[v2,1/1] drivers: watchdog: stm32_iwdg: set WDOG_HW_RUNNING at probe
diff mbox series

Message ID 20191122082442.21695-1-christophe.roullier@st.com
State Superseded
Headers show
Series
  • [v2,1/1] drivers: watchdog: stm32_iwdg: set WDOG_HW_RUNNING at probe
Related show

Commit Message

Christophe ROULLIER Nov. 22, 2019, 8:24 a.m. UTC
If the watchdog hardware is already enabled during the boot process,
when the Linux watchdog driver loads, it should start/reset the watchdog
and tell the watchdog framework. As a result, ping can be generated from
the watchdog framework (if CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED is set),
until the userspace watchdog daemon takes over control

Fixes:4332d113c66a ("watchdog: Add STM32 IWDG driver")

Signed-off-by: Christophe Roullier <christophe.roullier@st.com>
---
Changes since v1:
According to Guenter
I follow the guidance from intel-mid_wdt.c
and I added test to check if CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED is set
because we need to be flexible (depends on customer config, but watchdog
not always start by bootloader (Uboot, ..))
I've tested some config and it is working as expected:
Watchdog enable in Uboot + HANDLE_BOOT_ENABLE is not set + daemon watchdog in userland ON ==> No reset IWDG2
Watchdog enable in Uboot + HANDLE_BOOT_ENABLE is not set ==> Reset IWDG2
Watchdog enable in Uboot + HANDLE_BOOT_ENABLE=y ==> No reset IWDG2
Watchdog enable in Uboot + HANDLE_BOOT_ENABLE=y + daemon watchdog in userland ON puis OFF ==> Reset IWDG2
Watchdog disable in Uboot + HANDLE_BOOT_ENABLE is not set ==> No reset IWDG2
Watchdog disable in Uboot + HANDLE_BOOT_ENABLE=y ==> No reset IWDG2
Watchdog disable in Uboot + HANDLE_BOOT_ENABLE=y + daemon watchdog in userland ON ==> No reset IWDG2
Watchdog disable in Uboot + HANDLE_BOOT_ENABLE=y + daemon watchdog in userland ON puis OFF ==> Reset IWDG2

Thanks,
Christophe

 drivers/watchdog/stm32_iwdg.c | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)

Comments

Ahmad Fatoum Nov. 22, 2019, 10:28 a.m. UTC | #1
Hello Christophe,

On 11/22/19 9:24 AM, Christophe Roullier wrote:
> +	/*
> +	 * In case of CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED is set
> +	 * (Means U-Boot/bootloaders leaves the watchdog running)
> +	 * When we get here we should make a decision to prevent
> +	 * any side effects before user space daemon will take care of it.
> +	 * The best option, taking into consideration that there is no
> +	 * way to read values back from hardware, is to enforce watchdog
> +	 * being run with deterministic values.

What about the "ONF: Watchdog enable status bit" in the IWDG_SR register?

Cheers
Ahmad
Guenter Roeck Nov. 22, 2019, 10:45 a.m. UTC | #2
On 11/22/19 12:24 AM, Christophe Roullier wrote:
> If the watchdog hardware is already enabled during the boot process,
> when the Linux watchdog driver loads, it should start/reset the watchdog
> and tell the watchdog framework. As a result, ping can be generated from
> the watchdog framework (if CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED is set),
> until the userspace watchdog daemon takes over control
> 
> Fixes:4332d113c66a ("watchdog: Add STM32 IWDG driver")
> 
> Signed-off-by: Christophe Roullier <christophe.roullier@st.com>
> ---
> Changes since v1:
> According to Guenter
> I follow the guidance from intel-mid_wdt.c
> and I added test to check if CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED is set
> because we need to be flexible (depends on customer config, but watchdog
> not always start by bootloader (Uboot, ..))
> I've tested some config and it is working as expected:
> Watchdog enable in Uboot + HANDLE_BOOT_ENABLE is not set + daemon watchdog in userland ON ==> No reset IWDG2
> Watchdog enable in Uboot + HANDLE_BOOT_ENABLE is not set ==> Reset IWDG2
> Watchdog enable in Uboot + HANDLE_BOOT_ENABLE=y ==> No reset IWDG2
> Watchdog enable in Uboot + HANDLE_BOOT_ENABLE=y + daemon watchdog in userland ON puis OFF ==> Reset IWDG2
> Watchdog disable in Uboot + HANDLE_BOOT_ENABLE is not set ==> No reset IWDG2
> Watchdog disable in Uboot + HANDLE_BOOT_ENABLE=y ==> No reset IWDG2
> Watchdog disable in Uboot + HANDLE_BOOT_ENABLE=y + daemon watchdog in userland ON ==> No reset IWDG2
> Watchdog disable in Uboot + HANDLE_BOOT_ENABLE=y + daemon watchdog in userland ON puis OFF ==> Reset IWDG2
> 
> Thanks,
> Christophe
> 
>   drivers/watchdog/stm32_iwdg.c | 21 +++++++++++++++++++++
>   1 file changed, 21 insertions(+)
> 
> diff --git a/drivers/watchdog/stm32_iwdg.c b/drivers/watchdog/stm32_iwdg.c
> index a3a329011a06..7f454a6e17ba 100644
> --- a/drivers/watchdog/stm32_iwdg.c
> +++ b/drivers/watchdog/stm32_iwdg.c
> @@ -50,6 +50,9 @@
>   #define TIMEOUT_US	100000
>   #define SLEEP_US	1000
>   
> +static bool handle_boot_enabled =
> +	IS_ENABLED(CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED);
> +
>   struct stm32_iwdg_data {
>   	bool has_pclk;
>   	u32 max_prescaler;
> @@ -262,6 +265,24 @@ static int stm32_iwdg_probe(struct platform_device *pdev)
>   	watchdog_set_nowayout(wdd, WATCHDOG_NOWAYOUT);
>   	watchdog_init_timeout(wdd, 0, dev);
>   
> +	/*
> +	 * In case of CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED is set
> +	 * (Means U-Boot/bootloaders leaves the watchdog running)
> +	 * When we get here we should make a decision to prevent
> +	 * any side effects before user space daemon will take care of it.
> +	 * The best option, taking into consideration that there is no
> +	 * way to read values back from hardware, is to enforce watchdog
> +	 * being run with deterministic values.
> +	 */
> +	if (handle_boot_enabled) {

