Re: [PATCH 2/2] watchdog: gpio_wdt: add "start-at-boot" feature

[Francesco Zanella]

On 22/04/21 20:05, Guenter Roeck wrote:
> On Thu, Apr 22, 2021 at 06:28:40PM +0200, Francesco Zanella wrote:
>>
>>
>> On 21/04/21 18:42, Guenter Roeck wrote:
>>> On Wed, Apr 21, 2021 at 06:26:21PM +0200, Francesco Zanella wrote:
>>>> If "start-at-boot" property is present in the device tree, start pinging
>>>> hw watchdog at probe, in order to take advantage of kernel configs:
>>>> - WATCHDOG_HANDLE_BOOT_ENABLED: Avoid possible reboot if hw watchdog was
>>>>   been enabled before the kernel (by uboot for example) and userspace
>>>>   doesn't take control of /dev/watchdog in time;
>>>> - WATCHDOG_OPEN_TIMEOUT: Reboot if userspace doesn't take control of
>>>>   /dev/watchdog within the timeout.
>>>>
>>>> Signed-off-by: Francesco Zanella <francesco.zanella@xxxxxxxxx>
>>>> ---
>>>>  drivers/watchdog/gpio_wdt.c | 6 +++++-
>>>>  1 file changed, 5 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/watchdog/gpio_wdt.c b/drivers/watchdog/gpio_wdt.c
>>>> index 0923201ce874..1e6f0322ab7a 100644
>>>> --- a/drivers/watchdog/gpio_wdt.c
>>>> +++ b/drivers/watchdog/gpio_wdt.c
>>>> @@ -31,6 +31,7 @@ struct gpio_wdt_priv {
>>>>  	struct gpio_desc	*gpiod;
>>>>  	bool			state;
>>>>  	bool			always_running;
>>>> +	bool			start_at_boot;
>>>>  	unsigned int		hw_algo;
>>>>  	struct watchdog_device	wdd;
>>>>  };
>>>> @@ -147,6 +148,9 @@ static int gpio_wdt_probe(struct platform_device *pdev)
>>>>  	priv->always_running = of_property_read_bool(np,
>>>>  						     "always-running");
>>>>  
>>>> +	priv->start_at_boot = of_property_read_bool(np,
>>>> +						    "start-at-boot");
>>>> +
>>>>  	watchdog_set_drvdata(&priv->wdd, priv);
>>>>  
>>>>  	priv->wdd.info		= &gpio_wdt_ident;
>>>> @@ -161,7 +165,7 @@ static int gpio_wdt_probe(struct platform_device *pdev)
>>>>  
>>>>  	watchdog_stop_on_reboot(&priv->wdd);
>>>>  
>>>> -	if (priv->always_running)
>>>> +	if (priv->always_running || priv->start_at_boot)
>>>>  		gpio_wdt_start(&priv->wdd);
>>>
>>> So the only real difference to always_running is that always_running
>>> doesn't stop the watchdog on close but keeps it running.
>>>
>>> Does that really warrant another property ? Why not just use
>>> always-running ?
>>>
>>> The special use case of being able to stop the watchdog doesn't seem
>>> to be worth the trouble. Please explain your use case.
>>>
>>> Guenter
>>>
>>
>> You got the point.
>> I would like to be able to stop the watchdog when I want.
>> From my point of view always_running is used in the very special
>> case when the wdt chip can't be stopped.
>> I want a normal wdt that can be stopped (for example during a system
>> update), but I want it to start at boot for the 2 reasons that I
>> explained before:
>> - I need continuity in hw wdt pinging because I start in uboot,
>>   so I take advantage of WATCHDOG_HANDLE_BOOT_ENABLED that makes
>>   the kernel to do that job; but it is triggered only if it is
>>   started at probe, if I'm not wrong.
> 
> That depends. If the driver can read the current state (ie if
> it can detect if the watchdog is running) it can report it
> to the watchdog core accordingly. That would be the preferred
> mechanism. Everything else is just a workaround for bad hardware
> (bad as in "it doesn't report its state").
> 
> Guenter
> 
>> - I would like to monitor with the wdt also the kernel at boot,
>>   and more importantly handle the case when the userspace app
>>   doesn't take control of /dev/watchdog for whatever reason
>>   within the timeout set with WATCHDOG_OPEN_TIMEOUT.
>>
>> Hope that this explain my target.
>> Thanks for your attention,
>>
>> Francesco

Right, I agree with you that's the optimal way, but we are talking
about a low cost, simple GPIO WDT, that has only an input GPIO as
an interface with the system... how can it report its state if the
only way to enable/disable it is a particular state of the GPIO
that we are going to pilot?
I think that this driver was born for that kind of low cost chips
(I'm using a SGM706 for example), why can't we let it take
advantage of a useful kernel mechanism?

Thank you,

Francesco