1 .. SPDX-License-Identifier: BSD-3-Clause
2 Copyright(c) 2010-2014 Intel Corporation.
4 .. include:: <isonum.txt>
9 This chapter describes the packages required to compile the DPDK.
11 BIOS Setting Prerequisite on x86
12 --------------------------------
14 For the majority of platforms, no special BIOS settings are needed to use basic DPDK functionality.
15 However, for additional HPET timer and power management functionality,
16 and high performance of small packets, BIOS setting changes may be needed.
17 Consult the section on :ref:`Enabling Additional Functionality <Enabling_Additional_Functionality>`
18 for more information on the required changes.
20 Compilation of the DPDK
21 -----------------------
23 **Required Tools and Libraries:**
27 The setup commands and installed packages needed on various systems may be different.
28 For details on Linux distributions and the versions tested, please consult the DPDK Release Notes.
30 * General development tools including a supported C compiler such as gcc (version 4.9+) or clang (version 3.4+),
31 and ``pkg-config`` or ``pkgconf`` to be used when building end-user binaries against DPDK.
33 * For RHEL/Fedora systems these can be installed using ``dnf groupinstall "Development Tools"``
34 * For Ubuntu/Debian systems these can be installed using ``apt install build-essential``
35 * For Alpine Linux, ``apk add alpine-sdk bsd-compat-headers libexecinfo-dev``
39 pkg-config 0.27, supplied with RHEL-7,
40 does not process the Libs.private section correctly,
41 resulting in statically linked applications not being linked properly.
42 Use an updated version of ``pkg-config`` or ``pkgconf`` instead when building applications
44 * Python 3.5 or later.
46 * Meson (version 0.49.2+) and ninja
48 * ``meson`` & ``ninja-build`` packages in most Linux distributions
49 * If the packaged version is below the minimum version, the latest versions
50 can be installed from Python's "pip" repository: ``pip3 install meson ninja``
52 * ``pyelftools`` (version 0.22+)
54 * For Fedora systems it can be installed using ``dnf install python-pyelftools``
55 * For RHEL/CentOS systems it can be installed using ``pip3 install pyelftools``
56 * For Ubuntu/Debian it can be installed using ``apt install python3-pyelftools``
57 * For Alpine Linux, ``apk add py3-elftools``
59 * Library for handling NUMA (Non Uniform Memory Access).
61 * ``numactl-devel`` in RHEL/Fedora;
62 * ``libnuma-dev`` in Debian/Ubuntu;
63 * ``numactl-dev`` in Alpine Linux
67 Please ensure that the latest patches are applied to third party libraries
68 and software to avoid any known vulnerabilities.
73 * Intel\ |reg| C++ Compiler (icc). For installation, additional libraries may be required.
74 See the icc Installation Guide found in the Documentation directory under the compiler installation.
76 * IBM\ |reg| Advance ToolChain for Powerlinux. This is a set of open source development tools and runtime libraries
77 which allows users to take leading edge advantage of IBM's latest POWER hardware features on Linux. To install
78 it, see the IBM official installation document.
80 **Additional Libraries**
82 A number of DPDK components, such as libraries and poll-mode drivers (PMDs) have additional dependencies.
83 For DPDK builds, the presence or absence of these dependencies will be automatically detected
84 enabling or disabling the relevant components appropriately.
86 In each case, the relevant library development package (``-devel`` or ``-dev``) is needed to build the DPDK components.
88 For libraries the additional dependencies include:
90 * libarchive: for some unit tests using tar to get their resources.
92 * libelf: to compile and use the bpf library.
94 For poll-mode drivers, the additional dependencies for each driver can be
95 found in that driver's documentation in the relevant DPDK guide document,
96 e.g. :doc:`../nics/index`
98 Running DPDK Applications
99 -------------------------
101 To run a DPDK application, some customization may be required on the target machine.
108 * Kernel version >= 4.4
110 The kernel version required is based on the oldest long term stable kernel available
111 at kernel.org when the DPDK version is in development.
112 Compatibility for recent distribution kernels will be kept, notably RHEL/CentOS 7.
114 The kernel version in use can be checked using the command::
118 * glibc >= 2.7 (for features related to cpuset)
120 The version can be checked using the ``ldd --version`` command.
122 * Kernel configuration
124 In the Fedora OS and other common distributions, such as Ubuntu, or Red Hat Enterprise Linux,
125 the vendor supplied kernel configurations can be used to run most DPDK applications.
