Embedded Linux Slides

Embedded Linux system development
Embedded Linux system

development
Free Electrons
Free Electrons
Embedded Linux
Developers
Copyright 2004-2015, Free Electrons.

Creative Commons BY-SA 3.0 license.
Latest update: July 19, 2015.
Document updates and sources:
http://free- electrons.com/doc/training/embeddedlinux
Corrections, suggestions, contributions and translations are welcome!
Send them to feedback@free-electrons.com
Free Electrons - Embedded Linux, kernel, drivers and Android - Development, consulting, training and support. http://free-electrons.com
1/540
Rights to copy
Copyright 2004-2015, Free Electrons
License: Creative Commons Attribution - Share Alike 3.0

http://creativecommons.org/licenses/by-sa/3.0/legalcode
You are free:
I to copy, distribute, display, and perform the work
I to make derivative works
I to make commercial use of the work
Under the following conditions:
I Attribution. You must give the original author credit.
I Share Alike. If you alter, transform, or build upon this work, you may distribute
the resulting work only under a license identical to this one.
I For any reuse or distribution, you must make clear to others the license terms of
this work.
I Any of these conditions can be waived if you get permission from the copyright
holder.
Your fair use and other rights are in no way affected by the above.
2/540
Hyperlinks in the document

There are many hyperlinks in the document
I
Regular hyperlinks:
http://kernel.org/
Kernel documentation links:

Documentation/kmemcheck.txt
Links to kernel source files and directories:

drivers/input
include/linux/fb.h
Links to the declarations, definitions and instances of kernel

symbols (functions, types, data, structures):
platform_get_irq()
GFP_KERNEL
struct file_operations
3/540
Free Electrons at a glance

I
Engineering company created in 2004

(not a training company!)
Locations: Orange, Toulouse, Lyon (France)
Serving customers all around the world

See http://free-electrons.com/company/customers/
Head count: 9
Only Free Software enthusiasts!
Focus: Embedded Linux, Linux kernel, Android Free Software

/ Open Source for embedded and real-time systems.
Activities: development, training, consulting, technical

support.
Added value: get the best of the user and development

community and the resources it offers.
4/540
Free Electrons on-line resources

I
All our training materials:

http://free-electrons.com/docs/
Technical blog:
http://free-electrons.com/blog/
Quarterly newsletter:
http://lists.freeelectrons.com/mailman/listinfo/newsletter
News and discussions (Google +):
https://plus.google.com/+FreeElectronsDevelopers
News and discussions (LinkedIn):
http://linkedin.com/groups/Free-Electrons-4501089
Quick news (Twitter):
http://twitter.com/free_electrons
Linux Cross Reference - browse Linux kernel sources on-line:
http://lxr.free-electrons.com
5/540
Generic course information
Generic course
information
Free Electrons
Embedded Linux
Developers
Free Electrons

6/540
Hardware used in this training session

Using Atmel SAMA5D3 Xplained boards in all practical labs
I SAMA5D36 (Cortex A5) CPU from Atmel
I USB powered!
I 256 MB DDR2 RAM, 256 MB NAND
flash
I 2 Ethernet ports (Gigabit + 100 Mbit)
I 2 USB 2.0 host, 1 USB device
I 1 MMC/SD slot
I 3.3 V serial port (like Beaglebone Black)
I Misc: Arduino R3-compatible header,
JTAG, buttons, LEDs

I Currently sold at 69 EUR by Mouser
(V.A.T. not included)
Board and CPU documentation, design files, software:

http://www.atmel.com/sama5d3xplained
7/540
Shopping list: hardware for this course
I Atmel SAMA5D3 Xplained board - Available from
Atmel and multiple distributors (Mouser, Digikey...)

See http://www.atmel.com/tools/ATSAMA5D3XPLD.aspx
I USB Serial Cable - Female ends: Olimex:
http://j.mp/18Hk8yF
I Logitech USB H340 audio headsets
http://www.logitech.com/en-us/product/11608
I An SD card with at least 128 MB of capacity
8/540
Participate!
During the lectures...
I
Dont hesitate to ask questions. Other people in the audience

may have similar questions too.
This helps the trainer to detect any explanation that wasnt

clear or detailed enough.
Dont hesitate to share your experience, for example to

compare Linux / Android with other operating systems used
in your company.
Your point of view is most valuable, because it can be similar

to your colleagues and different from the trainers.
Your participation can make our session more interactive and

make the topics easier to learn.
9/540
Practical lab guidelines

During practical labs...
I
We cannot support more than 8 workstations at once (each

with its board and equipment). Having more would make the
whole class progress slower, compromising the coverage of the
whole training agenda (exception for public sessions: up to 10
people).
So, if you are more than 8 participants, please form up to 8

working groups.
Open the electronic copy of your lecture materials, and use it

throughout the practical labs to find the slides you need again.
Dont copy and paste from the PDF slides.

The slides contain UTF-8 characters that look the same as
ASCII ones, but wont be understood by shells or compilers.
10/540
Cooperate!
As in the Free Software and Open Source community, cooperation

during practical labs is valuable in this training session:
I
If you complete your labs before other people, dont hesitate

to help other people and investigate the issues they face. The
faster we progress as a group, the more time we have to
explore extra topics.
Explain what you understood to other participants when

needed. It also helps to consolidate your knowledge.
Dont hesitate to report potential bugs to your instructor.
Dont hesitate to look for solutions on the Internet as well.
11/540
Command memento sheet

I
This memento sheet gives

command examples for the most
typical needs (looking for files,
extracting a tar archive...)
It saves us 1 day of UNIX / Linux

command line training.
Our best tip: in the command line

shell, always hit the Tab key to
complete command names and file
paths. This avoids 95% of typing
mistakes.
Get an electronic copy on

http://free-electrons.com/
doc/training/embeddedlinux/command_memento.pdf
12/540
vi basic commands
I
The vi editor is very useful to

make quick changes to files in an
embedded target.
Though not very user friendly at

first, vi is very powerful and its
main 15 commands are easy to
learn and are sufficient for 99% of
everyones needs!
Get an electronic copy on

doc/training/embeddedlinux/vi_memento.pdf
You can also take the quick tutorial

by running vimtutor. This is a
worthy investment!
13/540
Introduction to Embedded Linux
Introduction to
Embedded Linux
Free Electrons
Embedded Linux
Developers
Free Electrons

14/540
Birth of free software
1983, Richard Stallman, GNU project and the free software

concept. Beginning of the development of gcc, gdb, glibc and
other important tools
1991, Linus Torvalds, Linux kernel project, a Unix-like

operating system kernel. Together with GNU software and
many other open-source components: a completely free
operating system, GNU/Linux
1995, Linux is more and more popular on server systems
2000, Linux is more and more popular on embedded

systems
2008, Linux is more and more popular on mobile devices
2010, Linux is more and more popular on phones
15/540
Free software?
A program is considered free when its license offers to all its

users the following four freedoms
I
I
I
I
Freedom
Freedom
Freedom
Freedom
to
to
to
to
run the software for any purpose

study the software and to change it
redistribute copies
distribute copies of modified versions
These freedoms are granted for both commercial and

non-commercial use
They imply the availability of source code, software can be

modified and distributed to customers
Good match for embedded systems!
16/540
What is embedded Linux?
Embedded Linux is the usage of the

Linux kernel and various
open-source components in
embedded systems
17/540
Advantages of Linux and open-source

for embedded systems
18/540
Re-using components
I
The key advantage of Linux and open-source in embedded

systems is the ability to re-use components
The open-source ecosystem already provides many

components for standard features, from hardware support to
network protocols, going through multimedia, graphic,
cryptographic libraries, etc.
As soon as a hardware device, or a protocol, or a feature is

wide-spread enough, high chance of having open-source
components that support it.
Allows to quickly design and develop complicated products,

based on existing components.
No-one should re-develop yet another operating system kernel,

TCP/IP stack, USB stack or another graphical toolkit library.
Allows to focus on the added value of your product.
19/540
Low cost
Free software can be duplicated on as many devices as you

want, free of charge.
If your embedded system uses only free software, you can

reduce the cost of software licenses to zero. Even the
development tools are free, unless you choose a commercial
embedded Linux edition.
Allows to have a higher budget for the hardware or to

increase the companys skills and knowledge
20/540
Full control
With open-source, you have the source code for all

components in your system
Allows unlimited modifications, changes, tuning, debugging,

optimization, for an unlimited period of time
Without lock-in or dependency from a third-party vendor
To be true, non open-source components must be avoided

when the system is designed and developed
Allows to have full control over the software part of your

system
21/540
Quality
Many open-source components are widely used, on millions of

systems
Usually higher quality than what an in-house development can

produce, or even proprietary vendors
Of course, not all open-source components are of good

quality, but most of the widely-used ones are.
Allows to design your system with high-quality

components at the foundations
22/540
Eases testing of new features
Open-source being freely available, it is easy to get a piece of

software and evaluate it
Allows to easily study several options while making a choice
Much easier than purchasing and demonstration procedures

needed with most proprietary products
Allows to easily explore new possibilities and solutions
23/540
Community support
Open-source software components are developed by

communities of developers and users
This community can provide a high-quality support: you can

directly contact the main developers of the component you
are using. The likelyhood of getting an answer doesnt depend
what company you work for.
Often better than traditional support, but one needs to

understand how the community works to properly use the
community support possibilities
Allows to speed up the resolution of problems when

developing your system
24/540
Taking part into the community

I
Possibility of taking part into the development community of

some of the components used in the embedded systems: bug
reporting, test of new versions or features, patches that fix
bugs or add new features, etc.
Most of the time the open-source components are not the

core value of the product: its the interest of everybody to
contribute back.
For the engineers: a very motivating way of being recognized

outside the company, communication with others in the same
field, opening of new possibilities, etc.
For the managers: motivation factor for engineers, allows

the company to be recognized in the open-source community
and therefore get support more easily and be more attractive
to open-source developers
25/540
A few examples of embedded systems

running Linux
26/540
Personal routers
27/540
Television
28/540
Point of sale terminal
29/540
Laser cutting machine
30/540
Viticulture machine
31/540
Embedded hardware for Linux

systems
32/540
Processor and architecture (1)
The Linux kernel and most other architecture-dependent

components support a wide range of 32 and 64 bits
architectures
I
I
I
I
I
I
I
I
x86 and x86-64, as found on PC platforms, but also embedded

systems (multimedia, industrial)
ARM, with hundreds of different SoC (multimedia, industrial)
PowerPC (mainly real-time, industrial applications)
MIPS (mainly networking applications)
SuperH (mainly set top box and multimedia applications)
Blackfin (DSP architecture)
Microblaze (soft-core for Xilinx FPGA)
Coldfire, SCore, Tile, Xtensa, Cris, FRV, AVR32, M32R
33/540
Processor and architecture (2)
Both MMU and no-MMU architectures are supported, even

though no-MMU architectures have a few limitations.
Linux is not designed for small microcontrollers.
Besides the toolchain, the bootloader and the kernel, all other
components are generally architecture-independent
34/540
RAM and storage
RAM: a very basic Linux system can work within 8 MB of

RAM, but a more realistic system will usually require at least
32 MB of RAM. Depends on the type and size of applications.
Storage: a very basic Linux system can work within 4 MB of
storage, but usually more is needed.
I
Flash storage is supported, both NAND and NOR flash, with

specific filesystems
Block storage including SD/MMC cards and eMMC is
supported
Not necessarily interesting to be too restrictive on the amount

of RAM/storage: having flexibility at this level allows to
re-use as many existing components as possible.
35/540
Communication
The Linux kernel has support for many common

communication buses
I
I
I
I
I
I
I2C
SPI
CAN
1-wire
SDIO
USB
And also extensive networking support

I
I
I
Ethernet, Wifi, Bluetooth, CAN, etc.

IPv4, IPv6, TCP, UDP, SCTP, DCCP, etc.
Firewalling, advanced routing, multicast
36/540
Types of hardware platforms

I
Evaluation platforms from the SoC vendor. Usually

expensive, but many peripherals are built-in. Generally
unsuitable for real products.
Component on Module, a small board with only

CPU/RAM/flash and a few other core components, with
connectors to access all other peripherals. Can be used to
build end products for small to medium quantities.
Community development platforms, to make a particular

SoC popular and easily available. These are ready-to-use and
low cost, but usually have less peripherals than evaluation
platforms. To some extent, can also be used for real products.
Custom platform. Schematics for evaluation boards or

development platforms are more and more commonly freely
available, making it easier to develop custom platforms.
37/540
Criteria for choosing the hardware

I
Make sure the hardware you plan to use is already supported

by the Linux kernel, and has an open-source bootloader,
especially the SoC youre targeting.
Having support in the official versions of the projects (kernel,

bootloader) is a lot better: quality is better, and new versions
are available.
Some SoC vendors and/or board vendors do not contribute

their changes back to the mainline Linux kernel. Ask them to
do so, or use another product if you can. A good measurement
is to see the delta between their kernel and the official one.
Between properly supported hardware in the official

Linux kernel and poorly-supported hardware, there will
be huge differences in development time and cost.
38/540
Embedded Linux system architecture
39/540
Global architecture
40/540
Software components
I
Cross-compilation toolchain
I
Bootloader
I
Contains the process and memory management, network stack,

device drivers and provides services to user space applications
C library
I
Started by the hardware, responsible for basic initialization,

loading and executing the kernel
Linux Kernel
I
Compiler that runs on the development machine, but generates

code for the target
The interface between the kernel and the user space

applications
Libraries and applications

I
Third-party or in-house
41/540
Embedded Linux work

Several distinct tasks are needed when deploying embedded Linux
in a product:
I
Board Support Package development

I
System integration
I
A BSP contains a bootloader and kernel with the suitable

device drivers for the targeted hardware
Purpose of our Kernel Development training
Integrate all the components, bootloader, kernel, third-party
libraries and applications and in-house applications into a
working system
Purpose of this training
Development of applications
I
Normal Linux applications, but using specifically chosen

libraries
42/540
Embedded Linux development environment
Embedded Linux
development
environment
Free Electrons
Embedded Linux
Developers
Free Electrons

43/540
Embedded Linux solutions
Two ways to switch to embedded Linux

I
Use solutions provided and supported by vendors like

MontaVista, Wind River or TimeSys. These solutions come
with their own development tools and environment. They use
a mix of open-source components and proprietary tools.
Use community solutions. They are completely open,
supported by the community.
In Free Electrons training sessions, we do not promote a

particular vendor, and therefore use community solutions
I
However, knowing the concepts, switching to vendor solutions

will be easy
44/540
OS for Linux development

I
We strongly recommend to use Linux as the desktop operating

system to embedded Linux developers, for multiple reasons.
All community tools are developed and designed to run on

Linux. Trying to use them on other operating systems
(Windows, Mac OS X) will lead to trouble, and their usage on
these systems is generally not supported by community
developers.
As Linux also runs on the embedded device, all the knowledge

gained from using Linux on the desktop will apply similarly to
the embedded device.
45/540
Desktop Linux distribution

I
Any good and sufficiently recent Linux

desktop distribution can be used for the
development workstation
I
Ubuntu, Debian, Fedora, openSUSE,

Red Hat, etc.
We have chosen Ubuntu, as it is a widely

used and easy to use desktop Linux
distribution
The Ubuntu setup on the training laptops

has intentionally been left untouched after
the normal installation process. Learning
embedded Linux is also about learning the
tools needed on the development
workstation!
46/540
Linux root and non-root users
Linux is a multi-user operating system

I
The root user is the administrator, and it can do privileged

operations such as: mounting filesystems, configuring the
network, creating device files, changing the system
configuration, installing or removing software
All other users are unprivileged, and cannot perform these
administrator-level operations
On an Ubuntu system, it is not possible to log in as root,

only as a normal user.
The system has been configured so that the user account

created first is allowed to run privileged operations through a
program called sudo.
I
Example: sudo mount /dev/sda2 /mnt/disk
47/540
Software packages
I
The distribution mechanism for software in GNU/Linux is

different from the one in Windows
Linux distributions provides a central and coherent way of

installing, updating and removing applications and libraries:
packages
Packages contains the application or library files, and
associated meta-information, such as the version and the
dependencies
.deb on Debian and Ubuntu, .rpm on Red Hat, Fedora,

openSUSE
Packages are stored in repositories, usually on HTTP or FTP

servers
You should only use packages from official repositories for

your distribution, unless strictly required.
48/540
Managing software packages (1)
Instructions for Debian based GNU/Linux systems

(Debian, Ubuntu...)
I
Package repositories are specified in

/etc/apt/sources.list
To update package repository lists:

sudo apt-get update
To find the name of a package to install, the best is to use the

search engine on http://packages.debian.org or on
http://packages.ubuntu.com. You may also use:
apt-cache search <keyword>
49/540
Managing software packages (2)

I
To install a given package:

sudo apt-get install <package>
To remove a given package:

sudo apt-get remove <package>
To install all available package updates:

sudo apt-get dist-upgrade
Get information about a package:

apt-cache show <package>
Graphical interfaces
I
I
Synaptic for GNOME

KPackageKit for KDE
Further details on package management:

http://www.debian.org/doc/manuals/apt-howto/
50/540
Host vs. target

I
When doing embedded development, there is always a split

between
I
The host, the development workstation, which is typically a

powerful PC
The target, which is the embedded system under development
They are connected by various means: almost always a serial

line for debugging purposes, frequently an Ethernet
connection, sometimes a JTAG interface for low-level
debugging
51/540
Serial line communication program
An essential tool for embedded development is a serial line

communication program, like HyperTerminal in Windows.
There are multiple options available in Linux: Minicom,

Picocom, Gtkterm, Putty, etc.
In this training session, we recommend using the simplest of
them: picocom
I
I
I
Installation with sudo apt-get install picocom

Run with picocom -b BAUD_RATE /dev/SERIAL_DEVICE
Exit with Control-A Control-X
SERIAL_DEVICE is typically
I
I
ttyUSBx for USB to serial converters

ttySx for real serial ports
52/540
Command line tips

I
Using the command line is mandatory for many operations

needed for embedded Linux development
It is a very powerful way of interacting with the system, with

which you can save a lot of time.
Some useful tips
I
I
You can use several tabs in the Gnome Terminal

Remember that you can use relative paths (for example:
../../linux) in addition to absolute paths (for example:
/home/user)
In a shell, hit [Control] [r], then a keyword, will search
through the command history. Hit [Control] [r] again to
search backwards in the history
You can copy/paste paths directly from the file manager to the
terminal by drag-and-drop.
53/540
Practical lab - Training Setup
Prepare your lab environment

I
Download the lab archive
Enforce correct permissions
54/540
Cross-compiling toolchains
Cross-compiling
toolchains
Free Electrons
Embedded Linux
Developers
Free Electrons

55/540
Definition and Components
56/540
Definition (1)
The usual development tools available on a GNU/Linux

workstation is a native toolchain
This toolchain runs on your workstation and generates code

for your workstation, usually x86
For embedded system development, it is usually impossible or
not interesting to use a native toolchain
I
I
I
The target is too restricted in terms of storage and/or memory

The target is very slow compared to your workstation
You may not want to install all development tools on your
target.
Therefore, cross-compiling toolchains are generally used.

They run on your workstation but generate code for your
target.
57/540
Definition (2)
58/540
Machines in build procedures
Three machines must be distinguished when discussing

toolchain creation
I
I
I
The build machine, where the toolchain is built.

The host machine, where the toolchain will be executed.
The target machine, where the binaries created by the
toolchain are executed.
Four common build types are possible for toolchains
59/540
Different toolchain build procedures
60/540
Components
61/540
Binutils
Binutils is a set of tools to generate and manipulate binaries

for a given CPU architecture
I
I
I
I
as, the assembler, that generates binary code from assembler

source code
ld, the linker
ar, ranlib, to generate .a archives, used for libraries
objdump, readelf, size, nm, strings, to inspect binaries.
Very useful analysis tools!
strip, to strip useless parts of binaries in order to reduce their
size
http://www.gnu.org/software/binutils/
GPL license
62/540
Kernel headers (1)

I
The C library and compiled

programs needs to interact with
the kernel
I
I
I
Available system calls and their

numbers
Constant definitions
Data structures, etc.
Therefore, compiling the C library

requires kernel headers, and many
applications also require them.
Available in <linux/...> and

<asm/...> and a few other
directories corresponding to the
ones visible in include/ in the
kernel sources
63/540
Kernel headers (2)

I
System call numbers, in <asm/unistd.h>

#define __NR_exit
#define __NR_fork
#define __NR_read
1
2
3
Constant definitions, here in <asm-generic/fcntl.h>,

included from <asm/fcntl.h>, included from
<linux/fcntl.h>
#define O_RDWR 00000002
Data structures, here in <asm/stat.h>

struct stat {
unsigned long st_dev;
unsigned long st_ino;
[...]
};
64/540
Kernel headers (3)
The kernel to user space ABI is backward compatible

I
Binaries generated with a toolchain using kernel headers older

than the running kernel will work without problem, but wont
be able to use the new system calls, data structures, etc.
Binaries generated with a toolchain using kernel headers newer
than the running kernel might work on if they dont use the
recent features, otherwise they will break
Using the latest kernel headers is not necessary, unless access
to the new kernel features is needed
The kernel headers are extracted from the kernel sources using
the headers_install kernel Makefile target.
65/540
GCC
GNU Compiler Collection, the famous free

software compiler
Can compile C, C++, Ada, Fortran, Java,

Objective-C, Objective-C++, and generate code
for a large number of CPU architectures,
including ARM, AVR, Blackfin, CRIS, FRV,
M32, MIPS, MN10300, PowerPC, SH, v850,
i386, x86 64, IA64, Xtensa, etc.
http://gcc.gnu.org/
Available under the GPL license, libraries under

the LGPL.
66/540
C library
The C library is an essential component of

a Linux system
I
Interface between the applications and

the kernel
Provides the well-known standard C API
to ease application development
Several C libraries are available:

glibc, uClibc, eglibc, dietlibc, newlib, etc.
The choice of the C library must be made

at the time of the cross-compiling
toolchain generation, as the GCC compiler
is compiled against a specific C library.
67/540
C Libraries
68/540
glibc
License: LGPL
C library from the GNU project
Designed for performance, standards

compliance and portability
Found on all GNU / Linux host systems
Of course, actively maintained
Quite big for small embedded systems:

approx 2.5 MB on ARM (version 2.9 libc: 1.5 MB, libm: 750 KB)
http://www.gnu.org/software/libc/
69/540
uClibc
I
http://www.uclibc.org/
License: LGPL
Lightweight C library for small embedded systems
I
I
High configurability: many features can be enabled or disabled

through a menuconfig interface
Works only with Linux/uClinux, works on most embedded
architectures
No guaranteed binary compatibility. May need to recompile
applications when the library configuration changes.
Focus on size rather than performance
Small compile time
News: project revived by the uClibc-ng project, making

frequent releases again, and now supported by Buildroot (see
http://uclibc-ng.org/).
70/540
uClibc (2)
I
Most of the applications compile with uClibc. This applies to

all applications used in embedded systems.
Size (arm): 4 times smaller than glibc!
I
uClibc 0.9.30.1: approx. 600 KB (libuClibc: 460 KB, libm:

96KB)
glibc 2.9: approx 2.5 MB
Some features not available or limited: priority-inheritance

mutexes, NPTL support is very new, fixed Name Service
Switch functionality, etc.
Used on a large number of production embedded products,

including consumer electronic devices
Supported by all commercial embedded Linux providers (proof

of maturity).
71/540
eglibc
I
Embedded glibc, LGPL license too
Was a variant of glibc, better adapted to the

needs of embedded systems, because of past
disagreement with glibc maintainers.
eglibcs goals included reduced footprint,

configurable components, better support for
cross-compilation and cross-testing.
Could be built without support for NIS, locales,

IPv6, and many other features.
Fortunately for eglibc, the glibc maintainer has

changed and its features are now included in
glibc. Version 2.19 (Feb. 2014) was the last
release.
http:
//en.wikipedia.org/wiki/Embedded_GLIBC
72/540
Honey, I shrunk the programs!
Executable size comparison on ARM, tested with eglibc 2.15

and uClibc 0.9.33.2
Plain hello world program (stripped):

helloworld
static dynamic
uClibc
18kB
2.5kB
uClibc with Thumb-2
14kB
2.4kB
eglibc with Thumb-2 361kB
2.7kB
Busybox (stripped):
busybox
uClibc
uClibc with Thumb-2
eglibc with Thumb-2
static
750kB
533kB
934kB
dynamic
603kB
439kB
444kB
73/540
Other smaller C libraries
Several other smaller C libraries have been developed, but

none of them have the goal of allowing the compilation of
large existing applications
They need specially written programs and applications

Choices:
I
I
Dietlibc, http://www.fefe.de/dietlibc/. Approximately

70 KB.
Newlib, http://sourceware.org/newlib/
Klibc, http://www.kernel.org/pub/linux/libs/klibc/,
designed for use in an initramfs or initrd at boot time.
74/540
Toolchain Options
75/540
ABI
I
When building a toolchain, the ABI used to generate binaries

needs to be defined
ABI, for Application Binary Interface, defines the calling

conventions (how function arguments are passed, how the
return value is passed, how system calls are made) and the
organization of structures (alignment, etc.)
All binaries in a system must be compiled with the same ABI,

and the kernel must understand this ABI.
On ARM, two main ABIs: OABI and EABI
Nowadays everybody uses EABI
On MIPS, several ABIs: o32, o64, n32, n64
http://en.wikipedia.org/wiki/Application_Binary_
Interface
76/540
Floating point support

I
Some processors have a floating point unit, some others do

not.
I
For example, many ARMv4 and ARMv5 CPUs do not have a

floating point unit. Since ARMv7, a VFP unit is mandatory.
For processors having a floating point unit, the toolchain

should generate hard float code, in order to use the floating
point instructions directly
For processors without a floating point unit, two solutions
I
Generate hard float code and rely on the kernel to emulate the
floating point instructions. This is very slow.
Generate soft float code, so that instead of generating floating
point instructions, calls to a user space library are generated
Decision taken at toolchain configuration time
Also possible to configure which floating point unit should be

used
77/540
CPU optimization flags
A set of cross-compiling tools is specific to a CPU

architecture (ARM, x86, MIPS, PowerPC)
However, with the -march=, -mcpu=, -mtune= options, one
can select more precisely the target CPU type
I
At the toolchain compilation time, values can be chosen.

They are used:
I
For example, -march=armv7 -mcpu=cortex-a8
As the default values for the cross-compiling tools, when no

other -march, -mcpu, -mtune options are passed
To compile the C library
Even if the C library has been compiled for armv5t, it doesnt

prevent from compiling other programs for armv7
78/540
Obtaining a Toolchain
79/540
Building a toolchain manually

Building a cross-compiling toolchain by yourself is a difficult and
painful task! Can take days or weeks!
I
Lots of details to learn: many components to build,

complicated configuration
Lots of decisions to make (such as C library version, ABI,

floating point mechanisms, component versions)
Need kernel headers and C library sources
Need to be familiar with current gcc issues and patches on

your platform
Useful to be familiar with building and configuring tools
See the Crosstool-NG docs/ directory for details on how

toolchains are built.
80/540
Get a pre-compiled toolchain

I
Solution that many people choose

I
I
Advantage: it is the simplest and most convenient solution

Drawback: you cant fine tune the toolchain to your needs
Determine what toolchain you need: CPU, endianness, C

library, component versions, ABI, soft float or hard float, etc.
Check whether the available toolchains match your

requirements.
Possible choices
I
I
I
Sourcery CodeBench toolchains

Linaro toolchains
Toolchains packaged by your distribution
Ubuntu example:
sudo apt-get install gcc-arm-linux-gnueabi
More references at http://elinux.org/Toolchains
81/540
Sourcery CodeBench
I
CodeSourcery was a company with extended expertise on free

software toolchains. It has been bought by Mentor Graphics.
They sell toolchains with support, but also provide a Lite

version, that is free and usable for commercial products
They have toolchains available for
I
I
I
I
MIPS
NIOS-II
AMD64
Hexagon
They stopped support for ARM, PowerPC, SuperH, x86!
Be sure to use the GNU/Linux versions. The EABI versions

are for bare-metal development (no operating system)
http://mentor.com/embedded-software/sourcerytools/sourcery-codebench/editions/lite-edition/
82/540
Linaro toolchains
Linaro contributes to improving mainline gcc on ARM,

in particular by hiring CodeSourcery developers.
For people who cant wait for the next releases of gcc,
Linaro releases modified sources of stable releases of
gcc, with these optimizations for ARM (mainly for
recent Cortex A CPUs).
As any gcc release, these sources can be used by build

tools to build their own binary toolchains (Buildroot,
OpenEmbedded...). This allows to support glibc,
uClibc and eglibc.
https://wiki.linaro.org/WorkingGroups/ToolChain
83/540
Installing and using a pre-compiled toolchain
Follow the installation procedure proposed by the vendor
Usually, it is simply a matter of extracting a tarball wherever

you want.
Then, add the path to toolchain binaries in your PATH:

export PATH=/path/to/toolchain/bin/:$PATH
Finally, compile your applications

PREFIX-gcc -o foobar foobar.c
PREFIX depends on the toolchain configuration, and allows to

distinguish cross-compilation tools from native compilation
utilities
84/540
Toolchain building utilities
Another solution is to use utilities that automate the process of

building the toolchain
I
Same advantage as the pre-compiled toolchains: you dont

need to mess up with all the details of the build process
But also offers more flexibility in terms of toolchain

configuration, component version selection, etc.
They also usually contain several patches that fix known

issues with the different components on some architectures
Multiple tools with identical principle: shell scripts or Makefile

that automatically fetch, extract, configure, compile and
install the different components
85/540
Toolchain building utilities (2)
Crosstool-ng
I
I
I
Rewrite of the older Crosstool, with a menuconfig-like

configuration system
Feature-full: supports uClibc, glibc, eglibc, hard and soft float,
many architectures
Actively maintained
http://crosstool-ng.org/
86/540
Toolchain building utilities (3)

Many root filesystem build systems also allow the construction of a
cross-compiling toolchain
I Buildroot
I
PTXdist
I
Makefile-based. Can build (e)glibc, uClibc and musl based

toolchains, for a wide range of architectures.
http://www.buildroot.net
Makefile-based, uClibc or glibc, maintained mainly by
Pengutronix
http://pengutronix.de/software/ptxdist/
OpenEmbedded / Yocto
I
I
I
A featureful, but more complicated build system

http://www.openembedded.org/
https://www.yoctoproject.org/
87/540
Crosstool-NG: installation and usage

I
Installation of Crosstool-NG can be done system-wide, or just

locally in the source directory. For local installation:
./configure --enable-local
make
make install
Some sample configurations for various architectures are

available in samples, they can be listed using
./ct-ng list-samples
To load a sample configuration

./ct-ng <sample-name>
To adjust the configuration

./ct-ng menuconfig
To build the toolchain

./ct-ng build
88/540
Toolchain contents
I
The cross compilation tool binaries, in bin/

I
One or several sysroot, each containing

I
I
This directory can be added to your PATH to ease usage of the

toolchain
The C library and related libraries, compiled for the target
The C library headers and kernel headers
There is one sysroot for each variant: toolchains can be

multilib if they have several copies of the C library for different
configurations (for example: ARMv4T, ARMv5T, etc.)
I
CodeSourcery ARM toolchain are multilib, the sysroots are in

arm-none-linux-gnueabi/libc/,
arm-none-linux-gnueabi/libc/armv4t/,
arm-none-linux-gnueabi/libc/thumb2
Crosstool-NG toolchains can be multilib too (still
experimental), otherwise the sysroot is in
arm-unknown-linux-uclibcgnueabi/sysroot
89/540
Practical lab - Using Crosstool-NG
Time to build your toolchain

I
Configure Crosstool-NG
Run it to build your own

cross-compiling toolchain
90/540
Bootloaders
Free Electrons
Bootloaders
Embedded Linux
Developers
Free Electrons

91/540
Bootloaders
Boot Sequence
92/540
Bootloaders
The bootloader is a piece of code responsible for

I
I
I
I
Basic hardware initialization

Loading of an application binary, usually an operating system
kernel, from flash storage, from the network, or from another
type of non-volatile storage.
Possibly decompression of the application binary
Execution of the application
Besides these basic functions, most bootloaders provide a shell

with various commands implementing different operations.
I
Loading of data from storage or network, memory inspection,

hardware diagnostics and testing, etc.
93/540
Bootloaders on x86 (1)

I
The x86 processors are typically bundled on a

board with a non-volatile memory containing a
program, the BIOS.
This program gets executed by the CPU after
reset, and is responsible for basic hardware
initialization and loading of a small piece of code
from non-volatile storage.
I
This piece of code is usually the first 512 bytes

of a storage device
This piece of code is usually a 1st stage

bootloader, which will load the full bootloader
itself.
The bootloader can then offer all its features. It

typically understands filesystem formats so that
the kernel file can be loaded directly from a
normal filesystem.
94/540
Bootloaders on x86 (2)
GRUB, Grand Unified Bootloader, the most powerful one.

http://www.gnu.org/software/grub/
I
Can read many filesystem formats to load the kernel image and
the configuration, provides a powerful shell with various
commands, can load kernel images over the network, etc.
See our dedicated presentation for details:
http://free-electrons.com/docs/grub/
Syslinux, for network and removable media booting (USB key,

CD-ROM)
http://www.kernel.org/pub/linux/utils/boot/syslinux/
95/540
Booting on embedded CPUs: case 1

I
When powered, the CPU starts executing code

at a fixed address
There is no other booting mechanism provided

by the CPU
The hardware design must ensure that a NOR

flash chip is wired so that it is accessible at the
address at which the CPU starts executing
instructions
The first stage bootloader must be programmed

at this address in the NOR
NOR is mandatory, because it allows random

access, which NAND doesnt allow
Not very common anymore (unpractical, and

requires NOR flash)
96/540
Booting on embedded CPUs: case 2

I
The CPU has an integrated boot code in ROM

I
I
This boot code is able to load a first stage bootloader from a

storage device into an internal SRAM (DRAM not initialized
yet)
I
Storage device can typically be: MMC, NAND, SPI flash,

UART (transmitting data over the serial line), etc.
The first stage bootloader is

I
I
BootROM on AT91 CPUs, ROM code on OMAP, etc.

Exact details are CPU-dependent
Limited in size due to hardware constraints (SRAM size)

Provided either by the CPU vendor or through community
projects
This first stage bootloader must initialize DRAM and other

hardware devices and load a second stage bootloader into
RAM
97/540
Booting on ARM Atmel AT91
I RomBoot: tries to find a valid bootstrap image
from various storage sources, and load it into

SRAM (DRAM not initialized yet). Size limited
to 4 KB. No user interaction possible in standard
boot mode.
I AT91Bootstrap: runs from SRAM. Initializes the
DRAM, the NAND or SPI controller, and loads

the secondary bootloader into RAM and starts it.
No user interaction possible.
I U-Boot: runs from RAM. Initializes some other
hardware devices (network, USB, etc.). Loads the

kernel image from storage or network to RAM
and starts it. Shell with commands provided.
I Linux Kernel: runs from RAM. Takes over the
system completely (bootloaders no longer exists).
98/540
Booting on ARM TI OMAP3

I ROM Code: tries to find a valid bootstrap image

SRAM or RAM (RAM can be initialized by ROM
code through a configuration header). Size
limited to <64 KB. No user interaction possible.
I X-Loader or U-Boot: runs from SRAM.
Initializes the DRAM, the NAND or MMC

controller, and loads the secondary bootloader
into RAM and starts it. No user interaction
possible. File called MLO.

and starts it. Shell with commands provided. File
called u-boot.bin or u-boot.img.

99/540
Booting on Marvell SoC
I ROM Code: tries to find a valid bootstrap image

RAM. The RAM configuration is described in a
CPU-specific header, prepended to the bootloader
image.

and starts it. Shell with commands provided. File
called u-boot.kwb.
100/540
Generic bootloaders for embedded CPUs

I
We will focus on the generic part, the main bootloader,

offering the most important features.
There are several open-source generic bootloaders.
Here are the most popular ones:
I
U-Boot, the universal bootloader by Denx

The most used on ARM, also used on PPC, MIPS, x86, m68k,
NIOS, etc. The de-facto standard nowadays. We will study it
in detail.
http://www.denx.de/wiki/U-Boot
Barebox, a new architecture-neutral bootloader, written as a
successor of U-Boot. Better design, better code, active
development, but doesnt yet have as much hardware support
as U-Boot.
http://www.barebox.org
There are also a lot of other open-source or proprietary

bootloaders, often architecture-specific
I
RedBoot, Yaboot, PMON, etc.
101/540
Bootloaders
The U-boot bootloader
102/540
U-Boot
U-Boot is a typical free software project
I
License: GPLv2 (same as Linux)
Freely available at http://www.denx.de/wiki/U-Boot
Documentation available at
http://www.denx.de/wiki/U-Boot/Documentation
The latest development source code is available in a Git

repository:
http://git.denx.de/?p=u-boot.git;a=summary
Development and discussions happen around an open

mailing-list http://lists.denx.de/pipermail/u-boot/
Since the end of 2008, it follows a fixed-interval release

schedule. Every three months, a new version is released.
Versions are named YYYY.MM.
103/540
U-Boot configuration
I
I
Get the source code from the website, and uncompress it

The include/configs/ directory contains one configuration
file for each supported board
I
It defines the CPU type, the peripherals and their

configuration, the memory mapping, the U-Boot features that
should be compiled in, etc.
It is a simple .h file that sets C pre-processor constants. See
the README file for the documentation of these constants. This
file can also be adjusted to add or remove features from
U-Boot (commands, etc.).
Assuming that your board is already supported by U-Boot,

there should be one entry corresponding to your board in the
boards.cfg file.
I
I
Run ./tools/genboardscfg.py to generate it.

Or just look in the configs/ directory.
104/540
U-Boot configuration file excerpt
/* CPU configuration */
#define CONFIG_ARMV7 1
#define CONFIG_OMAP 1
#define CONFIG_OMAP34XX 1
#define CONFIG_OMAP3430 1
#define CONFIG_OMAP3_IGEP0020 1
[...]
/* Memory configuration */
#define CONFIG_NR_DRAM_BANKS 2
#define PHYS_SDRAM_1 OMAP34XX_SDRC_CS0
#define PHYS_SDRAM_1_SIZE (32 << 20)
#define PHYS_SDRAM_2 OMAP34XX_SDRC_CS1
[...]
/* USB configuration */
#define CONFIG_MUSB_UDC 1
#define CONFIG_USB_OMAP3 1
#define CONFIG_TWL4030_USB 1
[...]
/* Available commands and features */

#define CONFIG_CMD_CACHE
#define CONFIG_CMD_EXT2
#define CONFIG_CMD_FAT
#define CONFIG_CMD_I2C
#define CONFIG_CMD_MMC
#define CONFIG_CMD_NAND
#define CONFIG_CMD_NET
#define CONFIG_CMD_DHCP
#define CONFIG_CMD_PING
#define CONFIG_CMD_NFS
#define CONFIG_CMD_MTDPARTS
[...]
105/540
Configuring and compiling U-Boot
U-Boot must be configured before being compiled

I
I
make BOARDNAME_config
Where BOARDNAME is the name of the board, as visible in the
boards.cfg file (first column).
Make sure that the cross-compiler is available in PATH
Compile U-Boot, by specifying the cross-compiler prefix.

Example, if your cross-compiler executable is arm-linux-gcc:
make CROSS_COMPILE=arm-linux-
The main result is a u-boot.bin file, which is the U-Boot

image. Depending on your specific platform, there may be
other specialized images: u-boot.img, u-boot.kwb, MLO,
etc.
106/540
Installing U-Boot
I
U-Boot must usually be installed in flash memory to be

executed by the hardware. Depending on the hardware, the
installation of U-Boot is done in a different way:
I
The CPU provides some kind of specific boot monitor with

which you can communicate through serial port or USB using
a specific protocol
The CPU boots first on removable media (MMC) before
booting from fixed media (NAND). In this case, boot from
MMC to reflash a new version
U-Boot is already installed, and can be used to flash a new
version of U-Boot. However, be careful: if the new version of
U-Boot doesnt work, the board is unusable
The board provides a JTAG interface, which allows to write to
the flash memory remotely, without any system running on the
board. It also allows to rescue a board if the bootloader
doesnt work.
107/540
U-boot prompt
I
I
Connect the target to the host through a serial console

Power-up the board. On the serial console, you will see
something like:
U-Boot 2013.04 (May 29 2013 - 10:30:21)

OMAP36XX/37XX-GP ES1.2, CPU-OPP2, L3-165MHz, Max CPU Clock 1 Ghz
IGEPv2 + LPDDR/NAND
I2C:
ready
DRAM: 512 MiB
NAND: 512 MiB
MMC:
OMAP SD/MMC: 0
Die ID #255000029ff800000168580212029011
Net:
smc911x-0
U-Boot #
I
The U-Boot shell offers a set of commands. We will study the

most important ones, see the documentation for a complete
reference or the help command.
108/540
Information commands
Flash information (NOR and SPI flash)
U-Boot> flinfo
DataFlash:AT45DB021
Nb pages: 1024
Page Size: 264
Size= 270336 bytes
Logical address: 0xC0000000
Area 0: C0000000 to C0001FFF (RO) Bootstrap
Area 1: C0002000 to C0003FFF Environment
Area 2: C0004000 to C0041FFF (RO) U-Boot
NAND flash information
U-Boot> nand info
Device 0: nand0, sector size 128 KiB
Page size
2048 b
OOB size
64 b
Erase size
131072 b
Version details
U-Boot> version
U-Boot 2013.04 (May 29 2013 - 10:30:21)
109/540
Important commands (1)

I
The exact set of commands depends on the U-Boot

configuration
help and help command
boot, runs the default boot command, stored in bootcmd
bootm <address>, starts a kernel image loaded at the given

address in RAM
ext2load, loads a file from an ext2 filesystem to RAM
And also ext2ls to list files, ext2info for information
fatload, loads a file from a FAT filesystem to RAM

I
And also fatls and fatinfo
tftp, loads a file from the network to RAM
ping, to test the network
110/540
Important commands (2)
loadb, loads, loady, load a file from the serial line to RAM
usb, to initialize and control the USB subsystem, mainly used

for USB storage devices such as USB keys
mmc, to initialize and control the MMC subsystem, used for

SD and microSD cards
nand, to erase, read and write contents to NAND flash
erase, protect, cp, to erase, modify protection and write to

NOR flash
md, displays memory contents. Can be useful to check the

contents loaded in memory, or to look at hardware registers.
mm, modifies memory contents. Can be useful to modify

directly hardware registers, for testing purposes.
111/540
Environment variables commands (1)
U-Boot can be configured through environment variables,

which affect the behavior of the different commands.
Environment variables are loaded from flash to RAM at

U-Boot startup, can be modified and saved back to flash for
persistence
There is a dedicated location in flash (or in MMC storage) to

store the U-Boot environment, defined in the board
configuration file
112/540
Environment variables commands (2)
Commands to manipulate environment variables:

I
printenv
Shows all variables
printenv <variable-name>
Shows the value of a variable
setenv <variable-name> <variable-value>

Changes the value of a variable, only in RAM
editenv <variable-name>
Edits the value of a variable, only in RAM
saveenv
Saves the current state of the environment to flash
113/540
Environment variables commands - Example
u-boot # printenv
baudrate=19200
ethaddr=00:40:95:36:35:33
netmask=255.255.255.0
ipaddr=10.0.0.11
serverip=10.0.0.1
stdin=serial
stdout=serial
stderr=serial
u-boot # printenv serverip
serverip=10.0.0.1
u-boot # setenv serverip 10.0.0.100
u-boot # saveenv
114/540
Important U-Boot env variables

I
bootcmd, contains the command that U-Boot will

automatically execute at boot time after a configurable delay,
if the process is not interrupted
bootargs, contains the arguments passed to the Linux kernel,

covered later
serverip, the IP address of the server that U-Boot will

contact for network related commands
ipaddr, the IP address that U-Boot will use
netmask, the network mask to contact the server
ethaddr, the MAC address, can only be set once
bootdelay, the delay in seconds before which U-Boot runs

bootcmd
autostart, if yes, U-Boot starts automatically an image that

has been loaded into memory
115/540
Scripts in environment variables
Environment variables can contain small scripts, to execute

several commands and test the results of commands.
I
I
I
I
I
Useful to automate booting or upgrade processes

Several commands can be chained using the ; operator
Tests can be done using
if command ; then ... ; else ... ; fi
Scripts are executed using run <variable-name>
You can reference other variables using ${variable-name}
Example
I
setenv mmc-boot if fatload mmc 0 80000000 boot.

ini; then source; else if fatload mmc 0 80000000
uImage; then run mmc-bootargs; bootm; fi; fi
116/540
Transferring files to the target
U-Boot is mostly used to load and boot a kernel image, but it

also allows to change the kernel image and the root filesystem
stored in flash.
Files must be exchanged between the target and the
development workstation. This is possible:
I
Through the network if the target has an Ethernet connection,

and U-Boot contains a driver for the Ethernet chip. This is the
fastest and most efficient solution.
Through a USB key, if U-Boot supports the USB controller of
your platform
Through a SD or microSD card, if U-Boot supports the MMC
controller of your platform
Through the serial port
117/540
TFTP
Network transfer from the development workstation to U-Boot

on the target takes place through TFTP
I
I
A TFTP server is needed on the development workstation

I
I
I
Trivial File Transfer Protocol

Somewhat similar to FTP, but without authentication and over
UDP
sudo apt-get install tftpd-hpa
All files in /var/lib/tftpboot are then visible through TFTP
A TFTP client is available in the tftp-hpa package, for
testing
A TFTP client is integrated into U-Boot

I
I
Configure the ipaddr and serverip environment variables

Use tftp <address> <filename> to load a file
118/540
U-boot mkimage
I
The kernel image that U-Boot loads and boots must be

prepared, so that a U-Boot specific header is added in front of
the image
I
This header gives details such as the image size, the expected
load address, the compression type, etc.
This is done with a tool that comes in U-Boot, mkimage
Debian / Ubuntu: just install the u-boot-tools package.
Or, compile it by yourself: simply configure U-Boot for any

board of any architecture and compile it. Then install
mkimage:
cp tools/mkimage /usr/local/bin/
The special target uImage of the kernel Makefile can then be

used to generate a kernel image suitable for U-Boot.
119/540
Practical lab - U-Boot
Time to start the practical lab!

I
Communicate with the board using

a serial console
Configure, build and install U-Boot
Learn U-Boot commands
Set up TFTP communication with

the board
120/540
Linux kernel introduction
Linux kernel
introduction
Free Electrons
Embedded Linux
Developers
Free Electrons

121/540
Linux features
122/540
History
The Linux kernel is one component of a system, which also

requires libraries and applications to provide features to end
users.
The Linux kernel was created as a hobby in 1991 by a Finnish
student, Linus Torvalds.
I
Linux quickly started to be used as the kernel for free software

operating systems
Linus Torvalds has been able to create a large and dynamic

developer and user community around Linux.
Nowadays, more than one thousand people contribute to each

kernel release, individuals or companies big and small.
123/540
Linux kernel key features
Portability and hardware

support. Runs on most
architectures.
Security. It cant hide its

flaws. Its code is reviewed
by many experts.
Scalability. Can run on

super computers as well as
on tiny devices (4 MB of
RAM is enough).
Stability and reliability.
Modularity. Can include

only what a system needs
even at run time.
Easy to program. You can

learn from existing code.
Many useful resources on
the net.
Compliance to standards
and interoperability.
Exhaustive networking
support.
124/540
Linux kernel in the system
125/540
Linux kernel main roles
Manage all the hardware resources: CPU, memory, I/O.
Provide a set of portable, architecture and hardware

independent APIs to allow user space applications and
libraries to use the hardware resources.
Handle concurrent accesses and usage of hardware
resources from different applications.
Example: a single network interface is used by multiple user

space applications through various network connections. The
kernel is responsible to multiplex the hardware resource.
126/540
System calls
The main interface between the kernel and user space is the
set of system calls
About 300 system calls that provide the main kernel services
I
File and device operations, networking operations,

inter-process communication, process management, memory
mapping, timers, threads, synchronization primitives, etc.
This interface is stable over time: only new system calls can
be added by the kernel developers
This system call interface is wrapped by the C library, and

user space applications usually never make a system call
directly but rather use the corresponding C library function
127/540
Pseudo filesystems
Linux makes system and kernel information available in user

space through pseudo filesystems, sometimes also called
virtual filesystems
Pseudo filesystems allow applications to see directories and

files that do not exist on any real storage: they are created
and updated on the fly by the kernel
The two most important pseudo filesystems are
proc, usually mounted on /proc:

Operating system related information (processes, memory
management parameters...)
sysfs, usually mounted on /sys:
Representation of the system as a set of devices and buses.
Information about these devices.
128/540
Inside the Linux kernel
129/540
Linux license
The whole Linux sources are Free Software released under the
GNU General Public License version 2 (GPL v2).
For the Linux kernel, this basically implies that:
I
When you receive or buy a device with Linux on it, you should
receive the Linux sources, with the right to study, modify and
redistribute them.
When you produce Linux based devices, you must release the
sources to the recipient, with the same rights, with no
restriction.
130/540
Supported hardware architectures
See the arch/ directory in the kernel sources
Minimum: 32 bit processors, with or without MMU, and gcc

support
32 bit architectures (arch/ subdirectories)

Examples: arm, avr32, blackfin, c6x, m68k, microblaze,
mips, score, sparc, um
64 bit architectures:
Examples: alpha, arm64, ia64, tile
32/64 bit architectures

Examples: powerpc, x86, sh, sparc
Find details in kernel sources: arch/<arch>/Kconfig,

arch/<arch>/README, or Documentation/<arch>/
131/540
Linux versioning scheme and

development process
132/540
Until 2.6 (1)
One stable major branch every 2 or 3 years

I
I
One development branch to integrate new functionalities and

major changes
I
I
I
Identified by an even middle number

Examples: 1.0.x, 2.0.x, 2.2.x, 2.4.x
Identified by an odd middle number

Examples: 2.1.x, 2.3.x, 2.5.x
After some time, a development version becomes the new base
version for the stable branch
Minor releases once in while: 2.2.23, 2.5.12, etc.
133/540
Until 2.6 (2)
134/540
Changes since Linux 2.6
Since 2.6.0, kernel developers have been able to introduce

lots of new features one by one on a steady pace, without
having to make disruptive changes to existing subsystems.
Since then, there has been no need to create a new

development branch massively breaking compatibility with the
stable branch.
Thanks to this, more features are released to users at a

faster pace.
135/540
3.x stable branch
From 2003 to 2011, the official kernel versions were named

2.6.x.
Linux 3.0 was released in July 2011

This is only a change to the numbering scheme
I
I
Official kernel versions are now named 3.x (3.0, 3.1, 3.2,
etc.)
Stabilized versions are named 3.x.y (3.0.2, 3.4.3, etc.)
It effectively only removes a digit compared to the previous
numbering scheme
136/540
New development model
Using merge and bug fixing windows
137/540
New development model - Details
After the release of a 3.x version (for example), a two-weeks

merge window opens, during which major additions are
merged.
The merge window is closed by the release of test version

3.(x+1)-rc1
The bug fixing period opens, for 6 to 10 weeks.
At regular intervals during the bug fixing period,

3.(x+1)-rcY test versions are released.
When considered sufficiently stable, kernel 3.(x+1) is

released, and the process starts again.
138/540
More stability for the kernel source tree

I
Issue: bug and security fixes only released

for most recent stable kernel versions.
Some people need to have a recent kernel,

but with long term support for security
updates.
You could get long term support from a

commercial embedded Linux provider.
You could reuse sources for the kernel

used in Ubuntu Long Term Support
releases (5 years of free security updates).
The http://kernel.org front page

shows which versions will be supported for
some time (up to 2 or 3 years), and which
ones wont be supported any more
(EOL: End Of Life)
139/540
Whats new in each Linux release?

I
The official list of changes for each Linux release is just a

huge list of individual patches!
commit aa6e52a35d388e730f4df0ec2ec48294590cc459
Author: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Date:
Wed Jul 13 11:29:17 2011 +0200
at91: at91-ohci: support overcurrent notification
Several USB power switches (AIC1526 or MIC2026) have a digital output
that is used to notify that an overcurrent situation is taking
place. This digital outputs are typically connected to GPIO inputs of
the processor and can be used to be notified of these overcurrent
situations.
Therefore, we add a new overcurrent_pin[] array in the at91_usbh_data
structure so that boards can tell the AT91 OHCI driver which pins are
used for the overcurrent notification, and an overcurrent_supported
boolean to tell the driver whether overcurrent is supported or not.
The code has been largely borrowed from ohci-da8xx.c and
ohci-s3c2410.c.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Nicolas Ferre <nicolas.ferre@atmel.com>
Very difficult to find out the key changes and to get the global
picture out of individual changes.
Fortunately, there are some useful resources available

I
I
I
http://wiki.kernelnewbies.org/LinuxChanges
http://lwn.net
http://linuxfr.org, for French readers
140/540
Linux kernel sources
141/540
Location of kernel sources

I
The official versions of the Linux kernel, as released by Linus

Torvalds, are available at http://www.kernel.org
I
I
Many chip vendors supply their own kernel sources

I
I
I
These versions follow the development model of the kernel

However, they may not contain the latest development from a
specific area yet. Some features in development might not be
ready for mainline inclusion yet.
Focusing on hardware support first
Can have a very important delta with mainline Linux
Useful only when mainline hasnt caught up yet.
Many kernel sub-communities maintain their own kernel, with

usually newer but less stable features
I
Architecture communities (ARM, MIPS, PowerPC, etc.),

device drivers communities (I2C, SPI, USB, PCI, network,
etc.), other communities (real-time, etc.)
No official releases, only development trees are available.
142/540
Getting Linux sources
The kernel sources are available from

http://kernel.org/pub/linux/kernel as full tarballs
(complete kernel sources) and patches (differences between
two kernel versions).
However, more and more people use the git version control
system. Absolutely needed for kernel development!
I
Fetch the entire kernel sources and history

git clone git://git.kernel.org/pub/scm/linux/
kernel/git/torvalds/linux.git
Create a branch that starts at a specific stable version
git checkout -b <name-of-branch> v3.11
Web interface available at http://git.kernel.org/cgit/
linux/kernel/git/torvalds/linux.git/tree/.
Read more about Git at http://git-scm.com/
143/540
Linux kernel size (1)

I
Linux 3.10 sources:

Raw size: 573 MB (43,000 files, approx 15,800,000 lines)
gzip compressed tar archive: 105 MB
bzip2 compressed tar archive: 83 MB (better)
xz compressed tar archive: 69 MB (best)
Minimum Linux 3.17 compiled kernel size, booting on the

ARM Versatile board (hard drive on PCI, ext2 filesystem, ELF
executable support, framebuffer console and input devices):
876 KB (compressed), 2.3 MB (raw)
Why are these sources so big?

Because they include thousands of device drivers, many
network protocols, support many architectures and
filesystems...
The Linux core (scheduler, memory management...) is pretty

small!
144/540
Linux kernel size (2)
As of kernel version 3.10.

I
drivers/: 49.4%
arch/: 21.9%
fs/: 6.0%
include/: 4.7%
sound/: 4.4%
Documentation/: 4.0%
net/: 3.9%
firmware/: 1.0%
kernel/: 1.0%
tools/: 0.9%
scripts/: 0.5%
mm/: 0.5%
crypto/: 0.4%
security/: 0.4%
lib/: 0.4%
block/: 0.2%
...
145/540
Getting Linux sources

I
Full tarballs
I
Contain the complete kernel sources: long to download and

uncompress, but must be done at least once
Example:
http://www.kernel.org/pub/linux/kernel/v3.x/linux3.10.9.tar.xz
Extract command:
tar xf linux-3.10.9.tar.xz
Incremental patches between versions

I
It assumes you already have a base version and you apply the
correct patches in the right order. Quick to download and
apply
Examples:
http://www.kernel.org/pub/linux/kernel/v3.x/patch-3.10.xz
(3.9 to 3.10)
http://www.kernel.org/pub/linux/kernel/v3.x/patch-3.10.9.xz
(3.10 to 3.10.9)
All previous kernel versions are available in

http://kernel.org/pub/linux/kernel/
146/540
Patch
I
A patch is the difference between two source trees

I
I
I
Computed with the diff tool, or with more elaborate version

control systems
They are very common in the open-source community

Excerpt from a patch:
diff -Nru a/Makefile b/Makefile

--- a/Makefile 2005-03-04 09:27:15 -08:00
+++ b/Makefile 2005-03-04 09:27:15 -08:00
@@ -1,7 +1,7 @@
VERSION = 2
PATCHLEVEL = 6
SUBLEVEL = 11
-EXTRAVERSION =
+EXTRAVERSION = .1
NAME=Woozy Numbat
# *DOCUMENTATION*
147/540
Contents of a patch
I
One section per modified file, starting with a header

diff -Nru a/Makefile b/Makefile
--- a/Makefile 2005-03-04 09:27:15 -08:00
+++ b/Makefile 2005-03-04 09:27:15 -08:00
One sub-section per modified part of the file, starting with

header with the affected line numbers
@@ -1,7 +1,7 @@
Three lines of context before the change

VERSION = 2
PATCHLEVEL = 6
SUBLEVEL = 11
The change itself

-EXTRAVERSION =
+EXTRAVERSION = .1
Three lines of context after the change

NAME=Woozy Numbat
# *DOCUMENTATION*
148/540
Using the patch command

The patch command:
I
Takes the patch contents on its standard input
Applies the modifications described by the patch into the

current directory
patch usage examples:

I
patch -p<n> < diff_file
cat diff_file | patch -p<n>
xzcat diff_file.xz | patch -p<n>
bzcat diff_file.bz2 | patch -p<n>
zcat diff_file.gz | patch -p<n>

Notes:
I
I
I
n: number of directory levels to skip in the file paths

You can reverse apply a patch with the -R option
You can test a patch with --dry-run option
149/540
Applying a Linux patch

I
Two types of Linux patches:

I
Either to be applied to the previous stable version

(from 3.<x-1> to 3.x)
Or implementing fixes to the current stable version
(from 3.x to 3.x.y)
Can be downloaded in gzip, bzip2 or xz (much smaller)

compressed files.
Always produced for n=1
(thats what everybody does... do it too!)
Need to run the patch command inside the kernel source
directory
Linux patch command line example:
cd linux-3.9
xzcat ../patch-3.10.xz | patch -p1
xzcat ../patch-3.10.9.xz | patch -p1
cd ..; mv linux-3.9 linux-3.10.9
150/540
Practical lab - Kernel sources
Time to start the practical lab!

I
Get the Linux kernel sources
Apply patches
151/540
Kernel configuration
152/540
Kernel configuration and build system
The kernel configuration and build system is based on

multiple Makefiles
One only interacts with the main Makefile, present at the

top directory of the kernel source tree
Interaction takes place
I
I
using the make tool, which parses the Makefile

through various targets, defining which action should be done
(configuration, compilation, installation, etc.). Run
make help to see all available targets.
Example
I
I
cd linux-3.6.x/
make <target>
153/540
Kernel configuration (1)
The kernel contains thousands of device drivers, filesystem

drivers, network protocols and other configurable items
Thousands of options are available, that are used to

selectively compile parts of the kernel source code
The kernel configuration is the process of defining the set of

options with which you want your kernel to be compiled
The set of options depends
I
I
On your hardware (for device drivers, etc.)

On the capabilities you would like to give to your kernel
(network capabilities, filesystems, real-time, etc.)
154/540
Kernel configuration (2)
The configuration is stored in the .config file at the root of

kernel sources
I
I
I
I
Simple text file, key=value style
As options have dependencies, typically never edited by hand,

but through graphical or text interfaces:
make xconfig, make gconfig (graphical)
make menuconfig, make nconfig (text)
You can switch from one to another, they all load/save the
same .config file, and show the same set of options
To modify a kernel in a GNU/Linux distribution: the

configuration files are usually released in /boot/, together
with kernel images: /boot/config-3.2.0-31-generic
155/540
Kernel or module?
I
The kernel image is a single file, resulting from the linking

of all object files that correspond to features enabled in the
configuration
I
I
This is the file that gets loaded in memory by the bootloader

All included features are therefore available as soon as the
kernel starts, at a time where no filesystem exists
Some features (device drivers, filesystems, etc.) can however

be compiled as modules
I
These are plugins that can be loaded/unloaded dynamically to

add/remove features to the kernel
Each module is stored as a separate file in the filesystem,
and therefore access to a filesystem is mandatory to use
modules
This is not possible in the early boot procedure of the kernel,
because no filesystem is available
156/540
Kernel option types
There are different types of options

I
bool options, they are either

I
I
tristate options, they are either

I
I
I
I
I
I
true (to include the feature in the kernel) or

false (to exclude the feature from the kernel)
true (to include the feature in the kernel image) or
module (to include the feature as a kernel module) or
false (to exclude the feature)
int options, to specify integer values

hex options, to specify hexadecimal values
string options, to specify string values
157/540
Kernel option dependencies
There are dependencies between kernel options
For example, enabling a network driver requires the network

stack to be enabled
Two types of dependencies
depends on dependencies. In this case, option A that depends

on option B is not visible until option B is enabled
select dependencies. In this case, with option A depending
on option B, when option A is enabled, option B is
automatically enabled
make xconfig allows to see all options, even the ones that
cannot be selected because of missing dependencies. In this
case, they are displayed in gray
158/540
make xconfig
make xconfig
I
The most common graphical interface to configure the kernel.
Make sure you read

help -> introduction: useful options!
File browser: easier to load configuration files
Search interface to look for parameters
Required Debian / Ubuntu packages: libqt4-dev g++

(libqt3-mt-dev for older kernel releases)
159/540
make xconfig screenshot
160/540
make xconfig search interface

Looks for a keyword in the parameter name. Allows to select or
unselect found parameters.
161/540
Kernel configuration options
162/540
Corresponding .config file excerpt

Options are grouped by sections and are prefixed with CONFIG_.
#
# CD-ROM/DVD Filesystems
#
CONFIG_ISO9660_FS=m
CONFIG_JOLIET=y
CONFIG_ZISOFS=y
CONFIG_UDF_FS=y
CONFIG_UDF_NLS=y
#
# DOS/FAT/NT Filesystems
#
# CONFIG_MSDOS_FS is not set
# CONFIG_VFAT_FS is not set
CONFIG_NTFS_FS=m
# CONFIG_NTFS_DEBUG is not set
CONFIG_NTFS_RW=y
163/540
make gconfig
make gconfig
I
GTK based graphical

configuration interface.
Functionality similar to that
of make xconfig.
Just lacking a search

functionality.
Required Debian packages:

libglade2-dev
164/540
make menuconfig
make menuconfig
I
Useful when no graphics are

available. Pretty convenient
too!
Same interface found in

other tools: BusyBox,
Buildroot...

libncurses-dev
165/540
make nconfig
make nconfig
I
A newer, similar text

interface
More user friendly (for

example, easier to access
help information).

libncurses-dev
166/540
make oldconfig
make oldconfig
I
Needed very often!
Useful to upgrade a .config file from an earlier kernel release
Issues warnings for configuration parameters that no longer

exist in the new kernel.
Asks for values for new parameters (while xconfig and

menuconfig silently set default values for new parameters).
If you edit a .config file by hand, its strongly recommended to

run make oldconfig afterwards!
167/540
Undoing configuration changes
A frequent problem:
I
After changing several kernel configuration settings, your

kernel no longer works.
If you dont remember all the changes you made, you can get
back to your previous configuration:
$ cp .config.old .config
All the configuration interfaces of the kernel (xconfig,

menuconfig, oldconfig...) keep this .config.old backup
copy.
168/540
Configuration per architecture
The set of configuration options is architecture dependent

I
I
Some configuration options are very architecture-specific

Most of the configuration options (global kernel options,
network subsystem, filesystems, most of the device drivers) are
visible in all architectures.
By default, the kernel build system assumes that the kernel is

being built for the host architecture, i.e. native compilation
The architecture is not defined inside the configuration, but at

a higher level
We will see later how to override this behaviour, to allow the

configuration of kernels for a different architecture
169/540
Compiling and installing the kernel

for the host system
170/540
Kernel compilation
I
make
I
I
in the main kernel source directory

Remember to run multiple jobs in parallel if you have multiple
CPU cores. Example: make -j 4
No need to run as root!
Generates
I
vmlinux, the raw uncompressed kernel image, in the ELF

format, useful for debugging purposes, but cannot be booted
arch/<arch>/boot/*Image, the final, usually compressed,
kernel image that can be booted
I
bzImage for x86, zImage for ARM, vmImage.gz for Blackfin,

etc.
arch/<arch>/boot/dts/*.dtb, compiled Device Tree files

(on some architectures)
All kernel modules, spread over the kernel source tree, as .ko
files.
171/540
Kernel installation
I
make install
I
Installs
I
Does the installation for the host system by default, so needs

to be run as root. Generally not used when compiling for an
embedded system, as it installs files on the development
workstation.
/boot/vmlinuz-<version>
Compressed kernel image. Same as the one in
arch/<arch>/boot
/boot/System.map-<version>
Stores kernel symbol addresses
/boot/config-<version>
Kernel configuration for this version
Typically re-runs the bootloader configuration utility to take

the new kernel into account.
172/540
Module installation
make modules_install
I
Does the installation for the host system by default, so needs

to be run as root
Installs all modules in /lib/modules/<version>/

I
kernel/
Module .ko (Kernel Object) files, in the same directory
structure as in the sources.
modules.alias
Module aliases for module loading utilities. Example line:
alias sound-service-?-0 snd_mixer_oss
modules.dep
Module dependencies
modules.symbols
Tells which module a given symbol belongs to.
173/540
Kernel cleanup targets
Clean-up generated files (to force

re-compilation):
make clean
Remove all generated files. Needed when

switching from one architecture to another.
Caution: it also removes your .config file!
make mrproper
Also remove editor backup and patch reject files

(mainly to generate patches):
make distclean
174/540
Cross-compiling the kernel
175/540
Cross-compiling the kernel
When you compile a Linux kernel for another CPU architecture

I
Much faster than compiling natively, when the target system

is much slower than your GNU/Linux workstation.
Much easier as development tools for your GNU/Linux

workstation are much easier to find.
To make the difference with a native compiler, cross-compiler

executables are prefixed by the name of the target system,
architecture and sometimes library. Examples:
mips-linux-gcc, the prefix is mips-linuxarm-linux-gnueabi-gcc, the prefix is arm-linux-gnueabi-
176/540
Specifying cross-compilation (1)
The CPU architecture and cross-compiler prefix are defined through

the ARCH and CROSS_COMPILE variables in the toplevel Makefile.
I
ARCH is the name of the architecture. It is defined by the

name of the subdirectory in arch/ in the kernel sources
I
Example: arm if you want to compile a kernel for the arm

architecture.
CROSS_COMPILE is the prefix of the cross compilation tools

I
Example: arm-linux- if your compiler is arm-linux-gcc
177/540
Specifying cross-compilation (2)

Two solutions to define ARCH and CROSS_COMPILE:
I
Pass ARCH and CROSS_COMPILE on the make command line:

make ARCH=arm CROSS_COMPILE=arm-linux- ...
Drawback: it is easy to forget to pass these variables when
you run any make command, causing your build and
configuration to be screwed up.
Define ARCH and CROSS_COMPILE as environment variables:

export ARCH=arm
export CROSS_COMPILE=arm-linuxDrawback: it only works inside the current shell or terminal.
You could put these settings in a file that you source every
time you start working on the project. If you only work on a
single architecture with always the same toolchain, you could
even put these settings in your ~/.bashrc file to make them
permanent and visible from any terminal.
178/540
Predefined configuration files
Default configuration files available, per board or per-CPU

family
I
They are stored in arch/<arch>/configs/, and are just

minimal .config files
This is the most common way of configuring a kernel for
embedded platforms
Run make help to find if one is available for your platform
To load a default configuration file, just run

make acme_defconfig
I
This will overwrite your existing .config file!
To create your own default configuration file

I
I
make savedefconfig, to create a minimal configuration file

mv defconfig arch/<arch>/configs/myown_defconfig
179/540
Configuring the kernel
After loading a default configuration file, you can adjust the

configuration to your needs with the normal xconfig,
gconfig or menuconfig interfaces
As the architecture is different from your host architecture
I
Some options will be different from the native configuration

(processor and architecture specific options, specific drivers,
etc.)
Many options will be identical (filesystems, network protocols,
architecture-independent drivers, etc.)
180/540
Device Tree
I
Many embedded architectures have a lot of non-discoverable

hardware.
Depending on the architecture, such hardware is either

described using C code directly within the kernel, or using a
special hardware description language in a Device Tree.
ARM, PowerPC, OpenRISC, ARC, Microblaze are examples of

architectures using the Device Tree.
A Device Tree Source, written by kernel developers, is
compiled into a binary Device Tree Blob, passed at boot time
to the kernel.
There is one different Device Tree for each board/platform

supported by the kernel, available in
arch/arm/boot/dts/<board>.dtb.
The bootloader must load both the kernel image and the
Device Tree Blob in memory before starting the kernel.
181/540
Building and installing the kernel

I
I
Run make
Copy the final kernel image to the target storage
I
make install is rarely used in embedded development, as

the kernel image is a single file, easy to handle
I
can be uImage, zImage, vmlinux, bzImage in

arch/<arch>/boot
copying the Device Tree Blob might be necessary as well, they
are available in arch/<arch>/boot/dts
It is however possible to customize the make install behaviour

in arch/<arch>/boot/install.sh
make modules_install is used even in embedded

development, as it installs many modules and description files
I
I
make INSTALL_MOD_PATH=<dir>/ modules_install

The INSTALL_MOD_PATH variable is needed to install the
modules in the target root filesystem instead of your host root
filesystem.
182/540
Booting with U-Boot

I
I
Recent versions of U-Boot can boot the zImage binary.

Older versions require a special kernel image format: uImage
I
uImage is generated from zImage using the mkimage tool. It

is done automatically by the kernel make uImage target.
On some ARM platforms, make uImage requires passing a
LOADADDR environment variable, which indicates at which
physical memory address the kernel will be executed.
In addition to the kernel image, U-Boot can also pass a

Device Tree Blob to the kernel.
The typical boot process is therefore:
1. Load zImage or uImage at address X in memory
2. Load <board>.dtb at address Y in memory
3. Start the kernel with bootz X - Y or bootm X - Y
The - in the middle indicates no initramfs
183/540
Kernel command line

I
In addition to the compile time configuration, the kernel

behaviour can be adjusted with no recompilation using the
kernel command line
The kernel command line is a string that defines various
arguments to the kernel
I
I
I
I
It is very important for system configuration

root= for the root filesystem (covered later)
console= for the destination of kernel messages
Many more exist. The most important ones are documented in
Documentation/kernel-parameters.txt in kernel sources.
This kernel command line is either

I
Passed by the bootloader. In U-Boot, the contents of the

bootargs environment variable is automatically passed to the
kernel
Built into the kernel, using the CONFIG_CMDLINE option.
184/540
Practical lab - Kernel cross-compiling
Set up the cross-compiling

environment
Configure and cross-compile the

kernel for an arm platform
On this platform, interact with the

bootloader and boot your kernel
185/540
Using kernel modules
186/540
Advantages of modules
Modules make it easy to develop drivers without rebooting:

load, test, unload, rebuild, load...
Useful to keep the kernel image size to the minimum

(essential in GNU/Linux distributions for PCs).
Also useful to reduce boot time: you dont spend time

initializing devices and kernel features that you only need later.
Caution: once loaded, have full control and privileges in the

system. No particular protection. Thats why only the root
user can load and unload modules.
187/540
Module dependencies
Some kernel modules can depend on other modules, which

need to be loaded first.
Example: the usb-storage module depends on the

scsi_mod, libusual and usbcore modules.
Dependencies are described in

/lib/modules/<kernel-version>/modules.dep
This file is generated when you run make modules_install.
188/540
Kernel log
When a new module is loaded, related information is available in
the kernel log.
I
The kernel keeps its messages in a circular buffer (so that it

doesnt consume more memory with many messages)
Kernel log messages are available through the dmesg

command (diagnostic message)
Kernel log messages are also displayed in the system console

(console messages can be filtered by level using the loglevel
kernel parameter, or completely disabled with the quiet
parameter).
Note that you can write to the kernel log from user space too:
echo "<n>Debug info" > /dev/kmsg
189/540
Module utilities (1)
modinfo <module_name>
modinfo <module_path>.ko
Gets information about a module: parameters, license,
description and dependencies.
Very useful before deciding to load a module or not.
sudo insmod <module_path>.ko

Tries to load the given module. The full path to the module
object file must be given.
190/540
Understanding module loading issues
When loading a module fails, insmod often doesnt give you

enough details!
Details are often available in the kernel log.
Example:
$ sudo insmod ./intr_monitor.ko
insmod: error inserting './intr_monitor.ko': -1 Device or resource busy
$ dmesg
[17549774.552000] Failed to register handler for irq channel 2
191/540
sudo modprobe <module_name>

Most common usage of modprobe: tries to load all the
modules the given module depends on, and then this module.
Lots of other options are available. modprobe automatically
looks in /lib/modules/<version>/ for the object file
corresponding to the given module name.
lsmod
Displays the list of loaded modules
Compare its output with the contents of /proc/modules!
192/540
sudo rmmod <module_name>

Tries to remove the given module.
Will only be allowed if the module is no longer in use (for
example, no more processes opening a device file)
sudo modprobe -r <module_name>

Tries to remove the given module and all dependent modules
(which are no longer needed after removing the module)
193/540
Passing parameters to modules

I
Find available parameters:

modinfo snd-intel8x0m
Through insmod:
sudo insmod ./snd-intel8x0m.ko index=-2
Through modprobe:
Set parameters in /etc/modprobe.conf or in any file in
/etc/modprobe.d/:
options snd-intel8x0m index=-2
Through the kernel command line, when the driver is built

statically into the kernel:
snd-intel8x0m.index=-2
I
I
I
snd-intel8x0m is the driver name

index is the driver parameter name
-2 is the driver parameter value
194/540
Check module parameter values
How to find the current values for the parameters of a loaded

module?
I
Check /sys/module/<name>/parameters.
There is one file per parameter, containing the parameter

value.
195/540
Useful reading
Linux Kernel in a Nutshell, Dec 2006
I
By Greg Kroah-Hartman, OReilly

http://www.kroah.com/lkn/
A good reference book and guide on

configuring, compiling and managing the
Linux kernel sources.
Freely available on-line!

