time issues

YONETANI Tomokazu qhwt+dfly at les.ath.cx
Wed Oct 27 23:28:48 PDT 2004


On Wed, Oct 13, 2004 at 06:51:46AM -0700, walt wrote:
> On Tue, 12 Oct 2004, Matthew Dillon wrote:
> > :> Hi,
> > :>
> > :> Studbolt mentioned this on IRC and when I checked in the logs I found that I
> > :> have the same problem. It seems that the time is way off sometimes:
> > :
> > :> This seems to happen with or without TIMER_USE_1.
> > :>
> > :> Any ideas? Anyone experiencing the same?
> > :
> > :Now that you mention it, my machine is ten minutes fast now.  It was
> > :only two minutes fast early this morning.  No TIMER_USE_1 here.
> > :Today is the first day I've seen this, though.
> >
> >     I will begin running a comprehensive cron-based test on my
> >     test boxes to see if I can duplicate it.
> 
> The machine kept perfect time overnight, however.  I believe that the
> only 24-hour period I ever noticed the time problem is when I was
> running your vfs9 patch.  As of yesterday afternoon I went back to
> a HEAD kernel to test your latest commits, and so far no time problem.

I've seen something similar to this with HEAD just before the last big
slip of DragonFly_Stable to HEAD, but I haven't manageged to reproduce it
since then. I can't say whether the system clock was completely dead, or that
it just slowed down.

By the way, is TIMER_USE_1 safe for anyone? Only tested with SMP, but:

- I've only seen the slow down of system clock on kernel *without* TIMER_USE_1,
  but now I can't reproduce it. The system clock never goes fast *without*
  TIMER_USE_1.

- on kernel *with* TIMER_USE_1 on the same machine, the system clock
  goes crazy(about 10 times faster than the real time) a few minutes after
  staring buildworld. If I reboot without running ntpdate, the clock goes
  back to normal(I mean the hard clock doesn't seem to be affected).
  The machine has 4 logical CPUs(dual Xeon with HTT enabled). If I turn off
  HTT in the BIOS, the number of logical CPU becomes 2, and this time
  the system clock will not be as fast as it did with HTT enabled
  (like 5 times or so faster than the real time).





More information about the Bugs mailing list