Re: [PATCH 18/18] ipu3: Add driver for dummy INT3472 ACPI device

[Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>]

On Thu, Dec 03, 2020 at 12:37:12PM +0000, Dan Scally wrote:
> On 02/12/2020 15:08, Andy Shevchenko wrote:
> > On Wed, Dec 02, 2020 at 02:42:28PM +0200, Laurent Pinchart wrote:
> >> On Wed, Dec 02, 2020 at 01:09:56PM +0200, Sakari Ailus wrote:
> >>> On Tue, Dec 01, 2020 at 08:37:58PM +0200, Laurent Pinchart wrote:
> > 
> > ...
> > 
> >> I think we should consider ACPI to be a hack in the first place :-)
> > 
> > I feel that about DT (and all chaos around it) but it's not a topic here.
> > 
> >>> Could this be just one more platform device for each of the three cases (or
> >>> one for the two latter; I'm not quite sure yet)?
> >>
> >> Using MFD for this seems a bit overkill to me. I won't care much as I
> >> won't maintain those drivers, but the current situation is complex
> >> enough, it was hard for me to understand how things worked. Adding yet
> >> another layer with another platform device won't make it any simpler.
> >>
> >> If we want to split this in two, I'd rather have a tps68470 driver on
> >> one side, without ACPI op region support, but registering regulators,
> >> GPIOs and clocks (without using separate drivers and devices for these
> >> three features), and an INT3472 driver on the other side, with all the
> >> ACPI glue and hacks. The tps68470 code could possibly even be structured
> >> in such a way that it would be used as a library by the INT3472 driver
> >> instead of requiring a separate platform device.
> > 
> > I'm afraid TPS68470 is MFD in hardware and its representation in the MFD is
> > fine. What we need is to move IN3472 pieces out from it.
> > 
> > And I agree with your proposal in general.
> 
> Way back when I first joined this project we thought we needed i2c
> drivers for driving the tps68470's clks and regulators. Tsuchiya found
> some in an old Intel tree; they needed some minor tweaks but nothing
> drastic. And I think they're designed to work with the mfd driver that's
> already in the kernel.
> 
> So, can we do this by just checking (in a new
> platform/x86/intel_skl_int3472.c) for a CLDB buffer in the PMIC, and
> calling devm_mfd_add_devices() with either the GPIO and OpRegion drivers
> (if no CLDB buffer found) or with the GPIO, clk and regulator drivers
> (If there's a CLDB and it's not a discrete PMIC). Or else, using the
> code from this patch directly in the platform driver if the CLDB says
> it's a discrete PMIC?

Lee, who is a maintainer of MFD, is quite sensitive about this.
I don't think he would approve this (however I see 8 drivers
that are using MFD API out of drivers/mfd).

-- 
With Best Regards,
Andy Shevchenko