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** and **Mellanox Bluefield** families of 10/25/40/50/100 Gb/s
11 adapters as well as their virtual functions (VF) in SR-IOV context.
13 Information and documentation about these adapters can be found on the
14 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
15 `Mellanox community <http://community.mellanox.com/welcome>`__.
17 There is also a `section dedicated to this poll mode driver
18 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
22 Due to external dependencies, this driver is disabled by default. It must
23 be enabled manually by setting ``CONFIG_RTE_LIBRTE_MLX5_PMD=y`` and
26 Implementation details
27 ----------------------
29 Besides its dependency on libibverbs (that implies libmlx5 and associated
30 kernel support), librte_pmd_mlx5 relies heavily on system calls for control
31 operations such as querying/updating the MTU and flow control parameters.
33 For security reasons and robustness, this driver only deals with virtual
34 memory addresses. The way resources allocations are handled by the kernel
35 combined with hardware specifications that allow it to handle virtual memory
36 addresses directly ensure that DPDK applications cannot access random
37 physical memory (or memory that does not belong to the current process).
39 This capability allows the PMD to coexist with kernel network interfaces
40 which remain functional, although they stop receiving unicast packets as
41 long as they share the same MAC address.
42 This means legacy linux control tools (for example: ethtool, ifconfig and
43 more) can operate on the same network interfaces that owned by the DPDK
46 Enabling librte_pmd_mlx5 causes DPDK applications to be linked against
52 - Multi arch support: x86_64, POWER8, ARMv8, i686.
53 - Multiple TX and RX queues.
54 - Support for scattered TX and RX frames.
55 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
56 - Several RSS hash keys, one for each flow type.
57 - Default RSS operation with no hash key specification.
58 - Configurable RETA table.
59 - Support for multiple MAC addresses.
63 - RX CRC stripping configuration.
65 - Multicast promiscuous mode.
66 - Hardware checksum offloads.
67 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
71 - KVM and VMware ESX SR-IOV modes are supported.
72 - RSS hash result is supported.
73 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
74 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
76 - Statistics query including Basic, Extended and per queue.
78 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE, MPLSoGRE, MPLSoUDP.
79 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
84 - For secondary process:
86 - Forked secondary process not supported.
87 - All mempools must be initialized before rte_eth_dev_start().
89 - Flow pattern without any specific vlan will match for vlan packets as well:
91 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
92 Meaning, the flow rule::
94 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
96 Will only match vlan packets with vid=3. and the flow rules::
98 flow create 0 ingress pattern eth / ipv4 / end ...
102 flow create 0 ingress pattern eth / vlan / ipv4 / end ...
104 Will match any ipv4 packet (VLAN included).
106 - A multi segment packet must have less than 6 segments in case the Tx burst function
107 is set to multi-packet send or Enhanced multi-packet send. Otherwise it must have
108 less than 50 segments.
110 - Count action for RTE flow is **only supported in Mellanox OFED**.
112 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
113 to 0 are not supported.
115 - VXLAN TSO and checksum offloads are not supported on VM.
117 - L3 VXLAN and VXLAN-GPE tunnels cannot be supported together with MPLSoGRE and MPLSoUDP.
119 - VF: flow rules created on VF devices can only match traffic targeted at the
120 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
124 MAC addresses not already present in the bridge table of the associated
125 kernel network device will be added and cleaned up by the PMD when closing
126 the device. In case of ungraceful program termination, some entries may
127 remain present and should be removed manually by other means.
129 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
130 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
131 ol_flags. As the mempool for the external buffer is managed by PMD, all the
132 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
133 the external buffers will be freed by PMD and the application which still
134 holds the external buffers may be corrupted.
136 - If Multi-Packet Rx queue is configured (``mprq_en``) and Rx CQE compression is
137 enabled (``rxq_cqe_comp_en``) at the same time, RSS hash result is not fully
138 supported. Some Rx packets may not have PKT_RX_RSS_HASH.
143 MLX5 supports various of methods to report statistics:
145 Port statistics can be queried using ``rte_eth_stats_get()``. The port statistics are through SW only and counts the number of packets received or sent successfully by the PMD.
147 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.
