Re: 回覆: 回覆: [net-next 4/4] net: ftgmac100: add RGMII delay for AST2600

[Krzysztof Kozlowski <krzk@kernel.org>]

On 18/03/2025 14:51, Andrew Lunn wrote:
> On Tue, Mar 18, 2025 at 05:34:08AM +0000, Jacky Chou wrote:
>> Hi Andrew,
>>
>> Thank you for your reply.
>>
>>>> The RGMII delay of AST2600 has a lot of steps can be configured.
>>>
>>> Are they uniformly space? Then it should be a simple formula to calculate? Or
>>> a lookup table?
>>
>> There are fixed delay values by step. I list below.
>> AST2600 MAC0/1 one step delay = 45 ps
>> AST2600 MAC2/3 one step delay = 250 ps
> 
> That is messy.
> 
>> I calculate all step and emulate them.
>> The dt-binding will be like below.
>> rx-internal-delay-ps:
>>     description:
>>       Setting this property to a non-zero number sets the RX internal delay
>>       for the MAC. ... skip ...
>>     enum:
>>       [45, 90, 135, 180, 225, 250, 270, 315, 360, 405, 450, 495, 500, 540, 585, 630, 675, 
>>        720, 750, 765, 810, 855, 900, 945, 990, 1000, 1035, 1080, 1125, 1170, 1215, 1250, 
>>        1260, 1305, 1350, 1395, 1440, 1500, 1750, 2000, 2250, 2500, 2750, 3000, 3250, 3500, 
>>        3750, 4000, 4250, 4500, 4750, 5000, 5250, 5500, 5750, 6000, 6250, 6500, 6750, 7000, 
>>        7250, 7500, 7750, 8000]
> 
> Can the hardware do 0 ps?
> 
> So this list is a superset of both 45ps and 250ps steps?

git grep multipleOf:

e.g.
oneOf:
 - minimum: 45
   maximum: ...
   multipleOf: 45
 - minimum: 1500
   maximum: ...
   multipleOf: 250

> 
> Lets see what the DT Maintainers say, but it could be you need two
> different compatibles for mac0/1 to mac2/3 because they are not
> actually compatible! You can then have a list per compatible.
If this is the only, *only* difference, then just go with vendor
property matching register value... but oh, wait, how person reading and
writing the DTS would understand if "0x2" means 90 ps or 1750 ps? I
don't see how the original binding was helping here in total. Just
moving the burden from driver developer to DTS developer. :/

If different instances are not the same, means the devices are not the
same, so two compatibles seem reasonable.

Best regards,
Krzysztof