[PATCH V3] clk: palmas: add clock driver for palmas

[Nishanth Menon]

On Tue, Oct 8, 2013 at 3:43 PM, Stephen Warren <swarren at wwwdotorg.org> wrote:
> On 10/08/2013 11:33 AM, Nishanth Menon wrote:
>> On 10/08/2013 12:08 PM, Stephen Warren wrote:
>>> On 10/08/2013 10:14 AM, Nishanth Menon wrote:
>>>> On 10/08/2013 09:39 AM, Laxman Dewangan wrote:
>>>>> Thanks Nishanth for review.
>>>>>
>>>>> On Tuesday 08 October 2013 06:59 PM, Nishanth Menon wrote:
>>>>>> On 10/08/2013 08:21 AM, Laxman Dewangan wrote:
>>>>>>> Palmas devices has two clock output CLK32K_KG and CLK32K_KG_AUDIO
>>>>>> not all palmas devices have 2 clocks - example: tps659038
>>>>>
>>>>> This is for generic palmas and I have seen it for TPS65913, TPS65914,
>>>>> TPS80036. If the generic one is not compatible then it need to add
>>>>> device specific and at that time, it is require to update the binding
>>>>> document accordingly.
>>>>
>>>> ?? you do have two clocks inside the device they should be represented
>>>> as two compatible entities - that simplifies everyone's life.
>>>
>>> I think the terminology you're using here is quite confusing.
>>>
>>> Are you talking about having two different compatible values for two
>>> different HW designs, where those different designs implement different
>>> sets of clocks (which makes sense), or two different DT nodes for two
>>> different clocks (which IMHO doesn't always, unless those different
>>> clocks *truly* are separate IP blocks with completely independent
>>> register regions, and where those IP blocks are likely to be re-used
>>> as-is in other chips).
>>
>> clk32k and clk32k_audio are two different resources and since these
>> are two different resource instances - a "compatible" matching an
>> actual device is my suggestion.
>
> The fact that two clocks are two different resources isn't at all
> relevant to DT structure. HW module design is what's relevant.
>
>> clk32k and clk32k_audio are two different resources because they have
>> their specific set of controls registers and may even be independently
>> present in a Palmas variant.
>
> That's a better argument, assuming that: The registers for those two
> clocks aren't randomly interleaved with other registers within the HW
> module. That would imply that the clock registers aren't independant HW
> blocks.

You would be unpleasantly surprised if register offsets is a standard
to determine if IP blocks are independent instances or not. it is just
integration level decision, and unfortunately, not all h/w integration
guys care a lot about s/w :(.

>
>> To highlight this: The example of tps659038 where clk32k is not
>> present, but clk32k_audio is present (and happens to be disabled by
>> default - thanks to an OTP on the chip - on platform like DRA7-evm, it
>> is used to for 32k clk for wlan -currently hacked in u-boot using
>> plain i2c writes[1] - yes it is yucky).
>
> That can easily be handled by having separate compatible values for a
> monolithic overall Palmas or Palmas-clock node/HW-block. The fact that
> different chips are different doesn't, in and of itself, need to
> influence whether the different clocks are represented as different DT
> nodes.

So, the fact that a clock does not exist on a variation of palmas is
not sufficient proof that the blocks are independent.

I am not sure, given the lack of public documentation, short of
sharing rtl (which I cant ;) ), i might let this debate flame out..

>
>> Obviously, there are many ways to implement this. based on the current
>> implementation, it indicates that if i create a node with
>> "ti,palmas-clk" -i'd create two clocks - that is wrong for tps659038.
>>
>> Now (with the current approach), if I have to create a one clock for
>> tps659038, i have to fix the for adding clock providers, add up
>> "ti,tps659038-clk" etc.. it is doable - but IMHO, I dont need to do it
>> with only the relevant nodes in dts.
>>
>> Further, it has no way to indicate that device X uses clock Y using
>> clocks =<&xyz> either.
>
> Sorry, I just don't understand that.
>
> If a clock provider provides two clocks, it could number then e.g. 0 and
> 1. Clock consumers would reference those IDs. If a different chip that
> uses the same binding only supports one of those two clocks, just have
> the driver return an error if the DT attempts to use/reference the
> invalid clock ID; nothign could be simpler.

fair enough.

Regards,
Nishanth Menon