Hi Richard,

On Fri, 2022-03-04 at 12:35 +0100, Richard Weinberger via Xenomai
wrote:
> Hello,
> 
> I'm investigating into a cyclictest issue where it reports negative values 
> just like this commit states:
> https://source.denx.de/Xenomai/xenomai/-/commit/3069a7bbc48559f4d2a6184d6159b771f193a4e9

Ipipe or Dovetail? Does running "autotune" help?

> 
> Since I'm on Xenomai 3.1.2 the said commit is already included.
> But I have a hard time to understand what problem the commit actually fixes.
> Why is the old calculation wrong?

old code: In case the second wrapped we have
ts1.nsec = <small> (we consider t1 to be the later point in time)
ts2.nsec = <big>

diff += t1 - t2 is now negative. Right?

> 
> PREEMPT_RT's rt-test does it the same way:
> https://git.kernel.org/pub/scm/utils/rt-tests/rt-tests.git/tree/src/include/rt-utils.h#n60
> 
> Except that it casts tv_secs/tv_nsecs to int first, which shouldn't make a 
> difference.
> 
> Thanks,
> //richard
> 

Reply via email to