[PATCHv2 5/5] ARM: picoxcell: implement watchdog restart

Jamie Iles jamie at jamieiles.com
Sat Dec 17 19:05:23 EST 2011


Hi Arnd,

On Sat, Dec 17, 2011 at 09:17:00PM +0100, Arnd Bergmann wrote:
> On Saturday 17 December 2011 13:42:37 Jamie Iles wrote:
> > Allow the platform to be restarted by triggering the watchdog to expire
> > with the shortest possible expiry.  This should reset the CPU core and
> > all on-chip peripherals.
> > 
> > Signed-off-by: Jamie Iles <jamie at jamieiles.com>
> 
> Hi Jamie,
> 
> > @@ -23,6 +24,22 @@
> >  
> >  #include "common.h"
> >  
> > +#define WDT_CTRL_REG_EN_MASK	(1 << 0)
> > +#define WDT_CTRL_REG_OFFS	(0x00)
> > +#define WDT_TIMEOUT_REG_OFFS	(0x04)
> > +static void __iomem *wdt_regs;
> > +
> > +static void picoxcell_setup_restart(void)
> > +{
> > +	struct device_node *np = of_find_compatible_node(NULL, NULL,
> > +							 "snps,dw-apb-wdg");
> > +	if (WARN(!np, "unable to setup watchdog restart"))
> > +		return;
> > +
> > +	wdt_regs = of_iomap(np, 0);
> > +	WARN(!wdt_regs, "failed to remap watchdog regs");
> > +}
> 
> Can you do this at restart time? I think it would be nicer to
> avoid the global variable and init function and instead have
> everything in a single function call for restart
> 
> > +static void picoxcell_wdt_restart(char mode, const char *cmd)
> > +{
> > +	/*
> > +	 * Configure the watchdog to reset with the shortest possible timeout
> > +	 * and give it chance to do the reset.
> > +	 */
> > +	if (wdt_regs) {
> > +		__raw_writel(WDT_CTRL_REG_EN_MASK, wdt_regs + WDT_CTRL_REG_OFFS);
> > +		__raw_writel(0, wdt_regs + WDT_TIMEOUT_REG_OFFS);
> > +		mdelay(500);
> > +	}
> > +}
> 
> Does this work if the real watchdog is still running, on a preemtible kernel?

Yes, this sets the timeout period to about 3us on this platform so 
there's no way the driver should be pinging it that fast even if it was 
still active.

> Also, please always use writel_relaxed instead of __raw_writel, and try to use
> msleep instead of mdelay by default. If msleep doesn't work here because it
> actually causes the system not to be reset, just add a comment.

OK fair point.  I was a bit wary of doing too much in the actual reboot 
time as it wasn't clear what services are available at that time, but 
all of your changes work so I'll go ahead and implement them.

Thanks,

Jamie



More information about the linux-arm-kernel mailing list