At the end of February, 2025 I wrote a rough application to the German Sovereign Tech Agency to build an open community-based boot loader distribution. The goal of the project is to follow the model of the kernel and Linux distributions to reshape boot loaders and therefore change the foundation of nearly all modern computers.

See the linuxbootchain project on Gitlab. Of particular interest is the documentation in doc/ and the Gitlab issues.

Distribution developers have been trying to get away from packaged based Linux distributions (e.g. docker/podman, ostree, Flatpack/Snap, coreOS, mkosi) for years, but adoption is at best fractured. Embedded systems have been shipping image based platforms since Linux began to become popular. However the tooling is complex and heavily vendor specific. If the history of Linux as an operating system is not taken for granted, we can imagine what it would be like if the same approach was taken for microcontrollers (e.g. Zepyhr) or boot loaders.

If a company today wants to create a product they have to deal with the overly complex mammoth that is Android, or use leaner tools like Buildroot or Yocto that require designing a platform from scratch. Similarly, desktop/laptop users or those working on server infrastructure are faced with a complex and fractured landscape.

There are countless technologies available, but they often require specialized knowledge to use. Boot loaders should be just like the kernel, which provides enormous functionality through a fairly simple configuration system and built/deployed by Linux distributions. A open boot loader community that is responsible for developing, building and releasing binaries like the kernel and various Linux distributions would allow the multitude of features that exist in the community to actually be used, without hours of labor and expert knowledge.

Raspberry Pi was founded by a couple with a educational focus, but always with an eye towards business success. It is now a over £1 billion publicly traded company. Despite imperfect hardware design, investment in outlandish custom IP, and superfluous platform specific software, it succeed because it provided something accessible that built a community.

I would like to develop, build and deploy open source boot loaders to LVFS for embedded boards that allows booting mainline distributions.

Challenges

The proposal does not require writing new software, since it depends on existing projects, but naturally there are a number of challenges combining all of the software in a way that is easy for users to use.

• Developing a Kernel configuration to minimize size while maintaining hardware compatibility and userspace dependencies
  • Kconfig fragments are not more of a hack than a real solution, with small changes require complete rebuilds and no tooling to make incremental changes
  • Identify and describe kernel common configuration options as they relate to the boot loader use case and minimizing the kernel
• When supporting new boards having a setup that quickly allows for making configuration changes and testing the changes. For example, a fully kernel rebuild, reflashing an SD card and inserting it into the target takes long enough to make
• Systemd (optional) configuration to mimize size while supporting necessary features
• Kexec isn’t as well maintained in the kernel as some other subsystems and features. In particular it would be nice to improve hardware initialization debugging including documentation
• Dependency on EFI/ACPI implementations rather devicetree/standard-less implementations. This is mostly avoided by using kexec, but some issues exists (e.g. Fedora installer checks the platform and assumes aarch64 is always EFI)

Goals

Complete a full proof of concept including:

• RockPro64 with eMMC module and NVMe
• U-Boot SPL (Buildroot)
• A LinuxBoot implementation (e.g. Buildroot, systemd, u-root, kexec-tools, …) that is small enought to fit in the 16MB flash (without redudency, which can come later with further optimization)
• Successfully boot the Fedora Anaconda installer image from USB and an official Fedora installation from eMMC and NVMe
• Test Fedora Anaconda installer with new web UI interface for headless embedded boards
• Extensive documentation. The proof-of-concept will implement one boot chain, but there are main valid options and understanding each of them is a very complex topic

From there:

• Basic CI/CD and versioning
• Upload and testing with fwupd/LVFS
• Conference presentation and/or YouTube presentation(s)
• Collaborate with OFSC/OSFF
• Collaborate with userspace (e.g. systemd, mkosi, All Systems Go, …)
• Collaborate with Buildroot
• Add support for additional boards
• Document vendor boot ROMs starting with supported features (e.g. booting from SPI flash to network) and high level secure boot design

Reply from STA

I heard back on March 4 that the application was not eligible. Although it was not explicit, I think it is safe to assume because project is just getting started and they focus on large well established projects. Despite the decision, there was some interest expressed by one of their team members about learning more about the project.