On Wed, Sep 29, 2021 at 3:35 PM Fabio Estevam <festevam@xxxxxxxxx> wrote:
>
> On Wed, Sep 29, 2021 at 1:10 PM Fabio Estevam <festevam@xxxxxxxxx> wrote:
> >
> > Hi Petr and Sergey,
> >
> > I know this has been reported before [1] and [2], but I am still
> > observing the deadlock below
> > on an imx6q board since commit dbdda842fe96 ("printk: Add console
> > owner and waiter logic to load balance console writes").
> >
> > To reproduce it:
> >
> > # echo t > /proc/sysrq-trigger
> >
> > [ 20.325246] ======================================================
> > [ 20.325252] WARNING: possible circular locking dependency detected
> > [ 20.325260] 5.15.0-rc2-next-20210924-00004-gd2d6e664f29f-dirty #163
>
> If I move the __imx_uart_rxint() block outside the spin_lock() like this:
>
> diff --git a/drivers/tty/serial/imx.c b/drivers/tty/serial/imx.c
> index 8b121cd869e9..c94704f5dd99 100644
> --- a/drivers/tty/serial/imx.c
> +++ b/drivers/tty/serial/imx.c
> @@ -955,13 +955,6 @@ static irqreturn_t imx_uart_int(int irq, void *dev_id)
> if ((ucr4 & UCR4_OREN) == 0)
> usr2 &= ~USR2_ORE;
>
> - if (usr1 & (USR1_RRDY | USR1_AGTIM)) {
> - imx_uart_writel(sport, USR1_AGTIM, USR1);
> -
> - __imx_uart_rxint(irq, dev_id);
> - ret = IRQ_HANDLED;
> - }
> -
> if ((usr1 & USR1_TRDY) || (usr2 & USR2_TXDC)) {
> imx_uart_transmit_buffer(sport);
> ret = IRQ_HANDLED;
> @@ -993,6 +986,13 @@ static irqreturn_t imx_uart_int(int irq, void *dev_id)
>
> spin_unlock(&sport->port.lock);
>
> + if (usr1 & (USR1_RRDY | USR1_AGTIM)) {
> + imx_uart_writel(sport, USR1_AGTIM, USR1);
> +
> + __imx_uart_rxint(irq, dev_id);
> + ret = IRQ_HANDLED;
> + }
> +
> return ret;
> }
>
> Then the problem does not happen anymore.
>
> Is this a proper fix?
It seems that a proper fix would be to use the same approach as in
5697df7322fe ("serial: fsl_lpuart: split sysrq handling):
I will send a proper patch with this solution.
Thanks
