Ipmi(4) under NetBSD-5 seems to have problems

[buhrow%lothlorien.nfbcal.org@localhost (Brian Buhrow)]

        Hello.  I've been using the ipmi(4) driver since NetBSD-3 with good
success.  That is, until NetBSD-5.  Under NetBSD-5, with the new smp
polling changes, it seems that the impi(4) driver can't reliably talk to
the bmc hardware and get reasonable results.  Further, using the watchdog
timer on some bmc hardware causes the bmc hardware to reset, and the host
itself to crash.
        Without using the watchdog timer, the ipmi(4) driver reports spurious
out-of-range conditions on various sensors periodically.
Output such as the following is seen:

ipmi0: critical over limit on 'Fan 2A'
ipmi0: normal state on 'Fan 2A'

        Under NetBSD-4 on the same hardware, several examples of different
hardware, I see no such trouble.
        Has anyone else seen this issue, and if so, have you figured out a
fix?  It seems like either the polling interval is too quick under the new
timing framework of NetBSD-5, or the results are getting corrupted somehow
between the hardware and the driver.
        I've seen this under NetBSD/I386 and NetBSD/amd64, NetBSD-5.0-stable
as of May 15, 2009.  About to try with NetBSD-5-stable as of today, July
10, 2009, but don't have those results yet.

Any thoughts anyone might have on this would be greatly appreciated.
-thanks
-Brian