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 - VLAN set PCP offload is not supported on existing headers.
137 - VLAN set VID offload is not supported on existing headers.
139 - A multi segment packet must have not more segments than reported by dev_infos_get()
140 in tx_desc_lim.nb_seg_max field. This value depends on maximal supported Tx descriptor
141 size and ``txq_inline_min`` settings and may be from 2 (worst case forced by maximal
142 inline settings) to 58.
144 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
145 to 0 are not supported.
147 - VXLAN TSO and checksum offloads are not supported on VM.
149 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
151 - VF: flow rules created on VF devices can only match traffic targeted at the
152 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
156 MAC addresses not already present in the bridge table of the associated
157 kernel network device will be added and cleaned up by the PMD when closing
158 the device. In case of ungraceful program termination, some entries may
159 remain present and should be removed manually by other means.
161 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
162 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
163 ol_flags. As the mempool for the external buffer is managed by PMD, all the
164 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
165 the external buffers will be freed by PMD and the application which still
166 holds the external buffers may be corrupted.
168 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
169 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
170 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
172 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
173 and allmulticast mode are both set to off.
174 To receive IPv6 Multicast messages on VM, explicitly set the relevant
175 MAC address using rte_eth_dev_mac_addr_add() API.
177 - The amount of descriptors in Tx queue may be limited by data inline settings.
178 Inline data require the more descriptor building blocks and overall block
179 amount may exceed the hardware supported limits. The application should
180 reduce the requested Tx size or adjust data inline settings with
181 ``txq_inline_max`` and ``txq_inline_mpw`` devargs keys.
183 - E-Switch decapsulation Flow:
185 - can be applied to PF port only.
186 - must specify VF port action (packet redirection from PF to VF).
187 - optionally may specify tunnel inner source and destination MAC addresses.
189 - E-Switch encapsulation Flow:
191 - can be applied to VF ports only.
192 - must specify PF port action (packet redirection from VF to PF).
194 - ICMP/ICMP6 code/type matching, IP-in-IP and MPLS flow matching are all
195 mutually exclusive features which cannot be supported together
196 (see :ref:`mlx5_firmware_config`).
200 - Requires DevX to be enabled.
201 - KEEP_CRC offload cannot be supported with LRO.
202 - The first mbuf length, without head-room, must be big enough to include the
208 MLX5 supports various methods to report statistics:
210 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.
212 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.
214 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.
222 These options can be modified in the ``.config`` file.
224 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
226 Toggle compilation of librte_pmd_mlx5 itself.
228 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
230 Build PMD with additional code to make it loadable without hard
231 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
232 on the target system.
234 In this mode, their presence is still required for it to run properly,
235 however their absence won't prevent a DPDK application from starting (with
236 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
237 missing with ``ldd(1)``.
239 It works by moving these dependencies to a purpose-built rdma-core "glue"
240 plug-in which must either be installed in a directory whose name is based
241 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
242 standard location for the dynamic linker (e.g. ``/lib``) if left to the
243 default empty string (``""``).
245 This option has no performance impact.
247 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
249 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
250 in the PMD shared library or the executable static binary.
252 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
254 Toggle debugging code and stricter compilation flags. Enabling this option
255 adds additional run-time checks and debugging messages at the cost of
260 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
261 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
262 64. Default armv8a configuration of make build and meson build set it to 128
263 then brings performance degradation.
265 Environment variables
266 ~~~~~~~~~~~~~~~~~~~~~
270 A list of directories in which to search for the rdma-core "glue" plug-in,
271 separated by colons or semi-colons.
273 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
274 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
275 since ``LD_LIBRARY_PATH`` has no effect in this case.
277 - ``MLX5_SHUT_UP_BF``
279 Configures HW Tx doorbell register as IO-mapped.
281 By default, the HW Tx doorbell is configured as a write-combining register.
282 The register would be flushed to HW usually when the write-combining buffer
283 becomes full, but it depends on CPU design.
285 Except for vectorized Tx burst routines, a write memory barrier is enforced
286 after updating the register so that the update can be immediately visible to
289 When vectorized Tx burst is called, the barrier is set only if the burst size
290 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
291 variable will bring better latency even though the maximum throughput can
294 Run-time configuration
295 ~~~~~~~~~~~~~~~~~~~~~~
297 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
298 because it is affected by their state. Forcing them down prevents packets
301 - **ethtool** operations on related kernel interfaces also affect the PMD.
