CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer

linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed

* CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer_interrupt occurs
       [not found] <AM0PR03MB480425D5999E0D08DAB30204BB8E0@AM0PR03MB4804.eurprd03.prod.outlook.com>
@ 2019-01-04 12:42 ` Tom Putzeys
  2019-01-07 10:26   ` Peter Zijlstra
  0 siblings, 1 reply; 10+ messages in thread
From: Tom Putzeys @ 2019-01-04 12:42 UTC (permalink / raw)
  To: mingo, peterz; +Cc: linux-kernel

Dear Ingo and Peter,

I would like to report a possible bug in the CFS scheduler causing a dead lock. 

We suspect this bug to have caused intermittent yet highly-persistent system freezes on our quad-core SMP systems.

We noticed the problem on 4.1.17 preempt-rt but we suspect the problematic code is not linked to the preempt-rt patch and is also present in the latest 4.20 kernel.

The problem concerns the use of spin_lock to lock cfs_b in a situation where the spin lock is used in an interrupt handler:
-  __run_hrtimer (in kernel/time/hrtimer.c) calls fn(timer) with IRQ's enabled. This can call sched_cfs_period_timer() (in kernel/sched/fair.c) which locks cfs_b. 
- the hard IRQ smp_apic_timer_interrupt can then occur. It can call ttwu_queue() which grabs the spin lock for its CPU run queue and can then try to enqueue a task via the CFS scheduler.
- this can call check_enqueue_throttle() which can call assign_cfs_rq_runtime() which tries to obtain the cfs_b lock. It is now blocked.

The cfs_b lock uses spin_lock and so was not intended for use inside a hard irq but the CFS scheduler does just that when it uses a hrtimer_interrupt to wake up and enqueue work. Our initial impression is that  the cfs_b needs to be locked using spin_lock_irqsave.

My colleague Mike Pearce has submitted a bug report on Bugzilla 3 weeks ago: https://bugzilla.kernel.org/show_bug.cgi?id=201993

We would appreciate any feedback.

Kind regards,

Tom

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer_interrupt occurs
  2019-01-04 12:42 ` CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer_interrupt occurs Tom Putzeys
@ 2019-01-07 10:26   ` Peter Zijlstra
  2019-01-07 12:28     ` Mike Galbraith
  0 siblings, 1 reply; 10+ messages in thread
From: Peter Zijlstra @ 2019-01-07 10:26 UTC (permalink / raw)
  To: Tom Putzeys
  Cc: mingo, linux-kernel, Sebastian Andrzej Siewior, Thomas Gleixner

On Fri, Jan 04, 2019 at 12:42:27PM +0000, Tom Putzeys wrote:
> Dear Ingo and Peter,
> 
> I would like to report a possible bug in the CFS scheduler causing a
> dead lock. 
> 
> We suspect this bug to have caused intermittent yet highly-persistent
> system freezes on our quad-core SMP systems.
> 
> We noticed the problem on 4.1.17 preempt-rt but we suspect the
> problematic code is not linked to the preempt-rt patch and is also
> present in the latest 4.20 kernel.
> 
> The problem concerns the use of spin_lock to lock cfs_b in a situation
> where the spin lock is used in an interrupt handler:

> -  __run_hrtimer (in kernel/time/hrtimer.c) calls fn(timer) with IRQ's
> enabled. This can call sched_cfs_period_timer() (in
> kernel/sched/fair.c) which locks cfs_b. 

Hurmph, that's the softirq timer handling. And that is a nasty subtle
difference in context between softirq and hardirq timers.

Also, upstream doesn't use HRTIMER_MODE_SOFT here, but I suppose -rt
forces everything !HARD into SOFT.

> - the hard IRQ smp_apic_timer_interrupt can then occur. It can call
> ttwu_queue() which grabs the spin lock for its CPU run queue and can
> then try to enqueue a task via the CFS scheduler.

> - this can call check_enqueue_throttle() which can call
> assign_cfs_rq_runtime() which tries to obtain the cfs_b lock. It is
> now blocked.
> 
> The cfs_b lock uses spin_lock and so was not intended for use inside a
> hard irq but the CFS scheduler does just that when it uses a
> hrtimer_interrupt to wake up and enqueue work. Our initial impression
> is that  the cfs_b needs to be locked using spin_lock_irqsave.

