[ntp:questions] leap announcement hangups and kernel discipline

Dave Hart hart at ntp.org
Mon Jul 2 15:36:22 UTC 2012


On Mon, Jul 2, 2012 at 10:28 AM, Marco Marongiu <brontolinux at gmail.com> wrote:
> Hi all
>
> I assume everyone has read the reports about the mess connected to the
> announcement of a leap second and bugs in Linux and Java. Anyway,
> reports suggest that the calls of adjtimex by ntpd, related to the leap
> second announcement, made the Linux kernel hang in heavy load conditions.
>
> My question is: would that have happened if the kernel discipline was
> disabled with "disable kernel"?

The Java problem likely would have happened either way, as one way or
another the clock would be stepped back one second.  What Linux calls
adjtimex in the kernel is known to ntpd, libc, and the rest of the
world as ntp_adjtime.  With "disable kernel" that interface is either
never called (older ntpd) or called only to set the frequency
correction to zero once at ntpd startup (changed in 4.2.7 sometime).
In either case, the Linux livelock couldn't occur as the faulty code
wouldn't be exercised.  The same is true regarding the older Linux
deadlock bug hit at the moment of insertion while trying to log the
message with the :60 reference.

Cheers,
Dave Hart


More information about the questions mailing list