ceph/src/seastar/dpdk/doc/guides/nics/mlx4.rst

   1 ..  BSD LICENSE
   2     Copyright 2012-2015 6WIND S.A.
   3
   4     Redistribution and use in source and binary forms, with or without
   5     modification, are permitted provided that the following conditions
   6     are met:
   7
   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
  13     distribution.
  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.
  17
  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.
  29
  30 MLX4 poll mode driver library
  31 =============================
  32
  33 The MLX4 poll mode driver library (**librte_pmd_mlx4**) implements support
  34 for **Mellanox ConnectX-3** and **Mellanox ConnectX-3 Pro** 10/40 Gbps adapters
  35 as well as their virtual functions (VF) in SR-IOV context.
  36
  37 Information and documentation about this family of adapters can be found on
  38 the `Mellanox website <http://www.mellanox.com>`_. Help is also provided by
  39 the `Mellanox community <http://community.mellanox.com/welcome>`_.
  40
  41 There is also a `section dedicated to this poll mode driver
  42 <http://www.mellanox.com/page/products_dyn?product_family=209&mtag=pmd_for_dpdk>`_.
  43
  44 .. note::
  45
  46    Due to external dependencies, this driver is disabled by default. It must
  47    be enabled manually by setting ``CONFIG_RTE_LIBRTE_MLX4_PMD=y`` and
  48    recompiling DPDK.
  49
  50 Implementation details
  51 ----------------------
  52
  53 Most Mellanox ConnectX-3 devices provide two ports but expose a single PCI
  54 bus address, thus unlike most drivers, librte_pmd_mlx4 registers itself as a
  55 PCI driver that allocates one Ethernet device per detected port.
  56
  57 For this reason, one cannot white/blacklist a single port without also
  58 white/blacklisting the others on the same device.
  59
  60 Besides its dependency on libibverbs (that implies libmlx4 and associated
  61 kernel support), librte_pmd_mlx4 relies heavily on system calls for control
  62 operations such as querying/updating the MTU and flow control parameters.
  63
  64 For security reasons and robustness, this driver only deals with virtual
  65 memory addresses. The way resources allocations are handled by the kernel
  66 combined with hardware specifications that allow it to handle virtual memory
  67 addresses directly ensure that DPDK applications cannot access random
  68 physical memory (or memory that does not belong to the current process).
  69
  70 This capability allows the PMD to coexist with kernel network interfaces
  71 which remain functional, although they stop receiving unicast packets as
  72 long as they share the same MAC address.
  73
  74 Compiling librte_pmd_mlx4 causes DPDK to be linked against libibverbs.
  75
  76 Features
  77 --------
  78
  79 - RSS, also known as RCA, is supported. In this mode the number of
  80   configured RX queues must be a power of two.
  81 - VLAN filtering is supported.
  82 - Link state information is provided.
  83 - Promiscuous mode is supported.
  84 - All multicast mode is supported.
  85 - Multiple MAC addresses (unicast, multicast) can be configured.
  86 - Scattered packets are supported for TX and RX.
  87 - Inner L3/L4 (IP, TCP and UDP) TX/RX checksum offloading and validation.
  88 - Outer L3 (IP) TX/RX checksum offloading and validation for VXLAN frames.
  89 - Secondary process TX is supported.
  90
  91 Limitations
  92 -----------
  93
  94 - RSS hash key cannot be modified.
  95 - RSS RETA cannot be configured
  96 - RSS always includes L3 (IPv4/IPv6) and L4 (UDP/TCP). They cannot be
  97   dissociated.
  98 - Hardware counters are not implemented (they are software counters).
  99 - Secondary process RX is not supported.
