2 Copyright 2015 6WIND S.A.
4 Redistribution and use in source and binary forms, with or without
5 modification, are permitted provided that the following conditions
8 * Redistributions of source code must retain the above copyright
9 notice, this list of conditions and the following disclaimer.
10 * Redistributions in binary form must reproduce the above copyright
11 notice, this list of conditions and the following disclaimer in
12 the documentation and/or other materials provided with the
14 * Neither the name of 6WIND S.A. nor the names of its
15 contributors may be used to endorse or promote products derived
16 from this software without specific prior written permission.
18 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19 "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
21 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
22 OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
23 SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
24 LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25 DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26 THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27 (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
28 OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
33 The MLX5 poll mode driver library (**librte_pmd_mlx5**) provides support for
34 **Mellanox ConnectX-4** and **Mellanox ConnectX-4 Lx** families of
35 10/25/40/50/100 Gb/s adapters as well as their virtual functions (VF) in
38 Information and documentation about these adapters can be found on the
39 `Mellanox website <http://www.mellanox.com>`__. Help is also provided by the
40 `Mellanox community <http://community.mellanox.com/welcome>`__.
42 There is also a `section dedicated to this poll mode driver
43 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`__.
47 Due to external dependencies, this driver is disabled by default. It must
48 be enabled manually by setting ``CONFIG_RTE_LIBRTE_MLX5_PMD=y`` and
51 Implementation details
52 ----------------------
54 Besides its dependency on libibverbs (that implies libmlx5 and associated
55 kernel support), librte_pmd_mlx5 relies heavily on system calls for control
56 operations such as querying/updating the MTU and flow control parameters.
58 For security reasons and robustness, this driver only deals with virtual
59 memory addresses. The way resources allocations are handled by the kernel
60 combined with hardware specifications that allow it to handle virtual memory
61 addresses directly ensure that DPDK applications cannot access random
62 physical memory (or memory that does not belong to the current process).
64 This capability allows the PMD to coexist with kernel network interfaces
65 which remain functional, although they stop receiving unicast packets as
66 long as they share the same MAC address.
68 Enabling librte_pmd_mlx5 causes DPDK applications to be linked against
74 - Multiple TX and RX queues.
75 - Support for scattered TX and RX frames.
76 - IPv4, IPv6, TCPv4, TCPv6, UDPv4 and UDPv6 RSS on any number of queues.
77 - Several RSS hash keys, one for each flow type.
78 - Configurable RETA table.
79 - Support for multiple MAC addresses.
83 - RX CRC stripping configuration.
85 - Multicast promiscuous mode.
86 - Hardware checksum offloads.
87 - Flow director (RTE_FDIR_MODE_PERFECT and RTE_FDIR_MODE_PERFECT_MAC_VLAN).
88 - Secondary process TX is supported.
93 - KVM and VMware ESX SR-IOV modes are not supported yet.
94 - Inner RSS for VXLAN frames is not supported yet.
95 - Port statistics through software counters only.
96 - Hardware checksum offloads for VXLAN inner header are not supported yet.
97 - Secondary process RX is not supported.
105 These options can be modified in the ``.config`` file.
107 - ``CONFIG_RTE_LIBRTE_MLX5_PMD`` (default **n**)
109 Toggle compilation of librte_pmd_mlx5 itself.
111 - ``CONFIG_RTE_LIBRTE_MLX5_DEBUG`` (default **n**)
113 Toggle debugging code and stricter compilation flags. Enabling this option
114 adds additional run-time checks and debugging messages at the cost of
117 - ``CONFIG_RTE_LIBRTE_MLX5_SGE_WR_N`` (default **4**)
119 Number of scatter/gather elements (SGEs) per work request (WR). Lowering
120 this number improves performance but also limits the ability to receive
121 scattered packets (packets that do not fit a single mbuf). The default
122 value is a safe tradeoff.
124 - ``CONFIG_RTE_LIBRTE_MLX5_MAX_INLINE`` (default **0**)
126 Amount of data to be inlined during TX operations. Improves latency.
127 Can improve PPS performance when PCI backpressure is detected and may be
128 useful for scenarios involving heavy traffic on many queues.
130 Since the additional software logic necessary to handle this mode can
131 lower performance when there is no backpressure, it is not enabled by
134 - ``CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE`` (default **8**)
136 Maximum number of cached memory pools (MPs) per TX queue. Each MP from
137 which buffers are to be transmitted must be associated to memory regions
138 (MRs). This is a slow operation that must be cached.