303 - ``rxq_cqe_comp_en`` parameter [int]
305 A nonzero value enables the compression of CQE on RX side. This feature
306 allows to save PCI bandwidth and improve performance. Enabled by default.
310 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
311 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
313 - ``rxq_cqe_pad_en`` parameter [int]
315 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
316 is aligned with the size of a cacheline of the core. If cacheline size is
317 128B, the CQE size is configured to be 128B even though the device writes
318 only 64B data on the cacheline. This is to avoid unnecessary cache
319 invalidation by device's two consecutive writes on to one cacheline.
320 However in some architecture, it is more beneficial to update entire
321 cacheline with padding the rest 64B rather than striding because
322 read-modify-write could drop performance a lot. On the other hand,
323 writing extra data will consume more PCIe bandwidth and could also drop
324 the maximum throughput. It is recommended to empirically set this
325 parameter. Disabled by default.
329 - CPU having 128B cacheline with ConnectX-5 and BlueField.
331 - ``rxq_pkt_pad_en`` parameter [int]
333 A nonzero value enables padding Rx packet to the size of cacheline on PCI
334 transaction. This feature would waste PCI bandwidth but could improve
335 performance by avoiding partial cacheline write which may cause costly
336 read-modify-copy in memory transaction on some architectures. Disabled by
341 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
342 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
344 - ``mprq_en`` parameter [int]
346 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
347 configured as Multi-Packet RQ if the total number of Rx queues is
348 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
351 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
352 by posting a single large buffer for multiple packets. Instead of posting a
353 buffers per a packet, one large buffer is posted in order to receive multiple
354 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
355 and each stride receives one packet. MPRQ can improve throughput for
356 small-packet traffic.
358 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
359 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
360 configure large stride size enough to accommodate max_rx_pkt_len as long as
361 device allows. Note that this can waste system memory compared to enabling Rx
362 scatter and multi-segment packet.
364 - ``mprq_log_stride_num`` parameter [int]
366 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
367 strides can reduce PCIe traffic further. If configured value is not in the
368 range of device capability, the default value will be set with a warning
369 message. The default value is 4 which is 16 strides per a buffer, valid only
370 if ``mprq_en`` is set.
372 The size of Rx queue should be bigger than the number of strides.
374 - ``mprq_max_memcpy_len`` parameter [int]
376 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
377 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
378 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
379 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
380 A mempool for external buffers will be allocated and managed by PMD. If Rx
381 packet is externally attached, ol_flags field of the mbuf will have
382 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
383 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
385 - ``rxqs_min_mprq`` parameter [int]
387 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
388 greater or equal to this value. The default value is 12, valid only if
391 - ``txq_inline`` parameter [int]
393 Amount of data to be inlined during TX operations. This parameter is
394 deprecated and converted to the new parameter ``txq_inline_max`` providing
395 partial compatibility.
397 - ``txqs_min_inline`` parameter [int]
399 Enable inline data send only when the number of TX queues is greater or equal
402 This option should be used in combination with ``txq_inline_max`` and
403 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
405 If this option is not specified the default value 16 is used for BlueField
406 and 8 for other platforms
408 The data inlining consumes the CPU cycles, so this option is intended to
409 auto enable inline data if we have enough Tx queues, which means we have
410 enough CPU cores and PCI bandwidth is getting more critical and CPU
411 is not supposed to be bottleneck anymore.
413 The copying data into WQE improves latency and can improve PPS performance
414 when PCI back pressure is detected and may be useful for scenarios involving
415 heavy traffic on many queues.
417 Because additional software logic is necessary to handle this mode, this
418 option should be used with care, as it may lower performance when back
419 pressure is not expected.
421 If inline data are enabled it may affect the maximal size of Tx queue in
422 descriptors because the inline data increase the descriptor size and
423 queue size limits supported by hardware may be exceeded.
425 - ``txq_inline_min`` parameter [int]
427 Minimal amount of data to be inlined into WQE during Tx operations. NICs
428 may require this minimal data amount to operate correctly. The exact value
429 may depend on NIC operation mode, requested offloads, etc. It is strongly
430 recommended to omit this parameter and use the default values. Anyway,
431 applications using this parameter should take into consideration that
432 specifying an inconsistent value may prevent the NIC from sending packets.
