svn commit: r360033 - head/tests/sys/kqueue/libkqueue

[Kyle Evans]

On Thu, Apr 16, 2020 at 9:33 PM Ian Lepore <ian at freebsd.org> wrote:
>
> On Fri, 2020-04-17 at 02:22 +0000, Kyle Evans wrote:
> > Author: kevans
> > Date: Fri Apr 17 02:22:15 2020
> > New Revision: 360033
> > URL: https://svnweb.freebsd.org/changeset/base/360033
> >
> > Log:
> >   tests: kqueue: use a more precise timer for the NOTE_ABSTIME test
> >
> >   Originally noticed while attempting to run the kqueue tests under
> >   qemu-user-static, this apparently just happens sometimes when running in a
> >   jail in general -- the timer will fire off "too early," but it's really just
> >   the result of imprecise measurements (noted by cem).
> >
> >   Kicking this over to NOTE_USECONDS still tests the correct thing while
> >   allowing it to work more consistently; a basic sanity test reveals that we
> >   often end up coming in just less than 200 microseconds after the timer
> >   fired off.
> >
> >   MFC after:  3 days
> >
> > Modified:
> >   head/tests/sys/kqueue/libkqueue/timer.c
> >
> > Modified: head/tests/sys/kqueue/libkqueue/timer.c
> > ==============================================================================
> > --- head/tests/sys/kqueue/libkqueue/timer.c   Fri Apr 17 02:21:46 2020        (r360032)
> > +++ head/tests/sys/kqueue/libkqueue/timer.c   Fri Apr 17 02:22:15 2020        (r360033)
> > @@ -216,17 +216,17 @@ test_abstime(void)
> >  {
> >      const char *test_id = "kevent(EVFILT_TIMER, EV_ONESHOT, NOTE_ABSTIME)";
> >      struct kevent kev;
> > -    time_t start;
> > -    time_t stop;
> > -    const int timeout = 3;
> > +    long end, start, stop;
> > +    const int timeout_sec = 3;
> >
> >      test_begin(test_id);
> >
> >      test_no_kevents();
> >
> > -    start = time(NULL);
> > +    start = now();
> > +    end = start + SEC_TO_US(timeout_sec);
> >      EV_SET(&kev, vnode_fd, EVFILT_TIMER, EV_ADD | EV_ONESHOT,
> > -      NOTE_ABSTIME | NOTE_SECONDS, start + timeout, NULL);
> > +      NOTE_ABSTIME | NOTE_USECONDS, end, NULL);
> >      if (kevent(kqfd, &kev, 1, NULL, 0, NULL) < 0)
> >          err(1, "%s", test_id);
> >
> > @@ -235,10 +235,10 @@ test_abstime(void)
> >      kev.data = 1;
> >      kev.fflags = 0;
> >      kevent_cmp(&kev, kevent_get(kqfd));
> > -    stop = time(NULL);
> > -    if (stop < start + timeout)
> > -        err(1, "too early %jd %jd", (intmax_t)stop, (intmax_t)(start + timeout));
> >
> > +    stop = now();
> > +    if (stop < end)
> > +        err(1, "too early %jd %jd", (intmax_t)stop, (intmax_t)end);
> >      /* Check if the event occurs again */
> >      sleep(3);
> >      test_no_kevents();
>
> Not caused by you, but this change made me notice:  using 'long' to
> hold the number of microseconds since the unix epoch will only work on
> amd64 and arm64.  Everything involved with that needs to use uint64_t
> to work on all arches.
>
> -- Ian
>

Good point, I'll go fix that after arichardson's in-flight change near
these tests lands.

Thanks,

Kyle Evans