1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright 2012 6WIND S.A.
3 Copyright 2015 Mellanox Technologies, Ltd
5 MLX4 poll mode driver library
6 =============================
8 The MLX4 poll mode driver library (**librte_pmd_mlx4**) implements support
9 for **Mellanox ConnectX-3** and **Mellanox ConnectX-3 Pro** 10/40 Gbps adapters
10 as well as their virtual functions (VF) in SR-IOV context.
12 Information and documentation about this family of adapters can be found on
13 the `Mellanox website <http://www.mellanox.com>`_. Help is also provided by
14 the `Mellanox community <http://community.mellanox.com/welcome>`_.
16 There is also a `section dedicated to this poll mode driver
17 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`_.
21 Due to external dependencies, this driver is disabled by default. It must
22 be enabled manually by setting ``CONFIG_RTE_LIBRTE_MLX4_PMD=y`` and
25 Implementation details
26 ----------------------
28 Most Mellanox ConnectX-3 devices provide two ports but expose a single PCI
29 bus address, thus unlike most drivers, librte_pmd_mlx4 registers itself as a
30 PCI driver that allocates one Ethernet device per detected port.
32 For this reason, one cannot white/blacklist a single port without also
33 white/blacklisting the others on the same device.
35 Besides its dependency on libibverbs (that implies libmlx4 and associated
36 kernel support), librte_pmd_mlx4 relies heavily on system calls for control
37 operations such as querying/updating the MTU and flow control parameters.
39 For security reasons and robustness, this driver only deals with virtual
40 memory addresses. The way resources allocations are handled by the kernel
41 combined with hardware specifications that allow it to handle virtual memory
42 addresses directly ensure that DPDK applications cannot access random
43 physical memory (or memory that does not belong to the current process).
45 This capability allows the PMD to coexist with kernel network interfaces
46 which remain functional, although they stop receiving unicast packets as
47 long as they share the same MAC address.
49 Compiling librte_pmd_mlx4 causes DPDK to be linked against libibverbs.
57 These options can be modified in the ``.config`` file.
59 - ``CONFIG_RTE_LIBRTE_MLX4_PMD`` (default **n**)
61 Toggle compilation of librte_pmd_mlx4 itself.
63 - ``CONFIG_RTE_IBVERBS_LINK_DLOPEN`` (default **n**)
65 Build PMD with additional code to make it loadable without hard
66 dependencies on **libibverbs** nor **libmlx4**, which may not be installed
69 In this mode, their presence is still required for it to run properly,
70 however their absence won't prevent a DPDK application from starting (with
71 ``CONFIG_RTE_BUILD_SHARED_LIB`` disabled) and they won't show up as
72 missing with ``ldd(1)``.
74 It works by moving these dependencies to a purpose-built rdma-core "glue"
75 plug-in which must either be installed in a directory whose name is based
76 on ``CONFIG_RTE_EAL_PMD_PATH`` suffixed with ``-glue`` if set, or in a
77 standard location for the dynamic linker (e.g. ``/lib``) if left to the
78 default empty string (``""``).
80 This option has no performance impact.
82 - ``CONFIG_RTE_IBVERBS_LINK_STATIC`` (default **n**)
84 Embed static flavour of the dependencies **libibverbs** and **libmlx4**
85 in the PMD shared library or the executable static binary.
87 - ``CONFIG_RTE_LIBRTE_MLX4_DEBUG`` (default **n**)
89 Toggle debugging code and stricter compilation flags. Enabling this option
90 adds additional run-time checks and debugging messages at the cost of
98 A list of directories in which to search for the rdma-core "glue" plug-in,
99 separated by colons or semi-colons.
101 Only matters when compiled with ``CONFIG_RTE_IBVERBS_LINK_DLOPEN``
102 enabled and most useful when ``CONFIG_RTE_EAL_PMD_PATH`` is also set,
103 since ``LD_LIBRARY_PATH`` has no effect in this case.
105 Run-time configuration
106 ~~~~~~~~~~~~~~~~~~~~~~
108 - librte_pmd_mlx4 brings kernel network interfaces up during initialization
109 because it is affected by their state. Forcing them down prevents packets
112 - **ethtool** operations on related kernel interfaces also affect the PMD.
