[PATCH 00/15 v4] tick/sched: Refactor idle cputime accounting

[PATCH 00/15 v4] tick/sched: Refactor idle cputime accounting

[Frederic Weisbecker]

Hi,

After the issue reported here:

        https://lore.kernel.org/all/20251210083135.3993562-1-jackzxcui1989@163.com/

It occurs that the idle cputime accounting is a big mess that
accumulates within two concurrent statistics, each having their own
shortcomings:

* The accounting for online CPUs which is based on the delta between
  tick_nohz_start_idle() and tick_nohz_stop_idle().

  Pros:
       - Works when the tick is off

       - Has nsecs granularity

  Cons:
       - Account idle steal time but doesn't substract it from idle
         cputime.

       - Assumes CONFIG_IRQ_TIME_ACCOUNTING by not accounting IRQs but
         the IRQ time is simply ignored when
         CONFIG_IRQ_TIME_ACCOUNTING=n

       - The windows between 1) idle task scheduling and the first call
         to tick_nohz_start_idle() and 2) idle task between the last
         tick_nohz_stop_idle() and the rest of the idle time are
         blindspots wrt. cputime accounting (though mostly insignificant
         amount)

       - Relies on private fields outside of kernel stats, with specific
         accessors.

* The accounting for offline CPUs which is based on ticks and the
  jiffies delta during which the tick was stopped.

  Pros:
       - Handles steal time correctly

       - Handle CONFIG_IRQ_TIME_ACCOUNTING=y and
         CONFIG_IRQ_TIME_ACCOUNTING=n correctly.

       - Handles the whole idle task

       - Accounts directly to kernel stats, without midlayer accumulator.

   Cons:
       - Doesn't elapse when the tick is off, which doesn't make it
         suitable for online CPUs.

       - Has TICK_NSEC granularity (jiffies)

       - Needs to track the dyntick-idle ticks that were accounted and
         substract them from the total jiffies time spent while the tick
         was stopped. This is an ugly workaround.

Having two different accounting for a single context is not the only
problem: since those accountings are of different natures, it is
possible to observe the global idle time going backward after a CPU goes
offline, as reported by Xin Zhao.

Clean up the situation with introducing a hybrid approach that stays
coherent, fixes the backward jumps and works for both online and offline
CPUs:

* Tick based or native vtime accounting operate before the tick is
  stopped and resumes once the tick is restarted.

* When the idle loop starts, switch to dynticks-idle accounting as is
  done currently, except that the statistics accumulate directly to the
  relevant kernel stat fields.

* Private dyntick cputime accounting fields are removed.

* Works on both online and offline case.

* Move most of the relevant code to the common sched/cputime subsystem

* Handle CONFIG_IRQ_TIME_ACCOUNTING=n correctly such that the
  dynticks-idle accounting still elapses while on IRQs.

* Correctly substract idle steal cputime from idle time

Changes since v3 (among which a lot of relevant reviews from Sashiko):

- Add new tags

- Rebase on latest -rc1

- Add "tick/sched: Fix TOCTOU in nohz idle time fetch" (Sashiko)

- Fix buggy state refetch in kcpustat_cpu_fetch_vtime() (Sashiko)

- Fix build issue on powerpc (Christophe Leroy)

- Fix s390 lost steal time occuring on idle IRQs (call vtime_flush() on
  vtime_account_hardirq() and vtime_account_softirq()) (Sashiko)

- Fix build issue on s390

- Fix uninitialized idle_sleeptime_seq (Sashiko)

- Fix irqtime being disabled or enabled in the middle of an idle IRQ
  (Sashiko)
  
- Fix tick restart and then restop in the same idle loop (Sashiko)

- Fix "sched/cputime: Handle idle irqtime gracefully" changelog (Sashiko)

- Fix idle steal time substracted from the wrong index between idle and
  iowait kcpustat. (Sashiko)

git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git
	timers/core-v4

HEAD: e64ba052ce04e363ff76d3cb8bedc5f812188acb
Thanks,
	Frederic
---

Frederic Weisbecker (15):
      tick/sched: Fix TOCTOU in nohz idle time fetch
      sched/idle: Handle offlining first in idle loop
      sched/cputime: Remove superfluous and error prone kcpustat_field() parameter
      sched/cputime: Correctly support generic vtime idle time
      powerpc/time: Prepare to stop elapsing in dynticks-idle
      s390/time: Prepare to stop elapsing in dynticks-idle
      tick/sched: Unify idle cputime accounting
      tick/sched: Remove nohz disabled special case in cputime fetch
      tick/sched: Move dyntick-idle cputime accounting to cputime code
      tick/sched: Remove unused fields
      tick/sched: Account tickless idle cputime only when tick is stopped
      tick/sched: Consolidate idle time fetching APIs
      sched/cputime: Provide get_cpu_[idle|iowait]_time_us() off-case
      sched/cputime: Handle idle irqtime gracefully
      sched/cputime: Handle dyntick-idle steal time correctly

 arch/powerpc/kernel/time.c         |  41 +++++
 arch/s390/include/asm/idle.h       |   2 +
 arch/s390/kernel/idle.c            |   5 +-
 arch/s390/kernel/vtime.c           |  75 ++++++++-
 drivers/cpufreq/cpufreq.c          |  29 +---
 drivers/cpufreq/cpufreq_governor.c |   6 +-
 drivers/macintosh/rack-meter.c     |   2 +-
 fs/proc/stat.c                     |  40 +----
 fs/proc/uptime.c                   |   8 +-
 include/linux/kernel_stat.h        |  76 +++++++--
 include/linux/tick.h               |   4 -
 include/linux/vtime.h              |  22 ++-
 kernel/rcu/tree.c                  |   9 +-
 kernel/rcu/tree_stall.h            |   7 +-
 kernel/sched/core.c                |   6 +-
 kernel/sched/cputime.c             | 308 +++++++++++++++++++++++++++++++------
 kernel/sched/idle.c                |  13 +-
 kernel/time/tick-sched.c           | 212 ++++++-------------------
 kernel/time/tick-sched.h           |  12 --
 kernel/time/timer_list.c           |   6 +-
 scripts/gdb/linux/timerlist.py     |   4 -
 21 files changed, 529 insertions(+), 358 deletions(-)

[PATCH 01/15] tick/sched: Fix TOCTOU in nohz idle time fetch

[Frederic Weisbecker]

When the nohz idle time is fetched, the current clock timestamp is taken
outside the seqcount, which can result in such a race as reported by
Sashiko:

    get_cpu_sleep_time_us()                 tick_nohz_start_idle()
    -----------------------                 ---------------------
    now = ktime_get()
                                            write_seqcount_begin(idle_sleeptime_seq);
                                            idle_entrytime = ktime_get()
                                            tick_sched_flag_set(ts, TS_FLAG_IDLE_ACTIVE);
                                            write_seqcount_end(&ts->idle_sleeptime_seq);
    read_seqcount_begin(idle_sleeptime_seq)
    delta = now - idle_entrytime);
    //!! But now < idle_entrytime
    idle = *sleeptime +  delta;
    read_seqcount_retry(&ts->idle_sleeptime_seq, seq)

Here the read side fetches the timestamp before the write side and its
update. As a result the time delta computed on the read side is negative
(ktime_t is signed) and breaks the cputime monotonicity guarantee.

This could possibly be fixed with reading the current clock timestamp
inside the seqcount but the reader overhead might then increase. Also
simply checking that the current timestamp is above the idle entry time
is enough to prevent any issue of the like.

Reported-by: Sashiko
Fixes: 620a30fa0bd1 ("timers/nohz: Protect idle/iowait sleep time under seqcount")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 kernel/time/tick-sched.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index cbbb87a0c6e7..171393367b5c 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -797,15 +797,16 @@ static u64 get_cpu_sleep_time_us(struct tick_sched *ts, ktime_t *sleeptime,
 		*last_update_time = ktime_to_us(now);
 
 	do {
+		ktime_t delta = 0;
+
 		seq = read_seqcount_begin(&ts->idle_sleeptime_seq);
 
 		if (tick_sched_flag_test(ts, TS_FLAG_IDLE_ACTIVE) && compute_delta) {
-			ktime_t delta = ktime_sub(now, ts->idle_entrytime);
-
-			idle = ktime_add(*sleeptime, delta);
-		} else {
-			idle = *sleeptime;
+			if (now > ts->idle_entrytime)
+				delta = ktime_sub(now, ts->idle_entrytime);
 		}
+
+		idle = ktime_add(*sleeptime, delta);
 	} while (read_seqcount_retry(&ts->idle_sleeptime_seq, seq));
 
 	return ktime_to_us(idle);
-- 
2.53.0

[PATCH 02/15] sched/idle: Handle offlining first in idle loop

[Frederic Weisbecker]

Offline handling happens from within the inner idle loop,
after the beginning of dyntick cputime accounting, nohz idle
load balancing and TIF_NEED_RESCHED polling.

This is not necessary and even buggy because:

* There is no dyntick handling to do. And calling tick_nohz_idle_enter()
  messes up with the struct tick_sched reset that was performed on
  tick_sched_timer_dying().

* There is no nohz idle balancing to do.

* Polling on TIF_RESCHED is irrelevant at this stage, there are no more
  tasks allowed to run.

* No need to check if need_resched() before offline handling since
  stop_machine is done and all per-cpu kthread should be done with
  their job.

Therefore move the offline handling at the beginning of the idle loop.
This will also ease the idle cputime unification later by not elapsing
idle time while offline through the call to:

	tick_nohz_idle_enter() -> tick_nohz_start_idle()

Reviewed-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Reviewed-by: Shrikanth Hegde <sshegde@linux.ibm.com>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 kernel/sched/idle.c | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
index a83be0c834dd..aa7e3dc59856 100644
--- a/kernel/sched/idle.c
+++ b/kernel/sched/idle.c
@@ -280,6 +280,14 @@ static void do_idle(void)
 	int cpu = smp_processor_id();
 	bool got_tick = false;
 
+	if (cpu_is_offline(cpu)) {
+		local_irq_disable();
+		/* All per-CPU kernel threads should be done by now. */
+		WARN_ON_ONCE(need_resched());
+		cpuhp_report_idle_dead();
+		arch_cpu_idle_dead();
+	}
+
 	/*
 	 * Check if we need to update blocked load
 	 */
@@ -331,11 +339,6 @@ static void do_idle(void)
 		 */
 		local_irq_disable();
 
-		if (cpu_is_offline(cpu)) {
-			cpuhp_report_idle_dead();
-			arch_cpu_idle_dead();
-		}
-
 		arch_cpu_idle_enter();
 		rcu_nocb_flush_deferred_wakeup();
 
-- 
2.53.0

[PATCH 03/15] sched/cputime: Remove superfluous and error prone kcpustat_field() parameter

[Frederic Weisbecker]

The first parameter to kcpustat_field() is a pointer to the cpu kcpustat
to be fetched from. This parameter is error prone because a copy to a
kcpustat could be passed by accident instead of the original one. Also
the kcpustat structure can already be retrieved with the help of the
mandatory CPU argument.

Remove the needless parameter.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Reviewed-by: Shrikanth Hegde <sshegde@linux.ibm.com>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 drivers/cpufreq/cpufreq_governor.c | 6 +++---
 drivers/macintosh/rack-meter.c     | 2 +-
 include/linux/kernel_stat.h        | 8 +++-----
 kernel/rcu/tree.c                  | 9 +++------
 kernel/rcu/tree_stall.h            | 7 +++----
 kernel/sched/cputime.c             | 5 ++---
 6 files changed, 15 insertions(+), 22 deletions(-)

diff --git a/drivers/cpufreq/cpufreq_governor.c b/drivers/cpufreq/cpufreq_governor.c
index 86f35e451914..3c4a1f9af3ae 100644
--- a/drivers/cpufreq/cpufreq_governor.c
+++ b/drivers/cpufreq/cpufreq_governor.c
@@ -105,7 +105,7 @@ void gov_update_cpu_data(struct dbs_data *dbs_data)
 			j_cdbs->prev_cpu_idle = get_cpu_idle_time(j, &j_cdbs->prev_update_time,
 								  dbs_data->io_is_busy);
 			if (dbs_data->ignore_nice_load)
-				j_cdbs->prev_cpu_nice = kcpustat_field(&kcpustat_cpu(j), CPUTIME_NICE, j);
+				j_cdbs->prev_cpu_nice = kcpustat_field(CPUTIME_NICE, j);
 		}
 	}
 }
@@ -165,7 +165,7 @@ unsigned int dbs_update(struct cpufreq_policy *policy)
 		j_cdbs->prev_cpu_idle = cur_idle_time;
 
 		if (ignore_nice) {
-			u64 cur_nice = kcpustat_field(&kcpustat_cpu(j), CPUTIME_NICE, j);
+			u64 cur_nice = kcpustat_field(CPUTIME_NICE, j);
 
 			idle_time += div_u64(cur_nice - j_cdbs->prev_cpu_nice, NSEC_PER_USEC);
 			j_cdbs->prev_cpu_nice = cur_nice;
@@ -539,7 +539,7 @@ int cpufreq_dbs_governor_start(struct cpufreq_policy *policy)
 		j_cdbs->prev_load = 0;
 
 		if (ignore_nice)
-			j_cdbs->prev_cpu_nice = kcpustat_field(&kcpustat_cpu(j), CPUTIME_NICE, j);
+			j_cdbs->prev_cpu_nice = kcpustat_field(CPUTIME_NICE, j);
 	}
 
 	gov->start(policy);
diff --git a/drivers/macintosh/rack-meter.c b/drivers/macintosh/rack-meter.c
index 8a1e2c08b096..26cb93191ede 100644
--- a/drivers/macintosh/rack-meter.c
+++ b/drivers/macintosh/rack-meter.c
@@ -87,7 +87,7 @@ static inline u64 get_cpu_idle_time(unsigned int cpu)
 		 kcpustat->cpustat[CPUTIME_IOWAIT];
 
 	if (rackmeter_ignore_nice)
-		retval += kcpustat_field(kcpustat, CPUTIME_NICE, cpu);
+		retval += kcpustat_field(CPUTIME_NICE, cpu);
 
 	return retval;
 }
diff --git a/include/linux/kernel_stat.h b/include/linux/kernel_stat.h
index b97ce2df376f..dd020ecaf67b 100644
--- a/include/linux/kernel_stat.h
+++ b/include/linux/kernel_stat.h
@@ -100,14 +100,12 @@ static inline unsigned long kstat_cpu_irqs_sum(unsigned int cpu)
 }
 
