[RFC] Teach dwarf2-frame.c about StackGhost

[RFC] Teach dwarf2-frame.c about StackGhost

[Mark Kettenis]

One of the things that still keeps me from enabling the DWARF2
unwinder on SPARC is the fact that it doesn't play nice with
StackGhost.  Here is an attempt to make dwarf2-frame.c deal with it by
introducing yet another register rule.  People might object since this
is only sort of architecture independent code.  However it seems that
this is the cleanest way to solve this.

If there's no opposition, I'll check this in next weekend.

Mark


Index: ChangeLog
from  Mark Kettenis  <kettenis@gnu.org>

	* dwarf2-frame.h (enum dwarf2_frame_reg_rule): Add
	DWARF2_FRAME_REG_SAVED_WCOOKIE.
	* dwarf2-frame.c: Include "target.h".
	(execute_cfa_program): Handle StackGhost.
	(dwarf2_frame_prev_register): Likewise.
	* Makefile.in (dwarf2-frame.o): Update dependencies.

Index: dwarf2-frame.c
===================================================================
RCS file: /cvs/src/src/gdb/dwarf2-frame.c,v
retrieving revision 1.50
diff -u -p -r1.50 dwarf2-frame.c
--- dwarf2-frame.c 26 Mar 2005 15:20:42 -0000 1.50
+++ dwarf2-frame.c 2 Apr 2005 22:14:29 -0000
@@ -32,6 +32,7 @@
 #include "symtab.h"
 #include "objfiles.h"
 #include "regcache.h"
+#include "target.h"
 
 #include "gdb_assert.h"
 #include "gdb_string.h"
@@ -462,11 +463,14 @@ bad CFI data; mismatched DW_CFA_restore_
 		    fs->regs.reg[reg].how = DWARF2_FRAME_REG_SAVED_REG;
 		    fs->regs.reg[reg].loc.reg = reg + 16;
 		  }
-		for (reg = 16; reg < 32; reg++)
+		for (reg = 16; reg < 31; reg++)
 		  {
 		    fs->regs.reg[reg].how = DWARF2_FRAME_REG_SAVED_OFFSET;
 		    fs->regs.reg[reg].loc.offset = (reg - 16) * size;
 		  }
+		/* We may have to apply a StackGhost cookie.  */
+		fs->regs.reg[31].how = DWARF2_FRAME_REG_SAVED_WCOOKIE;
+		fs->regs.reg[31].loc.offset = 15 * size;
 	      }
 	      break;
 
@@ -836,6 +840,7 @@ dwarf2_frame_prev_register (struct frame
       break;
 
     case DWARF2_FRAME_REG_SAVED_OFFSET:
+    case DWARF2_FRAME_REG_SAVED_WCOOKIE:
       *optimizedp = 0;
       *lvalp = lval_memory;
       *addrp = cache->cfa + cache->reg[regnum].loc.offset;
@@ -915,6 +920,7 @@ dwarf2_frame_prev_register (struct frame
       if (valuep)
         {
           CORE_ADDR pc = cache->reg[regnum].loc.offset;
+	  int regnum;
 
           regnum = DWARF2_REG_TO_REGNUM (cache->retaddr_reg.loc.reg);
           pc += frame_unwind_register_unsigned (next_frame, regnum);
@@ -925,6 +931,32 @@ dwarf2_frame_prev_register (struct frame
     default:
       internal_error (__FILE__, __LINE__, _("Unknown register rule."));
     }
+
+  /* Handle StackGhost.  */
+  if (cache->reg[regnum].how == DWARF2_FRAME_REG_SAVED_WCOOKIE)
+    {
+      int size = register_size (gdbarch, regnum);
+      struct target_ops *ops = &current_target;
+      char *buf = alloca (size);
+      int len;
+
+      len = target_read_partial (ops, TARGET_OBJECT_WCOOKIE, NULL,
+				 buf, 0, size);
+      if (len != -1)
+	{
+	  ULONGEST wcookie = extract_unsigned_integer (buf, len);
+
+	  gdb_assert (len == size);
+
+	  *lvalp = not_lval;
+	  *addrp = 0;
+	  if (valuep)
+	    {
+	      ULONGEST value = extract_unsigned_integer (valuep, size);
+	      store_unsigned_integer (valuep, size, value ^ wcookie);
+	    }
+	}
+    }
 }
 
 static const struct frame_unwind dwarf2_frame_unwind =
Index: dwarf2-frame.h
===================================================================
RCS file: /cvs/src/src/gdb/dwarf2-frame.h,v
retrieving revision 1.8
diff -u -p -r1.8 dwarf2-frame.h
--- dwarf2-frame.h 25 Mar 2005 16:51:40 -0000 1.8
+++ dwarf2-frame.h 2 Apr 2005 22:14:29 -0000
@@ -55,7 +55,9 @@ enum dwarf2_frame_reg_rule
      used internally by GDB.  */
   DWARF2_FRAME_REG_RA,		/* Return Address.  */
   DWARF2_FRAME_REG_RA_OFFSET,	/* Return Address with offset.  */
-  DWARF2_FRAME_REG_CFA		/* Call Frame Address.  */
+  DWARF2_FRAME_REG_CFA,		/* Call Frame Address.  */
+  DWARF2_FRAME_REG_SAVED_WCOOKIE/* Like DWARF2_FRAME_SAVED_OFFSET, but
+				   apply StackGhost cookie.  */
 };
 
