Re: 2.6.22-rc[23]: blinking capslock led, stuck keys?

["Indan Zupancic" <indan@nul.nu>]

On Fri, June 15, 2007 07:41, Dmitry Torokhov wrote:
> Does the patch below help?

Didn't try it yet, but will tomorrow.


> Input: atkbd - throttle LED switching
>
> On some boxes keyboard controllers are too slow to withstand
> continuous flow of requests to turn keyboard LEDs on and off
> and start losing some keypresses or even all of them.
>
> Delay executing of LED switching request if we had another one
> withing 50 ms thus easing load on the controller.
>
> Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
> ---
>
>  drivers/input/keyboard/atkbd.c |   40 ++++++++++++++++++++++++++--------------
>  1 files changed, 26 insertions(+), 14 deletions(-)
>
> Index: work/drivers/input/keyboard/atkbd.c
> ===================================================================
> --- work.orig/drivers/input/keyboard/atkbd.c
> +++ work/drivers/input/keyboard/atkbd.c
> @@ -219,7 +219,8 @@ struct atkbd {
>  	unsigned long time;
>  	unsigned long err_count;
>
> -	struct work_struct event_work;
> +	struct delayed_work event_work;
> +	unsigned long event_jiffies;
>  	struct mutex event_mutex;
>  	unsigned long event_mask;
>  };
> @@ -565,7 +566,7 @@ static int atkbd_set_leds(struct atkbd *
>
>  static void atkbd_event_work(struct work_struct *work)
>  {
> -	struct atkbd *atkbd = container_of(work, struct atkbd, event_work);
> +	struct atkbd *atkbd = container_of(work, struct atkbd, event_work.work);
>
>  	mutex_lock(&atkbd->event_mutex);
>
> @@ -579,12 +580,30 @@ static void atkbd_event_work(struct work
>  }
>
>  /*
> + * Schedule switch for execution. We need to throttle requests,
> + * otherwise keyboard may become unresponsive.
> + */
> +static void atkbd_schedule_event_work(struct atkbd *atkbd, int event_bit)
> +{
> +	unsigned long delay = msecs_to_jiffies(50);
> +
> +	if (time_after(jiffies, atkbd->event_jiffies + delay))
> +		delay = 0;
> +
> +	atkbd->event_jiffies = jiffies;
> +	set_bit(event_bit, &atkbd->event_mask);
> +	wmb();
> +	schedule_delayed_work(&atkbd->event_work, delay);
> +}

I don't know whether schedule_delayed_work() requeues event_work, or if
it adds more work, but both seem to give wrong behaviour: In the first case
event_work can be postponed forever if atkbd_schedule_event_work() is
called repeatedly each time within 50 ms, and for the second case there's a
delay added, but the number of times the LED is switched stays the same,
so it's not being throttled.

Greetings,

Indan