hrtimer: prevent migration of per CPU hrtimers

Impact: per CPU hrtimers can be migrated from a dead CPU

The hrtimer code has no knowledge about per CPU timers, but we need to
prevent the migration of such timers and warn when such a timer is
active at migration time.

Explicitely mark the timers as per CPU and use a more understandable
mode descriptor for the interrupts safe unlocked callback mode, which
is used by hrtimer_sleeper and the scheduler code.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

5 files changed, 40 insertions, 19 deletions

diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
index bdd88df1b4e..2f245fe63bd 100644
--- a/include/linux/hrtimer.h
+++ b/include/linux/hrtimer.h
@@ -47,14 +47,22 @@ enum hrtimer_restart {
  *	HRTIMER_CB_IRQSAFE:		Callback may run in hardirq context
  *	HRTIMER_CB_IRQSAFE_NO_RESTART:	Callback may run in hardirq context and
  *					does not restart the timer
- *	HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:	Callback must run in hardirq context
- *					Special mode for tick emultation
+ *	HRTIMER_CB_IRQSAFE_PERCPU:	Callback must run in hardirq context
+ *					Special mode for tick emulation and
+ *					scheduler timer. Such timers are per
+ *					cpu and not allowed to be migrated on
+ *					cpu unplug.
+ *	HRTIMER_CB_IRQSAFE_UNLOCKED:	Callback should run in hardirq context
+ *					with timer->base lock unlocked
+ *					used for timers which call wakeup to
+ *					avoid lock order problems with rq->lock
  */
 enum hrtimer_cb_mode {
 	HRTIMER_CB_SOFTIRQ,
 	HRTIMER_CB_IRQSAFE,
 	HRTIMER_CB_IRQSAFE_NO_RESTART,
-	HRTIMER_CB_IRQSAFE_NO_SOFTIRQ,
+	HRTIMER_CB_IRQSAFE_PERCPU,
+	HRTIMER_CB_IRQSAFE_UNLOCKED,
 };
 
 /*
diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c
index ace723dd1e5..cdec83e722f 100644
--- a/kernel/hrtimer.c
+++ b/kernel/hrtimer.c
@@ -672,13 +672,14 @@ static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
 			 */
 			BUG_ON(timer->function(timer) != HRTIMER_NORESTART);
 			return 1;
-		case HRTIMER_CB_IRQSAFE_NO_SOFTIRQ:
+		case HRTIMER_CB_IRQSAFE_PERCPU:
+		case HRTIMER_CB_IRQSAFE_UNLOCKED:
 			/*
 			 * This is solely for the sched tick emulation with
 			 * dynamic tick support to ensure that we do not
 			 * restart the tick right on the edge and end up with
 			 * the tick timer in the softirq ! The calling site
-			 * takes care of this.
+			 * takes care of this. Also used for hrtimer sleeper !
 			 */
 			debug_hrtimer_deactivate(timer);
 			return 1;
@@ -1245,7 +1246,8 @@ static void __run_hrtimer(struct hrtimer *timer)
 	timer_stats_account_hrtimer(timer);
 
 	fn = timer->function;
-	if (timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ) {
+	if (timer->cb_mode == HRTIMER_CB_IRQSAFE_PERCPU ||
+	    timer->cb_mode == HRTIMER_CB_IRQSAFE_UNLOCKED) {
 		/*
 		 * Used for scheduler timers, avoid lock inversion with
 		 * rq->lock and tasklist_lock.
@@ -1452,7 +1454,7 @@ void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
 	sl->timer.function = hrtimer_wakeup;
 	sl->task = task;
 #ifdef CONFIG_HIGH_RES_TIMERS
-	sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ;
+	sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED;
 #endif
 }
 
@@ -1592,7 +1594,7 @@ static void __cpuinit init_hrtimers_cpu(int cpu)
 #ifdef CONFIG_HOTPLUG_CPU
 
 static int migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
-				struct hrtimer_clock_base *new_base)
+				struct hrtimer_clock_base *new_base, int dcpu)
 {
 	struct hrtimer *timer;
 	struct rb_node *node;
@@ -1604,6 +1606,18 @@ static int migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
 		debug_hrtimer_deactivate(timer);
 
 		/*
+		 * Should not happen. Per CPU timers should be
+		 * canceled _before_ the migration code is called
+		 */
+		if (timer->cb_mode == HRTIMER_CB_IRQSAFE_PERCPU) {
+			__remove_hrtimer(timer, old_base,
+					 HRTIMER_STATE_INACTIVE, 0);
+			WARN(1, "hrtimer (%p %p)active but cpu %d dead\n",
+			     timer, timer->function, dcpu);
+			continue;
+		}
+
+		/*
 		 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
 		 * timer could be seen as !active and just vanish away
 		 * under us on another CPU
@@ -1619,12 +1633,11 @@ static int migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
 		/*
 		 * Happens with high res enabled when the timer was
 		 * already expired and the callback mode is
-		 * HRTIMER_CB_IRQSAFE_NO_SOFTIRQ
-		 * (hrtimer_sleeper). The enqueue code does not move
-		 * them to the soft irq pending list for
-		 * performance/latency reasons, but in the migration
-		 * state, we need to do that otherwise we end up with
-		 * a stale timer.
+		 * HRTIMER_CB_IRQSAFE_UNLOCKED (hrtimer_sleeper). The
+		 * enqueue code does not move them to the soft irq
+		 * pending list for performance/latency reasons, but
+		 * in the migration state, we need to do that
+		 * otherwise we end up with a stale timer.
 		 */
 		if (timer->state == HRTIMER_STATE_MIGRATE) {
 			timer->state = HRTIMER_STATE_PENDING;
@@ -1682,7 +1695,7 @@ static void migrate_hrtimers(int cpu)
 
 	for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
 		if (migrate_hrtimer_list(&old_base->clock_base[i],
-					 &new_base->clock_base[i]))
+					 &new_base->clock_base[i], cpu))
 			raise = 1;
 	}
 
diff --git a/kernel/sched.c b/kernel/sched.c
index 13dd2db9fb2..ad1962dc0aa 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -201,7 +201,7 @@ void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
 	hrtimer_init(&rt_b->rt_period_timer,
 			CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 	rt_b->rt_period_timer.function = sched_rt_period_timer;
-	rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ;
+	rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED;
 }
 
 static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
@@ -1119,7 +1119,7 @@ static void init_rq_hrtick(struct rq *rq)
 
 	hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 	rq->hrtick_timer.function = hrtick;
-	rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ;
+	rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU;
 }
 #else
 static inline void hrtick_clear(struct rq *rq)
diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index 39019b3f762..cb02324bdb8 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -625,7 +625,7 @@ void tick_setup_sched_timer(void)
 	 */
 	hrtimer_init(&ts->sched_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS);
 	ts->sched_timer.function = tick_sched_timer;
-	ts->sched_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ;
+	ts->sched_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU;
 
 	/* Get the next period (per cpu) */
 	ts->sched_timer.expires = tick_init_jiffy_update();
diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c
index bb948e52ce2..db58fb66a13 100644
--- a/kernel/trace/trace_sysprof.c
+++ b/kernel/trace/trace_sysprof.c
@@ -202,7 +202,7 @@ static void start_stack_timer(int cpu)
 
 	hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 	hrtimer->function = stack_trace_timer_fn;
-	hrtimer->cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ;
+	hrtimer->cb_mode = HRTIMER_CB_IRQSAFE_PERCPU;
 
 	hrtimer_start(hrtimer, ns_to_ktime(sample_period), HRTIMER_MODE_REL);
 }