Re: [PATCH bpf-next] bpf: Disable -Wmissing-declarations for globally-linked kfuncs

[David Vernet <void@manifault.com>]

On Wed, Aug 16, 2023 at 08:48:16PM -0700, Alexei Starovoitov wrote:
> On Wed, Aug 16, 2023 at 8:38 PM Yonghong Song <yonghong.song@linux.dev> wrote:
> >
> >
> >
> > On 8/16/23 8:06 AM, David Vernet wrote:
> > > We recently got an lkp warning about missing declarations, as in e.g.
> > > [0]. This warning is largely redundant with -Wmissing-prototypes, which
> > > we already disable for kfuncs that have global linkage and are meant to
> > > be exported in BTF, and called from BPF programs. Let's also disable
> > > -Wmissing-declarations for kfuncs. For what it's worth, I wasn't able to
> > > reproduce the warning even on W <= 3, so I can't actually be 100% sure
> > > this fixes the issue.
> > >
> > > [0]: https://lore.kernel.org/all/202308162115.Hn23vv3n-lkp@intel.com/
> >
> > Okay, I just got a similar email to [0] which complains
> >    bpf_obj_new_impl, ..., bpf_cast_to_kern_ctx
> > missing declarations.
> >
> > In the email, the used compiler is
> > compiler: gcc-7 (Ubuntu 7.5.0-6ubuntu2) 7.5.0
> >
> > Unfortunately, I did not have gcc-7 to verify this.
> > Also, what is the minimum gcc version kernel supports? 5.1?
> 
> pahole and BTF might be broken in such old GCC too.
> Maybe we should add:
> config BPF_SYSCALL
>         depends on GCC_VERSION >= 90000 || CLANG_VERSION >= 130000

It seems prudent to formally declare minimum compiler versions. Though
modern gcc and clang also support -Wmissing-declarations, so maybe we
should merge this patch regardless? Just unfortunate to have to add even
more boilerplate just to get the compiler off our backs.