Re: [PATCH bpf-next 1/3] bpf: Add __bpf_kfunc tag for marking kernel functions as kfuncs

[David Vernet]

On Fri, Jan 06, 2023 at 05:04:02PM -0800, Alexei Starovoitov wrote:
> On Fri, Jan 6, 2023 at 11:51 AM David Vernet <void@xxxxxxxxxxxxx> wrote:
> >
> > kfuncs are functions defined in the kernel, which may be invoked by BPF
> > programs. They may or may not also be used as regular kernel functions,
> > implying that they may be static (in which case the compiler could e.g.
> > inline it away), or it could have external linkage, but potentially be
> > elided in an LTO build if a function is observed to never be used, and
> > is stripped from the final kernel binary.
> >
> > We therefore require some convenience macro that kfunc developers can
> > use just add to their kfuncs, and which will prevent all of the above
> > issues from happening. This is in contrast with what we have today,
> > where some kfunc definitions have "noinline", some have "__used", and
> > others are static and have neither.
> >
> > In addition to providing the obvious correctness benefits, having such a
> > macro / tag also provides the following advantages:
> >
> > - Giving an easy and intuitive thing to query for if people are looking
> >   for kfuncs, as Christoph suggested at the kernel maintainers summit
> >   (https://lwn.net/Articles/908464/). This is currently possible by
> >   grepping for BTF_ID_FLAGS(func, but having something more self
> >   describing would be useful as well.
> >
> > - In the future, the tag can be expanded with other useful things such
> >   as the ability to suppress -Wmissing-prototype for the kfuncs rather
> >   than requiring developers to surround the kfunc with __diags to
> >   suppress the warning (this requires compiler support that as far as I
> >   know currently does not exist).
> 
> Have you considered doing bpf_kfunc_start/bpf_kfunc_end ?
> The former would include:
> __diag_push(); __diag_ignore_all(); __used noinline

Yeah that's certainly an option. The downside is that all functions
within scope of the __diag_push() will be affected, and sometimes we mix
kfuncs with non-kfuncs (including e.g. static helper functions that are
used by the kfuncs themselves). -Wmissing-prototypes isn't a big deal,
but __used and noinline are kind of unfortunate. Not a big deal though,
it'll just result in a few extra __bpf_kfuncs_start() and
__bpf_kfuncs_end() sprinkled throughout to avoid them being included.
The upside is of course that we can get rid of the __diag_push()'es we
currently have to prevent -Wmissing-prototypes.

Wdyt? I do like the idea of getting rid of those ugly __diag_push()'es.
And we could always go back to using a __bpf_kfunc macro if and when
compilers ever support using attributes to ignore warnings for specific
functions.

> 
> Also how about using bpf_kfunc on the same line ?
> Then 'git grep' will be easier.

Sure, if we keep this approach I'll do this in v2.