Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Embedded Linux Labs

Download as pdf or txt
Download as pdf or txt
You are on page 1of 57

Embedded Linux System Development

Microchip SAMA5D3 variant

Practical Labs

https://bootlin.com

January 30, 2020


Embedded Linux System Development

About this document


Updates to this document can be found on https://bootlin.com/doc/training/embedded-
linux.
This document was generated from LaTeX sources found on https://github.com/bootlin/
training-materials.
More details about our training sessions can be found on https://bootlin.com/training.

Copying this document


© 2004-2020, Bootlin, https://bootlin.com.
This document is released under the terms of the Creative Commons CC BY-SA
3.0 license . This means that you are free to download, distribute and even modify
it, under certain conditions.
Corrections, suggestions, contributions and translations are welcome!

2 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Training setup
Download files and directories used in practical labs

Install lab data


For the different labs in this course, your instructor has prepared a set of data (kernel images,
kernel configurations, root filesystems and more). Download and extract its tarball from a
terminal:
cd
wget https://bootlin.com/doc/training/embedded-linux/embedded-linux-labs.tar.xz
tar xvf embedded-linux-labs.tar.xz

Lab data are now available in an embedded-linux-labs directory in your home directory. This
directory contains directories and files used in the various practical labs. It will also be used as
working space, in particular to keep generated files separate when needed.
You are now ready to start the real practical labs!

Install extra packages


Feel free to install other packages you may need for your development environment. In particular,
we recommend to install your favorite text editor and configure it to your taste. The favorite
text editors of embedded Linux developers are of course Vim and Emacs, but there are also
plenty of other possibilities, such as GEdit, Qt Creator, CodeBlocks, Geany, etc.
It is worth mentioning that by default, Ubuntu comes with a very limited version of the vi
editor. So if you would like to use vi, we recommend to use the more featureful version by
installing the vim package.

More guidelines
Can be useful throughout any of the labs
• Read instructions and tips carefully. Lots of people make mistakes or waste time because
they missed an explanation or a guideline.
• Always read error messages carefully, in particular the first one which is issued. Some
people stumble on very simple errors just because they specified a wrong file path and
didn’t pay enough attention to the corresponding error message.
• Never stay stuck with a strange problem more than 5 minutes. Show your problem to
your colleagues or to the instructor.
• You should only use the root user for operations that require super-user privileges, such
as: mounting a file system, loading a kernel module, changing file ownership, configuring
the network. Most regular tasks (such as downloading, extracting sources, compiling...)
can be done as a regular user.

© 2004-2020 Bootlin, CC BY-SA license 3


Embedded Linux System Development

• If you ran commands from a root shell by mistake, your regular user may no longer be
able to handle the corresponding generated files. In this case, use the chown -R command
to give the new files back to your regular user.
Example: chown -R myuser.myuser linux/

4 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Building a cross-compiling toolchain


Objective: Learn how to compile your own cross-compiling toolchain
for the uClibc C library

After this lab, you will be able to:


• Configure the crosstool-ng tool
• Execute crosstool-ng and build up your own cross-compiling toolchain

Setup
Go to the $HOME/embedded-linux-labs/toolchain directory.

Install needed packages


Install the packages needed for this lab:
sudo apt install build-essential git autoconf bison flex \
texinfo help2man gawk libtool-bin libncurses5-dev

Getting Crosstool-ng
Let’s download the sources of Crosstool-ng, through its git source repository, and switch to a
commit that we have tested:
git clone https://github.com/crosstool-ng/crosstool-ng.git
cd crosstool-ng/
git checkout eb65ba65

Installing Crosstool-ng
We can either install Crosstool-ng globally on the system, or keep it locally in its download
directory. We’ll choose the latter solution. As documented at http://crosstool-ng.github.
io/docs/install/#hackers-way, do:
./bootstrap
./configure --enable-local
make
Then you can get Crosstool-ng help by running
./ct-ng help

Configure the toolchain to produce


A single installation of Crosstool-ng allows to produce as many toolchains as you want, for
different architectures, with different C libraries and different versions of the various components.

© 2004-2020 Bootlin, CC BY-SA license 5


Embedded Linux System Development

Crosstool-ng comes with a set of ready-made configuration files for various typical setups:
Crosstool-ng calls them samples. They can be listed by using ./ct-ng list-samples.
We will start with the arm-cortexa5-linux-uclibcgnueabihf sample. It can be loaded by
issuing:
./ct-ng arm-cortexa5-linux-uclibcgnueabihf
Then, to refine the configuration, let’s run the menuconfig interface:
./ct-ng menuconfig
In Path and misc options:
• Change Maximum log level to see to DEBUG so that we can have more details on what
happened during the build in case something went wrong.
In Toolchain options:
• Set Tuple's vendor string to training.
• Set Tuple's alias to arm-linux. This way, we will be able to use the compiler as arm-
linux-gcc instead of arm-training-linux-uclibcgnueabihf-gcc, which is much longer
to type.
In C-library:
• Enable IPv6 support. That’s because of Buildroot (which we will use later), which doesn’t
accept to use toolchains without IPv6 support.
In Debug facilities, disable every option. Some of these options will be useful in a real
toolchain, but in our labs, we will do debugging work with another toolchain anyway. Hence,
not compiling debugging features here will reduce toolchain building time.
Explore the different other available options by traveling through the menus and looking at the
help for some of the options. Don’t hesitate to ask your trainer for details on the available
options. However, remember that we tested the labs with the configuration described above.
You might waste time with unexpected issues if you customize the toolchain configuration.

Produce the toolchain


Nothing is simpler:
./ct-ng build
The toolchain will be installed by default in $HOME/x-tools/. That’s something you could have
changed in Crosstool-ng’s configuration.
And wait!

Known issues

Source archives not found on the Internet

It is frequent that Crosstool-ng aborts because it can’t find a source archive on the Internet,
when such an archive has moved or has been replaced by more recent versions. New Crosstool-ng
versions ship with updated URLs, but in the meantime, you need work-arounds.

6 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

If this happens to you, what you can do is look for the source archive by yourself on the Internet,
and copy such an archive to the src directory in your home directory. Note that even source
archives compressed in a different way (for example, ending with .gz instead of .bz2) will be
fine too. Then, all you have to do is run ./ct-ng build again, and it will use the source archive
that you downloaded.

Testing the toolchain


You can now test your toolchain by adding $HOME/x-tools/arm-training-linux-uclibcgnueabihf/
bin/ to your PATH environment variable and compiling the simple hello.c program in your main
lab directory with arm-linux-gcc.
You can use the file command on your binary to make sure it has correctly been compiled for
the ARM architecture.

Cleaning up
Do this only if you have limited storage space. In case you made a mistake in the toolchain
configuration, you may need to run Crosstool-ng again. Keeping generated files would save a
significant amount of time.
To save about 7 GB of storage space, do a ./ct-ng clean in the Crosstool-NG source direc-
tory. This will remove the source code of the different toolchain components, as well as all the
generated files that are now useless since the toolchain has been installed in $HOME/x-tools.

© 2004-2020 Bootlin, CC BY-SA license 7


Embedded Linux System Development

Bootloader - U-Boot
Objectives: Set up serial communication, compile and install the
U-Boot bootloader, use basic U-Boot commands, set up TFTP com-
munication with the development workstation.

As the bootloader is the first piece of software executed by a hardware platform, the installation
procedure of the bootloader is very specific to the hardware platform. There are usually two
cases:

• The processor offers nothing to ease the installation of the bootloader, in which case the
JTAG has to be used to initialize flash storage and write the bootloader code to flash.
Detailed knowledge of the hardware is of course required to perform these operations.

• The processor offers a monitor, implemented in ROM, and through which access to the
memories is made easier.

The Xplained board, which uses the SAMA5D3 SoCs, falls into the second category. The
monitor integrated in the ROM reads the MMC/SD card to search for a valid bootloader before
looking at the internal NAND flash for a bootloader. In case nothing is available, it will operate
in a fallback mode, that will allow to use an external tool to reflash some bootloader through
USB. Therefore, either by using an MMC/SD card or that fallback mode, we can start up a
SAMA5D3-based board without having anything installed on it.

Downloading Microchip’s flashing tool

Go to the ~/embedded-linux-labs/bootloader directory.

We’re going to use that fallback mode, and its associated tool, sam-ba.

We first need to download this tool, from Microchip’s website1 .

wget https://ww1.microchip.com/downloads/en/DeviceDoc/sam-ba_2.15.zip
unzip sam-ba_2.15.zip

Setting up serial communication with the board

Plug the USB-to-serial cable on the Xplained board. The blue end of the cable is going to GND
on J23, red on RXD and green on TXD. When plugged in your computer, a serial port should
appear, /dev/ttyUSB0.

1 In case this website is down, you can also find this tool on https://bootlin.com/labs/tools/.

8 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

You can also see this device appear by looking at the output of dmesg.

To communicate with the board through the serial port, install a serial communication program,
such as picocom:

sudo apt install picocom

You also need to make your user belong to the dialout group to be allowed to write to the serial
console:

sudo adduser $USER dialout

Important: for the group change to be effective, in Ubuntu 18.04, you have to completely reboot
the system 2 . A workaround is to run newgrp dialout, but it is not global. You have to run it
in each terminal.

Run picocom -b 115200 /dev/ttyUSB0, to start serial communication on /dev/ttyUSB0, with a


baudrate of 115200.

You can now power-up the board by connecting the micro-USB cable to the board, and to your
PC at the other end. If a system was previously installed on the board, you should be able to
interact with it through the serial line.

If you wish to exit picocom, press [Ctrl][a] followed by [Ctrl][x].

AT91Bootstrap Setup

The boot process is done in two steps with the ROM monitor trying to execute a first piece of
software, called AT91Bootstrap, from its internal SRAM, that will initialize the DRAM, load
U-Boot that will in turn load Linux and execute it.

As far as bootloaders are concerned, the layout of the NAND flash will look like:

2 As explained on https://askubuntu.com/questions/1045993/after-adding-a-group-logoutlogin-is-not-

enough-in-18-04/.

© 2004-2020 Bootlin, CC BY-SA license 9


Embedded Linux System Development

• Offset 0x0 for the first stage bootloader is dictated by the hardware: the ROM code of the
SAMA5D3 looks for a bootloader at offset 0x0 in the NAND flash.
• Offset 0x40000 for the second stage bootloader is decided by the first stage bootloader.
This can be changed by changing the AT91Bootstrap configuration.
• Offset 0xc0000 of the U-Boot environment is decided by U-Boot. This can be changed by
modifying the U-Boot configuration.
The first item to compile is AT91Bootstrap that you can fetch from Microchip’s GitHub account:
git clone https://github.com/linux4sam/at91bootstrap.git
cd at91bootstrap
git checkout v3.8.9
Then, we first need to configure the build system for our setup. We’re going to need a few pieces
of information for this:
• Which board you want to run AT91Bootstrap on
• Which device should AT91Bootstrap will be stored on
• What component you want AT91Boostrap to load
You can get the list of the supported boards by listing the board directory. You’ll see that in
each of these folders, we have a bunch of defconfig files, that are the supported combinations.
In our case, using the Atmel SAMA5D3 Xplained board, we will load U-Boot, from NAND flash
on (nf in the defconfig file names).
After finding the right defconfig file, load it using make <defconfig_filename> (just the file
name, without the directory part).
In recent versions of AT91Bootstrap, you can now run make menuconfig to explore options
available in this program.
The next thing to do is to specify the cross-compiler prefix (the part before gcc in the cross-
compiler executable name):
export CROSS_COMPILE=arm-linux-
Last but not least, install the python package that the Makefile for AT91Bootstrap will try to
invoke.

10 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

You can now start compiling using make3 .


