Re: [PATCH 1/3] leds: triggers: add support for RGB triggers

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

On 03/23/2016 05:36 PM, Heiner Kallweit wrote:
> Am 23.03.2016 um 17:02 schrieb Jacek Anaszewski:
>> On 03/23/2016 12:57 PM, Heiner Kallweit wrote:
>>> Am 23.03.2016 um 09:32 schrieb Jacek Anaszewski:
>>>> On 03/22/2016 11:06 PM, Heiner Kallweit wrote:
>>>>> Am 22.03.2016 um 17:00 schrieb Jacek Anaszewski:
>>>>>> On 03/22/2016 12:47 PM, Heiner Kallweit wrote:
>>>>>>> Am 22.03.2016 um 09:05 schrieb Jacek Anaszewski:
>>>>>>>> On 03/21/2016 06:34 PM, Heiner Kallweit wrote:
>>>>>>>>> Am 21.03.2016 um 16:35 schrieb Jacek Anaszewski:
>>>>>>>>>> On 03/19/2016 08:11 PM, Heiner Kallweit wrote:
>>>>>>>>>>> Am 18.03.2016 um 14:10 schrieb Jacek Anaszewski:
>>>>>>>>>>>> On 03/17/2016 08:53 PM, Heiner Kallweit wrote:
>>>>>>>>>>>>> Am 17.03.2016 um 14:41 schrieb Jacek Anaszewski:
>>>>>>>>>>>>>> Hi Heiner,
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> On 03/13/2016 06:14 PM, Heiner Kallweit wrote:
>>>>>>>>>>>>>>> Add basic support for RGB triggers. Triggers with flag LED_TRIG_CAP_RGB
>>>>>>>>>>>>>>> set are available to RGB LED devices only.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
>>>>>>>>>>>>>>> ---
>>>>>>>>>>>>>>>          drivers/leds/led-triggers.c | 15 ++++++++++++---
>>>>>>>>>>>>>>>          include/linux/leds.h        |  3 +++
>>>>>>>>>>>>>>>          2 files changed, 15 insertions(+), 3 deletions(-)
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> diff --git a/drivers/leds/led-triggers.c b/drivers/leds/led-triggers.c
>>>>>>>>>>>>>>> index 2181581..3ccf88b 100644
>>>>>>>>>>>>>>> --- a/drivers/leds/led-triggers.c
>>>>>>>>>>>>>>> +++ b/drivers/leds/led-triggers.c
>>>>>>>>>>>>>>> @@ -30,6 +30,13 @@ static LIST_HEAD(trigger_list);
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>           /* Used by LED Class */
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> +static inline bool led_trig_check_rgb(struct led_trigger *trig,
>>>>>>>>>>>>>>> +                      struct led_classdev *led_cdev)
>>>>>>>>>>>>>>> +{
>>>>>>>>>>>>>>> +    return !(trig->flags & LED_TRIG_CAP_RGB) ||
>>>>>>>>>>>>>>> +           led_cdev->flags & LED_DEV_CAP_RGB;
>>>>>>>>>>>>>>> +}
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Could you explain what is the purpose of this function?
>>>>>>>>>>>>>> What actually do we want to check here?
>>>>>>>>>>>>>>
>>>>>>>>>>>>> Triggers using RGB functionality can't be used with non-RGB LED's.
>>>>>>>>>>>>> This check checks for such unsupported combinations:
>>>>>>>>>>>>> It returns false if the trigger uses RGB functionality but LED doesn't
>>>>>>>>>>>>> support the RGB extension.
>>>>>>>>>>>>
>>>>>>>>>>>> We need more meaningful name for it. Maybe led_trigger_is_supported() ?
>>>>>>>>>>>> And let's make it no-op for !CONFIG_LEDS_CLASS_RGB case.
>>>>>>>>>>>>
>>>>>>>>>>> OK, led_trigger_is_supported() is better.
>>>>>>>>>>>
>>>>>>>>>>> Making the function a no-op in the non-RGB case would have some impact:
>>>>>>>>>>> We'd have to make sure that all public trigger functions are a de-facto no-op
>>>>>>>>>>> for RGB triggers (at least register / unregister). Means we would need
>>>>>>>>>>> something like this in each public trigger function:
>>>>>>>>>>>
>>>>>>>>>>> #if !IS_ENABLED(CONFIG_LEDS_CLASS_RGB)
>>>>>>>>>>>          if (trig->flags & LED_TRIG_CAP_RGB))
>>>>>>>>>>>              return;
>>>>>>>>>>> #endif
>>>>>>>>>>>
>>>>>>>>>>> I think this would add a lot of overhead and therefore IMHO it's better to
>>>>>>>>>>> not make the check function a no-op.
>>>>>>>>>>
>>>>>>>>>> Wouldn't it suffice to make the no-op returning true?
>>>>>>>>>> Preventing RGB trigger registration for non-RGB LED class configuration
>>>>>>>>>> seems to be different thing, also to be considered.
>>>>>>>>>>
>>>>>>>>> No, it's not sufficient. Let's say the RGB extension is disabled and we have a RGB trigger.
>>>>>>>>> The check is a no-op now (returns always true), therefore the RGB trigger would be displayed
>>>>>>>>> in the list of available triggers also for all non-RGB LED's.
>>>>>>>>
>>>>>>>> If RGB trigger was made dependent on LED RGB class, then the related
>>>>>>>> Kconfig symbol would remain undefined in !CONFIG_LEDS_CLASS_RGB case.
>>>>>>>>
>>>>>>> Making a RGB trigger dependent on LED RGB class would mean to enclose all calls to trigger
>>>>>>> functions in the RGB trigger like this:
>>>>>>> #if IS_ENABLED(CONFIG_LEDS_CLASS_RGB)
>>>>>>>        trigger_function()
>>>>>>> #endif
>>>>>>
>>>>>> You probably think about the case when we have two triggers in
>>>>>> single module, like in the planned {rgb-}heartbeat case?
>>>>>>
>>>>>> If so this is an argument for having RGB triggers in separate files.
>>>>>>
>>>>> I mean the case of triggers implemented outside drivers/leds. There the trigger code
>>>>> often is not separated from other functionality (e.g. drivers/tty/vt/keyboard.c)
>>>>> and it's not directly under our (LED core) control.
>>>>>
>>>>>>> This would apply to led_trigger_(un)register, led_trigger_event, led_trigger_blink, etc.
>>>>>>> And I think it wouldn't be too nice to force other kernel modules wanting to implement
>>>>>>> a RGB trigger to add these conditional compile statements.
>>>>>>
>>>>>> What other modules do you have on mind? LED triggers are implemented in
>>>>>> their own files.
>>>>>>
>>>>> That's true for the triggers under drivers/leds/trigger, but not necessarily for triggers
>>>>> implemented in other parts of the kernel.
>>>>
>>>> In this case surrounding all the trigger implementation  with
>>>> IS_ENABLED(CONFIG_LEDS_CLASS_RGB) guard would do.
>>>
>>> Yes, that's what would need to be done. But IMHO it's not nice to force trigger implementations
>>> in other parts of the kernel to guard each trigger-related call this way.
>>
>> My main objection is that led_trigger_is_supported() would be redundant
>> in led_trigger_store() and led_trigger_show() for non-RGB LED subsystem
>> configuration.
>>
> Yes, it's redundant for non-RGB configurations. But it affects sysfs access only
> and overhead / impact should be minimal to negligible

