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.
66 - Multicast promiscuous mode.
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.
85 - For secondary process:
87 - Forked secondary process not supported.
88 - All mempools must be initialized before rte_eth_dev_start().
89 - External memory unregistered in EAL memseg list cannot be used for DMA
90 unless such memory has been registered by ``mlx5_mr_update_ext_mp()`` in
91 primary process and remapped to the same virtual address in secondary
92 process. If the external memory is registered by primary process but has
93 different virtual address in secondary process, unexpected error may happen.
95 - Flow pattern without any specific vlan will match for vlan packets as well:
97 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
98 Meaning, the flow rule::
100 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
102 Will only match vlan packets with vid=3. and the flow rules::
104 flow create 0 ingress pattern eth / ipv4 / end ...
108 flow create 0 ingress pattern eth / vlan / ipv4 / end ...
110 Will match any ipv4 packet (VLAN included).
112 - A multi segment packet must have less than 6 segments in case the Tx burst function
113 is set to multi-packet send or Enhanced multi-packet send. Otherwise it must have
114 less than 50 segments.
116 - Count action for RTE flow is **only supported in Mellanox OFED**.
118 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
119 to 0 are not supported.
121 - VXLAN TSO and checksum offloads are not supported on VM.
123 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
125 - VF: flow rules created on VF devices can only match traffic targeted at the
126 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
130 MAC addresses not already present in the bridge table of the associated
131 kernel network device will be added and cleaned up by the PMD when closing
132 the device. In case of ungraceful program termination, some entries may
133 remain present and should be removed manually by other means.
135 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
136 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
137 ol_flags. As the mempool for the external buffer is managed by PMD, all the
138 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
139 the external buffers will be freed by PMD and the application which still
140 holds the external buffers may be corrupted.
142 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
143 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
144 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
146 - IPv6 Multicast messages are not supported on VM, while promiscuous mode
147 and allmulticast mode are both set to off.
148 To receive IPv6 Multicast messages on VM, explicitly set the relevant
149 MAC address using rte_eth_dev_mac_addr_add() API.
151 - E-Switch VXLAN tunnel is not supported together with outer VLAN.
153 - E-Switch Flows with VNI pattern must include the VXLAN decapsulation action.
155 - E-Switch VXLAN decapsulation Flow:
157 - can be applied to PF port only.
158 - must specify VF port action (packet redirection from PF to VF).
159 - must specify tunnel outer UDP local (destination) port, wildcards not allowed.
160 - must specify tunnel outer VNI, wildcards not allowed.
161 - must specify tunnel outer local (destination) IPv4 or IPv6 address, wildcards not allowed.
162 - optionally may specify tunnel outer remote (source) IPv4 or IPv6, wildcards or group IPs allowed.
163 - optionally may specify tunnel inner source and destination MAC addresses.
165 - E-Switch VXLAN encapsulation Flow:
167 - can be applied to VF ports only.
168 - must specify PF port action (packet redirection from VF to PF).
169 - must specify the VXLAN item with tunnel outer parameters.
170 - must specify the tunnel outer VNI in the VXLAN item.
171 - must specify the tunnel outer remote (destination) UDP port in the VXLAN item.
172 - must specify the tunnel outer local (source) IPv4 or IPv6 in the , this address will locally (with scope link) assigned to the outer network interace, wildcards not allowed.
173 - must specify the tunnel outer remote (destination) IPv4 or IPv6 in the VXLAN item, group IPs allowed.
174 - must specify the tunnel outer destination MAC address in the VXLAN item, this address will be used to create neigh rule.
179 MLX5 supports various of methods to report statistics:
181 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.
183 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.
185 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.
193 These options can be modified in the ``.config`` file.
195 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
197 Toggle compilation of librte_pmd_mlx5 itself.
199 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
201 Build PMD with additional code to make it loadable without hard
202 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
203 on the target system.
205 In this mode, their presence is still required for it to run properly,
206 however their absence won't prevent a DPDK application from starting (with
207 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
208 missing with ``ldd(1)``.
210 It works by moving these dependencies to a purpose-built rdma-core "glue"
211 plug-in which must either be installed in a directory whose name is based
212 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
213 standard location for the dynamic linker (e.g. ``/lib``) if left to the
214 default empty string (``""``).
