Re: [PATCH 00/18] dt-bindings: irqchip: include generic schema

[Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>]

On 21/03/2022 19:53, Rob Herring wrote:
> On Thu, Mar 17, 2022 at 12:55:24PM +0100, Krzysztof Kozlowski wrote:
>> Hi,
>>
>> The DTS patches can be picked up independently.
>>
>> Best regards,
>> Krzysztof
>>
>> Krzysztof Kozlowski (18):
>>   ARM: dts: nspire: use lower case hex addresses in node unit addresses
>>   ARM: dts: ox820: align interrupt controller node name with dtschema
>>   ARM: dts: socfpga: align interrupt controller node name with dtschema
>>   dt-bindings: irqchip: actions,owl-sirq: include generic schema
>>   dt-bindings: irqchip: fsl: include generic schema
>>   dt-bindings: irqchip: ingenic: include generic schema
>>   dt-bindings: irqchip: intel,ixp4xx: include generic schema
>>   dt-bindings: irqchip: kontron,sl28cpld: include generic schema
>>   dt-bindings: irqchip: loongson: include generic schema
>>   dt-bindings: irqchip: microchip,eic: include generic schema
>>   dt-bindings: irqchip: mrvl,intc: include generic schema
>>   dt-bindings: irqchip: mstar,mst-intc: include generic schema
>>   dt-bindings: irqchip: mti,gic: include generic schema
>>   dt-bindings: irqchip: nuvoton,wpcm450-aic: include generic schema
>>   dt-bindings: irqchip: realtek,rtl-intc: include generic schema
>>   dt-bindings: irqchip: renesas: include generic schema
>>   dt-bindings: irqchip: sifive: include generic schema
>>   dt-bindings: irqchip: ti: include generic schema
> 
> I'm somewhat on the fence about these. Originally only devices with a 
> bus or child nodes included a common schema. For 'simple' providers 
> with mainly a '#<provider>-cells' property, we had to set the 
> constraints on the number of cells anyways, so referencing another 
> schema doesn't save anything. It is nice to declare the 'class' of the 
> device though.
> 
> It makes the schema be applied twice (if the node name matches). That's 
> not all bad because it finds cases of wrong node name. However, 
> sometimes we have devices which are multiple providers and can't set the 
> node name. So those can't reference interrupt-controller.yaml.

Indeed there were cases of MSI and interrupt controller.

> 
> It also means that 'interrupt-map' for example is now valid. That could 
> be fixed by splitting 'interrupt-map' related properties to its own 
> schema. 

Yes, that's a side effect. Maybe my series should be abandoned.

Best regards,
Krzysztof