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 - Support for multiple MAC addresses.
64 - RX CRC stripping configuration.
65 - Promiscuous mode on PF and VF.
66 - Multicast promiscuous mode on PF and VF.
67 - Hardware checksum offloads.
68 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
70 - Flow API, including :ref:`flow_isolated_mode`.
72 - KVM and VMware ESX SR-IOV modes are supported.
73 - RSS hash result is supported.
74 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
75 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
77 - Statistics query including Basic, Extended and per queue.
79 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP, IP-in-IP.
80 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
81 - NIC HW offloads: encapsulation (vxlan, gre, mplsoudp, mplsogre), NAT, routing, TTL
82 increment/decrement, count, drop, mark. For details please see :ref:`Supported hardware offloads using rte_flow API`.
83 - Flow insertion rate of more then million flows per second, when using Direct Rules.
84 - Support for multiple rte_flow groups.
90 - For secondary process:
92 - Forked secondary process not supported.
93 - External memory unregistered in EAL memseg list cannot be used for DMA
94 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
95 primary process and remapped to the same virtual address in secondary
96 process. If the external memory is registered by primary process but has
97 different virtual address in secondary process, unexpected error may happen.
99 - Flow pattern without any specific vlan will match for vlan packets as well:
101 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
102 Meaning, the flow rule::
104 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
106 Will only match vlan packets with vid=3. and the flow rules::
108 flow create 0 ingress pattern eth / ipv4 / end ...
112 flow create 0 ingress pattern eth / vlan / ipv4 / end ...
114 Will match any ipv4 packet (VLAN included).
116 - A multi segment packet must have less than 6 segments in case the Tx burst function
117 is set to multi-packet send or Enhanced multi-packet send. Otherwise it must have
118 less than 50 segments.
120 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
121 to 0 are not supported.
123 - VXLAN TSO and checksum offloads are not supported on VM.
125 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
127 - VF: flow rules created on VF devices can only match traffic targeted at the
128 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
132 MAC addresses not already present in the bridge table of the associated
133 kernel network device will be added and cleaned up by the PMD when closing
134 the device. In case of ungraceful program termination, some entries may
135 remain present and should be removed manually by other means.
137 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
138 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
139 ol_flags. As the mempool for the external buffer is managed by PMD, all the
140 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
141 the external buffers will be freed by PMD and the application which still
142 holds the external buffers may be corrupted.
144 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
145 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
146 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
148 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
149 and allmulticast mode are both set to off.
150 To receive IPv6 Multicast messages on VM, explicitly set the relevant
151 MAC address using rte_eth_dev_mac_addr_add() API.
153 - E-Switch decapsulation Flow:
155 - can be applied to PF port only.
156 - must specify VF port action (packet redirection from PF to VF).
157 - optionally may specify tunnel inner source and destination MAC addresses.
159 - E-Switch encapsulation Flow:
161 - can be applied to VF ports only.
162 - must specify PF port action (packet redirection from VF to PF).
164 - ICMP/ICMP6 code/type matching cannot be supported togeter with IP-in-IP tunnel.
169 MLX5 supports various of methods to report statistics:
171 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.
173 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.
175 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.
183 These options can be modified in the ``.config`` file.
185 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
187 Toggle compilation of librte_pmd_mlx5 itself.
189 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
191 Build PMD with additional code to make it loadable without hard
192 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
193 on the target system.
195 In this mode, their presence is still required for it to run properly,
196 however their absence won't prevent a DPDK application from starting (with
197 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
198 missing with ``ldd(1)``.
200 It works by moving these dependencies to a purpose-built rdma-core "glue"
201 plug-in which must either be installed in a directory whose name is based
202 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
203 standard location for the dynamic linker (e.g. ``/lib``) if left to the
204 default empty string (``""``).
206 This option has no performance impact.
208 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
210 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
211 in the PMD shared library or the executable static binary.
213 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
215 Toggle debugging code and stricter compilation flags. Enabling this option
216 adds additional run-time checks and debugging messages at the cost of
221 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
222 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
223 64. Default armv8a configuration of make build and meson build set it to 128
224 then brings performance degradation.
226 Environment variables
227 ~~~~~~~~~~~~~~~~~~~~~
231 A list of directories in which to search for the rdma-core "glue" plug-in,
232 separated by colons or semi-colons.
234 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
235 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
236 since ``LD_LIBRARY_PATH`` has no effect in this case.
