clockevents: Introduce CLOCK_EVT_STATE_ONESHOT_STOPPED

When no timers/hrtimers are pending, the expiry time is set to a special value:
'KTIME_MAX'. This normally happens with NO_HZ_{IDLE|FULL} in both LOWRES/HIGHRES modes.

When 'expiry == KTIME_MAX', we either cancel the 'tick-sched' hrtimer (NOHZ_MODE_HIGHRES) or skip reprogramming clockevent device (NOHZ_MODE_LOWRES). But, the clockevent device is already reprogrammed from the tick-handler for next tick.

As the clock event device is programmed in ONESHOT mode it will at least fire one more time (unnecessarily). Timers on few implementations (like arm_arch_timer, etc.) only support PERIODIC mode and their drivers emulate ONESHOT over that. Which means that on these platforms we will get spurious interrupts periodically (at last programmed interval rate, normally tick rate).

In order to avoid spurious interrupts, the clockevent device should be stopped or its interrupts should be masked.

A simple (yet hacky) solution to get this fixed could be: update hrtimer_force_reprogram() to always reprogram clockevent device and update clockevent drivers to STOP generating events (or delay it to max time) when 'expires' is set to KTIME_MAX. But the drawback here is that every clockevent driver has to be hacked for this particular case and its very easy for new ones to miss this.

However, Thomas suggested to add an optional state ONESHOT_STOPPED to solve this
problem: lkml.org/lkml/2014/5/9/508.

This patch adds support for ONESHOT_STOPPED state in clockevents core. It will only be available to drivers that implement the state-specific callbacks instead of the legacy ->set_mode() callback.

Signed-off-by Viresh Kumar <[email protected]>

---
 include/linux/clockchips.h |  7 ++++++-
 kernel/time/clockevents.c  | 14 +++++++++++++-
 kernel/time/timer_list.c   |  6 ++++++
 3 files changed, 25 insertions(+), 2 deletions(-)

include/linux/clockchips.h +6 -1

index e20232c3320a..33ad247f8662 100644
@@ -51,12 +51,15 @@ enum clock_event_mode {
  *		reached from DETACHED or SHUTDOWN.
  * ONESHOT:	Device is programmed to generate event only once. Can be reached
  *		from DETACHED or SHUTDOWN.
+ * ONESHOT_STOPPED: Device was programmed in ONESHOT mode and is temporarily
+ *		    stopped.
  */
 enum clock_event_state {
 	CLOCK_EVT_STATE_DETACHED = 0,
 	CLOCK_EVT_STATE_SHUTDOWN,
 	CLOCK_EVT_STATE_PERIODIC,
 	CLOCK_EVT_STATE_ONESHOT,
+	CLOCK_EVT_STATE_ONESHOT_STOPPED,
 };
 
 /*

@@ -103,6 +106,7 @@ enum clock_event_state {
  * @set_mode:		legacy set mode function, only for modes <= CLOCK_EVT_MODE_RESUME.
  * @set_state_periodic:	switch state to periodic, if !set_mode
  * @set_state_oneshot:	switch state to oneshot, if !set_mode
+ * @set_state_oneshot_stopped: switch state to oneshot_stopped, if !set_mode
  * @set_state_shutdown:	switch state to shutdown, if !set_mode
  * @tick_resume:	resume clkevt device, if !set_mode
  * @broadcast:		function to broadcast events

@@ -136,12 +140,13 @@ struct clock_event_device {
 	 * State transition callback(s): Only one of the two groups should be
 	 * defined:
 	 * - set_mode(), only for modes <= CLOCK_EVT_MODE_RESUME.
-	 * - set_state_{shutdown|periodic|oneshot}(), tick_resume().
+	 * - set_state_{shutdown|periodic|oneshot|oneshot_stopped}(), tick_resume().
 	 */
 	void			(*set_mode)(enum clock_event_mode mode,
 					    struct clock_event_device *);
 	int			(*set_state_periodic)(struct clock_event_device *);
 	int			(*set_state_oneshot)(struct clock_event_device *);
+	int			(*set_state_oneshot_stopped)(struct clock_event_device *);
 	int			(*set_state_shutdown)(struct clock_event_device *);
 	int			(*tick_resume)(struct clock_event_device *);
 

kernel/time/clockevents.c +13 -1

index 73689df1e4b8..04f6c3433f8e 100644
@@ -138,6 +138,17 @@ static int __clockevents_set_state(struct clock_event_device *dev,
 			return -ENOSYS;
 		return dev->set_state_oneshot(dev);
 
