84 KiB

Raw Blame History

% Libreboot 10-year anniversary % Leah Rowe % 12 December 2023

I'm very proud of the work done in Libreboot, both by myself and by others who I work with. Many people make Libreboot possible, be it direct contributors to the project, or the countless individuals and companies that work on all the upstream projects used in Libreboot, projects like coreboot. I'm very pleased now to announce the 10-year anniversary of the project.

I've decided to celebrate this event, by reflecting on the history of the project during all these years. This article will go into depth about Libreboot history. This article complements the pre-existing contrib page, which attempts to list the numerous contributions made by people to Libreboot, over the years.

Introduction

The Libreboot project provides free, open source (libre) boot firmware based on coreboot, replacing proprietary BIOS/UEFI firmware on specific Intel/AMD x86 and ARM based motherboards, including laptop and desktop computers. It initialises the hardware (e.g. memory controller, CPU, peripherals) and starts a bootloader for your operating system. Linux and BSD are well-supported. The goal of Libreboot is to provide well-tested binary releases of coreboot, along with a fully automated build system, so as to make coreboot easy to use for non-technical users, who believe in the principles of the Free Software movement - Libreboot is also more secure and generally works more reliably than most vendor firmware.

As of today, Libreboot is 10 years old. The very first release was Libreboot 20131212, released on 12 December 2013, and today is 12 December 2023, exactly 10 years after that first release.

I actually wasn't sure what to do for this day, and I contemplated doing nothing at all. In the end, I've decided to simply acknowledge it, and reflect on the last 10 years of history. So, I spent all of today writing this article, which is being published just before midnight, in the final hour of 12 December 2023, relative to UK timezone which is (on this day) UTC+0000.

You can otherwise read the Libreboot git repositories, the contrib page, and you can find historical articles all over the internet pertaining to the subject. This article, that you are reading now, is written solely from my own perspective, through my eyes, describing project history the way I see it, since I've been here all this time.

Let's get started!

How did Libreboot start?

Why did it start?

I'm a free software activist. First and foremost, my focus is on using as much Free Software myself, and I want to help other people achieve a level of software freedom themselves. Many people reading this article today will know that Libreboot no longer strictly aligns to FSF policy, but I still very much believe in the ideology behind it. I merely apply it in different, more pragmatic ways - but this page is about the history of the project, so I have to put myself in the headspace I was in, during each development year relative to when Libreboot started.

Early history

The name Libreboot actually existed from about February 2014 onward. The libreboot.org domain name was registered on 26 January 2014. I initially started Libreboot as a commercial project, then unnamed, hosted on my company website (nowadays called Minifree Ltd). I sell computers with Libreboot preinstalled, but I started providing coreboot-based services in mid-2013. People started noticing what I was doing around August or so, in 2013. Back then, Libreboot did not exist, not even anything that I later referred to as Libreboot, I just provided coreboot pre-compiled and pre-flashed.

One of the people who noticed me, reported my company to the Free Software Foundation who were, at that time, interested in providing a Libreboot-like project. They approached me, asking if I'd like my company to be promoted, if I complied with certain policies of theirs. Long story short, I was interested and turned my work into a formalised project. I still remember who told the FSF, but I won't say his name, for the sake of his privacy. It was Joshua Gay, FSF's licensing and compliance manager, who initially contacted me, expressing his own excitement about the prospect. Joshua is the one who initially reviewed Libreboot for compliance, and worked with me on getting Minifree products to be RYF-endorsed; RYF is the FSF's certification program, for companies selling free software oriented hardware. Joshua also taught me a lot about licensing, especially the pitfalls of the GPL when it comes to compliance. He taught me what to watch out for. Many of Libreboot's practises today are still inspired by his initial guidance, for which I remain grateful; he was an excellent teacher.

When I first started this work, coreboot had already existed for 14 years, having been founded in the year 1999 as LinuxBIOS, renaming to coreboot in the mid-2000s. However, coreboot wasn't (and still isn't) made for end users. Coreboot's infrastructure is intended for use by other developers, who work on boot firmware. Nothing user-friendly (for regular non-technical users) existed.

When Libreboot started, the first goal was to solve this problem. Libreboot's mission today is exactly the same as it was then: to provide free boot firmware that normal people can easily use. Something that normal users don't have to think about, that Just Works, with regular updates. Clean, simple and professional, with clear documentation. Coreboot documentation today is much better, but it was very poor quality in 2013 (MediaWiki site with lots of errant pages, poorly organised).

It came with a twist: Libreboot initially complied with the FSF's own RYF criteria, and the GNU FSDG (Free System Distribution Guidelines). Under this policy, no binary blobs are allowed. I believed in it then, because in practise it seemed like a good idea and didn't do any harm. It would later become a liability, for reasons laid out in Libreboot's modern Binary Blob Reduction Policy - a policy that was enacted during November 2022, in the Libreboot project. This meant that Libreboot could only support a small subset of hardware from coreboot because, although it is nominally a free software project, coreboot does require binary blobs on some boards. More on this later!

Since the project was nameless until February 2014, early promotion of my work was talking only about my company and about the FSF endorsement of it; Minifree used to be called Gluglug, hosted at gluglug.org.uk (you can check it in the Wayback Machine. The shop itself was on shop.gluglug.org.uk). For example, this video talked about the ThinkPad X60 running Libreboot 20131212, but I only retroactively named that to Libreboot in February 2014, so that video is talking about my company, which is where you could download my source tarballs at the time.

2013-2014: early days

When Libreboot first started, it only provided very basic releases: pre compiled ROM images, with source code, but there was no automated build system. The build system evolved over time.

The first releases of Libreboot supported only the ThinkPad X60. Later, T60 support and MacBook 2,1 support were added. This first year of the project was spent on building solid infrastructure, specifically documentation and an automated build system. The design of Libreboot revolves around scripts that automatically download, patch, configure and then compile various codebases, so you can think of it like a coreboot distribution; Libreboot's build system is essentially a source-based package manager, for coreboot components.

The very first Libreboot releases only provided a GNU GRUB payload. The GRUB bootloader had been adapted for coreboot many years before then, but there weren't any ready-made coreboot images available that used it, and it wasn't very well-developed for end users. GRUB is a lot more complex than many people realise, containing a small kernel, shell, utilities, even its own subset of a standard libc. It even has its own malloc functions, and it's quite decent. Then there are modules, basically small programs that run inside GRUB. The GRUB shell even has things like cat and ls, adapted for it. GRUB configuration files behave similarly to BASH scripts, with similar syntax and overall logic flow/style. It is an extremely configurable bootloader.

GRUB revolves around the shell, which is basically an extension of the Bash shell, adapted for use in a Multiboot context. GRUB is the reference Multiboot implementation, which boots Linux. Libreboot provides GRUB so that it can directly boot Linux kernels. GRUB has many advanced features, such as GPG and LUKS support, and it has support for virtually all of the major file systems in use today.

Libreboot logo

Of note, though it doesn't fit anywhere else in the article, so I'll just say it now:

I went on the GNU mailing lists during 2014, asking people to make a logo for the project. Several people then emailed me their submissions. I finally picked the logo design that Libreboot now uses, and has used since then. That design was created by a person named Marcus Moeller, for use in the Libreboot project.

Why GRUB?

Libreboot has always, at least on x86, preferred GRUB as a coreboot payload.

Libreboot's initial no-binary-blob policy meant that SeaBIOS could not be used. SeaBIOS then was the most popular payload, and the default one in coreboot, at least on x86. Although SeaBIOS is free software, it did not have a free video BIOS back then. Coreboot's native video initialisation was developed at around that time, for Intel video chipsets, providing a basic framebuffer, but no video BIOS services, so operating systems like Windows could not boot (and they still cannot, on most Libreboot setups).

Nowadays, SeaBIOS comes with a feature called SeaVGABIOS, which piggybacks off of the coreboot framebuffer, providing a basic video BIOS and initialisation on top of the coreboot framebuffer, but this capability did not exist in 2013 when Libreboot started. So, SeaBIOS could not be used in Libreboot then. Libreboot provided just the coreboot framebuffer, then loaded GRUB, which then loaded Linux; Linux itself could interact with the coreboot framebuffer, and the Intel video driver (i915) sets modes directly, so it doesn't need BIOS INT10H or UEFI GOP services for video-related functions. In Linux, your video drivers just write to a framebuffer directly in memory. No abstractions. This feature, called KMS (Kernel Mode Setting) was later ported to the BSDs, first made stable in 2016; that means, that when Libreboot first started, it could only boot Linux.

