md_apply_fix3 bug on arm-linux

[Richard Earnshaw]

> A user reports that "mov r0, #UNDEFINED_SYMBOL" assembles to "mov r0,
> #0" without emitting any relocation or diagnostic, which clearly isn't
> right.
> 
> Also, this code:
> 
> .word 0
> foo:
> mov r0, #foo
> 
> assembles to "mov r0, #4", again with no relocation, which also seems
> fairly bogus.
> 
> A patch like the one below seems to clear it up.  Does this look right? 
> I guess I should also generate some cases for the testsuite.
> 
> Thanks
> 
> p.
> 
> 
> 2004-04-05  Philip Blundell  <philb@gnu.org>
> 
> 	* config/tc-arm.c (md_apply_fix3): Decline to handle IMMEDIATE
> 	relocs involving symbols.

Argh!!  md_apply_fix3 makes my brain hurt...

After a long stare at the code I wonder whether the bug is that the 
BFD_RELOC_ARM_IMMEDIATE case unconditionally sets fixP->fx_done; but I 
think for all the cases that matter the earlier code

  if (fixP->fx_addsy == 0 && !fixP->fx_pcrel)
    fixP->fx_done = 1;

will have already done this.

> 
> --- gas/config/tc-arm.c	2004-04-05 19:40:42.000000000 +0100
> +++ gas/config/tc-arm.c	2004-04-05 19:40:36.000000000 +0100
> @@ -12190,6 +12167,10 @@
>    switch (fixP->fx_r_type)
>      {
>      case BFD_RELOC_ARM_IMMEDIATE:
> +      /* If a symbol is involved, we must leave this to the linker.  */
> +      if (fixP->fx_addsy)
> +	return;
> +
>        newimm = validate_immediate (value);
>        temp = md_chars_to_number (buf, INSN_SIZE);
>  
>