Create embedded Linux images with Docker from scratch

Embedded Linux systems are so common these days and they can be found in set-top boxes, smart TVs, network devices (like routers, 3G/4G USB modems), etc. For the developers, a major concern with these systems is to create their own custom Linux distribution based on their needs. This includes creating the image for the development or the final product. Making the image usually consists of compiling the Linux kernel, device tree binaries, compiling bootloader, and customizing the Linux file system. These steps may become frustrating if you want to test something or trying to rebuild the image over and over. On the other side creating the environment for cross-compiling is a little complicated as you need to set a special environment for the ARM toolchain.

There are many automated tools to achieve this like Buildroot and Yocto Project. They usually give you a minimal version of the Linux image and are a good choice for the final image that intended to use in the production environment, but for the development environment, they are not always the best choice because of lacking package manager or external tools.
Another choice is to create your own custom Linux image by compiling the kernel, bootloader, file system, and finally bring them all together and put it into the device SD card or flash storage. With the Docker, you can do this very straight forward and you don't need to mess around with your host. Moreover, it can be used even on Mac or Windows machines.

I created a simple Docker project to achieve this. It can be customized to use with any embedded Linux board and can be developed easily. I've tested it with BeagleBone Black but with a little change can be used on other educational development boards like Raspberry Pi, Nano Pi, etc. You can also use it on your own customized embedded Linux board but you need to provide your configuration file for the kernel or bootloader.
All you need to do is to provide the resource binaries for Linux kernel, Bootloader, and filesystem then start the build with a single command docker-compose up. The output is a standard image file that can be written on the SD card easily. Also, it uses Linaro toolchain for cross-compilation so it is fast and reliable.

Although the use of the project is quite easy and straight forward I've made a step by step guide. So let's go into details and how to set it up.

To use this project there is no need to set up any special environment or meet any special requirements.
I've used BeagleBone Black as the target system with Ubuntu 18.04 and Ti's Linux Kernel 4.19.94.

Features

• Based on Docker so can be run on any machine without any requirements.
• Build Kernel, DTS, modules, and headers from scratch.
• Build Bootloader (u-boot) from the source.
• Can be combined with any Linux distro filesystem.
• Create image file (.img) as output file for easy writing to SD card.
• High-speed build with help of using Linaro toolchain and Docker engine.
• Can be used with any Embedded Linux system with little configuration change.

Requirements

• Docker and Docker Compose
• Download resource files and put them in resources directory:
  • Linaro cross compile toolchain.
  • You can get the binaries from here.
  • Linux Kernel source for your board.
  • For BeagleBone Black you can use the official kernel
  • u-boot source code).
  • A base rootfs Linux distro.
  • You can grab Ubuntu or Debian distro from here.
  • Ubuntu 18.04.3 used in this project.
  • As the compile process may take longer, you need to increase the Docker engine resources:
  • Windows
  • Mac

How to use

1. Make sure that you meet the requirements and put the necessary files into the resources directory:

   • For example:

     project
     └───resources
     │   │   armhf-rootfs-ubuntu-bionic.tar
     │   │   gcc-linaro-7.5.0-2019.12-x86_64_arm-linux-gnueabihf.tar
     │   │   linux-4.19.94-ti-rt-r41.tar.gz
     │   │   u-boot-2020.04-rc3.tar.gz
   

2. Make sure that you decompress the Linux rootfs to get the right compressed file. The files that contain all the Linux rootfs inside its root directory (e.g. armhf-rootfs-ubuntu-bionic.tar).

3. Set the Dockerfile parameters in docker-compose.yml:

   • kernel_version: kernel version in X.Y.Z format
   • kernel_filename: kernel source file path
   • kernel_defconfig: kernel configuration file
   • uboot_filename: u-boot source file path
   • uboot_defconfig: u-boot configuration file
   • rootfs_filename: Linux Root file system path
   • linaro_filename: Linaro toolchain file path

4. Bring up the console and navigate to project root directory and execute docker-compose up --build.

5. If the process complete successfully you should get the image file sd_image.img in the output directory.

Contribution

Project source code can be found here. Reporting any bugs or suggestions for improvement would be appreciated.