At the end of the compilation, you should have a file called sama5d3_xplained-nandflashboot-
uboot-*.bin, in the binaries folder.
In order to flash it, we need to do a few things. First, remove the NAND CS jumper on the board.
It’s next to the pin header closest to the Micro-USB plug. Now, press the RESET button. On
the serial port, you should see RomBoot.
Put the jumper back.
Then, start sam-ba_64, running the executable from where it was extracted. You’ll get a small
window. Select the ttyACM0 connection, and the at91sama5d3x-xplained board. Hit Connect.
You need to:
• Hit the NANDFlash tab
• In the Scripts choices, select Enable NandFlash and hit Execute
• Select Erase All, and execute the command
• Then, select and execute Enable OS PMECC parameters in order to change the NAND ECC
4
parameters to what RomBOOT expects.
• Finally, send the image we just compiled using the command Send Boot File
AT91Bootstrap should be flashed now, keep sam-ba open, and move to the next section.

U-Boot setup
Download U-Boot:
wget ftp://ftp.denx.de/pub/u-boot/u-boot-2017.09.tar.bz2
More recent versions may also work, but we have not tested them.
Extract the source archive and get an understanding of U-Boot’s configuration and compilation
steps by reading the README file, and specifically the Building the Software section.
Basically, you need to:
• Set the CROSS_COMPILE environment variable;
• Run make <NAME>_defconfig, where the list of available configurations can be found in
the configs/ directory. There are two flavors of the Xplained configuration: one to run
from the SD card (sama5d3_xplained_mmc) and one to run from the NAND flash (sama5d3_
xplained_nandflash). Since we’re going to boot on the NAND, use the latter.
• Now that you have a valid initial configuration, you can now run make menuconfig to
further edit your bootloader features.
• In recent versions of U-Boot and for some boards, you will need to have the Device Tree
compiler:
sudo apt install device-tree-compiler
• Finally, run make, which should build U-Boot.
3 You can speed up the compiling by using the -jX option with make, where X is the number of parallel jobs

used for compiling. Twice the number of CPU cores is a good value.
4 ECC means Error Correcting Code. If we don’t have the same ECC scheme as the one RomBOOT expects,

RomBOOT will think that the NAND contents are corrupted.

© 2004-2020 Bootlin, CC BY-SA license 11


Embedded Linux System Development

Shrinking U-Boot
Look at the size of the u-boot.bin binary. According to the above flash layout, the U-Boot
binary is supposed to fit between flash offset 0x40000 and offset 0xc0000, corresponding to a
maximum size of 524288 bytes. Is u-boot.bin bigger than this maximum?
The first offset is what AT91Bootstrap expects (though it can be changed in AT91Bootstrap’s
configuration). The second one, corresponding to where U-Boot stores its environment settings,
is board dependent but apparently cannot be changed through make menuconfig.
To avoid recompiling AT91Bootstrap, we propose to compile U-Boot with less features, to make
its binary small. That’s probably something you will do too during real-life projects.
So, in U-Boot sources, run make menuconfig, look for and disable the below options:5 .
• ext4 options
• nfs options
• USB options
• SPL options
• XIMG support
• FIT (Flattened Image Tree) support
• CMD_ELF option
• dhcp command support
• Regular expression support (REGEX)
• loadb support
• CMD_MII option
Now, recompile U-Boot and check that u-boot.bin is smaller than our maximum size.

Flashing U-Boot
Now, in sam-ba, in the Send File Name field, set the path to the u-boot.bin that was just
compiled, and set the address to 0x40000. Click on the Send File button.
You can now exit sam-ba.

Testing U-Boot
Reset the board and check that it boots your new bootloaders. You can verify this by checking
the build dates:
AT91Bootstrap 3.8.9 (Mon Oct 30 2017 16:09:08 (UTC+0100))

NAND: ONFI flash detected


NAND: Manufacturer ID: 0x2c Chip ID: 0xda
NAND: Page Bytes: 2048, Spare Bytes: 64
NAND: ECC Correctability Bits: 4, ECC Sector Bytes: 512
NAND: Disable On-Die ECC
5 For each option, don’t hesitate to use help information to find out what it is about

12 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

NAND: Initialize PMECC params, cap: 4, sector: 512


NAND: Image: Copy 0xa0000 bytes from 0x40000 to 0x26f00000
NAND: Done to load image
<debug_uart>

U-Boot 2017.09 (Oct 30 2017 - 16:20:29 +0100)

CPU: SAMA5D36
Crystal frequency: 12 MHz
CPU clock : 528 MHz
Master clock : 132 MHz
DRAM: 256 MiB
NAND: 256 MiB
MMC: Atmel mci: 0, Atmel mci: 1
*** Warning - bad CRC, using default environment

In: serial@ffffee00
Out: serial@ffffee00
Err: serial@ffffee00
Net:
Error: ethernet@f0028000 address not set.
No ethernet found.
Hit any key to stop autoboot: 0
Interrupt the countdown to enter the U-Boot shell:
=>
In U-Boot, type the help command, and explore the few commands available.

Setting up Ethernet communication


Later on, we will transfer files from the development workstation to the board using the TFTP
protocol, which works on top of an Ethernet connection.
To start with, install and configure a TFTP server on your development workstation, as detailed
in the bootloader slides.
With a network cable, connect the Ethernet port labelled ETH0/GETH of your board to the
one of your computer. If your computer already has a wired connection to the network, your
instructor will provide you with a USB Ethernet adapter. A new network interface should
appear on your Linux system.
Find the name of this interface by typing:
ifconfig -a
The network interface name is likely to be enxxx6 . If you have a pluggable Ethernet device, it’s
easy to identify as it’s the one that shows up after pluging in the device.
Then, instead of configuring the host IP address from NetWork Manager’s graphical interface,
let’s do it through its command line interface, which is so much easier to use:
nmcli con add type ethernet ifname en... ip4 192.168.0.1/24
6 Following the Predictable Network Interface Names convention: https://www.freedesktop.org/wiki/
Software/systemd/PredictableNetworkInterfaceNames/

© 2004-2020 Bootlin, CC BY-SA license 13


Embedded Linux System Development

Now, configure the network on the board in U-Boot by setting the ipaddr and serverip envi-
ronment variables:
setenv ipaddr 192.168.0.100
setenv serverip 192.168.0.1
The first time you use your board, you also need to set the MAC address in U-boot:
setenv ethaddr 12:34:56:ab:cd:ef
In case the board was previously configured in a different way, we also turn off automatic booting
after commands that can be used to copy a kernel to RAM:
setenv autostart no
To make these settings permanent, save the environment:
saveenv
Now reset your board7 .
You can then test the TFTP connection. First, put a small text file in the directory exported
through TFTP on your development workstation. Then, from U-Boot, do:
tftp 0x22000000 textfile.txt
The tftp command should have downloaded the textfile.txt file from your development
workstation into the board’s memory at location 0x220000008 .
You can verify that the download was successful by dumping the contents of the memory:
md 0x22000000
We will see in the next labs how to use U-Boot to download, flash and boot a kernel.

Rescue binaries
If you have trouble generating binaries that work properly, or later make a mistake that causes
you to loose your bootloader binaries, you will find working versions under data/ in the current
lab directory.

7 Resetting your board is needed to make your ethaddr permanent, for obscure reasons. If you don’t, U-boot

will complain that ethaddr is not set.


8 This location is part of the board DRAM. If you want to check where this value comes from, you can check the

Atmel SAMA5D3 datasheet at http://ww1.microchip.com/downloads/en/DeviceDoc/Atmel-11121-32-bit-Cortex-


A5-Microcontroller-SAMA5D3_Datasheet_B.pdf. It’s a big document (more than 1,900 pages). In this document,
look for Memory Mapping and you will find the SoC memory map. You will see that the address range for the
memory controller (DDRC S) starts at 0x20000000 and ends at 0x3fffffff. This shows that the 0x22000000
address is within the address range for RAM. You can also try with other values in the same address range,
knowing that our board only has 256 MB of RAM (that’s 0x10000000, so the physical RAM probably ends at
0x30000000).

14 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Kernel sources
Objective: Learn how to get the kernel sources and patch them.

After this lab, you will be able to:


• Get the kernel sources from the official location
• Apply kernel patches

Setup
Create the $HOME/embedded-linux-labs/kernel directory and go into it.

Get the sources


Go to the Linux kernel web site (https://kernel.org/) and identify the latest stable version.
Just to make sure you know how to do it, check the version of the Linux kernel running on your
machine.
We will use linux-4.19.x, which this lab was tested with.
To practice with the patch command later, download the full 4.18 sources. Unpack the archive,
which creates a linux-4.18 directory. Remember that you can use wget <URL> on the command
line to download files.

Apply patches
Download the patch files corresponding to the latest 4.19 stable release: a first patch to move
from 4.18 to 4.19 and if one exists, a second patch to move from 4.19 to 4.19.x.
Without uncompressing them to a separate file, apply the patches to the Linux source directory.
View one of the patch files with vi or gvim (if you prefer a graphical editor), to understand the
information carried by such a file. How are described added or removed files?
Rename the linux-4.18 directory to linux-4.19.<x>.

© 2004-2020 Bootlin, CC BY-SA license 15


Embedded Linux System Development

Kernel - Cross-compiling
Objective: Learn how to cross-compile a kernel for an ARM target
platform.

After this lab, you will be able to:


• Set up a cross-compiling environment
• Configure the kernel Makefile accordingly
• Cross compile the kernel for the Microchip SAMA5D3 Xplained ARM board
• Use U-Boot to download the kernel
• Check that the kernel you compiled starts the system

Setup
Go to the $HOME/embedded-linux-labs/kernel directory.

Target system
We are going to cross-compile and boot a Linux kernel for the Microchip SAMA5D3 Xplained
board.

Kernel sources
We will re-use the kernel sources downloaded and patched in the previous lab.

Cross-compiling environment setup


To cross-compile Linux, you need to have a cross-compiling toolchain. We will use the cross-
compiling toolchain that we previously produced, so we just need to make it available in the
PATH:
export PATH=$HOME/x-tools/arm-training-linux-uclibcgnueabihf/bin:$PATH
Also, don’t forget to either:
• Define the value of the ARCH and CROSS_COMPILE variables in your environment (using
export)
• Or specify them on the command line at every invocation of make, i.e: make ARCH=...
CROSS_COMPILE=... <target>

Linux kernel configuration


By running make help, find the proper Makefile target to configure the kernel for the Xplained
board (hint: the default configuration is not named after the board, but after the SoC name).

16 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Once found, use this target to configure the kernel with the ready-made configuration.
Don’t hesitate to visualize the new settings by running make xconfig afterwards!
In the kernel configuration, as an experiment, change the kernel compression from Gzip to XZ.
This compression algorithm is far more efficient than Gzip, in terms of compression ratio, at the
expense of a higher decompression time.

Cross compiling
At this stage, you need to install the libssl-dev package to compile the kernel.
You’re now ready to cross-compile your kernel. Simply run:
make
and wait a while for the kernel to compile. Don’t forget to use make -j<n> if you have multiple
cores on your machine!
Look at the end of the kernel build output to see which file contains the kernel image. You can
also see the Device Tree .dtb files which got compiled. Find which .dtb file corresponds to your
board.
Copy the linux kernel image and DTB files to the TFTP server home directory.