127 For other kernel builds, options which should be enabled for DPDK include:
131 * PROC_PAGE_MONITOR support
133 * HPET and HPET_MMAP configuration options should also be enabled if HPET support is required.
134 See the section on :ref:`High Precision Event Timer (HPET) Functionality <High_Precision_Event_Timer>` for more details.
136 .. _linux_gsg_hugepages:
138 Use of Hugepages in the Linux Environment
139 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
141 Hugepage support is required for the large memory pool allocation used for packet buffers
142 (the HUGETLBFS option must be enabled in the running kernel as indicated the previous section).
143 By using hugepage allocations, performance is increased since fewer pages are needed,
144 and therefore less Translation Lookaside Buffers (TLBs, high speed translation caches),
145 which reduce the time it takes to translate a virtual page address to a physical page address.
146 Without hugepages, high TLB miss rates would occur with the standard 4k page size, slowing performance.
148 Reserving Hugepages for DPDK Use
149 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
151 The reservation of hugepages can be performed at run time.
152 This is done by echoing the number of hugepages required
153 to a ``nr_hugepages`` file in the ``/sys/kernel/`` directory
154 corresponding to a specific page size (in Kilobytes).
155 For a single-node system, the command to use is as follows
156 (assuming that 1024 of 2MB pages are required)::
158 echo 1024 > /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages
160 On a NUMA machine, the above command will usually divide the number of hugepages
161 equally across all NUMA nodes (assuming there is enough memory on all NUMA nodes).
162 However, pages can also be reserved explicitly on individual NUMA nodes
163 using a ``nr_hugepages`` file in the ``/sys/devices/`` directory::
165 echo 1024 > /sys/devices/system/node/node0/hugepages/hugepages-2048kB/nr_hugepages
166 echo 1024 > /sys/devices/system/node/node1/hugepages/hugepages-2048kB/nr_hugepages
168 The tool ``dpdk-hugepages.py`` can be used to manage hugepages.
172 Some kernel versions may not allow reserving 1 GB hugepages at run time,
173 so reserving them at boot time may be the only option.
174 Please see below for instructions.
178 In the general case, reserving hugepages at run time is perfectly fine,
179 but in use cases where having lots of physically contiguous memory is required,
180 it is preferable to reserve hugepages at boot time,
181 as that will help in preventing physical memory from becoming heavily fragmented.
183 To reserve hugepages at boot time, a parameter is passed to the Linux kernel on the kernel command line.
185 For 2 MB pages, just pass the hugepages option to the kernel. For example, to reserve 1024 pages of 2 MB, use::
189 For other hugepage sizes, for example 1G pages, the size must be specified explicitly and
190 can also be optionally set as the default hugepage size for the system.
191 For example, to reserve 4G of hugepage memory in the form of four 1G pages, the following options should be passed to the kernel::
193 default_hugepagesz=1G hugepagesz=1G hugepages=4
197 The hugepage sizes that a CPU supports can be determined from the CPU flags on Intel architecture.
198 If pse exists, 2M hugepages are supported; if pdpe1gb exists, 1G hugepages are supported.
199 On IBM Power architecture, the supported hugepage sizes are 16MB and 16GB.
203 For 64-bit applications, it is recommended to use 1 GB hugepages if the platform supports them.
205 In the case of a dual-socket NUMA system,
206 the number of hugepages reserved at boot time is generally divided equally between the two sockets
207 (on the assumption that sufficient memory is present on both sockets).
209 See the Documentation/admin-guide/kernel-parameters.txt file in your Linux source tree for further details of these and other kernel options.
211 Using Hugepages with the DPDK
212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
214 If secondary process support is not required, DPDK is able to use hugepages
215 without any configuration by using "in-memory" mode.
216 Please see :doc:`linux_eal_parameters` for more details.
218 If secondary process support is required,
219 mount points for hugepages need to be created.
220 On modern Linux distributions, a default mount point for hugepages
221 is provided by the system and is located at ``/dev/hugepages``.
222 This mount point will use the default hugepage size
223 set by the kernel parameters as described above.
225 However, in order to use hugepage sizes other than the default, it is necessary
226 to manually create mount points for those hugepage sizes (e.g. 1GB pages).
228 To make the hugepages of size 1GB available for DPDK use,
229 following steps must be performed::
232 mount -t hugetlbfs pagesize=1GB /mnt/huge
234 The mount point can be made permanent across reboots, by adding the following line to the ``/etc/fstab`` file::
236 nodev /mnt/huge hugetlbfs pagesize=1GB 0 0