Linux as a payload wasn't very well developed in practise, even though Linux in flash was the original aim of the coreboot project, and the systems that Libreboot supported in 2013 only had 2MB of flash, which could not fit Linux; the Heads project did not exist either! However, GRUB could (and still can) fit in such a tiny space. In fact, even to this day, a compressed GRUB payload can be optimised to fit within 512KB, though the Libreboot version adds a few additional modules so the compressed size in flash is about 600KB (at least for configurations that were provided on 12 December 2023).

Tianocore did exist then, but Tianocore is bloat. It has always been policy not to provide Tianocore, because of this. Code==bugs, and Tianocore is extremely over-engineered; the UEFI specification is completely redundant, because BSD and Linux kernels run on bare metal anyway (without making use of any BIOS or UEFI services). Linux/BSD video drivers set modes directly, without needing to use BIOS/UEFI-based services e.g. BIOS INT10H or UEFI GOP.

Other payloads weren't really suitable either, so: Libreboot provided GRUB right from the start. GRUB provides a special build target for coreboot, where GRUB provides all of its own drivers and runs on the bare metal, without relying on abstractions such as BIOS or UEFI.

GRUB wasn't very useable either. Back then, it was basically assumed that if you used the GRUB payload, you just had to reconfigure it in flash, every time you wanted to boot another Linux distro, or otherwise manually boot from the GRUB shell. This was in fact how GRUB was used. However, GRUB otherwise worked perfectly, and had (still has) support for the coreboot framebuffer, and text mode. It is the preferred payload, for our purposes. It was there and just worked. Libreboot then spent a lot of time tweaking GRUB, configuring it so that it is as user-friendly as possible, roughly matching the user experience that is found on most BIOS/UEFI systems, at least for booting Linux and BSD.

Libreboot's innovation was that it made GRUB user-friendly as a payload. Over time, GRUB was configured in clever ways. These days, Libreboot's GRUB is well-configured and can auto-boot almost any Linux distribution. The way it works is by systematically scanning for GRUB configurations on many different partition layouts, including encrypted (LUKS), to then load a GRUB config provided by the installed distro. GRUB also has a parser in it, that lets you show ISOLINUX/SYSLINUX menus, converted on the fly for use with GRUB. Nowadays, many UEFI-based distros also provide GRUB on installation media, which means that many Linux distro installers come with GRUB, making installers even easier to boot from Libreboot.

Libreboot's GRUB configuration is unique. Even with coreboot's earlier work, nothing quite like Libreboot's GRUB config exists that pre-dated it. The way Libreboot configures GRUB is extremely unique. Several other projects now exist inspired by it, such as Canoeboot. GRUB also has several flaws, but it mostly Just Works. Libreboot made it work well, for the casual coreboot user.

With the Libreboot version of GRUB, you can normally use it intuitively, where most users who are used to UEFI/BIOS systems can just use the GRUB payload, without knowing much about it. In fact, many users are (to this day) still surprised that they have GRUB in the flash. They install their favourite Linux distro, unmodified, and it just boots, needing no manual intervention or customisation whatsoever, so long as the distro provided a GRUB configuration during installation. Libreboot did that!

The only real downside is that GRUB isn't very good at booting BSD systems, though it does have BSD partition support and it has code tailed to specific types of BSD kernels, but it usually cannot be relied upon; that said, GRUB does have GELI support and can load a FreeBSD kernel. If configured properly, it might work. However, Libreboot provides SeaBIOS these days and SeaBIOS is what is recommended to boot BSD systems in Libreboot, where you would then execute the bootloader provided by your BSD system.

Another benefit of using GRUB (as a payload) is that, when installing Linux, you no longer need to install anything bootloader-related on your HDD/SSD. Although excluding this usually requires re-configuring grub.cfg in flash, Libreboot does provide this capability.

Automatic build system

One of Libreboot's other main innovations, in addition to providing binary releases, was to provide an automated build system interface. With it, the user could run a few simple commands, to build entire sets of ROM images.

By contrast, configuring coreboot on your own has always been quite complicated. The average user would look at coreboot and become overwhelmed. It's easy to just run make menuconfig in coreboot, pick a board, come out of that and type make, but this only gets you a very basic setup (with SeaBIOS), and this assumes that no other components are needed e.g. extra firmware that coreboot does not provide. Especially for more modern systems, there are many moving parts that the user otherwise has to deal with, whereas Libreboot automates all of it. Such automation did not exist anywhere, prior to Libreboot. The only noteworthy exception would have been the Chromebook products, where Google did(does) provide updates, but without so much community focus as Libreboot and without quite the same user-accessible build system automation; their infrastructure is designed so that vendors can sell more Chromebooks.

Libreboot takes everything needed for each board, auto-downloading, auto-patching, auto-configuring and auto-building the various software components. Nowadays, this automation is very sophisticated, and can be done with only a single command per board.

During 2014, this automation was already established, though it wasn't as well-developed then, and a lot of manual user intervention was still required; in other words, it wasn't full automation. It was more like 5 commands instead of 1; but coreboot was (still is) like 15+ commands, depending.

A lot of Libreboot users today can't easily appreciate how much easier Libreboot makes everything for them, because of how entrenched it is, at least on what it supports. There are many coreboot users today, who have only ever set it up by way of Libreboot installation. Many such users are people who would never use coreboot on their own. Libreboot's build system just works, to the extent that people today even take it for granted. It works so well that you forget it exists, which is the essence of all good user interface design.

Before 2014, nothing like Libreboot ever existed. Coreboot existed, but nothing that is easy to use based on it. Nothing quite as developed as Libreboot, for end users. Libreboot's main innovation was (and is) precisely this, that it automates the process of building ROM images, and then provides releases.

Everything in Libreboot is automated, and that is the entire purpose of it, to provide such automation. It's the same concept as a Linux distribution, but in the context of boot firmware. A project like Debian automatically builds many codebases from all over the internet, putting it all together to provide an operating system that users can simply install. Libreboot applies that same concept to coreboot, so it is a coreboot distribution.

Libreboot was the first coreboot distro ever, and today is still the most popular one. Several similar projects now exist, inspired by the example that Libreboot had set, all those years ago!

You can learn more about the build system by reading lbmk documentation.

late 2014-2016: expansion

Libreboot up until late 2014 was mainly a proof of concept. It worked, but not well. The precursor of what became Libreboot's automated build system, began development from early 2014 onward.

Almost all of 2014 was spent entirely on the build system, and on documentation, without much additional focus on hardware support; I always knew then that it would later expand, so I focused solely on the infrastructure.

GM45 hardware (e.g. ThinkPad X200)

I was approached by another developer, named Steve Shenton, around October of 2014. I'd expressed interest in the ThinkPad X200, as an upgrade to the X60 offering in Libreboot. Coreboot supported the X200 at that time, done by coreboot developer Vladimir Serbinenko based on Nico Huber's work on the Roda RK9.

The Intel Management Engine was the only issue. At that time, nothing like me_cleaner (which strips out code from the ME, usually leaving only the bringup code intact) existed. On the GM45 machines, Intel ME is usually disabled on most vendors, but Lenovo provided it because their machines came with Intel Gigabit Ethernet, which requires an Intel Flash Descriptor and Gigabit Ethernet config in the flash; Intel's standard setup here is to then provide the Intel ME, in two separate versions: a 1.5MB version without AMT, and a 5MB version with AMT.

The X200 was always possible to use without Intel ME, because the ME only handled AMT and TPM on that platform, which are not required for boot, but the only known way at that time was to flash a descriptorless setup, which meant that you would lose use of the Intel Gigabit Ethernet. This didn't seem all that useful for Libreboot users, who expect everything to just work.

Steve reverse engineered the Intel Flash Descriptor, finding disable bits in it that put the Intel ME into a permanent reset loop. In this configuration, the ME bootrom does not load any firmware from the flash. The full ME firmware otherwise goes in the main boot flash, alongside the BIOS software (e.g. coreboot). In Libreboot, the Intel ME firmware is excluded on GM45 hardware.

