On Mon, May 14, 2012 at 02:38:20PM +0300, Peter Ujfalusi wrote:
The child (or drivers for the functionality) only needs small update for compatible_of, and new chip access wrapper.
At the very least anything that's using interrupts through the device core will also need to know how those have been wired out into the chip top level, and of course if anything happened to move about in the register map or there are any changes in available functionality (which would be unsurprising, perhaps an additional line output or something for example) those will also need to be dealt with.
Besides, if that's all the function drivers need to look at they can just look at their parent node instead of looking at the child node, problem sorted.
You might be right that the resulting dts section for the chip is just represents the current MD stack, but if I want to prepare for the future this is the way I think is the best for us.
Again, if I could see how this supported non-trivial reuse I'd be right with you but I'm just not seeing how this would allow reuse on another chip without rework of all existing device trees or what it hides about the chip top level. The code changes to drop the child nodes from the device tree should be trivial. This in itself seems to me like a clear sign that the abstraction in the device tree isn't actually providing useful generalisation of the chip.
I'm just not seeing anything in the device tree that abstracts the chip top level from the function drivers; if there were then it'd be very much easier to see a general use for this.