238 - ``MLX5_SHUT_UP_BF``
240 Configures HW Tx doorbell register as IO-mapped.
242 By default, the HW Tx doorbell is configured as a write-combining register.
243 The register would be flushed to HW usually when the write-combining buffer
244 becomes full, but it depends on CPU design.
246 Except for vectorized Tx burst routines, a write memory barrier is enforced
247 after updating the register so that the update can be immediately visible to
250 When vectorized Tx burst is called, the barrier is set only if the burst size
251 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
252 variable will bring better latency even though the maximum throughput can
255 Run-time configuration
256 ~~~~~~~~~~~~~~~~~~~~~~
258 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
259 because it is affected by their state. Forcing them down prevents packets
262 - **ethtool** operations on related kernel interfaces also affect the PMD.
264 - ``rxq_cqe_comp_en`` parameter [int]
266 A nonzero value enables the compression of CQE on RX side. This feature
267 allows to save PCI bandwidth and improve performance. Enabled by default.
271 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
272 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
274 - ``rxq_cqe_pad_en`` parameter [int]
276 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
277 is aligned with the size of a cacheline of the core. If cacheline size is
278 128B, the CQE size is configured to be 128B even though the device writes
279 only 64B data on the cacheline. This is to avoid unnecessary cache
280 invalidation by device's two consecutive writes on to one cacheline.
281 However in some architecture, it is more beneficial to update entire
282 cacheline with padding the rest 64B rather than striding because
283 read-modify-write could drop performance a lot. On the other hand,
284 writing extra data will consume more PCIe bandwidth and could also drop
285 the maximum throughput. It is recommended to empirically set this
286 parameter. Disabled by default.
290 - CPU having 128B cacheline with ConnectX-5 and BlueField.
292 - ``rxq_pkt_pad_en`` parameter [int]
294 A nonzero value enables padding Rx packet to the size of cacheline on PCI
295 transaction. This feature would waste PCI bandwidth but could improve
296 performance by avoiding partial cacheline write which may cause costly
297 read-modify-copy in memory transaction on some architectures. Disabled by
302 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
303 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
305 - ``mprq_en`` parameter [int]
307 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
308 configured as Multi-Packet RQ if the total number of Rx queues is
309 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
312 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
313 by posting a single large buffer for multiple packets. Instead of posting a
314 buffers per a packet, one large buffer is posted in order to receive multiple
315 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
316 and each stride receives one packet. MPRQ can improve throughput for
317 small-packet traffic.
319 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
320 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
321 configure large stride size enough to accommodate max_rx_pkt_len as long as
322 device allows. Note that this can waste system memory compared to enabling Rx
323 scatter and multi-segment packet.
325 - ``mprq_log_stride_num`` parameter [int]
327 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
328 strides can reduce PCIe traffic further. If configured value is not in the
329 range of device capability, the default value will be set with a warning
330 message. The default value is 4 which is 16 strides per a buffer, valid only
331 if ``mprq_en`` is set.
333 The size of Rx queue should be bigger than the number of strides.
335 - ``mprq_max_memcpy_len`` parameter [int]
337 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
338 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
339 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
340 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
341 A mempool for external buffers will be allocated and managed by PMD. If Rx
342 packet is externally attached, ol_flags field of the mbuf will have
343 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
344 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
346 - ``rxqs_min_mprq`` parameter [int]
348 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
349 greater or equal to this value. The default value is 12, valid only if
352 - ``txq_inline`` parameter [int]
354 Amount of data to be inlined during TX operations. This parameter is
355 deprecated and converted to the new parameter ``txq_inline_max`` providing
356 partial compatibility.
358 - ``txqs_min_inline`` parameter [int]
360 Enable inline data send only when the number of TX queues is greater or equal
363 This option should be used in combination with ``txq_inline_max`` and
364 ``txq_inline_mpw`` below and does not affect ``txq_inline_min`` settings above.
366 If this option is not specified the default value 16 is used for BlueField
367 and 8 for other platforms
369 The data inlining consumes the CPU cycles, so this option is intended to
370 auto enable inline data if we have enough Tx queues, which means we have
371 enough CPU cores and PCI bandwidth is getting more critical and CPU
372 is not supposed to be bottleneck anymore.
374 The copying data into WQE improves latency and can improve PPS performance
375 when PCI back pressure is detected and may be useful for scenarios involving
376 heavy traffic on many queues.
