On Mon, Jan 04, 2021 at 09:24:49PM +0000, Mark Brown wrote:

On Mon, Jan 04, 2021 at 07:38:21PM +0100, Krzysztof Kozlowski wrote:

quoted

On Mon, Jan 04, 2021 at 06:27:34PM +0000, Mark Brown wrote:

quoted

quoted

We can indicate the presence of features without adding new compatible
strings, that's just encoding the way Linux currently divides things up
into the bindings.  For example having an extcon property seems like it
should be enough to figure out if we're using extcon.

quoted

It won't be enough because MFD will create device for extcon and bind
the driver. The same for the charger. We have a board where max8997 is
used only as PMIC (providing regulators) without battery and USB
connectivity.

I'm not sure I follow, sorry?  Either the core driver can parse the
bindings enough to know what children it has or (probably better) it can
instantiate the children unconditionally and then the function drivers
can figure out if they need to do anything.

Currently the MFD parent/core driver will instantiate children
unconditionally.  It would have to be adapted. With proposed bindings -
nothing to change.  MFD core already does the thing.

The point is that function drivers should not be even bound, should not
start to probe. Otherwise if they probe and fail, they will pollute the
dmesg/probe log with failure. With the failure coming from looking for
missing of_node or any other condition from parent/core driver.

quoted

Another point, is that this reflects the real hardware. The same as we
model entire SoC as multiple children of soc node (with their own
properties), here we represent smaller chip which also has
sub-components.

Components we're calling things like "extcon"...

I am rather thinking about charger, but yes, extcon as well. Either you
have USB socket (and want to use associated logic) or not.

Best regards,
Krzysztof