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** and **Mellanox
10 ConnectX-5** families of 10/25/40/50/100 Gb/s adapters as well as their
11 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.
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 - Configurable RETA table.
58 - Support for multiple MAC addresses.
62 - RX CRC stripping configuration.
64 - Multicast promiscuous mode.
65 - Hardware checksum offloads.
66 - Flow director (RTE_FDIR_MODE_PERFECT, RTE_FDIR_MODE_PERFECT_MAC_VLAN and
70 - KVM and VMware ESX SR-IOV modes are supported.
71 - RSS hash result is supported.
72 - Hardware TSO for generic IP or UDP tunnel, including VXLAN and GRE.
73 - Hardware checksum Tx offload for generic IP or UDP tunnel, including VXLAN and GRE.
75 - Statistics query including Basic, Extended and per queue.
77 - Tunnel types: VXLAN, L3 VXLAN, VXLAN-GPE, GRE.
78 - Tunnel HW offloads: packet type, inner/outer RSS, IP and UDP checksum verification.
83 - For secondary process:
85 - Forked secondary process not supported.
86 - All mempools must be initialized before rte_eth_dev_start().
88 - Flow pattern without any specific vlan will match for vlan packets as well:
90 When VLAN spec is not specified in the pattern, the matching rule will be created with VLAN as a wild card.
91 Meaning, the flow rule::
93 flow create 0 ingress pattern eth / vlan vid is 3 / ipv4 / end ...
95 Will only match vlan packets with vid=3. and the flow rules::
97 flow create 0 ingress pattern eth / ipv4 / end ...
101 flow create 0 ingress pattern eth / vlan / ipv4 / end ...
103 Will match any ipv4 packet (VLAN included).
105 - A multi segment packet must have less than 6 segments in case the Tx burst function
106 is set to multi-packet send or Enhanced multi-packet send. Otherwise it must have
107 less than 50 segments.
109 - Count action for RTE flow is **only supported in Mellanox OFED**.
111 - Flows with a VXLAN Network Identifier equal (or ends to be equal)
112 to 0 are not supported.
114 - VXLAN TSO and checksum offloads are not supported on VM.
116 - VF: flow rules created on VF devices can only match traffic targeted at the
117 configured MAC addresses (see ``rte_eth_dev_mac_addr_add()``).
121 MAC addresses not already present in the bridge table of the associated
122 kernel network device will be added and cleaned up by the PMD when closing
123 the device. In case of ungraceful program termination, some entries may
124 remain present and should be removed manually by other means.
126 - When Multi-Packet Rx queue is configured (``mprq_en``), a Rx packet can be
127 externally attached to a user-provided mbuf with having EXT_ATTACHED_MBUF in
128 ol_flags. As the mempool for the external buffer is managed by PMD, all the
129 Rx mbufs must be freed before the device is closed. Otherwise, the mempool of
130 the external buffers will be freed by PMD and the application which still
131 holds the external buffers may be corrupted.
136 MLX5 supports various of methods to report statistics:
138 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.
140 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.
142 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.
150 These options can be modified in the ``.config`` file.
152 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
154 Toggle compilation of librte_pmd_mlx5 itself.
156 - ``CONFIG_RTE_LIBRTE_MLX5_DLOPEN_DEPS`` (default **n**)
158 Build PMD with additional code to make it loadable without hard
159 dependencies on **libibverbs** nor **libmlx5**, which may not be installed
160 on the target system.
162 In this mode, their presence is still required for it to run properly,
163 however their absence won't prevent a DPDK application from starting (with
164 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
165 missing with ``ldd(1)``.
167 It works by moving these dependencies to a purpose-built rdma-core "glue"
168 plug-in which must either be installed in a directory whose name is based
169 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
170 standard location for the dynamic linker (e.g. ``/lib``) if left to the
171 default empty string (``""``).
173 This option has no performance impact.
175 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
177 Toggle debugging code and stricter compilation flags. Enabling this option
178 adds additional run-time checks and debugging messages at the cost of
181 Environment variables
182 ~~~~~~~~~~~~~~~~~~~~~
186 A list of directories in which to search for the rdma-core "glue" plug-in,
187 separated by colons or semi-colons.