216 This option has no performance impact.
218 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
220 Embed static flavour of the dependencies **libibverbs** and **libmlx5**
221 in the PMD shared library or the executable static binary.
223 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
225 Toggle debugging code and stricter compilation flags. Enabling this option
226 adds additional run-time checks and debugging messages at the cost of
231 For Bluefield, target should be set to ``arm64-bluefield-linux-gcc``. This
232 will enable ``CONFIG_RTE_LIBRTE_MLX5_PMD`` and set ``RTE_CACHE_LINE_SIZE`` to
233 64. Default armv8a configuration of make build and meson build set it to 128
234 then brings performance degradation.
236 Environment variables
237 ~~~~~~~~~~~~~~~~~~~~~
241 A list of directories in which to search for the rdma-core "glue" plug-in,
242 separated by colons or semi-colons.
244 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
245 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
246 since ``LD_LIBRARY_PATH`` has no effect in this case.
248 - ``MLX5_SHUT_UP_BF``
250 Configures HW Tx doorbell register as IO-mapped.
252 By default, the HW Tx doorbell is configured as a write-combining register.
253 The register would be flushed to HW usually when the write-combining buffer
254 becomes full, but it depends on CPU design.
256 Except for vectorized Tx burst routines, a write memory barrier is enforced
257 after updating the register so that the update can be immediately visible to
260 When vectorized Tx burst is called, the barrier is set only if the burst size
261 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
262 variable will bring better latency even though the maximum throughput can
265 Run-time configuration
266 ~~~~~~~~~~~~~~~~~~~~~~
268 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
269 because it is affected by their state. Forcing them down prevents packets
272 - **ethtool** operations on related kernel interfaces also affect the PMD.
274 - ``rxq_cqe_comp_en`` parameter [int]
276 A nonzero value enables the compression of CQE on RX side. This feature
277 allows to save PCI bandwidth and improve performance. Enabled by default.
281 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and Bluefield.
282 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and Bluefield.
284 - ``rxq_cqe_pad_en`` parameter [int]
286 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
287 is aligned with the size of a cacheline of the core. If cacheline size is
288 128B, the CQE size is configured to be 128B even though the device writes
289 only 64B data on the cacheline. This is to avoid unnecessary cache
290 invalidation by device's two consecutive writes on to one cacheline.
291 However in some architecture, it is more beneficial to update entire
292 cacheline with padding the rest 64B rather than striding because
293 read-modify-write could drop performance a lot. On the other hand,
294 writing extra data will consume more PCIe bandwidth and could also drop
295 the maximum throughput. It is recommended to empirically set this
296 parameter. Disabled by default.
300 - CPU having 128B cacheline with ConnectX-5 and Bluefield.
302 - ``rxq_pkt_pad_en`` parameter [int]
304 A nonzero value enables padding Rx packet to the size of cacheline on PCI
305 transaction. This feature would waste PCI bandwidth but could improve
306 performance by avoiding partial cacheline write which may cause costly
307 read-modify-copy in memory transaction on some architectures. Disabled by
312 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and Bluefield.
313 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5, ConnectX-6 and Bluefield.
315 - ``mprq_en`` parameter [int]
317 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
318 configured as Multi-Packet RQ if the total number of Rx queues is
319 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
322 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
323 by posting a single large buffer for multiple packets. Instead of posting a
324 buffers per a packet, one large buffer is posted in order to receive multiple
325 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
326 and each stride receives one packet. MPRQ can improve throughput for
327 small-packet traffic.
329 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
330 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
331 configure large stride size enough to accommodate max_rx_pkt_len as long as
332 device allows. Note that this can waste system memory compared to enabling Rx
333 scatter and multi-segment packet.
335 - ``mprq_log_stride_num`` parameter [int]
337 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
338 strides can reduce PCIe traffic further. If configured value is not in the
339 range of device capability, the default value will be set with a warning
340 message. The default value is 4 which is 16 strides per a buffer, valid only
341 if ``mprq_en`` is set.