434 If ``txq_inline_min`` key is present the specified value (may be aligned
435 by the driver in order not to exceed the limits and provide better descriptor
436 space utilization) will be used by the driver and it is guaranteed that
437 requested amount of data bytes are inlined into the WQE beside other inline
438 settings. This key also may update ``txq_inline_max`` value (default
439 or specified explicitly in devargs) to reserve the space for inline data.
441 If ``txq_inline_min`` key is not present, the value may be queried by the
442 driver from the NIC via DevX if this feature is available. If there is no DevX
443 enabled/supported the value 18 (supposing L2 header including VLAN) is set
444 for ConnectX-4 and ConnectX-4LX, and 0 is set by default for ConnectX-5
445 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
448 For ConnectX-4 NIC, driver does not allow specifying value below 18
449 (minimal L2 header, including VLAN), error will be raised.
451 For ConnectX-4LX NIC, it is allowed to specify values below 18, but
452 it is not recommended and may prevent NIC from sending packets over
455 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
456 Multi-Packet Write), because last one does not support partial packet inlining.
457 This is not very critical due to minimal data inlining is mostly required
458 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
460 - ``txq_inline_max`` parameter [int]
462 Specifies the maximal packet length to be completely inlined into WQE
463 Ethernet Segment for ordinary SEND method. If packet is larger than specified
464 value, the packet data won't be copied by the driver at all, data buffer
465 is addressed with a pointer. If packet length is less or equal all packet
466 data will be copied into WQE. This may improve PCI bandwidth utilization for
467 short packets significantly but requires the extra CPU cycles.
469 The data inline feature is controlled by number of Tx queues, if number of Tx
470 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
471 is engaged, if there are not enough Tx queues (which means not enough CPU cores
472 and CPU resources are scarce), data inline is not performed by the driver.
473 Assigning ``txqs_min_inline`` with zero always enables the data inline.
475 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
476 by the driver in order not to exceed the limit (930 bytes) and to provide better
477 WQE space filling without gaps, the adjustment is reflected in the debug log.
479 - ``txq_inline_mpw`` parameter [int]
481 Specifies the maximal packet length to be completely inlined into WQE for
482 Enhanced MPW method. If packet is large the specified value, the packet data
483 won't be copied, and data buffer is addressed with pointer. If packet length
484 is less or equal, all packet data will be copied into WQE. This may improve PCI
485 bandwidth utilization for short packets significantly but requires the extra
488 The data inline feature is controlled by number of TX queues, if number of Tx
489 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
490 is engaged, if there are not enough Tx queues (which means not enough CPU cores
491 and CPU resources are scarce), data inline is not performed by the driver.
492 Assigning ``txqs_min_inline`` with zero always enables the data inline.
494 The default ``txq_inline_mpw`` value is 268. The specified value may be adjusted
495 by the driver in order not to exceed the limit (930 bytes) and to provide better
496 WQE space filling without gaps, the adjustment is reflected in the debug log.
497 Due to multiple packets may be included to the same WQE with Enhanced Multi
498 Packet Write Method and overall WQE size is limited it is not recommended to
499 specify large values for the ``txq_inline_mpw``.
501 - ``txqs_max_vec`` parameter [int]
503 Enable vectorized Tx only when the number of TX queues is less than or
504 equal to this value. This parameter is deprecated and ignored, kept
505 for compatibility issue to not prevent driver from probing.
507 - ``txq_mpw_hdr_dseg_en`` parameter [int]
509 A nonzero value enables including two pointers in the first block of TX
510 descriptor. The parameter is deprecated and ignored, kept for compatibility
513 - ``txq_max_inline_len`` parameter [int]
515 Maximum size of packet to be inlined. This limits the size of packet to
516 be inlined. If the size of a packet is larger than configured value, the
517 packet isn't inlined even though there's enough space remained in the
518 descriptor. Instead, the packet is included with pointer. This parameter
519 is deprecated and converted directly to ``txq_inline_mpw`` providing full
520 compatibility. Valid only if eMPW feature is engaged.
