Linux stores time internally, regardless of your hardware clock (a.k.a. RTC). That means your system can show one time when you run
date (Linux clock), and a different time when you run
hwclock (hardware clock).
Usually, you would want to load the time from the hardware to Linux when the machine boots (with
hwclock -hctosys), and when the machine goes down, you want to store your pretty accurate time (you do use
ntpd, don't you? ;)) - back to the hardware clock, with
Now, what happens if your system dies, reboots abnormally, etc? The clock doesn't get synchronized to hardware. On next boot, you might have a large clock skew, because the hardware clock is not that accurate... next time you'll say "I need my system to start with
ntpd, because otherwise,
ntpd might not sync the time at all, because the delta on the current clock and the real time is too big". The problem with that approach is... what happens if you happen to sync to a machine which by itself is not in sync, and then your
ntp will never have the right time?
So in order to avoid all that, it's good to keep the right time synchronized to the hardware at all times. What the kernel option you're asking about seems to do, is to note if your time is actually synchronized with NTP (there is a way to know that...) - and if it is - sync that time regularly from Linux time (a.k.a. System Time), to the hardware clock, so it will be very accurate at all times, including sudden system crashes.