Problems with gettimeofday_ns()

["Julio M. Merino Vidal" <jmerino at ac dot upc dot edu>]

Hello,

[ Please CC me any replies as I am not currently subscribed to this  
list. ]

I am using SystemTap to obtain traces of process scheduling.  This is  
on a machine with a SMT-enabled processor (a Cell).  To do this, I  
have added some markers to the kernel and then use small probes in  
stap to print a line mentioning when the probe fired, what was it and  
any parameters I may be interested in.

To add a timestamp to each line I was using gettimeofday_ns().  I  
have found, however, that I receive events with duplicate timestamps,  
which is OK if it is not supposed to be strict monotonically increasing.

But the big problem is that sometimes I receive events in the  
incorrect order!  That is, I place a marker B after a marker A in the  
code, and I know it is impossible for B to appear before A in a  
trace.  However, sometimes, I get B before A in the final trace with  
the timings (supposedly) reversed.  I also print the CPU that fired  
the events, and both are happening from the same virtual CPU, so the  
possibility of both CPUs using different timings is not relevant.

If I change the gettimeofday_ns call to get_cycles or to a custom  
counter instead, things go perfectly.

Any idea about why this is happening?  May it be a bug in stap, or am  
I missing something in the way gettimeofday works?

Thanks.