Re: [PATCH] serial: 8250: 8250_omap: Support native RS485

[Lukas Wunner <lukas@wunner.de>]

On Thu, Oct 06, 2022 at 12:37:43PM +0530, Vignesh Raghavendra wrote:
> On 04/10/22 7:15 pm, Bin Liu wrote:
> > On Mon, Oct 03, 2022 at 09:42:24PM +0200, Lukas Wunner wrote:
> >> On Mon, Oct 03, 2022 at 10:10:59AM -0500, Bin Liu wrote:
> >>> On Tue, Sep 27, 2022 at 02:10:01PM +0200, Lukas Wunner wrote:
> >>>> Recent TI Sitara SoCs such as AM64/AM65 have gained the ability to
> >>>> automatically assert RTS when data is transmitted, obviating the need
> >>>> to emulate this functionality in software.
> >>>>
> >>>> The feature is controlled through new DIR_EN and DIR_POL bits in the
> >>>> Mode Definition Register 3.  For details see page 8783 and 8890 of the
> >>>> AM65 TRM:  https://www.ti.com/lit/ug/spruid7e/spruid7e.pdf
> >> [...]
> >>>> -	if (up->port.rs485.flags & SER_RS485_ENABLED)
> >>>> +	if (priv->habit & UART_HAS_NATIVE_RS485)
> >>>> +		serial_out(up, UART_OMAP_MDR3, priv->mdr3);
> >>>
> >>> This makes the NATIVE_RS485 always used if the SoC supports it, but
> >>>
> >>>> +	else if (up->port.rs485.flags & SER_RS485_ENABLED)
> >>>>  		serial8250_em485_stop_tx(up);
> >>>
> >>> there are cases em485 should be used even if SoC supports NATIVE_RS485.
> >>> For example:
> >>> - the design has pinmux conflict and the RTS pin has to be used for
> >>>   something else. Then a GPIO pin would be used for RS485 DE control;
> >>> - the design requires customized delay_rts_before_send or
> >>>   delay_rts_after_send which NATIVE_RS485 doesn't support;
> >>>
> >>> So we might need an option for such usecases. A device tree flag?
> >>
> >> I expect those cases to be rare, hence do not see the need to
> > 
> > Maybe rare, but I know some projects use GPIO for DE control.
> > 
> >> address them right from the start.  Support for falling back
> > 
> > So I think missing it is a regression, because this patch forces to
> > use the RTS pin for DE control, this breaks the existing projects which
> > use GPIO for RTS/DE or have customized delay_rts_{before,after}_send.
> 
> I agree with Bin. This patch as such can break DTs which intend to use
> GPIO based DE.
> Quick fix would be to simply look for presence of rts-gpios property in
> DT, if so fallback to registering serial8250_em485_*. This should avoid
> regressing DTs using GPIO for DE control.

I've updated the patch to check for

  mctrl_gpio_to_gpiod(up->gpios, UART_GPIO_RTS)

as well as user-requested delays that exceed the fixed hardware delays.
The driver falls back to software emulation in that case.

Updated patch is on this development branch and is currently being
compile-tested by Intel's 0-day bot:

https://github.com/l1k/linux/commits/rs485_sitara

Thanks,

Lukas