Skip to contentSkip to conversation

Clarification regarding multicolor leds

DISCUSSION Luca Weiss 6 participants 8 messages Oct 23, 2020 – Oct 26, 2020 LED

lore
Share & tools
lore

Luca Weiss

author
#0 lore raw

Hi,
I'm currently experimenting with the qcom lpg[0] which is a driver for the rgb notification led found on e.g. Snapdragon 801 devices (and many more), specifically my example is about the Fairphone 2 (msm8974-fairphone-fp2).

[0] https://lore.kernel.org/lkml/[email protected]/

My dts is looking like the following (abbreviated):

    [in lpg node]
    multi-led {
        color = <LED_COLOR_ID_MULTI>;
        function = LED_FUNCTION_STATUS;
        ....
    };

I'm comparing this to the PinePhone where the leds are defined as follows:

    [in gpio-leds node]
    blue {
        function = LED_FUNCTION_INDICATOR;
        color = <LED_COLOR_ID_BLUE>;
    };
    
    green {
        function = LED_FUNCTION_INDICATOR;
        color = <LED_COLOR_ID_GREEN>;
    };
    
    red {
        function = LED_FUNCTION_INDICATOR;
        color = <LED_COLOR_ID_RED>;
    };

(sidenote: the LED_FUNCTION_INDICATOR should probably also be LED_FUNCTION_STATUS there; the dts was made before the better descriptions for the defines have been added)

This gets the following directories created in /sys/class/leds/:

blue:indicator
green:indicator
red:indicator

But with the multicolor led on the Fairphone 2 only a directory with the name of "multi-led" gets created which I would have expected to be "multicolor:status" instead.

What's further confusing me is that include/dt-bindings/leds/common.h states (reformatted for clarity):

/* For multicolor LEDs */
#define LED_COLOR_ID_MULTI 8

/* For multicolor LEDs that can do arbitrary color, so this would include RGBW and similar */
#define LED_COLOR_ID_RGB 9

It sounds like these comments are the wrong way around as it says that RGB does RGBW & others while MULTI is normal RGB?

I have also found this commit[1] while browsing through lkml which seems to validate my suspicions that _ID_RGB should be used normally? This commit seems have been applied early September but hasn't been merged in the 5.10 merge window?

[1] https://git.kernel.org/pub/scm/linux/kernel/git/pavel/linux-leds.git/
commit/?h=for-next&id=3d93edc77515c6f51fa9bbbe2185e2ec32bad024

But drivers/leds/led-core.c also states "We want to label LEDs that can produce full range of colors as RGB, not multicolor" - not sure what "full range" means here.

Thanks for reading through my long email and I'd appreciate any clarification on the situation!
Regards,
Luca

#1 lore raw

Hello Luca,

Quoted 3 lines · Luca Weiss

I'm currently experimenting with the qcom lpg[0] which is a driver for the rgb notification led found on e.g. Snapdragon 801 devices (and many more), specifically my example is about the Fairphone 2 (msm8974-fairphone-fp2).

Great to hear someone is interested in mainline support for Fairphone. I just bought a used FP2 on ebay. :-)

Quoted 34 lines

[0] https://lore.kernel.org/lkml/[email protected]/

My dts is looking like the following (abbreviated):

    [in lpg node]
    multi-led {
        color = <LED_COLOR_ID_MULTI>;
        function = LED_FUNCTION_STATUS;
        ....
    };

I'm comparing this to the PinePhone where the leds are defined as follows:

    [in gpio-leds node]
    blue {
        function = LED_FUNCTION_INDICATOR;
        color = <LED_COLOR_ID_BLUE>;
    };
    
    green {
        function = LED_FUNCTION_INDICATOR;
        color = <LED_COLOR_ID_GREEN>;
    };
    
    red {
        function = LED_FUNCTION_INDICATOR;
        color = <LED_COLOR_ID_RED>;
    };

(sidenote: the LED_FUNCTION_INDICATOR should probably also be LED_FUNCTION_STATUS there; the dts was made before the better descriptions for the defines have been added)

Quoted 5 lines

This gets the following directories created in /sys/class/leds/:

blue:indicator
green:indicator
red:indicator

That's right. From Linux point of view these behave like three independent LEDs. It's fully up to userspace to handle this.

Quoted 4 lines

But with the multicolor led on the Fairphone 2 only a directory with the name of "multi-led" gets created which I would have expected to be "multicolor:status" instead.

