hwmon: (pmbus) Move boolean error condition check to generating code

0-day rightfully complains about a sometimes uninitialized variable
in pmbus_get_boolean().

drivers/hwmon/pmbus/pmbus_core.c:903:13: warning:
		variable 'ret' is used uninitialized whenever 'if' condition is true
	} else if (!s1 || !s2) {

While that is technically true, it won't be hit in the field since the
condition indicates a programming error. Move the check of that condition
into the code generating the attribute entry, and refuse generating the
attribute if the condition is true. Swap the condition check in
pmbus_get_boolean() to ensure that static analyzers don't get a hiccup
(because we check if s1 and s2 are NULL, static analyzers may believe
that they can be NULL independently of each other).
Reported-by: Nkernel test robot <lkp@intel.com>
Cc: Alex Qiu <xqiu@google.com>
Reviewed-by: NAlex Qiu <xqiu@google.com>
Signed-off-by: NGuenter Roeck <linux@roeck-us.net>

drivers/hwmon/pmbus/pmbus_core.c

@@ -898,11 +898,7 @@ static int pmbus_get_boolean(struct i2c_client *client, struct pmbus_boolean *b,
 		pmbus_update_sensor_data(client, s2);
 
 	regval = status & mask;
-	if (!s1 && !s2) {
-		ret = !!regval;
-	} else if (!s1 || !s2) {
-		WARN(1, "Bad boolean descriptor %p: s1=%p, s2=%p\n", b, s1, s2);
-	} else {
+	if (s1 && s2) {
 		s64 v1, v2;
 
 		if (s1->data < 0) {
@@ -917,6 +913,8 @@ static int pmbus_get_boolean(struct i2c_client *client, struct pmbus_boolean *b,
 		v1 = pmbus_reg2data(data, s1);
 		v2 = pmbus_reg2data(data, s2);
 		ret = !!(regval && v1 >= v2);
+	} else {
+		ret = !!regval;
 	}
 unlock:
 	mutex_unlock(&data->update_lock);
@@ -1044,6 +1042,9 @@ static int pmbus_add_boolean(struct pmbus_data *data,
 	struct pmbus_boolean *boolean;
 	struct sensor_device_attribute *a;
 
+	if (WARN((s1 && !s2) || (!s1 && s2), "Bad s1/s2 parameters\n"))
+		return -EINVAL;
+
 	boolean = devm_kzalloc(data->dev, sizeof(*boolean), GFP_KERNEL);
 	if (!boolean)
 		return -ENOMEM;