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 - VLAN pop offload command:
129 - Flow rules having a VLAN pop offload command as one of their actions and
130 are lacking a match on VLAN as one of their items are not supported.
131 - The command is not supported on egress traffic.
133 - VLAN push offload is not supported on ingress traffic.
135 - A multi segment packet must have not more segments than reported by dev_infos_get()
136 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
137 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
138 inline settings) to 58.
140 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
141 to 0 are not supported.
143 - VXLAN TSO and checksum offloads are not supported on VM.
145 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
147 - VF: flow rules created on VF devices can only match traffic targeted at the
148 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
152 MAC addresses not already present in the bridge table of the associated
153 kernel network device will be added and cleaned up by the PMD when closing
154 the device. In case of ungraceful program termination, some entries may
155 remain present and should be removed manually by other means.
157 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
158 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
159 ol_flags. As the mempool for the external buffer is managed by PMD, all the
160 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
161 the external buffers will be freed by PMD and the application which still
162 holds the external buffers may be corrupted.
164 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
165 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
166 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
168 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
169 and allmulticast mode are both set to off.
170 To receive IPv6 Multicast messages on VM, explicitly set the relevant
171 MAC address using rte_eth_dev_mac_addr_add() API.
173 - The amount of descriptors in Tx queue may be limited by data inline settings.
174 Inline data require the more descriptor building blocks and overall block
175 amount may exceed the hardware supported limits. The application should
176 reduce the requested Tx size or adjust data inline settings with
177 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
179 - E-Switch decapsulation Flow:
181 - can be applied to PF port only.
182 - must specify VF port action (packet redirection from PF to VF).
183 - optionally may specify tunnel inner source and destination MAC addresses.
185 - E-Switch encapsulation Flow:
187 - can be applied to VF ports only.
188 - must specify PF port action (packet redirection from VF to PF).
190 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
191 mutually exclusive features which cannot be supported together
192 (see :ref:`mlx5_firmware_config`).
196 - Requires DevX to be enabled.
197 - KEEP_CRC offload cannot be supported with LRO.
198 - The first mbuf length, without head-room, must be big enough to include the
204 MLX5 supports various methods to report statistics:
206 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.
208 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.
210 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.
218 These options can be modified in the ``.config`` file.
220 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
222 Toggle compilation of librte_pmd_mlx5 itself.
224 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
226 Build PMD with additional code to make it loadable without hard
227 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
228 on the target system.
230 In this mode, their presence is still required for it to run properly,
231 however their absence won't prevent a DPDK application from starting (with
232 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
233 missing with ``ldd(1)``.
235 It works by moving these dependencies to a purpose-built rdma-core "glue"
236 plug-in which must either be installed in a directory whose name is based
237 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
238 standard location for the dynamic linker (e.g. ``/lib``) if left to the
239 default empty string (``""``).
241 This option has no performance impact.
243 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
245 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
246 in the PMD shared library or the executable static binary.
248 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
250 Toggle debugging code and stricter compilation flags. Enabling this option
251 adds additional run-time checks and debugging messages at the cost of
256 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
257 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
258 64. Default armv8a configuration of make build and meson build set it to 128
259 then brings performance degradation.
261 Environment variables
262 ~~~~~~~~~~~~~~~~~~~~~
266 A list of directories in which to search for the rdma-core "glue" plug-in,
267 separated by colons or semi-colons.
269 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
270 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
271 since ``LD_LIBRARY_PATH`` has no effect in this case.
273 - ``MLX5_SHUT_UP_BF``
275 Configures HW Tx doorbell register as IO-mapped.
277 By default, the HW Tx doorbell is configured as a write-combining register.
278 The register would be flushed to HW usually when the write-combining buffer
279 becomes full, but it depends on CPU design.
281 Except for vectorized Tx burst routines, a write memory barrier is enforced
282 after updating the register so that the update can be immediately visible to
285 When vectorized Tx burst is called, the barrier is set only if the burst size
286 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
287 variable will bring better latency even though the maximum throughput can
290 Run-time configuration
291 ~~~~~~~~~~~~~~~~~~~~~~
293 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
294 because it is affected by their state. Forcing them down prevents packets
297 - **ethtool** operations on related kernel interfaces also affect the PMD.