I worked with Steve between October 2014 to mid-January 2015. Steve provided instructions for disabling the Intel ME in the ICH9M flash descriptor, and wrote a proof of concept utility that would take a factory ROM image dump, for instance X200 LenovoBIOS dump, and create a file containing just the descriptor and Gigabit Ethernet config, but without the Intel ME. That utility was called ich9deblob.

Under Steve's guidance, I wrote a utility that would generate an ICH9M descriptor from scratch. I then studied the datasheets for Intel's gigabit NIC, writing a tool to generate the Ethernet config from scratch. I had a new program written by around mid-January 2015, and I called it ich9gen. You can read about ich9gen and ich9deblob on the ich9utils page. The ich9gen utility was used to provide blobless configurations on the ThinkPad X200 machines.

Based on this, and extensive amounts of documentation was written during the same period, I then provided the Libreboot 20150124 release on 24 January 2015. It added ThinkPad X200, X200S and X200 Tablet support.

2015: T400, T500, W500... R400

Coreboot lacked supported for other GM45 thinkpads at this time, when X200 support was first added to Libreboot. Testing was later done on these other machines. In particular, Timothy Pearson (of Raptor Engineering) added support for switchable graphics, enabling video initialisation on versions of these boards that have both Intel and AMD graphics. Without such support, no video init was available unless you used a board that only had the Intel GPU. A hardware mux is present on the ones with AMD graphics, enabling you to still use just the Intel one, or switch to the AMD one.

So, during 2015, Libreboot added support for many newer laptops. ASUS KCMA-D8 and KGPE-D16 support was also first added to Libreboot during this year, worked on by Timothy Pearson and integrated into Libreboot.

2015-2016: Major build system improvements

The build system was heavily improved during this period, improving the automation and greatly expanding the supported features.

During this time, Paul Kocialkowski got involved with the project and used his expertise to integrate Depthcharge as a payload. This is the bootloader used on Google Chromebooks; he has previously worked on Replicant and several other projects. The ASUS Chromebook C201 was added during this time as a proof of concept, which is a Rockchip RK3288-based machine (ARM). SeaBIOS wasn't available, nor was GRUB(at that time, for ARM). Libreboot today uses U-Boot (adapted for use as a coreboot payload) on ARM machines, but that work hadn't been done in 2016.

In general, Libreboot's first years (2014-2016) were spent writing extensive amounts of end user documentation, and implementing the automated build system, so as to support many machines. Libreboot's design is kept as efficient as possible, so that it can run on very inexpensive infrastructure, with many more releases; the main focus has always been to further extend the number of machines that are supported, and to reduce the amount of maintenance required, so I try to keep everything very simple.

Several other desktop boards were also added during this period, such as the Gigabyte GA-G41M-ES2L board that was ported to coreboot by Damien Zammit. He was compensated for his work; I paid him 4000 AUD for the trouble. Arthur Heymans contributed several ports aswell, and today still does extensive work on coreboot - both Damien and Arthur left the Libreboot project, after its departure from GNU (more on this next).

This period was a watershed moment in general, for the Libreboot project. Many people contributed to the project, during this period. You can check the contrib page for a list of people, and check git logs.

2016: GNU Libreboot

Yes. That is also covered here. Naturally, Libreboot was becoming very popular and I looked for ways to extend that. Back in 2015 and most of 2016, I saw GNU membership as a natural step for the project to take. Libreboot had a mailing list at that time, hosted on the GNU Savannah infrastructure, so it seemed only logical to give it a go.

Evaluation and accession

Libreboot was briefly a GNU project, between 14 April 2016 to 15 September 2016. I had proposed this during the summer of 2015, and worked with the GNU Eval team lead by Mike Gerwitz.

Libreboot's build system and documentation used a very peculiar design, which did not fit well into GNU's design policies. GNU had several technical requirements for projects, such as use of Texinfo files for documentation, to be generated into HTML by a static site generator. At that time, Libreboot's website was written in static HTML, manually. It did not use a static site generator of any kind.

On the coding side of things, many GNU programs use things like GNU Autoconf for configuration, in build systems, and GNU has a certain coding style that is preferred. To speed up entry into GNU, many of the requirements were overlooked since, ideologically speaking, Libreboot aligned with the GNU Free System Distribution Guidelines (FSDG) at that time, so Libreboot officially joined GNU on 14 April 2016, under the assumption that it would gradually start integrating with the GNU infrastructure.

Three Libreboot releases were released, during its GNU membership, namely: Libreboot 20160818, 20160902 and 20160907.

Initial outcomes

Libreboot's popularity reached great heights during this time, greatly expanding and attracting many new developers. Joining GNU accelerated this further, though it came with certain drawbacks.

At that time, I believed in the GNU FSDG policy, and I saw GNU membership as beneficial to the Libreboot project, not because I was interested in GNU for any technical merit; Libreboot is boot firmware, whereas GNU has always been about providing an operating system with the libc and many useful applications. Libreboot did not really fit into the GNU infrastructure, but it aligned with the ideology both in theory and in practise, and the prestige of it at that time would attract more help.

I later realised otherwise, regarding the ideology; more on this later, in the sections about osboot, and the osboot/libreboot merge, and events after that.

Regret

Anyway: during Libreboot's membership, I started to realise that many of the technical requirements imposed on Libreboot were unworkable, especially the GNU coding standards, and in general how a GNU project is supposed to be run, which turned out to be quite inflexible for the Libreboot project. GNU projects tend to be highly collaborative, with decisions taken on a collective level, and not just on behalf of the given project, but with thought also given to how it integrates with the rest of GNU; the assumption is that there would be a complete GNU system, everything all integrated very tightly together. Libreboot wasn't (and isn't) like that, because the knowledge is highly specialised and it pertains to a lower level than GNU (that of the boot firmware).

RMS, a man who had no technical merit to lead the Libreboot project, began trying to tell me what to do. I started noticing that the GNU project was not run very democratically, or transparently either; as a maintainer, I had access to the gnu-prog-discuss mailing list, which is otherwise unavailable to the public, and I saw the true nature of it. Individual projects, even maintainers, had little to no oversight or say in how RMS or the GNU Advisory Committee made decisions, but they could impose policy on projects. In practise, many projects ignore the leadership of RMS and the committee, and they don't use GNU's stagnant infrastructure such as Savannah. Most modern GNU projects have their own management, their own infrastructure and their own practises, where their only connection to GNU is the prestige of the GNU brand name itself. However, that mailing list existed, and RMS (plus others) used it to leverage other GNU projects. Prior to joining, I saw it as a great and noble organisation, but my experience as a maintainer within it showed me that it was actually just the personal fiefdom of a very small group of people, lead by RMS.

I could begin to see a future very clearly where I was no longer in control of the Libreboot project. I'll provide an example: one of the last threads I saw, before being kicked off of the list due to Libreboot's departure, was a thread started by RMS where he told all the GNU projects that they should start implementing reproducible builds, in all of their build systems; this is a noble goal in and of itself, but it was started based on the whim of one man, who would not implement the policy himself. RMS has no coding experience anymore, because he mostly stopped doing any real programming ever since the 90s, yet he had the audacity to try to claim authority over my project? He does not even know how to install Linux! - Libreboot's membership was a gift to GNU, and I don't try to rub things in people's faces, but he owned me a debt of gratitude. Instead, he showed me contempt. However, I cannot reveal the emails due to privacy concerns (I would never publish without consent).

Libreboot is a lifelong passion of mine. It's one of the best things I've ever done, and I have no intention of ever letting it go, something I realise could happen with GNU membership. There is another freedom that exists, namely the freedom to say no. I reclaimed this freedom, by removing Libreboot from GNU.

I realised that if Libreboot stayed in GNU for too long, then I would start to lose control over the project as it became more associated. I had been deciding already, early into Libreboot's membership, whether to reverse the decision and take Libreboot out of GNU, thereby making it an independent project again. You could try to assert control by doing good work, but eventually everything grows, and who will the other people who join answer to; you? No, I saw definite coup possibility in the future. RMS could have put someone else in charge, and I'd seem ungrateful if I resisted or challenged it, people would side with him.

