Skip to contentSkip to conversation

tick/sched: Forward timer even in nohz mode

PATCH In review Scott Wood 3 participants 8 messages Dec 16, 2019 – Jan 8, 2020

Subsystems (2)
lore
Share & tools
lore

Scott Wood

author
#0 lore raw

PATCH 1/4 tick/sched: Forward timer even in nohz mode

Currently when exiting nohz, the expiry will be forwarded as if we had just run the timer. If we re-enter nohz before this new expiry, and exit after, this forwarding will happen again. If this load pattern recurs the tick can be indefinitely postponed.

To avoid this, use last_tick as-is rather than calling hrtimer_forward(). However, in some cases the tick *will* have just run (despite being "stopped"), and leading to double timer execution.

To avoid that, forward the timer after every tick (regardless of nohz status) and keep last_tick up-to-date. During nohz, last_tick will reflect what the expiry would have been if not in nohz mode.

---
 kernel/time/tick-sched.c | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)
kernel/time/tick-sched.c +5 -6
index 8b192e67aabc..8936b604dd6c 100644
@@ -642,9 +642,6 @@ static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
 	hrtimer_cancel(&ts->sched_timer);
 	hrtimer_set_expires(&ts->sched_timer, ts->last_tick);
 
-	/* Forward the time to expire in the future */
-	hrtimer_forward(&ts->sched_timer, now, tick_period);
-
 	if (ts->nohz_mode == NOHZ_MODE_HIGHRES) {
 		hrtimer_start_expires(&ts->sched_timer,
 				      HRTIMER_MODE_ABS_PINNED_HARD);
@@ -1207,12 +1204,13 @@ static void tick_nohz_handler(struct clock_event_device *dev)
 
 	tick_sched_do_timer(ts, now);
 	tick_sched_handle(ts, regs);
+	hrtimer_forward(&ts->sched_timer, now, tick_period);
+	ts->last_tick = hrtimer_get_expires(&ts->sched_timer);
 
 	/* No need to reprogram if we are running tickless  */
 	if (unlikely(ts->tick_stopped))
 		return;
 
-	hrtimer_forward(&ts->sched_timer, now, tick_period);
 	tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1);
 }
 
@@ -1311,12 +1309,13 @@ static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer)
 	else
 		ts->next_tick = 0;
 
+	hrtimer_forward(timer, now, tick_period);
+	ts->last_tick = hrtimer_get_expires(&ts->sched_timer);
+
 	/* No need to reprogram if we are in idle or full dynticks mode */
 	if (unlikely(ts->tick_stopped))
 		return HRTIMER_NORESTART;
 
-	hrtimer_forward(timer, now, tick_period);
-
 	return HRTIMER_RESTART;
 }
 
Signature · 1 line
1.8.3.1
author
#1 lore raw

PATCH 2/4 tick/sched: Set last_tick in init paths

This eliminates the need to save last_tick on nohz entry.

---
 kernel/time/tick-sched.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
kernel/time/tick-sched.c +2 -1
index 8936b604dd6c..59e663e240fc 100644
@@ -794,7 +794,6 @@ static void tick_nohz_stop_tick(struct tick_sched *ts, int cpu)
 		calc_load_nohz_start();
 		quiet_vmstat();
 
-		ts->last_tick = hrtimer_get_expires(&ts->sched_timer);
 		ts->tick_stopped = 1;
 		trace_tick_stop(1, TICK_DEP_MASK_NONE);
 	}
@@ -1248,6 +1247,7 @@ static void tick_nohz_switch_to_nohz(void)
 
 	hrtimer_set_expires(&ts->sched_timer, next);
 	hrtimer_forward_now(&ts->sched_timer, tick_period);
+	ts->last_tick = hrtimer_get_expires(&ts->sched_timer);
 	tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1);
 	tick_nohz_activate(ts, NOHZ_MODE_LOWRES);
 }
@@ -1355,6 +1355,7 @@ void tick_setup_sched_timer(void)
 	}
 
 	hrtimer_forward(&ts->sched_timer, now, tick_period);
+	ts->last_tick = hrtimer_get_expires(&ts->sched_timer);
 	hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED_HARD);
 	tick_nohz_activate(ts, NOHZ_MODE_HIGHRES);
 }
Signature · 1 line
1.8.3.1
author
#2 lore raw

PATCH 3/4 sched/core: Don't skip remote tick for idle cpus

In reply to Scott Wood

This will be used in the next patch to get a loadavg update from nohz cpus. The delta check is skipped because idle_sched_class doesn't update se.exec_start.

