inconsistent sigtramp code in mips target

[Kevin Nomura <nomura@netapp.com>]

gdb 6.0, file gdb/mips-tdep.c, function mips_find_saved_regs:

There are calls to "set_reg_offset" in a loop, followed by
some individual calls after the loop.  The latter have
no effect if the register was processed in the loop
(see the code for set_reg_offset).  SP_REGNUM is like this,
but PC_REGNUM is not, because of the particular values of
these constants (ugh).


  if ((get_frame_type (fci) == SIGTRAMP_FRAME))
    {
      for (ireg = 0; ireg < MIPS_NUMREGS; ireg++)
        {
          CORE_ADDR reg_position = (get_frame_base (fci) + SIGFRAME_REGSAVE_OFF
                                    + ireg * SIGFRAME_REG_SIZE);
          set_reg_offset (saved_regs, ireg, reg_position);
        }
      for (ireg = 0; ireg < MIPS_NUMREGS; ireg++)
        {
          CORE_ADDR reg_position = (get_frame_base (fci)
                                    + SIGFRAME_FPREGSAVE_OFF
                                    + ireg * SIGFRAME_REG_SIZE);
          set_reg_offset (saved_regs, FP0_REGNUM + ireg, reg_position);
        }

      set_reg_offset (saved_regs, PC_REGNUM, get_frame_base (fci) + SIGFRAME_PC_OFF);
      /* SP_REGNUM, contains the value and not the address.  */
      set_reg_offset (saved_regs, SP_REGNUM, get_frame_base (fci));
      return;
    }


The logical patch would be to move the special cases up front,
but I have no way to test this out (no access to a MIPS UNIX
platform).

On the other hand this isn't purely academic.  I came across this
because I need to define a custom sigtramp frame for our own
MIPS embedded platform.  It was baffling for a while that my
modification to the SP_REGNUM override had no effect.