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
-extern u64 kcpustat_field(struct kernel_cpustat *kcpustat,
-			  enum cpu_usage_stat usage, int cpu);
+extern u64 kcpustat_field(enum cpu_usage_stat usage, int cpu);
 extern void kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu);
 #else
-static inline u64 kcpustat_field(struct kernel_cpustat *kcpustat,
-				 enum cpu_usage_stat usage, int cpu)
+static inline u64 kcpustat_field(enum cpu_usage_stat usage, int cpu)
 {
-	return kcpustat->cpustat[usage];
+	return kcpustat_cpu(cpu).cpustat[usage];
 }
 
 static inline void kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu)
diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 55df6d37145e..3cbf79bee976 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -969,14 +969,11 @@ static int rcu_watching_snap_recheck(struct rcu_data *rdp)
 		if (rcu_cpu_stall_cputime && rdp->snap_record.gp_seq != rdp->gp_seq) {
 			int cpu = rdp->cpu;
 			struct rcu_snap_record *rsrp;
-			struct kernel_cpustat *kcsp;
-
-			kcsp = &kcpustat_cpu(cpu);
 
 			rsrp = &rdp->snap_record;
-			rsrp->cputime_irq     = kcpustat_field(kcsp, CPUTIME_IRQ, cpu);
-			rsrp->cputime_softirq = kcpustat_field(kcsp, CPUTIME_SOFTIRQ, cpu);
-			rsrp->cputime_system  = kcpustat_field(kcsp, CPUTIME_SYSTEM, cpu);
+			rsrp->cputime_irq     = kcpustat_field(CPUTIME_IRQ, cpu);
+			rsrp->cputime_softirq = kcpustat_field(CPUTIME_SOFTIRQ, cpu);
+			rsrp->cputime_system  = kcpustat_field(CPUTIME_SYSTEM, cpu);
 			rsrp->nr_hardirqs = kstat_cpu_irqs_sum(cpu) + arch_irq_stat_cpu(cpu);
 			rsrp->nr_softirqs = kstat_cpu_softirqs_sum(cpu);
 			rsrp->nr_csw = nr_context_switches_cpu(cpu);
diff --git a/kernel/rcu/tree_stall.h b/kernel/rcu/tree_stall.h
index b67532cb8770..cf7ae51cba40 100644
--- a/kernel/rcu/tree_stall.h
+++ b/kernel/rcu/tree_stall.h
@@ -479,7 +479,6 @@ static void print_cpu_stat_info(int cpu)
 {
 	struct rcu_snap_record rsr, *rsrp;
 	struct rcu_data *rdp = per_cpu_ptr(&rcu_data, cpu);
-	struct kernel_cpustat *kcsp = &kcpustat_cpu(cpu);
 
 	if (!rcu_cpu_stall_cputime)
 		return;
@@ -488,9 +487,9 @@ static void print_cpu_stat_info(int cpu)
 	if (rsrp->gp_seq != rdp->gp_seq)
 		return;
 
-	rsr.cputime_irq     = kcpustat_field(kcsp, CPUTIME_IRQ, cpu);
-	rsr.cputime_softirq = kcpustat_field(kcsp, CPUTIME_SOFTIRQ, cpu);
-	rsr.cputime_system  = kcpustat_field(kcsp, CPUTIME_SYSTEM, cpu);
+	rsr.cputime_irq     = kcpustat_field(CPUTIME_IRQ, cpu);
+	rsr.cputime_softirq = kcpustat_field(CPUTIME_SOFTIRQ, cpu);
+	rsr.cputime_system  = kcpustat_field(CPUTIME_SYSTEM, cpu);
 
 	pr_err("\t         hardirqs   softirqs   csw/system\n");
 	pr_err("\t number: %8lld %10d %12lld\n",
diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index fbf31db0d2f3..caaaf0a04ced 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -975,10 +975,9 @@ static int kcpustat_field_vtime(u64 *cpustat,
 	return 0;
 }
 
-u64 kcpustat_field(struct kernel_cpustat *kcpustat,
-		   enum cpu_usage_stat usage, int cpu)
+u64 kcpustat_field(enum cpu_usage_stat usage, int cpu)
 {
-	u64 *cpustat = kcpustat->cpustat;
+	u64 *cpustat = kcpustat_cpu(cpu).cpustat;
 	u64 val = cpustat[usage];
 	struct rq *rq;
 	int err;
-- 
2.53.0

[PATCH 04/15] sched/cputime: Correctly support generic vtime idle time

[Frederic Weisbecker]

Currently whether generic vtime is running or not, the idle cputime is
fetched from the nohz accounting.

However generic vtime already does its own idle cputime accounting. Only
the kernel stat accessors are not plugged to support it.

Read the idle generic vtime cputime when it's running, this will allow
to later more clearly split nohz and vtime cputime accounting.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 include/linux/vtime.h    |  9 +++++++--
 kernel/sched/cputime.c   | 38 +++++++++++++++++++++++++++++---------
 kernel/time/tick-sched.c | 12 +++++++++---
 3 files changed, 45 insertions(+), 14 deletions(-)

diff --git a/include/linux/vtime.h b/include/linux/vtime.h
index 29dd5b91dd7d..336875bea767 100644
--- a/include/linux/vtime.h
+++ b/include/linux/vtime.h
@@ -10,7 +10,6 @@
  */
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
 extern void vtime_account_kernel(struct task_struct *tsk);
-extern void vtime_account_idle(struct task_struct *tsk);
 #endif /* !CONFIG_VIRT_CPU_ACCOUNTING */
 
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
@@ -27,7 +26,13 @@ static inline void vtime_guest_exit(struct task_struct *tsk) { }
 static inline void vtime_init_idle(struct task_struct *tsk, int cpu) { }
 #endif
 
+static inline bool vtime_generic_enabled_cpu(int cpu)
+{
+	return context_tracking_enabled_cpu(cpu);
+}
+
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
+extern void vtime_account_idle(struct task_struct *tsk);
 extern void vtime_account_irq(struct task_struct *tsk, unsigned int offset);
 extern void vtime_account_softirq(struct task_struct *tsk);
 extern void vtime_account_hardirq(struct task_struct *tsk);
@@ -74,7 +79,7 @@ static inline bool vtime_accounting_enabled(void)
 
 static inline bool vtime_accounting_enabled_cpu(int cpu)
 {
-	return context_tracking_enabled_cpu(cpu);
+	return vtime_generic_enabled_cpu(cpu);
 }
 
 static inline bool vtime_accounting_enabled_this_cpu(void)
diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index caaaf0a04ced..815d0f772cae 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -773,9 +773,9 @@ void vtime_guest_exit(struct task_struct *tsk)
 }
 EXPORT_SYMBOL_GPL(vtime_guest_exit);
 
-void vtime_account_idle(struct task_struct *tsk)
+static void __vtime_account_idle(struct vtime *vtime)
 {
-	account_idle_time(get_vtime_delta(&tsk->vtime));
+	account_idle_time(get_vtime_delta(vtime));
 }
 
 void vtime_task_switch_generic(struct task_struct *prev)
@@ -784,7 +784,7 @@ void vtime_task_switch_generic(struct task_struct *prev)
 
 	write_seqcount_begin(&vtime->seqcount);
 	if (vtime->state == VTIME_IDLE)
-		vtime_account_idle(prev);
+		__vtime_account_idle(vtime);
 	else
 		__vtime_account_kernel(prev, vtime);
 	vtime->state = VTIME_INACTIVE;
@@ -926,6 +926,7 @@ static int kcpustat_field_vtime(u64 *cpustat,
 				int cpu, u64 *val)
 {
 	struct vtime *vtime = &tsk->vtime;
+	struct rq *rq = cpu_rq(cpu);
 	unsigned int seq;
 
 	do {
@@ -967,6 +968,14 @@ static int kcpustat_field_vtime(u64 *cpustat,
 			if (state == VTIME_GUEST && task_nice(tsk) > 0)
 				*val += vtime->gtime + vtime_delta(vtime);
 			break;
+		case CPUTIME_IDLE:
+			if (state == VTIME_IDLE && !atomic_read(&rq->nr_iowait))
+				*val += vtime_delta(vtime);
+			break;
+		case CPUTIME_IOWAIT:
+			if (state == VTIME_IDLE && atomic_read(&rq->nr_iowait) > 0)
+				*val += vtime_delta(vtime);
+			break;
 		default:
 			break;
 		}
@@ -1029,8 +1038,8 @@ static int kcpustat_cpu_fetch_vtime(struct kernel_cpustat *dst,
 		*dst = *src;
 		cpustat = dst->cpustat;
 
-		/* Task is sleeping, dead or idle, nothing to add */
-		if (state < VTIME_SYS)
+		/* Task is sleeping or dead, nothing to add */
+		if (state < VTIME_IDLE)
 			continue;
 
 		delta = vtime_delta(vtime);
@@ -1039,15 +1048,17 @@ static int kcpustat_cpu_fetch_vtime(struct kernel_cpustat *dst,
 		 * Task runs either in user (including guest) or kernel space,
 		 * add pending nohz time to the right place.
 		 */
-		if (state == VTIME_SYS) {
+		switch (state) {
+		case VTIME_SYS:
 			cpustat[CPUTIME_SYSTEM] += vtime->stime + delta;
-		} else if (state == VTIME_USER) {
+			break;
+		case VTIME_USER:
 			if (task_nice(tsk) > 0)
 				cpustat[CPUTIME_NICE] += vtime->utime + delta;
 			else
 				cpustat[CPUTIME_USER] += vtime->utime + delta;
-		} else {
-			WARN_ON_ONCE(state != VTIME_GUEST);
+			break;
+		case VTIME_GUEST:
 			if (task_nice(tsk) > 0) {
 				cpustat[CPUTIME_GUEST_NICE] += vtime->gtime + delta;
 				cpustat[CPUTIME_NICE] += vtime->gtime + delta;
@@ -1055,6 +1066,15 @@ static int kcpustat_cpu_fetch_vtime(struct kernel_cpustat *dst,
 				cpustat[CPUTIME_GUEST] += vtime->gtime + delta;
 				cpustat[CPUTIME_USER] += vtime->gtime + delta;
 			}
+			break;
+		case VTIME_IDLE:
+			if (atomic_read(&cpu_rq(cpu)->nr_iowait) > 0)
+				cpustat[CPUTIME_IOWAIT] += delta;
+			else
+				cpustat[CPUTIME_IDLE] += delta;
+			break;
+		default:
+			WARN_ON_ONCE(1);
 		}
 	} while (read_seqcount_retry(&vtime->seqcount, seq));
 
diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index 171393367b5c..597c3a0682e7 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -783,9 +783,10 @@ static void tick_nohz_start_idle(struct tick_sched *ts)
 	sched_clock_idle_sleep_event();
 }
 
-static u64 get_cpu_sleep_time_us(struct tick_sched *ts, ktime_t *sleeptime,
+static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx, ktime_t *sleeptime,
 				 bool compute_delta, u64 *last_update_time)
 {
+	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
 	ktime_t now, idle;
 	unsigned int seq;
 
@@ -796,6 +797,11 @@ static u64 get_cpu_sleep_time_us(struct tick_sched *ts, ktime_t *sleeptime,
 	if (last_update_time)
 		*last_update_time = ktime_to_us(now);
 
+	if (vtime_generic_enabled_cpu(cpu)) {
+		idle = kcpustat_field(idx, cpu);
+		return ktime_to_us(idle);
+	}
+
 	do {
 		ktime_t delta = 0;
 
@@ -834,7 +840,7 @@ u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
 {
 	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
 
-	return get_cpu_sleep_time_us(ts, &ts->idle_sleeptime,
+	return get_cpu_sleep_time_us(cpu, CPUTIME_IDLE, &ts->idle_sleeptime,
 				     !nr_iowait_cpu(cpu), last_update_time);
 }
 EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
@@ -860,7 +866,7 @@ u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
 {
 	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
 
-	return get_cpu_sleep_time_us(ts, &ts->iowait_sleeptime,
+	return get_cpu_sleep_time_us(cpu, CPUTIME_IOWAIT, &ts->iowait_sleeptime,
 				     nr_iowait_cpu(cpu), last_update_time);
 }
 EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us);
-- 
2.53.0

[PATCH 05/15] powerpc/time: Prepare to stop elapsing in dynticks-idle

[Frederic Weisbecker]

Currently the tick subsystem stores the idle cputime accounting in
private fields, allowing cohabitation with architecture idle vtime
accounting. The former is fetched on online CPUs, the latter on offline
CPUs.

For consolidation purpose, architecture vtime accounting will continue
to account the cputime but will make a break when the idle tick is
stopped. The dyntick cputime accounting will then be relayed by the tick
subsystem so that the idle cputime is still seen advancing coherently
even when the tick isn't there to flush the idle vtime.

Prepare for that and introduce three new APIs which will be used in
subsequent patches:

_ vtime_dynticks_start() is deemed to be called when idle enters in
  dyntick mode. The idle cputime that elapsed so far is accumulated.

- vtime_dynticks_stop() is deemed to be called when idle exits from
  dyntick mode. The vtime entry clocks are fast-forward to current time
  so that idle accounting restarts elapsing from now.

- vtime_reset() is deemed to be called from dynticks idle IRQ entry to
  fast-forward the clock to current time so that the IRQ time is still
  accounted by vtime while nohz cputime is paused.

