Re: CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer_interrupt occurs

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Jan 07, 2019 at 01:28:34PM +0100, Mike Galbraith wrote:
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 960ad0ce77d7..420624c49f38 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -5007,9 +5007,9 @@ void init_cfs_bandwidth(struct cfs_bandwidth *cfs_b)
>  	cfs_b->period = ns_to_ktime(default_cfs_period());
>  
>  	INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq);
> -	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED);
> +	hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_ABS_PINNED_HARD);
>  	cfs_b->period_timer.function = sched_cfs_period_timer;
> -	hrtimer_init(&cfs_b->slack_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
> +	hrtimer_init(&cfs_b->slack_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_HARD);
>  	cfs_b->slack_timer.function = sched_cfs_slack_timer;
>  }

Right, that should sort it. But I'm not sure this is the best solution
though. That cfs-runtime crud can (IIRC) iterate lists etc.. so running
it from the softirq isn't a bad idea. We just need to fix that locking
up a bit.

Something a wee bit like so perhaps..

---
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 13776fac7b74..3cfe26aa098a 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4566,7 +4566,7 @@ static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
 		struct rq *rq = rq_of(cfs_rq);
 		struct rq_flags rf;
 
-		rq_lock(rq, &rf);
+		rq_lock_irqsave(rq, &rf);
 		if (!cfs_rq_throttled(cfs_rq))
 			goto next;
 
@@ -4583,7 +4583,7 @@ static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
 			unthrottle_cfs_rq(cfs_rq);
 
 next:
-		rq_unlock(rq, &rf);
+		rq_unlock_irqrestore(rq, &rf);
 
 		if (!remaining)
 			break;
@@ -4599,7 +4599,7 @@ static u64 distribute_cfs_runtime(struct cfs_bandwidth *cfs_b,
  * period the timer is deactivated until scheduling resumes; cfs_b->idle is
  * used to track this state.
  */
-static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun)
+static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun, unsigned long flags)
 {
 	u64 runtime, runtime_expires;
 	int throttled;
@@ -4641,11 +4641,11 @@ static int do_sched_cfs_period_timer(struct cfs_bandwidth *cfs_b, int overrun)
 	while (throttled && cfs_b->runtime > 0 && !cfs_b->distribute_running) {
 		runtime = cfs_b->runtime;
 		cfs_b->distribute_running = 1;
-		raw_spin_unlock(&cfs_b->lock);
+		raw_spin_unlock_irqrestore(&cfs_b->lock, flags);
 		/* we can't nest cfs_b->lock while distributing bandwidth */
 		runtime = distribute_cfs_runtime(cfs_b, runtime,
 						 runtime_expires);
-		raw_spin_lock(&cfs_b->lock);
+		raw_spin_lock_irqsave(&cfs_b->lock, flags);
 
 		cfs_b->distribute_running = 0;
 		throttled = !list_empty(&cfs_b->throttled_cfs_rq);
@@ -4754,17 +4754,18 @@ static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq)
 static void do_sched_cfs_slack_timer(struct cfs_bandwidth *cfs_b)
 {
 	u64 runtime = 0, slice = sched_cfs_bandwidth_slice();
+	unsigned long flags;
 	u64 expires;
 
 	/* confirm we're still not at a refresh boundary */
-	raw_spin_lock(&cfs_b->lock);
+	raw_spin_lock_irqsave(&cfs_b->lock, flags);
 	if (cfs_b->distribute_running) {
-		raw_spin_unlock(&cfs_b->lock);
+		raw_spin_unlock_irqrestore(&cfs_b->lock, flags);
 		return;
 	}
 
 	if (runtime_refresh_within(cfs_b, min_bandwidth_expiration)) {
-		raw_spin_unlock(&cfs_b->lock);
+		raw_spin_unlock_irqrestore(&cfs_b->lock, flags);
 		return;
 	}
 
@@ -4775,18 +4776,18 @@ static void do_sched_cfs_slack_timer(struct cfs_bandwidth *cfs_b)
 	if (runtime)
 		cfs_b->distribute_running = 1;
 
-	raw_spin_unlock(&cfs_b->lock);
+	raw_spin_unlock_irqrestore(&cfs_b->lock, flags);
 
 	if (!runtime)
 		return;
 
 	runtime = distribute_cfs_runtime(cfs_b, runtime, expires);
 
-	raw_spin_lock(&cfs_b->lock);
+	raw_spin_lock_irqsave(&cfs_b->lock, flags);
 	if (expires == cfs_b->runtime_expires)
 		lsub_positive(&cfs_b->runtime, runtime);
 	cfs_b->distribute_running = 0;
-	raw_spin_unlock(&cfs_b->lock);
+	raw_spin_unlock_irqrestore(&cfs_b->lock, flags);
 }
 
 /*
@@ -4864,20 +4865,21 @@ static enum hrtimer_restart sched_cfs_period_timer(struct hrtimer *timer)
 {
 	struct cfs_bandwidth *cfs_b =
 		container_of(timer, struct cfs_bandwidth, period_timer);
+	unsigned long flags;
 	int overrun;
 	int idle = 0;
 
-	raw_spin_lock(&cfs_b->lock);
+	raw_spin_lock_irqsave(&cfs_b->lock, flags);
 	for (;;) {
 		overrun = hrtimer_forward_now(timer, cfs_b->period);
 		if (!overrun)
 			break;
 
-		idle = do_sched_cfs_period_timer(cfs_b, overrun);
+		idle = do_sched_cfs_period_timer(cfs_b, overrun, flags);
 	}
 	if (idle)
 		cfs_b->period_active = 0;
-	raw_spin_unlock(&cfs_b->lock);
+	raw_spin_unlock_irqrestore(&cfs_b->lock, flags);
 
 	return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
 }