rcu: Make expedited GPs handle CPU 0 being offline

Currently, the parallelized initialization of expedited grace periods uses
the workqueue associated with each rcu_node structure's ->grplo field.
This works fine unless that CPU is offline.  This commit therefore uses
the CPU corresponding to the lowest-numbered online CPU, or just queues
the work on WORK_CPU_UNBOUND if there are no online CPUs corresponding
to this rcu_node structure.

Note that this patch uses cpu_is_offline() instead of the usual approach
of checking bits in the rcu_node structure's ->qsmaskinitnext field.  This
is safe because preemption is disabled across both the cpu_is_offline()
check and the call to queue_work_on().
Signed-off-by: Boqun Feng <boqun.feng@gmail.com>
[ paulmck: Disable preemption to close offline race window. ]
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
[ paulmck: Apply Peter Zijlstra feedback on CPU selection. ]
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>

kernel/rcu/tree_exp.h

@@ -472,6 +472,7 @@ static void sync_rcu_exp_select_node_cpus(struct work_struct *wp)
 static void sync_rcu_exp_select_cpus(struct rcu_state *rsp,
 				     smp_call_func_t func)
 {
+	int cpu;
 	struct rcu_node *rnp;
 
 	trace_rcu_exp_grace_period(rsp->name, rcu_exp_gp_seq_endval(rsp), TPS("reset"));
@@ -493,7 +494,13 @@ static void sync_rcu_exp_select_cpus(struct rcu_state *rsp,
 			continue;
 		}
 		INIT_WORK(&rnp->rew.rew_work, sync_rcu_exp_select_node_cpus);
-		queue_work_on(rnp->grplo, rcu_par_gp_wq, &rnp->rew.rew_work);
+		preempt_disable();
+		cpu = cpumask_next(rnp->grplo - 1, cpu_online_mask);
+		/* If all offline, queue the work on an unbound CPU. */
+		if (unlikely(cpu > rnp->grphi))
+			cpu = WORK_CPU_UNBOUND;
+		queue_work_on(cpu, rcu_par_gp_wq, &rnp->rew.rew_work);
+		preempt_enable();
 		rnp->exp_need_flush = true;
 	}