Re: binutils 2.19.92 linker broke with --as-needed flag

[Vincent Wen <wenvincent90 at gmail dot com>]

Eric,

Thanks a lot for clarification!

Alan,

So I am testing the following patch per our discussion. Please let me
know if anything missing.

Thanks,

Vincent

--- a/bfd/elfxx-mips.c    2013-02-10 16:50:56.000000000 -0800
+++ b/bfd/elfxx-mips.c    2013-02-10 17:14:06.000000000 -0800
@@ -4681,12 +4681,9 @@
   h->non_elf = 0;
   h->def_regular = 1;
   h->type = STT_OBJECT;
+  h->other = STV_HIDDEN;
   elf_hash_table (info)->hgot = h;

-  if (info->shared
-      && ! bfd_elf_link_record_dynamic_symbol (info, h))
-    return FALSE;
-
   amt = sizeof (struct mips_got_info);
   g = bfd_alloc (abfd, amt);
   if (g == NULL)

On Thu, Mar 28, 2013 at 3:02 PM, Eric Christopher <echristo@gmail.com> wrote:
> On Wed, Mar 27, 2013 at 7:36 PM, Alan Modra <amodra@gmail.com> wrote:
>> On Wed, Mar 27, 2013 at 06:54:39PM -0700, Vincent Wen wrote:
>>> Alan,
>>>
>>> Who is the MIPS maintainer?
>>
>> Theoretically Eric Christopher, but binutils/MAINTAINERS lists an
>> old email address and it's been a while since we've seen him around
>> here.  Eric, can you comment on the code below?
>>
>
> Yeah, it's been a while. Sorry about that. I'll update everything at some point
> though I'm no longer doing anything with MIPS (or mn10300) so while I can
> pitch in on occasion I could also move myself to write-after-approval
> if you'd like.
>
>>> On Wed, Mar 27, 2013 at 5:02 PM, Alan Modra <amodra@gmail.com> wrote:
>>> > On Wed, Mar 27, 2013 at 10:24:51AM -0700, Vincent Wen wrote:
>>> >> Alan,
>>> >>
>>> >> I'd like to know whether the patch for this issue is accepted or not?
>>> >
>>> > That really is up to one of the MIPS maintainers.  As I said in
>>> > http://sourceware.org/ml/binutils/2013-02/msg00159.html the MIPS backend
>>> > wants _GLOBAL_OFFSET_TABLE_ dynamic for some reason.
>>> >
>>> >   bh = NULL;
>>> >   if (! (_bfd_generic_link_add_one_symbol
>>> >          (info, abfd, "_GLOBAL_OFFSET_TABLE_", BSF_GLOBAL, s,
>>> >           0, NULL, FALSE, get_elf_backend_data (abfd)->collect, &bh)))
>>> >     return FALSE;
>>> >
>>> >   h = (struct elf_link_hash_entry *) bh;
>>> >   h->non_elf = 0;
>>> >   h->def_regular = 1;
>>> >   h->type = STT_OBJECT;
>>> >   elf_hash_table (info)->hgot = h;
>>> >
>>> >   if (info->shared
>>> >       && ! bfd_elf_link_record_dynamic_symbol (info, h))
>>> >     return FALSE;
>>> >
>>> > I think that is just plain wrong, but I don't know the history of this
>>> > code.
>
> It's been that way since the original revision and that's a long long time
> ago. :) I think I noticed it at one point and thought it odd, but I'm not sure
> what it was actually for... some oddity on SGI machines that depended
> upon a dynamic _GLOBAL_OFFSET_TABLE_ that only appeared if there
> were entries? If there's anything weird the easiest assumption is "SGI
> was weird".
>
> Sorry I couldn't be more help.
>
> -eric