Before this change, if last IO-blocked task wakes up
on a different CPU, the original CPU may stay idle for much longer,
and the entire time it stays idle is accounted as iowait time.

This change adds struct tick_sched::iowait_exittime member.
On entry to idle, it is set to KTIME_MAX.
Last IO-blocked task, if migrated, sets it to current time.
Note that this can happen only once per each idle period:
new iowaiting tasks can't magically appear on idle CPU's rq.

If iowait_exittime is set, then (iowait_exittime - idle_entrytime)
gets accounted as iowait, and the remaining (now - iowait_exittime)
as "true" idle.

Run-tested: /proc/stat counters no longer go backwards.

Signed-off-by: Denys Vlasenko <[email protected]>
Cc: Frederic Weisbecker <[email protected]>
Cc: Hidetoshi Seto <[email protected]>
Cc: Fernando Luis Vazquez Cao <[email protected]>
Cc: Tetsuo Handa <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Arjan van de Ven <[email protected]>
Cc: Oleg Nesterov <[email protected]>
---
 include/linux/tick.h     |  2 ++
 kernel/sched/core.c      | 14 +++++++++++
 kernel/time/tick-sched.c | 64 ++++++++++++++++++++++++++++++++++++++++--------
 3 files changed, 70 insertions(+), 10 deletions(-)

diff --git a/include/linux/tick.h b/include/linux/tick.h
index 4de1f9e..1bf653e 100644
--- a/include/linux/tick.h
+++ b/include/linux/tick.h
@@ -67,6 +67,7 @@ struct tick_sched {
        ktime_t                         idle_exittime;
        ktime_t                         idle_sleeptime;
        ktime_t                         iowait_sleeptime;
+       ktime_t                         iowait_exittime;
        seqcount_t                      idle_sleeptime_seq;
        ktime_t                         sleep_length;
        unsigned long                   last_jiffies;
@@ -140,6 +141,7 @@ extern void tick_nohz_irq_exit(void);
 extern ktime_t tick_nohz_get_sleep_length(void);
 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);
+extern void tick_nohz_iowait_to_idle(int cpu);
 
 # else /* !CONFIG_NO_HZ_COMMON */
 static inline int tick_nohz_tick_stopped(void)
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 268a45e..ffea757 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -4218,7 +4218,14 @@ void __sched io_schedule(void)
        current->in_iowait = 1;
        schedule();
        current->in_iowait = 0;
+#ifdef CONFIG_NO_HZ_COMMON
+       if (atomic_dec_and_test(&rq->nr_iowait)) {
+               if (raw_smp_processor_id() != cpu_of(rq))
+                       tick_nohz_iowait_to_idle(cpu_of(rq));
+       }
+#else
        atomic_dec(&rq->nr_iowait);
+#endif
        delayacct_blkio_end();
 }
 EXPORT_SYMBOL(io_schedule);
@@ -4234,7 +4241,14 @@ long __sched io_schedule_timeout(long timeout)
        current->in_iowait = 1;
        ret = schedule_timeout(timeout);
        current->in_iowait = 0;
+#ifdef CONFIG_NO_HZ_COMMON
+       if (atomic_dec_and_test(&rq->nr_iowait)) {
+               if (raw_smp_processor_id() != cpu_of(rq))
+                       tick_nohz_iowait_to_idle(cpu_of(rq));
+       }
+#else
        atomic_dec(&rq->nr_iowait);
+#endif
        delayacct_blkio_end();
        return ret;
 }
diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index 47ed7cf..d78c942 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -408,15 +408,27 @@ static void tick_nohz_update_jiffies(ktime_t now)
 
 static void tick_nohz_stop_idle(struct tick_sched *ts, ktime_t now)
 {
-       ktime_t delta;
+       ktime_t delta, entry, end;
 
