On Mon, Nov 09, 2015 at 11:41:20AM -0600, Andrew F. Davis wrote:

On 11/06/2015 03:16 PM, Mark Brown wrote:

quoted

There are cases where it's useful where we're abstracting something and
gaining some meaningful reuse.  This really does not appear to be one of
those cases, there are no parameters in the DT and the compatible string
is the full device name.

As before I see no reason to make that call now and limit ourselves.

To repeat *yet* *again* the point is that putting the current Linux
driver model into the DT is limiting our future selves.

quoted

You do not need to populate it.  There is no value in populating it and
as previously discussed putting the Linux driver model into DT can be
actively harmful if we change our idea of how we should model things.

The dev passed to regulator_register needs to have of_node populated for
your OF init_data helper to work. Devices with OF tables can just pass
their own dev. Others have to use their parents' nodes, this is a
workaround, OF devices should be probed with their of_node pre-populated.

This is not a workaroud, the only reason you think it is a workaround is
the desire to directly represent the Linux device model in the DT.

quoted

quoted

quoted

Please stop this.  I don't understand why you are pushing so hard to put
the Linux device model representation of the device into DT but it's
getting very repetitive.

quoted

quoted

I'm not pushing anything, this is how other sub-nodes of MFD devices are

quoted

Every time we go through this we finish the discussion and then you come
back with yet another excuse for trying to push the current Linux device
model into the DT or another version of the patch with the same problem.

I keep finding different problems, do you expect me to ignore them?

You are making minor restatements of the same thing over and over again
which ignore the main feedback.

quoted

The fact that other people have merged imperfect code into the kernel is
not a good reason to merge even more of it when we have better tools.
Looking at that binding I'm seeing no reason why any of the subfunctions
should have compatible strings (and if we're going down the route you're
trying to go down we really ought to have something in the binding for
at least an interrupt controller in there as well...).

These are not "subfunctions" they are full drivers, they only need
register accessors passed in, they do not call the core and the core
does not call them.

To repeat *yet* *again* they are groupings of functionality which happen
to represent the way Linux models devices right now.  There's no
generality in there, it's just a dump of the current Linux model of the
functions into the DT.

If your problem is with the DT binding for this or other MFDs, then
nack *them* and explain to everyone why what they are doing is wrong
and why regulators should be special cases. Blocking the regulator
drivers to force a change in DT is not going to fix this issue.

Of course this is a negative review of the binding!  What on earth did
you think my feedback meant?  The driver and the binding go together.