Re: [PATCH v4 2/2] dt-bindings: nand: meson: refine Amlogic NAND controller driver

[Liang Yang <liang.yang@amlogic.com>]

Hi Miquel,

On 2022/4/20 20:16, Miquel Raynal wrote:
> [ EXTERNAL EMAIL ]
> 
> Hi Liang,
> 
>>>> +maintainers:
>>>> +  - liang.yang@amlogic.com
>>>> +
>>>> +properties:
>>>> +  compatible:
>>>> +    enum:
>>>> +      - "amlogic,meson-gxl-nfc"
>>>> +      - "amlogic,meson-axg-nfc"
>>>> +
>>>> +  reg:
>>>> +    maxItems: 2
>>>> +
>>>> +  '#address-cells':
>>>> +    const: 1
>>>
>>> Not sure this property is needed.
>> this is for the subnode, such as nand@0.
> 
> Yes but if you refer to nand-controller.yaml you no longer need these.
ok, i will try it.
> 
>>>    
>>>> +
>>>> +  '#size-cells':
>>>> +    const: 0
>>>
>>> Ditto. Plus, this one looks wrong anyway.
>> this is for the subnode, such as nand@0. do you mean s/''/""/?
> 
> Sorry, this is not "wrong anyway", my fault. But still, you don't need
> this property for the same reason as above.
ok.
> 
>>>    
>>>> +
>>>> +  reg-names:
>>>> +    items:
>>>> +      - const: nfc
>>>> +      - const: emmc
>>>
>>> Why do you need the emmc register map? Do you really need to perform a
>>> register access there?
>> yes, we have to access the emmc register map. because the NFC clock comes from SDEMMC_CLOCK register.
> 
> But if it's a clock you should get the clock and call
> clk_prepare_enable(), you don't need to poke directly in the registers.
> Do you?

no, it doesn't. it is special and the reason why need to implement a MMC 
sub clock driver previously. also we don't implement it as a clock 
provider in drivers/clk/meson/, because the SDEMMC_CLOCK register is 
internal in MCI controller.
> 
>>>> +examples:
>>>> +  - |
>>>> +    #include <dt-bindings/clock/axg-clkc.h>
>>>> +    #include <dt-bindings/interrupt-controller/arm-gic.h>
>>>> +    apb {
>>>> +      #address-cells = <2>;
>>>> +      #size-cells = <2>;
>>>
>>> Not sure you need this upper node in the example.
>> use the upper node to indicate the "#address-cells" and "#size-cells". if i do not do that, dt_binding_check will report:
>>    ".....reg:0: [0, 30720, 0, 256] is too long" and
>>    ".....reg:1: [0, 28672, 0, 2048] is too long".
> 
> ok, maybe, I'll let bindings maintainer review that.

ok. thanks.

> 
>>>    
>>>> +      nand-controller@7800 {
>>>> +        #address-cells = <1>;
>>>> +        #size-cells = <0>;
>>>> +        compatible = "amlogic,meson-axg-nfc";
>>>> +        reg = <0x0 0x7800 0x0 0x100>,
>>>> +              <0x0 0x7000 0x0 0x800>;
>>>> +        reg-names = "nfc", "emmc";
>>>> +
>>>> +        interrupts = <GIC_SPI 34 IRQ_TYPE_EDGE_RISING>;
>>>> +        clocks = <&clkc CLKID_SD_EMMC_C>,
>>>> +                 <&clkc CLKID_FCLK_DIV2>;
>>>> +        clock-names = "core", "device";
>>>> +
>>>> +      };
>>>> +    };
>>>> +...
> 
> 
> Thanks,
> Miquèl
> 
> .