`thumbv6m-none-eabi` inline asm `clobber_abi("C")` fails with `register 'LR' allocated for constraint '{r14}' does not match required type`

[kevinmehall]

The following minimized example fails to compile on the latest nightly, but works on stable and the previous nightly:

#![no_std]

#[no_mangle]
pub fn test() {
    unsafe { core::arch::asm!("nop", clobber_abi("C")) }
}

#[panic_handler]
fn panic(_: &core::panic::PanicInfo) -> ! {
    loop {}
}

Current nightly:

❯ rustc +nightly-2026-09-22 --version
rustc 1.100.0-nightly (1303417c4 2026-09-21)
                                                                                                                                                                                      
❯ rustc +nightly-2026-09-22 --target thumbv6m-none-eabi --crate-type staticlib -o repro.a repro.rs
error: register 'LR' allocated for constraint '{r14}' does not match required type
 --> repro.rs:6:32
  |
6 |     unsafe { core::arch::asm!("nop", clobber_abi("C")) }
  |                                ^^^

error: aborting due to 1 previous error

Prior nightly:

❯ rustc +nightly-2026-09-21 --version                                                             
rustc 1.100.0-nightly (bba531001 2026-09-20)

❯ rustc +nightly-2026-09-21 --target thumbv6m-none-eabi --crate-type staticlib -o repro.a repro.rs

Stable:

❯ rustc +stable --version                                                             
rustc 1.98.1 (48a229cea 2026-09-01)
                                                                                                                                                                                          
❯ rustc +stable --target thumbv6m-none-eabi --crate-type staticlib -o repro.a repro.rs

This can also be reproduced with the examples of embassy-rp and zeptos, and likely anything that calls into the RP2040 boot ROM for flash operations.

[zachs18]

searched nightlies: from nightly-2026-09-21 to nightly-2026-09-23
regressed nightly: nightly-2026-09-22
searched commit range: bba5310...1303417
regressed commit: f45772e

bisected with cargo-bisect-rustc v0.6.11

Host triple: x86_64-unknown-linux-gnu
Reproduce with:

cargo bisect-rustc --preserve --with-src --without-cargo -- b --target thumbv6m-none-eabi -Zbuild-std 

Currently testing with rust-lang/cargo-bisect-rustc#438 to check which PR in the rollup #163095 may have caused the regression. #163062 seems likely.

Edit: Yes, cargo-bisect-rustc said dbcd41d (edit 2: and Reverting 3de0624 seems to fix the regression), so seems like #163062 caused the regression. CC @malezjaa @folkertdev

[malezjaa]

Well for now, I think we can revert the ARM r14 change and keep the AArch64 x30 change. That should fix the regression. I can look for a proper fix for {r14} later today.

[folkertdev]

Hmm yeah, {r14} works for arm but not for thumb, I think that needs further LLVM changes.

Can you revert the arm part and add a test for thumb (that would have caught this regression)?