Re: [PATCH v1 1/1] driver: watchdog: Remove mtk_wdt_stop() in probe() to prevent the system freeze and it doesn't reboot by watchdog problem

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

On 12/29/20 12:39 AM, Freddy.Hsin wrote:
> On Fri, 2020-12-25 at 22:27 +0800, Guenter Roeck wrote:
>> On 12/24/20 11:30 PM, Freddy Hsin wrote:
>>> From: "freddy.hsin" <freddy.hsin@mediatek.com>
>>>
>>> Before user space daemon start to access the watchdog device,
>>> there is a time interval that watchdog is disabled in the
>>> original flow. If the system freezing at this interval, it
>>> cannot be rebooted by watchdog hardware automatically.
>>>
>>> In order to solve this problem, the watchdog hardware should be
>>> kept working, and start hrtimer in framework to ping it by
>>> setting max_hw_heartbeat_ms and HW_RUNNING used in
>>> watchdog_need_worker to determine whether the worker should be
>>> started or not
>>>
>>> Change-Id: I6a041b0922888a90011d7538ee804d80bc8d15ea
>>> Signed-off-by: freddy.hsin <freddy.hsin@mediatek.com>
>>> ---
>>>  drivers/watchdog/mtk_wdt.c |   22 ++++++++++++++++++----
>>>  1 file changed, 18 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/drivers/watchdog/mtk_wdt.c b/drivers/watchdog/mtk_wdt.c
>>> index d6a6393..62f08cd 100644
>>> --- a/drivers/watchdog/mtk_wdt.c
>>> +++ b/drivers/watchdog/mtk_wdt.c
>>> @@ -195,6 +195,19 @@ static int mtk_wdt_set_timeout(struct watchdog_device *wdt_dev,
>>>  	return 0;
>>>  }
>>>  
>>> +static void mtk_wdt_init(struct watchdog_device *wdt_dev)
>>> +{
>>> +	struct mtk_wdt_dev *mtk_wdt = watchdog_get_drvdata(wdt_dev);
>>> +	void __iomem *wdt_base;
>>> +
>>> +	wdt_base = mtk_wdt->wdt_base;
>>> +
>>> +	if (readl(wdt_base + WDT_MODE) & WDT_MODE_EN) {
>>> +		set_bit(WDOG_HW_RUNNING, &wdt_dev->status);
>>> +		mtk_wdt_set_timeout(wdt_dev, wdt_dev->timeout);
>>> +	}
>>> +}
>>> +
>>>  static int mtk_wdt_stop(struct watchdog_device *wdt_dev)
>>>  {
>>>  	struct mtk_wdt_dev *mtk_wdt = watchdog_get_drvdata(wdt_dev);
>>> @@ -266,16 +279,17 @@ static int mtk_wdt_probe(struct platform_device *pdev)
>>>  	mtk_wdt->wdt_dev.timeout = WDT_MAX_TIMEOUT;
>>>  	mtk_wdt->wdt_dev.max_timeout = WDT_MAX_TIMEOUT;
>>
>> No longer needed if max_hw_heartbeat_ms is set.
> 
> Got it. it will be removed in V2 patch
> 
>>
>>>  	mtk_wdt->wdt_dev.min_timeout = WDT_MIN_TIMEOUT;
>>> +	mtk_wdt->wdt_dev.max_hw_heartbeat_ms = (WDT_MAX_TIMEOUT - 1) * 1000;
>>
>> This needs explanation. Why WDT_MAX_TIMEOUT - 1 ?
> 
> In watchdog_need_worker(), there is a condition for starting worker (hm
> && watchdog_active(wdd) && t > hm), so I think in the case
> that watchdog_active is true, the t (timeout) should be greater than
> hm (max_hw_heartbeat_ms) or the worker will not work. Shouldn't I
> consider for this case?
> 

You are claiming that the watchdog core is buggy. Make your case there, not here.

Guenter

>>
>>>  	mtk_wdt->wdt_dev.parent = dev;
>>>  
>>> +	watchdog_set_drvdata(&mtk_wdt->wdt_dev, mtk_wdt);
>>> +
>>> +	mtk_wdt_init(&mtk_wdt->wdt_dev);
>>> +
>>
>> Setting the chip (with the maximum timeout) ...
>>
> Got it. mtk_wdt_init will be moved after watchdog_init_timeout in V2
> 
> BRs,
> FreddyHsin
> 
>>>  	watchdog_init_timeout(&mtk_wdt->wdt_dev, timeout, dev);
>>
>> just before initializing/setting the real timeout doesn't really make sense.
>>
>>>  	watchdog_set_nowayout(&mtk_wdt->wdt_dev, nowayout);
>>>  	watchdog_set_restart_priority(&mtk_wdt->wdt_dev, 128);
>>>  
>>> -	watchdog_set_drvdata(&mtk_wdt->wdt_dev, mtk_wdt);
>>> -
>>> -	mtk_wdt_stop(&mtk_wdt->wdt_dev);
>>> -
>>>  	watchdog_stop_on_reboot(&mtk_wdt->wdt_dev);
>>>  	err = devm_watchdog_register_device(dev, &mtk_wdt->wdt_dev);
>>>  	if (unlikely(err))
>>>
>>
>