Re: [PATCH] Move DONT_USE_BOOTSTRAP_MAP defininition todl-machine.h

[Andreas Jaeger <aj at suse dot de>]

Jakub Jelinek <jakub@redhat.com> writes:

> On Thu, Aug 22, 2002 at 08:07:26PM -0400, Roland McGrath wrote:
>> I just experimented with making a static variable in rtld.c and redefining
>> GL using that locally, with strong_alias to define _rtld_global.  It seems
>> to work.  Can platforms like Alpha and SPARC avoid the relocs for local
>> symbols?
>
> Ok, Alpha can do it, just tested:
> static int i;
> int j __attribute__((visibility("hidden")));
> int k[100] __attribute__((visibility("hidden"), section(".sdata")));
> int foo (void)
> {
>   return i;
> }
> int bar (void)
> {
>   return j;
> }
> int baz (void)
> {
>   return k[0];
> }
>
> on CVS head, unfortunately it doesn't do this in visibility
> backport for 3.2 (yet, I'll have a look at this next weak).
> On SPARC, it is not possible until binutils/gcc/glibc are
> changed for the proposed new relocs.
> IA-64 is ok with both 3.2 backport and 3.3 mainline.
>
> So it IMHO looks like we need configury test for this (which will be
> unfortunately arch dependend, and have a default for not tested architecture
> (use bootstrap map)).

What should I test on x86-64?

Btw. elf/ld.so crashes also on x86-64 with current CVS.  Is this the
same problem?

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj