Hi Pavel,
On 9/1/19 12:23 PM, Pavel Machek wrote:
> On Fri 2019-08-30 18:08:20, Andy Shevchenko wrote:
>> sscanf() is a heavy one and moreover requires additional boundary checks.
>> Convert driver to use kstrtox() and replace kstrtoul() by kstrtobool()
>> in gpio_trig_inverted_store().
>>
>> While here, check the desired brightness against maximum defined for
>> a certain LED.
>
> One change per patch, please.
>
> Because this one will not end well.
>
>> @@ -60,10 +60,10 @@ static ssize_t gpio_trig_brightness_store(struct device *dev,
>> unsigned desired_brightness;
>> int ret;
>>
>> - ret = sscanf(buf, "%u", &desired_brightness);
>> - if (ret < 1 || desired_brightness > 255) {
>> + ret = kstrtouint(buf, 10, &desired_brightness);
>> + if (ret || desired_brightness > gpio_data->led->max_brightness) {
>> dev_err(dev, "invalid value\n");
>> - return -EINVAL;
>> + return ret ? ret : -EINVAL;
>> }
>
> We have people writing 255 into brightness, because that's what we
> used to do even for on/off LEDS. It is expected to work even for leds
> with max_brightness of 1.
>
> So... we want to saturate here, not return -EINVAL. (And we will
> eventually want to switch on/off leds to max_brightness = 1...)
Good point. We shouldn't fail here but proceed similarly as in case
of setting brightness for a LED in led_set_brightness_nosleep(), i.e.
here it should be:
desired_brightness = min(desired_brightness,
gpio_data->led->->max_brightness);
So the condition should be limited to checking error code.
>> @@ -86,16 +86,13 @@ static ssize_t gpio_trig_inverted_store(struct device *dev,
>> {
>> struct led_classdev *led = led_trigger_get_led(dev);
>> struct gpio_trig_data *gpio_data = led_trigger_get_drvdata(dev);
>> - unsigned long inverted;
>> + bool inverted;
>> int ret;
>>
>> - ret = kstrtoul(buf, 10, &inverted);
>> - if (ret < 0)
>> + ret = kstrtobool(buf, &inverted);
>> + if (ret)
>> return ret;
>>
>> - if (inverted > 1)
>> - return -EINVAL;
>> -
>> gpio_data->inverted = inverted;
>>
>> /* After inverting, we need to update the LED. */
>
> So, this accepted 0/1. Now it also accepts true false and many other pairs.
>
> Which... might be ok. But probably should be separated.
>
> Best regards,
> Pavel
>
>
--
Best regards,
Jacek Anaszewski
