Withdraw the claim extern "C-cmse-nonsecure-*" always matches extern "C"

[workingjubilee]

We currently claim that extern "C-cmse-nonsecure-*" ABIs will always match extern "C", but that seems... optimistic when one considers that extern "C" is ambiguous enough to be redefined in ways we may not want the Cortex M Security Extensions ABIs to mirror. If some configuration, feature, or other platform quirk that applied to Arm CPUs with CMSE would modify the extern "C" ABI, it does not seem like we should guarantee that also applies to the extern "cmse-nonsecure-*" ABIs. Anything involving target modifiers that might affect register availability or usage could make us liars if, for instance, clang decides those apply to normal C functions but not ones with the CMSE attributes, but we still want to have interop with the C compiler.

We simply do not control enough of the factors involved to both force these ABIs to match and still provide useful interop, so we shouldn't implicitly promise they do. We should leave this judgement call to the decisions of platform experts who can afford to keep up with the latest news from Cambridge, instead of enshrining today's hopeful guess forever in Rust's permitted ABIs.

It's a bit weird anyways.

• The attributes are __attribute__((cmse_nonsecure_call)) and __attribute__((cmse_nonsecure_entry)), so the obvious choice is extern "cmse-nonsecure-call" and extern "cmse-nonsecure-entry".
• We do not prefix any other ABI that reflects (or even is) a C ABI with "C-", with the exception of the Rust-defined extern "C-unwind", e.g. we do not have extern "C-aapcs" or extern "C-sysv64".

Tracking issues:

• Tracking Issue for the extern "C-cmse-nonsecure-entry" ABI #75835
• Tracking Issue for the extern "C-cmse-nonsecure-call" ABI #81391

[rustbot]

r? @compiler-errors

rustbot has assigned @compiler-errors.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[rustbot]

HIR ty lowering was modified

cc @fmease

These commits modify compiler targets.
(See the Target Tier Policy.)

Some changes occurred in diagnostic error codes

cc @GuillaumeGomez

This PR changes a file inside tests/crashes. If a crash was fixed, please move into the corresponding ui subdir and add 'Fixes #' to the PR description to autoclose the issue upon merge.

[folkertdev]

Fair enough.

AAPCS is used as a base, but https://arm-software.github.io/acle/cmse/cmse.html describes some of deviations from it (e.g. secure can't assume that unsecure properly sign/zero extended integers). Also actually we implement the LLVM interpretation of the specification (which just disallows using the stack for argument passing altogether).

[workingjubilee]

Yeah, there was speculation in the original thread, as it led to the suggestion of "C-cmse-nonsecure-*", that there might be "interop with other calling conventions". But the ABI is so constrained, its arity topping out at ~4, due to only a few registers and no stack, that I can't see any real room for improvement there. Nor difference, really.

[bors]

☔ The latest upstream changes (presumably #142181) made this pull request unmergeable. Please resolve the merge conflicts.