selftests/bpf: fix test_progs send_signal flakiness with nmi mode
Subsystems (3)
PATCH 1/1 selftests/bpf: fix test_progs send_signal flakiness with nmi mode
Alexei observed that test_progs send_signal may fail if run with command line "./test_progs" and the tests will pass if just run "./test_progs -n 40".
I observed similar issue with nmi subtest failure and added a delay 100 us in Commit ab8b7f0cb358 ("tools/bpf: Add self tests for bpf_send_signal_thread()") and the problem is gone for me. But the issue still exists in Alexei's testing environment.
The current code uses sample_freq = 50 (50 events/second), which may not be enough. But if the sample_freq value is larger than sysctl kernel/perf_event_max_sample_rate, the perf_event_open syscall will fail.
This patch changed nmi perf testing to use sample_period = 1, which means trying to sampling every event. This seems fixing the issue.
--- tools/testing/selftests/bpf/prog_tests/send_signal.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)
tools/testing/selftests/bpf/prog_tests/send_signal.c +1 -5
@@ -76,9 +76,6 @@ static void test_send_signal_common(struct perf_event_attr *attr,if (CHECK (!skel,"skel_open_and_load" ,"skeleton open_and_load failed\n" )) goto skel_open_load_failure; -/* add a delay for child thread to ramp up */ -usleep (100 ); -if (!attr) { err =test_send_signal_kern__attach (skel);if (CHECK (err,"skel_attach" ,"skeleton attach failed\n" )) {
@@ -155,8 +152,7 @@ static void test_send_signal_perf(bool signal_thread)static void test_send_signal_nmi (bool signal_thread) {struct perf_event_attr attr = { - .sample_freq =50 , - .freq =1 , + .sample_period =1 , .type =PERF_TYPE_HARDWARE , .config =PERF_COUNT_HW_CPU_CYCLES , };
Signature · 1 line
2.17.1
Quoted 16 lines · Yonghong Song
Alexei observed that
test_progssend_signalmay fail if run with command line "./test_progs" and the tests will pass if just run "./test_progs -n 40".I observed similar issue with nmi subtest failure and added a delay 100 us in Commit
ab8b7f0cb358("tools/bpf: Add self tests forbpf_send_signal_thread()") and the problem is gone for me. But the issue still exists in Alexei's testing environment.The current code uses
sample_freq= 50 (50 events/second), which may not be enough. But if thesample_freqvalue is larger than sysctlkernel/perf_event_max_sample_rate, theperf_event_opensyscall will fail.
Quoted 7 lines
This patch changed nmi perf testing to use
sample_period= 1, which means trying to sampling every event. This seems fixing the issue.Fixes ab8b7f0cb358 ("tools/bpf: Add self tests for bpf_send_signal_thread()")Signed-off-by Yonghong Song <[email protected]>---
Good not to have to rely on arbitrary timeout!
Quoted 30 lines
tools/testing/selftests/bpf/prog_tests/send_signal.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-)tools/testing/selftests/bpf/prog_tests/send_signal.c +1
-6@@ -76,9 +76,6 @@ static void test_send_signal_common(struct perf_event_attr *attr,if (CHECK (!skel,"skel_open_and_load" ,"skeleton open_and_load failed\n" )) goto skel_open_load_failure; -/* add a delay for child thread to ramp up */ -usleep (100 ); -if (!attr) { err =test_send_signal_kern__attach (skel);if (CHECK (err,"skel_attach" ,"skeleton attach failed\n" )) {@@ -155,8 +152,7 @@ static void test_send_signal_perf(bool signal_thread)static void test_send_signal_nmi (bool signal_thread) {struct perf_event_attr attr = { - .sample_freq =50 , - .freq =1 , + .sample_period =1 , .type =PERF_TYPE_HARDWARE , .config =PERF_COUNT_HW_CPU_CYCLES , }; --2.17.1
Quoted 22 lines · Andrii Nakryiko
Quoted 20 lines · Yonghong Song
Alexei observed that
test_progssend_signalmay fail if run with command line "./test_progs" and the tests will pass if just run "./test_progs -n 40".I observed similar issue with nmi subtest failure and added a delay 100 us in Commit
ab8b7f0cb358("tools/bpf: Add self tests forbpf_send_signal_thread()") and the problem is gone for me. But the issue still exists in Alexei's testing environment.The current code uses
sample_freq= 50 (50 events/second), which may not be enough. But if thesample_freqvalue is larger than sysctlkernel/perf_event_max_sample_rate, theperf_event_opensyscall will fail.This patch changed nmi perf testing to use
sample_period= 1, which means trying to sampling every event. This seems fixing the issue.
Quoted 5 lines
Quoted 3 lines
Fixes ab8b7f0cb358 ("tools/bpf: Add self tests for bpf_send_signal_thread()")Signed-off-by Yonghong Song <[email protected]>---
Good not to have to rely on arbitrary timeout!
Indeed.
Applied. Thanks