Currently, if someone tries to set the thermal sensor type to an
unsupported value, subsequent accesses to the chip may temporarily
show the sensor in question as disabled. Use a temporary variable
and only update the cached value on success, to prevent such
confusion.
Signed-off-by: Jean Delvare <khali@xxxxxxxxxxxx>
---
drivers/hwmon/it87.c | 20 ++++++++++----------
1 file changed, 10 insertions(+), 10 deletions(-)
--- linux-2.6.34-rc3.orig/drivers/hwmon/it87.c 2010-04-09 10:29:27.000000000 +0200
+++ linux-2.6.34-rc3/drivers/hwmon/it87.c 2010-04-09 10:34:29.000000000 +0200
@@ -539,15 +539,14 @@ static ssize_t set_sensor(struct device
struct it87_data *data = dev_get_drvdata(dev);
long val;
+ u8 reg;
if (strict_strtol(buf, 10, &val) < 0)
return -EINVAL;
- mutex_lock(&data->update_lock);
-
- data->sensor = it87_read_value(data, IT87_REG_TEMP_ENABLE);
- data->sensor &= ~(1 << nr);
- data->sensor &= ~(8 << nr);
+ reg = it87_read_value(data, IT87_REG_TEMP_ENABLE);
+ reg &= ~(1 << nr);
+ reg &= ~(8 << nr);
if (val == 2) { /* backwards compatibility */
dev_warn(dev, "Sensor type 2 is deprecated, please use 4 "
"instead\n");
@@ -555,13 +554,14 @@ static ssize_t set_sensor(struct device
}
/* 3 = thermal diode; 4 = thermistor; 0 = disabled */
if (val == 3)
- data->sensor |= 1 << nr;
+ reg |= 1 << nr;
else if (val == 4)
- data->sensor |= 8 << nr;
- else if (val != 0) {
- mutex_unlock(&data->update_lock);
+ reg |= 8 << nr;
+ else if (val != 0)
return -EINVAL;
- }
+
+ mutex_lock(&data->update_lock);
+ data->sensor = reg;
it87_write_value(data, IT87_REG_TEMP_ENABLE, data->sensor);
mutex_unlock(&data->update_lock);
return count;
--
Jean Delvare
_______________________________________________
lm-sensors mailing list
lm-sensors@xxxxxxxxxxxxxx
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
