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.
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.
89 - For secondary process:
91 - Forked secondary process not supported.
92 - External memory unregistered in EAL memseg list cannot be used for DMA
93 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
94 primary process and remapped to the same virtual address in secondary
95 process. If the external memory is registered by primary process but has
96 different virtual address in secondary process, unexpected error may happen.
98 - Flow pattern without any specific vlan will match for vlan packets as well:
100 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
101 Meaning, the flow rule::
103 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
105 Will only match vlan packets with vid=3. and the flow rules::
107 flow create 0 ingress pattern eth / ipv4 / end ...
111 flow create 0 ingress pattern eth / vlan / ipv4 / end ...
113 Will match any ipv4 packet (VLAN included).
115 - A multi segment packet must have less than 6 segments in case the Tx burst function
116 is set to multi-packet send or Enhanced multi-packet send. Otherwise it must have
117 less than 50 segments.
119 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
120 to 0 are not supported.
122 - VXLAN TSO and checksum offloads are not supported on VM.
124 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
126 - VF: flow rules created on VF devices can only match traffic targeted at the
127 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
131 MAC addresses not already present in the bridge table of the associated
132 kernel network device will be added and cleaned up by the PMD when closing
133 the device. In case of ungraceful program termination, some entries may
134 remain present and should be removed manually by other means.
136 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
137 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
138 ol_flags. As the mempool for the external buffer is managed by PMD, all the
139 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
140 the external buffers will be freed by PMD and the application which still
141 holds the external buffers may be corrupted.
143 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
144 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
145 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
147 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
148 and allmulticast mode are both set to off.
149 To receive IPv6 Multicast messages on VM, explicitly set the relevant
150 MAC address using rte_eth_dev_mac_addr_add() API.
152 - E-Switch decapsulation Flow:
154 - can be applied to PF port only.
155 - must specify VF port action (packet redirection from PF to VF).
156 - optionally may specify tunnel inner source and destination MAC addresses.
158 - E-Switch encapsulation Flow:
160 - can be applied to VF ports only.
161 - must specify PF port action (packet redirection from VF to PF).
166 MLX5 supports various of methods to report statistics:
168 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.
170 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.
172 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.
180 These options can be modified in the ``.config`` file.
182 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
184 Toggle compilation of librte_pmd_mlx5 itself.
186 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
188 Build PMD with additional code to make it loadable without hard
189 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
190 on the target system.
192 In this mode, their presence is still required for it to run properly,
193 however their absence won't prevent a DPDK application from starting (with
194 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
195 missing with ``ldd(1)``.
197 It works by moving these dependencies to a purpose-built rdma-core "glue"
198 plug-in which must either be installed in a directory whose name is based
199 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
200 standard location for the dynamic linker (e.g. ``/lib``) if left to the
201 default empty string (``""``).
203 This option has no performance impact.
205 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
207 Embed static flavor of the dependencies **libibverbs** and **libmlx5**
208 in the PMD shared library or the executable static binary.
210 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
212 Toggle debugging code and stricter compilation flags. Enabling this option
213 adds additional run-time checks and debugging messages at the cost of
218 For BlueField, target should be set to ``arm64-bluefield-linux-gcc``. This
219 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
220 64. Default armv8a configuration of make build and meson build set it to 128
221 then brings performance degradation.
223 Environment variables
224 ~~~~~~~~~~~~~~~~~~~~~
228 A list of directories in which to search for the rdma-core "glue" plug-in,
229 separated by colons or semi-colons.
231 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
232 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
233 since ``LD_LIBRARY_PATH`` has no effect in this case.
235 - ``MLX5_SHUT_UP_BF``
237 Configures HW Tx doorbell register as IO-mapped.
239 By default, the HW Tx doorbell is configured as a write-combining register.
240 The register would be flushed to HW usually when the write-combining buffer
241 becomes full, but it depends on CPU design.
