[RFA] Locate sparc64 arguments correctly

[RFA] Locate sparc64 arguments correctly

[David S. Miller]

Two problems:

1) Debugging information encodes LOC_ARG/LOC_REF_ARG offsets
   with the Sparc64 stack bias included, we keep track of the
   frame pointer with the stack bias removed on sparc64.

   sparc64_frame_args_address takes care of that.

2) REG_STRUCT_HAS_ADDR was wrong, structs larger than 16 bytes
   are passed by reference.

No regression test changes, likely because these things aren't
hit by the testsuite.

2002-04-20  David S. Miller  <davem@redhat.com>

	* config/sparc/tm-sp64.h (REG_STRUCT_HAS_ADDR): Structs are passed
	by reference when they are greater than 16 bytes in size, not 32.
	* sparc-tdep.c (sparc_reg_struct_has_addr): Likewise.

	* config/sparc/tm-sp64.h (FRAME_ARGS_ADDRESS): Define.
	* sparc-tdep.c (sparc64_frame_args_address): New function.
	(sparc_gdbarch_init): Use it for sparc64, else use
	default_frame_address.

--- config/sparc/tm-sp64.h.~1~	Fri Apr 19 23:02:03 2002
+++ config/sparc/tm-sp64.h	Sat Apr 20 01:55:05 2002
@@ -223,6 +223,9 @@ CORE_ADDR sparc64_push_arguments (int,
 #undef  STACK_ALIGN
 #define STACK_ALIGN(ADDR) (((ADDR) + 15 ) & -16)
 
+#undef FRAME_ARGS_ADDRESS
+#define FRAME_ARGS_ADDRESS(FI)	((FI)->frame - 2047)
+
 /* Initializer for an array of names of registers.
    There should be NUM_REGS strings in this initializer.  */
 /* Some of these registers are only accessible from priviledged mode.
@@ -263,7 +266,7 @@ CORE_ADDR sparc64_push_arguments (int,
 }
 
 #undef REG_STRUCT_HAS_ADDR
-#define REG_STRUCT_HAS_ADDR(gcc_p,type) (TYPE_LENGTH (type) > 32)
+#define REG_STRUCT_HAS_ADDR(gcc_p,type) (TYPE_LENGTH (type) > 16)
 
 extern CORE_ADDR sparc64_read_sp ();
 extern CORE_ADDR sparc64_read_fp ();
--- sparc-tdep.c.~1~	Sat Apr 20 01:46:27 2002
+++ sparc-tdep.c	Sat Apr 20 01:52:38 2002
@@ -2798,6 +2798,16 @@ sparc64_register_byte (int regno)
     return 64 * 8 + (regno - 80) * 8;
 }
 
+/* Debugging information stores LOC_ARG/LOC_REF_ARG offsets with the
+   sparc64 stack bias present, this undoes that so that users of
+   FRAME_ARGS_ADDRESS use the right location.  */
+
+static CORE_ADDR
+sparc64_frame_args_address (struct frame_info *fi)
+{
+  return fi->frame - 2047;
+}
+
 /* Advance PC across any function entry prologue instructions to reach
    some "real" code.  SKIP_PROLOGUE_FRAMELESS_P advances the PC past
    some of the prologue, but stops as soon as it knows that the
@@ -2886,7 +2896,7 @@ int
 sparc_reg_struct_has_addr (int gcc_p, struct type *type)
 {
   if (GDB_TARGET_IS_SPARC64)
-    return (TYPE_LENGTH (type) > 32);
+    return (TYPE_LENGTH (type) > 16);
   else
     return (gcc_p != 1);
 }
@@ -3128,6 +3139,7 @@ sparc_gdbarch_init (struct gdbarch_info 
       set_gdbarch_use_struct_convention (gdbarch, 
 					 sparc64_use_struct_convention);
       set_gdbarch_write_sp (gdbarch, sparc64_write_sp);
+      set_gdbarch_frame_args_address (gdbarch, sparc64_frame_args_address);
       tdep->y_regnum = SPARC64_Y_REGNUM;
       tdep->fp_max_regnum = SPARC_FP0_REGNUM + 48;
       tdep->intreg_size = 8;

