Re: VxWorks time()

[Ralph Lange <Ralph.Lange@bessy.de>]

Note that there's always been stuff in base to do the time sync:

In 3.14, the code in src/libCom/osi/os/vxWorks/iocClock.c will create a 
thread that does a regular NTP sync with the server specified by the 
environment var EPICS_TS_NTP_INET (default: boot server). Watch for a 
thread named "syncNTP".

In 3.13, drvTS was part of base, so calling "TSconfigure" with no 
arguments from the startup script (making the IOC a "soft timing slave") 
would have the IOC do a regular NTP sync with the server specified by 
the same environment var (default: boot server). Set "TSdriverDebug=10" 
on the vxWorks shell to see what it's doing.

The main purpose of generalTime is to allow for multiple time sources to 
be configured, with the IOC switching between the sources on 
availability to always automatically provide the most accurate time 
possible.

Cheers,
Ralph


Thompson, David H. wrote:
>  
> The time(0) will return the number of seconds that the IOC has been
> running unless you set it.  Unless the ioc has been up for a year ctime
> would convert that into a date in 1970 since time(0) is 1/1/70 00:00
> GMT.
>
> Out kernels make a call to sntpTimeGet() and then a call to
> clock_settime() before running the startup script.  That gets it right
> for a while but it will drift.  You can make these calls from the top of
> the startup script so long as you malloc a block of memory to hold a
> struct timespec to pass to the two calls.
>
> The generalTime driver may do the same thing.
>