On Tue, Nov 08, 2016 at 07:05:26PM +0530, Laxman Dewangan wrote:

On Tuesday 08 November 2016 06:59 PM, Linus Walleij wrote:

quoted

On Tue, Nov 8, 2016 at 11:20 AM, Laxman Dewangan [off-list ref] wrote:

quoted

On Tuesday 08 November 2016 03:45 PM, Linus Walleij wrote:

quoted

If you can *actually* change the volatage, it needs to be modeled
as a (fixed voltage?) regulator, not as a custom property for the pin
control attributes. I guess you definiately need the regulator framework
to accumulate and infer the different consumer requirements anyway
in that case.

The PMIC voltage output is changed via regulator calls.
Here, we need to have two configruations for given voltage level of
interface:
* One at IO voltage from PMIC via regulator call to change votlage of IO
rail.
* Second, configure the IO pad register to tell the IO voltage level so that
it can configured internally for that level.

I understand! (I think.)

Thanks,

quoted

But then the two things (A) changing the regulator voltage and (B) changing
the pin setting need to happen at the same time do they
not?

Now you're just hardcoding something into these device tree properties
and hoping that the regulators will somehow be set up in accordance to
what you set up for the pads in the device tree, correct?

There is two types of configuration in given platform, the IO voltage does
not get change (fixed in given platform) and in some of cases, get change
dynamically like SDIO3.0 where the voltage switches to 3.3V and 1.8V.

Yes, it can be integrated with the regulator handle and then it can call the
required configurations through notifier and regulator_get_voltage().
But I think it is too much complex for the static configurations. This
mandate also to populate the regulator handle and all power tree.

It looks as if regulator notifiers should be able to support whatever
use-cases we may have (I suspect that we really only need pre- and post-
voltage-change notifications.

The simple way for static configuration (case where voltage does not get
change), just take the power tree IO voltage from DT and configure the IO
pad control register.

For static configurations where we have a regulator along with a pinmux
setting for the I/O voltage we could potentially run into issues where
both settings don't match. How would we handle that?

For dynamic case, there is some sequence need to be followed based on
voltage direction change (towards lower or towards higher) for the voltage
change and the IO pad voltage configuration and it is simple to do it from
client driver.

Are patches available for this? It'd be useful to know how this will end
up being used in order to come up with the best solution.

In general I think it's good to send a complete series of patches, even
if it's long and spans multiple subsystems. As it is it's difficult for
anyone else (myself included) to understand where this is headed, which
makes it more complicated than necessary to get at the final solution.

quoted

To me it seems like the pins/pads should all have an <&phandle> to
the regulator controlling its voltage output, in the device tree.

In the Linux kernel, the driver has to regulator_[bulk_]get() this for
each pin, check the voltage with regulator_get_voltage() and set up
this according to the supplied voltage.

The driver then ideally should subscribe to regulator voltage notifier
events to change the setting if the voltage changes. I guess. But
atleast the first step seems inevitable: get the voltage from a regulator.

Else there is no dependency between the regulator and its consumer.

So what your pins need is a regulator phandle, not a magic value to
be poked into a register, hoping things will match up.

I understand that this is a simple quick-and-dirty solution but it is
not the right solution.

Yaah, the static power tree configuration is much simple in this approach
without having regulator drivers and support.

Integrating with regulator driver can be done here also.

I like to have both approach, through pinmux DT and also from regulator. So
based on the platform, if regulator supported then populate required
properties in DT for regulator else go on standard pinmux DT way (for
non-regulator cases).

Again, it would be best to see this in actual use. Right now it's not
clear that we'll even have both cases. Implementing both approaches will
mean potentially unused code.

On another note: in my experience you seldom need both cases, since the
dynamic configuration is the hard one, and the static configuration will
usually be a special case of the dynamic configuration.

Thierry