Re: [PATCH v2 2/2] leds: lp50xx: Add the LP50XX family of the RGB LED driver

[Dan Murphy <dmurphy@ti.com>]

Jacek

On 1/22/19 3:39 PM, Jacek Anaszewski wrote:
> Hi all,
> 
> On 1/20/19 7:42 AM, Vesa Jääskeläinen wrote:
>> Hi Dan,
>>
>> On 18/01/2019 15.58, Dan Murphy wrote:
>>> Jacek
>>>
>>> On 1/18/19 7:45 AM, Dan Murphy wrote:
>>>> Jacek
>>>>
>>>> On 1/17/19 3:10 PM, Jacek Anaszewski wrote:
>>>>> Hi Dan,
>>>>>
>>>>> On 1/16/19 7:41 PM, Dan Murphy wrote:
>>>>>> Hello
>>>>>>
>>>>>> On 1/16/19 4:55 AM, Pavel Machek wrote:
>>>>>>> Hi!
>>>>>>>
>>>>>>>> On 1/15/19 4:22 PM, Pavel Machek wrote:
>>>>>>>>> Hi!
>>>>>>>>>
>>>>>>>>>>> +The 24-bit RGB value passed in follows the pattern 0xXXRRGGBB
>>>>>>>>>>> +XX - Do not care ignored by the driver
>>>>>>>>>>> +RR - is the 8 bit Red LED value
>>>>>>>>>>> +GG - is the 8 bit Green LED value
>>>>>>>>>>> +BB - is the 8 bit Blue LED value
>>>>>>>>>>> +
>>>>>>>>>>> +Example:
>>>>>>>>>>> +LED module output 4 of the LP5024 will be a yellow color:
>>>>>>>>>>> +echo 0xe6de00 > /sys/class/leds/lp5024\:led4_mod/color
>>>>>>>>>>> +
>>>>>>>>>>> +LED module output 4 of the LP5024 will be dimmed 50%:
>>>>>>>>>>> +echo 0x80 > /sys/class/leds/lp5024\:led4_mod/brightness
>>>>>>>>>>> +
>>>>>>>>>>> +LED banked RGBs of the LP5036 will be a white color:
>>>>>>>>>>> +echo 0xffffff > /sys/class/leds/lp5036\:led_banked/color
>>>>>>>>>>
>>>>>>>>>> This part with example cans remain in Documentation/leds if you
>>>>>>>>>>> like.
>>>>>>>>>
>>>>>>>>> Does it actually work like that on hardware?
>>>>>>>>
>>>>>>>> What?
>>>>>>>
>>>>>>> If you do echo 0xffffff > /sys/class/leds/lp5036\:led_banked/color,
>>>>>>> does it actually produce white? With all the different RGB modules
>>>>>>> manufacturers can use with lp5024P?
>>>>>>>
>>>>>>> If you do echo 0xe6de00 > /sys/class/leds/lp5024\:led4_mod/color, does
>>>>>>> it actually produce yellow, with all the different RGB modules
>>>>>>> manufacturers can use with lp5024P?
>>>>>>>
>>>>>>
>>>>>> I believe the answer to the general questions is no for any RGB cluster and driver out there.
>>>>>> Because if you set the same values on each and every RGB device out there you will get varying shades of the color.
>>>>>> But for this device yes the color does appear to be yellow to me versus what was displayed on my monitor by the HSL picker.
>>>>>> But everyone interprets colors differently.
>>>>>>
>>>>>> If you write the same value for yellow or white on a droid 4 and the N900 do they produce the same color side by side?
>>>>>> Most probably not.
>>>>>>
>>>>>> As you pointed out the PWM needs to be modified to obtain the correct white color to account for LED and other device constraints.
>>>>>>
>>>>>> But we need to take into account the light pipe.  Pools nowadays have RGB LED spot lights in them.  It can
>>>>>> be set to white.  On my pool right off the lens the color has a purplish hue to it.  As the light is diffracted into
>>>>>> the pool the color becomes white.  The pool is clear.  When I add chemicals to the pool and make it cloudy
>>>>>> and turn on the lights the color off the lens is now white.  This is an example on a large scale but the issue
>>>>>> scales down to the hand helds and smart home applications.
>>>>>>
>>>>>> If the cluster is piped through a flexible optic 0xffffff may produce the "white" you want on its output.
>>>>>>
>>>>>> So an expectation of certain color without proper piping based on a single RGB value may be a little unreasonable.
>>>>>> There may need to be a way to attenuate the values based on the hardware aspect of the equation ie light pipe (or lack thereof) and LED vendor.
>>>>>> So if we write 0xffffff to the RGB driver the driver could adjust the intensity of the individual LEDs based on the diffraction
>>>>>> coefficients.
>>>>>>
>>>>>> I also think that is an unreasonable expectation here that writing a single value to any LED RGB driver would produce
>>>>>> a "rest of the world" absolute color.  Maybe it can produce something similar but not identical.
>>>>>> As you indicated in the requirements there is more involved here then just the LED and the values written.
>>>>>> The colors should be close but may not be identical.
>>>>>>
>>>>>> A 10 year old N900 should not be considered the gold standard for color production due to advancements in LED,
>>>>>> light pipe and LED driver technology.
>>>>>> The single package RGB clusters on the board I am testing is about the size of a single RGB LED from 10 years ago.
>>>>>>
>>>>>> I agree that the interface developed should work on the device but the algorithm derived to obtain the color needs to have
