Re: DCD as an alternate /dev device?

[AGC <agcarver+netbsd%acarver.net@localhost>]

On 8/24/2011 01:01, Alexander Carver wrote:
> So I think I'm almost to a solution for the PPS issue with ntpd. I
> actually got the PPS refclock working but I can't use it with the GPS
> refclock.
>
> One person mentioned that on his system running FreeBSD (on a Soekris),
> the TX/RX portion of the serial port is cuad1 (symlinked to /dev/gps0)
> but the DCD signal is on some type of GPIO (in his case elan-mmcr
> symlinked to /dev/pps0). However both of them point to the same physical
> serial port.
>
>
> Is there something similar in NetBSD?

A bit of extra information:

The system has /dev/ttyb symlinked to /dev/pps1 and /dev/gps1.

The current configuration of ntpd 4.2.6p3 (compiled from source) has:

#Pool NTP
server          us.pool.ntp.org prefer
server          us.pool.ntp.org prefer

#PPS_ATOM refclock flag2 DCD clear, flag3 kernel PPS
#opens /dev/pps1
server          127.127.22.1
fudge           127.127.22.1  flag2 1 flag3 1 refid PPS

This segment gets added if I try to use the GPS data as a clock:

#GPS_NMEA refclock mode 1 $GPRMC only
#opens /dev/gps1
server         127.127.20.1  minpoll 4 mode 1 noselect
fudge          127.127.20.1  time2 0.245 refid GPS


Adding the GPS_NMEA refclock kills PPS entirely.  It seems they're both 
trying to access the serial port simultaneously although PPS_ATOM only 
wants to see DCD and GPS_NMEA wants TX/RX.

From some of the discussions on the ntpd list it appears that there 
should be an ability to open the port twice.  However, the discussions 
it also appears that the port is actually split in two pieces by the 
underlying OS.  The TX/RX is available via one device and the signals 
(especially DCD) is available by a separate device.  One user commented 
that FreeBSD happens to perform this breakout.

I'm seeing the PPS_ATOM clock transmit something each second when it's 
actually working and I will need to investigate that.  But, other than 
that, I'm told that having both GPS_NMEA and PPS_ATOM pointing to the 
same physical port is common and normally works fine.  But in this case 
something is getting it stuck.

As of right now, I have enabled only PPS_ATOM and left GPS_NMEA fully 
disabled.  I would like to have GPS_NMEA working again so that I can 
always have some type of clock data even if the network to the other 
servers dies.  I'm just failing to understand why the two aren't working 
with each other when it is apparently a fairly common configuration.

NB:  It seems that dual access to a single port is also the behavior 
when the GPS_NMEA refclock has its PPS code activated (via flag3 for 
kernel PPS).  The PPS code tries to access the port again even though 
it's already open for GPS data.  Consequently the GPS data goes through 
but the PPS data via the kernel or direct dies.