[Intel-gfx] [PATCH v4 3/3] PM/runtime:Replace jiffies based accounting with ktime based accounting

[Ulf Hansson]

On Thu, 20 Dec 2018 at 15:17, Vincent Guittot
<vincent.guittot at linaro.org> wrote:
>
> From: Thara Gopinath <thara.gopinath at linaro.org>
>
> This patch replaces jiffies based accounting for runtime_active_time
> and runtime_suspended_time with ktime base accounting. This makes the
> runtime debug counters inline with genpd and other pm subsytems which
> uses ktime based accounting.
>
> Signed-off-by: Thara Gopinath <thara.gopinath at linaro.org>
> [move from ktime to raw nsec]
> Signed-off-by: Vincent Guittot <vincent.guittot at linaro.org>
> ---
>  drivers/base/power/runtime.c | 17 +++++++++--------
>  drivers/base/power/sysfs.c   | 11 ++++++++---
>  include/linux/pm.h           |  6 +++---
>  3 files changed, 20 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
> index e695544..f700524 100644
> --- a/drivers/base/power/runtime.c
> +++ b/drivers/base/power/runtime.c
> @@ -64,8 +64,8 @@ static int rpm_suspend(struct device *dev, int rpmflags);
>   */
>  void update_pm_runtime_accounting(struct device *dev)
>  {
> -       unsigned long now = jiffies;
> -       unsigned long delta;
> +       u64 now = ktime_to_ns(ktime_get());
> +       u64 delta;
>
>         delta = now - dev->power.accounting_timestamp;
>
> @@ -75,9 +75,9 @@ void update_pm_runtime_accounting(struct device *dev)
>                 return;
>
>         if (dev->power.runtime_status == RPM_SUSPENDED)
> -               dev->power.suspended_jiffies += delta;
> +               dev->power.suspended_time += delta;
>         else
> -               dev->power.active_jiffies += delta;
> +               dev->power.active_time += delta;
>  }
>
>  static void __update_runtime_status(struct device *dev, enum rpm_status status)
> @@ -88,17 +88,18 @@ static void __update_runtime_status(struct device *dev, enum rpm_status status)
>
>  u64 pm_runtime_suspended_time(struct device *dev)
>  {
> -       unsigned long flags, time;
> +       u64 time;
> +       unsigned long flags;
>
>         spin_lock_irqsave(&dev->power.lock, flags);
>
>         update_pm_runtime_accounting(dev);
>
> -       time = dev->power.suspended_jiffies;
> +       time = dev->power.suspended_time;
>
>         spin_unlock_irqrestore(&dev->power.lock, flags);
>
> -       return jiffies_to_nsecs(time);
> +       return time;
>  }
>  EXPORT_SYMBOL_GPL(pm_runtime_suspended_time);
>
> @@ -1503,7 +1504,7 @@ void pm_runtime_init(struct device *dev)
>         dev->power.request_pending = false;
>         dev->power.request = RPM_REQ_NONE;
>         dev->power.deferred_resume = false;
> -       dev->power.accounting_timestamp = jiffies;
> +       dev->power.accounting_timestamp = ktime_to_ns(ktime_get());

This change worries me a bit, but it may not be a problem in practice.

pm_runtime_init() is called when devices get initialized, via
device_initialize(). If timekeeping has not been initialized prior a
call to ktime_get() is done, it will fail and causing a NULL pointer
deference or something along those lines.

In other words, do we know that device_initialize() is always called
after timekeeping has been initialized during boot?

[...]

Kind regards
Uffe