[PATCH v2 05/11] ARM: dt: tegra30: Add device node for APB MISC

[Stephen Warren]

On 01/04/2013 04:56 AM, Laxman Dewangan wrote:
> On Friday 04 January 2013 09:30 AM, Stephen Warren wrote:
>> On 01/03/2013 08:23 PM, Prashant Gaikwad wrote:
>>> On Friday 04 January 2013 08:35 AM, Stephen Warren wrote:
>>>> On 01/03/2013 06:48 PM, Prashant Gaikwad wrote:
>>>>> On Thursday 03 January 2013 09:41 PM, Stephen Warren wrote:
>> ...
>>>>>> OK. It sounds like we need a true APB MISC driver then, to
>>>>>> abstract the
>>>>>> differences; the clock driver really shouldn't be touching the APB
>>>>>> MISC
>>>>>> registers in all likelihood, unless a subset of the sections you
>>>>>> mention
>>>>>> above are truly dedicated to clock functionality.
>>>>> I don't think it is a good idea to create a driver for APB MISC, all
>>>>> registers are used by different drivers.
>>>> Well, it's even worse to have a bunch of other drivers randomly trample
>>>> on a set of registers they don't own.
>>>>
>>>>> Only chip id revision registers are used in clock driver.
>>>> There are already global variables exposed by the Tegra fuse driver;
>>>> can
>>>> you just read those?
>>> It is not about variables or some value, we have to read some apb
>>> register to flush the write operation in apb bus before we disable
>>> peripheral clock.
>>> We are using chip id revision register for this purpose.
>> Ah. That's definitely not something the clock driver should be doing
>> directly. It's probably OK to add a custom Tegra-specific function to
>> some file in arch/arm/mach-tegra to implement this. Even better would be
>> a full bus driver for the APB bus, but that's probably too much bloat
>> for now.
> 
> I think individual driver should take care of flushing the write
> operation inplace of clock driver.
> Atleast I moved flushing to i2c and spi for these drivers. Polluting
> clock driver here does not make sense here.

That does seem like a reasonable assertion.

Still, I believe the clock driver currently does this already today, so
removing it might be a regression. I think we want to:

a) Maintain this feature in the clock driver rework for now.
b) Audit and fix any device drivers.
c) Remove the feature from the clock driver.

Of course, if (b) is so easy to do that you don't need to do (a) or (c)
at all, that's fine too, but I believe Prashant is under time pressure
to get the common clock rework done before his vacation.