Great companion to the printed book for
easy electronic searches!
Available as single PDF file on
community/kernel/lkn/
Our rating: 2 stars
196/540
Linux Root Filesystem
Linux Root
Filesystem
Free Electrons
Embedded Linux
Developers
Free Electrons

197/540
Principle and solutions
198/540
Filesystems
I
Filesystems are used to organize data in directories and files

on storage devices or on the network. The directories and files
are organized as a hierarchy
In Unix systems, applications and users see a single global

hierarchy of files and directories, which can be composed of
several filesystems.
Filesystems are mounted in a specific location in this
hierarchy of directories
When a filesystem is mounted in a directory (called mount

point), the contents of this directory reflects the contents of
the storage device
When the filesystem is unmounted, the mount point is empty
again.
This allows applications to access files and directories easily,

regardless of their exact storage location
199/540
Filesystems (2)
Create a mount point, which is just a directory

$ mkdir /mnt/usbkey
It is empty
$ ls /mnt/usbkey
$
Mount a storage device in this mount point

$ mount -t vfat /dev/sda1 /mnt/usbkey
$
You can access the contents of the USB key

$ ls /mnt/usbkey
docs prog.c picture.png movie.avi
$
200/540
mount / umount
mount allows to mount filesystems

I
I
I
I
mount -t type device mountpoint

type is the type of filesystem
device is the storage device, or network location to mount
mountpoint is the directory where files of the storage device
or network location will be accessible
mount with no arguments shows the currently mounted
filesystems
umount allows to unmount filesystems

I
This is needed before rebooting, or before unplugging a USB

key, because the Linux kernel caches writes in memory to
increase performance. umount makes sure that these writes are
committed to the storage.
201/540
Root filesystem
I
A particular filesystem is mounted at the root of the hierarchy,

identified by /
This filesystem is called the root filesystem

As mount and umount are programs, they are files inside a
filesystem.
They are not accessible before mounting at least one

filesystem.
As the root filesystem is the first mounted filesystem, it

cannot be mounted with the normal mount command
It is mounted directly by the kernel, according to the root=

kernel option
When no root filesystem is available, the kernel panics

Please append a correct "root=" boot option
Kernel panic - not syncing: VFS: Unable to mount root fs on unknown block(0,0)
202/540
Location of the root filesystem
It can be mounted from different locations

I
I
I
I
I
I
From the partition of a hard disk

From the partition of a USB key
From the partition of an SD card
From the partition of a NAND flash chip or similar type of
storage device
From the network, using the NFS protocol
From memory, using a pre-loaded filesystem (by the
bootloader)
etc.
It is up to the system designer to choose the configuration for

the system, and configure the kernel behaviour with root=
203/540
Mounting rootfs from storage devices
Partitions of a hard disk or USB key

I
Partitions of an SD card
I
root=/dev/sdXY, where X is a letter indicating the device,

and Y a number indicating the partition
/dev/sdb2 is the second partition of the second disk drive
(either USB key or ATA hard drive)
root=/dev/mmcblkXpY, where X is a number indicating the
device and Y a number indicating the partition
/dev/mmcblk0p2 is the second partition of the first device
Partitions of flash storage

I
I
root=/dev/mtdblockX, where X is the partition number

/dev/mtdblock3 is the fourth partition of a NAND flash chip
(if only one NAND flash chip is present)
204/540
Mounting rootfs over the network (1)

Once networking works, your root filesystem could be a directory
on your GNU/Linux development host, exported by NFS (Network
File System). This is very convenient for system development:
I
Makes it very easy to update files on the root filesystem,

without rebooting. Much faster than through the serial port.
Can have a big root filesystem even if you dont have support
for internal or external storage yet.
The root filesystem can be huge. You can even build native
compiler tools and build all the tools you need on the target
itself (better to cross-compile though).
205/540
On the development workstation side, a NFS server is needed

I
Install an NFS server (example: Debian, Ubuntu)

sudo apt-get install nfs-kernel-server
Add the exported directory to your /etc/exports file:

/home/tux/rootfs 192.168.1.111(rw,no_root_squash,
no_subtree_check)
I
I
192.168.1.111 is the client IP address

rw,no_root_squash,no_subtree_check are the NFS server
options for this directory export.
Start or restart your NFS server (example: Debian, Ubuntu)

sudo /etc/init.d/nfs-kernel-server restart
206/540
I
I
On the target system

The kernel must be compiled with
I
I
I
CONFIG_NFS_FS=y (NFS support)

CONFIG_IP_PNP=y (configure IP at boot time)
CONFIG_ROOT_NFS=y (support for NFS as rootfs)
The kernel must be booted with the following parameters:

I
I
I
root=/dev/nfs (we want rootfs over NFS)

ip=192.168.1.111 (target IP address)
nfsroot=192.168.1.110:/home/tux/rootfs/ (NFS server
details)
207/540
208/540
rootfs in memory: initramfs (1)

I
I
I
It is also possible to have the root filesystem integrated into

the kernel image
It is therefore loaded into memory together with the kernel
This mechanism is called initramfs
I
It integrates a compressed archive of the filesystem into the

kernel image
Variant: the compressed archive can also be loaded separately
by the bootloader.
It is useful for two cases

I
Fast booting of very small root filesystems. As the filesystem is

completely loaded at boot time, application startup is very fast.
As an intermediate step before switching to a real root
filesystem, located on devices for which drivers not part of the
kernel image are needed (storage drivers, filesystem drivers,
network drivers). This is always used on the kernel of
desktop/server distributions to keep the kernel image size
reasonable.
209/540
210/540

I
The contents of an initramfs are defined at the kernel

configuration level, with the CONFIG_INITRAMFS_SOURCE
option
I
I
I
Can be the path to a directory containing the root filesystem

contents
Can be the path to a cpio archive
Can be a text file describing the contents of the initramfs
(see documentation for details)
The kernel build process will automatically take the contents

of the CONFIG_INITRAMFS_SOURCE option and integrate the
root filesystem into the kernel image
Details (in kernel sources):

Documentation/filesystems/ramfs-rootfs-initramfs.txt
Documentation/early-userspace/README
211/540
Contents
212/540
Root filesystem organization
The organization of a Linux root filesystem in terms of

directories is well-defined by the Filesystem Hierarchy
Standard
http://www.linuxfoundation.org/collaborate/
workgroups/lsb/fhs
Most Linux systems conform to this specification
I
I
Applications expect this organization

It makes it easier for developers and users as the filesystem
organization is similar in all systems
213/540
Important directories (1)
/bin Basic programs

/boot Kernel image (only when the kernel is loaded from a
filesystem, not common on non-x86 architectures)
/dev Device files (covered later)
/etc System-wide configuration
/home Directory for the users home directories
/lib Basic libraries
/media Mount points for removable media
/mnt Mount points for static media
/proc Mount point for the proc virtual filesystem
214/540
Important directories (2)
/root Home directory of the root user

/sbin Basic system programs
/sys Mount point of the sysfs virtual filesystem
/tmp Temporary files
/usr
/usr/bin Non-basic programs

/usr/lib Non-basic libraries
/usr/sbin Non-basic system programs
/var Variable data files. This includes spool directories

and files, administrative and logging data, and
transient and temporary files
215/540
Separation of programs and libraries

I
Basic programs are installed in /bin and /sbin and basic

libraries in /lib
All other programs are installed in /usr/bin and /usr/sbin

and all other libraries in /usr/lib
In the past, on Unix systems, /usr was very often mounted

over the network, through NFS
In order to allow the system to boot when the network was

down, some binaries and libraries are stored in /bin, /sbin
and /lib
/bin and /sbin contain programs like ls, ifconfig, cp,

bash, etc.
/lib contains the C library and sometimes a few other basic

libraries
All other programs and libraries are in /usr
216/540
Device Files
217/540
Devices
One of the kernel important role is to allow applications to

access hardware devices
In the Linux kernel, most devices are presented to user space
applications through two different abstractions
I
I
Character device
Block device
Internally, the kernel identifies each device by a triplet of

information
I
I
I
Type (character or block)

Major (typically the category of device)
Minor (typically the identifier of the device)
218/540
Types of devices
Block devices
I
A device composed of fixed-sized blocks, that can be read and

written to store data
Used for hard disks, USB keys, SD cards, etc.
Character devices
I
Originally, an infinite stream of bytes, with no beginning, no

end, no size. The pure example: a serial port.
Used for serial ports, terminals, but also sound cards, video
acquisition devices, frame buffers
Most of the devices that are not block devices are represented
as character devices by the Linux kernel
219/540
Devices: everything is a file
A very important Unix design decision was to represent most

of the system objects as files
It allows applications to manipulate all system objects with

the normal file API (open, read, write, close, etc.)
So, devices had to be represented as files to the applications
This is done through a special artifact called a device file
It is a special type of file, that associates a file name visible to

user space applications to the triplet (type, major, minor) that
the kernel understands
All device files are by convention stored in the /dev directory
220/540
Device files examples

Example of device files in a Linux system
$ ls -l /dev/ttyS0 /dev/tty1 /dev/sda1 /dev/sda2
brw-rw---- 1 root disk
8, 1 2011-05-27 08:56
brw-rw---- 1 root disk
8, 2 2011-05-27 08:56
crw------- 1 root root
4, 1 2011-05-27 08:57
crw-rw---- 1 root dialout 4, 64 2011-05-27 08:56
crw-rw-rw- 1 root root
1, 5 2011-05-27 08:56
/dev/zero
/dev/sda1
/dev/sda2
/dev/tty1
/dev/ttyS0
/dev/zero
Example C code that uses the usual file API to write data to a
serial port
int fd;
fd = open("/dev/ttyS0", O_RDWR);
write(fd, "Hello", 5);
close(fd);
221/540
Creating device files
On a basic Linux system, the device files have to be created

manually using the mknod command
I
I
I
mknod /dev/<device> [c|b] major minor

Needs root privileges
Coherency between device files and devices handled by the
kernel is left to the system developer
On more elaborate Linux systems, mechanisms can be added

to create/remove them automatically when devices appear
and disappear
I
I
devtmpfs virtual filesystem, since kernel 2.6.32

udev daemon, solution used by desktop and server Linux
systems
mdev program, a lighter solution than udev
222/540
Pseudo Filesystems
223/540
proc virtual filesystem

I
I
The proc virtual filesystem exists since the beginning of Linux

It allows
I
The kernel to expose statistics about running processes in the

system
The user to adjust at runtime various system parameters about
process management, memory management, etc.
The proc filesystem is used by many standard user space

applications, and they expect it to be mounted in /proc
Applications such as ps or top would not work without the

proc filesystem
Command to mount /proc:

mount -t proc nodev /proc
Documentation/filesystems/proc.txt in the kernel

sources
man proc
224/540
proc contents
I
One directory for each running process in the system

I
I
I
/proc/<pid>
cat /proc/3840/cmdline
It contains details about the files opened by the process, the
CPU and memory usage, etc.
/proc/interrupts, /proc/devices, /proc/iomem,

/proc/ioports contain general device-related information
/proc/cmdline contains the kernel command line

/proc/sys contains many files that can be written to to
adjust kernel parameters
They are called sysctl. See Documentation/sysctl/ in

kernel sources.
Example
echo 3 > /proc/sys/vm/drop_caches
225/540
sysfs filesystem
I
The sysfs filesystem is a feature integrated in the 2.6 Linux

kernel
It allows to represent in user space the vision that the kernel

has of the buses, devices and drivers in the system
It is useful for various user space applications that need to list

and query the available hardware, for example udev or mdev.
All applications using sysfs expect it to be mounted in the

/sys directory
Command to mount /sys:

mount -t sysfs nodev /sys
$ ls /sys/
block bus class dev devices firmware
fs kernel module power
226/540
Minimal filesystem
227/540
Basic applications
I
I
In order to work, a Linux system needs at least a few

applications
An init application, which is the first user space application
started by the kernel after mounting the root filesystem
I
I
I
I
The kernel tries to run /sbin/init, /bin/init, /etc/init

and /bin/sh.
In the case of an initramfs, it will only look for /init.
Another path can be supplied by the rdinit kernel argument.
If none of them are found, the kernel panics and the boot
process is stopped.
The init application is responsible for starting all other user
space applications and services
Usually a shell, to allow a user to interact with the system

Basic Unix applications, to copy files, move files, list files
(commands like mv, cp, mkdir, cat, etc.)
These basic components have to be integrated into the root
filesystem to make it usable
228/540
Overall booting process
229/540
Overall booting process with initramfs
230/540
Busybox
Free Electrons
Busybox
Embedded Linux
Developers
Free Electrons

231/540
Why Busybox?
I
A Linux system needs a basic set of programs to work

I
I
I
In normal Linux systems, these programs are provided by

different projects
I
I
I
An init program
A shell
Various basic utilities for file manipulation and system
configuration
coreutils, bash, grep, sed, tar, wget, modutils, etc. are

all different projects
A lot of different components to integrate
Components not designed with embedded systems constraints
in mind: they are not very configurable and have a wide range
of features
Busybox is an alternative solution, extremely common on

embedded systems
232/540
General purpose toolbox: BusyBox
Rewrite of many useful Unix command line utilities

I
Integrated into a single project, which makes it easy to work

with
Designed with embedded systems in mind: highly configurable,
no unnecessary features
All the utilities are compiled into a single executable,

/bin/busybox
I
Symbolic links to /bin/busybox are created for each

application integrated into Busybox
For a fairly featureful configuration, less than 500 KB

(statically compiled with uClibc) or less than 1 MB (statically
compiled with glibc).
http://www.busybox.net/
233/540
BusyBox commands!
Commands available in BusyBox 1.13

[, [[, addgroup, adduser, adjtimex, ar, arp, arping, ash, awk, basename, bbconfig, bbsh,
brctl, bunzip2, busybox, bzcat, bzip2, cal, cat, catv, chat, chattr, chcon, chgrp, chmod,
chown, chpasswd, chpst, chroot, chrt, chvt, cksum, clear, cmp, comm, cp, cpio, crond, crontab,
cryptpw, cttyhack, cut, date, dc, dd, deallocvt, delgroup, deluser, depmod, devfsd, df,
dhcprelay, diff, dirname, dmesg, dnsd, dos2unix, dpkg, dpkg_deb, du, dumpkmap, dumpleases,
e2fsck, echo, ed, egrep, eject, env, envdir, envuidgid, ether_wake, expand, expr, fakeidentd,
false, fbset, fbsplash, fdflush, fdformat, fdisk, fetchmail, fgrep, find, findfs, fold, free,
freeramdisk, fsck, fsck_minix, ftpget, ftpput, fuser, getenforce, getopt, getsebool, getty,
grep, gunzip, gzip, halt, hd, hdparm, head, hexdump, hostid, hostname, httpd, hush, hwclock,
id, ifconfig, ifdown, ifenslave, ifup, inetd, init, inotifyd, insmod, install, ip, ipaddr,
ipcalc, ipcrm, ipcs, iplink, iproute, iprule, iptunnel, kbd_mode, kill, killall, killall5,
klogd, lash, last, length, less, linux32, linux64, linuxrc, ln, load_policy, loadfont,
loadkmap, logger, login, logname, logread, losetup, lpd, lpq, lpr, ls, lsattr, lsmod, lzmacat,
makedevs, man, matchpathcon, md5sum, mdev, mesg, microcom, mkdir, mke2fs, mkfifo, mkfs_minix,
mknod, mkswap, mktemp, modprobe, more, mount, mountpoint, msh, mt, mv, nameif, nc, netstat,
nice, nmeter, nohup, nslookup, od, openvt, parse, passwd, patch, pgrep, pidof, ping, ping6,
pipe_progress, pivot_root, pkill, poweroff, printenv, printf, ps, pscan, pwd, raidautorun,
rdate, rdev, readahead, readlink, readprofile, realpath, reboot, renice, reset, resize,
restorecon, rm, rmdir, rmmod, route, rpm, rpm2cpio, rtcwake, run_parts, runcon, runlevel,
runsv, runsvdir, rx, script, sed, selinuxenabled, sendmail, seq, sestatus, setarch,
setconsole, setenforce, setfiles, setfont, setkeycodes, setlogcons, setsebool, setsid,
setuidgid, sh, sha1sum, showkey, slattach, sleep, softlimit, sort, split, start_stop_daemon,
stat, strings, stty, su, sulogin, sum, sv, svlogd, swapoff, swapon, switch_root, sync, sysctl,
syslogd, tac, tail, tar, taskset, tcpsvd, tee, telnet, telnetd, test, tftp, tftpd, time, top,
touch, tr, traceroute, true, tty, ttysize, tune2fs, udhcpc, udhcpd, udpsvd, umount, uname,
uncompress, unexpand, uniq, unix2dos, unlzma, unzip, uptime, usleep, uudecode, uuencode,
vconfig, vi, vlock, watch, watchdog, wc, wget, which, who, whoami, xargs, yes, zcat, zcip
234/540
Applet highlight: Busybox init
Busybox provides an implementation of an init program
Simpler than the init implementation found on desktop/server

systems: no runlevels are implemented
A single configuration file: /etc/inittab
Each line has the form <id>::<action>:<process>
Allows to run services at startup, and to make sure that

certain services are always running on the system
See examples/inittab in Busybox for details on the

configuration
235/540
Applet highlight - BusyBox vi
If you are using BusyBox, adding vi support only adds 20K.

(built with shared libraries, using uClibc).
You can select which exact features to compile in.
Users hardly realize that they are using a lightweight vi

version!
Tip: you can learn vi on the desktop, by running the

vimtutor command.
236/540
Configuring BusyBox
I
I
Get the latest stable sources from http://busybox.net

Configure BusyBox (creates a .config file):
I
make defconfig
Good to begin with BusyBox.
Configures BusyBox with all options for regular users.
make allnoconfig
Unselects all options. Good to configure only what you need.
make xconfig (graphical, needs the libqt3-mt-dev

package)
or make menuconfig (text)
Same configuration interfaces as the ones used by the Linux
kernel (though older versions are used).
237/540
BusyBox make xconfig
You can choose:

I
the commands
to compile,
and even the

command
options and
features that
you need!
238/540
Compiling BusyBox
I
Set the cross-compiler prefix in the configuration interface:

BusyBox Settings -> Build Options > Cross Compiler prefix
Example: arm-linuxSet the installation directory in the configuration interface:
BusyBox Settings -> Installation Options > BusyBox installation prefix
Add the cross-compiler path to the PATH environment
variable:
export PATH=/usr/xtools/arm-unknown-linuxuclibcgnueabi/bin:$PATH
Compile BusyBox:
make
Install it (this creates a Unix directory structure symbolic links
to the busybox executable):
make install
239/540
Practical lab - A tiny embedded system
Make Linux boot on a directory on

your workstation, shared by NFS
Create and configure a minimalistic

Linux embedded system
Install and use BusyBox
System startup with /sbin/init
Set up a simple web interface
Use shared libraries
240/540
Block filesystems
Free Electrons
Block filesystems
Embedded Linux
Developers
Free Electrons

241/540
Block vs. flash

I
Storage devices are classified in two main types: block

devices and flash devices
I
Block devices can be read and written to on a per-block

basis, without erasing.
I
I
They are handled by different subsystems and different

filesystems
Hard disks, floppy disks, RAM disks

USB keys, Compact Flash, SD card: these are based on flash
storage, but have an integrated controller that emulates a
block device, managing and erasing flash sectors in a
transparent way.
Raw flash devices are driven by a controller on the SoC.

They can be read, but writing requires erasing, and often
occurs on a larger size than the block size.
I
NOR flash, NAND flash
242/540
Block device list
The list of all block devices available in the system can be

found in /proc/partitions
$ cat /proc/partitions
major minor #blocks name
179
179
179
8
8
8
0
1
2
0
1
2
3866624
73712
3792896
976762584
1060258
975699742
mmcblk0
mmcblk0p1
mmcblk0p2
sda
sda1
sda2
And also in /sys/block/
243/540
Traditional block filesystems
Traditional filesystems
I
Can be left in a non-coherent state after a system crash or

sudden poweroff, which requires a full filesystem check after
reboot.
ext2: traditional Linux filesystem

(repair it with fsck.ext2)
vfat: traditional Windows filesystem

(repair it with fsck.vfat on GNU/Linux or Scandisk on
Windows)
244/540
Journaled filesystems
Designed to stay in a
correct state even
after system crashes
or a sudden poweroff
All writes are first

described in the
journal before being
committed to files
245/540
Filesystem recovery after crashes
Thanks to the
journal, the filesystem
is never left in a
corrupted state
Recently saved data

could still be lost
246/540
Journaled block filesystems

Journaled filesystems
I
ext3: ext2 with journal extension

ext4: newest version in the family with many improvements.
Btrfs (Butter FS)

The next generation. Great performance. Now used in major
GNU/Linux distros.
The Linux kernel supports many other filesystems: reiserFS,

JFS, XFS, etc. Each of them have their own characteristics,
but are more oriented towards server or scientific workloads.
Its easy to switch filesystems. The best is to try each and find
out which yields the best performance on your own system.
We recommend ext2 for very small partitions (< 5 MB), because

other filesystems need too much space for metadata (ext3 and
ext4 need about 1 MB for a 4 MB partition).
247/540
Creating ext2/ext3/ext4 volumes
To create an empty ext2/ext3/ext4 filesystem on a block

device or inside an already-existing image file
I
I
I
I
mkfs.ext2
mkfs.ext3
mkfs.ext4
mkfs.ext2
/dev/hda3
/dev/sda2
/dev/sda3
disk.img
To create a filesystem image from a directory containing all

your files and directories
I
I
I
Use the genext2fs tool, from the package of the same name
genext2fs -d rootfs/ rootfs.img
Your image is then ready to be transferred to your block device
248/540
Mounting filesystem images

I
Once a filesystem image has been created, one can access and
modifies its contents from the development workstation, using
the loop mechanism
Example:
genext2fs -d rootfs/ rootfs.img
mkdir /tmp/tst
mount -t ext2 -o loop rootfs.img /tmp/tst
In the /tmp/tst directory, one can access and modify the

contents of the rootfs.img file.
This is possible thanks to loop, which is a kernel driver that

emulates a block device with the contents of a file.
Do not forget to run umount before using the filesystem

image!
249/540
F2FS
http://en.wikipedia.org/wiki/F2FS
I
Filesystem optimized for block devices based on NAND flash
Available in the mainline Linux kernel
Benchmarks: best performer on flash devices most of the time:

See http://lwn.net/Articles/520003/
Technical details: http://lwn.net/Articles/518988/
250/540
Squashfs
Squashfs: http://squashfs.sourceforge.net
I
Read-only, compressed filesystem for block devices. Fine for

parts of a filesystem which can be read-only (kernel,
binaries...)
Great compression rate and read access performance
Used in most live CDs and live USB distributions
Supports LZO compression for better performance on

embedded systems with slow CPUs (at the expense of a
slightly degraded compression rate)
Now supports the XZ algorithm, for a much better

compression rate, at the expense of higher CPU usage and
time.
Benchmarks: (roughly 3 times smaller than ext3, and 2-4 times

faster)
http://elinux.org/Squash_Fs_Comparisons
251/540
Squashfs - How to use

I
I
Need to install the squashfs-tools package

Creation of the image
I
Installation of the image

I
I
On your workstation, create your filesystem image:

mksquashfs rootfs/ rootfs.sqfs
Caution: if the image already exists remove it first,
or use the -noappend option.
Lets assume your partition on the target is in /dev/sdc1
Copy the filesystem image on the device
dd if=rootfs.sqfs of=/dev/sdc1
Be careful when using dd to not overwrite the incorrect
partition!
Mount your filesystem:

mount -t squashfs /dev/sdc1 /mnt/root
252/540
tmpfs
Not a block filesystem of course!
Perfect to store temporary data in RAM: system log files,
connection data, temporary files...
I
tmpfs configuration:
File systems -> Pseudo filesystems
Lives in the Linux file cache. Doesnt waste RAM: unlike
ramdisks, no need to copy files to the file cache, grows and
shrinks to accommodate stored files. Saves RAM: can swap
out pages to disk when needed.
How to use: choose a name to distinguish the various tmpfs

instances you could have. Examples:
mount -t tmpfs varrun /var/run
mount -t tmpfs udev /dev
See Documentation/filesystems/tmpfs.txt in kernel sources.

253/540
Mixing read-only and read-write filesystems

Good idea to split your block storage into:
I
A compressed read-only partition

(Squashfs)
Typically used for the root filesystem
(binaries, kernel...).
Compression saves space. Read-only
access protects your system from mistakes
and data corruption.
A read-write partition with a journaled

filesystem (like ext3)
Used to store user or configuration data.
Guarantees filesystem integrity after
power off or crashes.
Ram storage for temporary files (tmpfs)
254/540
Issues with flash-based block storage
Flash storage made available only through a block interface.
Hence, no way to access a low level flash interface and use the
Linux filesystems doing wear leveling.
No details about the layer (Flash Translation Layer) they use.

Details are kept as trade secrets, and may hide poor
implementations.
Can use flashbench

(https://github.com/bradfa/flashbench) to find out
the erase block size and optimize filesystem formating.
Not knowing about the wear leveling algorithm, it is highly

recommended to limit the number of writes to these devices.
255/540
Practical lab - Block filesystems
Creating partitions on your block

storage
Booting your system with a mix of

filesystems: SquashFS for the root
filesystem (including applications),
ext3 for configuration and user
data, and tmpfs for temporary
system files.
256/540
Flash filesystems
Free Electrons
Flash filesystems
Embedded Linux
Developers
Free Electrons

257/540
Block devices vs flash devices: reminder
Block devices:
I
I
I
Allow for random data access using fixed size blocks

Do not require special care when writing on the media
Block size is relatively small (minimum 512 bytes, can be
increased for performance reasons)
Considered as reliable (if the storage media is not, some
hardware or software parts are supposed to make it reliable)
Flash devices:
I
I
Allow for random data access too

Require special care before writing on the media (erasing the
region you are about to write on)
Erase, write and read operation might not use the same block
size
Reliability depends on the flash technology
258/540
NAND flashes: how they work ?
Encode bits with Voltage levels
Start with all bits set to 1
Programming implies changing some bits from 1 to 0
Restoring bits to 1 is done via the ERASE operation
Programming and erasing is not done on a per bit or per byte

basis
Organization
I
I
Page: minimum unit for PROGRAM operation

Block: minimum unit for ERASE operation
259/540
NAND flashes: organization
260/540
NAND flashes: constraints

I
Reliability
I
I
I
Lifetime
I
I
I
I
Far less reliable than NOR flashes

Reliability depends on the NAND flash technology (SLC, MLC)
Require additional mechanisms to recover from bit flips: ECC
(Error Corrector Code)
ECC information stored in the OOB (Out-of-band area)
Short lifetime compared to other storage media
Lifetime depends on the NAND flash technology (SLC, MLC):
between 1000000 and 1000 erase cycles per block
Wear leveling mechanisms are required
Bad block detection/handling required too
Despite the number of constraints brought by NAND they are

widely used in embedded systems for several reasons:
I
I
I
Cheaper than other flash technologies

Provide high capacity storage
Provide good performance (both in read and write access)
261/540
NAND flashes: ECC

I
ECC partly address the reliability problem on NAND flashes
Operates on blocks (usually 512 or 1024 bytes)
ECC data are stored in the OOB area

Three algorithms:
I
I
I
Hamming: can fixup a single bit per block

Reed-Solomon: can fixup several bits per block
BCH: can fixup several bits per block
BCH and Reed-Solomon strengths depends on the size

allocated for ECC data, which in turn depends on the OOB
size
NAND manufacturers specify the required ECC strength in

their datasheets: ignoring these requirements might
compromise data integrity
262/540
The MTD subsystem (1)

I
I
I
I
I
I
MTD stands for Memory Technology Devices

Generic subsystem dealing with all type of storage media that
are not fitting in the block subsystem
Supported media types: RAM, ROM, NOR flash, NAND
flash, Dataflash
Independent of the communication interface (drivers available
for parallel, SPI, direct memory mapping, ...)
Abstract storage media characteristics and provide a simple
API to access MTD devices
MTD device characteristics exposed to users:
I
I
I
I
I
erasesize: minimum erase size unit

writesize: minimum write size unit
oobsize: extra size to store metadata or ECC data
size: device size
flags: information about device type and capabilities
Various kind of MTD users: file-systems, block device

emulation layers, userspace interfaces, ...
263/540
The MTD subsystem (2)
264/540
MTD partitioning
I
MTD devices are usually partitioned

I
Unlike block devices, which contains their own partition table,

the partitioning of MTD devices is described externally
I
I
I
It allows to use different areas of the flash for different

purposes: read-only filesystem, read-write filesystem, backup
areas, bootloader area, kernel area, etc.
Specified in the board Device Tree

Hard-coded into the kernel code (if no Device Tree)
Specified through the kernel command line
Each partition becomes a separate MTD device

I
I
Different from block device labeling (hda3, sda2)

/dev/mtd1 is either the second partition of the first flash
device, or the first partition of the second flash device
Note that the master MTD device (the device those partitions
belongs to) is not exposed in /dev
265/540
U-Boot: manipulating NAND devices

I
U-Boot provides a set of commands to manipulate NAND

devices
Grouped under the nand command
I
I
I
I
nand info: show available NAND devices and their

characteristics
nand device [dev]: select or display the active NAND
device
nand read[.option] <addr> <offset|part> <size>
read data from the NAND
nand write[.option] <addr> <offset|part> <size>
write data on the NAND
Use nand write.trimffs to avoid writing empty pages
(those filled with 0xff)
nand erase <offset> <size> erase a NAND region
nand erase.part <part> erase a NAND partition
more commands for debugging purpose
266/540
U-Boot: manipulating NOR devices (1)
U-Boot provides a set of commands to manipulate NOR

devices
Memory mapped NOR devices
I
I
I
I
flinfo [devid] display information of all NOR devices or a

specific one if devid is provided
cp.[bwl] <src> <target> <count> read/write data
frm/to the NOR device
erase <start> <end> or erase <start> +<len> erase a
memory region
erase bank <bankid> erase a memory bank
erase all erase all banks
protect on|off <range-description> protect a memory
range
267/540
U-Boot: manipulating NOR devices (2)
SPI NOR devices

I
I
I
I
Grouped under the sf command

sf probre [[bus:]cs] [hz] [mode] probe a NOR device
on
sf read|write <addr> <offset> <len> read/write data
from/to a SPI NOR
sf erase <offset> +<len> erase a memory region
sf update <addr> <offset> <len> erase + write
operation
268/540
U-Boot: using MTD partitions

I
I
I
U-Boot provides a way to define mtd partitions on top of flash

devices
Using partitions eases the manipulation of flash devices, which
might be particularly error prone
Requires two environment variables:
I
mtdids attaches an mtdid to a flash device. Example:

mtdids=<devid>=<mtdid>[,<devid>=<mtdid>]
I
I
I
devid: device identifier retrieved with nand info or flinfo

mtdid: mtd identifier (should match linux MTD device name)
mtdid is displayed when booting the linux kernel
mtdparts defines the partitions for the different devices
Use the mtdparts command to setup the configuration

specified by the mtdids and mtdparts variables
Use flash specific commands (depends on your flash device
type), and pass partition names instead of numerical offsets
Example: nand erase <partname>
269/540
U-Boot: defining MTD partitions (1)
mtdparts format: <mtdid>:<partition>[,partition]
partition format: <size>[@offset](<name>)[ro]
mtdid is the MTD device name assigned by linux
mtdid is displayed at boot (and can be retrieved using dmesg)
NAND device: Manufacturer ID: 0x2c, Chip ID: 0xbc (Micron NAND 512MiB 1,8V 16-bit)
Creating 5 MTD partitions on "omap2-nand.0":
0x000000000000-0x000000080000 : "X-Loader"
0x000000080000-0x000000200000 : "U-Boot"
0x000000200000-0x000000280000 : "Environment"
0x000000280000-0x000000580000 : "Kernel"
0x000000580000-0x000020000000 : "File System"
270/540
U-Boot: defining MTD partitions (2)

I
Example:
mtdids=nand0=omap2-nand.0
mtdparts=mtdparts=omap2-nand.0:512k(XLoader)ro,1536k(UBoot)ro,512k(Environment),4m(Kernel),-(RootFS)
Weve just defined 5 partitions in the omap2-nand.0 device:
I
I
I
I
I
I
I
1st stage bootloader (512 KiB, read-only)

U-Boot (1536 KiB, read-only)
U-Boot environment (512 KiB)
Kernel (4 MiB)
Root filesystem (Remaining space)
Partition sizes must be multiple of the erase block size.

