On Wed, 4 Apr 2012 12:17:54 -0700, Olof Johansson [off-list ref] wrote:

On Tue, Apr 03, 2012 at 09:32:57AM +0200, Nicolas Ferre wrote:

quoted

From: Ludovic Desroches <redacted>

Having a unit-address different from 0 causes bad memory node detection
during early_init_dt_scan_memory.

Sounds like a bug in the generic code to me? But specifying a unit
address on a non-ambiguous device node isn't necessary anyway, so this
patch is still a reasonable thing to do.

It's actually a 'feature' of the way we're including skeleton.dtsi.
skeleton.dtsi has already set the node name, so if the next file
doesn't use exactly the same node name then the .dtb ends up with two
memory nodes, which is really bad.

It real problem is the dtc tooling.

g.