149 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.
157 These options can be modified in the ``.config`` file.
159 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
161 Toggle compilation of librte_pmd_mlx5 itself.
163 - ``CONFIG_RTE_LIBRTE_MLX5_DLOPEN_DEPS`` (default **n**)
165 Build PMD with additional code to make it loadable without hard
166 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
167 on the target system.
169 In this mode, their presence is still required for it to run properly,
170 however their absence won't prevent a DPDK application from starting (with
171 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
172 missing with ``ldd(1)``.
174 It works by moving these dependencies to a purpose-built rdma-core "glue"
175 plug-in which must either be installed in a directory whose name is based
176 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
177 standard location for the dynamic linker (e.g. ``/lib``) if left to the
178 default empty string (``""``).
180 This option has no performance impact.
182 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
184 Toggle debugging code and stricter compilation flags. Enabling this option
185 adds additional run-time checks and debugging messages at the cost of
188 Environment variables
189 ~~~~~~~~~~~~~~~~~~~~~
193 A list of directories in which to search for the rdma-core "glue" plug-in,
194 separated by colons or semi-colons.
196 Only matters when compiled with ``CONFIG_RTE_LIBRTE_MLX5_DLOPEN_DEPS``
197 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
198 since ``LD_LIBRARY_PATH`` has no effect in this case.
200 - ``MLX5_PMD_ENABLE_PADDING``
202 Enables HW packet padding in PCI bus transactions.
204 When packet size is cache aligned and CRC stripping is enabled, 4 fewer
205 bytes are written to the PCI bus. Enabling padding makes such packets
208 In cases where PCI bandwidth is the bottleneck, padding can improve
211 This is disabled by default since this can also decrease performance for
212 unaligned packet sizes.
214 - ``MLX5_SHUT_UP_BF``
216 Configures HW Tx doorbell register as IO-mapped.
218 By default, the HW Tx doorbell is configured as a write-combining register.
219 The register would be flushed to HW usually when the write-combining buffer
220 becomes full, but it depends on CPU design.
222 Except for vectorized Tx burst routines, a write memory barrier is enforced
223 after updating the register so that the update can be immediately visible to
226 When vectorized Tx burst is called, the barrier is set only if the burst size
227 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
228 variable will bring better latency even though the maximum throughput can
231 Run-time configuration
232 ~~~~~~~~~~~~~~~~~~~~~~
234 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
235 because it is affected by their state. Forcing them down prevents packets
238 - **ethtool** operations on related kernel interfaces also affect the PMD.
240 - ``rxq_cqe_comp_en`` parameter [int]
242 A nonzero value enables the compression of CQE on RX side. This feature
243 allows to save PCI bandwidth and improve performance. Enabled by default.
247 - x86_64 with ConnectX-4, ConnectX-4 LX, ConnectX-5 and Bluefield.
248 - POWER8 and ARMv8 with ConnectX-4 LX, ConnectX-5 and Bluefield.
250 - ``rxq_cqe_pad_en`` parameter [int]
252 A nonzero value enables 128B padding of CQE on RX side. The size of CQE
253 is aligned with the size of a cacheline of the core. If cacheline size is
254 128B, the CQE size is configured to be 128B even though the device writes
255 only 64B data on the cacheline. This is to avoid unnecessary cache
256 invalidation by device's two consecutive writes on to one cacheline.
257 However in some architecture, it is more beneficial to update entire
258 cacheline with padding the rest 64B rather than striding because
259 read-modify-write could drop performance a lot. On the other hand,
260 writing extra data will consume more PCIe bandwidth and could also drop
261 the maximum throughput. It is recommended to empirically set this
262 parameter. Disabled by default.
266 - CPU having 128B cacheline with ConnectX-5 and Bluefield.
268 - ``mprq_en`` parameter [int]
270 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
271 configured as Multi-Packet RQ if the total number of Rx queues is
272 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
275 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
276 by posting a single large buffer for multiple packets. Instead of posting a
277 buffers per a packet, one large buffer is posted in order to receive multiple
278 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
279 and each stride receives one packet. MPRQ can improve throughput for
280 small-packet tarffic.
