https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=288345

--- Comment #32 from Craig Leres <[email protected]> ---
Yesterday I upgraded to 14.3-RELEASE-p8 and rebuilt and rebooted the
GENERIC-DEBUG kernel I had been running on my build server for the last few
days.

My build server has a cron job to poudriere build ~1400 package starting at
midnight. I've been using a nagios plugin that runs via nrpe to detect
persistent processes with DJ flags shown by ps. At 00:49 today the test flagged
a problem. Somehow it resolved six minutes later. Another problem was reported
at 1:22am, it cleared up eight minutes later. I checked syslog and the only
thing I see see is:

    Jan 28 00:04:34 zinc.ee.lbl.gov kernel: : witness_lock_list_get: witness
exhausted

which doesn't really match the first window. Is this at all interesting?

I think these are the things missing from GENERIC-DEBUG that were in my
standard custom kernel config:

    device        pps
    options       PPS_SYNC
    options       P1003_1B_MQUEUE
    device        disc
    device        pf
    device        pflog
    options       KDB_UNATTENDED
    options       DEVICE_POLLING
    options       DCONS_POLL_HZ=1000

Is it worth adding these to kernel I'm running here? I doubt making disc and pf
part of the kernel instead of using loadable drivers is a problem. And I don't
think the pps options are problematic, this isn't one of my systems running ntp
with a gps. Is it possible the deadlock is due to device polling? 

(Or were the DJ state processes just transients and I'm reading too much into
seeing them briefly exist?)

-- 
You are receiving this mail because:
You are the assignee for the bug.

Reply via email to