Load and boot the kernel using U-Boot


We will use TFTP to load the kernel image on the Xplained board:
• On your workstation, copy the zImage and DTB files to the directory exposed by the
TFTP server.
• On the target (in the U-Boot prompt), load zImage from TFTP into RAM at address
0x21000000:
tftp 0x21000000 zImage
• Now, also load the DTB file into RAM at address 0x22000000:
tftp 0x22000000 at91-sama5d3_xplained.dtb
• Boot the kernel with its device tree:
bootz 0x21000000 - 0x22000000
You should see Linux boot and finally panicking. This is expected: we haven’t provided a
working root filesystem for our device yet.
You can now automate all this every time the board is booted or reset. Reset the board, and
specify a different bootcmd:
setenv bootcmd 'tftp 0x21000000 zImage; tftp 0x22000000 at91-sama5d3_xplained.dtb; bootz 0x21000000 - 0x22000000'
saveenv

Flashing the kernel and DTB in NAND flash


In order to let the kernel boot on the board autonomously, we can flash the kernel image and
DTB in the NAND flash available on the Xplained board.
After storing the first stage bootloader, U-boot and its environment variables, we will keep
special areas in NAND flash for the DTB and Linux kernel images:

© 2004-2020 Bootlin, CC BY-SA license 17


Embedded Linux System Development

So, let’s start by erasing the corresponding 128 KiB of NAND flash for the DTB:
nand erase 0x140000 0x20000
(NAND offset) (size)
Then, let’s erase the 5 MiB of NAND flash for the kernel image:
nand erase 0x160000 0x500000
Then, copy the DTB and kernel binaries from TFTP into memory, using the same addresses as
before.
Then, flash the DTB and kernel binaries:
nand write 0x22000000 0x140000 0x20000
(RAM addr) (NAND offset) (size)
nand write 0x21000000 0x160000 0x500000
Power your board off and on, to clear RAM contents. We should now be able to load the DTB
and kernel image from NAND and boot with:
nand read 0x22000000 0x140000 0x20000
(RAM addr) (offset) (size)
nand read 0x21000000 0x160000 0x500000
bootz 0x21000000 - 0x22000000
Write a U-Boot script that automates the DTB + kernel download and flashing procedure.
You are now ready to modify bootcmd to boot the board from flash. But first, save the settings
for booting from tftp:
setenv bootcmdtftp ${bootcmd}
This will be useful to switch back to tftp booting mode later in the labs.
Finally, using editenv bootcmd, adjust bootcmd so that the Xplained board starts using the
kernel in flash.
Now, reset the board to check that it boots in the same way from NAND flash. Check that this
is really your own version of the kernel that’s running9

9 Look at the kernel log. You will find the kernel version number as well as the date when it was compiled.

That’s very useful to check that you’re not loading an older version of the kernel instead of the one that you’ve
just compiled.

18 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Tiny embedded system with Busy-


Box
Objective: making a tiny yet full featured embedded system

After this lab, you will:


• be able to configure and build a Linux kernel that boots on a directory on your workstation,
shared through the network by NFS.
• be able to create and configure a minimalistic root filesystem from scratch (ex nihilo, out
of nothing, entirely hand made...) for the Xplained board.
• understand how small and simple an embedded Linux system can be.
• be able to install BusyBox on this filesystem.
• be able to create a simple startup script based on /sbin/init.
• be able to set up a simple web interface for the target.

Lab implementation
While (s)he develops a root filesystem for a device, a developer needs to make frequent changes
to the filesystem contents, like modifying scripts or adding newly compiled programs.
It isn’t practical at all to reflash the root filesystem on the target every time a change is made.
Fortunately, it is possible to set up networking between the development workstation and the
target. Then, workstation files can be accessed by the target through the network, using NFS.
Unless you test a boot sequence, you no longer need to reboot the target to test the impact of
script or application updates.

Setup
Go to the $HOME/embedded-linux-labs/tinysystem/ directory.

© 2004-2020 Bootlin, CC BY-SA license 19


Embedded Linux System Development

Kernel configuration
We will re-use the kernel sources from our previous lab, in $HOME/embedded-linux-labs/kernel/.
In the kernel configuration built in the previous lab, verify that you have all options needed for
booting the system using a root filesystem mounted over NFS, and if necessary, enable them
and rebuild your kernel.

Setting up the NFS server


Create a nfsroot directory in the current lab directory. This nfsroot directory will be used to
store the contents of our new root filesystem.
Install the NFS server by installing the nfs-kernel-server package if you don’t have it yet.
Once installed, edit the /etc/exports file as root to add the following line, assuming that the
IP address of your board will be 192.168.0.100:
/home/<user>/embedded-linux-labs/tinysystem/nfsroot 192.168.0.100(rw,no_root_squash,no_subtree_check)

Of course, replace <user> by your actual user name.


Make sure that the path and the options are on the same line. Also make sure that there is no
space between the IP address and the NFS options, otherwise default options will be used for
this IP address, causing your root filesystem to be read-only.
Then, restart the NFS server:
sudo service nfs-kernel-server restart

Booting the system


First, boot the board to the U-Boot prompt. Before booting the kernel, we need to tell it that
the root filesystem should be mounted over NFS, by setting some kernel parameters.
Use the following U-Boot command to do so, in just 1 line
setenv bootargs root=/dev/nfs ip=192.168.0.100:::::eth0
nfsroot=192.168.0.1:/home/<user>/embedded-linux-labs/tinysystem/nfsroot,nfsvers=3 rw
Once again, replace <user> by your actual user name.
Of course, you need to adapt the IP addresses to your exact network setup. Save the environment
variables (with saveenv).
You will later need to make changes to the bootargs value. Don’t forget you can do this with
the editenv command.
Now, boot your system. The kernel should be able to mount the root filesystem over NFS:
VFS: Mounted root (nfs filesystem) on device 0:14.
If the kernel fails to mount the NFS filesystem, look carefully at the error messages in the console.
If this doesn’t give any clue, you can also have a look at the NFS server logs in /var/log/syslog.
However, at this stage, the kernel should stop because of the below issue:
[ 7.476715] devtmpfs: error mounting -2
This happens because the kernel is trying to mount the devtmpfs filesystem in /dev/ in the root
filesystem. To address this, create a dev directory under nfsroot and reboot.

20 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Now, the kernel should complain for the last time, saying that it can’t find an init application:
Kernel panic - not syncing: No working init found. Try passing init= option to kernel.
See Linux Documentation/init.txt for guidance.

Obviously, our root filesystem being mostly empty, there isn’t such an application yet. In the
next paragraph, you will add Busybox to your root filesystem and finally make it usable.

Root filesystem with Busybox


Download the sources of the latest BusyBox 1.31.x release.
To configure BusyBox, we won’t be able to use make xconfig, which is currently broken for
BusyBox in Ubuntu 18.04, because it requires an old version of the Qt library.
So, let’s use make menuconfig.
Now, configure BusyBox with the configuration file provided in the data/ directory (remember
that the Busybox configuration file is .config in the Busybox sources).
If you don’t use the BusyBox configuration file that we provide, at least, make sure you build
BusyBox statically! Compiling Busybox statically in the first place makes it easy to set up the
system, because there are no dependencies on libraries. Later on, we will set up shared libraries
and recompile Busybox.
Build BusyBox using the toolchain that you used to build the kernel.
Going back to the BusyBox configuration interface specify the installation directory for Busy-
Box10 . It should be the path to your nfsroot directory.
Now run make install to install BusyBox in this directory.
Try to boot your new system on the board. You should now reach a command line prompt,
allowing you to execute the commands of your choice.

Virtual filesystems
Run the ps command. You can see that it complains that the /proc directory does not exist.
The ps command and other process-related commands use the proc virtual filesystem to get
their information from the kernel.
From the Linux command line in the target, create the proc, sys and etc directories in your
root filesystem.
Now mount the proc virtual filesystem. Now that /proc is available, test again the ps command.
Note that you can also now halt your target in a clean way with the halt command, thanks to
proc being mounted11 .

System configuration and startup


The first user space program that gets executed by the kernel is /sbin/init and its configuration
file is /etc/inittab.
In the BusyBox sources, read details about /etc/inittab in the examples/inittab file.
10 You will find this setting in Settings -> Install Options -> BusyBox installation prefix.
11 halt can find the list of mounted filesystems in /proc/mounts, and unmount each of them in a clean way
before shutting down.

© 2004-2020 Bootlin, CC BY-SA license 21


Embedded Linux System Development

Then, create a /etc/inittab file and a /etc/init.d/rcS startup script declared in /etc/
inittab. In this startup script, mount the /proc and /sys filesystems.
Any issue after doing this?

Starting the shell in a proper terminal


Before the shell prompt, you probably noticed the below warning message:
/bin/sh: can't access tty; job control turned off
This happens because the shell specified in the /etc/inittab file in started by default in /dev/
console:
::askfirst:/bin/sh
When nothing is specified before the leading ::, /dev/console is used. However, while this
device is fine for a simple shell, it is not elaborate enough to support things such as job control
([Ctrl][c] and [Ctrl][z]), allowing to interrupt and suspend jobs.
So, to get rid of the warning message, we need init to run /bin/sh in a real terminal device:
ttyS0::askfirst:/bin/sh
Reboot the system and the message will be gone!

Switching to shared libraries


Take the hello.c program supplied in the lab data directory. Cross-compile it for ARM,
dynamically-linked with the libraries12 , and run it on the target.
You will first encounter a very misleading not found error, which is not because the hello
executable is not found, but because something else is not found using the attempt to execute
this executable. What’s missing is the ld-uClibc.so.0 executable, which is the dynamic linker
required to execute any program compiled with shared libraries. Using the find command (see
examples in your command memento sheet), look for this file in the toolchain install directory,
and copy it to the lib/ directory on the target.
Then, running the executable again and see that the loader executes and finds out which shared
libraries are missing.
If you still get the same error message, work, just try again a few seconds later. Such a delay can
be needed because the NFS client can take a little time (at most 30-60 seconds) before seeing
the changes made on the NFS server.
Similarly, find the missing libraries in the toolchain and copy them to lib/ on the target.
Once the small test program works, we are going to recompile Busybox without the static
compilation option, so that Busybox takes advantages of the shared libraries that are now
present on the target.
Before doing that, measure the size of the busybox executable.
Then, build Busybox with shared libraries, and install it again on the target filesystem. Make
sure that the system still boots and see how much smaller the busybox executable got.
12 Invoke your cross-compiler in the same way you did during the toolchain lab

22 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Implement a web interface for your device


Replicate data/www/ to the /www directory in your target root filesystem.
Now, run the BusyBox http server from the target command line:
/usr/sbin/httpd -h /www/
It will automatically background itself.
If you use a proxy, configure your host browser so that it doesn’t go through the proxy to connect
to the target IP address, or simply disable proxy usage. Now, test that your web interface works
well by opening http://192.168.0.100 on the host.
See how the dynamic pages are implemented. Very simple, isn’t it?

© 2004-2020 Bootlin, CC BY-SA license 23


Embedded Linux System Development

Filesystems - Block file systems


Objective: configure and boot an embedded Linux system relying on
block storage

After this lab, you will be able to:


• Manage partitions on block storage.
• Produce file system images.
• Configure the kernel to use these file systems
• Use the tmpfs file system to store temporary files

Goals
After doing the A tiny embedded system lab, we are going to copy the filesystem contents to the
SD card. The filesystem will be split into several partitions, and your sama5d3 X-plained board
will be booted with this SD card, without using NFS anymore.

Setup
Throughout this lab, we will continue to use the root filesystem we have created in the $HOME/
embedded-linux-labs/tinysystem/nfsroot directory, which we will progressively adapt to use
block filesystems.

Filesystem support in the kernel


Recompile your kernel with support for SquashFS and ext413 .
Update your kernel image in NAND flash.
Boot your board with this new kernel and on the NFS filesystem you used in this previous lab.
Now, check the contents of /proc/filesystems. You should see that ext4 and SquashFS are
now supported.

Prepare the SD card


We’re going to use an SD card for our block device.
Plug the SD card your instructor gave you on your workstation. Type the dmesg command to
see which device is used by your workstation. In case the device is /dev/mmcblk0, you will see
something like
[46939.425299] mmc0: new high speed SDHC card at address 0007
[46939.427947] mmcblk0: mmc0:0007 SD16G 14.5 GiB
13 Basic configuration options for these filesystems will be sufficient. No need for things like extended attributes.

24 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

The device file name may be different (such as /dev/sdb if the card reader is connected to a
USB bus (either inside your PC or using a USB card reader).
In the following instructions, we will assume that your SD card is seen as /dev/mmcblk0 by your
PC workstation.
Type the mount command to check your currently mounted partitions. If SD partitions are
mounted, unmount them:
$ sudo umount /dev/mmcblk0*
Then, clear possible SD card contents remaining from previous training sessions (only the first
megabytes matter):
$ sudo dd if=/dev/zero of=/dev/mmcblk0 bs=1M count=256
Now, let’s use the cfdisk command to create the partitions that we are going to use:
$ sudo cfdisk /dev/mmcblk0
If cfdisk asks you to Select a label type, choose dos. This corresponds to traditional par-
titions tables that DOS/Windows would understand. gpt partition tables are needed for disks
bigger than 2 TB.
In the cfdisk interface, delete existing partitions, then create three primary partitions, starting
from the beginning, with the following properties:
• One partition, 64MB big, with the FAT16 partition type.
• One partition, 8 MB big14 , that will be used for the root filesystem. Due to the geometry
of the device, the partition might be larger than 8 MB, but it does not matter. Keep the
Linux type for the partition.
• One partition, that fills the rest of the SD card, that will be used for the data filesystem.
Here also, keep the Linux type for the partition.
Press Write when you are done.
To make sure that partition definitions are reloaded on your workstation, remove the SD card
and insert it again.

