Timing question

["Jonas Persson" <jonas.persson@comsys.se>]

Hi,

I'm measuring the timing of my (RT-) processes using gethrtime(). Basically, the timing loop looks like this:

hrtime_t a,b,c;
pthread_make_periodic_np(pthread_self(),gethrtime(), periodicity);
while (1)
{
	a = gethrtime();
	pthread_wait_np();
	b = gethrtime();

	// do something useful
	c = gethrtime();

	// send a,b,c to user space via RTfifo
}

When comparing (b-a), which should be the time between context-switch-out and switch-in, I am getting unreasonable values - 42806400 ns - no matter how i set the period time of the process. I have tried with a very slow 99 Hz and a more reasonable 9900 Hz, the results are the same... what is happening here?

I am running this on an SMP machine, albeit only with one processor (866MHz PIII). gethrtimeres() reports 32. The linux kernel is compiled for SMP, so I guess an APIC is used for timing.

I tried to read the TSC with the following piece of code,

extern __inline__ unsigned long long int get_rdtsc()
{
  unsigned long long int x;
  __asm__ __volatile__ (".byte 0x0f, 0x31" : "=A" (x)) ;
  return x;
}

but I get even more bogus values using that. Is the TSC reset by linux or rtlinux during for example scheduling?

Regards, Jonas

______________________________________________________________________
 Jonas Persson                    E-mail: jonas.persson@comsys.se
 Project Manager, M.Sc C.Sc&E     http://www.comsys.se/
 Comsys AB                        Phone: +46-(0)46-286 3500
 Forskningsbyn IDEON              Fax: +46-(0)46-286 3505
 S-223 70 Lund, Sweden