Obviously it's named after the node label. If I read Documentation/devicetree/bindings/leds/leds-class-multicolor.yaml correctly, that's how it is supposed to be named?

Quoted 23 lines

What's further confusing me is that include/dt-bindings/leds/common.h states (reformatted for clarity):

/* For multicolor LEDs */
#define LED_COLOR_ID_MULTI 8

/* For multicolor LEDs that can do arbitrary color, so this would include RGBW and similar */
#define LED_COLOR_ID_RGB 9

It sounds like these comments are the wrong way around as it says that RGB does RGBW & others while MULTI is normal RGB?

I have also found this commit[1] while browsing through lkml which seems to validate my suspicions that _ID_RGB should be used normally? This commit seems have been applied early September but hasn't been merged in the 5.10 merge window?

[1] https://git.kernel.org/pub/scm/linux/kernel/git/pavel/linux-leds.git/
commit/?h=for-next&id=3d93edc77515c6f51fa9bbbe2185e2ec32bad024

Quoted 6 lines

But drivers/leds/led-core.c also states "We want to label LEDs that can produce full range of colors as RGB, not multicolor" - not sure what "full range" means here.

Thanks for reading through my long email and I'd appreciate any clarification on the situation!

I just read
https://www.kernel.org/doc/html/latest/leds/leds-class-multicolor.html
and apart from formatting and inter-doc-link issues due to not used markup features, I can understand how it's supposed to be used from userspace.

However, multicolor is still quite new, maybe drivers/leds/TODO gives some hints? It seems to me some open issues are already known. ;-)

HTH & Greets
Alex

Signature · 4 lines
/"\ ASCII RIBBON | »With the first link, the chain is forged. The first
\ / CAMPAIGN     | speech censured, the first thought forbidden, the
 X  AGAINST      | first freedom denied, chains us all irrevocably.«
/ \ HTML MAIL    | (Jean-Luc Picard, quoting Judge Aaron Satie)
#2 lore raw

Hi!

Quoted 7 lines · Luca Weiss
Quoted 3 lines

I'm currently experimenting with the qcom lpg[0] which is a driver for the rgb notification led found on e.g. Snapdragon 801 devices (and many more), specifically my example is about the Fairphone 2 (msm8974-fairphone-fp2).

Great to hear someone is interested in mainline support for Fairphone. I just bought a used FP2 on ebay. :-)

You may want to cc [email protected].

And yes, I'd like to see usable phone running mainline. I currently have Droid 4, which is close.

Quoted 10 lines
Quoted 10 lines

[0] https://lore.kernel.org/lkml/[email protected]/

My dts is looking like the following (abbreviated):

    [in lpg node]
    multi-led {
        color = <LED_COLOR_ID_MULTI>;
        function = LED_FUNCTION_STATUS;
        ....
    };

Make it _ID_RGB.

Quoted 4 lines
Quoted 4 lines

(sidenote: the LED_FUNCTION_INDICATOR should probably also be LED_FUNCTION_STATUS there; the dts was made before the better descriptions for the defines have been added)

Quoted 8 lines
Quoted 5 lines

This gets the following directories created in /sys/class/leds/:

blue:indicator
green:indicator
red:indicator

That's right. From Linux point of view these behave like three independent LEDs. It's fully up to userspace to handle this.

Yes. Eventually we'll want to switch Pinephone to newer interface.

Quoted 6 lines
Quoted 6 lines

What's further confusing me is that include/dt-bindings/leds/common.h states (reformatted for clarity):

/* For multicolor LEDs */
#define LED_COLOR_ID_MULTI 8

Quoted 6 lines
Quoted 6 lines

/* For multicolor LEDs that can do arbitrary color, so this would include RGBW and similar */
#define LED_COLOR_ID_RGB 9

It sounds like these comments are the wrong way around as it says that RGB does RGBW & others while MULTI is normal RGB?

Multi is for strange stuff like Blue-UV combination. Avoid.

Best regards,
Pavel

Signature · 1 line
http://www.livejournal.com/~pavelmachek
author
#3 lore raw
In reply to Alexander Dahl

Hi Alex,

Quoted 50 lines · On Samstag, 24. Oktober 2020 08:42:38 CEST Alexander Dahl wrote:

Hello Luca,

Quoted 4 lines · Luca Weiss

I'm currently experimenting with the qcom lpg[0] which is a driver for the rgb notification led found on e.g. Snapdragon 801 devices (and many more), specifically my example is about the Fairphone 2 (msm8974-fairphone-fp2).

