Re: [RFC PATCH] led: add Cycle LED trigger.

["Gaël PORTAY" <g.portay@overkiz.com>]

On Jun 20, 2013, at 7:58 PM, Bryan Wu <cooloney@gmail.com> wrote:

> On Thu, Jun 20, 2013 at 2:44 AM, Gaël PORTAY <g.portay@overkiz.com> wrote:
>> On 19/06/2013 00:05, Joe Perches wrote:
>>>
>>> On Tue, 2013-06-18 at 18:24 +0200, Gaël PORTAY wrote:
>>>>
>>>> Currently, none of available triggers supports playing with the LED
>>>> brightness
>>>> level.  The cycle trigger provides a way to define custom brightness
>>>> cycle.
>>>> For example, it is easy to customize the cycle to mock up the rhythm of
>>>> human
>>>> breathing which is a nice cycle to tell the user the system is doing
>>>> something.
>>>
>>> I think maybe this is a userspace thing, but here's a
>>> trivial comment or two
>>>
>>>
>>>> +static int cycle_start(struct cycle_trig_data *data)
>>>> +{
>>>> +       unsigned long flags;
>>>> +
>>>> +       if (hrtimer_active(&data->timer))
>>>> +               return -EINVAL;
>>>> +
>>>> +       spin_lock_irqsave(&data->lock, flags);
>>>> +       data->plot_index = 0;
>>>> +       data->cycle_count = 0;
>>>> +       hrtimer_start(&data->timer, ktime_get(), HRTIMER_MODE_ABS);
>>>> +       spin_unlock_irqrestore(&data->lock, flags);
>>>> +
>>>> +       return 1;
>>>
>>> Maybe return 0 on success
>>>
>>>> +static ssize_t cycle_control_store(struct device *dev,
>>>> +                                   struct device_attribute *attr,
>>>> +                                   const char *buf, size_t size)
>>>> +{
>>>> +       struct led_classdev *led_cdev = dev_get_drvdata(dev);
>>>> +       struct cycle_trig_data *data = led_cdev->trigger_data;
>>>> +
>>>> +       if (strncmp(buf, "start", sizeof("start") - 1) == 0)
>>>> +               cycle_start(data);
>>>> +       else if (strncmp(buf, "stop", sizeof("stop") - 1) == 0)
>>>> +               cycle_stop(data);
>>>> +       else if (strncmp(buf, "reset", sizeof("reset") - 1) == 0)
>>>> +               cycle_reset(data);
>>>> +       else if (strncmp(buf, "pause", sizeof("stop") - 1) == 0)
>>>> +               cycle_pause(data);
>>>> +       else if (strncmp(buf, "resume", sizeof("resume") - 1) == 0)
>>>> +               cycle_resume(data);
>>>> +       else
>>>> +               return -EINVAL;
>>>
>>> I think strcasecmp better than strncmp
>>>
>>>> +static ssize_t cycle_rawplot_store(struct device *dev,
>>>> +                                  struct device_attribute *attr,
>>>> +                                  const char *buf, size_t size)
>>>> +{
>>>
>>> []
>>> +       plot = kzalloc(size, GFP_KERNEL);
>>> +       if (plot) {
>>> +               hrtimer_cancel(&data->timer);
>>>
>>> Ick.
>>>
>>>        if (!plot)
>>>                return -ENOMEM;
>>>
>>>        etc...
>>>
>> Hello,
>>
>> Thanks a lot for your review. I took your remarks into consideration and
>> fixed the mistakes.
>>
>> About the kernel/user space discussion, I'd rather keep the cycle trigger
>> implementation in the kernel space,
>> because it implies brightness change every 10-100ms or less. This 
>> leads to
>> lots of context switches, and I'm not
>> even sure the user space can handle such timings accurately.
>>
>> Could you detail your concerns about adding this driver in the kernel?
>>
>> Best Regards,
>> Gaël
>
> Hi Gaël,
>
> Is that possible to extend an existing leds trigger like ledtrig-timer
> or other triggers instead of creating a new one?
>
> Thanks,
> -Bryan

Hi Bryan,

I'm sorry but the cycle trigger interface is too much different from 
already defined triggers. That's why I have decided to create a new one.

The cycle trigger purposes are:
  - to control a cycle (start/stop pause/resume reset), and
  - to play with the brightness level

But it's possible to implement the timer and heartbeat triggers using 
the interface of the cycle trigger.

Yours sincerely,
Gaël