378 Because additional software logic is necessary to handle this mode, this
379 option should be used with care, as it may lower performance when back
380 pressure is not expected.
382 - ``txq_inline_min`` parameter [int]
384 Minimal amount of data to be inlined into WQE during Tx operations. NICs
385 may require this minimal data amount to operate correctly. The exact value
386 may depend on NIC operation mode, requested offloads, etc.
388 If ``txq_inline_min`` key is present the specified value (may be aligned
389 by the driver in order not to exceed the limits and provide better descriptor
390 space utilization) will be used by the driver and it is guaranteed the
391 requested data bytes are inlined into the WQE beside other inline settings.
392 This keys also may update ``txq_inline_max`` value (default of specified
393 explicitly in devargs) to reserve the space for inline data.
395 If ``txq_inline_min`` key is not present, the value may be queried by the
396 driver from the NIC via DevX if this feature is available. If there is no DevX
397 enabled/supported the value 18 (supposing L2 header including VLAN) is set
398 for ConnectX-4, value 58 (supposing L2-L4 headers, required by configurations
399 over E-Switch) is set for ConnectX-4 Lx, and 0 is set by default for ConnectX-5
400 and newer NICs. If packet is shorter the ``txq_inline_min`` value, the entire
403 For the ConnectX-4 and ConnectX-4 Lx NICs driver does not allow to set
404 this value below 18 (minimal L2 header, including VLAN).
406 Please, note, this minimal data inlining disengages eMPW feature (Enhanced
407 Multi-Packet Write), because last one does not support partial packet inlining.
408 This is not very critical due to minimal data inlining is mostly required
409 by ConnectX-4 and ConnectX-4 Lx, these NICs do not support eMPW feature.
411 - ``txq_inline_max`` parameter [int]
413 Specifies the maximal packet length to be completely inlined into WQE
414 Ethernet Segment for ordinary SEND method. If packet is larger than specified
415 value, the packet data won't be copied by the driver at all, data buffer
416 is addressed with a pointer. If packet length is less or equal all packet
417 data will be copied into WQE. This may improve PCI bandwidth utilization for
418 short packets significantly but requires the extra CPU cycles.
420 The data inline feature is controlled by number of Tx queues, if number of Tx
421 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
422 is engaged, if there are not enough Tx queues (which means not enough CPU cores
423 and CPU resources are scarce), data inline is not performed by the driver.
424 Assigning ``txqs_min_inline`` with zero always enables the data inline.
426 The default ``txq_inline_max`` value is 290. The specified value may be adjusted
427 by the driver in order not to exceed the limit (930 bytes) and to provide better
428 WQE space filling without gaps, the adjustment is reflected in the debug log.
430 - ``txq_inline_mpw`` parameter [int]
432 Specifies the maximal packet length to be completely inlined into WQE for
433 Enhanced MPW method. If packet is large the specified value, the packet data
434 won't be copied, and data buffer is addressed with pointer. If packet length
435 is less or equal, all packet data will be copied into WQE. This may improve PCI
436 bandwidth utilization for short packets significantly but requires the extra
439 The data inline feature is controlled by number of TX queues, if number of Tx
440 queues is larger than ``txqs_min_inline`` key parameter, the inline feature
441 is engaged, if there are not enough Tx queues (which means not enough CPU cores
442 and CPU resources are scarce), data inline is not performed by the driver.
443 Assigning ``txqs_min_inline`` with zero always enables the data inline.
445 The default ``txq_inline_mpw`` value is 188. The specified value may be adjusted
446 by the driver in order not to exceed the limit (930 bytes) and to provide better
447 WQE space filling without gaps, the adjustment is reflected in the debug log.
448 Due to multiple packets may be included to the same WQE with Enhanced Multi
449 Packet Write Method and overall WQE size is limited it is not recommended to
450 specify large values for the ``txq_inline_mpw``.
452 - ``txqs_max_vec`` parameter [int]
454 Enable vectorized Tx only when the number of TX queues is less than or
455 equal to this value. This parameter is deprecated and ignored, kept
456 for compatibility issue to not prevent driver from probing.
458 - ``txq_mpw_hdr_dseg_en`` parameter [int]
460 A nonzero value enables including two pointers in the first block of TX
461 descriptor. The parameter is deprecated and ignored, kept for compatibility
464 - ``txq_max_inline_len`` parameter [int]
466 Maximum size of packet to be inlined. This limits the size of packet to
467 be inlined. If the size of a packet is larger than configured value, the
468 packet isn't inlined even though there's enough space remained in the
469 descriptor. Instead, the packet is included with pointer. This parameter
470 is deprecated and converted directly to ``txq_inline_mpw`` providing full
471 compatibility. Valid only if eMPW feature is engaged.