243 Except for vectorized Tx burst routines, a write memory barrier is enforced
244 after updating the register so that the update can be immediately visible to
247 When vectorized Tx burst is called, the barrier is set only if the burst size
248 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
249 variable will bring better latency even though the maximum throughput can
252 Run-time configuration
253 ~~~~~~~~~~~~~~~~~~~~~~
255 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
256 because it is affected by their state. Forcing them down prevents packets
259 - **ethtool** operations on related kernel interfaces also affect the PMD.
261 - ``rxq_cqe_comp_en`` parameter [int]
263 A nonzero value enables the compression of CQE on RX side. This feature
264 allows to save PCI bandwidth and improve performance. Enabled by default.
268 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
269 - POWER9 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
271 - ``rxq_cqe_pad_en`` parameter [int]
273 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
274 is aligned with the size of a cacheline of the core. If cacheline size is
275 128B, the CQE size is configured to be 128B even though the device writes
276 only 64B data on the cacheline. This is to avoid unnecessary cache
277 invalidation by device's two consecutive writes on to one cacheline.
278 However in some architecture, it is more beneficial to update entire
279 cacheline with padding the rest 64B rather than striding because
280 read-modify-write could drop performance a lot. On the other hand,
281 writing extra data will consume more PCIe bandwidth and could also drop
282 the maximum throughput. It is recommended to empirically set this
283 parameter. Disabled by default.
287 - CPU having 128B cacheline with ConnectX-5 and BlueField.
289 - ``rxq_pkt_pad_en`` parameter [int]
291 A nonzero value enables padding Rx packet to the size of cacheline on PCI
292 transaction. This feature would waste PCI bandwidth but could improve
293 performance by avoiding partial cacheline write which may cause costly
294 read-modify-copy in memory transaction on some architectures. Disabled by
299 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
300 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField.
302 - ``mprq_en`` parameter [int]
304 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
305 configured as Multi-Packet RQ if the total number of Rx queues is
306 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
309 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
310 by posting a single large buffer for multiple packets. Instead of posting a
311 buffers per a packet, one large buffer is posted in order to receive multiple
312 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
313 and each stride receives one packet. MPRQ can improve throughput for
314 small-packet traffic.
316 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
317 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
318 configure large stride size enough to accommodate max_rx_pkt_len as long as
319 device allows. Note that this can waste system memory compared to enabling Rx
320 scatter and multi-segment packet.
322 - ``mprq_log_stride_num`` parameter [int]
324 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
325 strides can reduce PCIe traffic further. If configured value is not in the
326 range of device capability, the default value will be set with a warning
327 message. The default value is 4 which is 16 strides per a buffer, valid only
328 if ``mprq_en`` is set.
330 The size of Rx queue should be bigger than the number of strides.
332 - ``mprq_max_memcpy_len`` parameter [int]
334 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
335 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
336 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
337 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
338 A mempool for external buffers will be allocated and managed by PMD. If Rx
339 packet is externally attached, ol_flags field of the mbuf will have
340 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
341 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
343 - ``rxqs_min_mprq`` parameter [int]
345 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
346 greater or equal to this value. The default value is 12, valid only if
349 - ``txq_inline`` parameter [int]
351 Amount of data to be inlined during TX operations. Improves latency.
352 Can improve PPS performance when PCI back pressure is detected and may be
353 useful for scenarios involving heavy traffic on many queues.
355 Because additional software logic is necessary to handle this mode, this
356 option should be used with care, as it can lower performance when back
357 pressure is not expected.
359 - ``txqs_min_inline`` parameter [int]
361 Enable inline send only when the number of TX queues is greater or equal
364 This option should be used in combination with ``txq_inline`` above.
366 On ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and BlueField without
369 - Disabled by default.
370 - In case ``txq_inline`` is set recommendation is 4.
372 On ConnectX-5, ConnectX-6 and BlueField with Enhanced MPW:
374 - Set to 8 by default.
376 - ``txqs_max_vec`` parameter [int]
378 Enable vectorized Tx only when the number of TX queues is less than or
379 equal to this value. Effective only when ``tx_vec_en`` is enabled.