>>>>>> a hardware aspect to the calculation.
>>>>>>
>>>>>>>>> Is it supposed to support "normal" RGB colors as seen on monitors?
>>>>>>>>
>>>>>>>> Monitors are not an application for this part.
>>>>>>>
>>>>>>> You did not answer the question. When you talk about yellow, is it
>>>>>>> same yellow the rest of world talks about?
>>>>>>>
>>>>>>
>>>>>> See above.  It is close to what was on my monitor displayed.
>>>>>>
>>>>>>>>> Because 100% PWM on all channels does not result in white on hardware
>>>>>>>>> I have.
>>>>>>>>
>>>>>>>> I don't know I am usually blinded by the light and have no diffuser over
>>>>>>>> the LEDs to disperse the light so when I look I see all 3 colors.
>>>>>>>
>>>>>>> How can we have useful discussion about colors when you don't see the
>>>>>>> colors?
>>>>>>>
>>>>>>> Place a piece of paper over the LEDs....
>>>>>>>
>>>>>>
>>>>>> Good suggestion for a rough test.
>>>>>>
>>>>>>>>> But...
>>>>>>>>>
>>>>>>>>> I believe we should have a reasonable design before we do something
>>>>>>>>> like this. There's no guarantee someone will not use lp50xx with just
>>>>>>>>> the white LEDs for example. How will this work? Plus existing hardware
>>>>>>>>> already uses three separate LEDs for RGB LED. Why not provide same
>>>>>>>>> interface?
>>>>>>>>
>>>>>>>> Which existing hardware?  Are they using this part?
>>>>>>>
>>>>>>> Nokia N900. They are not using this part, but any interface we invent
>>>>>>> should work there, too.
>>>>>>>
>>>>>>
>>>>>> Yes a common interface would be nice with some sort of hardware tuning coefficient.
>>>>>>
>>>>>>>> <rant>
>>>>>>>> Why are we delaying getting the RGB framework or HSV in?
>>>>>>>> I would rather design against something you want instead of having
>>>>>>>> everyone complain about every implementation I post.
>>>>>>>> </rant>
>>>>>>>
>>>>>>> Because you insist on creating new kernel interfaces, when existing
>>>>>>> interfaces work, and are doing that badly.
>>>>>>>
>>>>>>> Because your patches are of lower quality than is acceptable for linux
>>>>>>> kernel.
>>>>>>>
>>>>>>> Because you don't seem to be reading the emails.
>>>>>>>
>>>>>>> I sent list of requirements for RGB led support. This does not meet
>>>>>>> them.
>>>>>>>
>>>>>>
>>>>>> Sigh.  You did not answer my question.
>>>>>>
>>>>>> Your requirements seem to be centered around monitors but that is only one application of the current
>>>>>> RGB LED landscape.
>>>>>>
>>>>>> I am willing to work with you on the HSV and adapting the LP50xx part to this framework.
>>>>>> Or any RGB framework for that matter.  I still don't agree with the kernel needing to declare colors
>>>>>>    maybe color capabilities but not specific colors.
>>>>>
>>>>> Dan, if you have a bandwidth for LED RGB class implementation
>>>>> then please go ahead. It would be good to compare colors produced
>>>>> by software HSV->RGB algorithm to what can be achieved with
>>>>> LEDn_BRIGHTNESS feature.
>>>>>
>>>>> The requirements for LED RGB class as I would see it:
>>>>>
>>>>> sysfs interface:
>>>>>
>>>>> brightness-model: space separated list of available options:
>>>>> - rgb (default):
>>>>>    - creates color file with "red green blue" decimal values
>>>>
>>>> What about other colored LEDs?  Presenting RGB for an Amber LED does not seem right.
>>>> Should the LED color come from the DT?
>>>>
>>>
>>> I thought about this, other non-RGB LEDs would not use the RGB framework.
>>> But should they have the same interfaces as RGB?
>>>
>>> Should PWM control be a global interface?
>>
>> In order to being able to set multi color element led at one go I would recommend using then model:
>>
>> color_names: "red green blue white"
>>
>> echo "32 43 0 128" > color
>>
>> This way all elements would be set at same time from user space point of view.
>>
>> This of course requires that it is part of the physical/logical led that is being controlled. If it is a separate thing then it would logically be differently controlled mono color led.
>>
>> If you look what kinds of leds are available lets say from digikey you get all kinds of combos:
>>
>> - red, green, blue
>> - red, green, blue, amber
>> - red, green, blue, white
>> - red, green, blue, white - cool
>> - red, green, blue, white - neutral
>> - red, green, blue, white - warm
>> - red, orange
>> - purple, ultraviolet
>> - amber, blue
>> - amber, blue, cyan, green, red, violet, white - cool
>> - amber, blue, green
>> - amber, green, blue
>> - and then lots of single special colors
> 
> It suggested me another solution. Instead of LED RGB class
> we would have LED multi-color class.
> 