343 The size of Rx queue should be bigger than the number of strides.
345 - ``mprq_max_memcpy_len`` parameter [int]
347 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
348 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
349 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
350 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
351 A mempool for external buffers will be allocated and managed by PMD. If Rx
352 packet is externally attached, ol_flags field of the mbuf will have
353 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
354 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
356 - ``rxqs_min_mprq`` parameter [int]
358 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
359 greater or equal to this value. The default value is 12, valid only if
362 - ``txq_inline`` parameter [int]
364 Amount of data to be inlined during TX operations. Improves latency.
365 Can improve PPS performance when PCI back pressure is detected and may be
366 useful for scenarios involving heavy traffic on many queues.
368 Because additional software logic is necessary to handle this mode, this
369 option should be used with care, as it can lower performance when back
370 pressure is not expected.
372 - ``txqs_min_inline`` parameter [int]
374 Enable inline send only when the number of TX queues is greater or equal
377 This option should be used in combination with ``txq_inline`` above.
379 On ConnectX-4, ConnectX-4 LX, ConnectX-5, ConnectX-6 and Bluefield without
382 - Disabled by default.
383 - In case ``txq_inline`` is set recommendation is 4.
385 On ConnectX-5, ConnectX-6 and Bluefield with Enhanced MPW:
387 - Set to 8 by default.
389 - ``txqs_max_vec`` parameter [int]
391 Enable vectorized Tx only when the number of TX queues is less than or
392 equal to this value. Effective only when ``tx_vec_en`` is enabled.
396 - Set to 8 by default on ARMv8.
397 - Set to 4 by default otherwise.
401 - Set to 16 by default.
403 - ``txq_mpw_en`` parameter [int]
405 A nonzero value enables multi-packet send (MPS) for ConnectX-4 Lx and
406 enhanced multi-packet send (Enhanced MPS) for ConnectX-5, ConnectX-6 and Bluefield.
407 MPS allows the TX burst function to pack up multiple packets in a
408 single descriptor session in order to save PCI bandwidth and improve
409 performance at the cost of a slightly higher CPU usage. When
410 ``txq_inline`` is set along with ``txq_mpw_en``, TX burst function tries
411 to copy entire packet data on to TX descriptor instead of including
412 pointer of packet only if there is enough room remained in the
413 descriptor. ``txq_inline`` sets per-descriptor space for either pointers
414 or inlined packets. In addition, Enhanced MPS supports hybrid mode -
415 mixing inlined packets and pointers in the same descriptor.
417 This option cannot be used with certain offloads such as ``DEV_TX_OFFLOAD_TCP_TSO,
418 DEV_TX_OFFLOAD_VXLAN_TNL_TSO, DEV_TX_OFFLOAD_GRE_TNL_TSO, DEV_TX_OFFLOAD_VLAN_INSERT``.
419 When those offloads are requested the MPS send function will not be used.
421 It is currently only supported on the ConnectX-4 Lx, ConnectX-5, ConnectX-6 and Bluefield
422 families of adapters.
423 On ConnectX-4 Lx the MPW is considered un-secure hence disabled by default.
424 Users which enable the MPW should be aware that application which provides incorrect
425 mbuf descriptors in the Tx burst can lead to serious errors in the host including, on some cases,
427 On ConnectX-5, ConnectX-6 and Bluefield the MPW is secure and enabled by default.
429 - ``txq_mpw_hdr_dseg_en`` parameter [int]
431 A nonzero value enables including two pointers in the first block of TX
432 descriptor. This can be used to lessen CPU load for memory copy.
434 Effective only when Enhanced MPS is supported. Disabled by default.
436 - ``txq_max_inline_len`` parameter [int]
438 Maximum size of packet to be inlined. This limits the size of packet to
439 be inlined. If the size of a packet is larger than configured value, the
440 packet isn't inlined even though there's enough space remained in the
441 descriptor. Instead, the packet is included with pointer.
443 Effective only when Enhanced MPS is supported. The default value is 256.
445 - ``tx_vec_en`` parameter [int]
447 A nonzero value enables Tx vector on ConnectX-5, ConnectX-6 and Bluefield NICs if the number of
448 global Tx queues on the port is less than ``txqs_max_vec``.
