[ntp:questions] Linux kernel wedging 30-Jun UTC in adjtimex()/ntp_adjtime()

Dave Hart hart at ntp.org
Sun Jul 1 03:29:29 UTC 2012


On Sat, Jun 30, 2012 at 21:43 UTC, Dave Hart <hart at ntp.org> wrote:
> Linux kernels 2.6.26 and later are reported at risk of livelock the
> day preceding the leap insertion.  I haven't found the first version
> number with the fix.

http://news.ycombinator.com/item?id=4183122 indicates 3.4 is the first
corrected version, and links to the change that introduced the bug in
2008, its fix in mid-March, and the integration of that fix into the
mainstream Linux kernel at the end of March.

>  This is distinct from the earlier bug logging
> the actual insertion event.
>
> http://serverfault.com/questions/403732/anyone-else-experiencing-high-rates-of-linux-server-crashes-today
>
> and mentioned there:
>
> http://lkml.indiana.edu/hypermail/linux/kernel/1203.1/04598.html
>
> The Red Hat knowledge base article wasn't updated a moment ago when I checked.

Separately there are reports of servers running Java spinning
consuming CPU starting at the leap insertion.  Stepping the clock
marginally (by reading then setting the clock to the reading) is
reported to fix it, even without restarting Java or ntpd.

Cheers,
Dave Hart


More information about the questions mailing list