On 20/03/14 20:49, Laurent Pinchart wrote:

quoted

The CPU is the _controlling_ component - it's the component that has to
configure the peripherals so they all talk to each other in the right
way.  Therefore, the view of it needs to be CPU centric.

If we were providing a DT description for consumption by some other
device in the system, then the view should be as seen from that device
instead.

Think about this.  Would you describe a system starting at, say, the
system keyboard, and branching all the way through just becuase that's
how you interact with it, or would you describe it from the CPUs point
of view because that's what has to be in control of the system.

DT has been designed to represent a control-based view of the system. It does 
so pretty well using its tree-based model. However, it doesn't have a native 
way to represent a flow-based graph, hence the OF graph solution we're 
discussing. The whole point of this proposal is to represent the topology of 
the media device, not how each entity is controlled.

I agree with Laurent here. I think this is an important point to keep in
mind. We already describe the control graph in the DT via the
parent-child relationships. There's no point in describing the same
thing again with the graph links being discussed.

 Tomi