[PATCH -v2 04/17] sched/core: Wait for tasks being pushed away on hotplug

[Peter Zijlstra <peterz@infradead.org>]

From: Thomas Gleixner <tglx@linutronix.de>

RT kernels need to ensure that all tasks which are not per CPU kthreads
have left the outgoing CPU to guarantee that no tasks are force migrated
within a migrate disabled section.

There is also some desire to (ab)use fine grained CPU hotplug control to
clear a CPU from active state to force migrate tasks which are not per CPU
kthreads away for power control purposes.

Add a mechanism which waits until all tasks which should leave the CPU
after the CPU active flag is cleared have moved to a different online CPU.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
 kernel/sched/core.c  |   40 +++++++++++++++++++++++++++++++++++++++-
 kernel/sched/sched.h |    4 ++++
 2 files changed, 43 insertions(+), 1 deletion(-)

--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -6900,8 +6900,21 @@ static bool balance_push(struct rq *rq)
 	 * Both the cpu-hotplug and stop task are in this case and are
 	 * required to complete the hotplug process.
 	 */
-	if (is_per_cpu_kthread(push_task))
+	if (is_per_cpu_kthread(push_task)) {
+		/*
+		 * If this is the idle task on the outgoing CPU try to wake
+		 * up the hotplug control thread which might wait for the
+		 * last task to vanish. The rcuwait_active() check is
+		 * accurate here because the waiter is pinned on this CPU
+		 * and can't obviously be running in parallel.
+		 */
+		if (!rq->nr_running && rcuwait_active(&rq->hotplug_wait)) {
+			raw_spin_unlock(&rq->lock);
+			rcuwait_wake_up(&rq->hotplug_wait);
+			raw_spin_lock(&rq->lock);
+		}
 		return false;
+	}
 
 	get_task_struct(push_task);
 	/*
@@ -6934,6 +6947,20 @@ static void balance_push_set(int cpu, bo
 	rq_unlock_irqrestore(rq, &rf);
 }
 
+/*
+ * Invoked from a CPUs hotplug control thread after the CPU has been marked
+ * inactive. All tasks which are not per CPU kernel threads are either
+ * pushed off this CPU now via balance_push() or placed on a different CPU
+ * during wakeup. Wait until the CPU is quiescent.
+ */
+static void balance_hotplug_wait(void)
+{
+	struct rq *rq = this_rq();
+
+	rcuwait_wait_event(&rq->hotplug_wait, rq->nr_running == 1,
+			   TASK_UNINTERRUPTIBLE);
+}
+
 #else
 
 static inline bool balance_push(struct rq *rq)
@@ -6945,6 +6972,10 @@ static void balance_push_set(int cpu, bo
 {
 }
 
+static inline void balance_hotplug_wait(void)
+{
+}
+
 #endif /* CONFIG_HOTPLUG_CPU */
 
 void set_rq_online(struct rq *rq)
@@ -7099,6 +7130,10 @@ int sched_cpu_deactivate(unsigned int cp
 		return ret;
 	}
 	sched_domains_numa_masks_clear(cpu);
+
+	/* Wait for all non per CPU kernel threads to vanish. */
+	balance_hotplug_wait();
+
 	return 0;
 }
 
@@ -7339,6 +7374,9 @@ void __init sched_init(void)
 
 		rq_csd_init(rq, &rq->nohz_csd, nohz_csd_func);
 #endif
+#ifdef CONFIG_HOTPLUG_CPU
+		rcuwait_init(&rq->hotplug_wait);
+#endif
 #endif /* CONFIG_SMP */
 		hrtick_rq_init(rq);
 		atomic_set(&rq->nr_iowait, 0);
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1004,6 +1004,10 @@ struct rq {
 
 	/* This is used to determine avg_idle's max value */
 	u64			max_idle_balance_cost;
+
+#ifdef CONFIG_HOTPLUG_CPU
+	struct rcuwait		hotplug_wait;
+#endif
 #endif /* CONFIG_SMP */
 
 #ifdef CONFIG_IRQ_TIME_ACCOUNTING