Message-ID: <8c5d2d85-24ac-2ff9-8512-f669063edf4c@gmail.com>
Date:   Mon, 17 Jul 2017 23:08:46 +0200
From:   Jacek Anaszewski <jacek.anaszewski@...il.com>
To:     Bjorn Andersson <bjorn.andersson@...aro.org>
Cc:     Richard Purdie <rpurdie@...ys.net>, Pavel Machek <pavel@....cz>,
        Rob Herring <robh+dt@...nel.org>,
        Mark Rutland <mark.rutland@....com>,
        linux-kernel@...r.kernel.org, linux-leds@...r.kernel.org,
        linux-arm-msm@...r.kernel.org, devicetree@...r.kernel.org,
        Fenglin Wu <fenglinw@...eaurora.org>
Subject: Re: [PATCH v2 3/3] DT: leds: Add Qualcomm Light Pulse Generator
 binding

On 07/17/2017 06:44 AM, Bjorn Andersson wrote:
> On Sun 16 Jul 11:49 PDT 2017, Jacek Anaszewski wrote:
>> On 07/15/2017 12:45 AM, Bjorn Andersson wrote:
>>> diff --git a/Documentation/devicetree/bindings/leds/leds-qcom-lpg.txt b/Documentation/devicetree/bindings/leds/leds-qcom-lpg.txt
>>> new file mode 100644
>>> index 000000000000..cc9ffee6586b
>>> --- /dev/null
>>> +++ b/Documentation/devicetree/bindings/leds/leds-qcom-lpg.txt
>>> @@ -0,0 +1,145 @@
>>> +Binding for Qualcomm Light Pulse Generator
>>> +
>>> +The Qualcomm Light Pulse Generator consists of three different hardware blocks;
>>
>> Is there a freely available documentation thereof?
>>
> 
> The only publicly available Qualcomm PMIC documentation that I'm aware
> of only have the PWM hardware block, so it will be possible to use this
> driver but with limited functionality.

I asked because having an access to the doc would speed up the
contribution process a lot I think. Of course we will manage to make it
also basing on the details you're providing us with, but it will take
a bit longer, taking into account the device complexity.

> [..]
>>> += Light Pulse Generator (LPG)
>>> +The Light Pulse Generator can operate either as a standard PWM controller or in
>>> +a more advanced lookup-table based mode. These are described separately below.
>>
>> Why a user would prefer one option over the other? I assume that both
>> controllers offer at least slightly different capabilities.
>> If so, then it could be the driver which would decide which one fits
>> better for the requested LED class device configuration.
>>
> 
> I have never seen this hardware block been used as a PWM, but I imagine
> it to be used when someone has another driver that expects to be able to
> use the PWM API to control an output.

We have already leds-pwm driver and related DT bindings. We could think
of making some part of leds-pwm code reusable. I'd skip it for now
though.

> In this case the node would need a #pwm-cells property, which it doesn't
> when it's acting as a LED and it wouldn't make much sense to expose the
> pin as a LED at the same time.
> 
> Perhaps I overthought this? Maybe I should just leave the PWM mode out
> for now and it could be added in the future?

Sounds reasonable.

> [..]
>>> +&spmi_bus {
>>> +	pmic@3 {
>>> +		compatible = "qcom,pmi8994", "qcom,spmi-pmic";
>>> +		reg = <0x3 SPMI_USID>;
>>> +		#address-cells = <1>;
>>> +		#size-cells = <0>;
>>> +
>>> +		pmi8994_lpg_lut: lpg-lut@...0 {
>>> +			compatible = "qcom,spmi-lpg-lut";
>>> +			reg = <0xb000>;
>>> +
>>> +			qcom,lut-size = <24>;
>>> +		};
>>> +
>>> +		lpg@...0 {
>>> +			compatible = "qcom,spmi-lpg";
>>> +			reg = <0xb200>;
>>> +
>>> +			cell-index = <2>;
>>> +
>>> +			label = "lpg:green:user0";
>>> +
>>> +			qcom,tri-led = <&pmi8994_tri_led 1>;
>>> +			qcom,lut = <&pmi8994_lpg_lut>;
>>> +
>>> +			default-state = "on";
>>> +		};
>>> +
>>> +		pmi8994_tri_led: tri-led@...0 {
>>> +			compatible = "qcom,spmi-tri-led";
>>> +			reg = <0xd000>;
>>> +
>>> +			qcom,power-source = <1>;
>>> +		};
>>
>> Such a design is uncommon for LED class DT bindings. It should
>> suffice to have a single DT LED node per LED. I have an impression
>> that you're exposing too many hardware details here.
>> You can use led-sources property (see Documentation/devicetree/bindings
>> /leds/common.txt and drivers/leds/leds-max77693.c where it is used).
>>
> 
> The LUT is shared among the (up to) 8 LPG blocks, so while I did
> consider just including the LUT in each LPG block it didn't look nice
> and I had to implement the LUT as a singleton in the driver itself.
> 
> The TRILED is only one of the available current sinks in the PMIC that
> can be driven by the LPG; the other one I use so far is a special GPIO
> pin acting as a current sink.
> 
> Also the power-source configuration is shared among the three channels
> of the TRILED, so it doesn't really make sense to put this configuration
> in the LPG blocks themselves.

I'll mention led-sources once again. Probably it could be of help here.

> 
> And note that these are different blocks within the Qualcomm PMIC, with
> my design only one of them is actually representing the LED instance.

Maybe the core of the driver should be placed in MFD subsystem then?

>> It is also not clear to me why single green color LED presented here
>> would have to use tri-led sink? I suppose that the sink is predestined
>> for three-color LEDs.
>>
> 
> The board I'm working on (DragonBoard820c) has 4 green LEDs, the first 3
> are connected to the 3 channels of the TRILED and the fourth is
> connected to a special GPIO in current sink mode. But I choose to
> shorted the example to one channel.
> 
> So I end up having one LUT node, four LPG nodes, one TRILED and one GPIO
> node and the user space is presented with a unified interface to all
> four.

Generally I'd prefer to have a single LED class driver for this device,
or alternatively a LED class driver for a LED cell of MFD device.
DT bindings would define which hw blocks are LED related.
All routing related issues should be solvable with use of led-sources
property.

-- 
Best regards,
Jacek Anaszewski

Powered by blists - more mailing lists