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 by default. It must
24 be enabled manually by setting ``CONFIG_RTE_LIBRTE_MLX5_PMD=y`` and
27 Implementation details
28 ----------------------
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 it 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 Enabling librte_pmd_mlx5 causes DPDK applications to be linked against
53 - Multi arch support: x86_64, POWER8, ARMv8, i686.
54 - Multiple TX and RX queues.
55 - Support for scattered TX and RX frames.
56 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
57 - Several RSS hash keys, one for each flow type.
58 - Default RSS operation with no hash key specification.
59 - Configurable RETA table.
60 - Link flow control (pause frame).
61 - Support for multiple MAC addresses.
65 - RX CRC stripping configuration.
66 - Promiscuous mode on PF and VF.
67 - Multicast promiscuous mode on PF and VF.
68 - Hardware checksum offloads.
69 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
71 - Flow API, including :ref:`flow_isolated_mode`.
73 - KVM and VMware ESX SR-IOV modes are supported.
74 - RSS hash result is supported.
75 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
76 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
78 - Statistics query including Basic, Extended and per queue.
80 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP.
81 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
82 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
83 increment/decrement, count, drop, mark. For details please see :ref:`Supported hardware offloads using rte_flow API`.
84 - Flow insertion rate of more then million flows per second, when using Direct Rules.
85 - Support for multiple rte_flow groups.
91 - For secondary process:
93 - Forked secondary process not supported.
94 - External memory unregistered in EAL memseg list cannot be used for DMA
95 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
96 primary process and remapped to the same virtual address in secondary
97 process. If the external memory is registered by primary process but has
98 different virtual address in secondary process, unexpected error may happen.
100 - Flow pattern without any specific vlan will match for vlan packets as well:
102 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
103 Meaning, the flow rule::
105 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
107 Will only match vlan packets with vid=3. and the flow rules::
109 flow create 0 ingress pattern eth / ipv4 / end ...
113 flow create 0 ingress pattern eth / vlan / ipv4 / end ...
115 Will match any ipv4 packet (VLAN included).
117 - A multi segment packet must have less than 6 segments in case the Tx burst function
118 is set to multi-packet send or Enhanced multi-packet send. Otherwise it must have
119 less than 50 segments.
121 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
122 to 0 are not supported.
124 - VXLAN TSO and checksum offloads are not supported on VM.
126 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
128 - VF: flow rules created on VF devices can only match traffic targeted at the
129 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
133 MAC addresses not already present in the bridge table of the associated
134 kernel network device will be added and cleaned up by the PMD when closing
135 the device. In case of ungraceful program termination, some entries may
136 remain present and should be removed manually by other means.
138 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
139 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
140 ol_flags. As the mempool for the external buffer is managed by PMD, all the
141 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
142 the external buffers will be freed by PMD and the application which still
143 holds the external buffers may be corrupted.
145 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
146 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
147 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
149 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
150 and allmulticast mode are both set to off.
151 To receive IPv6 Multicast messages on VM, explicitly set the relevant
152 MAC address using rte_eth_dev_mac_addr_add() API.
154 - E-Switch decapsulation Flow:
156 - can be applied to PF port only.
157 - must specify VF port action (packet redirection from PF to VF).
158 - optionally may specify tunnel inner source and destination MAC addresses.
160 - E-Switch encapsulation Flow:
162 - can be applied to VF ports only.
163 - must specify PF port action (packet redirection from VF to PF).
165 - ICMP/ICMP6 code/type matching cannot be supported togeter with IP-in-IP tunnel.
169 - KEEP_CRC offload cannot be supported with LRO.
170 - The first mbuf length, without head-room, must be big enough to include the
176 MLX5 supports various of methods to report statistics:
178 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.
180 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.
182 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.
190 These options can be modified in the ``.config`` file.
192 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
194 Toggle compilation of librte_pmd_mlx5 itself.
196 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
198 Build PMD with additional code to make it loadable without hard
199 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
200 on the target system.
202 In this mode, their presence is still required for it to run properly,
203 however their absence won't prevent a DPDK application from starting (with
204 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
205 missing with ``ldd(1)``.
207 It works by moving these dependencies to a purpose-built rdma-core "glue"
208 plug-in which must either be installed in a directory whose name is based
209 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
210 standard location for the dynamic linker (e.g. ``/lib``) if left to the
211 default empty string (``""``).
