powerpc: don't mess with r2 in copy_thread() and friends

kernel_thread() callbacks are *not* in modules and are not going to
be there.  And it's not even read in ppc32 ret_from_kernel_thread(),
so no need to bother with it there either.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

arch/powerpc/kernel/entry_64.S

@@ -373,7 +373,6 @@ _GLOBAL(ret_from_fork)
 _GLOBAL(ret_from_kernel_thread)
 	bl	.schedule_tail
 	REST_NVGPRS(r1)
-	REST_GPR(2,r1)
 	mtlr	r14
 	mr	r3,r15
 	blrl

arch/powerpc/kernel/process.c

@@ -751,12 +751,10 @@ int copy_thread(unsigned long clone_flags, unsigned long usp,
 		childregs->gpr[1] = sp + sizeof(struct pt_regs);
 #ifdef CONFIG_PPC64
 		childregs->gpr[14] = *(unsigned long *)usp;
-		childregs->gpr[2] = ((unsigned long *)usp)[1],
 		clear_tsk_thread_flag(p, TIF_32BIT);
 		childregs->softe = 1;
 #else
 		childregs->gpr[14] = usp;	/* function */
-		childregs->gpr[2] = (unsigned long) p;
 #endif
 		childregs->gpr[15] = arg;
 		p->thread.regs = NULL;	/* no user register state */