You can use sizes in hexadecimal too. Remember the below
sizes:
0x20000 = 128k, 0x100000 = 1m, 0x1000000 = 16m
ro lists the partition as read only
- is used to use all the remaining space.
271/540
Linux: MTD devices interface with userspace
MTD devices are visible in /proc/mtd
The userspace only see MTD partitions, not the flash device
under those partitions
The mtdchar driver creates a character device for each MTD
device/partition of the system
I
I
Usually named /dev/mtdX or /dev/mtdXro, major 90. Even

minors for read-write access (/dev/mtdX), odd minors for
read-only access (/dev/mtdXro)
Provide ioctl() to erase and manage the flash
Used by the mtd-utils utilities
272/540
Linux: definition of MTD partitions (1)

The Device Tree is the standard place to define MTD partitions for
boards with Device Tree support.
Example from arch/arm/boot/dts/omap3-igep0020.dts:
nand@0,0 {
linux,mtd-name= "micron,mt29c4g96maz";
[...]
partition@0 {
label = "SPL";
reg = <0 0x100000>;
};
partition@0x80000 {
label = "U-Boot";
reg = <0x100000 0x180000>;
};
[...]
partition@0x780000 {
label = "Filesystem";
reg = <0x680000 0x1f980000>;
};
273/540
Linux: definition of MTD partitions (2)

For boards or platforms that do not use the Device Tree, MTD
partitions can be defined in the kernel. Legacy example from
arch/arm/mach-omap2/board-igep0020.c (removed in 3.13):
static struct mtd_partition igep2_flash_partitions[] = {
{
.name
= "X-Loader",
.offset = 0,
.size
= 2 * (64*(2*2048))
},
{
.name
= "U-Boot",
.offset = MTDPART_OFS_APPEND,
.size
= 6 * (64*(2*2048)),
},
[...]
{
.name
= "File System",
.offset = MTDPART_OFS_APPEND,
.size
= MTDPART_SIZ_FULL,
},
};
274/540
Linux: modifying MTD partitions (3)
MTD partitions can fortunately be defined through the kernel

command line.
U-Boot and linux are using the same syntax to describe the
MTD partitions
Directly pass the mtdparts environment variable defined in

u-boot to the kernel cmdline
Be careful: when modifying the partition layout you might

impact MTD users which have already stored data on existing
partitions: e.g. a file system might complain about data
corruption if you change your partition layout after the file
system initialization
275/540
Linux: userspace flash management tools
mtd-utils is a set of utilities to manipulate MTD devices

I
I
I
I
I
mtdinfo to get detailed information about an MTD device

flash_erase to partially or completely erase a given MTD
device
flashcp to write to NOR flash
nandwrite to write to NAND flash
Flash filesystem image creation tools: mkfs.jffs2,
mkfs.ubifs, ubinize, etc.
Usually available as the mtd-utils package in your

distribution
Most commands now also available in BusyBox
See http://www.linux-mtd.infradead.org/
276/540
Flash wear leveling (1)
Wear leveling consists in distributing writes over the whole

flash device to avoid quickly reaching the maximum number
of erase cycles on blocks that are written really often
Can be done in:
I
I
the filesystem layer (JFFS2, YAFFS/YAFFS2, ...)

an intermediate layer dedicated to wear leveling (UBI)
The wear leveling implementation is what makes your flash

lifetime good or not
277/540
Flash wear leveling (2)
Flash users should also take the limited lifetime of flash

devices into account by taking additional precautions
I
I
I
Do not use your flash storage as swap area (rare in embedded

systems anyway)
Mount your filesystems as read-only, or use read-only
filesystems (SquashFS), whenever possible.
Keep volatile files in RAM (tmpfs)
Dont use the sync mount option (commits writes
immediately). Use the fsync() system call for per-file
synchronization.
278/540
Flash file-systems
Standard file systems are meant to work on block devices
Specific file systems have been developed to deal flash

constraints
These file systems are relying on the MTD layer to access

flash chips
There are several legacy flash filesystems which might be

useful for specific usage: JFFS2, YAFFS2.
Nowadays, UBI/UBIFS is the de facto standard for medium to

large capacity NANDs (above 128MB)
279/540
Legacy flash filesystems: JFFS2
Supports on the fly compression
Wear leveling, power failure resistant
Available in the official Linux kernel
Boot time depends on the filesystem size:

doesnt scale well for large partitions.
http://www.linuxmtd.infradead.org/doc/jffs2.html
280/540
Legacy flash filesystems: YAFFS2
Mainly supports NAND flash
No compression
Wear leveling, power failure resistant
Fast boot time
Not part of the official Linux kernel: code

only available separately
(Dual GPL / Proprietary license for non
Linux operating systems)
http://www.yaffs.net/
281/540
UBI/UBIFS
Aimed at replacing JFFS2 by addressing

its limitations
Design choices:
I
I
I
Split the wear leveling and filesystem

layers
Add some flexibility
Focus on scalability, performance and
reliability
Drawback: introduces noticeable

overhead, especially when used on small
devices or partitions
282/540
UBI (1)
Unsorted Block Images
I
http://www.linux-mtd.infradead.org/doc/ubi.html
Volume management system on top of MTD devices (similar

to what LVM provides for block devices)
Allows to create multiple logical volumes and spread writes

across all physical blocks
Takes care of managing the erase blocks and wear leveling.

Makes filesystems easier to implement
Wear leveling can operate on the whole storage, not only on

individual partitions (strong advantage)
Volumes can be dynamically resized or, on the opposite, can

be read-only (static)
283/540
UBI (2)
284/540
UBI: internals
I
UBI is storing its metadata

in-band
In each MTD erase block
I
One page is reserved to

count the number of erase
cycles
Another page is reserved
to attach the erase block
to a UBI volume
The remaining pages are
used to store payload data
If the device supports

subpage write, the EC and
VID headers can be stored
on the same page
285/540
UBI: good practice
UBI is responsible for distributing writes all over the flash

device: the more space you assign to a partition attached to
the UBI layer the more efficient the wear leveling will be
If you need partitioning, use UBI volumes not MTD partitions
Some partitions will still have to be MTD partitions: e.g. the

bootloaders and bootloader environments
If you need extra MTD partitions, try to group them at the

end or the beginning of the flash device
286/540
UBI layout: bad example
287/540
UBI layout: good example
288/540
UBIFS
Unsorted Block Images File System

I
http://www.linux-mtd.infradead.org/doc/ubifs.html
The filesystem part of the UBI/UBIFS couple
Works on top of UBI volumes
Journaling file system providing better performance than

JFFS2 and addressing its scalability issues
See this paper for more technical details about UBIFS

internals http://www.linuxmtd.infradead.org/doc/ubifs_whitepaper.pdf
289/540
Linux: UBI host tools (1)
ubinize is the only host tool for the UBI layer
Creates a UBI image to be flashed on an MTD partition

Takes the following arguments:
I
I
I
-o <output-file-path> path to the output image file

-p <peb-size> the PEB size (MTD erase block size)
-m <min-io-size> the minimum write unit size (e.g. MTD
write size)
-s <subpage-size> subpage size, only needed if both your
flash and your flash controller are supporting subpage writes
The last argument is a path to a UBI image description file
Example:
ubinize -o ubi.img -p 16KiB -m 512 -s 256 cfg.ini
290/540
Linux: UBI host tools (2)

I
ubinize config file can contain several sections
Each section is describing a UBI volume
Example:
[kernel-volume]
mode=ubi
image=zImage
vol_id=1
vol_type=static
vol_name=kernel
vol_alignment=1
I
[rootfs-volume]
mode=ubi
image=rootfs.squashfs
vol_id=2
vol_type=static
vol_name=rootfs
vol_alignment=1
[data-volume]
mode=ubi
image=data.ubifs
vol_id=3
vol_size=30MiB
vol_type=dynamic
vol_name=data
vol_flags=autoresize
vol_alignment=1
Build systems like Buildroot can run ubinize automatically!
291/540
U-Boot: UBI tools
Grouped under the ubi command

I
I
I
I
I
ubi part <part-name> attach an MTD partition to the UBI

layer
ubi info [layout] display UBI device information (or
volume information if the layout string is passed
ubi check <vol-name> check if a volume exists
ubi readvol <dest-addr> <vol-name> [<size>] read
volume content
U-Boot also provides tools to update the UBI device content
Using them is highly discouraged (the U-Boot UBI
implementation is not entirely stable, and using commands
that do not touch the UBI metadata is safer)
I
I
I
ubi createvol <vol-name> [<size>] [<type>]

ubi removevol <vol-name>
ubi writevol <src-addr> <vol-name> <size>
292/540
Linux: UBI target tools (1)
Tools used on the target to dynamically create and modify

UBI elements
UBI device management:
I
ubiformat <MTD-device-id> format an MTD partition and

preserve Erase Counter information if any
ubiattach -m <MTD-device-id> /dev/ubi_ctrl attach
an MTD partition/device to the UBI layer, and create a UBI
device
ubidetach -m <MTD-device-id> /dev/ubi_ctrl detach
an MTD partition/device from the UBI layer, and remove this
associated UBI device
293/540
Linux: UBI target tools (2)
UBI volume management:

I
ubimkvol /dev/ubi<UBI-device-id> -N <name> s <size> create a new volume. Use -m in place of

-s <size> if you want to assign all the remaining to this
volume
ubirmvol /dev/ubi<UBI-device-id> -N <name> delete
an UBI volume
ubiupdatevol /dev/ubi<UBI-device-id>_<UBI-volid> [-s <size>] <vol-image-file> update volume
content
ubirsvol /dev/ubi<UBI-device-id> -N <name> s <size> resize an UBI volume
ubirename /dev/ubi<UBI-device-id>_<UBI-volid> <old-name> <new-size> rename a UBI volume
294/540
Linux: UBIFS host tools
UBIFS filesystems images can be created using the

mkfs.ubifs utility
I
mkfs.ubifs -m 4096 -e 252KiB -c 1000 r rootfs/ ubifs.img

I
-m 4096, minimal I/O size

(see /sys/class/mtd/mtdx/writesize).
-e 252KiB, logical erase block size (smaller than PEB size,
look at dmesg)
-c 1000, maximum number of logical erase blocks. Details:
http://linux-mtd.infradead.org/faq/ubifs.html#L_max_leb_cnt
Once created
I
Can be written to a UBI volume from the target using

ubiupdatevol
Or, can be included in a UBI image (using ubinize on the
host)
295/540
Linux: UBIFS target tools
No specific tools are required to manipulate a UBIFS

filesystem
Mounting a UBIFS filesystem is done with mount:

mount -t ubifs <ubi-device-id>:<volumename> <mount-point>
Example:
mount -t ubifs ubi0:data /data
296/540
Linux: UBI image creation workflow
297/540
Linux: flashing UBI/UBIFS howtos
See http://free-electrons.com/blog/creatingflashing-ubi-ubifs-images/ for details about creating

UBI and UBIFS images.
See http://www.linuxmtd.infradead.org/doc/ubi.html#L_flasher_algo for

what is required when flashing UBI images.
298/540
Linux: Using a UBIFS filesystem as a rootfs
You just have to pass the following information on the kernel

cmdline:
I
ubi.mtd=1: attach /dev/mtd1 to the UBI layer and create

ubi0
rootfstype=ubifs root=ubi0:rootfs: mount the rootfs
volume on ubi0 as a UBIFS filesystem
Example:
rootfstype=ubifs ubi.mtd=1 root=ubi0:rootfs
299/540
Flash filesystem benchmarks
http://elinux.org/Flash_Filesystem_Benchmarks
I jffs2
I
I
yaffs2
I
Worst performance
Requires CONFIG_SUMMARY to have acceptable boot time
Good performance, but not in mainline Linux
ubifs
I
I
Best solution and performance for medium and big partitions

Too much metadata overhead small partitions (only case when
yaffs2 and jffs2 are still useful)
300/540
Linux: Block emulation layers
Sometimes we need block devices to re-use existing block

filesystems
Particularly useful for read-only block filesystems like squashfs

Linux provides two block emulation layers:
I
I
mtdblock: block devices emulated on top of MTD devices

ubiblock: block devices emulated on top of UBI volumes
For read access, using emulated block devices is exactly the

same as using regular block devices
Even if supported through the mtdblock emulation layer,
writing on emulated block devices is highly discouraged
I
the emulation layer does not properly deal with wear leveling
and data retention issues occurring on the flash media
301/540
Linux: mtdblock
The mtdblock layer creates a block device for each MTD

device of the system
Usually named /dev/mtdblockX, major 31. Minor is the

number of the MTD device
Allows read/write block-level access. But bad blocks are not

handled, and no wear leveling is done for writes.
For historical reasons JFFS2 filesystems require a block device

to be mounted
Do not write on mtdblock devices
302/540
Linux: ubiblock
I
I
I
Preferred over mtdblock if UBI is available (UBI accounts for

data retention and wear leveling issues, while MTD does not)
The ubiblock layer creates read-only block devices on
demand
The user specifies which static volumes he would like to
attach to ubiblock
I
Through the cmdline: by passing

ubi.block=<ubi-dev-id>,<volume-name>
Using the ubiblock utility provided by mtd-utils:
ubiblock --create <ubi-volume-dev-file>
Usually named /dev/ubiblockX_Y, where X is the UBI

device id and Y is the UBI volume id
Major is dynamically assigned and stay the same for all
ubiblock devices
Minor is the id assigned to the ubiblock device (depends on
the total number of ubiblock devices and the creation order).
303/540
Useful reading
Managing flash storage with Linux:

http://free-electrons.com/blog/managing-flashstorage-with-linux/
Documentation on the linux-mtd website:

http://www.linux-mtd.infradead.org/
304/540
Practical lab - Flash Filesystems
Creating partitions in your internal

flash storage
Creating a UBI image with several

volumes and flashing it from
U-Boot
Manipulating UBI volumes from

linux
305/540
Embedded Linux
system
development
Free Electrons
Embedded Linux
Developers
Free Electrons

306/540
Contents
I
I
Using open-source components

Tools for the target device
I
I
I
I
I
I
I
Networking
System utilities
Language interpreters
Audio, video and multimedia
Graphical toolkits
Databases
Web browsers
System building
Commercial tool sets and distributions
307/540
Leveraging open-source components

in an Embedded Linux system
308/540
Third party libraries and applications
One of the advantages of embedded Linux is the wide range

of third-party libraries and applications that one can leverage
in its product
I
They are freely available, freely distributable, and thanks to

their open-source nature, they can be analyzed and modified
according to the needs of the project
However, efficiently re-using these components is not always

easy. One must:
I
I
I
I
Find these components

Choose the most appropriate ones
Cross-compile them
Integrate them in the embedded system and with the other
applications
309/540
Find existing components
Free Software Directory

http://directory.fsf.org
Look at other embedded Linux products, and see what their

components are
Look at the list of software packaged by embedded Linux
build systems
These are typically chosen for their suitability to embedded

systems
Ask the community or Google
This presentation will also feature a list of components for

common needs
310/540
Choosing components
I
Not all free software components are necessarily good to

re-use. One must pay attention to:
I
Vitality of the developer and user communities. This vitality

ensures long-term maintenance of the component, and
relatively good support. It can be measured by looking at the
mailing-list traffic and the version control system activity.
Quality of the component. Typically, if a component is
already available through embedded build systems, and has a
dynamic user community, it probably means that the quality is
relatively good.
License. The license of the component must match your
licensing constraints. For example, GPL libraries cannot be
used in proprietary applications.
Technical requirements. Of course, the component must
match your technical requirements. But dont forget that you
can improve the existing components if a feature is missing!
311/540
Licenses (1)
I
All software that are under a free software license give four
freedoms to all users
I
I
I
I
Freedom
Freedom
Freedom
Freedom
to
to
to
to
use
study
copy
modify and distribute modified copies
See http://www.gnu.org/philosophy/free-sw.html for

a definition of Free Software
Open Source software, as per the definition of the Open

Source Initiative, are technically similar to Free Software in
terms of freedoms
See http://www.opensource.org/docs/osd for the

definition of Open Source Software
312/540
Licenses (2)
I
Free Software licenses fall in two main categories

I
I
The concept of copyleft is to ask for reciprocity in the

freedoms given to a user.
The result is that when you receive a software under a
copyleft free software license and distribute modified versions
of it, you must do so under the same license
I
I
The copyleft licenses

The non-copyleft licenses
Same freedoms to the new users

Its an incentive to contribute back your changes instead of
keeping them secret
Non-copyleft licenses have no such requirements, and

modified versions can be kept proprietary, but they still require
attribution
313/540
GPL
I
I
GNU General Public License

Covers around 55% of the free software projects
I
Is a copyleft license
I
I
Including the Linux kernel, Busybox and many applications

Requires derivative works to be released under the same license
Programs linked with a library released under the GPL must
also be released under the GPL
Some programs covered by version 2 (Linux kernel, Busybox

and others)
More and more programs covered by version 3, released in
2007
I
Major change for the embedded market: the requirement that

the user must be able to run the modified versions on the
device, if the device is a consumer device
314/540
GPL: redistribution
I
No obligation when the software is not distributed

I
It is then authorized to distribute binary versions, if one of the

following conditions is met:
I
You can keep your modifications secret until the product

delivery
Convey the binary with a copy of the source on a physical

medium
Convey the binary with a written offer valid for 3 years that
indicates how to fetch the source code
Convey the binary with the network address of a location
where the source code can be found
See section 6. of the GPL license
In all cases, the attribution and the license must be preserved

I
See section 4. and 5.
315/540
LGPL
I
GNU Lesser General Public License
Covers around 10% of the free software projects

A copyleft license
I
I
Used instead of the GPL for most of the libraries, including

the C libraries
I
Modified versions must be released under the same license

But, programs linked against a library under the LGPL do not
need to be released under the LGPL and can be kept
proprietary.
However, the user must keep the ability to update the library
independently from the program. Dynamic linking is the
easiest solution. Statically linked executables are only possible
if the developer provides a way to relink with an update (with
source code or linkable object files).
Some exceptions: MySQL, or Qt <= 4.4
Also available in two versions, v2 and v3
316/540
Licensing: examples
I
You make modifications to the Linux kernel (to add drivers or

adapt to your board), to Busybox, U-Boot or other GPL
software
I
You make modifications to the C library or any other LGPL

library
I
You must release the modified versions under the same license
You create an application that relies on LGPL libraries

I
You must release the modified versions under the same license,
and be ready to distribute the source code to your customers
You can keep your application proprietary, but you must link
dynamically with the LGPL libraries
You make modifications to a non-copyleft licensed software

I
You can keep your modifications proprietary, but you must still
credit the authors
317/540
Non-copyleft licenses
A large family of non-copyleft licenses that are relatively

similar in their requirements
A few examples
I
I
I
I
I
Apache license (around 4%)

BSD license (around 6%)
MIT license (around 4%)
X11 license
Artistic license (around 9 %)
318/540
BSD license
Copyright (c) <year>, <copyright holder>

All rights reserved.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are
met:
* Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
* Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
* Neither the name of the <organization> nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
[...]
319/540
Is this free software?
Most of the free software projects are covered by 10

well-known licenses, so it is fairly easy for the majority of
project to get a good understanding of the license
Otherwise, read the license text
Check Free Software Foundations opinion

http://www.fsf.org/licensing/licenses/
Check Open Source Initiatives opinion

http://www.opensource.org/licenses
320/540
Respect free software licenses

I
Free Software is not public domain software, the distributors

have obligations due to the licenses
I
Free Software licenses have been enforced successfully in

courts. Organizations which can help:
I
Before using a free software component, make sure the license

matches your project constraints
Make sure to keep a complete list of the free software
packages you use, the original version you used and to keep
your modifications and adaptations well-separated from the
original version
Conform to the license requirements before shipping the
product to the customers
Software Freedom Law Center,

http://www.softwarefreedom.org/
Software Freedom Conservancy,
http://sfconservancy.org/
Ask your legal department!
321/540
Keeping changes separate (1)

I
When integrating existing open-source components in your

project, it is sometimes needed to make modifications to them
I
Instead of mixing these changes, it is much better to keep

them separate from the original component version
I
Better integration, reduced footprint, bug fixes, new features,

etc.
If the component needs to be upgraded, easier to know what

modifications were made to the component
If support from the community is requested, important to
know how different the component were using is from the
upstream version
Makes contributing the changes back to the community
possible
It is even better to keep the various changes made on a given

component separate
I
Easier to review and to update to newer versions
322/540
Keeping changes separate (2)
The simplest solution is to use Quilt

I
Quilt is a tool that allows to maintain a stack of patches over

source code
Makes it easy to add, remove modifications from a patch, to
add and remove patches from stack and to update them
The stack of patches can be integrated into your version
control system
https://savannah.nongnu.org/projects/quilt/
Another solution is to use a version control system

I
Import the original component version into your version

control system
Maintain your changes in a separate branch
323/540
Tools for the target device:

Networking
324/540
ssh server and client: Dropbear
http://matt.ucc.asn.au/dropbear/dropbear.html
I
Very small memory footprint ssh server for embedded systems
Satisfies most needs. Both client and server!
Size: 110 KB, statically compiled with uClibc on i386.

(OpenSSH client and server: approx 1200 KB, dynamically
compiled with glibc on i386)
Useful to:
I
I
Get a remote console on the target device

Copy files to and from the target device (scp or rsync -e ssh).
An alternative to OpenSSH, used on desktop and server

systems.
325/540
Benefits of a web server interface
Many network enabled devices can just have a network interface

I
Examples: modems / routers, IP cameras, printers...
No need to develop drivers and applications for computers

connected to the device. No need to support multiple
operating systems!
Just need to develop static or dynamic HTML pages (possibly

with powerful client-side JavaScript).
Easy way of providing access to device information and
parameters.
Reduced hardware costs (no LCD, very little storage space

needed)
326/540
Web servers
I
BusyBox http server: http://busybox.net

I
Tiny: only adds 9 K to BusyBox (dynamically

linked with glibc on i386, with all features
enabled.)
Sufficient features for many devices with a web
interface, including CGI, http authentication
and script support (like PHP, with a separate
interpreter).
License: GPL
lighttpd: http://lighttpd.net
Low footprint server good at managing high
loads.
May be useful in embedded systems too
Other possibilities: Boa, thttpd, nginx, etc
327/540
Network utilities (1)
avahi is an implementation of Multicast DNS Service

Discovery, that allows programs to publish and discover
services on a local network
bind, a DNS server
iptables, the user space tools associated to the Linux firewall,

Netfilter
iw and wireless tools, the user space tools associated to

Wireless devices
netsnmp, implementation of the SNMP protocol
openntpd, implementation of the Network Time Protocol, for

clock synchronization
openssl, a toolkit for SSL and TLS connections
328/540
Network utilities (2)
pppd, implementation of the Point to Point Protocol, used

for dial-up connections
samba, implements the SMB and CIFS protocols, used by

Windows to share files and printers
coherence, a UPnP/DLNA implementation
vsftpd, proftpd, FTP servers
329/540
Tools for the target device: System

utilities
330/540
System utilities
dbus, an inter-application object-oriented communication bus
gpsd, a daemon to interpret and share GPS data
libraw1394, raw access to Firewire devices
libusb, a user space library for accessing USB devices without

writing an in-kernel driver
Utilities for kernel subsystems: i2c-tools for I2C, input-tools

for input, mtd-utils for MTD devices, usbutils for USB
devices
331/540
Tools for the target device: Language

interpreters
332/540
Language interpreters
Interpreters for the most common scripting languages are

available. Useful for
I
I
I
Application development
Web services development
Scripting
Languages supported
I
I
I
I
I
I
Lua
Python
Perl
Ruby
TCL
PHP
333/540
Tools for the target device: Audio,

video and multimedia
334/540
Audio, video and multimedia
GStreamer, a multimedia framework

I
I
Allows to decode/encode a wide variety of codecs.