213 This option has no performance impact.
215 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
217 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
218 in the PMD shared library or the executable static binary.
220 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
222 Toggle debugging code and stricter compilation flags. Enabling this option
223 adds additional run-time checks and debugging messages at the cost of
228 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
229 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
230 64. Default armv8a configuration of make build and meson build set it to 128
231 then brings performance degradation.
233 Environment variables
234 ~~~~~~~~~~~~~~~~~~~~~
238 A list of directories in which to search for the rdma-core "glue" plug-in,
239 separated by colons or semi-colons.
241 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
242 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
243 since ``LD_LIBRARY_PATH`` has no effect in this case.
245 - ``MLX5_SHUT_UP_BF``
247 Configures HW Tx doorbell register as IO-mapped.
249 By default, the HW Tx doorbell is configured as a write-combining register.
250 The register would be flushed to HW usually when the write-combining buffer
251 becomes full, but it depends on CPU design.
253 Except for vectorized Tx burst routines, a write memory barrier is enforced
254 after updating the register so that the update can be immediately visible to
257 When vectorized Tx burst is called, the barrier is set only if the burst size
258 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
259 variable will bring better latency even though the maximum throughput can
262 Run-time configuration
263 ~~~~~~~~~~~~~~~~~~~~~~
265 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
266 because it is affected by their state. Forcing them down prevents packets
269 - **ethtool** operations on related kernel interfaces also affect the PMD.
271 - ``rxq_cqe_comp_en`` parameter [int]
273 A nonzero value enables the compression of CQE on RX side. This feature
274 allows to save PCI bandwidth and improve performance. Enabled by default.
278 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
279 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
281 - ``rxq_cqe_pad_en`` parameter [int]
283 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
284 is aligned with the size of a cacheline of the core. If cacheline size is
285 128B, the CQE size is configured to be 128B even though the device writes
286 only 64B data on the cacheline. This is to avoid unnecessary cache
287 invalidation by device's two consecutive writes on to one cacheline.
288 However in some architecture, it is more beneficial to update entire
289 cacheline with padding the rest 64B rather than striding because
290 read-modify-write could drop performance a lot. On the other hand,
291 writing extra data will consume more PCIe bandwidth and could also drop
292 the maximum throughput. It is recommended to empirically set this
293 parameter. Disabled by default.
297 - CPU having 128B cacheline with ConnectX-5 and BlueField.
299 - ``rxq_pkt_pad_en`` parameter [int]
301 A nonzero value enables padding Rx packet to the size of cacheline on PCI
302 transaction. This feature would waste PCI bandwidth but could improve
303 performance by avoiding partial cacheline write which may cause costly
304 read-modify-copy in memory transaction on some architectures. Disabled by
309 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
310 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
312 - ``mprq_en`` parameter [int]
314 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
315 configured as Multi-Packet RQ if the total number of Rx queues is
316 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
319 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
320 by posting a single large buffer for multiple packets. Instead of posting a
321 buffers per a packet, one large buffer is posted in order to receive multiple
322 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
323 and each stride receives one packet. MPRQ can improve throughput for
324 small-packet traffic.
326 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
327 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
328 configure large stride size enough to accommodate max_rx_pkt_len as long as
329 device allows. Note that this can waste system memory compared to enabling Rx
330 scatter and multi-segment packet.
332 - ``mprq_log_stride_num`` parameter [int]
334 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
335 strides can reduce PCIe traffic further. If configured value is not in the
336 range of device capability, the default value will be set with a warning
337 message. The default value is 4 which is 16 strides per a buffer, valid only
338 if ``mprq_en`` is set.
340 The size of Rx queue should be bigger than the number of strides.
342 - ``mprq_max_memcpy_len`` parameter [int]
344 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
345 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
346 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
347 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
348 A mempool for external buffers will be allocated and managed by PMD. If Rx
349 packet is externally attached, ol_flags field of the mbuf will have
350 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
351 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
353 - ``rxqs_min_mprq`` parameter [int]
355 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
356 greater or equal to this value. The default value is 12, valid only if
359 - ``txq_inline`` parameter [int]
361 Amount of data to be inlined during TX operations. This parameter is
362 deprecated and converted to the new parameter ``txq_inline_max`` providing
363 partial compatibility.
