On Thu, Oct 10, 2024 at 4:24 PM Ihor Solodrai <ihor.solodrai@xxxxx> wrote: > > Recently perf_link test started unreliably failing on libbpf CI: > * https://github.com/libbpf/libbpf/actions/runs/11260672407/job/31312405473 > * https://github.com/libbpf/libbpf/actions/runs/11260992334/job/31315514626 > * https://github.com/libbpf/libbpf/actions/runs/11263162459/job/31320458251 > > Part of the test is running a dummy loop for a while and then checking > for a counter incremented by the test program. > > Instead of waiting for an arbitrary number of loop iterations once, > check for the test counter in a loop and use get_time_ns() helper to > enforce a 100ms timeout. > > Signed-off-by: Ihor Solodrai <ihor.solodrai@xxxxx> > --- > tools/testing/selftests/bpf/prog_tests/perf_link.c | 14 ++++++++++++-- > 1 file changed, 12 insertions(+), 2 deletions(-) > > diff --git a/tools/testing/selftests/bpf/prog_tests/perf_link.c b/tools/testing/selftests/bpf/prog_tests/perf_link.c > index 3a25f1c743a1..c1f13d77c464 100644 > --- a/tools/testing/selftests/bpf/prog_tests/perf_link.c > +++ b/tools/testing/selftests/bpf/prog_tests/perf_link.c > @@ -4,8 +4,12 @@ > #include <pthread.h> > #include <sched.h> > #include <test_progs.h> > +#include "testing_helpers.h" > #include "test_perf_link.skel.h" > > +#define BURN_TIMEOUT_MS 100 > +#define BURN_TIMEOUT_NS BURN_TIMEOUT_MS * 1000000 > + > static void burn_cpu(void) > { > volatile int j = 0; > @@ -63,8 +67,14 @@ void serial_test_perf_link(void) > ASSERT_GT(info.prog_id, 0, "link_prog_id"); > > /* ensure we get at least one perf_event prog execution */ > - burn_cpu(); > - ASSERT_GT(skel->bss->run_cnt, 0, "run_cnt"); > + __u64 timeout_time_ns = get_time_ns() + BURN_TIMEOUT_NS; nit: this is not really a kernel C89 code style (even though we don't enforce that in selftests, but still), can you please move variable *declaration* up to the top of the function? > + while (1) { nit: true pw-bot: cr > + burn_cpu(); > + if (skel->bss->run_cnt > 0) > + break; > + if (!ASSERT_LT(get_time_ns(), timeout_time_ns, "run_cnt_timeout")) > + break; > + } > > /* perf_event is still active, but we close link and BPF program > * shouldn't be executed anymore > -- > 2.39.5 (Apple Git-154) >
