[Avila] watchdog on avila (GW-2348)

[Tim Harvey]

The ixp4xx watchdog is a CPU watchdog which does a chip reset but not a
peripheral reset (PCI, Flash, etc).  Being a CPU watchdog its pretty inferior
to a true watchdog in my experience as it will not detect CPU spins (which
KASSERT's in arm-linux are typically configured as I might mention)

The Gateworks Avila platform boards have a DS1819 used as a pushbutton
debouncer and reset monitor as well as a strobed watchdog.  The DS1819 resets
the board if not strobed within 1.12secs (which makes it somewhat tricky to
tickle quick enough and a bit difficult to use as a watchdog).  Note that the
various avila boards route the DS1819 differently (bummer).  On the GW2348 the
strobe is connected to GPIO14 via R11 which is unloaded by default.  GPIO14 is
configured by redboot as a clock, therefore you must unconfigure it as a clock
to strobe it manually.  On the GW2347 strobe is connected to GPIO12 through R92
which is unloaded by default (as GPIO14 is needed as a pciclock on this board)
and GPIO9 is a buffer enabling that line.  Note that if you load the resistors
to enable the strobe to the DS1819 you may end up with issues with the
Gateworks JTAG flasher as in some cases (I think the 2348 case) it won't tickle
the strobe while flashing (so you need to remove the resistor if you need to
JTAG it).

For my use I wrote a driver for the 2347/2348 boards that implemented the
standard watchdog API but then realized that the typical 1sec heartbeat from
userspace was typically not enough to keep the watchdog tickled reliably so I
implemented a heartbeat in my driver.  I still ended up with issues on certain
kernel versions during certain system calls where the timer in my kernel driver
was not getting enough attention.

Hope this helps,

Tim

--- Sebastian G <sebastian.gra at gmail.com> wrote:

> 2008/4/10, David Acker <dacker at roinet.com>:
> 
> > >
> >  We had been told the following:
> >  "The system reset (from our DS1819) reset generator is routed to both the
> > processor and the flash devices #RP pin. ... the board does contain a
> > hardware watchdog which can be enabled by loading R11 with a zero ohm
> > resistor (location on bottomside by CF connector). You will then need to
> > strobe CPU GPIO14."
> >
> >  We are using the ixp4xx_wdt.c watchdog in linux.  Can the folks at
> > gateworks confirm if this watchdog will reset the flash or if we need to
> use
> > a different one?
> >  -Ack
> >
> Yes, I have also found  that information about the other watchdog. For
> the time being we are using the ixp4xx_wdt watchdog but it does not
> seem to reset the flash. However, if there is another watchdog and it
> could be used and it is confirmed the otherone is better, then why the
> R11 resistor is not assembled by default? What about the software for
> DS1819 watchdog? It probably needs to be written, if it is not enabled
> by default?
> 
> Sebastian
> 
> _______________________________________________
> Avila mailing list
> Avila at lists.gateworks.com
> http://lists.gateworks.com/cgi-bin/mailman/listinfo/avila
>