Re: [RFC PATCH 1/2] ACPI / PNP: Don't add "enumeration_by_parent" devices

[John Garry]

On 30/04/2018 11:46, Lee Jones wrote:
> On Mon, 30 Apr 2018, John Garry wrote:
>
> > > > > > So we using the mfd_cell to match child devices using _HID. At a glance, I
> > > > > > don't actually see other drivers to use mfd_cell_acpi_match.pnpid .
> > > > > >
> > > > > > Anyway we don't use static tables as we need to update the resources of the
> > > > > > cell dynamically. However I do look at a driver like intel_quark_i2c_gpio.c,
> > > > > > and this dynamically modifies the value of global mfd_cell array here:
> > > > > > https://elixir.bootlin.com/linux/latest/source/drivers/mfd/intel_quark_i2c_gpio.c#L266
> > > > > >
> > > > > > I know the cell array is only used at probe time, but this did not look to
> > > > > > be good standard practice to me.
> > > > >
> > > > > Lots of drivers do this to supply dynamic data.  If there is no other
> > > > > sane way of providing such data, it's fine to do.  Although each
> > > > > situation should be dealt with on a case-by-case basis.
> > > >
> > > > Hi Lee,
> > > >
> > > > Thanks for your input.
> > > >
> > > > I do see others drivers which use dynamic mem for the mfd_cells (like
> > > > cros_ec_dev.c), so what we're doing in this driver already is not totally
> > > > unchartered territory. But creating the MFD cells from the ACPI table could
> > > > be ...
> > >
> > > Right.  I don't normally like mixing platform data technologies (MFD,
> > > ACPI and DT).  I normally NACK patches which take information from
> > > Device Tree and populate MFD cells with it.  ACPI would be the same I
> > > guess.
> >
> > Oh, well that is what we have in this driver. So what's the preferred
> > approach? Just not use MFD model at all if ACPI/DT needs to be scanned for
> > data to create the cells?
>
> I've just seen the driver - yuk!
>
> Why are you using the MFD API outside of MFD anyway?

Hi Lee,

This goes back a bit. The original thread was here:
https://lkml.org/lkml/2017/6/13/581

Essentially a method was required to model this host to support platform 
device children for ACPI FW, and this did the job. In hindsight I think 
that there was a misunderstanding in recommending MFD since the devices 
attached are not fixed - hence the dynamic part.

Cheers,
John
>