watchdog: Add Broadcom BCM2708 watchdog timer driver

[Guenter Roeck]

On Sun, Mar 24, 2013 at 03:06:59PM +0100, Lubomir Rintel wrote:
> On Fri, 2013-03-22 at 06:56 -0700, Guenter Roeck wrote:
> 
> Thank you for your response!
> 
> On Fri Mar 22 09:56:01 EDT 2013, Guenter Roeck wrote:
> > On Fri, Mar 22, 2013 at 12:55:07PM -0000, Lubomir Rintel wrote:
> ...
> > > +	writel_relaxed(PM_PASSWORD | (cur & PM_RSTC_WRCFG_CLR) |
> > > +		  PM_RSTC_WRCFG_FULL_RESET, wdt_regs + PM_RSTC);
> > > +
> > Nitpick - I prefer people to use the recommended continuation line style,
> > but that is really up to the maintainer to decide.
> 
> Well, I intended to comply with Documentation/CodingStyle, are you referring to 
> it? I fail to understand what to do to be more compliant and could not really 
> identify a style that would be consistently used across the kernel source. 
> Should I cut then second line into two smaller parts that would be aligned with 
> right line end?
> 
I was referring to line continuation aligned with '(', such as

	writel_relaxed(PM_PASSWORD | (cur & PM_RSTC_WRCFG_CLR) |
		       PM_RSTC_WRCFG_FULL_RESET, wdt_regs + PM_RSTC);

I don't recall how deep the indentation was - does that not fit ?

> ...
> > > +static int bcm2835_wdt_set_timeout(struct watchdog_device *wdog, unsigned int t)
> > > +{
> > > +	wdog->timeout = t;
> > 
> > No need to update the actual chip timeout ?
> 
> No need to, watchdog core applies the new timeout by pinging the device (see 
> below for what happens when this driver is pinged).
> 
> See: WDIOC_SETTIMEOUT in drivers/watchdog/watchdog_dev.c
> 
Ok, makes sense.

> ...
> > > +static struct watchdog_ops bcm2835_wdt_ops = {
> > > +	.owner =	THIS_MODULE,
> > > +	.start =	bcm2835_wdt_start,
> > > +	.stop =		bcm2835_wdt_stop,
> > > +	.set_timeout =	bcm2835_wdt_set_timeout,
> > > +	.get_timeleft =	bcm2835_wdt_get_timeleft,
> > 
> > No separate ping function ?
> 
> The watchdog documentation core states:
> 
>   "Most hardware that does not support this as a separate function uses the
>   start function to restart the watchdog timer hardware. And that's also what
>   the watchdog timer driver core does."
> 
> This indeed applies to this driver.
> 
Ok.

> ...
> > > +	if (WARN(!wdt_regs, "failed to remap watchdog regs"))
> > > +		return -ENODEV;
> > 
> > WARN seems to be a bit extreme. Is this necessary ?
> 
> Probably not. I'll replace it with dev_err() instead.
> 
> > > +	dev_info(dev, "Broadcom BCM2835 watchdog timer");
> > > +
> > > +	watchdog_init_timeout(&bcm2835_wdt_wdd, heartbeat, dev);
> > 
> > Since heartbeat is by default set to -1, which is interpreted as unsigned
> > int, I would expect this call to return -EINVAL, leaving the default timeout
> > undefined. Is this really what you want ?
> 
> Well, I looked into orion-wdt for an example how to initialize the default 
> timeout, but failed to understand it correctly. I thought that watchdog core 
> picks a sensible value upon getting -1, which is incorrect. They in fact use 
> initialize timeout with maximal value, and use a fall-through vi EINVAL to leave 
> it untouched if it was not overridden. I'll do the same thing now.
> 
Some user level documentation states that the default timeout for all drivers
would be 60 seconds. Unfortunately, that is not correct, as all drivers do
what they want. I myself use it as a guideline, ie use a 60 seconds default
unless there is a good reason to use another default value (eg if the
chip only supports a lower maximum).

> > > +	watchdog_set_nowayout(&bcm2835_wdt_wdd, nowayout);
> > > +	return watchdog_register_device(&bcm2835_wdt_wdd);
> > 
> > Leaking iomap if this fails.
> 
> Oops. Fixing.
> 
> > Would be nice to have something like devm_of_iomap ...
> 
> That sounds sound to me. Sent out a separate patch implementing it, and I'll 
> modify this if it gets merged.
> 
Excellent!

Thanks,
Guenter