Re: [gpsd-dev] Port to QNX: PPS without SHM?

[Hal Murray]

address@hidden said:
> Hal, Since I last wrote I spent a lot of time getting familiar with ntpd
> configuration and I was successful getting it to lock well using gpsd ng
> (refclock 46) and the "atomic" pps (refclock 22). However in this
> configuration we are seeing ntpd lock up about once a day, stuck on a  mutex
> lock. The problem is correlated with use of the gpsd ng refclock.  Have you
> ever encountered this sort of problem and/or do you have any  strategies for
> tracking it down? There appear to be many many places in  the ntpd codebase
> where mutex locks are used. 

I haven't seen anything like that.  Can you get a stack trace?  If you want 
to run ntpd from gdb, you have to use the -n flag.


> I should point out that at this point, both gpsd and ntpd have local
> modifications. It might make sense for me to back out my PPS mods to  gpsd,
> for example, to see if that makes any difference. The ntpd mods so  far
> should be just steps necessary for porting. 

Nothing you do to gpsd should be able to make ntpd hang.

ntpd is essentially single threaded so "stuck on a  mutex" is pretty 
suspicious since there aren't many of them.  I assume you are talking about 
sem_wait or sem_timedwait.  They only occur once in ntp-dev.  That's in the 
handoff to/from another thread to do DNS lookups.


-- 
These are my opinions.  I hate spam.