On Thu, Jun 4, 2020 at 12:47 AM Alan Modra <amodra@gmail.com> wrote:
>
> On Wed, Jun 03, 2020 at 11:55:08AM -0700, H.J. Lu wrote:
> > --- a/bfd/elfxx-mips.c
> > +++ b/bfd/elfxx-mips.c
> > @@ -8618,7 +8618,6 @@ _bfd_mips_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
> >        struct elf_link_hash_entry *h;
> >        bfd_boolean can_make_dynamic_p;
> >        bfd_boolean call_reloc_p;
> > -      bfd_boolean constrain_symbol_p;
> >
> >        r_symndx = ELF_R_SYM (abfd, rel->r_info);
> >        r_type = ELF_R_TYPE (abfd, rel->r_info);
> > @@ -8653,12 +8652,6 @@ _bfd_mips_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
> >        and if pointer equality therefore doesn't matter.  */
> >        call_reloc_p = FALSE;
> >
> > -      /* Set CONSTRAIN_SYMBOL_P if we need to take the relocation
> > -      into account when deciding how to define the symbol.
> > -      Relocations in nonallocatable sections such as .pdr and
> > -      .debug* should have no effect.  */
> > -      constrain_symbol_p = ((sec->flags & SEC_ALLOC) != 0);
> > -
>
> So "normal" sections will result in constrain_symbol_p being true.
>
> >        switch (r_type)
> >       {
> >       case R_MIPS_CALL16:
> > @@ -8740,21 +8733,10 @@ _bfd_mips_elf_check_relocs (bfd *abfd, struct bfd_link_info *info,
> >       case R_MIPS_NONE:
> >       case R_MIPS_JALR:
> >       case R_MICROMIPS_JALR:
> > -       /* These relocations have empty fields and are purely there to
> > -          provide link information.  The symbol value doesn't matter.  */
> > -       constrain_symbol_p = FALSE;
> > -       break;
> > -
>
> But in for some special relocs it should be false.  I think removing
> constrain_symbol_p is just wrong.  You should keep it but initialise
> to true.

Done.

Here is the updated patch.   OK for master?

-- 
H.J.