365 - ``txqs_min_inline`` parameter [int]
367 Enable inline data send only when the number of TX queues is greater or equal
370 This option should be used in combination with ``txq_inline_max`` and
371 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
373 If this option is not specified the default value 16 is used for BlueField
374 and 8 for other platforms
376 The data inlining consumes the CPU cycles, so this option is intended to
377 auto enable inline data if we have enough Tx queues, which means we have
378 enough CPU cores and PCI bandwidth is getting more critical and CPU
379 is not supposed to be bottleneck anymore.
381 The copying data into WQE improves latency and can improve PPS performance
382 when PCI back pressure is detected and may be useful for scenarios involving
383 heavy traffic on many queues.
385 Because additional software logic is necessary to handle this mode, this
386 option should be used with care, as it may lower performance when back
387 pressure is not expected.
389 - ``txq_inline_min`` parameter [int]
391 Minimal amount of data to be inlined into WQE during Tx operations. NICs
392 may require this minimal data amount to operate correctly. The exact value
393 may depend on NIC operation mode, requested offloads, etc.
395 If ``txq_inline_min`` key is present the specified value (may be aligned
396 by the driver in order not to exceed the limits and provide better descriptor
397 space utilization) will be used by the driver and it is guaranteed the
398 requested data bytes are inlined into the WQE beside other inline settings.
399 This keys also may update ``txq_inline_max`` value (default of specified
400 explicitly in devargs) to reserve the space for inline data.
402 If ``txq_inline_min`` key is not present, the value may be queried by the
403 driver from the NIC via DevX if this feature is available. If there is no DevX
404 enabled/supported the value 18 (supposing L2 header including VLAN) is set
405 for ConnectX-4, value 58 (supposing L2-L4 headers, required by configurations
406 over E-Switch) is set for ConnectX-4 Lx, and 0 is set by default for ConnectX-5
407 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
410 For the ConnectX-4 and ConnectX-4 Lx NICs driver does not allow to set
411 this value below 18 (minimal L2 header, including VLAN).
413 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
414 Multi-Packet Write), because last one does not support partial packet inlining.
415 This is not very critical due to minimal data inlining is mostly required
416 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
418 - ``txq_inline_max`` parameter [int]
420 Specifies the maximal packet length to be completely inlined into WQE
421 Ethernet Segment for ordinary SEND method. If packet is larger than specified
422 value, the packet data won't be copied by the driver at all, data buffer
423 is addressed with a pointer. If packet length is less or equal all packet
424 data will be copied into WQE. This may improve PCI bandwidth utilization for
425 short packets significantly but requires the extra CPU cycles.
427 The data inline feature is controlled by number of Tx queues, if number of Tx
428 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
429 is engaged, if there are not enough Tx queues (which means not enough CPU cores
430 and CPU resources are scarce), data inline is not performed by the driver.
431 Assigning ``txqs_min_inline`` with zero always enables the data inline.
433 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
434 by the driver in order not to exceed the limit (930 bytes) and to provide better
435 WQE space filling without gaps, the adjustment is reflected in the debug log.
437 - ``txq_inline_mpw`` parameter [int]
439 Specifies the maximal packet length to be completely inlined into WQE for
440 Enhanced MPW method. If packet is large the specified value, the packet data
441 won't be copied, and data buffer is addressed with pointer. If packet length
442 is less or equal, all packet data will be copied into WQE. This may improve PCI
443 bandwidth utilization for short packets significantly but requires the extra
446 The data inline feature is controlled by number of TX queues, if number of Tx
447 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
448 is engaged, if there are not enough Tx queues (which means not enough CPU cores
449 and CPU resources are scarce), data inline is not performed by the driver.
450 Assigning ``txqs_min_inline`` with zero always enables the data inline.
452 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
453 by the driver in order not to exceed the limit (930 bytes) and to provide better
454 WQE space filling without gaps, the adjustment is reflected in the debug log.
455 Due to multiple packets may be included to the same WQE with Enhanced Multi
456 Packet Write Method and overall WQE size is limited it is not recommended to
457 specify large values for the ``txq_inline_mpw``.
459 - ``txqs_max_vec`` parameter [int]
461 Enable vectorized Tx only when the number of TX queues is less than or
462 equal to this value. This parameter is deprecated and ignored, kept
463 for compatibility issue to not prevent driver from probing.
