Re: [PATCH 3/3] riscv: dts: spacemit: define a SPI controller node

[Alex Elder <elder@riscstar.com>]

On 9/18/25 11:22 AM, Alex Elder wrote:
> On 9/18/25 9:33 AM, Yao Zi wrote:
>>> .. em, so the SPI will use pdma, then probably you should also adjust 
>>> Kconfig to
>>> select PDMA driver?
>> The driver seems to depend on the generic DMA engine API only, IOW,
>> theoretically it should work with other DMA controller as well. And it's
>> even capable to operate without DMA (see k1_spi_dma_setup()).
>>
>> Dependency to PDMA really doesn't seem something should be enforced in
>> Kconfig: it doesn't exist in code level, and the driver is actually more
>> flexible.
> 
> You're right on both points.  The code doesn't *require* PDMA to
> operate correctly (to my knowledge).
> 
> Yixun, what do you think?
> 
>                      -Alex

I did some experiments.  Currently, if I build a kernel with
SPI_SPACEMIT_K1 enabled (module or built-in) but MMP_PDMA
not set, the K1 SPI driver probe doesn't complete.  The reason
is that dma_request_chan() returns -EPROBE_DEFER, not "knowing"
that the needed driver will never show up.

For now I have added a call to IS_ENABLED(CONFIG_MMP_PDMA) in
devm_k1_spi_dma_setup(), and if it's not enabled it will
return 0 (to indicate "all is well, but we won't use DMA").

That doesn't allow for a different DMA option, but it does
allow the driver to work without an explicit dependency
on the MMP_PDMA (via Kconfig).

					-Alex