Home Original page

as for sh-any-elf (weak symbol)

kaz Kojima kkojima@rr.iij4u.or.jp
Thu Oct 26 07:08:00 GMT 2000 
Hi,

It seems that as for sh-any-elf target gives a wrong valued
in place relocation for the reference to the weak data symbol.
For example,

        .data
        .align 2
d0:
        .long   1
        .global w0
w0:
        .long   0
        .weak w0
        .text
        .align 5
f:
        mov.l   .L3,r1
        mov.l   @r1,r0
        rts     
        nop
        .align 2
.L3:
        .long   w0

is assembled to

00000000 <f>:
   0:   d1 01           mov.l   8 <f+0x8>,r1    ! 0xfffffffc
   2:   60 12           mov.l   @r1,r0
   4:   00 0b           rts
   6:   00 09           nop
   8:   ff ff           .word 0xffff
   a:   ff fc           .word 0xfffc

with a relocation

  Offset    Info  Type            Symbol's Value  Symbol's Name          Addend
  00000008  00601 R_SH_DIR32            00000004  w0                        + 0

and ld gives

00001000 <f>:
    1000:       d1 01           mov.l   1008 <f+0x8>,r1 ! 0x108c
    1002:       60 12           mov.l   @r1,r0
    1004:       00 0b           rts
    1006:       00 09           nop
    1008:       00 00           .word 0x0000
    100a:       10 8c           mov.l   r8,@(48,r0)

which returns the content of d0 not of w0.

    11: 0000108c     0 NOTYPE  LOCAL  DEFAULT    2 d0
    12: 00001000     0 NOTYPE  LOCAL  DEFAULT    1 f
    13: 00001090     0 NOTYPE  WEAK   DEFAULT    2 w0

If the line .weak w0 is deleted from the source, everything
is ok.  I guess that the code

  /* The function adjust_reloc_syms won't convert a reloc against a weak
     symbol into a reloc against a section, but bfd_install_relocation
     will screw up if the symbol is defined, so we have to adjust val here
     to avoid the screw up later.  */
  if (fixP->fx_addsy != NULL
      && S_IS_WEAK (fixP->fx_addsy))
    val -= S_GET_VALUE  (fixP->fx_addsy);

in function md_apply_fix of gas/config/tc-sh.c causes this.

--
kazumoto Kojima
Mail	kkojima@rr.iij4u.or.jp
URL	http://rr.iij4u.or.jp/~kkojima
More information about the Binutils mailing list