Add -Zfunction-return={keep,thunk-extern} option

[ojeda]

This is intended to be used for Linux kernel RETHUNK builds.

With this commit (optionally backported to Rust 1.73.0), plus a patched Linux kernel to pass the flag, I get a RETHUNK build with Rust enabled that is objtool-warning-free and is able to boot in QEMU and load a sample Rust kernel module.

Issue: #116853.

[rustbot]

r? @wesleywiser

(rustbot has picked a reviewer for you, use r? to override)

[ojeda]

Note that LLVM ignores the attribute in non-x86, but I have not replicated Clang's check that this only works in x86 for the CLI flag for the moment (should it be part of the target spec?).

Also, I linked the feature request issue as the tracking one in the docs -- we can reuse it changing the OP message, I assume, but please let me know if it is best to create a new one.

[ojeda]

If there is a way to get the effective code model, it would be nice to use it.

[Urgau]

Maybe check sess.target.options.code_model ?

[ojeda]

That covers even less, no? i.e. code_model() covers codegen and target options.

[ojeda]

I added a folder but that still counts against the tidy limit in tests/ui. Should I move the folder to e.g. tests/ui/invalid-compile-flags or increase the limit by 1?

[ojeda]

One more note: GCC and Clang give an error if one tries to use function-return at all, i.e. even with keep, on non-x86. However, in this PR keep is allowed for all architectures.

I did it like this because it seemed simpler (no Option) and more flexible, especially since (in the case of this flag) other architectures could support particular modes, e.g. s390 expolines (https://github.com/llvm/llvm-project/blob/49af6502c6dcb4a7f7520178bd14df396f78240c/clang/lib/Frontend/CompilerInvocation.cpp#L1942).

But perhaps consistency with GCC/Clang is better?

[ojeda]

I added a folder but that still counts against the tidy limit in tests/ui. Should I move the folder to e.g. tests/ui/invalid-compile-flags or increase the limit by 1?

I moved them inside the other folder, and noticed one in invalid/ should be in that invalid-compile-flags/ folder: #116975.

[bors]

☔ The latest upstream changes (presumably #117405) made this pull request unmergeable. Please resolve the merge conflicts.

[ojeda]

Resolved trivial formatting conflict.

[wesleywiser]

Thanks for already getting the tracking issue set up @ojeda!

@bors r+

[ojeda]

@wesleywiser Thanks for the very quick review and have a nice holiday (I saw the other PR by chance when looking for this one :)

[wesleywiser]

Thanks! ☺️