Also accumulated vtime won't be flushed from dyntick-idle ticks to avoid
accounting twice the idle cputime, along with nohz accounting.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Reviewed-by: Shrikanth Hegde <sshegde@linux.ibm.com>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 arch/powerpc/kernel/time.c | 41 ++++++++++++++++++++++++++++++++++++++
 include/linux/vtime.h      |  6 ++++++
 2 files changed, 47 insertions(+)

diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c
index 4bbeb8644d3d..18506740f4a4 100644
--- a/arch/powerpc/kernel/time.c
+++ b/arch/powerpc/kernel/time.c
@@ -376,6 +376,47 @@ void vtime_task_switch(struct task_struct *prev)
 		acct->starttime = acct0->starttime;
 	}
 }
+
+#ifdef CONFIG_NO_HZ_COMMON
+/**
+ * vtime_reset - Fast forward vtime entry clocks
+ *
+ * Called from dynticks idle IRQ entry to fast-forward the clocks to current time
+ * so that the IRQ time is still accounted by vtime while nohz cputime is paused.
+ */
+void vtime_reset(void)
+{
+	struct cpu_accounting_data *acct = get_accounting(current);
+
+	acct->starttime = mftb();
+#ifdef CONFIG_ARCH_HAS_SCALED_CPUTIME
+	acct->startspurr = read_spurr(acct->starttime);
+#endif
+}
+
+/**
+ * vtime_dyntick_start - Inform vtime about entry to idle-dynticks
+ *
+ * Called when idle enters in dyntick mode. The idle cputime that elapsed so far
+ * is accumulated and the tick subsystem takes over the idle cputime accounting.
+ */
+void vtime_dyntick_start(void)
+{
+	vtime_account_idle(current);
+}
+
+/**
+ * vtime_dyntick_stop - Inform vtime about exit from idle-dynticks
+ *
+ * Called when idle exits from dyntick mode. The vtime entry clocks are
+ * fast-forward to current time so that idle accounting restarts elapsing from
+ * now.
+ */
+void vtime_dyntick_stop(void)
+{
+	vtime_reset();
+}
+#endif /* CONFIG_NO_HZ_COMMON */
 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 
 void __no_kcsan __delay(unsigned long loops)
diff --git a/include/linux/vtime.h b/include/linux/vtime.h
index 336875bea767..5516ad549e60 100644
--- a/include/linux/vtime.h
+++ b/include/linux/vtime.h
@@ -37,11 +37,17 @@ extern void vtime_account_irq(struct task_struct *tsk, unsigned int offset);
 extern void vtime_account_softirq(struct task_struct *tsk);
 extern void vtime_account_hardirq(struct task_struct *tsk);
 extern void vtime_flush(struct task_struct *tsk);
+extern void vtime_reset(void);
+extern void vtime_dyntick_start(void);
+extern void vtime_dyntick_stop(void);
 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
 static inline void vtime_account_irq(struct task_struct *tsk, unsigned int offset) { }
 static inline void vtime_account_softirq(struct task_struct *tsk) { }
 static inline void vtime_account_hardirq(struct task_struct *tsk) { }
 static inline void vtime_flush(struct task_struct *tsk) { }
+static inline void vtime_reset(void) { }
+static inline void vtime_dyntick_start(void) { }
+static inline void vtime_dyntick_stop(void) { }
 #endif
 
 /*
-- 
2.53.0

[PATCH 06/15] s390/time: Prepare to stop elapsing in dynticks-idle

[Frederic Weisbecker]

Currently the tick subsystem stores the idle cputime accounting in
private fields, allowing cohabitation with architecture idle vtime
accounting. The former is fetched on online CPUs, the latter on offline
CPUs.

For consolidation purpose, architecture vtime accounting will continue
to account the cputime but will make a break when the idle tick is
stopped. The dyntick cputime accounting will then be relayed by the tick
subsystem so that the idle cputime is still seen advancing coherently
even when the tick isn't there to flush the idle vtime.

Prepare for that and introduce three new APIs which will be used in
subsequent patches:

_ vtime_dynticks_start() is deemed to be called when idle enters in
  dyntick mode. The idle cputime that elapsed so far is accumulated
  and accounted. Also idle time accounting is ignored.

- vtime_dynticks_stop() is deemed to be called when idle exits from
  dyntick mode. The vtime entry clocks are fast-forward to current time
  so that idle accounting restarts elapsing from now. Also idle time
  accounting is resumed.

- vtime_reset() is deemed to be called from dynticks idle IRQ entry to
  fast-forward the clock to current time so that the IRQ time is still
  accounted by vtime while nohz cputime is paused.

Also accumulated vtime won't be flushed from dyntick-idle ticks to avoid
accounting twice the idle cputime, along with nohz accounting.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Co-developed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 arch/s390/include/asm/idle.h |  2 +
 arch/s390/kernel/idle.c      |  5 ++-
 arch/s390/kernel/vtime.c     | 75 ++++++++++++++++++++++++++++++++----
 3 files changed, 73 insertions(+), 9 deletions(-)

diff --git a/arch/s390/include/asm/idle.h b/arch/s390/include/asm/idle.h
index 32536ee34aa0..e4ad09a22400 100644
--- a/arch/s390/include/asm/idle.h
+++ b/arch/s390/include/asm/idle.h
@@ -8,10 +8,12 @@
 #ifndef _S390_IDLE_H
 #define _S390_IDLE_H
 
+#include <linux/percpu-defs.h>
 #include <linux/types.h>
 #include <linux/device.h>
 
 struct s390_idle_data {
+	bool	      idle_dyntick;
 	unsigned long idle_count;
 	unsigned long idle_time;
 	unsigned long clock_idle_enter;
diff --git a/arch/s390/kernel/idle.c b/arch/s390/kernel/idle.c
index 1f1b06b6b4ef..4685d7c5bc51 100644
--- a/arch/s390/kernel/idle.c
+++ b/arch/s390/kernel/idle.c
@@ -31,7 +31,10 @@ void account_idle_time_irq(void)
 	/* Account time spent with enabled wait psw loaded as idle time. */
 	__atomic64_add(idle_time, &idle->idle_time);
 	__atomic64_add_const(1, &idle->idle_count);
-	account_idle_time(cputime_to_nsecs(idle_time));
+
+	/* Dyntick idle time accounted by nohz/scheduler */
+	if (!idle->idle_dyntick)
+		account_idle_time(cputime_to_nsecs(idle_time));
 }
 
 void noinstr arch_cpu_idle(void)
diff --git a/arch/s390/kernel/vtime.c b/arch/s390/kernel/vtime.c
index bf48744d0912..d1102a6f80bd 100644
--- a/arch/s390/kernel/vtime.c
+++ b/arch/s390/kernel/vtime.c
@@ -17,6 +17,7 @@
 #include <asm/vtimer.h>
 #include <asm/vtime.h>
 #include <asm/cpu_mf.h>
+#include <asm/idle.h>
 #include <asm/smp.h>
 
 #include "entry.h"
@@ -110,6 +111,16 @@ static void account_system_index_scaled(struct task_struct *p, u64 cputime,
 	account_system_index_time(p, cputime_to_nsecs(cputime), index);
 }
 
+static inline void vtime_reset_last_update(struct lowcore *lc)
+{
+	asm volatile(
+		"	stpt	%0\n"	/* Store current cpu timer value */
+		"	stckf	%1"	/* Store current tod clock value */
+		: "=Q" (lc->last_update_timer),
+		  "=Q" (lc->last_update_clock)
+		: : "cc");
+}
+
 /*
  * Update process times based on virtual cpu times stored by entry.S
  * to the lowcore fields user_timer, system_timer & steal_clock.
@@ -121,17 +132,16 @@ static int do_account_vtime(struct task_struct *tsk)
 
 	timer = lc->last_update_timer;
 	clock = lc->last_update_clock;
-	asm volatile(
-		"	stpt	%0\n"	/* Store current cpu timer value */
-		"	stckf	%1"	/* Store current tod clock value */
-		: "=Q" (lc->last_update_timer),
-		  "=Q" (lc->last_update_clock)
-		: : "cc");
+
+	vtime_reset_last_update(lc);
+
 	clock = lc->last_update_clock - clock;
 	timer -= lc->last_update_timer;
 
 	if (hardirq_count())
 		lc->hardirq_timer += timer;
+	else if (in_serving_softirq())
+		lc->softirq_timer += timer;
 	else
 		lc->system_timer += timer;
 
@@ -231,14 +241,63 @@ EXPORT_SYMBOL_GPL(vtime_account_kernel);
 
 void vtime_account_softirq(struct task_struct *tsk)
 {
-	get_lowcore()->softirq_timer += vtime_delta();
+	if (!__this_cpu_read(s390_idle.idle_dyntick))
+		get_lowcore()->softirq_timer += vtime_delta();
+	else
+		vtime_flush(tsk);
 }
 
 void vtime_account_hardirq(struct task_struct *tsk)
 {
-	get_lowcore()->hardirq_timer += vtime_delta();
+	if (!__this_cpu_read(s390_idle.idle_dyntick)) {
+		get_lowcore()->hardirq_timer += vtime_delta();
+	} else {
+		/*
+		 * In dynticks mode, the idle cputime is accounted by the nohz
+		 * subsystem. Therefore the s390 timer/clocks are reset on IRQ
+		 * entry and steal time must be accounted now.
+		 */
+		vtime_flush(tsk);
+	}
 }
 
+#ifdef CONFIG_NO_HZ_COMMON
+/**
+ * vtime_reset - Fast forward vtime entry clocks
+ *
+ * Called from dynticks idle IRQ entry to fast-forward the clocks to current time
+ * so that the IRQ time is still accounted by vtime while nohz cputime is paused.
+ */
+void vtime_reset(void)
+{
+	vtime_reset_last_update(get_lowcore());
+}
+
+/**
+ * vtime_dyntick_start - Inform vtime about entry to idle-dynticks
+ *
+ * Called when idle enters in dyntick mode. The idle cputime that elapsed so far
+ * is flushed and the tick subsystem takes over the idle cputime accounting.
+ */
+void vtime_dyntick_start(void)
+{
+	__this_cpu_write(s390_idle.idle_dyntick, true);
+	vtime_flush(current);
+}
+
+/**
+ * vtime_dyntick_stop - Inform vtime about exit from idle-dynticks
+ *
+ * Called when idle exits from dyntick mode. The vtime entry clocks are
+ * fast-forward to current time and idle accounting resumes.
+ */
+void vtime_dyntick_stop(void)
+{
+	vtime_reset_last_update(get_lowcore());
+	__this_cpu_write(s390_idle.idle_dyntick, false);
+}
+#endif /* CONFIG_NO_HZ_COMMON */
+
 /*
  * Sorted add to a list. List is linear searched until first bigger
  * element is found.
-- 
2.53.0

[PATCH 07/15] tick/sched: Unify idle cputime accounting

[Frederic Weisbecker]

The non-vtime dynticks-idle cputime accounting is a big mess that
accumulates within two concurrent statistics, each having their own
shortcomings:

* The accounting for online CPUs which is based on the delta between
  tick_nohz_start_idle() and tick_nohz_stop_idle().

  Pros:
       - Works when the tick is off

       - Has nsecs granularity

  Cons:
       - Account idle steal time but doesn't substract it from idle
         cputime.

       - Assumes CONFIG_IRQ_TIME_ACCOUNTING by not accounting IRQs but
         the IRQ time is simply ignored when
         CONFIG_IRQ_TIME_ACCOUNTING=n

       - The windows between 1) idle task scheduling and the first call
         to tick_nohz_start_idle() and 2) idle task between the last
         tick_nohz_stop_idle() and the rest of the idle time are
         blindspots wrt. cputime accounting (though mostly insignificant
         amount)

       - Relies on private fields outside of kernel stats, with specific
         accessors.

* The accounting for offline CPUs which is based on ticks and the
  jiffies delta during which the tick was stopped.

  Pros:
       - Handles steal time correctly

       - Handle CONFIG_IRQ_TIME_ACCOUNTING=y and
         CONFIG_IRQ_TIME_ACCOUNTING=n correctly.

       - Handles the whole idle task

       - Accounts directly to kernel stats, without midlayer accumulator.

   Cons:
       - Doesn't elapse when the tick is off, which doesn't make it
         suitable for online CPUs.

       - Has TICK_NSEC granularity (jiffies)

       - Needs to track the dyntick-idle ticks that were accounted and
         substract them from the total jiffies time spent while the tick
         was stopped. This is an ugly workaround.

Having two different accounting for a single context is not the only
problem: since those accountings are of different natures, it is
possible to observe the global idle time going backward after a CPU goes
offline.

Clean up the situation with introducing a hybrid approach that stays
coherent and works for both online and offline CPUs:

* Tick based or native vtime accounting operate before the idle loop
  is entered and resume once the idle loop prepares to exit.

* When the idle loop starts, switch to dynticks-idle accounting as is
  done currently, except that the statistics accumulate directly to the
  relevant kernel stat fields.

* Private dyntick cputime accounting fields are removed.

* Works on both online and offline case.

Further improvement will include:

* Only switch to dynticks-idle cputime accounting when the tick actually
  goes in dynticks mode.

* Handle CONFIG_IRQ_TIME_ACCOUNTING=n correctly such that the
  dynticks-idle accounting still elapses while on IRQs.

