rcu_ctrlblk.lock is used to read the ->cur and ->next_pending
atomically in __rcu_process_callbacks(). It can be replaced
by a couple of memory barriers.

rcu_start_batch:
	rcp->next_pending = 0;
	smp_wmb();
	rcp->cur++;

__rcu_process_callbacks:
	rdp->batch = rcp->cur + 1;
	smp_rmb();
	if (!rcp->next_pending)
		rcu_start_batch(rcp, rsp, 1);

This way, if __rcu_process_callbacks() sees incremented ->cur value,
it must also see that ->next_pending == 0 (or rcu_start_batch() is
already in progress on another cpu).
Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>