Re: [gpsd-dev] gpsd processor load

[Greg Troxel]

address@hidden (Eric S. Raymond) writes:

> Looks like you have a NetBSD-specific issue.  I've been watching gpsd
> in top for a while and it generally wanders between 0.3%-0.7% of CPU.
> I've never seen it go above 1.3%
>
> Given the time pattern of its inputs, gpsd spends most of every second
> waiting on select(2).  ckuethe is right; something likely is awry in
> your select.

Nope, it's not select.  I ran ktrace, and the thread that reads from the
device is fine.  The pps/shm thred is in a tight loop doing
gettimeofday.  The problem is that the loop in gpsd_ppsmonitor has very
little body content if both TIOCMIWAIT and SYS_TIMEPPS are both
undefined.  NetBSD has sys/timepps.h, but it needs sys/types.h, so it's
undefined.  As I read the code, it will be in an infinite tight loop
calling gettimeofday and then logging a debug message if neither method
is available.

The new code disables pps if TIOCMIWAIT is not present.  I'm not sure
that's right, as kpps might be able to work, but I will figure it out
before asking for that to change.

pgpp4S5mDQ0S9.pgp
Description: PGP signature