 100
 101 Configuration
 102 -------------
 103
 104 Compilation options
 105 ~~~~~~~~~~~~~~~~~~~
 106
 107 These options can be modified in the ``.config`` file.
 108
 109 - ``CONFIG_RTE_LIBRTE_MLX4_PMD`` (default **n**)
 110
 111   Toggle compilation of librte_pmd_mlx4 itself.
 112
 113 - ``CONFIG_RTE_LIBRTE_MLX4_DEBUG`` (default **n**)
 114
 115   Toggle debugging code and stricter compilation flags. Enabling this option
 116   adds additional run-time checks and debugging messages at the cost of
 117   lower performance.
 118
 119 - ``CONFIG_RTE_LIBRTE_MLX4_SGE_WR_N`` (default **4**)
 120
 121   Number of scatter/gather elements (SGEs) per work request (WR). Lowering
 122   this number improves performance but also limits the ability to receive
 123   scattered packets (packets that do not fit a single mbuf). The default
 124   value is a safe tradeoff.
 125
 126 - ``CONFIG_RTE_LIBRTE_MLX4_MAX_INLINE`` (default **0**)
 127
 128   Amount of data to be inlined during TX operations. Improves latency but
 129   lowers throughput.
 130
 131 - ``CONFIG_RTE_LIBRTE_MLX4_TX_MP_CACHE`` (default **8**)
 132
 133   Maximum number of cached memory pools (MPs) per TX queue. Each MP from
 134   which buffers are to be transmitted must be associated to memory regions
 135   (MRs). This is a slow operation that must be cached.
 136
 137   This value is always 1 for RX queues since they use a single MP.
 138
 139 - ``CONFIG_RTE_LIBRTE_MLX4_SOFT_COUNTERS`` (default **1**)
 140
 141   Toggle software counters. No counters are available if this option is
 142   disabled since hardware counters are not supported.
 143
 144 Environment variables
 145 ~~~~~~~~~~~~~~~~~~~~~
 146
 147 - ``MLX4_INLINE_RECV_SIZE``
 148
 149   A nonzero value enables inline receive for packets up to that size. May
 150   significantly improve performance in some cases but lower it in
 151   others. Requires careful testing.
 152
 153 Run-time configuration
 154 ~~~~~~~~~~~~~~~~~~~~~~
 155
 156 - The only constraint when RSS mode is requested is to make sure the number
 157   of RX queues is a power of two. This is a hardware requirement.
 158
 159 - librte_pmd_mlx4 brings kernel network interfaces up during initialization
 160   because it is affected by their state. Forcing them down prevents packets
 161   reception.
 162
 163 - **ethtool** operations on related kernel interfaces also affect the PMD.
 164
 165 - ``port`` parameter [int]
 166
 167   This parameter provides a physical port to probe and can be specified multiple
 168   times for additional ports. All ports are probed by default if left
 169   unspecified.
 170
 171 Kernel module parameters
 172 ~~~~~~~~~~~~~~~~~~~~~~~~
 173
 174 The **mlx4_core** kernel module has several parameters that affect the
 175 behavior and/or the performance of librte_pmd_mlx4. Some of them are described
 176 below.
 177
 178 - **num_vfs** (integer or triplet, optionally prefixed by device address
 179   strings)
 180
 181   Create the given number of VFs on the specified devices.
 182
 183 - **log_num_mgm_entry_size** (integer)
 184
 185   Device-managed flow steering (DMFS) is required by DPDK applications. It is
 186   enabled by using a negative value, the last four bits of which have a
 187   special meaning.
 188
 189   - **-1**: force device-managed flow steering (DMFS).
 190   - **-7**: configure optimized steering mode to improve performance with the
 191     following limitation: VLAN filtering is not supported with this mode.
 192     This is the recommended mode in case VLAN filter is not needed.
 193
 194 Prerequisites
 195 -------------
 196
 197 This driver relies on external libraries and kernel drivers for resources
 198 allocations and initialization. The following dependencies are not part of
 199 DPDK and must be installed separately:
 200
 201 - **libibverbs**
 202
 203   User space verbs framework used by librte_pmd_mlx4. This library provides
 204   a generic interface between the kernel and low-level user space drivers
 205   such as libmlx4.
