Re: [PATCH v8 1/1] leds: LED driver for TI LP3952 6-Channel Color LED

[Jacek Anaszewski <j.anaszewski@samsung.com>]

Hi Tony,

On 07/06/2016 05:05 PM, Tony wrote:
> Thank you for the comments Jacek and Mika.
>
> On 06/07/16 14:33, Jacek Anaszewski wrote:
>
>
>>> +static int lp3952_set_brightness(struct led_classdev *cdev,
>>> +                 enum led_brightness value)
>>> +{
>>> +    unsigned int reg, shift_val;
>>> +    struct lp3952_ctrl_hdl *led = container_of(cdev,
>>> +                           struct lp3952_ctrl_hdl,
>>> +                           cdev);
>>> +    struct lp3952_led_array *priv = (struct lp3952_led_array
>>> *)led->priv;
>>> +
>>> +    dev_dbg(cdev->dev, "Brightness request: %d on %d\n", value,
>>> +        led->channel);
>>> +
>>> +    if (value == LED_OFF) {
>>> +        lp3952_on_off(priv, led->channel, LED_OFF);
>>> +        return 0;
>>> +    }
>>> +
>>> +    if (led->channel > LP3952_RED_1) {
>>> +        dev_err(cdev->dev, " %s Invalid LED requested", __func__);
>>> +        return -EINVAL;
>>> +    }
>>> +
>>> +    if (led->channel >= LP3952_BLUE_1) {
>>> +        reg = LP3952_REG_RGB1_MAX_I_CTRL;
>>> +        shift_val = (led->channel - LP3952_BLUE_1) * 2;
>>> +    } else {
>>> +        reg = LP3952_REG_RGB2_MAX_I_CTRL;
>>> +        shift_val = led->channel * 2;
>>> +    }
>>> +
>>> +    /* Enable the LED in case it is not enabled already */
>>> +    lp3952_on_off(priv, led->channel, 1);
>>> +
>>> +    return regmap_update_bits(priv->regmap, reg, 3 << shift_val,
>>> +                  --value << shift_val);
>>
>> Here you have two separate calls that change the device state.
>> I think that mutex protection is required here to assure that
>> the device will not be left in an inconsistent state, due to
>> the concurrent calls from other processes.
>
> Sorry, I did not quite understand.
>
> Did you mean 'regmap_update_bits' here and the one within
> 'lp3952_on_off'? If there isn't a lock assigned by driver, a mutex or
> spinlock is assigned based on fast_io support by regmap during init.

The lock protects an access to the I2C bus for a single transmission.
When there are two registers to set then current process can be
preempted after first write by the second process which overwrites
the register with a new value. The first one after being woken up
performs second register write, being unaware that the instruction it
completed in the previous step has been just overwritten.

It applies to the situations where there is a common register
setting affecting all LEDs, e.g. power down mode or so, but after
consulting the documentation I figured out that this is not the
case.

>>
>>> +}
>>> +
>>> +static int lp3952_get_label(char *dest, const char *label, struct
>>> device *dev)
>>> +{
>>> +    int ret;
>>> +    const union acpi_object *obj;
>>> +    struct acpi_device *adev = ACPI_COMPANION(dev);
>>> +
>>> +    ret = acpi_dev_get_property(adev, label, ACPI_TYPE_STRING, &obj);
>>> +    if (!ret)
>>> +        strncpy(dest, obj->string.pointer, obj->string.length + 1);
>>> +
>>> +    return ret;
>>> +}
>>> +
>>> +static int lp3952_register_led_classdev(struct lp3952_led_array *priv)
>>> +{
>>> +    int i, ret = -ENODEV;
>>> +    static const char *led_name_hdl[LP3952_LED_ALL] = {
>>> +        "blue2",
>>> +        "green2",
>>> +        "red2",
>>> +        "blue1",
>>> +        "green1",
>>> +        "red1"
>>> +    };

Why actually do you need this array? Couldn't you provide the LED
identifier in the DSD entry corresponding to a LED? You can follow
the scheme used in Device Tree by introducing "reg" property.

>>> +    for (i = 0; i < LP3952_LED_ALL; i++) {
>>> +        priv->leds[i].cdev.brightness = LED_OFF;
>>> +        priv->leds[i].cdev.max_brightness = LP3952_BRIGHT_MAX;
>>> +        priv->leds[i].cdev.brightness_set_blocking =
>>> +                lp3952_set_brightness;
>>> +        priv->leds[i].channel = i;
>>> +        priv->leds[i].priv = priv;
>>> +
>>> +        ret = lp3952_get_label(priv->leds[i].name, led_name_hdl[i],
>>> +                       &priv->client->dev);
>>> +        if (ret)
>>> +            break;
>>
>> You're assuming here that all LEDs will always be present, which
>> doesn't necessarily have to be true. How about just skipping
>> the LED and moving to the next one if given label was not found?
>> You could move this check to the beginning of the loop then.
>
> If we skip leds, there could be problem, if all the leds were
> skipped(For example, no ACPI name data present).

If all the LEDs were skipped, then no LED class device would be
registered. In such a case you could return an error from
lp3952_probe().

>
> At present, if a label is not present, the probe will fail.
> How about reverting to a default name (eg: "lp3952:blue2") if led name
> read fails?

There is no point in registering the LED if its DSD entry is not
valid. In case of Device Tree we fail LED registration in such
cases.

> This would also cover cases in which none of the led names
> were read properly.

Whole driver probing should fail then.

>
>>
>>> +
>>> +        priv->leds[i].cdev.name = priv->leds[i].name;
>>> +
>>> +        ret = devm_led_classdev_register(&priv->client->dev,
>>> +                         &priv->leds[i].cdev);
>>> +        if (ret < 0) {
>>> +            dev_err(&priv->client->dev,
>>> +                "couldn't register LED %s\n",
>>> +                priv->leds[i].cdev.name);
>>> +            break;
>>> +        }
>>> +    }
>>> +    return ret;
>>> +}
>>> +
> --
> To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>
>


-- 
Best regards,
Jacek Anaszewski