* Correctly substract idle steal cputime from idle time

Reported-by: Xin Zhao <jackzxcui1989@163.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 include/linux/kernel_stat.h | 24 ++++++++++---
 include/linux/vtime.h       |  7 +++-
 kernel/sched/cputime.c      | 62 ++++++++++++++++-----------------
 kernel/time/tick-sched.c    | 69 ++++++++++---------------------------
 4 files changed, 75 insertions(+), 87 deletions(-)

diff --git a/include/linux/kernel_stat.h b/include/linux/kernel_stat.h
index dd020ecaf67b..ba65aad308a1 100644
--- a/include/linux/kernel_stat.h
+++ b/include/linux/kernel_stat.h
@@ -34,6 +34,9 @@ enum cpu_usage_stat {
 };
 
 struct kernel_cpustat {
+#ifdef CONFIG_NO_HZ_COMMON
+	int idle_dyntick;
+#endif
 	u64 cpustat[NR_STATS];
 };
 
@@ -99,6 +102,20 @@ static inline unsigned long kstat_cpu_irqs_sum(unsigned int cpu)
 	return kstat_cpu(cpu).irqs_sum;
 }
 
+#ifdef CONFIG_NO_HZ_COMMON
+extern void kcpustat_dyntick_start(void);
+extern void kcpustat_dyntick_stop(void);
+static inline bool kcpustat_idle_dyntick(void)
+{
+	return __this_cpu_read(kernel_cpustat.idle_dyntick);
+}
+#else
+static inline bool kcpustat_idle_dyntick(void)
+{
+	return false;
+}
+#endif /* CONFIG_NO_HZ_COMMON */
+
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
 extern u64 kcpustat_field(enum cpu_usage_stat usage, int cpu);
 extern void kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu);
@@ -113,7 +130,7 @@ static inline void kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu)
 	*dst = kcpustat_cpu(cpu);
 }
 
-#endif
+#endif /* !CONFIG_VIRT_CPU_ACCOUNTING_GEN */
 
 extern void account_user_time(struct task_struct *, u64);
 extern void account_guest_time(struct task_struct *, u64);
@@ -127,14 +144,13 @@ extern u64 get_idle_time(struct kernel_cpustat *kcs, int cpu);
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 static inline void account_process_tick(struct task_struct *tsk, int user)
 {
-	vtime_flush(tsk);
+	if (!kcpustat_idle_dyntick())
+		vtime_flush(tsk);
 }
 #else
 extern void account_process_tick(struct task_struct *, int user);
 #endif
 
-extern void account_idle_ticks(unsigned long ticks);
-
 #ifdef CONFIG_SCHED_CORE
 extern void __account_forceidle_time(struct task_struct *tsk, u64 delta);
 #endif
diff --git a/include/linux/vtime.h b/include/linux/vtime.h
index 5516ad549e60..9dc25b04a119 100644
--- a/include/linux/vtime.h
+++ b/include/linux/vtime.h
@@ -31,6 +31,11 @@ static inline bool vtime_generic_enabled_cpu(int cpu)
 	return context_tracking_enabled_cpu(cpu);
 }
 
+static inline bool vtime_generic_enabled_this_cpu(void)
+{
+	return context_tracking_enabled_this_cpu();
+}
+
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 extern void vtime_account_idle(struct task_struct *tsk);
 extern void vtime_account_irq(struct task_struct *tsk, unsigned int offset);
@@ -90,7 +95,7 @@ static inline bool vtime_accounting_enabled_cpu(int cpu)
 
 static inline bool vtime_accounting_enabled_this_cpu(void)
 {
-	return context_tracking_enabled_this_cpu();
+	return vtime_generic_enabled_this_cpu();
 }
 
 extern void vtime_task_switch_generic(struct task_struct *prev);
diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index 815d0f772cae..a5733789e0bd 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -414,16 +414,30 @@ static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
 	}
 }
 
-static void irqtime_account_idle_ticks(int ticks)
-{
-	irqtime_account_process_tick(current, 0, ticks);
-}
 #else /* !CONFIG_IRQ_TIME_ACCOUNTING: */
-static inline void irqtime_account_idle_ticks(int ticks) { }
 static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick,
 						int nr_ticks) { }
 #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
 
+#ifdef CONFIG_NO_HZ_COMMON
+void kcpustat_dyntick_start(void)
+{
+	if (!vtime_generic_enabled_this_cpu()) {
+		vtime_dyntick_start();
+		__this_cpu_write(kernel_cpustat.idle_dyntick, 1);
+	}
+}
+
+void kcpustat_dyntick_stop(void)
+{
+	if (!vtime_generic_enabled_this_cpu()) {
+		__this_cpu_write(kernel_cpustat.idle_dyntick, 0);
+		vtime_dyntick_stop();
+		steal_account_process_time(ULONG_MAX);
+	}
+}
+#endif /* CONFIG_NO_HZ_COMMON */
+
 /*
  * Use precise platform statistics if available:
  */
@@ -437,11 +451,15 @@ void vtime_account_irq(struct task_struct *tsk, unsigned int offset)
 		vtime_account_hardirq(tsk);
 	} else if (pc & SOFTIRQ_OFFSET) {
 		vtime_account_softirq(tsk);
-	} else if (!IS_ENABLED(CONFIG_HAVE_VIRT_CPU_ACCOUNTING_IDLE) &&
-		   is_idle_task(tsk)) {
-		vtime_account_idle(tsk);
+	} else if (!kcpustat_idle_dyntick()) {
+		if (!IS_ENABLED(CONFIG_HAVE_VIRT_CPU_ACCOUNTING_IDLE) &&
+		    is_idle_task(tsk)) {
+			vtime_account_idle(tsk);
+		} else {
+			vtime_account_kernel(tsk);
+		}
 	} else {
-		vtime_account_kernel(tsk);
+		vtime_reset();
 	}
 }
 
@@ -483,6 +501,9 @@ void account_process_tick(struct task_struct *p, int user_tick)
 	if (vtime_accounting_enabled_this_cpu())
 		return;
 
+	if (kcpustat_idle_dyntick())
+		return;
+
 	if (irqtime_enabled()) {
 		irqtime_account_process_tick(p, user_tick, 1);
 		return;
@@ -504,29 +525,6 @@ void account_process_tick(struct task_struct *p, int user_tick)
 		account_idle_time(cputime);
 }
 
-/*
- * Account multiple ticks of idle time.
- * @ticks: number of stolen ticks
- */
-void account_idle_ticks(unsigned long ticks)
-{
-	u64 cputime, steal;
-
-	if (irqtime_enabled()) {
-		irqtime_account_idle_ticks(ticks);
-		return;
-	}
-
-	cputime = ticks * TICK_NSEC;
-	steal = steal_account_process_time(ULONG_MAX);
-
-	if (steal >= cputime)
-		return;
-
-	cputime -= steal;
-	account_idle_time(cputime);
-}
-
 /*
  * Adjust tick based cputime random precision against scheduler runtime
  * accounting.
diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index 597c3a0682e7..c3efd3583cf9 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -285,8 +285,6 @@ static void tick_sched_handle(struct tick_sched *ts, struct pt_regs *regs)
 	if (IS_ENABLED(CONFIG_NO_HZ_COMMON) &&
 	    tick_sched_flag_test(ts, TS_FLAG_STOPPED)) {
 		touch_softlockup_watchdog_sched();
-		if (is_idle_task(current))
-			ts->idle_jiffies++;
 		/*
 		 * In case the current tick fired too early past its expected
 		 * expiration, make sure we don't bypass the next clock reprogramming
@@ -753,8 +751,12 @@ static void tick_nohz_update_jiffies(ktime_t now)
 
 static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now)
 {
+	u64 *cpustat = kcpustat_this_cpu->cpustat;
 	ktime_t delta;
 
+	if (vtime_generic_enabled_this_cpu())
+		return;
+
 	if (WARN_ON_ONCE(!tick_sched_flag_test(ts, TS_FLAG_IDLE_ACTIVE)))
 		return;
 
@@ -762,9 +764,9 @@ static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now)
 
 	write_seqcount_begin(&ts->idle_sleeptime_seq);
 	if (nr_iowait_cpu(smp_processor_id()) > 0)
-		ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta);
+		cpustat[CPUTIME_IOWAIT] = ktime_add(cpustat[CPUTIME_IOWAIT], delta);
 	else
-		ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta);
+		cpustat[CPUTIME_IDLE] = ktime_add(cpustat[CPUTIME_IDLE], delta);
 
 	ts->idle_entrytime = now;
 	tick_sched_flag_clear(ts, TS_FLAG_IDLE_ACTIVE);
@@ -775,18 +777,21 @@ static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now)
 
 static void tick_nohz_start_idle(struct tick_sched *ts)
 {
+	if (vtime_generic_enabled_this_cpu())
+		return;
+
 	write_seqcount_begin(&ts->idle_sleeptime_seq);
 	ts->idle_entrytime = ktime_get();
 	tick_sched_flag_set(ts, TS_FLAG_IDLE_ACTIVE);
 	write_seqcount_end(&ts->idle_sleeptime_seq);
-
 	sched_clock_idle_sleep_event();
 }
 
-static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx, ktime_t *sleeptime,
+static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx,
 				 bool compute_delta, u64 *last_update_time)
 {
 	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
+	u64 *cpustat = kcpustat_cpu(cpu).cpustat;
 	ktime_t now, idle;
 	unsigned int seq;
 
@@ -812,7 +817,7 @@ static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx, ktime_t *slee
 				delta = ktime_sub(now, ts->idle_entrytime);
 		}
 
-		idle = ktime_add(*sleeptime, delta);
+		idle = ktime_add(cpustat[idx], delta);
 	} while (read_seqcount_retry(&ts->idle_sleeptime_seq, seq));
 
 	return ktime_to_us(idle);
@@ -838,9 +843,7 @@ static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx, ktime_t *slee
  */
 u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
 {
-	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
-
-	return get_cpu_sleep_time_us(cpu, CPUTIME_IDLE, &ts->idle_sleeptime,
+	return get_cpu_sleep_time_us(cpu, CPUTIME_IDLE,
 				     !nr_iowait_cpu(cpu), last_update_time);
 }
 EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
@@ -864,9 +867,7 @@ EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
  */
 u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
 {
-	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
-
-	return get_cpu_sleep_time_us(cpu, CPUTIME_IOWAIT, &ts->iowait_sleeptime,
+	return get_cpu_sleep_time_us(cpu, CPUTIME_IOWAIT,
 				     nr_iowait_cpu(cpu), last_update_time);
 }
 EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us);
@@ -1279,10 +1280,8 @@ void tick_nohz_idle_stop_tick(void)
 		ts->idle_sleeps++;
 		ts->idle_expires = expires;
 
-		if (!was_stopped && tick_sched_flag_test(ts, TS_FLAG_STOPPED)) {
-			ts->idle_jiffies = ts->last_jiffies;
+		if (!was_stopped && tick_sched_flag_test(ts, TS_FLAG_STOPPED))
 			nohz_balance_enter_idle(cpu);
-		}
 	} else {
 		tick_nohz_retain_tick(ts);
 	}
@@ -1311,6 +1310,7 @@ void tick_nohz_idle_enter(void)
 	WARN_ON_ONCE(ts->timer_expires_base);
 
 	tick_sched_flag_set(ts, TS_FLAG_INIDLE);
+	kcpustat_dyntick_start();
 	tick_nohz_start_idle(ts);
 
 	local_irq_enable();
@@ -1436,37 +1436,12 @@ unsigned long tick_nohz_get_idle_calls_cpu(int cpu)
 	return ts->idle_calls;
 }
 
-static void tick_nohz_account_idle_time(struct tick_sched *ts,
-					ktime_t now)
-{
-	unsigned long ticks;
-
-	ts->idle_exittime = now;
-
-	if (vtime_accounting_enabled_this_cpu())
-		return;
-	/*
-	 * We stopped the tick in idle. update_process_times() would miss the
-	 * time we slept, as it does only a 1 tick accounting.
-	 * Enforce that this is accounted to idle !
-	 */
-	ticks = jiffies - ts->idle_jiffies;
-	/*
-	 * We might be one off. Do not randomly account a huge number of ticks!
-	 */
-	if (ticks && ticks < LONG_MAX)
-		account_idle_ticks(ticks);
-}
-
 void tick_nohz_idle_restart_tick(void)
 {
 	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 
-	if (tick_sched_flag_test(ts, TS_FLAG_STOPPED)) {
-		ktime_t now = ktime_get();
-		tick_nohz_restart_sched_tick(ts, now);
-		tick_nohz_account_idle_time(ts, now);
-	}
+	if (tick_sched_flag_test(ts, TS_FLAG_STOPPED))
+		tick_nohz_restart_sched_tick(ts, ktime_get());
 }
 
 static void tick_nohz_idle_update_tick(struct tick_sched *ts, ktime_t now)
@@ -1475,8 +1450,6 @@ static void tick_nohz_idle_update_tick(struct tick_sched *ts, ktime_t now)
 		__tick_nohz_full_update_tick(ts, now);
 	else
 		tick_nohz_restart_sched_tick(ts, now);
-
-	tick_nohz_account_idle_time(ts, now);
 }
 
 /**
@@ -1518,6 +1491,7 @@ void tick_nohz_idle_exit(void)
 
 	if (tick_stopped)
 		tick_nohz_idle_update_tick(ts, now);
+	kcpustat_dyntick_stop();
 
 	local_irq_enable();
 }
@@ -1655,20 +1629,15 @@ void tick_setup_sched_timer(bool hrtimer)
 void tick_sched_timer_dying(int cpu)
 {
 	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
-	ktime_t idle_sleeptime, iowait_sleeptime;
 	unsigned long idle_calls, idle_sleeps;
 
 	/* This must happen before hrtimers are migrated! */
 	if (tick_sched_flag_test(ts, TS_FLAG_HIGHRES))
 		hrtimer_cancel(&ts->sched_timer);
 
-	idle_sleeptime = ts->idle_sleeptime;
-	iowait_sleeptime = ts->iowait_sleeptime;
 	idle_calls = ts->idle_calls;
 	idle_sleeps = ts->idle_sleeps;
 	memset(ts, 0, sizeof(*ts));
-	ts->idle_sleeptime = idle_sleeptime;
-	ts->iowait_sleeptime = iowait_sleeptime;
 	ts->idle_calls = idle_calls;
 	ts->idle_sleeps = idle_sleeps;
 }
-- 
2.53.0

[PATCH 08/15] tick/sched: Remove nohz disabled special case in cputime fetch

[Frederic Weisbecker]

Even when nohz is not runtime enabled, the dynticks idle cputime
accounting can run and the common idle cputime accessors are still
relevant.

