[PATCH v3 1/4] ASoC: dt-bindings: lpc32xx: Add lpc32xx i2s DT binding

[Piotr Wojtaszczyk]

On Mon, Jun 17, 2024 at 5:48 PM Krzysztof Kozlowski <krzk at kernel.org> wrote:
>
> On 17/06/2024 16:04, Piotr Wojtaszczyk wrote:
> >>
> >>> It's used by snd_soc_dai_init_dma_data() in [PATCH v3 4/4] to give the
> >>> dmaengine a
> >>> hint which dma config to use. The LPC32xx doesn't have yet a dmamux driver like
> >>
> >> and if I change driver platform data to foo and bar, does the DTS work? No.
> >
> > They shouldn't change the same way as expected dma-names shouldn't change.
> > Lots of drivers expect the dma-names to be "rx", "tx"
> >
> >>
> >>> lpc18xx-dmamux.c therefore it still uses platform data entries for
> >>> pl08x dma channels
> >>> and 'SND_DMAENGINE_PCM_FLAG_NO_DT | SND_DMAENGINE_PCM_FLAG_COMPAT'
> >>> flags in the devm_snd_dmaengine_pcm_register().
> >>> Typically instead of this platform data you would use regular 'dma'
> >>> and 'dma-names' if it had
> >>> proper dmamux driver like lpc18xx-dmamux.c
> >>
> >> Exactly. Use these.
> >
> > Then I need to write a lpc32xx dma mux driver, device tree binding for
> > it and adjust the
> > LPC32xx I2S driver for it. Is this a hard requirement to accept this
> > patch set for the
> > legacy LPC32xx SoC?
>
> I do not see at all analogy with dma-names. dma-names are used ONLY by
> the consumer to pick up proper property "dmas" from DT. They are not
> passed to DMA code. They are not used to configure DMA provider at all.
>
> You parse string from DT and pass it further as DMA filtering code. This
> is abuse of hardware description for programming your driver and their
> dependencies.
>
> Why you cannot hard-code them?
>
> Sorry, to be clear: NAK

That's fine, clear answers are always good.
I considered to hardcode this as it was in the first version of the patch set
but LPC32XX has two I2S interfaces which use different DMA signals
and mux settings and I really didn't want to pick the virtual DMA channel
name based on hardcoded I2S node name therefore I thought having a DT
property to select proper dma channel is a better solution.

-- 
Piotr Wojtaszczyk
Timesys