I know I'm right, and I don't regret the fact that Libreboot left GNU; I do regret how it left, and I'll have more to say on this in the next section. I had the wisdom then to know what might happen in the future. I'd seen it before, with other projects (for example GnuTLS, whose founder and author did all the work and later took it out of GNU - something that RMS tried to prevent, and only failed by chance). I did initially offer for Libreboot to re-join GNU, mostly because I felt regret about the manner in which it left, but not the fact alone that it did leave. Again, more on this will be written in the next sections. My re-join proposal was neither accepted nor rejected; RMS simply ignored it, and did not respond. I offered for it to re-join because I felt bad about everyone else I'd affected during its later departure, but I'd hoped RMS would say no. So I gambled on it; making such an offer reliably absolved me of any guilt, mitigating my past actions entirely, but he didn't have to accept it, and he didn't. If he had accepted it, Libreboot in 2023 would be unrecognisable today, at least compared to the Libreboot project that does now exist, in 2023.

So I thought at that point: was Libreboot's membership really worth it? I concluded that it wasn't necessarily a bad idea, not a liability for the project, at least in the short term. I basically continued running the project the same way I always did, but under the GNU name, while being wary of my decision. I had begun to contemplate whether I should remove Libreboot from the GNU orbit.

In retrospect, I believe Libreboot's membership was always doomed from the start. The way I operate, and the way I think, is incompatible with their style of leadership, and with the expectations put upon GNU projects. Actually, I only half-expected that they'd even accept Libreboot, at the time. I didn't even know it had become a GNU project, on the day it became one, which was 14 April 2016; I was quite preoccupied with many other things during that month, and I only noticed RMS's acceptance email on the 18th of April in that year. He just sent the email on a whim, and Libreboot had become part of GNU, just like that - even in the days after the 14th, but before 18th, I had been thinking about the changes that Mike Gerwitz wanted me to make first, which had even been discussed during that time, only for RMS to sidestep all of it and accept Libreboot early; yet another example of the power that he held.

Needless to say, I was stunned. And now, I will write about Libreboot's subsequent exit from the GNU project:

Exit

Later, a member of the FSF staff got fired, who happened to be transgender, and I had been friends with that person. I myself had only recently come out as trans (I came out publicly as trans during May 2016), and I wasn't very secure about myself at that time. Needless to stay, I wasn't very stable, and I took the friend's firing personally. I publicly accused the FSF of firing the person on discriminatory grounds, accusations which later turned out to be false, for which I apologised, but this went on for about 6 months.

That incident alone wasn't solely my reason for removing Libreboot from GNU, and I probably would have done so anyway, but it served as a sufficient catalyst that motivated me into action, though the way I handled it was regrettable. Rather regrettably, I went on an angry tirade against the entire apparatus of the FSF and the GNU project, denouncing them both, an act which I know now made no sense whatsoever, because only one person was responsible for the firing; and even if my diatribes were justified, they did not belong on the Libreboot website. I should have put them on a personal blog, not the homepage of a major free/opensource software project that I'd poured 3 years of blood, sweat and tears into. What I did, nearly destroyed the Libreboot project.

Because of that incident, I announced on 16 September 2016 that Libreboot is no longer part of GNU. GNU did not officially have a mechanism for projects to leave, and that's still true to this day. It has an evaluation process for projects to join, and then integrate with the GNU infrastructure. They do have a tendency to list projects as unmaintained or decommissioned if no longer developed within GNU, but they do not have language pertaining to active projects that leave, that stay active.

GNU's response

Libreboot's decision to leave was not without precedent. GnuTLS also left, as did Nano (Nano later re-joined), though GnuTLS's departure was never officially acknowledged; it's still non-GNU to this day, but the GNU project still lists it as an official member project. GNU initially did not accept Libreboot's departure. The GNU Advisory Committee, under Stallman's leadership, went around trying to find people who could take my place as leader of the project. In other words: they had the goal of keeping GNU Libreboot going, despite my intentions as leader of the project. I regarded this as their intent to create a hostile fork of Libreboot, under the same name, and I still hold them to account for it today, even if other actions that I did take were regrettable in the way that I did them. They had no moral or technical authority over the project.

In the event, nobody else wanted to take my place under GNU auspices, and RMS finally acknowledged Libreboot's departure, during January of 2017. At the time, I wore this badge with pride. GnuTLS had left, but their leaving was never officially acknowledged, and was in fact opposed by the GNU Advisory Committee; it is even still listed today as a GNU project, when it has de facto not been that way for many years. RMS published this acknowledgement, on the GNU mailing lists, for everyone to see.

I still, to this day, believe that Libreboot's membership was flawed. It did not need to join in the first place, regardless of ideology. It was already thriving as an independent project, but GNU membership made no sense due to the very nature of the project: as an aggregate of other codebases, it could be anything at any given time, which would make auditing very difficult. It'd be like Debian joining GNU, to become GNU Debian. GNU Debian GNU/Linux. This may seem hilarious, in fact, but consider that Libreboot provides a modified version of GRUB, which itself is a GNU project.

GNU Libreboot just didn't quite feel right, on any level, but this realisation was only made after the fact. If there are infinite universes, ours is the only one that had GNU Libreboot.

An alternative might have been to just provide a deblobbed coreboot in GNU, which is something I also considered at the time, but that would have required maintaining a full coreboot repository, merging and overseeing a lot more patches from upstream, and diverging heavily. The way Libreboot's deblobbing worked was that it just deleted the blobs, and (by way of configuration) avoided all boards except those that could be booted entirely blob-free in the flash. This method required far less maintenance - the original blob-free policy of Libreboot has been continued, as of 2023, in a new project I maintain called Canoeboot - I'll have more to say about this, and Libreboot's newer Binary Blob Reduction Policy, later in the article.

2017-2021: Post-GNU years

GNU cuts Libreboot loose

Libreboot's departure from GNU came about 1 week after a big stable release, the Libreboot 20160907 release. I pretty much burned out after that release, especially during the Librexit saga (which is what I called Libreboot's departure from GNU, in my head, at the time).

The Libreboot project pretty much ground to a halt, until around April or May of 2017. I had a lot of other things going on in my life at the time, especially transition-related (I was preparing financially for gender reassignment surgery in Thailand, which was to be performed on me on 1 August 2018 - I did in fact go there and get it done, on that day, but I will not talk about it further in this article).

In any case: tensions calmed, after January 2017 and the dust settled. I remained angry, bitter in fact, but I did calm down after a while. I can look back at these years now with pride; I did the right thing, even if I caused a lot of trouble while doing it, trouble that I had to later correct, and atone for. It's a bittersweet realisation, but it is what it is.

Apology to GNU

Alyssa Rosenzweig, who later founded the Panfrost project and nowadays works on free graphics drivers for Asahi Linux, briefly joined the Libreboot project during mid-2017. She had a stern talk with me, telling me the folly of my ways. I'd already been reconsidering my actions, but her talks with me made me finally see sense.

I learned by April 2017 that my allegations against the FSF were false. The employee was fired on other grounds. Even if my allegations were true, my public reaction to it was wholly inappropriate. I used my position then to attack the entire FSF and GNU apparatus, when in reality the firing had only been decided upon by one man, namely John Sullivan who was the FSF's executive director. Even if my allegations were true, I did a lot of damage to the reputation of the FSF and the GNU project, and to Libreboot, and although it wasn't my intention at that time, it brought a lot of undue attention upon the person in question, who only wanted privacy.

I felt bad about all of it, so I publicly recanted my allegations on 2 April 2017, and stepped down as leader of the project. You can read about the event here: https://www.reddit.com/r/linux/comments/639ob1/libreboot_no_longer_opposes_the_gnu_project_or/

Also of note: Alyssa contributed to Libreboot a custom-written static site generator, converting its static HTML into Markdown files for documentation on the website, then generating pages with Pandoc. This static site generator is included in the original Libreboot git repository, and it later became the basis for my own Untitled Static Site Generator which libreboot.org now uses. Alyssa acted as a sort of spokesperson during this period, acting in my place and I briefly gave her sysadmin privileges for the Libreboot hosting infrastructure, during that year (2017), because she was helping me improve my infrastructure - for example, she taught me how to use Nginx, where I'd previously used lighttpd as Libreboot's web server.

And that is the greatest irony of all. One of my main concerns about Libreboot's GNU membership was that I would lose control of the project, but I soon stepped down and handed it over to other people, after its departure; I'll have more to say about this, in later sections. I was absent from leadership during 2017 to 2021, though I retained exclusive control of the domain name and infrastructure, while fully delegating all day to day development, maintenance and user support to other people - I later took over the project again in 2021, a topic which I will also cover, later in this article.

