| Message ID | 20220525113113.9811-1-chrubis@suse.cz |
|---|---|
| State | Accepted |
| Headers | show |
| Series | syscalls/timer_settime03: Scale interval with clock precision | expand |
On Wed, May 25, 2022 at 7:29 PM Cyril Hrubis <chrubis@suse.cz> wrote: > What the test does is to: > > - set initial expiration in the past > - set very small interval value > - expect the timer to overrun immediatelly many times > to trigger timer overrun counter overflow > > However the test has harcoded expectation that the kernel timers have > 1ns resolution. And while that is true for many modern hardware high > resolution timers are generally not always present. > > The test tried to cope with that by adding kernel requirement for > CONFIG_HIGH_RES_TIMERS=y however that does not necessarily mean that the > high resolution hardware is present or that the drivers are loaded. > This only means that the support has been compiled in the kernel. > > So instead of disabling the test when kernel timers have lower precision > we scale the timer interval so that the inverval length divided by the > timer precision is constant i.e. handler_delay. > > Fixes #925 > > Signed-off-by: Cyril Hrubis <chrubis@suse.cz> > Reviewed-by: Li Wang <liwang@redhat.com>
Hi Cyril, Li,
Reviewed-by: Petr Vorel <pvorel@suse.cz>
Could we merge this before release?
Do we need more testing? (some testing already done on by reported on
https://github.com/linux-test-project/ltp/issues/925)
Kind regards,
Petr
Hi! > Reviewed-by: Petr Vorel <pvorel@suse.cz> > > Could we merge this before release? I guess that this is pretty safe, it does not change the test behavior on high resolution timers and fixes false possitive without them. So I would vote for merging it now.
> > Could we merge this before release? > > I guess that this is pretty safe, it does not change the test behavior > on high resolution timers and fixes false possitive without them. So I > would vote for merging it now. > Yes, tested also good from my side. Pushed, thanks!
diff --git a/testcases/kernel/syscalls/timer_settime/timer_settime03.c b/testcases/kernel/syscalls/timer_settime/timer_settime03.c index 4597bf74e..a828f63d3 100644 --- a/testcases/kernel/syscalls/timer_settime/timer_settime03.c +++ b/testcases/kernel/syscalls/timer_settime/timer_settime03.c @@ -32,6 +32,8 @@ static timer_t timer; static volatile int handler_called, overrun, saved_errno; +static struct timespec realtime_resolution; + static void sighandler(int sig LTP_ATTRIBUTE_UNUSED) { struct itimerspec spec; @@ -61,6 +63,11 @@ static void setup(void) SAFE_SIGNAL(SIGUSR1, sighandler); SAFE_TIMER_CREATE(CLOCK_REALTIME, &sev, &timer); + + SAFE_CLOCK_GETRES(CLOCK_REALTIME, &realtime_resolution); + + tst_res(TINFO, "CLOCK_REALTIME resolution %lins", + (long)realtime_resolution.tv_nsec); } static void run(void) @@ -81,9 +88,9 @@ static void run(void) /* spec.it_value = now - 1.4 * max overrun value */ /* IOW, overflow will land right in the middle of negative range */ - spec.it_value.tv_sec -= handler_delay / 100000000; + spec.it_value.tv_sec -= (handler_delay / 100000000) * realtime_resolution.tv_nsec; spec.it_value.tv_nsec -= nsec; - spec.it_interval.tv_nsec = 1; + spec.it_interval.tv_nsec = realtime_resolution.tv_nsec; SAFE_TIMER_SETTIME(timer, TIMER_ABSTIME, &spec, NULL); while (!handler_called); @@ -115,10 +122,6 @@ static struct tst_test test = { .test_all = run, .setup = setup, .cleanup = cleanup, - .needs_kconfigs = (const char *[]) { - "CONFIG_HIGH_RES_TIMERS=y", - NULL - }, .tags = (const struct tst_tag[]) { {"linux-git", "78c9c4dfbf8c"}, {"CVE", "2018-12896"},
What the test does is to: - set initial expiration in the past - set very small interval value - expect the timer to overrun immediatelly many times to trigger timer overrun counter overflow However the test has harcoded expectation that the kernel timers have 1ns resolution. And while that is true for many modern hardware high resolution timers are generally not always present. The test tried to cope with that by adding kernel requirement for CONFIG_HIGH_RES_TIMERS=y however that does not necessarily mean that the high resolution hardware is present or that the drivers are loaded. This only means that the support has been compiled in the kernel. So instead of disabling the test when kernel timers have lower precision we scale the timer interval so that the inverval length divided by the timer precision is constant i.e. handler_delay. Fixes #925 Signed-off-by: Cyril Hrubis <chrubis@suse.cz> --- .../syscalls/timer_settime/timer_settime03.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-)