I would expect lockdep you also complain about this, but yes, something
like that. I was very much expecting this to run with IRQs disabled (and
it does on mainline afaict).

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer_interrupt occurs
  2019-01-07 10:26   ` Peter Zijlstra
@ 2019-01-07 12:28     ` Mike Galbraith
  2019-01-07 12:52       ` Peter Zijlstra
  0 siblings, 1 reply; 10+ messages in thread
From: Mike Galbraith @ 2019-01-07 12:28 UTC (permalink / raw)
  To: Peter Zijlstra, Tom Putzeys
  Cc: mingo, linux-kernel, Sebastian Andrzej Siewior, Thomas Gleixner

On Mon, 2019-01-07 at 11:26 +0100, Peter Zijlstra wrote:
> 
> I would expect lockdep you also complain about this...

And grumble it did.

commit df7e8acc0c9a84979a448d215b8ef889efe4ac5a
Author: Mike Galbraith <efault@gmx.de>
Date:   Fri May 4 08:14:38 2018 +0200

    sched/fair: Fix CFS bandwidth control lockdep DEADLOCK report
    
    CFS bandwidth control yields the inversion gripe below, moving
    handling quells it.
    
    |========================================================
    |WARNING: possible irq lock inversion dependency detected
    |4.16.7-rt1-rt #2 Tainted: G            E
    |--------------------------------------------------------
    |sirq-hrtimer/0/15 just changed the state of lock:
    | (&cfs_b->lock){+...}, at: [<000000009adb5cf7>] sched_cfs_period_timer+0x28/0x140
    |but this lock was taken by another, HARDIRQ-safe lock in the past: (&rq->lock){-...}
    |and interrupts could create inverse lock ordering between them.
    |other info that might help us debug this:
    | Possible interrupt unsafe locking scenario:
    |       CPU0                    CPU1
    |       ----                    ----
    |  lock(&cfs_b->lock);
    |                               local_irq_disable();
    |                               lock(&rq->lock);
    |                               lock(&cfs_b->lock);
    |  <Interrupt>
    |    lock(&rq->lock);
    
    Cc: stable-rt@vger.kernel.org
    Acked-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
    Signed-off-by: Mike Galbraith <efault@gmx.de>
    Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

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;
 }
 

^ permalink raw reply related	[flat|nested] 10+ messages in thread

* Re: CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer_interrupt occurs
  2019-01-07 12:28     ` Mike Galbraith
@ 2019-01-07 12:52       ` Peter Zijlstra
  2019-01-08  5:30         ` Mike Galbraith
                           ` (3 more replies)
  0 siblings, 4 replies; 10+ messages in thread
From: Peter Zijlstra @ 2019-01-07 12:52 UTC (permalink / raw)
  To: Mike Galbraith
  Cc: Tom Putzeys, mingo, linux-kernel, Sebastian Andrzej Siewior,
	Thomas Gleixner

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;
 }

^ permalink raw reply related	[flat|nested] 10+ messages in thread

* Re: CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer_interrupt occurs
  2019-01-07 12:52       ` Peter Zijlstra
@ 2019-01-08  5:30         ` Mike Galbraith
  2019-01-08  9:06           ` Peter Zijlstra
  2019-01-21 11:37         ` [tip:sched/core] sched/fair: Robustify CFS-bandwidth timer locking tip-bot for Peter Zijlstra
                           ` (2 subsequent siblings)
  3 siblings, 1 reply; 10+ messages in thread
From: Mike Galbraith @ 2019-01-08  5:30 UTC (permalink / raw)
  To: Peter Zijlstra
  Cc: Tom Putzeys, mingo, linux-kernel, Sebastian Andrzej Siewior,
	Thomas Gleixner

On Mon, 2019-01-07 at 13:52 +0100, Peter Zijlstra wrote:
> 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..

I plugged that into 4.19-rt along with revert of hard irq context, and
it (as expected) does the trick.