2017-2021: the great rewrite

Initial work, and why

While Libreboot was a member of the GNU project, I did a thought experiment: what if Libreboot's build system and general infrastructure did adhere to GNU standards?

I attempted to answer this, by actually doing it. I commissioned work, on the Libreboot mailing list (which existed at the time, and was hosted by GNU), to convert the website's static HTML files. Work was underway for converting the entire documentation to Texinfo files, GNU's preferred format for documents. This was later scrapped, when I decided to accept Alyssa's work instead; she converted the website to use Markdown instead, with a custom static site generator.

Paper build system

I simultaneously commissioned a re-write of the build system. There wasn't anything seriously wrong with the one I wrote, but I wanted to see if it could be done better, so I went along with this. Paul Kocialkowski began the work. Paul was at that time a Libreboot developer. He is listed on the contrib page.

Paul's build system, the so-called Paper build system, was provided by him as a proof of concept. It initially only supported CrOS-based devices, for example the ASUS Chromebook C201, integrating coreboot, Depthcharge and various ChromeOS utils that were needed.

The goal of the build system re-write was to make Libreboot much more flexible and generally more configurable. Libreboot's build system worked very well at that time, but a lot of functionalities were hardcoded; the entire design was basically geared towards providing a useful sane default configuration per board, in the quickest way possible, so as to provide releases for end users. Reconfiguration required code modification. It was basically similar to the suckless design philosophy.

I was initially impressed by Paul's work. He got a proof of concept done in only a few months, writing thousands and thousands of lines of code. Indeed, his build system design was much more complex. Too complex, but its flaws would not be realised until later.

All development history on Paper can be found in the old pre-2021 Libreboot git repository, here (roll back by a few commits to see the code, because the code was nuked in the last commits, adding a notice that the repository had been split into multiple smaller repositories): https://notabug.org/libreboot/obsolete-repository-preserved-for-historical-purposes/

I keep that repository there for archival, but it is no longer developed. I took over the project again in 2021, and scrapped the rewrite. More on this later in the article!

Andrew Robbins and Sebastian Grzywna

Sebastian had joined the project during early 2016, advising about hardware and he made quite a few useful code contributions at first. For example, he added support for 16MB IFD configurations in ich9gen.

Some time around mid-2017, another person named Andrew Robbins joined the Libreboot project. His ambition was to expand upon Paul's work in the Paper build system, adding support for x86 machines.

As we discovered very early on, the Paper design was very, very complex and required a lot more heavy maintenance than the original Libreboot build system. The Paper design was the antithesis of the original Libreboot design (now named lbmk, but back then it was just called the old libreboot build system).

The Paper design was implemented with the assumption: the user must be able to reconfigure absolutely anything, without ever touching code. Sort of like when you have a game engine, and then a scripting engine and editors, and the like, in videogame development. Everything completely modularised, where one component could work independently of the other. If I had to make an analogy, it was sort of like the monolith vs microkernel argument, paper being the same sort of mentality for the latter.

I was AWOL for years

I wasn't active in the project during these years, at all. I first was prepping mentally and financially for gender reassignment surgery scheduled for late August 2018; then I spent a year recovering from that. And lots of other things. A lot of events, most/all of which won't be covered in detail here, were going on in my life, such that I never had the time nor mental energy to get involved in Libreboot. And then the COVID pandemic threw me off for a further year. I was planning a takeover of Libreboot in early 2020, but ended up doing it in 2021 instead. I had discussed the possibility publicly, several times during 2020, but nobody took me seriously at the time.

During my hiatus, I did still keep up to date with developments in the project. I occasionally engaged in code review, helping out where I could, but I wasn't in charge. One of the problems I had during this period, was that the other people I put in charge were regularly hostile to my ideas, and seemingly did not want my help at all. They wanted to maintain everything themselves, their way. Which, fair enough, and for a while they did actually do a lot of work. Good work, even. The problem was not their competence, or passion, both of which they had in abundance, but the build system design that they were working on was inherently unmaintainable; to put it most succinctly, it was crap. But they wanted to do that, and I wasn't ready to fully take over again, so I left them to it, hoping they'd prove me wrong.

When I stepped down, the project had adapted a formalised "democratic" governance policy, implementing a horizontal hierarchy form of collective management. In practise, the former BDFL leadership under me was replaced by basically the same thing, only it was two people; Andrew and Sebastian called all the shots. They would regularly turn away code contributions, and Sebastian in particular was often extremely rude to users on the IRC channel, acting in a very elitist manner.

Another general problem that Sebastian's leadership had was, that everything was always later. You want X new feature added? Wait until after next release. When's the next release? Soon. That account of events is quite reductive, but it more or less sums everything up perfectly. The 2017-2021 leadership under Andrew and Sebastian had no focus. It was overly ambitious, trying to focus on and perfect everything, but the scope of the work required at least 5 developers not 2, but those extra 3 developers weren't welcome in the project. They tried to control everything themselves. It was their Libreboot. They were trying to do what I do: lead a project. But they weren't competent at leading it. The total lack of releases during that time is evidence of it. I provided them with all the infrastructure they needed, funding too (from my own pocket), they had an existing prestige (Libreboot was and still is very popular) to fall behind, all they had to do was finish their work. They never finished it, because the design of the Paper build system was unworkable.

But that's how it all was, for many years. I regard Libreboot as having been a dead project, during this period. All (all) of the work put into it at that time, was a complete waste of energy and effort.

Paper never worked

The rewrite largely failed. During 2017-2021, Libreboot no longer had any releases. The Paper build system never reliably built any ROM images, due to its design; it wasn't actually designed to build anything on its own. The whole point of it was to provide a large framework for other people to make their own configs, in any number of permutations, on whatever hardware they liked, with many different payloads and machines supported.

In other words, they were re-creating the coreboot build system. Poorly. The sad thing is that, in retrospect, those two could have helped coreboot directly, working on its own build system. Many of the features they worked on would have been suitable to upstream, and all coreboot distros (not just Libreboot) could have benefited, but I digress.

It was an extremely complex design. The old build system was still provided, by the script named oldbuild and the directory named resources/ - if I remove those in a branch, and then run sloccount . (using David Wheeler's sloccount utility), I see:

SLOC	Directory	SLOC-by-Language (Sorted)
7874    tools           sh=7874
5343    projects        sh=3278,ansic=2065
2359    libs            sh=2359
472     www             ansic=369,sh=103
282     top_dir         sh=282

In total, 13896 of shell scripts and several thousand lines of C for various utils (most of the C sloccount is ich9utils).

If you actually look at the Paper system, almost all of it is covering utilities that don't need to be included in the project, many of them ChromeOS-related; modern Libreboot (using my design, the lbmk build system), uses U-Boot, courtesy of work done by Alper Nebi Yasak (more on this later). I compile U-Boot with the same generic script that I maintain, which is less than 300 sloc in size, and compiles SeaBIOS, coreboot and U-Boot, without any project-specific logic for the most part, and it doesn't just clean. It also configures, updates configs, fetches, patches... it does it all.

For comparison, here is modern lbmk (the current Libreboot build system today on 12 December 2023):

SLOC	Directory	SLOC-by-Language (Sorted)
1410    script          sh=1410
1007    util            ansic=601,python=406
321     include         sh=321
127     top_dir         sh=127
0       config          (none)

I will write more about lbmk later in the article, but what you need to know for now is: lbmk supports more boards than Paper (a lot more), and actually does a lot more, but the design is extremely efficient. The Paper design results in an extreme amount of code repetition, where each module (e.g. packaged util) is handled by scripts conforming to a uniform design, but each script is tweaked per task. On the one hand, that offers Paper some flexibility but a project like Libreboot is only ever managed by a few people, and the Paper design was very overwhelming to new contributors.

Actually, I won't write much about lbmk at all. You can just read the lbmk documentation. Paper also has in excess of 100 scripts. The lbmk design, at least on 12 December 2023, is currently 12 scripts. Just 12, but don't be fooled by simplicity. It can do everything Paper can.

The lbmk design is so light because it generalises everything as much as possible. The lbmk design is about 1800 lines of shell script and a couple hundred lines of C; ich9utils is no longer included, because lbmk instead includes the ifd/gbe configs generated by it, directly, and many other utils are now handled (downloaded, patched and compiled) by the shell scripts. So, the modern lbmk design does/can do everything Paper does/can do, but in 90% fewer lines of code.

