[LinuxPPS] Occasionally ntpd stops receiving NMEA data

[Udo van den Heuvel]

Philip M. White wrote:
> On Mon, Jun 05, 2006 at 07:47:18AM +0200, Udo van den Heuvel wrote:
>> Or?
> 
> Looks like both, because NMEA stops scrolling in ntpd debug mode, and
> ntpd marks the GPS device as being completely unusable.  If PPS was
> still available, I'd expect that ntpd would treat it in the same manner
> it treats the SHM driver: we can still synchronize with it, just not
> read the wall clock time.

Dunno how the NMEA driver treats just a PPS signal.
It expetcs a NMEA stri g to correlate the time to the PPS I would guess.

>> You have quite a lot of sentences enabled.
> 
> Yes. :)  I know I don't need all of them, but can't hurt.  I've always
> had them enabled.

Hmm. OK.

>> Please verify using the /proc interface.
>> # cat /proc/pps/00/assert
>> 1149486295.999996675 #478873
>>
>> (repeat, do you the numbers change?)
> 
> Indeed, the numbers do NOT change.  The 'assert' node's data is:
> 1149484412.1091000 #135
> where #135 is one greater than the last visible number in `dmesg'.

Hmm, no big deal I guess.
So the LinuxPPS api stops receiving PPS as well.
So it is not just ntpd that stops receiving.

>>> When I Ctrl-C the ntpd process (which is running in the foreground with
>>> the -d switch), it takes a while for the process to die, like it's stuck
>>> waiting for something.
>> That is abnormal?
> 
> When everything is working fine, pressing Ctrl-C ends ntpd instantly.
> So yes, this appears to be abnormal.

But there is no handshaking going on. Just two RS232 data lines and a
ground. So stopping should be very quick.

>> Dunno. I am still running the (patched now) versio I made for 2.6.16.x.
>> Please reduce the number of NMEA sentences. Just one should be enough.
> 
> Which one is required?  GPRMC?

I believe so, see the driver page for driver 20 I referenced a few mails
ago.

> Like I said earlier, I've *always* had these enabled without any
> problems.  The problems started after upgrading my LinuxPPS version, and
> the problem is pretty severe -- the GPS device in ntpd dies just minutes
> after starting it.  So, I couldn't have not noticed it before. :)

Strange. I just run that one sentence and a slightly different LinuxPPS
api (older..) but have never seen your issue.

>> (it is for me, I use the default sentence that the NMEA driver uses).
>> Please verify PPS using the /proc interface.
>> What kernel/npt are you using?
> 
> Right now I am using kernel 2.6.16.19 with LinuxPPS 2.1.0 that's
> available in Rodolfo's patches/ directory.  It patched, compiled, and
> loaded cleanly.

If you reboot to the older kernel (with oldr LinuxPPS) the problem goes
away?
If so you might have something to report to Rodolfo.