Supports hardware encoders and decoders through plugins,
proprietary/specific plugins are often provided by SoC vendors.
alsa-lib, the user space tools associated to the ALSA sound

kernel subsystem
Directly using encoding and decoding libraries, if you decide

not to use GStreamer:
libavcodec, libogg, libtheora, libvpx, flac, libvorbis, libmad,
libsndfile, speex, etc.
335/540
Tools for the target device: Graphical

toolkits
336/540
Graphical toolkits: Low-level

solutions and layers
337/540
DirectFB
I
Low-level graphical library

I
I
I
I
I
Lines, rectangles, triangles drawing and filling

Blitting, flipping
Text drawing
Windows and transparency
Image loading and video display
But also handles input event handling: mice, keyboards,

joysticks, touchscreens, etc.
Provides accelerated graphic operations on a few hardware

platforms
Single-application by default, but multiple applications can

share the framebuffer thanks to fusion
License: LGPL 2.1
http://www.directfb.org
338/540
DirectFB: architecture
339/540
DirectFB: usage
Multimedia applications
I
For example the Disko framework, for set-top box related

applications
Simple graphical applications

I
I
Industrial control
Device control with limited number of widgets
Visualization applications
As a lower layer for higher-level graphical libraries
340/540
DirectFB: screenshot
341/540
X.org - KDrive
I
Stand-alone simplified version of the X server,

for embedded systems
I
I
Formerly know as Tiny-X

Kdrive is integrated in the official X.org server
Works on top of the Linux frame buffer, thanks

to the Xfbdev variant of the server
Real X server
I
Fully supports the X11 protocol: drawing, input

event handling, etc.
Allows to use any existing X11 application or
library
Actively developed and maintained.
X11 license
http://www.x.org
342/540
Kdrive: architecture
343/540
Kdrive: usage
Can be directly programmed using Xlib / XCB

I
I
Low-level graphic library

Probably doesnt make sense since DirectFB is a more
lightweight solution for an API of roughly the same level (no
widgets)
Or, usually used with a toolkit on top of it

I
I
I
I
Gtk
Qt
Enlightment Foundation Libraries
Others: Fltk, WxEmbedded, etc
344/540
Wayland
Intended to be a simpler replacement for X
Wayland is a protocol for a compositor to talk to

its clients as well as a C library implementation
of that protocol.
Weston: a minimal and fast reference

implementation of a Wayland compositor, and is
suitable for many embedded and mobile use
cases.
Still experimental. Porting Gtk and Qt to

Wayland is not complete yet, for example.
http://wayland.freedesktop.org/
345/540
Wayland: architecture
346/540
Graphical toolkits: High-level

solutions
347/540
Gtk
The famous toolkit, providing

widget-based high-level APIs to develop
graphical applications
Standard API in C, but bindings exist for

various languages: C++, Python, etc.
Works on top of X.org.
No windowing system, a lightweight

window manager needed to run several
applications. Possible solution: Matchbox.
License: LGPL
http://www.gtk.org
348/540
Gtk stack components
Glib, core infrastructure

I
I
Object-oriented infrastructure GObject

Event loop, threads, asynchronous queues, plug-ins, memory
allocation, I/O channels, string utilities, timers, date and time,
internationalization, simple XML parser, regular expressions
Data types: memory slices and chunks, linked lists, arrays,
trees, hash tables, etc.
Pango, internationalization of text handling
ATK, accessibility toolkit
Cairo, vector graphics library
Gtk+, the widget library itself
The Gtk stack is a complete framework to develop applications
349/540
Gtk examples (1)
Openmoko phone interface
350/540
Gtk examples (2)
Maemo tablet / phone interface

GTK is losing traction, however: Mer, the descendent of Maemo,
is now implemented in EFL (see next slides).
351/540
Qt (1)
The other famous toolkit, providing widget-based high-level

APIs to develop graphical applications
Implemented in C++
I
I
the C++ library is required on the target system

standard API in C++, but with bindings for other languages
Works either on top of

I
I
I
Framebuffer
X11
DirectFB back-end integrated in version 4.4, which allows to
take advantage of the acceleration provided by DirectFB
drivers
352/540
Qt (2)
Qt is more than just a graphical toolkit, it also offers a

complete development framework: data structures, threads,
network, databases, XML, etc.
See our presentation Qt for non graphical applications

presentation at ELCE 2011 (Thomas Petazzoni):
http://j.mp/W4PK85
Qt Embedded has an integrated windowing system, allowing

several applications to share the same screen
Very well documented
Since version 4.5, available under the LGPL, allowing

proprietary applications
353/540
Qts usage
Qt on the Netflix player by Roku

Qt on the Dash Express
navigation system
354/540
Other less frequent solutions
Enlightenment Foundation Libraries (EFL)

I
I
I
Very powerful. Supported by Samsung, Intel and Free.fr.

Work on top of X or Wayland.
http://www.enlightenment.org/p.php?p=about/efl
355/540
Tools for the target device:

Databases
356/540
Lightweight database - SQLite
http://www.sqlite.org
I
SQLite is a small C library that implements a self-contained,

embeddable, lightweight, zero-configuration SQL database
engine
The database engine of choice for embedded Linux systems
I
I
Can be used as a normal library

Can be directly embedded into a application, even a
proprietary one since SQLite is released in the public domain
357/540
Tools for the target device: Web

browsers
358/540
WebKit
http://webkit.org/
I
Web browser engine. Application framework that

can be used to develop web browsers.
License: portions in LGPL and others in BSD.

Proprietary applications allowed.
Used by many web browsers: Safari, iPhone and

Android default browsers ... Google Chrome now
uses a fork of its WebCore component). Used by
e-mail clients too to render HTML:
http://trac.webkit.org/wiki/
Applications%20using%20WebKit
Multiple graphical back-ends: Qt4, GTK, EFL...
You could use it to create your custom browser.
359/540
Example of components used in real

devices
360/540
Industrial applications
In many industrial applications, the system is only responsible

for monitoring and control a device
Such a system is usually relatively simple in terms of
components
I
I
I
I
Kernel
BusyBox
C library
Applications relying directly on the C library, sometimes using
the real-time capabilities of the Linux kernel
Sometimes a Web server for remote control, or another server
implementing a custom protocol
361/540
Digital Photo Frame: requirements
Example taken from a conference of Matt Porter, Embedded

Alley at ELC 2008
Hardware: ARM SoC with DSP, audio, 800x600 LCD,

MMC/SD, NAND, buttons, speakers
The photo frame must be able to
I
I
I
I
I
I
Display to the LCD

Detect SD card insertion, notify applications of the insertion so
that applications can build a catalog of the pictures on the SD
card
Modern 3D GUI with nice transitions
Navigation through buttons
Support audio playback (MP3, playlists, ID3 tag)
JPEG resizing and rotation
362/540
Digital Photo Frame: components (1)
Base system
I
I
I
Components present in virtually all embedded Linux systems

The U-Boot bootloader
Linux Kernel
I
I
I
I
Drivers for SD/MMC, framebuffer, sound, input devices
Busybox
Build system, in this case was OpenEmbedded
Components: u-boot, Linux, busybox
363/540
Event handling to detect SD card insertion

I
udev, that receives events from the kernel, creates device

nodes, and sends events to hal
hal, which maintains a database of available devices and
provides a dbus API
dbus to connect hal with the application. The application
subscribes to hal event through dbus and gets notified when
they are triggered
Components: udev, hal, dbus
364/540
JPEG display
I
I
I
libjpeg to decode the pictures

jpegtran to resize and rotate them
FIM (Fbi Improved) for dithering
MP3 support
I
I
I
I
libmad for playing

libid3 for ID3 tags reading
libm3u to support playlists
Used vendor-provided components to leverage the DSP to play
MP3
365/540
3D interface
I
I
Vincent, an open-source implementation of OpenGL ES

Clutter, higher-level API to develop 3D applications
Application itself
I
I
I
I
I
I
Manages media events

Uses the JPEG libraries to decode and render pictures
Receives Linux input events from buttons and draws
OpenGL-based UI developed with Clutter
Manage a user-defined configuration
Play the music with the MP3-related libraries
Display photo slideshow
366/540
System building
367/540
System building: goal and solutions
Goal
I
Integrate all the software components, both third-party and

in-house, into a working root filesystem
It involves the download, extraction, configuration, compilation
and installation of all components, and possibly fixing issues
and adapting configuration files
Several solutions
I
I
I
Manually
System building tools
Distributions or ready-made filesystems
368/540
System building: manually
Manually building a target system involves downloading,

configuring, compiling and installing all the components of the
system.
All the libraries and dependencies must be configured,

compiled and installed in the right order.
Sometimes, the build system used by libraries or applications

is not very cross-compile friendly, so some adaptations are
necessary.
There is no infrastructure to reproduce the build from scratch,

which might cause problems if one component needs to be
changed, if somebody else takes over the project, etc.
369/540
System building: manually (2)
Manual system building is not recommended for production

projects
However, using automated tools often requires the developer

to dig into specific issues
Having a basic understanding of how a system can be built
manually is therefore very useful to fix issues encountered with
automated tools
We will first study manual system building, and during a

practical lab, create a system using this method
Then, we will study the automated tools available, and use one
of them during a lab
370/540
System foundations
A basic root file system needs at least

I
A traditional directory hierarchy, with /bin, /etc, /lib,

/root, /usr/bin, /usr/lib, /usr/share, /usr/sbin,
/var, /sbin
A set of basic utilities, providing at least the init program, a
shell and other traditional Unix command line tools. This is
usually provided by Busybox
The C library and the related libraries (thread, math, etc.)
installed in /lib
A few configuration files, such as /etc/inittab, and
initialization scripts in /etc/init.d
On top of this foundation common to most embedded Linux

system, we can add third-party or in-house components
371/540
Target and build spaces
The system foundation, Busybox and C library, are the core of

the target root filesystem
However, when building other components, one must
distinguish two directories
I
The target space, which contains the target root filesystem,

everything that is needed for execution of the application
The build space, which will contain a lot more files than the
target space, since it is used to keep everything needed to
compile libraries and applications. So we must keep the
headers, documentation, and other configuration files
372/540
Build systems
Each open-source component comes with a mechanism to

configure, compile and install it
I
A basic Makefile
I
A build system based on the Autotools

I
As this is the most common build system, we will study it in

details
CMake, http://www.cmake.org/
I
Need to read the Makefile to understand how it works and

how to tweak it for cross-compilation
Newer and simpler than the autotools. Used by large projects

such as KDE or Second Life
Scons, http://www.scons.org/
Waf, http://code.google.com/p/waf/
Other manual build systems
373/540
Autotools and friends

I
A family of tools, which associated together form a complete

and extensible build system
I
autoconf is used to handle the configuration of the software

package
automake is used to generate the Makefiles needed to build
the software package
pkgconfig is used to ease compilation against already installed
shared libraries
libtool is used to handle the generation of shared libraries in a
system-independent way
Most of these tools are old and relatively complicated to use,

but they are used by a majority of free software packages
today. One must have a basic understanding of what they do
and how they work.
374/540
automake / autoconf / autoheader
375/540
automake / autoconf
Files written by the developer

I
The configure script and the Makefile.in files are

generated by autoconf and automake respectively.
I
I
configure.in describes the configuration options and the

checks done at configure time
Makefile.am describes how the software should be built
They should never be modified directly

They are usually shipped pre-generated in the software
package, because there are several versions of autoconf and
automake, and they are not completely compatible
The Makefile files are generated at configure time, before

compiling
I
They are never shipped in the software package.
376/540
Configuring and compiling: native case
The traditional steps to configure and compile an autotools

based package are
I
Configuration of the package

./configure
Compilation of the package
make
Installation of the package
make install
Additional arguments can be passed to the ./configure

script to adjust the component configuration.
Only the make install needs to be done as root if the

installation should take place system-wide
377/540
Configuring and compiling: cross case (1)

I
For cross-compilation, things are a little bit more complicated.
At least some of the environment variables AR, AS, LD, NM, CC,
GCC, CPP, CXX, STRIP, OBJCOPY must be defined to point to
the proper cross-compilation tools. The host tuple is also by
default used as prefix.
configure script arguments:
I
I
--host: mandatory but a bit confusing. Corresponds to the

target platform the code will run on. Example:
--host=arm-linux
--build: build system. Automatically detected.
--target is only for tools generating code.
It is recommended to pass the --prefix argument. It defines

from which location the software will run in the target
environment. Usually, /usr is fine.
378/540
Configuring and compiling: cross case (2)

I
If one simply runs make install, the software will be

installed in the directory passed as --prefix. For
cross-compiling, one must pass the DESTDIR argument to
specify where the software must be installed.
Making the distinction between the prefix (as passed with

--prefix at configure time) and the destination directory (as
passed with DESTDIR at installation time) is very important.
Example:
export PATH=/usr/local/arm-linux/bin:$PATH
export CC=arm-linux-gcc
export STRIP=arm-linux-strip
./configure --host=arm-linux --prefix=/usr
make
make DESTDIR=$HOME/work/rootfs install
379/540
Installation (1)
I
The autotools based software packages provide both a

install and install-strip make targets, used to install
the software, either stripped or unstripped.
For applications, the software is usually installed in

<prefix>/bin, with configuration files in <prefix>/etc
and data in <prefix>/share/<application>/
The case of libraries is a little more complicated:
I
I
I
I
In <prefix>/lib, the library itself (a .so.<version>), a few

symbolic links, and the libtool description file (a .la file)
The pkgconfig description file in <prefix>/lib/pkgconfig
Include files in <prefix>/include/
Sometimes a <libname>-config program in <prefix>/bin
Documentation in <prefix>/share/man or
<prefix>/share/doc/
380/540
Installation (2)
Contents of usr/lib after installation of libpng and zlib
I libpng libtool description files
./lib/libpng12.la
./lib/libpng.la -> libpng12.la
I libpng static version
./lib/libpng12.a
./lib/libpng.a -> libpng12.a
I libpng dynamic version
./lib/libpng.so.3.32.0
./lib/libpng12.so.0.32.0
./lib/libpng12.so.0 -> libpng12.so.0.32.0
./lib/libpng12.so -> libpng12.so.0.32.0
./lib/libpng.so -> libpng12.so
./lib/libpng.so.3 -> libpng.so.3.32.0
I libpng pkg-config description files
./lib/pkgconfig/libpng12.pc
./lib/pkgconfig/libpng.pc -> libpng12.pc
I zlib dynamic version
./lib/libz.so.1.2.3
./lib/libz.so -> libz.so.1.2.3
./lib/libz.so.1 -> libz.so.1.2.3
381/540
Installation in the build and target spaces
From all these files, everything except documentation is

necessary to build an application that relies on libpng.
I
However, only the library .so binaries in <prefix>/lib and

some symbolic links are needed to execute the application on
the target.
I
These files will go into the build space
Only these files will go in the target space
The build space must be kept in order to build other

applications or recompile existing applications.
382/540
pkg-config
I
I
I
I
pkg-config is a tool that allows to query a small database to

get information on how to compile programs that depend on
libraries
The database is made of .pc files, installed by default in
<prefix>/lib/pkgconfig/.
pkg-config is used by the configure script to get the
library configurations
It can also be used manually to compile an application:
arm-linux-gcc -o test test.c $(pkg-config -libs --cflags thelib)
By default, pkg-config looks in /usr/lib/pkgconfig for
the *.pc files, and assumes that the paths in these files are
correct.
PKG_CONFIG_PATH allows to set another location for the *.pc
files and PKG_CONFIG_SYSROOT_DIR to prepend a prefix to
the paths mentioned in the .pc files.
383/540
Lets find the libraries

I
When compiling an application or a library that relies on other

libraries, the build process by default looks in /usr/lib for
libraries and /usr/include for headers.
The first thing to do is to set the CFLAGS and LDFLAGS

environment variables:
export CFLAGS=-I/my/build/space/usr/include/
export LDFLAGS=-L/my/build/space/usr/lib
The libtool files (.la files) must be modified because they

include the absolute paths of the libraries:
- libdir=/usr/lib
+ libdir=/my/build/space/usr/lib
The PKG_CONFIG_PATH environment variable must be set to

the location of the .pc files and the
PKG_CONFIG_SYSROOT_DIR variable must be set to the build
space directory.
384/540
Practical lab - Manual cross-compiling
Manually cross-compiling
applications and libraries
Learning about common techniques

and issues.
385/540
System building tools: principle
Different tools are available to automate the process of

building a target system, including the kernel, and sometimes
the toolchain.
They automatically download, configure, compile and install

all the components in the right order, sometimes after
applying patches to fix cross-compiling issues.
They already contain a large number of packages, that should

fit your main requirements, and are easily extensible.
The build becomes reproducible, which allows to easily change

the configuration of some components, upgrade them, fix
bugs, etc.
386/540
Available system building tools

Large choice of tools
I
Buildroot, developed by the community

http://www.buildroot.net
PTXdist, developed by Pengutronix

http://pengutronix.de/software/ptxdist/
OpenWRT, originally a fork of Buildroot for wireless routers, now a

more generic project
http://www.openwrt.org
LTIB. Good support for Freescale boards, but small community

http://ltib.org/
OpenEmbedded, more flexible but also far more complicated

http://www.openembedded.org, its industrialized version Yocto
Project and vendor-specific derivatives such as Arago.
See our dedicated course and training materials:
http://free-electrons.com/training/yocto/.
Vendor specific tools (silicon vendor or embedded Linux vendor)
387/540
Buildroot (1)
I
Allows to build a toolchain, a root filesystem image with many

applications and libraries, a bootloader and a kernel image
I
Or any combination of the previous items
Supports building uClibc, glibc, eglibc and musl toolchains,

either built by Buildroot, or external
Over 1200+ applications or libraries integrated, from basic

utilities to more elaborate software stacks: X.org, GStreamer,
Qt, Gtk, WebKit, Python, PHP, etc.
Good for small to medium embedded systems, with a fixed set
of features
I
I
No support for generating packages (.deb or .ipk)

Needs complete rebuild for most configuration changes.
Active community, releases published every 3 months.
388/540
Buildroot (2)
Configuration takes place through

a *config interface similar to the
kernel
make menuconfig
Allows to define
I
I
I
I
I
Architecture and specific CPU

Toolchain configuration
Set of applications and libraries
to integrate
Filesystem images to generate
Kernel and bootloader
configuration
Build by just running

make
389/540
Buildroot: adding a new package (1)
A package allows to integrate a user application or library to

Buildroot
Each package has its own directory (such as
package/gqview). This directory contains:
I
A Config.in file (mandatory), describing the configuration

options for the package. At least one is needed to enable the
package. This file must be sourced from package/Config.in
A gqview.mk file (mandatory), describing how the package is
built.
Patches (optional). Each file of the form gqview-*.patch
will be applied as a patch.
390/540
Buildroot: adding a new package (2)
For a simple package with a single configuration option to

enable/disable it, the Config.in file looks like:
config BR2_PACKAGE_GQVIEW
bool "gqview"
depends on BR2_PACKAGE_LIBGTK2
help
GQview is an image viewer for Unix operating systems
http://prdownloads.sourceforge.net/gqview
I
It must be sourced from package/Config.in:
source "package/gqview/Config.in"
391/540
Buildroot: adding new package (3)

I
Create the gqview.mk file to describe the build steps
GQVIEW_VERSION = 2.1.5
GQVIEW_SOURCE = gqview-$(GQVIEW_VERSION).tar.gz
GQVIEW_SITE = http://prdownloads.sourceforge.net/gqview
GQVIEW_DEPENDENCIES = host-pkgconf libgtk2
GQVIEW_CONF_ENV = LIBS="-lm"
$(eval $(autotools-package))
I
The package directory and the prefix of all variables must be

identical to the suffix of the main configuration option
BR2_PACKAGE_GQVIEW
The autotools-package infrastructure knows how to build

autotools packages. A more generic generic-package
infrastructure is available for packages not using the autotools
as their build system.
392/540
OpenEmbedded / Yocto Project

I
The most versatile and powerful embedded Linux build system

I
I
A collection of recipes (.bb files)

A tool that processes the recipes: bitbake
Integrates 2000+ application and libraries, is highly

configurable, can generate binary packages to make the
system customizable, supports multiple versions/variants of
the same package, no need for full rebuild when the
configuration is changed.
Configuration takes place by editing various configuration files
Good for larger embedded Linux systems, or people looking

for more configurability and extensibility
Drawbacks: very steep learning curve, very long first build.
393/540
Distributions - Debian
Debian GNU/Linux, http://www.debian.org

I
Provides the easiest environment for quickly building

prototypes and developing applications. Countless
runtime and development packages available.
But probably too costly to maintain and

unnecessarily big for production systems.
Available on ARM (armel, armhf, arm64), MIPS

and PowerPC architectures
Software is compiled natively by default.
394/540
Distributions - Others
Fedora
I
http://fedoraproject.org/wiki/
Architectures/ARM
Supported on various recent ARM boards

(such as Beaglebone Black). Pidora
supports Raspberry Pi too.
Supports QEMU emulated ARM boards

too (Versatile Express board)
Shipping the same version as for desktops!
Ubuntu
I
Had some releases for ARM mobile

multimedia devices, but stopped at
version 12.04. Now focusing on ARM
servers only.
395/540
Embedded distributions
Distributions designed for specific types of devices

I
Android: http://www.android.com/
Googles distribution for phones and tablet PCs.
Except the Linux kernel, very different user space
than other Linux distributions. Very successful, lots
of applications available (many proprietary).
Angstr
om:
http://www.angstrom-distribution.org/
Targets PDAs and webpads (Siemens Simpad...)
Binaries available for arm little endian.
396/540
Application frameworks
Not real distributions you can download. Instead, they
implement middleware running on top of the Linux kernel
and allowing to develop applications.
I
Mer: http://merproject.org/
Fork from the Meego project.
Targeting mobile devices.
Supports x86, ARM and MIPS.
See http://en.wikipedia.org/wiki/Mer_
(software_distribution)
Tizen: https://www.tizen.org/
Targeting smartphones, tablets, netbooks, smart TVs
and In Vehicle Infotainment devices.
Supported by big phone manufacturers and operators
HTML5 base application framework.
Likely to compete against Android!
See http://en.wikipedia.org/wiki/Tizen
397/540
System emulator: QEMU
http://qemu.org
Fast processor emulator
I
Useful to develop root filesystems and applications when the

hardware is not available yet (binary compatible development
boards are also possible).
Emulates the processor and various peripherals (System

emulation): x86, ppc, arm, sparc, mips, m68k...
Many different ARM boards supported:

qemu-system-arm -M ?
QEMU also allows to just run binaries for other architectures

(User emulation). Used inside some building tools.
398/540
Practical lab - Buildroot
Rebuild the same system, this time

with Buildroot.
See how easier it gets!
399/540
Commercial embedded Linux

solutions
400/540
Commercial embedded Linux solutions
Caution: commercial doesnt mean proprietary!

I Vendors play fair with the GPL and do make their source code
available to their users, and most of the time, eventually to
the community.
I
As long as they distribute the sources to their users, the GPL

doesnt require vendors to share their sources with any third
party.
Graphical toolkits developed by the vendors are usually

proprietary, trying to make it easier to create and embedded
Linux systems.
Major players: Wind River, Montavista, TimeSys
401/540
Commercial solution strengths

I
Technical advantages
I
Well tested and supported

kernel and tool versions
Often supporting patches
not supported by the
mainline kernel yet
(example: real-time
patches)
Complete development tool

sets: kernels, toolchains,
utilities, binaries for
impressive lists of target
platforms
Integrated utilities for
automatic kernel image,
initramfs and filesystem
generation.
Graphical developments
tools
Development tools available

on multiple platforms: GNU
/ Linux, Solaris, Windows...
Support services
Useful if you dont have

your own support
resources
Long term support
commitment, even for
versions considered as
obsolete by the
community, but not by
your users!
402/540
Commercial or community solutions?