Re: [RFA] Locate sparc64 arguments correctly

[Michael Snyder]

"David S. Miller" wrote:
> 
> Two problems:
> 
> 1) Debugging information encodes LOC_ARG/LOC_REF_ARG offsets
>    with the Sparc64 stack bias included, we keep track of the
>    frame pointer with the stack bias removed on sparc64.
> 
>    sparc64_frame_args_address takes care of that.

This is a good idea, but FRAME_ARGS_ADDRESS is multi-arched.
Can you use that implementation?


> 2) REG_STRUCT_HAS_ADDR was wrong, structs larger than 16 bytes
>    are passed by reference.

Have you verified that this is the case on Solaris?

> No regression test changes, likely because these things aren't
> hit by the testsuite.

How about extending testsuite/gdb.base/structs.[c exp]
so that they cover this case?

> 2002-04-20  David S. Miller  <davem@redhat.com>
> 
>         * config/sparc/tm-sp64.h (REG_STRUCT_HAS_ADDR): Structs are passed
>         by reference when they are greater than 16 bytes in size, not 32.
>         * sparc-tdep.c (sparc_reg_struct_has_addr): Likewise.
> 
>         * config/sparc/tm-sp64.h (FRAME_ARGS_ADDRESS): Define.
>         * sparc-tdep.c (sparc64_frame_args_address): New function.
>         (sparc_gdbarch_init): Use it for sparc64, else use
>         default_frame_address.
> 
> --- config/sparc/tm-sp64.h.~1~  Fri Apr 19 23:02:03 2002
> +++ config/sparc/tm-sp64.h      Sat Apr 20 01:55:05 2002
> @@ -223,6 +223,9 @@ CORE_ADDR sparc64_push_arguments (int,
>  #undef  STACK_ALIGN
>  #define STACK_ALIGN(ADDR) (((ADDR) + 15 ) & -16)
> 
> +#undef FRAME_ARGS_ADDRESS
> +#define FRAME_ARGS_ADDRESS(FI) ((FI)->frame - 2047)
> +
>  /* Initializer for an array of names of registers.
>     There should be NUM_REGS strings in this initializer.  */
>  /* Some of these registers are only accessible from priviledged mode.
> @@ -263,7 +266,7 @@ CORE_ADDR sparc64_push_arguments (int,
>  }
> 
>  #undef REG_STRUCT_HAS_ADDR
> -#define REG_STRUCT_HAS_ADDR(gcc_p,type) (TYPE_LENGTH (type) > 32)
> +#define REG_STRUCT_HAS_ADDR(gcc_p,type) (TYPE_LENGTH (type) > 16)
> 
>  extern CORE_ADDR sparc64_read_sp ();
>  extern CORE_ADDR sparc64_read_fp ();
> --- sparc-tdep.c.~1~    Sat Apr 20 01:46:27 2002
> +++ sparc-tdep.c        Sat Apr 20 01:52:38 2002
> @@ -2798,6 +2798,16 @@ sparc64_register_byte (int regno)
>      return 64 * 8 + (regno - 80) * 8;
>  }
> 
> +/* Debugging information stores LOC_ARG/LOC_REF_ARG offsets with the
> +   sparc64 stack bias present, this undoes that so that users of
> +   FRAME_ARGS_ADDRESS use the right location.  */
> +
> +static CORE_ADDR
> +sparc64_frame_args_address (struct frame_info *fi)
> +{
> +  return fi->frame - 2047;
> +}
> +
>  /* Advance PC across any function entry prologue instructions to reach
>     some "real" code.  SKIP_PROLOGUE_FRAMELESS_P advances the PC past
>     some of the prologue, but stops as soon as it knows that the
> @@ -2886,7 +2896,7 @@ int
>  sparc_reg_struct_has_addr (int gcc_p, struct type *type)
>  {
>    if (GDB_TARGET_IS_SPARC64)
> -    return (TYPE_LENGTH (type) > 32);
> +    return (TYPE_LENGTH (type) > 16);
>    else
>      return (gcc_p != 1);
>  }
> @@ -3128,6 +3139,7 @@ sparc_gdbarch_init (struct gdbarch_info
>        set_gdbarch_use_struct_convention (gdbarch,
>                                          sparc64_use_struct_convention);
>        set_gdbarch_write_sp (gdbarch, sparc64_write_sp);
> +      set_gdbarch_frame_args_address (gdbarch, sparc64_frame_args_address);
>        tdep->y_regnum = SPARC64_Y_REGNUM;
>        tdep->fp_max_regnum = SPARC_FP0_REGNUM + 48;
>        tdep->intreg_size = 8;

