Re: [PATCH 1/1] perf bpf: Move BPF disassembly routines to separate file to avoid clash with capstone bpf headers

[Arnaldo Carvalho de Melo <acme@kernel.org>]

On Thu, Aug 01, 2024 at 08:18:23AM -0700, Ian Rogers wrote:
> On Wed, Jul 31, 2024 at 11:51 AM Arnaldo Carvalho de Melo
> <acme@kernel.org> wrote:
> >
> > On Wed, Jul 31, 2024 at 10:07:50AM -0700, Namhyung Kim wrote:
> > > On Wed, Jul 31, 2024 at 8:12 AM Arnaldo Carvalho de Melo
> [snip]
> > > I think this can be gated by LIBBFD and LIBBPF config, but not sure
> > > it can express the both requirements easily.
> >
> > It is possible, but as discussed with Ian in another message, the
> > cheapest way to do it was, I think, like I did, in the end it is just a
> > stub returning an error when those libraries are not linked with perf.
> >
> > When we manage to implement the disassembly of BPF using something other
> > than libbfd we can spend more time in this area, I think.
> 
> Not following this, can you show an example perf command. There is a
> BPF disassembler already in the kernel tree:
> https://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools-next.git/tree/kernel/bpf/disasm.c?h=perf-tools-next
> and in use in bpftool that we use in the perf build.

Well, looking at the files in this patch shows it, here, more precisely:


tools/perf/util/disasm_bpf.c

#else // defined(HAVE_LIBBFD_SUPPORT) && defined(HAVE_LIBBPF_SUPPORT)
int symbol__disassemble_bpf(struct symbol *sym __maybe_unused, struct annotate_args *args __maybe_unused)
{
        return SYMBOL_ANNOTATE_ERRNO__NO_LIBOPCODES_FOR_BPF;
}
#endif // defined(HAVE_LIBBFD_SUPPORT) && defined(HAVE_LIBBPF_SUPPORT)

- Arnaldo