Re: [PATCH v2 1/2] leds: Add Intel Cherry Trail Whiskey Cove PMIC LEDs

[Jacek Anaszewski <jacek.anaszewski@gmail.com>]

Hi Pavel,

On 2/14/19 1:28 PM, Pavel Machek wrote:
> Hi!
> 
> Jacek, could we get you to comment here? I'd prefer "hardware" trigger...

What prevents use from using pattern trigger with its hw_pattern file?

Do you remember drivers/leds/leds-sc27xx-bltc.c and its breathing mode,
for which pattern trigger was introduced?

>>>> That assumes that breathing is the default setting on all hardware
>>>> and again I don't see why not to export this functionality to
>>>
>>> Save the status on boot, then restore it on rmmod/reboot/poweroff? :-).
>>
>> Which works until the system freezes one time. I believe that
>> if we are going to do a LED driver for the charging LED on these
>> devices, we MUST offer a way to put it back in its original
>> state, even if the state is foo-barred at bootup.
>>
>>>> userspace. Just because something does not fit in the existing
>>>> API is IMHO not a good reason to not expose it to userspace.
>>>>
>>>> I suggest that we deal with this special case by adding 3 custom
>>>> sysfs attributes:
>>>>
>>>> 1) "mode" which when read, prints, e.g. :
>>>> manual [on-when-charging]
>>>
>>> While this allows _user on console_ to control everything using echo,
>>> it is not suitable for applications trying to control LEDs.
>>>
>>> As there's nothing special about the case here, I believe we should
>>> have generic solution here.
>>>
>>> My preffered solution would be "hardware" trigger that leaves the LED
>>> in hardware control.
>>
>> As you explained in the parts which I snipped, there are many
>> devices which have a similar choice for a LED being under hw or
>> user control. I can see how this looks like a trigger and how we
>> could use the trigger API for this.
>>
>> I believe though, that if we implement a "virtual" (for lack of
>> a better word) trigger for this, that this should be done in the
>> LED core. I can envision this working like this:
> 
> Agreed about the LED core.
> 
>> 1) Add a:
>>
>> hw_control_set(struct led_classdev *led_cdev, bool enable_hw_control);
>>
>> Callback to struct led_classdev which when implemented by a driver
>> like the current PMIC LED controller would do what it says.
>>
>> 2) Have the core create and register a virtual hardware trigger the
>> first time a LED cdev which has this callback gets registered.
>>
>> When configured as the trigger for this LED device this trigger calls
>> hw_control_set(cdev, true) and when unregistered calls
>> hw_control_set(cdev, false)
>>
>> Taking a quick look at the trigger code, a problem with this is
>> that normally any trigger can work with any led device, so this
>> "hardware" trigger will show up in the list of possible
>> triggers for each device.
>>
>> This problem can be solved by making the activate method for the
>> hardware trigger check the classdev has a hw_control_set callback
>> and if not return -EINVAL, or maybe -ENXIO but still this is somewhat
>> inconsistent with other triggers, which AFAIK work with any LED.
> 
> I guess other option is to modify core so that it does not list
> "hardware" trigger for leds that don't support it.

Pattern trigger will not show hw_pattern file if pattern_set/get ops
are not provided.

>>> Alternatively I could imagine "hardware" sysfs attribute, containing
>>> 0/1, with 0==software controlled, 1==hardware controlled.
>>
>> Hmm, maybe call it "hardware_controlled" instead ? Otherwise this
>> would work for me and I would personally prefer this solution. This
>> could even be done in the LED core using the hw_control_set callback
>> I proposed, to make sure it is handled consistently between devices.
> 
> This should be in LED core, yes.
> 
>>> The rest of attributes would have to be Cove-specific, yes (but still
>>> should fit with the rest of LED subsystem).
>>
>> Right, I see that the triggers attribute already uses the fmt where
>> on "cat" all options are listed and the current active one has [] around it,
>> so I think the pattern and frequency attributes I proposed should work
>> well, although thinking more about this I believe the freq. attribute should
>> be called pattern_freq to make clear it applies to blinking / breathing
>> set through the pattern attribute.
> 
> Take a look at blinking trigger. It can already do hardware
> acceleration, but uses different format than what you proposed.
> 
> Best regards,
> 									Pavel
> 

-- 
Best regards,
Jacek Anaszewski