 206
 207   It allows slow and privileged operations (context initialization, hardware
 208   resources allocations) to be managed by the kernel and fast operations to
 209   never leave user space.
 210
 211 - **libmlx4**
 212
 213   Low-level user space driver library for Mellanox ConnectX-3 devices,
 214   it is automatically loaded by libibverbs.
 215
 216   This library basically implements send/receive calls to the hardware
 217   queues.
 218
 219 - **Kernel modules** (mlnx-ofed-kernel)
 220
 221   They provide the kernel-side verbs API and low level device drivers that
 222   manage actual hardware initialization and resources sharing with user
 223   space processes.
 224
 225   Unlike most other PMDs, these modules must remain loaded and bound to
 226   their devices:
 227
 228   - mlx4_core: hardware driver managing Mellanox ConnectX-3 devices.
 229   - mlx4_en: Ethernet device driver that provides kernel network interfaces.
 230   - mlx4_ib: InifiniBand device driver.
 231   - ib_uverbs: user space driver for verbs (entry point for libibverbs).
 232
 233 - **Firmware update**
 234
 235   Mellanox OFED releases include firmware updates for ConnectX-3 adapters.
 236
 237   Because each release provides new features, these updates must be applied to
 238   match the kernel modules and libraries they come with.
 239
 240 .. note::
 241
 242    Both libraries are BSD and GPL licensed. Linux kernel modules are GPL
 243    licensed.
 244
 245 Currently supported by DPDK:
 246
 247 - Mellanox OFED **4.0-2.0.0.0**.
 248 - Firmware version **2.40.7000**.
 249 - Supported architectures:  **x86_64** and **POWER8**.
 250
 251 Getting Mellanox OFED
 252 ~~~~~~~~~~~~~~~~~~~~~
 253
 254 While these libraries and kernel modules are available on OpenFabrics
 255 Alliance's `website <https://www.openfabrics.org/>`_ and provided by package
 256 managers on most distributions, this PMD requires Ethernet extensions that
 257 may not be supported at the moment (this is a work in progress).
 258
 259 `Mellanox OFED
 260 <http://www.mellanox.com/page/products_dyn?product_family=26&mtag=linux_sw_drivers>`_
 261 includes the necessary support and should be used in the meantime. For DPDK,
 262 only libibverbs, libmlx4, mlnx-ofed-kernel packages and firmware updates are
 263 required from that distribution.
 264
 265 .. note::
 266
 267    Several versions of Mellanox OFED are available. Installing the version
 268    this DPDK release was developed and tested against is strongly
 269    recommended. Please check the `prerequisites`_.
 270
 271 Supported NICs
 272 --------------
 273
 274 * Mellanox(R) ConnectX(R)-3 Pro 40G MCX354A-FCC_Ax (2*40G)
 275
 276 Usage example
 277 -------------
 278
 279 This section demonstrates how to launch **testpmd** with Mellanox ConnectX-3
 280 devices managed by librte_pmd_mlx4.
 281
 282 #. Load the kernel modules:
 283
 284    .. code-block:: console
 285
 286       modprobe -a ib_uverbs mlx4_en mlx4_core mlx4_ib
 287
 288    Alternatively if MLNX_OFED is fully installed, the following script can
 289    be run:
 290
 291    .. code-block:: console
 292
 293       /etc/init.d/openibd restart
 294
 295    .. note::
 296
 297       User space I/O kernel modules (uio and igb_uio) are not used and do
 298       not have to be loaded.
