[PATCHv5 2/2] INPUT: Introduce generic trigger/LED pairs to input LEDs

[Dmitry Torokhov]

Hi Samuel,

On Mon, Jan 05, 2015 at 12:45:01AM +0100, Samuel Thibault wrote:
> Dmitry Torokhov, le Sun 04 Jan 2015 15:28:38 -0800, a écrit :
> > I'd rather we did not have a separate config option for this. Do we really need to
> > support case where LEDs are disabled?
> 
> I don't really mind.
> 
> > I'd rather stub it out instead of providing 2 separate code paths.
> 
> Ok.
> 
> > > +/* LED state change for some keyboard, notify that keyboard.  */
> > > +static void perdevice_input_led_set(struct led_classdev *cdev,
> > > +			  enum led_brightness brightness)
> > > +{
> > > +	struct input_dev *dev;
> > > +	struct led_classdev *leds;
> > > +	int led;
> > > +
> > > +	dev = cdev->dev->platform_data;
> > 
> > Umm, platform data is not the best place for storing this. Why not drvdata?
> 
> Just because it didn't exist when I wrote the code :)
> Ok.
> 
> > > +/* A new input device with potential LEDs to connect.  */
> > > +int input_led_connect(struct input_dev *dev)
> > > +{
> > > +	int i, error = -ENOMEM;
> > > +	struct led_classdev *leds;
> > > +	struct led_trigger *triggers;
> > > +
> > > +	leds = kcalloc(LED_CNT, sizeof(*leds), GFP_KERNEL);
> > > +	if (!leds)
> > > +		goto err;
> > 
> > Why do we allocate all possible led's for every device?
> 
> Ah, right, that was making things simpler, but it could be
> squeezed.  I'm just afraid of one thing: may dev->ledbit change after
> input_register_device?  It seems that at least uinput somehow permits
> this.  It then means having to store the number of actually created LEDs
> and triggers alongside.

The input device's capabilities (with the exception of keymap) should
not change after device registration. If uinput allows that we should
fix it there.

Thanks.

> 
> > > +	dev->leds = leds;
> > > +
> > > +	triggers = kcalloc(LED_CNT, sizeof(*triggers), GFP_KERNEL);
> > > +	if (!triggers)
> > > +		goto err;
> > > +	dev->triggers = triggers;
> > 
> > Hmm, maybe having per-device triggers is a bit of overkill and we could have
> > just "input-numl", "input-capsl", etc.
> 
> No, that won't work, notably for evdev access: we have to respect the
> per-device semantic.
> 
> > > +	/* No issue so far, we can register for real.  */
> > > +	for (i = 0; i < LED_CNT; i++)
> > > +		if (leds[i].name) {
> > > +			led_classdev_register(&dev->dev, &leds[i]);
> > > +			leds[i].dev->platform_data = dev;
> > > +			led_trigger_register(&triggers[i]);
> > 
> > We need error handling here.
> 
> Right.
> 
> Samuel

-- 
Dmitry