Great to hear someone is interested in mainline support for Fairphone. I just bought a used FP2 on ebay. :-)

Quoted 40 lines

[0]
https://lore.kernel.org/lkml/20201021201224.3430546-1-bjorn.andersson@lin
aro.org/>
My dts is looking like the following (abbreviated):

    [in lpg node]
    multi-led {
    
        color = <LED_COLOR_ID_MULTI>;
        function = LED_FUNCTION_STATUS;
        ....
    
    };

I'm comparing this to the PinePhone where the leds are defined as follows:

    [in gpio-leds node]
    blue {
    
        function = LED_FUNCTION_INDICATOR;
        color = <LED_COLOR_ID_BLUE>;
    
    };
    
    green {
    
        function = LED_FUNCTION_INDICATOR;
        color = <LED_COLOR_ID_GREEN>;
    
    };
    
    red {
    
        function = LED_FUNCTION_INDICATOR;
        color = <LED_COLOR_ID_RED>;
    
    };

(sidenote: the LED_FUNCTION_INDICATOR should probably also be LED_FUNCTION_STATUS there; the dts was made before the better descriptions for the defines have been added)

Quoted 7 lines
Quoted 4 lines

This gets the following directories created in /sys/class/leds/:
blue:indicator
green:indicator
red:indicator

That's right. From Linux point of view these behave like three independent LEDs. It's fully up to userspace to handle this.

Exactly, that I understand well.

Quoted 9 lines
Quoted 4 lines

But with the multicolor led on the Fairphone 2 only a directory with the name of "multi-led" gets created which I would have expected to be "multicolor:status" instead.

Obviously it's named after the node label. If I read Documentation/devicetree/bindings/leds/leds-class-multicolor.yaml correctly, that's how it is supposed to be named?

That's from the i read from the documentation as well.

How is user space supposed to get function and/or color from the led? I don't see it exposed in user space - apart from the directory name (label) in the leds-gpio example.

This is what I get in sysfs for the multicolor led with the lpg driver:

    brightness
    device -> ../../../fc4cf000.spmi:pm8941@1:lpg
    max_brightness
    multi_index
    multi_intensity
    power
    subsystem -> ../../../../../../../../../class/leds
    trigger
    uevent
Quoted 6 lines
Quoted 6 lines

What's further confusing me is that include/dt-bindings/leds/common.h states (reformatted for clarity):

/* For multicolor LEDs */
#define LED_COLOR_ID_MULTI 8

Quoted 7 lines
Quoted 7 lines

/* For multicolor LEDs that can do arbitrary color, so this would include RGBW and similar */
#define LED_COLOR_ID_RGB 9

It sounds like these comments are the wrong way around as it says that RGB does RGBW & others while MULTI is normal RGB?

Based on Pavel's reply my RGB LED should be _ID_RGB.
It would be great if this comment would be clarified because honestly still don't know what "arbitrary color" should mean; and add an example to the _ID_MULTI one that it only should be used for "strange stuff like Blue-UV combination".

Quoted 13 lines
Quoted 13 lines

I have also found this commit[1] while browsing through lkml which seems to
validate my suspicions that _ID_RGB should be used normally? This commit seems have been applied early September but hasn't been merged in the 5.10 merge window?

[1] https://git.kernel.org/pub/scm/linux/kernel/git/pavel/linux-leds.git/
commit/?h=for-next&id=3d93edc77515c6f51fa9bbbe2185e2ec32bad024

But drivers/leds/led-core.c also states "We want to label LEDs that can produce full range of colors as RGB, not multicolor" - not sure what "full range" means here.

Quoted 9 lines
Quoted 2 lines

Thanks for reading through my long email and I'd appreciate any clarification on the situation!

I just read
https://www.kernel.org/doc/html/latest/leds/leds-class-multicolor.html
and apart from formatting and inter-doc-link issues due to not used markup features, I can understand how it's supposed to be used from userspace.

Even in the doc the directory name is "multicolor:status" exposing color & function..

Curiously when setting the (deprecated) property "label" to "rgb:status" the directory name in sysfs becomes rgb:status - so is it a bug that this doesn't happen automatically? Feels like it at least.

I've looked into the code yesterday a bit and found that "struct led_init_data *init_data" in led_classdev_register_ext is always passed as NULL for multicolor leds from devm_led_classdev_multicolor_register which seems to cause this to happen.

Quoted 5 lines

However, multicolor is still quite new, maybe drivers/leds/TODO gives some hints? It seems to me some open issues are already known. ;-)

HTH & Greets
Alex

Regards
Luca

#4 lore raw

Luca

Quoted 74 lines · Luca Weiss

Hi Alex,

Quoted 50 lines · On Samstag, 24. Oktober 2020 08:42:38 CEST Alexander Dahl wrote:

Hello Luca,

Quoted 4 lines · Luca Weiss

I'm currently experimenting with the qcom lpg[0] which is a driver for the rgb notification led found on e.g. Snapdragon 801 devices (and many more), specifically my example is about the Fairphone 2 (msm8974-fairphone-fp2).

Great to hear someone is interested in mainline support for Fairphone. I just bought a used FP2 on ebay. :-)

Quoted 40 lines

[0]
https://lore.kernel.org/lkml/20201021201224.3430546-1-bjorn.andersson@lin
aro.org/>
My dts is looking like the following (abbreviated):

     [in lpg node]
     multi-led {
     
         color = <LED_COLOR_ID_MULTI>;
         function = LED_FUNCTION_STATUS;
         ....
     
     };

I'm comparing this to the PinePhone where the leds are defined as follows:

     [in gpio-leds node]
     blue {
     
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_BLUE>;
     
     };
     
     green {
     
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_GREEN>;
     
     };
     
     red {
     
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_RED>;
     
     };

(sidenote: the LED_FUNCTION_INDICATOR should probably also be LED_FUNCTION_STATUS there; the dts was made before the better descriptions for the defines have been added)

Quoted 6 lines
Quoted 4 lines

This gets the following directories created in /sys/class/leds/:
blue:indicator
green:indicator
red:indicator

That's right. From Linux point of view these behave like three independent LEDs. It's fully up to userspace to handle this.

Exactly, that I understand well.

Quoted 7 lines
Quoted 3 lines

But with the multicolor led on the Fairphone 2 only a directory with the name of "multi-led" gets created which I would have expected to be "multicolor:status" instead.

Obviously it's named after the node label. If I read Documentation/devicetree/bindings/leds/leds-class-multicolor.yaml correctly, that's how it is supposed to be named?

That's from the i read from the documentation as well.

How is user space supposed to get function and/or color from the led? I don't see it exposed in user space - apart from the directory name (label) in the leds-gpio example.

Quoted 6 lines

This is what I get in sysfs for the multicolor led with the lpg driver:

     brightness
     device -> ../../../fc4cf000.spmi:pm8941@1:lpg
     max_brightness
     multi_index

If you refer to the ABI doc you see that by reading the multi_index file gives you the colors contained within the file.

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/ABI/testing/sysfs-class-led-multicolor

Dan

#5 lore raw
In reply to Luca Weiss

Hi Luca,

Quoted 40 lines · Luca Weiss

Hi,
I'm currently experimenting with the qcom lpg[0] which is a driver for the rgb notification led found on e.g. Snapdragon 801 devices (and many more), specifically my example is about the Fairphone 2 (msm8974-fairphone-fp2).

[0] https://lore.kernel.org/lkml/[email protected]/

My dts is looking like the following (abbreviated):

     [in lpg node]
     multi-led {
         color = <LED_COLOR_ID_MULTI>;
         function = LED_FUNCTION_STATUS;
         ....
     };

I'm comparing this to the PinePhone where the leds are defined as follows:

     [in gpio-leds node]
     blue {
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_BLUE>;
     };
     
     green {
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_GREEN>;
     };
     
     red {
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_RED>;
     };

(sidenote: the LED_FUNCTION_INDICATOR should probably also be LED_FUNCTION_STATUS there; the dts was made before the better descriptions for the defines have been added)

This gets the following directories created in /sys/class/leds/:

Quoted 7 lines

blue:indicator
green:indicator
red:indicator

But with the multicolor led on the Fairphone 2 only a directory with the name of "multi-led" gets created which I would have expected to be "multicolor:status" instead.

This is because the driver from patch set [0] does not use *ext() multicolor registration API, but follows old-fashion LED name initialization via 'name' property of struct led_classdev, which is initialized to DT 'label' value or DT node name if the former is absent.

Signature · 2 lines
Best regards,
Jacek Anaszewski
author
#6 lore raw

Hi Jacek

Quoted 54 lines · On Samstag, 24. Oktober 2020 21:34:38 CET Jacek Anaszewski wrote:

Hi Luca,

Quoted 51 lines · Luca Weiss

Hi,
I'm currently experimenting with the qcom lpg[0] which is a driver for the rgb notification led found on e.g. Snapdragon 801 devices (and many more), specifically my example is about the Fairphone 2 (msm8974-fairphone-fp2).

[0]
https://lore.kernel.org/lkml/20201021201224.3430546-1-bjorn.andersson@lin
aro.org/>
My dts is looking like the following (abbreviated):

     [in lpg node]
     multi-led {
     
         color = <LED_COLOR_ID_MULTI>;
         function = LED_FUNCTION_STATUS;
         ....
     
     };

I'm comparing this to the PinePhone where the leds are defined as follows:

     [in gpio-leds node]
     blue {
     
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_BLUE>;
     
     };
     
     green {
     
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_GREEN>;
     
     };
     
     red {
     
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_RED>;
     
     };

(sidenote: the LED_FUNCTION_INDICATOR should probably also be LED_FUNCTION_STATUS there; the dts was made before the better descriptions for the defines have been added)

This gets the following directories created in /sys/class/leds/:
blue:indicator
green:indicator
red:indicator

Quoted 8 lines
Quoted 3 lines

But with the multicolor led on the Fairphone 2 only a directory with the name of "multi-led" gets created which I would have expected to be "multicolor:status" instead.

This is because the driver from patch set [0] does not use *ext() multicolor registration API, but follows old-fashion LED name initialization via 'name' property of struct led_classdev, which is initialized to DT 'label' value or DT node name if the former is absent.

Right, I remember now having done something similar with my sgm3140 flash led driver half a year ago, thanks for the reminder!
With the qcom-lpg driver using the _ext function it works as expected.

@Bjorn I've prepared a diff for your patches at
https://public.z3ntu.xyz/tmp/lpg.diff where
1. LED_COLOR_ID_MULTI is replaced with LED_COLOR_ID_RGB based on these emails
2. The example in the documentation uses LED_FUNCTION_STATUS (see description for the constant in include/dt-bindings/leds/common.h)
3. The _ext registration api now gets used, see e.g. commit 5b4b723c483f

Thanks for your help, really appreciate it!
Regards
Luca

#7 lore raw
In reply to Jacek Anaszewski
Quoted 45 lines · Jacek Anaszewski

Hi Luca,

Quoted 42 lines · Luca Weiss

Hi,
I'm currently experimenting with the qcom lpg[0] which is a driver for the rgb notification led found on e.g. Snapdragon 801 devices (and many more), specifically my example is about the Fairphone 2 (msm8974-fairphone-fp2).

[0] https://lore.kernel.org/lkml/[email protected]/

My dts is looking like the following (abbreviated):

     [in lpg node]
     multi-led {
         color = <LED_COLOR_ID_MULTI>;
         function = LED_FUNCTION_STATUS;
         ....
     };

I'm comparing this to the PinePhone where the leds are defined as follows:

     [in gpio-leds node]
     blue {
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_BLUE>;
     };
     green {
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_GREEN>;
     };
     red {
         function = LED_FUNCTION_INDICATOR;
         color = <LED_COLOR_ID_RED>;
     };

(sidenote: the LED_FUNCTION_INDICATOR should probably also be LED_FUNCTION_STATUS there; the dts was made before the better descriptions for the defines have been added)

This gets the following directories created in /sys/class/leds/:

blue:indicator
green:indicator
red:indicator

Quoted 9 lines
Quoted 3 lines

But with the multicolor led on the Fairphone 2 only a directory with the name of "multi-led" gets created which I would have expected to be "multicolor:status" instead.

This is because the driver from patch set [0] does not use *ext() multicolor registration API, but follows old-fashion LED name initialization via 'name' property of struct led_classdev, which is initialized to DT 'label' value or DT node name if the former is absent.

Sorry, I had missed this advancement. I will update the LPG patches to add this and to switch to expect LED_COLOR_ID_RGB instead of multi.

Thanks,
Bjorn