This sloccount comparison is valid, comparing Paper from 2021 (I ended all development of it in 2021), versus lbmk in December 2023. I saw as early as 2017, when Paper was in very early development stages, that the design was completely unworkable due to complexity, versus the scope of the project which is to provide firmware images for end users.

I was going to scrap Paper in 2017, and revive my work on the original design, but as I said: I had other real-life concerns going on, and had to go on hiatus for many years. I let development of Paper continue during all those years, in the hope that it would be stabilised at some point, and I had intended to take it over myself at some point; but it was too complex.

I would later take over the Libreboot project. This is what I will cover, in the next sections. Paul stopped working on Libreboot after around late 2017, leaving the work solely in the hands of Andrew and Sebastian.

late 2020: osboot

I started the osboot project in December 2020. Check osboot.org on the Wayback Machine. Osboot was initially called Retroboot, but soon renamed to osboot, so also check retroboot.org on the wayback machine.

Osboot started as an experiment: what if Libreboot hardware support could be greatly expanded, by adopting a more pragmatic policy? That was answered, with the Binary Blob Reduction Policy - today, that is the Libreboot policy, since osboot later merged with Libreboot, in November of 2022, but it's the same policy that osboot used.

I was commissioned by a Minifree customer to provide a ThinkPad X230. I already knew then that the X230 (an Ivy Bridge platform) could boot entirely blob-free in coreboot, except for the neutered Intel ME (using me_cleaner) and microcode updates. To my mind, it seemed acceptable in terms of software freedom. A solid, all-round very useful machine.

I had a problem: what firmware would I provide? Skulls? Heads? This customer wanted something easy. Skulls is great but lacks a build system as advanced as Libreboot's (Paper or lbmk), though it does have one that is reminiscent of ours, and they standardize on SeaBIOS. They didn't (and still don't) have very good documentation, in my view. Then there is Heads: an excellent project, but quite excessive for this customer's needs. In fact, the customer specifically wanted UEFI at that time.

I decided to fork the Libreboot 20160907 build system, during December 2020, for this customer. The 20160907 build system had many flaws, and it was 4 years unmaintained at that point; the Paper build system was nominally still in development, and the official build system of Libreboot, but not developed by myself. Although I understood the Paper design well, I didn't like the code. I still hate it. I hate that design, passionately. I wanted something clean, and efficient.

Thus, osboot was born. It initially began with Tianocore integrated in it, using MrChromebox's branch for coreboot. MrChromebox is a coreboot distro that provides Tianocore as a payload, on many Chromebooks, so I leaned on his expertise to provide experimental Tianocore support, in my Libreboot 20160907 fork. I completely gutted about half of the build system in that Libreboot release, when forking it, because most of it no longer worked for modern upstreams like coreboot and GRUB, both of which had gone through massive changes over the years.

One of Libreboot's earlier design flaws was that parts of the build system ran completely standalone, while relying on work done by other parts of the build system, while not checking automatically whether other steps had been done. For example, the final preparation of a ROM image did not check whether GRUB had been built. So you had to know the build system well, and know which parts of it to run, and in what precise order. The Libreboot documentation used to cover this, telling you how to use it in different ways, depending on the desired configuration. This was fixed, very early on in the osboot project.

Osboot greatly improved the configurability of everything. For example, it enabled you to more easily mix and match different payload combinations per mainboard, whereas the Libreboot 20160907 design only allowed one payload type, per target, and it contained a lot of target-specific logic that was hardcoded, baked into the build system's design - even in December 2020 when osboot started, I envisioned that it would one day support many more boards, whereas Libreboot only supported like 15 mainboards in 2016.

The osboot redesign, relative to Libreboot 20160907, made handling of source tree revisions and patches much cleaner, especially for multi-tree projects like coreboot, where multiple revisions could be used depending on mainboard.

Of course, the first osboot revision also updated to a modern coreboot revision at that time, which (in that month, December 2020) was coreboot 4.14.

Although I only did osboot initially for the X230, for a customer, it took weeks of work to get the build system up to date, and I spent time updating all of the documentation too. At that point, I realised what I had actually done: I had gotten back into development again. Moreover, I found the experience fun again, and I found that I once again had the energy.

In short, I had my mojo back. My then 4-year hiatus was over.

December 2020 Libreboot takeover

Perhaps getting too carried away too soon, I actually took over the Libreboot project again in late December 2020. I was rapidly working on adding all the Libreboot boards to osboot, and I would have just very quickly updated the deblob scripts; I acted on this desire too soon, running under the gun.

I actually did remove Sebastian and Andrew's access to the Libreboot infrastructure, for a few hours, before re-instating them - I wasn't ready for a Libreboot takeover yet. I needed more time to polish everything. Doing a tiny release for 1 customer, on 1 machine (the X230) was all well and good, but I decided that I had time to polish it more.

March 2021 Libreboot takeover

I wanted to get rid of Sebastian and Andrew for some time, at that point. Too many years had gone by, without any releases in Libreboot, and the Paper build system was only growing more complex; it was completely unworkable, and their time was up. They failed. An unwritten rule in the new constitution of 2017 was that they would competently run the project; I entrusted this to them, in that year. They failed, miserably. Despite their clear passion and desire to do good for the project, they were adamant.

I had tried to sell to them the idea of scrapping the Paper build system (the re-write), instead opting for my cleanly revived build system based on Libreboot 20160907, which at that time was called osbmk (osboot make). They would have worked under me on it. They refused this, but asserted that Libreboot would have two projects within it: one run by me, using my newly updated design, and one with theirs. Now, I ask: why should I accept/enable the authority of those who clearly had none? I was at a point then that I knew my work was viable, and that I would have a release soon.

I'd had too many arguments with them over the years, largely technical in nature. They regularly derided my own work, even the improved build system design and all the rapid early progress I'd already made with the new osboot project; I actually did make some temporary releases of osboot (tarballs, with ROM images and source code) in December 2020, but deleted them, because they weren't intended for general use. I was only testing everything.

Sebastian had a lot of knowledge about hardware, but did almost no coding himself; he left that all to Andrew. And Andrew hadn't done any substantial work in over 6 months at that time. And the work that was done, was still far from complete. I anticipated that there would be years left before the Paper design would yield a stable release, so I came to the conclusion:

I should not carry around so much baggage. I had a clean new design, that worked.

So I removed all their access to everything, and took over the Libreboot project. Naturally, they both started denouncing me, but it didn't matter: I was in charge. I was (and still am) grateful for their efforts, but Libreboot must have stable releases, and there must be a clear direction. Having two projects within it would only confuse people, and I was already addressing many of the previous flaws in my design, fixing them one by one.

However, I only had osboot then. I started actually working on a new Libreboot release after I took over.

2021 (lib)reboot

New Libreboot repositories

The original Libreboot git repository can be found here: https://notabug.org/libreboot/obsolete-repository-preserved-for-historical-purposes (if this link ever goes offline, I'll mirror that repository to another git site, or self-host it)

From mid-2021 onward, the Libreboot project was revived. The original repository, linked above, contained the build system, documentation and images, along with the static site generator that generates the Libreboot website (from Markdown files). All in the same repository.

In mid-2021, Libreboot was split into smaller repositories:

lbmk: the build system
lbwww: the website (markdown files)
lbww-img: images
various other repositories, containing random utilities
untitled: the static site generator, which creates the website, by creating HTML files from the Markdown files in lbwww

Untitled Static Site Generator

Untitled Static Site Generator was going to be part of Libreboot officially, but by that time, I had forked the Libreboot static site generator many times, for other websites that I also host. Therefore, the static site generator was split into its own standalone project, independent of Libreboot. It is hosted here. It combines all of the features of the forks, and it can handle many different websites simultaneously. Many changes were made, to make Untitled more reliable and user-friendly than the original scripts. The original static site generator, that Untitled was forked from, was submitted to the Libreboot project by Alyssa Rosenzweig in 2017.

The osboot project greatly expanded, during this time. It never made any releases officially, but it provided an excellent proof of concept, and it became the basis for modern day Libreboot.

May 2021 Libreboot release

Somewhat annoyingly, it coincided with the Freenode/Libera drama at that time, when the latter forked from Freenode after a hostile takeover. Thankfully, that drama ended quickly and we ended up with Libera being the clear winner.

