Re: Modem control lines for RTSCTS hardware flow control via rts-gpio and cts-gpio with IMX

[Richard Genoud <richard.genoud@gmail.com>]

Le 25/01/2022 à 00:56, Tim Harvey a écrit :
> On Thu, Jan 13, 2022 at 10:19 PM Tomasz Moń <tomasz.mon@camlingroup.com> wrote:
>>
>> On 14.01.2022 04:08, Tim Harvey wrote:
>>> So I believe in order to support using gpios for rts/cts in the imx
>>> uart driver I must find the right place to call imx_uart_rts_active
>>> and imx_uart_rts_inactive when the FIFO is not full and full
>>> respectively. I'm not that familiar with the Linux uart driver
>>> framework - am I on the right track and if so any ideas where this is
>>> best done?
>>
>> It is not really the driver (and thus FIFO level), but rather the amount
>> of free space in tty buffer (checked by Line Discipline workqueue) that
>> determines when to throttle (set RTS inactive). This mostly works fine,
>> but fails [1] when the RX interrupt frequency is too high [2].
>>
>> The throttle/unthrottle request, when termios CRTSCTS is set, is seen by
>> the driver as the call to .set_mctrl (imx_uart_set_mctrl) with TIOCM_RTS
>> bit cleared/set in mctrl parameter. Currently imx_uart_set_mctrl() only
>> controls the UCR2_CTS and UCR2_CTSC bits based on mctrl.
>>
>> To support your case you would most likely have to add the gpio handling
>> in imx_uart_set_mctrl(). However, I am unaware what other issues you
>> might encounter (i.e. if it is not done there yet simply because nobody
>> had that use case or if there is some deeper problem).
>>
>> [1] https://lore.kernel.org/linux-serial/10e723c0-a28b-de0d-0632-0bd250478313@camlingroup.com/
>> [2] https://lore.kernel.org/linux-serial/20220104103203.2033673-1-tomasz.mon@camlingroup.com/
>>
>> Best Regards,
>> Tomasz Mon
>>
> 
> Tomasz,
> 
> Thanks for the info. Currently imx_uart_set_mctrl calls mctrl_gpio_set
> which does toggle the rts-gpio per mctrl & TIOCM_RTS.
> 
> Also, there's something in
> Documentation/devicetree/bindings/serial/serial.yaml that puzzles me:
> 
> if:
>    required:
>      - uart-has-rtscts
> then:
>    properties:
>      cts-gpios: false
>      rts-gpios: false
> 
> That would seem to indicate to me that if you define 'uart-has-rtscts'
> you should not be defining 'cts-gpios' or 'rts-gpios' but I found that
> when I omitted 'uart-has-rtscts' I could no longer enable hardware
> flow control. Is my understanding of the yaml wrong or is this just
> not accurate?
Yes, it seems that you're right
cf older textual description :
https://elixir.bootlin.com/linux/v4.20.17/source/Documentation/devicetree/bindings/serial/serial.txt

Regards,
Richard