V3 vs. AIX

[Mark Mitchell]

I think there's probably little Benjamin can do about this problem.

David, I'm using GAS per your suggestion (as you mentioned, it is
otherwise the case that the IBM assembler falls over).

However, I think GAS is perhaps doing something wrong.  Either that or
AIX ld.  Or something.

We are faulting in the middle of the cross-TOC glue code.  In
particular:

    1004b4a8:   81 82 02 60     l       r12,608(r2)
                        1004b4aa: R_TOC _ZNSs10_M_replaceIPcEERSsSt17__normal_iteratorIS0_SsES3_T_S4_St20forward_iterator_tag+0xffffffffdffe16e8
    1004b4ac:   90 41 00 14     st      r2,20(r1)
    1004b4b0:   80 0c 00 00     l       r0,0(r12)
    1004b4b4:   80 4c 00 04     l       r2,4(r12)
  
Here, we fault when doing the load at 1004b4b0.  Here:

  p (void*) $r12
  $24 = (void *) 0xcabf0

And:

  p (void*) $r2
  $27 = (void *) 0xf0d296b8

Apparently, this is not a happy value.  Any ideas what could be going
wrong here?

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com