+	case CLOCK_EVT_STATE_ONESHOT_STOPPED:
+		/* Core internal bug */
+		if (WARN_ONCE(dev->state != CLOCK_EVT_STATE_ONESHOT,
+			      "Current state: %d\n", dev->state))
+			return -EINVAL;
+
+		if (dev->set_state_oneshot_stopped)
+			return dev->set_state_oneshot_stopped(dev);
+		else
+			return -ENOSYS;
+
 	default:
 		return -ENOSYS;
 	}

@@ -449,7 +460,8 @@ static int clockevents_sanity_check(struct clock_event_device *dev)
 	if (dev->set_mode) {
 		/* We shouldn't be supporting new modes now */
 		WARN_ON(dev->set_state_periodic || dev->set_state_oneshot ||
-			dev->set_state_shutdown || dev->tick_resume);
+			dev->set_state_shutdown || dev->tick_resume ||
+			dev->set_state_oneshot_stopped);
 
 		BUG_ON(dev->mode != CLOCK_EVT_MODE_UNUSED);
 		return 0;

kernel/time/timer_list.c +0 -0

index 05aa5590106a..98d77969ba85 100644
@@ -251,6 +251,12 @@ print_tickdevice(struct seq_file *m, struct tick_device *td, int cpu)
 			SEQ_printf(m, "\n");
 		}

When no timers/hrtimers are pending, the expiry time is set to a special value:
'KTIME_MAX'. This normally happens with NO_HZ_{IDLE|FULL} in both LOWRES/HIGHRES modes.

When 'expiry == KTIME_MAX', we either cancel the 'tick-sched' hrtimer (NOHZ_MODE_HIGHRES) or skip reprogramming clockevent device (NOHZ_MODE_LOWRES). But, the clockevent device is already reprogrammed from the tick-handler for next tick.

As the clock event device is programmed in ONESHOT mode it will at least fire one more time (unnecessarily). Timers on few implementations (like arm_arch_timer, etc.) only support PERIODIC mode and their drivers emulate ONESHOT over that. Which means that on these platforms we will get spurious interrupts periodically (at last programmed interval rate, normally tick rate).

In order to avoid spurious interrupts, the clockevent device should be stopped or its interrupts should be masked.

A simple (yet hacky) solution to get this fixed could be: update hrtimer_force_reprogram() to always reprogram clockevent device and update clockevent drivers to STOP generating events (or delay it to max time) when 'expires' is set to KTIME_MAX. But the drawback here is that every clockevent driver has to be hacked for this particular case and its very easy for new ones to miss this.

However, Thomas suggested to add an optional state ONESHOT_STOPPED to solve this
problem: lkml.org/lkml/2014/5/9/508.

This patch adds support for ONESHOT_STOPPED state in clockevents core. It will only be available to drivers that implement the state-specific callbacks instead of the legacy ->set_mode() callback.

Signed-off-by Viresh Kumar <[email protected]>

Next commit would update clockevents core to switch clockevent devices to ONESHOT_STOPPED state to avoid getting spurious interrupts on a tickless CPU.

Before programming the clockevent device for next event, we must switch its state to ONESHOT.

Switch state back to ONESHOT from ONESHOT_STOPPED at following places:

1.) NOHZ_MODE_LOWRES Mode

Timers & hrtimers are dependent on tick for their working in this mode and the only place from where clockevent device is programmed is the tick-code.

Two routines can restart ticks in LOWRES mode: tick_nohz_restart() and tick_nohz_stop_sched_tick().

2.) NOHZ_MODE_HIGHRES Mode

Tick & timers are dependent on hrtimers for their working in this mode and the only place from where clockevent device is programmed is the hrtimer-code.