Re: [RFA] Locate sparc64 arguments correctly

[Kevin Buettner]

On Apr 20,  2:09am, David S. Miller wrote:

> 1) Debugging information encodes LOC_ARG/LOC_REF_ARG offsets
>    with the Sparc64 stack bias included, we keep track of the
>    frame pointer with the stack bias removed on sparc64.
> 
>    sparc64_frame_args_address takes care of that.

I agree that this is the problem, but, when you do this...

> +/* Debugging information stores LOC_ARG/LOC_REF_ARG offsets with the
> +   sparc64 stack bias present, this undoes that so that users of
> +   FRAME_ARGS_ADDRESS use the right location.  */
> +
> +static CORE_ADDR
> +sparc64_frame_args_address (struct frame_info *fi)
> +{
> +  return fi->frame - 2047;
> +}

...does the output from ``info frame'' still look reasonable?

Re: [RFA] Locate sparc64 arguments correctly

[David S. Miller]

   From: Michael Snyder <msnyder@redhat.com>
   Date: Tue, 23 Apr 2002 12:41:20 -0700

   "David S. Miller" wrote:
   > Two problems:
   > 
   > 1) Debugging information encodes LOC_ARG/LOC_REF_ARG offsets
   >    with the Sparc64 stack bias included, we keep track of the
   >    frame pointer with the stack bias removed on sparc64.
   > 
   >    sparc64_frame_args_address takes care of that.
   
   This is a good idea, but FRAME_ARGS_ADDRESS is multi-arched.
   Can you use that implementation?
   
I do this in my patches, as per:
   
   > --- sparc-tdep.c.~1~    Sat Apr 20 01:46:27 2002
   > +++ sparc-tdep.c        Sat Apr 20 01:52:38 2002
   > @@ -2798,6 +2798,16 @@ sparc64_register_byte (int regno)
   >      return 64 * 8 + (regno - 80) * 8;
   >  }
   > 
   > +/* Debugging information stores LOC_ARG/LOC_REF_ARG offsets with the
   > +   sparc64 stack bias present, this undoes that so that users of
   > +   FRAME_ARGS_ADDRESS use the right location.  */
   > +
   > +static CORE_ADDR
   > +sparc64_frame_args_address (struct frame_info *fi)
   > +{
   > +  return fi->frame - 2047;
   > +}
   > +

I have to do the tm-sp64.h non-multi-arch'd version too in order
to prevent potential breakage until Sparc is %100 multi-arch'd.

   > 2) REG_STRUCT_HAS_ADDR was wrong, structs larger than 16 bytes
   >    are passed by reference.
   
   Have you verified that this is the case on Solaris?
   
This is what the GCC backend uses, and is what the Sysv4 Sparc 64-bit
ABI specifies.

   > No regression test changes, likely because these things aren't
   > hit by the testsuite.
   
   How about extending testsuite/gdb.base/structs.[c exp]
   so that they cover this case?
   
