smp/hotplug: Differentiate the AP completion between up and down

With lockdep-crossrelease we get deadlock reports that span cpu-up and
cpu-down chains. Such deadlocks cannot possibly happen because cpu-up
and cpu-down are globally serialized.

takedown_cpu()
irq_lock_sparse()
wait_for_completion(&st->done)

cpuhp_thread_fun
cpuhp_up_callback
cpuhp_invoke_callback
irq_affinity_online_cpu
irq_local_spare()
irq_unlock_sparse()
complete(&st->done)

Now that we have consistent AP state, we can trivially separate the
AP completion between up and down using st->bringup.

Signed-off-by: Peter Zijlstra (Intel) <[e-mail]>
Signed-off-by: Thomas Gleixner <[e-mail]>
Acked-by: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
Cc: [e-mail]
Link: https://lkml.kernel.org/r/[e-mail]

kernel/cpu.c

···
  * @bringup:	Single callback bringup or teardown selector
  * @cb_state:	The state for a single callback (install/uninstall)
  * @result:	Result of the operation
- * @done:	Signal completion to the issuer of the task
+ * @done_up:	Signal completion to the issuer of the task for cpu-up
+ * @done_down:	Signal completion to the issuer of the task for cpu-down
  */
 struct cpuhp_cpu_state {
 	enum cpuhp_state	state;
···
 	struct hlist_node	*last;
 	enum cpuhp_state	cb_state;
 	int			result;
-	struct completion	done;
+	struct completion	done_up;
+	struct completion	done_down;
 #endif
 };
 
···
 	 * purposes as that state is handled explicitly in cpu_down.
 	 */
 	return state > CPUHP_BRINGUP_CPU && state != CPUHP_TEARDOWN_CPU;
-}
-
-/*
- * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
- */
-static bool cpuhp_is_atomic_state(enum cpuhp_state state)
-{
-	return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
 }
 
 static struct cpuhp_step *cpuhp_get_step(enum cpuhp_state state)
···
 }
 
 #ifdef CONFIG_SMP
+static inline void wait_for_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
+{
+	struct completion *done = bringup ? &st->done_up : &st->done_down;
+	wait_for_completion(done);
+}
+
+static inline void complete_ap_thread(struct cpuhp_cpu_state *st, bool bringup)
+{
+	struct completion *done = bringup ? &st->done_up : &st->done_down;
+	complete(done);
+}
+
+/*
+ * The former STARTING/DYING states, ran with IRQs disabled and must not fail.
+ */
+static bool cpuhp_is_atomic_state(enum cpuhp_state state)
+{
+	return CPUHP_AP_IDLE_DEAD <= state && state < CPUHP_AP_ONLINE;
+}
+
 /* Serializes the updates to cpu_online_mask, cpu_present_mask */
 static DEFINE_MUTEX(cpu_add_remove_lock);
 bool cpuhp_tasks_frozen;
···
 	smp_mb();
 	st->should_run = true;
 	wake_up_process(st->thread);
-	wait_for_completion(&st->done);
+	wait_for_ap_thread(st, st->bringup);
 }
 
 static int cpuhp_kick_ap(struct cpuhp_cpu_state *st, enum cpuhp_state target)
···
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
 
 	/* Wait for the CPU to reach CPUHP_AP_ONLINE_IDLE */
-	wait_for_completion(&st->done);
+	wait_for_ap_thread(st, true);
 	if (WARN_ON_ONCE((!cpu_online(cpu))))
 		return -ECANCELED;
 
···
 {
 	struct cpuhp_cpu_state *st = per_cpu_ptr(&cpuhp_state, cpu);
 
-	init_completion(&st->done);
+	init_completion(&st->done_up);
+	init_completion(&st->done_down);
 }
 
 static int cpuhp_should_run(unsigned int cpu)
···
 	cpuhp_lock_release(bringup);
 
 	if (!st->should_run)
-		complete(&st->done);
+		complete_ap_thread(st, bringup);
 }
 
 /* Invoke a single callback on a remote cpu */
···
 	 *
 	 * Wait for the stop thread to go away.
 	 */
-	wait_for_completion(&st->done);
+	wait_for_ap_thread(st, false);
 	BUG_ON(st->state != CPUHP_AP_IDLE_DEAD);
 
 	/* Interrupts are moved away from the dying cpu, reenable alloc/free */
···
 {
 	struct cpuhp_cpu_state *st = arg;
 
-	complete(&st->done);
+	complete_ap_thread(st, false);
 }
 
 void cpuhp_report_idle_dead(void)
···
 		return;
 
 	st->state = CPUHP_AP_ONLINE_IDLE;
-	complete(&st->done);
+	complete_ap_thread(st, true);
 }
 
 /* Requires cpu_add_remove_lock to be held */