[PATCH] tests: posix: use cpu clock for sleep

[Yury Khrustalev]

On Mon, Feb 02, 2026 at 04:42:03PM +0100, Florian Weimer wrote:
> * Yury Khrustalev:
> 
> > On some emulated targets sleep may result in inconsistent wait
> > times which will lead to the failure of the tst-chmod test.
> >
> > To account for this we use the CLOCK_PROCESS_CPUTIME_ID clock ID
> > while also consuming CPU time by repeatedly calling clock_gettime.
> >
> > ---
> > Passes regression on aarch64 and x86. OK for trunk?
> > Base commit: 67494ceb5d
> >
> > ---
> >  posix/tst-chmod.c | 19 +++++++++++++++++--
> >  1 file changed, 17 insertions(+), 2 deletions(-)
> >
> > diff --git a/posix/tst-chmod.c b/posix/tst-chmod.c
> > index 80a69dddd4..620055080c 100644
> > --- a/posix/tst-chmod.c
> > +++ b/posix/tst-chmod.c
> > @@ -26,6 +26,7 @@
> >  #include <string.h>
> >  #include <unistd.h>
> >  #include <sys/stat.h>
> > +#include <time.h>
> >  
> >  #include <support/xunistd.h>
> >  
> > @@ -37,6 +38,20 @@
> >      goto fail;								      \
> >    } while (0)
> >  
> > +/* On some emulated targets sleep may result in inconsistent wait
> > +   times which will lead to the failure of this test.  To account
> > +   for this we use the CLOCK_PROCESS_CPUTIME_ID clock ID while
> > +   also consuming CPU time by repeatedly calling clock_gettime. */
> > +static void
> > +test_sleep (unsigned long seconds)
> > +{
> > +  struct timespec ts;
> > +  clock_gettime (CLOCK_PROCESS_CPUTIME_ID, &ts);
> > +  unsigned long end = ts.tv_sec + seconds;
> > +  while (ts.tv_sec < end)
> > +    clock_gettime (CLOCK_PROCESS_CPUTIME_ID, &ts);
> > +}
> 
> This can wait a very short time only if tv_nsec is close to the maximum.

Right, good point, I should amend this code to make it more accurate in
terms of nanoseconds.

> How accurate is CLOCK_PROCESS_CPUTIME_ID on emulated targets?

I thought that accuracy was not very important for this test as long as 
the delay is substantial for stat to see time difference.

> Should we
> busy-wait and also wait for the wall clock time to increase?

Calling clock_gettime() is a sort of busy-waiting. This functions is
supposed to work very much like sleep() on any normal system in terms
of observable time delays (save for nanosecond inaccuracy).

Thanks,
Yury