News #3: OutRun

This is Anchorboot News #3, a semi-regular update post about the state of the project. I’ve had another problematic month where I could do little to no work. But in the time I could, I’ve continued to work on the coreboot side of things: finishing and sending patches upstream for review, and deliberating on how to fix things based on reviews. Here’s the details:

Coreboot Gerrit

Coreboot uses the Gerrit Code Review software to manage code contributions. Their processes are a bit different than the usual mailing list or pull request workflows that I have been accustomed to. I took a bit of time to learn it as best as I could, and integrate what it expects from me into the way I work with git repositories.

One weird thing is that Gerrit essentially treats each commit as independent regardless of history, and merely shows a “relation chain” in the sidebar. There’s no easy way to talk about or link to an entire series like a cover letter or a pull request description, so I’ll just try to link to the most prominent patch in each.

QEMU Bochs Display

I had managed to get the Bochs display to work on other architectures, and I have sent those patches upstream as CB:80376. Part of that relation chain is CB:80372 which adds stub implementations for x86-style port I/O instructions, to help clean things up. I tried to avoid using legacy VGA resources on non-x86 architectures so that I don’t have to implement those and port the entire VGA stack to three different architectures, but I’m having a feeling that I’ll be required to do it.

RAM Detection on QEMU

For RAM detection, I have managed to write an ad-hoc parser to get memory information from the flat device-tree binary as a utility function, and uploaded it as CB:80322, along with patches for using it on ARM64 and RISC-V virtual machines. Most of the reviews for that are about refactoring useful parts into separate functions (whereas I wanted to avoid extending an API I didn’t like), so there’s more to do.

There’s also the question of using device-tree information in other places, so maybe I should try to push for a switch to a more common implementation like libfdt.

There was another RAM detection issue on ARM64 QEMU VMs where the existing RAM probing mechanism could only detect up to 1GiB of RAM due to how MMU is initialized, which I got fixed with CB:80321.

QEMU Firmware Configuration Driver

Using device-tree for RAM detection means I don’t have an immediate need for the QEMU Firmware Configuration device, but I didn’t want to leave that work incomplete. I managed to incorporate the changes I needed for other architectures into the existing code, and uploaded it as CB:80369.

ARMv7 “virt” mainboard

I had ended up creating a new mainboard for the ARMv7 “virt” platform, which I uploaded upstream as a work-in-progress CB:80831, at least until I can get the other patches merged.

PCIe Support on QEMU RISC-V

I had managed to get PCIe support working on RISC-V QEMU virtual machine enough to use the Bochs display with my other patches, which I uploaded as CB:80378.

U-Boot Payload for QEMU RISC-V

I have found an earlier attempt to use U-Boot as a coreboot payload for RISC-V QEMU builds, CB:77486. It’s a nice base for coreboot build integration which can easily be adapted to ARM architectures as well.

However, that still relies on U-Boot being built for a specific board, where it’s necessary to port U-Boot to each board one wants to use it on, which I’m trying to avoid by making generic ports on the U-Boot side for ARM. It also lacks support for a framebuffer that would be passed on to the payload, which my patches above will help fix.