299 - ``rxq_cqe_comp_en`` parameter [int]
301 A nonzero value enables the compression of CQE on RX side. This feature
302 allows to save PCI bandwidth and improve performance. Enabled by default.
306 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
307 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
309 - ``rxq_cqe_pad_en`` parameter [int]
311 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
312 is aligned with the size of a cacheline of the core. If cacheline size is
313 128B, the CQE size is configured to be 128B even though the device writes
314 only 64B data on the cacheline. This is to avoid unnecessary cache
315 invalidation by device's two consecutive writes on to one cacheline.
316 However in some architecture, it is more beneficial to update entire
317 cacheline with padding the rest 64B rather than striding because
318 read-modify-write could drop performance a lot. On the other hand,
319 writing extra data will consume more PCIe bandwidth and could also drop
320 the maximum throughput. It is recommended to empirically set this
321 parameter. Disabled by default.
325 - CPU having 128B cacheline with ConnectX-5 and BlueField.
327 - ``rxq_pkt_pad_en`` parameter [int]
329 A nonzero value enables padding Rx packet to the size of cacheline on PCI
330 transaction. This feature would waste PCI bandwidth but could improve
331 performance by avoiding partial cacheline write which may cause costly
332 read-modify-copy in memory transaction on some architectures. Disabled by
337 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
338 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
340 - ``mprq_en`` parameter [int]
342 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
343 configured as Multi-Packet RQ if the total number of Rx queues is
344 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
347 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
348 by posting a single large buffer for multiple packets. Instead of posting a
349 buffers per a packet, one large buffer is posted in order to receive multiple
350 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
351 and each stride receives one packet. MPRQ can improve throughput for
352 small-packet traffic.
354 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
355 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
356 configure large stride size enough to accommodate max_rx_pkt_len as long as
357 device allows. Note that this can waste system memory compared to enabling Rx
358 scatter and multi-segment packet.
360 - ``mprq_log_stride_num`` parameter [int]
362 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
363 strides can reduce PCIe traffic further. If configured value is not in the
364 range of device capability, the default value will be set with a warning
365 message. The default value is 4 which is 16 strides per a buffer, valid only
366 if ``mprq_en`` is set.
368 The size of Rx queue should be bigger than the number of strides.
370 - ``mprq_max_memcpy_len`` parameter [int]
372 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
373 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
374 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
375 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
376 A mempool for external buffers will be allocated and managed by PMD. If Rx
377 packet is externally attached, ol_flags field of the mbuf will have
378 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
379 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
381 - ``rxqs_min_mprq`` parameter [int]
383 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
384 greater or equal to this value. The default value is 12, valid only if
387 - ``txq_inline`` parameter [int]
389 Amount of data to be inlined during TX operations. This parameter is
390 deprecated and converted to the new parameter ``txq_inline_max`` providing
391 partial compatibility.
393 - ``txqs_min_inline`` parameter [int]
395 Enable inline data send only when the number of TX queues is greater or equal
398 This option should be used in combination with ``txq_inline_max`` and
399 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
401 If this option is not specified the default value 16 is used for BlueField
402 and 8 for other platforms
404 The data inlining consumes the CPU cycles, so this option is intended to
405 auto enable inline data if we have enough Tx queues, which means we have
406 enough CPU cores and PCI bandwidth is getting more critical and CPU
407 is not supposed to be bottleneck anymore.
409 The copying data into WQE improves latency and can improve PPS performance
410 when PCI back pressure is detected and may be useful for scenarios involving
411 heavy traffic on many queues.
413 Because additional software logic is necessary to handle this mode, this
414 option should be used with care, as it may lower performance when back
415 pressure is not expected.
417 If inline data are enabled it may affect the maximal size of Tx queue in
418 descriptors because the inline data increase the descriptor size and
419 queue size limits supported by hardware may be exceeded.
