Re: [PATCH bpf-next 2/4] bpftool: add support for subskeletons

[Delyan Kratunov <delyank@fb.com>]

On Thu, 2022-03-03 at 10:52 -0800, Delyan Kratunov wrote:
> > 
> > > 
> > > +               map_type_id = btf__find_by_name_kind(btf,
> > > bpf_map__section_name(map), BTF_KIND_DATASEC);
> > 
> > if we set obj_name to "", bpf_map__name() should return ELF section
> > name here, so no need to expose this as an API
> > 
> > 
> > oh, but also bpf_map__btf_value_type_id() should give you this ID directly
> 
> TIL, that's not obvious at all. There's a few places in gen.c that could be
> simplified then - find_type_for_map goes through slicing the complete name and
> walking over every BTF type to match on the slice. Is there some compatibility
> reason to do that or is btf_value_type_id always there?

Unfortunately, the internal datasec maps have value_type_id = key_value_type_id
= 0 i.e. void, so bpf_map__btf_value_type_id won't work out of the box.

I haven't looked if that's a bug all the way in clang-emitted object or
somewhere further on.

-- Delyan