This is the mail archive of the
binutils@sources.redhat.com
mailing list for the binutils project.
[PATCH] Fix broken -pie on ppc32
On Fri, Feb 20, 2004 at 11:06:24AM +1030, Alan Modra wrote:
> On Thu, Feb 19, 2004 at 04:07:06PM -0800, Geoff Keating wrote:
> > Jakub Jelinek <jakub@redhat.com> writes:
> > [about undefined weak syms]
> > > Should ld in this case (on all arches) just skip the dynamic relocation
> > > and store 0?
> >
> > I believe that the right thing to do is to create the dynamic
> > relocation. ld should assume that an undefined weak symbol is
> > actually defined in a shared library.
>
> I think so too, but this is a gray area. The handling of weak symbols
> isn't fully defined in the gABI ELF spec, so you could make a case for
> all undefined weaks being resolved to zero at link time. In any case,
> we ought to be consistent in our handling of undefined weaks. Since
> we currently make them dynamic in other cases, I think we should here
> too.
Like this?
2004-02-20 Jakub Jelinek <jakub@redhat.com>
* elf32-ppc.c (allocate_dynrelocs): Create dynsym for undef weak
symbols used in PIE relocs.
--- bfd/elf32-ppc.c.jj 2004-01-19 00:08:02.000000000 +0100
+++ bfd/elf32-ppc.c 2004-02-20 13:30:13.947305152 +0100
@@ -3189,6 +3189,18 @@ allocate_dynrelocs (struct elf_link_hash
if (ELF_ST_VISIBILITY (h->other) != STV_DEFAULT
&& h->root.type == bfd_link_hash_undefweak)
eh->dyn_relocs = NULL;
+
+ /* Make sure undefined weak symbols are output as a dynamic symbol
+ in PIEs. */
+ if (info->pie
+ && eh->dyn_relocs != NULL
+ && h->dynindx == -1
+ && h->root.type == bfd_link_hash_undefweak
+ && (h->elf_link_hash_flags & ELF_LINK_FORCED_LOCAL) == 0)
+ {
+ if (! bfd_elf32_link_record_dynamic_symbol (info, h))
+ return FALSE;
+ }
}
else if (ELIMINATE_COPY_RELOCS)
{
Jakub