Commercial distributions and tool sets
I
Best if you dont have your own support resources and have a
sufficient budget
Help focusing on your primary job: making an embedded

device.
You can even subcontract driver development to the vendor
Community distributions and tools

I
Best if you are on a tight budget
Best if you are willing to build your own embedded Linux

expertise, investigate issues by yourselves, and train your own
support resources.
In any case, your products are based on Free Software!
403/540
Embedded Linux application development
Embedded Linux
application
development
Free Electrons
Embedded Linux
Developers
Free Electrons

404/540
Contents
I
I
Source management
I
I
Integrated development environments (IDEs)

Version control systems
Debugging and analysis tools

I
I
I
Developing applications on embedded Linux

Building your applications
Debuggers
Memory checkers
System analysis
Development environments
I
Developing on Windows
405/540
Developing applications on embedded

Linux
406/540
An embedded Linux system is just a normal Linux system,

with usually a smaller selection of components
In terms of application development, developing on embedded

Linux is exactly the same as developing on a desktop Linux
system
All existing skills can be re-used, without any particular

adaptation
All existing libraries, either third-party or in-house, can be
integrated into the embedded Linux system
Taking into account, of course, the limitation of the embedded

systems in terms of performance, storage and memory
407/540
Programming language
The default programming language for system-level

application in Linux is usually C
I
C++ can be used for larger applications

I
I
The C++ library must be added to the system

Some libraries, including Qt, are developed in C++ so they
need the C++ library on the system anyway
Scripting languages can also be useful for quick application

development, web applications or scripts
I
The C library is already present on your system, nothing to add
But they require an interpreter on the embedded system and

have usually higher memory consumption and slightly lower
performance
Languages: Python, Perl, Lua, Ada, Fortran, etc.
408/540
C library or higher-level libraries?
For many applications, the C library already provides a

relatively large set of features
I
file and device I/O, networking, threads and synchronization,

inter-process communication
Thoroughly described in the glibc manual, or in any Linux
system programming book
However, the API carries a lot of history and is not necessarily
easy to grasp for new comers
Therefore, using a higher level framework, such as Qt or the

Gtk stack, might be a good idea
I
These frameworks are not only graphical libraries, their core is

separate from the graphical part
But of course, these libraries have some memory and storage
footprint, in the order of a few megabytes
409/540
Building your applications
For simple applications that do not need to be really portable

or provide compile-time configuration options, a simple
Makefile will be sufficient
For more complicated applications, or if you want to be able
to run your application on a desktop Linux PC and on the
target device, using a build system is recommended
I
I
autotools is ancient, complicated but very widely used.

We recommend to invest in CMake instead: modern, simpler,
smaller but growing user base.
The QT library is a special case, since it comes with its own

build system for applications, called qmake.
410/540
Simple Makefile (1)

I
Case of an application that only uses the C library, contains

two source files and generates a single binary
CROSS_COMPILE?=arm-linuxCC=$(CROSS_COMPILE)gcc
OBJS=foo.o bar.o
all: foobar
foobar: $(OBJS)
$(CC) -o $@ $^
clean:
$(RM) -f foobar $(OBJS)
411/540
Simple Makefile (2)

I
Case of an application that uses the Glib and the GPS libraries
CROSS_COMPILE?=arm-linuxLIBS=libgps glib-2.0
OBJS=foo.o bar.o
CC=$(CROSS_COMPILE)gcc
CFLAGS=$(shell pkg-config --cflags $(LIBS))
LDFLAGS=$(shell pkg-config --libs $(LIBS))
all: foobar
foobar: $(OBJS)
$(CC) -o $@ $^ $(LDFLAGS)
clean:
$(RM) -f foobar $(OBJS)
412/540
Integrated Development
Environments (IDE)
413/540
KDevelop
http://kdevelop.org
I
A full featured IDE!
License: GPL
Supports many languages: Ada, C, C++,

Database, Java, Perl, PHP, Python, Ruby, Shell
Supports many kinds of projects: KDE, but also

GTK, Gnome, kernel drivers, embedded (Opie)...
Many features: editor, syntax highlighting, code

completion, compiler interface, debugger
interface, file manager, class browser...
Nice overview:
http://en.wikipedia.org/wiki/Kdevelop
414/540
KDevelop screenshot
Ruby debugger
415/540
Eclipse (1)
http://www.eclipse.org/
I
An extensible, plug-in based software

development kit, typically used for creating IDEs.
Supported by the Eclipse foundation, a

non-profit consortium of major software industry
vendors (IBM, Intel, Borland, Nokia, Wind
River, Zend, Computer Associates...).
Free Software license (Eclipse Public License).

Incompatible with the GPL.
Supported platforms: GNU/Linux, Unix,

Windows
Extremely popular: created a lot of attraction.
416/540
Eclipse (2)
Eclipse is actually a platform composed of many projects:

http://www.eclipse.org/projects/
I
Some projects are dedicated to integrating into Eclipse

features useful for embedded developers (cross-compilation,
remote development, remote debugging, etc.)
The platform is used by major embedded Linux software

vendors for their (proprietary) system development kits:
MontaVista DevRocket, TimeSys TimeStorm, Wind River
Workbench, Sysgo ELinOS.
Used by free software build systems and development

environments too, such as Yocto and Buildroot.
Eclipse is a huge project. It would require an entire training session!
417/540
Other popular solutions

I
Many embedded Linux developers

simply use Vim or Emacs. They
can integrate with debuggers,
source code browsers such as
cscope, offer syntax highlighting
and more.
Geany is an easy-to-use graphical

code editor.
CodeBlocks is also quite popular,

since its also available on the
Windows platform.
Vim
Emacs
All these editors are available in most

Linux distributions, simply install them
and try them out!
418/540
419/540
Real projects cant do without them

I
Allow multiple developers to contribute on the same project.

Each developer can see the latest changes from the others, or
choose to stick with older versions of some components.
Allow to keep track of changes, and revert them if needed.
Allow developers to have their own development branch

(branching)
Supposed to help developers resolving conflicts with different

branches (merging)
420/540
Traditional version control systems

Rely on a central repository. The most popular open-source ones:
I CVS - Concurrent Versions System
I
I
I
Still quite popular in enterprise contexts. Almost no longer

exists in the open-source community.
Should no longer be used for new projects
http:
//en.wikipedia.org/wiki/Concurrent_Versions_System
Subversion
I
I
I
Created as a replacement of CVS, removing many of its

limitations.
Commits on several files, proper renaming support, better
performance, etc.
The user interface is very similar to CVS
http://en.wikipedia.org/wiki/Subversion_(software)
421/540
Distributed source control systems (1)
No longer have a central repository

I
More adapted to the way the Free Software community

develops software and organizes
Allows each developer to have a full local history of the

project, to create local branches. Makes each developers work
easier.
People get working copies from other peoples working copies,

and exchange changes between themselves. Branching and
merging is made easier.
Make it easier for new developers to join, making their own

experiments without having to apply for repository access.
422/540
Distributed source control systems (2)
Git
I
I
I
Initially designed and developed by Linus Torvalds for Linux

kernel development
Extremely popular in the community, and used by more and
more projects (kernel, U-Boot, Barebox, uClibc, GNOME,
X.org, etc.)
Outstanding performance, in particular in big projects
http://en.wikipedia.org/wiki/Git_(software)
Mercurial
I
I
I
Another system, created with the same goals as Git.

Used by some big projects too
http://en.wikipedia.org/wiki/Mercurial
http://en.wikipedia.org/wiki/Version_control_systems#
Distributed_revision_control
423/540
Debuggers
424/540
GDB
The GNU Project Debugger
http://www.gnu.org/software/gdb/
I
The debugger on GNU/Linux, available for most

embedded architectures.
Supported languages: C, C++, Pascal,

Objective-C, Fortran, Ada...
Console interface (useful for remote debugging).
Graphical front-ends available.
Can be used to control the execution of a

program, set breakpoints or change internal
variables. You can also use it to see what a
program was doing when it crashed (by loading
its memory image, dumped into a core file).
See also http://en.wikipedia.org/wiki/Gdb

425/540
GDB crash course
A few useful GDB commands

I
break foobar
puts a breakpoint at the entry of function foobar()
break foobar.c:42
puts a breakpoint in foobar.c, line 42
print var or print task->files[0].fd
prints the variable var, or a more complicated reference. GDB
can also nicely display structures with all their members
continue
continue the execution
next
continue to the next line, stepping over function calls
step
continue to the next line, entering into subfunctions
426/540
GDB graphical front-ends
DDD - Data Display Debugger

http://www.gnu.org/software/ddd/
A popular graphical front-end, with advanced data plotting
capabilities.
GDB/Insight
http://sourceware.org/insight/
From the GDB maintainers.
KDbg
http://www.kdbg.org/
Another front-end, for the K Display Environment.
Integration with other IDEs: Eclipse, Emacs, KDevelop, etc.
427/540
Remote debugging
428/540
Remote debugging
I
In a non-embedded environment, debugging takes place using

gdb or one of its front-ends.
gdb has direct access to the binary and libraries compiled with
debugging symbols.
However, in an embedded context, the target platform

environment is often too limited to allow direct debugging
with gdb (2.4 MB on x86).
Remote debugging is preferred
gdb is used on the development workstation, offering all its

features.
gdbserver is used on the target system (only 100 KB on arm).
429/540
Remote debugging: architecture
430/540
Remote debugging: usage
On the target, run a program through gdbserver.

Program execution will not start immediately.
gdbserver localhost:<port> <executable> <args>
gdbserver /dev/ttyS0 <executable> <args>
Otherwise, attach gdbserver to an already running program:

gdbserver --attach localhost:<port> <pid>
Then, on the host, run the ARCH-linux-gdb program,

and use the following gdb commands:
I
To connect to the target:

gdb> target remote <ip-addr>:<port> (networking)
gdb> target remote /dev/ttyS0 (serial link)
To tell gdb where shared libraries are:
gdb> set sysroot <library-path> (without lib/)
431/540
Post mortem analysis

I
When an application crashes due to a segmentation fault and

the application was not under control of a debugger, we get
no information about the crash
Fortunately, Linux can generate a core file that contains the

image of the application memory at the moment of the crash,
and gdb can use this core file to let us analyze the state of the
crashed application
On the target
Use ulimit -c unlimited to enable the generation of a core

file when a crash occurs
On the host
I
After the crash, transfer the core file from the target to the
host, and run
ARCH-linux-gdb -c core-file application-binary
432/540
Memory checkers
433/540
DUMA
Detect Unintended Memory Access
http://duma.sourceforge.net/
I
Fork and replacement for Electric Fence
Stops your program on the exact instruction that overruns or

underruns a malloc() memory buffer.
GDB will then display the source-code line that causes the
bug.
Works by using the virtual-memory hardware to create a

red-zone at the border of each buffer - touch that, and your
program stops.
Works on any platform supported by Linux, whatever the CPU

(provided virtual memory support is available).
434/540
Valgrind (1)
http://valgrind.org/
I
GNU GPL Software suite for debugging and

profiling programs.
Supported platforms: Linux on x86, x86 64,

ppc32, ppc64 and arm (armv7 only: Cortex A8,
A9 and A5)
Can detect many memory management and

threading bugs.
Profiler: provides information helpful to speed

up your program and reduce its memory usage.
The most popular tool for this usage. Even used

by projects with hundreds of programmers.
435/540
Valgrind (2)
Can be used to run any program, without the

need to recompile it.
Example usage
valgrind --leak-check=yes ls -la
Works by adding its own instrumentation to your

code and then running in on its own virtual cpu
core.
Significantly slows down execution, but still fine
for testing!
More details on http://valgrind.org/info/

and http://valgrind.org/docs/manual/
coregrind_core.html#howworks
436/540
System analysis
437/540
strace
System call tracer
http://sourceforge.net/projects/strace/
I
Available on all GNU/Linux systems

Can be built by your cross-compiling toolchain generator.
Even easier: drop a ready-made static binary for your

architecture, just when you need it. See
http://git.free-electrons.com/users/michaelopdenacker/static-binaries/tree/strace
Allows to see what any of your processes is doing:

accessing files, allocating memory...
Often sufficient to find simple bugs.
Usage:
strace <command> (starting a new process)
strace -p <pid> (tracing an existing process)
See man strace for details.

438/540
strace example output
> strace cat Makele

execve("/bin/cat", ["cat", "Makele"], [/* 38 vars */]) = 0
brk(0) = 0x98b4000
access("/etc/ld.so.nohwcap", F_OK) = -1 ENOENT (No such le or directory)
mmap2(NULL, 8192, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb7f85000
access("/etc/ld.so.preload", R_OK) = -1 ENOENT (No such le or directory)
open("/etc/ld.so.cache", O_RDONLY) = 3
fstat64(3, {st_mode=S_IFREG|0644, st_size=111585, ...}) = 0
mmap2(NULL, 111585, PROT_READ, MAP_PRIVATE, 3, 0) = 0xb7f69000
close(3) = 0
access("/etc/ld.so.nohwcap", F_OK) = -1 ENOENT (No such le or directory)
open("/lib/tls/i686/cmov/libc.so.6", O_RDONLY) = 3
read(3, "\177ELF\1\1\1\0\0\0\0\0\0\0\0\0\3\0\3\0\1\0\0\0\320h\1\0004\0\0\0\344"..., 512) = 512
fstat64(3, {st_mode=S_IFREG|0755, st_size=1442180, ...}) = 0
mmap2(NULL, 1451632, PROT_READ|PROT_EXEC, MAP_PRIVATE|MAP_DENYWRITE, 3, 0) = 0xb7e06000
mprotect(0xb7f62000, 4096, PROT_NONE) = 0
mmap2(0xb7f63000, 12288, PROT_READ|PROT_WRITE,
MAP_PRIVATE|MAP_FIXED|MAP_DENYWRITE, 3, 0x15c) = 0xb7f63000
mmap2(0xb7f66000, 9840, PROT_READ|PROT_WRITE,
MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0) = 0xb7f66000
close(3) = 0
Hint: follow the open file descriptors returned by open().

This tells you what files system calls are run on.
439/540
ltrace
A tool to trace library calls used by a program and all the signals it
receives
I
Very useful complement to strace, which shows only system

calls.
Of course, works even if you dont have the sources
Allows to filter library calls with regular expressions, or just by

a list of function names.
Manual page: http://linux.die.net/man/1/ltrace
See http://en.wikipedia.org/wiki/Ltrace for details
440/540
ltrace example output
ltrace nedit index.html

sscanf(0x8274af1, 0x8132618, 0x8248640, 0xbfaadfe8, 0) = 1
sprintf("const 0", "const %d", 0) = 7
strcmp("startScan", "const 0") = 1
strcmp("ScanDistance", "const 0") = -1
strcmp("const 200", "const 0") = 1
strcmp("$list_dialog_button", "const 0") = -1
strcmp("$shell_cmd_status", "const 0") = -1
strcmp("$read_status", "const 0") = -1
strcmp("$search_end", "const 0") = -1
strcmp("$string_dialog_button", "const 0") = -1
strcmp("$rangeset_list", "const 0") = -1
strcmp("$calltip_ID", "const 0") = -1
441/540
ltrace summary
Example summary at the end of the ltrace output (-c option)
Process 17019 detached
% time
seconds usecs/call
calls
errors syscall
------ ----------- ----------- --------- --------- ---------------100.00
0.000050
50
1
set_thread_area
0.00
0.000000
0
48
read
0.00
0.000000
0
44
write
0.00
0.000000
0
80
63 open
0.00
0.000000
0
19
close
0.00
0.000000
0
1
execve
0.00
0.000000
0
2
2 access
0.00
0.000000
0
3
brk
0.00
0.000000
0
1
munmap
0.00
0.000000
0
1
uname
0.00
0.000000
0
1
mprotect
0.00
0.000000
0
19
mmap2
0.00
0.000000
0
50
46 stat64
0.00
0.000000
0
18
fstat64
------ ----------- ----------- --------- --------- ---------------100.00
0.000050 288 111 total
442/540
OProfile
http://oprofile.sourceforge.net
I
A system-wide profiling tool
Can collect statistics like the top users of the CPU.
Works without having the sources.
Requires a kernel patch to access all features, but is already

available in a standard kernel.
Requires more investigation to see how it works.
Ubuntu/Debian packages: oprofile, oprofile-gui
443/540
Callgrind / KCachegrind
Cachegrind / Callgrind: part of the Valgrind tool suite

Collects function call statistics and call graphs. Useful to
know in which functions most time is spent.
KCachegrind: http://kcachegrind.sourceforge.net/
An amazing visualizer for Cachegrind / Callgrind data.
KCachegrind can also import data from other profilers (such

as OProfile), and from profiling output from Python, Perl and
PHP.
444/540
KCachegrind screenshot
445/540
Practical lab - App. development and debugging
I
Compile your own application with

the ncurses library
Remote debugging
I
Set up remote debugging tools on

the target: strace, ltrace
and gdbserver.
Debug a simple application running

on the target using remote
debugging
446/540
Developing on Windows
447/540
Developing on Windows!?
Using a GNU/Linux workstation is the easiest way to create

software for GNU/Linux or embedded Linux
I
You use the same tools and environment as all community

developers do. Much fewer issues you are the only one to face.
You get familiar with the system. Essential for understanding

issues.
However, some developers have no choice: Windows is the only

desktop OS allowed in their company.
448/540
Cygwin
http://cygwin.com/
Linux (POSIX)-like environment for Windows
I
2 components:
Linux API emulation layer: cygwin1.dll
A collection of tools originally found in
GNU/Linux
Allows to compile and run many GNU/Linux

programs on Windows: shells, compiler, http
servers, X Window, GTK...
Very easy to install. Can choose which tools to

download and install.
For embedded Linux system developers: makes it

possible to use GNU toolchains (compiled for
Windows) required to build Linux binaries
(kernel, libraries or applications).
449/540
Cygwin limitations
Cygwin is not a complete substitute for a real GNU/Linux system.

I
Almost all developers work on GNU/Linux or on another Unix

platform (typically BSD). Dont expect them to test that their
tools build on Windows with Cygwin.
The number of Cygwin users is quite small.

You may be the first to face or report building issues on this
platform for a given compiler or tool version.
Cygwin is very slow.
So, the best solution is to run Linux inside Windows!
450/540
VirtualBox
http://virtualbox.org from Oracle

I
PC emulation solution available on Windows,

GNU/Linux, OS X and Solaris.
2 licenses:
I
Open Source Edition (OSE): GPL license. Most

features (except USB 2.0 support). Binaries
available for all platforms now.
Proprietary: free of cost for personal use and
evaluation. With USB 2.0 support and misc
other non-essential features.
Based on QEMUs core engine, very good

performance.
See http://en.wikipedia.org/wiki/VirtualBox
451/540
Real-time in embedded Linux systems
Real-time in
embedded Linux
systems
Free Electrons
Embedded Linux
Developers
Free Electrons

452/540
Introduction
453/540
Embedded Linux and real time
Due to its advantages, Linux and open-source software are

more and more commonly used in embedded applications
However, some applications also have real-time constraints

They, at the same time, want to
Get all the nice advantages of Linux: hardware support,

components re-use, low cost, etc.
Get their real-time constraints met
454/540
Embedded Linux and real time
I
I
Linux is an operating system part of the large Unix family

It was originally designed as a time-sharing system
I
The main goal was to get the best throughput from the
available hardware, by making the best possible usage of
resources (CPU, memory, I/O)
Time determinism was not taken into account
On the opposite, real-time constraints imply time

determinism, even at the expense of lower global throughput
Best throughput and time determinism are contradictory

requirements
455/540
Linux and real-time approaches (1)
Over time, two major approaches have been taken to bring

real-time requirements into Linux
Approach 1
I
Improve the Linux kernel itself so that it matches real-time

requirements, by providing bounded latencies, real-time APIs,
etc.
Approach taken by the mainline Linux kernel and the
PREEMPT RT project.
Approach 2
I
Add a layer below the Linux kernel that will handle all the
real-time requirements, so that the behaviour of Linux doesnt
affect real-time tasks.
Approach taken by RTLinux, RTAI and Xenomai
456/540
Linux and real-time approaches (2)
An alternative approach is to use specific hardware to run real-time

work on:
I
Dedicating a CPU core to a real-time OS or to a real-time

application, using some kind of hypervizor.
Running real-time work on an FPGA
Running real-time work on a dedicated microcontroller. For

example, the TI AM335x CPU (used in the Beaglebone Black)
has a Programmable Real-Time Unit and Industrial
Communication Subsystem (PRU-ICSS), which can be used
for real-time processing.
457/540
Improving the main Linux kernel with

PREEMPT RT
458/540
Understanding latency
I
When developing real-time applications with a system such as

Linux, the typical scenario is the following
I
An event from the physical world happens and gets notified to

the CPU by means of an interrupt
The interrupt handler recognizes and handles the event, and
then wake-up the user space task that will react to this event
Some time later, the user space task will run and be able to
react to the physical world event
Real-time is about providing guaranteed worst case latencies

for this reaction time, called latency
459/540
Linux kernel latency components
kernel latency = interrupt latency + handler duration + scheduler

latency + scheduler duration
460/540
Interrupt latency
461/540
Sources of interrupt latency

I
One of the concurrency prevention mechanism used in the

kernel is the spinlock
It has several variants, but one of the variant commonly used
to prevent concurrent accesses between a process context and
an interrupt context works by disabling interrupts
Critical sections protected by spinlocks, or other section in
which interrupts are explicitly disabled will delay the beginning
of the execution of the interrupt handler
I
The duration of these critical sections is unbounded
Other possible source: shared interrupts
462/540
Interrupt handler duration
463/540
Interrupt handler implementation

I
In Linux, many interrupt handlers are split in two parts

I
A top-half, started by the CPU as soon as interrupts are

enabled. It runs with the interrupt line disabled and is
supposed to complete as quickly as possible.
A bottom-half, scheduled by the top-half, which starts after all
pending top-halves have completed their execution.
Therefore, for real-time critical interrupts, bottom-halves

shouldnt be used: their execution is delayed by all other
interrupts in the system.
464/540
Scheduler latency
465/540
Understanding preemption (1)
The Linux kernel is a preemptive operating system
When a task runs in user space mode and gets interrupted by

an interruption, if the interrupt handler wakes up another
task, this task can be scheduled as soon as we return from the
interrupt handler.
466/540
Understanding preemption (2)
However, when the interrupt comes while the task is executing

a system call, this system call has to finish before another task
can be scheduled.
By default, the Linux kernel does not do kernel preemption.
This means that the time before which the scheduler will be
called to schedule another task is unbounded.
467/540
Scheduler duration
468/540
Other non-deterministic mechanisms
Outside of the critical path detailed previously, other

non-deterministic mechanisms of Linux can affect the
execution time of real-time tasks
Linux is highly based on virtual memory, as provided by an

MMU, so that memory is allocated on demand. Whenever an
application accesses code or data for the first time, it is
loaded on demand, which can creates huge delays.
Many C library services or kernel services are not designed

with real-time constraints in mind.
469/540
Priority inversion
A process with a low priority might hold a lock needed by a higher
priority process, effectively reducing the priority of this process.
Things can be even worse if a middle priority process uses the CPU.
470/540
Interrupt handler priority
In Linux, interrupt handlers are executed directly by the CPU

interrupt mechanisms, and not under control of the Linux
scheduler. Therefore, all interrupt handlers have a higher priority
than all tasks running on the system.
471/540
The PREEMPT RT project

I
Long-term project lead by Linux kernel developers Ingo

Molnar, Thomas Gleixner and Steven Rostedt
I
https://rt.wiki.kernel.org
The goal is to gradually improve the Linux kernel regarding

real-time requirements and to get these improvements merged
into the mainline kernel
PREEMPT RT development works very closely with the
mainline development
Many of the improvements designed, developed and debugged

inside PREEMPT RT over the years are now part of the
mainline Linux kernel
I
The project is a long-term branch of the Linux kernel that

ultimately should disappear as everything will have been
merged
472/540
Improvements in the mainline kernel
From the PREEMPT RT project

I Since the beginning of 2.6
I
I
I
O(1) scheduler
Kernel preemption
Better POSIX real-time
API support
Since 2.6.18
I
Priority inheritance
support for mutexes
Since 2.6.21
I
Since 2.6.30
I
High-resolution timers
Threaded interrupts
Since 2.6.33
I
Spinlock annotations
473/540
New preemption options in Linux 2.6
2 new preemption models offered by standard Linux 2.6:
474/540
1st option: no forced preemption
CONFIG_PREEMPT_NONE
Kernel code (interrupts, exceptions, system calls) never preempted.
Default behavior in standard kernels.
I
Best for systems making intense computations, on which

overall throughput is key.
Best to reduce task switching to maximize CPU and cache

usage (by reducing context switching).
Still benefits from some Linux 2.6 improvements: O(1)

scheduler, increased multiprocessor safety (work on RT
preemption was useful to identify hard to find SMP bugs).
Can also benefit from a lower timer frequency (100 Hz instead

of 250 or 1000).
475/540
2nd option: voluntary kernel preemption
CONFIG_PREEMPT_VOLUNTARY
Kernel code can preempt itself
I
Typically for desktop systems, for quicker application reaction

to user input.
Adds explicit rescheduling points throughout kernel code.
Minor impact on throughput.
Used in: Ubuntu Desktop 13.04, Ubuntu Server 12.04
476/540
3rd option: preemptible kernel
CONFIG_PREEMPT
Most kernel code can be involuntarily preempted at any time.
When a process becomes runnable, no more need to wait for kernel
code (typically a system call) to return before running the
scheduler.
I
Exception: kernel critical sections (holding spinlocks). In a

case you hold a spinlock on a uni-processor system, kernel
preemption could run another process, which would loop
forever if it tried to acquire the same spinlock.
Typically for desktop or embedded systems with latency

requirements in the milliseconds range.
Still a relatively minor impact on throughput.
477/540
Priority inheritance
One classical solution to the priority inversion problem is

called priority inheritance
I
The idea is that when a task of a low priority holds a lock

requested by a higher priority task, the priority of the first task
gets temporarily raised to the priority of the second task: it
has inherited its priority.
In Linux, since 2.6.18, mutexes support priority inheritance
In user space, priority inheritance must be explicitly enabled

on a per-mutex basis.
478/540
High resolution timers
The resolution of the timers used to be bound to the

resolution of the regular system tick
I
I
I
Usually 100 Hz or 250 Hz, depending on the architecture and

the configuration
A resolution of only 10 ms or 4 ms.
Increasing the regular system tick frequency is not an option as
it would consume too many resources
The high-resolution timers infrastructure, merged in 2.6.21,

allows to use the available hardware timers to program
interrupts at the right moment.
I
Hardware timers are multiplexed, so that a single hardware

timer is sufficient to handle a large number of
software-programmed timers.
Usable directly from user space using the usual timer APIs
479/540
Threaded interrupts
I
To solve the interrupt inversion problem, PREEMPT RT has

introduced the concept of threaded interrupts
The interrupt handlers run in normal kernel threads, so that

the priorities of the different interrupt handlers can be
configured
The real interrupt handler, as executed by the CPU, is only in

charge of masking the interrupt and waking-up the
corresponding thread
The idea of threaded interrupts also allows to use sleeping

spinlocks (see later)
Merged since 2.6.30, the conversion of interrupt handlers to

threaded interrupts is not automatic: drivers must be modified
In PREEMPT RT, all interrupt handlers are switched to

threaded interrupts
480/540
The future of the PREEMPT RT patchset
July 2014 status: the project is stalled because of lack of

funding. Thomas Gleixner is still the maintainer but lacks
time for further development and mainlining efforts. He must
work on other topics that pay the bills!
December 2014: the latest release is 3.14.10-rt7 (while

mainline is 3.18).
Contact (and contract) Thomas if you need further features

and support for newer kernels.
See http://lwn.net/Articles/604695/
481/540
PREEMPT RT specifics
482/540
CONFIG PREEMPT RT (1)
The PREEMPT RT patch adds a new level of preemption,

called CONFIG_PREEMPT_RT
This level of preemption replaces all kernel spinlocks by
mutexes (or so-called sleeping spinlocks)
I
Instead of providing mutual exclusion by disabling interrupts

and preemption, they are just normal locks: when contention
happens, the process is blocked and another one is selected by
the scheduler.
Works well with threaded interrupts, since threads can block,
while usual interrupt handlers could not.
Some core, carefully controlled, kernel spinlocks remain as
normal spinlocks.
483/540
CONFIG PREEMPT RT (2)
With CONFIG_PREEMPT_RT, virtually all kernel code becomes

preemptible
I
An interrupt can occur at any time, when returning from the

interrupt handler, the woken up process can start immediately.
This is the last big part of PREEMPT RT that isnt fully in

the mainline kernel yet
I
Part of it has been merged in 2.6.33: the spinlock annotations.

