diff mbox series

: watchdog: Allow watchdog to remain enabled after probe

Message ID 20250206093956.46593-1-regis.dargent@gmail.com (mailing list archive)
State New
Headers show
Series : watchdog: Allow watchdog to remain enabled after probe | expand

Commit Message

Regis Dargent Feb. 6, 2025, 9:39 a.m. UTC
---
The sunxi_wdt watchdog unconditionally stops the watchdog during probe (on my Allwinner H616).

What I want to achieve with this patch is to start the watchdog in the bootloader (either manually or automatically), then boot Linux.
The watchdog is about 16sec timeout maximum, while the full boot to userland lasts about 90sec, and I want the board to reset if, eg.
the rootfs cannot be mounted. So I need the watchdog to be handled by the kernel during boot (which it can do pretty well).

The thing is, the current driver stops the watchdog during probe, so it does not run during boot, and it also does not manages the "status"
field, so the kernel would know that it must handle the HW watchdog.
This avoids automatic reboot in case a problem occurs during boot (and for example handling in the bootloader).

The driver should detect if the HW watchdog is already running during probe and set its appropriate status bit to allow the kernel to handle the watchdog pings itself.
The call to sunxi_wdt_start/stop allows for proper driver and device configuration.
By default, the kernel will then ping the HW watchdog at apropriate frequency, but the user can have it stop after a time with open_timeout parameter.

 drivers/watchdog/sunxi_wdt.c | 20 +++++++++++++++++++-
 1 file changed, 19 insertions(+), 1 deletion(-)

Comments

Chen-Yu Tsai Feb. 6, 2025, 10:14 a.m. UTC | #1
On Thu, Feb 6, 2025 at 5:40 PM Regis Dargent <regis.dargent@gmail.com> wrote:

Please format your commit subject correctly. See previous commits for
examples.

Please provide a commit message including justification for the change.
The message below already makes sense, though it could be made more
concise.

Please also include your Signed-off-by.

> ---
> The sunxi_wdt watchdog unconditionally stops the watchdog during probe (on my Allwinner H616).
>
> What I want to achieve with this patch is to start the watchdog in the bootloader (either manually or automatically), then boot Linux.
> The watchdog is about 16sec timeout maximum, while the full boot to userland lasts about 90sec, and I want the board to reset if, eg.
> the rootfs cannot be mounted. So I need the watchdog to be handled by the kernel during boot (which it can do pretty well).
>
> The thing is, the current driver stops the watchdog during probe, so it does not run during boot, and it also does not manages the "status"
> field, so the kernel would know that it must handle the HW watchdog.
> This avoids automatic reboot in case a problem occurs during boot (and for example handling in the bootloader).
>
> The driver should detect if the HW watchdog is already running during probe and set its appropriate status bit to allow the kernel to handle the watchdog pings itself.
> The call to sunxi_wdt_start/stop allows for proper driver and device configuration.
> By default, the kernel will then ping the HW watchdog at apropriate frequency, but the user can have it stop after a time with open_timeout parameter.

                                                           ^ appropriate

Please wrap your messages under 80 characters per line.

>  drivers/watchdog/sunxi_wdt.c | 20 +++++++++++++++++++-
>  1 file changed, 19 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/watchdog/sunxi_wdt.c b/drivers/watchdog/sunxi_wdt.c
> index b85354a99582..20fe7da445ea 100644
> --- a/drivers/watchdog/sunxi_wdt.c
> +++ b/drivers/watchdog/sunxi_wdt.c
> @@ -192,6 +192,16 @@ static int sunxi_wdt_start(struct watchdog_device *wdt_dev)
>         return 0;
>  }
>
> +static int sunxi_wdt_enabled(struct sunxi_wdt_dev *wdt)

Make it return bool.

> +{
> +       u32 reg;
> +       void __iomem *wdt_base = wdt->wdt_base;
> +       const struct sunxi_wdt_reg *regs = wdt->wdt_regs;
> +
> +       reg = readl(wdt_base + regs->wdt_mode);
> +       return (reg & WDT_MODE_EN);

          return !!(reg & WDT_MODE_EN);
> +}
> +
>  static const struct watchdog_info sunxi_wdt_info = {
>         .identity       = DRV_NAME,
>         .options        = WDIOF_SETTIMEOUT |
> @@ -268,6 +278,11 @@ static int sunxi_wdt_probe(struct platform_device *pdev)
>         sunxi_wdt->wdt_dev.max_timeout = WDT_MAX_TIMEOUT;
>         sunxi_wdt->wdt_dev.min_timeout = WDT_MIN_TIMEOUT;
>         sunxi_wdt->wdt_dev.parent = dev;
> +       if (sunxi_wdt_enabled(sunxi_wdt)) {
> +               set_bit(WDOG_HW_RUNNING, &sunxi_wdt->wdt_dev.status);

Maybe also read back the current timeout? This should be done after
watchdog_init_timeout(), so ...

