8.5 KiB

Raw Blame History

Welcome to the coreboot documentation

This is the developer documentation for coreboot. It is built from Markdown files in the Documentation directory in the source code.

Purpose of coreboot

coreboot is a project to develop open source boot firmware for various architectures. Its design philosophy is to do the bare minimum necessary to ensure that hardware is usable and then pass control to a different program called the payload.

Separation of concerns

The payload can then provide user interfaces, file system drivers, various policies etc. to load the OS. Through this separation of concerns coreboot maximizes reusability of the complicated and fundamental hardware initialization routines across many different use cases, no matter if they provide standard interfaces or entirely custom boot flows.

Popular payloads in use with coreboot are SeaBIOS, which provides PCBIOS services, Tianocore, which provides UEFI services, GRUB2, the bootloader used by many Linux distributions, or depthcharge, a custom boot loader used on Chromebooks.

No resident services (if possible)

Ideally coreboot completely hands over control to the payload with no piece of coreboot remaining resident in the system, or even available for callback. Given the reality of contemporary computer design, there's often a small piece that survives for the whole runtime of the computer. It runs in a highly privileged CPU mode (e.g. SMM on x86) and provides some limited amount of services to the OS. But here, too, coreboot aims to keep everything at the minimum possible, both in scope (e.g. services provided) and code size.

No specification of its own

coreboot uses a very minimal interface to the payload, and otherwise doesn't impose any standards on the ecosystem. This is made possible by separating out concerns (interfaces and resident services are delegated to the payload), but it's also a value that is deeply ingrained in the project. We fearlessly rip out parts of the architecture and remodel it when a better way of doing the same was identified.

One tree for everything

Another difference to various other firmware projects is that we try to avoid fragmentation: the traditional development model of firmware is one of "set and forget" in which some code base is copied, adapted for the purpose at hands, shipped and only touched again if there's an important fix to do.

All newer development happens on another copy of some code base without flowing back to any older copy, and so normally there's a huge amount of fragmentation.

In coreboot, we try to keep everything in a single source tree, and lift up older devices when we change something fundamentally. That way, new and old devices benefit alike from new development in the common parts.

There's a downside to that: Some devices might have no maintainer anymore who could ensure that coreboot is still functional for them after a big rework, or maybe a rework even requires knowledge that doesn't exist anymore within the project (for example because the developer moved on to do something else).

In this case, we announce the deprecation of the device and defer the big rework until the deprecation period passed, typically 6-12 months. This gives interested developers a chance to step in and bring devices up to latest standards.

While without this deprecation mechanism we could inflate the number of supported devices (probably 300+), only a tiny fraction of them would even work, which helps nobody.

Scope of the coreboot project

coreboot as a project is closer to the Linux kernel than to most user level programs. One place where this becomes apparent is the distribution mechanism: The project itself only provides source code and does not ship ready-to-install coreboot-based firmware binaries.

What the project distributes, even if - strictly speaking - it's not part of the project, is a collection of vendor binaries (that we call "blobs") that are redistributable. They cover the parts of hardware init that we haven't managed to open up, and while some hardware requires them, there's still hardware that can boot without any such binary components.

The build system can integrate them into the build automatically if required, but that requires explicit opt-in and downloads a separate repository to ensure that the distinction remains clear.

There are various distributions, some shipping coreboot with their hardware (e.g. Purism or Chromebooks), others providing after-market images for various devices (e.g. Libreboot, MrChromebox).

If you want to use coreboot on your system, that's great!

Please note that the infrastructure around coreboot.org is built for development purposes. We gladly help out users through our communication channels, but we also expect a "firmware developer mindset": If compiling your own firmware and, at some point, recovering from a bad flash by hooking wires onto chips in your computer sounds scary to you, you're right, as it is.

If that's way beyond your comfort zone, consider looking into the various distributions, as they typically provide pre-tested binaries which massively reduces the risk that the binary you write to flash is one that won't boot the system (with the consequence that to get it to work again, you'll need to attach various tools to various chips)

The coreboot community

If you're interested in getting your hands dirty (incl. potentially wiring up an external flasher to your computer), you've come to the right place!

We have various forums where we discuss and coordinate our activities, review patches, and help out each other. To help promote a positive atmosphere, we established a Code of Conduct. We invested a lot of time to balance it out, so please keep it in mind when engaging with the coreboot community.

Every now and then, coreboot is present in one way or another at conferences. If you're around, come and say hello!

Getting the source code

coreboot is primarily developed in the git version control system, using Gerrit to manage contributions and code review.

In general we try to keep the master branch in the repository functional for all hardware we support. So far, the only guarantee we can make is that the master branch will (nearly) always build for all boards in a standard configuration.

However, we're continually working on improvements to our infrastructure to get better in that respect, e.g. by setting up boot testing facilities. This is obviously more complex than regular integration testing, so progress is slow.

What our releases mean

We also schedule two source code releases every year, around April and October. These releases see some very limited testing and mostly serve as synchronization points for deprecation notices and for other projects such as external distributions.

This approach and terminology differs somewhat from how other projects handle releases where releases are well-tested artifacts and the development repository tends to be unstable. The "rolling release" model of some projects, for example OpenBSD, is probably the closest cousin of our approach.

Contents:

8.5 KiB Raw Blame History