Re: [PATCH/RFC v3 5/5] media: Add registration helpers for V4L2 flash sub-devices

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

Hi Sakari,

On 04/23/2014 05:24 PM, Sakari Ailus wrote:
> Hi Jacek,
>
> On Thu, Apr 17, 2014 at 10:26:44AM +0200, Jacek Anaszewski wrote:
>> Hi Sakari,
>>
>> Thanks for the review.
>>
>> On 04/16/2014 08:21 PM, Sakari Ailus wrote:
>>> Hi Jacek,
>>>
>>> Thanks for the update!
>>>
>> [...]
>>>> +static inline enum led_brightness v4l2_flash_intensity_to_led_brightness(
>>>> +					struct led_ctrl *config,
>>>> +					u32 intensity)
>>>
>>> Fits on a single line.
>>>
>>>> +{
>>>> +	return intensity / config->step;
>>>
>>> Shouldn't you first decrement the minimum before the division?
>>
>> Brightness level 0 means that led is off. Let's consider following case:
>>
>> intensity - 15625
>> config->step - 15625
>> intensity / config->step = 1 (the lowest possible current level)
>
> In V4L2 controls the minimum is not off, and zero might not be a possible
> value since minimum isn't divisible by step.
>
> I wonder how to best take that into account.

I've assumed that in MODE_TORCH a led is always on. Switching
the mode to MODE_FLASH or MODE_OFF turns the led off.
This way we avoid the problem with converting 0 uA value to
led_brightness, as available torch brightness levels start from
the minimum current level value and turning the led off is
accomplished on transition to MODE_OFF or MODE_FLASH, by
calling brightness_set op with led_brightness = 0.

[...]

>>>> +/*
>>>> + * V4L2 subdev internal operations
>>>> + */
>>>> +
>>>> +static int v4l2_flash_open(struct v4l2_subdev *sd, struct v4l2_subdev_fh *fh)
>>>> +{
>>>> +	struct v4l2_flash *v4l2_flash = v4l2_subdev_to_v4l2_flash(sd);
>>>> +	struct led_classdev *led_cdev = v4l2_flash->led_cdev;
>>>> +
>>>> +	mutex_lock(&led_cdev->led_lock);
>>>> +	v4l2_call_flash_op(sysfs_lock, led_cdev);
>>>
>>> Have you thought about device power management yet?
>>
>> Having in mind that the V4L2 Flash sub-device is only a wrapper
>> for LED driver, shouldn't power management be left to the
>> drivers?
>
> How does the LED controller driver know it needs to power the device up in
> that case?
>
> I think an s_power() op which uses PM runtime to set the power state until
> V4L2 sub-device switches to it should be enough. But I'm fine leaving it out
> from this patchset.
>

This solution looks reasonable.

Regards,
Jacek Anaszewski