Data partition on the SD card


Using the mkfs.ext4 create a journaled file system on the third partition of the SD card:
sudo mkfs.ext4 -L data -E nodiscard /dev/mmcblk0p3
• -L assigns a volume name to the partition
• -E nodiscard disables bad block discarding. While this should be a useful option for cards
with bad blocks, skipping this step saves long minutes in SD cards.
Now, mount this new partition and move the contents of the /www/upload/files directory (in
your target root filesystem) into it. The goal is to use the third partition of the SD card as the
storage for the uploaded images.
Connect the SD card to your board. You should see the partitions in /proc/partitions.
14 For the needs of our system, the partition could even be much smaller, and 1 MB would be enough. However,

with the 8 GB SD cards that we use in our labs, 8 MB will be the smallest partition that cfdisk will allow you
to create.

© 2004-2020 Bootlin, CC BY-SA license 25


Embedded Linux System Development

Mount this third partition on /www/upload/files.


Once this works, modify the startup scripts in your root filesystem to do it automatically at
boot time.
Reboot your target system and with the mount command, check that /www/upload/files is
now a mount point for the third SD card partition. Also make sure that you can still upload
new images, and that these images are listed in the web interface.

Adding a tmpfs partition for log files


For the moment, the upload script was storing its log file in /www/upload/files/upload.log.
To avoid seeing this log file in the directory containing uploaded files, let’s store it in /var/log
instead.
Add the /var/log/ directory to your root filesystem and modify the startup scripts to mount a
tmpfs filesystem on this directory. You can test your tmpfs mount command line on the system
before adding it to the startup script, in order to be sure that it works properly.
Modify the www/cgi-bin/upload.cfg configuration file to store the log file in /var/log/upload.
log. You will lose your log file each time you reboot your system, but that’s OK in our system.
That’s what tmpfs is for: temporary data that you don’t need to keep across system reboots.
Reboot your system and check that it works as expected.

Making a SquashFS image


We are going to store the root filesystem in a SquashFS filesystem in the second partition of the
SD card.
In order to create SquashFS images on your host, you need to install the squashfs-tools
package. Now create a SquashFS image of your NFS root directory.
Finally, using the dd command, copy the file system image to the second partition of the SD
card.

Booting on the SquashFS partition


In the U-boot shell, configure the kernel command line to use the second partition of the SD
card as the root file system. Also add the rootwait boot argument, to wait for the SD card
to be properly initialized before trying to mount the root filesystem. Since the SD cards are
detected asynchronously by the kernel, the kernel might try to mount the root filesystem too
early without rootwait.
Check that your system still works. Congratulations if it does!

Store the kernel image and DTB on the SD card


You’ll first need to format the first partition, using:
sudo mkfs.vfat -F 16 -n boot /dev/mmcblk0p1
It will create a new FAT16 partition, called boot. Remove and plug the SD card back in. You
can now copy the kernel image and Device Tree to it.

26 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

You now need to adjust the bootcmd of U-Boot so that it loads the kernel and DTB from the
SD card instead of loading them from the NAND.
In U-boot, you can load a file from a FAT filesystem using a command like
fatload mmc 0:1 0x21000000 filename
Which will load the file named filename from the first partition of the device handled by the
first MMC controller to the system memory at the address 0x21000000.

Going further
At this point our board still uses the bootloaders (at91bootstrap and U-Boot) stored in the
NAND flash. Let’s try to have everything on our SD card. The ROM code can load the first
stage bootloader from an MMC or SD card, from a file named boot.bin located in the first FAT
partition. U-Boot will be stored as u-boot.bin.
For this you will need to recompile at91bootstrap (you’ll need to switch to version 3.8.12) to
support booting from an SD card. Then recompile U-Boot after reconfiguring it with its MMC
configuration (we previously used the configuration for running from NAND flash).

© 2004-2020 Bootlin, CC BY-SA license 27


Embedded Linux System Development

Filesystems - Flash file systems


Objective: Understand flash and flash file systems usage and their
integration on the target

After this lab, you will be able to:


• Prepare filesystem images and flash them.
• Define partitions in embedded flash storage.

Setup
Stay in $HOME/embedded-linux-labs/tinysystem. Install the mtd-utils package, which will be
useful to create UBIFS and UBI images.

Goals
Instead of using an external SD card as in the previous lab, we will make our system use its
internal flash storage.
We will create an MTD partition to be attached to the UBI layer (the partitions previously used
to store the kernel image and the DTB should be merged with this UBI partition).
The kernel and DTB images will be stored in two separate static (read-only) UBI volumes.
The root filesystem will be a UBI volume storing a UBIFS filesystem mounted read-only, the web
server upload data will be stored in another UBI volume storing a UBIFS filesystem mounted
read/write. These volumes will be dynamic volumes and will be 16 MiB large.
Which gives the following layout:

Enabling NAND flash and filesystems


First, make sure your kernel has support for UBI and UBIFS, and also the option allowing us
to pass the partition table through the command line: (CONFIG_MTD_CMDLINE_PARTS).

28 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Recompile your kernel if needed. We will update your kernel image on flash in the next section.

Filesystem image preparation


To prepare filesystem images, we won’t use what you stored on the SD card during the previous
lab. We will get back to the filesystem sources that you have in $HOME/embedded-linux-labs/
tinysystem/nfsroot.
To run mkfs.ubifs, you will need to find the Logical Erase Block (LEB) size that UBI will use.
To find out this information, simply run nand info in U-Boot:
• The Erase size is 128KB, which is the size of the Physical Erase Block
• Both the Page size and subpagesize are 2KB, which means this NAND doesn’t support
sub-pages.
Therefore, the size of one LEB is the size of the PEB minus the size of two pages: 128 KB - 2
* 2 KB, i.e 124 KB.
Knowing that the data and rootfs UBI volumes will be 16 MiB big, you can now divide their
total size by the LEB size, to compute the maximum of LEBs that they will contain. That’s
the last parameter (-c) that you need to pass to mkfs.ubifs.
You can now prepare a UBIFS filesystem image containing the files stored in the www/upload/
files directory.
15
Modify the etc/init.d/rcS file under nfsroot to mount a UBI volume called data on www/
upload/files.
Once done, create a UBIFS image of your root filesystem.

UBI image preparation


Create a ubinize config file where you will define the 4 volumes described above, then use the
ubinize tool to generate your UBI image.
Warning: do not use the autoresize flag (vol_flags=autoresize): U-Boot corrupts the UBI
metadata when trying to expand the volume.
Remember that some of these volumes are static (read-only) and some are not.

MTD partitioning and flashing


Look at the default MTD partitions in the kernel log. They do not match the way we wish to
organize our flash storage. Therefore, we will define our own partitions at boot time, on the
kernel command line.
Redefine the partitions in U-Boot using the mtdids and mtdparts environment variables. Once
done, execute the mtdparts command and check the partition definitions in the output of this
command.
You can now safely erase the UBI partition without risking any corruption on other partitions.
Download the UBI image (using tftp) you have created in the previous section and flash it on
the UBI partition.
15 We will create it when running ubinize in the next section

© 2004-2020 Bootlin, CC BY-SA license 29


Embedded Linux System Development

When flashing the UBI image, use the trimffs version of the command nand write16 .

Loading kernel and DTB images from UBI and booting it


From U-Boot, retrieve the kernel and DTB images from their respective UBI volumes and try
to boot them. If it works, you can modify your bootcmd accordingly.
Set the bootargs variable so that:
• The mtdparts environment variable contents are passed to the kernel through its command
line.
• The UBI partition is automatically attached to the UBI layer at boot time
• The root filesystem is mounted from the root volume, and is mounted read-only (kernel
parameter ro).
Boot the target, and check that your system still works as expected. Your root filesystem should
be mounted read-only, while the data filesystem should be mounted read-write, allowing you to
upload data using the web server.

Going further
Using squashfs for the root filesystem

Root filesystems are often a sensitive part of your system, and you don’t want it to be corrupted,
hence some people decide to use a read-only file system for their rootfs and use another file system
to store their auxiliary data.
squashfs is one of these read-only file systems. However, squashfs expects to be mounted on a
block device.
Use the ubiblk layer to emulate a read-only block device on top of a static UBI volume to mount
a squashfs filesystem as the root filesystem:
• First create a squashfs image with your rootfs contents
• Then create a new static volume to store your squashfs and update it with your squashfs
image
• Enable and setup the ubiblk layer
• Boot on your new rootfs

Atomic update

UBI also provides an atomic update feature, which is particularly useful if you need to safely
upgrade sensitive parts of your system (kernel, DTB or rootfs).
Duplicate the kernel volume and create a U-Boot script to fallback on the second kernel volume
if the first one is corrupted:
16 The command nand write.trimffs skips the blank sectors instead of writing them. It is needed because the
algorithm used by the hardware ECC for the SAMA5D3 SoC generates a checksum with bytes different from
0xFF if the page is blank. Linux only checks the page, and if it is blank it doesn’t erase it, but as the OOB is not
blank it leads to ECC errors. More generally it is not recommended writing more than one time on a page and
its OOB even if the page is blank.

30 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

• First create a new static volume to store your kernel backup


• Flash a valid kernel on the backup volume
• Modify your bootcmd to fallback to the backup volume if the first one is corrupted
• Now try to update the kernel volume and interrupt the process before it has finished and
see what happens (unplug the platform)
• Create a shell script to automate kernel updates (executed in Linux). Be careful, this
script should also handle the case where the backup volume has been corrupted (copy the
contents of the kernel volume into the backup one)

© 2004-2020 Bootlin, CC BY-SA license 31


Embedded Linux System Development

Third party libraries and applica-


tions
Objective: Learn how to leverage existing libraries and applications:
how to configure, compile and install them

To illustrate how to use existing libraries and applications, we will extend the small root filesys-
tem built in the A tiny embedded system lab to add the ALSA libraries and tools and an audio
playback application using these libraries.

We’ll see that manually re-using existing libraries is quite tedious, so that more automated
procedures are necessary to make it easier. However, learning how to perform these operations
manually will significantly help you when you face issues with more automated tools.

Audio support in the Kernel

Recompile your kernel with audio support. The options we want are: CONFIG_SOUND, CONFIG_SND,
CONFIG_SND_USB and CONFIG_SND_USB_AUDIO.

At this stage, the easiest solution to update your kernel is probably to get back to copying it
to RAM from tftp. Anyway, we will have to modify U-Boot environment variables, as we are
going to switch back to NFS booting anyway.

Make sure that your board still boots with this new kernel.

Figuring out library dependencies

We’re going to integrate the alsa-utils and ogg123 executables. As most software components,
they in turn depend on other libraries, and these dependencies are different depending on the
configuration chosen for them. In our case, the dependency chain for alsa-utils is quite simple,
it only depends on the alsa-lib library.

The dependencies are a bit more complex for ogg123. It is part of vorbis-tools, that depend
on libao and libvorbis. libao in turn depends on alsa-lib, and libvorbis on libogg.

libao, alsa-utils and alsa-lib are here to abstract the use of ALSA, one of the Audio Subsys-
tems supported in Linux. vorbis-tools, libvorbis and libogg are used to handle the audio
files encoded using the Ogg codec, which is quite common.

So, we end up with the following dependency tree:

32 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Of course, all these libraries rely on the C library, which is not mentioned here, because it is
already part of the root filesystem built in the A tiny embedded system lab. You might wonder
how to figure out this dependency tree by yourself. Basically, there are several ways, that can
be combined:

• Read the library documentation, which often mentions the dependencies;

• Read the help message of the configure script (by running ./configure --help).

• By running the configure script, compiling and looking at the errors.

