On 1 October 2014 21:50, Rafael J. Wysocki [off-list ref] wrote:

On Wednesday, October 01, 2014 06:18:58 PM Sylwester Nawrocki wrote:

quoted

On 01/10/14 16:41, Ulf Hansson wrote:

quoted

At ->probe() it's common practice for drivers/subsystems to bring their
devices to full power and without depending on CONFIG_PM_RUNTIME.

We could also expect that drivers/subsystems requires their device's
corresponding PM domains to be powered, to successfully complete a
->probe() sequence.

Align to the behavior above, by ensuring all PM domains are powered
prior initialization of a generic PM domain.

Do note, since the generic PM domain will try to power off unused PM
domains at late_init, there should be no increased power consumption
over time, but potentially during boot.

Wouldn't it be a better idea to power on the power domains which are
turned off only when CONFIG_PM_RUNTIME is not enabled ? I had a plan
to submit a patch doing that but unfortunately this has fallen through
the cracks. At the moment mach-exynos/pm_domains.c is not even built in
when CONFIG_PM_RUNTIME is disabled.

Yes, that's the approach I also intend to take in the next step.

But, it's not that simple. Since this requires a mechanism for drivers
to bring their device's PM domains into power state prior doing probe.
We don't have such today. I do have some ideas about this, but I think
we need to keep that as a separate discussion.

quoted

I don't like the behaviour introduced in this patch to be the default,
i.e. turning all possible power domains during boot sequence, even if
some are not used and not needed. While we're trying to decrease the
power consumption in any possible way this doesn't help at all.

This will hit only during boot, until late_init. Unless you have a
platform that keeps rebooting all the time, is this really a big
worry?

Still, I certainly agree that we should strive for a solution where
it's possible to leave PM domains powered off at init. It's should be
too hard to support this from genpd point of view, but
drivers/subsystems will need some adaptations.

Agreed (as stated before).

And I'm wondering why that comment of mine was ignored?

Sorry, if missed to comment of that. I guess I have at this point.

Kind regards
Uffe