Re: [PATCH net-next 1/2] dt-bindings: net: tja11xx: add nxp,refclk_in property

[Krzysztof Kozlowski]

On 19/08/2022 15:44, Andrew Lunn wrote:
> On Fri, Aug 19, 2022 at 02:37:36PM +0300, Krzysztof Kozlowski wrote:
>> On 19/08/2022 12:37, Wei Fang wrote:
>>>>
>>>>> +          in RMII mode. This clock signal is provided by the PHY and is
>>>>> +          typically derived from an external 25MHz crystal. Alternatively,
>>>>> +          a 50MHz clock signal generated by an external oscillator can be
>>>>> +          connected to pin REF_CLK. A third option is to connect a 25MHz
>>>>> +          clock to pin CLK_IN_OUT. So, the REF_CLK should be configured
>>>>> +          as input or output according to the actual circuit connection.
>>>>> +          If present, indicates that the REF_CLK will be configured as
>>>>> +          interface reference clock input when RMII mode enabled.
>>>>> +          If not present, the REF_CLK will be configured as interface
>>>>> +          reference clock output when RMII mode enabled.
>>>>> +          Only supported on TJA1100 and TJA1101.
>>>>
>>>> Then disallow it on other variants.
>>>>
>>>> Shouldn't this be just "clocks" property?
>>>>
>>>>
>>> This property is to configure the pin REF_CLK of PHY as a input pin through phy register,
>>> indicates that the REF_CLK signal is provided by an external oscillator. so I don't think it's a
>>> "clock" property.
>>
>> clocks, not clock.
>>
>> You just repeated pieces of description as an counter-argument, so this
>> does not explain anything.
>>
>> If it is external oscillator shouldn't it be represented in DTS and then
>> obtained by driver (clk_get + clk_prepare_enable)? Otherwise how are you
>> sure that clock is actually enabled? And the lack of presence of the
>> external clock means it is derived from PHY?
> 
> Using the common clock framework has been discussed in the past. But
> no PHY actually does this. When the SoC provides the clock, a few PHYs
> do make use of the common clock framework as clock consumers to ensure
> the clock is ticking.

IOW, all DTSes would have a fixed clock stub without any logic usable by
Common CF (like enabling)?

> Plus, as the description says, this pin can be either a clock producer
> or a consumer. I don't think the common clock code allows this. It is
> also not something you negotiate between the MAC and PHY. The hardware
> designer typically decides based on the MAC and PHY actually used. So
> this is a fixed hardware property.

Indeed.

Anyway the property name and typo need fixes.
Best regards,
Krzysztof