[RFA] Save SSE registers to generated core files, without breaking other architectures

[RFA] Save SSE registers to generated core files, without breaking other architectures

[Daniel Jacobowitz]

How does this patch look?  The original problem was that we can not rely on
HAVE_PTRACE_FPXREGSET, because the presence of PTRACE_FPXREGSET doesn't
imply that GDB has a fill_fpxregset function available.  I updated it to
check for a separate flag.  It's not ideal, but until we separate regset
handling into an appropriate vector, it's the best we can do (I think).

OK?

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

2002-04-02  Daniel Jacobowitz  <drow@mvista.com>

	* config/i386/tm-linux.h: Define FILL_FPXREGSET.
	* gregset.h: If FILL_FPXREGSET is defined, provide
	gdb_fpxregset_t, supply_fpxregset, and fill_fpxregset.
	* linux-proc.c (linux_do_thread_registers): If FILL_FPXREGSET
	is defined, call fill_fpxregset.

Index: gregset.h
===================================================================
RCS file: /cvs/src/src/gdb/gregset.h,v
retrieving revision 1.5
diff -u -p -r1.5 gregset.h
--- gregset.h	2002/02/24 22:14:33	1.5
+++ gregset.h	2002/04/02 19:51:15
@@ -52,5 +52,18 @@ extern void supply_fpregset (gdb_fpregse
 extern void fill_gregset (gdb_gregset_t *gregs, int regno);
 extern void fill_fpregset (gdb_fpregset_t *fpregs, int regno);
 
+#ifdef FILL_FPXREGSET
+/* Linux/i386: Copy register values between GDB's internal register cache
+   and the i386 extended floating point registers.  */
+
+#ifndef GDB_FPXREGSET_T
+#define GDB_FPXREGSET_T elf_fpxregset_t
+#endif
+
+typedef GDB_FPXREGSET_T gdb_fpxregset_t;
+
+extern void supply_fpxregset (gdb_fpxregset_t *fpxregs);
+extern void fill_fpxregset (gdb_fpxregset_t *fpxregs, int regno);
+#endif
 
 #endif
Index: linux-proc.c
===================================================================
RCS file: /cvs/src/src/gdb/linux-proc.c,v
retrieving revision 1.8
diff -u -p -r1.8 linux-proc.c
--- linux-proc.c	2002/03/25 19:47:41	1.8
+++ linux-proc.c	2002/04/02 19:51:15
@@ -167,6 +167,9 @@ linux_do_thread_registers (bfd *obfd, pt
 {
   gdb_gregset_t gregs;
   gdb_fpregset_t fpregs;
+#ifdef FILL_FPXREGSET
+  gdb_fpxregset_t fpxregs;
+#endif
   unsigned long merged_pid = ptid_get_tid (ptid) << 16 | ptid_get_pid (ptid);
 
   fill_gregset (&gregs, -1);
@@ -183,6 +186,14 @@ linux_do_thread_registers (bfd *obfd, pt
 					      note_size, 
 					      &fpregs, 
 					      sizeof (fpregs));
+#ifdef FILL_FPXREGSET
+  fill_fpxregset (&fpxregs, -1);
+  note_data = (char *) elfcore_write_prxfpreg (obfd, 
+					       note_data, 
+					       note_size, 
+					       &fpxregs, 
+					       sizeof (fpxregs));
+#endif
   return note_data;
 }
 
Index: config/i386/tm-linux.h
===================================================================
RCS file: /cvs/src/src/gdb/config/i386/tm-linux.h,v
retrieving revision 1.17
diff -u -p -r1.17 tm-linux.h
--- tm-linux.h	2002/02/24 22:56:05	1.17
+++ tm-linux.h	2002/04/02 19:51:46
@@ -26,6 +26,7 @@
 #define I386_GNULINUX_TARGET
 #define HAVE_I387_REGS
 #ifdef HAVE_PTRACE_GETFPXREGS
+#define FILL_FPXREGSET
 #define HAVE_SSE_REGS
 #endif
 

Re: [RFA] Save SSE registers to generated core files, without breaking other architectures

[Michael Snyder]

Daniel Jacobowitz wrote:
> 
> How does this patch look?  The original problem was that we can not rely on
> HAVE_PTRACE_FPXREGSET, because the presence of PTRACE_FPXREGSET doesn't
> imply that GDB has a fill_fpxregset function available.  I updated it to
> check for a separate flag.  It's not ideal, but until we separate regset
> handling into an appropriate vector, it's the best we can do (I think).
> 
> OK?