421 - ``txq_inline_min`` parameter [int]
423 Minimal amount of data to be inlined into WQE during Tx operations. NICs
424 may require this minimal data amount to operate correctly. The exact value
425 may depend on NIC operation mode, requested offloads, etc. It is strongly
426 recommended to omit this parameter and use the default values. Anyway,
427 applications using this parameter should take into consideration that
428 specifying an inconsistent value may prevent the NIC from sending packets.
430 If ``txq_inline_min`` key is present the specified value (may be aligned
431 by the driver in order not to exceed the limits and provide better descriptor
432 space utilization) will be used by the driver and it is guaranteed that
433 requested amount of data bytes are inlined into the WQE beside other inline
434 settings. This key also may update ``txq_inline_max`` value (default
435 or specified explicitly in devargs) to reserve the space for inline data.
437 If ``txq_inline_min`` key is not present, the value may be queried by the
438 driver from the NIC via DevX if this feature is available. If there is no DevX
439 enabled/supported the value 18 (supposing L2 header including VLAN) is set
440 for ConnectX-4 and ConnectX-4LX, and 0 is set by default for ConnectX-5
441 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
444 For ConnectX-4 NIC, driver does not allow specifying value below 18
445 (minimal L2 header, including VLAN), error will be raised.
447 For ConnectX-4LX NIC, it is allowed to specify values below 18, but
448 it is not recommended and may prevent NIC from sending packets over
451 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
452 Multi-Packet Write), because last one does not support partial packet inlining.
453 This is not very critical due to minimal data inlining is mostly required
454 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
456 - ``txq_inline_max`` parameter [int]
458 Specifies the maximal packet length to be completely inlined into WQE
459 Ethernet Segment for ordinary SEND method. If packet is larger than specified
460 value, the packet data won't be copied by the driver at all, data buffer
461 is addressed with a pointer. If packet length is less or equal all packet
462 data will be copied into WQE. This may improve PCI bandwidth utilization for
463 short packets significantly but requires the extra CPU cycles.
465 The data inline feature is controlled by number of Tx queues, if number of Tx
466 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
467 is engaged, if there are not enough Tx queues (which means not enough CPU cores
468 and CPU resources are scarce), data inline is not performed by the driver.
469 Assigning ``txqs_min_inline`` with zero always enables the data inline.
471 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
472 by the driver in order not to exceed the limit (930 bytes) and to provide better
473 WQE space filling without gaps, the adjustment is reflected in the debug log.
475 - ``txq_inline_mpw`` parameter [int]
477 Specifies the maximal packet length to be completely inlined into WQE for
478 Enhanced MPW method. If packet is large the specified value, the packet data
479 won't be copied, and data buffer is addressed with pointer. If packet length
480 is less or equal, all packet data will be copied into WQE. This may improve PCI
481 bandwidth utilization for short packets significantly but requires the extra
484 The data inline feature is controlled by number of TX queues, if number of Tx
485 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
486 is engaged, if there are not enough Tx queues (which means not enough CPU cores
487 and CPU resources are scarce), data inline is not performed by the driver.
488 Assigning ``txqs_min_inline`` with zero always enables the data inline.
490 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
491 by the driver in order not to exceed the limit (930 bytes) and to provide better
492 WQE space filling without gaps, the adjustment is reflected in the debug log.
493 Due to multiple packets may be included to the same WQE with Enhanced Multi
494 Packet Write Method and overall WQE size is limited it is not recommended to
495 specify large values for the ``txq_inline_mpw``.
497 - ``txqs_max_vec`` parameter [int]
499 Enable vectorized Tx only when the number of TX queues is less than or
500 equal to this value. This parameter is deprecated and ignored, kept
501 for compatibility issue to not prevent driver from probing.
503 - ``txq_mpw_hdr_dseg_en`` parameter [int]
505 A nonzero value enables including two pointers in the first block of TX
506 descriptor. The parameter is deprecated and ignored, kept for compatibility
509 - ``txq_max_inline_len`` parameter [int]
511 Maximum size of packet to be inlined. This limits the size of packet to
512 be inlined. If the size of a packet is larger than configured value, the
513 packet isn't inlined even though there's enough space remained in the
514 descriptor. Instead, the packet is included with pointer. This parameter
515 is deprecated and converted directly to ``txq_inline_mpw`` providing full
516 compatibility. Valid only if eMPW feature is engaged.
