Re: [avr] gas support for cfi info

[Anitha Boyapati <anitha.boyapati@gmail.com>]

2011/2/18 Richard Henderson <rth@redhat.com>
>
> On 02/17/2011 08:04 AM, Richard Henderson wrote:
> >> The other scenario is - how about functions with signals/interrupts?
> >> The compiler will give an ICE compiling a function as below:
> >>
> >> void my_interrupt_handler() __attribute__ (("interrupt"));
> >
> > It's a bug in the avr backend; the enable_interrupt insn should not be
> > marked as frame-related.  We should fix this separately.
>
> Looking at this closer, it seem that the FRAME_RELATED markers in avr.c
> were sprinkled at random without really knowing what is going on.

This is quite a clean up! I am yet to reach the changes in expand_epilogue.

>  There
> were 2 places where UNSPEC_VOLATILE insns were marked frame related,
> somehow expecting the unwinder to do something magical.
>

Just to be on same page, these are the 2 places:

1. gen_enable_interrupt()
2. gen_call_prologue_saves()


For the latter, can you  explain why adding reg notes is required?

+      add_reg_note (insn, REG_CFA_ADJUST_CFA,
+		    gen_rtx_SET (VOIDmode,
+				 (frame_pointer_needed
+				  ? frame_pointer_rtx : stack_pointer_rtx),
+				 plus_constant (stack_pointer_rtx,
+						-(size + live_seq))));
+

(The comment does say that this is to describe the effect of
UNSPEC_VOLATILE, but how reg notes help?)


> The following cleans all that up.  There are a couple of odd points:
>
>  (1) SREG and RAMPZ cannot be annotated as saved in the frame of an
>      interrupt function without allocating hard register numbers for
>      these registers.  Not to be confused with "real" registers,
>      these need a number in the same way that SP_L and SP_H have
>      register numbers.
>
>      I've simply ignored unwind info for these for now.
>

In addition to the above, there are atleast 4 more registers RAMPX,
RAMPY, RAMPD and EIND. (Actually the support for these registers is in
the form of patches which aren't upstream yet)


>  (2) At present it's possible to use epilogue_restores without
>      having used prologue_saves.  I.e. use epilogue_restores with
>      an inline prologue.  The problem being that the inline prologue
>      uses an HImode push of REG_Y (i.e. r29 first), whereas the code
>      in prologue_saves pushes r28 first and epilogue_restores is
>      written to expect that.
>

Is this is the line being referred to ?

if (frame_pointer_needed)
  {
...
   /* Push frame pointer.  */
insn = emit_move_insn (pushword, frame_pointer_rtx);


On the whole, when the patch is applied, an ICE is longer generated
during the compilation of interrupt and signal functions.  I am yet to
see the changes due to ARG_POINTER_CFA_OFFSET(FNDECL) definition and
the complete unwind info incase of interrupts.

Anitha