522 - ``txq_mpw_en`` parameter [int]
524 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
525 ConnectX-6 and BlueField. eMPW allows the TX burst function to pack up multiple
526 packets in a single descriptor session in order to save PCI bandwidth and improve
527 performance at the cost of a slightly higher CPU usage. When ``txq_inline_mpw``
528 is set along with ``txq_mpw_en``, TX burst function copies entire packet
529 data on to TX descriptor instead of including pointer of packet.
531 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
532 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
533 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
534 option or reported by the NIC, the eMPW feature is disengaged.
536 - ``tx_vec_en`` parameter [int]
538 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6 and BlueField
539 NICs if the number of global Tx queues on the port is less than
540 ``txqs_max_vec``. The parameter is deprecated and ignored.
542 - ``rx_vec_en`` parameter [int]
544 A nonzero value enables Rx vector if the port is not configured in
545 multi-segment otherwise this parameter is ignored.
549 - ``vf_nl_en`` parameter [int]
551 A nonzero value enables Netlink requests from the VF to add/remove MAC
552 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
553 Otherwise the relevant configuration must be run with Linux iproute2 tools.
554 This is a prerequisite to receive this kind of traffic.
556 Enabled by default, valid only on VF devices ignored otherwise.
558 - ``l3_vxlan_en`` parameter [int]
560 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
561 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
562 parameter. This is a prerequisite to receive this kind of traffic.
566 - ``dv_flow_en`` parameter [int]
568 A nonzero value enables the DV flow steering assuming it is supported
573 - ``dv_esw_en`` parameter [int]
575 A nonzero value enables E-Switch using Direct Rules.
577 Enabled by default if supported.
579 - ``mr_ext_memseg_en`` parameter [int]
581 A nonzero value enables extending memseg when registering DMA memory. If
582 enabled, the number of entries in MR (Memory Region) lookup table on datapath
583 is minimized and it benefits performance. On the other hand, it worsens memory
584 utilization because registered memory is pinned by kernel driver. Even if a
585 page in the extended chunk is freed, that doesn't become reusable until the
586 entire memory is freed.
590 - ``representor`` parameter [list]
592 This parameter can be used to instantiate DPDK Ethernet devices from
593 existing port (or VF) representors configured on the device.
595 It is a standard parameter whose format is described in
596 :ref:`ethernet_device_standard_device_arguments`.
598 For instance, to probe port representors 0 through 2::
602 - ``max_dump_files_num`` parameter [int]
604 The maximum number of files per PMD entity that may be created for debug information.
605 The files will be created in /var/log directory or in current directory.
607 set to 128 by default.
609 - ``lro_timeout_usec`` parameter [int]
611 The maximum allowed duration of an LRO session, in micro-seconds.
612 PMD will set the nearest value supported by HW, which is not bigger than
613 the input ``lro_timeout_usec`` value.
614 If this parameter is not specified, by default PMD will set
615 the smallest value supported by HW.
617 .. _mlx5_firmware_config:
619 Firmware configuration
620 ~~~~~~~~~~~~~~~~~~~~~~
622 Firmware features can be configured as key/value pairs.
624 The command to set a value is::
626 mlxconfig -d <device> set <key>=<value>
628 The command to query a value is::
630 mlxconfig -d <device> query | grep <key>
632 The device name for the command ``mlxconfig`` can be either the PCI address,
633 or the mst device name found with::
637 Below are some firmware configurations listed.
643 value: 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
649 - maximum number of SR-IOV virtual functions::
653 - enable DevX (required by Direct Rules and other features)::
657 - aggressive CQE zipping::
661 - L3 VXLAN and VXLAN-GPE destination UDP port::
664 IP_OVER_VXLAN_PORT=<udp dport>
666 - enable IP-in-IP tunnel flow matching::
668 FLEX_PARSER_PROFILE_ENABLE=0
670 - enable MPLS flow matching::
672 FLEX_PARSER_PROFILE_ENABLE=1
674 - enable ICMP/ICMP6 code/type fields matching::
676 FLEX_PARSER_PROFILE_ENABLE=2
681 This driver relies on external libraries and kernel drivers for resources
682 allocations and initialization. The following dependencies are not part of
683 DPDK and must be installed separately:
687 User space Verbs framework used by librte_pmd_mlx5. This library provides
688 a generic interface between the kernel and low-level user space drivers
691 It allows slow and privileged operations (context initialization, hardware
692 resources allocations) to be managed by the kernel and fast operations to
693 never leave user space.
