Re: new NTP/PPS patch available

Theodore Y. Ts'o (
Fri, 6 Dec 1996 14:56:16 -0500

From: (Harald Koenig)
Date: Fri, 6 Dec 1996 09:30:07 +0100 (MET)

> No, I was talking about adding such an option. Yes, the disadvantage is
> that you will be taking a time-stamp for every interrupt on that irq
> (instead of just when the carrier detect line changes). But the place
> where you are doing the time-stamp will be much cleaner, and result in
> more accurate times for the PPS calculations.

first for CPUs without cycle counter where you have to read the timer
counter etc. it's an absolute nono because of the slow timer access
(at least 5 or more in/outs)

I'm assuming here that the irq isn't being shared with any other serial
port, and is only being used for the radio clock. In that case, you're
going to be have to read the time from the counter, etc. in any case.
The only question is where you do it. You might as well do it as close
to the beginning of the interrupt processing as possible.

Note that this is on a per-IRQ basis only! And I think it's reasonable
to assume that you won't be trying to share the serial port with any
other ports, since that's when you get into lots of problems with
variable amounts of time needed to poll several UARTs.

- Ted