Re: [RFC 0/7] Qualcomm SMEM, SMD, RPM and regulators

[Bjorn Andersson <bjorn.andersson@sonymobile.com>]

On Tue 30 Sep 06:49 PDT 2014, Kumar Gala wrote:

> 
> On Sep 29, 2014, at 7:34 PM, Bjorn Andersson <Bjorn.Andersson@sonymobile.com> wrote:
> 
> > All Qualcomm platforms implements a shared heap among the processors in the
> > SoC, used for sharing data with other parts of the system.
> > 
> > One consumer of items from this heap is the "Shared Memory Driver", a ring
> > buffer based point-to-point communication mechanism used to send either stream
> > or packet based data to remote processors.
> > 
> > Starting with 8x74 this system is used to talk to the Resource Power Manager
> > (RPM), a power efficient "coprocessor" with responsibility of aggregate votes
> > from the various systems in the SoC related to regulators, clocks and bus
> > frequencies.
> > 
> > The PMIC regulators and root clocks in these platforms are only accessible via
> > the RPM, so to get access to these we need the full chain of smem, smd, rpm and
> > a regulator driver implemented. And that is exactly what this series provides.
> > 
> > 
> > A key outstanding question is where in the tree we should put the
> > implementation, for now I dropped them in drivers/soc/qcom but that's only
> > because I don't know where to put it otherwise. I have not found any equivalent
> > of the SMEM driver, SMD resembles mailbox and rpmsg - but comments in that
> > patch on why it's neither.
> > 
> > RPM is a mfd and regulator is a regulator :)
> 
> I still don’t see why RPM support for either A-family or B-family should
> exist in MFD vis drivers/soc/qcom.  What benefit is there in putting this in
> MFD?
> 
> I think both A and B-family support should be in drivers/soc/qcom for the
> current time being until we determine there is some framework that makes more
> sense in the future.  I almost see RPM more like a bus controller than
> anything else.  Something like an I2C bus controller that than has some set
> of devices off of that bus.
> 

When you look at what functionality the RPM exposes it has very much in common
with a PMIC. So after looking at this back and forth for months I think MFD is
a nice fit.

As with all the other pmics we could create a new subsystem (drivers/pmic?) for
this kind of devices that exposes variable size registers for children to read
and write.

But if you can convince the maintainers about that then we have a whole bunch
of stuff in mfd etc that we should move out, so let's not put this in
qcom-staging just for the sake of it.

Regards,
Bjorn