[PATCH 2/8] clk: keystone: Add gate control clock driver

[Santosh Shilimkar]

On Tuesday 20 August 2013 10:22 PM, Mike Turquette wrote:
> Quoting Santosh Shilimkar (2013-08-20 15:54:15)
>> On Tuesday 20 August 2013 06:41 PM, Mike Turquette wrote:
>>> Quoting Santosh Shilimkar (2013-08-20 14:55:56)
>>>> On Tuesday 20 August 2013 05:30 PM, Mike Turquette wrote:
>>
>> [...]
>>
>>>>>> They are bit different w.r.t OMAP. LPSC itself is the clock control of the
>>>>>> IP. The LPSC number in the bindings is actually the specific number which
>>>>>> is used to reach to the address space of the clock control. One can view
>>>>>> that one as clock control register index.
>>>>>
>>>>> Thanks for the information. I have a further question about then: are
>>>>> the LPSC clocks really module clocks that belong to the IP that they are
>>>>> gating?
>>>>>
>>>> LPSC controls the clock enable/disable to the IP/module so answer is yes.
>>>> In certain cases LPSC controls clock to more than one IP as well.
>>>>
>>>>> If so then they could be defined within the node defining their parent
>>>>> IP. That might be enough to get rid of the LPSC index value. Again I
>>>>> might be over-engineering it. Just trying to get an understanding.
>>>>>
>>>> Am not sure I follow you here on not having the LPSC index. Sorry. 
>>>
>>> How are the 'reg' property and the 'lpsc' property related? Does the
>>> lpsc property modify the register address used to access the clock
>>> control bits?
>>>
>> Yes it does. Currently all nodes use fix address and then lpsc is
>> used as an index.
> 
> Ok cool. Well the reason I brought that up was because I even had the
> idea to define these module clocks within the module nodes that own them
> in DT. I am way outside of my DT knowledge at this point but I wonder
> if the following type of binding is possible:
> 
> module: module at 4a308200 {
> 	#address-cells = <1>;
> 	#size-cells = <1>;
> 	reg = <0x4a308200 0x1000>;
> 
> 	clock {
> 		#clock-cells = <0>;
> 		compatible = "keystone,psc-clk";
> 		clocks = <&chipclk3>;
> 		clock-output-names = "debugss_trc";
> 		reg = <0x0256>;
> 		pd = <1>;
> 	};
> };
> 
> Again, my DT knowledge is pretty limited, but could the reg property of
> the clock be directly affected by the parent node? That seems like it
> could nicely model what the hardware is really doing.
> 
The module(I assume you mean IP here) reg address space is separate than
that used for clock control so that doesn't fit as such. Traditionally
clock controls even though targeted for specific modules sits in different
control as at least seen on OMAP and Keystone. OCP wrappers on OMAP
were bit of exceptions but they were little bit of glue logic without
much control within the address space.

>> But I think we can do better by just using the
>> right(offset) address in the reg property. Will have a look at it
>> and see what I can do here.
> 
> This also solves the problem nicely.  Thanks for putting up with my
> silly questions ;-)
> 
You asked right and good questions.

Regards,
Santosh