465 - ``txq_mpw_hdr_dseg_en`` parameter [int]
467 A nonzero value enables including two pointers in the first block of TX
468 descriptor. The parameter is deprecated and ignored, kept for compatibility
471 - ``txq_max_inline_len`` parameter [int]
473 Maximum size of packet to be inlined. This limits the size of packet to
474 be inlined. If the size of a packet is larger than configured value, the
475 packet isn't inlined even though there's enough space remained in the
476 descriptor. Instead, the packet is included with pointer. This parameter
477 is deprecated and converted directly to ``txq_inline_mpw`` providing full
478 compatibility. Valid only if eMPW feature is engaged.
480 - ``txq_mpw_en`` parameter [int]
482 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
483 ConnectX-6 and BlueField. eMPW allows the TX burst function to pack up multiple
484 packets in a single descriptor session in order to save PCI bandwidth and improve
485 performance at the cost of a slightly higher CPU usage. When ``txq_inline_mpw``
486 is set along with ``txq_mpw_en``, TX burst function copies entire packet
487 data on to TX descriptor instead of including pointer of packet.
489 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
490 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
491 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
492 option or reported by the NIC, the eMPW feature is disengaged.
494 - ``tx_vec_en`` parameter [int]
496 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6 and BlueField
497 NICs if the number of global Tx queues on the port is less than
498 ``txqs_max_vec``. The parameter is deprecated and ignored.
500 - ``rx_vec_en`` parameter [int]
502 A nonzero value enables Rx vector if the port is not configured in
503 multi-segment otherwise this parameter is ignored.
507 - ``vf_nl_en`` parameter [int]
509 A nonzero value enables Netlink requests from the VF to add/remove MAC
510 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
511 Otherwise the relevant configuration must be run with Linux iproute2 tools.
512 This is a prerequisite to receive this kind of traffic.
514 Enabled by default, valid only on VF devices ignored otherwise.
516 - ``l3_vxlan_en`` parameter [int]
518 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
519 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
520 parameter. This is a prerequisite to receive this kind of traffic.
524 - ``dv_flow_en`` parameter [int]
526 A nonzero value enables the DV flow steering assuming it is supported
531 - ``dv_esw_en`` parameter [int]
533 A nonzero value enables E-Switch using Direct Rules.
535 Enabled by default if supported.
537 - ``mr_ext_memseg_en`` parameter [int]
539 A nonzero value enables extending memseg when registering DMA memory. If
540 enabled, the number of entries in MR (Memory Region) lookup table on datapath
541 is minimized and it benefits performance. On the other hand, it worsens memory
542 utilization because registered memory is pinned by kernel driver. Even if a
543 page in the extended chunk is freed, that doesn't become reusable until the
544 entire memory is freed.
548 - ``representor`` parameter [list]
550 This parameter can be used to instantiate DPDK Ethernet devices from
551 existing port (or VF) representors configured on the device.
553 It is a standard parameter whose format is described in
554 :ref:`ethernet_device_standard_device_arguments`.
556 For instance, to probe port representors 0 through 2::
560 - ``max_dump_files_num`` parameter [int]
562 The maximum number of files per PMD entity that may be created for debug information.
563 The files will be created in /var/log directory or in current directory.
565 set to 128 by default.
567 - ``lro_timeout_usec`` parameter [int]
569 The maximum allowed duration of an LRO session, in micro-seconds.
570 PMD will set the nearest value supported by HW, which is not bigger than
571 the input ``lro_timeout_usec`` value.
572 If this parameter is not specified, by default PMD will set
573 the smallest value supported by HW.
575 Firmware configuration
576 ~~~~~~~~~~~~~~~~~~~~~~
578 Firmware features can be configured as key/value pairs.
580 The command to set a value is::
582 mlxconfig -d <device> set <key>=<value>
584 The command to query a value is::
586 mlxconfig -d <device> query | grep <key>
588 The device name for the command ``mlxconfig`` can be either the PCI address,
589 or the mst device name found with::