I was thinking the same thing this morning.  But I was thinking that the RGB
class should be an additional class to stand on its own or can register to the
multi color class.

> Sysfs interface design:
> -----------------------
> 
> colors: directory containing files that map to
>         the brightness of particular LEDs; there
>         would be predefined color names that LED class
>         driver should map iouts to, e.g.:

Filling in the missing ideas with questions.

Is it a directory or a file?  If it is a directory does that not break the current
directory label model? 

so the path would be /sys/class/leds/colors ?  (It is probably not this but needs clarification)
How would this look if I had 2 of the same color LEDs?  The Beagle bone black has 2 Blue LEDs.
They are assigned to different triggers and have different directories.  Both are GPIO controlled.

Or are you saying it would be something like (More then likely this is what you intended)
/sys/class/leds/input4::numlock/colors? 
Maybe it is mandated that "multi" be added to the label when the class is registered so the caller
knows that this is a multi color class and not a single LED color class.

What about providing a file called colors_raw which takes in the raw decimal values to obtain other color
variants when RGB is only available?  And this can also present a single write to the kernel with the new
color values.

I am not a fan of hard coding preset colors as you can see there are to many of them and variations of the color.
In addition this severely limits the ability of the user.  Unless we stick to primary colors only and not secondary
colors.
Defining and hard coding hte colors can get out of control and not maintainable moving forward.  Especially
if we start adding defines like white_warm, white_neutral and other variations to the color.

What about IR LEDs used for night vision products?  Do these fall into the multi color class?
We do have a driver I submitted that had an IR LED and a White LED combined.  It was created against the
flash class but it could be a multi color LED as well.

How would traversing through the full color palette work if you were to want to produce a multi
color ring like the LP50xx where the pattern can traverse from one end of the color spectrum and back?
Or in a product like the gaming keyboards that will change color or change backlight brightness?