The reason the regressions didn't change turned out to not be because
the testsuite didn't test it, it was because the current sparc64 arg
passing code would fix up these case by itself.

Franks a lot,
David S. Miller
davem@redhat.com

Re: [RFA] Locate sparc64 arguments correctly

[David S. Miller]

   From: Kevin Buettner <kevinb@redhat.com>
   Date: Tue, 23 Apr 2002 16:09:40 -0700

   I agree that this is the problem, but, when you do this...
   
   > +/* Debugging information stores LOC_ARG/LOC_REF_ARG offsets with the
   > +   sparc64 stack bias present, this undoes that so that users of
   > +   FRAME_ARGS_ADDRESS use the right location.  */
   > +
   > +static CORE_ADDR
   > +sparc64_frame_args_address (struct frame_info *fi)
   > +{
   > +  return fi->frame - 2047;
   > +}
   
   ...does the output from ``info frame'' still look reasonable?
   
You're right, it does the wrong thing for info frame.

This indicates that FRAME_ARGS_ADDRESS is being used for two entirely
different purposes.  One is to interpret debugging information, the
other is to print arguments on the stack via info frame.

This particular case would be cured by adjusting what we use for
FRAME_ARGS_SKIP on a 64-bit sparc target such that the STACK bias is
calcelled back out.  I will make that modification to my patch and
resubmit.

Thanks for pointing this out.

Re: [RFA] Locate sparc64 arguments correctly

[Kevin Buettner]

On Apr 24,  8:23am, David S. Miller wrote:

>    From: Kevin Buettner <kevinb@redhat.com>
>    Date: Tue, 23 Apr 2002 16:09:40 -0700
> 
>    I agree that this is the problem, but, when you do this...
>    
>    > +/* Debugging information stores LOC_ARG/LOC_REF_ARG offsets with the
>    > +   sparc64 stack bias present, this undoes that so that users of
>    > +   FRAME_ARGS_ADDRESS use the right location.  */
>    > +
>    > +static CORE_ADDR
>    > +sparc64_frame_args_address (struct frame_info *fi)
>    > +{
>    > +  return fi->frame - 2047;
>    > +}
>    
>    ...does the output from ``info frame'' still look reasonable?
>    
> You're right, it does the wrong thing for info frame.
> 
> This indicates that FRAME_ARGS_ADDRESS is being used for two entirely
> different purposes.  One is to interpret debugging information, the
> other is to print arguments on the stack via info frame.
> 
> This particular case would be cured by adjusting what we use for
> FRAME_ARGS_SKIP on a 64-bit sparc target such that the STACK bias is
> calcelled back out.  I will make that modification to my patch and
> resubmit.

Hmm... I'm not sure this will work.  FWIW, when I looked at this problem
a few weeks back, I concluded that LOC_REF_ARG shouldn't really be used
when the debug info contains information about which basereg to use.

I came up with this patch instead.  My testing was using dwarf2 debug
info though.  I haven't taken a look yet to see if it'll work with
stabs.

	* symtab.h (LOC_BASEREG_REF_ARG): New constant for
	``enum address_class''.
	* dwarf2read.c (new_symbol): Use LOC_BASEREG_REF_ARG instead of
	LOC_REF_ARG.
	* ax-gdb.c (gen_var_ref): Add case for LOC_BASEREG_REF_ARG.
	* buildsym.c (LOC_BASEREG_REF_ARG): Likewise.
	* ch-exp.c (ch_lex): Likewise.
	* findvar.c (symbol_needs_frame, read_var_value): Likewise.
	* m2-exp.y (yylex): Likewise.
	* printcmd.c (address_info, print_frame_args): Likewise.
	* stack.c (print_frame_args): Likewise.
	* symmisc.c (print_symbol): Likewise.
	* symtab.c (lookup_block_symbol): Likewise.
	* tracepoint.c (collect_symbol, add_local_symbols, scope_info):
	Likewise.
	* gdbtk-stack.c (gdb_block_vars, gdb_get_blocks, gdb_get_vars_command):
	Likewise.
	* mi/mi-cmd-stack.c (list_args_or_locals): Likewise.

