Re: [PATCH 08/10] ARM: dts: exynos: use proper 'dma-channels/requests' properties

[Marek Szyprowski]

On 28.04.2022 12:00, Krzysztof Kozlowski wrote:
> On 28/04/2022 11:57, Marek Szyprowski wrote:
>> On 28.04.2022 11:54, Krzysztof Kozlowski wrote:
>>> On 28/04/2022 11:50, Marek Szyprowski wrote:
>>>> On 27.04.2022 17:58, Krzysztof Kozlowski wrote:
>>>>> pl330 DMA controller bindings documented 'dma-channels' and
>>>>> 'dma-requests' properties (without leading hash sign), so fix the DTS to
>>>>> match the bindings.
>>>>>
>>>>> Reported-by: Rob Herring <robh@xxxxxxxxxx>
>>>>> Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx>
>>>> Are those properties really needed for PL330 driver on Exynos SoCs? I've
>>>> removed them and I still see the proper values read from registers and
>>>> reported in the log (Exynos4210):
>>>>
>>>> dma-pl330 12680000.dma-controller: Loaded driver for PL330 DMAC-141330
>>>> dma-pl330 12680000.dma-controller:       DBUFF-32x4bytes Num_Chans-8
>>>> Num_Peri-32 Num_Events-32
>>>> dma-pl330 12690000.dma-controller: Loaded driver for PL330 DMAC-141330
>>>> dma-pl330 12690000.dma-controller:       DBUFF-32x4bytes Num_Chans-8
>>>> Num_Peri-32 Num_Events-32
>>>> dma-pl330 12850000.dma-controller: Loaded driver for PL330 DMAC-141330
>>>> dma-pl330 12850000.dma-controller:       DBUFF-64x8bytes Num_Chans-8
>>>> Num_Peri-1 Num_Events-32
>>>>
>>>> I also don't see any code that would read those properties. IMHO they
>>>> should be simply removed at all, at least for the PL330 related nodes.
>>> In current Linux implementation they indeed are not used. Nothing parses
>>> them. However:
>>> 1. They describe (hopefully correct) the hardware.
>>> 2. They might be used by other implementations of pl330 driver.
>>>
>>> I would not remove them from existing sources, but indeed maybe there is
>>> no need to add for new files.
>> What's the point in having dt properties duplicating data that might be
>> read from the driver registers?
> Hm, indeed, there is no point in this. Since they are read from
> registers, what was the idea behind in commit 42cf20980cde?

#dma-cells is indeed required, but the rest seems to be the cargo-cult 
of some kind.


Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland