| Message ID | 20200314002743.3782677-1-andriin@fb.com |
|---|---|
| State | Accepted |
| Delegated to: | BPF Maintainers |
| Headers | show |
| Series | [v2,bpf-next] selftests/bpf: fix nanosleep for real this time | expand |
On Fri, Mar 13, 2020 at 05:27:43PM -0700, Andrii Nakryiko wrote: > Amazingly, some libc implementations don't call __NR_nanosleep syscall from > their nanosleep() APIs. Hammer it down with explicit syscall() call and never > get back to it again. Also simplify code for timespec initialization. > > I verified that nanosleep is called w/ printk and in exactly same Linux image > that is used in Travis CI. So it should both sleep and call correct syscall. > > v1->v2: > - math is too hard, fix usec -> nsec convertion (Martin); > - test_vmlinux has explicit nanosleep() call, convert that one as well. Acked-by: Martin KaFai Lau <kafai@fb.com>
On 3/14/20 1:27 AM, Andrii Nakryiko wrote: > Amazingly, some libc implementations don't call __NR_nanosleep syscall from > their nanosleep() APIs. Hammer it down with explicit syscall() call and never > get back to it again. Also simplify code for timespec initialization. > > I verified that nanosleep is called w/ printk and in exactly same Linux image > that is used in Travis CI. So it should both sleep and call correct syscall. > > v1->v2: > - math is too hard, fix usec -> nsec convertion (Martin); > - test_vmlinux has explicit nanosleep() call, convert that one as well. > > Fixes: 4e1fd25d19e8 ("selftests/bpf: Fix usleep() implementation") > Signed-off-by: Andrii Nakryiko <andriin@fb.com> Applied, thanks!
diff --git a/tools/testing/selftests/bpf/prog_tests/vmlinux.c b/tools/testing/selftests/bpf/prog_tests/vmlinux.c index 04939eda1325..72310cfc6474 100644 --- a/tools/testing/selftests/bpf/prog_tests/vmlinux.c +++ b/tools/testing/selftests/bpf/prog_tests/vmlinux.c @@ -11,7 +11,7 @@ static void nsleep() { struct timespec ts = { .tv_nsec = MY_TV_NSEC }; - (void)nanosleep(&ts, NULL); + (void)syscall(__NR_nanosleep, &ts, NULL); } void test_vmlinux(void) diff --git a/tools/testing/selftests/bpf/test_progs.c b/tools/testing/selftests/bpf/test_progs.c index f85a06512541..dc12fd0de1c2 100644 --- a/tools/testing/selftests/bpf/test_progs.c +++ b/tools/testing/selftests/bpf/test_progs.c @@ -35,16 +35,12 @@ struct prog_test_def { */ int usleep(useconds_t usec) { - struct timespec ts; - - if (usec > 999999) { - ts.tv_sec = usec / 1000000; - ts.tv_nsec = usec % 1000000; - } else { - ts.tv_sec = 0; - ts.tv_nsec = usec; - } - return nanosleep(&ts, NULL); + struct timespec ts = { + .tv_sec = usec / 1000000, + .tv_nsec = (usec % 1000000) * 1000, + }; + + return syscall(__NR_nanosleep, &ts, NULL); } static bool should_run(struct test_selector *sel, int num, const char *name)
Amazingly, some libc implementations don't call __NR_nanosleep syscall from their nanosleep() APIs. Hammer it down with explicit syscall() call and never get back to it again. Also simplify code for timespec initialization. I verified that nanosleep is called w/ printk and in exactly same Linux image that is used in Travis CI. So it should both sleep and call correct syscall. v1->v2: - math is too hard, fix usec -> nsec convertion (Martin); - test_vmlinux has explicit nanosleep() call, convert that one as well. Fixes: 4e1fd25d19e8 ("selftests/bpf: Fix usleep() implementation") Signed-off-by: Andrii Nakryiko <andriin@fb.com> --- tools/testing/selftests/bpf/prog_tests/vmlinux.c | 2 +- tools/testing/selftests/bpf/test_progs.c | 16 ++++++---------- 2 files changed, 7 insertions(+), 11 deletions(-)