I agree.

>>> Also it might happen
>>> that a trigger is implemented w/o this guarding and w/o informing you.
>>> Then this (RGB) trigger would show up also for all non-RGB LED's.
>>
>> It is likely that it wouldn't compile without led-rgb-core.o.
>>
> It would compile because the only relevant difference between a RGB and a non-RGB trigger is a flag
> being set in struct led_trigger.

RGB trigger would probably need to use some led-rgb-core API, e.g. as
in case of led_trigger_range_event() from your patch - we've already
agreed about moving most of its internals to led-rgb-core.c

>>> I still think that not making led_trigger_is_supported() a no-op in the non-RGB case is a small
>>> price for preventing such potential issues.
>>
>> We could avoid the issues by adding a guard in led_trigger_register(),
>> that would prevent RGB trigger registration in !CONFIG_LEDS_CLASS_RGB
>> case.
>>
> With "preventing registration" most likely you mean registering being a no-op.

Actually I mean checking if trigger is supported by current LED
subsystem configuration, i.e. we will need to use
led_is_trigger_supported() in led_trigger_register(). This is another
argument for this API to be no-op only if !CONFIG_LEDS_TRIGGERS.
To conclude - I agree that it shouldn't be no-op in
!CONFIG_LEDS_CLASS_RGB case.

