Rollup merge of #131755 - jfrimmel:avr-rjmp-offset-regression-test, r=jieyouxu

Regression test for AVR `rjmp` offset

This adds a regression test for #129301 by minimizing the code in the linked issue and putting it into a `#![no_core]`-compatible format, so that it can easily be used within an `rmake`-test. This needs to be a `rmake` test (opposed to a `tests/assembly` one), since the linked issue describes, that the problem only occurs if the code is directly compiled. Note, that `lld` is used instead of `avr-gcc`; see the [comments](https://github.com/rust-lang/rust/pull/131755#issuecomment-2416469675) [below](https://github.com/rust-lang/rust/pull/131755#issuecomment-2417160045).
Closes #129301.

To show, that the test actually catches the wrong behavior, this can be tested with a faulty rustc:
```bash
$ rustup install nightly-2024-08-19
$ rustc +nightly-2024-08-19 -C opt-level=s -C panic=abort --target avr-unknown-gnu-atmega328 -Clinker=build/x86_64-unknown-linux-gnu/ci-llvm/bin/lld -Clink-arg='--entry=main' -o compiled tests/run-make/avr-rjmp-offset/avr-rjmp-offsets.rs
$ llvm-objdump -d compiled | grep '<main>' -A 6
000110b4 <main>:
   110b4: 81 e0         ldi     r24, 0x1
   110b6: 92 e0         ldi     r25, 0x2
   110b8: 85 b9         out     0x5, r24
   110ba: 95 b9         out     0x5, r25
   110bc: fe cf         rjmp    .-4
```
One can see, that the wrong label offset (`4` instead of `6`) is produced, which would trigger an assertion in the test case.

This would be a good candidate for using the `minicore` proposed in #130693. Since this is not yet merged, there is a FIXME.

r? Patryk27
I think, you are the yet-to-be official target maintainer, hence I'll assign to you.

`@rustbot` label +O-AVR

0 files changed, 0 insertions, 0 deletions