Only hrtimer_reprogram() is responsible for programming the clockevent device for next event, if the clockevent device is stopped earlier. And updating that alone is sufficient here.

To make sure we haven't missed any corner case, add a WARN() for the case where we try to reprogram clockevent device while we aren't configured in ONESHOT_STOPPED state.

Signed-off-by Viresh Kumar <[email protected]>

---
 kernel/time/clockevents.c |  4 ++++
 kernel/time/hrtimer.c     |  5 +++++
 kernel/time/tick-sched.c  | 14 +++++++++++++-
 3 files changed, 22 insertions(+), 1 deletion(-)

kernel/time/clockevents.c +4 -0

index 04f6c3433f8e..e95896fd21fd 100644
@@ -335,6 +335,10 @@ int clockevents_program_event(struct clock_event_device *dev, ktime_t expires,
 	if (dev->state == CLOCK_EVT_STATE_SHUTDOWN)
 		return 0;
 
+	/* We must be in ONESHOT state here */
+	WARN_ONCE(dev->state != CLOCK_EVT_STATE_ONESHOT, "Current state: %d\n",
+		  dev->state);
+
 	/* Shortcut for clockevent devices that can deal with ktime. */
 	if (dev->features & CLOCK_EVT_FEAT_KTIME)
 		return dev->set_next_ktime(expires, dev);

kernel/time/hrtimer.c +5 -0

index bee0c1f78091..045ba7e2be6c 100644
@@ -566,6 +566,7 @@ static int hrtimer_reprogram(struct hrtimer *timer,
 {
 	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
 	ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
+	struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev);
 	int res;
 
 	WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);

@@ -610,6 +611,10 @@ static int hrtimer_reprogram(struct hrtimer *timer,
 	if (cpu_base->hang_detected)
 		return 0;
 
+	/* Switchback to ONESHOT state */
+	if (unlikely(dev->state == CLOCK_EVT_STATE_ONESHOT_STOPPED))
+		clockevents_set_state(dev, CLOCK_EVT_STATE_ONESHOT);
+
 	/*
 	 * Clockevents returns -ETIME, when the event was in the past.
 	 */

kernel/time/tick-sched.c +9 -1

index a4c4edac4528..47c04edd07df 100644
@@ -694,8 +694,14 @@ static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts,
 			/* Check, if the timer was already in the past */
 			if (hrtimer_active(&ts->sched_timer))
 				goto out;
-		} else if (!tick_program_event(expires, 0))
+		} else {
+			/* Switchback to ONESHOT state */
+			if (unlikely(dev->state == CLOCK_EVT_STATE_ONESHOT_STOPPED))
+				clockevents_set_state(dev, CLOCK_EVT_STATE_ONESHOT);
+
+			if (!tick_program_event(expires, 0))
 				goto out;
+		}
 		/*
 		 * We are past the event already. So we crossed a
 		 * jiffie boundary. Update jiffies and raise the

@@ -873,6 +879,8 @@ ktime_t tick_nohz_get_sleep_length(void)
 
 static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
 {
+	struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev);
+
 	hrtimer_cancel(&ts->sched_timer);
 	hrtimer_set_expires(&ts->sched_timer, ts->last_tick);
 

Only hrtimer_reprogram() is responsible for programming the clockevent device for next event, if the clockevent device is stopped earlier. And updating that alone is sufficient here.

+++ b/kernel/time/hrtimer.c

kernel/time/hrtimer.c +1 -0

@@ -566,6 +566,7 @@ static int hrtimer_reprogram(struct hrtimer *timer,
 {
 	struct hrtimer_cpu_base *cpu_base = this_cpu_ptr(&hrtimer_bases);
 	ktime_t expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
+	struct clock_event_device *dev = __this_cpu_read(tick_cpu_device.evtdev);
 	int res;
 
 	WARN_ON_ONCE(hrtimer_get_expires_tv64(timer) < 0);

kernel/time/hrtimer.c static int hrtimer_reprogram(struct hrtimer *timer, · line 620

