On Thu, 2013-07-11 at 15:56 +0100, Grant Likely wrote:

quoted

+contiguous-memory {
+
+       (name): region@(base-address) {
+               reg = <(baseaddr) (size)>;
+               (linux,default-contiguous-region);
+               device = <&device_0 &device_1 ...>
+       };
+};

Okay, I've gone and read all of the backlog on the 3 versions of the
patch series, and I think I understand the issues. I actually think it
was better off to have the regions specified as children of the memory
node. I understand the argument about how would firmware know what
size the kernel wants and that it would be better to have a kernel
parameter to override the default. However, it is also reasonable for
the kernel to be provided with a default amount of CMA based on the
usage profile of the device. In that regard it is absolutely
appropriate to put the CMA region data into the memory node. I don't
think /chosen is the right place for that.

Picking up on that old thread after the rant I just posted ... Grant,
your proposal is all wrong.

First we already had a proposal for reserved memory, which due to the
complete lack of comment, we actually merged support for in powerpc in
3.11, second, do NOT make that a child of "memory". See the email I just
posted today for more details about the breakage in that proposal.

I'm advocating a revert at this stage.

Ben.