Re: [RFC PATCH v2 10/10] watchdog: bd70528: Initial support for ROHM BD70528 watchdog block

[Guenter Roeck]

On Mon, Jan 28, 2019 at 10:13:47AM +0200, Matti Vaittinen wrote:
> On Mon, Jan 28, 2019 at 10:00:35AM +0200, Matti Vaittinen wrote:
> > On Sat, Jan 26, 2019 at 08:36:14AM -0800, Guenter Roeck wrote:
> > > On 1/25/19 3:06 AM, Matti Vaittinen wrote:
> > > > +/* Max time we can set is 1 hour, 59 minutes and 59 seconds */
> > > > +#define WDT_MAX_MS ((2 * 60 * 60 - 1) * 1000)
> > > > +/* Minimum time is 1 second */
> > > > +#define WDT_MIN_MS 1000
> > > > +#define DEFAULT_TIMEOUT 60
> > > > +
> > > > +static int bd70528_wdt_probe(struct platform_device *pdev)
> > > > +{
> > > > +	struct bd70528 *bd70528;
> > > > +	struct wdtbd70528 *w;
> > > > +	int ret;
> > > > +	unsigned int reg;
> > > > +	struct watchdog_device *wdt;
> > > > +
> > > > +	bd70528 = dev_get_drvdata(pdev->dev.parent);
> > > > +	if (!bd70528) {
> > > > +		dev_err(&pdev->dev, "No MFD driver data\n");
> > > > +		return -EINVAL;
> > > > +	}
> > > > +	w = devm_kzalloc(&pdev->dev, sizeof(*w), GFP_KERNEL);
> > > > +	if (!w)
> > > > +		return -ENOMEM;
> > > > +	w->bd = bd70528;
> > > 
> > > Unless I am missing something, only the mutex and the regmap pointer
> > > are used from struct bd70528. With that in mind, it might be desirable
> > > to copy those pointers to struct wdtbd70528 to avoid the additional
> > > dereferencing at runtime.
> > 
> > You are not missing anyting. If we ever add support for another PMIC
> > variant we will probably be using also the chip_type. Untill then only
> > the mutex and regmap. (And maybe the of_node if we have any RTC
> > properties in DT). So at this point we just use regmap and mutex. I will
> > change this.
> 
> Oh, but actually we are also using the WDT state setting function
> provided by MFD. And this is taking the struct bd70528 as parameter. So
> maybe we can keep it like this afterall? 
> 
Ok.

Guenter

> > 
> > > > +	w->dev = &pdev->dev;
> > > > +
> > > > +	wdt = devm_kzalloc(&pdev->dev, sizeof(*wdt), GFP_KERNEL);
> > > > +	if (!wdt)
> > > > +		return -ENOMEM;
> > > > +
> > > 
> > > struct watchdog_device can be part of struct wdtbd70528. Two separate allocations
> > > are not necessary.
> > 
> > Correct. Thanks for pointing that out. I'll simplify this.
> > 
> > > > +	wdt->info = &bd70528_wdt_info;
> > > > +	wdt->ops =  &bd70528_wdt_ops;
> > > > +	wdt->min_hw_heartbeat_ms = WDT_MIN_MS;
> > > > +	wdt->max_hw_heartbeat_ms = WDT_MAX_MS;
> > > > +	wdt->parent = pdev->dev.parent;
> > > > +	wdt->timeout = DEFAULT_TIMEOUT;
> > > > +	watchdog_set_drvdata(wdt, w);
> > > > +	watchdog_init_timeout(wdt, 0, pdev->dev.parent);
> > > > +
> > > > +	ret = bd70528_wdt_set_timeout(wdt, wdt->timeout);
> > > > +	if (ret) {
> > > > +		dev_err(&pdev->dev, "Failed to set the watchdog timeout\n");
> > > > +		return ret;
> > > > +	}
> > > > +
> > > > +	mutex_lock(&bd70528->rtc_timer_lock);
> > > > +	ret = regmap_read(bd70528->chip.regmap, BD70528_REG_WDT_CTRL, &reg);
> > > > +	mutex_unlock(&bd70528->rtc_timer_lock);
> > > > +
> > > 
> > > I don't see the point for the above mutex locks. This is just reading a
> > > single register. regmap itself provides locking for that already.
> > 
> > It has a purpose here - but I'd better add a comment. We want to get the
> > initial state of WDG here. If the probe is executed when RTC time is being
> > set we may read the state just when RTC is (temporarily) disabling WDT -
> > and we might tell the WDT core that WDT is disabled - even if it is
> > actually enabled. The mutex prevents us from reading the WDT state when
> > RTC is being set.
> >  
> > Br,
> > 	Matti Vaittinen