Re: [RFC][PATCH] new timeofday core subsystem (v.A0)

[Ulrich Windl]

On 15 Sep 2004 at 11:48, john stultz wrote:

> NTP adjustments need to be applied smoothly and consistently as time
> progresses. Adjustment to the NTP parameters occur at do_adjtimex and at

Actually I think (constant) NTP adjustments need to be applied any time. That's 
why I even tried to correct the TSC interpolation between ticks (If you visit the 
PPSkit patch, that's the "nanoscale" thing). The "constant adjustment values" are 
updated regularly also (per tick, per second, per NTP update).

The basic thought was: If the time between two ticks is not 1/HZ the TSC 
interpolation (that is calibrated against the timer ticks) also will not 
interpolate between 1/HZ ticks. I assumed the TSC interpolations is wrong by the 
same percentage the timer interrupts are. Thus I tried to reversely apply the 
frequency error learnt through NTP even to the TSC interpolation (in addition to 
the time increments every tick). My code however has the problem that I ran out of 
significant bits during several muliplies and divides, while supporting that 
maximum imaginable frequency range for TSC. Even with the best oscillator, it 
won't reach single nanosecond precision as I don't have enough bits for them.

Regards,
Ulrich


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/