On 02/18/2013 04:30 AM, Wei Ni wrote:

Add functions to support using dt node with args to get sensor.

You need to write a device tree binding document to explain this.

+struct thermal_sensor *get_sensor_by_node(struct node_args *np_args)
+{
+	struct thermal_sensor *pos;

"pos" isn't a great variable name. Why not use "sensor", or just the
"ts" variable you have right below?

+	struct thermal_sensor *ts = NULL;
+	struct node_args *args;
+
+	mutex_lock(&sensor_list_lock);
+	for_each_thermal_sensor(pos) {
+		args = &pos->np_args;
+		if (args->np) {
+			if ((args->np == np_args->np) &&
+			   (args->index == np_args->index)) {
+				ts = pos;
+				break;

Replace those 2 lines with "goto out;".

+			}
+		}
+	}

here, add:

	ts = NULL;
out:

That way, you can use "ts" as the loop iteration variable.

This whole patch rather assumes that all DT nodes can identify their
exposed thermal sensors using an index in a single DT cell. That's not
very flexible. All other DT bindings work like this:

Provider of a service indicates how many DT cells are in the object
(GPIO, IRQ, thermal sensors) specifier:

sensor1: lm90 at 1c {
	...
	#thermal-sensor-cells = <1>;
};

Each consumer of a service imports it by referencing it:

thermal-zone {
	...
	sensors = <&sensor1 0>;
};

The driver for LM90 provides an "of_xlate" function which receives a
struct of_phandle_args and outputs/returns whatever Linux-internal
identification/representation of the object is required. For example, see:

include/linux/pwm.h:161:	struct pwm_device *	(*of_xlate)(struct pwm_chip *pc,

This allows each providing object's DT binding to define its own value
of #thermal-sensor-cells, as suited for its own requirements, and allows
each driver to implement the mapping from DT to internal ID in whatever
way is necessary.

Now, many bindings/drivers might just end up using a common simple
implementation. That's why functions such as of_pwm_simple_xlate() or
of_gpio_simple_xlate() exist.