Not sure what color LEDs the keyboard manufacturers place on their keyboards but does the interface design
capable of doing this?

https://www.youtube.com/watch?v=pfKv3g2FeBE
or something like this

https://www.youtube.com/watch?v=PO3auX3f5C4

The LP5036 has this capability.

>   - red
>   - green
>   - blue
>   - white
>   - sync: accepts "write" and "read", which executes
>           write/readout to/from a device respectively.
> 

What are these above, the values or the files under the colors directory?

I am assuming they are files.

Are they mandatory or optional?

> brightness-model: defines brightness level to color configuration
>                   mapping
>   - "hardware": for devices with feature like LEDn_BRIGHTNESS of lp50xx
>   - "rgb-<hue>": available only when all three red,green,blue colors
>                  are present in the colors directory.
>                  <hue> is a placeholder for given DT hue presets.
>   - "rgbw-<hue>": Same as above, but available only when white color
>                  (any of amber or white-{cool|neutral|warm} can be
>                  mapped to white) is also available. In this mode
>                  max_brightness equals num-of-hue-presets + 1, and
>                  max_brightness, when set, turns the "white" LED on

Why do we need white combined here?  Should this not be its own entity?

Again I don't like limiting the color palette from the DT.  I think that the
user space can see what colors are available for that device and makes its own
decision on what color to present.

For the RGBw what about RGB amber and RGB purple.  Are the white LEDs always part of the
same function trying to be achieved by the system designer?  The RGB can be used to denote
notification status and the white can be used to denote that a charger is connected.  Motorola
Droid did this.


>   - "rgb-linear": I'm not sure if it should be available - it will
>                   have unpredictable results
> 
> brightness: sets/reads brightness in the way specific to the
>             current brightness-model. When more colors are available
>             (e.g. amber, blue, cyan, green, red, violet, white), they
>             are not touched by write to brightness).
> 
> HSV->RGB conversion is left entirely to the userspace, which can set any
> color with use of the proposed interface.
> 
> Let's agree on sysfs interface first, and only after that move
> to the DT details.
> 

DT's are meant to describe hardware and not describe the product.  Unless Rob does not see
an issue with defining product capabilities in the DT then we should keep that out of the DT.

Dan

> Best regards,
> Jacek Anaszewski
> 
>> Better models even list some properties about how different color elements behave.
>>
>> I would expect red, green, blue to be the most common case. So automating for that for chip like what you are working now is reasonable.
>>
>> For simple case where we can expect red, green, blue mapping we can auto-configure the multi color element led with "red", "green" and "blue" color elements.
>>
>> But TI also have other kind of driver chips that could be used in "more freely" enabling use all of those combos.
>>
>> If user would have "amber", "green" and "blue" then we would need to add this definition in devicetree in order for user space to be able to recognize that now we have different config.
>>
>> Or if user would have most complex of those then in devicetree you would configure 7 color element led like "amber", "blue", "cyan", "green", "red", "violet", "white" as color elements.
>>
>> Then you could set them in one go:
>> echo "32 76 43 2 43 76 54" > color
>>
>> Now in this special case if we would have hsl brightness model and lets say it only knows red, green and blue. Then it would pick red, green and blue color element names and only adjust those others would be linear should there be non-zero value.
>>
>> In this case however you might be better doing it in user space as it might be preferable to control all values let say in sRGB space and then let the software transform result to all color elements and then setup entries to kernel. In here the ICC color profile might be the way to go to model your let and let user space software do the fancier math.
>>
>> As this latest example is the fanciest I would like to remind that the red, green, blue is the most common. I would make their life easier and then when you go to more advanced setups then you are are already mentally ready to do more complex things :)
>>
>> As an example in our case we can do with "red", "green", and "red", "green", "blue" combos fill all our current requirements we have. Perhaps white could be added as one thing in future (white is currently used only in lcd module's background).
>>
>> Latest proposal I have for the interface now lets you do all of those things so it should be easy enough for most cases and still flexes to more complex cases while keeping kernel space simple.
>>
>> Thanks,
>> Vesa Jääskeläinen
>>
> 
> 


-- 
------------------
Dan Murphy