[PATCH] arm64: xilinx: Move address/size-cells to proper locations

[Krzysztof Kozlowski]

On 11/01/2024 09:48, Michal Simek wrote:
> 
> 
> On 1/11/24 09:18, Krzysztof Kozlowski wrote:
>> On 11/01/2024 09:10, Michal Simek wrote:
>>>
>>>
>>> On 1/11/24 08:56, Krzysztof Kozlowski wrote:
>>>> On 11/01/2024 08:09, Michal Simek wrote:
>>>>>
>>>>>
>>>>> On 1/10/24 22:27, Krzysztof Kozlowski wrote:
>>>>>> On 10/01/2024 14:35, Michal Simek wrote:
>>>>>>> Move cells to board dtsi files from generic zynqmp.dtsi. Changes are
>>>>>>> related to qspi, spi, nand, i2c and ethernet nodes.
>>>>>>>
>>>>>>> All errors are generated when dtbs are compiled with W=1.
>>>>>>>
>>>>>>
>>>>>> I don't see any errors on some other platforms, like Samsung. Isn't the
>>>>>> actual problem that you do not disable the nodes (I2C, SPI etc) in DTSI?
>>>>>
>>>>> On i2c node. Bus is present on the board but it can end in a connector or device
>>>>> which we don't have OS/bootloader drivers for. But we have drivers using i2c
>>>>> tools or u-boot i2c probe. It means that transition should happen.
>>>>> On i2c interesting is that W=1 is not able to report issues when you have i2c
>>>>> mux described like this
>>>>>
>>>>>                    i2c at 0 {
>>>>>                            #address-cells = <1>;
>>>>>                            #size-cells = <0>;
>>>>>                            reg = <0>;
>>>>>                            /* HPC0_IIC */
>>>>
>>>> I understand and it is quite common, but it does not explain the case.
>>>> Your bus should still be disabled in DTSI and enabled in DTS for these
>>>> cases.
>>>>
>>>> And how exactly do you solve the warning for above case?
>>>
>>> I had address/size-cells in dtsi (entire SoC) and then just enabled the whole
>>> i2c bus(for above reason) in board dts file without any childs there.
>>> Then W=1 reported that there are address/size-cells without child.
>>>
>>> Another way how to solve it would be to simply delete address/size cells in
>>> these cases.
>>
>> But your device without children sill have address/size cells, doesn't
>> it? Or does it mean you did not add it to such cases?
>>
>> Anyway, I have exactly the same case exynos5800-peach-pi.dts with i2c_2
>> and no W=1 warnings. I really do not think your solution is correct.
>>
>> Also, address/size cells are properties of the SoC, so rather DTSI.
> 
> Let me take a look at i2c case again. For gems it should be clear that I need to 
> remove it because when there is mdio node I am getting errors like this.
> 
> arch/arm64/boot/dts/xilinx/zynqmp.dtsi:627.27-641.5: Warning 
> (avoid_unnecessary_addr_size): /axi/ethernet at ff0e0000: unnecessary 
> #address-cells/#size-cells without "ranges" or child "reg" property

Yes, which also synchronizes your DTS with kernel:
See: eb2f7ff7de56 ("arm64: xilinx: Remove address/size-cells from gem
nodes")

BTW, Instead of working on U-Boot clone of DTS, I would suggest spending
effort on re-using in kernel DTS (see Sumit's work using rebased-DTS tree).


Best regards,
Krzysztof