189 Only matters when compiled with ``CONFIG_RTE_LIBRTE_MLX5_DLOPEN_DEPS``
190 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
191 since ``LD_LIBRARY_PATH`` has no effect in this case.
193 - ``MLX5_PMD_ENABLE_PADDING``
195 Enables HW packet padding in PCI bus transactions.
197 When packet size is cache aligned and CRC stripping is enabled, 4 fewer
198 bytes are written to the PCI bus. Enabling padding makes such packets
201 In cases where PCI bandwidth is the bottleneck, padding can improve
204 This is disabled by default since this can also decrease performance for
205 unaligned packet sizes.
207 - ``MLX5_SHUT_UP_BF``
209 Configures HW Tx doorbell register as IO-mapped.
211 By default, the HW Tx doorbell is configured as a write-combining register.
212 The register would be flushed to HW usually when the write-combining buffer
213 becomes full, but it depends on CPU design.
215 Except for vectorized Tx burst routines, a write memory barrier is enforced
216 after updating the register so that the update can be immediately visible to
219 When vectorized Tx burst is called, the barrier is set only if the burst size
220 is not aligned to MLX5_VPMD_TX_MAX_BURST. However, setting this environmental
221 variable will bring better latency even though the maximum throughput can
224 Run-time configuration
225 ~~~~~~~~~~~~~~~~~~~~~~
227 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
228 because it is affected by their state. Forcing them down prevents packets
231 - **ethtool** operations on related kernel interfaces also affect the PMD.
233 - ``rxq_cqe_comp_en`` parameter [int]
235 A nonzero value enables the compression of CQE on RX side. This feature
236 allows to save PCI bandwidth and improve performance. Enabled by default.
240 - x86_64 with ConnectX-4, ConnectX-4 LX and ConnectX-5.
241 - POWER8 and ARMv8 with ConnectX-4 LX and ConnectX-5.
243 - ``mprq_en`` parameter [int]
245 A nonzero value enables configuring Multi-Packet Rx queues. Rx queue is
246 configured as Multi-Packet RQ if the total number of Rx queues is
247 ``rxqs_min_mprq`` or more and Rx scatter isn't configured. Disabled by
250 Multi-Packet Rx Queue (MPRQ a.k.a Striding RQ) can further save PCIe bandwidth
251 by posting a single large buffer for multiple packets. Instead of posting a
252 buffers per a packet, one large buffer is posted in order to receive multiple
253 packets on the buffer. A MPRQ buffer consists of multiple fixed-size strides
254 and each stride receives one packet. MPRQ can improve throughput for
255 small-packet tarffic.
257 When MPRQ is enabled, max_rx_pkt_len can be larger than the size of
258 user-provided mbuf even if DEV_RX_OFFLOAD_SCATTER isn't enabled. PMD will
259 configure large stride size enough to accommodate max_rx_pkt_len as long as
260 device allows. Note that this can waste system memory compared to enabling Rx
261 scatter and multi-segment packet.
263 - ``mprq_log_stride_num`` parameter [int]
265 Log 2 of the number of strides for Multi-Packet Rx queue. Configuring more
266 strides can reduce PCIe tarffic further. If configured value is not in the
267 range of device capability, the default value will be set with a warning
268 message. The default value is 4 which is 16 strides per a buffer, valid only
269 if ``mprq_en`` is set.
271 The size of Rx queue should be bigger than the number of strides.
273 - ``mprq_max_memcpy_len`` parameter [int]
275 The maximum length of packet to memcpy in case of Multi-Packet Rx queue. Rx
276 packet is mem-copied to a user-provided mbuf if the size of Rx packet is less
277 than or equal to this parameter. Otherwise, PMD will attach the Rx packet to
278 the mbuf by external buffer attachment - ``rte_pktmbuf_attach_extbuf()``.
279 A mempool for external buffers will be allocated and managed by PMD. If Rx
280 packet is externally attached, ol_flags field of the mbuf will have
281 EXT_ATTACHED_MBUF and this flag must be preserved. ``RTE_MBUF_HAS_EXTBUF()``
282 checks the flag. The default value is 128, valid only if ``mprq_en`` is set.
