Re: [PATCH v1] mfd: Add support for Merrifield Basin Cove PMIC

[Andy Shevchenko]

On Thu, Apr 04, 2019 at 10:03:14AM +0100, Lee Jones wrote:
> On Thu, 04 Apr 2019, Andy Shevchenko wrote:
> > On Thu, Apr 04, 2019 at 08:03:57AM +0100, Lee Jones wrote:
> > > On Thu, 04 Apr 2019, Lee Jones wrote:
> > > > On Tue, 02 Apr 2019, Andy Shevchenko wrote:
> > > > > On Tue, Apr 02, 2019 at 06:12:11AM +0100, Lee Jones wrote:
> > > > > > On Mon, 18 Mar 2019, Andy Shevchenko wrote:

> > > > > > Although succinct, dragging values from one platform device into
> > > > > > another doesn't sound that neat.
> > > > > 
> > > > > So, how to split resources given in one _physical_ multi-functional device to
> > > > > several of them?  Isn't it what MFD framework for?
> > > > > 
> > > > > Any other approach here? I'm all ears!
> > > > 
> > > > From the child:
> > > > 
> > > >   platform_get_irq(dev->parent, CLIENT_ID);
> > 
> > So, instead of keeping a fragile approach in one driver, we will spread this
> > to all of them.
> 
> No, the fragileness goes away with implicit definitions of IDs.

Did you mean "explicit"?
Something like we need to have a shared map of those indices?

> > > > > > Also, since the ordering of the
> > > > > > devices is critical in this implementation, it also comes across as
> > > > > > fragile.
> > > > > 
> > > > > How fragile? In ACPI we don't have IRQ labeling scheme. Index is used for that.
> > > > > 
> > > > > > Any reason why ACPI can't register all of the child devices, or for
> > > > > > the child devices to obtain their IRQ directly from the tables?
> > > > > 
> > > > > And how are we supposed to enumerated them taking into consideration single
> > > > > ACPI ID given?
> > > > 
> > > > This question was a little whimsical, since I have no idea how the
> > > > ACPI tables you're working with are laid out.
> > 
> > There is one device node with several IRQ and other resources.
> > In pseudo code:
> > 
> > 	device node {
> > 		device ID,
> > 		IRQ 0,
> > 		IRQ 1,
> > 		...
> > 		MMIO 0,
> > 		...
> > 	}
> 
> Sure.  Thanks for the explanation.
> 
> Very well.  I guess it's not too bad as it is.

It represent real hardware 1:1.
Just out of curiosity how this case can be described in DT?

-- 
With Best Regards,
Andy Shevchenko