[PATCH v2 1/3] dt-bindings: dma: snps,dw-axi-dmac: Add reset items

Walker Chen walker.chen at starfivetech.com
Fri Feb 24 07:23:03 PST 2023



On 2023/2/24 21:21, Krzysztof Kozlowski wrote:
> On 24/02/2023 14:09, Walker Chen wrote:
>> 
>> 
>> On 2023/2/24 18:51, Krzysztof Kozlowski wrote:
>>> On 24/02/2023 11:14, Walker Chen wrote:
>>>>>>    resets:
>>>>>> -    maxItems: 1
>>>>>> +    maxItems: 2
>>>>>
>>>>> This breaks ABI and all other users. Test your changes before sending.
>>>>
>>>> I think 'minItems' should be added here. So like this:
>>>> resets:
>>>>   minItems: 1
>>>>   maxItems: 2
>>>>
>>>> Other platform/users will not be affected by this.
>>>
>>> Which will allow two resets on all platforms. Is this correct for these
>>> platforms? Do they have two resets?
>>>
>> In kernel 6.2, only two platforms use this DMA controller (see 'arch/arc/boot/dts/hsdk.dts' and 'arch/riscv/boot/dts/canaan/k210.dtsi').
> 
> What about all out-of-tree platforms, bootloaders and FW?
> 
>> There is one reset on k210, while there is no reset of DMA on hsdk at all.
>> If here minItems with value 1 is added and the value of maxItems is changed to 2, after my testing,
>> whether it is one reset or two resets, even no reset, there is no errors occur when doing dtbs_check,
> 
> Yeah, I know how this works.
> 
>> the DMA initialization shall not be affected either on their platforms.
> 
> I asked whether the hardware physically have two resets. dtbs_check
> checks the DTS, not the hardware. You know allow two resets for each
> other variant. What's more, you call first reset axi for all variants.
> This does not look correct, e.g. k210.dtsi does not indicate this is axi
> reset line but reset for entire block.
> 
> Thus your change should be constrained per your variant (min/max in
> top-level, allOf:if:then allowing two lines for you, disallowing for
> others).
> 
I see what you mean. Thank you for your guidance!
I will use allOf:if:then structure to indicate different case for resets in next version of patch.

Best regards,
Walker 




More information about the linux-riscv mailing list