284 - ``rxqs_min_mprq`` parameter [int]
286 Configure Rx queues as Multi-Packet RQ if the total number of Rx queues is
287 greater or equal to this value. The default value is 12, valid only if
290 - ``txq_inline`` parameter [int]
292 Amount of data to be inlined during TX operations. Improves latency.
293 Can improve PPS performance when PCI back pressure is detected and may be
294 useful for scenarios involving heavy traffic on many queues.
296 Because additional software logic is necessary to handle this mode, this
297 option should be used with care, as it can lower performance when back
298 pressure is not expected.
300 - ``txqs_min_inline`` parameter [int]
302 Enable inline send only when the number of TX queues is greater or equal
305 This option should be used in combination with ``txq_inline`` above.
307 On ConnectX-4, ConnectX-4 LX and ConnectX-5 without Enhanced MPW:
309 - Disabled by default.
310 - In case ``txq_inline`` is set recommendation is 4.
312 On ConnectX-5 with Enhanced MPW:
314 - Set to 8 by default.
316 - ``txq_mpw_en`` parameter [int]
318 A nonzero value enables multi-packet send (MPS) for ConnectX-4 Lx and
319 enhanced multi-packet send (Enhanced MPS) for ConnectX-5. MPS allows the
320 TX burst function to pack up multiple packets in a single descriptor
321 session in order to save PCI bandwidth and improve performance at the
322 cost of a slightly higher CPU usage. When ``txq_inline`` is set along
323 with ``txq_mpw_en``, TX burst function tries to copy entire packet data
324 on to TX descriptor instead of including pointer of packet only if there
325 is enough room remained in the descriptor. ``txq_inline`` sets
326 per-descriptor space for either pointers or inlined packets. In addition,
327 Enhanced MPS supports hybrid mode - mixing inlined packets and pointers
328 in the same descriptor.
330 This option cannot be used with certain offloads such as ``DEV_TX_OFFLOAD_TCP_TSO,
331 DEV_TX_OFFLOAD_VXLAN_TNL_TSO, DEV_TX_OFFLOAD_GRE_TNL_TSO, DEV_TX_OFFLOAD_VLAN_INSERT``.
332 When those offloads are requested the MPS send function will not be used.
334 It is currently only supported on the ConnectX-4 Lx and ConnectX-5
335 families of adapters. Enabled by default.
337 - ``txq_mpw_hdr_dseg_en`` parameter [int]
339 A nonzero value enables including two pointers in the first block of TX
340 descriptor. This can be used to lessen CPU load for memory copy.
342 Effective only when Enhanced MPS is supported. Disabled by default.
344 - ``txq_max_inline_len`` parameter [int]
346 Maximum size of packet to be inlined. This limits the size of packet to
347 be inlined. If the size of a packet is larger than configured value, the
348 packet isn't inlined even though there's enough space remained in the
349 descriptor. Instead, the packet is included with pointer.
351 Effective only when Enhanced MPS is supported. The default value is 256.
353 - ``tx_vec_en`` parameter [int]
355 A nonzero value enables Tx vector on ConnectX-5 only NIC if the number of
356 global Tx queues on the port is lesser than MLX5_VPMD_MIN_TXQS.
358 This option cannot be used with certain offloads such as ``DEV_TX_OFFLOAD_TCP_TSO,
359 DEV_TX_OFFLOAD_VXLAN_TNL_TSO, DEV_TX_OFFLOAD_GRE_TNL_TSO, DEV_TX_OFFLOAD_VLAN_INSERT``.
360 When those offloads are requested the MPS send function will not be used.
362 Enabled by default on ConnectX-5.
364 - ``rx_vec_en`` parameter [int]
366 A nonzero value enables Rx vector if the port is not configured in
367 multi-segment otherwise this parameter is ignored.
371 - ``vf_nl_en`` parameter [int]
373 A nonzero value enables Netlink requests from the VF to add/remove MAC
374 addresses or/and enable/disable promiscuous/all multicast on the Netdevice.
375 Otherwise the relevant configuration must be run with Linux iproute2 tools.
376 This is a prerequisite to receive this kind of traffic.
378 Enabled by default, valid only on VF devices ignored otherwise.
380 - ``l3_vxlan_en`` parameter [int]
382 A nonzero value allows L3 VXLAN and VXLAN-GPE flow creation. To enable
383 L3 VXLAN or VXLAN-GPE, users has to configure firmware and enable this
384 parameter. This is a prerequisite to receive this kind of traffic.