---
 kernel/sched/core.c | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)
kernel/sched/core.c +10 -8
index 90e4b00ace89..dfb8ea801700 100644
@@ -3668,22 +3668,24 @@ static void sched_tick_remote(struct work_struct *work)
 	 * statistics and checks timeslices in a time-independent way, regardless
 	 * of when exactly it is running.
 	 */
-	if (idle_cpu(cpu) || !tick_nohz_tick_stopped_cpu(cpu))
+	if (!tick_nohz_tick_stopped_cpu(cpu))
 		goto out_requeue;
 
 	rq_lock_irq(rq, &rf);
 	curr = rq->curr;
-	if (is_idle_task(curr) || cpu_is_offline(cpu))
+	if (cpu_is_offline(cpu))
 		goto out_unlock;
 
 	update_rq_clock(rq);
-	delta = rq_clock_task(rq) - curr->se.exec_start;
 
-	/*
-	 * Make sure the next tick runs within a reasonable
-	 * amount of time.
-	 */
-	WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3);
+	if (!is_idle_task(curr)) {
+		/*
+		 * Make sure the next tick runs within a reasonable
+		 * amount of time.
+		 */
+		delta = rq_clock_task(rq) - curr->se.exec_start;
+		WARN_ON_ONCE(delta > (u64)NSEC_PER_SEC * 3);
+	}
 	curr->sched_class->task_tick(rq, curr, 0);
 
 out_unlock:
Signature · 1 line
1.8.3.1
author
#3 lore raw

PATCH 4/4 timers/nohz: Update nohz load in remote tick

In reply to Scott Wood

From: Peter Zijlstra <[email protected]>

The way loadavg is tracked during nohz only pays attention to the load upon entering nohz. This can be particularly noticeable if full nohz is entered while non-idle, and then the cpu goes idle and stays that way for a long time.

Use the remote tick to ensure that full nohz cpus report their deltas within a reasonable time.

[swood: added changelog and removed recheck of stopped tick]

---
This patch was provided by Peter in an unsigned email reply -- Peter, can I get a signoff?

---
 include/linux/sched/nohz.h |  2 ++
 kernel/sched/core.c        |  4 +++-
 kernel/sched/loadavg.c     | 33 +++++++++++++++++++++++----------
 3 files changed, 28 insertions(+), 11 deletions(-)
include/linux/sched/nohz.h +2 -0
index 1abe91ff6e4a..6d67e9a5af6b 100644
@@ -15,9 +15,11 @@ static inline void nohz_balance_enter_idle(int cpu) { }
 
 #ifdef CONFIG_NO_HZ_COMMON
 void calc_load_nohz_start(void);
+void calc_load_nohz_remote(struct rq *rq);
 void calc_load_nohz_stop(void);
 #else
 static inline void calc_load_nohz_start(void) { }
+static inline void calc_load_nohz_remote(struct rq *rq) { }
 static inline void calc_load_nohz_stop(void) { }
 #endif /* CONFIG_NO_HZ_COMMON */
 
kernel/sched/core.c +3 -1
index dfb8ea801700..2e4a505e48af 100644
@@ -3676,6 +3676,7 @@ static void sched_tick_remote(struct work_struct *work)
 	if (cpu_is_offline(cpu))
 		goto out_unlock;
 
+	curr = rq->curr;
 	update_rq_clock(rq);
 
 	if (!is_idle_task(curr)) {
@@ -3688,10 +3689,11 @@ static void sched_tick_remote(struct work_struct *work)
 	}
 	curr->sched_class->task_tick(rq, curr, 0);
 
+	calc_load_nohz_remote(rq);
 out_unlock:
 	rq_unlock_irq(rq, &rf);
-
 out_requeue:
+
 	/*
 	 * Run the remote tick once per second (1Hz). This arbitrary
 	 * frequency is large enough to avoid overload but short enough
kernel/sched/loadavg.c +23 -10
index 28a516575c18..de22da666ac7 100644
@@ -231,16 +231,11 @@ static inline int calc_load_read_idx(void)
 	return calc_load_idx & 1;
 }
 
-void calc_load_nohz_start(void)
+static void calc_load_nohz_fold(struct rq *rq)
 {
-	struct rq *this_rq = this_rq();
 	long delta;
 
-	/*
-	 * We're going into NO_HZ mode, if there's any pending delta, fold it
-	 * into the pending NO_HZ delta.
-	 */
-	delta = calc_load_fold_active(this_rq, 0);
+	delta = calc_load_fold_active(rq, 0);
 	if (delta) {
 		int idx = calc_load_write_idx();
 
@@ -248,6 +243,24 @@ void calc_load_nohz_start(void)
 	}
 }
 
+void calc_load_nohz_start(void)
+{
+	/*
+	 * We're going into NO_HZ mode, if there's any pending delta, fold it
+	 * into the pending NO_HZ delta.
+	 */
+	calc_load_nohz_fold(this_rq());
+}
+
+/*
+ * Keep track of the load for NOHZ_FULL, must be called between
+ * calc_load_nohz_{start,stop}().
+ */
+void calc_load_nohz_remote(struct rq *rq)
+{
+	calc_load_nohz_fold(rq);
+}
+
 void calc_load_nohz_stop(void)
 {
 	struct rq *this_rq = this_rq();
@@ -268,7 +281,7 @@ void calc_load_nohz_stop(void)
 		this_rq->calc_load_update += LOAD_FREQ;
 }
 