> I'm afraid we'd need the same also in all other public trigger functions, because it may cause
> problems if registering is a no-op and we call e.g. led_trigger_event then (not being a no-op).
> That's what I meant when I wrote earlier in this thread that we might need something like this
> in all exported trigger functions:
>
> #if !IS_ENABLED(CONFIG_LEDS_CLASS_RGB)
> 	if (trig->flags & LED_TRIG_CAP_RGB))
> 		return;
> #endif
>
> And this seems to be much more overhead than the one check in sysfs access not being a no-op
> in the non-RGB case.
>
>>>> In the aformentioned drivers/tty/vt/keyboard.c we have even more generic
>>>> IS_ENABLED(CONFIG_LEDS_TRIGGERS) guard anyway.
>>>>
>>>>>>> Alternatively, as mentioned before, we would have to add this to all public trigger functions:
>>>>>>>
>>>>>>> #if !IS_ENABLED(CONFIG_LEDS_CLASS_RGB)
>>>>>>>        if (trig->flags & LED_TRIG_CAP_RGB))
>>>>>>>            return;
>>>>>>> #endif
>>>>>>> I think this would add significant overhead w/o gaining really something.
>>>>>>>
>>>>>>>>> We could maximum remove the "|| led_cdev->flags & LED_DEV_CAP_RGB" from the check if
>>>>>>>>> the RGB extension is disabled. But it's open whether this minimal gain in a non-critical
>>>>>>>>> code path justifies this.
>>>>>>>>>
>>>>>>>>>>>>>>>          ssize_t led_trigger_store(struct device *dev, struct device_attribute *attr,
>>>>>>>>>>>>>>>                  const char *buf, size_t count)
>>>>>>>>>>>>>>>          {
>>>>>>>>>>>>>>> @@ -52,12 +59,12 @@ ssize_t led_trigger_store(struct device *dev, struct device_attribute *attr,
>>>>>>>>>>>>>>>              down_read(&triggers_list_lock);
>>>>>>>>>>>>>>>              list_for_each_entry(trig, &trigger_list, next_trig) {
>>>>>>>>>>>>>>>                  if (sysfs_streq(buf, trig->name)) {
>>>>>>>>>>>>>>> +            if (!led_trig_check_rgb(trig, led_cdev))
>>>>>>>>>>>>>>> +                break;
>>>>>>>>>>>>>
>>>>>>>>>>>>> Check for the case that userspace wants to set a RGB trigger for a non-RGB LED via sysfs.
>>>>>>>>>>>>>
>>>>>>>>>>>>>>>                      down_write(&led_cdev->trigger_lock);
>>>>>>>>>>>>>>>                      led_trigger_set(led_cdev, trig);
>>>>>>>>>>>>>>>                      up_write(&led_cdev->trigger_lock);
>>>>>>>>>>>>>>> -
>>>>>>>>>>>>>>> -            up_read(&triggers_list_lock);
>>>>>>>>>>>>>>> -            goto unlock;
>>>>>>>>>>>>>>> +            break;
>>>>>>>>>>>>
>>>>>>>>>>>> This seems to be an unrelated cleanup. Please submit it separately.
>>>>>>>>>>>>
>>>>>>>>>>> OK
>>>>>>>>>>>
>>>>>>>>>>>>>>>                  }
>>>>>>>>>>>>>>>              }
>>>>>>>>>>>>>>>              up_read(&triggers_list_lock);
>>>>>>>>>>>>>>> @@ -84,6 +91,8 @@ ssize_t led_trigger_show(struct device *dev, struct device_attribute *attr,
>>>>>>>>>>>>>>>                  len += sprintf(buf+len, "none ");
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>              list_for_each_entry(trig, &trigger_list, next_trig) {
>>>>>>>>>>>>>>> +        if (!led_trig_check_rgb(trig, led_cdev))
>>>>>>>>>>>>>>> +            continue;
>>>>>>>>>>>>>
>>>>>>>>>>>>> Omit RGB triggers when listing the available triggers for a non-RGB LED via sysfs.
>>>>>>>>>>>>>
>>>>>>>>>>>>>>>                  if (led_cdev->trigger && !strcmp(led_cdev->trigger->name,
>>>>>>>>>>>>>>>                                      trig->name))
>>>>>>>>>>>>>>>                      len += sprintf(buf+len, "[%s] ", trig->name);
>>>>>>>>>>>>>>> diff --git a/include/linux/leds.h b/include/linux/leds.h
>>>>>>>>>>>>>>> index 58e22e6..07eb074 100644
>>>>>>>>>>>>>>> --- a/include/linux/leds.h
>>>>>>>>>>>>>>> +++ b/include/linux/leds.h
>>>>>>>>>>>>>>> @@ -248,6 +248,9 @@ enum led_brightness led_hsv_to_rgb(enum led_brightness hsv);
>>>>>>>>>>>>>>>          struct led_trigger {
>>>>>>>>>>>>>>>              /* Trigger Properties */
>>>>>>>>>>>>>>>              const char     *name;
>>>>>>>>>>>>>>> +    u8        flags;
>>>>>>>>>>>>>>> +#define LED_TRIG_CAP_RGB    BIT(0)
>>>>>>>>>>>>>>> +
>>>>>>>>>>>>>>>              void        (*activate)(struct led_classdev *led_cdev);
>>>>>>>>>>>>>>>              void        (*deactivate)(struct led_classdev *led_cdev);
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>
>>>
>>>
>>
>>
>
>
>


-- 
Best regards,
Jacek Anaszewski