Compiling For Embedded Debian Target Systems

5 minute read

Recently I bought a Raspberry Pi 3 Model B since I wanted to evaluate the different strategies of compiling source code for such an ARM powered device.

The probably most popular operating system for the Raspberry Pi is Raspbian. From my point of view this is a very reasonable choice and therefore I prepared a flash card according to this instructions.

Since my two sons (4 and 6 years old) helped me to unpack and assemble everything it was now their turn to watch a “bob the builder” on this brand new “computer”.

For this reason I had no access to the Raspberry Pi and this gave me some time to think about a good benchmark.

I decided that I want to recompile and repackage openssl. According to sloccount it consists of about 310k lines of C and some 4k lines of C++ code. Furthermore it also contains perl, assembler, shell and lisp code.

Due to the above mentioned shortage of Raspberry Pi’s I opted for a test run on my notebook (equipped with Ubuntu 16.04, amd64). The Raspbian that I have installed to the “bob the builder” device is based on Debian jessie. Therefore I decided to to build a Debian jessie LXC container for my first experiments:

mkdir debian-amd64
cd debian-amd64
edi config init debian-jessie-amd64 debian-jessie-amd64
sudo edi -v lxc configure debian-jessie-amd64 debian-jessie-amd64-develop.yml

For the next few steps we will enter the container (password is ChangeMe!, please check this documentation if you are interested in more details) and install some additional software:

lxc exec debian-jessie-amd64 -- login $USER
sudo apt install dpkg-dev devscripts m4 bc vim

Now it is time to adjust /etc/apt/sources.list within the container by adding a line such as deb-src jessie main.

The following commands will fetch the source code of openssl for the current Debian installation (in our case openssl-1.0.1t):

sudo apt update
cd edi-workspace
apt-get source libssl1.0.0
cd openssl-1.0.1t

Finally it is time to start the Debian package build:

time debuild -us -uc

It took slightly less than 5 minutes to get everything done on my notebook that is powered by an i7-7500U (Kaby Lake) CPU and the result were two amd64 Debian packages that are obviously not installable on the Raspberry Pi due to its ARM architecture.

Native Compilation on the Raspberry Pi

As the sun was shining I had an easy game to convince my sons to stop watching “bob the builder” and go out instead. Now the Raspberry Pi shortage was resolved and I reproduced the above steps on the Raspberry Pi.

After the command

time debuild -us -uc

I joined my sons outside.

Later in the evening I had a look at the result: The job was done after about 35 minutes resulting in two Raspbian compatible Debian packages. Honestly, I was surprised that it did not take even longer. The ARM Cortex-A53 CPU of the Raspberry Pi 3 Model B seems to be quite powerful.

The fact that the compilation on my notebook is seven times faster than on the Raspberry Pi made me think about an additional compiling approach:

Native Compilation within an Emulated Container

This time we will build an armhf Raspbian container that runs on the Intel notebook thanks to QEMU:

mkdir raspbian-armhf
cd raspbian-armhf
edi config init raspbian-jessie-armhf raspbian-jessie-armhf
sudo edi -v lxc configure raspbian-jessie-armhf raspbian-jessie-armhf-develop.yml

The setup of the raspbian-jessie-armhf container takes considerably longer than for the debian-jessie-amd64 container due to the emulation.

After the successful creation we reproduce the same steps as for the debian-jessie-amd64 container.

Finally we run the build within this container:

time debuild -us -uc

As expected the build produced two Raspbian compatible Debian packages. However - to my big disappointment - it took more than 65 minutes!

It was now time to scratch my head and remember my past when I was working with cross compilers:

Cross Compilation Within a Native Container

We return into our debian_jessie_amd64 container and pimp it with a cross compiler. As of Debian jessie cross compilers are not part of the main repository and therefore we have to activate an additional repository:

First we add an additional line to our /etc/apt/sources.list which reads as follows: deb jessie main. Then we install the required archive key:

curl | sudo apt-key add -

The security aware reader will notice that we should actually fetch such keys from a https source.

Many people have done a great job that will allow us to proceed with the next step: A Debian based installation is aware of multiple architectures and therefore we can enable armhf as a foreign architecture:

sudo dpkg --add-architecture armhf

This will allow us to install armhf libraries alongside the amd64 libraries.

It is now time to update the apt cache:

sudo apt update

Finally we are ready to install the gcc based cross toolchain:

sudo apt install crossbuild-essential-armhf

Again we want to build openssl and therefore we re-fetch the source code:

cd edi-workspace
mv openssl-1.0.1t openssl-1.0.1t.old
apt-get source libssl1.0.0
cd openssl-1.0.1t

And anxiously waiting for the result we start the cross compilation:

export DEB_BUILD_OPTIONS=nocheck; time debuild -us -uc -aarmhf

Great - only about 5 minutes! This build was seven times faster than on the Raspberry Pi 3 Model B!

Unfortunately it is too early to celebrate. The Raspbian binaries are built with slightly different compiler options than those we have just been applying right now. Therefore do not install those packages on your Raspbian installation!

Bob would now ask: “Can we fix it?” And the whole team knows the answer: “Yes we can!”

I will show the necessary steps in my next blog post!


Cross compilation can speed up your development cycles for embedded devices. The Debian/Ubuntu community has done a great job by introducing multiarch within recent versions of their distributions. It makes cross compiling very sleek.

If you stumble upon a piece of software that persistently refuses to get cross compiled - unfortunately this might happen - you are still able to compile it within an emulated container or on the target hardware.

While cross compilation is typically done within a chroot, we use LXC/LXD containers instead. The overhead introduced by LXC/LXD is negligible and the added value is huge if you are planning to do additional stuff within the container. Please check Stéphane Graber’s website if you want to learn more about LXC/LXD.


Leave a comment