[PATCH] Use PT_IO ptrace request if available

[PATCH] Use PT_IO ptrace request if available

[Mark Kettenis]

The attached patch makes use of the PT_IO ptrace request that's
available in *BSD for data transfers to/from the inferior.  It's a
couple of weeks ago now that my patch that fixes the FreeBSD
5.0-CURRENT kernel went in, and I hope that OpenBSD and NetBSD don't
have any problems.  We'll probably find out soon :-).  Note that if
you run FreeBSD 5.0-CURRENT from before October 16 and after March 16,
your machine will probably lock up when using GDB.  However, since
-CURRENT is a development version, I think it's acceptable to ask
people to upgrade.

It would be great if we could get Linux to support the same interface.

Committed.

Mark

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

	* infptrace.c (child_xfer_memory): Make use of the new PT_IO
	request that's available in *BSD.

Index: infptrace.c
===================================================================
RCS file: /cvs/src/src/gdb/infptrace.c,v
retrieving revision 1.21
diff -u -p -r1.21 infptrace.c
--- infptrace.c 19 Jan 2002 03:32:36 -0000 1.21
+++ infptrace.c 8 Nov 2002 23:45:18 -0000
@@ -514,6 +514,37 @@ child_xfer_memory (CORE_ADDR memaddr, ch
   PTRACE_XFER_TYPE *buffer;
   struct cleanup *old_chain = NULL;
 
+#ifdef PT_IO
+  /* OpenBSD 3.1, NetBSD 1.6 and FreeBSD 5.0 have a new PT_IO request
+     that promises to be much more efficient in reading and writing
+     data in the traced process's address space.  */
+
+  {
+    struct ptrace_io_desc piod;
+
+    /* NOTE: We assume that there are no distinct address spaces for
+       instruction and data.  */
+    piod.piod_op = write ? PIOD_WRITE_D : PIOD_READ_D;
+    piod.piod_offs = (void *) memaddr;
+    piod.piod_addr = myaddr;
+    piod.piod_len = len;
+
+    if (ptrace (PT_IO, PIDGET (inferior_ptid), (caddr_t) &piod, 0) == -1)
+      {
+	/* If the PT_IO request is somehow not supported, fallback on
+           using PT_WRITE_D/PT_READ_D.  Otherwise we will return zero
+           to indicate failure.  */
+	if (errno != EINVAL)
+	  return 0;
+      }
+    else
+      {
+	/* Return the actual number of bytes read or written.  */
+	return piod.piod_len;
+      }
+  }
+#endif
+
   /* Allocate buffer of that many longwords.  */
   if (len < GDB_MAX_ALLOCA)
     {

Re: [PATCH] Use PT_IO ptrace request if available

[Daniel Jacobowitz]

On Sat, Nov 09, 2002 at 12:58:02AM +0100, Mark Kettenis wrote:
> The attached patch makes use of the PT_IO ptrace request that's
> available in *BSD for data transfers to/from the inferior.  It's a
> couple of weeks ago now that my patch that fixes the FreeBSD
> 5.0-CURRENT kernel went in, and I hope that OpenBSD and NetBSD don't
> have any problems.  We'll probably find out soon :-).  Note that if
> you run FreeBSD 5.0-CURRENT from before October 16 and after March 16,
> your machine will probably lock up when using GDB.  However, since
> -CURRENT is a development version, I think it's acceptable to ask
> people to upgrade.
> 
> It would be great if we could get Linux to support the same interface.

We've got read covered, from /proc/<pid>/mem; I'm going to implement
that when I get a chance.

I suppose I could implement PT_IO the way you did... I was planning on
a Sun-style PTRACE_READDATA (SPARC Linux already has this) but it's not
possible on i386 (syscall with five arguments).  I'm not sure it's
worth it; it's generally reading that we do in bulk, not writing.

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer