On Thu, Oct 27, 2022 at 09:12:22PM -0700, Zev Weiss wrote:

> I can see why it might look that way, but I'd argue it's actually not.  The
> systems this is intended to support provide power to entirely separate
> external devices -- think of a power distribution unit that might have
> arbitrary things plugged into it.  It seems to me like a property of the
> hardware that those things shouldn't have their power supply turned off (or
> on) just because a controller in the PDU rebooted.

We don't turn things off on reboot?  We don't do anything in particular
on reboot...

> > I guess it easy
> > to understand in case of Linux which disables unclaimed regulators
> > during. But what if other system/firmware does not behave like that?

> In this case, then no change would be needed -- a system that (unlike Linux)
> doesn't twiddle regulator state on its own would just continue to not do
> that.

We don't turn unclaimed regulators off until userspace has had a chance
to start, if there's some problem with system integrators arranging to
do this we can look into how that works, for example making the delay
tunable.  I don't think this is really meaningfully different from a
driver deciding to turn things off from a binding point of view.

> > And what is the "external actor"? OS is not an external actor?

> It's admittedly a bit vague, but I couldn't think of a clearer way to
> express what is a sort of nebulous concept -- essentially, some entity
> outside the "driver" (or analogous software component) using the information
> in the device-tree.  In many common cases this would essentially mean "a
> human user", since in the PDU-like systems I'm targeting here the only thing
> that should ever be deciding to turn the regulator on or off is an operator
> logged in to the system to manually enable or disable an outlet.  I was
> aiming to leave the wording a bit more general though, since in some other
> context I could imagine some other piece of software toggling things
> automatedly (e.g. lights getting turned on and off on a schedule or
> something, if that's what happens to be plugged in).

This is policy stuff, it doesn't translate into DTs at all.

> > I could not get the problem you want to solve with this property - I
> > looked at cover letter and at commit msg.

> The problem is that a driver deciding on its own to enable or disable the
> regulator (e.g. during boot or shutdown) would be a critical failure for the
> kind of systems I'm aiming to support.

If the driver is doing something like this it should be addressed in the
driver.

> > I can only imagine that you want to keep regulator on, after last its
> > user disappears... but for what purpose? Do you expect that after system
> > shutdown the pin will stay high so regulator will be also on? If so, you
> > need hardware design, e.g. with some pull up (if control is over GPIO).

> As described above, the regulators involved here (in these sorts of PDU-like
> systems) provide power for external systems and devices.  It is critical
> that the controller's boot and shutdown sequences not alter the state of the
> regulator.

This really sounds like a full stack system integration problem, not
something that can be resolved with one software component.