Re: Should regulator core support parsing OF based fwnode? - Jean-Jacques Hiblot

From: Jean-Jacques Hiblot <jjhiblot@ti.com>
To: Mark Brown <broonie@kernel.org>,
	Jacek Anaszewski <jacek.anaszewski@gmail.com>
Cc: Sebastian Reichel <sebastian.reichel@collabora.com>,
	<pavel@ucw.cz>, <robh+dt@kernel.org>, <mark.rutland@arm.com>,
	<lee.jones@linaro.org>, <daniel.thompson@linaro.org>,
	<linux-kernel@vger.kernel.org>, <dri-devel@lists.freedesktop.org>,
	<tomi.valkeinen@ti.com>, <dmurphy@ti.com>,
	<linux-leds@vger.kernel.org>, Liam Girdwood <lgirdwood@gmail.com>
Subject: Re: Should regulator core support parsing OF based fwnode?
Date: Fri, 4 Oct 2019 15:33:13 +0200	[thread overview]
Message-ID: <b6318ba5-e76e-dc1c-6921-a702abf6749c@ti.com> (raw)
In-Reply-To: <20191004113942.GB4866@sirena.co.uk>

On 04/10/2019 13:39, Mark Brown wrote:
> On Thu, Oct 03, 2019 at 10:27:26PM +0200, Jacek Anaszewski wrote:
>> On 10/3/19 9:41 PM, Mark Brown wrote:
>>> Why would we want to do that?  We'd continue to support only DT systems,
>>> just with code that's less obviously DT only and would need to put
>>> checks in.  I'm not seeing an upside here.
>> For instance few weeks ago we had a patch [0] in the LED core switching
>> from using struct device's of_node property to fwnode for conveying
>> device property data. And this transition to fwnode property API can be
>> observed as a frequent pattern across subsystems.
> For most subsystems the intent is to reuse DT bindings on embedded ACPI
> systems via _DSD.
>
>> Recently there is an ongoing effort aiming to add generic support for
>> handling regulators in the LED core [1], but it turns out to require
>> bringing back initialization of of_node property for
>> devm_regulator_get_optional() to work properly.
> Consumers should just be able to request a regulator without having to
> worry about how that's being provided - they should have no knowledge at
> all of firmware bindings or platform data for defining this.  If they
> do that suggests there's an abstraction issue somewhere, what makes you
> think that doing something with of_node is required?

The regulator core accesses consumer->of_node to get a phandle to a 
regulator's node. The trouble arises from the fact that the LED core 
does not populate of_node anymore, instead it populates fwnode. This 
allows the LED core to be agnostic of ACPI or OF to get the properties 
of a LED.

IMO it is better to populate both of_node and fwnode in the LED core at 
the moment. It has already been fixed this way for the platform driver 
[0], MTD [1] and PCI-OF [2].

>
> Further, unless you have LEDs that work without power you probably
> shouldn't be using _get_optional() for their supply.  That interface is
> intended only for supplies that may be physically absent.

Not all LEDs have a regulator to provide the power. The power can be 
supplied by the LED controller for example.

[0] f94277af03ead0d3bf2 of/platform: Initialise dev->fwnode appropriately

[1] c176c6d7e932662668 mfd: core: Set fwnode for created devices

[2] 9b099a6c75e4ddceea PCI: OF: Initialize dev->fwnode appropriately