140 This value is always 1 for RX queues since they use a single MP.
142 Environment variables
143 ~~~~~~~~~~~~~~~~~~~~~
145 - ``MLX5_ENABLE_CQE_COMPRESSION``
147 A nonzero value lets ConnectX-4 return smaller completion entries to
148 improve performance when PCI backpressure is detected. It is most useful
149 for scenarios involving heavy traffic on many queues.
151 Since the additional software logic necessary to handle this mode can
152 lower performance when there is no backpressure, it is not enabled by
155 - ``MLX5_PMD_ENABLE_PADDING``
157 Enables HW packet padding in PCI bus transactions.
159 When packet size is cache aligned and CRC stripping is enabled, 4 fewer
160 bytes are written to the PCI bus. Enabling padding makes such packets
163 In cases where PCI bandwidth is the bottleneck, padding can improve
166 This is disabled by default since this can also decrease performance for
167 unaligned packet sizes.
169 Run-time configuration
170 ~~~~~~~~~~~~~~~~~~~~~~
172 - librte_pmd_mlx5 brings kernel network interfaces up during initialization
173 because it is affected by their state. Forcing them down prevents packets
176 - **ethtool** operations on related kernel interfaces also affect the PMD.
181 This driver relies on external libraries and kernel drivers for resources
182 allocations and initialization. The following dependencies are not part of
183 DPDK and must be installed separately:
187 User space Verbs framework used by librte_pmd_mlx5. This library provides
188 a generic interface between the kernel and low-level user space drivers
191 It allows slow and privileged operations (context initialization, hardware
192 resources allocations) to be managed by the kernel and fast operations to
193 never leave user space.
197 Low-level user space driver library for Mellanox ConnectX-4 devices,
198 it is automatically loaded by libibverbs.
200 This library basically implements send/receive calls to the hardware
203 - **Kernel modules** (mlnx-ofed-kernel)
205 They provide the kernel-side Verbs API and low level device drivers that
206 manage actual hardware initialization and resources sharing with user
209 Unlike most other PMDs, these modules must remain loaded and bound to
212 - mlx5_core: hardware driver managing Mellanox ConnectX-4 devices and
213 related Ethernet kernel network devices.
214 - mlx5_ib: InifiniBand device driver.
215 - ib_uverbs: user space driver for Verbs (entry point for libibverbs).
217 - **Firmware update**
219 Mellanox OFED releases include firmware updates for ConnectX-4 adapters.
221 Because each release provides new features, these updates must be applied to
222 match the kernel modules and libraries they come with.
226 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
229 Currently supported by DPDK:
231 - Mellanox OFED **3.1-1.0.3**, **3.1-1.5.7.1** or **3.2-2.0.0.0** depending
234 The following features are supported with version **3.1-1.5.7.1** and
237 - IPv6, UPDv6, TCPv6 RSS.
238 - RX checksum offloads.
241 The following features are supported with version **3.2-2.0.0.0** and
247 - RX CRC stripping configuration.
249 - Minimum firmware version:
251 With MLNX_OFED **3.1-1.0.3**:
253 - ConnectX-4: **12.12.1240**
254 - ConnectX-4 Lx: **14.12.1100**
256 With MLNX_OFED **3.1-1.5.7.1**:
258 - ConnectX-4: **12.13.0144**
259 - ConnectX-4 Lx: **14.13.0144**
261 With MLNX_OFED **3.2-2.0.0.0**:
263 - ConnectX-4: **12.14.2036**
264 - ConnectX-4 Lx: **14.14.2036**
266 Getting Mellanox OFED
267 ~~~~~~~~~~~~~~~~~~~~~
269 While these libraries and kernel modules are available on OpenFabrics
270 Alliance's `website <https://www.openfabrics.org/>`__ and provided by package
271 managers on most distributions, this PMD requires Ethernet extensions that
272 may not be supported at the moment (this is a work in progress).
275 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux>`__
276 includes the necessary support and should be used in the meantime. For DPDK,
277 only libibverbs, libmlx5, mlnx-ofed-kernel packages and firmware updates are
278 required from that distribution.
282 Several versions of Mellanox OFED are available. Installing the version
283 this DPDK release was developed and tested against is strongly
284 recommended. Please check the `prerequisites`_.
289 Compared to librte_pmd_mlx4 that implements a single RSS configuration per
290 port, librte_pmd_mlx5 supports per-protocol RSS configuration.