388 Firmware configuration
389 ~~~~~~~~~~~~~~~~~~~~~~
391 - L3 VXLAN and VXLAN-GPE destination UDP port
393 .. code-block:: console
395 mlxconfig -d <mst device> set IP_OVER_VXLAN_EN=1
396 mlxconfig -d <mst device> set IP_OVER_VXLAN_PORT=<udp dport>
398 Verify configurations are set:
400 .. code-block:: console
402 mlxconfig -d <mst device> query | grep IP_OVER_VXLAN
403 IP_OVER_VXLAN_EN True(1)
404 IP_OVER_VXLAN_PORT <udp dport>
409 This driver relies on external libraries and kernel drivers for resources
410 allocations and initialization. The following dependencies are not part of
411 DPDK and must be installed separately:
415 User space Verbs framework used by librte_pmd_mlx5. This library provides
416 a generic interface between the kernel and low-level user space drivers
419 It allows slow and privileged operations (context initialization, hardware
420 resources allocations) to be managed by the kernel and fast operations to
421 never leave user space.
425 Low-level user space driver library for Mellanox ConnectX-4/ConnectX-5
426 devices, it is automatically loaded by libibverbs.
428 This library basically implements send/receive calls to the hardware
433 They provide the kernel-side Verbs API and low level device drivers that
434 manage actual hardware initialization and resources sharing with user
437 Unlike most other PMDs, these modules must remain loaded and bound to
440 - mlx5_core: hardware driver managing Mellanox ConnectX-4/ConnectX-5
441 devices and related Ethernet kernel network devices.
442 - mlx5_ib: InifiniBand device driver.
443 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
445 - **Firmware update**
447 Mellanox OFED releases include firmware updates for ConnectX-4/ConnectX-5
450 Because each release provides new features, these updates must be applied to
451 match the kernel modules and libraries they come with.
455 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
461 Either RDMA Core library with a recent enough Linux kernel release
462 (recommended) or Mellanox OFED, which provides compatibility with older
465 RMDA Core with Linux Kernel
466 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
468 - Minimal kernel version : v4.14 or the most recent 4.14-rc (see `Linux installation documentation`_)
469 - Minimal rdma-core version: v15+ commit 0c5f5765213a ("Merge pull request #227 from yishaih/tm")
470 (see `RDMA Core installation documentation`_)
472 .. _`Linux installation documentation`: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/plain/Documentation/admin-guide/README.rst
473 .. _`RDMA Core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
478 - Mellanox OFED version: **4.2, 4.3**.
481 - ConnectX-4: **12.21.1000** and above.
482 - ConnectX-4 Lx: **14.21.1000** and above.
483 - ConnectX-5: **16.21.1000** and above.
484 - ConnectX-5 Ex: **16.21.1000** and above.
486 While these libraries and kernel modules are available on OpenFabrics
487 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
488 managers on most distributions, this PMD requires Ethernet extensions that
489 may not be supported at the moment (this is a work in progress).
492 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__
493 includes the necessary support and should be used in the meantime. For DPDK,
494 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
495 required from that distribution.
499 Several versions of Mellanox OFED are available. Installing the version
500 this DPDK release was developed and tested against is strongly
501 recommended. Please check the `prerequisites`_.