To configure, compile and install all the components of our system, we’re going to start from
the bottom of the tree with alsa-lib, then continue with alsa-utils, libao, libogg, and libvorbis, to
finally compile vorbis-tools.

Preparation
For our cross-compilation work, we will need two separate spaces:

• A staging space in which we will directly install all the packages: non-stripped versions
of the libraries, headers, documentation and other files needed for the compilation. This
staging space can be quite big, but will not be used on our target, only for compiling
libraries or applications;

• A target space, in which we will only copy the required files from the staging space: binaries
and libraries, after stripping, configuration files needed at runtime, etc. This target space
will take a lot less space than the staging space, and it will contain only the files that are
really needed to make the system work on the target.

To sum up, the staging space will contain everything that’s needed for compilation, while the
target space will contain only what’s needed for execution.

So, in $HOME/embedded-linux-labs/thirdparty, create two directories: staging and target.

For the target, we need a basic system with BusyBox and initialization scripts. We will re-use
the system built in the A tiny embedded system lab, so copy this system in the target directory:

cp -a $HOME/embedded-linux-labs/tinysystem/nfsroot/* target/

Note that for this lab, a lot of typing will be required. To save time typing, we advise you to
copy and paste commands from the electronic version of these instructions.

© 2004-2020 Bootlin, CC BY-SA license 33


Embedded Linux System Development

Testing
Make sure the target/ directory is exported by your NFS server to your board by modifying
/etc/exports and restarting your NFS server.
Make your board boot from this new directory through NFS.

alsa-lib
alsa-lib is a library supposed to handle the interaction with the ALSA subsystem. It is available
at http://alsa-project.org. Download version 1.1.6, and extract it in $HOME/embedded-linux-
labs/thirdparty/.
By looking at the configure script, we see that it has been generated by autoconf (the header
contains a sentence like Generated by GNU Autoconf 2.69). Most of the time, autoconf comes
with automake, that generates Makefiles from Makefile.am files. So alsa-lib uses a rather com-
mon build system. Let’s try to configure and build it:
./configure
make
You can see that the files are getting compiled with gcc, which generates code for x86 and not
for the target platform. This is obviously not what we want, so we clean-up the object and tell
the configure script to use the ARM cross-compiler:
make clean
CC=arm-linux-gcc ./configure
Of course, the arm-linux-gcc cross-compiler must be in your PATH prior to running the configure
script. The CC environment variable is the classical name for specifying the compiler to use.
Quickly, you should get an error saying:
checking whether we are cross compiling... configure: error: in `.../thirdparty/alsa-lib-1.1.6':
configure: error: cannot run C compiled programs.
If you meant to cross compile, use `--host'.
See `config.log' for more details

If you look at the config.log file, you can see that the configure script compiles a binary with
the cross-compiler and then tries to run it on the development workstation. This is a rather
usual thing to do for a configure script, and that’s why it tests so early that it’s actually doable,
and bails out if not.
Obviously, it cannot work in our case, and the scripts exits. The job of the configure script
is to test the configuration of the system. To do so, it tries to compile and run a few sample
applications to test if this library is available, if this compiler option is supported, etc. But in
our case, running the test examples is definitely not possible.
We need to tell the configure script that we are cross-compiling, and this can be done using
the --build and --host options, as described in the help of the configure script:
System types:
--build=BUILD configure for building on BUILD [guessed]
--host=HOST cross-compile to build programs to run on HOST [BUILD]
The --build option allows to specify on which system the package is built, while the --host
option allows to specify on which system the package will run. By default, the value of the
--build option is guessed and the value of --host is the same as the value of the --build
option. The value is guessed using the ./config.guess script, which on your system should

34 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

return i686-pc-linux-gnu. See http://www.gnu.org/software/autoconf/manual/html_node/


Specifying-Names.html for more details on these options.
So, let’s override the value of the --host option:
CC=arm-linux-gcc ./configure --host=arm-linux
The configure script should end properly now, and create a Makefile. Run the make command,
which should run just fine.
Look at the result of compiling in src/.libs: a set of object files and a set of libasound.so*
files.
The libasound.so* files are a dynamic version of the library. The shared library itself is
libasound.so.2.0.0, it has been generated by the following command line:
arm-linux-gcc -shared conf.o confmisc.o input.o output.o \
async.o error.o dlmisc.o socket.o shmarea.o \
userfile.o names.o -lm -ldl -lpthread -lrt \
-Wl,-soname -Wl,libasound.so.2 -o libasound.so.2.0.0
And creates the symbolic links libasound.so and libasound.so.2.
ln -s libasound.so.2.0.0 libasound.so.2
ln -s libasound.so.2.0.0 libasound.so
These symlinks are needed for two different reasons:
• libasound.so is used at compile time when you want to compile an application that is
dynamically linked against the library. To do so, you pass the -lLIBNAME option to the
compiler, which will look for a file named lib<LIBNAME>.so. In our case, the compilation
option is -lasound and the name of the library file is libasound.so. So, the libasound.so
symlink is needed at compile time;
• libasound.so.2 is needed because it is the SONAME of the library. SONAME stands for
Shared Object Name. It is the name of the library as it will be stored in applications linked
against this library. It means that at runtime, the dynamic loader will look for exactly
this name when looking for the shared library. So this symbolic link is needed at runtime.
To know what’s the SONAME of a library, you can use:
arm-linux-readelf -d libasound.so.2.0.0
and look at the (SONAME) line. You’ll also see that this library needs the C library, because of
the (NEEDED) line on libc.so.0.
The mechanism of SONAME allows to change the library without recompiling the applications
linked with this library. Let’s say that a security problem is found in the alsa-lib release that
provides libasound 2.0.0, and fixed in the next alsa-lib release, which will now provide libasound
2.0.1.
You can just recompile the library, install it on your target system, change the libasound.so.2
link so that it points to libasound.so.2.0.1 and restart your applications. And it will work,
because your applications don’t look specifically for libasound.so.2.0.0 but for the SONAME
libasound.so.2.
However, it also means that as a library developer, if you break the ABI of the library, you must
change the SONAME: change from libasound.so.2 to libasound.so.3.
Finally, the last step is to tell the configure script where the library is going to be installed.
Most configure scripts consider that the installation prefix is /usr/local/ (so that the library

© 2004-2020 Bootlin, CC BY-SA license 35


Embedded Linux System Development

is installed in /usr/local/lib, the headers in /usr/local/include, etc.). But in our system,


we simply want the libraries to be installed in the /usr prefix, so let’s tell the configure script
about this:

CC=arm-linux-gcc ./configure --host=arm-linux --disable-python --prefix=/usr


make

For this library, this option may not change anything to the resulting binaries, but for safety, it
is always recommended to make sure that the prefix matches where your library will be running
on the target system.

Do not confuse the prefix (where the application or library will be running on the target system)
from the location where the application or library will be installed on your host while building
the root filesystem.

For example, libasound will be installed in $HOME/embedded-linux-labs/thirdparty/target/


usr/lib/ because this is the directory where we are building the root filesystem, but once our
target system will be running, it will see libasound in /usr/lib.

The prefix corresponds to the path in the target system and never on the host. So, one
should never pass a prefix like $HOME/embedded-linux-labs/thirdparty/target/usr, otherwise
at runtime, the application or library may look for files inside this directory on the target
system, which obviously doesn’t exist! By default, most build systems will install the application
or library in the given prefix (/usr or /usr/local), but with most build systems (including
autotools), the installation prefix can be overridden, and be different from the configuration
prefix.

We now only have the installation process left to do.

First, let’s make the installation in the staging space:

make DESTDIR=$HOME/embedded-linux-labs/thirdparty/staging install

Now look at what has been installed by alsa-lib:

• Some configuration files in /usr/share/alsa

• The headers in /usr/include

• The shared library and its libtool (.la) file in /usr/lib

• A pkgconfig file in /usr/lib/pkgconfig. We’ll come back to these later

Finally, let’s install the library in the target space:

1. Create the target/usr/lib directory, it will contain the stripped version of the library

2. Copy the dynamic version of the library. Only libasound.so.2 and libasound.so.2.0.0
are needed, since libasound.so.2 is the SONAME of the library and libasound.so.2.0.0
is the real binary:

• cp -a staging/usr/lib/libasound.so.2* target/usr/lib

3. Strip the library:

• arm-linux-strip target/usr/lib/libasound.so.2.0.0

And we’re done with alsa-lib!

36 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Alsa-utils
Download alsa-utils from the ALSA offical webpage. We tested the lab with version 1.1.6.
Once uncompressed, we quickly discover that the alsa-utils build system is based on the autotools,
so we will work once again with a regular configure script.
As we’ve seen previously, we will have to provide the prefix and host options and the CC variable:
CC=arm-linux-gcc ./configure --host=arm-linux --prefix=/usr
Now, we should quiclky get an error in the execution of the configure script:
checking for libasound headers version >= 1.0.27... not present.
configure: error: Sufficiently new version of libasound not found.
Again, we can check in config.log what the configure script is trying to do:
configure:7146: checking for libasound headers version >= 1.0.27
configure:7208: arm-linux-gcc -c -g -O2 conftest.c >&5
conftest.c:12:10: fatal error: alsa/asoundlib.h: No such file or directory

Of course, since alsa-utils uses alsa-lib, it includes its header file! So we need to tell the C
compiler where the headers can be found: there are not in the default directory /usr/include/,
but in the /usr/include directory of our staging space. The help text of the configure script
says:
CPPFLAGS C/C++/Objective C preprocessor flags,
e.g. -I<include dir> if you have headers
in a nonstandard directory <include dir>
Let’s use it:
CPPFLAGS=-I$HOME/embedded-linux-labs/thirdparty/staging/usr/include \
CC=arm-linux-gcc \
./configure --host=arm-linux --prefix=/usr
Now, it should stop a bit later, this time with the error:
checking for libasound headers version >= 1.0.27... found.
checking for snd_ctl_open in -lasound... no
configure: error: No linkable libasound was found.
The configure script tries to compile an application against libasound (as can be seen from
the -lasound option): alsa-utils uses alsa-lib, so the configure script wants to make sure this
library is already installed. Unfortunately, the ld linker doesn’t find it. So, let’s tell the linker
where to look for libraries using the -L option followed by the directory where our libraries
are (in staging/usr/lib). This -L option can be passed to the linker by using the LDFLAGS at
configure time, as told by the help text of the configure script:
LDFLAGS linker flags, e.g. -L<lib dir> if you have
libraries in a nonstandard directory <lib dir>
Let’s use this LDFLAGS variable:
LDFLAGS=-L$HOME/embedded-linux-labs/thirdparty/staging/usr/lib \
CPPFLAGS=-I$HOME/embedded-linux-labs/thirdparty/staging/usr/include \
CC=arm-linux-gcc \
./configure --host=arm-linux --prefix=/usr