Remove the nohz disabled special case accordingly.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 kernel/time/tick-sched.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index c3efd3583cf9..cb235ec7d2d6 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -795,9 +795,6 @@ static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx,
 	ktime_t now, idle;
 	unsigned int seq;
 
-	if (!tick_nohz_active)
-		return -1;
-
 	now = ktime_get();
 	if (last_update_time)
 		*last_update_time = ktime_to_us(now);
@@ -839,7 +836,7 @@ static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx,
  * This time is measured via accounting rather than sampling,
  * and is as accurate as ktime_get() is.
  *
- * Return: -1 if NOHZ is not enabled, else total idle time of the @cpu
+ * Return: -1 if generic vtime is enabled, else total idle time of the @cpu
  */
 u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
 {
@@ -863,7 +860,7 @@ EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
  * This time is measured via accounting rather than sampling,
  * and is as accurate as ktime_get() is.
  *
- * Return: -1 if NOHZ is not enabled, else total iowait time of @cpu
+ * Return: -1 if generic vtime is enabled, else total iowait time of @cpu
  */
 u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
 {
-- 
2.53.0

[PATCH 09/15] tick/sched: Move dyntick-idle cputime accounting to cputime code

[Frederic Weisbecker]

Although the dynticks-idle cputime accounting is necessarily tied to
the tick subsystem, the actual related accounting code has no business
residing there and should be part of the scheduler cputime code.

Move away the relevant pieces and state machine to where they belong.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 include/linux/kernel_stat.h |  14 +++-
 kernel/sched/core.c         |   6 +-
 kernel/sched/cputime.c      | 148 ++++++++++++++++++++++++++++++--
 kernel/time/tick-sched.c    | 163 +++++++-----------------------------
 4 files changed, 188 insertions(+), 143 deletions(-)

diff --git a/include/linux/kernel_stat.h b/include/linux/kernel_stat.h
index ba65aad308a1..9343353ac7a3 100644
--- a/include/linux/kernel_stat.h
+++ b/include/linux/kernel_stat.h
@@ -35,9 +35,12 @@ enum cpu_usage_stat {
 
 struct kernel_cpustat {
 #ifdef CONFIG_NO_HZ_COMMON
-	int idle_dyntick;
+	bool		idle_dyntick;
+	bool		idle_elapse;
+	seqcount_t	idle_sleeptime_seq;
+	u64		idle_entrytime;
 #endif
-	u64 cpustat[NR_STATS];
+	u64		cpustat[NR_STATS];
 };
 
 struct kernel_stat {
@@ -103,8 +106,11 @@ static inline unsigned long kstat_cpu_irqs_sum(unsigned int cpu)
 }
 
 #ifdef CONFIG_NO_HZ_COMMON
-extern void kcpustat_dyntick_start(void);
-extern void kcpustat_dyntick_stop(void);
+extern void kcpustat_dyntick_start(u64 now);
+extern void kcpustat_dyntick_stop(u64 now);
+extern void kcpustat_irq_enter(u64 now);
+extern void kcpustat_irq_exit(u64 now);
+
 static inline bool kcpustat_idle_dyntick(void)
 {
 	return __this_cpu_read(kernel_cpustat.idle_dyntick);
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index da20fb6ea25a..0cfc027a955d 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -5517,7 +5517,11 @@ void sched_exec(void)
 }
 
 DEFINE_PER_CPU(struct kernel_stat, kstat);
-DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat);
+DEFINE_PER_CPU(struct kernel_cpustat, kernel_cpustat) = {
+#ifdef CONFIG_NO_HZ_COMMON
+	.idle_sleeptime_seq = SEQCNT_ZERO(kernel_cpustat.idle_sleeptime_seq)
+#endif
+};
 
 EXPORT_PER_CPU_SYMBOL(kstat);
 EXPORT_PER_CPU_SYMBOL(kernel_cpustat);
diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index a5733789e0bd..4c00163b74b9 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -2,6 +2,7 @@
 /*
  * Simple CPU accounting cgroup controller
  */
+#include <linux/sched/clock.h>
 #include <linux/sched/cputime.h>
 #include <linux/tsacct_kern.h>
 #include "sched.h"
@@ -420,22 +421,155 @@ static inline void irqtime_account_process_tick(struct task_struct *p, int user_
 #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
 
 #ifdef CONFIG_NO_HZ_COMMON
-void kcpustat_dyntick_start(void)
+static void kcpustat_idle_stop(struct kernel_cpustat *kc, u64 now)
 {
-	if (!vtime_generic_enabled_this_cpu()) {
-		vtime_dyntick_start();
-		__this_cpu_write(kernel_cpustat.idle_dyntick, 1);
-	}
+	u64 *cpustat = kc->cpustat;
+	u64 delta;
+
+	if (!kc->idle_elapse)
+		return;
+
+	delta = now - kc->idle_entrytime;
+
+	write_seqcount_begin(&kc->idle_sleeptime_seq);
+	if (nr_iowait_cpu(smp_processor_id()) > 0)
+		cpustat[CPUTIME_IOWAIT] += delta;
+	else
+		cpustat[CPUTIME_IDLE] += delta;
+
+	kc->idle_entrytime = now;
+	kc->idle_elapse = false;
+	write_seqcount_end(&kc->idle_sleeptime_seq);
 }
 
-void kcpustat_dyntick_stop(void)
+static void kcpustat_idle_start(struct kernel_cpustat *kc, u64 now)
 {
+	write_seqcount_begin(&kc->idle_sleeptime_seq);
+	kc->idle_entrytime = now;
+	kc->idle_elapse = true;
+	write_seqcount_end(&kc->idle_sleeptime_seq);
+}
+
+void kcpustat_dyntick_stop(u64 now)
+{
+	struct kernel_cpustat *kc = kcpustat_this_cpu;
+
 	if (!vtime_generic_enabled_this_cpu()) {
-		__this_cpu_write(kernel_cpustat.idle_dyntick, 0);
+		WARN_ON_ONCE(!kc->idle_dyntick);
+		kcpustat_idle_stop(kc, now);
+		kc->idle_dyntick = false;
 		vtime_dyntick_stop();
 		steal_account_process_time(ULONG_MAX);
 	}
 }
+
+void kcpustat_dyntick_start(u64 now)
+{
+	struct kernel_cpustat *kc = kcpustat_this_cpu;
+
+	if (!vtime_generic_enabled_this_cpu()) {
+		vtime_dyntick_start();
+		kc->idle_dyntick = true;
+		kcpustat_idle_start(kc, now);
+	}
+}
+
+void kcpustat_irq_enter(u64 now)
+{
+	struct kernel_cpustat *kc = kcpustat_this_cpu;
+
+	if (!vtime_generic_enabled_this_cpu())
+		kcpustat_idle_stop(kc, now);
+}
+
+void kcpustat_irq_exit(u64 now)
+{
+	struct kernel_cpustat *kc = kcpustat_this_cpu;
+
+	if (!vtime_generic_enabled_this_cpu())
+		kcpustat_idle_start(kc, now);
+}
+
+static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx,
+				 bool compute_delta, u64 *last_update_time)
+{
+	struct kernel_cpustat *kc = &kcpustat_cpu(cpu);
+	u64 *cpustat = kc->cpustat;
+	unsigned int seq;
+	ktime_t now;
+	u64 idle;
+
+	now = ktime_get();
+	if (last_update_time)
+		*last_update_time = ktime_to_us(now);
+
+	if (vtime_generic_enabled_cpu(cpu)) {
+		idle = kcpustat_field(idx, cpu);
+		goto to_us;
+	}
+
+	do {
+		seq = read_seqcount_begin(&kc->idle_sleeptime_seq);
+
+		idle = cpustat[idx];
+		if (kc->idle_elapse && compute_delta && now > kc->idle_entrytime)
+			idle += (now - kc->idle_entrytime);
+	} while (read_seqcount_retry(&kc->idle_sleeptime_seq, seq));
+
+to_us:
+	do_div(idle, NSEC_PER_USEC);
+
+	return idle;
+}
+
+/**
+ * get_cpu_idle_time_us - get the total idle time of a CPU
+ * @cpu: CPU number to query
+ * @last_update_time: variable to store update time in. Do not update
+ * counters if NULL.
+ *
+ * Return the cumulative idle time (since boot) for a given
+ * CPU, in microseconds. Note that this is partially broken due to
+ * the counter of iowait tasks that can be remotely updated without
+ * any synchronization. Therefore it is possible to observe backward
+ * values within two consecutive reads.
+ *
+ * This time is measured via accounting rather than sampling,
+ * and is as accurate as ktime_get() is.
+ *
+ * Return: -1 if generic vtime is enabled, else total idle time of the @cpu
+ */
+u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
+{
+	return get_cpu_sleep_time_us(cpu, CPUTIME_IDLE,
+				     !nr_iowait_cpu(cpu), last_update_time);
+}
+EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
+
+/**
+ * get_cpu_iowait_time_us - get the total iowait time of a CPU
+ * @cpu: CPU number to query
+ * @last_update_time: variable to store update time in. Do not update
+ * counters if NULL.
+ *
+ * Return the cumulative iowait time (since boot) for a given
+ * CPU, in microseconds. Note this is partially broken due to
+ * the counter of iowait tasks that can be remotely updated without
+ * any synchronization. Therefore it is possible to observe backward
+ * values within two consecutive reads.
+ *
+ * This time is measured via accounting rather than sampling,
+ * and is as accurate as ktime_get() is.
+ *
+ * Return: -1 if generic vtime is enabled, else total iowait time of @cpu
+ */
+u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
+{
+	return get_cpu_sleep_time_us(cpu, CPUTIME_IOWAIT,
+				     nr_iowait_cpu(cpu), last_update_time);
+}
+EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us);
+
 #endif /* CONFIG_NO_HZ_COMMON */
 
 /*
diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index cb235ec7d2d6..fa03cf7b3cec 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -749,126 +749,6 @@ static void tick_nohz_update_jiffies(ktime_t now)
 	touch_softlockup_watchdog_sched();
 }
 
-static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now)
-{
-	u64 *cpustat = kcpustat_this_cpu->cpustat;
-	ktime_t delta;
-
-	if (vtime_generic_enabled_this_cpu())
-		return;
-
-	if (WARN_ON_ONCE(!tick_sched_flag_test(ts, TS_FLAG_IDLE_ACTIVE)))
-		return;
-
-	delta = ktime_sub(now, ts->idle_entrytime);
-
-	write_seqcount_begin(&ts->idle_sleeptime_seq);
-	if (nr_iowait_cpu(smp_processor_id()) > 0)
-		cpustat[CPUTIME_IOWAIT] = ktime_add(cpustat[CPUTIME_IOWAIT], delta);
-	else
-		cpustat[CPUTIME_IDLE] = ktime_add(cpustat[CPUTIME_IDLE], delta);
-
-	ts->idle_entrytime = now;
-	tick_sched_flag_clear(ts, TS_FLAG_IDLE_ACTIVE);
-	write_seqcount_end(&ts->idle_sleeptime_seq);
-
-	sched_clock_idle_wakeup_event();
-}
-
-static void tick_nohz_start_idle(struct tick_sched *ts)
-{
-	if (vtime_generic_enabled_this_cpu())
-		return;
-
-	write_seqcount_begin(&ts->idle_sleeptime_seq);
-	ts->idle_entrytime = ktime_get();
-	tick_sched_flag_set(ts, TS_FLAG_IDLE_ACTIVE);
-	write_seqcount_end(&ts->idle_sleeptime_seq);
-	sched_clock_idle_sleep_event();
-}
-
-static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx,
-				 bool compute_delta, u64 *last_update_time)
-{
-	struct tick_sched *ts = &per_cpu(tick_cpu_sched, cpu);
-	u64 *cpustat = kcpustat_cpu(cpu).cpustat;
-	ktime_t now, idle;
-	unsigned int seq;
-
-	now = ktime_get();
-	if (last_update_time)
-		*last_update_time = ktime_to_us(now);
-
-	if (vtime_generic_enabled_cpu(cpu)) {
-		idle = kcpustat_field(idx, cpu);
-		return ktime_to_us(idle);
-	}
-
-	do {
-		ktime_t delta = 0;
-
-		seq = read_seqcount_begin(&ts->idle_sleeptime_seq);
-
-		if (tick_sched_flag_test(ts, TS_FLAG_IDLE_ACTIVE) && compute_delta) {
-			if (now > ts->idle_entrytime)
-				delta = ktime_sub(now, ts->idle_entrytime);
-		}
-
-		idle = ktime_add(cpustat[idx], delta);
-	} while (read_seqcount_retry(&ts->idle_sleeptime_seq, seq));
-
-	return ktime_to_us(idle);
-
-}
-
-/**
- * get_cpu_idle_time_us - get the total idle time of a CPU
- * @cpu: CPU number to query
- * @last_update_time: variable to store update time in. Do not update
- * counters if NULL.
- *
- * Return the cumulative idle time (since boot) for a given
- * CPU, in microseconds. Note that this is partially broken due to
- * the counter of iowait tasks that can be remotely updated without
- * any synchronization. Therefore it is possible to observe backward
- * values within two consecutive reads.
- *
- * This time is measured via accounting rather than sampling,
- * and is as accurate as ktime_get() is.
- *
- * Return: -1 if generic vtime is enabled, else total idle time of the @cpu
- */
-u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
-{
-	return get_cpu_sleep_time_us(cpu, CPUTIME_IDLE,
-				     !nr_iowait_cpu(cpu), last_update_time);
-}
-EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
-
-/**
- * get_cpu_iowait_time_us - get the total iowait time of a CPU
- * @cpu: CPU number to query
- * @last_update_time: variable to store update time in. Do not update
- * counters if NULL.
- *
- * Return the cumulative iowait time (since boot) for a given
- * CPU, in microseconds. Note this is partially broken due to
- * the counter of iowait tasks that can be remotely updated without
- * any synchronization. Therefore it is possible to observe backward
- * values within two consecutive reads.
- *
- * This time is measured via accounting rather than sampling,
- * and is as accurate as ktime_get() is.
- *
- * Return: -1 if generic vtime is enabled, else total iowait time of @cpu
- */
-u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
-{
-	return get_cpu_sleep_time_us(cpu, CPUTIME_IOWAIT,
-				     nr_iowait_cpu(cpu), last_update_time);
-}
-EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us);
-
 /* Simplified variant of hrtimer_forward_now() */
 static ktime_t tick_forward_now(ktime_t expires, ktime_t now)
 {
@@ -1289,6 +1169,20 @@ void tick_nohz_idle_retain_tick(void)
 	tick_nohz_retain_tick(this_cpu_ptr(&tick_cpu_sched));
 }
 