697 Low-level user space driver library for Mellanox
698 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
701 This library basically implements send/receive calls to the hardware
706 They provide the kernel-side Verbs API and low level device drivers that
707 manage actual hardware initialization and resources sharing with user
710 Unlike most other PMDs, these modules must remain loaded and bound to
713 - mlx5_core: hardware driver managing Mellanox
714 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
716 - mlx5_ib: InifiniBand device driver.
717 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
719 - **Firmware update**
721 Mellanox OFED/EN releases include firmware updates for
722 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
724 Because each release provides new features, these updates must be applied to
725 match the kernel modules and libraries they come with.
729 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
735 Either RDMA Core library with a recent enough Linux kernel release
736 (recommended) or Mellanox OFED/EN, which provides compatibility with older
739 RDMA Core with Linux Kernel
740 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
742 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
743 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
744 (see `RDMA Core installation documentation`_)
745 - When building for i686 use:
747 - rdma-core version 18.0 or above built with 32bit support.
748 - Kernel version 4.14.41 or above.
750 - Starting with rdma-core v21, static libraries can be built::
753 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
756 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
757 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
759 If rdma-core libraries are built but not installed, DPDK makefile can link them,
760 thanks to these environment variables:
762 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
763 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
764 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
769 - Mellanox OFED version: ** 4.5, 4.6** /
770 Mellanox EN version: **4.5, 4.6**
773 - ConnectX-4: **12.21.1000** and above.
774 - ConnectX-4 Lx: **14.21.1000** and above.
775 - ConnectX-5: **16.21.1000** and above.
776 - ConnectX-5 Ex: **16.21.1000** and above.
777 - ConnectX-6: **20.99.5374** and above.
778 - BlueField: **18.25.1010** and above.
780 While these libraries and kernel modules are available on OpenFabrics
781 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
782 managers on most distributions, this PMD requires Ethernet extensions that
783 may not be supported at the moment (this is a work in progress).
786 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
788 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
789 include the necessary support and should be used in the meantime. For DPDK,
790 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
791 required from that distribution.
795 Several versions of Mellanox OFED/EN are available. Installing the version
796 this DPDK release was developed and tested against is strongly
797 recommended. Please check the `prerequisites`_.
802 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
803 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
804 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
805 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
806 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
807 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
808 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
809 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
810 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
811 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
812 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
813 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
814 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
815 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
816 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
817 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
818 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
819 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
820 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
822 Quick Start Guide on OFED/EN
823 ----------------------------
825 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
828 2. Install the required libraries and kernel modules either by installing
829 only the required set, or by installing the entire Mellanox OFED/EN::
831 ./mlnxofedinstall --upstream-libs --dpdk
833 3. Verify the firmware is the correct one::
837 4. Verify all ports links are set to Ethernet::
839 mlxconfig -d <mst device> query | grep LINK_TYPE
843 Link types may have to be configured to Ethernet::
845 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
847 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
849 For hypervisors, verify SR-IOV is enabled on the NIC::
851 mlxconfig -d <mst device> query | grep SRIOV_EN
854 If needed, configure SR-IOV::
856 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
857 mlxfwreset -d <mst device> reset
859 5. Restart the driver::
861 /etc/init.d/openibd restart
865 service openibd restart
867 If link type was changed, firmware must be reset as well::
869 mlxfwreset -d <mst device> reset
871 For hypervisors, after reset write the sysfs number of virtual functions
874 To dynamically instantiate a given number of virtual functions (VFs)::
876 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
878 6. Compile DPDK and you are ready to go. See instructions on
879 :ref:`Development Kit Build System <Development_Kit_Build_System>`
881 Enable switchdev mode
882 ---------------------
884 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
885 Representor is a port in DPDK that is connected to a VF in such a way
886 that assuming there are no offload flows, each packet that is sent from the VF
887 will be received by the corresponding representor. While each packet that is
888 sent to a representor will be received by the VF.
