Re: [PATCH tty] serial: 8250: Ignore flow control on suspend/resume with no_console_suspend

[John Ogness <john.ogness@linutronix.de>]

On 2026-07-07, sashiko.dev wrote:
>> -	/* Wait up to 1s for flow control if necessary */
>> -	if (uart_console_hwflow_active(&up->port)) {
>> +	/*
>> +	 * Wait up to 1s for flow control if necessary.
>> +	 * When 'no_console_suspend' is active (in the window between
>> +	 * suspend() and resume()), flow control is temporarily ignored
>> +	 * because the canary workaround is not reliable in all situations,
>> +	 * leading to flow control timeouts for every character.
>> +	 */
>> +	if (uart_console_hwflow_active(&up->port) && !up->canary) {
>
> This isn't a bug introduced by this patch, but does this incomplete
> fix still allow flow control timeouts if the UART loses power during
> suspend?

Yes, and this is fine. More below...

> Looking at serial8250_console_write(), if power is lost, the canary is
> successfully matched, the UART is restored, and the canary is cleared
> to 0:
>
> serial8250_console_write() {
>     ...
>     if (up->canary && (up->canary != serial_port_in(port, UART_SCR))) {
>         serial8250_console_restore(up);
>         up->canary = 0;
>     }
>     ...
> }
>
> Since this happens before the actual characters are printed, won't
> subsequent calls to wait_for_xmitr() see !up->canary as true and
> re-enable the flow control wait loop for each character?

Yes, and that is correct behavior. When serial8250_console_restore() is
called (right before clearing @canary), hardware flow control is
correctly re-configured, thus hardware flow control will work correctly
and it is fine for wait_for_xmitr() to poll.

John