> ---
> 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;
>  }

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer_interrupt occurs
  2019-01-08  5:30         ` Mike Galbraith
@ 2019-01-08  9:06           ` Peter Zijlstra
  2019-01-08 11:05             ` Sebastian Andrzej Siewior
  0 siblings, 1 reply; 10+ messages in thread
From: Peter Zijlstra @ 2019-01-08  9:06 UTC (permalink / raw)
  To: Mike Galbraith
  Cc: Tom Putzeys, mingo, linux-kernel, Sebastian Andrzej Siewior,
	Thomas Gleixner

On Tue, Jan 08, 2019 at 06:30:59AM +0100, Mike Galbraith wrote:
> On Mon, 2019-01-07 at 13:52 +0100, Peter Zijlstra wrote:
> > 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..
> 
> I plugged that into 4.19-rt along with revert of hard irq context, and
> it (as expected) does the trick.

Much appreciated; I queued it as the below.

---
Subject: sched/fair: Robustify CFS-bandwidth timer locking
From: Peter Zijlstra <peterz@infradead.org>
Date: Mon, 7 Jan 2019 13:52:31 +0100

Traditionally hrtimer callbacks were run with IRQs disabled, but with
the introduction of HRTIMER_MODE_SOFT it is possible they run from
SoftIRQ context, which does _NOT_ have IRQs disabled.

Allow for the CFS bandwidth timers (period_timer and slack_timer) to
be ran from SoftIRQ context; this entails removing the assumption that
IRQs are already disabled from the locking.

While mainline doesn't strictly need this, -RT forces all timers not
explicitly marked with MODE_HARD into MODE_SOFT and trips over this.
And marking these timers as MODE_HARD doesn't make sense as they're
not required for RT operation and can potentially be quite expensive.

Cc: Ingo Molnar <mingo@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Reported-by: Tom Putzeys <tom.putzeys@be.atlascopco.com>
Tested-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20190107125231.GE14122@hirez.programming.kicks-ass.net
---
 kernel/sched/fair.c |   30 ++++++++++++++++--------------
 1 file changed, 16 insertions(+), 14 deletions(-)

--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4566,7 +4566,7 @@ static u64 distribute_cfs_runtime(struct
 		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
 			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
  * 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(str
 	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_r
 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(str
 	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_pe
 {
 	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;
 }

^ permalink raw reply	[flat|nested] 10+ messages in thread

* Re: CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer_interrupt occurs
  2019-01-08  9:06           ` Peter Zijlstra
@ 2019-01-08 11:05             ` Sebastian Andrzej Siewior
  0 siblings, 0 replies; 10+ messages in thread
From: Sebastian Andrzej Siewior @ 2019-01-08 11:05 UTC (permalink / raw)
  To: Peter Zijlstra
  Cc: Mike Galbraith, Tom Putzeys, mingo, linux-kernel, Thomas Gleixner

On 2019-01-08 10:06:25 [+0100], Peter Zijlstra wrote:
> Much appreciated; I queued it as the below.
thank you, queued, too.

Sebastian

^ permalink raw reply	[flat|nested] 10+ messages in thread

* [tip:sched/core] sched/fair: Robustify CFS-bandwidth timer locking
  2019-01-07 12:52       ` Peter Zijlstra
  2019-01-08  5:30         ` Mike Galbraith
@ 2019-01-21 11:37         ` tip-bot for Peter Zijlstra
  2019-01-21 13:53         ` tip-bot for Peter Zijlstra
  2019-01-27 11:36         ` tip-bot for Peter Zijlstra
  3 siblings, 0 replies; 10+ messages in thread
From: tip-bot for Peter Zijlstra @ 2019-01-21 11:37 UTC (permalink / raw)
  To: linux-tip-commits
  Cc: torvalds, hpa, tglx, efault, bigeasy, tom.putzeys, peterz, mingo,
	linux-kernel

Commit-ID:  b733c2d2f2810ec8556d2d711d1b95f491bd7697
Gitweb:     https://git.kernel.org/tip/b733c2d2f2810ec8556d2d711d1b95f491bd7697
Author:     Peter Zijlstra <peterz@infradead.org>
AuthorDate: Mon, 7 Jan 2019 13:52:31 +0100
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Mon, 21 Jan 2019 11:27:55 +0100

sched/fair: Robustify CFS-bandwidth timer locking

Traditionally hrtimer callbacks were run with IRQs disabled, but with
the introduction of HRTIMER_MODE_SOFT it is possible they run from
SoftIRQ context, which does _NOT_ have IRQs disabled.

