Fix loongarch64 ABI support for C procedure call

[yxd-ym]

See: https://github.com/loongson/la-abi-specs/blob/release/lapcs.adoc

[GabrielBarbosaGV]

Hello, all! Apologies if this question is misplaced or inappropriate, but I thought I'd ask here given that the discussion section is disabled.

I've come across this issue while in search of contributor friendly ones, due to my very brief experience with open-source (actually, in my case, what I only later learned to be source-available) development. It didn't take long for me to notice that I simply do not know enough at this point to contribute, at least in cases such as this, as I do not fully understand the codebase. This leads into my question, which is, what should I study to be in shape for matters of this kind?

My first instinct was to look for mentions of loongarch(64)? within the codebase, bringing me to loongarch64.zig and this auto-generated loongarch.zig. I was more attracted to the first option, and so inspected it. At first, the syscalls seemed rather alien to me, but a brief look into inline assembly within Zig's documentation cleared things up. Even having seen that, however, I could not determine if the issue would be found there, realizing that it'd be no good if I did not know what ABI support for procedure calls meant. From what I gathered of it, it is the hardware dependent communication layer between code formats. This indicated to me that the file I first referenced was probably what I was looking for, which led me to fork the project to be able to better navigate it.

Having forked the codebase, I could not find direct mentions of the syscalls defined in that file, but did see it being imported in linux.zig. From this point on, I could hypothesize on the problems by cross-referencing the linked .adoc with the syscalls, but think those would be mostly conjectures rather than valuable insights. My further questions in line with this are how the error was found, and what its symptoms are so that I can pinpoint whether it's been fixed through the changes I perform. I'd also like to know if what I deduced even is correct, if posing questions such as this in issues is OK, if I need a machine with the specified architecture, and if it would also be OK to ask more in case other relevant thoughts come to mind.

Many thanks!

[yxd-ym]

I created this issue based on the comment here: #21667 (comment)

If this issue is solved, c_abi tests mentioned above should all pass on loongarch64.

[GabrielBarbosaGV]

Ah, OK! That means I'd need loongarch64 handy to run the tests on the architecture and check if all works well, right?

[gabeuehlein]

@GabrielBarbosaGV

You would at least need a loongarch64 VM (you can use QEMU to emulate loongarch64) and then run the tests on that. If you have a loongarch64-based device, I'd prefer that, because it would run quite a bit faster.

As for the "proper" place to ask this, there are quite a few Zig groups online. Most of them are documented here. I don't have a presence on any of them, but I would look into the ZSF's Zulip server or #zig at libera.chat for contributing to Zig (preferrably Zulip).