518 - ``txq_mpw_en`` parameter [int]
520 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
521 ConnectX-6 and BlueField. eMPW allows the TX burst function to pack up multiple
522 packets in a single descriptor session in order to save PCI bandwidth and improve
523 performance at the cost of a slightly higher CPU usage. When ``txq_inline_mpw``
524 is set along with ``txq_mpw_en``, TX burst function copies entire packet
525 data on to TX descriptor instead of including pointer of packet.
527 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
528 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
529 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
530 option or reported by the NIC, the eMPW feature is disengaged.
532 - ``tx_vec_en`` parameter [int]
534 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6 and BlueField
535 NICs if the number of global Tx queues on the port is less than
536 ``txqs_max_vec``. The parameter is deprecated and ignored.
538 - ``rx_vec_en`` parameter [int]
540 A nonzero value enables Rx vector if the port is not configured in
541 multi-segment otherwise this parameter is ignored.
545 - ``vf_nl_en`` parameter [int]
547 A nonzero value enables Netlink requests from the VF to add/remove MAC
548 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
549 Otherwise the relevant configuration must be run with Linux iproute2 tools.
550 This is a prerequisite to receive this kind of traffic.
552 Enabled by default, valid only on VF devices ignored otherwise.
554 - ``l3_vxlan_en`` parameter [int]
556 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
557 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
558 parameter. This is a prerequisite to receive this kind of traffic.
562 - ``dv_flow_en`` parameter [int]
564 A nonzero value enables the DV flow steering assuming it is supported
569 - ``dv_esw_en`` parameter [int]
571 A nonzero value enables E-Switch using Direct Rules.
573 Enabled by default if supported.
575 - ``mr_ext_memseg_en`` parameter [int]
577 A nonzero value enables extending memseg when registering DMA memory. If
578 enabled, the number of entries in MR (Memory Region) lookup table on datapath
579 is minimized and it benefits performance. On the other hand, it worsens memory
580 utilization because registered memory is pinned by kernel driver. Even if a
581 page in the extended chunk is freed, that doesn't become reusable until the
582 entire memory is freed.
586 - ``representor`` parameter [list]
588 This parameter can be used to instantiate DPDK Ethernet devices from
589 existing port (or VF) representors configured on the device.
591 It is a standard parameter whose format is described in
592 :ref:`ethernet_device_standard_device_arguments`.
594 For instance, to probe port representors 0 through 2::
598 - ``max_dump_files_num`` parameter [int]
600 The maximum number of files per PMD entity that may be created for debug information.
601 The files will be created in /var/log directory or in current directory.
603 set to 128 by default.
605 - ``lro_timeout_usec`` parameter [int]
607 The maximum allowed duration of an LRO session, in micro-seconds.
608 PMD will set the nearest value supported by HW, which is not bigger than
609 the input ``lro_timeout_usec`` value.
610 If this parameter is not specified, by default PMD will set
611 the smallest value supported by HW.
613 .. _mlx5_firmware_config:
615 Firmware configuration
616 ~~~~~~~~~~~~~~~~~~~~~~
618 Firmware features can be configured as key/value pairs.
620 The command to set a value is::
622 mlxconfig -d <device> set <key>=<value>
624 The command to query a value is::
626 mlxconfig -d <device> query | grep <key>
628 The device name for the command ``mlxconfig`` can be either the PCI address,
629 or the mst device name found with::
633 Below are some firmware configurations listed.
639 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
645 - maximum number of SR-IOV virtual functions::
649 - enable DevX (required by Direct Rules and other features)::
653 - aggressive CQE zipping::
657 - L3 VXLAN and VXLAN-GPE destination UDP port::
660 IP_OVER_VXLAN_PORT=<udp dport>
662 - enable IP-in-IP tunnel flow matching::
664 FLEX_PARSER_PROFILE_ENABLE=0
666 - enable MPLS flow matching::
668 FLEX_PARSER_PROFILE_ENABLE=1
670 - enable ICMP/ICMP6 code/type fields matching::
672 FLEX_PARSER_PROFILE_ENABLE=2
677 This driver relies on external libraries and kernel drivers for resources
678 allocations and initialization. The following dependencies are not part of
679 DPDK and must be installed separately:
683 User space Verbs framework used by librte_pmd_mlx5. This library provides
684 a generic interface between the kernel and low-level user space drivers
687 It allows slow and privileged operations (context initialization, hardware
688 resources allocations) to be managed by the kernel and fast operations to
689 never leave user space.
