On Thu, 28 Apr 2016, John Syne wrote:
> > On Apr 28, 2016, at 1:14 AM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
> > It's not a race condition. By stopping the kernel you make the clocksource
> > wrap around resulting in a negative delta.
> That makes sense, however it seems that clocksource wraps every 172 seconds,
> which doesnʼt make sense to me. Given a 32 bit number, I would expect
> clocksource to wrap every 2,147 seconds. What am I missing?
That would mean the clocksource runs at exactly one GHz. clocksource !=
timekeeping units.
The math here is:
nsec = delta * tk->mult + tk->xtime_nsec;
So if delta is large enough, nsec might become negative.
nsec >>= tk->shift;
So because nsec is s64 the compiler preserves the signedness.
Can you please print delta, tk->mult, tk->shift and tk->xtime_nsec for that
case?
Thanks,
tglx
