink requests from the VF to add/remove MAC
520 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
521 Otherwise the relevant configuration must be run with Linux iproute2 tools.
522 This is a prerequisite to receive this kind of traffic.
524 Enabled by default, valid only on VF devices ignored otherwise.
526 - ``l3_vxlan_en`` parameter [int]
528 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
529 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
530 parameter. This is a prerequisite to receive this kind of traffic.
534 - ``dv_flow_en`` parameter [int]
536 A nonzero value enables the DV flow steering assuming it is supported
541 - ``dv_esw_en`` parameter [int]
543 A nonzero value enables E-Switch using Direct Rules.
545 Enabled by default if supported.
547 - ``mr_ext_memseg_en`` parameter [int]
549 A nonzero value enables extending memseg when registering DMA memory. If
550 enabled, the number of entries in MR (Memory Region) lookup table on datapath
551 is minimized and it benefits performance. On the other hand, it worsens memory
552 utilization because registered memory is pinned by kernel driver. Even if a
553 page in the extended chunk is freed, that doesn't become reusable until the
554 entire memory is freed.
558 - ``representor`` parameter [list]
560 This parameter can be used to instantiate DPDK Ethernet devices from
561 existing port (or VF) representors configured on the device.
563 It is a standard parameter whose format is described in
564 :ref:`ethernet_device_standard_device_arguments`.
566 For instance, to probe port representors 0 through 2::
570 - ``max_dump_files_num`` parameter [int]
572 The maximum number of files per PMD entity that may be created for debug information.
573 The files will be created in /var/log directory or in current directory.
575 set to 128 by default.
577 - ``lro_timeout_usec`` parameter [int]
579 The maximum allowed duration of an LRO session, in micro-seconds.
580 PMD will set the nearest value supported by HW, which is not bigger than
581 the input ``lro_timeout_usec`` value.
582 If this parameter is not specified, by default PMD will set
583 the smallest value supported by HW.
585 Firmware configuration
586 ~~~~~~~~~~~~~~~~~~~~~~
588 Firmware features can be configured as key/value pairs.
590 The command to set a value is::
592 mlxconfig -d <device> set <key>=<value>
594 The command to query a value is::
596 mlxconfig -d <device> query | grep <key>
598 The device name for the command ``mlxconfig`` can be either the PCI address,
599 or the mst device name found with::
603 Below are some firmware configurations listed.
609 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
615 - maximum number of SR-IOV virtual functions::
619 - aggressive CQE zipping::
623 - L3 VXLAN and VXLAN-GPE destination UDP port::
626 IP_OVER_VXLAN_PORT=<udp dport>
628 - enable IP-in-IP tunnel flow matching::
630 FLEX_PARSER_PROFILE_ENABLE=0
632 - enable ICMP/ICMP6 code/type fields matching::
634 FLEX_PARSER_PROFILE_ENABLE=2
639 This driver relies on external libraries and kernel drivers for resources
640 allocations and initialization. The following dependencies are not part of
641 DPDK and must be installed separately:
645 User space Verbs framework used by librte_pmd_mlx5. This library provides
646 a generic interface between the kernel and low-level user space drivers
649 It allows slow and privileged operations (context initialization, hardware
650 resources allocations) to be managed by the kernel and fast operations to
651 never leave user space.
655 Low-level user space driver library for Mellanox
656 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
659 This library basically implements send/receive calls to the hardware
664 They provide the kernel-side Verbs API and low level device drivers that
665 manage actual hardware initialization and resources sharing with user
668 Unlike most other PMDs, these modules must remain loaded and bound to
671 - mlx5_core: hardware driver managing Mellanox
672 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
674 - mlx5_ib: InifiniBand device driver.
675 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
677 - **Firmware update**
679 Mellanox OFED/EN releases include firmware updates for
680 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
682 Because each release provides new features, these updates must be applied to
683 match the kernel modules and libraries they come with.
687 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
693 Either RDMA Core library with a recent enough Linux kernel release
694 (recommended) or Mellanox OFED/EN, which provides compatibility with older
697 RDMA Core with Linux Kernel
698 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
700 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
701 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
702 (see `RDMA Core installation documentation`_)
703 - When building for i686 use:
705 - rdma-core version 18.0 or above built with 32bit support.
706 - Kernel version 4.14.41 or above.
708 - Starting with rdma-core v21, static libraries can be built::
711 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
714 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
715 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
717 If rdma-core libraries are built but not installed, DPDK makefile can link them,
718 thanks to these environment variables:
720 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
721 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
722 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
727 - Mellanox OFED version: ** 4.5, 4.6** /
728 Mellanox EN version: **4.5, 4.6**
731 - ConnectX-4: **12.21.1000** and above.
732 - ConnectX-4 Lx: **14.21.1000** and above.
733 - ConnectX-5: **16.21.1000** and above.
