Re: Clang/LLVM 'main' regression testing and -Wthread-safety

[Nathan Chancellor <nathan@kernel.org>]

On Wed, Feb 11, 2026 at 12:54:40PM +0100, Miguel Ojeda wrote:
> On Wed, Feb 11, 2026 at 11:59 AM Marco Elver <elver@google.com> wrote:
> >
> > I recall there was a CI system that did this, but lost track of what's
> > the latest.
> 
> I think you may be referring to
> https://github.com/ClangBuiltLinux/continuous-integration2.

Indeed.

> By the way, some time ago I asked Aaron (Cc'd) whether we could get
> the kernel into upstream LLVM's CI system, i.e. to at least build a
> minimal kernel configuration of a fixed, tagged release for one or two
> architectures. Would that help you?
> 
> (I also suggested Jose (Cc'd) the same, for GCC.)
> 
> The idea is that it helps both sides. It serves to catch issues very
> early on (including with subtle things like diagnostics), and thus
> avoids having to play catch up later to fix them (compiler) and/or
> workaround them (kernel) in either side.
> 
> To be clear, this wouldn't replace the other CIs that run out of tree,
> like Nathan's and the tests I do for Rust stuff etc., i.e. it is not
> about catching every compiler issue or testing the kernel as a
> product.
> 
> For context, this all happened because I asked the same to upstream
> Rust and it was a success there. In that case, the goal was to catch
> issues early with the unstable features we use, but it helps with
> other things too, and thus I think it would help just the same for
> Clang and GCC. So far, we only got pinged a few times, so it has not
> been a lot of work, and it did catch a couple issues (which sometimes
> may end with a preemptive kernel patch or fixing the upstream PR).
> 
> We also did it for upstream `bindgen`, and hopefully it will happen
> for `rustc_codegen_gcc` (currently we have an out-of-tree one) and GCC
> Rust eventually.
> 
> Anyway, this is a bit of a tangent, but I thought I could raise it up
> again and see what comes out of it...

It has always been on my backburner to try and get a build of the Linux
kernel as part of LLVM's continuous integration but I have never had the
time to sit down and interface with folks to see what exactly that
entails (I figured it would involve integration with LLVM's build bot
infrastructure but I have not dug in super hard to the documentation).
An arm64 and x86_64 defconfig build would probably catch quite a bit of
initial breakage and it would be easy to pin Linux's version and only
bump it when thoroughly tested.

Cheers,
Nathan