506 * Mellanox(R) ConnectX(R)-4 10G MCX4111A-XCAT (1x10G)
507 * Mellanox(R) ConnectX(R)-4 10G MCX4121A-XCAT (2x10G)
508 * Mellanox(R) ConnectX(R)-4 25G MCX4111A-ACAT (1x25G)
509 * Mellanox(R) ConnectX(R)-4 25G MCX4121A-ACAT (2x25G)
510 * Mellanox(R) ConnectX(R)-4 40G MCX4131A-BCAT (1x40G)
511 * Mellanox(R) ConnectX(R)-4 40G MCX413A-BCAT (1x40G)
512 * Mellanox(R) ConnectX(R)-4 40G MCX415A-BCAT (1x40G)
513 * Mellanox(R) ConnectX(R)-4 50G MCX4131A-GCAT (1x50G)
514 * Mellanox(R) ConnectX(R)-4 50G MCX413A-GCAT (1x50G)
515 * Mellanox(R) ConnectX(R)-4 50G MCX414A-BCAT (2x50G)
516 * Mellanox(R) ConnectX(R)-4 50G MCX415A-GCAT (2x50G)
517 * Mellanox(R) ConnectX(R)-4 50G MCX416A-BCAT (2x50G)
518 * Mellanox(R) ConnectX(R)-4 50G MCX416A-GCAT (2x50G)
519 * Mellanox(R) ConnectX(R)-4 50G MCX415A-CCAT (1x100G)
520 * Mellanox(R) ConnectX(R)-4 100G MCX416A-CCAT (2x100G)
521 * Mellanox(R) ConnectX(R)-4 Lx 10G MCX4121A-XCAT (2x10G)
522 * Mellanox(R) ConnectX(R)-4 Lx 25G MCX4121A-ACAT (2x25G)
523 * Mellanox(R) ConnectX(R)-5 100G MCX556A-ECAT (2x100G)
524 * Mellanox(R) ConnectX(R)-5 Ex EN 100G MCX516A-CDAT (2x100G)
526 Quick Start Guide on OFED
527 -------------------------
529 1. Download latest Mellanox OFED. For more info check the `prerequisites`_.
532 2. Install the required libraries and kernel modules either by installing
533 only the required set, or by installing the entire Mellanox OFED:
535 .. code-block:: console
537 ./mlnxofedinstall --upstream-libs --dpdk
539 3. Verify the firmware is the correct one:
541 .. code-block:: console
545 4. Verify all ports links are set to Ethernet:
547 .. code-block:: console
549 mlxconfig -d <mst device> query | grep LINK_TYPE
553 Link types may have to be configured to Ethernet:
555 .. code-block:: console
557 mlxconfig -d <mst device> set LINK_TYPE_P1/2=1/2/3
559 * LINK_TYPE_P1=<1|2|3> , 1=Infiniband 2=Ethernet 3=VPI(auto-sense)
561 For hypervisors verify SR-IOV is enabled on the NIC:
563 .. code-block:: console
565 mlxconfig -d <mst device> query | grep SRIOV_EN
568 If needed, set enable the set the relevant fields:
570 .. code-block:: console
572 mlxconfig -d <mst device> set SRIOV_EN=1 NUM_OF_VFS=16
573 mlxfwreset -d <mst device> reset
575 5. Restart the driver:
577 .. code-block:: console
579 /etc/init.d/openibd restart
583 .. code-block:: console
585 service openibd restart
587 If link type was changed, firmware must be reset as well:
589 .. code-block:: console
591 mlxfwreset -d <mst device> reset
593 For hypervisors, after reset write the sysfs number of virtual functions
596 To dynamically instantiate a given number of virtual functions (VFs):
598 .. code-block:: console
600 echo [num_vfs] > /sys/class/infiniband/mlx5_0/device/sriov_numvfs
602 6. Compile DPDK and you are ready to go. See instructions on
603 :ref:`Development Kit Build System <Development_Kit_Build_System>`
608 1. Configure aggressive CQE Zipping for maximum performance:
610 .. code-block:: console
612 mlxconfig -d <mst device> s CQE_COMPRESSION=1
614 To set it back to the default CQE Zipping mode use:
616 .. code-block:: console
618 mlxconfig -d <mst device> s CQE_COMPRESSION=0
620 2. In case of virtualization:
622 - Make sure that hypervisor kernel is 3.16 or newer.
623 - Configure boot with ``iommu=pt``.
625 - Make sure to allocate a VM on huge pages.
626 - Make sure to set CPU pinning.
628 3. Use the CPU near local NUMA node to which the PCIe adapter is connected,
629 for better performance. For VMs, verify that the right CPU
630 and NUMA node are pinned according to the above. Run:
632 .. code-block:: console
636 to identify the NUMA node to which the PCIe adapter is connected.
638 4. If more than one adapter is used, and root complex capabilities allow
639 to put both adapters on the same NUMA node without PCI bandwidth degradation,
640 it is recommended to locate both adapters on the same NUMA node.