 299
 300 #. Make sure Ethernet interfaces are in working order and linked to kernel
 301    verbs. Related sysfs entries should be present:
 302
 303    .. code-block:: console
 304
 305       ls -d /sys/class/net/*/device/infiniband_verbs/uverbs* | cut -d / -f 5
 306
 307    Example output:
 308
 309    .. code-block:: console
 310
 311       eth2
 312       eth3
 313       eth4
 314       eth5
 315
 316 #. Optionally, retrieve their PCI bus addresses for whitelisting:
 317
 318    .. code-block:: console
 319
 320       {
 321           for intf in eth2 eth3 eth4 eth5;
 322           do
 323               (cd "/sys/class/net/${intf}/device/" && pwd -P);
 324           done;
 325       } |
 326       sed -n 's,.*/\(.*\),-w \1,p'
 327
 328    Example output:
 329
 330    .. code-block:: console
 331
 332       -w 0000:83:00.0
 333       -w 0000:83:00.0
 334       -w 0000:84:00.0
 335       -w 0000:84:00.0
 336
 337    .. note::
 338
 339       There are only two distinct PCI bus addresses because the Mellanox
 340       ConnectX-3 adapters installed on this system are dual port.
 341
 342 #. Request huge pages:
 343
 344    .. code-block:: console
 345
 346       echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages/nr_hugepages
 347
 348 #. Start testpmd with basic parameters:
 349
 350    .. code-block:: console
 351
 352       testpmd -l 8-15 -n 4 -w 0000:83:00.0 -w 0000:84:00.0 -- --rxq=2 --txq=2 -i
 353
 354    Example output:
 355
 356    .. code-block:: console
 357
 358       [...]
 359       EAL: PCI device 0000:83:00.0 on NUMA socket 1
 360       EAL:   probe driver: 15b3:1007 librte_pmd_mlx4
 361       PMD: librte_pmd_mlx4: PCI information matches, using device "mlx4_0" (VF: false)
 362       PMD: librte_pmd_mlx4: 2 port(s) detected
 363       PMD: librte_pmd_mlx4: port 1 MAC address is 00:02:c9:b5:b7:50
 364       PMD: librte_pmd_mlx4: port 2 MAC address is 00:02:c9:b5:b7:51
 365       EAL: PCI device 0000:84:00.0 on NUMA socket 1
 366       EAL:   probe driver: 15b3:1007 librte_pmd_mlx4
 367       PMD: librte_pmd_mlx4: PCI information matches, using device "mlx4_1" (VF: false)
 368       PMD: librte_pmd_mlx4: 2 port(s) detected
 369       PMD: librte_pmd_mlx4: port 1 MAC address is 00:02:c9:b5:ba:b0
 370       PMD: librte_pmd_mlx4: port 2 MAC address is 00:02:c9:b5:ba:b1
 371       Interactive-mode selected
 372       Configuring Port 0 (socket 0)
 373       PMD: librte_pmd_mlx4: 0x867d60: TX queues number update: 0 -> 2
 374       PMD: librte_pmd_mlx4: 0x867d60: RX queues number update: 0 -> 2
 375       Port 0: 00:02:C9:B5:B7:50
 376       Configuring Port 1 (socket 0)
 377       PMD: librte_pmd_mlx4: 0x867da0: TX queues number update: 0 -> 2
 378       PMD: librte_pmd_mlx4: 0x867da0: RX queues number update: 0 -> 2
 379       Port 1: 00:02:C9:B5:B7:51
 380       Configuring Port 2 (socket 0)
 381       PMD: librte_pmd_mlx4: 0x867de0: TX queues number update: 0 -> 2
 382       PMD: librte_pmd_mlx4: 0x867de0: RX queues number update: 0 -> 2
 383       Port 2: 00:02:C9:B5:BA:B0
 384       Configuring Port 3 (socket 0)
 385       PMD: librte_pmd_mlx4: 0x867e20: TX queues number update: 0 -> 2
 386       PMD: librte_pmd_mlx4: 0x867e20: RX queues number update: 0 -> 2
 387       Port 3: 00:02:C9:B5:BA:B1
 388       Checking link statuses...
 389       Port 0 Link Up - speed 10000 Mbps - full-duplex
 390       Port 1 Link Up - speed 40000 Mbps - full-duplex
 391       Port 2 Link Up - speed 10000 Mbps - full-duplex
 392       Port 3 Link Up - speed 40000 Mbps - full-duplex
 393       Done
 394       testpmd>