Validate use of parameters in naked functions

[tmiasko]

• Reject use of parameters inside naked function body.
• Reject use of patterns inside function parameters, to emphasize role
  of parameters a signature declaration (mirroring existing behaviour
  for function declarations) and avoid generating code introducing
  specified bindings.

Closes issues below by considering input to be ill-formed.

Closes#75922.
Closes#77848.
Closes#79350.

[rust-highfive]

r? @lcnr

(rust-highfive has picked a reviewer for you, use r? to override)

[lcnr]

impl looks good, don't know enough about how #[naked] should be used though

maybe r? @bjorn3 as they get recommended by github?

[tmiasko]

The overall direction of this changes is to constrain the use of naked functions towards cases that can be reliably supported. The next step would be to permit only assembly inside the naked functions.

cc @joshtriplett@Amanieu

[bjorn3]

@lcnr

maybe r? @bjorn3 as they get recommended by github?

I probably got recommended as I recently changed rustc_interface. I don't think I am a good reviewer for this. r? @Amanieu maybe?

@tmiasko

Reject use of parameters inside naked function body.

Maybe it should be allowed for asm! though I am not sure.

Reject use of patterns inside function parameters, to emphasize role
of parameters a signature declaration (mirroring existing behaviour
for function declarations) and avoid generating code introducing
specified bindings.

Makes sense.

[Amanieu]

Maybe it should be allowed for asm! though I am not sure.

No, you're supposed to access parameters directly through registers or the stack based on the function ABI.

[Amanieu]

Should we consider adding additional checks to this pass, such as the requirement that a naked function only contain a single asm! block with only const and sym operands?

[tmiasko]

Should we consider adding additional checks to this pass, such as the requirement that a naked function only contain a single asm! block with only const and sym operands?

Yes, I think we should have additional checks along those lines. I didn't think through all the details to propose one yet. There is also a question of compatibility with llvm_asm! which is often used together with core::intrinsics::unreachable(), e.g., code in compiler-builtins.

[Amanieu]

@bors r+

[bors]

📌 Commit 22d3431 has been approved by Amanieu

[bors]

⌛ Testing commit 22d3431 with merge b48cafd...

[bors]

☀️ Test successful - checks-actions
Approved by: Amanieu
Pushing b48cafd to master...