Building Rustc on m68k

[FractalFir]

I am currently working on building rustc for m68k(since that seems like a more achievable short-term goal). Many of the roadblocks I'll hit on m68k will also likely be shared with other targets - this issue could server as a reference for that.

Here is a list of currently-known issues:

• - linux-raw-sys does not support m68k. This crate is needed by rustc, and a lot of other crates. Proper support will require a moderate amount of work, but it something that can be done now.

• - The rustix crate does not support m68k. (more specifically, the iocl::const module needs the right values for syscall arguments).

• - The const assert in https://github.com/rust-lang/rust/blob/ce5fdd7d42aba9a2925692e11af2bd39cf37798a/compiler/rustc_data_structures/src/tagged_ptr.rs#L163 fails. EDIT: the problem is the alignment of usize. Rust uses those bits to store type tags. If the aligement is less than 4, this fails. I have a fix ready to be sent & merged.

• - The bootstrap script fails to compile native GCC for m68k. This prevents it from building any codegen backends or tools later down the line. This is skipable with some fiddling.

• - When building other tools(eg. cargo) the bootstrap script "forgets" where the std for m68k is.

So, as of toady, with some workarounds, I can get pretty far into the cross-compilation process - a bit more work, and we may be able to get rustc running on m68k.

@glaubitz I think this issue might be of interest to you.

[FractalFir]

For now, it seems like those 4 are the only remaining issues that stop us from getting rustc on that plaform.

Working around them, I got a build of rustc for m68k.

[glaubitz]

For now, it seems like those 4 are the only remaining issues that stop us from getting rustc on that plaform.

Working around them, I got a build of rustc for m68k.

Wow, amazing! I guess you can also then add support for alpha, hppa and sh4 ;-).

[glaubitz]

FWIW, in case you're running into alignment issues with m68k, you can use 32-bit alignment as this is the alignment we want to use in the future on Linux/m68k and which is also the default alignment according to the official ELF ABI specification.

[FractalFir]

Wow, amazing! I guess you can also then add support for alpha, hppa and sh4 ;-).

Sadly, any upstream support will have to wait for an MCP about adding the alpha / hppa / sh4 target. There is nothing I can do in this area: I can't maintain those targets, and I have to wait until the interested parties create a MCP.

[glaubitz]

Wow, amazing! I guess you can also then add support for alpha, hppa and sh4 ;-).

Sadly, any upstream support will have to wait for an MCP about adding the alpha / hppa / sh4 target. There is nothing I can do in this area: I can't maintain those targets, and I have to wait until the interested parties create a MCP.

Can you elaborate on this a little more? Has it become more complicated adding new targets? What's an MCP?

Back when I added m68k-unknown-linux-gnu, powerpc-unknown-linux-gnuspe and sparc-unknown-linux-gnu it was a matter of creating a PR.

[FractalFir]

I sent a message about this here - guess it just got buried.

The TLDR is that adding a tier 3 target requires a MCP, like this one, for example.

A MCP issue is kind of like the PRs you'd send before(I think, unless the process changed a lot). If nobody voices concern about adding Alpha, and one of the Rust devs agrees to help facilitate/oversee the work, the MCP gets accepted after 10 days.

At that point, we have all the blessings to start compiler-side work, and get it merged. This way, we also avoid the repeat of the hppa situation - the target can't be removed as easily.

[glaubitz]

On Jul 27, 2025, at 9:15 PM, Fractal Fir ***@***.***> wrote:FractalFir left a comment (rust-lang/rustc_codegen_gcc#744) I sent a message about this here - guess it just got buried. The TLDR is that adding a tier 3 target requires a MCP, like this one, for example. This issue is kind of like the PRs you'd send before. If nobody voices concern about adding Alpha, and one of the Rust devs agrees to help facilitate the work, the MCP gets accepted after 10 days. At that point, we have all the blessings to start compiler-side work, and get it merged. This way, we also avoid the repeat of the hppa situation - the target can't be removed as easily.Message ID: ***@***.***> OK, thanks a lot for the explanation!If you could write down all the necessary steps for adding a completely new GCC-based target to rustc, I’d already be more than happy.Once you got rustc on m68k working and we have a rough guide for adding new GCC targets, I’ll be happy to take care of alpha, hppa and sh4.

[glaubitz]

@FractalFir Could you share the m68k rustc binaries with me for testing?

And did you also manage to build cargo?

[FractalFir]

Building cargo ATM - will see if it works out. Once that is done, I will share the tar.gz files with you.

Also, you I think you will need to build libgccjit from our repo. That contains the needed patches.

[FractalFir]

cargo depends on openSSL, and that is causing a lot of build problems.

For now, here is the Rust for m68k without cargo: https://drive.google.com/file/d/1jF30QJUbDHqvRfjzwXrg4K-oTpLEGqO-/view?usp=sharing

[antoyo]

With this rustc, can you successfully compile a "Hello, world!" (using the std) for m68k?

[FractalFir]

Not yet - I don't have libgccjit.so for m68k.

[FractalFir]

With a mock libgccjit.so, I am able to check the no_core test file on m68k.

So, we now know that everything up to codegen works.

Codegen fails, due to GenericArgKind not being sufficiently aligned.

assert_eq!(align_of_val(&*lt.0.0) & TAG_MASK, 0);

[glaubitz]

With a mock libgccjit.so, I am able to check the no_core test file on m68k.

So, we now know that everything up to codegen works.

Codegen fails, due to GenericArgKind not being sufficiently aligned.

assert_eq!(align_of_val(&*lt.0.0) & TAG_MASK, 0);

Well, I was actually expecting that. It's the main reason why I'm working to switch Linux/m68k to 32-bit alignment. The linked wiki page contains a patch to switch GCC to 32-bit alignment on Linux/m68k.

Please patch rustc to use 32-bit alignment on m68k-unknown-linux-gnu by default and don't bother about the broken alignment.

Once you have everything in place, I will take care of the alignment discussion with GCC upstream.

[dkm]

I'm only lurking here, but when the build is easily reproducible and working, feel free to ping me to add it to compiler explorer (I tried 3 years ago compiler-explorer/compiler-explorer#3760 without success, but would be happy to resurrect the task)