Re: [PATCH 2/3] perf symbols: Try reading the symbol table with libbfd

[RÃmi Bernon]

On 2020-06-24 15:25, Jiri Olsa wrote:
> On Wed, Jun 24, 2020 at 12:30:40PM +0200, Remi Bernon wrote:
>
> SNIP
>
> > +
> > +		symbols__insert(&dso->symbols, symbol);
> > +	}
> > +#ifdef bfd_get_section
> > +#undef bfd_asymbol_section
> > +#endif
> > +
> > +	symbols__fixup_end(&dso->symbols);
> > +	symbols__fixup_duplicate(&dso->symbols);
> > +	dso->adjust_symbols = 1;
> > +
> > +	err = 0;
> > +out_free:
> > +	free(symbols);
> > +out_close:
> > +	bfd_close(abfd);
> > +	return err;
> > +}
> > +#endif
> > +
> >   static bool dso__is_compatible_symtab_type(struct dso *dso, bool kmod,
> >   					   enum dso_binary_type type)
> >   {
> > @@ -1695,6 +1816,7 @@ int dso__load(struct dso *dso, struct map *map)
> >   		bool next_slot = false;
> >   		bool is_reg;
> >   		bool nsexit;
> > +		int bfdrc = -1;
> >   		int sirc = -1;
> >   
> >   		enum dso_binary_type symtab_type = binary_type_symtab[i];
> > @@ -1713,12 +1835,19 @@ int dso__load(struct dso *dso, struct map *map)
> >   			nsinfo__mountns_exit(&nsc);
> >   
> >   		is_reg = is_regular_file(name);
> > +#ifdef HAVE_LIBBFD_SUPPORT
> >   		if (is_reg)
> > +			bfdrc = dso__load_bfd_symbols(dso, name);
> > +#endif
>
> so the comment says that we try bfd when the libelf fails,
> but the code seems to do the oposite.. starts with bfd
> and skip libelf if it's ok
>
> jirka

Ah, sorry I forgot to reword the commit comment to match the changes, 
thanks for noticing. It should probably be something like that:

  Trying to read symbols on non-ELF binaries with libbfd, when
  supported, makes it possible for perf to report symbols and
  annotations for Windows applications running under Wine.

As you suggested, the libbfd code now tries to detect the binary format, 
 and early rejects if ELF is detected, so it then falls back to the 
libelf code path.
--
RÃmi Bernon <rbernon@xxxxxxxxxxxxxxx>