734 - ConnectX-5 Ex: **16.21.1000** and above.
735 - ConnectX-6: **20.99.5374** and above.
736 - BlueField: **18.25.1010** and above.
738 While these libraries and kernel modules are available on OpenFabrics
739 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
740 managers on most distributions, this PMD requires Ethernet extensions that
741 may not be supported at the moment (this is a work in progress).
744 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
746 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
747 include the necessary support and should be used in the meantime. For DPDK,
748 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
749 required from that distribution.
753 Several versions of Mellanox OFED/EN are available. Installing the version
754 this DPDK release was developed and tested against is strongly
755 recommended. Please check the `prerequisites`_.
760 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
761 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
762 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
763 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
764 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
765 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
766 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
767 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
768 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
769 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
770 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
771 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
772 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
773 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
774 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
775 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
776 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
777 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
778 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
780 Quick Start Guide on OFED/EN
781 ----------------------------
783 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
786 2. Install the required libraries and kernel modules either by installing
787 only the required set, or by installing the entire Mellanox OFED/EN::
789 ./mlnxofedinstall --upstream-libs --dpdk
791 3. Verify the firmware is the correct one::
795 4. Verify all ports links are set to Ethernet::
797 mlxconfig -d <mst device> query | grep LINK_TYPE
801 Link types may have to be configured to Ethernet::
803 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
805 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
807 For hypervisors, verify SR-IOV is enabled on the NIC::
809 mlxconfig -d <mst device> query | grep SRIOV_EN
812 If needed, configure SR-IOV::
814 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
815 mlxfwreset -d <mst device> reset
817 5. Restart the driver::
819 /etc/init.d/openibd restart
823 service openibd restart
825 If link type was changed, firmware must be reset as well::
827 mlxfwreset -d <mst device> reset
829 For hypervisors, after reset write the sysfs number of virtual functions
832 To dynamically instantiate a given number of virtual functions (VFs)::
834 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
836 6. Compile DPDK and you are ready to go. See instructions on
837 :ref:`Development Kit Build System <Development_Kit_Build_System>`
839 Enable switchdev mode
840 ---------------------
842 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
843 Representor is a port in DPDK that is connected to a VF in such a way
844 that assuming there are no offload flows, each packet that is sent from the VF
845 will be received by the corresponding representor. While each packet that is
846 sent to a representor will be received by the VF.
847 This is very useful in case of SRIOV mode, where the first packet that is sent
848 by the VF will be received by the DPDK application which will decide if this
849 flow should be offloaded to the E-Switch. After offloading the flow packet
850 that the VF that are matching the flow will not be received any more by
851 the DPDK application.
853 1. Enable SRIOV mode::
855 mlxconfig -d <mst device> set SRIOV_EN=true
857 2. Configure the max number of VFs::
859 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
863 mlxfwreset -d <mst device> reset
865 3. Configure the actual number of VFs::
867 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
869 4. Unbind the device (can be rebind after the switchdev mode)::
871 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
873 5. Enbale switchdev mode::
875 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
880 1. Configure aggressive CQE Zipping for maximum performance::
882 mlxconfig -d <mst device> s CQE_COMPRESSION=1
884 To set it back to the default CQE Zipping mode use::
886 mlxconfig -d <mst device> s CQE_COMPRESSION=0
888 2. In case of virtualization:
890 - Make sure that hypervisor kernel is 3.16 or newer.
891 - Configure boot with ``iommu=pt``.
893 - Make sure to allocate a VM on huge pages.
894 - Make sure to set CPU pinning.
896 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
897 for better performance. For VMs, verify that the right CPU
898 and NUMA node are pinned according to the above. Run::
902 to identify the NUMA node to which the PCIe adapter is connected.
904 4. If more than one adapter is used, and root complex capabilities allow
905 to put both adapters on the same NUMA node without PCI bandwidth degradation,
906 it is recommended to locate both adapters on the same NUMA node.
907 This in order to forward packets from one to the other without
908 NUMA performance penalty.
910 5. Disable pause frames::
912 ethtool -A <netdev> rx off tx off
914 6. Verify IO non-posted prefetch is disabled by default. This can be checked
915 via the BIOS configuration. Please contact you server provider for more
916 information about the settings.
920 On some machines, depends on the machine integrator, it is beneficial
921 to set the PCI max read request parameter to 1K. This can be
922 done in the following way:
924 To query the read request size use::
926 setpci -s <NIC PCI address> 68.w
928 If the output is different than 3XXX, set it by::
930 setpci -s <NIC PCI address> 68.w=3XXX
932 The XXX can be different on different systems. Make sure to configure
933 according to the setpci output.
935 7. To minimize overhead of searching Memory Regions:
937 - '--socket-mem' is recommended to pin memory by predictable amount.
938 - Configure per-lcore cache when creating Mempools for packet buffer.
