On 06/13/2017 04:11 AM, Ulf Hansson wrote:

[...]

quoted

quoted

Isn't the off_time the summary of the states' genpd->states[i].active_time?

Yes. it is.

quoted

Then we could do that computation quite easily at the point when
producing the debugfs output instead.

Are yu suggesting doing away with off_time parameter all together. It is
possible.

Yes, I think that makes sense.

quoted

quoted

quoted

+       }
+       genpd->accounting_time = now;
+}
+

[...]

quoted

quoted

quoted

@@ -774,6 +795,7 @@ static void genpd_sync_power_off(struct generic_pm_domain *genpd, bool use_lock,
        if (_genpd_power_off(genpd, false))
                return;

+       genpd_update_accounting(genpd);

Timekeeping gets disabled at a certain point during suspend. Therefore
you can't update accounting here, as this function gets called in this
stages.

I will remove the update of accounting at this point. But then the
statistics  can be skewed for some power domains atleast. Other option
is to restart the accounting every time after device suspend and resume.

I would suggest to start simple and just avoid updating the accounting
in these cases. Then we can look into improvements, if we think there
is a need for it.

What do you think?

I do agree. Will post the updated series.

Regards
Thara

[...]

Kind regards
Uffe

-- 
Regards
Thara