The spinlocks that must remain as spinning spinlocks are now
differentiated from spinlocks that can be converted to sleeping
spinlocks. This has reduced a lot the PREEMPT RT patch
size!
484/540
Threaded interrupts
The mechanism of threaded interrupts in PREEMPT RT is

still different from the one merged in mainline
In PREEMPT RT, all interrupt handlers are unconditionally

converted to threaded interrupts.
This is a temporary solution, until interesting drivers in

mainline get gradually converted to the new threaded
interrupt API that has been merged in 2.6.30.
485/540
Setting up PREEMPT RT
486/540
PREEMPT RT setup (1)
PREEMPT RT is delivered as a patch against the mainline

kernel
I
Best to have a board supported by the mainline kernel,

otherwise the PREEMPT RT patch may not apply and may
require some adaptations
At the moment, only even releases of the Linux kernel are

supported: 3.0, 3.2, 3.4, 3.6, 3.8, 3.10, 3.12, 3.14.
Quick set up
I
Download the latest PREEMPT RT patch

from http://www.kernel.org/pub/linux/kernel/projects/rt/
Download and extract the corresponding mainline kernel

version
Apply the patch to the mainline kernel tree
487/540
PREEMPT RT setup (2)
In the kernel configuration, be sure to enable

I
I
CONFIG_PREEMPT_RT
High-resolution timers
Compile your kernel, and boot
You are now running the real-time Linux kernel
Of course, some system configuration remains to be done, in

particular setting appropriate priorities to the interrupt
threads, which depend on your application.
488/540
Real-time application development

with PREEMPT RT
489/540
Development and compilation
No special library is needed, the POSIX real-time API is part

of the standard C library
The glibc or eglibc C libraries are recommended, as support
for some real-time features is not available in uClibc yet
I
Compile a program
I
Priority inheritance mutexes or NPTL on some architectures,

for example
ARCH-linux-gcc -o myprog myprog.c -lrt
To get the documentation of the POSIX API

I
I
Install the manpages-posix-dev package

Run man function-name
490/540
Process, thread?
I
I
Confusion about the terms process, thread and task

In Unix, a process is created using fork() and is composed of
I
I
I
An address space, which contains the program code, data,

stack, shared libraries, etc.
One thread, that starts executing the main() function.
Upon creation, a process contains one thread
Additional threads can be created inside an existing process,

using pthread_create()
I
They run in the same address space as the initial thread of the
process
They start executing a function passed as argument to
pthread_create()
491/540
Process, thread: kernel point of view
The kernel represents each thread running in the system by a

structure of type task_struct
From a scheduling point of view, it makes no difference

between the initial thread of a process and all additional
threads created dynamically using pthread_create()
492/540
Creating threads
I
Linux support the POSIX thread API
To create a new thread
pthread_create(pthread_t *thread, pthread_attr_t *attr,

void *(*routine)(void*), void *arg);
I
The new thread will run in the same address space, but will
be scheduled independently
Exiting from a thread
pthread_exit(void *value_ptr);
I
Waiting for the termination of a thread
pthread_join(pthread_t *thread, void **value_ptr);
493/540
Scheduling classes (1)
I
I
The Linux kernel scheduler support different scheduling classes

The default class, in which processes are started by default is
a time-sharing class
I
I
All processes, regardless of their priority, get some CPU time

The proportion of CPU time they get is dynamic and affected
by the nice value, which ranges from -20 (highest) to 19
(lowest). Can be set using the nice or renice commands
The real-time classes SCHED_FIFO and SCHED_RR

I
The highest priority process gets all the CPU time, until it
blocks.
In SCHED_RR, round-robin scheduling between the processes of
the same priority. All must block before lower priority
processes get CPU time.
Priorities ranging from 0 (lowest) to 99 (highest)
494/540
An existing program can be started in a specific scheduling

class with a specific priority using the chrt command line tool
I
Example: chrt -f 99 ./myprog
The sched_setscheduler() API can be used to change the

scheduling class and priority of a process
int sched_setscheduler(pid_t pid, int policy,

const struct sched_param *param);
I
I
policy can be SCHED_OTHER, SCHED_FIFO, SCHED_RR, etc.

param is a structure containing the priority
495/540

I
The priority can be set on a per-thread basis when a thread is

created
struct sched_param parm;

pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_attr_setinheritsched(&attr, PTHREAD_EXPLICIT_SCHED);
pthread_attr_setschedpolicy(&attr, SCHED_FIFO);
parm.sched_priority = 42;
pthread_attr_setschedparam(&attr, &parm);
I
Then the thread can be created using pthread_create(),

passing the attr structure.
Several other attributes can be defined this way: stack size,

etc.
496/540
Memory locking
In order to solve the non-determinism introduced by virtual

memory, memory can be locked
I
I
mlockall(MCL_CURRENT | MCL_FUTURE);
I
Guarantee that the system will keep it allocated

Guarantee that the system has pre-loaded everything into
memory
Locks all the memory of the current address space, for
currently mapped pages and pages mapped in the future
Other, less useful parts of the API: munlockall, mlock,

munlock.
Watch out for non-currently mapped pages
I
I
Stack pages
Dynamically-allocated memory
497/540
Mutexes
I
Allows mutual exclusion between two threads in the same

address space
Initialization/destruction
pthread_mutex_init(pthread_mutex_t *mutex,
const pthread_mutexattr_t *mutexattr);
pthread_mutex_destroy(pthread_mutex_t *mutex);
I
Lock/unlock
pthread_mutex_lock(pthread_mutex_t *mutex);
pthread_mutex_unlock(pthread_mutex_t *mutex);
I
Priority inheritance must be activated explicitly
pthread_mutexattr_t attr;
pthread_mutexattr_init (&attr);
pthread_mutexattr_setprotocol(&attr, PTHREAD_PRIO_INHERIT);
498/540
Timers
Timer creation
timer_create(clockid_t clockid, struct sigevent *evp,

timer_t *timerid);
I
clockid is usually CLOCK_MONOTONIC. sigevent defines

what happens upon timer expiration: send a signal or start a
function in a new thread. timerid is the returned timer
identifier.
Configure the timer for expiration at a given time
timer_settime(timer_t timerid, int flags,

struct itimerspec *newvalue,
struct itimerspec *oldvalue);
499/540
Timers (2)
Delete a timer
timer_delete(timer_t timerid)
I
Get the resolution of a clock, clock_getres
Other functions: timer_getoverrun(), timer_gettime()
500/540
Signals
I
I
Signals are asynchronous notification mechanisms

Notification occurs either
I
By the call of a signal handler. Be careful with the limitations

of signal handlers!
By being unblocked from the sigwait(), sigtimedwait() or
sigwaitinfo() functions. Usually better.
Signal behaviour can be configured using sigaction()
The mask of blocked signals can be changed with

pthread_sigmask()
Delivery of a signal using pthread_kill() or tgkill()
All signals between SIGRTMIN and SIGRTMAX, 32 signals under

Linux.
501/540
Inter-process communication
Semaphores
I
I
Message queues
I
I
Usable between different processes using named semaphores

sem_open(), sem_close(), sem_unlink(), sem_init(),
sem_destroy(), sem_wait(), sem_post(), etc.
Allows processes to exchange data in the form of messages.
mq_open(), mq_close(), mq_unlink(), mq_send(),
mq_receive(), etc.
Shared memory
I
Allows processes to communicate by sharing a segment of

memory
shm_open(), ftruncate(), mmap(), munmap(), close(),
shm_unlink()
502/540
Debugging latencies in
PREEMPT RT
503/540
ftrace - Kernel function tracer
Infrastructure that can be used for debugging or analyzing

latencies and performance issues in the kernel.
I
Very well documented in

Documentation/trace/ftrace.txt
Negligible overhead when tracing is not enabled at run-time.
Can be used to trace any kernel function!
504/540
Using ftrace
Tracing information available through the debugfs virtual fs

(CONFIG_DEBUG_FS in the Kernel Hacking section)
Mount this filesystem as follows:

mount -t debugfs nodev /sys/kernel/debug
When tracing is enabled (see the next slides), tracing

information is available in /sys/kernel/debug/tracing.
Check available tracers in

/sys/kernel/debug/tracing/available_tracers
505/540
Scheduling latency tracer

CONFIG_SCHED_TRACER (Kernel Hacking section)
I
Maximum recorded time between waking up a top priority task and

its scheduling on a CPU, expressed in us.
Check that wakeup is listed in

/sys/kernel/debug/tracing/available_tracers
To select, reset and enable this tracer:
echo wakeup > /sys/kernel/debug/tracing/current_tracer

echo 0 > /sys/kernel/debug/tracing/tracing_max_latency
echo 1 > /sys/kernel/debug/tracing/tracing_enabled
I
Let your system run, in particular real-time tasks.

Dummy example: chrt -f 5 sleep 1
Disable tracing:
echo 0 > /sys/kernel/debug/tracing/tracing_enabled

I
Read the maximum recorded latency and the corresponding trace:

cat /sys/kernel/debug/tracing/tracing_max_latency
506/540
Exemple of worst-case latencies
Tests done with cyclictest

I
x86 Geode 500 Mhz: 96 us
x86 Pentium Dual-core T4500 2.3 Ghz: 36 us
ARM Freescale i.MX35 533 Mhz: 120 us
ARM Marvell 88FR131 1200 Mhz: 54 us
See more results at https://www.osadl.org/QA-FarmRealtime.qa-farm-about.0.html
507/540
Useful reading
About real-time support in the standard Linux kernel

I
Inside the RT patch, Steven Rostedt, Red Hat, ELC 2013

http://elinux.org/images/b/ba/Elc2013_Rostedt.pdf
Video: http://j.mp/1apUtu6
The Real-Time Linux Wiki: http://rt.wiki.kernel.org

The Wiki Web for the CONFIG PREEMPT RT community,
and real-time Linux in general.
Contains nice and useful documents!
See also our books page.
508/540
Real-time extensions to the Linux

kernel
509/540
Linux real-time extensions

Three generations
I
RTLinux
RTAI
Xenomai
A common principle
I
Add an extra layer between

the hardware and the Linux
kernel, to manage real-time
tasks separately.
510/540
RTLinux
First real-time extension for Linux, created by Victor Yodaiken.

I
Nice, but the author filed a software patent covering the

addition of real-time support to general operating systems as
implemented in RTLinux!
Its Open Patent License drew many developers away and

frightened users. Community projects like RTAI and Xenomai
now attract most developers and users.
February, 2007: RTLinux rights sold to Wind River. Today, no

longer advertised by Wind River.
Project completely dead.
511/540
RTAI
http://www.rtai.org/ - Real-Time Application Interface for
Linux
I
Created in 1999, by Prof. Paolo Mantegazza (long time

contributor to RTLinux), Dipartimento di Ingegneria
Aerospaziale Politecnico di Milano (DIAPM).
Community project. Significant user base. Attracted

contributors frustrated by the RTLinux legal issues.
Only really actively maintained on x86
May offer slightly better latencies than Xenomai, at the

expense of a less maintainable and less portable code base
Since RTAI is not really maintained on ARM and other

embedded architectures, our presentation is focused on
Xenomai.
512/540
Xenomai project
http://www.xenomai.org/
I
Started in 2001 as a project aiming at emulating

traditional RTOS.
Initial goals: facilitate the porting of programs to

GNU / Linux.
Initially related to the RTAI project (as the RTAI

/ fusion branch), now independent.
Skins mimicking the APIs of traditional RTOS

such as VxWorks, pSOS+, and VRTXsa as well
as the POSIX API, and a native API.
Aims at working both as a co-kernel and on top

of PREEMPT RT in future upstream Linux
versions.
Will never be merged in the mainline kernel.
513/540
Xenomai architecture
514/540
The Adeos interrupt pipeline abstraction

I
I
From the Adeos point of view, guest OSes are prioritized

domains.
For each event (interrupts, exceptions, syscalls, etc...), the
various domains may handle the event or pass it down the
pipeline.
515/540
Xenomai features
Factored real-time core with skins implementing various

real-time APIs
Seamless support for hard real-time in user space
No second-class citizen, all ports are equivalent feature-wise
Xenomai support is as much as possible independent from the

Linux kernel version (backward and forward compatible when
reasonable)
Each Xenomai branch has a stable user/kernel ABI
Timer system based on hardware high-resolution timers
Per-skin time base which may be periodic
RTDM skin allowing to write real-time drivers
516/540
Xenomai user space real-time support
Xenomai supports real-time in user space on 5 architectures,

including 32 and 64 bits variants.
Two modes are defined for a thread
I
the primary mode, where the thread is handled by the Xenomai

scheduler
the secondary mode, when it is handled by the Linux scheduler.
Thanks to the services of the Adeos I-pipe service, Xenomai

system calls are defined.
I
A thread migrates from secondary mode to primary mode

when such a system call is issued
It migrates from primary mode to secondary mode when a
Linux system call is issued, or to handle gracefully exceptional
events such as exceptions or Linux signals.
517/540
Life of a Xenomai application

I
Xenomai applications are started like normal Linux processes,

they are initially handled by the Linux scheduler and have
access to all Linux services
After their initialization, they declare themselves as real-time
applications, which migrates them to primary mode. In this
mode:
I
They are scheduled directly by the Xenomai scheduler, so they

have the real-time properties offered by Xenomai
They dont have access to any Linux service, otherwise they
get migrated back to secondary mode and loose all real-time
properties
They can only use device drivers that are implemented in
Xenomai, not the ones of the Linux kernel
Need to implement device drivers in Xenomai, and to split

real-time and non real-time parts of your applications.
518/540
Real Time Driver Model (RTDM)

I
An approach to unify the interfaces for developing device

drivers and associated applications under real-time Linux
I
Allows to develop in kernel space:

I
I
An API very similar to the native Linux kernel driver API

Character-style device drivers
Network-style device drivers
See the whitepaper on

http://www.xenomai.org/documentation/xenomai2.6/pdf/RTDM-and-Applications.pdf
Current notable RTDM based drivers:
I
I
I
I
Serial port controllers;

RTnet UDP/IP stack;
RT socket CAN, drivers for CAN controllers;
Analogy, fork of the Comedi project, drivers for acquisition
cards.
519/540
Setting up Xenomai
520/540
How to build Xenomai
Download Xenomai sources at

http://download.gna.org/xenomai/stable/
Download one of the Linux versions supported by this release

(see ksrc/arch/<arch>/patches/)
Since version 2.0, split kernel/user building model.
Kernel uses a script called script/prepare-kernel.sh

which integrates Xenomai kernel-space support in the Linux
sources.
Run the kernel configuration menu.
521/540
Linux options for Xenomai configuration
522/540
Xenomai user space support
User space libraries are compiled using the traditional

autotools
I
./configure --host=arm-linux && make && make

DESTDIR=/your/rootfs/ install
Xenomai installs pkg-config files which helps you to compile

your own programs against the Xenomai libraries.
See Xenomais examples directory.
Installation details may be found in the README.INSTALL

guide.
523/540
Developing applications on Xenomai
524/540
The POSIX skin

I
The POSIX skin allows to recompile without changes a

traditional POSIX application so that instead of using Linux
real-time services, it uses Xenomai services
I
http://www.xenomai.org/index.php/Porting_POSIX_
applications_to_Xenomai
Clocks and timers, condition variables, message queues,
mutexes, semaphores, shared memory, signals, thread
management
Good for existing code or programmers familiar with the
POSIX API
Of course, if the application uses any Linux service that isnt

available in Xenomai, it will switch back to secondary mode
To link an application against the POSIX skin
CFL=`pkg-config --cflags libxenomai_posix`

LDF=`pkg-config --libs libxenomai_posix`
ARCH-gcc $CFL -o rttest rttest.c $LDF
525/540
Communication with a normal task

I
If a Xenomai real-time application using the POSIX skin

wishes to communicate with a separate non-real-time
application, it must use the rtipc mechanism
In the Xenomai application, create an IPCPROTO_XDDP socket
socket(AF_RTIPC, SOCK_DGRAM, IPCPROTO_XDDP);

setsockopt(s, SOL_RTIPC, XDDP_SETLOCALPOOL,
&poolsz, sizeof(poolsz));
memset(&saddr, 0, sizeof(saddr));
saddr.sipc_family = AF_RTIPC;
saddr.sipc_port = PORTX;
ret = bind(s, (struct sockaddr *)&saddr, sizeof(saddr));
I
And then the normal socket API sendto() / recvfrom()
In the Linux application

I
I
Open /dev/rtpPORTX, where PORTX is the XDDP port

Use read() and write()
526/540
The native API (1)

I
A Xenomai specific API for developing real-time tasks

I
http://xenomai.org/documentation/branches/v2.4.x/
pdf/Native-API-Tour-rev-C.pdf
Usable both in user space and kernel space. Development of
tasks in user space is the preferred way.
More coherent and more flexible API than the POSIX API.
Easier to learn and understand. Certainly the way to go for
new applications.
Applications should include <native/service.h>, where

service can be alarm, buffer, cond, event, heap, intr,
misc, mutex, pipe, queue, sem, task, timer
To compile applications:
CFL=`pkg-config --cflags libxenomai_native`

LDF=`pkg-config --libs libxenomai_native`
ARCH-gcc $CFL -o rttest rttest.c $LDF
527/540
The native API (2)

I
Task management services

I
Counting semaphore services

I
rt_sem_create(), rt_sem_delete(), rt_sem_p(),

rt_sem_v(), etc.
Message queue services

I
rt_task_create(), rt_task_start(),
rt_task_suspend(), rt_task_resume(),
rt_task_delete(), rt_task_join(), etc.
rt_queue_create(), rt_queue_delete(),
rt_queue_alloc(), rt_queue_free(), rt_queue_send(),
rt_queue_receive(), etc.
Mutex services
I
rt_mutex_create(), rt_mutex_delete(),
rt_mutex_acquire(), rt_mutex_release(), etc.
528/540
The native API (3)
Alarm services
I
Memory heap services

I
rt_alarm_create(), rt_alarm_delete(),
rt_alarm_start(), rt_alarm_stop(), rt_alarm_wait(),
etc.
Allows to share memory between processes and/or to
pre-allocate a pool of memory
rt_heap_create(), rt_heap_delete(), rt_heap_alloc(),
rt_heap_bind()
Condition variable services

I
rt_cond_create(), rt_cond_delete(),
rt_cond_signal(), rt_cond_broadcast(),
rt_cond_wait(), etc.
529/540
Xenomai and normal task communication
I
I
Using rt pipes
In the native Xenomai application, use the Pipe API
I
rt_pipe_create(), rt_pipe_delete(),
rt_pipe_receive(), rt_pipe_send(), rt_pipe_alloc(),
rt_pipe_free()
In the normal Linux application

I
Open the corresponding /dev/rtpX file, the minor is specified

at rt_pipe_create() time
Then, just read() and write() to the opened file
530/540
Xenomai worst-case latencies results
ARM OMAP5432 1.1 GHz: 24 us
ARM OMAP4430 1 GHz: 23 us
ARM OMAP3530 720 MHz: 44 us
ARM SAMA5D3 528 MHz: 51 us
ARM AT91RM9200 180 MHz: 181 us
x86 Atom 1.6 GHz: 35 us
Geode LX800 processor at 500 MHz: 55 us
See results at
http://xenomai.org/~gch/core-3.14-latencies/
531/540
Books
Building Embedded Linux Systems, OReilly

By Karim Yaghmour, Jon Masters,
Gilad Ben-Yossef, Philippe Gerum and others
(including Michael Opdenacker), August 2008
A nice coverage of Xenomai (Philippe Gerum)

and the RT patch (Steven Rostedt)
http://oreilly.com/catalog/9780596529680/
532/540
Organizations
http://www.osadl.org
Open Source Automation Development Lab (OSADL)
Targets machine and plant control systems. Most member
companies are German (Thomas Gleixner is on board).
Supports the use of PREEMPT RT and Xenomai and
contributes to these projects. Shares useful documentation
and resources.
They also organize a yearly Real Time Linux Workshop.
533/540
Practical lab - Real-time - Scheduling latency
Check clock accuracy.
Start processes with real-time

priority.
Build a real-time application

against the standard POSIX
real-time API, and against
Xenomais POSIX skin
Compare scheduling latency on

your system, between a standard
kernel and a kernel with Xenomai.
534/540
References
Free Electrons
References
Embedded Linux
Developers
Free Electrons

535/540
Books
I
Embedded Linux Primer, Second Edition,

Prentice Hall
By Christopher Hallinan, October 2010
Covers a very wide range of interesting topics.
http://j.mp/17NYxBP
Building Embedded Linux Systems, OReilly

By Karim Yaghmour, Jon Masters, Gilad Ben-Yossef,
Philippe Gerum and others (including Michael
Opdenacker), August 2008
http://oreilly.com/catalog/9780596529680/
Embedded Linux System Design and

Development
P. Raghavan, A. Lad, S. Neelakandan, Auerbach,
Dec. 2005. Very good coverage of the POSIX
programming API (still up to date).
http://j.mp/19X8iu2
536/540
Web sites
ELinux.org, http://elinux.org, a Wiki entirely dedicated

to embedded Linux. Lots of topics covered: real-time,
filesystem, multimedia, tools, hardware platforms, etc.
Interesting to explore to discover new things.
LWN, http://lwn.net, very interesting news site about

Linux in general, and specifically about the kernel. Weekly
newsletter, available for free after one week for non-paying
visitors.
Linux Gizmos, http://linuxgizmos.com, a news site

about embedded Linux, mainly oriented on hardware
platforms related news.
537/540
International conferences
Useful conferences featuring embedded Linux and kernel topics
I
Embedded Linux Conference:

http://embeddedlinuxconference.com/
Organized by the Linux Foundation: California (San
Francisco, Spring), in Europe (Fall). Very interesting kernel
and user space topics for embedded systems developers.
Presentation slides freely available
Linux Plumbers, http://linuxplumbersconf.org

Conference on the low-level plumbing of Linux: kernel, audio,
power management, device management, multimedia, etc.
FOSDEM: http://fosdem.org (Brussels, February)

For developers. Presentations about system development.
Dont miss our free conference videos on http://freeelectrons.com/community/videos/conferences/!
538/540
Last slides
Free Electrons
Last slides
Embedded Linux
Developers
Free Electrons

539/540
Last slide
Thank you!
And may the Source be with you
540/540

Embedded Linux Slides

Uploaded by

Copyright:

Available Formats

Embedded Linux Slides

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Embedded Linux Slides

Uploaded by

Copyright:

Available Formats

Embedded Linux system development

Embedded Linux system

Copyright 2004-2015, Free Electrons.

Copyright 2004-2015, Free Electrons

License: Creative Commons Attribution - Share Alike 3.0

Hyperlinks in the document

Kernel documentation links:

Links to kernel source files and directories:

Links to the declarations, definitions and instances of kernel

Free Electrons at a glance

Engineering company created in 2004

Locations: Orange, Toulouse, Lyon (France)

Serving customers all around the world

Focus: Embedded Linux, Linux kernel, Android Free Software

Activities: development, training, consulting, technical

Added value: get the best of the user and development

Free Electrons on-line resources

All our training materials:

Generic course information

Copyright 2004-2015, Free Electrons.

Hardware used in this training session

JTAG, buttons, LEDs

(V.A.T. not included)

Board and CPU documentation, design files, software:

Shopping list: hardware for this course

I Atmel SAMA5D3 Xplained board - Available from

Atmel and multiple distributors (Mouser, Digikey...)

Dont hesitate to ask questions. Other people in the audience

This helps the trainer to detect any explanation that wasnt

Dont hesitate to share your experience, for example to

Your point of view is most valuable, because it can be similar

Your participation can make our session more interactive and

Practical lab guidelines

We cannot support more than 8 workstations at once (each

So, if you are more than 8 participants, please form up to 8

Open the electronic copy of your lecture materials, and use it

Dont copy and paste from the PDF slides.

As in the Free Software and Open Source community, cooperation

If you complete your labs before other people, dont hesitate

Explain what you understood to other participants when

Dont hesitate to report potential bugs to your instructor.

Dont hesitate to look for solutions on the Internet as well.

Command memento sheet

This memento sheet gives

It saves us 1 day of UNIX / Linux

Our best tip: in the command line

Get an electronic copy on

The vi editor is very useful to

Though not very user friendly at

Get an electronic copy on

You can also take the quick tutorial

Introduction to Embedded Linux

Copyright 2004-2015, Free Electrons.

Birth of free software

1983, Richard Stallman, GNU project and the free software

1991, Linus Torvalds, Linux kernel project, a Unix-like

1995, Linux is more and more popular on server systems

2000, Linux is more and more popular on embedded

2008, Linux is more and more popular on mobile devices

2010, Linux is more and more popular on phones