Re: [PATCH dwarves v2 0/2] dwarf_loader: Fix --lang_exclude on merged CUs

[Alan Maguire <almagui@kernel.org>]

On 11/5/26 4:48 PM, Alan Derk wrote:
> On Wed, Nov 4, 2026 at 7:38 AM Alan Maguire wrote:
>> Thanks for doing this work! Sorry for the delay in getting to it, I'm
>> looking at the patches now.
> 
> Of course. Please take your time on the review. And thanks for the
> background on the Rust BTF work.
>

Changes are applied, thanks for fixing these issues! I'm thinking we should
probably add a CI job that builds an LTO vmlinux; I'll look at doing that
shortly.
 
>> We've been working on a solution; essentially, the idea is to detect
>> such cases and fix up the type graph in pahole using a new
>> "rust_bindgen_replace" feature so that the final BTF does not have to
>> carry such duplicates. It will only run for Rust CUs, [...]
> 
> Patches 1-2 are general --lang_exclude fixes (both reproducers are
> C++/C). The Kconfig follow-up keeps --lang_exclude=rust, so it is a
> stopgap: Rust stays out of BTF, as it does without LTO, until
> rust_bindgen_replace brings Rust types in. I realize it may not be
> worth it, though it would -- if I understand correctly -- unblock RUST with
> LTO and BTF in the meantime.
> 
> To check how this series of patches fits with your branch, I applied it on
> top of pahole-next-btf-rust (b5e46f8; it applies cleanly) and tried both on
> the 7.2.9 kernels I tested the series on (x86_64 ThinLTO, full LTO
> and distributed ThinLTO, arm64 ThinLTO) and a non-LTO x86_64 build,
> booting each in qemu and loading rust_minimal, rust_print and
> kobject-example:
> 
> - With kbuild's flags (--lang_exclude=rust), the branch alone fails to
>   build the Rust sample modules on the LTO kernels, as patch 1
>   describes. With the series on top, all five kernels build, boot and
>   load the three modules with BTF.
> - With rust_bindgen_replace and nothing excluded, the series changes
>   nothing: vmlinux and module BTF are byte-identical with and without
>   it.
> 
> One thing I noticed along the way, since it's the merged-CU language
> that patch 1 deals with: under LTO the merged CU has one language,
> taken from its first CU (a C or asm file for vmlinux), so
> cu__is_rust() is false and rust_bindgen_replace doesn't run on
> vmlinux. Without LTO it replaced 564 types as expected.
> 

Ah, that's a great point. May have to rethink that part. Thanks again
for giving the changes a try!

Alan