282 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
283 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
284 configure large stride size enough to accommodate max_rx_pkt_len as long as
285 device allows. Note that this can waste system memory compared to enabling Rx
286 scatter and multi-segment packet.
288 - ``mprq_log_stride_num`` parameter [int]
290 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
291 strides can reduce PCIe tarffic further. If configured value is not in the
292 range of device capability, the default value will be set with a warning
293 message. The default value is 4 which is 16 strides per a buffer, valid only
294 if ``mprq_en`` is set.
296 The size of Rx queue should be bigger than the number of strides.
298 - ``mprq_max_memcpy_len`` parameter [int]
300 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
301 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
302 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
303 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
304 A mempool for external buffers will be allocated and managed by PMD. If Rx
305 packet is externally attached, ol_flags field of the mbuf will have
306 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
307 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
309 - ``rxqs_min_mprq`` parameter [int]
311 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
312 greater or equal to this value. The default value is 12, valid only if
315 - ``txq_inline`` parameter [int]
317 Amount of data to be inlined during TX operations. Improves latency.
318 Can improve PPS performance when PCI back pressure is detected and may be
319 useful for scenarios involving heavy traffic on many queues.
321 Because additional software logic is necessary to handle this mode, this
322 option should be used with care, as it can lower performance when back
323 pressure is not expected.
325 - ``txqs_min_inline`` parameter [int]
327 Enable inline send only when the number of TX queues is greater or equal
330 This option should be used in combination with ``txq_inline`` above.
332 On ConnectX-4, ConnectX-4 LX, ConnectX-5 and Bluefield without
335 - Disabled by default.
336 - In case ``txq_inline`` is set recommendation is 4.
338 On ConnectX-5 and Bluefield with Enhanced MPW:
340 - Set to 8 by default.
342 - ``txqs_max_vec`` parameter [int]
344 Enable vectorized Tx only when the number of TX queues is less than or
345 equal to this value. Effective only when ``tx_vec_en`` is enabled.
349 - Set to 8 by default on ARMv8.
350 - Set to 4 by default otherwise.
354 - Set to 16 by default.
356 - ``txq_mpw_en`` parameter [int]
358 A nonzero value enables multi-packet send (MPS) for ConnectX-4 Lx and
359 enhanced multi-packet send (Enhanced MPS) for ConnectX-5 and Bluefield.
360 MPS allows the TX burst function to pack up multiple packets in a
361 single descriptor session in order to save PCI bandwidth and improve
362 performance at the cost of a slightly higher CPU usage. When
363 ``txq_inline`` is set along with ``txq_mpw_en``, TX burst function tries
364 to copy entire packet data on to TX descriptor instead of including
365 pointer of packet only if there is enough room remained in the
366 descriptor. ``txq_inline`` sets per-descriptor space for either pointers
367 or inlined packets. In addition, Enhanced MPS supports hybrid mode -
368 mixing inlined packets and pointers in the same descriptor.
370 This option cannot be used with certain offloads such as ``DEV_TX_OFFLOAD_TCP_TSO,
371 DEV_TX_OFFLOAD_VXLAN_TNL_TSO, DEV_TX_OFFLOAD_GRE_TNL_TSO, DEV_TX_OFFLOAD_VLAN_INSERT``.
372 When those offloads are requested the MPS send function will not be used.
374 It is currently only supported on the ConnectX-4 Lx, ConnectX-5 and Bluefield
375 families of adapters.
376 On ConnectX-4 Lx the MPW is considered un-secure hence disabled by default.
377 Users which enable the MPW should be aware that application which provides incorrect
378 mbuf descriptors in the Tx burst can lead to serious errors in the host including, on some cases,
380 On ConnectX-5 and Bluefield the MPW is secure and enabled by default.
382 - ``txq_mpw_hdr_dseg_en`` parameter [int]
384 A nonzero value enables including two pointers in the first block of TX
385 descriptor. This can be used to lessen CPU load for memory copy.
387 Effective only when Enhanced MPS is supported. Disabled by default.
389 - ``txq_max_inline_len`` parameter [int]
391 Maximum size of packet to be inlined. This limits the size of packet to
392 be inlined. If the size of a packet is larger than configured value, the
393 packet isn't inlined even though there's enough space remained in the
394 descriptor. Instead, the packet is included with pointer.