383 - Set to 8 by default on ARMv8.
384 - Set to 4 by default otherwise.
388 - Set to 16 by default.
390 - ``txq_mpw_en`` parameter [int]
392 A nonzero value enables multi-packet send (MPS) for ConnectX-4 Lx and
393 enhanced multi-packet send (Enhanced MPS) for ConnectX-5, ConnectX-6 and BlueField.
394 MPS allows the TX burst function to pack up multiple packets in a
395 single descriptor session in order to save PCI bandwidth and improve
396 performance at the cost of a slightly higher CPU usage. When
397 ``txq_inline`` is set along with ``txq_mpw_en``, TX burst function tries
398 to copy entire packet data on to TX descriptor instead of including
399 pointer of packet only if there is enough room remained in the
400 descriptor. ``txq_inline`` sets per-descriptor space for either pointers
401 or inlined packets. In addition, Enhanced MPS supports hybrid mode -
402 mixing inlined packets and pointers in the same descriptor.
404 This option cannot be used with certain offloads such as ``DEV_TX_OFFLOAD_TCP_TSO,
405 DEV_TX_OFFLOAD_VXLAN_TNL_TSO, DEV_TX_OFFLOAD_GRE_TNL_TSO, DEV_TX_OFFLOAD_VLAN_INSERT``.
406 When those offloads are requested the MPS send function will not be used.
408 It is currently only supported on the ConnectX-4 Lx, ConnectX-5, ConnectX-6 and BlueField
409 families of adapters.
410 On ConnectX-4 Lx the MPW is considered un-secure hence disabled by default.
411 Users which enable the MPW should be aware that application which provides incorrect
412 mbuf descriptors in the Tx burst can lead to serious errors in the host including, on some cases,
414 On ConnectX-5, ConnectX-6 and BlueField the MPW is secure and enabled by default.
416 - ``txq_mpw_hdr_dseg_en`` parameter [int]
418 A nonzero value enables including two pointers in the first block of TX
419 descriptor. This can be used to lessen CPU load for memory copy.
421 Effective only when Enhanced MPS is supported. Disabled by default.
423 - ``txq_max_inline_len`` parameter [int]
425 Maximum size of packet to be inlined. This limits the size of packet to
426 be inlined. If the size of a packet is larger than configured value, the
427 packet isn't inlined even though there's enough space remained in the
428 descriptor. Instead, the packet is included with pointer.
430 Effective only when Enhanced MPS is supported. The default value is 256.
432 - ``tx_vec_en`` parameter [int]
434 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6 and BlueField NICs if the number of
435 global Tx queues on the port is less than ``txqs_max_vec``.
437 This option cannot be used with certain offloads such as ``DEV_TX_OFFLOAD_TCP_TSO,
438 DEV_TX_OFFLOAD_VXLAN_TNL_TSO, DEV_TX_OFFLOAD_GRE_TNL_TSO, DEV_TX_OFFLOAD_VLAN_INSERT``.
439 When those offloads are requested the MPS send function will not be used.
441 Enabled by default on ConnectX-5, ConnectX-6 and BlueField.
443 - ``rx_vec_en`` parameter [int]
445 A nonzero value enables Rx vector if the port is not configured in
446 multi-segment otherwise this parameter is ignored.
450 - ``vf_nl_en`` parameter [int]
452 A nonzero value enables Netlink requests from the VF to add/remove MAC
453 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
454 Otherwise the relevant configuration must be run with Linux iproute2 tools.
455 This is a prerequisite to receive this kind of traffic.
457 Enabled by default, valid only on VF devices ignored otherwise.
459 - ``l3_vxlan_en`` parameter [int]
461 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
462 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
463 parameter. This is a prerequisite to receive this kind of traffic.
