Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Jacek Anaszewski <jacek.anaszewski@gmail.com>
Subject: Re: [PATCH 2/2] leds: lp5024: Add the LP5024/18 RGB LED driver
To:     =?UTF-8?B?VmVzYSBKw6TDpHNrZWzDpGluZW4=?= <dachaac@gmail.com>,
        Dan Murphy <dmurphy@ti.com>, Pavel Machek <pavel@ucw.cz>
Cc:     robh+dt@kernel.org, devicetree@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-leds@vger.kernel.org
References: <20181219162626.12297-1-dmurphy@ti.com>
 <20181219162626.12297-3-dmurphy@ti.com> <20181219193455.GA21159@amd>
 <8740cfd6-a6b5-ad27-313b-984a9febf18a@ti.com> <20181219201047.GA23448@amd>
 <fd624cc1-9997-1cd5-c5cc-62e6897997cf@ti.com>
 <54f28115-0a7d-8e9c-3bec-6e91fb3981ec@gmail.com>
 <cd90f179-6c16-823f-496f-27f141e9c322@ti.com>
 <a72c11c0-949b-bdbc-df2c-4490e04a1065@gmail.com>
 <986b5105-2fdb-bd25-7c8a-ca8fd1ade821@gmail.com>
Message-ID: <7f205102-e854-f1cb-cc03-1307d1cddc87@gmail.com>
Date:   Thu, 3 Jan 2019 23:05:42 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.4.0
MIME-Version: 1.0
In-Reply-To: <986b5105-2fdb-bd25-7c8a-ca8fd1ade821@gmail.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Vesa,

Thank you for sharing your ideas.

Please find my comment below.

On 1/1/19 2:45 PM, Vesa Jääskeläinen wrote:
> Hi All,
> 
> On 20/12/2018 14.40, Vesa Jääskeläinen wrote:
>> Idea was to define preset colors in device tree as an example when you 
>> are dealing with multi-color LEDs without PWM. In that case you only 
>> have GPIOs to control and then have a problem what does those GPIO's 
>> mean.
>>
>> With preset definitions one can use color names to act as a shortcut 
>> to configure GPIO's to proper state for that particular color.
>>
>> For more flexible setups where you have PWM or such control you have 
>> larger space of available colors. In this case you need to somehow 
>> define also meaning of those controls.
>>
>> Also we may not have LED with only red, green and blue elements. There 
>> might in example be amber, ultraviolet, white elements.
>>
>> This is where device tree is concerned. It helps us craft the logical 
>> definition for LED so that we can control it from user space in common 
>> way.
>>
>> Now the next problem then is how does user space work then.
>>
>> For multi-color LEDs it it important to change the color atomically so 
>> that no wrong colors are being shown as user space got interrupted 
>> when controlling it.
>>
>> Also we have brightness setting that would be useful for PWM 
>> controlled LEDs.
>>
>> Setting color is easy when you use preset names then you only need to 
>> deal with brightness value (eg. RGB -> HSV * brightness -> RGB). Of 
>> course here additional problem is other color elements are they then 
>> scaled according to brightness value?.
>>
>> Setting color as "raw" values is then next problem. In order to do it 
>> atomically it needs to be one atomic activation and could be eg. one 
>> write to "color" sysfs entry with combination of all color elements 
>> and perhaps additionally also brightness value. Next question is then 
>> what is the format for such entry then? What are the value ranges? In 
>> here we can utilize device tree definition to help define what kind of 
>> LED we do have and what kind of capabilities it does have.
>>
>> Additional problem risen also in discussion was non-linearity of some 
>> control mechanisms vs. perceived color. So there might be a need for 
>> curve mapping similarly what is with backlight control and that would 
>> be defined either in device tree and possibly in user space if there 
>> is a need for that. I suppose golden curve definition in device tree 
>> should be good enough.
>>
>> Then there was additional discussion about possible animation support 
>> but I would leave that for future design as that would then be 
>> utilizing the same framework.
>>
>> I suppose color space handling and that kind of stuff should be in 
>> some led core functionality and then raw control should be part of 
>> physical led driver.
>>
>> I was planning to play with it during holiday season but lets see how 
>> it goes. Feel free to also experiment with the idea.
> 
> I was playing with this and got some results with PWM LED driver. I 
> would like to get feedback now even thou it is not yet ready for patch 
> sending.
> 
> They still need more work but the idea can be seen here:
> https://github.com/vesajaaskelainen/linux/tree/wip-multi-color-led
> 
> This branch is now based on Linux kernel 4.20 release.
> 
> Consider that branch as volatile as I will forcibly update it when there 
> are updates.
> 
>  From there specifically in commits (while they last):
> 
> drivers: leds: Add core support for multi color element LEDs
> https://github.com/vesajaaskelainen/linux/commit/55d553906d0a158591435bb6323a318462079d59 
> 
> 
> WIP: drivers: leds: leds-pwm: Add multi color element LED support.
> https://github.com/vesajaaskelainen/linux/commit/efccef08cbf3b2e1e49b95b69ff81cd380519fe3 
> 
> 
> What is there now:
> 
> - led-core supports color elements
> - led-class supports users space configuration
> - both led-core and led-class are driver agnostic so they should be 
> treated as generic code.
> - leds-pwm: my testing code with PWM led.
> - no HSV support for brightness as there could be multiple color 
> elements out from traditional red-green-blue space or odd combinations 
> of colors and they are a bit hard to map to HSV formula (and it needs 
> fixed point math).
> - no color presets that could be optionally be selected
> - when I configure led trigger to heartbeat it actually blinks with 
> color specified -- thou trigger gets zeroed out with one sets new color 
> or brightness as that was previous functionality with brightness.
> - some documentation added
> - code should pass checkpatch
> 
> What I was planning to do next:
> 
> - cleanup PWM LED driver so that it works with and without 
> LED_MULTI_COLOR_LED being defined.
> - improve documentation
> - try out how my other device behaves which have dual color element LED 
> controlled with GPIO's and see how it would integrate to gpio-led driver.
> 
> I would like to get feedback on:
> - Device tree idea
> - Internal logic
> - Should the trigger be really reseted when one changes value of 
> brightness? I would think it should function like setting brightness 
> entry from sysfs would set current brightness for trigger when it is 
> lit. Setting color should change color and brightness and it should be 
> active from there one until trigger is disabled from trigger sysfs node.
> 
> My testing device has RGB LED with all color elements controlled with 
> individual PWM channels from TI's AM335x's integrated PWM controller.
> 
> In device tree I have following:
> 
>      multi-color-leds {
>          compatible = "pwm-leds";
> 
>          status-led {
>              label = "status";
> 
>              element-red {
>                  pwms = <&ehrpwm0 0 100000 0>;
>              };
>              element-green {
>                  pwms = <&ehrpwm1 0 100000 0>;
>              };
>              element-blue {
>                  pwms = <&ehrpwm1 1 100000 0>;
>              };
>          };
>      };
> 
> For my second test device I was planning to replace "pwms" with "gpios" 
> or such entries.
> 
> In user space one can use it like:
> 
> # --- start of snippet ---
> 
> hostname ~ # cd /sys/class/leds/
> hostname leds # ls
> status
> hostname leds # cd status
> hostname status # ls
> brightness      color           device          max_brightness 
> max_color       power           subsystem       trigger         uevent
> hostname status # cat color
> brightness=0 red=0 green=0 blue=0

