Re: [PATCH 2 1/9] serial: core: move RS485 configuration tasks from drivers into core

From: Lukas Wunner <lukas@wunner.de>
To: Lino Sanfilippo <LinoSanfilippo@gmx.de>
Cc: gregkh@linuxfoundation.org, jirislaby@kernel.org,
	u.kleine-koenig@pengutronix.de, linux@armlinux.org.uk,
	richard.genoud@gmail.com, nicolas.ferre@microchip.com,
	alexandre.belloni@bootlin.com, ludovic.desroches@microchip.com,
	shawnguo@kernel.org, s.hauer@pengutronix.de,
	kernel@pengutronix.de, festevam@gmail.com, linux-imx@nxp.com,
	mcoquelin.stm32@gmail.com, alexandre.torgue@foss.st.com,
	linux-serial@vger.kernel.org, linux-kernel@vger.kernel.org,
	linux-arm-kernel@lists.infradead.org,
	linux-stm32@st-md-mailman.stormreply.com
Subject: Re: [PATCH 2 1/9] serial: core: move RS485 configuration tasks from drivers into core
Date: Thu, 17 Feb 2022 12:33:54 +0100	[thread overview]
Message-ID: <20220217113354.GA7826@wunner.de> (raw)
In-Reply-To: <20220216001803.637-2-LinoSanfilippo@gmx.de>

On Wed, Feb 16, 2022 at 01:17:55AM +0100, Lino Sanfilippo wrote:
> --- a/drivers/tty/serial/serial_core.c
> +++ b/drivers/tty/serial/serial_core.c
> @@ -1282,8 +1282,26 @@ static int uart_set_rs485_config(struct uart_port *port,
>  	if (copy_from_user(&rs485, rs485_user, sizeof(*rs485_user)))
>  		return -EFAULT;
>  
> +	/* pick sane settings if the user hasn't */
> +	if (!(rs485.flags & SER_RS485_RTS_ON_SEND) ==
> +	    !(rs485.flags & SER_RS485_RTS_AFTER_SEND)) {
> +		rs485.flags |= SER_RS485_RTS_ON_SEND;
> +		rs485.flags &= ~SER_RS485_RTS_AFTER_SEND;
> +	}

The policy you're enforcing here is:  If settings are nonsensical,
always default to active-high polarity.

However some drivers currently enforce a completely different policy:
E.g. with 8250_lpc18xx.c, if SER_RS485_RTS_ON_SEND is set, use active-high
(and fix up SER_RS485_RTS_AFTER_SEND), else use active-low polarity.
This yields a different result compared to your policy in case both bits
are cleared.

Similarly, sc16is7xx.c defaults to active-low if SER_RS485_RTS_AFTER_SEND
is set, else active-high polarity.  This yields a different result compared
to your policy in case both bits are set.

You risk breaking existing user space applications with this change
if those applications specify nonsensical polarity settings.

I happen to have created a similar commit to this one a month ago
and I came to the conclusion that all one can do is offer a library
function uart_sanitize_rs485_mode() which drivers may elect to call
if that helper's policy is identical to what they're doing now:

https://github.com/l1k/linux/commit/637984111e42

> +
> +	rs485.delay_rts_before_send = min_t(unsigned int,
> +					    rs485.delay_rts_before_send,
> +					    SER_RS485_MAX_RTS_DELAY);
> +	rs485.delay_rts_after_send = min_t(unsigned int,
> +					   rs485.delay_rts_after_send,
> +					   SER_RS485_MAX_RTS_DELAY);

Nonsensical delays may not only be passed in from user space via ioctl(),
but through the device tree.  That's another reason to use a library
function:  It can be called both from the ioctl() as well as after (or in)
uart_get_rs485_mode().

> +	/* Return clean padding area to userspace */
> +	memset(rs485.padding, 0, sizeof(rs485.padding));

Unlike the polarity and delay handling, this one makes sense.

Thanks,

Lukas