473 - ``txq_mpw_en`` parameter [int]
475 A nonzero value enables Enhanced Multi-Packet Write (eMPW) for ConnectX-5,
476 ConnectX-6 and BlueField. eMPW allows the TX burst function to pack up multiple
477 packets in a single descriptor session in order to save PCI bandwidth and improve
478 performance at the cost of a slightly higher CPU usage. When ``txq_inline_mpw``
479 is set along with ``txq_mpw_en``, TX burst function copies entire packet
480 data on to TX descriptor instead of including pointer of packet.
482 The Enhanced Multi-Packet Write feature is enabled by default if NIC supports
483 it, can be disabled by explicit specifying 0 value for ``txq_mpw_en`` option.
484 Also, if minimal data inlining is requested by non-zero ``txq_inline_min``
485 option or reported by the NIC, the eMPW feature is disengaged.
487 - ``tx_vec_en`` parameter [int]
489 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6 and BlueField
490 NICs if the number of global Tx queues on the port is less than
491 ``txqs_max_vec``. The parameter is deprecated and ignored.
493 - ``rx_vec_en`` parameter [int]
495 A nonzero value enables Rx vector if the port is not configured in
496 multi-segment otherwise this parameter is ignored.
500 - ``vf_nl_en`` parameter [int]
502 A nonzero value enables Netlink requests from the VF to add/remove MAC
503 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
504 Otherwise the relevant configuration must be run with Linux iproute2 tools.
505 This is a prerequisite to receive this kind of traffic.
507 Enabled by default, valid only on VF devices ignored otherwise.
509 - ``l3_vxlan_en`` parameter [int]
511 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
512 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
513 parameter. This is a prerequisite to receive this kind of traffic.
517 - ``dv_flow_en`` parameter [int]
519 A nonzero value enables the DV flow steering assuming it is supported
524 - ``dv_esw_en`` parameter [int]
526 A nonzero value enables E-Switch using Direct Rules.
528 Enabled by default if supported.
530 - ``mr_ext_memseg_en`` parameter [int]
532 A nonzero value enables extending memseg when registering DMA memory. If
533 enabled, the number of entries in MR (Memory Region) lookup table on datapath
534 is minimized and it benefits performance. On the other hand, it worsens memory
535 utilization because registered memory is pinned by kernel driver. Even if a
536 page in the extended chunk is freed, that doesn't become reusable until the
537 entire memory is freed.
541 - ``representor`` parameter [list]
543 This parameter can be used to instantiate DPDK Ethernet devices from
544 existing port (or VF) representors configured on the device.
546 It is a standard parameter whose format is described in
547 :ref:`ethernet_device_standard_device_arguments`.
549 For instance, to probe port representors 0 through 2::
553 - ``max_dump_files_num`` parameter [int]
555 The maximum number of files per PMD entity that may be created for debug information.
556 The files will be created in /var/log directory or in current directory.
558 set to 128 by default.
560 - ``lro_timeout_usec`` parameter [int]
562 The maximum allowed duration of an LRO session, in micro-seconds.
563 PMD will set the nearest value supported by HW, which is not bigger than
564 the input ``lro_timeout_usec`` value.
565 If this parameter is not specified, by default PMD will set
566 the smallest value supported by HW.
568 Firmware configuration
569 ~~~~~~~~~~~~~~~~~~~~~~
571 - L3 VXLAN and VXLAN-GPE destination UDP port
573 .. code-block:: console
575 mlxconfig -d <mst device> set IP_OVER_VXLAN_EN=1
576 mlxconfig -d <mst device> set IP_OVER_VXLAN_PORT=<udp dport>
578 Verify configurations are set:
580 .. code-block:: console
582 mlxconfig -d <mst device> query | grep IP_OVER_VXLAN
583 IP_OVER_VXLAN_EN True(1)
584 IP_OVER_VXLAN_PORT <udp dport>
586 - enable ICMP/ICMP6's code/type field matching
588 .. code-block:: console
590 mlxconfig -d <mst device> set FLEX_PARSER_PROFILE_ENABLE=2
592 Verify configurations are set:
594 .. code-block:: console
596 mlxconfig -d <mst device> query | grep FLEX_PARSER_PROFILE_ENABLE
597 FLEX_PARSER_PROFILE_ENABLE 2
599 - IP-in-IP tunnel enable
601 .. code-block:: console
603 mlxconfig -d <mst device> set FLEX_PARSER_PROFILE_ENABLE=0
605 Verify configurations are set:
607 .. code-block:: console
609 mlxconfig -d <mst device> query | grep FLEX_PARSER_PROFILE_ENABLE
610 FLEX_PARSER_PROFILE_ENABLE 0
615 This driver relies on external libraries and kernel drivers for resources
616 allocations and initialization. The following dependencies are not part of
617 DPDK and must be installed separately:
621 User space Verbs framework used by librte_pmd_mlx5. This library provides
622 a generic interface between the kernel and low-level user space drivers
625 It allows slow and privileged operations (context initialization, hardware
626 resources allocations) to be managed by the kernel and fast operations to
627 never leave user space.
631 Low-level user space driver library for Mellanox
632 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
635 This library basically implements send/receive calls to the hardware
640 They provide the kernel-side Verbs API and low level device drivers that
641 manage actual hardware initialization and resources sharing with user
644 Unlike most other PMDs, these modules must remain loaded and bound to
647 - mlx5_core: hardware driver managing Mellanox
648 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
650 - mlx5_ib: InifiniBand device driver.
651 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
653 - **Firmware update**
655 Mellanox OFED/EN releases include firmware updates for
656 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
658 Because each release provides new features, these updates must be applied to
659 match the kernel modules and libraries they come with.
663 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
669 Either RDMA Core library with a recent enough Linux kernel release
670 (recommended) or Mellanox OFED/EN, which provides compatibility with older
673 RDMA Core with Linux Kernel
674 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
676 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
677 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
678 (see `RDMA Core installation documentation`_)
679 - When building for i686 use:
681 - rdma-core version 18.0 or above built with 32bit support.
682 - Kernel version 4.14.41 or above.
684 - Starting with rdma-core v21, static libraries can be built::
687 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
690 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
691 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
693 If rdma-core libraries are built but not installed, DPDK makefile can link them,
694 thanks to these environment variables:
696 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
697 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
698 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
703 - Mellanox OFED version: ** 4.5, 4.6** /
704 Mellanox EN version: **4.5, 4.6**
707 - ConnectX-4: **12.21.1000** and above.
708 - ConnectX-4 Lx: **14.21.1000** and above.
709 - ConnectX-5: **16.21.1000** and above.
710 - ConnectX-5 Ex: **16.21.1000** and above.
711 - ConnectX-6: **20.99.5374** and above.
712 - BlueField: **18.25.1010** and above.
714 While these libraries and kernel modules are available on OpenFabrics
715 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
716 managers on most distributions, this PMD requires Ethernet extensions that
717 may not be supported at the moment (this is a work in progress).
720 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
722 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
723 include the necessary support and should be used in the meantime. For DPDK,
724 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
725 required from that distribution.
729 Several versions of Mellanox OFED/EN are available. Installing the version
730 this DPDK release was developed and tested against is strongly
731 recommended. Please check the `prerequisites`_.
736 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
737 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
738 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
739 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
740 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
741 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
742 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
743 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
744 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
745 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
746 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
747 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
748 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
749 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
750 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
751 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
752 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
753 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
754 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
756 Quick Start Guide on OFED/EN
757 ----------------------------
759 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
762 2. Install the required libraries and kernel modules either by installing
763 only the required set, or by installing the entire Mellanox OFED/EN:
765 .. code-block:: console
767 ./mlnxofedinstall --upstream-libs --dpdk
769 3. Verify the firmware is the correct one:
771 .. code-block:: console
775 4. Verify all ports links are set to Ethernet:
777 .. code-block:: console
779 mlxconfig -d <mst device> query | grep LINK_TYPE
783 Link types may have to be configured to Ethernet:
785 .. code-block:: console
787 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
789 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
791 For hypervisors verify SR-IOV is enabled on the NIC:
793 .. code-block:: console
795 mlxconfig -d <mst device> query | grep SRIOV_EN
798 If needed, set enable the set the relevant fields:
800 .. code-block:: console
802 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
803 mlxfwreset -d <mst device> reset
805 5. Restart the driver:
807 .. code-block:: console
809 /etc/init.d/openibd restart
813 .. code-block:: console
815 service openibd restart
817 If link type was changed, firmware must be reset as well:
819 .. code-block:: console
821 mlxfwreset -d <mst device> reset
823 For hypervisors, after reset write the sysfs number of virtual functions
826 To dynamically instantiate a given number of virtual functions (VFs):
828 .. code-block:: console
830 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
832 6. Compile DPDK and you are ready to go. See instructions on
833 :ref:`Development Kit Build System <Development_Kit_Build_System>`
835 Enable switchdev mode
836 ---------------------
838 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
839 Representor is a port in DPDK that is connected to a VF in such a way
840 that assuming there are no offload flows, each packet that is sent from the VF
841 will be received by the corresponding representor. While each packet that is
842 sent to a representor will be received by the VF.
