For i386 with TSC, the kernel calibrates how much CPU cycles will fit
between two timer interrupts. That value corresponds to 10000
microseconds. Ideally.
In practice however the timer interrupts do not happen exactly every
10000 us (for hardware reasons). When interpolating time between ticks
that calibration value is used.
When using NTP (or adjusting tick manually) the value added every tick
may be different from 10000us.
If that value is larger, the time seems to jump ahead at the beginning
of each tick; if the value is smaller, the time may seem to get stuck,
get slow, or jump back at the beginning of a new tick.
Therefore I added experimental code to scale the value used for tick
interpolation according to these corrections. As it seems to me, the
clock quality improves, and the performance penalty only appears when
the correction value changes.
I haven't done the non-TSC case or other architectures. For
microseconds it may seem neglectible, but not for nanoseconds.
If anybody has an interesting opinion on this, please Mail.
Regards,
Ulrich
P.S. Not subscribed here.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue May 15 2001 - 21:00:23 EST