asm.rs assumes that multiple global_asm statements are concatenated

[jannic]

In https://github.com/rust-embedded/riscv/blob/master/riscv-rt/src/asm.rs there are multiple global_asm! macro calls (generated from multiple cfg_global_asm! macro calls`) with code that only make sense if the generated machine code is a single contiguous block.

However, as far as I can tell, rust doesn't guarantee that.

To the contrary, https://dirname.github.io/rust-std-doc/unstable-book/library-features/global-asm.html contains the following language: "However, you should not rely on assembler state (e.g. assembler macros) defined in one global_asm! to be available in another one. It is implementation-defined whether the multiple usages are concatenated into one or assembled separately."

(Of course, even if assembled separately, the generated code could still be contiguous. I just didn't find any guarantee that it will.)

In the rust-embedded matrix channel, @Dirbaio suggested to use a single cfg_global_asm! block. To avoid unwieldy repetitive cfg statements, the cfg_global_asm! macro could be extended to allow for nested blocks:

/// Parse cfg attributes inside a global_asm call.
macro_rules! cfg_global_asm {
    {@inner, [$($x:tt)*], } => {
        global_asm!{$($x)*}
    };
    (@inner, [$($x:tt)*], #[cfg($meta:meta)] { $($y:tt)* } $($rest:tt)*) => {
        #[cfg($meta)]
        cfg_global_asm!{@inner, [$($x)*], $($y)* $($rest)*}
        #[cfg(not($meta))]
        cfg_global_asm!{@inner, [$($x)*], $($rest)*}
    };
    (@inner, [$($x:tt)*], #[cfg($meta:meta)] $asm:literal, $($rest:tt)*) => {
        #[cfg($meta)]
        cfg_global_asm!{@inner, [$($x)* $asm,], $($rest)*}
        #[cfg(not($meta))]
        cfg_global_asm!{@inner, [$($x)*], $($rest)*}
    };
    {@inner, [$($x:tt)*], $asm:literal, $($rest:tt)*} => {
        cfg_global_asm!{@inner, [$($x)* $asm,], $($rest)*}
    };
    {$($asms:tt)*} => {
        cfg_global_asm!{@inner, [], $($asms)*}
    };
}

untested, could be used like this:

cfg_global_asm!(
    "blah",
    #[cfg(foo)]
    "blah",
    #[cfg(bar)] {
        "blah",
        #[cfg(foo)]
        "blah",
    }
)

(https://libera.catirclogs.org/rust-embedded/2025-07-01#38366632)

[romancardenas]

Thanks! Will try this soon