GNUBoot
/
coreboot-libre-fam15h-rdimm
2
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
coreboot-libre-fam15h-rdimm/Documentation/contributing/project_ideas.md

9.3 KiB
Raw Permalink Blame History

Project Ideas

This section collects ideas to improve coreboot and related projects and should serve as a pool of ideas for people who want to enter the field of firmware development but need some guidance what to work on.

These tasks can be adopted as part of programs like Google Summer of Code or by motivated individuals outside such programs.

Each entry should outline what would be done, the benefit it brings to the project, the pre-requisites, both in knowledge and parts. They should also list people interested in supporting people who want to work on them - since we started building this list for Google Summer of Code, we'll adopt its term for those people and call them mentors.

The requirements for each project aim for productive work on the project, but it's always possible to learn them "on the job". If you have any doubt if you can bring yourself up to speed in a required time frame (e.g. for GSoC), feel free to ask in the community or the mentors listed with the projects. We can then try together to figure out if you're a good match for a project, even when requirements might not all be met.

Provide toolchain binaries

Our crossgcc subproject provides a uniform compiler environment for working on coreboot and related projects. Sadly, building it takes hours, which is a bad experience when trying to build coreboot the first time.

Provide packages/installers of our compiler toolchain for Linux distros, Windows, Mac OS. For Windows, this should also include the environment (shell, make, ...).

The scripts to generate these packages should be usable on a Linux host, as that's what we're using for our automated build testing system that we could extend to provide current packages going forward. This might include automating some virtualization system (eg. QEMU or CrosVM) for non-Linux builds or Docker for different Linux distributions.

Requirements

  • coreboot knowledge: Should know how to build coreboot images and where the compiler comes into play in our build system.
  • other knowledge: Should know how packages or installers for their target OS work. Knowledge of the GCC build system is a big plus
  • hardware requirements: Nothing special

Mentors

Support Power9/Power8 in coreboot

There are some basic PPC64 stubs in coreboot, and there's open hardware in TALOS2 and its family. While they already have fully open source firmware, coreboot support adds a unified story for minimal firmware across architectures.

Requirements

  • coreboot knowledge: Should be familiar with making chipset level changes to the code.
  • other knowledge: A general idea of the Power architecture, the more, the better
  • hardware requirements: QEMU Power bring-up exists, and even if it probably needs to be fixed up, that shouldn't be an exceedingly large task. For everything else, access to real Power8/9 hardware and recovery tools (e.g. for external flashing) is required.

Mentors

Support QEMU AArch64 or MIPS

Having QEMU support for the architectures coreboot can boot helps with some (limited) compatibility testing: While QEMU generally doesn't need much hardware init, any CPU state changes in the boot flow will likely be quite close to reality.

That could be used as a baseline to ensure that changes to architecture code doesn't entirely break these architectures

Requirements

  • coreboot knowledge: Should know the general boot flow in coreboot.
  • other knowledge: This will require knowing how the architecture typically boots, to adapt the coreboot payload interface to be appropriate and, for example, provide a device tree in the platform's typical format.
  • hardware requirements: since QEMU runs practically everywhere and needs no recovery mechanism, these are suitable projects when no special hardware is available.

Mentors

Add Kernel Address Sanitizer functionality to coreboot

The Kernel Address Sanitizer (KASAN) is a runtime dynamic memory error detector. The idea is to check every memory access (variables) for its validity during runtime and find bugs like stack overflow or out-of-bounds accesses. Implementing this stub into coreboot like "Undefined behavior sanitizer support" would help to ensure code quality and make the runtime code more robust.

Requirements

  • knowledge in the coreboot build system and the concept of stages
  • the KASAN feature can be improved in a way so that the memory space needed during runtime is not on a fixed address provided during compile time but determined during runtime. For this to achieve a small patch to the GCC will be helpful. Therefore minor GCC knowledge would be beneficial.
  • Implementation can be initially done in QEMU and improved on different mainboards and platforms

Mentors

Port payloads to ARM, AArch64, MIPS or RISC-V

While we have a rather big set of payloads for x86 based platforms, all other architectures are rather limited. Improve the situation by porting a payload to one of the platforms, for example GRUB2, U-Boot (the UI part), Tianocore, yabits, FILO, or Linux-as-Payload.

Since this is a bit of a catch-all idea, an application to GSoC should pick a combination of payload and architecture to support.

Requirements

  • coreboot knowledge: Should know the general boot flow in coreboot
  • other knowledge: It helps to be familiar with the architecture you want to work on.
  • hardware requirements: Much of this can be done in QEMU or other emulators, but the ability to test on real hardware is a plus.

Mentors

Fully support building coreboot with the Clang compiler

Most coreboot code is written in C, and it would be useful to support a second compiler suite in addition to gcc. Clang is another popular compiler suite and the build system generally supports building coreboot with it, but firmware is a rather special situation and we need to adjust coreboot and Clang some more to get usable binaries out of that combination.

The goal would be to get the emulation targets to boot reliably first, but also to support real hardware. If you don't have hardware around, you likely will find willing testers for devices they own and work from their bug reports.

Requirements

  • coreboot knowledge: Have a general concept of the build system
  • Clang knowledge: It may be necessary to apply minor modifications to Clang itself, but at least there will be Clang-specific compiler options etc to adapt, so some idea how compilers work and how to modify their behavior is helpful.
  • hardware requirements: If you have your own hardware that is already supported by coreboot that can be a good test target, but you will debug other people's hardware, too.
  • debugging experience: It helps if you know how to get the most out of a bug report, generate theories, build patches to test them and figure out what's going on from the resulting logs.

Mentors

Make coreboot coverity clean

coreboot and several other of our projects are automatically tested using Synopsys' free "Coverity Scan" service. While some fare pretty good, like em100 at 0 known defects, there are still many open issues in other projects, most notably coreboot itself (which is also the largest codebase).

Not all of the reports are actual issues, but the project benefits a lot if the list of unhandled reports is down to 0 because that provides a baseline when future changes reintroduce new issues: it's easier to triage and handle a list of 5 issues rather than more than 350.

This project would be going through all reports and handling them appropriately: Figure out if reports are valid or not and mark them as such. For valid reports, provide patches to fix the underlying issue.

Mentors

Extend Ghidra to support analysis of firmware images

Ghidra is a recently released cross-platform disassembler and decompiler that is extensible through plugins. Make it useful for firmware related work: Automatically parse formats (eg. by integrating UEFITool, cbfstool, decompressors), automatically identify 16/32/64bit code on x86/amd64, etc.

Learn hardware behavior from I/O and memory access logs

SerialICE is a tool to trace the behavior of executable code like firmware images. One result of that is a long log file containing the accesses to hardware resources.

It would be useful to have a tool that assists a developer-analyst in deriving knowledge about hardware from such logs. This likely can't be entirely automatic, but a tool that finds patterns and can propagate them across the log (incrementially raising the log from plain I/O accesses to a high-level description of driver behavior) would be of great use.

This is a research-heavy project.

Requirements

  • Driver knowledge: Somebody working on this should be familiar with how hardware works (eg. MMIO based register access, index/data port accesses) and how to read data sheets.
  • Machine Learning: ML techniques may be useful to find structure in traces.

Mentors