Index: ax-gdb.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/ax-gdb.c,v
retrieving revision 1.26
diff -u -p -r1.26 ax-gdb.c
--- ax-gdb.c	2002/02/06 17:15:19	1.26
+++ ax-gdb.c	2002/04/24 16:22:33
@@ -573,6 +573,15 @@ gen_var_ref (struct agent_expr *ax, stru
       value->kind = axs_lvalue_memory;
       break;
 
+    case LOC_BASEREG_REF_ARG:	/* relative to some base register,
+                                   dereferenced */
+      ax_reg (ax, SYMBOL_BASEREG (var));
+      gen_sym_offset (ax, var);
+      /* Don't assume any particular pointer size.  */
+      gen_fetch (ax, lookup_pointer_type (builtin_type_void));
+      value->kind = axs_lvalue_memory;
+      break;
+
     case LOC_TYPEDEF:
       error ("Cannot compute value of typedef `%s'.",
 	     SYMBOL_SOURCE_NAME (var));
Index: buildsym.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/buildsym.c,v
retrieving revision 2.112
diff -u -p -r2.112 buildsym.c
--- buildsym.c	2002/01/22 17:41:09	2.112
+++ buildsym.c	2002/04/24 16:22:33
@@ -284,6 +284,7 @@ finish_block (struct symbol *symbol, str
 		case LOC_REGPARM:
 		case LOC_REGPARM_ADDR:
 		case LOC_BASEREG_ARG:
+		case LOC_BASEREG_REF_ARG:
 		case LOC_LOCAL_ARG:
 		  nparams++;
 		  break;
@@ -320,6 +321,7 @@ finish_block (struct symbol *symbol, str
 		    case LOC_REGPARM:
 		    case LOC_REGPARM_ADDR:
 		    case LOC_BASEREG_ARG:
+		    case LOC_BASEREG_REF_ARG:
 		    case LOC_LOCAL_ARG:
 		      TYPE_FIELD_TYPE (ftype, iparams) = SYMBOL_TYPE (sym);
 		      TYPE_FIELD_ARTIFICIAL (ftype, iparams) = 0;
Index: ch-exp.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/ch-exp.c,v
retrieving revision 2.28
diff -u -p -r2.28 ch-exp.c
--- ch-exp.c	2002/01/13 00:32:15	2.28
+++ ch-exp.c	2002/04/24 16:22:33
@@ -2156,6 +2156,7 @@ ch_lex (void)
 	    case LOC_LOCAL_ARG:
 	    case LOC_BASEREG:
 	    case LOC_BASEREG_ARG:
+	    case LOC_BASEREG_REF_ARG:
 	      if (innermost_block == NULL
 		  || contained_in (block_found, innermost_block))
 		{
Index: dwarf2read.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/dwarf2read.c,v
retrieving revision 2.79
diff -u -p -r2.79 dwarf2read.c
--- dwarf2read.c	2002/02/18 02:40:36	2.79
+++ dwarf2read.c	2002/04/24 16:22:35
@@ -4383,9 +4383,8 @@ new_symbol (struct die_info *die, struct
 		{
 		  if (isderef)
 		    {
-		      if (basereg != frame_base_reg)
-			complain (&dwarf2_complex_location_expr);
-		      SYMBOL_CLASS (sym) = LOC_REF_ARG;
+		      SYMBOL_CLASS (sym) = LOC_BASEREG_REF_ARG;
+		      SYMBOL_BASEREG (sym) = DWARF2_REG_TO_REGNUM (basereg);
 		    }
 		  else
 		    {
Index: findvar.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/findvar.c,v
retrieving revision 1.137.10.1
diff -u -p -r1.137.10.1 findvar.c
--- findvar.c	2002/02/27 03:27:18	1.137.10.1
+++ findvar.c	2002/04/24 16:22:35
@@ -378,6 +378,7 @@ symbol_read_needs_frame (struct symbol *
     case LOC_LOCAL_ARG:
     case LOC_BASEREG:
     case LOC_BASEREG_ARG:
+    case LOC_BASEREG_REF_ARG:
     case LOC_THREAD_LOCAL_STATIC:
       return 1;
 
