[PATCH 1/3] stop_machine: ensure that a queued callback will be called before cpu_stop_park()

[Oleg Nesterov <oleg@redhat.com>]

cpu_stop_queue_work() checks stopper->enabled before it queues the
work, but ->enabled == T can only guarantee cpu_stop_signal_done()
if we race with cpu_down().

This is not enough for stop_two_cpus() or stop_machine(), they will
deadlock if multi_cpu_stop() won't be called by one of the target
CPU's. stop_machine/stop_cpus are fine, they rely on stop_cpus_mutex.
But stop_two_cpus() has to check cpu_active() to avoid the same race
with hotplug, and this check is very unobvious and probably not even
correct if we race with cpu_up().

Change cpu_down() pass to clear ->enabled before cpu_stopper_thread()
flushes the pending ->works and returns with KTHREAD_SHOULD_PARK set.

Note also that smpboot_thread_call() calls cpu_stop_unpark() which
sets enabled == T at CPU_ONLINE stage, so this CPU can't go away until
cpu_stopper_thread() is called at least once. This all means that if
cpu_stop_queue_work() succeeds, we know that work->fn() will be called.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---
 include/linux/stop_machine.h |    1 +
 kernel/cpu.c                 |    2 +-
 kernel/stop_machine.c        |   23 +++++++++++++----------
 3 files changed, 15 insertions(+), 11 deletions(-)

diff --git a/include/linux/stop_machine.h b/include/linux/stop_machine.h
index 414d924..7b76362 100644
--- a/include/linux/stop_machine.h
+++ b/include/linux/stop_machine.h
@@ -33,6 +33,7 @@ void stop_one_cpu_nowait(unsigned int cpu, cpu_stop_fn_t fn, void *arg,
 			 struct cpu_stop_work *work_buf);
 int stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg);
 int try_stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg);
+void stop_machine_park(int cpu);
 
 #else	/* CONFIG_SMP */
 
diff --git a/kernel/cpu.c b/kernel/cpu.c
index 1f4566c..8a7225a 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -342,7 +342,7 @@ static int __ref take_cpu_down(void *_param)
 	/* Give up timekeeping duties */
 	tick_handover_do_timer();
 	/* Park the stopper thread */
-	kthread_park(current);
+	stop_machine_park((long)param->hcpu);
 	return 0;
 }
 
diff --git a/kernel/stop_machine.c b/kernel/stop_machine.c
index 12484e5..6a40209 100644
--- a/kernel/stop_machine.c
+++ b/kernel/stop_machine.c
@@ -452,6 +452,18 @@ repeat:
 	}
 }
 
+void stop_machine_park(int cpu)
+{
+	struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
+	/*
+	 * Lockless. cpu_stopper_thread() will take stopper->lock and flush
+	 * the pending works before it parks, until then it is fine to queue
+	 * the new works.
+	 */
+	stopper->enabled = false;
+	kthread_park(stopper->thread);
+}
+
 extern void sched_set_stop_task(int cpu, struct task_struct *stop);
 
 static void cpu_stop_create(unsigned int cpu)
@@ -462,17 +474,8 @@ static void cpu_stop_create(unsigned int cpu)
 static void cpu_stop_park(unsigned int cpu)
 {
 	struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
-	struct cpu_stop_work *work, *tmp;
-	unsigned long flags;
 
-	/* drain remaining works */
-	spin_lock_irqsave(&stopper->lock, flags);
-	list_for_each_entry_safe(work, tmp, &stopper->works, list) {
-		list_del_init(&work->list);
-		cpu_stop_signal_done(work->done, false);
-	}
-	stopper->enabled = false;
-	spin_unlock_irqrestore(&stopper->lock, flags);
+	WARN_ON(!list_empty(&stopper->works));
 }
 
 static void cpu_stop_unpark(unsigned int cpu)
-- 
1.5.5.1