693 Low-level user space driver library for Mellanox
694 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
697 This library basically implements send/receive calls to the hardware
702 They provide the kernel-side Verbs API and low level device drivers that
703 manage actual hardware initialization and resources sharing with user
706 Unlike most other PMDs, these modules must remain loaded and bound to
709 - mlx5_core: hardware driver managing Mellanox
710 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
712 - mlx5_ib: InifiniBand device driver.
713 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
715 - **Firmware update**
717 Mellanox OFED/EN releases include firmware updates for
718 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
720 Because each release provides new features, these updates must be applied to
721 match the kernel modules and libraries they come with.
725 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
731 Either RDMA Core library with a recent enough Linux kernel release
732 (recommended) or Mellanox OFED/EN, which provides compatibility with older
735 RDMA Core with Linux Kernel
736 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
738 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
739 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
740 (see `RDMA Core installation documentation`_)
741 - When building for i686 use:
743 - rdma-core version 18.0 or above built with 32bit support.
744 - Kernel version 4.14.41 or above.
746 - Starting with rdma-core v21, static libraries can be built::
749 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
752 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
753 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
755 If rdma-core libraries are built but not installed, DPDK makefile can link them,
756 thanks to these environment variables:
758 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
759 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
760 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
765 - Mellanox OFED version: ** 4.5, 4.6** /
766 Mellanox EN version: **4.5, 4.6**
769 - ConnectX-4: **12.21.1000** and above.
770 - ConnectX-4 Lx: **14.21.1000** and above.
771 - ConnectX-5: **16.21.1000** and above.
772 - ConnectX-5 Ex: **16.21.1000** and above.
773 - ConnectX-6: **20.99.5374** and above.
774 - BlueField: **18.25.1010** and above.
776 While these libraries and kernel modules are available on OpenFabrics
777 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
778 managers on most distributions, this PMD requires Ethernet extensions that
779 may not be supported at the moment (this is a work in progress).
782 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
784 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
785 include the necessary support and should be used in the meantime. For DPDK,
786 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
787 required from that distribution.
791 Several versions of Mellanox OFED/EN are available. Installing the version
792 this DPDK release was developed and tested against is strongly
793 recommended. Please check the `prerequisites`_.
798 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
799 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
800 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
801 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
802 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
803 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
804 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
805 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
806 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
807 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
808 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
809 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
810 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
811 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
812 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
813 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
814 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
815 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
816 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
818 Quick Start Guide on OFED/EN
819 ----------------------------
821 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
824 2. Install the required libraries and kernel modules either by installing
825 only the required set, or by installing the entire Mellanox OFED/EN::
827 ./mlnxofedinstall --upstream-libs --dpdk
829 3. Verify the firmware is the correct one::
833 4. Verify all ports links are set to Ethernet::
835 mlxconfig -d <mst device> query | grep LINK_TYPE
839 Link types may have to be configured to Ethernet::
841 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
843 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
845 For hypervisors, verify SR-IOV is enabled on the NIC::
847 mlxconfig -d <mst device> query | grep SRIOV_EN
850 If needed, configure SR-IOV::
852 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
853 mlxfwreset -d <mst device> reset
855 5. Restart the driver::
857 /etc/init.d/openibd restart
861 service openibd restart
863 If link type was changed, firmware must be reset as well::
865 mlxfwreset -d <mst device> reset
867 For hypervisors, after reset write the sysfs number of virtual functions
870 To dynamically instantiate a given number of virtual functions (VFs)::
872 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
874 6. Compile DPDK and you are ready to go. See instructions on
875 :ref:`Development Kit Build System <Development_Kit_Build_System>`
877 Enable switchdev mode
878 ---------------------
880 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
881 Representor is a port in DPDK that is connected to a VF in such a way
882 that assuming there are no offload flows, each packet that is sent from the VF
883 will be received by the corresponding representor. While each packet that is
884 sent to a representor will be received by the VF.
