Re: [PATCH v4] watchdog: add new parameter to start the watchdog on module insertion

From: Guenter Roeck
Date: Sun Oct 10 2021 - 21:26:07 EST


On Fri, Sep 24, 2021 at 03:35:09PM +0200, Flavio Suligoi wrote:
> The new parameter "start_enabled" starts the watchdog at the same time
> of the module insertion.
> This feature is very useful in embedded systems, to avoid cases where
> the system hangs before reaching userspace.
>
> This feature can be enabled in the kernel config, so it can be also
> used when the watchdog driver is build as "built-in".
>
> This parameter involves the "core" section of the watchdog driver;
> in this way it is common for all the watchdog hardware implementations.
>
> Signed-off-by: Flavio Suligoi <f.suligoi@xxxxxxx>
> ---
>
> v4: - improve code legibility in returned error checking
> - stop the watchdog if driver registration fails
> v3: - add check for the returned error of wdog start function
> v2: - check WDOG_HW_RUNNING before starting watchdog;
> - remove useless comments in commit text, watchdog-parameters.rst and
> Kconfig;
> v1: - first version;
>
> Documentation/watchdog/watchdog-parameters.rst | 3 +++
> drivers/watchdog/Kconfig | 9 +++++++++
> drivers/watchdog/watchdog_core.c | 17 +++++++++++++++++
> 3 files changed, 29 insertions(+)
>
> diff --git a/Documentation/watchdog/watchdog-parameters.rst b/Documentation/watchdog/watchdog-parameters.rst
> index 223c99361a30..7780d0c1fb4a 100644
> --- a/Documentation/watchdog/watchdog-parameters.rst
> +++ b/Documentation/watchdog/watchdog-parameters.rst
> @@ -21,6 +21,9 @@ watchdog core:
> timeout. Setting this to a non-zero value can be useful to ensure that
> either userspace comes up properly, or the board gets reset and allows
> fallback logic in the bootloader to try something else.
> + start_enabled:
> + Watchdog is started on module insertion. This option can be also
> + selected by kernel config (default=kernel config parameter).
>
> -------------------------------------------------
>
> diff --git a/drivers/watchdog/Kconfig b/drivers/watchdog/Kconfig
> index b81fe4f7d434..19be709e01ac 100644
> --- a/drivers/watchdog/Kconfig
> +++ b/drivers/watchdog/Kconfig
> @@ -47,6 +47,15 @@ config WATCHDOG_NOWAYOUT
> get killed. If you say Y here, the watchdog cannot be stopped once
> it has been started.
>
> +config WATCHDOG_START_ENABLED
> + bool "Start watchdog on module insertion"
> + help
> + Say Y if you want to start the watchdog at the same time when the
> + driver is loaded.
> + This feature is very useful in embedded systems, to avoid cases where
> + the system could hang before reaching userspace.
> + This parameter applies to all watchdog drivers.
> +
> config WATCHDOG_HANDLE_BOOT_ENABLED
> bool "Update boot-enabled watchdog until userspace takes over"
> default y
> diff --git a/drivers/watchdog/watchdog_core.c b/drivers/watchdog/watchdog_core.c
> index 3fe8a7edc252..d9211fea45d7 100644
> --- a/drivers/watchdog/watchdog_core.c
> +++ b/drivers/watchdog/watchdog_core.c
> @@ -44,6 +44,11 @@ static int stop_on_reboot = -1;
> module_param(stop_on_reboot, int, 0444);
> MODULE_PARM_DESC(stop_on_reboot, "Stop watchdogs on reboot (0=keep watching, 1=stop)");
>
> +static bool start_enabled = IS_ENABLED(CONFIG_WATCHDOG_START_ENABLED);
> +module_param(start_enabled, bool, 0444);
> +MODULE_PARM_DESC(start_enabled, "Start watchdog on module insertion (default="
> + __MODULE_STRING(IS_ENABLED(CONFIG_WATCHDOG_START_ENABLED)) ")");
> +
> /*
> * Deferred Registration infrastructure.
> *
> @@ -252,6 +257,15 @@ static int __watchdog_register_device(struct watchdog_device *wdd)
> * corrupted in a later stage then we expect a kernel panic!
> */
>
> + /* If required, start the watchdog immediately */
> + if (start_enabled && !watchdog_hw_running(wdd)) {
> + set_bit(WDOG_HW_RUNNING, &wdd->status);
> + ret = wdd->ops->start(wdd);
> + if (ret)
> + return ret;
> + pr_info("Watchdog enabled\n");
> + }
> +
I am not convinced that this is the best location to start the watchdog.
Maybe that should be done in watchdog_cdev_register().

> /* Use alias for watchdog id if possible */
> if (wdd->parent) {
> ret = of_alias_get_id(wdd->parent->of_node, "watchdog");
> @@ -356,6 +370,9 @@ int watchdog_register_device(struct watchdog_device *wdd)
> mutex_unlock(&wtd_deferred_reg_mutex);
>
> if (ret) {
> + if (start_enabled && watchdog_hw_running(wdd) &&
> + wdd->ops->stop)
> + wdd->ops->stop(wdd);

This code stops the watchdog if watchdog registration failed even and
especially if it was already running when the module was inserted.
This changes semantics and is thus not aceptable.

> dev_str = wdd->parent ? dev_name(wdd->parent) :
> (const char *)wdd->info->identity;
> pr_err("%s: failed to register watchdog device (err = %d)\n",