Re: [PATCH] MIPS bfd: abi 64 ld -r

[Alexandre Oliva <aoliva at redhat dot com>]

On Sep  4, 2002, Alexandre Oliva <aoliva@redhat.com> wrote:

>> I would like to see the exact case so I can try to reproduce it here.

> I'll try to come up with something for you soonish.

Ok, here comes the obvious testcase:

.data
foo:
.word bar
bar:

assemble it with the IRIX assembler (n32), and you get:

0000000000000000 <.data>:
   0:   00000004        sllv    $zero,$zero,$zero
                        0: R_MIPS_32    .data
               ^ note the in-place addend,    ^ nothing here

now link this with the GNU linker, before installing the patch, and
you get:

0000000010100098 <_fdata>:
    10100098:   10100098        beq     $zero,$s0,101002fc <__bss_start+0x244>
                      ^^  oops, where did the +4 go?

but if you link it with the IRIX linker, you get:

0000000010014000 <.data>:
    10014000:   10014004        beq     $zero,$at,0x10024014
                       ^ here it is

I've just noticed, with this old build of GNU ld I still had around on
the IRIX box, that `ld -r'ing the object file produced by GNU as
produces even worse results:

ld-new: test.o: relocation size mismatch in t.os section .data

I don't have an IRIX-hosted newer native toolchain to verify whether
this problem is already fixed, or if the patch has fixed it, but ld -r
seems to have worked for me lately.  At least it generates an object
file at the end; I still don't know whether the result of the linking
is usable :-)

-- 
Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
Red Hat GCC Developer                 aoliva@{redhat.com, gcc.gnu.org}
CS PhD student at IC-Unicamp        oliva@{lsd.ic.unicamp.br, gnu.org}
Free Software Evangelist                Professional serial bug killer