467 - ``dv_flow_en`` parameter [int]
469 A nonzero value enables the DV flow steering assuming it is supported
474 - ``dv_esw_en`` parameter [int]
476 A nonzero value enables E-Switch using Direct Rules.
478 Enabled by default if supported.
480 - ``mr_ext_memseg_en`` parameter [int]
482 A nonzero value enables extending memseg when registering DMA memory. If
483 enabled, the number of entries in MR (Memory Region) lookup table on datapath
484 is minimized and it benefits performance. On the other hand, it worsens memory
485 utilization because registered memory is pinned by kernel driver. Even if a
486 page in the extended chunk is freed, that doesn't become reusable until the
487 entire memory is freed.
491 - ``representor`` parameter [list]
493 This parameter can be used to instantiate DPDK Ethernet devices from
494 existing port (or VF) representors configured on the device.
496 It is a standard parameter whose format is described in
497 :ref:`ethernet_device_standard_device_arguments`.
499 For instance, to probe port representors 0 through 2::
503 Firmware configuration
504 ~~~~~~~~~~~~~~~~~~~~~~
506 - L3 VXLAN and VXLAN-GPE destination UDP port
508 .. code-block:: console
510 mlxconfig -d <mst device> set IP_OVER_VXLAN_EN=1
511 mlxconfig -d <mst device> set IP_OVER_VXLAN_PORT=<udp dport>
513 Verify configurations are set:
515 .. code-block:: console
517 mlxconfig -d <mst device> query | grep IP_OVER_VXLAN
518 IP_OVER_VXLAN_EN True(1)
519 IP_OVER_VXLAN_PORT <udp dport>
524 This driver relies on external libraries and kernel drivers for resources
525 allocations and initialization. The following dependencies are not part of
526 DPDK and must be installed separately:
530 User space Verbs framework used by librte_pmd_mlx5. This library provides
531 a generic interface between the kernel and low-level user space drivers
534 It allows slow and privileged operations (context initialization, hardware
535 resources allocations) to be managed by the kernel and fast operations to
536 never leave user space.
540 Low-level user space driver library for Mellanox
541 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices, it is automatically loaded
544 This library basically implements send/receive calls to the hardware
549 Minimalistic Netlink library mainly relied on to manage E-Switch flow
550 rules (i.e. those with the "transfer" attribute and typically involving
555 They provide the kernel-side Verbs API and low level device drivers that
556 manage actual hardware initialization and resources sharing with user
559 Unlike most other PMDs, these modules must remain loaded and bound to
562 - mlx5_core: hardware driver managing Mellanox
563 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices and related Ethernet kernel
565 - mlx5_ib: InifiniBand device driver.
566 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
568 - **Firmware update**
570 Mellanox OFED/EN releases include firmware updates for
571 ConnectX-4/ConnectX-5/ConnectX-6/BlueField adapters.
573 Because each release provides new features, these updates must be applied to
574 match the kernel modules and libraries they come with.
578 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
584 Either RDMA Core library with a recent enough Linux kernel release
585 (recommended) or Mellanox OFED/EN, which provides compatibility with older
588 RDMA Core with Linux Kernel
589 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
591 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
592 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
593 (see `RDMA Core installation documentation`_)
594 - When building for i686 use:
596 - rdma-core version 18.0 or above built with 32bit support.
597 - Kernel version 4.14.41 or above.
599 - Starting with rdma-core v21, static libraries can be built::
602 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
605 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
606 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
608 If rdma-core libraries are built but not installed, DPDK makefile can link them,
609 thanks to these environment variables:
611 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
612 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
613 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
618 - Mellanox OFED version: ** 4.5, 4.6** /
619 Mellanox EN version: **4.5, 4.6**
622 - ConnectX-4: **12.21.1000** and above.
623 - ConnectX-4 Lx: **14.21.1000** and above.
624 - ConnectX-5: **16.21.1000** and above.
625 - ConnectX-5 Ex: **16.21.1000** and above.
626 - ConnectX-6: **20.99.5374** and above.
627 - BlueField: **18.25.1010** and above.
