On Thu, Jul 25, 2013 at 11:58:16PM +0100, Russell King - ARM Linux wrote:
> On Thu, Jul 25, 2013 at 08:16:04PM +0100, Mark Brown wrote:

> > This seems mostly fine, though it may be best to keep kirkwood-dma as a
> > separate module for the benefit of the S/PDIF support when it gets added
> > - I had a look at the implementation Russell has and it looks like it
> > can be added as a separate interface.

> You wouldn't want I2S and SPDIF to be separate modules though - they're
> the same hardware but different output stream formatters attached to the
> DMA FIFO output.  Don't forget the requirements concerning the simultaneous
> use of I2S and SPDIF - these "output formatters" must both be enabled and
> disabled in unison when concurrent use is required - both bits must be
> set or cleared together with a single register write.

OK, I knew they both needed to know about each other and to share some
stuff but I figured it was reasonable to compile out the S/PDIF support
if only using I2S or similar - so long as they're coupled at runtime it
should be fine.

> > Should the name be done as dev_name() for the interface (I don't know if
> > there is ever more than one)?

> Getting away from "kirkwood-i2s" would be sensible, because it may not be
> just "i2s" in this hardware block.  The documentation calls this an "audio
> controller" but I guess "kirkwood-pcm" would be a reasonable compromise,
> even though it has a separate AC'97 block which could also be construed
> as being "pcm".

I was thinking just pick a name at runtime based on what was doing the
instantiation which would avoid hard coding anything in the DMA driver.