at91: watchdog: Can't get watchdog to trigger

[Yang, Wenyou]

> -----Original Message-----
> From: Bryan Evenson [mailto:bevenson at melinkcorp.com]
> Sent: 2013年8月17日 4:37
> To: Yang, Wenyou; linux-arm-kernel at lists.infradead.org
> Cc: Ferre, Nicolas
> Subject: RE: at91: watchdog: Can't get watchdog to trigger
> 
> 
> 
> > -----Original Message-----
> > From: Yang, Wenyou [mailto:Wenyou.Yang at atmel.com]
> > Sent: Thursday, August 15, 2013 10:23 PM
> > To: Bryan Evenson; linux-arm-kernel at lists.infradead.org
> > Cc: Ferre, Nicolas
> > Subject: RE: at91: watchdog: Can't get watchdog to trigger
> >
> >
> >
> > > -----Original Message-----
> > > From: Bryan Evenson [mailto:bevenson at melinkcorp.com]
> > > Sent: 2013年8月15日 9:42
> > > To: linux-arm-kernel at lists.infradead.org
> > > Cc: Ferre, Nicolas; Yang, Wenyou
> > > Subject: at91: watchdog: Can't get watchdog to trigger
> > >
> > > All,
> > >
> > > I am using Atmel's stable branch of the Linux kernel
> > > (https://github.com/linux4sam/linux-at91.git, branch linux-3.6.9-
> > at91,
> > > back at commit 5fe012d4aee98ef82df66e6763882c735902ed0f) on a
> > > AT91SAM9G25 based board.  I am having issues getting the watchdog to
> > > trigger with this version of the kernel.
> > >
> > > First, I've verified that the watchdog is enabled in the bootstrap
> > and
> > > U-Boot.  I see the line:
> > >
> > > at91sam9_wdt: enabled (heartbeat=15 sec, nowayout=0)
> > >
> > > during boot showing that the watchdog is enabled.  I've tried
> > > triggering the watchdog by writing a single character to it from the
> > > command line (echo "0" >
> > > /dev/watchdog) and by using the two watchdog sample applications
> > under
> > > Documentation/watchdog.  In every case I see the line:
> > >
> > > watchdog watchdog0: watchdog did not stop!
> > >
> > > But the board does not reset.  Just to make sure, I also set
> > > CONFIG_WATCHDOG_NOWAYOUT in the kernel configuration and I still
> have
> > > the same results.
> > >
> > > I have my kernel configuration and the device tree files for my build
> > > over at https://github.com/evensonbryan/at91sam9g25-dt.git.  Other
> > > than the device tree files and the configuration, I have not made any
> > > kernel modifications.  What do I have setup wrong?
> >
> > This is the at91_wdt controller inherent property, once the watchdog is
> > started, it can't be stopped.
> > Because the Watchdog Mode Register (WDT_MR) can be written only once.
> > Only a processor reset resets it, as said in the datasheet.
> >
> > So the driver is implemented as following, 1./ If the \dev\watchdog
> > device is opened by the user space application, then it is the
> > application's responsibility to feed the watchdog. If the application
> > does not feed the watchdog, it will trigger RESET.
> > 2./ If the device isn't opened or close by the application, The
> > responsibility to feed the watchdog is turned to the driver.
> >
> > For your case,
> > echo "0" > /dev/watchdog, it will arise the following actions 1./ open
> > \dev\watchdog device, -->The responsibility to feed the watchdog is
> > turned to the user space application 2./ start the watchdog controller
> > 3./ ping the watchdog controller 4./ stop the watchdog controller -->
> > Since the watchdog can't be stopped, so output: watchdog watchdog0:
> > watchdog did not stop!
> > 5./ close \dev\watchdog device --> The responsibility to feed the
> > watchdog is turned to the Kernel Space
> >
> > So, the watchdog doesn't trigger the RESET.
> >
> > Documentation/watchdog/src/ watchdog-simple.c: after opening
> > \dev\watchdog device, it is always feed the watchdog with the 10
> > seconds interval.
> > Documentation/watchdog/src/ watchdog-test.c: after opening
> > \dev\watchdog device, it is always ping the watchdog with the 1 seconds
> > interval.
> 
> This is different than what I've seen with other versions of the kernel.  In the
> past, if my application didn't write the magic character to the watchdog before
> closing /dev/watchdog (or if it didn't close /dev/watchdog), the watchdog
> would reset the system.  For example, if I run the watchdog-simple.c test
> under my 2.6.39 kernel and ctrl-c to kill it, the watchdog resets my board.  
I double checked it, ctrl-c will invoke watchdog_release to close the /dev/watchdog. 

> But,
> with this 3.6.9 kernel that action does not reset my board.  I did just run a test
> where I opened /dev/watchdog, wrote one character to it, and started and
> infinite loop.  In that case, my board reset as it should.
> 
> Is this a intentional change in how the watchdog interacts with userspace, or
> was it always supposed to work the way it does in the 3.6.9 kernel?
I think it will work in this way. It is more reasonable.

Best Regards,
Wenyou Yang

> 
> Thanks,
> Bryan
> 
> >
> > Bye
> >
> > Wenyou Yang
> >
> > >
> > > Thanks,
> > > Bryan