Re: [RFC 3/3] am35xx: pm: Hook-up with TPS65023

[Nishanth Menon <nm@ti.com>]

Premi, Sanjeev wrote, on 03/08/2011 05:56 PM:
[...]
>>>>>
>>>>> Add glue logic to hook-up AM35x processors
>>>>> with TPS65023.
>>>> It seems that you are not really using Voltage layer for any
>> interaction
>>>> with TPS65023 as you are not using VP and VC. Then what is the
>>> purpose of
>>>> registering this PMIC with Voltage layer. I fail to understand the
>>> purpose
>>>> of this patch series.
>>>
>>> [sp] Then, can you suggest how do I get the AM35x EVM to boot? Given
>>> the
>>>       current limitation of all voltage related data being "extracted"
>> from
>>>       the voltage layer - which expects only TWLx0y0 PMICs.
>> Pls use regulator framework for setting the voltage for your PMIC.
>
> [sp] If you follow the current framework, voltage.c is ingrained with
> OMAP3 initialization and same holds good for opp[_3xxx_data].c.
>
> It would have been great if current implementation allowed pluggable
> interface to a PMIC. As it stands today, it isn't.
Partly true - there are PMIC abstractions that have been attempted in 
current voltage layer - but only TWL is used - so yep, implementation 
probably needs evolution and yes again that TPS here is a perfect choice 
to make the implementation generic and scalable.

>
> Until implemented, workarounds have to be put in place to get platform(s)
> working. Though patch series is being done for AM35x with TPS65023. The
> similar workaround would be required for OMAP3 with TPS65023 (for example).
>
> ...and I expect workaround to be simple and unobtrusive.
>
> See:
> core_initcall(omap_voltage_early_init);
> device_initcall(omap3_opp_init);
> late_initcall(omap2_common_pm_late_init);

Thinking from a generic soln perspective, lets try and split this into 
multiple issues:
a) OPP and Voltage layer voltages - these need to be PMIC aware as well. 
See my comment on http://marc.info/?l=linux-omap&m=129955003611548&w=2 
-> essentially means that pmic_voltage information should be registered 
earlier to opp init

b) split up structure information for voltage layer - this should be 
done in a manner to make PMIC, Board and OMAP SoC information independent.

c) Ability to plug in multiple PMICs in two manners:
    i) use PMIC with VC/VP/SR combinations.
    ii) use PMIC which is plugged on regulator frameworks.

If anyone is attempting cleanups, it might be a good idea to base on the 
accepted cleanups from pm-core branch which is planned for 39-rc1 to 
prevent any surprises ;)

-- 
Regards,
Nishanth Menon