© 2004-2020 Bootlin, CC BY-SA license 37


Embedded Linux System Development

Once again, it should fail a bit further down the tests, this time complaining about a missing
curses helper header. curses or ncurses is a graphical framework to design UIs in the terminal.
This is only used by alsamixer, one of the tools provided by alsa-utils, that we are not going to
use. Hence, we can just disable the build of alsamixer.
Of course, if we wanted it, we would have had to build ncurses first, just like we built alsa-lib.
We will also need to disable support for xmlto that generates the documentation.
LDFLAGS=-L$HOME/embedded-linux-labs/thirdparty/staging/usr/lib \
CPPFLAGS=-I$HOME/embedded-linux-labs/thirdparty/staging/usr/include \
CC=arm-linux-gcc \
./configure --host=arm-linux --prefix=/usr \
--disable-alsamixer --disable-xmlto
Then, run the compilation with make. Hopefully, it works!
Let’s now begin the installation process. Before really installing in the staging directory, let’s
install in a dummy directory, to see what’s going to be installed (this dummy directory will not
be used afterwards, it is only to verify what will be installed before polluting the staging space):
make DESTDIR=/tmp/alsa-utils/ install
The DESTDIR variable can be used with all Makefiles based on automake. It allows to override
the installation directory: instead of being installed in the configuration prefix directory, the
files will be installed in DESTDIR/configuration-prefix.
Now, let’s see what has been installed in /tmp/alsa-utils/ (run tree /tmp/alsa-utils):
./lib/udev/rules.d/90-alsa-restore.rules
./usr/bin/aseqnet
./usr/bin/aseqdump
./usr/bin/arecordmidi
./usr/bin/aplaymidi
./usr/bin/aconnect
./usr/bin/alsaloop
./usr/bin/speaker-test
./usr/bin/iecset
./usr/bin/aplay
./usr/bin/amidi
./usr/bin/amixer
./usr/bin/alsaucm
./usr/sbin/alsaconf
./usr/sbin/alsactl
./usr/share/sounds/alsa/Side_Left.wav
./usr/share/sounds/alsa/Rear_Left.wav
./usr/share/sounds/alsa/Noise.wav
./usr/share/sounds/alsa/Front_Right.wav
./usr/share/sounds/alsa/Front_Center.wav
./usr/share/sounds/alsa/Side_Right.wav
./usr/share/sounds/alsa/Rear_Right.wav
./usr/share/sounds/alsa/Rear_Center.wav
./usr/share/sounds/alsa/Front_Left.wav
./usr/share/locale/ru/LC_MESSAGES/alsaconf.mo
./usr/share/locale/ja/LC_MESSAGES/alsaconf.mo
./usr/share/locale/ja/LC_MESSAGES/alsa-utils.mo
./usr/share/locale/fr/LC_MESSAGES/alsa-utils.mo

38 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

./usr/share/locale/de/LC_MESSAGES/alsa-utils.mo
./usr/share/man/fr/man8/alsaconf.8
./usr/share/man/man8/alsaconf.8
./usr/share/man/man1/aseqnet.1
./usr/share/man/man1/aseqdump.1
./usr/share/man/man1/arecordmidi.1
./usr/share/man/man1/aplaymidi.1
./usr/share/man/man1/aconnect.1
./usr/share/man/man1/alsaloop.1
./usr/share/man/man1/speaker-test.1
./usr/share/man/man1/iecset.1
./usr/share/man/man1/aplay.1
./usr/share/man/man1/amidi.1
./usr/share/man/man1/amixer.1
./usr/share/man/man1/alsactl.1
./usr/share/alsa/speaker-test/sample_map.csv
./usr/share/alsa/init/ca0106
./usr/share/alsa/init/hda
./usr/share/alsa/init/test
./usr/share/alsa/init/info
./usr/share/alsa/init/help
./usr/share/alsa/init/default
./usr/share/alsa/init/00main

So, we have:

• The udev rules in lib/udev

• The alsa-utils binaries in /usr/bin and /usr/sbin

• Some sound samples in /usr/share/sounds

• The various translations in /usr/share/locale

• The manual pages in /usr/share/man/, explaining how to use the various tools

• Some configuration samples in /usr/share/alsa.

Now, let’s make the installation in the staging space:

make DESTDIR=$HOME/embedded-linux-labs/thirdparty/staging/ install

Then, let’s install only the necessary files in the target space, manually:

cd ..
cp -a staging/usr/bin/a* staging/usr/bin/speaker-test target/usr/bin/
cp -a staging/usr/sbin/alsa* target/usr/sbin
arm-linux-strip target/usr/bin/a*
arm-linux-strip target/usr/bin/speaker-test
arm-linux-strip target/usr/sbin/alsactl
mkdir -p target/usr/share/alsa/pcm
cp -a staging/usr/share/alsa/alsa.conf* target/usr/share/alsa/
cp -a staging/usr/share/alsa/cards target/usr/share/alsa/
cp -a staging/usr/share/alsa/pcm/default.conf target/usr/share/alsa/pcm/

And we’re finally done with alsa-utils!

© 2004-2020 Bootlin, CC BY-SA license 39


Embedded Linux System Development

Now test that all is working fine by running the speaker-test util on your board, with the
headset provided by your instructor plugged in. You may need to add the missing libraries from
the toolchain install directory.
Caution: don’t copy the dmix.conf file. speaker-test will tell you that it cannot find this file,
but it won’t work if you copy this file from the staging area.
The sound you get will be mainly noise (as what you would get by running speaker-test on
your PCs). At least, sound output is showing some signs of life! It will get much better when
we play samples with ogg123.

libogg
Now, let’s work on libogg. Download the 1.3.3 version from http://xiph.org and extract it.
Configuring libogg is very similar to the configuration of the previous libraries:
CC=arm-linux-gcc ./configure --host=arm-linux --prefix=/usr
Of course, compile the library:
make
Installation to the staging space can be done using the classical DESTDIR mechanism:
make DESTDIR=$HOME/embedded-linux-labs/thirdparty/staging/ install
And finally, only install manually in the target space the files needed at runtime:
cd ..
cp -a staging/usr/lib/libogg.so.0* target/usr/lib/
arm-linux-strip target/usr/lib/libogg.so.0.8.3
Done with libogg!

libvorbis
Libvorbis is the next step. Grab the 1.3.6 version from http://xiph.org and uncompress it.
Once again, the libvorbis build system is a nice example of what can be done with a good usage
of the autotools. Cross-compiling libvorbis is very easy, and almost identical to what we’ve seen
with alsa-utils. First, the configure step:
CC=arm-linux-gcc \
./configure --host=arm-linux --prefix=/usr
It will fail with:
configure: error: Ogg >= 1.0 required !
By running ./configure --help, you will find the --with-ogg-libraries and --with-ogg-
includes options. Use these:
CC=arm-linux-gcc ./configure --host=arm-linux --prefix=/usr \
--with-ogg-includes=$HOME/embedded-linux-labs/thirdparty/staging/usr/include \
--with-ogg-libraries=$HOME/embedded-linux-labs/thirdparty/staging/usr/lib
Then, compile the library:
make

40 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Install it in the staging space:


make DESTDIR=$HOME/embedded-linux-labs/thirdparty/staging/ install
And install only the required files in the target space:
cd ..
cp -a staging/usr/lib/libvorbis.so.0* target/usr/lib/
arm-linux-strip target/usr/lib/libvorbis.so.0.4.8
cp -a staging/usr/lib/libvorbisfile.so.3* target/usr/lib/
arm-linux-strip target/usr/lib/libvorbisfile.so.3.3.7
And we’re done with libvorbis!

libao
Now, let’s work on libao. Download the 1.2.0 version from http://xiph.org and extract it.
Configuring libao is once again fairly easy, and similar to every sane autotools based build
system:
LDFLAGS=-L$HOME/embedded-linux-labs/thirdparty/staging/usr/lib \
CPPFLAGS=-I$HOME/embedded-linux-labs/thirdparty/staging/usr/include \
CC=arm-linux-gcc ./configure --host=arm-linux \
--prefix=/usr
Of course, compile the library:
make
Installation to the staging space can be done using the classical DESTDIR mechanism:
make DESTDIR=$HOME/embedded-linux-labs/thirdparty/staging/ install
And finally, install manually the only needed files at runtime in the target space:
cd ..
cp -a staging/usr/lib/libao.so.4* target/usr/lib/
arm-linux-strip target/usr/lib/libao.so.4.1.0
We will also need the alsa plugin that is loaded dynamically by libao at startup:
mkdir -p target/usr/lib/ao/plugins-4/
cp -a staging/usr/lib/ao/plugins-4/libalsa.so target/usr/lib/ao/plugins-4/
Done with libao!

vorbis-tools
Finally, thanks to all the libraries we compiled previously, all the dependencies are ready. We
can now build the vorbis tools themselves. Download the 1.4.0 version from the official website,
at http://xiph.org/. As usual, extract the tarball.
Before starting the configuration, let’s have a look at the available options by running ./
configure --help. Many options are available. We see that we can, in addition to the usual
autotools configuration options:
• Enable/Disable the various tools that are going to be built: ogg123, oggdec, oggenc, etc.
• Enable or disable support for various other codecs: FLAC, Speex, etc.

© 2004-2020 Bootlin, CC BY-SA license 41


Embedded Linux System Development

• Enable or disable the use of various libraries that can optionally be used by the vorbis
tools
So, let’s begin with our usual configure line:
LDFLAGS=-L$HOME/embedded-linux-labs/thirdparty/staging/usr/lib \
CPPFLAGS=-I$HOME/embedded-linux-labs/thirdparty/staging/usr/include \
CC=arm-linux-gcc \
./configure --host=arm-linux --prefix=/usr
At the end, you should see the following warning:
configure: WARNING: Prerequisites for ogg123 not met, ogg123 will be skipped.
Please ensure that you have POSIX threads, libao, and (optionally) libcurl
libraries and headers present if you would like to build ogg123.
Which is unfortunate, since we precisely want ogg123.
If you look back at the script output, you should see that at some point that it tests for libao
and fails to find it:
checking for AO... no
configure: WARNING: libao too old; >= 1.0.0 required
If you look into the config.log file now, you should find something like:
configure:22343: checking for AO
configure:22351: $PKG_CONFIG --exists --print-errors "ao >= 1.0.0"
Package ao was not found in the pkg-config search path.
Perhaps you should add the directory containing `ao.pc'
to the PKG_CONFIG_PATH environment variable
No package 'ao' found
In this case, the configure script uses the pkg-config system to get the configuration parameters
to link the library against libao. By default, pkg-config looks in /usr/lib/pkgconfig/ for .pc
files, and because the libao-dev package is probably not installed in your system the configure
script will not find libao library that we just compiled.
It would have been worse if we had the package installed, because it would have detected and
used our host package to compile libao, which, since we’re cross-compiling, is a pretty bad thing
to do.
This is one of the biggest issue with cross-compilation: mixing host and target libraries, because
build systems have a tendency to look for libraries in the default paths.
So, now, we must tell pkg-config to look inside the /usr/lib/pkgconfig/ directory of our staging
space. This is done through the PKG_CONFIG_LIBDIR environment variable, as explained in the
manual page of pkg-config.
Moreover, the .pc files contain references to paths. For example, in $HOME/embedded-linux-
labs/thirdparty/staging/usr/lib/pkgconfig/ao.pc, we can see:
prefix=/usr
exec_prefix=${prefix}
libdir=${exec_prefix}/lib
includedir=${prefix}/include
[...]
Libs: -L${libdir} -lao
Cflags: -I${includedir}

42 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

So we must tell pkg-config that these paths are not absolute, but relative to our staging space.
This can be done using the PKG_CONFIG_SYSROOT_DIR environment variable.
Then, let’s run the configuration of the vorbis-tools again, passing the PKG_CONFIG_LIBDIR and
PKG_CONFIG_SYSROOT_DIR environment variables:
LDFLAGS=-L$HOME/embedded-linux-labs/thirdparty/staging/usr/lib \
CPPFLAGS=-I$HOME/embedded-linux-labs/thirdparty/staging/usr/include \
PKG_CONFIG_LIBDIR=$HOME/embedded-linux-labs/thirdparty/staging/usr/lib/pkgconfig \
PKG_CONFIG_SYSROOT_DIR=$HOME/embedded-linux-labs/thirdparty/staging \
CC=arm-linux-gcc \
./configure --host=arm-linux --prefix=/usr

Now, the configure script should end properly, we can now start the compilation:
make
It should fail with the following cryptic message:
make[2]: Entering directory '/home/tux/embedded-linux-labs/thirdparty/vorbis-tools-1.4.0/ogg123'
if arm-linux-gcc -DSYSCONFDIR=\"/usr/etc\" -DLOCALEDIR=\"/usr/share/locale\" -DHAVE_CONFIG_H -I. -I. -I.. -I/usr/include -I../
then mv -f ".deps/audio.Tpo" ".deps/audio.Po"; else rm -f ".deps/audio.Tpo"; exit 1; fi
In file included from audio.c:22:
/usr/include/stdio.h:27:10: fatal error: bits/libc-header-start.h: No such file or directory

