Re: [RFC PATCH] watchdog: renesas_wdt: support handover from bootloader

[Guenter Roeck <linux@roeck-us.net>]

On Mon, Apr 15, 2019 at 12:52:01PM +0200, Wolfram Sang wrote:
> Support an already running watchdog by checking its enable bit and set
> up the status accordingly before registering the device.
> 
> Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
> ---
> 
> This patch was tested using a Renesas Salvator XS board (R-Car M3N). It works.
> However, there is a small window where the watchdog clock is disabled, namely
> after the MSSR clock driver initializes it until RuntimePM of the watchdog
> driver takes over. If the system hangs in this window, bad luck. So, I'd think
> it makes sense to have this clock either always-on or to keep the state which
> came from the firmware. Geert, what do you think?
> 
>  drivers/watchdog/renesas_wdt.c | 15 +++++++++++++--
>  1 file changed, 13 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/watchdog/renesas_wdt.c b/drivers/watchdog/renesas_wdt.c
> index 565dbc1ec638..37d757288b22 100644
> --- a/drivers/watchdog/renesas_wdt.c
> +++ b/drivers/watchdog/renesas_wdt.c
> @@ -179,6 +179,7 @@ static int rwdt_probe(struct platform_device *pdev)
>  	struct clk *clk;
>  	unsigned long clks_per_sec;
>  	int ret, i;
> +	u8 csra;
>  
>  	if (rwdt_blacklisted(&pdev->dev))
>  		return -ENODEV;
> @@ -198,8 +199,8 @@ static int rwdt_probe(struct platform_device *pdev)
>  	pm_runtime_enable(&pdev->dev);
>  	pm_runtime_get_sync(&pdev->dev);
>  	priv->clk_rate = clk_get_rate(clk);
> -	priv->wdev.bootstatus = (readb_relaxed(priv->base + RWTCSRA) &
> -				RWTCSRA_WOVF) ? WDIOF_CARDRESET : 0;
> +	csra = readb_relaxed(priv->base + RWTCSRA);
> +	priv->wdev.bootstatus = csra & RWTCSRA_WOVF ? WDIOF_CARDRESET : 0;
>  	pm_runtime_put(&pdev->dev);
>  
>  	if (!priv->clk_rate) {
> @@ -237,6 +238,16 @@ static int rwdt_probe(struct platform_device *pdev)
>  	/* This overrides the default timeout only if DT configuration was found */
>  	watchdog_init_timeout(&priv->wdev, 0, &pdev->dev);
>  
> +	if (csra & RWTCSRA_TME) {
> +		/* Ensure properly initialized dividers */
> +		rwdt_start(&priv->wdev);
> +		set_bit(WDOG_HW_RUNNING, &priv->wdev.status);
> +		//FIXME: We are missing pm_runtime_put in some code paths to
> +		// to balance PM calls. We first need to decide if we maybe
> +		// should have the RWDT clock always-on or if using RPM for
> +		// clock management is OK.

Maybe I am missing something, but ..

Is handover even possible if the clock is controlled by clock management ?
Seems to me the clock would then be turned off through pm, which effectively
turns off the watchdog. So it will be off between clock/pm initialization
and the above code, meaning wdt handover from the boot loader is for all
practical purposes useless if the kernel gets stuck in between.

Thanks,
Guenter

> +	}
> +
>  	ret = watchdog_register_device(&priv->wdev);
>  	if (ret < 0)
>  		goto out_pm_disable;
> -- 
> 2.11.0
>