292 Since ``testpmd`` defaults to IP RSS mode and there is currently no
293 command-line parameter to enable additional protocols (UDP and TCP as well
294 as IP), the following commands must be entered from its CLI to get the same
295 behavior as librte_pmd_mlx4:
297 .. code-block:: console
300 > port config all rss all
306 This section demonstrates how to launch **testpmd** with Mellanox ConnectX-4
307 devices managed by librte_pmd_mlx5.
309 #. Load the kernel modules:
311 .. code-block:: console
313 modprobe -a ib_uverbs mlx5_core mlx5_ib
315 Alternatively if MLNX_OFED is fully installed, the following script can
318 .. code-block:: console
320 /etc/init.d/openibd restart
324 User space I/O kernel modules (uio and igb_uio) are not used and do
325 not have to be loaded.
327 #. Make sure Ethernet interfaces are in working order and linked to kernel
328 verbs. Related sysfs entries should be present:
330 .. code-block:: console
332 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
336 .. code-block:: console
343 #. Optionally, retrieve their PCI bus addresses for whitelisting:
345 .. code-block:: console
348 for intf in eth2 eth3 eth4 eth5;
350 (cd "/sys/class/net/${intf}/device/" && pwd -P);
353 sed -n 's,.*/\(.*\),-w \1,p'
357 .. code-block:: console
364 #. Request huge pages:
366 .. code-block:: console
368 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
370 #. Start testpmd with basic parameters:
372 .. code-block:: console
374 testpmd -c 0xff00 -n 4 -w 05:00.0 -w 05:00.1 -w 06:00.0 -w 06:00.1 -- --rxq=2 --txq=2 -i
378 .. code-block:: console
381 EAL: PCI device 0000:05:00.0 on NUMA socket 0
382 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
383 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_0" (VF: false)
384 PMD: librte_pmd_mlx5: 1 port(s) detected
385 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fe
386 EAL: PCI device 0000:05:00.1 on NUMA socket 0
387 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
388 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_1" (VF: false)
389 PMD: librte_pmd_mlx5: 1 port(s) detected
390 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:ff
391 EAL: PCI device 0000:06:00.0 on NUMA socket 0
392 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
393 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_2" (VF: false)
394 PMD: librte_pmd_mlx5: 1 port(s) detected
395 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fa
396 EAL: PCI device 0000:06:00.1 on NUMA socket 0
397 EAL: probe driver: 15b3:1013 librte_pmd_mlx5
398 PMD: librte_pmd_mlx5: PCI information matches, using device "mlx5_3" (VF: false)
399 PMD: librte_pmd_mlx5: 1 port(s) detected
400 PMD: librte_pmd_mlx5: port 1 MAC address is e4:1d:2d:e7:0c:fb
401 Interactive-mode selected
402 Configuring Port 0 (socket 0)
403 PMD: librte_pmd_mlx5: 0x8cba80: TX queues number update: 0 -> 2
404 PMD: librte_pmd_mlx5: 0x8cba80: RX queues number update: 0 -> 2
405 Port 0: E4:1D:2D:E7:0C:FE
406 Configuring Port 1 (socket 0)
407 PMD: librte_pmd_mlx5: 0x8ccac8: TX queues number update: 0 -> 2
408 PMD: librte_pmd_mlx5: 0x8ccac8: RX queues number update: 0 -> 2
409 Port 1: E4:1D:2D:E7:0C:FF
410 Configuring Port 2 (socket 0)
411 PMD: librte_pmd_mlx5: 0x8cdb10: TX queues number update: 0 -> 2
412 PMD: librte_pmd_mlx5: 0x8cdb10: RX queues number update: 0 -> 2
413 Port 2: E4:1D:2D:E7:0C:FA
414 Configuring Port 3 (socket 0)
415 PMD: librte_pmd_mlx5: 0x8ceb58: TX queues number update: 0 -> 2
416 PMD: librte_pmd_mlx5: 0x8ceb58: RX queues number update: 0 -> 2
417 Port 3: E4:1D:2D:E7:0C:FB
418 Checking link statuses...
419 Port 0 Link Up - speed 40000 Mbps - full-duplex
420 Port 1 Link Up - speed 40000 Mbps - full-duplex
421 Port 2 Link Up - speed 10000 Mbps - full-duplex
422 Port 3 Link Up - speed 10000 Mbps - full-duplex