Re: [libunwind] problems parsing dwarf frame info on amd64 optimized code

[Max Asbock]

On Friday 12 March 2004 11:40, David Mosberger wrote:
> >>>>> On Fri, 12 Mar 2004 10:37:02 -0800, Max Asbock <address@hidden> said:
> 
>   Max> In apply_reg_state() this could translate to:
> 
>   Max> for (i = 0; i < DWARF_NUM_PRESERVED_REGS; ++i)
>   Max> {
>   Max> switch ((dwarf_where_t) rs->reg[i].where)
>   Max> {
>   Max> ......
>   Max> case DWARF_WHERE_SAME:
>   Max> if (i == <REGNUM_SP>)
>   Max> if (ret = dwarf_put(c, c->loc[i], cfa))
>   Max> return ret;
>   Max> .......
> 
> 
> That may work for this particular case, but it's almost certainly
> wrong: it will modify the initial machine-state, which is not what you
> want.  Instead, I think you'll need to use c->cfa "in the right
> places" (I'm not sure yet what the right places are...).
> 
Then how about:     
 if (rs->reg[DWARF_CFA_REG_COLUMN].where == DWARF_WHERE_REG)
   {
      if ((rs->reg[DWARF_CFA_REG_COLUMN].val == <REGNUM_SP>)
           && (rs->reg[<REGNUM_SP>].where == DWARF_WHERE_SAME))
        cfa = c->cfa;
     else
        read the value for the cfa from the register
  }

This works as well.

max