889 This is very useful in case of SRIOV mode, where the first packet that is sent
890 by the VF will be received by the DPDK application which will decide if this
891 flow should be offloaded to the E-Switch. After offloading the flow packet
892 that the VF that are matching the flow will not be received any more by
893 the DPDK application.
895 1. Enable SRIOV mode::
897 mlxconfig -d <mst device> set SRIOV_EN=true
899 2. Configure the max number of VFs::
901 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
905 mlxfwreset -d <mst device> reset
907 3. Configure the actual number of VFs::
909 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
911 4. Unbind the device (can be rebind after the switchdev mode)::
913 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
915 5. Enbale switchdev mode::
917 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
922 1. Configure aggressive CQE Zipping for maximum performance::
924 mlxconfig -d <mst device> s CQE_COMPRESSION=1
926 To set it back to the default CQE Zipping mode use::
928 mlxconfig -d <mst device> s CQE_COMPRESSION=0
930 2. In case of virtualization:
932 - Make sure that hypervisor kernel is 3.16 or newer.
933 - Configure boot with ``iommu=pt``.
935 - Make sure to allocate a VM on huge pages.
936 - Make sure to set CPU pinning.
938 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
939 for better performance. For VMs, verify that the right CPU
940 and NUMA node are pinned according to the above. Run::
944 to identify the NUMA node to which the PCIe adapter is connected.
946 4. If more than one adapter is used, and root complex capabilities allow
947 to put both adapters on the same NUMA node without PCI bandwidth degradation,
948 it is recommended to locate both adapters on the same NUMA node.
949 This in order to forward packets from one to the other without
950 NUMA performance penalty.
952 5. Disable pause frames::
954 ethtool -A <netdev> rx off tx off
956 6. Verify IO non-posted prefetch is disabled by default. This can be checked
957 via the BIOS configuration. Please contact you server provider for more
958 information about the settings.
962 On some machines, depends on the machine integrator, it is beneficial
963 to set the PCI max read request parameter to 1K. This can be
964 done in the following way:
966 To query the read request size use::
968 setpci -s <NIC PCI address> 68.w
970 If the output is different than 3XXX, set it by::
972 setpci -s <NIC PCI address> 68.w=3XXX
974 The XXX can be different on different systems. Make sure to configure
975 according to the setpci output.
977 7. To minimize overhead of searching Memory Regions:
979 - '--socket-mem' is recommended to pin memory by predictable amount.
980 - Configure per-lcore cache when creating Mempools for packet buffer.
981 - Refrain from dynamically allocating/freeing memory in run-time.
983 .. _mlx5_offloads_support:
985 Supported hardware offloads
986 ---------------------------
988 .. table:: Minimal SW/HW versions for queue offloads
990 ============== ===== ===== ========= ===== ========== ==========
991 Offload DPDK Linux rdma-core OFED firmware hardware
992 ============== ===== ===== ========= ===== ========== ==========
993 common base 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
994 checksums 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
995 Rx timestamp 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
996 TSO 17.11 4.14 16 4.2-1 12.21.1000 ConnectX-4
997 LRO 19.08 N/A N/A 4.6-4 16.25.6406 ConnectX-5
998 ============== ===== ===== ========= ===== ========== ==========
1000 .. table:: Minimal SW/HW versions for rte_flow offloads
1002 +-----------------------+-----------------+-----------------+
1003 | Offload | with E-Switch | with vNIC |
1004 +=======================+=================+=================+
1005 | Count | | DPDK 19.05 | | DPDK 19.02 |
1006 | | | OFED 4.6 | | OFED 4.6 |
1007 | | | rdma-core 24 | | rdma-core 23 |
1008 | | | ConnectX-5 | | ConnectX-5 |
1009 +-----------------------+-----------------+-----------------+
1010 | Drop / Queue / RSS | | DPDK 19.05 | | DPDK 18.11 |
1011 | | | OFED 4.6 | | OFED 4.5 |
1012 | | | rdma-core 24 | | rdma-core 23 |
1013 | | | ConnectX-5 | | ConnectX-4 |
1014 +-----------------------+-----------------+-----------------+
1015 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
1016 | (VXLAN / NVGRE / RAW) | | OFED 4.6-2 | | OFED 4.6 |
1017 | | | rdma-core 24 | | rdma-core 23 |
1018 | | | ConnectX-5 | | ConnectX-5 |
1019 +-----------------------+-----------------+-----------------+
1020 | | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
1021 | | (set_ipv4_src / | | OFED 4.6-2 | | OFED 4.6-2 |
1022 | | set_ipv4_dst / | | rdma-core 24 | | rdma-core 23 |