        /* Updates the per cpu time idle statistics counters */
        write_seqcount_begin(&ts->idle_sleeptime_seq);
-       delta = ktime_sub(now, ts->idle_entrytime);
-       if (ts->idle_active == 2)
+       entry = ts->idle_entrytime;
+       delta = ktime_sub(now, entry);
+       if (ts->idle_active == 2) {
+               end = ts->iowait_exittime;
+               if (end.tv64 != KTIME_MAX) {
+                       /*
+                        * Last iowaiting task on our rq was woken up on other 
CPU
+                        * sometime in the past, it updated ts->iowait_exittime.
+                        */
+                       delta = ktime_sub(now, end);
+                       ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, 
delta);
+                       delta = ktime_sub(end, entry);
+               }
                ts->iowait_sleeptime = ktime_add(ts->iowait_sleeptime, delta);
-       else
+       } else {
                ts->idle_sleeptime = ktime_add(ts->idle_sleeptime, delta);
+       }
        ts->idle_active = 0;
        write_seqcount_end(&ts->idle_sleeptime_seq);
 
@@ -430,6 +442,7 @@ static ktime_t tick_nohz_start_idle(struct tick_sched *ts)
        write_seqcount_begin(&ts->idle_sleeptime_seq);
        ts->idle_entrytime = now;
        ts->idle_active = nr_iowait_cpu(smp_processor_id()) ? 2 : 1;
+       ts->iowait_exittime.tv64 = KTIME_MAX;
        write_seqcount_end(&ts->idle_sleeptime_seq);
 
        sched_clock_idle_sleep_event();
@@ -437,6 +450,16 @@ static ktime_t tick_nohz_start_idle(struct tick_sched *ts)
        return now;
 }
 
+void tick_nohz_iowait_to_idle(int cpu)
+{
+       struct tick_sched *ts = tick_get_tick_sched(cpu);
+       ktime_t now = ktime_get();
+
+       write_seqcount_begin(&ts->idle_sleeptime_seq);
+       ts->iowait_exittime = now;
+       write_seqcount_end(&ts->idle_sleeptime_seq);
+}
+
 /**
  * get_cpu_idle_time_us - get the total idle time of a cpu
  * @cpu: CPU number to query
@@ -465,13 +488,26 @@ u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
                *last_update_time = ktime_to_us(now);
 
        do {
-               ktime_t delta;
+               ktime_t start, delta, iowait_exit;
 
                seq = read_seqcount_begin(&ts->idle_sleeptime_seq);
                idle = ts->idle_sleeptime;
-               if (ts->idle_active == 1) {
-                       delta = ktime_sub(now, ts->idle_entrytime);
+               if (ts->idle_active /* either 1 or 2 */) {
+                       start = ts->idle_entrytime;
+                       if (ts->idle_active == 2) {
+                               /* This idle period started as "iowait idle" */
+                               iowait_exit = ts->iowait_exittime;
+                               if (iowait_exit.tv64 == KTIME_MAX)
+                                       goto skip; /* and it still is */
+                               /*
+                                * This CPU used to be "iowait idle", but 
iowait task
+                                * has migrated. The rest of idle time is "true 
idle":
+                                */
+                               start = iowait_exit;
+                       }
+                       delta = ktime_sub(now, start);
                        idle = ktime_add(idle, delta);
+ skip: ;
                }
        } while (read_seqcount_retry(&ts->idle_sleeptime_seq, seq));
 
@@ -507,13 +543,21 @@ u64 get_cpu_iowait_time_us(int cpu, u64 *last_update_time)
                *last_update_time = ktime_to_us(now);
 
        do {
-               ktime_t delta;
+               ktime_t delta, end;
 
                seq = read_seqcount_begin(&ts->idle_sleeptime_seq);
                iowait = ts->iowait_sleeptime;
                if (ts->idle_active == 2) {
-                       delta = ktime_sub(now, ts->idle_entrytime);
-                       iowait = ktime_add(ts->iowait_sleeptime, delta);
+                       /*
+                        * If last iowaiting task on our rq was woken up on 
other CPU
+                        * sometime in the past, it updated ts->iowait_exittime.
+                        * Otherwise, ts->iowait_exittime == KTIME_MAX.
+                        */
+                       end = ts->iowait_exittime;
+                       if (end.tv64 == KTIME_MAX)
+                               end = now;
+                       delta = ktime_sub(end, ts->idle_entrytime);
+                       iowait = ktime_add(iowait, delta);
                }
        } while (read_seqcount_retry(&ts->idle_sleeptime_seq, seq));
 
-- 
1.8.1.4

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Reply via email to