Re: bfd/elf.c assertion for multiple relocations to same section

[Joe Lawrence <joe dot lawrence at redhat dot com>]

On 4/16/19 7:19 PM, Alan Modra wrote:
> On Tue, Apr 16, 2019 at 05:13:08PM -0400, Joe Lawrence wrote:
> > I was wondering if anyone could explain the assertion in bfd/elf.c ::
> > bfd_section_from_shdr() where it is processing a SHT_REL/SHT_RELA section
> > and decides to return failure if *p_hdr is already set non-NULL:
> >
> > 	/* PR 17512: file: 0b4f81b7.  */
> > 	if (*p_hdr != NULL)
> > 	  goto fail;
>
> The BFD library is incapable of handling two rela sections with
> identical sh_info values *as relocation sections*.  It might be
> reasonable to modify BFD for the case you describe by creating a
> normal section for the second rela section, as is done for other
> cases.  See the comment starting "If this is an alloc section" a few
> lines above the lines you show here.  No interpretation of the second
> rela section contents would then be done.

Hi Alan,

Thanks for pointing out that source-code comment about handling 
relocation sections and ordinary sections:

   /* If this is an alloc section in an executable or shared
      library, or the reloc section does not use the main symbol
      table we don't treat it as a reloc section.  BFD can't
      adequately represent such a section, so at least for now,
      we don't try.  We just present it as a normal section.  We
      also can't use it as a reloc section if it points to the
      null section, an invalid section, another reloc section, or
      its sh_link points to the null section.  */

From this and your comment, if I understand correctly: we've created a 
valid ELF file that cannot have a BFD representation since we've doubled 
up on relocation sections that share the same sh_info value.  However, 
an ELF file that contained additional ordinary sections may share 
sh_info values and BFD will be happy as it will not process them as 
relocation sections.

Thanks,

-- Joe