Fixing stack backtraces on 26-bit ARM

Fixing stack backtraces on 26-bit ARM

[Ben Harris]

At present, stack backtraces don't work on 26-bit ARM systems because R15 
pulled out of stack frames doesn't get its PSR bits stripped out, so PC 
ends up appearing to point somewhere insane.  This is a simple patch (from 
NetBSD CVS) to apply ADDR_BITS_REMOVE to values pulled from stack frames 
to generate PC.  This patch may have been approved once before, in 2001, 
but it was never applied.

2006-10-19  Ben Harris  <bjh21@NetBSD.org>

	* arm-tdep.c (arm_unwind_pc): Use ADDR_BITS_REMOVE.

--- gdb/arm-tdep.c      2 Jul 2006 21:04:40 -0000       1.3
+++ gdb/arm-tdep.c      14 Oct 2006 10:42:53 -0000      1.4
@@ -1056,7 +1056,7 @@ arm_unwind_pc (struct gdbarch *gdbarch, 
 {
   CORE_ADDR pc;
   pc = frame_unwind_register_unsigned (this_frame, ARM_PC_REGNUM);
-  return IS_THUMB_ADDR (pc) ? UNMAKE_THUMB_ADDR (pc) : pc;
+  return ADDR_BITS_REMOVE (pc);
 }
 
 static CORE_ADDR

-- 
Ben Harris                                                   <bjh21@NetBSD.org>
Portmaster, NetBSD/acorn26               <http://www.NetBSD.org/Ports/acorn26/>

Re: Fixing stack backtraces on 26-bit ARM

[Richard Earnshaw]

On Thu, 2006-10-19 at 20:46 +0100, Ben Harris wrote:
> At present, stack backtraces don't work on 26-bit ARM systems because R15 
> pulled out of stack frames doesn't get its PSR bits stripped out, so PC 
> ends up appearing to point somewhere insane.  This is a simple patch (from 
> NetBSD CVS) to apply ADDR_BITS_REMOVE to values pulled from stack frames 
> to generate PC.  This patch may have been approved once before, in 2001, 
> but it was never applied.
> 
> 2006-10-19  Ben Harris  <bjh21@NetBSD.org>
> 
> 	* arm-tdep.c (arm_unwind_pc): Use ADDR_BITS_REMOVE.
> 
> --- gdb/arm-tdep.c      2 Jul 2006 21:04:40 -0000       1.3
> +++ gdb/arm-tdep.c      14 Oct 2006 10:42:53 -0000      1.4
> @@ -1056,7 +1056,7 @@ arm_unwind_pc (struct gdbarch *gdbarch, 
>  {
>    CORE_ADDR pc;
>    pc = frame_unwind_register_unsigned (this_frame, ARM_PC_REGNUM);
> -  return IS_THUMB_ADDR (pc) ? UNMAKE_THUMB_ADDR (pc) : pc;
> +  return ADDR_BITS_REMOVE (pc);
>  }
>  
>  static CORE_ADDR

I think it would be better to just call arm_addr_bits_remove() directly
(more efficient within the back-end); but otherwise this is OK.

R.