On Thu, Mar 18, 2010 at 9:47 AM, Anton Vorontsov
[off-list ref] wrote:

Hi Grant,

On Thu, Mar 18, 2010 at 09:22:50AM -0600, Grant Likely wrote:

quoted

Both dev_archdata.of_node and of_device.node are duplications of the
device.of_node value. =A0This patch removes them.

Yeah, they're plain duplications since you introduced dev.of_node.
I wonder what was the problem with using dev.archdata.of_node?
Why dev.of_node is better?

CONFIG_OF support is not going to be an arch-specific thing any
longer.  The code is being generalized, and I'm removing as many
things as possible that arch code needs to add to enable CONFIG_OF.
That includes the dev_archdata element.

The impact of moving of_node from dev_archdata to device is pretty
small anyway.  Most current users are getting the device node from
of_device.node instead of archdata.  The number of dev_archdata users
is comparatively small.

Also, by using dev.of_node directly you have to introduce ugly
#ifdefs in the non-OF code (as in i2c patch), which you don't
need with transparent archdata and accessors, which you've just
removed:

The #ifdefs are only needed in the i2c code because the i2c API
doesn't currently support separate allocation and registration of i2c
devices.  With separate allocation and registration, the of_i2c code
could set the device node pointer directly without touching the common
i2c code at all (like how of_register_spi_devices handles it).  I do
plan to write a patch to do this, but that is a task for another patch
series.

g.