@@ -526,6 +527,23 @@ addresses have not been bound by the dyn
 	  error ("Value of base register not available.");
 	addr = value_as_address (regval);
 	addr += SYMBOL_VALUE (var);
+	break;
+      }
+
+    case LOC_BASEREG_REF_ARG:
+      {
+	struct value *regval;
+	CORE_ADDR argref;
+	struct value *ref;
+
+	regval = value_from_register (lookup_pointer_type (type),
+				      SYMBOL_BASEREG (var), frame);
+	if (regval == NULL)
+	  error ("Value of base register not available.");
+	argref = value_as_address (regval);
+	argref += SYMBOL_VALUE (var);
+	ref = value_at (lookup_pointer_type (type), argref, NULL);
+	addr = value_as_address (ref);
 	break;
       }
 
Index: m2-exp.y
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/m2-exp.y,v
retrieving revision 2.60
diff -u -p -r2.60 m2-exp.y
--- m2-exp.y	2001/03/26 19:54:18	2.60
+++ m2-exp.y	2002/04/24 16:22:35
@@ -1029,6 +1029,7 @@ yylex ()
        case LOC_LOCAL_ARG:
        case LOC_BASEREG:
        case LOC_BASEREG_ARG:
+       case LOC_BASEREG_REF_ARG:
        case LOC_CONST:
        case LOC_CONST_BYTES:
        case LOC_OPTIMIZED_OUT:
Index: printcmd.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/printcmd.c,v
retrieving revision 1.191
diff -u -p -r1.191 printcmd.c
--- printcmd.c	2002/02/06 17:15:20	1.191
+++ printcmd.c	2002/04/24 16:22:35
@@ -1246,6 +1246,11 @@ address_info (char *exp, int from_tty)
 		       val, REGISTER_NAME (basereg));
       break;
 
+    case LOC_BASEREG_REF_ARG:
+      printf_filtered ("a reference argument at offset %ld from register %s",
+		       val, REGISTER_NAME (basereg));
+      break;
+
     case LOC_TYPEDEF:
       printf_filtered ("a typedef");
       break;
@@ -1840,6 +1845,7 @@ print_frame_args (struct symbol *func, s
 	    case LOC_REGPARM_ADDR:
 	    case LOC_LOCAL_ARG:
 	    case LOC_BASEREG_ARG:
+	    case LOC_BASEREG_REF_ARG:
 	      break;
 
 	    /* Other types of symbols we just skip over.  */
Index: stack.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/stack.c,v
retrieving revision 1.149
diff -u -p -r1.149 stack.c
--- stack.c	2002/02/18 02:40:36	1.149
+++ stack.c	2002/04/24 16:22:37
@@ -1399,6 +1399,7 @@ print_frame_arg_vars (register struct fr
 	case LOC_REGPARM:
 	case LOC_REGPARM_ADDR:
 	case LOC_BASEREG_ARG:
+	case LOC_BASEREG_REF_ARG:
 	  values_printed = 1;
 	  fputs_filtered (SYMBOL_SOURCE_NAME (sym), stream);
 	  fputs_filtered (" = ", stream);
Index: symmisc.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/symmisc.c,v
retrieving revision 1.92
diff -u -p -r1.92 symmisc.c
--- symmisc.c	2001/12/03 20:48:52	1.92
+++ symmisc.c	2002/04/24 16:22:37
@@ -678,6 +678,11 @@ print_symbol (PTR args)
 			    SYMBOL_VALUE (symbol), SYMBOL_BASEREG (symbol));
 	  break;
 