843 This is very useful in case of SRIOV mode, where the first packet that is sent
844 by the VF will be received by the DPDK application which will decide if this
845 flow should be offloaded to the E-Switch. After offloading the flow packet
846 that the VF that are matching the flow will not be received any more by
847 the DPDK application.
849 1. Enable SRIOV mode:
851 .. code-block:: console
853 mlxconfig -d <mst device> set SRIOV_EN=true
855 2. Configure the max number of VFs:
857 .. code-block:: console
859 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
863 .. code-block:: console
865 mlxfwreset -d <mst device> reset
867 3. Configure the actual number of VFs:
869 .. code-block:: console
871 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
873 4. Unbind the device (can be rebind after the switchdev mode):
875 .. code-block:: console
877 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
879 5. Enbale switchdev mode:
881 .. code-block:: console
883 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
888 1. Configure aggressive CQE Zipping for maximum performance:
890 .. code-block:: console
892 mlxconfig -d <mst device> s CQE_COMPRESSION=1
894 To set it back to the default CQE Zipping mode use:
896 .. code-block:: console
898 mlxconfig -d <mst device> s CQE_COMPRESSION=0
900 2. In case of virtualization:
902 - Make sure that hypervisor kernel is 3.16 or newer.
903 - Configure boot with ``iommu=pt``.
905 - Make sure to allocate a VM on huge pages.
906 - Make sure to set CPU pinning.
908 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
909 for better performance. For VMs, verify that the right CPU
910 and NUMA node are pinned according to the above. Run:
912 .. code-block:: console
916 to identify the NUMA node to which the PCIe adapter is connected.
918 4. If more than one adapter is used, and root complex capabilities allow
919 to put both adapters on the same NUMA node without PCI bandwidth degradation,
920 it is recommended to locate both adapters on the same NUMA node.
921 This in order to forward packets from one to the other without
922 NUMA performance penalty.
924 5. Disable pause frames:
926 .. code-block:: console
928 ethtool -A <netdev> rx off tx off
930 6. Verify IO non-posted prefetch is disabled by default. This can be checked
931 via the BIOS configuration. Please contact you server provider for more
932 information about the settings.
936 On some machines, depends on the machine integrator, it is beneficial
937 to set the PCI max read request parameter to 1K. This can be
938 done in the following way:
940 To query the read request size use:
942 .. code-block:: console
944 setpci -s <NIC PCI address> 68.w
946 If the output is different than 3XXX, set it by:
948 .. code-block:: console
950 setpci -s <NIC PCI address> 68.w=3XXX
952 The XXX can be different on different systems. Make sure to configure
953 according to the setpci output.
955 7. To minimize overhead of searching Memory Regions:
957 - '--socket-mem' is recommended to pin memory by predictable amount.
958 - Configure per-lcore cache when creating Mempools for packet buffer.
959 - Refrain from dynamically allocating/freeing memory in run-time.