396 Effective only when Enhanced MPS is supported. The default value is 256.
398 - ``tx_vec_en`` parameter [int]
400 A nonzero value enables Tx vector on ConnectX-5 and Bluefield NICs if the number of
401 global Tx queues on the port is less than ``txqs_max_vec``.
403 This option cannot be used with certain offloads such as ``DEV_TX_OFFLOAD_TCP_TSO,
404 DEV_TX_OFFLOAD_VXLAN_TNL_TSO, DEV_TX_OFFLOAD_GRE_TNL_TSO, DEV_TX_OFFLOAD_VLAN_INSERT``.
405 When those offloads are requested the MPS send function will not be used.
407 Enabled by default on ConnectX-5 and Bluefield.
409 - ``rx_vec_en`` parameter [int]
411 A nonzero value enables Rx vector if the port is not configured in
412 multi-segment otherwise this parameter is ignored.
416 - ``vf_nl_en`` parameter [int]
418 A nonzero value enables Netlink requests from the VF to add/remove MAC
419 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
420 Otherwise the relevant configuration must be run with Linux iproute2 tools.
421 This is a prerequisite to receive this kind of traffic.
423 Enabled by default, valid only on VF devices ignored otherwise.
425 - ``l3_vxlan_en`` parameter [int]
427 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
428 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
429 parameter. This is a prerequisite to receive this kind of traffic.
433 - ``dv_flow_en`` parameter [int]
435 A nonzero value enables the DV flow steering assuming it is supported
440 - ``representor`` parameter [list]
442 This parameter can be used to instantiate DPDK Ethernet devices from
443 existing port (or VF) representors configured on the device.
445 It is a standard parameter whose format is described in
446 :ref:`ethernet_device_standard_device_arguments`.
448 For instance, to probe port representors 0 through 2::
452 Firmware configuration
453 ~~~~~~~~~~~~~~~~~~~~~~
455 - L3 VXLAN and VXLAN-GPE destination UDP port
457 .. code-block:: console
459 mlxconfig -d <mst device> set IP_OVER_VXLAN_EN=1
460 mlxconfig -d <mst device> set IP_OVER_VXLAN_PORT=<udp dport>
462 Verify configurations are set:
464 .. code-block:: console
466 mlxconfig -d <mst device> query | grep IP_OVER_VXLAN
467 IP_OVER_VXLAN_EN True(1)
468 IP_OVER_VXLAN_PORT <udp dport>
473 This driver relies on external libraries and kernel drivers for resources
474 allocations and initialization. The following dependencies are not part of
475 DPDK and must be installed separately:
479 User space Verbs framework used by librte_pmd_mlx5. This library provides
480 a generic interface between the kernel and low-level user space drivers
483 It allows slow and privileged operations (context initialization, hardware
484 resources allocations) to be managed by the kernel and fast operations to
485 never leave user space.
489 Low-level user space driver library for Mellanox
490 ConnectX-4/ConnectX-5/Bluefield devices, it is automatically loaded
493 This library basically implements send/receive calls to the hardware
498 Minimalistic Netlink library mainly relied on to manage E-Switch flow
499 rules (i.e. those with the "transfer" attribute and typically involving
504 They provide the kernel-side Verbs API and low level device drivers that
505 manage actual hardware initialization and resources sharing with user
508 Unlike most other PMDs, these modules must remain loaded and bound to
511 - mlx5_core: hardware driver managing Mellanox
512 ConnectX-4/ConnectX-5/Bluefield devices and related Ethernet kernel
514 - mlx5_ib: InifiniBand device driver.
515 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
517 - **Firmware update**
519 Mellanox OFED releases include firmware updates for
520 ConnectX-4/ConnectX-5/Bluefield adapters.
522 Because each release provides new features, these updates must be applied to
523 match the kernel modules and libraries they come with.
527 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
533 Either RDMA Core library with a recent enough Linux kernel release
534 (recommended) or Mellanox OFED, which provides compatibility with older
537 RMDA Core with Linux Kernel
538 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
540 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
541 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
542 (see `RDMA Core installation documentation`_)
543 - When building for i686 use:
545 - rdma-core version 18.0 or above built with 32bit support.
