[RFC] should we remove soc subnode in all berlin SoC's dts if SOC_BUS is selected

[jszhang@marvell.com (Jisheng Zhang)]

On Fri, 26 Sep 2014 05:32:29 -0700
Arnd Bergmann <arnd@arndb.de> wrote:

> On Friday 26 September 2014 19:39:45 Jisheng Zhang wrote:
> > We want to export machine name, soc family, chip revision and chipid to
> > userspace via. the SOC BUS interface, i.e call soc_device_register()
> > 
> > It works perfectly, but there's one side effect: every device nodes will
> > have an additional level. For example:
> > 
> > /sys/devices/soc.0/xxx.sdhci will be changed to
> > 
> > /sys/devices/soc0/soc.0/xxx.sdhci
> > 
> > This seems a bit ugly. I have one solution: remove "soc" subnodes in all
> > berlin's dts, then we will have 
> > 
> > /sys/devices/soc0/xxx.sdhci
> > 
> > Is there any better solution? 
> > 
> 
> As of a recent discussion, we decided that the soc_device_register is now
> purely informational and not meant to be used as the root for all the
> other devices.
> 
> The structure in the DTS is fine and will continue to do the right thing.
> 
> Originally the intention was that soc_device_register would be used to
> register the soc device  that corresponds to your /soc device node in DT,
> and then you'd populate the other devices below, but we never implemented
> a nice way to call that, so you don't need to bother with it.
> 
> Instead, you will now have /sys/devices/soc0/ for the soc_device you
> register manually, and /sys/devices/soc.0/ to correspond with the DT node
> and continues to have the child devices. This is still a bit confusing, but
> should be consistent across all (new) platforms.
> 

Got it. Thanks for the detailed explanations. I will follow this solution both
in mainline and mrvl internal berlin kernel tree.

Thanks very much,
Jisheng