114 - ``port`` parameter [int]
116 This parameter provides a physical port to probe and can be specified multiple
117 times for additional ports. All ports are probed by default if left
120 Kernel module parameters
121 ~~~~~~~~~~~~~~~~~~~~~~~~
123 The **mlx4_core** kernel module has several parameters that affect the
124 behavior and/or the performance of librte_pmd_mlx4. Some of them are described
127 - **num_vfs** (integer or triplet, optionally prefixed by device address
130 Create the given number of VFs on the specified devices.
132 - **log_num_mgm_entry_size** (integer)
134 Device-managed flow steering (DMFS) is required by DPDK applications. It is
135 enabled by using a negative value, the last four bits of which have a
138 - **-1**: force device-managed flow steering (DMFS).
139 - **-7**: configure optimized steering mode to improve performance with the
140 following limitation: VLAN filtering is not supported with this mode.
141 This is the recommended mode in case VLAN filter is not needed.
146 - CRC stripping is supported by default and always reported as "true".
147 The ability to enable/disable CRC stripping requires OFED version
148 4.3-1.5.0.0 and above or rdma-core version v18 and above.
150 - TSO (Transmit Segmentation Offload) is supported in OFED version
156 This driver relies on external libraries and kernel drivers for resources
157 allocations and initialization. The following dependencies are not part of
158 DPDK and must be installed separately:
160 - **libibverbs** (provided by rdma-core package)
162 User space verbs framework used by librte_pmd_mlx4. This library provides
163 a generic interface between the kernel and low-level user space drivers
166 It allows slow and privileged operations (context initialization, hardware
167 resources allocations) to be managed by the kernel and fast operations to
168 never leave user space.
170 - **libmlx4** (provided by rdma-core package)
172 Low-level user space driver library for Mellanox ConnectX-3 devices,
173 it is automatically loaded by libibverbs.
175 This library basically implements send/receive calls to the hardware
180 They provide the kernel-side verbs API and low level device drivers that
181 manage actual hardware initialization and resources sharing with user
184 Unlike most other PMDs, these modules must remain loaded and bound to
187 - mlx4_core: hardware driver managing Mellanox ConnectX-3 devices.
188 - mlx4_en: Ethernet device driver that provides kernel network interfaces.
189 - mlx4_ib: InifiniBand device driver.
190 - ib_uverbs: user space driver for verbs (entry point for libibverbs).
192 - **Firmware update**
194 Mellanox OFED releases include firmware updates for ConnectX-3 adapters.
196 Because each release provides new features, these updates must be applied to
197 match the kernel modules and libraries they come with.
201 Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
204 Depending on system constraints and user preferences either RDMA core library
205 with a recent enough Linux kernel release (recommended) or Mellanox OFED,
206 which provides compatibility with older releases.
208 Current RDMA core package and Linux kernel (recommended)
209 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
211 - Minimal Linux kernel version: 4.14.
212 - Minimal RDMA core version: v15 (see `RDMA core installation documentation`_).
214 - Starting with rdma-core v21, static libraries can be built::
217 CFLAGS=-fPIC cmake -DIN_PLACE=1 -DENABLE_STATIC=1 -GNinja ..
220 .. _`RDMA core installation documentation`: https://raw.githubusercontent.com/linux-rdma/rdma-core/master/README.md
222 If rdma-core libraries are built but not installed, DPDK makefile can link them,
223 thanks to these environment variables:
225 - ``EXTRA_CFLAGS=-I/path/to/rdma-core/build/include``
226 - ``EXTRA_LDFLAGS=-L/path/to/rdma-core/build/lib``
227 - ``PKG_CONFIG_PATH=/path/to/rdma-core/build/lib/pkgconfig``
229 .. _Mellanox_OFED_as_a_fallback:
231 Mellanox OFED as a fallback
232 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
234 - `Mellanox OFED`_ version: **4.4, 4.5**.
235 - firmware version: **2.42.5000** and above.