546 - Kernel version 4.14.41 or above.
548 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
549 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
554 - Mellanox OFED version: **4.3, 4.4**.
557 - ConnectX-4: **12.21.1000** and above.
558 - ConnectX-4 Lx: **14.21.1000** and above.
559 - ConnectX-5: **16.21.1000** and above.
560 - ConnectX-5 Ex: **16.21.1000** and above.
561 - Bluefield: **18.99.3950** and above.
563 While these libraries and kernel modules are available on OpenFabrics
564 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
565 managers on most distributions, this PMD requires Ethernet extensions that
566 may not be supported at the moment (this is a work in progress).
569 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__
570 includes the necessary support and should be used in the meantime. For DPDK,
571 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
572 required from that distribution.
576 Several versions of Mellanox OFED are available. Installing the version
577 this DPDK release was developed and tested against is strongly
578 recommended. Please check the `prerequisites`_.
583 Minimal version for libmnl is **1.0.3**.
585 As a dependency of the **iproute2** suite, this library is often installed
586 by default. It is otherwise readily available through standard system
589 Its development headers must be installed in order to compile this PMD.
590 These packages are usually named **libmnl-dev** or **libmnl-devel**
591 depending on the Linux distribution.
596 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
597 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
598 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
599 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
600 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
601 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
602 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
603 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
604 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
605 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
606 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
607 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
608 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
609 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
610 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
611 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
612 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
613 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
614 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
616 Quick Start Guide on OFED
617 -------------------------
619 1. Download latest Mellanox OFED. For more info check the `prerequisites`_.
622 2. Install the required libraries and kernel modules either by installing
623 only the required set, or by installing the entire Mellanox OFED:
625 .. code-block:: console
627 ./mlnxofedinstall --upstream-libs --dpdk
629 3. Verify the firmware is the correct one:
631 .. code-block:: console
635 4. Verify all ports links are set to Ethernet:
637 .. code-block:: console
639 mlxconfig -d <mst device> query | grep LINK_TYPE
643 Link types may have to be configured to Ethernet:
645 .. code-block:: console
647 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
649 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
651 For hypervisors verify SR-IOV is enabled on the NIC:
653 .. code-block:: console
655 mlxconfig -d <mst device> query | grep SRIOV_EN
658 If needed, set enable the set the relevant fields:
660 .. code-block:: console
662 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
663 mlxfwreset -d <mst device> reset
665 5. Restart the driver:
667 .. code-block:: console
669 /etc/init.d/openibd restart
673 .. code-block:: console
675 service openibd restart
677 If link type was changed, firmware must be reset as well:
679 .. code-block:: console
681 mlxfwreset -d <mst device> reset
683 For hypervisors, after reset write the sysfs number of virtual functions
686 To dynamically instantiate a given number of virtual functions (VFs):
688 .. code-block:: console
690 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
692 6. Compile DPDK and you are ready to go. See instructions on
693 :ref:`Development Kit Build System <Development_Kit_Build_System>`
698 1. Configure aggressive CQE Zipping for maximum performance:
700 .. code-block:: console
702 mlxconfig -d <mst device> s CQE_COMPRESSION=1
704 To set it back to the default CQE Zipping mode use:
706 .. code-block:: console
708 mlxconfig -d <mst device> s CQE_COMPRESSION=0
710 2. In case of virtualization:
712 - Make sure that hypervisor kernel is 3.16 or newer.
713 - Configure boot with ``iommu=pt``.
715 - Make sure to allocate a VM on huge pages.
716 - Make sure to set CPU pinning.
718 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
719 for better performance. For VMs, verify that the right CPU
720 and NUMA node are pinned according to the above. Run:
722 .. code-block:: console
726 to identify the NUMA node to which the PCIe adapter is connected.
728 4. If more than one adapter is used, and root complex capabilities allow
729 to put both adapters on the same NUMA node without PCI bandwidth degradation,
730 it is recommended to locate both adapters on the same NUMA node.
