[patch/mips] Use unwind in mips_get_saved_register

[patch/mips] Use unwind in mips_get_saved_register

[Andrew Cagney]

[-- Attachment #1: Type: text/plain, Size: 553 bytes --]

Hello,

The attached is ``two steps forward, one step back'' patch.

It modifies mips_get_saved_register() so that it uses the new unwind 
code.  The new unwind code automatically handles the fetching of 
registers in dummy-frames.  Unfortunatly, the change doesn't fix the 
problem of the MIPS fiddling memory register reads dependant on the 
current ABI.

I think, over all it is a step forward though.  It at least helps to 
make it possible to convert the MIPS to generic dummy frames.

-o32 -64 and -n32 all show no regressions.

committed,
Andrew

[-- Attachment #2: diffs --]
[-- Type: text/plain, Size: 2903 bytes --]

Fri Sep 13 07:42:09 2002  Andrew Cagney  <cagney@redhat.com>

	* mips-tdep.c (mips_get_saved_register): Re-hack using
 	frame_register_unwind.

Index: mips-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/mips-tdep.c,v
retrieving revision 1.121
diff -u -r1.121 mips-tdep.c
--- mips-tdep.c	10 Sep 2002 21:31:18 -0000	1.121
+++ mips-tdep.c	13 Sep 2002 17:21:13 -0000
@@ -5426,59 +5426,47 @@
 
 static void
 mips_get_saved_register (char *raw_buffer,
-			 int *optimized,
+			 int *optimizedp,
 			 CORE_ADDR *addrp,
 			 struct frame_info *frame,
 			 int regnum,
-			 enum lval_type *lval)
+			 enum lval_type *lvalp)
 {
-  CORE_ADDR addr;
+  CORE_ADDR addrx;
+  enum lval_type lvalx;
+  int optimizedx;
+  int realnum;
 
   if (!target_has_registers)
     error ("No registers.");
 
-  /* Normal systems don't optimize out things with register numbers.  */
-  if (optimized != NULL)
-    *optimized = 0;
-  addr = find_saved_register (frame, regnum);
-  if (addr != 0)
+  /* Make certain that all needed parameters are present.  */
+  if (addrp == NULL)
+    addrp = &addrx;
+  if (lvalp == NULL)
+    lvalp = &lvalx;
+  if (optimizedp == NULL)
+    optimizedp = &optimizedx;
+  frame_register_unwind (get_next_frame (frame), regnum, optimizedp, lvalp,
+			 addrp, &realnum, raw_buffer);
+  /* FIXME: cagney/2002-09-13: This is just so bad.  The MIPS should
+     have a pseudo register range that correspons to the ABI's, rather
+     than the ISA's, view of registers.  These registers would then
+     implicitly describe their size and hence could be used without
+     the below munging.  */
+  if ((*lvalp) == lval_memory)
     {
-      if (lval != NULL)
-	*lval = lval_memory;
-      if (regnum == SP_REGNUM)
-	{
-	  if (raw_buffer != NULL)
-	    {
-	      /* Put it back in target format.  */
-	      store_address (raw_buffer, REGISTER_RAW_SIZE (regnum),
-			     (LONGEST) addr);
-	    }
-	  if (addrp != NULL)
-	    *addrp = 0;
-	  return;
-	}
       if (raw_buffer != NULL)
 	{
-	  LONGEST val;
 	  if (regnum < 32)
-	    /* Only MIPS_SAVED_REGSIZE bytes of GP registers are
-               saved. */
-	    val = read_memory_integer (addr, MIPS_SAVED_REGSIZE);
-	  else
-	    val = read_memory_integer (addr, REGISTER_RAW_SIZE (regnum));
-	  store_address (raw_buffer, REGISTER_RAW_SIZE (regnum), val);
+	    {
+	      /* Only MIPS_SAVED_REGSIZE bytes of GP registers are
+		 saved. */
+	      LONGEST val = read_memory_integer ((*addrp), MIPS_SAVED_REGSIZE);
+	      store_address (raw_buffer, REGISTER_RAW_SIZE (regnum), val);
+	    }
 	}
     }
-  else
-    {
-      if (lval != NULL)
-	*lval = lval_register;
-      addr = REGISTER_BYTE (regnum);
-      if (raw_buffer != NULL)
-	read_register_gen (regnum, raw_buffer);
-    }
-  if (addrp != NULL)
-    *addrp = addr;
 }
 
 /* Immediately after a function call, return the saved pc.

Re: [patch/mips] Use unwind in mips_get_saved_register

[Kevin Buettner]

On Sep 13,  1:55pm, Andrew Cagney wrote:

> The attached is ``two steps forward, one step back'' patch.
> 
> It modifies mips_get_saved_register() so that it uses the new unwind 
> code.  The new unwind code automatically handles the fetching of 
> registers in dummy-frames.  Unfortunatly, the change doesn't fix the 
> problem of the MIPS fiddling memory register reads dependant on the 
> current ABI.

I posted a similar patch last month:

    http://sources.redhat.com/ml/gdb-patches/2002-08/msg00198.html

At that time you rejected it for the reasons given in:

    http://sources.redhat.com/ml/gdb-patches/2002-08/msg00214.html
    http://sources.redhat.com/ml/gdb-patches/2002-08/msg00235.html

Please explain what has changed between then and now which makes it
acceptable to now use frame_register_unwind()?

Puzzled,

Kevin

Re: [patch/mips] Use unwind in mips_get_saved_register

[Andrew Cagney]

> On Sep 13,  1:55pm, Andrew Cagney wrote:
> 
> 
>> The attached is ``two steps forward, one step back'' patch.
>> 
>> It modifies mips_get_saved_register() so that it uses the new unwind 
>> code.  The new unwind code automatically handles the fetching of 
>> registers in dummy-frames.  Unfortunatly, the change doesn't fix the 
>> problem of the MIPS fiddling memory register reads dependant on the 
>> current ABI.
> 
> 
> I posted a similar patch last month:
> 
>     http://sources.redhat.com/ml/gdb-patches/2002-08/msg00198.html
> 
> At that time you rejected it for the reasons given in:
> 
>     http://sources.redhat.com/ml/gdb-patches/2002-08/msg00214.html
>     http://sources.redhat.com/ml/gdb-patches/2002-08/msg00235.html
> 
> Please explain what has changed between then and now which makes it
> acceptable to now use frame_register_unwind()?

I know.  See:
http://sources.redhat.com/ml/gdb-patches/2002-09/msg00223.html

At the time of your original patch, the only motivation was the 
elimination of frame_register_unwind() --- on its own, I don't see any 
benefit.  Since then, as part of converting the MIPS to generic dummy 
frames (a needed and huge forward step), I've found that I need to use 
the generic unwind code to at least keep things working.  You'll note 
that the code includes some very blunt comments that make it clear where 
any bugs are.

enjoy,
Andrew