[ntp:questions] Re: NEMA driver on Fedora Core 3 or 4

David L. Mills mills at udel.edu
Sat Nov 19 15:05:45 UTC 2005


I have been a very shrill advocate to have the drivers cease using 
per-driver code in favor of the PPSAPI interface and refclock_atom.c 
driver, but with limited success. Be adivised the PPSAPI interface, with 
or without the atom driver, does not have to use kernel modifications 
and can survive nicely with whatever interface is available for the 
pariticular machine. That's the intended functionality of the PPSAPI 
interface. However, some drivers continue using platform-specific 
interfaces which have long since been deprecated. These drivers are not 
supported and will not work on machines without the supported interface.


Hal Murray wrote:
> My query from a month ago...
>>I've got a GPSClock 200.  The serial port is off by 500 or 600 ms.
>>That's running FC 4.  I've got a similar setup at work running FC 3.
>>It has the same quirk.
> I think I have figured out what's going on...
> The NMEA driver has this code in the nmea_receive routine:
> #ifdef HAVE_PPSAPI
>         /*
>          * If the PPSAPI is working, rather use its timestamps.
>          * assume that the PPS occurs on the second so blow any msec
>          */
>         if (nmea_pps(up, &rd_tmp) == 1) {
>                 pp->lastrec = up->tstamp = rd_tmp;
>                 pp->nsec = 0;
>         }
> #endif /* HAVE_PPSAPI */
> The new FC 3 and 4 boxes that I'm having troubles with don't
> have the PPS kernel mods so that code gets run on the good/normal
> case but not on my new kernels that are causing troubles.
> Is that a bug or feature?  Should other refclocks do the same thing?
> Is smashing the nsec the right thing?  Assuming the second is right,
> won't that make it look like all the NMEA samples arrive at exactly
> the right time?  How can you correct the clock if the samples all
> appear to arrive at the right time?  (I know, use the PPS/ATOM driver,
> but it needs to know that the seconds are close enough.)

More information about the questions mailing list