Re: [PATCH RFC v2 1/5] dt-bindings: clock: histb-clock: Add missing common clock and Hi3798MV200 specific clock definition

[Krzysztof Kozlowski]

On 20/02/2024 18:29, Yang Xiwen wrote:
> On 2/21/2024 1:06 AM, Krzysztof Kozlowski wrote:
>> On 20/02/2024 17:31, Yang Xiwen wrote:
>>> On 2/21/2024 12:25 AM, Krzysztof Kozlowski wrote:
>>>> On 20/02/2024 17:19, Yang Xiwen wrote:
>>>>> On 2/21/2024 12:13 AM, Krzysztof Kozlowski wrote:
>>>>>> On 20/02/2024 15:06, Yang Xiwen wrote:
>>>>>>> On 2/20/2024 6:10 PM, Krzysztof Kozlowski wrote:
>>>>>>>> On 17/02/2024 13:52, Yang Xiwen via B4 Relay wrote:
>>>>>>>>> From: Yang Xiwen <forbidden405@xxxxxxxxxxx>
>>>>>>>>>
>>>>>>>>> According to the datasheet, some clocks are missing, add their
>>>>>>>>> definitions first.
>>>>>>>>>
>>>>>>>>> Some aliases for hi3798mv200 are also introduced.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Yang Xiwen <forbidden405@xxxxxxxxxxx>
>>>>>>>>> ---
>>>>>>>>>      include/dt-bindings/clock/histb-clock.h | 21 +++++++++++++++++++++
>>>>>>>>>      1 file changed, 21 insertions(+)
>>>>>>>>>
>>>>>>>>> diff --git a/include/dt-bindings/clock/histb-clock.h b/include/dt-bindings/clock/histb-clock.h
>>>>>>>>> index e64e5770ada6..68a53053586a 100644
>>>>>>>>> --- a/include/dt-bindings/clock/histb-clock.h
>>>>>>>>> +++ b/include/dt-bindings/clock/histb-clock.h
>>>>>>>>> @@ -58,6 +58,27 @@
>>>>>>>>>      #define HISTB_USB3_UTMI_CLK1		48
>>>>>>>>>      #define HISTB_USB3_PIPE_CLK1		49
>>>>>>>>>      #define HISTB_USB3_SUSPEND_CLK1		50
>>>>>>>>> +#define HISTB_SDIO1_BIU_CLK		51
>>>>>>>>> +#define HISTB_SDIO1_CIU_CLK		52
>>>>>>>>> +#define HISTB_SDIO1_DRV_CLK		53
>>>>>>>>> +#define HISTB_SDIO1_SAMPLE_CLK		54
>>>>>>>>> +#define HISTB_ETH0_PHY_CLK		55
>>>>>>>>> +#define HISTB_ETH1_PHY_CLK		56
>>>>>>>>> +#define HISTB_WDG0_CLK			57
>>>>>>>>> +#define HISTB_USB2_UTMI0_CLK		HISTB_USB2_UTMI_CLK
>>>>>>>> Why? It's anyway placed oddly, the entries are ordered by number/value.
>>>>>>> So this is somewhat broken at the beginning. It named after
>>>>>>> histb-clock.h but actually they are all clocks for Hi3798CV200 SoC. For
>>>>>>> Hi3798MV200(also a HiSTB SoC), there is one additional UTMI clock.
>>>>>>>
>>>>>>>
>>>>>>> What solution do you prefer? rename UTMI_CLK to UTMI0_CLK, add UTMI1_CLK
>>>>>>> after it and increment all the indexes after it? Then the diff would be
>>>>>>> very ugly.
>>>>>> I still don't understand what is the problem you are trying to solve
>>>>>> here. Your commit msg says add missing ID, but that ID -
>>>>>> HISTB_USB2_UTMI_CLK - is already there.
>>>>>>
>>>>>> I also do not get why there is a need to rename anything.
>>>>> Because there are two USB2_UTMI_CLKs in total, at least for Hi3798MV200.
>>>>> UTMI1 is missing here. For other HiSTB SoCs, there could be even more.
>>>> My comment was under UTMI0. We do not talk about UTMI1...
>>>>
>>>>> If we add USB2_UTMI1_CLK, it looks silly to keep USB2_UTMI_CLK without
>>>>> renaming it to UTMI0. Just like all the other clocks. E.g.
>>>>> I2Cn_CLK(n=0,1,2,3,4) etc.., so the same for USB2_UTMI_CLK.
>>>> Then place it next to old name and explain why it is deprecated with
>>>> comment.
>>>
>>> Do we need to keep the old name? I can fix all the users (only
>>> hi3798cv200.dtsi) in next version and drop this name directly. Is that
>> All users in all projects? That might be tricky. And even for Linux
>> kernel, how can you do it in a bisectable way? Just keep old name.
>>
>>
>>> okay? Should i insert UTMI1_CLK to the middle and re-index all the
>>> macros after it? Or simply add it to the tail?
>> Bindings and header constants are ABI, so you cannot change them.
> 
> 
> This file should be renamed to hi3798cv200-clock.h, it shouldn't be 
> called histb-clock.h from the beginning. Now I have to workaround this 
> in a dirty way. What if another HiSTB SoC has 3 or more UTMI_CLKs? Do we 
> need to add more definitions to the end of the file? The file is gonna 

That's not a big problem, but indeed shows poor design of the driver and
bindings.

> to be more and more unreadable with scattered clock definitions.
> 
> 
> Do you think it's acceptable to create a new header file instead? I 

This depends on the purpose of it. In general every SoC follows that
concept - binding headers are per given clock controller, not even per SoC.

> think we don't need a generic histb-clock.h. Each SoC should maintain 
> their own clock indexes header file. Maybe we can rename it to 
> hi3798cv200-clock.h and include it from a new histb-clock.h (also mark 
> this generic header file deprecated and only for hi3798cv200). Then I'll 
> create hi3798mv200-clock.h header file instead. So we don't have to 
> workaround this.


Best regards,
Krzysztof