Quoted 11 lines · Viresh Kumar

+++ b/kernel/time/hrtimer.c

kernel/time/hrtimer.c +4 -0

@@ -566,6 +566,7 @@ static int hrtimer_reprogram(struct hrtimer *timer,
 {
 
+     /* Switchback to ONESHOT state */
+     if (unlikely(dev->state == CLOCK_EVT_STATE_ONESHOT_STOPPED))
+             clockevents_set_state(dev, CLOCK_EVT_STATE_ONESHOT);
+
      /*
       * Clockevents returns -ETIME, when the event was in the past.
       */

Should we not do this in tick_program_event() instead? Note that there are a few more places that call that, the two in the hrtimer_interrupt() should be safe because if we're handling the interrupt its cannot be stopped anyhow.

hrtimer_force_reprogram() seems to need the annotation regardless.

Furthermore, by putting it in tick_program_event() you also don't need to fixup tick_nohz_restart().

Or am I completely missing something?

Quoted 1 line

Or am I completely missing something?

So yes, if we would have done that in tick_program_event(), it would have been a single place for doing this change..

But, when Thomas ranted [1] at me on this earlier, he said:

"
No, we are not doing a state change behind the scene and a magic restore.

2B) Implement the ONESHOT_STOPPED logic and make sure all of the core
code is aware of it.
"

Quoted 7 lines · Viresh Kumar

Quoted 1 line

Or am I completely missing something?

So yes, if we would have done that in tick_program_event(), it would have been a single place for doing this change..

But, when Thomas ranted [1] at me on this earlier, he said:

Quoted 8 lines

"
No, we are not doing a state change behind the scene and a magic restore.

2B) Implement the ONESHOT_STOPPED logic and make sure all of the core
code is aware of it.
"

lkml.org didn't work for me, alternative link:

http://marc.info/?l=linux-kernel&m=139966616803683&w=2

So I've read that (several times) and I thing Thomas meant something else.

So I think he disliked what you did to the clockevent layer, not so much you touching tick_program_event(). But the last_state thing (which was broken), and you imposing the SHUTDOWN policy for everybody.

But with the optional ONESHOT_STOPPED state both those are gone, and we'd end up with the much simpler patch below.

Further note that tick-oneshot is the natural place to do this, that is the glue layer between the (oneshot) clockevents stuff and the timer stuff. Pulling clockevents into hrtimers feels wrong.

Ingo, do you agree with that after reading Thomas' email?

Clockevent device can now be stopped by switching to ONESHOT_STOPPED state, to avoid getting spurious interrupts on a tickless CPU.

Switch state to ONESHOT_STOPPED at following places:

1.) NOHZ_MODE_LOWRES Mode

Timers & hrtimers are dependent on tick for their working in this mode and the only place from where clockevent device is programmed is the tick-code.

In LOWRES mode we skip reprogramming the clockevent device in tick_nohz_stop_sched_tick() if expires == KTIME_MAX. In addition to that we must also switch the clockevent device to ONESHOT_STOPPED state to avoid all spurious interrupts that may follow.

2.) NOHZ_MODE_HIGHRES Mode

Tick & timers are dependent on hrtimers for their working in this mode and the only place from where clockevent device is programmed is the hrtimer-code.

There are two places here from which we reprogram the clockevent device or skip reprogramming it on expires == KTIME_MAX.

Signed-off-by Viresh Kumar <[email protected]>

---
 kernel/time/hrtimer.c    | 46 ++++++++++++++++++++++++++++++++++++++++++----
 kernel/time/tick-sched.c |  3 +++
 2 files changed, 45 insertions(+), 4 deletions(-)

kernel/time/hrtimer.c +42 -4

index 045ba7e2be6c..89d4b593dfc0 100644
@@ -543,8 +543,19 @@ hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
 	if (cpu_base->hang_detected)
 		return;
 
-	if (cpu_base->expires_next.tv64 != KTIME_MAX)
+	if (cpu_base->expires_next.tv64 != KTIME_MAX) {
 		tick_program_event(cpu_base->expires_next, 1);
+	} else {
+		struct clock_event_device *dev =
+			__this_cpu_read(tick_cpu_device.evtdev);
+		/*
+		 * Don't need clockevent device anymore, stop it.
+		 *
+		 * We reach here only for NOHZ_MODE_HIGHRES mode and we are
+		 * guaranteed that no timers/hrtimers are enqueued on this cpu.
+		 */
+		clockevents_set_state(dev, CLOCK_EVT_STATE_ONESHOT_STOPPED);
+	}
 }
 
 /*

@@ -1312,9 +1323,36 @@ void hrtimer_interrupt(struct clock_event_device *dev)
 	cpu_base->in_hrtirq = 0;
 	raw_spin_unlock(&cpu_base->lock);
 
-	/* Reprogramming necessary ? */
-	if (expires_next.tv64 == KTIME_MAX ||
-	    !tick_program_event(expires_next, 0)) {
+	if (expires_next.tv64 == KTIME_MAX) {
+		struct clock_event_device *dev =
+			__this_cpu_read(tick_cpu_device.evtdev);
+
+		cpu_base->hang_detected = 0;
+
+		/*
+		 * Don't need clockevent device anymore, stop it.
+		 *
+		 * We reach here only for NOHZ_MODE_HIGHRES mode and we are
+		 * guaranteed that no timers/hrtimers are enqueued on this cpu.
+		 *
+		 * Most of the scenarios will be covered by similar code
+		 * present in hrtimer_force_reprogram(), as we always try to
+		 * evaluate tick requirement on idle/irq exit and cancel
+		 * tick-sched hrtimer when tick isn't required anymore.
+		 *
+		 * It is required here as well as a special case.
+		 *
+		 * Last hrtimer fires on a tickless CPU and doesn't rearm
+		 * itself. tick_nohz_irq_exit() reevaluates next event and it
+		 * gets expires == KTIME_MAX. Because tick was already stopped,
+		 * and last expires == new_expires, we return early. And the
+		 * clockevent device is never stopped.
+		 */
+		clockevents_set_state(dev, CLOCK_EVT_STATE_ONESHOT_STOPPED);
+		return;
+	}
+
+	if (!tick_program_event(expires_next, 0)) {
 		cpu_base->hang_detected = 0;
 		return;
 	}

kernel/time/tick-sched.c +0 -0

index 47c04edd07df..ff271a26fa4d 100644

 kernel/time/hrtimer.c    | 46 ++++++++++++++++++++++++++++++++++++++++++----
 kernel/time/tick-sched.c |  3 +++
 2 files changed, 45 insertions(+), 4 deletions(-)

kernel/time/hrtimer.c +17 -4

index 045ba7e2be6c..89d4b593dfc0 100644
@@ -543,8 +543,19 @@ hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base, int skip_equal)
 	if (cpu_base->hang_detected)
 		return;
 
-	if (cpu_base->expires_next.tv64 != KTIME_MAX)
+	if (cpu_base->expires_next.tv64 != KTIME_MAX) {
 		tick_program_event(cpu_base->expires_next, 1);
+	} else {
+		struct clock_event_device *dev =
+			__this_cpu_read(tick_cpu_device.evtdev);
 
+		clockevents_set_state(dev, CLOCK_EVT_STATE_ONESHOT_STOPPED);
+	}
 }
 
 /*

@@ -1312,9 +1323,36 @@ void hrtimer_interrupt(struct clock_event_device *dev)
 	cpu_base->in_hrtirq = 0;
 	raw_spin_unlock(&cpu_base->lock);
 
-	/* Reprogramming necessary ? */
-	if (expires_next.tv64 == KTIME_MAX ||
-	    !tick_program_event(expires_next, 0)) {
+	if (expires_next.tv64 == KTIME_MAX) {
+		struct clock_event_device *dev =
+			__this_cpu_read(tick_cpu_device.evtdev);
+
+		cpu_base->hang_detected = 0;
+
+		clockevents_set_state(dev, CLOCK_EVT_STATE_ONESHOT_STOPPED);
+		return;
+	}
+
+	if (!tick_program_event(expires_next, 0)) {
 		cpu_base->hang_detected = 0;
 		return;
 	}

kernel/time/tick-sched.c +0 -0

index 47c04edd07df..ff271a26fa4d 100644

Should we teach tick_program_event() about KTIME_MAX instead of adding it to its callers?