Well, I like it (and thanks for working on it), 
but Andrew might have something to say about adding
a new macro.   Andrew?

> 
> --
> Daniel Jacobowitz                           Carnegie Mellon University
> MontaVista Software                         Debian GNU/Linux Developer
> 
> 2002-04-02  Daniel Jacobowitz  <drow@mvista.com>
> 
>         * config/i386/tm-linux.h: Define FILL_FPXREGSET.
>         * gregset.h: If FILL_FPXREGSET is defined, provide
>         gdb_fpxregset_t, supply_fpxregset, and fill_fpxregset.
>         * linux-proc.c (linux_do_thread_registers): If FILL_FPXREGSET
>         is defined, call fill_fpxregset.
> 
> Index: gregset.h
> ===================================================================
> RCS file: /cvs/src/src/gdb/gregset.h,v
> retrieving revision 1.5
> diff -u -p -r1.5 gregset.h
> --- gregset.h   2002/02/24 22:14:33     1.5
> +++ gregset.h   2002/04/02 19:51:15
> @@ -52,5 +52,18 @@ extern void supply_fpregset (gdb_fpregse
>  extern void fill_gregset (gdb_gregset_t *gregs, int regno);
>  extern void fill_fpregset (gdb_fpregset_t *fpregs, int regno);
> 
> +#ifdef FILL_FPXREGSET
> +/* Linux/i386: Copy register values between GDB's internal register cache
> +   and the i386 extended floating point registers.  */
> +
> +#ifndef GDB_FPXREGSET_T
> +#define GDB_FPXREGSET_T elf_fpxregset_t
> +#endif
> +
> +typedef GDB_FPXREGSET_T gdb_fpxregset_t;
> +
> +extern void supply_fpxregset (gdb_fpxregset_t *fpxregs);
> +extern void fill_fpxregset (gdb_fpxregset_t *fpxregs, int regno);
> +#endif
> 
>  #endif
> Index: linux-proc.c
> ===================================================================
> RCS file: /cvs/src/src/gdb/linux-proc.c,v
> retrieving revision 1.8
> diff -u -p -r1.8 linux-proc.c
> --- linux-proc.c        2002/03/25 19:47:41     1.8
> +++ linux-proc.c        2002/04/02 19:51:15
> @@ -167,6 +167,9 @@ linux_do_thread_registers (bfd *obfd, pt
>  {
>    gdb_gregset_t gregs;
>    gdb_fpregset_t fpregs;
> +#ifdef FILL_FPXREGSET
> +  gdb_fpxregset_t fpxregs;
> +#endif
>    unsigned long merged_pid = ptid_get_tid (ptid) << 16 | ptid_get_pid (ptid);
> 
>    fill_gregset (&gregs, -1);
> @@ -183,6 +186,14 @@ linux_do_thread_registers (bfd *obfd, pt
>                                               note_size,
>                                               &fpregs,
>                                               sizeof (fpregs));
> +#ifdef FILL_FPXREGSET
> +  fill_fpxregset (&fpxregs, -1);
> +  note_data = (char *) elfcore_write_prxfpreg (obfd,
> +                                              note_data,
> +                                              note_size,
> +                                              &fpxregs,
> +                                              sizeof (fpxregs));
> +#endif
>    return note_data;
>  }
> 
> Index: config/i386/tm-linux.h
> ===================================================================
> RCS file: /cvs/src/src/gdb/config/i386/tm-linux.h,v
> retrieving revision 1.17
> diff -u -p -r1.17 tm-linux.h
> --- tm-linux.h  2002/02/24 22:56:05     1.17
> +++ tm-linux.h  2002/04/02 19:51:46
> @@ -26,6 +26,7 @@
>  #define I386_GNULINUX_TARGET
>  #define HAVE_I387_REGS
>  #ifdef HAVE_PTRACE_GETFPXREGS
> +#define FILL_FPXREGSET
>  #define HAVE_SSE_REGS
>  #endif
>

Re: [RFA] Save SSE registers to generated core files, without breaking other architectures

[Andrew Cagney]

> Daniel Jacobowitz wrote:
> 
>> 
>> How does this patch look?  The original problem was that we can not rely on
>> HAVE_PTRACE_FPXREGSET, because the presence of PTRACE_FPXREGSET doesn't
>> imply that GDB has a fill_fpxregset function available.  I updated it to
>> check for a separate flag.  It's not ideal, but until we separate regset
>> handling into an appropriate vector, it's the best we can do (I think).
>> 
>> OK?
> 
> 
> Well, I like it (and thanks for working on it), 
> but Andrew might have something to say about adding
> a new macro.   Andrew?


