Re: DT DMA channel binding for Tegra I2S

On 11/23/2011 05:26 PM, Stephen Warren wrote:

Rob Herring wrote at Wednesday, November 23, 2011 2:55 PM:

quoted

On 11/23/2011 03:25 PM, Stephen Warren wrote:

quoted

Rob, Grant,

Re: http://www.spinics.net/lists/arm-kernel/msg148899.html

Background: Tegra's DMA controller needs to be told which DMA "request
select" to use for each transfer. The identifies which peripheral to
transfer to/from.

The Tegra I2S driver needs to know its own "request select" value to
pass to the audio driver DMA path. The board files typically provide
this to the driver as an IORESOURCE_DMA. I don't think there's any
standard binding that creates such a resource when instantiating a
platform device from device tree, and the discussions I found when
looking for one didn't seem to reach conclusion that there should be
one. So, I proposed the following property in the I2S driver's DT
binding for this:

I think this is a case of every platform being different and
IORESOURCE_DMA is not too widely used. Even on PPC, there's not a
standard way.

quoted

Doc:

- dma-channel : The Tegra DMA controller's channel ID for this I2S controller

Example:

i2s@70002800 {
	compatible = "nvidia,tegra20-i2s";
	reg = <0x70002800 0x200>;
	interrupts = < 45 >;
	dma-channel = < 2 >;
};

Does that look reasonable? Or, should I pursue some more standardized
solution?

I had some discussion with Thomas Abraham about this for the pl330 dma:

http://lists.ozlabs.org/pipermail/devicetree-discuss/2011-September/008269.html

So I would just add a phandle to the dma controller here.

So you mean just modify the binding as follows:

apbdma: dma@NNNNN { ... };

i2s@70002800 {
	dma-channel = < &apbdma, 2 >;
};

... and leave the code I quoted as-is?

Right.

Although, I think you are possibly missing some other properties for
i2s mode like word size and master/slave mode. I think the ideal case
is that a single ASoC platform driver for DT could handle multiple
SoCs.

I /think/ the rest of the patch you linked isn't really relevant to Tegra
at present, since it isn't ported to the DMA engine (yet?).

Yes, of course..

Rob