[PATCH v3] tty: serial: meson: hard LOCKUP on crtscts mode

[Neil Armstrong]

On 12/10/2023 17:25, Pavel Krasavin wrote:
> From: Pavel Krasavin <pkrasavin at imaqliq.com>
> 
> There might be hard lockup if we set crtscts mode on port without RTS/CTS configured:
> 
> # stty -F /dev/ttyAML6 crtscts; echo 1 > /dev/ttyAML6; echo 2 > /dev/ttyAML6
> [   95.890386] rcu: INFO: rcu_preempt detected stalls on CPUs/tasks:
> [   95.890857] rcu:     3-...0: (201 ticks this GP) idle=e33c/1/0x4000000000000000 softirq=5844/5846 fqs=4984
> [   95.900212] rcu:     (detected by 2, t=21016 jiffies, g=7753, q=296 ncpus=4)
> [   95.906972] Task dump for CPU 3:
> [   95.910178] task:bash            state:R  running task     stack:0     pid:205   ppid:1      flags:0x00000202
> [   95.920059] Call trace:
> [   95.922485]  __switch_to+0xe4/0x168
> [   95.925951]  0xffffff8003477508
> [   95.974379] watchdog: Watchdog detected hard LOCKUP on cpu 3
> [   95.974424] Modules linked in: 88x2cs(O) rtc_meson_vrtc
> 
> Possible solution would be to not allow to setup crtscts on such port.
> 
> Tested on S905X3 based board.
> 
> Signed-off-by: Pavel Krasavin <pkrasavin at imaqliq.com>
> ---
> v3: "From:" line added to the mail
> v2: https://lore.kernel.org/lkml/OF950BEF72.7F425944-ON00258A46.00488A76-00258A46.00497D44@gdc.ru/
> braces for single statement removed according to Dmitry's note
> v1: https://lore.kernel.org/lkml/OF28B2B8C9.5BC0CD28-ON00258A46.0037688F-00258A46.0039155B@gdc.ru/
> ---
> 
> --- a/drivers/tty/serial/meson_uart.c	2023-10-12 15:44:02.410538523 +0300
> +++ b/drivers/tty/serial/meson_uart.c	2023-10-12 15:58:06.242395253 +0300
> @@ -380,10 +380,14 @@ static void meson_uart_set_termios(struc
>   	else
>   		val |= AML_UART_STOP_BIT_1SB;
>   
> -	if (cflags & CRTSCTS)
> -		val &= ~AML_UART_TWO_WIRE_EN;
> -	else
> +	if (cflags & CRTSCTS) {
> +		if (port->flags & UPF_HARD_FLOW)
> +			val &= ~AML_UART_TWO_WIRE_EN;
> +		else
> +			termios->c_cflag &= ~CRTSCTS;
> +	} else {
>   		val |= AML_UART_TWO_WIRE_EN;
> +	}
>   
>   	writel(val, port->membase + AML_UART_CONTROL);
>   
> @@ -705,6 +709,7 @@ static int meson_uart_probe(struct platf
>   	u32 fifosize = 64; /* Default is 64, 128 for EE UART_0 */
>   	int ret = 0;
>   	int irq;
> +	bool has_rtscts;
>   
>   	if (pdev->dev.of_node)
>   		pdev->id = of_alias_get_id(pdev->dev.of_node, "serial");
> @@ -732,6 +737,7 @@ static int meson_uart_probe(struct platf
>   		return irq;
>   
>   	of_property_read_u32(pdev->dev.of_node, "fifo-size", &fifosize);
> +	has_rtscts = of_property_read_bool(pdev->dev.of_node, "uart-has-rtscts");
>   
>   	if (meson_ports[pdev->id]) {
>   		return dev_err_probe(&pdev->dev, -EBUSY,
> @@ -762,6 +768,8 @@ static int meson_uart_probe(struct platf
>   	port->mapsize = resource_size(res_mem);
>   	port->irq = irq;
>   	port->flags = UPF_BOOT_AUTOCONF | UPF_LOW_LATENCY;
> +	if (has_rtscts)
> +		port->flags |= UPF_HARD_FLOW;
>   	port->has_sysrq = IS_ENABLED(CONFIG_SERIAL_MESON_CONSOLE);
>   	port->dev = &pdev->dev;
>   	port->line = pdev->id;

Looks fine, I expected the tty core to check uart-has-rtscts or UPF_HARD_FLOW, but seems I was naive.

LGTM
Reviewed-by: Neil Armstrong <neil.armstrong at linaro.org>