-static long calc_load_nohz_fold(void)
+static long calc_load_nohz_read(void)
 {
 	int idx = calc_load_read_idx();
 	long delta = 0;
@@ -323,7 +336,7 @@ static void calc_global_nohz(void)
 }
 #else /* !CONFIG_NO_HZ_COMMON */
 
-static inline long calc_load_nohz_fold(void) { return 0; }
+static inline long calc_load_nohz_read(void) { return 0; }
 static inline void calc_global_nohz(void) { }
 
 #endif /* CONFIG_NO_HZ_COMMON */
@@ -346,7 +359,7 @@ void calc_global_load(unsigned long ticks)
 	/*
 	 * Fold the 'old' NO_HZ-delta to include all NO_HZ CPUs.
 	 */
-	delta = calc_load_nohz_fold();
+	delta = calc_load_nohz_read();
 	if (delta)
 		atomic_long_add(delta, &calc_load_tasks);
 
Signature · 1 line
1.8.3.1
maintainer
#4 lore raw

Re: [PATCH 4/4] timers/nohz: Update nohz load in remote tick

Quoted 7 lines · Scott Wood

From: Peter Zijlstra <[email protected]>

The way loadavg is tracked during nohz only pays attention to the load upon entering nohz. This can be particularly noticeable if full nohz is entered while non-idle, and then the cpu goes idle and stays that way for a long time.

Quoted 8 lines

Use the remote tick to ensure that full nohz cpus report their deltas within a reasonable time.

[swood: added changelog and removed recheck of stopped tick]

---
This patch was provided by Peter in an unsigned email reply -- Peter, can I get a signoff?

Here goes, glad it works :-)

author
#5 lore raw
In reply to Scott Wood
Quoted 9 lines · Scott Wood

Currently when exiting nohz, the expiry will be forwarded as if we had just run the timer. If we re-enter nohz before this new expiry, and exit after, this forwarding will happen again. If this load pattern recurs the tick can be indefinitely postponed.

To avoid this, use last_tick as-is rather than calling hrtimer_forward(). However, in some cases the tick *will* have just run (despite being "stopped"), and leading to double timer execution.

Quoted 8 lines

To avoid that, forward the timer after every tick (regardless of nohz status) and keep last_tick up-to-date. During nohz, last_tick will reflect what the expiry would have been if not in nohz mode.

---
 kernel/time/tick-sched.c | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

Any comments on these patches?

Thanks,
Scott

#6 lore raw
In reply to Scott Wood
Quoted 4 lines · Scott Wood

Currently when exiting nohz, the expiry will be forwarded as if we had just run the timer. If we re-enter nohz before this new expiry, and exit after, this forwarding will happen again. If this load pattern recurs the tick can be indefinitely postponed.

I must be missing something but I don't see why that would be a problem. Indeed the tick can be indefinitely postponed but that's as long as it's not needed. As soon as it's needed (timer callback expired, RCU, ...), the tick will be retained and it will eventually fire.

kernel/time/tick-sched.c static void tick_nohz_restart(struct tick_sched *ts, ktime_t now) · line 647

Quoted 7 lines
kernel/time/tick-sched.c +0 -3
@@ -642,9 +642,6 @@ static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
 	hrtimer_cancel(&ts->sched_timer);
 	hrtimer_set_expires(&ts->sched_timer, ts->last_tick);
 
-	/* Forward the time to expire in the future */
-	hrtimer_forward(&ts->sched_timer, now, tick_period);
-

By doing that, you may program a past tick and thus add a useless interrupt at each idle exit.

Thanks.

author
#7 lore raw
Quoted 10 lines · Frederic Weisbecker
Quoted 4 lines · Scott Wood

Currently when exiting nohz, the expiry will be forwarded as if we had just run the timer. If we re-enter nohz before this new expiry, and exit after, this forwarding will happen again. If this load pattern recurs the tick can be indefinitely postponed.

I must be missing something but I don't see why that would be a problem. Indeed the tick can be indefinitely postponed but that's as long as it's not needed. As soon as it's needed (timer callback expired, RCU, ...), the tick will be retained and it will eventually fire.

My main concern was the loadavg recording, which does not prevent entering nohz. However, if we're going into and out of nohz then the call to calc_load_nohz_start() in tick_nohz_stop_tick() should take care of it, so I guess we only need patches 3 and 4, if there's nothing else in the tick that doesn't either prevent nohz or get handled during the transition (psi_task_tick?).

-Scott