Re: [PATCH v4 2/3] PM / sleep: Go direct_complete if driver has no callbacks

["Rafael J. Wysocki" <rjw@rjwysocki.net>]

On Thursday, July 16, 2015 10:47:51 AM Tomeu Vizoso wrote:
> On 16 July 2015 at 02:41, Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> > On Wednesday, July 15, 2015 02:47:50 PM Alan Stern wrote:
> >> On Wed, 15 Jul 2015, Tomeu Vizoso wrote:
> >>
> >> > If a suitable prepare callback cannot be found for a given device and
> >> > its driver has no PM callbacks at all, assume that it can go direct to
> >> > complete when the system goes to sleep.
> >> >
> >> > The reason for this is that there's lots of devices in a system that do
> >> > no PM at all and there's no reason for them to prevent their ancestors
> >> > to do direct_complete if they can support it.
> >> >
> >> > Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
> >> > ---
> >> >
> >> >  drivers/base/power/main.c | 17 +++++++++++++++++
> >> >  1 file changed, 17 insertions(+)
> >> >
> >> > diff --git a/drivers/base/power/main.c b/drivers/base/power/main.c
> >> > index 1710c26ba097..edda3f233c7c 100644
> >> > --- a/drivers/base/power/main.c
> >> > +++ b/drivers/base/power/main.c
> >> > @@ -1540,6 +1540,21 @@ int dpm_suspend(pm_message_t state)
> >> >     return error;
> >> >  }
> >> >
> >> > +static bool driver_has_no_pm_callbacks(struct device_driver *drv)
> >> > +{
> >> > +   if (!drv->pm)
> >> > +           return true;
> >> > +
> >> > +   return !drv->pm->prepare &&
> >> > +          !drv->pm->suspend &&
> >> > +          !drv->pm->suspend_late &&
> >> > +          !drv->pm->suspend_noirq &&
> >> > +          !drv->pm->resume_noirq &&
> >> > +          !drv->pm->resume_early &&
> >> > +          !drv->pm->resume &&
> >> > +          !drv->pm->complete;
> >> > +}
> >>
> >> This isn't exactly what I meant.  We also need to check the dev_pm_ops
> >> fields in dev->pm_domain, dev->type, dev->class, and dev->bus.  Only if
> >> _all_ of these callbacks are missing should we use direct_complete.
> >
> > Also checking that on every suspend is kind of wasteful, because those things
> > do not change very often.
> 
> Do you have any suggestion on when would be a good time to do that
> check? device_pm_sleep_init() and device_pm_add() are unfortunately
> too early.

The time to check that is (a) when the device is registered (for the bus
type/class/device type), (b) when it is added to (or removed from) a PM
domain and (c) when a driver is bound to (unbound from) it.

> Alternatively we could check once on the first suspend and cache it,

That information may be stale by the time we suspend next time.

> but I'm not sure that complexity would be worth it.

I don't think that adding extra overhead to *every* suspend can be
justified easily, however.


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.