Re: [PATCH v2] dt-bindings: spi: convert Freescale DSPI to dt-schema

[Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx>]

On 15/11/2022 15:19, Vladimir Oltean wrote:
> On Tue, Nov 15, 2022 at 03:08:37PM +0100, Krzysztof Kozlowski wrote:
>> On 15/11/2022 14:59, Vladimir Oltean wrote:
>>> On Tue, Nov 15, 2022 at 02:46:21PM +0100, Krzysztof Kozlowski wrote:
>>>>> +$id: http://devicetree.org/schemas/spi/fsl,spi-fsl-dspi.yaml
>>>>
>>>> Why second "fsl" in file name? It does not patch compatibles and
>>>> duplicates the vendor. We do not have compatibles "nxp,imx6-nxp".
>>>
>>> Ok, which file name would be good then? There are 9 different (all SoC
>>> specific) compatible strings, surely the convention of naming the file
>>> after a compatible string has some limitations...
>>
>> If all DSPI blocks fit here, then maybe: fsl,dspi.yaml
>>
>> fsl,spi-dspi.yaml is also a bit redundant.
> 
> Ok, fsl,dspi.yaml and fsl,dspi-peripheral-props.yaml, and MAINTAINERS
> entry for fsl,dspi*.yaml?

Yes.

> 
>>>>> +properties:
>>>>> +  compatible:
>>>>> +    description:
>>>>> +      Some integrations can have a single compatible string containing their
>>>>> +      SoC name (LS1012A, LS1021A, ...). Others require their SoC compatible
>>>>> +      string, plus a fallback compatible string (either on LS1021A or on
>>>>> +      LS2085A).
>>>>
>>>> Why? The fsl,ls1012a-dspi device is either compatible with
>>>> fsl,ls1021a-v1.0-dspi or not. It cannot be both - compatible and not
>>>> compatible.
>>>
>>> LS1012A is compatible with LS1021A to the extent that it works when
>>> treated like a LS1021A. LS1012A has a FIFO size of 8 SPI words, LS1021A
>>> of just 4. Treating it like LS1021A means roughly half the performance,
>>> but it still works.
>>>
>>> I didn't invent any of this. When I took over the driver, there were
>>> device trees like this all over the place:
>>>
>>> 		dspi: spi@2100000 {
>>> 			compatible = "fsl,ls1012a-dspi", "fsl,ls1021a-v1.0-dspi";
>>
>> Which looks ok...
>>
>>> 			#address-cells = <1>;
>>> 			#size-cells = <0>;
>>> 			reg = <0x0 0x2100000 0x0 0x10000>;
>>> 			interrupts = <0 64 IRQ_TYPE_LEVEL_HIGH>;
>>> 			clock-names = "dspi";
>>> 			clocks = <&clockgen QORIQ_CLK_PLATFORM_PLL
>>> 					    QORIQ_CLK_PLL_DIV(1)>;
>>> 			spi-num-chipselects = <5>;
>>> 			big-endian;
>>> 			status = "disabled";
>>> 		};
>>>
>>> but the Linux driver pre-~5.7 always relied on the fallback compatible
>>> string (LS1021A in this case). I'm working with what's out in the field,
>>> haven't changed a thing there.
>>
>> The driver matters less (except ABI), but anyway it confirms the case -
>> fallback is expected always.  Why the fallback should be removed if the
>> devices are compatible (including halved performance)?
> 
> I don't think I said the fallback should be removed? I think you're
> talking about a typo/braino I made, which puts the LS1012A both in the
> bucket of SoCs with a single compatible strings required, as well as in
> that with fallback required. Obviously both can't be true... I didn't
> mean LS1012A but VF610.

To be clear: ls1012a, ls1028a and lx2160a should be either followed by
compatible or not. Cannot be both.

Best regards,
Krzysztof