1023 | | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
1024 | | set_ipv6_dst / | | |
1025 | | set_tp_src / | | |
1026 | | set_tp_dst / | | |
1029 | | set_mac_src / | | |
1030 | | set_mac_dst) | | |
1031 +-----------------------+-----------------+-----------------+
1032 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1033 | | | OFED 4.6-4 | | OFED 4.6-4 |
1034 | | | rdma-core 24 | | N/A |
1035 | | | ConnectX-5 | | ConnectX-5 |
1036 +-----------------------+-----------------+-----------------+
1037 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1038 | | | OFED 4.6 | | OFED 4.5 |
1039 | | | rdma-core 24 | | rdma-core 23 |
1040 | | | ConnectX-5 | | ConnectX-4 |
1041 +-----------------------+-----------------+-----------------+
1042 | Port ID | | DPDK 19.05 | | N/A |
1043 | | | OFED 4.6 | | N/A |
1044 | | | rdma-core 24 | | N/A |
1045 | | | ConnectX-5 | | N/A |
1046 +-----------------------+-----------------+-----------------+
1047 | | VLAN | | DPDK 19.11 | | DPDK 19.11 |
1048 | | (of_pop_vlan / | | OFED 4.6-4 | | OFED 4.6-4 |
1049 | | of_push_vlan / | | ConnectX-5 | | ConnectX-5 |
1050 | | of_set_vlan_pcp / | | |
1051 | | of_set_vlan_vid) | | |
1052 +-----------------------+-----------------+-----------------+
1057 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1058 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1060 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1061 command-line parameter to enable additional protocols (UDP and TCP as well
1062 as IP), the following commands must be entered from its CLI to get the same
1063 behavior as librte_pmd_mlx4::
1066 > port config all rss all
1072 This section demonstrates how to launch **testpmd** with Mellanox
1073 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1075 #. Load the kernel modules::
1077 modprobe -a ib_uverbs mlx5_core mlx5_ib
1079 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1082 /etc/init.d/openibd restart
1086 User space I/O kernel modules (uio and igb_uio) are not used and do
1087 not have to be loaded.
1089 #. Make sure Ethernet interfaces are in working order and linked to kernel
1090 verbs. Related sysfs entries should be present::
1092 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1101 #. Optionally, retrieve their PCI bus addresses for whitelisting::
1104 for intf in eth2 eth3 eth4 eth5;
1106 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1109 sed -n 's,.*/\(.*\),-w \1,p'
1118 #. Request huge pages::
1120 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1122 #. Start testpmd with basic parameters::
1124 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
1129 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1130 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1131 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1132 PMD: librte_pmd_mlx5: 1 port(s) detected
1133 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1134 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1135 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1136 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1137 PMD: librte_pmd_mlx5: 1 port(s) detected
1138 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1139 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1140 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1141 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1142 PMD: librte_pmd_mlx5: 1 port(s) detected
1143 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1144 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1145 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1146 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1147 PMD: librte_pmd_mlx5: 1 port(s) detected
1148 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1149 Interactive-mode selected
1150 Configuring Port 0 (socket 0)
1151 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1152 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1153 Port 0: E4:1D:2D:E7:0C:FE
1154 Configuring Port 1 (socket 0)
1155 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1156 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1157 Port 1: E4:1D:2D:E7:0C:FF
1158 Configuring Port 2 (socket 0)
1159 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1160 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1161 Port 2: E4:1D:2D:E7:0C:FA
1162 Configuring Port 3 (socket 0)
1163 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1164 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1165 Port 3: E4:1D:2D:E7:0C:FB
1166 Checking link statuses...
1167 Port 0 Link Up - speed 40000 Mbps - full-duplex
1168 Port 1 Link Up - speed 40000 Mbps - full-duplex
1169 Port 2 Link Up - speed 10000 Mbps - full-duplex
1170 Port 3 Link Up - speed 10000 Mbps - full-duplex