[V4] serial: core: Do stop_rx in suspend path for console if console_suspend is disabled

For the case of console_suspend disabled, if back to back suspend/resume
test is executed, at the end of test, sometimes console would appear to
be frozen not responding to input. This would happen because, during
resume, rx transactions can come in before system is ready, malfunction
of rx happens in turn resulting in console appearing to be stuck.

Do a stop_rx in suspend sequence to prevent this.

Signed-off-by: Vijaya Krishna Nivarthi <quic_vnivarth@quicinc.com>
---
v4: moved the change to serial core to apply for all drivers
v3: swapped the order of conditions to be more human readable
v2: restricted patch to contain only stop_rx in suspend sequence
v1: intial patch contained 2 additional unrelated changes in vicinity
---
 drivers/tty/serial/serial_core.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

Hi,

On 16.05.2022 11:20, Vijaya Krishna Nivarthi wrote:
> For the case of console_suspend disabled, if back to back suspend/resume
> test is executed, at the end of test, sometimes console would appear to
> be frozen not responding to input. This would happen because, during
> resume, rx transactions can come in before system is ready, malfunction
> of rx happens in turn resulting in console appearing to be stuck.
>
> Do a stop_rx in suspend sequence to prevent this.
>
> Signed-off-by: Vijaya Krishna Nivarthi <quic_vnivarth@quicinc.com>
> ---
> v4: moved the change to serial core to apply for all drivers
> v3: swapped the order of conditions to be more human readable
> v2: restricted patch to contain only stop_rx in suspend sequence
> v1: intial patch contained 2 additional unrelated changes in vicinity
> ---

This patch landed recently in linux-next as commit c9d2325cdb92 
("serial: core: Do stop_rx in suspend path for console if 
console_suspend is disabled").

Unfortunately it breaks console operation on my test systems after 
system suspend/resume cycle if 'no_console_suspend' kernel parameter is 
present. System properly resumes from suspend, the console displays all 
the messages and even command line prompt, but then doesn't react on any 
input. If I remove the 'no_console_suspend' parameter, the console is 
again operational after system suspend/resume cycle. Before this patch 
it worked fine regardless the 'no_console_suspend' parameter.

If this matters, the test system is ARM 32bit Samsung Exynos5422-based 
Odroid XU3lite board.

>   drivers/tty/serial/serial_core.c | 11 +++++++++--
>   1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/tty/serial/serial_core.c b/drivers/tty/serial/serial_core.c
> index 82a1770..9a85b41 100644
> --- a/drivers/tty/serial/serial_core.c
> +++ b/drivers/tty/serial/serial_core.c
> @@ -2211,9 +2211,16 @@ int uart_suspend_port(struct uart_driver *drv, struct uart_port *uport)
>   	}
>   	put_device(tty_dev);
>   
> -	/* Nothing to do if the console is not suspending */
> -	if (!console_suspend_enabled && uart_console(uport))
> +	/*
> +	 * Nothing to do if the console is not suspending
> +	 * except stop_rx to prevent any asynchronous data
> +	 * over RX line. Re-start_rx, when required, is
> +	 * done by set_termios in resume sequence
> +	 */
> +	if (!console_suspend_enabled && uart_console(uport)) {
> +		uport->ops->stop_rx(uport);
>   		goto unlock;
> +	}
>   
>   	uport->suspended = 1;
>   

Best regards

Hi,

On 01.06.2022 13:24, Vijaya Krishna Nivarthi wrote:
> On 5/24/2022 5:24 PM, Marek Szyprowski wrote:
>> On 23.05.2022 23:32, Marek Szyprowski wrote:
>>> Hi,
>>>
>>> On 16.05.2022 11:20, Vijaya Krishna Nivarthi wrote:
>>>> For the case of console_suspend disabled, if back to back 
>>>> suspend/resume
>>>> test is executed, at the end of test, sometimes console would 
>>>> appear to
>>>> be frozen not responding to input. This would happen because, during
>>>> resume, rx transactions can come in before system is ready, 
>>>> malfunction
>>>> of rx happens in turn resulting in console appearing to be stuck.
>>>>
>>>> Do a stop_rx in suspend sequence to prevent this.
>>>>
>>>> Signed-off-by: Vijaya Krishna Nivarthi <quic_vnivarth@quicinc.com>
>>>> ---
>>>> v4: moved the change to serial core to apply for all drivers
>>>> v3: swapped the order of conditions to be more human readable
>>>> v2: restricted patch to contain only stop_rx in suspend sequence
>>>> v1: intial patch contained 2 additional unrelated changes in vicinity
>>>> ---
>>> This patch landed recently in linux-next as commit c9d2325cdb92
>>> ("serial: core: Do stop_rx in suspend path for console if
>>> console_suspend is disabled").
>>>
>>> Unfortunately it breaks console operation on my test systems after
>>> system suspend/resume cycle if 'no_console_suspend' kernel parameter
>>> is present. System properly resumes from suspend, the console displays
>>> all the messages and even command line prompt, but then doesn't react
>>> on any input. If I remove the 'no_console_suspend' parameter, the
>>> console is again operational after system suspend/resume cycle. Before
>>> this patch it worked fine regardless the 'no_console_suspend' 
>>> parameter.
>>>
>>> If this matters, the test system is ARM 32bit Samsung Exynos5422-based
>>> Odroid XU3lite board.
>>
>> One more information. This issue can be easily reproduced with QEMU. It
>> happens both on ARM 32bit and ARM 64bit QEMU's 'virt' machines when
>> 'no_console_suspend' is added to kernel's cmdline.
>>
> Ideally, as comments indicate, the set_termios should have done 
> stop_rx at begin and start_rx at end to take care of this issue.
>
> This is probably missing in your driver. Can we check if this can be 
> fixed? 

Sure, just point me what need to be added in amba-pl011.c and 
samsung_tty.c. I've briefly compared the suspend/resume paths of those 
drivers with other drivers and I don't see anything missing there.

> OR other option is
>
> Add a start_rx in uart_resume_port after call to set_termios to handle 
> this scenario for other drivers.
>
> Please let me know if there are any concerns for this options.

IMHO this looks like an issue that affects lots of drivers and it should 
be handled in the core.

 > ...

Best regards

Hi,

On Fri, Jun 3, 2022 at 11:54 AM Vijaya Krishna Nivarthi
<quic_vnivarth@quicinc.com> wrote:
>
> >>> Add a start_rx in uart_resume_port after call to set_termios to handle
> >>> this scenario for other drivers.
>
> Since start_rx is not exposed it doesn't seem like we will be able to
> handle it in core.
>
> In your drivers, Can we add a call to stop_rx at begin of set_termios
> and then undo it at end?
>
> That would ensure that set_termios functionality is unaffected while
> fixing the broken cases?
>
> If that's not an option we will have to go back to a previous version of
> limiting the change to qcom driver.

How about this: add an optional start_rx() callback to "struct
uart_ops" and then only do your stop_rx() logic in uart_suspend_port()
if you'll be able to start it again (AKA if the start_rx() callback is
not NULL). That keeps the logic in the core.

-Doug