450 This option cannot be used with certain offloads such as ``DEV_TX_OFFLOAD_TCP_TSO,
451 DEV_TX_OFFLOAD_VXLAN_TNL_TSO, DEV_TX_OFFLOAD_GRE_TNL_TSO, DEV_TX_OFFLOAD_VLAN_INSERT``.
452 When those offloads are requested the MPS send function will not be used.
454 Enabled by default on ConnectX-5, ConnectX-6 and Bluefield.
456 - ``rx_vec_en`` parameter [int]
458 A nonzero value enables Rx vector if the port is not configured in
459 multi-segment otherwise this parameter is ignored.
463 - ``vf_nl_en`` parameter [int]
465 A nonzero value enables Netlink requests from the VF to add/remove MAC
466 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
467 Otherwise the relevant configuration must be run with Linux iproute2 tools.
468 This is a prerequisite to receive this kind of traffic.
470 Enabled by default, valid only on VF devices ignored otherwise.
472 - ``l3_vxlan_en`` parameter [int]
474 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
475 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
476 parameter. This is a prerequisite to receive this kind of traffic.
480 - ``dv_flow_en`` parameter [int]
482 A nonzero value enables the DV flow steering assuming it is supported
484 The DV flow steering is not supported on switchdev mode.
488 - ``representor`` parameter [list]
490 This parameter can be used to instantiate DPDK Ethernet devices from
491 existing port (or VF) representors configured on the device.
493 It is a standard parameter whose format is described in
494 :ref:`ethernet_device_standard_device_arguments`.
496 For instance, to probe port representors 0 through 2::
500 Firmware configuration
501 ~~~~~~~~~~~~~~~~~~~~~~
503 - L3 VXLAN and VXLAN-GPE destination UDP port
505 .. code-block:: console
507 mlxconfig -d <mst device> set IP_OVER_VXLAN_EN=1
508 mlxconfig -d <mst device> set IP_OVER_VXLAN_PORT=<udp dport>
510 Verify configurations are set:
512 .. code-block:: console
514 mlxconfig -d <mst device> query | grep IP_OVER_VXLAN
515 IP_OVER_VXLAN_EN True(1)
516 IP_OVER_VXLAN_PORT <udp dport>
521 This driver relies on external libraries and kernel drivers for resources
522 allocations and initialization. The following dependencies are not part of
523 DPDK and must be installed separately:
527 User space Verbs framework used by librte_pmd_mlx5. This library provides
528 a generic interface between the kernel and low-level user space drivers
531 It allows slow and privileged operations (context initialization, hardware
532 resources allocations) to be managed by the kernel and fast operations to
533 never leave user space.
537 Low-level user space driver library for Mellanox
538 ConnectX-4/ConnectX-5/ConnectX-6/Bluefield devices, it is automatically loaded
541 This library basically implements send/receive calls to the hardware
546 Minimalistic Netlink library mainly relied on to manage E-Switch flow
547 rules (i.e. those with the "transfer" attribute and typically involving
552 They provide the kernel-side Verbs API and low level device drivers that
553 manage actual hardware initialization and resources sharing with user
556 Unlike most other PMDs, these modules must remain loaded and bound to
559 - mlx5_core: hardware driver managing Mellanox
560 ConnectX-4/ConnectX-5/ConnectX-6/Bluefield devices and related Ethernet kernel
562 - mlx5_ib: InifiniBand device driver.
563 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
565 - **Firmware update**
567 Mellanox OFED/EN releases include firmware updates for
568 ConnectX-4/ConnectX-5/ConnectX-6/Bluefield adapters.
570 Because each release provides new features, these updates must be applied to
571 match the kernel modules and libraries they come with.
575 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
581 Either RDMA Core library with a recent enough Linux kernel release
582 (recommended) or Mellanox OFED/EN, which provides compatibility with older
585 RMDA Core with Linux Kernel
586 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
588 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
589 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
590 (see `RDMA Core installation documentation`_)
591 - When building for i686 use:
593 - rdma-core version 18.0 or above built with 32bit support.
594 - Kernel version 4.14.41 or above.
596 - Starting with rdma-core v21, static libraries can be built::
599 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
602 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
603 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
605 If rdma-core libraries are built but not installed, DPDK makefile can link them,
606 thanks to these environment variables:
608 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
609 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
610 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
615 - Mellanox OFED version: **4.4, 4.5** / Mellanox EN version: **4.5**
618 - ConnectX-4: **12.21.1000** and above.