885 This is very useful in case of SRIOV mode, where the first packet that is sent
886 by the VF will be received by the DPDK application which will decide if this
887 flow should be offloaded to the E-Switch. After offloading the flow packet
888 that the VF that are matching the flow will not be received any more by
889 the DPDK application.
891 1. Enable SRIOV mode::
893 mlxconfig -d <mst device> set SRIOV_EN=true
895 2. Configure the max number of VFs::
897 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
901 mlxfwreset -d <mst device> reset
903 3. Configure the actual number of VFs::
905 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
907 4. Unbind the device (can be rebind after the switchdev mode)::
909 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
911 5. Enbale switchdev mode::
913 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
918 1. Configure aggressive CQE Zipping for maximum performance::
920 mlxconfig -d <mst device> s CQE_COMPRESSION=1
922 To set it back to the default CQE Zipping mode use::
924 mlxconfig -d <mst device> s CQE_COMPRESSION=0
926 2. In case of virtualization:
928 - Make sure that hypervisor kernel is 3.16 or newer.
929 - Configure boot with ``iommu=pt``.
931 - Make sure to allocate a VM on huge pages.
932 - Make sure to set CPU pinning.
934 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
935 for better performance. For VMs, verify that the right CPU
936 and NUMA node are pinned according to the above. Run::
940 to identify the NUMA node to which the PCIe adapter is connected.
942 4. If more than one adapter is used, and root complex capabilities allow
943 to put both adapters on the same NUMA node without PCI bandwidth degradation,
944 it is recommended to locate both adapters on the same NUMA node.
945 This in order to forward packets from one to the other without
946 NUMA performance penalty.
948 5. Disable pause frames::
950 ethtool -A <netdev> rx off tx off
952 6. Verify IO non-posted prefetch is disabled by default. This can be checked
953 via the BIOS configuration. Please contact you server provider for more
954 information about the settings.
958 On some machines, depends on the machine integrator, it is beneficial
959 to set the PCI max read request parameter to 1K. This can be
960 done in the following way:
962 To query the read request size use::
964 setpci -s <NIC PCI address> 68.w
966 If the output is different than 3XXX, set it by::
968 setpci -s <NIC PCI address> 68.w=3XXX
970 The XXX can be different on different systems. Make sure to configure
971 according to the setpci output.
973 7. To minimize overhead of searching Memory Regions:
975 - '--socket-mem' is recommended to pin memory by predictable amount.
976 - Configure per-lcore cache when creating Mempools for packet buffer.
977 - Refrain from dynamically allocating/freeing memory in run-time.
979 .. _mlx5_offloads_support:
981 Supported hardware offloads
982 ---------------------------
984 .. table:: Minimal SW/HW versions for queue offloads
986 ============== ===== ===== ========= ===== ========== ==========
987 Offload DPDK Linux rdma-core OFED firmware hardware
988 ============== ===== ===== ========= ===== ========== ==========
989 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
990 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
991 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
992 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
993 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
994 ============== ===== ===== ========= ===== ========== ==========
996 .. table:: Minimal SW/HW versions for rte_flow offloads
998 +-----------------------+-----------------+-----------------+
999 | Offload | with E-Switch | with vNIC |
1000 +=======================+=================+=================+
1001 | Count | | DPDK 19.05 | | DPDK 19.02 |
1002 | | | OFED 4.6 | | OFED 4.6 |
1003 | | | rdma-core 24 | | rdma-core 23 |
1004 | | | ConnectX-5 | | ConnectX-5 |
1005 +-----------------------+-----------------+-----------------+
1006 | Drop / Queue / RSS | | DPDK 19.05 | | DPDK 18.11 |
1007 | | | OFED 4.6 | | OFED 4.5 |
1008 | | | rdma-core 24 | | rdma-core 23 |
1009 | | | ConnectX-5 | | ConnectX-4 |
1010 +-----------------------+-----------------+-----------------+
1011 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1012 | (VXLAN / NVGRE / RAW) | | OFED 4.6-2 | | OFED 4.6 |
1013 | | | rdma-core 24 | | rdma-core 23 |
1014 | | | ConnectX-5 | | ConnectX-5 |
1015 +-----------------------+-----------------+-----------------+
1016 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1017 | | (set_ipv4_src / | | OFED 4.6-2 | | OFED 4.6-2 |