This breaks one-value-per-file sysfs rule.

Regarding led_scale_color_elements() - I checked it in GIMP and
the results are not satisfactory when increasing brightness.
Even if we managed to fix it, the result would not be guaranteed
to be the same across all devices.

This is still the same problem.

I have another proposal, being a mix of what has been discussed so far:

    RGB LED class will expose following files:
    a) available by default:
      - red, green, blue
        Writing any of these file will result in writing corresponding
        device register.
      - color_space: it would accept color space, e,g. "hsv", that would
                     have to be supported by LED RGB core; setting color
                     space would create relevant files, e.g. for hsv
                     hue. saturation, brightness, and remove default ones
                     other "color spaces" could be defined in DT as
                     proposed by Vesa; reading this file would print
                     available color spaces

    b) available conditionally:
      - brightness
       It will be exposed by devices that have hardware support for
       changing color brightness, like lp5024, or it will be made
       available after setting relevant color space, like "hsv", or
       other color presets defined in DT

I think it will be flexible enough to meet everyone's needs.

Current triggers would work only when brightness file is available.

This is ad hoc design so it can have some logical flaws.

Best regards,
Jacek Anaszewski

> 
> # some notes about color value here.
> # - red, green, and blue are directly coming from device tree
> # definition. if one would add eg. element-white or element-uv entries
> # then they would appear here as well.
> # - brightness is hardcoded entry
> 
> hostname status # cat max_color
> brightness=255 red=255 green=255 blue=255
> hostname status # cat brightness
> 0
> hostname status # cat max_brightness
> 255
> 
> # Set color and brightness to maximum values eg. "white"
> hostname status # cat max_color > color
> 
> # Zero out red and green elements to make color blue
> hostname status # echo "red=0 green=0" > color
> 
> # Configure individual color elements
> hostname status # echo "red=55 green=22 blue=11" > color
> 
> # Start with max brightness and then dim it with each step
> hostname status # echo 255 > brightness
> hostname status # echo 128 > brightness
> hostname status # echo 32 > brightness
> 
> # And finally with brightness=0 LED is OFF
> hostname status # echo 0 > brightness
> 
> # --- end of snippet ---
> 
> About color presets...
> 
> I believe the optional color preset thing could work like:
> 
>      multi-color-leds {
>          compatible = "pwm-leds";
> 
>          status-led {
>              label = "status";
> 
>              element-red {
>                  pwms = <&ehrpwm0 0 100000 0>;
>              };
>              element-green {
>                  pwms = <&ehrpwm1 0 100000 0>;
>              };
>              element-blue {
>                  pwms = <&ehrpwm1 1 100000 0>;
>              };
> 
>              color-orange = 255,128,0;
>          };
>      };
> 
> Then in user space:
> 
> hostname status # echo "orange" > color
> 
> or with combined brightness:
> 
> hostname status # echo "brightness=255 orange" > color
> 
> But then again -- I am a bit unsure here myself is it a good idea. I 
> need to try out how the dual color element GPIO LED variant works first. 
> My current hunch is that we might not need the color presets as one can 
> now set all color elements in one go. It just changes a bit logic how it 
> is used.
> 
> With dual color element LED with GPIO control it might as well be:
> 
> # for orange:
> echo "brightness=255 red=255 green=128" > color
> # or max value as GPIO led doesn't really care:
> echo "brightness=255 red=255 green=255" > color
> 
> # for red:
> echo "brightness=255 red=255 green=0" > color
> 
> # for green:
> echo "brightness=255 red=0 green=255" > color
> 
> # for off:
> echo "brightness=0" > color
> # or
> echo 0 > brightness
> 
> Thanks,
> Vesa Jääskeläinen
>