237 .. _`Mellanox OFED`: http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux_sw_drivers
241 Several versions of Mellanox OFED are available. Installing the version
242 this DPDK release was developed and tested against is strongly
243 recommended. Please check the `prerequisites`_.
245 Installing Mellanox OFED
246 ^^^^^^^^^^^^^^^^^^^^^^^^
248 1. Download latest Mellanox OFED.
250 2. Install the required libraries and kernel modules either by installing
251 only the required set, or by installing the entire Mellanox OFED:
255 .. code-block:: console
257 ./mlnxofedinstall --dpdk --upstream-libs
259 For SR-IOV hypervisors use:
261 .. code-block:: console
263 ./mlnxofedinstall --dpdk --upstream-libs --enable-sriov --hypervisor
265 For SR-IOV virtual machine use:
267 .. code-block:: console
269 ./mlnxofedinstall --dpdk --upstream-libs --guest
271 3. Verify the firmware is the correct one:
273 .. code-block:: console
277 4. Set all ports links to Ethernet, follow instructions on the screen:
279 .. code-block:: console
283 5. Continue with :ref:`section 2 of the Quick Start Guide <QSG_2>`.
288 * Mellanox(R) ConnectX(R)-3 Pro 40G MCX354A-FCC_Ax (2*40G)
295 1. Set all ports links to Ethernet
297 .. code-block:: console
299 PCI=<NIC PCI address>
300 echo eth > "/sys/bus/pci/devices/$PCI/mlx4_port0"
301 echo eth > "/sys/bus/pci/devices/$PCI/mlx4_port1"
305 If using Mellanox OFED one can permanently set the port link
306 to Ethernet using connectx_port_config tool provided by it.
307 :ref:`Mellanox_OFED_as_a_fallback`:
311 2. In case of bare metal or hypervisor, configure optimized steering mode
312 by adding the following line to ``/etc/modprobe.d/mlx4_core.conf``:
314 .. code-block:: console
316 options mlx4_core log_num_mgm_entry_size=-7
320 If VLAN filtering is used, set log_num_mgm_entry_size=-1.
321 Performance degradation can occur on this case.
323 3. Restart the driver:
325 .. code-block:: console
327 /etc/init.d/openibd restart
331 .. code-block:: console
333 service openibd restart
335 4. Compile DPDK and you are ready to go. See instructions on
336 :ref:`Development Kit Build System <Development_Kit_Build_System>`
341 1. Verify the optimized steering mode is configured:
343 .. code-block:: console
345 cat /sys/module/mlx4_core/parameters/log_num_mgm_entry_size
347 2. Use the CPU near local NUMA node to which the PCIe adapter is connected,
348 for better performance. For VMs, verify that the right CPU
349 and NUMA node are pinned according to the above. Run:
351 .. code-block:: console
355 to identify the NUMA node to which the PCIe adapter is connected.
357 3. If more than one adapter is used, and root complex capabilities allow
358 to put both adapters on the same NUMA node without PCI bandwidth degradation,
359 it is recommended to locate both adapters on the same NUMA node.
360 This in order to forward packets from one to the other without
361 NUMA performance penalty.
363 4. Disable pause frames:
365 .. code-block:: console
367 ethtool -A <netdev> rx off tx off
369 5. Verify IO non-posted prefetch is disabled by default. This can be checked
370 via the BIOS configuration. Please contact you server provider for more
371 information about the settings.
375 On some machines, depends on the machine integrator, it is beneficial
376 to set the PCI max read request parameter to 1K. This can be
377 done in the following way:
379 To query the read request size use:
381 .. code-block:: console
383 setpci -s <NIC PCI address> 68.w
385 If the output is different than 3XXX, set it by:
387 .. code-block:: console
389 setpci -s <NIC PCI address> 68.w=3XXX
391 The XXX can be different on different systems. Make sure to configure
392 according to the setpci output.
394 6. To minimize overhead of searching Memory Regions:
396 - '--socket-mem' is recommended to pin memory by predictable amount.
397 - Configure per-lcore cache when creating Mempools for packet buffer.
398 - Refrain from dynamically allocating/freeing memory in run-time.
