RE: Problem with nbcon console and amba-pl011 serial port

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

On 2025-06-04, "Toshiyuki Sato (Fujitsu)" <fj6611ie@fujitsu.com> wrote:
> This is a proposed fix to force termination by returning false from
> nbcon_reacquire_nobuf when a panic occurs within pl011_console_write_thread.
>  (I believe this is similar to what John suggested in his previous reply.)
>
> While I couldn't reproduce the issue using sysrq-trigger in my environment
>  (It seemed that the panic was being executed before the thread processing),
> I did observe nbcon_reacquire_nobuf failing to complete when injecting an
> NMI (SError) during pl011_console_write_thread. 
> Applying this fix seems to have resolved the "SMP: failed to stop secondary
> CPUs" issue.
>
> This patch is for test.
> Modifications to imx and other drivers, as well as adding __must_check, 
> will likely be required.
>
> Michael, could you please test this fix in your environment?
>
> Regards,
> Toshiyuki Sato
>
> diff --git a/drivers/tty/serial/amba-pl011.c b/drivers/tty/serial/amba-pl011.c
> index 11d650975..c3a2b22e6 100644
> --- a/drivers/tty/serial/amba-pl011.c
> +++ b/drivers/tty/serial/amba-pl011.c
> @@ -2577,8 +2577,10 @@ pl011_console_write_thread(struct console *co, struct nbcon_write_context *wctxt
>  		}
>  	}
>  
> -	while (!nbcon_enter_unsafe(wctxt))
> -		nbcon_reacquire_nobuf(wctxt);
> +	while (!nbcon_enter_unsafe(wctxt)) {

I realize this is just a test patch. But for the real patch, there needs
to be some comment here about bailing out. On panic, the driver is
leaving the clock enabled and not restoring REG_CR. At a quick glance
this appears to be a bug. So the comment is needed to make it clear that
the driver is exiting on purpose without proper completion. And perhaps
also mentioning why this is safe during panic. Thanks.

> +		if (!nbcon_reacquire_nobuf(wctxt))
> +			return;
> +	}
>  
>  	while ((pl011_read(uap, REG_FR) ^ uap->vendor->inv_fr) & uap->vendor->fr_busy)
>  		cpu_relax();

John Ogness