>> * config/i386/tm-linux.h: Define FILL_FPXREGSET.

linux-proc.c is only built native so I think FILL_.. should only be 
defined for native builds => config/i386/nm-linux.h.

Having #ifdef code enabled via a tm.h file macro just feels wrong.

enjoy,
Andrew

Re: [RFA] Save SSE registers to generated core files, without breaking other architectures

[Daniel Jacobowitz]

On Sun, Apr 14, 2002 at 06:22:13PM -0400, Andrew Cagney wrote:
> >Daniel Jacobowitz wrote:
> >
> >>
> >>How does this patch look?  The original problem was that we can not rely 
> >>on
> >>HAVE_PTRACE_FPXREGSET, because the presence of PTRACE_FPXREGSET doesn't
> >>imply that GDB has a fill_fpxregset function available.  I updated it to
> >>check for a separate flag.  It's not ideal, but until we separate regset
> >>handling into an appropriate vector, it's the best we can do (I think).
> >>
> >>OK?
> >
> >
> >Well, I like it (and thanks for working on it), 
> >but Andrew might have something to say about adding
> >a new macro.   Andrew?
> 
> 
> >>* config/i386/tm-linux.h: Define FILL_FPXREGSET.
> 
> linux-proc.c is only built native so I think FILL_.. should only be 
> defined for native builds => config/i386/nm-linux.h.
> 
> Having #ifdef code enabled via a tm.h file macro just feels wrong.

And the function in question is even in i386-linux-nat.c.  nm-linux.h
sounds like the right place to put it.  OK with that change, then?

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: [RFA] Save SSE registers to generated core files, without breaking other architectures

[Andrew Cagney]

> linux-proc.c is only built native so I think FILL_.. should only be 
>> defined for native builds => config/i386/nm-linux.h.
>> 
>> Having #ifdef code enabled via a tm.h file macro just feels wrong.
> 
> 
> And the function in question is even in i386-linux-nat.c.  nm-linux.h
> sounds like the right place to put it.  OK with that change, then?

(MarkK is off line for a few weeks)

Well ...  Ok, is probably a bit strong :-)  But yes.

Andrew

Re: [RFA] Save SSE registers to generated core files, without breakingother architectures

[Michael Snyder]

Andrew Cagney wrote:
> 
> > linux-proc.c is only built native so I think FILL_.. should only be
> >> defined for native builds => config/i386/nm-linux.h.
> >>
> >> Having #ifdef code enabled via a tm.h file macro just feels wrong.
> >
> >
> > And the function in question is even in i386-linux-nat.c.  nm-linux.h
> > sounds like the right place to put it.  OK with that change, then?
> 
> (MarkK is off line for a few weeks)
> 
> Well ...  Ok, is probably a bit strong :-)  But yes.


So Daniel -- commit it, already!   ;-)
(I want to see this working!)

Re: [RFA] Save SSE registers to generated core files, without breakingother architectures

[Daniel Jacobowitz]

On Wed, Apr 24, 2002 at 02:34:52PM -0700, Michael Snyder wrote:
> Andrew Cagney wrote:
> > 
> > > linux-proc.c is only built native so I think FILL_.. should only be
> > >> defined for native builds => config/i386/nm-linux.h.
> > >>
> > >> Having #ifdef code enabled via a tm.h file macro just feels wrong.
> > >
> > >
> > > And the function in question is even in i386-linux-nat.c.  nm-linux.h
> > > sounds like the right place to put it.  OK with that change, then?
> > 
> > (MarkK is off line for a few weeks)
> > 
> > Well ...  Ok, is probably a bit strong :-)  But yes.
> 
> 
> So Daniel -- commit it, already!   ;-)
> (I want to see this working!)

Thanks for the nudge.  Committed.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer