Re: [PATCH bpf-next] selftests/bpf: fix build with new LLVM

[Anton Protopopov <a.s.protopopov@gmail.com>]

On 25/09/18 09:26AM, Alexei Starovoitov wrote:
> On Thu, Sep 18, 2025 at 9:21 AM Anton Protopopov
> <a.s.protopopov@gmail.com> wrote:
> >
> > On 25/09/18 08:02AM, Alexei Starovoitov wrote:
> > > On Thu, Sep 18, 2025 at 2:30 AM Anton Protopopov
> > > <a.s.protopopov@gmail.com> wrote:
> > > >
> > > > The progs/stream.c BPF program now uses arena helpers, so it includes
> > > > bpf_arena_common.h, which conflicts with the declarations generated
> > > > in vmlinux.h. This leads to the following build errors with the recent
> > > > LLVM:
> > > >
> > > >     In file included from progs/stream.c:8:
> > > >     .../tools/testing/selftests/bpf/bpf_arena_common.h:47:15: error: conflicting types for 'bpf_arena_alloc_pages'
> > > >        47 | void __arena* bpf_arena_alloc_pages(void *map, void __arena *addr, __u32 page_cnt,
> > > >           |               ^
> > > >     .../tools/testing/selftests/bpf/tools/include/vmlinux.h:229284:14: note: previous declaration is here
> > > >      229284 | extern void *bpf_arena_alloc_pages(void *p__map, void *addr__ign, u32 page_cnt, int node_id, u64 flags) __weak __ksym;
> > > >             |              ^
> > > >
> > > >     ... etc
> > >
> > > I suspect you're using old pahole.
> > > New one can transfer __arena tags into vmlinux.h
> >
> > Ok, TIL about CONFIG_PAHOLE_VERSION (before I've sent the patch,
> > I've updated the pahole, re-built the kernel, but didn't do `make
> > oldconfig` after updating pahole.)
> 
> Yeah. It's a footgun that few people are aware of :(
> I was bitten by it too.
> We have a small section about pahole in bpf_devel_QA.rst.
> We should probably expand it and list the common issues and how to fix them.
> Every week somebody sends a patch due to old pahole :(

And today I also learned about the "next" branch of the pahole repo.
The "master" branch was too old to generate kfunc header for a kfunc
compiled with a .cold part (namely, bpf_dynptr_slice).

As the "old pahole" thing happens to so many people, maybe there is
a way to in-source it into kernel?..