593 Below are some firmware configurations listed.
599 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
605 - maximum number of SR-IOV virtual functions::
609 - aggressive CQE zipping::
613 - L3 VXLAN and VXLAN-GPE destination UDP port::
616 IP_OVER_VXLAN_PORT=<udp dport>
618 - enable IP-in-IP tunnel flow matching::
620 FLEX_PARSER_PROFILE_ENABLE=0
622 - enable ICMP/ICMP6 code/type fields matching::
624 FLEX_PARSER_PROFILE_ENABLE=2
629 This driver relies on external libraries and kernel drivers for resources
630 allocations and initialization. The following dependencies are not part of
631 DPDK and must be installed separately:
635 User space Verbs framework used by librte_pmd_mlx5. This library provides
636 a generic interface between the kernel and low-level user space drivers
639 It allows slow and privileged operations (context initialization, hardware
640 resources allocations) to be managed by the kernel and fast operations to
641 never leave user space.
645 Low-level user space driver library for Mellanox
646 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
649 This library basically implements send/receive calls to the hardware
654 They provide the kernel-side Verbs API and low level device drivers that
655 manage actual hardware initialization and resources sharing with user
658 Unlike most other PMDs, these modules must remain loaded and bound to
661 - mlx5_core: hardware driver managing Mellanox
662 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
664 - mlx5_ib: InifiniBand device driver.
665 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
667 - **Firmware update**
669 Mellanox OFED/EN releases include firmware updates for
670 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
672 Because each release provides new features, these updates must be applied to
673 match the kernel modules and libraries they come with.
677 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
683 Either RDMA Core library with a recent enough Linux kernel release
684 (recommended) or Mellanox OFED/EN, which provides compatibility with older
687 RDMA Core with Linux Kernel
688 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
690 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
691 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
692 (see `RDMA Core installation documentation`_)
693 - When building for i686 use:
695 - rdma-core version 18.0 or above built with 32bit support.
696 - Kernel version 4.14.41 or above.
698 - Starting with rdma-core v21, static libraries can be built::
701 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
704 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
705 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
707 If rdma-core libraries are built but not installed, DPDK makefile can link them,
708 thanks to these environment variables:
710 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
711 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
712 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
717 - Mellanox OFED version: ** 4.5, 4.6** /
718 Mellanox EN version: **4.5, 4.6**
721 - ConnectX-4: **12.21.1000** and above.
722 - ConnectX-4 Lx: **14.21.1000** and above.
723 - ConnectX-5: **16.21.1000** and above.
724 - ConnectX-5 Ex: **16.21.1000** and above.
725 - ConnectX-6: **20.99.5374** and above.
726 - BlueField: **18.25.1010** and above.
728 While these libraries and kernel modules are available on OpenFabrics
729 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
730 managers on most distributions, this PMD requires Ethernet extensions that
731 may not be supported at the moment (this is a work in progress).
734 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
736 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
737 include the necessary support and should be used in the meantime. For DPDK,
738 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
739 required from that distribution.
743 Several versions of Mellanox OFED/EN are available. Installing the version
744 this DPDK release was developed and tested against is strongly
745 recommended. Please check the `prerequisites`_.
750 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
751 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
752 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
753 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
754 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
755 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
756 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
757 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
758 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
759 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
760 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
761 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
762 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
763 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
764 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
765 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
766 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
767 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
768 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
770 Quick Start Guide on OFED/EN
771 ----------------------------
773 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
776 2. Install the required libraries and kernel modules either by installing
777 only the required set, or by installing the entire Mellanox OFED/EN:
779 .. code-block:: console
781 ./mlnxofedinstall --upstream-libs --dpdk
783 3. Verify the firmware is the correct one:
785 .. code-block:: console
789 4. Verify all ports links are set to Ethernet:
791 .. code-block:: console
793 mlxconfig -d <mst device> query | grep LINK_TYPE
797 Link types may have to be configured to Ethernet:
799 .. code-block:: console
801 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
803 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
805 For hypervisors verify SR-IOV is enabled on the NIC:
807 .. code-block:: console
809 mlxconfig -d <mst device> query | grep SRIOV_EN
812 If needed, set enable the set the relevant fields:
814 .. code-block:: console
816 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
817 mlxfwreset -d <mst device> reset
819 5. Restart the driver:
821 .. code-block:: console
823 /etc/init.d/openibd restart
827 .. code-block:: console
829 service openibd restart
831 If link type was changed, firmware must be reset as well:
833 .. code-block:: console
835 mlxfwreset -d <mst device> reset
837 For hypervisors, after reset write the sysfs number of virtual functions
840 To dynamically instantiate a given number of virtual functions (VFs):
842 .. code-block:: console
844 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
846 6. Compile DPDK and you are ready to go. See instructions on
847 :ref:`Development Kit Build System <Development_Kit_Build_System>`
849 Enable switchdev mode
850 ---------------------
852 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
853 Representor is a port in DPDK that is connected to a VF in such a way
854 that assuming there are no offload flows, each packet that is sent from the VF
855 will be received by the corresponding representor. While each packet that is
856 sent to a representor will be received by the VF.