619 - ConnectX-4 Lx: **14.21.1000** and above.
620 - ConnectX-5: **16.21.1000** and above.
621 - ConnectX-5 Ex: **16.21.1000** and above.
622 - ConnectX-6: **20.99.5374** and above.
623 - Bluefield: **18.99.3950** and above.
625 While these libraries and kernel modules are available on OpenFabrics
626 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
627 managers on most distributions, this PMD requires Ethernet extensions that
628 may not be supported at the moment (this is a work in progress).
631 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__ and
633 <http://www.mellanox.com/page/products_dyn?product_family=27&mtag=linux>`__
634 include the necessary support and should be used in the meantime. For DPDK,
635 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
636 required from that distribution.
640 Several versions of Mellanox OFED/EN are available. Installing the version
641 this DPDK release was developed and tested against is strongly
642 recommended. Please check the `prerequisites`_.
647 Minimal version for libmnl is **1.0.3**.
649 As a dependency of the **iproute2** suite, this library is often installed
650 by default. It is otherwise readily available through standard system
653 Its development headers must be installed in order to compile this PMD.
654 These packages are usually named **libmnl-dev** or **libmnl-devel**
655 depending on the Linux distribution.
660 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
661 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
662 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
663 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
664 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
665 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
666 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
667 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
668 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
669 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
670 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
671 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
672 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
673 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
674 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
675 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
676 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
677 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
678 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
680 Quick Start Guide on OFED/EN
681 ----------------------------
683 1. Download latest Mellanox OFED/EN. For more info check the `prerequisites`_.
686 2. Install the required libraries and kernel modules either by installing
687 only the required set, or by installing the entire Mellanox OFED/EN:
689 .. code-block:: console
691 ./mlnxofedinstall --upstream-libs --dpdk
693 3. Verify the firmware is the correct one:
695 .. code-block:: console
699 4. Verify all ports links are set to Ethernet:
701 .. code-block:: console
703 mlxconfig -d <mst device> query | grep LINK_TYPE
707 Link types may have to be configured to Ethernet:
709 .. code-block:: console
711 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
713 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
715 For hypervisors verify SR-IOV is enabled on the NIC:
717 .. code-block:: console
719 mlxconfig -d <mst device> query | grep SRIOV_EN
722 If needed, set enable the set the relevant fields:
724 .. code-block:: console
726 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
727 mlxfwreset -d <mst device> reset
729 5. Restart the driver:
731 .. code-block:: console
733 /etc/init.d/openibd restart
737 .. code-block:: console
739 service openibd restart
741 If link type was changed, firmware must be reset as well:
743 .. code-block:: console
745 mlxfwreset -d <mst device> reset
747 For hypervisors, after reset write the sysfs number of virtual functions
750 To dynamically instantiate a given number of virtual functions (VFs):
752 .. code-block:: console
754 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
756 6. Compile DPDK and you are ready to go. See instructions on
757 :ref:`Development Kit Build System <Development_Kit_Build_System>`
762 1. Configure aggressive CQE Zipping for maximum performance:
764 .. code-block:: console
766 mlxconfig -d <mst device> s CQE_COMPRESSION=1
768 To set it back to the default CQE Zipping mode use:
770 .. code-block:: console
772 mlxconfig -d <mst device> s CQE_COMPRESSION=0
774 2. In case of virtualization:
776 - Make sure that hypervisor kernel is 3.16 or newer.
777 - Configure boot with ``iommu=pt``.
779 - Make sure to allocate a VM on huge pages.
780 - Make sure to set CPU pinning.
782 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
783 for better performance. For VMs, verify that the right CPU
784 and NUMA node are pinned according to the above. Run:
786 .. code-block:: console
790 to identify the NUMA node to which the PCIe adapter is connected.
792 4. If more than one adapter is used, and root complex capabilities allow
793 to put both adapters on the same NUMA node without PCI bandwidth degradation,
794 it is recommended to locate both adapters on the same NUMA node.
