rcu: Remove preemption disabling from expedited CPU selection

It turns out that it is queue_delayed_work_on() rather than
queue_work_on() that has difficulties when used concurrently with
CPU-hotplug removal operations.  It is therefore unnecessary to protect
CPU identification and queue_work_on() with preempt_disable().

This commit therefore removes the preempt_disable() and preempt_enable()
from sync_rcu_exp_select_cpus(), which has the further benefit of reducing
the number of changes that must be maintained in the -rt patchset.
Reported-by: NThomas Gleixner <tglx@linutronix.de>
Reported-by: NSebastian Siewior <bigeasy@linutronix.de>
Suggested-by: NBoqun Feng <boqun.feng@gmail.com>
Signed-off-by: NPaul E. McKenney <paulmck@linux.ibm.com>

kernel/rcu/tree_exp.h

@@ -448,7 +448,6 @@ static void sync_rcu_exp_select_cpus(void)
 			continue;
 		}
 		INIT_WORK(&rnp->rew.rew_work, sync_rcu_exp_select_node_cpus);
-		preempt_disable();
 		cpu = find_next_bit(&rnp->ffmask, BITS_PER_LONG, -1);
 		/* If all offline, queue the work on an unbound CPU. */
 		if (unlikely(cpu > rnp->grphi - rnp->grplo))
@@ -456,7 +455,6 @@ static void sync_rcu_exp_select_cpus(void)
 		else
 			cpu += rnp->grplo;
 		queue_work_on(cpu, rcu_par_gp_wq, &rnp->rew.rew_work);
-		preempt_enable();
 		rnp->exp_need_flush = true;
 	}