+	case LOC_BASEREG_REF_ARG:
+	  fprintf_filtered (outfile, "arg at 0x%lx from register %d",
+			    SYMBOL_VALUE (symbol), SYMBOL_BASEREG (symbol));
+	  break;
+
 	case LOC_TYPEDEF:
 	  break;
 
Index: symtab.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/symtab.c,v
retrieving revision 1.265
diff -u -p -r1.265 symtab.c
--- symtab.c	2002/02/26 22:18:34	1.265
+++ symtab.c	2002/04/24 16:22:38
@@ -1329,7 +1329,8 @@ lookup_block_symbol (register const stru
 		  SYMBOL_CLASS (sym) != LOC_REF_ARG &&
 		  SYMBOL_CLASS (sym) != LOC_REGPARM &&
 		  SYMBOL_CLASS (sym) != LOC_REGPARM_ADDR &&
-		  SYMBOL_CLASS (sym) != LOC_BASEREG_ARG)
+		  SYMBOL_CLASS (sym) != LOC_BASEREG_ARG &&
+		  SYMBOL_CLASS (sym) != LOC_BASEREG_REF_ARG)
 		{
 		  break;
 		}
Index: symtab.h
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/symtab.h,v
retrieving revision 1.167
diff -u -p -r1.167 symtab.h
--- symtab.h	2002/02/22 21:59:06	1.167
+++ symtab.h	2002/04/24 16:22:38
@@ -632,6 +632,14 @@ enum address_class
 
     LOC_BASEREG_ARG,
 
+    /* Value address is at SYMBOL_VALUE plus the value of the register
+       indicated by SYMBOL_BASEREG.  Note that this is like
+       LOC_BASEREG_ARG, except that an extra dereference is performed. 
+       It's also similar to LOC_REF_ARG, but the basereg value has
+       been made explicit.  */
+
+    LOC_BASEREG_REF_ARG,
+
     /* Value is at fixed address, but the address of the variable has
        to be determined from the minimal symbol table whenever the
        variable is referenced.
@@ -717,7 +725,7 @@ struct symbol
 
     union
       {
-	/* Used by LOC_BASEREG and LOC_BASEREG_ARG.  */
+	/* Used by LOC_BASEREG, LOC_BASEREG_ARG, and LOC_BASEREG_REF_ARG.  */
 	short basereg;
       }
     aux_value;
Index: tracepoint.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/tracepoint.c,v
retrieving revision 2.79
diff -u -p -r2.79 tracepoint.c
--- tracepoint.c	2002/02/26 22:18:34	2.79
+++ tracepoint.c	2002/04/24 16:22:39
@@ -1281,6 +1281,11 @@ collect_symbol (struct collection_list *
 	}
       add_memrange (collect, reg, offset, len);
       break;
+    case LOC_BASEREG_REF_ARG:
+      printf_filtered ("Sorry, don't know how to do LOC_BASEREG_REF_ARG yet.\n");
+      printf_filtered ("       (will not collect %s)\n",
+		       SYMBOL_NAME (sym));
+      break;
     case LOC_UNRESOLVED:
       printf_filtered ("Don't know LOC_UNRESOLVED %s\n", SYMBOL_NAME (sym));
       break;
@@ -1327,6 +1332,7 @@ add_local_symbols (struct collection_lis
 	    case LOC_REGPARM:
 	    case LOC_REGPARM_ADDR:
 	    case LOC_BASEREG_ARG:
+	    case LOC_BASEREG_REF_ARG:
 	      if (type == 'A')	/* collecting Arguments */
 		{
 		  count++;
@@ -2433,6 +2439,11 @@ scope_info (char *args, int from_tty)
 	      break;
 	    case LOC_BASEREG_ARG:
 	      printf_filtered ("an argument at offset %ld from register $%s",
+			       SYMBOL_VALUE (sym),
+			       REGISTER_NAME (SYMBOL_BASEREG (sym)));
+	      break;
+	    case LOC_BASEREG_REF_ARG:
+	      printf_filtered ("a reference argument at offset %ld from register $%s",
 			       SYMBOL_VALUE (sym),
 			       REGISTER_NAME (SYMBOL_BASEREG (sym)));
 	      break;
Index: gdbtk/generic/gdbtk-stack.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/gdbtk/generic/gdbtk-stack.c,v
retrieving revision 1.5
diff -u -p -r1.5 gdbtk-stack.c
--- gdbtk-stack.c	2002/02/08 22:28:04	1.5
+++ gdbtk-stack.c	2002/04/24 16:22:39
@@ -121,6 +121,7 @@ gdb_block_vars (ClientData clientData, T
 		case LOC_REGPARM_ADDR:    /* indirect register arg */
 		case LOC_LOCAL_ARG:	  /* stack arg             */
 		case LOC_BASEREG_ARG:	  /* basereg arg           */
+		case LOC_BASEREG_REF_ARG: /* basereg ref arg       */
 		case LOC_LOCAL:	          /* stack local           */
 		case LOC_BASEREG:	  /* basereg local         */
 		case LOC_STATIC:	  /* static                */
@@ -196,6 +197,7 @@ gdb_get_blocks (ClientData clientData, T
 		case LOC_REGPARM_ADDR:    /* indirect register arg */
 		case LOC_LOCAL_ARG:	  /* stack arg             */
 		case LOC_BASEREG_ARG:	  /* basereg arg           */
+		case LOC_BASEREG_REF_ARG: /* basereg ref arg       */
 