795 This in order to forward packets from one to the other without
796 NUMA performance penalty.
798 5. Disable pause frames:
800 .. code-block:: console
802 ethtool -A <netdev> rx off tx off
804 6. Verify IO non-posted prefetch is disabled by default. This can be checked
805 via the BIOS configuration. Please contact you server provider for more
806 information about the settings.
810 On some machines, depends on the machine integrator, it is beneficial
811 to set the PCI max read request parameter to 1K. This can be
812 done in the following way:
814 To query the read request size use:
816 .. code-block:: console
818 setpci -s <NIC PCI address> 68.w
820 If the output is different than 3XXX, set it by:
822 .. code-block:: console
824 setpci -s <NIC PCI address> 68.w=3XXX
826 The XXX can be different on different systems. Make sure to configure
827 according to the setpci output.
829 7. To minimize overhead of searching Memory Regions:
831 - '--socket-mem' is recommended to pin memory by predictable amount.
832 - Configure per-lcore cache when creating Mempools for packet buffer.
833 - Refrain from dynamically allocating/freeing memory in run-time.
838 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
839 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
841 Since ``testpmd`` defaults to IP RSS mode and there is currently no
842 command-line parameter to enable additional protocols (UDP and TCP as well
843 as IP), the following commands must be entered from its CLI to get the same
844 behavior as librte_pmd_mlx4:
846 .. code-block:: console
849 > port config all rss all
855 This section demonstrates how to launch **testpmd** with Mellanox
856 ConnectX-4/ConnectX-5/ConnectX-6/Bluefield devices managed by librte_pmd_mlx5.
858 #. Load the kernel modules:
860 .. code-block:: console
862 modprobe -a ib_uverbs mlx5_core mlx5_ib
864 Alternatively if MLNX_OFED/MLNX_EN is fully installed, the following script
867 .. code-block:: console
869 /etc/init.d/openibd restart
873 User space I/O kernel modules (uio and igb_uio) are not used and do
874 not have to be loaded.
876 #. Make sure Ethernet interfaces are in working order and linked to kernel
877 verbs. Related sysfs entries should be present:
879 .. code-block:: console
881 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
885 .. code-block:: console
892 #. Optionally, retrieve their PCI bus addresses for whitelisting:
894 .. code-block:: console
897 for intf in eth2 eth3 eth4 eth5;
899 (cd "/sys/class/net/${intf}/device/" && pwd -P);
902 sed -n 's,.*/\(.*\),-w \1,p'
906 .. code-block:: console
913 #. Request huge pages:
915 .. code-block:: console
917 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
919 #. Start testpmd with basic parameters:
921 .. code-block:: console
923 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
927 .. code-block:: console
930 EAL: PCI device 0000:05:00.0 on NUMA socket 0
931 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
932 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
933 PMD: librte_pmd_mlx5: 1 port(s) detected
934 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
935 EAL: PCI device 0000:05:00.1 on NUMA socket 0
936 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
937 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
938 PMD: librte_pmd_mlx5: 1 port(s) detected
939 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
940 EAL: PCI device 0000:06:00.0 on NUMA socket 0
941 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
942 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
943 PMD: librte_pmd_mlx5: 1 port(s) detected
944 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
945 EAL: PCI device 0000:06:00.1 on NUMA socket 0
946 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
947 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
948 PMD: librte_pmd_mlx5: 1 port(s) detected
949 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
950 Interactive-mode selected
951 Configuring Port 0 (socket 0)
952 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
953 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
954 Port 0: E4:1D:2D:E7:0C:FE
955 Configuring Port 1 (socket 0)
956 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
957 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
958 Port 1: E4:1D:2D:E7:0C:FF
959 Configuring Port 2 (socket 0)
960 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
961 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
962 Port 2: E4:1D:2D:E7:0C:FA
963 Configuring Port 3 (socket 0)
964 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
965 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
966 Port 3: E4:1D:2D:E7:0C:FB
967 Checking link statuses...
968 Port 0 Link Up - speed 40000 Mbps - full-duplex
969 Port 1 Link Up - speed 40000 Mbps - full-duplex
970 Port 2 Link Up - speed 10000 Mbps - full-duplex
971 Port 3 Link Up - speed 10000 Mbps - full-duplex