629 While these libraries and kernel modules are available on OpenFabrics
630 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
631 managers on most distributions, this PMD requires Ethernet extensions that
632 may not be supported at the moment (this is a work in progress).
635 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
637 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
638 include the necessary support and should be used in the meantime. For DPDK,
639 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
640 required from that distribution.
644 Several versions of Mellanox OFED/EN are available. Installing the version
645 this DPDK release was developed and tested against is strongly
646 recommended. Please check the `prerequisites`_.
651 Minimal version for libmnl is **1.0.3**.
653 As a dependency of the **iproute2** suite, this library is often installed
654 by default. It is otherwise readily available through standard system
657 Its development headers must be installed in order to compile this PMD.
658 These packages are usually named **libmnl-dev** or **libmnl-devel**
659 depending on the Linux distribution.
664 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
665 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
666 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
667 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
668 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
669 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
670 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
671 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
672 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
673 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
674 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
675 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
676 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
677 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
678 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
679 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
680 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
681 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
682 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
684 Quick Start Guide on OFED/EN
685 ----------------------------
687 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
690 2. Install the required libraries and kernel modules either by installing
691 only the required set, or by installing the entire Mellanox OFED/EN:
693 .. code-block:: console
695 ./mlnxofedinstall --upstream-libs --dpdk
697 3. Verify the firmware is the correct one:
699 .. code-block:: console
703 4. Verify all ports links are set to Ethernet:
705 .. code-block:: console
707 mlxconfig -d <mst device> query | grep LINK_TYPE
711 Link types may have to be configured to Ethernet:
713 .. code-block:: console
715 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
717 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
719 For hypervisors verify SR-IOV is enabled on the NIC:
721 .. code-block:: console
723 mlxconfig -d <mst device> query | grep SRIOV_EN
726 If needed, set enable the set the relevant fields:
728 .. code-block:: console
730 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
731 mlxfwreset -d <mst device> reset
733 5. Restart the driver:
735 .. code-block:: console
737 /etc/init.d/openibd restart
741 .. code-block:: console
743 service openibd restart
745 If link type was changed, firmware must be reset as well:
747 .. code-block:: console
749 mlxfwreset -d <mst device> reset
751 For hypervisors, after reset write the sysfs number of virtual functions
754 To dynamically instantiate a given number of virtual functions (VFs):
756 .. code-block:: console
758 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
760 6. Compile DPDK and you are ready to go. See instructions on
761 :ref:`Development Kit Build System <Development_Kit_Build_System>`
763 Enable switchdev mode
764 ---------------------
766 Switchdev mode is a mode in E-Switch, that binds between representor and VF.
767 Representor is a port in DPDK that is connected to a VF in such a way
768 that assuming there are no offload flows, each packet that is sent from the VF
769 will be received by the corresponding representor. While each packet that is
770 sent to a representor will be received by the VF.
771 This is very useful in case of SRIOV mode, where the first packet that is sent
772 by the VF will be received by the DPDK application which will decide if this
773 flow should be offloaded to the E-Switch. After offloading the flow packet
774 that the VF that are matching the flow will not be received any more by
775 the DPDK application.
777 1. Enable SRIOV mode:
779 .. code-block:: console
781 mlxconfig -d <mst device> set SRIOV_EN=true
783 2. Configure the max number of VFs:
785 .. code-block:: console
787 mlxconfig -d <mst device> set NUM_OF_VFS=<num of vfs>
791 .. code-block:: console
793 mlxfwreset -d <mst device> reset
795 3. Configure the actual number of VFs:
797 .. code-block:: console
799 echo <num of vfs > /sys/class/net/<net device>/device/sriov_numvfs
801 4. Unbind the device (can be rebind after the switchdev mode):
803 .. code-block:: console
805 echo -n "<device pci address" > /sys/bus/pci/drivers/mlx5_core/unbind
807 5. Enbale switchdev mode:
809 .. code-block:: console
811 echo switchdev > /sys/class/net/<net device>/compat/devlink/mode
816 1. Configure aggressive CQE Zipping for maximum performance:
818 .. code-block:: console
820 mlxconfig -d <mst device> s CQE_COMPRESSION=1
822 To set it back to the default CQE Zipping mode use:
824 .. code-block:: console
826 mlxconfig -d <mst device> s CQE_COMPRESSION=0
828 2. In case of virtualization:
830 - Make sure that hypervisor kernel is 3.16 or newer.