 		case LOC_LOCAL:	          /* stack local           */
 		case LOC_BASEREG:	  /* basereg local         */
@@ -341,6 +343,7 @@ gdb_get_vars_command (ClientData clientD
 	    case LOC_REGPARM_ADDR:	/* indirect register arg */
 	    case LOC_LOCAL_ARG:	/* stack arg             */
 	    case LOC_BASEREG_ARG:	/* basereg arg           */
+	    case LOC_BASEREG_REF_ARG:	/* basereg ref arg       */
 	      if (arguments)
 		Tcl_ListObjAppendElement (interp, result_ptr->obj_ptr,
 					  Tcl_NewStringObj (SYMBOL_NAME (sym), -1));
Index: mi/mi-cmd-stack.c
===================================================================
RCS file: /cvs/cvsfiles/devo/gdb/mi/mi-cmd-stack.c,v
retrieving revision 1.13
diff -u -p -r1.13 mi-cmd-stack.c
--- mi-cmd-stack.c	2002/02/06 17:15:23	1.13
+++ mi-cmd-stack.c	2002/04/24 16:22:39
@@ -248,6 +248,7 @@ list_args_or_locals (int locals, int val
 	    case LOC_REGPARM_ADDR:	/* indirect register arg */
 	    case LOC_LOCAL_ARG:	/* stack arg             */
 	    case LOC_BASEREG_ARG:	/* basereg arg           */
+	    case LOC_BASEREG_REF_ARG:	/* basereg ref arg       */
 	      if (!locals)
 		print_me = 1;
 	      break;

Re: [RFA] Locate sparc64 arguments correctly

[David S. Miller]

   From: Kevin Buettner <kevinb@redhat.com>
   Date: Wed, 24 Apr 2002 09:42:33 -0700
   
   Hmm... I'm not sure this will work.  FWIW, when I looked at this problem
   a few weeks back, I concluded that LOC_REF_ARG shouldn't really be used
   when the debug info contains information about which basereg to use.
   
   I came up with this patch instead.  My testing was using dwarf2 debug
   info though.  I haven't taken a look yet to see if it'll work with
   stabs.

I think my fix would work, however I like your patch a whole lot
better.

My testing would be with dwarf2 information as well, so I wouldn't be
able to provide better regression coverage in testing out your patch.