403 This section demonstrates how to launch **testpmd** with Mellanox ConnectX-3
404 devices managed by librte_pmd_mlx4.
406 #. Load the kernel modules:
408 .. code-block:: console
410 modprobe -a ib_uverbs mlx4_en mlx4_core mlx4_ib
412 Alternatively if MLNX_OFED is fully installed, the following script can
415 .. code-block:: console
417 /etc/init.d/openibd restart
421 User space I/O kernel modules (uio and igb_uio) are not used and do
422 not have to be loaded.
424 #. Make sure Ethernet interfaces are in working order and linked to kernel
425 verbs. Related sysfs entries should be present:
427 .. code-block:: console
429 ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
433 .. code-block:: console
440 #. Optionally, retrieve their PCI bus addresses for whitelisting:
442 .. code-block:: console
445 for intf in eth2 eth3 eth4 eth5;
447 (cd "/sys/class/net/${intf}/device/" && pwd -P);
450 sed -n 's,.*/\(.*\),-w \1,p'
454 .. code-block:: console
463 There are only two distinct PCI bus addresses because the Mellanox
464 ConnectX-3 adapters installed on this system are dual port.
466 #. Request huge pages:
468 .. code-block:: console
470 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
472 #. Start testpmd with basic parameters:
474 .. code-block:: console
476 testpmd -l 8-15 -n 4 -w 0000:83:00.0 -w 0000:84:00.0 -- --rxq=2 --txq=2 -i
480 .. code-block:: console
483 EAL: PCI device 0000:83:00.0 on NUMA socket 1
484 EAL: probe driver: 15b3:1007 librte_pmd_mlx4
485 PMD: librte_pmd_mlx4: PCI information matches, using device "mlx4_0" (VF: false)
486 PMD: librte_pmd_mlx4: 2 port(s) detected
487 PMD: librte_pmd_mlx4: port 1 MAC address is 00:02:c9:b5:b7:50
488 PMD: librte_pmd_mlx4: port 2 MAC address is 00:02:c9:b5:b7:51
489 EAL: PCI device 0000:84:00.0 on NUMA socket 1
490 EAL: probe driver: 15b3:1007 librte_pmd_mlx4
491 PMD: librte_pmd_mlx4: PCI information matches, using device "mlx4_1" (VF: false)
492 PMD: librte_pmd_mlx4: 2 port(s) detected
493 PMD: librte_pmd_mlx4: port 1 MAC address is 00:02:c9:b5:ba:b0
494 PMD: librte_pmd_mlx4: port 2 MAC address is 00:02:c9:b5:ba:b1
495 Interactive-mode selected
496 Configuring Port 0 (socket 0)
497 PMD: librte_pmd_mlx4: 0x867d60: TX queues number update: 0 -> 2
498 PMD: librte_pmd_mlx4: 0x867d60: RX queues number update: 0 -> 2
499 Port 0: 00:02:C9:B5:B7:50
500 Configuring Port 1 (socket 0)
501 PMD: librte_pmd_mlx4: 0x867da0: TX queues number update: 0 -> 2
502 PMD: librte_pmd_mlx4: 0x867da0: RX queues number update: 0 -> 2
503 Port 1: 00:02:C9:B5:B7:51
504 Configuring Port 2 (socket 0)
505 PMD: librte_pmd_mlx4: 0x867de0: TX queues number update: 0 -> 2
506 PMD: librte_pmd_mlx4: 0x867de0: RX queues number update: 0 -> 2
507 Port 2: 00:02:C9:B5:BA:B0
508 Configuring Port 3 (socket 0)
509 PMD: librte_pmd_mlx4: 0x867e20: TX queues number update: 0 -> 2
510 PMD: librte_pmd_mlx4: 0x867e20: RX queues number update: 0 -> 2
511 Port 3: 00:02:C9:B5:BA:B1
512 Checking link statuses...
513 Port 0 Link Up - speed 10000 Mbps - full-duplex
514 Port 1 Link Up - speed 40000 Mbps - full-duplex
515 Port 2 Link Up - speed 10000 Mbps - full-duplex
516 Port 3 Link Up - speed 40000 Mbps - full-duplex