831 - Configure boot with ``iommu=pt``.
833 - Make sure to allocate a VM on huge pages.
834 - Make sure to set CPU pinning.
836 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
837 for better performance. For VMs, verify that the right CPU
838 and NUMA node are pinned according to the above. Run:
840 .. code-block:: console
844 to identify the NUMA node to which the PCIe adapter is connected.
846 4. If more than one adapter is used, and root complex capabilities allow
847 to put both adapters on the same NUMA node without PCI bandwidth degradation,
848 it is recommended to locate both adapters on the same NUMA node.
849 This in order to forward packets from one to the other without
850 NUMA performance penalty.
852 5. Disable pause frames:
854 .. code-block:: console
856 ethtool -A <netdev> rx off tx off
858 6. Verify IO non-posted prefetch is disabled by default. This can be checked
859 via the BIOS configuration. Please contact you server provider for more
860 information about the settings.
864 On some machines, depends on the machine integrator, it is beneficial
865 to set the PCI max read request parameter to 1K. This can be
866 done in the following way:
868 To query the read request size use:
870 .. code-block:: console
872 setpci -s <NIC PCI address> 68.w
874 If the output is different than 3XXX, set it by:
876 .. code-block:: console
878 setpci -s <NIC PCI address> 68.w=3XXX
880 The XXX can be different on different systems. Make sure to configure
881 according to the setpci output.
883 7. To minimize overhead of searching Memory Regions:
885 - '--socket-mem' is recommended to pin memory by predictable amount.
886 - Configure per-lcore cache when creating Mempools for packet buffer.
887 - Refrain from dynamically allocating/freeing memory in run-time.
889 Supported hardware offloads using rte_flow API
890 ----------------------------------------------
892 .. _Supported hardware offloads using rte_flow API:
894 .. table:: Supported hardware offloads using rte_flow API
896 +-----------------------+-----------------+-----------------+
897 | Offload | E-Switch | NIC |
899 +=======================+=================+=================+
900 | Count | | DPDK 19.05 | | DPDK 19.02 |
901 | | | OFED 4.6 | | OFED 4.6 |
902 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
903 | | | ConnectX-5 | | ConnectX-5 |
904 +-----------------------+-----------------+-----------------+
905 | Drop / Queue / RSS | | DPDK 19.05 | | DPDK 18.11 |
906 | | | OFED 4.6 | | OFED 4.5 |
907 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
908 | | | ConnectX-5 | | ConnectX-4 |
909 +-----------------------+-----------------+-----------------+
910 | Encapsulation | | DPDK 19.05 | | DPDK 19.02 |
911 | (VXLAN / NVGRE / RAW) | | OFED 4.6.2 | | OFED 4.6 |
912 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
913 | | | ConnectX-5 | | ConnectX-5 |
914 +-----------------------+-----------------+-----------------+
915 | Header rewrite | | DPDK 19.05 | | DPDK 19.02 |
916 | (set_ipv4_src / | | OFED 4.6.2 | | OFED 4.6.2 |
917 | set_ipv4_dst / | | RDMA-CORE V24 | | RDMA-CORE V23 |
918 | set_ipv6_src / | | ConnectX-5 | | ConnectX-5 |
919 | set_ipv6_dst / | | |
924 | set_mac_src / | | |
926 +-----------------------+-----------------+-----------------+
927 | Jump | | DPDK 19.05 | | DPDK 19.02 |
928 | | | OFED 4.6.2 | | OFED 4.6.2 |
929 | | | RDMA-CORE V24 | | N/A |
930 | | | ConnectX-5 | | ConnectX-5 |
931 +-----------------------+-----------------+-----------------+
932 | Mark / Flag | | DPDK 19.05 | | DPDK 18.11 |
933 | | | OFED 4.6 | | OFED 4.5 |
934 | | | RDMA-CORE V24 | | RDMA-CORE V23 |
935 | | | ConnectX-5 | | ConnectX-4 |
936 +-----------------------+-----------------+-----------------+
937 | Port ID | | DPDK 19.05 | | N/A |
938 | | | OFED 4.6 | | N/A |
939 | | | RDMA-CORE V24 | | N/A |
940 | | | ConnectX-5 | | N/A |
941 +-----------------------+-----------------+-----------------+
943 * Minimum version for each component and nic.