731 This in order to forward packets from one to the other without
732 NUMA performance penalty.
734 5. Disable pause frames:
736 .. code-block:: console
738 ethtool -A <netdev> rx off tx off
740 6. Verify IO non-posted prefetch is disabled by default. This can be checked
741 via the BIOS configuration. Please contact you server provider for more
742 information about the settings.
746 On some machines, depends on the machine integrator, it is beneficial
747 to set the PCI max read request parameter to 1K. This can be
748 done in the following way:
750 To query the read request size use:
752 .. code-block:: console
754 setpci -s <NIC PCI address> 68.w
756 If the output is different than 3XXX, set it by:
758 .. code-block:: console
760 setpci -s <NIC PCI address> 68.w=3XXX
762 The XXX can be different on different systems. Make sure to configure
763 according to the setpci output.
765 7. To minimize overhead of searching Memory Regions:
767 - '--socket-mem' is recommended to pin memory by predictable amount.
768 - Configure per-lcore cache when creating Mempools for packet buffer.
769 - Refrain from dynamically allocating/freeing memory in run-time.
774 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
775 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
777 Since ``testpmd`` defaults to IP RSS mode and there is currently no
778 command-line parameter to enable additional protocols (UDP and TCP as well
779 as IP), the following commands must be entered from its CLI to get the same
780 behavior as librte_pmd_mlx4:
782 .. code-block:: console
785 > port config all rss all
791 This section demonstrates how to launch **testpmd** with Mellanox
792 ConnectX-4/ConnectX-5/Bluefield devices managed by librte_pmd_mlx5.
794 #. Load the kernel modules:
796 .. code-block:: console
798 modprobe -a ib_uverbs mlx5_core mlx5_ib
800 Alternatively if MLNX_OFED is fully installed, the following script can
803 .. code-block:: console
805 /etc/init.d/openibd restart
809 User space I/O kernel modules (uio and igb_uio) are not used and do
810 not have to be loaded.
812 #. Make sure Ethernet interfaces are in working order and linked to kernel
813 verbs. Related sysfs entries should be present:
815 .. code-block:: console
817 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
821 .. code-block:: console
828 #. Optionally, retrieve their PCI bus addresses for whitelisting:
830 .. code-block:: console
833 for intf in eth2 eth3 eth4 eth5;
835 (cd "/sys/class/net/${intf}/device/" && pwd -P);
838 sed -n 's,.*/\(.*\),-w \1,p'
842 .. code-block:: console
849 #. Request huge pages:
851 .. code-block:: console
853 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
855 #. Start testpmd with basic parameters:
857 .. code-block:: console
859 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
863 .. code-block:: console
866 EAL: PCI device 0000:05:00.0 on NUMA socket 0
867 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
868 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
869 PMD: librte_pmd_mlx5: 1 port(s) detected
870 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
871 EAL: PCI device 0000:05:00.1 on NUMA socket 0
872 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
873 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
874 PMD: librte_pmd_mlx5: 1 port(s) detected
875 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
876 EAL: PCI device 0000:06:00.0 on NUMA socket 0
877 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
878 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
879 PMD: librte_pmd_mlx5: 1 port(s) detected
880 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
881 EAL: PCI device 0000:06:00.1 on NUMA socket 0
882 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
883 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
884 PMD: librte_pmd_mlx5: 1 port(s) detected
885 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
886 Interactive-mode selected
887 Configuring Port 0 (socket 0)
888 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
889 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
890 Port 0: E4:1D:2D:E7:0C:FE
891 Configuring Port 1 (socket 0)
892 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
893 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
894 Port 1: E4:1D:2D:E7:0C:FF
895 Configuring Port 2 (socket 0)
896 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
897 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
898 Port 2: E4:1D:2D:E7:0C:FA
899 Configuring Port 3 (socket 0)
900 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
901 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
902 Port 3: E4:1D:2D:E7:0C:FB
903 Checking link statuses...
904 Port 0 Link Up - speed 40000 Mbps - full-duplex
905 Port 1 Link Up - speed 40000 Mbps - full-duplex
906 Port 2 Link Up - speed 10000 Mbps - full-duplex
907 Port 3 Link Up - speed 10000 Mbps - full-duplex