Re: [PATCH net-next v2 1/4] dt-bindings: net: snps,dwmac: per channel irq

[Krzysztof Kozlowski]

On 07/01/2024 21:10, Serge Semin wrote:
> On Fri, Jan 05, 2024 at 03:09:22PM +0800, Leong Ching Swee wrote:
>> From: Swee Leong Ching <leong.ching.swee@xxxxxxxxx>
>>
>> Add dt-bindings for per channel irq.
>>
>> Signed-off-by: Rohan G Thomas <rohan.g.thomas@xxxxxxxxx>
>> Signed-off-by: Swee Leong Ching <leong.ching.swee@xxxxxxxxx>
>> ---
>>  .../devicetree/bindings/net/snps,dwmac.yaml   | 24 +++++++++++++------
>>  1 file changed, 17 insertions(+), 7 deletions(-)
>>
>> diff --git a/Documentation/devicetree/bindings/net/snps,dwmac.yaml b/Documentation/devicetree/bindings/net/snps,dwmac.yaml
>> index 5c2769dc689a..e72dded824f4 100644
>> --- a/Documentation/devicetree/bindings/net/snps,dwmac.yaml
>> +++ b/Documentation/devicetree/bindings/net/snps,dwmac.yaml
>> @@ -103,17 +103,27 @@ properties:
>>  
>>    interrupts:
>>      minItems: 1
>> -    items:
>> -      - description: Combined signal for various interrupt events
>> -      - description: The interrupt to manage the remote wake-up packet detection
>> -      - description: The interrupt that occurs when Rx exits the LPI state
>> +    maxItems: 19
>>  
>>    interrupt-names:
>>      minItems: 1
>> +    maxItems: 19
>>      items:
>> -      - const: macirq
>> -      - enum: [eth_wake_irq, eth_lpi]
>> -      - const: eth_lpi
>> +      oneOf:
>> +        - description: Combined signal for various interrupt events
>> +          const: macirq
>> +        - description: The interrupt to manage the remote wake-up packet detection
>> +          const: eth_wake_irq
>> +        - description: The interrupt that occurs when Rx exits the LPI state
>> +          const: eth_lpi
>> +        - description: DMA Tx per-channel interrupt
>> +          pattern: '^dma_tx[0-7]?$'
>> +        - description: DMA Rx per-channel interrupt
>> +          pattern: '^dma_rx[0-7]?$'
>> +
>> +    allOf:
>> +      - contains:
>> +          const: macirq
> 
> In order to restore the v1 discussion around this change, here is my
> comment copied from there:
> 
>> As Rob correctly noted it's also better to make sure that 'macirq' is placed first
>> in the array. So instead of the constraint above I guess the next one would
>> make sure both the array has 'macirq' name and it's the first item:
>>
>> allOf:
>>   - maxItems: 34
>>     items:
>>       - const: macirq
> 
> Leong said it didn't work:
> https://lore.kernel.org/netdev/CH0PR11MB54904615B45E521DE6B1A7B3CF61A@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> 
> Rob, Krzysztof, Conor could you please clarify whether this change is ok the
> way it is or it would be better to preserve the stricter constraint
> and fix the DT-schema validation tool somehow?

First of all this change is not good, because commit msg explains
absolutely nothing why this is done and what exactly you want to achieve
here. The "what" part often is obvious from the code, but not in this
case. Are the per-channel IRQs conflicting with macirq or others? Are
they complementary (maxItems: 19 suggests that, though, but could be
mistake as well)? Do they affect all snps,dwmac derivatives or only some?

So many questions and zero answers in one liner commit msg!

Now about the problem, I think we should preserve the order, assuming
that these are complementary so first three must be defined. This
however could be done in the device schema referencing snps,dwmac. I
think I will repeat myself: I dislike this schema, because it mixes two
purposes: defining shared part and defining final device part. The code
in this patch is fine for a schema defining the shared part.

Therefore before we start growing this monstrosity into bigger one, I
think we should go back to the plans of reworking and cleaning it.

Best regards,
Krzysztof