> +       } else {
> +               clear_bit(WDOG_HW_RUNNING, &sunxi_wdt->wdt_dev.status);
> +       }

... you should just combine the if-else block with the one below.

>
>         watchdog_init_timeout(&sunxi_wdt->wdt_dev, timeout, dev);
>         watchdog_set_nowayout(&sunxi_wdt->wdt_dev, nowayout);
> @@ -275,7 +290,10 @@ static int sunxi_wdt_probe(struct platform_device *pdev)
>
>         watchdog_set_drvdata(&sunxi_wdt->wdt_dev, sunxi_wdt);
>
> -       sunxi_wdt_stop(&sunxi_wdt->wdt_dev);
> +       if (watchdog_hw_running(&sunxi_wdt->wdt_dev))
> +               sunxi_wdt_start(&sunxi_wdt->wdt_dev);
> +       else
> +               sunxi_wdt_stop(&sunxi_wdt->wdt_dev);

FWIW this is what the dw_wdt driver does as well.


Thanks
ChenYu

>         watchdog_stop_on_reboot(&sunxi_wdt->wdt_dev);
>         err = devm_watchdog_register_device(dev, &sunxi_wdt->wdt_dev);
> --
> 2.25.1
>

On Thu, Feb 6, 2025 at 5:40 PM Regis Dargent <regis.dargent@gmail.com> wrote:
>
> ---
> The sunxi_wdt watchdog unconditionally stops the watchdog during probe (on my Allwinner H616).
>
> What I want to achieve with this patch is to start the watchdog in the bootloader (either manually or automatically), then boot Linux.
> The watchdog is about 16sec timeout maximum, while the full boot to userland lasts about 90sec, and I want the board to reset if, eg.
> the rootfs cannot be mounted. So I need the watchdog to be handled by the kernel during boot (which it can do pretty well).
>
> The thing is, the current driver stops the watchdog during probe, so it does not run during boot, and it also does not manages the "status"
> field, so the kernel would know that it must handle the HW watchdog.
> This avoids automatic reboot in case a problem occurs during boot (and for example handling in the bootloader).
>
> The driver should detect if the HW watchdog is already running during probe and set its appropriate status bit to allow the kernel to handle the watchdog pings itself.
> The call to sunxi_wdt_start/stop allows for proper driver and device configuration.
> By default, the kernel will then ping the HW watchdog at apropriate frequency, but the user can have it stop after a time with open_timeout parameter.
>
>  drivers/watchdog/sunxi_wdt.c | 20 +++++++++++++++++++-
>  1 file changed, 19 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/watchdog/sunxi_wdt.c b/drivers/watchdog/sunxi_wdt.c
> index b85354a99582..20fe7da445ea 100644
> --- a/drivers/watchdog/sunxi_wdt.c
> +++ b/drivers/watchdog/sunxi_wdt.c
> @@ -192,6 +192,16 @@ static int sunxi_wdt_start(struct watchdog_device *wdt_dev)
>         return 0;
>  }
>
> +static int sunxi_wdt_enabled(struct sunxi_wdt_dev *wdt)
> +{
> +       u32 reg;
> +       void __iomem *wdt_base = wdt->wdt_base;
> +       const struct sunxi_wdt_reg *regs = wdt->wdt_regs;
> +
> +       reg = readl(wdt_base + regs->wdt_mode);
> +       return (reg & WDT_MODE_EN);
> +}
> +
>  static const struct watchdog_info sunxi_wdt_info = {
>         .identity       = DRV_NAME,
>         .options        = WDIOF_SETTIMEOUT |
> @@ -268,6 +278,11 @@ static int sunxi_wdt_probe(struct platform_device *pdev)
>         sunxi_wdt->wdt_dev.max_timeout = WDT_MAX_TIMEOUT;
>         sunxi_wdt->wdt_dev.min_timeout = WDT_MIN_TIMEOUT;
>         sunxi_wdt->wdt_dev.parent = dev;
> +       if (sunxi_wdt_enabled(sunxi_wdt)) {
> +               set_bit(WDOG_HW_RUNNING, &sunxi_wdt->wdt_dev.status);
> +       } else {
> +               clear_bit(WDOG_HW_RUNNING, &sunxi_wdt->wdt_dev.status);
> +       }
>
>         watchdog_init_timeout(&sunxi_wdt->wdt_dev, timeout, dev);
>         watchdog_set_nowayout(&sunxi_wdt->wdt_dev, nowayout);
> @@ -275,7 +290,10 @@ static int sunxi_wdt_probe(struct platform_device *pdev)
>
>         watchdog_set_drvdata(&sunxi_wdt->wdt_dev, sunxi_wdt);
>
> -       sunxi_wdt_stop(&sunxi_wdt->wdt_dev);
> +       if (watchdog_hw_running(&sunxi_wdt->wdt_dev))
> +               sunxi_wdt_start(&sunxi_wdt->wdt_dev);
> +       else
> +               sunxi_wdt_stop(&sunxi_wdt->wdt_dev);
>
>         watchdog_stop_on_reboot(&sunxi_wdt->wdt_dev);
>         err = devm_watchdog_register_device(dev, &sunxi_wdt->wdt_dev);
> --
> 2.25.1
>
Regis Dargent Feb. 14, 2025, 11:22 a.m. UTC | #2
I (tried) to fix commit format (still learning, please comment is something
is missing).