1018 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 23 |
1019 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1020 | | set_ipv6_dst / | | |
1021 | | set_tp_src / | | |
1022 | | set_tp_dst / | | |
1025 | | set_mac_src / | | |
1026 | | set_mac_dst) | | |
1027 +-----------------------+-----------------+-----------------+
1028 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1029 | | | OFED 4.6-4 | | OFED 4.6-4 |
1030 | | | rdma-core 24 | | N/A |
1031 | | | ConnectX-5 | | ConnectX-5 |
1032 +-----------------------+-----------------+-----------------+
1033 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1034 | | | OFED 4.6 | | OFED 4.5 |
1035 | | | rdma-core 24 | | rdma-core 23 |
1036 | | | ConnectX-5 | | ConnectX-4 |
1037 +-----------------------+-----------------+-----------------+
1038 | Port ID | | DPDK 19.05 | | N/A |
1039 | | | OFED 4.6 | | N/A |
1040 | | | rdma-core 24 | | N/A |
1041 | | | ConnectX-5 | | N/A |
1042 +-----------------------+-----------------+-----------------+
1043 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1044 | | (of_pop_vlan / | | OFED 4.6-4 | | OFED 4.6-4 |
1045 | | of_push_vlan) | | ConnectX-5 | | ConnectX-5 |
1046 +-----------------------+-----------------+-----------------+
1051 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1052 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1054 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1055 command-line parameter to enable additional protocols (UDP and TCP as well
1056 as IP), the following commands must be entered from its CLI to get the same
1057 behavior as librte_pmd_mlx4::
1060 > port config all rss all
1066 This section demonstrates how to launch **testpmd** with Mellanox
1067 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1069 #. Load the kernel modules::
1071 modprobe -a ib_uverbs mlx5_core mlx5_ib
1073 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1076 /etc/init.d/openibd restart
1080 User space I/O kernel modules (uio and igb_uio) are not used and do
1081 not have to be loaded.
1083 #. Make sure Ethernet interfaces are in working order and linked to kernel
1084 verbs. Related sysfs entries should be present::
1086 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1095 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1098 for intf in eth2 eth3 eth4 eth5;
1100 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1103 sed -n 's,.*/\(.*\),-w \1,p'
1112 #. Request huge pages::
1114 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1116 #. Start testpmd with basic parameters::
1118 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 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1124 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1125 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1126 PMD: librte_pmd_mlx5: 1 port(s) detected
1127 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1128 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1129 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1130 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1131 PMD: librte_pmd_mlx5: 1 port(s) detected
1132 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1133 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1134 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1135 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1136 PMD: librte_pmd_mlx5: 1 port(s) detected
1137 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1138 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1139 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1140 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1141 PMD: librte_pmd_mlx5: 1 port(s) detected
1142 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1143 Interactive-mode selected
1144 Configuring Port 0 (socket 0)
1145 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1146 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1147 Port 0: E4:1D:2D:E7:0C:FE
1148 Configuring Port 1 (socket 0)
1149 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1150 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1151 Port 1: E4:1D:2D:E7:0C:FF
1152 Configuring Port 2 (socket 0)
1153 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1154 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1155 Port 2: E4:1D:2D:E7:0C:FA
1156 Configuring Port 3 (socket 0)
1157 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1158 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1159 Port 3: E4:1D:2D:E7:0C:FB
1160 Checking link statuses...
1161 Port 0 Link Up - speed 40000 Mbps - full-duplex
1162 Port 1 Link Up - speed 40000 Mbps - full-duplex
1163 Port 2 Link Up - speed 10000 Mbps - full-duplex
1164 Port 3 Link Up - speed 10000 Mbps - full-duplex