On Thu, May 22, 2014 at 8:36 AM, Arnd Bergmann [off-list ref] wrote:

On Thursday 22 May 2014 15:30:39 Linus Walleij wrote:

quoted

quoted

quoted

+       parent = soc_device_to_device(soc_dev);

We really need to standardize this soc device stuff so that platforms
always have a parent soc device or not.

Arnds position is that the entire platform/board should not be
a "soc device" and that seems correct since the board is no SoC.

So I will spin this off into a SoC driver and push under
drivers/soc and bind to the SoC node in the device tree.

To be clear, this has nothing to do with whether this is strictly a SoC
or not. I would still want to see all the internal components of the
main chip under the soc node, and probably anything that is attached
to one of its interfaces under those nodes. The parts that should not
be under /soc are the ones that are not in the device hierarchy, such
as external clock generators.

Pretty much everything is an SOC these days. And there is really no way to tell what is the SOC boundary in DT. Perhaps it is just all devices handled by the top-level of_platform_populate call.

If this is the route we want to go, I don't think it should be left to the whim of the platforms. We should have some sort of default/generic SOC device until a more specific device is added or SOCs just add data to the generic device instead of making their own. With Pawel's vexpress clean-up that just got pulled into arm-soc for 3.16, we can now call of_platform_populate multiple times on the DT root. This means we can get rid of most of_platform_populate calls in platforms.