`options(att_syntax)` doesn't do anything for `global_asm!` and `naked_asm!`

[SabrinaJewson]

Edit(trevor): this doesn't need compiler_builtins to repro #143542 (comment)

I tried this code:

cargo new x
cd x
RUSTFLAGS="--emit=asm -Cllvm-args=-x86-asm-syntax=intel" cargo +nightly build -Zbuild-std

I expected to see this happen: A successful build, with Intel-syntax assembly emitted. (as per #48011)

Instead, this happened:

A lot of inline assembly syntax errors

   Compiling compiler_builtins v0.1.160 (~/.local/share/rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/compiler-builtins/compiler-builtins)
   Compiling rustc-std-workspace-core v1.99.0 (~/.local/share/rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/rustc-std-workspace-core)
   Compiling alloc v0.0.0 (~/.local/share/rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/alloc)
   Compiling panic_abort v0.0.0 (~/.local/share/rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/panic_abort)
   Compiling libc v0.2.174
   Compiling cfg-if v1.0.1
   Compiling memchr v2.7.5
   Compiling adler2 v2.0.1
   Compiling rustc-demangle v0.1.25
error: unknown token in expression
  |
note: instantiated into assembly here
 --> <inline asm>:8:20
  |
8 |             pushq  %rbp
  |                    ^

error: unknown token in expression
   |
note: instantiated into assembly here
  --> <inline asm>:11:20
   |
11 |             movq   %rsp, %rbp
   |                    ^

error: unknown token in expression
   |
note: instantiated into assembly here
  --> <inline asm>:14:20
   |
14 |             mov    %rax,%r11        // duplicate %rax as we're clobbering %r11
   |                    ^

error: unknown token in expression
   |
note: instantiated into assembly here
  --> <inline asm>:30:20
   |
30 |             cmp    $0x1000,%r11
   |                    ^

error: unknown token in expression
   |
note: instantiated into assembly here
  --> <inline asm>:33:20
   |
33 |             sub    $0x1000,%rsp
   |                    ^

error: unknown token in expression
   |
note: instantiated into assembly here
  --> <inline asm>:34:20
   |
34 |             test   %rsp,8(%rsp)
   |                    ^

error: unknown token in expression
   |
note: instantiated into assembly here
  --> <inline asm>:35:20
   |
35 |             sub    $0x1000,%r11
   |                    ^

error: unknown token in expression
   |
note: instantiated into assembly here
  --> <inline asm>:36:20
   |
36 |             cmp    $0x1000,%r11
   |                    ^

error: unknown token in expression
   |
note: instantiated into assembly here
  --> <inline asm>:42:20
   |
42 |             sub    %r11,%rsp
   |                    ^

error: unknown token in expression
   |
note: instantiated into assembly here
  --> <inline asm>:43:20
   |
43 |             test   %rsp,8(%rsp)
   |                    ^

error: unknown token in expression
   |
note: instantiated into assembly here
  --> <inline asm>:48:20
   |
48 |             add    %rax,%rsp
   |                    ^

   Compiling unwind v0.0.0 (~/.local/share/rustup/toolchains/nightly-x86_64-unknown-linux-gnu/lib/rustlib/src/rust/library/unwind)
error: could not compile `compiler_builtins` (lib) due to 11 previous errors
warning: build failed, waiting for other jobs to finish...

Meta

rustc --version --verbose:

rustc 1.90.0-nightly (667787527 2025-07-02)
binary: rustc
commit-hash: 6677875279b560442a07a08d5119b4cd6b3c5593
commit-date: 2025-07-02
host: x86_64-unknown-linux-gnu
release: 1.90.0-nightly
LLVM version: 20.1.7

[folkertdev]

That is surprising, given that LLVM is normally able to parse the att syntax and re-emit it as intel, e.g. https://godbolt.org/z/so1T1zM4n.

cc @tgross35 wasn't there a plan to use intel syntax everywhere for x86_64? The code it errors on is __rust_probestack btw. I can reproduce the issue there by running this in compiler-builtins:

RUSTFLAGS="--emit=asm -Cllvm-args=-x86-asm-syntax=intel" cargo +nightly build --no-default-features

So it's quite likely that __rust_probestack is actually untested.

[tgross35]

Maybe LLVM supports parsing inline asm but not global? Passing the llvm-arg eats it with this simple example too

core::arch::global_asm!("
    .globl func
    func:
        movq (%rdi), %rax
        retq
    ",
    options(att_syntax),
);

https://rust.godbolt.org/z/WjzjPa7Kj

I feel like I've hit this with cargo asm as well

[tgross35]

So it's quite likely that __rust_probestack is actually untested.

This only occurs with a specific flag, no? If it's the same thing that shows up when running cargo asm, I've been hitting this quite a while before any of the more recent changes to that function.

[folkertdev]

well the --no-default-features flag seemed to make a difference for me. Without it apparently skips that file entirely by default. But yeah I also got it to reproduce with earlier commits.

[tgross35]

🤔 okay, so options(att_syntax) seems to not actually do anything in global_asm, we're not telling LLVM that it's AT&T. Adding .att_syntax fixes the issue, but the option should be doing that for us.

So this should just be a fix to global_asm/naked_asm directives