961 Supported hardware offloads using rte_flow API
962 ----------------------------------------------
964 .. _Supported hardware offloads using rte_flow API:
966 .. table:: Supported hardware offloads using rte_flow API
968 +-----------------------+-----------------+-----------------+
969 | Offload | E-Switch | NIC |
971 +=======================+=================+=================+
972 | Count | | DPDK 19.05 | | DPDK 19.02 |
973 | | | OFED 4.6 | | OFED 4.6 |
974 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
975 | | | ConnectX-5 | | ConnectX-5 |
976 +-----------------------+-----------------+-----------------+
977 | Drop / Queue / RSS | | DPDK 19.05 | | DPDK 18.11 |
978 | | | OFED 4.6 | | OFED 4.5 |
979 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
980 | | | ConnectX-5 | | ConnectX-4 |
981 +-----------------------+-----------------+-----------------+
982 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
983 | (VXLAN / NVGRE / RAW) | | OFED 4.6.2 | | OFED 4.6 |
984 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
985 | | | ConnectX-5 | | ConnectX-5 |
986 +-----------------------+-----------------+-----------------+
987 | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
988 | (set_ipv4_src / | | OFED 4.6.2 | | OFED 4.6.2 |
989 | set_ipv4_dst / | | RDMA-CORE V24 | | RDMA-CORE V23 |
990 | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
991 | set_ipv6_dst / | | |
996 | set_mac_src / | | |
998 +-----------------------+-----------------+-----------------+
999 | Jump | | DPDK 19.05 | | DPDK 19.02 |
1000 | | | OFED 4.6.2 | | OFED 4.6.2 |
1001 | | | RDMA-CORE V24 | | N/A |
1002 | | | ConnectX-5 | | ConnectX-5 |
1003 +-----------------------+-----------------+-----------------+
1004 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
1005 | | | OFED 4.6 | | OFED 4.5 |
1006 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
1007 | | | ConnectX-5 | | ConnectX-4 |
1008 +-----------------------+-----------------+-----------------+
1009 | Port ID | | DPDK 19.05 | | N/A |
1010 | | | OFED 4.6 | | N/A |
1011 | | | RDMA-CORE V24 | | N/A |
1012 | | | ConnectX-5 | | N/A |
1013 +-----------------------+-----------------+-----------------+
1015 * Minimum version for each component and nic.
1020 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
1021 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
1023 Since ``testpmd`` defaults to IP RSS mode and there is currently no
1024 command-line parameter to enable additional protocols (UDP and TCP as well
1025 as IP), the following commands must be entered from its CLI to get the same
1026 behavior as librte_pmd_mlx4:
1028 .. code-block:: console
1031 > port config all rss all
1037 This section demonstrates how to launch **testpmd** with Mellanox
1038 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
1040 #. Load the kernel modules:
1042 .. code-block:: console
1044 modprobe -a ib_uverbs mlx5_core mlx5_ib
1046 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
1049 .. code-block:: console
1051 /etc/init.d/openibd restart
1055 User space I/O kernel modules (uio and igb_uio) are not used and do
1056 not have to be loaded.
1058 #. Make sure Ethernet interfaces are in working order and linked to kernel
1059 verbs. Related sysfs entries should be present:
1061 .. code-block:: console
1063 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
1067 .. code-block:: console
1074 #. Optionally, retrieve their PCI bus addresses for whitelisting:
1076 .. code-block:: console
1079 for intf in eth2 eth3 eth4 eth5;
1081 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1084 sed -n 's,.*/\(.*\),-w \1,p'
1088 .. code-block:: console
1095 #. Request huge pages:
1097 .. code-block:: console
1099 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1101 #. Start testpmd with basic parameters:
1103 .. code-block:: console
1105 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
1109 .. code-block:: console
1112 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1113 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1114 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1115 PMD: librte_pmd_mlx5: 1 port(s) detected
1116 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1117 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1118 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1119 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1120 PMD: librte_pmd_mlx5: 1 port(s) detected
1121 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1122 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1123 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1124 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1125 PMD: librte_pmd_mlx5: 1 port(s) detected
1126 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1127 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1128 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1129 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1130 PMD: librte_pmd_mlx5: 1 port(s) detected
1131 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1132 Interactive-mode selected
1133 Configuring Port 0 (socket 0)
1134 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1135 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1136 Port 0: E4:1D:2D:E7:0C:FE
1137 Configuring Port 1 (socket 0)
1138 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1139 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1140 Port 1: E4:1D:2D:E7:0C:FF
1141 Configuring Port 2 (socket 0)
1142 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1143 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1144 Port 2: E4:1D:2D:E7:0C:FA
1145 Configuring Port 3 (socket 0)
1146 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1147 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1148 Port 3: E4:1D:2D:E7:0C:FB
1149 Checking link statuses...
1150 Port 0 Link Up - speed 40000 Mbps - full-duplex
1151 Port 1 Link Up - speed 40000 Mbps - full-duplex
1152 Port 2 Link Up - speed 10000 Mbps - full-duplex
1153 Port 3 Link Up - speed 10000 Mbps - full-duplex