On Thu, Apr 24, 2014 at 08:45:58PM +0200, Denys Vlasenko wrote:
> 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 <dvlas...@redhat.com>
> Cc: Frederic Weisbecker <fweis...@gmail.com>
> Cc: Hidetoshi Seto <seto.hideto...@jp.fujitsu.com>
> Cc: Fernando Luis Vazquez Cao <fernando...@lab.ntt.co.jp>
> Cc: Tetsuo Handa <penguin-ker...@i-love.sakura.ne.jp>
> Cc: Thomas Gleixner <t...@linutronix.de>
> Cc: Ingo Molnar <mi...@kernel.org>
> Cc: Peter Zijlstra <pet...@infradead.org>
> Cc: Andrew Morton <a...@linux-foundation.org>
> Cc: Arjan van de Ven <ar...@linux.intel.com>
> Cc: Oleg Nesterov <o...@redhat.com>
> ---
> 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));
Note that even using seqlock doesn't alone help to fix the preemption issue
when the above may overwrite the exittime of the next last iowait task from
the old rq.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