You can notice that /usr/include is added to the include paths. Again, this is not what we
want because it contains includes for the host, not the target. It is coming from the autodetected
value for CURL_CFLAGS.
Add the --without-curl option to the configure invocation, restart the compilation.
The compilation may then fail with an error related to libm. While the code uses the function
from this library, the generated Makefile doesn’t give the right command line argument in order
to link against the libm.
If you look at the configure help, you can see
LIBS libraries to pass to the linker, e.g. -l<library>
And this is exactly what we are supposed to use to add new linker flags.
Add this to the configure command line to get
LDFLAGS=-L$HOME/embedded-linux-labs/thirdparty/staging/usr/lib \
CPPFLAGS=-I$HOME/embedded-linux-labs/thirdparty/staging/usr/include \
PKG_CONFIG_LIBDIR=$HOME/embedded-linux-labs/thirdparty/staging/usr/lib/pkgconfig \
PKG_CONFIG_SYSROOT_DIR=$HOME/embedded-linux-labs/thirdparty/staging \
LIBS=-lm \
CC=arm-linux-gcc \
./configure --host=arm-linux --prefix=/usr --without-curl
Finally, it builds!
Now, install the vorbis-tools to the staging space using:
make DESTDIR=$HOME/embedded-linux-labs/thirdparty/staging/ install
And then install them in the target space:
cd ..
cp -a staging/usr/bin/ogg* target/usr/bin
arm-linux-strip target/usr/bin/ogg*

© 2004-2020 Bootlin, CC BY-SA license 43


Embedded Linux System Development

You can now test that everything works! Run ogg123 on the sample file found in thirdparty/
data.
There should still be one missing C library object. Copy it, and you should get: +
ERROR: Failed to load plugin /usr/lib/ao/plugins-4/libalsa.so => dlopen() failed
=== Could not load default driver and no driver specified in config file. Exiting.
This error message is unfortunately not sufficient to figure out what’s going wrong. It’s a good
opportunity to use the strace utility to get more details about what’s going on. To do so, you
can used the one built by Crosstool-ng inside the toolchain target/usr/bin directory.
You can now run ogg123 through strace:
strace ogg123 /sample.ogg
You can see that the command fails to open the ld-uClibc.so.1 file:
open("/lib/ld-uClibc.so.1", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/lib/ld-uClibc.so.1", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/usr/lib/ld-uClibc.so.1", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/usr/X11R6/lib/ld-uClibc.so.1", O_RDONLY) = -1 ENOENT (No such file or directory)
open("/home/tux/embedded-linux-labs/thirdparty/staging/usr/lib/ld-uClibc.so.1", O_RDONLY) = -1 ENOENT (No
write(2, "ERROR: Failed to load plugin ", 29ERROR: Failed to load plugin ) = 29
write(2, "/usr/lib/ao/plugins-4/libalsa.so", 32/usr/lib/ao/plugins-4/libalsa.so) = 32
write(2, " => dlopen() failed\n", 20 => dlopen() failed
Now, look for ld-uClibc.so.1 in the toolchain. You can see that both ld-uClibc.so.1 and
ld-uClibc.so.0 are symbolic links to the same file. So, create the missing link under target/lib
and run ogg123 again.
Everything should work fine now. Enjoy the sound sample!
To finish this lab completely, and to be consistent with what we’ve done before, let’s strip the
libraries in target/lib:
arm-linux-strip target/lib/*

44 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Using a build system, example with


Buildroot
Objectives: discover how a build system is used and how it works,
with the example of the Buildroot build system. Build a Linux system
with libraries and make it work on the board.

Setup
Create the $HOME/embedded-linux-labs/buildroot directory and go into it.

Get Buildroot and explore the source code


The official Buildroot website is available at https://buildroot.org/. Download the latest
stable 2019.02 version which we have tested for this lab. Uncompress the tarball and go inside
the Buildroot source directory.
Several subdirectories or files are visible, the most important ones are:
• boot contains the Makefiles and configuration items related to the compilation of common
bootloaders (Grub, U-Boot, Barebox, etc.)
• configs contains a set of predefined configurations, similar to the concept of defconfig in
the kernel.
• docs contains the documentation for Buildroot. You can start reading buildroot.html
which is the main Buildroot documentation;
• fs contains the code used to generate the various root filesystem image formats
• linux contains the Makefile and configuration items related to the compilation of the
Linux kernel
• Makefile is the main Makefile that we will use to use Buildroot: everything works through
Makefiles in Buildroot;
• package is a directory that contains all the Makefiles, patches and configuration items to
compile the user space applications and libraries of your embedded Linux system. Have a
look at various subdirectories and see what they contain;
• system contains the root filesystem skeleton and the device tables used when a static /dev
is used;
• toolchain contains the Makefiles, patches and configuration items to generate the cross-
compiling toolchain.

Configure Buildroot
In our case, we would like to:

© 2004-2020 Bootlin, CC BY-SA license 45


Embedded Linux System Development

• Generate an embedded Linux system for ARM;


• Use an already existing external toolchain instead of having Buildroot generating one for
us;
• Integrate Busybox, alsa-utils and vorbis-tools in our embedded Linux system;
• Integrate the target filesystem into a tarball
To run the configuration utility of Buildroot, simply run:
make menuconfig
Set the following options. Don’t hesitate to press the Help button whenever you need more
details about a given option:
• Target options
– Target Architecture: ARM (little endian)
– Target Architecture Variant: cortex-A5
– Enable VFP extension support: Enabled
– Target ABI: EABIhf
– Floating point strategy: VFPv4-D16
• Toolchain
– Toolchain type: External toolchain
– Toolchain: Custom toolchain
– Toolchain path: use the toolchain you built: /home/<user>/x-tools/arm-training-
linux-uclibcgnueabihf (replace <user> by your actual user name)
– External toolchain gcc version: 8.x
– External toolchain kernel headers series: 4.16.x
– External toolchain C library: uClibc/uClibc-ng
– We must tell Buildroot about our toolchain configuration, so select Toolchain has
WCHAR support?, Toolchain has SSP support? and Toolchain has C++ support?.
Buildroot will check these parameters anyway.
• Target packages
– Keep BusyBox (default version) and keep the Busybox configuration proposed by
Buildroot;
– Audio and video applications
∗ Select alsa-utils
∗ ALSA utils selection
· Select alsactl
· Select alsamixer
∗ Select vorbis-tools
• Filesystem images
– Select tar the root filesystem

46 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Exit the menuconfig interface. Your configuration has now been saved to the .config file.

Generate the embedded Linux system


Just run:
make
Buildroot will first create a small environment with the external toolchain, then download,
extract, configure, compile and install each component of the embedded system.
All the compilation has taken place in the output/ subdirectory. Let’s explore its contents:
• build, is the directory in which each component built by Buildroot is extracted, and where
the build actually takes place
• host, is the directory where Buildroot installs some components for the host. As Buildroot
doesn’t want to depend on too many things installed in the developer machines, it installs
some tools needed to compile the packages for the target. In our case it installed pkg-config
(since the version of the host may be ancient) and tools to generate the root filesystem
image (genext2fs, makedevs, fakeroot).
• images, which contains the final images produced by Buildroot. In our case it’s just a
tarball of the filesystem, called rootfs.tar, but depending on the Buildroot configuration,
there could also be a kernel image or a bootloader image.
• staging, which contains the “build” space of the target system. All the target libraries,
with headers and documentation. It also contains the system headers and the C library,
which in our case have been copied from the cross-compiling toolchain.
• target, is the target root filesystem. All applications and libraries, usually stripped, are
installed in this directory. However, it cannot be used directly as the root filesystem, as
all the device files are missing: it is not possible to create them without being root, and
Buildroot has a policy of not running anything as root.

Run the generated system


Go back to the $HOME/embedded-linux-labs/buildroot/ directory. Create a new nfsroot di-
rectory that is going to hold our system, exported over NFS. Go into this directory, and untar
the rootfs using:
sudo tar xvf ../buildroot-2019.02/output/images/rootfs.tar
Add our nfsroot directory to the list of directories exported by NFS in /etc/exports, and make
sure the board uses it too.
Boot the board, and log in (root account, no password).
You should now have a shell, where you will be able to run ogg123 like you used to in the
previous lab.

Going further
• Add dropbear (SSH server and client) to the list of packages built by Buildroot and log to
your target system using an ssh client on your development workstation. Hint: you will
have to set a non-empty password for the root account on your target for this to work.

© 2004-2020 Bootlin, CC BY-SA license 47


Embedded Linux System Development

• Add a new package in Buildroot for the GNU Gtypist game. Read the Buildroot documen-
tation to see how to add a new package. Finally, add this package to your target system,
compile it and run it. The newest versions require a library that is not fully supported by
Buildroot, so you’d better stick with the latest version in the 2.8 series.
• Flash the new system on the flash of the board
– First, in buildroot, select the UBIFS filesystem image type.
– You’ll also need to provide buildroot some information on the underlying device
that will store the filesystem. In our case, the logical eraseblock size is 124KiB, the
minimum I/O unit size is 2048 and the Maximum logical eraseblock (LEB) count is
133.
– Then, once the image has been generated, update your rootfs volume.

48 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Application development
Objective: Compile and run your own ncurses application on the
target.

Setup
Go to the $HOME/embedded-linux-labs/appdev directory.

Compile your own application


We will re-use the system built during the Buildroot lab and add to it our own application.
In the lab directory the file app.c contains a very simple ncurses application. It is a simple
game where you need to reach a target using the arrow keys of your keyboard. We will compile
and integrate this simple application to our Linux system.
Buildroot has generated toolchain wrappers in output/host/usr/bin, which make it easier to
use the toolchain, since these wrappers pass some mandatory flags (especially the --sysroot
gcc flag, which tells gcc where to look for the headers and libraries).
Let’s add this directory to our PATH:
export PATH=$HOME/embedded-linux-labs/buildroot/buildroot-XXXX.YY/output/host/usr/bin:$PATH

Let’s try to compile the application:


arm-linux-gcc -o app app.c
It complains about undefined references to some symbols. This is normal, since we didn’t tell
the compiler to link with the necessary libraries. So let’s use pkg-config to query the pkg-
config database about the location of the header files and the list of libraries needed to build an
application against ncurses17 :
arm-linux-gcc -o app app.c $(pkg-config --libs --cflags ncurses)
You can see that ncurses doesn’t need anything in particular for the CFLAGS but you can have
a look at what is needed for libvorbis to get a feel of what it can look like:
pkg-config --libs --cflags vorbis
Our application is now compiled! Copy the generated binary to the NFS root filesystem (in the
root/ directory for example), start your system, and run your application!
You can also try to run it over ssh if you added ssh support to your target. Do you notice the
difference?

17 Again, output/host/usr/bin has a special pkg-config that automatically knows where to look, so it already

knows the right paths to find .pc files and their sysroot.

© 2004-2020 Bootlin, CC BY-SA license 49


Embedded Linux System Development

Remote application debugging


Objective: Use strace to diagnose program issues. Use gdbserver
and a cross-debugger to remotely debug an embedded application

Setup
Go to the $HOME/embedded-linux-labs/debugging directory. Create an nfsroot directory.

Debugging setup
Because of issues in gdb and ltrace in the uClibc version that we are using in our toolchain, we
will use a different toolchain in this lab, based on glibc.
As glibc has more complete features that lighter libraries, it looks like a good idea to do your
application debugging work with a glibc toolchain first, and then switch to lighter libraries once
your application and software stack is production ready.
Extract the Buildroot 2019.02 sources into the current directory.
Then, in the menuconfig interface, configure the target architecture as done previously but
configure the toolchain and target packages differently:
• In Toolchain:
– Toolchain type: External toolchain
– Toolchain: Custom Toolchain
– Toolchain origin: Toolchain to be downloaded and installed
– Toolchain URL: https://toolchains.bootlin.com/downloads/releases/toolchains/
armv7-eabihf/tarballs/armv7-eabihf--glibc--bleeding-edge-2018.07-3.tar.bz2
You can easily choose such a toolchain on https://toolchains.bootlin.com by se-
lecting the architecture, the C library and the compiler version you need. While you
can try with other versions, the above toolchain is known to make this lab work.
– External toolchain gcc version: 8.x
– External toolchain kernel headers series: 4.14.x
– External toochain C library: glibc/eglibc
– Select Toolchain has SSP support?
– Select Toolchain has RPC support?
– Select Toolchain has C++ support?
– Select Copy gdb server to the Target
• Target packages
– Debugging, profiling and benchmark

50 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

∗ Select ltrace
∗ Select strace
Now, build your root filesystem.
Go back to the $HOME/embedded-linux-labs/debugging directory and extract the buildroot-
2019.02/output/images/rootfs.tar archive in the nfsroot directory.
Add this directory to the /etc/exports file and restart nfs-kernel-server.
Boot your ARM board over NFS on this new filesystem, using the same kernel as before.