Anyway: I finally did get a new Libreboot release out, based on the work that I had started in osboot, but at that time still complying with the GNU Free System Distribution Guidelines, like 2016 Libreboot did.

I did not use any of the work in the Paper re-write. I used precisely zero lines of code from Sebastian and Andrew's work. I did it all myself. My decision to take over the project had been vindicated. I did it for the users. People had been complaining for years about the lack of a release. The Libreboot project was dead. A joke to everyone, and it was no longer my fault; it was theirs, for utterly failing to do any releases, and for running it into the ground.

I still feel anger to this day, about my own failure. I should have never stepped down, in 2017. I should have stuck it out, and carried on my work. There are many works that I and others are completing today, that would have been done then. Alas, hindsight is 20/20.

The Libreboot 20210522 release was the first release in about 5 years, at that point, and it added many new boards. It inherited all of the massive build system design improvements from osboot.

Osboot from 2021-2022

The osboot project continued, since December 2020 when it started. I started adding many new boards to it. The purpose of Osboot was the same as Libreboot, except that it had the Binary Blob Reduction Policy that Libreboot now uses, but back then, Libreboot adhered to GNU FSDG.

Goals of osboot

I wanted to start a project like osboot much earlier, in 2017. In fact, I was going to, but real life got in the way and I went on hiatus for a few years. I concluded as early as 2017 that the FSDG policy was a liability, especially after me_cleaner became available. I realised that there were many more machines out there, that I then considered acceptable from a software freedom perspective, even if they did not comply with FSF RYF or GNU FSDG; I regard the FSF policy as overly dogmatic, and I've always thought so, but in practise it didn't hurt anything because Intel ME was always a problem. I didn't know ME neutering was possible, until I found out about me_cleaner - I would never provide anyone with a setup that uses a non-neutered ME, because of the networking and backdoors provided in such setups. The me_cleaner project removes such backdoors from the Intel ME, resulting in a clean image that only contains the init firmware in the ME. In this configuration, the ME initialises itself but doesn't actually run anything, so it just sits there permanently in the init stage. Things like AMT no longer work, when you run me_cleaner.

To put it simply: at least on the Intel side, me_cleaner changed everything and opened up a lot more machines for me. But to do that, I had to abandon Libreboot policy. Hence, osboot was born. Libreboot policy was a liability to the free software movement, because it limited the amount of free software that people could easily use. Remember, nothing else as easy to use as Libreboot existed, within the coreboot community. I saw that many people have machines which can have most of the proprietary firmware replaced with free software, namely coreboot, but Libreboot set very arbitrary limits on what boards I could add. These restrictions became a liability to the project.

The plan therefore was to modify Libreboot policy, and adapt to this new reality, but I wasn't really sure, and at that time I wasn't sure whether my work would succeed; this is why I created a separate project, rather than directly taking back over Libreboot from the start. To my surprise, osboot started becoming popular within my communities, early on, so I kept it going for another 2 years or so, to see how it would go.

The feedback I kept getting was that most people saw essentially no difference between in and Libreboot, except that Libreboot had a bunch of limitations in it preventing people from using more hardware. People often got confused, asking what differs between the two projects. I decided later that they would merge.

The purpose of osboot was then to support every board from coreboot, which is more challenging, especially when dealing with code from the vendor that often must be added, in different ways depending on the machine.

At first, osboot was very hands-on. For example, you had to dump the original vendor BIOS to extract Intel ME, to then run me_cleaner, extracting the IFD, GBE and (neutered) ME region with ifdtool to then insert into the ROM image. Nowadays, that is completely automated, based on initial work done in osboot by Caleb La Grange. Caleb adapted logic from Heads that extracts Intel ME images from Lenovo UEFI firmware updates, thus eliminating the need for vendor firmware dumps prior to flashing; though, backing up the factory firmware is still recommended, on any mainboard.

I later expanded Caleb's work, making the logic much more generic, expandable more easily to other boards/vendors and platforms. I also added support for other files such as Intel MRC firmware and SMSC KBC1126 EC firmware (KBC1126 EC firmware extraction was only done in Libreboot, after it merged with osboot - more on this later)

November 2022: osboot/libreboot merger

I shut down the osboot project, formerly hosted on osboot.org, during November 2022. The website was redirected to Libreboot. I then added all extra documentation and code from osboot, into Libreboot, thus merging the two projects together, and Libreboot formally adopted osboot's Binary Blob Reduction Policy. I concluded that the two projects were essentially the same, and I'd decided during the two years of osboot's existence that it was vastly superior to Libreboot. I had a lot more fun working on it. Libreboot became a side hobby at that point. OSBoot was my primary focus.

This was a resounding success. It was done in a day. Caleb adapted the documentation, while I worked on the build system. We did it in a day.

My efforts in osboot and my extensive research revealed to me that almost everyone supported this move. Sure enough, I saw a relative lack of opposition to it; though, some of the more dogmatic members of the FSF were quite upset. This level of upset later caused.... well, that's what I'm going to cover next.

GNU Boot

The purpose of today's article has been to write a rigorous history section for the Libreboot project, because a lot of earlier history for the project wasn't available. Many of Libreboot's early years were turbulent, and I never expected then that the project would last 5 years, let alone 10.

Anyway: I actually don't need to go into detail about GNU Boot, because I already wrote about it in the GNU Boot article. I will summarise it here, and perhaps share additional details:

Initial hostile fork

During Libreplanet 2023, which is the FSF's annual conference, the FSF, through Denis Carikli who is part of their inner circle, announced a hostile fork of Libreboot. I anticipated a fork of Libreboot by the FSF, as a result of the osboot merger, and I was OK with it, but:

They tried to steal the name Libreboot, and sell themselves as the real Libreboot project. This, I could not tolerate. I already knew of their effort in December 2022, because its existence was leaked to me; they asked all of the people I'd worked with to that point, asking them to betray me and work for them instead. None were interested, and one of them leaked it to me.

So I was prepared. I expected they'd announce it during Libreplanet and I was right; Denis had a talk scheduled for 19 March 2023, a Sunday, about boot firmware, though the fork wasn't announced until he spoke, in the talk. But I knew. So I prepared.

I released Libreboot 20230319, several hours before his talk, and I started an aggressive countercoup, whereby Libreboot was developed much more aggressively than ever before, to completely outcompete them in every way. It worked. What's more, I did the unthinkable: I got them to back down. That almost never happens, as the FSF tends to be extremely dogmatic, too drunk on their own prestige; set in their ways.

They initially hosted their project on the official sourcehut instance run by Drew DeVault. I simply asked Drew to terminate their hosting. He complied. He gave them two weeks to rename their project, or be deleted, and they did not rename. Their website was hosted on sourcehut at the time, so their website went offline for a day; on that same day, I announced the Libreboot 20230423 release. Drew's assessment, matching my own, was that it would be inappropriate to host a hostile same-named fork of an active project. I now own access to https://sr.ht/~libreboot/, at least on 12 December 2023, and I even pay for it, though it is currently empty - I may very well use at a later date, to provide mailing lists for Libreboot.

If you run whois on libreboot.at (the domain name of their original fork site), you will see that it says GNU Hostmaster. Owned by the FSF. Same config as used for several other domains used by GNU projects - they were going to make another GNU Libreboot, against my will. This, in spite of the truce established in 2017.

I don't feel bad about what I did. What they did is widely considered to be a dick move in the free software movement; it is widely considered impolite to fork an active project and use the same name. The unwritten rule is always that you use a new name. It's more than that though: in their initial fork, and also in GNU Boot, they didn't just say: this is what we are, what we do and why you should use our product. No. They put a paragraph in their documentation, urging people to delete links to libreboot.org, and link to them. They were out for blood.

Fork attempt 2: GNU Boot

So I've responded in kind, ever since. Regardless of whether they succeed or whether they are competent, a thought exists in their head. A dream, you could say. Their dream is a world in which Leah Rowe and Libreboot no longer exist. They wanted to destroy me. It's evident in Denis's LP2023 talk, where he said he wanted to "continue" the Libreboot project. In his mind, there is no room for disagreement over policy; it's FSDG or nothing. I had to play by their rules, or go home.

During that time, and subsequently, I and others had repeatedly put pressure on them to rename. I personally came up with the name GNU Boot - and suggested that they use it. It's a name that I myself came up with several years prior, when I was considering whether to work for GNU again myself, but as an actual fork of coreboot. The name just works.

