gethrtime () returns 14 digit integer

overberg · 08-06-2009, 03:48 AM

I am reading gethrtime () the moment an interrupt occurred. I expected a nanosec int value but got a 14 digit value.

Am I perhaps doing something wrong?

Thanks.

My code is as follows:

Code:

void get_timestamp {
 hrtime_t my_timestamp;
 
 my_timestamp = gethrtime();
 rtl_printf("Interrupt Timestamp = %lld\n", long long my_timestamp);
}

fantas · 08-06-2009, 08:01 AM

The result sounds about right to me. The documentation of gethrtime tells that it is the time that is counted since some arbitrary point in the past (I would take a guess probably means the system start), and so a 14-digit nano-second value would fit right in here.

Let's take a look at nanoseconds, they are the 1 / 1000*1000*1000 part of a second, so basically by removing the right-most 9 digits from a decimal number you can get the full seconds. 14 minus 9 digits, there are 5 digits left, leaving a range of 10000 to 99999 seconds. Divided by 3600 you have the hours, leaving a range of 2.8 to 27.8 hours. If you started your system sometimes between these value in the past, this would support my theory about being relative to the system start.

If you want to verify correct timing you could take a time snapshot, e.g.

Code:

starttime = gethrtime();

// do something

stoptime = gethrtime() - starttime; // stoptime = nanoseconds elapsed since starttime

overberg · 08-12-2009, 02:42 AM

Thanks for the reply.

Below is data recorded using your suggestion.
It is not impossible, but unlikely that the delay (difference) can be around 250 milliseconds. We have recorded delays of 2.5 milliseconds in the past using an oscillioscope.

Start Time - Stop Time = Difference
29239402799104, 29239656855264, 254056160
29240402964256, 29240657030176, 254065920
29241403127104, 29241657177504, 254050400
29242403292192, 29242657352192, 254060000
29243403457216, 29243657498400, 254041184
29244403620160, 29244657672544, 254052384

dmail · 08-12-2009, 06:54 AM

Whilst the values returned by the function are nanoseconds the resolution of the timer is dependant on the hardware that you are using. Which would suggest to me that your resolution is 254 milliseconds.

Quote:

Although the units of hi-res time are always the same
(nanoseconds), the actual resolution is hardware dependent.
Hi-res time is guaranteed to be monotonic (it won't go back-
ward, it won't periodically wrap) and linear (it won't occa-
sionally speed up or slow down for adjustment, like the time
of day can), but not necessarily unique: two sufficiently
proximate calls may return the same value.

fantas · 08-13-2009, 10:55 AM

This would be my conclusion as well. Could it be that you can set the timer resolution at any arbitrary time ? I *think* I recall having done this some time back on an rt-patched kernel, but can't recall it offhand now.