Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:6 as permitted sender) client-ip=2620:137:e000::3:6;
Message-ID: <b8df5f0c-a3cd-4cad-b1c6-db89686464fc@rasmusvillemoes.dk>
Date:   Thu, 23 Nov 2023 14:48:52 +0100
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH 1/2] dt-bindings: serial: rs485: add rs485-mux-gpios
 binding
Content-Language: en-US, da
To:     Lukas Wunner <lukas@wunner.de>
Cc:     Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Jiri Slaby <jirislaby@kernel.org>,
        Rob Herring <robh+dt@kernel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>,
        Conor Dooley <conor+dt@kernel.org>, devicetree@vger.kernel.org,
        Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
        linux-kernel@vger.kernel.org, linux-serial@vger.kernel.org,
        Crescent CY Hsieh <crescentcy.hsieh@moxa.com>,
        Lino Sanfilippo <LinoSanfilippo@gmx.de>,
        =?UTF-8?Q?Ilpo_J=C3=A4rvinen?= <ilpo.jarvinen@linux.intel.com>,
        Peter Rosin <peda@axentia.se>
References: <20231120151056.148450-1-linux@rasmusvillemoes.dk>
 <20231120151056.148450-2-linux@rasmusvillemoes.dk>
 <20231122145344.GA18949@wunner.de>
 <e731c0a9-7a5c-41c3-87aa-d6937b99d01a@rasmusvillemoes.dk>
 <20231123103802.GA30056@wunner.de>
From:   Rasmus Villemoes <linux@rasmusvillemoes.dk>
In-Reply-To: <20231123103802.GA30056@wunner.de>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 23/11/2023 11.38, Lukas Wunner wrote:
> On Thu, Nov 23, 2023 at 11:07:16AM +0100, Rasmus Villemoes wrote:
>> On 22/11/2023 15.53, Lukas Wunner wrote:
>>> But if that patch gets accepted, we'd have *three* different modes:
>>> RS-232, RS-485, RS-422.  A single GPIO seems insufficient to handle that.
>>> You'd need at least two GPIOs.
>>
>> I don't see Crescent introducing any new gpio that needs to be handled.
>> In fact, I can't even see why from the perspective of the software that
>> rs422 isn't just rs232; there's no transmit enable pin that needs to be
>> handled. But maybe the uart driver does something different in rs422
>> mode; I assume he must have some update of some driver, since otherwise
>> the new rs422 bit should be rejected by the core. So I can't really see
>> the whole picture of that rs422 story.
> 
> The question is, could we conceivably have the need to support
> switching between the three modes RS-232, RS-485, RS-422.
> If yes, then the GPIO mux interface should probably allow for that.
> 
> As a case in point, the Siemens IOT 2040 has two serial ports
> which can be set to either of those three modes.  The signals
> are routed to the same D-sub socket, but the pins used are
> different.  See page 46 and 47 of this document:
> 
> https://cache.industry.siemens.com/dl/files/658/109741658/att_899623/v1/iot2000_operating_instructions_enUS_en-US.pdf
> 
> The driver for this product is 8250_exar.c.  It's an Intel-based
> product, so no devicetree, but it shows that such use cases exist.

OK. I did look at the mux-controller/mux-consumer bindings, but couldn't
really make heads or tails of it, and there aren't a whole lot of
examples in-tree. Also, the C API seems ... not quite what is needed
here. I realize that's not really anything to do with the best way to
describe the hardware, but that, plus the fact that the serial core
already handles a number of gpios controlling circuitry related to
rs485, was what made me go for one extra gpio.

How would a mux-consumer description look?

  mux-states = <&mux 0>, <&mux 1>;
  mux-state-names = "rs485", "rs232";

or should that be mux-controls? Would that be enough so that we're sure
that if and when a rs422 state is needed that could easily be
represented here?

Now implementation-wise, there's the complication that switching the mux
to/from rs485 mode must be done after/before the driver's ->rs485_config
is called, to avoid the transceiver temporarily being activated (thus
blocking/disturbing other traffic). That plus the need to mux_*_deselect
the old mode means the consumer (serial core in this case) ends up with
quite a lot of bookkeeping, and even more so taking error path into
consideration.

Rasmus