948 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
949 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
951 Since ``testpmd`` defaults to IP RSS mode and there is currently no
952 command-line parameter to enable additional protocols (UDP and TCP as well
953 as IP), the following commands must be entered from its CLI to get the same
954 behavior as librte_pmd_mlx4:
956 .. code-block:: console
959 > port config all rss all
965 This section demonstrates how to launch **testpmd** with Mellanox
966 ConnectX-4/ConnectX-5/ConnectX-6/BlueField devices managed by librte_pmd_mlx5.
968 #. Load the kernel modules:
970 .. code-block:: console
972 modprobe -a ib_uverbs mlx5_core mlx5_ib
974 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
977 .. code-block:: console
979 /etc/init.d/openibd restart
983 User space I/O kernel modules (uio and igb_uio) are not used and do
984 not have to be loaded.
986 #. Make sure Ethernet interfaces are in working order and linked to kernel
987 verbs. Related sysfs entries should be present:
989 .. code-block:: console
991 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
995 .. code-block:: console
1002 #. Optionally, retrieve their PCI bus addresses for whitelisting:
1004 .. code-block:: console
1007 for intf in eth2 eth3 eth4 eth5;
1009 (cd "/sys/class/net/${intf}/device/" && pwd -P);
1012 sed -n 's,.*/\(.*\),-w \1,p'
1016 .. code-block:: console
1023 #. Request huge pages:
1025 .. code-block:: console
1027 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
1029 #. Start testpmd with basic parameters:
1031 .. code-block:: console
1033 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
1037 .. code-block:: console
1040 EAL: PCI device 0000:05:00.0 on NUMA socket 0
1041 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1042 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
1043 PMD: librte_pmd_mlx5: 1 port(s) detected
1044 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
1045 EAL: PCI device 0000:05:00.1 on NUMA socket 0
1046 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1047 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
1048 PMD: librte_pmd_mlx5: 1 port(s) detected
1049 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
1050 EAL: PCI device 0000:06:00.0 on NUMA socket 0
1051 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1052 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
1053 PMD: librte_pmd_mlx5: 1 port(s) detected
1054 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
1055 EAL: PCI device 0000:06:00.1 on NUMA socket 0
1056 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
1057 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
1058 PMD: librte_pmd_mlx5: 1 port(s) detected
1059 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
1060 Interactive-mode selected
1061 Configuring Port 0 (socket 0)
1062 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
1063 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
1064 Port 0: E4:1D:2D:E7:0C:FE
1065 Configuring Port 1 (socket 0)
1066 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
1067 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
1068 Port 1: E4:1D:2D:E7:0C:FF
1069 Configuring Port 2 (socket 0)
1070 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
1071 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
1072 Port 2: E4:1D:2D:E7:0C:FA
1073 Configuring Port 3 (socket 0)
1074 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
1075 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
1076 Port 3: E4:1D:2D:E7:0C:FB
1077 Checking link statuses...
1078 Port 0 Link Up - speed 40000 Mbps - full-duplex
1079 Port 1 Link Up - speed 40000 Mbps - full-duplex
1080 Port 2 Link Up - speed 10000 Mbps - full-duplex
1081 Port 3 Link Up - speed 10000 Mbps - full-duplex