939 - Refrain from dynamically allocating/freeing memory in run-time.
941 .. _mlx5_offloads_support:
943 Supported hardware offloads using rte_flow API
944 ----------------------------------------------
946 .. table:: Supported hardware offloads using rte_flow API
948 +-----------------------+-----------------+-----------------+
949 | Offload | with E-Switch | with vNIC |
950 +=======================+=================+=================+
951 | Count | | DPDK 19.05 | | DPDK 19.02 |
952 | | | OFED 4.6 | | OFED 4.6 |
953 | | | rdma-core 24 | | rdma-core 23 |
954 | | | ConnectX-5 | | ConnectX-5 |
955 +-----------------------+-----------------+-----------------+
956 | Drop / Queue / RSS | | DPDK 19.05 | | DPDK 18.11 |
957 | | | OFED 4.6 | | OFED 4.5 |
958 | | | rdma-core 24 | | rdma-core 23 |
959 | | | ConnectX-5 | | ConnectX-4 |
960 +-----------------------+-----------------+-----------------+
961 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
962 | (VXLAN / NVGRE / RAW) | | OFED 4.6-2 | | OFED 4.6 |
963 | | | rdma-core 24 | | rdma-core 23 |
964 | | | ConnectX-5 | | ConnectX-5 |
965 +-----------------------+-----------------+-----------------+
966 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
967 | | (set_ipv4_src / | | OFED 4.6-2 | | OFED 4.6-2 |
968 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 23 |
969 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
970 | | set_ipv6_dst / | | |
971 | | set_tp_src / | | |
972 | | set_tp_dst / | | |
975 | | set_mac_src / | | |
976 | | set_mac_dst) | | |
977 +-----------------------+-----------------+-----------------+
978 | Jump | | DPDK 19.05 | | DPDK 19.02 |
979 | | | OFED 4.6-2 | | OFED 4.6-2 |
980 | | | rdma-core 24 | | N/A |
981 | | | ConnectX-5 | | ConnectX-5 |
982 +-----------------------+-----------------+-----------------+
983 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
984 | | | OFED 4.6 | | OFED 4.5 |
985 | | | rdma-core 24 | | rdma-core 23 |
986 | | | ConnectX-5 | | ConnectX-4 |
987 +-----------------------+-----------------+-----------------+
988 | Port ID | | DPDK 19.05 | | N/A |
989 | | | OFED 4.6 | | N/A |
990 | | | rdma-core 24 | | N/A |
991 | | | ConnectX-5 | | N/A |
992 +-----------------------+-----------------+-----------------+
997 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
998 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1000 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1001 command-line parameter to enable additional protocols (UDP and TCP as well
1002 as IP), the following commands must be entered from its CLI to get the same
1003 behavior as librte_pmd_mlx4::
1006 > port config all rss all
1012 This section demonstrates how to launch **testpmd** with Mellanox
1013 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1015 #. Load the kernel modules::
1017 modprobe -a ib_uverbs mlx5_core mlx5_ib
1019 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1022 /etc/init.d/openibd restart
1026 User space I/O kernel modules (uio and igb_uio) are not used and do
1027 not have to be loaded.
1029 #. Make sure Ethernet interfaces are in working order and linked to kernel
1030 verbs. Related sysfs entries should be present::
1032 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1041 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1044 for intf in eth2 eth3 eth4 eth5;
1046 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1049 sed -n 's,.*/\(.*\),-w \1,p'
1058 #. Request huge pages::
1060 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1062 #. Start testpmd with basic parameters::
1064 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
1069 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1070 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1071 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1072 PMD: librte_pmd_mlx5: 1 port(s) detected
1073 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1074 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1075 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1076 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1077 PMD: librte_pmd_mlx5: 1 port(s) detected
1078 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1079 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1080 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1081 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1082 PMD: librte_pmd_mlx5: 1 port(s) detected
1083 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1084 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1085 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1086 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1087 PMD: librte_pmd_mlx5: 1 port(s) detected
1088 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1089 Interactive-mode selected
1090 Configuring Port 0 (socket 0)
1091 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1092 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1093 Port 0: E4:1D:2D:E7:0C:FE
1094 Configuring Port 1 (socket 0)
1095 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1096 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1097 Port 1: E4:1D:2D:E7:0C:FF
1098 Configuring Port 2 (socket 0)
1099 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1100 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1101 Port 2: E4:1D:2D:E7:0C:FA
1102 Configuring Port 3 (socket 0)
1103 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1104 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1105 Port 3: E4:1D:2D:E7:0C:FB
1106 Checking link statuses...
1107 Port 0 Link Up - speed 40000 Mbps - full-duplex
1108 Port 1 Link Up - speed 40000 Mbps - full-duplex
1109 Port 2 Link Up - speed 10000 Mbps - full-duplex
1110 Port 3 Link Up - speed 10000 Mbps - full-duplex