Allow for the CFS bandwidth timers (period_timer and slack_timer) to
be ran from SoftIRQ context; this entails removing the assumption that
IRQs are already disabled from the locking.

While mainline doesn't strictly need this, -RT forces all timers not
explicitly marked with MODE_HARD into MODE_SOFT and trips over this.
And marking these timers as MODE_HARD doesn't make sense as they're
not required for RT operation and can potentially be quite expensive.

Reported-by: Tom Putzeys <tom.putzeys@be.atlascopco.com>
Tested-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20190107125231.GE14122@hirez.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 kernel/sched/fair.c | 30 ++++++++++++++++--------------
 1 file changed, 16 insertions(+), 14 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 385725eb3bd6..90c7a7bf45d3 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4565,7 +4565,7 @@ 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;
 
@@ -4582,7 +4582,7 @@ 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;
@@ -4598,7 +4598,7 @@ next:
  * 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;
@@ -4640,11 +4640,11 @@ 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);
@@ -4753,17 +4753,18 @@ 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;
 	}
 
@@ -4774,18 +4775,18 @@ 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);
 }
 
 /*
@@ -4863,20 +4864,21 @@ 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;
 }

^ permalink raw reply related	[flat|nested] 10+ messages in thread

* [tip:sched/core] sched/fair: Robustify CFS-bandwidth timer locking
  2019-01-07 12:52       ` Peter Zijlstra
  2019-01-08  5:30         ` Mike Galbraith
  2019-01-21 11:37         ` [tip:sched/core] sched/fair: Robustify CFS-bandwidth timer locking tip-bot for Peter Zijlstra
@ 2019-01-21 13:53         ` tip-bot for Peter Zijlstra
  2019-01-27 11:36         ` tip-bot for Peter Zijlstra
  3 siblings, 0 replies; 10+ messages in thread
From: tip-bot for Peter Zijlstra @ 2019-01-21 13:53 UTC (permalink / raw)
  To: linux-tip-commits
  Cc: torvalds, peterz, tom.putzeys, linux-kernel, efault, hpa, mingo,
	bigeasy, tglx

Commit-ID:  3cd126af79ed5a4d6b06eba63d3349e143a3bd3b
Gitweb:     https://git.kernel.org/tip/3cd126af79ed5a4d6b06eba63d3349e143a3bd3b
Author:     Peter Zijlstra <peterz@infradead.org>
AuthorDate: Mon, 7 Jan 2019 13:52:31 +0100
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Mon, 21 Jan 2019 14:40:28 +0100

sched/fair: Robustify CFS-bandwidth timer locking

Traditionally hrtimer callbacks were run with IRQs disabled, but with
the introduction of HRTIMER_MODE_SOFT it is possible they run from
SoftIRQ context, which does _NOT_ have IRQs disabled.

Allow for the CFS bandwidth timers (period_timer and slack_timer) to
be ran from SoftIRQ context; this entails removing the assumption that
IRQs are already disabled from the locking.

While mainline doesn't strictly need this, -RT forces all timers not
explicitly marked with MODE_HARD into MODE_SOFT and trips over this.
And marking these timers as MODE_HARD doesn't make sense as they're
not required for RT operation and can potentially be quite expensive.