857 This is very useful in case of SRIOV mode, where the first packet that is sent
858 by the VF will be received by the DPDK application which will decide if this
859 flow should be offloaded to the E-Switch. After offloading the flow packet
860 that the VF that are matching the flow will not be received any more by
861 the DPDK application.
863 1. Enable SRIOV mode:
865 .. code-block:: console
867 mlxconfig -d <mst device> set SRIOV_EN=true
869 2. Configure the max number of VFs:
871 .. code-block:: console
873 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
877 .. code-block:: console
879 mlxfwreset -d <mst device> reset
881 3. Configure the actual number of VFs:
883 .. code-block:: console
885 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
887 4. Unbind the device (can be rebind after the switchdev mode):
889 .. code-block:: console
891 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
893 5. Enbale switchdev mode:
895 .. code-block:: console
897 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
902 1. Configure aggressive CQE Zipping for maximum performance:
904 .. code-block:: console
906 mlxconfig -d <mst device> s CQE_COMPRESSION=1
908 To set it back to the default CQE Zipping mode use:
910 .. code-block:: console
912 mlxconfig -d <mst device> s CQE_COMPRESSION=0
914 2. In case of virtualization:
916 - Make sure that hypervisor kernel is 3.16 or newer.
917 - Configure boot with ``iommu=pt``.
919 - Make sure to allocate a VM on huge pages.
920 - Make sure to set CPU pinning.
922 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
923 for better performance. For VMs, verify that the right CPU
924 and NUMA node are pinned according to the above. Run:
926 .. code-block:: console
930 to identify the NUMA node to which the PCIe adapter is connected.
932 4. If more than one adapter is used, and root complex capabilities allow
933 to put both adapters on the same NUMA node without PCI bandwidth degradation,
934 it is recommended to locate both adapters on the same NUMA node.
935 This in order to forward packets from one to the other without
936 NUMA performance penalty.
938 5. Disable pause frames:
940 .. code-block:: console
942 ethtool -A <netdev> rx off tx off
944 6. Verify IO non-posted prefetch is disabled by default. This can be checked
945 via the BIOS configuration. Please contact you server provider for more
946 information about the settings.
950 On some machines, depends on the machine integrator, it is beneficial
951 to set the PCI max read request parameter to 1K. This can be
952 done in the following way:
954 To query the read request size use:
956 .. code-block:: console
958 setpci -s <NIC PCI address> 68.w
960 If the output is different than 3XXX, set it by:
962 .. code-block:: console
964 setpci -s <NIC PCI address> 68.w=3XXX
966 The XXX can be different on different systems. Make sure to configure
967 according to the setpci output.
969 7. To minimize overhead of searching Memory Regions:
971 - '--socket-mem' is recommended to pin memory by predictable amount.
972 - Configure per-lcore cache when creating Mempools for packet buffer.
973 - Refrain from dynamically allocating/freeing memory in run-time.
