On 02/17/2015 11:52 AM, Sebastian Hesselbarth wrote:

I2C mux pinctrl driver currently determines the number of sub-busses by
counting available pinctrl-names. Unfortunately, this requires each
incarnation of the devicetree node with different available sub-busses
to be rewritten.

Can you be more explicit about the problem here? Why does anything need 
to be re-written if a child node is disabled; presumably there's no need 
for the child bus numbers to be contiguous. In other words, with the 
example in the existing DT binding doc:

         i2cmux {
                 compatible = "i2c-mux-pinctrl";
...
                 pinctrl-names = "ddc", "pta", "idle";
                 pinctrl-0 = <&state_i2cmux_ddc>;
                 pinctrl-1 = <&state_i2cmux_pta>;
                 pinctrl-2 = <&state_i2cmux_idle>;

                 i2c at 0 {
                         reg = <0>;
...
                 i2c at 1 {
                         reg = <1>;
...

That would generate child busses 0 and 1. If I was to disable the i2c at 0 
node, then there would still be definitions for child busses 0 and 1 in 
the DT, it's just that child bus 0 wouldn't actually exist at run-time. 
I don't see what part of DT needs to be re-written to accomodate this?

This patch reworks i2c-mux-pinctrl driver to count the number of
available sub-nodes instead. The rework should be compatible to the old
way of probing for sub-busses and additionally allows to disable unused
sub-busses with standard DT property status = "disabled".

This also amends the corresponding devicetree binding documentation to
reflect the new functionality to disable unused sub-nodes. While at it,
also fix two references to binding documentation files that miss an "i2c-"
prefix.

-For each named state defined in the pinctrl-names property, an I2C child bus
-will be created. I2C child bus numbers are assigned based on the index into
-the pinctrl-names property.
+For each child node that is not disabled by a status != "okay", an I2C
+child bus will be created. I2C child bus numbers are assigned based on the
+order of child nodes.

I would have assumed that disabled sub-nodes was a global concept within 
DT, and so wouldn't be mentioned in the binding. It would just be a bug 
in the driver if it didn't ignore disabled sub-nodes.

-The only exception is that no bus will be created for a state named "idle". If
-such a state is defined, it must be the last entry in pinctrl-names. For
-example:
-
-	pinctrl-names = "ddc", "pta", "idle"  ->  ddc = bus 0, pta = bus 1
-	pinctrl-names = "ddc", "idle", "pta"  ->  Invalid ("idle" not last)
-	pinctrl-names = "idle", "ddc", "pta"  ->  Invalid ("idle" not last)
+There must be a corresponding pinctrl-names entry for each enabled child
+node at the position of the child node's "reg" property.

The addition there seems fine, but the existing text re: the idle state 
seems clearer in the original text.

  Whenever an access is made to a device on a child bus, the relevant pinctrl
  state will be programmed into hardware.

-If an idle state is defined, whenever an access is not being made to a device
-on a child bus, the idle pinctrl state will be programmed into hardware.
+Also, there can be an idle pinctrl state defined at the end of possible
+pinctrl states. If an idle state is defined, whenever an access is not being
+made to a device on a child bus, the idle pinctrl state will be programmed
+into hardware.