On Thursday 21 June 2012, Guennadi Liakhovetski wrote:

Indeed, this solution should be good enough, thanks! I'm not sure, whether 
making this multiplexing available requires any additional code to the 
generic DMA DT binding implementation. If it does - please, let's make 
this a part of the implementation.

It depends how the dma engine gets represented in Linux then: Either
we make the common dmaengine code handle multiple dma-engines with the
same of_node pointer, or you make the driver register a single dma-engine
to Linux that is backed by the three physical devices but just one
of_node pointer. I think either way will work, and someone has to
try it to determine which one is simpler.

It is also important to provide a flexible multiple channel per device 
configuration support to let slave drivers distinguish between different 
DMA channels, that they get back from the API. But I think this is a part 
of the current proposal and is being taken care of.

Yes, it is.

	Arnd