tile kgdb: fix bug in copy to gdb regs, and optimize memset

David Binderman pointed out that we were doing a full memset()
of the gdb register buffer and then doing a memcpy() to it that
was almost as big.  This commit optimizes that by only doing a
memset() of the registers that are intended to be zero.

While making this change I noticed that we were not copying the
link register (LR, number 55) due to a fencepost error in commit
f419e6f6 ("arch: tile: kernel: kgdb.c: Use memcpy() instead of
pointer copy one by one"), and I've corrected that as well.
Reported-by: David Binderman <dcb314@hotmail.com>
Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>

arch/tile/kernel/kgdb.c

@@ -126,15 +126,15 @@ void
 sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *task)
 {
 	struct pt_regs *thread_regs;
+	const int NGPRS = TREG_LAST_GPR + 1;
 
 	if (task == NULL)
 		return;
 
-	/* Initialize to zero. */
-	memset(gdb_regs, 0, NUMREGBYTES);
-
 	thread_regs = task_pt_regs(task);
-	memcpy(gdb_regs, thread_regs, TREG_LAST_GPR * sizeof(unsigned long));
+	memcpy(gdb_regs, thread_regs, NGPRS * sizeof(unsigned long));
+	memset(&gdb_regs[NGPRS], 0,
+	       (TILEGX_PC_REGNUM - NGPRS) * sizeof(unsigned long));
 	gdb_regs[TILEGX_PC_REGNUM] = thread_regs->pc;
 	gdb_regs[TILEGX_FAULTNUM_REGNUM] = thread_regs->faultnum;
 }