Re: [PATCH 1/2] Input: atmel_mxt_ts: Convert bindings to YAML and extend

From: Dmitry Torokhov <dmitry.torokhov@gmail.com>
To: Linus Walleij <linus.walleij@linaro.org>
Cc: Linux Input <linux-input@vger.kernel.org>,
	Nick Dyer <nick@shmanahar.org>,
	Stephan Gerhold <stephan@gerhold.net>,
	"open list:OPEN FIRMWARE AND FLATTENED DEVICE TREE BINDINGS" 
	<devicetree@vger.kernel.org>
Subject: Re: [PATCH 1/2] Input: atmel_mxt_ts: Convert bindings to YAML and extend
Date: Thu, 29 Oct 2020 09:25:31 -0700	[thread overview]
Message-ID: <20201029162531.GB2547185@dtor-ws> (raw)
In-Reply-To: <CACRpkdaieExkEyjE=+GbQTVKsk21ifH9mm+q4vengqgbQ=Jd_A@mail.gmail.com>

On Thu, Oct 29, 2020 at 02:47:36PM +0100, Linus Walleij wrote:
> On Wed, Oct 28, 2020 at 7:01 PM Dmitry Torokhov
> <dmitry.torokhov@gmail.com> wrote:
> > On Wed, Oct 28, 2020 at 11:17:10AM +0100, Linus Walleij wrote:
> 
> > > This converts the Armel MXT touchscreen bindings to YAML
> > > format and extends them with the following two properties:
> > >
> > > - vdda-supply: the optional analog supply voltage
> > > - vdd-supply: the optional digital supply voltage
> > >
> > > I also explained about the reset-gpios property that this
> > > better be flagged as active high (0) despite actually
> > > being active low, because all current device trees and
> > > drivers assume that this is the case and will actively
> > > drive the line low to assert RESET.
> >
> > I wonder if we should fix that in driver and in DTs instead of doing
> > this cludge...
> 
> Unfortunately I think there are deployed systems with flashed-in
> system descriptions depending on this bug in the system
> description already.
> 
> I am not thinking about device trees now, but instead ACPI
> chromebooks, that have their reset line flagged as whatever
> ACPI or DT-to-ACPI use to indicate an active high line.
> Despite being active low.

The only ARM Chromebook that exposed reset line to the kernel was RK3288
Asus Chromebook "Minnie". DTS specifies correct polarity (active low),
but uses different binding (atmel,reset-gpios) from the driver found
upstream (I have never reconciled Atmel driver we ship with Chromebooks
with the upstream one). DT there is also part of the kernel, not flashed
separately.

x86 Chromebooks do not export reset line or regulators to the kernel but
rather handle power up/down sequence in firmware (either at boot or
exposing ACPI power control methods that kernel invokes form ACPI power
domain code).

> 
> I could fix all the in-tree devicetrees and do it the natural way
> (I have certainly done so before) and then add a quirk if used
> with ACPI. But it's really risky. I'm afraid of regressions here.

Unless there are unofficial firmwares that rework power handling on some
x86 Chromebooks and we want to support them I'd rather we did not quirk.

Thanks.

-- 
Dmitry