 /* Register state.  */
Index: Makefile.in
===================================================================
RCS file: /cvs/src/src/gdb/Makefile.in,v
retrieving revision 1.710
diff -u -p -r1.710 Makefile.in
--- Makefile.in 31 Mar 2005 19:58:24 -0000 1.710
+++ Makefile.in 2 Apr 2005 22:14:31 -0000
@@ -1887,7 +1887,7 @@ dwarf2expr.o: dwarf2expr.c $(defs_h) $(s
 	$(gdbcore_h) $(elf_dwarf2_h) $(dwarf2expr_h)
 dwarf2-frame.o: dwarf2-frame.c $(defs_h) $(dwarf2expr_h) $(elf_dwarf2_h) \
 	$(frame_h) $(frame_base_h) $(frame_unwind_h) $(gdbcore_h) \
-	$(gdbtypes_h) $(symtab_h) $(objfiles_h) $(regcache_h) \
+	$(gdbtypes_h) $(symtab_h) $(objfiles_h) $(regcache_h) $(target_h) \
 	$(gdb_assert_h) $(gdb_string_h) $(complaints_h) $(dwarf2_frame_h)
 dwarf2loc.o: dwarf2loc.c $(defs_h) $(ui_out_h) $(value_h) $(frame_h) \
 	$(gdbcore_h) $(target_h) $(inferior_h) $(ax_h) $(ax_gdb_h) \

Re: [RFC] Teach dwarf2-frame.c about StackGhost

[Daniel Jacobowitz]

On Sun, Apr 03, 2005 at 12:18:42AM +0200, Mark Kettenis wrote:
> One of the things that still keeps me from enabling the DWARF2
> unwinder on SPARC is the fact that it doesn't play nice with
> StackGhost.  Here is an attempt to make dwarf2-frame.c deal with it by
> introducing yet another register rule.  People might object since this
> is only sort of architecture independent code.  However it seems that
> this is the cleanest way to solve this.

Ewww... this isn't even vaguely architecture independent code.  It
seems like it's in the wrong place; there's no reason StackGhost
must be restricted to register windows.

Here's an alternative that I think is cleaner (and will work... I
hope): create a sparc-specific unwinder.  Its sniffer can call the
dwarf2 sniffer; similarly for this_id and prev_register.  Then you can
handle the StackGhost cookie when you are requested to unwind the PC.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: [RFC] Teach dwarf2-frame.c about StackGhost

[Mark Kettenis]

   Date: Sat, 2 Apr 2005 19:34:43 -0500
   From: Daniel Jacobowitz <drow@false.org>

   On Sun, Apr 03, 2005 at 12:18:42AM +0200, Mark Kettenis wrote:
   > One of the things that still keeps me from enabling the DWARF2
   > unwinder on SPARC is the fact that it doesn't play nice with
   > StackGhost.  Here is an attempt to make dwarf2-frame.c deal with it by
   > introducing yet another register rule.  People might object since this
   > is only sort of architecture independent code.  However it seems that
   > this is the cleanest way to solve this.

   Ewww... this isn't even vaguely architecture independent code.  It
   seems like it's in the wrong place; there's no reason StackGhost
   must be restricted to register windows.

On the contrary.  The whole idea pretty much depends on the concept of
register windows.  Now the number of register window architectures is
fairly limited: SPARC, UltraSPARC and IA-64.

   Here's an alternative that I think is cleaner (and will work... I
   hope): create a sparc-specific unwinder.  Its sniffer can call the
   dwarf2 sniffer; similarly for this_id and prev_register.  Then you can
   handle the StackGhost cookie when you are requested to unwind the PC.

The problem here is that whether you need to apply the stackghost
cookie is highly dependent on where exactly %i7 is stored.  If it is
stored in the reserved stack slot then you need to apply the cookie,
otherwise you won't.  The way you suggests makes it hard to get at
that information.

Mark

Re: [RFC] Teach dwarf2-frame.c about StackGhost

[Daniel Jacobowitz]

On Sun, Apr 03, 2005 at 09:03:31PM +0200, Mark Kettenis wrote:
>    Date: Sat, 2 Apr 2005 19:34:43 -0500
>    From: Daniel Jacobowitz <drow@false.org>
> 
>    On Sun, Apr 03, 2005 at 12:18:42AM +0200, Mark Kettenis wrote:
>    > One of the things that still keeps me from enabling the DWARF2
>    > unwinder on SPARC is the fact that it doesn't play nice with
>    > StackGhost.  Here is an attempt to make dwarf2-frame.c deal with it by
>    > introducing yet another register rule.  People might object since this
>    > is only sort of architecture independent code.  However it seems that
>    > this is the cleanest way to solve this.
> 
>    Ewww... this isn't even vaguely architecture independent code.  It
>    seems like it's in the wrong place; there's no reason StackGhost
>    must be restricted to register windows.
> 
> On the contrary.  The whole idea pretty much depends on the concept of
> register windows.  Now the number of register window architectures is
> fairly limited: SPARC, UltraSPARC and IA-64.

I have just gone and looked up the details of StackGhost.  Obviously, I
misunderstood it.  Similar mechanisms are possible elsewhere, but not
this one.

>    Here's an alternative that I think is cleaner (and will work... I
>    hope): create a sparc-specific unwinder.  Its sniffer can call the
>    dwarf2 sniffer; similarly for this_id and prev_register.  Then you can
>    handle the StackGhost cookie when you are requested to unwind the PC.
> 
> The problem here is that whether you need to apply the stackghost
> cookie is highly dependent on where exactly %i7 is stored.  If it is
> stored in the reserved stack slot then you need to apply the cookie,
> otherwise you won't.  The way you suggests makes it hard to get at
> that information.

I think it would still be fairly straightforward - but if you don't,
then I withdraw my objection.

-- 
Daniel Jacobowitz
CodeSourcery, LLC