They did rename, to GNU Boot. You can learn more about GNU Boot, by reading the GNU Boot article.

The points raised in the GNU Boot article are still valid on today, 12 December 2023. GNU Boot 0.1 RC3 is imminent for release, on this day, and it is still based largely on Libreboot 20220710, with only superficial changes on top of it. it still has all the old, obsolete revisions for all projects, including coreboot. It still has all of the same bugs, that Libreboot has since fixed, especially during 2023.

Unlike with their first attempt, GNU Boot is fully hosted on the GNU Savannah infrastructure, as any proper GNU project should be. I respect the GNU Boot project more, because it is its own thing, that doesn't try to ride off of my coattails. However, my perception of them is permanently coloured by their initial hostile actions; and the only reason they ceased such actions was because I made it clear that they would not be allowed. The fact that they even tried, means they thought I would roll over, or otherwise fail to counteract it meaningfully. They still want to damage Libreboot.

So, my strategy has been to constantly develop Libreboot from now on, much more aggressively, and generally stay on top of my game. And I made no secret of this strategy; I've been pretty open about everything, throughout 2023.

Countercoup

My countercoup has essentially consisted of a single strategy: be the best, on a technical level. During all of 2023, roughly 5x as much work has gone into Libreboot compared to any other average development year (2017-2021 notwithstanding). Essentially, the FSF's strategy was to overwhelm Libreboot; they wanted to make a big splash about their fork, and get devs on board as quickly as possible, before Libreboot could develop much further under the new policy, because Libreboot had not yet fully asserted itself in this regard, at least so far as public recognition was concerned. So I doubled, tripled and quadrupled down, and maintained absolute laser focus from 19 March 2023 onward, a focus that I still maintain to this day - I have many plans for Libreboot. If you think 2023 has been crazy, 2024 will be even better. 2024 will be Libreboot's biggest development year yet. 2023 was only a warmup.

I've done three major audits to the Libreboot build system, during 2023. The build system has been re-engineered, 3 times, each time becoming infinitely more efficient.

Many new boards have been added. Many new features. Non-coreboot firmware is now supported; for example, the (libre) stm32-vserprog and pico-serprog code is now used by Libreboot, to provide serprog firmware on many STM32- and RP2040-based microcontroller boards; we use them as SPI flashers, but you can use them for many other purposes. They're quite handy little machines. The serprog firmware integration was done by Riku Viitanen, who also added many HP EliteBook machines to Libreboot during 2023.

Thanks to the work done by Nicholas Chin, several Dell Latitude laptops are now supported in Libreboot, which can be flashed without disassembly. We support GM45 and Ivy Bridge models, and we're adding many more (we have identified about 20 of them).

You can actually just read the Libreboot release announcements from 2023, if you want to know about all of the rapid progress made in Libreboot this year. It is quite simply impossible to adequately describe, in only a short few paragraphs, just how much Libreboot has improved during 2023. The development pace during all of 2023 has been completely insane. Here are all of the Libreboot release announcements from 2023:

I also previously wrote about each of the three audits during 2023:

and now:

Canoeboot

Purely for my own entertainment, I decided to re-create FSDG Libreboot myself. FSDG is the policy that GNU Boot uses, that Libreboot previously used, before it adopted the modern Binary Blob Reduction Policy - GNU Boot started, specifically because it opposed the new policy in Libreboot.

Well, GNU Boot seemed to be going nowhere fast. I monitor their Git activity daily, and their development pace is much slower than mine; slower than mine, even when I'm going slow. I thought to myself: what if a competently engineered solution existed, like GNU Boot but not?

And thus, Canoeboot was born. I still mainly develop Libreboot, but I spend a few hours after each release, bringing Canoeboot up to date and running the deblob scripts, to update the blob.list files in its build system.

Canoeboot is even listed on the FSF's Free Software Directory - Craig Topham approved it, he's the FSF's current licensing and compliance manager, at least on 12 December 2023.

The purpose of Canoeboot is to simply exist, complying with GNU Boot policy while being superior to it in every way, outcompeting it so fast that the GNU Boot project is constantly behind - it's done, specifically to demonstrate the superiority of Libreboot policy, by showing what Libreboot would have been, if it didn't adopt the new policy.

There is even a Canoeboot vs GNU Boot page which you can read. I keep it up to date, comparing the latest releases of both projects with each other.

The FSF failed in its coup. My countercoup was a success. I madly beat them at their own game. The FSF's strategy might have worked, if I hadn't been so vigilant; Libreboot's development pace was much more conservative at that point, and most of the works that have now been completed or are now underway, were not even started. Libreboot was very young into its existence, relative to the merging of the osboot project. They probably didn't expect any of the actions that I've taken, or the massive leapfrog year of development this has been for Libreboot. As far as I'm concerned, Libreboot has won 2023 - 2024 is the next battle.

And I will maintain Canoeboot. It is currently completely in sync with the latest Libreboot release, and it will keep staying in sync, relative to each new Libreboot release. I actually describe how this syncing is done, in great detail, on Canoeboot's about page - and the first Canoeboot release was done, based on the latest Libreboot release at that time, in only 2 days, with all of Libreboot's vast improvements in it compared to GNU Boot. As of this day, 12 December 2023, Canoeboot is about 1 year ahead of GNU Boot in terms of code development, and 2 years ahead on the writing of documentation. Conversely, GNU Boot is 1 year and 2 years out of date, in terms of code and documentation respectively.

2024 reconciliation intentions

The FSF started the coldboot war. Libreboot merely won it.

From 2024 onward, unless more hostilities develop from FSF/GNU's side, I intend to adopt a more conciliatory approach toward GNU Boot. I won the battle of 2023. I won the cold boot war, but the real battle is this: how do we get free boot firmware to non-technical end users, efficiently and reliably? The answer to that question is projects such as Libreboot, or indeed others like GNU Boot, Heads, Skulls, MrChromebox... you name it. Distros, designed similarly to Linux distros, but for building boot firmware instead.

Much of 2023 was spent counteracting the FSF's coup, because they were hostile to the Libreboot project, but I decided that I will avoid any such counter action from now on. I will stil develop Canoeboot, but my main focus is Libreboot. My conclusion is that, so long as my own efforts exist, and I keep working on everything, the GNU Boot project is no threat to Libreboot whatsoever.

Towards the end of 2023, there was in fact cooperation, between the GNU Boot and Libreboot projects, in the form of small patches; Denis Carikli sent a few patches and reports to Canoeboot and the Untitled Static Site Generator, and I did similar for the GNU boot project. The FSF themselves even decided to accept Canoeboot, on their Free Software Directory. See: https://directory.fsf.org/wiki/Canoeboot

My only issue with GNU Boot at the start of 2023 was that they wanted to replace the Libreboot project, by using the same name. They did try to destroy the Libreboot project, and take it for themselves. Things pretty much calmed towards the end of 2023, and now the two projects/communities operate to their own ends, not taking much stock of the other. If they don't bother me from now on, I won't pay any attention to them.

They are welcome to send me patches, and I may help them if I feel like it.

Last remarks

I've pretty much gone through the entire history, to the best of my ability. I probably did forget a few things, but these are the broad strokes. If you managed to read this far, I salute you. In any case, this article describes both the history and nature of the Libreboot project up to this day, 12 December 2023, marking the 10th anniversary of the Libreboot project, relative to its first release on 12 December 2013. I was planning a new release for this day, but I decided not to rush that; more testing is needed, and there are several new features/boards that I and others in the Libreboot project are working on.

If you made it this far, reading all the way to the end, I will now reward your patience and your courage with a cool easter egg:

The minifree.org domain name initially hosted a website that was teaching people how to use VPNs and Tor, and it contained information about how to be private on the internet. I later planned on making it into an FSDG-compliant fork of Puppy Linux - because Puppy is small, lightweight, and mini is another word for small, and it was to be a fully free distribution, containing no blobs, not even firmware. Minifree GNU/Linux. I later scrapped the plan, and realised that Minifree was also newspeak, that could mean Ministry of Freedom. The domain name was re-purposed to provide my company website, with the erstwhile gluglug domain then redirecting to it, and I registered Minifree Ltd in the UK. Minifree is how I fund the Libreboot project, by selling computers with Libreboot pre-installed. You can find it on minifree.org.

84 KiB Raw Blame History