Reported-by: Tom Putzeys <tom.putzeys@be.atlascopco.com>
Tested-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20190107125231.GE14122@hirez.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 kernel/sched/fair.c | 30 ++++++++++++++++--------------
 1 file changed, 16 insertions(+), 14 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index b1374fbddd0d..3b61e19b504a 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4565,7 +4565,7 @@ 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;
 
@@ -4582,7 +4582,7 @@ 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;
@@ -4598,7 +4598,7 @@ next:
  * 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;
@@ -4640,11 +4640,11 @@ 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);
@@ -4753,17 +4753,18 @@ 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;
 	}
 
@@ -4774,18 +4775,18 @@ 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);
 }
 
 /*
@@ -4863,20 +4864,21 @@ 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;
 }

^ permalink raw reply related	[flat|nested] 10+ messages in thread

* [tip:sched/core] sched/fair: Robustify CFS-bandwidth timer locking
  2019-01-07 12:52       ` Peter Zijlstra
                           ` (2 preceding siblings ...)
  2019-01-21 13:53         ` tip-bot for Peter Zijlstra
@ 2019-01-27 11:36         ` tip-bot for Peter Zijlstra
  3 siblings, 0 replies; 10+ messages in thread
From: tip-bot for Peter Zijlstra @ 2019-01-27 11:36 UTC (permalink / raw)
  To: linux-tip-commits
  Cc: tom.putzeys, mingo, efault, bigeasy, torvalds, tglx, peterz,
	linux-kernel, hpa

Commit-ID:  c0ad4aa4d8416a39ad262a2bd68b30acd951bf0e
Gitweb:     https://git.kernel.org/tip/c0ad4aa4d8416a39ad262a2bd68b30acd951bf0e
Author:     Peter Zijlstra <peterz@infradead.org>
AuthorDate: Mon, 7 Jan 2019 13:52:31 +0100
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Sun, 27 Jan 2019 12:29:37 +0100

sched/fair: Robustify CFS-bandwidth timer locking

Traditionally hrtimer callbacks were run with IRQs disabled, but with
the introduction of HRTIMER_MODE_SOFT it is possible they run from
SoftIRQ context, which does _NOT_ have IRQs disabled.

Allow for the CFS bandwidth timers (period_timer and slack_timer) to
be ran from SoftIRQ context; this entails removing the assumption that
IRQs are already disabled from the locking.

While mainline doesn't strictly need this, -RT forces all timers not
explicitly marked with MODE_HARD into MODE_SOFT and trips over this.
And marking these timers as MODE_HARD doesn't make sense as they're
not required for RT operation and can potentially be quite expensive.

Reported-by: Tom Putzeys <tom.putzeys@be.atlascopco.com>
Tested-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20190107125231.GE14122@hirez.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 kernel/sched/fair.c | 30 ++++++++++++++++--------------
 1 file changed, 16 insertions(+), 14 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index b1374fbddd0d..3b61e19b504a 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4565,7 +4565,7 @@ 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;
 
@@ -4582,7 +4582,7 @@ 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;
@@ -4598,7 +4598,7 @@ next:
  * 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;
@@ -4640,11 +4640,11 @@ 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);
@@ -4753,17 +4753,18 @@ 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;
 	}
 
@@ -4774,18 +4775,18 @@ 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);
 }
 
 /*
@@ -4863,20 +4864,21 @@ 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;
 }

^ permalink raw reply related	[flat|nested] 10+ messages in thread

end of thread, other threads:[~2019-01-27 11:36 UTC | newest]

Thread overview: 10+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <AM0PR03MB480425D5999E0D08DAB30204BB8E0@AM0PR03MB4804.eurprd03.prod.outlook.com>
2019-01-04 12:42 ` CFS scheduler: spin_lock usage causes dead lock when smp_apic_timer_interrupt occurs Tom Putzeys
2019-01-07 10:26   ` Peter Zijlstra
2019-01-07 12:28     ` Mike Galbraith
2019-01-07 12:52       ` Peter Zijlstra
2019-01-08  5:30         ` Mike Galbraith
2019-01-08  9:06           ` Peter Zijlstra
2019-01-08 11:05             ` Sebastian Andrzej Siewior
2019-01-21 11:37         ` [tip:sched/core] sched/fair: Robustify CFS-bandwidth timer locking tip-bot for Peter Zijlstra
2019-01-21 13:53         ` tip-bot for Peter Zijlstra
2019-01-27 11:36         ` tip-bot for Peter Zijlstra

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).