641 This in order to forward packets from one to the other without
642 NUMA performance penalty.
644 5. Disable pause frames:
646 .. code-block:: console
648 ethtool -A <netdev> rx off tx off
650 6. Verify IO non-posted prefetch is disabled by default. This can be checked
651 via the BIOS configuration. Please contact you server provider for more
652 information about the settings.
656 On some machines, depends on the machine integrator, it is beneficial
657 to set the PCI max read request parameter to 1K. This can be
658 done in the following way:
660 To query the read request size use:
662 .. code-block:: console
664 setpci -s <NIC PCI address> 68.w
666 If the output is different than 3XXX, set it by:
668 .. code-block:: console
670 setpci -s <NIC PCI address> 68.w=3XXX
672 The XXX can be different on different systems. Make sure to configure
673 according to the setpci output.
675 7. To minimize overhead of searching Memory Regions:
677 - '--socket-mem' is recommended to pin memory by predictable amount.
678 - Configure per-lcore cache when creating Mempools for packet buffer.
679 - Refrain from dynamically allocating/freeing memory in run-time.
684 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
685 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
687 Since ``testpmd`` defaults to IP RSS mode and there is currently no
688 command-line parameter to enable additional protocols (UDP and TCP as well
689 as IP), the following commands must be entered from its CLI to get the same
690 behavior as librte_pmd_mlx4:
692 .. code-block:: console
695 > port config all rss all
701 This section demonstrates how to launch **testpmd** with Mellanox
702 ConnectX-4/ConnectX-5 devices managed by librte_pmd_mlx5.
704 #. Load the kernel modules:
706 .. code-block:: console
708 modprobe -a ib_uverbs mlx5_core mlx5_ib
710 Alternatively if MLNX_OFED is fully installed, the following script can
713 .. code-block:: console
715 /etc/init.d/openibd restart
719 User space I/O kernel modules (uio and igb_uio) are not used and do
720 not have to be loaded.
722 #. Make sure Ethernet interfaces are in working order and linked to kernel
723 verbs. Related sysfs entries should be present:
725 .. code-block:: console
727 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
731 .. code-block:: console
738 #. Optionally, retrieve their PCI bus addresses for whitelisting:
740 .. code-block:: console
743 for intf in eth2 eth3 eth4 eth5;
745 (cd "/sys/class/net/${intf}/device/" && pwd -P);
748 sed -n 's,.*/\(.*\),-w \1,p'
752 .. code-block:: console
759 #. Request huge pages:
761 .. code-block:: console
763 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
765 #. Start testpmd with basic parameters:
767 .. code-block:: console
769 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
773 .. code-block:: console
776 EAL: PCI device 0000:05:00.0 on NUMA socket 0
777 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
778 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
779 PMD: librte_pmd_mlx5: 1 port(s) detected
780 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
781 EAL: PCI device 0000:05:00.1 on NUMA socket 0
782 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
783 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
784 PMD: librte_pmd_mlx5: 1 port(s) detected
785 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
786 EAL: PCI device 0000:06:00.0 on NUMA socket 0
787 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
788 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
789 PMD: librte_pmd_mlx5: 1 port(s) detected
790 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
791 EAL: PCI device 0000:06:00.1 on NUMA socket 0
792 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
793 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
794 PMD: librte_pmd_mlx5: 1 port(s) detected
795 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
796 Interactive-mode selected
797 Configuring Port 0 (socket 0)
798 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
799 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
800 Port 0: E4:1D:2D:E7:0C:FE
801 Configuring Port 1 (socket 0)
802 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
803 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
804 Port 1: E4:1D:2D:E7:0C:FF
805 Configuring Port 2 (socket 0)
806 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
807 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
808 Port 2: E4:1D:2D:E7:0C:FA
809 Configuring Port 3 (socket 0)
810 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
811 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
812 Port 3: E4:1D:2D:E7:0C:FB
813 Checking link statuses...
814 Port 0 Link Up - speed 40000 Mbps - full-duplex
815 Port 1 Link Up - speed 40000 Mbps - full-duplex
816 Port 2 Link Up - speed 10000 Mbps - full-duplex
817 Port 3 Link Up - speed 10000 Mbps - full-duplex