[PATCHv2 8/9] ARM: dts: omap4: add clkctrl nodes

[Tero Kristo]

On 20/03/17 16:35, Tony Lindgren wrote:
> * Tero Kristo <t-kristo at ti.com> [170320 06:27]:
>> On 17/03/17 23:41, Tero Kristo wrote:
>>> On 17/03/17 17:43, Tony Lindgren wrote:
>>>> * Tero Kristo <t-kristo at ti.com> [170317 02:12]:
>>>>> Add clkctrl nodes for OMAP4 SoC. These are going to be acting as
>>>>> replacement for part of the existing clock data and the existing
>>>>> clkctrl hooks under hwmod data.
>>>>
>>>> It seems to be actually this patch in the series that breaks booting
>>>> if the last patch in this series is not applied.
>>>
>>> Yea I guess we need to squash this and patch #9, I'll double check next
>>> week. Sorry about not catching this one.
>>
>> Yes, patch #8 and #9 must be squashed. What happens with patch #8 alone, is
>> that you get all the clkctrl clocks registered, but as nobody is using them,
>> the clock core disables them later on in clk_disable_unused.
>
> So how hard would it to do a mixed approach where we don't need
> to have everything in place in the dts in order to flip a SoC?
>
> This atomic transaction worries me..

Well, you could do few things.

- You can use clk_ignore_unused cmdline option between patch #8 and #9 
and it should work, but will most likely break PM.
- You can introduce the DTS changes on a clkctrl module basis, e.g. 
introduce the node for l4_per_clkctrl along with the clock node changes 
for the same
- Do some sort of split between DTS / vs. clkctrl data introduction 
within the clock driver, this forces the files to be in sync with each 
other though.
- Try parsing the DTs for matching clock handles while registering DT 
clocks, if not found, toss our a pr_warn and skip registering the clock. 
Will be very ineffective though as one needs to parse the whole DTS to 
find the clock entries. This clock parsing could be dropped later.

-Tero