Re: [PATCH RFC 1/2] regulator: core: Disable unused regulators with unknown status

[Konrad Dybcio]

On 10/9/23 22:21, Stephan Gerhold wrote:
> On Fri, Oct 06, 2023 at 11:11:48PM +0200, Konrad Dybcio wrote:
> > On 4.10.2023 16:17, Stephan Gerhold wrote:
> > > Some regulator drivers do not provide a way to check if the regulator is
> > > currently enabled or not. That does not necessarily mean that the
> > > regulator is always-on. For example, the regulators managed by the RPM
> > > firmware on Qualcomm platforms can be either on or off during boot but
> > > the initial state is not known. To sync the state the regulator should
> > > get either explicitly enabled or explicitly disabled.
> > >
> > > Enabling all regulators unconditionally is not safe, because we might
> > > not know which voltages are safe. The devices supplied by those
> > > regulators might also require a special power-up sequence where the
> > > regulators are turned on in a certain order or with specific delay.
> > >
> > > Disabling all unused regulators is safer. If the regulator is already
> > > off it will just stay that way. If the regulator is on, disabling it
> > > explicitly allows the firmware to turn it off for reduced power
> > > consumption.
> > >
> > > The regulator core already has functionality for disabling unused
> > > regulators. However, at the moment it assumes that all regulators where
> > > the .is_enabled() callback fails are actually off. There is no way to
> > > return a special value for the "unknown" state to explicitly ask for
> > > disabling those regulators.
> > >
> > > Some drivers (e.g. qcom-rpmh-regulator.c) return -EINVAL for the case
> > > where the initial status is unknown. Use that return code to assume the
> > > initial status is unknown and try to explicitly disable the regulator
> > > in that case.
> > >
> > > Signed-off-by: Stephan Gerhold <stephan.gerhold@xxxxxxxxxxxxxxx>
> > > ---
> > > Instead of -EINVAL we could also use a different return code to indicate
> > > the initial status is unknown. Or maybe there is some other option that
> > > would be easier? This is working for me but I'm sending it as RFC to get
> > > more feedback. :)
> >
> > -EOPNOTSUPP for "doesn't support getting is_enabled state"?
>
> The way it is implemented right now the Qualcomm SMD RPM regulator does
> actually support getting the .is_enabled() state. It is only unable to
> determine the initial state during boot. Once the regulator has been
> enabled by some consumer for the first time the .is_enabled() callback
> starts returning the expected results.
>
> Typically -EOPNOTSUPP is used when the driver callback (or similar) is
> not implemented at all. I'm not sure if using -EOPNOTSUPP for the
> "temporarily unable to determine state" purpose would be misleading.
I'd say EOPNOTSUPP is fair here because calling is_enabled in that 
context is not supported, but I guess it's up to Mark.

Konrad