Using strace
Now, go to the $HOME/embedded-linux-labs/debugging directory.
strace allows to trace all the system calls made by a process: opening, reading and writing files,
starting other processes, accessing time, etc. When something goes wrong in your application,
strace is an invaluable tool to see what it actually does, even when you don’t have the source
code.
Update the PATH:
export PATH=$HOME/embedded-linux-labs/debugging/buildroot-2019.02/output/host/bin:$PATH

With your cross-compiling toolchain compile the data/vista-emulator.c program, strip it with
arm-linux-strip, and copy the resulting binary to the /root directory of the root filesystem.
Back to target system, try to run the /root/vista-emulator program. It should hang indefi-
nitely!
Interrupt this program by hitting [Ctrl] [C].
Now, running this program again through the strace command and understand why it hangs.
You can guess it without reading the source code!
Now add what the program was waiting for, and now see your program proceed to another bug,
failing with a segmentation fault.

Using ltrace
Now run the program through ltrace.
Now you should see what the program does: it tries to consume as much system memory as it
can!
Also run the program through ltrace -c, to see what function call statistics this utility can
provide.
It’s also interesting to run the program again with strace. You will see that memory allocations
translate into mmap() system calls. That’s how you can recognize them when you’re using strace.

Using gdbserver
We are now going to use gdbserver to understand why the program segfaults.
Compile vista-emulator.c again with the -g option to include debugging symbols. This time,
just keep it on your workstation, as you already have the version without debugging symbols
on your target.

© 2004-2020 Bootlin, CC BY-SA license 51


Embedded Linux System Development

Then, on the target side, run vista-emulator under gdbserver. gdbserver will listen on a TCP
port for a connection from gdb, and will control the execution of vista-emulator according to
the gdb commands:
gdbserver localhost:2345 vista-emulator
On the host side, run arm-linux-gdb (also found in your toolchain):
arm-linux-gdb vista-emulator
You can also start the debugger through the ddd interface:
ddd --debugger arm-linux-gdb vista-emulator
gdb starts and loads the debugging information from the vista-emulator binary that has been
compiled with -g.
Then, we need to tell where to find our libraries, since they are not present in the default /lib
and /usr/lib directories on your workstation. This is done by setting the gdb sysroot variable
(on one line):
(gdb) set sysroot /home/<user>/embedded-linux-labs/debugging/
buildroot-2019.02/output/staging
Of course, replace <user> by your actual user name.
And tell gdb to connect to the remote system:
(gdb) target remote <target-ip-address>:2345
Then, use gdb as usual to set breakpoints, look at the source code, run the application step by
step, etc. Graphical versions of gdb, such as ddd can also be used in the same way. In our case,
we’ll just start the program and wait for it to hit the segmentation fault:
(gdb) continue
You could then ask for a backtrace to see where this happened:
(gdb) backtrace
This will tell you that the segmentation fault occurred in a function of the C library, called by
our program. This should help you in finding the bug in our application.

What to remember
During this lab, we learned that...
• It’s easy to study the behavior of programs and diagnose issues without even having the
source code, thanks to strace.
• You can leave a small gdbserver program (about 300 KB) on your target that allows to
debug target applications, using a standard gdb debugger on the development host.
• It is fine to strip applications and binaries on the target machine, as long as the programs
and libraries with debugging symbols are available on the development host.

52 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Real-time - Timers and scheduling


latency
Objective: Learn how to handle real-time processes and practice with
the different real-time modes. Measure scheduling latency.

After this lab, you will:


• Be able to check clock accuracy.
• Be able to start processes with real-time priority.
• Be able to build a real-time application against the standard POSIX real-time API, and
against Xenomai’s POSIX skin.
• Have compared scheduling latency on your system, between a standard kernel, a kernel
with PREEMPT_RT and a kernel with Xenomai.

Setup
Go to the $HOME/embedded-linux-labs/realtime/ directory.
Install the netcat package.

Root filesystem
Create an nfsroot directory.
To compare real-time latency between standard Linux and Xenomai, we are going to need a
root filesystem and a build environment that supports Xenomai.
Let’s build this with Buildroot.
Download and extract the Buildroot 2016.02 sources. As the latest version of Xenomai doesn’t
seem to work on the Xplained board (yet), we need an older version of Buildroot that will build
Xenomai 2.6.
Configure Buildroot with the following settings, using the / command in make menuconfig to
find parameters by their name:
• In Target:
– Target architecture: ARM (little endian)
– Target Architecture Variant: cortex-A5
• In Toolchain:
– Toolchain type: External toolchain
– Toolchain: Sourcery CodeBench ARM 2014.05
• In System configuration:

© 2004-2020 Bootlin, CC BY-SA license 53


Embedded Linux System Development

– in Run a getty (login prompt) after boot, TTY port: ttyS0

• In Target packages:

– Enable Show packages that are also provided by busybox. We need this to build
the standard netcat command, not provided in the default BusyBox configuration.

– In Debugging, profiling and benchmark, enable rt-tests. This will be a few ap-
plications to test real-time latency.

– In Networking applications, enable netcat

– In Real-Time, enable Xenomai Userspace:

∗ Enable Install testsuite

∗ Make sure that POSIX skin library and Native skin library18 are enabled.

Now, build your root filesystem.

As you are using a 64 bit distribution, Buildroot should also ask you to install 32 bit compatibility
packages to be able to execute the Sourcery CodeBench external toolchain:

sudo apt install libc6-i386 lib32stdc++6 lib32z1

At the end of the build job, extract the output/images/rootfs.tar archive in the nfsroot
directory.

The last thing to do is to add a few files that we will need in our tests:

cp data/* nfsroot/root

Downloading sources and patches


We will use a kernel version that is supported by the PREEMPT_RT patchset.

So, go to https://kernel.org/pub/linux/kernel/projects/rt/4.13/, download the latest patch


available in a single file.

Then go to http://kernel.org and download the exact version corresponding to the patch you
downloaded. At the time of this writing, this version was 4.13.10.

Compile a standard Linux kernel


Extract the sources of your 4.13.x kernel but don’t apply the PREEMPT_RT patches yet.

Configure your kernel for your Xplained board, and then make sure that the below settings
are disabled: CONFIG_PROVE_LOCKING, CONFIG_DEBUG_LOCK_ALLOC, CONFIG_DEBUG_MUTEXES and
CONFIG_DEBUG_SPINLOCK.

Also, for the moment, disable the CONFIG_HIGH_RES_TIMERS option which impact we want to
measure.

Boot the Xplained board by mounting the root filesystem that you built. As usual, login as
root, there is no password.
18 Needed by the Xenomai testsuite.

54 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Compiling with the POSIX RT library


The root filesystem was built with the GNU C library, because it has better support for the
POSIX RT API.
In our case, when we created this lab, uClibc didn’t support the clock_nanosleep function used
in our rttest.c program. uClibc also does not support priority inheritance on mutexes.
Therefore, we will need to compile our test application with the toolchain that Buildroot used.
Let’s configure our PATH to use this toolchain:
export PATH=$HOME/embedded-linux-labs/realtime/buildroot-YYYY.MM/output/host/usr/bin:$PATH

Have a look at the rttest.c source file available in root/ in the nfsroot/ directory. See how it
shows the resolution of the CLOCK_MONOTONIC clock.
Now compile this program:
arm-none-linux-gnueabi-gcc -o rttest rttest.c -lrt
Execute the program on the board. Is the clock resolution good or bad? Compare it to the
timer tick of your system, as defined by CONFIG_HZ.
Copy the results in a file, in order to be able to compare them with further results.
Obviously, this resolution will not provide accurate sleep times, and this is because our kernel
doesn’t use high-resolution timers. So let’s add back the CONFIG_HIGH_RES_TIMERS option in the
kernel configuration.
Recompile your kernel, boot your Xplained with the new version, and check the new resolution.
Better, isn’t it?

Testing the non-preemptible kernel


Now, do the following tests:
• Test the program with nothing special and write down the results.
• Test your program and at the same time, add some workload to the board, by running
/root/doload 300 > /dev/null 2>&1 & on the board, and using netcat 192.168.0.100
5566 on your workstation in order to flood the network interface of the Xplained board
(where 192.168.0.100 is the IP address of the Xplained board).
• Test your program again with the workload, but by running the program in the SCHED_FIFO
scheduling class at priority 99, using the chrt command.

Testing the preemptible kernel


Recompile your kernel with CONFIG_PREEMPT enabled, which enables kernel preemption (except
for critical sections protected by spinlocks).
Run the simple tests again with this new preemptible kernel and compare the results.

Compiling and testing the PREEMPT_RT kernel


Download the latest PREEMPT_RT kernel patch and apply it to your kernel sources.
Configure your kernel with CONFIG_PREEMPT_RT_FULL and boot it.

© 2004-2020 Bootlin, CC BY-SA license 55


Embedded Linux System Development

Repeat the tests and compare the results again. You should see a massive improvement in the
maximum latency.

Testing Xenomai scheduling latency


Stay in $HOME/embedded-linux-labs/realtime.
Download the 2.6.4 release of Xenomai (that’s what our version of Buildroot supports by de-
fault), and extract it.
As you can see in the ksrc/arch/arm/patches directory, the most recent Linux kernel version
supported by Xenomai for ARM is 3.14.17.
Then, download the 3.14.17 Linux sources (not the latest 3.14.x sources because the Xenomai
patches only apply to this exact version), and extract them.
Now, prepare our kernel for Xenomai compilation:
cd $HOME/embedded-linux-labs/realtime
./xenomai-2.6.4/scripts/prepare-kernel.sh --arch=arm \
--linux=linux-3.14.17 \
--adeos=xenomai-2.6.4/ksrc/arch/arm/patches/ipipe-core-3.14.17-arm-4.patch
Now, configure your kernel for SAMA5 boards, then start the kernel configuration interface,
and make sure that the below options are enabled, taking your time to read their description:
• CONFIG_XENOMAI
• CONFIG_XENO_DRIVERS_TIMERBENCH
• CONFIG_XENO_HW_UNLOCKED_SWITCH
Compile your kernel, using the same Sourcery CodeBench compiler as earlier in the lab19 .
While the kernel compiles, we can start to build our application against the Xenomai libraries.
We will need pkg-config built by Buildroot. So go in your Buildroot source directory, and force
Buildroot to build the host variant of pkg-config:
cd $HOME/embedded-linux-labs/realtime/buildroot-YYYY.MM/
make host-pkgconf
We can now compile rttest for the Xenomai POSIX skin:
cd $HOME/embedded-linux-labs/realtime/nfsroot/root
export PATH=$HOME/embedded-linux-labs/realtime/buildroot-YYYY.MM/output/host/usr/bin:$PATH
arm-none-linux-gnueabi-gcc -o rttest rttest.c \
$(pkg-config --libs --cflags libxenomai_posix)

Now boot the board with the new kernel.


Run the following commands on the board:
echo 0 > /proc/xenomai/latency
This will disable the timer compensation feature of Xenomai. This feature allows Xenomai to
adjust the timer programming to take into account the time the system needs to schedule a task
after being woken up by a timer. However, this feature needs to be calibrated specifically for
each system. By disabling this feature, we will have raw Xenomai results, that could be further
improved by doing proper calibration of this compensation mechanism.
19 Your own toolchain is too recent for the 3.14 kernel, which doesn’t support compiling with gcc5 yet. The

Sourcery CodeBench gcc version is 4.8.x.

56 © 2004-2020 Bootlin, CC BY-SA license


Embedded Linux System Development

Run the tests again, compare the results.

Testing Xenomai interrupt latency


Measure the interrupt latency with and without load, running the following command:
latency -t 2

© 2004-2020 Bootlin, CC BY-SA license 57

You might also like