Re: [PATCH] serial: imx: Use uart_port_lock_irqsave() instead of uart_port_lock()

[Lucas Stach <l.stach@pengutronix.de>]

Am Montag, dem 09.12.2024 um 20:47 +0800 schrieb Xiaolei Wang:
> When executing 'ehco mem > /sys/power/state', the following
> deadlock occurs. Since there is output during the serial
> port entering the suspend process, the suspend will be
> interrupted, resulting in the nesting of locks. Therefore,
> use uart_port_lock_irqsave instead of uart_port_unlock.
> 
> WARNING: inconsistent lock state
> 6.12.0-rc2-00002-g3c199ed5bd64-dirty #23 Not tainted
> --------------------------------
> inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage.
> sh/494 [HC0[0]:SC0[0]:HE1:SE1] takes:
> c4db5850 (&port_lock_key){?.-.}-{3:3}, at: imx_uart_enable_wakeup+0x14/0x254
> {IN-HARDIRQ-W} state was registered at:
>   lock_acquire+0x104/0x348
>   _raw_spin_lock+0x48/0x84
>   imx_uart_int+0x14/0x4dc
>   __handle_irq_event_percpu+0xac/0x2fc
>   handle_irq_event_percpu+0xc/0x40
>   handle_irq_event+0x38/0x8c
>   handle_fasteoi_irq+0xb4/0x1b8
>   handle_irq_desc+0x1c/0x2c
>   gic_handle_irq+0x6c/0xa0
>   generic_handle_arch_irq+0x2c/0x64
>   call_with_stack+0x18/0x20
>   __irq_svc+0x9c/0xbc
>   _raw_spin_unlock_irqrestore+0x2c/0x48
>   uart_write+0xd8/0x3a0
>   do_output_char+0x1a8/0x1e4
>   n_tty_write+0x224/0x440
>   file_tty_write.constprop.0+0x124/0x250
>   do_iter_readv_writev+0x100/0x1e0
>   vfs_writev+0xc4/0x448
>   do_writev+0x68/0xf8
>   ret_fast_syscall+0x0/0x1c
> irq event stamp: 31593
> hardirqs last  enabled at (31593): [<c1150e48>] _raw_spin_unlock_irqrestore+0x44/0x48
> hardirqs last disabled at (31592): [<c07f32f0>] clk_enable_lock+0x60/0x120
> softirqs last  enabled at (30334): [<c012d1d4>] handle_softirqs+0x2cc/0x478
> softirqs last disabled at (30325): [<c012d510>] __irq_exit_rcu+0x120/0x15c
> 
> other info that might help us debug this:
>  Possible unsafe locking scenario:
> 
>        CPU0
>        ----
>   lock(&port_lock_key);
>   <Interrupt>
>     lock(&port_lock_key);
> 
> Fixes: 3c199ed5bd64 ("serial: imx: Grab port lock in imx_uart_enable_wakeup()")
> Signed-off-by: Xiaolei Wang <xiaolei.wang@windriver.com>
> ---
>  drivers/tty/serial/imx.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/tty/serial/imx.c b/drivers/tty/serial/imx.c
> index 17f70e4bee43..4470bcb3ef19 100644
> --- a/drivers/tty/serial/imx.c
> +++ b/drivers/tty/serial/imx.c
> @@ -2691,8 +2691,9 @@ static void imx_uart_save_context(struct imx_port *sport)
>  static void imx_uart_enable_wakeup(struct imx_port *sport, bool on)
>  {
>  	u32 ucr3;
> +	unsigned long flags;
>  
> -	uart_port_lock(&sport->port);
> +	uart_port_lock_irqsave(&sport->port, &flags);

This function is only called from process context, so it should be fine
to use uart_port_lock_irq().

Regards,
Lucas

>  
>  	ucr3 = imx_uart_readl(sport, UCR3);
>  	if (on) {
> @@ -2714,7 +2715,7 @@ static void imx_uart_enable_wakeup(struct imx_port *sport, bool on)
>  		imx_uart_writel(sport, ucr1, UCR1);
>  	}
>  
> -	uart_port_unlock(&sport->port);
> +	uart_port_unlock_irqrestore(&sport->port, flags);
>  }
>  
>  static int imx_uart_suspend_noirq(struct device *dev)