975 Supported hardware offloads using rte_flow API
976 ----------------------------------------------
978 .. _Supported hardware offloads using rte_flow API:
980 .. table:: Supported hardware offloads using rte_flow API
982 +-----------------------+-----------------+-----------------+
983 | Offload | E-Switch | NIC |
985 +=======================+=================+=================+
986 | Count | | DPDK 19.05 | | DPDK 19.02 |
987 | | | OFED 4.6 | | OFED 4.6 |
988 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
989 | | | ConnectX-5 | | ConnectX-5 |
990 +-----------------------+-----------------+-----------------+
991 | Drop / Queue / RSS | | DPDK 19.05 | | DPDK 18.11 |
992 | | | OFED 4.6 | | OFED 4.5 |
993 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
994 | | | ConnectX-5 | | ConnectX-4 |
995 +-----------------------+-----------------+-----------------+
996 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
997 | (VXLAN / NVGRE / RAW) | | OFED 4.6.2 | | OFED 4.6 |
998 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
999 | | | ConnectX-5 | | ConnectX-5 |
1000 +-----------------------+-----------------+-----------------+
1001 | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1002 | (set_ipv4_src / | | OFED 4.6.2 | | OFED 4.6.2 |
1003 | set_ipv4_dst / | | RDMA-CORE V24 | | RDMA-CORE V23 |
1004 | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1005 | set_ipv6_dst / | | |
1006 | set_tp_src / | | |
1007 | set_tp_dst / | | |
1010 | set_mac_src / | | |
1011 | set_mac_dst) | | |
1012 +-----------------------+-----------------+-----------------+
1013 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1014 | | | OFED 4.6.2 | | OFED 4.6.2 |
1015 | | | RDMA-CORE V24 | | N/A |
1016 | | | ConnectX-5 | | ConnectX-5 |
1017 +-----------------------+-----------------+-----------------+
1018 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1019 | | | OFED 4.6 | | OFED 4.5 |
1020 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
1021 | | | ConnectX-5 | | ConnectX-4 |
1022 +-----------------------+-----------------+-----------------+
1023 | Port ID | | DPDK 19.05 | | N/A |
1024 | | | OFED 4.6 | | N/A |
1025 | | | RDMA-CORE V24 | | N/A |
1026 | | | ConnectX-5 | | N/A |
1027 +-----------------------+-----------------+-----------------+
1029 * Minimum version for each component and nic.
1034 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1035 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1037 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1038 command-line parameter to enable additional protocols (UDP and TCP as well
1039 as IP), the following commands must be entered from its CLI to get the same
1040 behavior as librte_pmd_mlx4:
1042 .. code-block:: console
1045 > port config all rss all
1051 This section demonstrates how to launch **testpmd** with Mellanox
1052 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1054 #. Load the kernel modules:
1056 .. code-block:: console
1058 modprobe -a ib_uverbs mlx5_core mlx5_ib
1060 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1063 .. code-block:: console
1065 /etc/init.d/openibd restart
1069 User space I/O kernel modules (uio and igb_uio) are not used and do
1070 not have to be loaded.
1072 #. Make sure Ethernet interfaces are in working order and linked to kernel
1073 verbs. Related sysfs entries should be present:
1075 .. code-block:: console
1077 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1081 .. code-block:: console
1088 #. Optionally, retrieve their PCI bus addresses for whitelisting:
1090 .. code-block:: console
1093 for intf in eth2 eth3 eth4 eth5;
1095 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1098 sed -n 's,.*/\(.*\),-w \1,p'
1102 .. code-block:: console
1109 #. Request huge pages:
1111 .. code-block:: console
1113 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1115 #. Start testpmd with basic parameters:
1117 .. code-block:: console
1119 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
1123 .. code-block:: console
1126 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1127 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1128 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1129 PMD: librte_pmd_mlx5: 1 port(s) detected
1130 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1131 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1132 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1133 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1134 PMD: librte_pmd_mlx5: 1 port(s) detected
1135 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1136 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1137 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1138 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1139 PMD: librte_pmd_mlx5: 1 port(s) detected
1140 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1141 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1142 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1143 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1144 PMD: librte_pmd_mlx5: 1 port(s) detected
1145 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1146 Interactive-mode selected
1147 Configuring Port 0 (socket 0)
1148 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1149 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1150 Port 0: E4:1D:2D:E7:0C:FE
1151 Configuring Port 1 (socket 0)
1152 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1153 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1154 Port 1: E4:1D:2D:E7:0C:FF
1155 Configuring Port 2 (socket 0)
1156 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1157 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1158 Port 2: E4:1D:2D:E7:0C:FA
1159 Configuring Port 3 (socket 0)
1160 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1161 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1162 Port 3: E4:1D:2D:E7:0C:FB
1163 Checking link statuses...
1164 Port 0 Link Up - speed 40000 Mbps - full-duplex
1165 Port 1 Link Up - speed 40000 Mbps - full-duplex
1166 Port 2 Link Up - speed 10000 Mbps - full-duplex
1167 Port 3 Link Up - speed 10000 Mbps - full-duplex