Patch: Fix ftag error when it is restored from fxsave in gdbserver on x86

Patch: Fix ftag error when it is restored from fxsave in gdbserver on x86

[Qunying Pan]

Hello,

This patch fixes ftag error when it is restored from fxsave in gdbserver
on x86. Acceptable? 

Thanks,

Qunying


--------------------

The testcase is:

	typedef long double doublest;

	doublest
	add_doublest (doublest u, doublest v)
	{
	  return u + v;
	}

	int
	main ()
	{
	  doublest z = 0.0;
	  doublest x=1.0,y=2.0;
	  z = add_doublest(x, y);
	  printf("%Lf, %Lf %Lf\n", x, y, z);
	  return 0;
	}


When debug with gdbserver, and set breakpoint at add_doublest, the
output z
will be set to nan instead of the expected output 3.0.



The patch is:

2005-10-31 Qunying Pan <qpan@mvista.com>
   * gdbserver/i387-fp.c (i387_fxsave_to_cache): Fixing ftag error when 
     it is restored from fxsave

diff -urNp src/gdb/gdbserver.orig/i387-fp.c src/gdb/gdbserver/i387-fp.c
--- src/gdb/gdbserver.orig/i387-fp.c    2005-06-12 18:59:22.000000000
-0700
+++ src/gdb/gdbserver/i387-fp.c 2005-10-31 16:53:29.000000000 -0800
@@ -247,7 +247,7 @@ i387_fxsave_to_cache (const void *buf)
   int i, top;
   int st0_regnum = find_regno ("st0");
   int xmm0_regnum = find_regno ("xmm0");
-  unsigned long val;
+  unsigned long val, val2;

   for (i = 0; i < 8; i++)
     supply_register (i + st0_regnum, ((char *) &fp->st_space[0]) + i *
16);
@@ -267,7 +267,8 @@ i387_fxsave_to_cache (const void *buf)

   /* Generate the form of ftag data that GDB expects.  */
   top = (fp->fstat >> 11) & 0x7;
-  val = 0;
+  val = fp->ftag;
+  val2 = 0;
   for (i = 7; i >= 0; i--)
     {
       int tag;
@@ -275,9 +276,9 @@ i387_fxsave_to_cache (const void *buf)
        tag = i387_ftag (fp, (i + 8 - top) % 8);
       else
        tag = 3;
-      val |= tag << (2 * i);
+      val2 |= tag << (2 * i);
     }
-  supply_register_by_name ("ftag", &val);
+  supply_register_by_name ("ftag", &val2);

   val = fp->fiseg & 0xFFFF;
   supply_register_by_name ("fiseg", &val);

Patch: Fix ftag error when it is restored from fxsave in gdbserver on x86

[Qunying Pan]

Hello,

Sent out to gdb-patches@sources.redhat.com yesterday, but did not get
through I guess. Try again to gdb-patches@sourceware.org.

This patch fixes ftag error when it is restored from fxsave in gdbserver
on x86. Acceptable? 

Thanks,

Qunying


--------------------

The testcase is:

	typedef long double doublest;

	doublest
	add_doublest (doublest u, doublest v)
	{
	  return u + v;
	}

	int
	main ()
	{
	  doublest z = 0.0;
	  doublest x=1.0,y=2.0;
	  z = add_doublest(x, y);
	  printf("%Lf, %Lf %Lf\n", x, y, z);
	  return 0;
	}


When debug with gdbserver, and set breakpoint at add_doublest in
cross-gdb, the output z will be set to nan instead of the expected
output 3.0.



The patch is:

2005-10-31 Qunying Pan <qpan@mvista.com>
   * gdbserver/i387-fp.c (i387_fxsave_to_cache): Fixing ftag error when 
     it is restored from fxsave

diff -urNp src/gdb/gdbserver.orig/i387-fp.c src/gdb/gdbserver/i387-fp.c
--- src/gdb/gdbserver.orig/i387-fp.c    2005-06-12 18:59:22.000000000
-0700
+++ src/gdb/gdbserver/i387-fp.c 2005-10-31 16:53:29.000000000 -0800
@@ -247,7 +247,7 @@ i387_fxsave_to_cache (const void *buf)
   int i, top;
   int st0_regnum = find_regno ("st0");
   int xmm0_regnum = find_regno ("xmm0");
-  unsigned long val;
+  unsigned long val, val2;

   for (i = 0; i < 8; i++)
     supply_register (i + st0_regnum, ((char *) &fp->st_space[0]) + i *
16);
@@ -267,7 +267,8 @@ i387_fxsave_to_cache (const void *buf)

   /* Generate the form of ftag data that GDB expects.  */
   top = (fp->fstat >> 11) & 0x7;
-  val = 0;
+  val = fp->ftag;
+  val2 = 0;
   for (i = 7; i >= 0; i--)
     {
       int tag;
@@ -275,9 +276,9 @@ i387_fxsave_to_cache (const void *buf)
        tag = i387_ftag (fp, (i + 8 - top) % 8);
       else
        tag = 3;
-      val |= tag << (2 * i);
+      val2 |= tag << (2 * i);
     }
-  supply_register_by_name ("ftag", &val);
+  supply_register_by_name ("ftag", &val2);

   val = fp->fiseg & 0xFFFF;
   supply_register_by_name ("fiseg", &val);

Re: Patch: Fix ftag error when it is restored from fxsave in gdbserver on x86

[Daniel Jacobowitz]

On Tue, Nov 01, 2005 at 09:10:57AM -0800, Qunying Pan wrote:
> Hello,
> 
> Sent out to gdb-patches@sources.redhat.com yesterday, but did not get
> through I guess. Try again to gdb-patches@sourceware.org.
> 
> This patch fixes ftag error when it is restored from fxsave in gdbserver
> on x86. Acceptable? 

I've only belatedly had a chance to look at this :-(  I ended up using
a different patch, but thank you for your work - it was easy to see
what I'd done (completely) wrong when I wrote the fxsave support.
There was a similar error in the other direction.

Tested on x86_64-linux using gdbserver, where it fixes a failure in
store.exp.

-- 
Daniel Jacobowitz
CodeSourcery

2007-03-30  Daniel Jacobowitz  <dan@codesourcery.com>

	* i387-fp.c (i387_cache_to_fxsave): Reinitialize val2 before use.
	(i387_fxsave_to_cache): Check fp->ftag while building ftag value.

Index: i387-fp.c
===================================================================
RCS file: /cvs/src/src/gdb/gdbserver/i387-fp.c,v
retrieving revision 1.5
diff -u -p -r1.5 i387-fp.c
--- i387-fp.c	9 Jan 2007 17:59:08 -0000	1.5
+++ i387-fp.c	30 Mar 2007 15:28:45 -0000
@@ -174,6 +174,7 @@ i387_cache_to_fxsave (void *buf)
   /* Convert to the simplifed tag form stored in fxsave data.  */
   collect_register_by_name ("ftag", &val);
   val &= 0xFFFF;
+  val2 = 0;
   for (i = 7; i >= 0; i--)
     {
       int tag = (val >> (i * 2)) & 3;
@@ -270,7 +271,7 @@ i387_fxsave_to_cache (const void *buf)
   for (i = 7; i >= 0; i--)
     {
       int tag;
-      if (val & (1 << i))
+      if (fp->ftag & (1 << i))
 	tag = i387_ftag (fp, (i + 8 - top) % 8);
       else
 	tag = 3;