+static void tick_nohz_clock_sleep(struct tick_sched *ts)
+{
+	tick_sched_flag_set(ts, TS_FLAG_IDLE_ACTIVE);
+	sched_clock_idle_sleep_event();
+}
+
+static void tick_nohz_clock_wakeup(struct tick_sched *ts)
+{
+	if (tick_sched_flag_test(ts, TS_FLAG_IDLE_ACTIVE)) {
+		tick_sched_flag_clear(ts, TS_FLAG_IDLE_ACTIVE);
+		sched_clock_idle_wakeup_event();
+	}
+}
+
 /**
  * tick_nohz_idle_enter - prepare for entering idle on the current CPU
  *
@@ -1303,12 +1197,11 @@ void tick_nohz_idle_enter(void)
 	local_irq_disable();
 
 	ts = this_cpu_ptr(&tick_cpu_sched);
-
 	WARN_ON_ONCE(ts->timer_expires_base);
-
 	tick_sched_flag_set(ts, TS_FLAG_INIDLE);
-	kcpustat_dyntick_start();
-	tick_nohz_start_idle(ts);
+	ts->idle_entrytime = ktime_get();
+	kcpustat_dyntick_start(ts->idle_entrytime);
+	tick_nohz_clock_sleep(ts);
 
 	local_irq_enable();
 }
@@ -1336,10 +1229,13 @@ void tick_nohz_irq_exit(void)
 {
 	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 
-	if (tick_sched_flag_test(ts, TS_FLAG_INIDLE))
-		tick_nohz_start_idle(ts);
-	else
+	if (tick_sched_flag_test(ts, TS_FLAG_INIDLE)) {
+		ts->idle_entrytime = ktime_get();
+		kcpustat_irq_exit(ts->idle_entrytime);
+		tick_nohz_clock_sleep(ts);
+	} else {
 		tick_nohz_full_update_tick(ts);
+	}
 }
 
 /**
@@ -1484,11 +1380,11 @@ void tick_nohz_idle_exit(void)
 		now = ktime_get();
 
 	if (idle_active)
-		tick_nohz_stop_idle(ts, now);
+		tick_nohz_clock_wakeup(ts);
 
 	if (tick_stopped)
 		tick_nohz_idle_update_tick(ts, now);
-	kcpustat_dyntick_stop();
+	kcpustat_dyntick_stop(now);
 
 	local_irq_enable();
 }
@@ -1545,9 +1441,14 @@ static inline void tick_nohz_irq_enter(void)
 
 	if (!tick_sched_flag_test(ts, TS_FLAG_STOPPED | TS_FLAG_IDLE_ACTIVE))
 		return;
+
 	now = ktime_get();
-	if (tick_sched_flag_test(ts, TS_FLAG_IDLE_ACTIVE))
-		tick_nohz_stop_idle(ts, now);
+
+	if (tick_sched_flag_test(ts, TS_FLAG_IDLE_ACTIVE)) {
+		tick_nohz_clock_wakeup(ts);
+		kcpustat_irq_enter(now);
+	}
+
 	/*
 	 * If all CPUs are idle we may need to update a stale jiffies value.
 	 * Note nohz_full is a special case: a timekeeper is guaranteed to stay
-- 
2.53.0

[PATCH 10/15] tick/sched: Remove unused fields

[Frederic Weisbecker]

Remove fields after the dyntick-idle cputime migration to scheduler
code.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 kernel/time/tick-sched.h       | 12 ------------
 kernel/time/timer_list.c       |  6 +-----
 scripts/gdb/linux/timerlist.py |  4 ----
 3 files changed, 1 insertion(+), 21 deletions(-)

diff --git a/kernel/time/tick-sched.h b/kernel/time/tick-sched.h
index b4a7822f495d..79b9252047b1 100644
--- a/kernel/time/tick-sched.h
+++ b/kernel/time/tick-sched.h
@@ -44,9 +44,7 @@ struct tick_device {
  *			to resume the tick timer operation in the timeline
  *			when the CPU returns from nohz sleep.
  * @next_tick:		Next tick to be fired when in dynticks mode.
- * @idle_jiffies:	jiffies at the entry to idle for idle time accounting
  * @idle_waketime:	Time when the idle was interrupted
- * @idle_sleeptime_seq:	sequence counter for data consistency
  * @idle_entrytime:	Time when the idle call was entered
  * @last_jiffies:	Base jiffies snapshot when next event was last computed
  * @timer_expires_base:	Base time clock monotonic for @timer_expires
@@ -55,9 +53,6 @@ struct tick_device {
  * @idle_expires:	Next tick in idle, for debugging purpose only
  * @idle_calls:		Total number of idle calls
  * @idle_sleeps:	Number of idle calls, where the sched tick was stopped
- * @idle_exittime:	Time when the idle state was left
- * @idle_sleeptime:	Sum of the time slept in idle with sched tick stopped
- * @iowait_sleeptime:	Sum of the time slept in idle with sched tick stopped, with IO outstanding
  * @tick_dep_mask:	Tick dependency mask - is set, if someone needs the tick
  * @check_clocks:	Notification mechanism about clocksource changes
  */
@@ -73,12 +68,10 @@ struct tick_sched {
 	struct hrtimer			sched_timer;
 	ktime_t				last_tick;
 	ktime_t				next_tick;
-	unsigned long			idle_jiffies;
 	ktime_t				idle_waketime;
 	unsigned int			got_idle_tick;
 
 	/* Idle entry */
-	seqcount_t			idle_sleeptime_seq;
 	ktime_t				idle_entrytime;
 
 	/* Tick stop */
@@ -90,11 +83,6 @@ struct tick_sched {
 	unsigned long			idle_calls;
 	unsigned long			idle_sleeps;
 
-	/* Idle exit */
-	ktime_t				idle_exittime;
-	ktime_t				idle_sleeptime;
-	ktime_t				iowait_sleeptime;
-
 	/* Full dynticks handling */
 	atomic_t			tick_dep_mask;
 
diff --git a/kernel/time/timer_list.c b/kernel/time/timer_list.c
index 427d7ddea3af..514802def1e0 100644
--- a/kernel/time/timer_list.c
+++ b/kernel/time/timer_list.c
@@ -152,14 +152,10 @@ static void print_cpu(struct seq_file *m, int cpu, u64 now)
 		P_flag(highres, TS_FLAG_HIGHRES);
 		P_ns(last_tick);
 		P_flag(tick_stopped, TS_FLAG_STOPPED);
-		P(idle_jiffies);
 		P(idle_calls);
 		P(idle_sleeps);
 		P_ns(idle_entrytime);
 		P_ns(idle_waketime);
-		P_ns(idle_exittime);
-		P_ns(idle_sleeptime);
-		P_ns(iowait_sleeptime);
 		P(last_jiffies);
 		P(next_timer);
 		P_ns(idle_expires);
@@ -256,7 +252,7 @@ static void timer_list_show_tickdevices_header(struct seq_file *m)
 
 static inline void timer_list_header(struct seq_file *m, u64 now)
 {
-	SEQ_printf(m, "Timer List Version: v0.10\n");
+	SEQ_printf(m, "Timer List Version: v0.11\n");
 	SEQ_printf(m, "HRTIMER_MAX_CLOCK_BASES: %d\n", HRTIMER_MAX_CLOCK_BASES);
 	SEQ_printf(m, "now at %Ld nsecs\n", (unsigned long long)now);
 	SEQ_printf(m, "\n");
diff --git a/scripts/gdb/linux/timerlist.py b/scripts/gdb/linux/timerlist.py
index 9fb3436a217c..744b032e4d38 100644
--- a/scripts/gdb/linux/timerlist.py
+++ b/scripts/gdb/linux/timerlist.py
@@ -90,14 +90,10 @@ def print_cpu(hrtimer_bases, cpu, max_clock_bases):
             text += f"  .{'nohz':15s}: {int(bool(ts['flags'] & TS_FLAG_NOHZ))}\n"
             text += f"  .{'last_tick':15s}: {ts['last_tick']}\n"
             text += f"  .{'tick_stopped':15s}: {int(bool(ts['flags'] & TS_FLAG_STOPPED))}\n"
-            text += f"  .{'idle_jiffies':15s}: {ts['idle_jiffies']}\n"
             text += f"  .{'idle_calls':15s}: {ts['idle_calls']}\n"
             text += f"  .{'idle_sleeps':15s}: {ts['idle_sleeps']}\n"
             text += f"  .{'idle_entrytime':15s}: {ts['idle_entrytime']} nsecs\n"
             text += f"  .{'idle_waketime':15s}: {ts['idle_waketime']} nsecs\n"
-            text += f"  .{'idle_exittime':15s}: {ts['idle_exittime']} nsecs\n"
-            text += f"  .{'idle_sleeptime':15s}: {ts['idle_sleeptime']} nsecs\n"
-            text += f"  .{'iowait_sleeptime':15s}: {ts['iowait_sleeptime']} nsecs\n"
             text += f"  .{'last_jiffies':15s}: {ts['last_jiffies']}\n"
             text += f"  .{'next_timer':15s}: {ts['next_timer']}\n"
             text += f"  .{'idle_expires':15s}: {ts['idle_expires']} nsecs\n"
-- 
2.53.0

[PATCH 11/15] tick/sched: Account tickless idle cputime only when tick is stopped

[Frederic Weisbecker]

There is no real point in switching to dyntick-idle cputime accounting
mode if the tick is not actually stopped. This just adds overhead,
notably fetching the GTOD, on each idle exit and each idle IRQ entry for
no reason during short idle trips.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 kernel/time/tick-sched.c | 50 +++++++++++++++++++++-------------------
 1 file changed, 26 insertions(+), 24 deletions(-)

diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index fa03cf7b3cec..c1ee0b256445 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -1157,8 +1157,10 @@ void tick_nohz_idle_stop_tick(void)
 		ts->idle_sleeps++;
 		ts->idle_expires = expires;
 
-		if (!was_stopped && tick_sched_flag_test(ts, TS_FLAG_STOPPED))
+		if (!was_stopped && tick_sched_flag_test(ts, TS_FLAG_STOPPED)) {
+			kcpustat_dyntick_start(ts->idle_entrytime);
 			nohz_balance_enter_idle(cpu);
+		}
 	} else {
 		tick_nohz_retain_tick(ts);
 	}
@@ -1200,7 +1202,6 @@ void tick_nohz_idle_enter(void)
 	WARN_ON_ONCE(ts->timer_expires_base);
 	tick_sched_flag_set(ts, TS_FLAG_INIDLE);
 	ts->idle_entrytime = ktime_get();
-	kcpustat_dyntick_start(ts->idle_entrytime);
 	tick_nohz_clock_sleep(ts);
 
 	local_irq_enable();
@@ -1230,9 +1231,10 @@ void tick_nohz_irq_exit(void)
 	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 
 	if (tick_sched_flag_test(ts, TS_FLAG_INIDLE)) {
-		ts->idle_entrytime = ktime_get();
-		kcpustat_irq_exit(ts->idle_entrytime);
 		tick_nohz_clock_sleep(ts);
+		ts->idle_entrytime = ktime_get();
+		if (tick_sched_flag_test(ts, TS_FLAG_STOPPED))
+			kcpustat_irq_exit(ts->idle_entrytime);
 	} else {
 		tick_nohz_full_update_tick(ts);
 	}
@@ -1333,8 +1335,17 @@ void tick_nohz_idle_restart_tick(void)
 {
 	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 
-	if (tick_sched_flag_test(ts, TS_FLAG_STOPPED))
-		tick_nohz_restart_sched_tick(ts, ktime_get());
+	if (tick_sched_flag_test(ts, TS_FLAG_STOPPED)) {
+		/*
+		 * Update entrytime here in case the tick restart is due to temporary
+		 * polling on forced broadcast. The tick may be stopped again later within
+		 * the same idle trip. The idle_entrytime was updated recently but make sure
+		 * no tiny amount of idle time is accounted twice.
+		 */
+		ts->idle_entrytime = ktime_get();
+		kcpustat_dyntick_stop(ts->idle_entrytime);
+		tick_nohz_restart_sched_tick(ts, ts->idle_entrytime);
+	}
 }
 
 static void tick_nohz_idle_update_tick(struct tick_sched *ts, ktime_t now)
@@ -1364,7 +1375,6 @@ static void tick_nohz_idle_update_tick(struct tick_sched *ts, ktime_t now)
 void tick_nohz_idle_exit(void)
 {
 	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
-	bool idle_active, tick_stopped;
 	ktime_t now;
 
 	local_irq_disable();
@@ -1373,18 +1383,13 @@ void tick_nohz_idle_exit(void)
 	WARN_ON_ONCE(ts->timer_expires_base);
 
 	tick_sched_flag_clear(ts, TS_FLAG_INIDLE);
-	idle_active = tick_sched_flag_test(ts, TS_FLAG_IDLE_ACTIVE);
-	tick_stopped = tick_sched_flag_test(ts, TS_FLAG_STOPPED);
+	tick_nohz_clock_wakeup(ts);
 
