Hello. I've just filed kern/41724 with patches to fix the ipmi(4)
driver when using the watchdog timer. I've submitted patches for
NetBSD-3.x, NetBSD-4.x and NetBSD-5.x. These patches solve the problem of
making the watchdog timer crash the system when the tickler is unable to
communicate with it. The problem is that the watchdog timer functions do
not honor the locks which are used when accessing sensor data. By insuring
that only one set of functions is accessing the bmc boards at once, all
errors tickling the timers go away, and the rate of spurious out of range
conditions drops as well, although it doesn't go away entirely. (It looks
like a new fix was just committed to the ipmi(4) driver in -current that
might address that issue for once and all.)
So, if someone who has the authority to commit these patches to the
various trees could look at this bug, and make the commits, I'd be very
appreciative. Also, I imagine that all those people who've been trying to
use the ipmi watchdog timer in frustration will also be greatful. I can't
see how this ever worked since the ipmi(4) driver was imported. It may be
that it's even broken under OpenBSD.
In any case, these patches work for me under all three versions of the OS.
-thanks
-Brian