[RFC PATCH 05/11] mfd: omap: control: core system control driver

[Eduardo Valentin]

Hello,

On Fri, Jun 01, 2012 at 02:43:54PM +0200, Cousson Benoit wrote:
> On 6/1/2012 2:30 PM, Shilimkar, Santosh wrote:
> >On Fri, Jun 1, 2012 at 7:29 PM, Tony Lindgren<tony at atomide.com>  wrote:
> >>* Cousson, Benoit<b-cousson at ti.com>  [120529 06:29]:
> >>>On 5/28/2012 1:35 PM, Eduardo Valentin wrote:
> >>>>>
> >>>>>Mmm, we can have up to 4 control module instances in OMAP4.
> >>>>>
> >>>>>Well, I'm not sure it worth considering them as separate devices. Is
> >>>>>that your plan as well?
> >>>>
> >>>>At least for now I was focusing on the ctrl_module_core ...
> >>>
> >>>OK, that's a good start already :-)
> >>>
> >>>>>But since they all have different base address, it will be trick to
> >>>>>handle them with only a single entry.
> >>>>
> >>>>Indeed. We can always add the support latter on.
> >>>>
> >>>>I am not sure what would be the best way to handle those instances though,
> >>>>and how they are going to expose APIs. Would need to have an instance bound
> >>>>to API set?
> >>>
> >>>We should not go to that path I guess. We should have an API at
> >>>children level independent of the underlying control module
> >>>partitioning.
> >>
> >>These should be separate driver instances for the control modules.
> >>
> >>And then the ioremapped area should ignore at least the padconf
> >>registers so drivers/pinctrl/pinctrl-simple can handle those. There
> >>should not be any dependencies to the SCM driver from pinctrl-simple,
> >>the core SCM driver can manage the clocks and trigger the save of
> >>padconf regs.
> >>
> >>Also we should allow MMC driver handle the MMC specific registers
> >>and USB driver(s) handle the USB specific registers if possible. Those
> >>should not live under drivers/mfd unless there are some dependencies
> >>other than trying to ioremap the whole SCM module instead of ioremapping
> >>in each driver.
> >>
> >>We can have a static map for the SCM, so ioremapping each driver
> >>individually should not be an issue.

Within this series I kept the original MFD setup in the original driver
found in the TI Android tree. But in fact, I agree with Tony here.

In general, ideally it would be good if we could setup this driver
family, having for each driver their respective resources;
a separated individual io area, IRQs, locking, APIs, etc.

But, the issue is to avoid having hard coded io areas in the code,
and move them  to auto generated data, DT/hwmod.

> >>
> >This sounds a good idea. With this we may not even need a core control
> >module drivers if all the individual drivers take care of the registers they
> >care. Mapping shouldn't be a problem as you mentioned.
> 
> We should keep the MFD for PM / OCP single port correctness. Other
> than that it will be mostly useless, indeed.

Then, I guess the idea of having 1 static mapping for the MFD core, either in DT or hwmod,
is to maintain the single OCP port we have today for SCM.

But, Benoit, IIRC, from our previous discussions, we won't have the io area for children
autogenerated anyway, as they don't have any OCP. I suppose this applies also for
clock data?

If children data won't be autogenerated, assuming we are relying on the autogenerated
data, it means that SCM core area is the boundary that will be ioremapped and protected.
But then we still have a flaw because children can still stomp between each other.
Not to mention non-children users and the pin ctrl simple.

That said, if we want to have these io areas properly assigned to their drivers,
we either have children area autogenerated, in DT / hwmod. Or we have them in the code.

Assuming you still want to have the MFD core, for matching w/ the OCP single port,
it would still need to have the resources for its children in the driver code.

Or we hack them in the DT entries. Like I already did in this series for the thermal IRQ :-)


> 
> Regards,
> Benoit
>