[AVR][PATCH] Add linker relaxation support / Fix 64 bit bug in gas rev. 5

[Nick Clifton]

Hi Björn,

> The simple test case
> 
>  .data
> here:
>  .set sym, here
>  .long sym
> 
> Output of objdump:
> 
> /tmp/test.o:     file format elf32-avr
> 
> RELOCATION RECORDS FOR [.data]:
> OFFSET   TYPE              VALUE 
> 00000000 R_AVR_32          sym

Are you using version of GAS with your relaxation patch applied ? 
Without it (but with linkrelax set to 1 in md_begin) I get this:

   RELOCATION RECORDS FOR [.data]:
   OFFSET   TYPE              VALUE
   00000000 R_AVR_32          .data

Ie the reloc has been transformed into a reloc against the .data section 
symbol.

Note - the .set pseudo-op does not create an alias, it creates an 
equivalence, so in your test case it creates a new symbol called "sym" 
whose value is the same as that of the symbol called "here".  So for 
example with your test case if I run "objdump -t test.o" I see:

   SYMBOL TABLE:
   00000000 l    d  .text  00000000 .text
   00000000 l    d  .data  00000000 .data
   00000000 l    d  .bss   00000000 .bss
   00000000 l       .data  00000000 here
   00000000 l       .data  00000000 sym

ie both "sym" and "here" exist as local symbols.

> Is it possible that in tc_gen_reloc after
> 
> *reloc->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
> 
> one would need some kind of recursion for the case where the value of the 
> symbol sym itself is taken from another symbol?

That should not be necessary.  What should have happened is that before 
tc_gen_reloc() is called any relocs against local symbols should have 
been either resolved or else transformed into relocs against global 
symbols (usually section symbols).  See adjust_reloc_syms() in 
gas/write.c.  I am guessing that in your case tc_fix_adjustable() is 
defined and not allowing the local symbol reference to be transformed 
into a section symbol reference.

Cheers
   Nick