You don't need that variable. Just use
	if (IS_ENABLED(CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED)) {
directly.

> +		ret = stm32_iwdg_start(wdd);
> +		if (ret)
> +			return ret;
> +
> +		/* Make sure the watchdog is serviced */
> +		set_bit(WDOG_HW_RUNNING, &wdd->status);
> +	}
> +
>   	ret = devm_watchdog_register_device(dev, wdd);
>   	if (ret)
>   		return ret;
>
Christophe ROULLIER Nov. 22, 2019, 1:15 p.m. UTC | #3
On 11/22/19 11:28 AM, Ahmad Fatoum wrote:
> Hello Christophe,
>
> On 11/22/19 9:24 AM, Christophe Roullier wrote:
>> +	/*
>> +	 * In case of CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED is set
>> +	 * (Means U-Boot/bootloaders leaves the watchdog running)
>> +	 * When we get here we should make a decision to prevent
>> +	 * any side effects before user space daemon will take care of it.
>> +	 * The best option, taking into consideration that there is no
>> +	 * way to read values back from hardware, is to enforce watchdog
>> +	 * being run with deterministic values.
> What about the "ONF: Watchdog enable status bit" in the IWDG_SR register?
>
> Cheers
> Ahmad
>
Hi,

Thanks Ahmad for your feedback, it is a mistake in our ref manual. This 
bit is not present

in our iwdg IP, we will update our documentation accordingly.

Regards,

Christophe
Ahmad Fatoum Nov. 22, 2019, 1:31 p.m. UTC | #4
Hello Christophe,

On 11/22/19 2:15 PM, Christophe ROULLIER wrote:
>> On 11/22/19 9:24 AM, Christophe Roullier wrote:
>>> +	/*
>>> +	 * In case of CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED is set
>>> +	 * (Means U-Boot/bootloaders leaves the watchdog running)
>>> +	 * When we get here we should make a decision to prevent
>>> +	 * any side effects before user space daemon will take care of it.
>>> +	 * The best option, taking into consideration that there is no
>>> +	 * way to read values back from hardware, is to enforce watchdog
>>> +	 * being run with deterministic values.
>> What about the "ONF: Watchdog enable status bit" in the IWDG_SR register?

> Thanks Ahmad for your feedback, it is a mistake in our ref manual. This 
> bit is not present
> 
> in our iwdg IP, we will update our documentation accordingly.

Thanks for the info. I think it would be nice to have this information in
the driver as well, something like:

* The best option, taking into consideration that that the ONF
* watchdog status bit is non-functional, is to enforce watchdog
* being run with deterministic values.

This way it's clear that it's a deliberate choice to do it this way.

Cheers
Ahmad

Patch
diff mbox series

diff --git a/drivers/watchdog/stm32_iwdg.c b/drivers/watchdog/stm32_iwdg.c
index a3a329011a06..7f454a6e17ba 100644
--- a/drivers/watchdog/stm32_iwdg.c
+++ b/drivers/watchdog/stm32_iwdg.c
@@ -50,6 +50,9 @@ 
 #define TIMEOUT_US	100000
 #define SLEEP_US	1000
 
+static bool handle_boot_enabled =
+	IS_ENABLED(CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED);
+
 struct stm32_iwdg_data {
 	bool has_pclk;
 	u32 max_prescaler;
@@ -262,6 +265,24 @@  static int stm32_iwdg_probe(struct platform_device *pdev)
 	watchdog_set_nowayout(wdd, WATCHDOG_NOWAYOUT);
 	watchdog_init_timeout(wdd, 0, dev);
 
+	/*
+	 * In case of CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED is set
+	 * (Means U-Boot/bootloaders leaves the watchdog running)
+	 * When we get here we should make a decision to prevent
+	 * any side effects before user space daemon will take care of it.
+	 * The best option, taking into consideration that there is no
+	 * way to read values back from hardware, is to enforce watchdog
+	 * being run with deterministic values.
+	 */
+	if (handle_boot_enabled) {
+		ret = stm32_iwdg_start(wdd);
+		if (ret)
+			return ret;
+
+		/* Make sure the watchdog is serviced */
+		set_bit(WDOG_HW_RUNNING, &wdd->status);
+	}
+
 	ret = devm_watchdog_register_device(dev, wdd);
 	if (ret)
 		return ret;