Re: [PATCH 1/2] dt-bindings: clk: Add binding for versal clocking wizard

[Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>]

On 03/10/2022 17:27, Michal Simek wrote:
>>
>> Exactly. The names xlnx,clocking-wizard and xlnx,clk-wizard-1.0 are
>> therefore not specific enough and mixing different devices.
> 
> And just to be clear these IPs can be combined with systems where the main cpu 
> can be Microblaze. I have also seen some vendors mixing RISC-V with Xilinx IPs.
> 
> Please look below.
>>
>>> And because this is fpga world none is really describing programmable logic by
>>> hand because it would take a look a lot of time. That's why I created long time
>>> ago device-tree generator (DTG) which gets design data and based on it generate
>>> device tree description. Newest version is available for example here.
>>> https://github.com/Xilinx/device-tree-xlnx
>>> There is also newer version called system device tree generato
>>> https://github.com/Xilinx/system-device-tree-xlnx
>>>
>>> Because of this infrastructure user will all the time get proper compatible
>>> string which is aligned with IP catalog.
>>
>> I don't think so. Let's skip for now "clk" and "clocking" differences
>> and assume both are "clocking". You have then compatibles:
>>
>> xlnx,clocking-wizard and xlnx,clocking-wizard-1.0
>>
>> and you said these are entirely different blocks.
>>
>> There is no way this creates readable DTS.
> 
> And I really thank you for this discussion to do it properly and have proper 
> compatible string and description for this block.
> 
> Shubhrajyoti: please correct me if I am wrong.
> 
> All Xilinx SOCs have programmable logic aligned with FPGAs. Zynq is based 28nm,
> ZynqMP (Ultrascale MPSOC) is based on 16nm and Versal is based on 7nm.
> 
> I think these clocking IPs are using low level primitives available in PL logic.
> Which means there is connection to fpga/pl technology instead of SOC family and 
> main cpu.

Then maybe the compatibles (and device names) should have that fpga/pl
technology used to differentiate between them?

> It can be of course said that if this is ZynqMP SOC that IP A is used. The same 
> for Versal SOC. But for soft cores this can't be said.
> 
> Would it be better to reflect PL technology in compatible string?

Yes, although we might misunderstand what PL technology is. 28/16/7 nm
is the size of transistor or the process. Even two different processes
can use same type of technology, e.g. FinFET:
https://en.wikipedia.org/wiki/14_nm_process
https://en.wikipedia.org/wiki/10_nm_process

You could have very similar (or even the same) designs done in 28 nm and
16 nm. Does it mean these are entirely different devices? Not
necessarily... Maybe they are, maybe not, but is the size of process
differentiating? I actually don't know what's there in 28/16/7, I am
just saying that number alone might not mean different technology.
Programming API could be the same, inputs/outputs could be the same.
Just the size of transistor is different...

> 
> xlnx,clocking-wizard-vX.X (used now for ZynqMP and previous families)
> xlnx,clocking-wizard-7nm-vX.X (or find better name names which reflect PL logic 
> type)

Best regards,
Krzysztof