Re: [PATCH V2] tty: serial: fsl_lpuart: don't break the on-going transfer when global reset

[Jiri Slaby <jirislaby@kernel.org>]

On 19. 10. 22, 13:07, Sherry Sun wrote:
> lpuart_global_reset() shouldn't break the on-going transmit engin, need

"engine"

> to recover the on-going data transfer after reset.
> 
> This can help earlycon here, since commit 60f361722ad2 ("serial:
> fsl_lpuart: Reset prior to registration") moved lpuart_global_reset()
> before uart_add_one_port(), earlycon is writing during global reset,
> as global reset will disable the TX and clear the baud rate register,
> which caused the earlycon cannot work any more after reset, needs to
> restore the baud rate and re-enable the transmitter to recover the
> earlycon write.
> 
> Fixes: bd5305dcabbc ("tty: serial: fsl_lpuart: do software reset for imx7ulp and imx8qxp")
> Signed-off-by: Sherry Sun <sherry.sun@nxp.com>
> ---
> Changes in V2:
> 1. The while loop may never exit as the stat and sfifo are not re-read inside
> the loop, fix that.
> ---
>   drivers/tty/serial/fsl_lpuart.c | 22 +++++++++++++++++++---
>   1 file changed, 19 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/tty/serial/fsl_lpuart.c b/drivers/tty/serial/fsl_lpuart.c
> index 67fa113f77d4..9a0781395b1f 100644
> --- a/drivers/tty/serial/fsl_lpuart.c
> +++ b/drivers/tty/serial/fsl_lpuart.c
> @@ -408,11 +408,9 @@ static int lpuart_global_reset(struct lpuart_port *sport)
>   {
>   	struct uart_port *port = &sport->port;
>   	void __iomem *global_addr;
> +	unsigned long tx_enable, bd;
>   	int ret;
>   
> -	if (uart_console(port))
> -		return 0;
> -
>   	ret = clk_prepare_enable(sport->ipg_clk);
>   	if (ret) {
>   		dev_err(sport->port.dev, "failed to enable uart ipg clk: %d\n", ret);
> @@ -420,11 +418,29 @@ static int lpuart_global_reset(struct lpuart_port *sport)
>   	}
>   
>   	if (is_imx7ulp_lpuart(sport) || is_imx8qxp_lpuart(sport)) {
> +		/*
> +		 * If the transmitter is used by earlycon, wait transmit engin complete

"wait for transmit engine to complete"

> +		 * and then reset

"." (add a period)

> +		 */
> +		tx_enable = lpuart32_read(port, UARTCTRL) & UARTCTRL_TE;
> +		if (tx_enable) {
> +			bd = lpuart32_read(&sport->port, UARTBAUD);
> +			while (!(lpuart32_read(port, UARTSTAT) & UARTSTAT_TC &&
> +				 lpuart32_read(port, UARTFIFO) & UARTFIFO_TXEMPT))
> +				cpu_relax();

If the HW is stuck, you make the kernel stuck too. What about 
read_poll_timeout_atomic(). Or at least a custom timeout.

> +		}
> +
>   		global_addr = port->membase + UART_GLOBAL - IMX_REG_OFF;
>   		writel(UART_GLOBAL_RST, global_addr);
>   		usleep_range(GLOBAL_RST_MIN_US, GLOBAL_RST_MAX_US);
>   		writel(0, global_addr);
>   		usleep_range(GLOBAL_RST_MIN_US, GLOBAL_RST_MAX_US);
> +
> +		/* Recover the transmitter for earlycon */
> +		if (tx_enable) {
> +			lpuart32_write(port, bd, UARTBAUD);
> +			lpuart32_write(port, UARTCTRL_TE, UARTCTRL);

I don't know the HW. Is it enough to write TE here? IOW shouldn't we 
preserve the whole register content from the above read here?

thanks,
-- 
js
suse labs