[alsa-devel] [PATCH 20/20] ASoC: fsl: add imx-sgtl5000 machine driver

[Stephen Warren]

On 03/05/2012 10:03 AM, Mark Brown wrote:
> * PGP Signed by an unknown key
> 
> On Mon, Mar 05, 2012 at 08:55:22AM -0800, Stephen Warren wrote:
>> Mark Brown wrote at Monday, March 05, 2012 6:14 AM:
> 
>>> The DMA and SSI drivers should have some relationship with each other if
>>> they're the same piece of hardware.  The machine driver certainly
>>> shouldn't be creating either of them.
> 
>> Note that in Tegra's case, the DMA HW and the DAI (I2S/SPDIF) HW are
>> entirely separate HW blocks and have entirely separate drivers. The DMA
>> engine is generic and not audio specific, so has a node that instantiates
>> just a standalone DMA device which has its own API. This doesn't
>> instantiate any kind of ALSA PCM driver. It doesn't make sense to have a
> 
> In this case (and probably in the i.MX case, at a guess the structure is
> the same) I'd just have the DAI drivers kick off registration of the
> ASoC DMA stuff, it's all part of the ASoC driver for the hardware really.

How would that work when there are multiple DAIs, e.g. on a system with
2 I2S and 1 SPDIF DAI and they all want to register the PCM driver? I
guess I could just add some utility function in the PCM driver file to
ensure that the platform device only gets created once, and probably
would also need to refcount it for when the DAIs get unloaded etc.

>> DT node to instantiate the ALSA PCM driver since the PCM driver is more
>> of an API bridge to the real DMA HW, and certainly not an actual HW
> 
> Yes, and a Linux driver model device is questionable too.
> 
>> Perhaps our regular DMA module driver should expose both its custom
>> interface for other clients, and register an ASoC PCM driver? That way
>> we could remove the PCM device registration from the ASoC machine
>> drivers.
> 
> That's another option, do it from the DMA driver instead of the DAI
> driver.  My main concern here is that it shouldn't be the machine driver
> as it's all internal to the CPU and not machine specific.

OK, not sure which way I prefer yet.

I'll file a bug internally for this so I don't forget about it...