-	if (idle_active || tick_stopped)
+	if (tick_sched_flag_test(ts, TS_FLAG_STOPPED)) {
 		now = ktime_get();
-
-	if (idle_active)
-		tick_nohz_clock_wakeup(ts);
-
-	if (tick_stopped)
+		kcpustat_dyntick_stop(now);
 		tick_nohz_idle_update_tick(ts, now);
-	kcpustat_dyntick_stop(now);
+	}
 
 	local_irq_enable();
 }
@@ -1439,15 +1444,13 @@ static inline void tick_nohz_irq_enter(void)
 	struct tick_sched *ts = this_cpu_ptr(&tick_cpu_sched);
 	ktime_t now;
 
-	if (!tick_sched_flag_test(ts, TS_FLAG_STOPPED | TS_FLAG_IDLE_ACTIVE))
+	tick_nohz_clock_wakeup(ts);
+
+	if (!tick_sched_flag_test(ts, TS_FLAG_STOPPED))
 		return;
 
 	now = ktime_get();
-
-	if (tick_sched_flag_test(ts, TS_FLAG_IDLE_ACTIVE)) {
-		tick_nohz_clock_wakeup(ts);
-		kcpustat_irq_enter(now);
-	}
+	kcpustat_irq_enter(now);
 
 	/*
 	 * If all CPUs are idle we may need to update a stale jiffies value.
@@ -1456,8 +1459,7 @@ static inline void tick_nohz_irq_enter(void)
 	 * rare case (typically stop machine). So we must make sure we have a
 	 * last resort.
 	 */
-	if (tick_sched_flag_test(ts, TS_FLAG_STOPPED))
-		tick_nohz_update_jiffies(now);
+	tick_nohz_update_jiffies(now);
 }
 
 #else
-- 
2.53.0

[PATCH 12/15] tick/sched: Consolidate idle time fetching APIs

[Frederic Weisbecker]

Fetching the idle cputime is available through a variety of accessors
all over the place depending on the different accounting flavours and
needs:

- idle vtime generic accounting can be accessed by kcpustat_field(),
  kcpustat_cpu_fetch(), get_idle/iowait_time() and
  get_cpu_idle/iowait_time_us()

- dynticks-idle accounting can only be accessed by get_idle/iowait_time()
  or get_cpu_idle/iowait_time_us()

- CONFIG_NO_HZ_COMMON=n idle accounting can be accessed by kcpustat_field()
  kcpustat_cpu_fetch(), or get_idle/iowait_time() but not by
  get_cpu_idle/iowait_time_us()

Moreover get_idle/iowait_time() relies on get_cpu_idle/iowait_time_us()
with a non-sensical conversion to microseconds and back to nanoseconds
on the way.

Start consolidating the APIs with removing get_idle/iowait_time() and
make kcpustat_field() and kcpustat_cpu_fetch() work for all cases.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 fs/proc/stat.c              | 40 +++---------------------
 fs/proc/uptime.c            |  8 ++---
 include/linux/kernel_stat.h | 34 ++++++++++++++++++---
 kernel/sched/cputime.c      | 61 ++++++++++++++++++++++++-------------
 4 files changed, 76 insertions(+), 67 deletions(-)

diff --git a/fs/proc/stat.c b/fs/proc/stat.c
index 8b444e862319..c00468a83f64 100644
--- a/fs/proc/stat.c
+++ b/fs/proc/stat.c
@@ -22,38 +22,6 @@
 #define arch_irq_stat() 0
 #endif
 
-u64 get_idle_time(struct kernel_cpustat *kcs, int cpu)
-{
-	u64 idle, idle_usecs = -1ULL;
-
-	if (cpu_online(cpu))
-		idle_usecs = get_cpu_idle_time_us(cpu, NULL);
-
-	if (idle_usecs == -1ULL)
-		/* !NO_HZ or cpu offline so we can rely on cpustat.idle */
-		idle = kcs->cpustat[CPUTIME_IDLE];
-	else
-		idle = idle_usecs * NSEC_PER_USEC;
-
-	return idle;
-}
-
-static u64 get_iowait_time(struct kernel_cpustat *kcs, int cpu)
-{
-	u64 iowait, iowait_usecs = -1ULL;
-
-	if (cpu_online(cpu))
-		iowait_usecs = get_cpu_iowait_time_us(cpu, NULL);
-
-	if (iowait_usecs == -1ULL)
-		/* !NO_HZ or cpu offline so we can rely on cpustat.iowait */
-		iowait = kcs->cpustat[CPUTIME_IOWAIT];
-	else
-		iowait = iowait_usecs * NSEC_PER_USEC;
-
-	return iowait;
-}
-
 static void show_irq_gap(struct seq_file *p, unsigned int gap)
 {
 	static const char zeros[] = " 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0";
@@ -105,8 +73,8 @@ static int show_stat(struct seq_file *p, void *v)
 		user		+= cpustat[CPUTIME_USER];
 		nice		+= cpustat[CPUTIME_NICE];
 		system		+= cpustat[CPUTIME_SYSTEM];
-		idle		+= get_idle_time(&kcpustat, i);
-		iowait		+= get_iowait_time(&kcpustat, i);
+		idle		+= cpustat[CPUTIME_IDLE];
+		iowait		+= cpustat[CPUTIME_IOWAIT];
 		irq		+= cpustat[CPUTIME_IRQ];
 		softirq		+= cpustat[CPUTIME_SOFTIRQ];
 		steal		+= cpustat[CPUTIME_STEAL];
@@ -146,8 +114,8 @@ static int show_stat(struct seq_file *p, void *v)
 		user		= cpustat[CPUTIME_USER];
 		nice		= cpustat[CPUTIME_NICE];
 		system		= cpustat[CPUTIME_SYSTEM];
-		idle		= get_idle_time(&kcpustat, i);
-		iowait		= get_iowait_time(&kcpustat, i);
+		idle		= cpustat[CPUTIME_IDLE];
+		iowait		= cpustat[CPUTIME_IOWAIT];
 		irq		= cpustat[CPUTIME_IRQ];
 		softirq		= cpustat[CPUTIME_SOFTIRQ];
 		steal		= cpustat[CPUTIME_STEAL];
diff --git a/fs/proc/uptime.c b/fs/proc/uptime.c
index b5343d209381..433aa947cd57 100644
--- a/fs/proc/uptime.c
+++ b/fs/proc/uptime.c
@@ -18,12 +18,8 @@ static int uptime_proc_show(struct seq_file *m, void *v)
 	int i;
 
 	idle_nsec = 0;
-	for_each_possible_cpu(i) {
-		struct kernel_cpustat kcs;
-
-		kcpustat_cpu_fetch(&kcs, i);
-		idle_nsec += get_idle_time(&kcs, i);
-	}
+	for_each_possible_cpu(i)
+		idle_nsec += kcpustat_field(CPUTIME_IDLE, i);
 
 	ktime_get_boottime_ts64(&uptime);
 	timens_add_boottime(&uptime);
diff --git a/include/linux/kernel_stat.h b/include/linux/kernel_stat.h
index 9343353ac7a3..3680519d7b2c 100644
--- a/include/linux/kernel_stat.h
+++ b/include/linux/kernel_stat.h
@@ -110,32 +110,59 @@ extern void kcpustat_dyntick_start(u64 now);
 extern void kcpustat_dyntick_stop(u64 now);
 extern void kcpustat_irq_enter(u64 now);
 extern void kcpustat_irq_exit(u64 now);
+extern u64 kcpustat_field_idle(int cpu);
+extern u64 kcpustat_field_iowait(int cpu);
 
 static inline bool kcpustat_idle_dyntick(void)
 {
 	return __this_cpu_read(kernel_cpustat.idle_dyntick);
 }
 #else
+static inline u64 kcpustat_field_idle(int cpu)
+{
+	return kcpustat_cpu(cpu).cpustat[CPUTIME_IDLE];
+}
+static inline u64 kcpustat_field_iowait(int cpu)
+{
+	return kcpustat_cpu(cpu).cpustat[CPUTIME_IOWAIT];
+}
+
 static inline bool kcpustat_idle_dyntick(void)
 {
 	return false;
 }
 #endif /* CONFIG_NO_HZ_COMMON */
 
+/* Fetch cputime values when vtime is disabled on a CPU */
+static inline u64 kcpustat_field_default(enum cpu_usage_stat usage, int cpu)
+{
+	if (usage == CPUTIME_IDLE)
+		return kcpustat_field_idle(cpu);
+	if (usage == CPUTIME_IOWAIT)
+		return kcpustat_field_iowait(cpu);
+	return kcpustat_cpu(cpu).cpustat[usage];
+}
+
+static inline void kcpustat_cpu_fetch_default(struct kernel_cpustat *dst, int cpu)
+{
+	*dst = kcpustat_cpu(cpu);
+	dst->cpustat[CPUTIME_IDLE] = kcpustat_field_idle(cpu);
+	dst->cpustat[CPUTIME_IOWAIT] = kcpustat_field_iowait(cpu);
+}
+
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
 extern u64 kcpustat_field(enum cpu_usage_stat usage, int cpu);
 extern void kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu);
 #else
 static inline u64 kcpustat_field(enum cpu_usage_stat usage, int cpu)
 {
-	return kcpustat_cpu(cpu).cpustat[usage];
+	return kcpustat_field_default(usage, cpu);
 }
 
 static inline void kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu)
 {
-	*dst = kcpustat_cpu(cpu);
+	kcpustat_cpu_fetch_default(dst, cpu);
 }
-
 #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_GEN */
 
 extern void account_user_time(struct task_struct *, u64);
@@ -145,7 +172,6 @@ extern void account_system_index_time(struct task_struct *, u64,
 				      enum cpu_usage_stat);
 extern void account_steal_time(u64);
 extern void account_idle_time(u64);
-extern u64 get_idle_time(struct kernel_cpustat *kcs, int cpu);
 
 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
 static inline void account_process_tick(struct task_struct *tsk, int user)
diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index 4c00163b74b9..c91fd67f93ea 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -490,24 +490,14 @@ void kcpustat_irq_exit(u64 now)
 		kcpustat_idle_start(kc, now);
 }
 
-static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx,
-				 bool compute_delta, u64 *last_update_time)
+static u64 kcpustat_field_dyntick(int cpu, enum cpu_usage_stat idx,
+				  bool compute_delta, u64 now)
 {
 	struct kernel_cpustat *kc = &kcpustat_cpu(cpu);
 	u64 *cpustat = kc->cpustat;
 	unsigned int seq;
-	ktime_t now;
 	u64 idle;
 
-	now = ktime_get();
-	if (last_update_time)
-		*last_update_time = ktime_to_us(now);
-
-	if (vtime_generic_enabled_cpu(cpu)) {
-		idle = kcpustat_field(idx, cpu);
-		goto to_us;
-	}
-
 	do {
 		seq = read_seqcount_begin(&kc->idle_sleeptime_seq);
 
@@ -516,12 +506,42 @@ static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx,
 			idle += (now - kc->idle_entrytime);
 	} while (read_seqcount_retry(&kc->idle_sleeptime_seq, seq));
 
-to_us:
-	do_div(idle, NSEC_PER_USEC);
-
 	return idle;
 }
 
+u64 kcpustat_field_idle(int cpu)
+{
+	return kcpustat_field_dyntick(cpu, CPUTIME_IDLE,
+				      !nr_iowait_cpu(cpu), ktime_get());
+}
+EXPORT_SYMBOL_GPL(kcpustat_field_idle);
+
+u64 kcpustat_field_iowait(int cpu)
+{
+	return kcpustat_field_dyntick(cpu, CPUTIME_IOWAIT,
+				      nr_iowait_cpu(cpu), ktime_get());
+}
+EXPORT_SYMBOL_GPL(kcpustat_field_iowait);
+
+static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx,
+				 bool compute_delta, u64 *last_update_time)
+{
+	ktime_t now = ktime_get();
+	u64 res;
+
+	if (vtime_generic_enabled_cpu(cpu))
+		res = kcpustat_field(idx, cpu);
+	else
+		res = kcpustat_field_dyntick(cpu, idx, compute_delta, now);
+
+	do_div(res, NSEC_PER_USEC);
+
+	if (last_update_time)
+		*last_update_time = ktime_to_us(now);
+
+	return res;
+}
+
 /**
  * get_cpu_idle_time_us - get the total idle time of a CPU
  * @cpu: CPU number to query
@@ -569,7 +589,6 @@ u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
 				     nr_iowait_cpu(cpu), last_update_time);
 }
 EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us);
-
 #endif /* CONFIG_NO_HZ_COMMON */
 
 /*
@@ -1123,8 +1142,8 @@ u64 kcpustat_field(enum cpu_usage_stat usage, int cpu)
 	struct rq *rq;
 	int err;
 
-	if (!vtime_accounting_enabled_cpu(cpu))
-		return val;
+	if (!vtime_generic_enabled_cpu(cpu))
+		return kcpustat_field_default(usage, cpu);
 
 	rq = cpu_rq(cpu);
 
@@ -1219,8 +1238,8 @@ void kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu)
 	struct rq *rq;
 	int err;
 
-	if (!vtime_accounting_enabled_cpu(cpu)) {
-		*dst = *src;
+	if (!vtime_generic_enabled_cpu(cpu)) {
+		kcpustat_cpu_fetch_default(dst, cpu);
 		return;
 	}
 
@@ -1233,7 +1252,7 @@ void kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu)
 		curr = rcu_dereference(rq->curr);
 		if (WARN_ON_ONCE(!curr)) {
 			rcu_read_unlock();
-			*dst = *src;
+			kcpustat_cpu_fetch_default(dst, cpu);
 			return;
 		}
 
-- 
2.53.0

[PATCH 13/15] sched/cputime: Provide get_cpu_[idle|iowait]_time_us() off-case

[Frederic Weisbecker]

The last reason why get_cpu_idle/iowait_time_us() may return -1 now is
if the config doesn't support nohz.

