Re: [PATCH 1/2] mfd: pm8x41: add support for Qualcomm 8x41 PMICs

[Courtney Cavin <courtney.cavin@sonymobile.com>]

On Fri, May 09, 2014 at 02:45:30PM +0200, Ivan T. Ivanov wrote:
[...]
> > > > 
> > > > I'm thinking we should probably have a generic compatible entry as well,
> > > > "qcom,pmic-qpnp" or similar.  We should still specify in the binding
> > > > that PMIC slaves specify a version-specific string as well as the
> > > > generic string.  That is, a slave should have:
> > > > 
> > > > 	compatible = "qcom,pm8841", "qcom,pmic-qpnp";
> > > > 
> > > > ...in case we would ever need to differentiate in the future.
> > > > 
> > > > (I recall that in a previous version I had done this, but I don't
> > > > remember why I had changed it..)
> > >
> > > I gave this some thought but came to the conclusion that there is no
> > > benefit of adding a generic compatible to a new binding.  Please clarify
> > > a use-case where this would be ... useful.
> > 
> > Having a generic compatible entry allows for easily supporting new PMICs
> > without having to add yet another vacuous entry in the ID table.  In
> > this case I think it's perfectly acceptable given that this driver isn't
> > really defining a programming model for a specific device, but rather
> > acting much more like a bus.
> > 
> > Requiring a specific PMIC listed before a generic one allows us an
> > escape hatch in the future if for some reason we need to add a quirk for
> > a specific PMIC.
> 
> Is there a conclusion on this issue? I am voting for generic name :-)
> "qcom,pm-qpnp".

Josh and I have discussed this offline, and I think we have come to the
conclusion that this should be a generic driver with only a generic
binding.  The current proposed name is "spmi-ext", as there is specific
functional relation to Qualcomm, PMICs or QPNP.

Further, the binding documentation should be specific to pm8[89]41 as
'mfd/pm8x41.txt', and should contain the compatibles:
	- "qcom,pm8941", "spmi-ext"
	- "qcom,pm8841", "spmi-ext"

This naming has been discussed to death, so a few more shed color
suggestions can't possibly hurt.

> Further complication is that several sub function drivers expect to
> runtime detect the exact version of the controller ("qcom, qpnp-iadc",
> "qcom, qpnp-vadc", "qcom, qpnp-linear-charger").  This is realized by the
> exported function of the driver "qcom, qpnp-revid". Would it be good
> idea to merge qpnp-revid and "qcom,pm-qpnp" driver?

Each block within the PMICs have--undocumented--version registers, so a
global version number is not particularly useful.  A good example of
this is the ADC code [1], as you mentioned.

-Courtney

[1] https://www.codeaurora.org/cgit/quic/la/kernel/msm-3.10/tree/drivers/thermal/qpnp-adc-tm.c#n469