I made the suggested code fixes and added a sunxi_wdt_read_timeout function;
it was a bit more touchy, as the HW timeout could be set to less than a second
(really?), so I had to manage max_hw_heartbeat_ms too.
diff mbox series

Patch

diff --git a/drivers/watchdog/sunxi_wdt.c b/drivers/watchdog/sunxi_wdt.c
index b85354a99582..20fe7da445ea 100644
--- a/drivers/watchdog/sunxi_wdt.c
+++ b/drivers/watchdog/sunxi_wdt.c
@@ -192,6 +192,16 @@  static int sunxi_wdt_start(struct watchdog_device *wdt_dev)
 	return 0;
 }
 
+static int sunxi_wdt_enabled(struct sunxi_wdt_dev *wdt)
+{
+	u32 reg;
+	void __iomem *wdt_base = wdt->wdt_base;
+	const struct sunxi_wdt_reg *regs = wdt->wdt_regs;
+
+	reg = readl(wdt_base + regs->wdt_mode);
+	return (reg & WDT_MODE_EN);
+}
+
 static const struct watchdog_info sunxi_wdt_info = {
 	.identity	= DRV_NAME,
 	.options	= WDIOF_SETTIMEOUT |
@@ -268,6 +278,11 @@  static int sunxi_wdt_probe(struct platform_device *pdev)
 	sunxi_wdt->wdt_dev.max_timeout = WDT_MAX_TIMEOUT;
 	sunxi_wdt->wdt_dev.min_timeout = WDT_MIN_TIMEOUT;
 	sunxi_wdt->wdt_dev.parent = dev;
+	if (sunxi_wdt_enabled(sunxi_wdt)) {
+		set_bit(WDOG_HW_RUNNING, &sunxi_wdt->wdt_dev.status);
+	} else {
+		clear_bit(WDOG_HW_RUNNING, &sunxi_wdt->wdt_dev.status);
+	}
 
 	watchdog_init_timeout(&sunxi_wdt->wdt_dev, timeout, dev);
 	watchdog_set_nowayout(&sunxi_wdt->wdt_dev, nowayout);
@@ -275,7 +290,10 @@  static int sunxi_wdt_probe(struct platform_device *pdev)
 
 	watchdog_set_drvdata(&sunxi_wdt->wdt_dev, sunxi_wdt);
 
-	sunxi_wdt_stop(&sunxi_wdt->wdt_dev);
+	if (watchdog_hw_running(&sunxi_wdt->wdt_dev))
+		sunxi_wdt_start(&sunxi_wdt->wdt_dev);
+	else
+		sunxi_wdt_stop(&sunxi_wdt->wdt_dev);
 
 	watchdog_stop_on_reboot(&sunxi_wdt->wdt_dev);
 	err = devm_watchdog_register_device(dev, &sunxi_wdt->wdt_dev);