The ad-hoc replacement solution by cpufreq is to compute jiffies minus
the whole busy cputime. Although the intention should provide a coherent
low resolution estimation of the idle and iowait time, the
implementation is buggy because jiffies don't start at 0.

Just provide instead a real get_cpu_[idle|iowait]_time_us() offcase.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 drivers/cpufreq/cpufreq.c   | 29 +----------------------------
 include/linux/kernel_stat.h |  3 +++
 include/linux/tick.h        |  4 ----
 kernel/sched/cputime.c      | 12 +++++++++---
 4 files changed, 13 insertions(+), 35 deletions(-)

diff --git a/drivers/cpufreq/cpufreq.c b/drivers/cpufreq/cpufreq.c
index 44eb1b7e7fc1..dda0d34d3c02 100644
--- a/drivers/cpufreq/cpufreq.c
+++ b/drivers/cpufreq/cpufreq.c
@@ -130,38 +130,11 @@ struct kobject *get_governor_parent_kobj(struct cpufreq_policy *policy)
 }
 EXPORT_SYMBOL_GPL(get_governor_parent_kobj);
 
-static inline u64 get_cpu_idle_time_jiffy(unsigned int cpu, u64 *wall)
-{
-	struct kernel_cpustat kcpustat;
-	u64 cur_wall_time;
-	u64 idle_time;
-	u64 busy_time;
-
-	cur_wall_time = jiffies64_to_nsecs(get_jiffies_64());
-
-	kcpustat_cpu_fetch(&kcpustat, cpu);
-
-	busy_time = kcpustat.cpustat[CPUTIME_USER];
-	busy_time += kcpustat.cpustat[CPUTIME_SYSTEM];
-	busy_time += kcpustat.cpustat[CPUTIME_IRQ];
-	busy_time += kcpustat.cpustat[CPUTIME_SOFTIRQ];
-	busy_time += kcpustat.cpustat[CPUTIME_STEAL];
-	busy_time += kcpustat.cpustat[CPUTIME_NICE];
-
-	idle_time = cur_wall_time - busy_time;
-	if (wall)
-		*wall = div_u64(cur_wall_time, NSEC_PER_USEC);
-
-	return div_u64(idle_time, NSEC_PER_USEC);
-}
-
 u64 get_cpu_idle_time(unsigned int cpu, u64 *wall, int io_busy)
 {
 	u64 idle_time = get_cpu_idle_time_us(cpu, io_busy ? wall : NULL);
 
-	if (idle_time == -1ULL)
-		return get_cpu_idle_time_jiffy(cpu, wall);
-	else if (!io_busy)
+	if (!io_busy)
 		idle_time += get_cpu_iowait_time_us(cpu, wall);
 
 	return idle_time;
diff --git a/include/linux/kernel_stat.h b/include/linux/kernel_stat.h
index 3680519d7b2c..512104b0ff49 100644
--- a/include/linux/kernel_stat.h
+++ b/include/linux/kernel_stat.h
@@ -133,6 +133,9 @@ static inline bool kcpustat_idle_dyntick(void)
 }
 #endif /* CONFIG_NO_HZ_COMMON */
 
+extern u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time);
+extern u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time);
+
 /* Fetch cputime values when vtime is disabled on a CPU */
 static inline u64 kcpustat_field_default(enum cpu_usage_stat usage, int cpu)
 {
diff --git a/include/linux/tick.h b/include/linux/tick.h
index 738007d6f577..1cf4651f09ad 100644
--- a/include/linux/tick.h
+++ b/include/linux/tick.h
@@ -139,8 +139,6 @@ extern bool tick_nohz_idle_got_tick(void);
 extern ktime_t tick_nohz_get_next_hrtimer(void);
 extern ktime_t tick_nohz_get_sleep_length(ktime_t *delta_next);
 extern unsigned long tick_nohz_get_idle_calls_cpu(int cpu);
-extern u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time);
-extern u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time);
 #else /* !CONFIG_NO_HZ_COMMON */
 #define tick_nohz_enabled (0)
 static inline bool tick_nohz_is_active(void) { return false; }
@@ -162,8 +160,6 @@ static inline ktime_t tick_nohz_get_sleep_length(ktime_t *delta_next)
 	*delta_next = TICK_NSEC;
 	return *delta_next;
 }
-static inline u64 get_cpu_idle_time_us(int cpu, u64 *unused) { return -1; }
-static inline u64 get_cpu_iowait_time_us(int cpu, u64 *unused) { return -1; }
 #endif /* !CONFIG_NO_HZ_COMMON */
 
 /*
diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index c91fd67f93ea..335d2c127763 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -522,6 +522,13 @@ u64 kcpustat_field_iowait(int cpu)
 				      nr_iowait_cpu(cpu), ktime_get());
 }
 EXPORT_SYMBOL_GPL(kcpustat_field_iowait);
+#else
+static u64 kcpustat_field_dyntick(int cpu, enum cpu_usage_stat idx,
+				  bool compute_delta, ktime_t now)
+{
+	return kcpustat_cpu(cpu).cpustat[idx];
+}
+#endif /* CONFIG_NO_HZ_COMMON */
 
 static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx,
 				 bool compute_delta, u64 *last_update_time)
@@ -557,7 +564,7 @@ static u64 get_cpu_sleep_time_us(int cpu, enum cpu_usage_stat idx,
  * This time is measured via accounting rather than sampling,
  * and is as accurate as ktime_get() is.
  *
- * Return: -1 if generic vtime is enabled, else total idle time of the @cpu
+ * Return: total idle time of the @cpu
  */
 u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
 {
@@ -581,7 +588,7 @@ EXPORT_SYMBOL_GPL(get_cpu_idle_time_us);
  * This time is measured via accounting rather than sampling,
  * and is as accurate as ktime_get() is.
  *
- * Return: -1 if generic vtime is enabled, else total iowait time of @cpu
+ * Return: total iowait time of @cpu
  */
 u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
 {
@@ -589,7 +596,6 @@ u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
 				     nr_iowait_cpu(cpu), last_update_time);
 }
 EXPORT_SYMBOL_GPL(get_cpu_iowait_time_us);
-#endif /* CONFIG_NO_HZ_COMMON */
 
 /*
  * Use precise platform statistics if available:
-- 
2.53.0

[PATCH 14/15] sched/cputime: Handle idle irqtime gracefully

[Frederic Weisbecker]

The dyntick-idle cputime accounting always assumes that IRQ time
accounting is enabled and consequently stops elapsing the idle time
during dyntick-idle IRQs.

This doesn't mix up well with disabled IRQ time accounting because then
idle IRQs become a cputime blind-spot. Also this feature is disabled
on most configurations and the overhead of pausing dyntick-idle
accounting while in idle IRQs could then be avoided.

Fix the situation with conditionally pausing dyntick-idle accounting
during idle IRQs only iff either native vtime (which does IRQ time
accounting) or generic IRQ time accounting are enabled.

Also make sure that the accumulated IRQ time is not accidentally
substracted from later accounting.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 kernel/sched/cputime.c | 20 +++++++++++++++++---
 1 file changed, 17 insertions(+), 3 deletions(-)

diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index 335d2c127763..94be22aa5cb6 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -47,7 +47,8 @@ static void irqtime_account_delta(struct irqtime *irqtime, u64 delta,
 	u64_stats_update_begin(&irqtime->sync);
 	cpustat[idx] += delta;
 	irqtime->total += delta;
-	irqtime->tick_delta += delta;
+	if (!kcpustat_idle_dyntick())
+		irqtime->tick_delta += delta;
 	u64_stats_update_end(&irqtime->sync);
 }
 
@@ -444,6 +445,10 @@ static void kcpustat_idle_stop(struct kernel_cpustat *kc, u64 now)
 
 static void kcpustat_idle_start(struct kernel_cpustat *kc, u64 now)
 {
+	/* Irqtime accounting might have been enabled in the middle of the IRQ */
+	if (kc->idle_elapse)
+		return;
+
 	write_seqcount_begin(&kc->idle_sleeptime_seq);
 	kc->idle_entrytime = now;
 	kc->idle_elapse = true;
@@ -478,7 +483,8 @@ void kcpustat_irq_enter(u64 now)
 {
 	struct kernel_cpustat *kc = kcpustat_this_cpu;
 
-	if (!vtime_generic_enabled_this_cpu())
+	if (!vtime_generic_enabled_this_cpu() &&
+	    (irqtime_enabled() || vtime_accounting_enabled_this_cpu()))
 		kcpustat_idle_stop(kc, now);
 }
 
@@ -486,7 +492,15 @@ void kcpustat_irq_exit(u64 now)
 {
 	struct kernel_cpustat *kc = kcpustat_this_cpu;
 
-	if (!vtime_generic_enabled_this_cpu())
+	/*
+	 * Generic vtime already does its own idle accounting.
+	 * But irqtime accounting or arch vtime which also accounts IRQs
+	 * need to pause nohz accounting. Resume nohz accounting as long
+	 * as the irqtime config is enabled to handle case where irqtime
+	 * accounting got runtime disabled in the middle of an IRQ.
+	 */
+	if (!vtime_generic_enabled_this_cpu() &&
+	    (IS_ENABLED(CONFIG_IRQ_TIME_ACCOUNTING) || vtime_accounting_enabled_this_cpu()))
 		kcpustat_idle_start(kc, now);
 }
 
-- 
2.53.0

[PATCH 15/15] sched/cputime: Handle dyntick-idle steal time correctly

[Frederic Weisbecker]

The dyntick-idle steal time is currently accounted when the tick
restarts but the stolen idle time is not subtracted from the idle time
that was already accounted. This is to avoid observing the idle time
going backward as the dyntick-idle cputime accessors can't reliably know
in advance the stolen idle time.

In order to maintain a forward progressing idle cputime while
subtracting idle steal time from it, keep track of the previously
accounted idle stolen time and substract it from _later_ idle cputime
accounting.

Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Tested-by: Shrikanth Hegde <sshegde@linux.ibm.com>
---
 include/linux/kernel_stat.h |  1 +
 kernel/sched/cputime.c      | 28 +++++++++++++++++++++++-----
 2 files changed, 24 insertions(+), 5 deletions(-)

diff --git a/include/linux/kernel_stat.h b/include/linux/kernel_stat.h
index 512104b0ff49..fce1392e2140 100644
--- a/include/linux/kernel_stat.h
+++ b/include/linux/kernel_stat.h
@@ -39,6 +39,7 @@ struct kernel_cpustat {
 	bool		idle_elapse;
 	seqcount_t	idle_sleeptime_seq;
 	u64		idle_entrytime;
+	u64		idle_stealtime[2];
 #endif
 	u64		cpustat[NR_STATS];
 };
diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
index 94be22aa5cb6..244b57417240 100644
--- a/kernel/sched/cputime.c
+++ b/kernel/sched/cputime.c
@@ -425,19 +425,32 @@ static inline void irqtime_account_process_tick(struct task_struct *p, int user_
 static void kcpustat_idle_stop(struct kernel_cpustat *kc, u64 now)
 {
 	u64 *cpustat = kc->cpustat;
-	u64 delta;
+	u64 delta, steal, steal_delta;
+	int iowait;
 
 	if (!kc->idle_elapse)
 		return;
 
+	iowait = nr_iowait_cpu(smp_processor_id()) > 0;
 	delta = now - kc->idle_entrytime;
+	steal = steal_account_process_time(delta);
 
+	/*
+	 * Record the idle time after substracting the steal time from
+	 * previous update sequence. Don't substract the steal time from
+	 * the current update sequence to avoid readers moving backward.
+	 */
 	write_seqcount_begin(&kc->idle_sleeptime_seq);
-	if (nr_iowait_cpu(smp_processor_id()) > 0)
+	steal_delta = min_t(u64, kc->idle_stealtime[iowait], delta);
+	delta -= steal_delta;
+	kc->idle_stealtime[iowait] -= steal_delta;
+
+	if (iowait)
 		cpustat[CPUTIME_IOWAIT] += delta;
 	else
 		cpustat[CPUTIME_IDLE] += delta;
 
+	kc->idle_stealtime[iowait] += steal;
 	kc->idle_entrytime = now;
 	kc->idle_elapse = false;
 	write_seqcount_end(&kc->idle_sleeptime_seq);
@@ -464,7 +477,6 @@ void kcpustat_dyntick_stop(u64 now)
 		kcpustat_idle_stop(kc, now);
 		kc->idle_dyntick = false;
 		vtime_dyntick_stop();
-		steal_account_process_time(ULONG_MAX);
 	}
 }
 
@@ -508,6 +520,7 @@ static u64 kcpustat_field_dyntick(int cpu, enum cpu_usage_stat idx,
 				  bool compute_delta, u64 now)
 {
 	struct kernel_cpustat *kc = &kcpustat_cpu(cpu);
+	int iowait = idx == CPUTIME_IOWAIT;
 	u64 *cpustat = kc->cpustat;
 	unsigned int seq;
 	u64 idle;
@@ -516,8 +529,13 @@ static u64 kcpustat_field_dyntick(int cpu, enum cpu_usage_stat idx,
 		seq = read_seqcount_begin(&kc->idle_sleeptime_seq);
 
 		idle = cpustat[idx];
-		if (kc->idle_elapse && compute_delta && now > kc->idle_entrytime)
-			idle += (now - kc->idle_entrytime);
+
+		if (kc->idle_elapse && compute_delta && now > kc->idle_entrytime) {
+			u64 delta = now - kc->idle_entrytime;
+
+			delta -= min_t(u64, kc->idle_stealtime[iowait], delta);
+			idle += delta;
+		}
 	} while (read_seqcount_retry(&kc->idle_sleeptime_seq, seq));
 
 	return idle;
-- 
2.53.0