Problem with _TOD_Set

[rtems]

During  our tests we had  a problem updating the internal Time of Day. 
 If we do a  rtems_bsdnet_synchronize_ntp()   call, we verify   that  sometimes the date  returned from    rtems_clock_get()   has the ticks field  greater then 1000 ms/_TOD_Ticks_per_second. 

If  we watch the _TOD_Set   function  (c/src/exec/score/src/coretodset.c) we can observe that the test on ticks_until_next_second is based on the old _TOD_Current.ticks:

 -------------------------------------------------------------------------------------------------------------------

        ticks_until_next_second = _TOD_Ticks_per_second;

        if ( ticks_until_next_second > _TOD_Current.ticks )     <----

            ticks_until_next_second -= _TOD_Current.ticks;       <----

        _TOD_Current             = *the_tod;

        _TOD_Seconds_since_epoch = seconds_since_epoch;

        _TOD_Is_set              = TRUE;

        _TOD_Activate( ticks_until_next_second );

 ------------------------------------------------------------------------------------------------------------------

 After the test,  '_TOD_Current.ticks' is updated with the new value stored in the parameter  'the_tod'  but obviously  previous calculated 'ticks_until _next second'  is   uncorrelated with the new value  '_TOD_Current'.

 

We avoided the problem updating  '_TOD_Current'  before the test:

 --------------------------------------------------------------------------------------------------------------------

      _TOD_Current             = *the_tod;

      _TOD_Seconds_since_epoch = seconds_since_epoch;

      _TOD_Is_set              = TRUE;

     ticks_until_next_second = _TOD_Ticks_per_second;

      if ( ticks_until_next_second > _TOD_Current.ticks )

        ticks_until_next_second -= _TOD_Current.ticks;

      _TOD_Activate( ticks_until_next_second );

 -------------------------------------------------------------------------------------------------------------------

 Before the patch we are unable to obtain a NTP synchronization between different CPUs on Ethernet network better then 1 Sec 

After the patch we are capable to obtain synchronization in term of 1 tick system CPU clock between CPUs. 

 Can anyone confirm the problem or is  there an error in our analysis?

  

Thanks in advance.

 Fabrizio Pirovano

 THYSYS s.r.l.

    Milan    Italy

pirovano.thysys at tiscalinet.it

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/users/attachments/20020917/04295c1f/attachment.html>