Re: [RFC v4 01/14] regulator: of: Add helper for getting all supplies

[Peter Chen <hzpeterchen@gmail.com>]

On Sun, Jun 12, 2016 at 03:29:01PM +0800, Peter Chen wrote:
> On Fri, Jun 10, 2016 at 12:30:56PM -0500, Rob Herring wrote:
> > On Thu, Jun 09, 2016 at 01:42:02PM +0200, Krzysztof Kozlowski wrote:
> > > On 06/09/2016 12:29 PM, Mark Brown wrote:
> > > > On Thu, Jun 09, 2016 at 11:44:18AM +0200, Krzysztof Kozlowski wrote:
> > > >> Few drivers have a need of getting regulator supplies without knowing
> > > >> their names:
> > > >> 1. The Simple Framebuffer driver works on setup provided by bootloader
> > > >>    (outside of scope of kernel);
> > > >> 2. Generic power sequence driver may be attached to any device node.
> > > >>
> > > >> Add a Device Tree helper for parsing "-supply" properties and returning
> > > >> allocated bulk regulator consumers.
> > > > 
> > > > I'm still very concerned that this is just an invitation to people to
> > > > write half baked regulator consumers and half baked DTs to go along with
> > > > it, making it a standard API that doesn't have big red flags on it that
> > > > will flag up when "normal" drivers use it is not good.  Right now this
> > > > just looks like a standard API and people are going to just start using
> > > > it.  If we are going to do this perhaps we need a separate header or
> > > > something to help flag this up.
> > > 
> > > No problem, I can move it to a special header.  Actually, if you dislike
> > > this as an API, it does not have to be in header at all.  I can just
> > > duplicate the simplefb code.
> > > 
> > > > In the case of power sequences I'd expect the sequences to perform
> > > > operations on named supplies - the core shouldn't know what the supplies
> > > > are but the thing specifying the sequence should.
> > > 
> > > Hm, so maybe passing names like:
> > > 
> > > usb3503@08 {
> > > 	reset-gpios = <&gpx3 5 GPIO_ACTIVE_HIGH>;
> > > 	initial-mode = <1>;
> > > 	vdd-supply = <&buck8_reg>;
> > > 	foo-supply = <&buck9_reg>;
> > > 
> > >         power-sequence;
> > > 	power-sequence-supplies = "vdd", "foo";
> > 
> > This alone would be fine as it is just one property, but then what's 
> > next? power-sequence-delay, power-sequence-clocks, etc. What if you 
> > need to express ordering relationship of supplies, clocks, gpios? We end 
> > up with a scripting language in DT and we don't want to have that.
> > 
> 
> Can we do things like below:
> 
> - DT describes hardware elements (clock, gpios, etc) for power sequence, and we
> need a node for power sequence.
> - Power sequence framework handles getting hardware elements.

Framework may do few things, since hardware elements are also different
for devices.

> - Power sequence platform driver handles special sequence for devices,
> and we can create some generic drivers for generic devices.
> 

So, my suggestion is do like mmc does (like this patch set does). The
reasons like belows:

- This piece of power sequence code needs to work like device driver, not
library, it is easy to manage resources using device driver.
- The device on the bus has still not been found, so this piece of code
can't be in device driver on each subsystem.
- We need to have a place for these power sequences drivers

Ideally, I hope it can work like regulator class, but it seems hard to
compatible with current mmc-pwrseq DT node.

-- 

Best Regards,
Peter Chen