diff --git a/kernel/rcu/tree_nocb.h b/kernel/rcu/tree_nocb.h
index 97b99cd0692323eda39013951ed4e01eb9382a11..16865475120ba38c741aae897a3dc8d99f95d0ee 100644
--- a/kernel/rcu/tree_nocb.h
+++ b/kernel/rcu/tree_nocb.h
@@ -554,13 +554,19 @@ static void __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_alldone,
 			rcu_nocb_unlock(rdp);
 			wake_nocb_gp_defer(rdp, RCU_NOCB_WAKE_LAZY,
 					   TPS("WakeLazy"));
-		} else if (!irqs_disabled_flags(flags)) {
+		} else if (!irqs_disabled_flags(flags) && cpu_online(rdp->cpu)) {
 			/* ... if queue was empty ... */
 			rcu_nocb_unlock(rdp);
 			wake_nocb_gp(rdp, false);
 			trace_rcu_nocb_wake(rcu_state.name, rdp->cpu,
 					    TPS("WakeEmpty"));
 		} else {
+			/*
+			 * Don't do the wake-up upfront on fragile paths.
+			 * Also offline CPUs can't call swake_up_one_online() from
+			 * (soft-)IRQs. Rely on the final deferred wake-up from
+			 * rcutree_report_cpu_dead()
+			 */
 			rcu_nocb_unlock(rdp);
 			wake_nocb_gp_defer(rdp, RCU_NOCB_WAKE,
 					   TPS("WakeEmptyIsDeferred"));