[PATCH bpf v2] selftests/bpf: skip nmi test when perf hw events are disabled

[PATCH bpf v2] selftests/bpf: skip nmi test when perf hw events are disabled

[Ilya Leoshkevich]

Some setups (e.g. virtual machines) might run with hardware perf events
disabled. If this is the case, skip the test_send_signal_nmi test.

Add a separate test involving a software perf event. This allows testing
the perf event path regardless of hardware perf event support.

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
---

v1->v2: Skip the test instead of using a software event.
Add a separate test with a software event.

 .../selftests/bpf/prog_tests/send_signal.c    | 33 ++++++++++++++++++-
 1 file changed, 32 insertions(+), 1 deletion(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/send_signal.c b/tools/testing/selftests/bpf/prog_tests/send_signal.c
index 67cea1686305..54218ee3c004 100644
--- a/tools/testing/selftests/bpf/prog_tests/send_signal.c
+++ b/tools/testing/selftests/bpf/prog_tests/send_signal.c
@@ -173,6 +173,18 @@ static int test_send_signal_tracepoint(void)
 	return test_send_signal_common(&attr, BPF_PROG_TYPE_TRACEPOINT, "tracepoint");
 }
 
+static int test_send_signal_perf(void)
+{
+	struct perf_event_attr attr = {
+		.sample_period = 1,
+		.type = PERF_TYPE_SOFTWARE,
+		.config = PERF_COUNT_SW_CPU_CLOCK,
+	};
+
+	return test_send_signal_common(&attr, BPF_PROG_TYPE_PERF_EVENT,
+				       "perf_sw_event");
+}
+
 static int test_send_signal_nmi(void)
 {
 	struct perf_event_attr attr = {
@@ -181,8 +193,26 @@ static int test_send_signal_nmi(void)
 		.type = PERF_TYPE_HARDWARE,
 		.config = PERF_COUNT_HW_CPU_CYCLES,
 	};
+	int pmu_fd;
+
+	/* Some setups (e.g. virtual machines) might run with hardware
+	 * perf events disabled. If this is the case, skip this test.
+	 */
+	pmu_fd = syscall(__NR_perf_event_open, &attr, 0 /* pid */,
+			 -1 /* cpu */, -1 /* group_fd */, 0 /* flags */);
+	if (pmu_fd == -1) {
+		if (errno == ENOENT) {
+			printf("%s:SKIP:no PERF_COUNT_HW_CPU_CYCLES\n",
+				__func__);
+			return 0;
+		}
+		/* Let the test fail with a more informative message */
+	} else {
+		close(pmu_fd);
+	}
 
-	return test_send_signal_common(&attr, BPF_PROG_TYPE_PERF_EVENT, "perf_event");
+	return test_send_signal_common(&attr, BPF_PROG_TYPE_PERF_EVENT,
+				       "perf_hw_event");
 }
 
 void test_send_signal(void)
@@ -190,6 +220,7 @@ void test_send_signal(void)
 	int ret = 0;
 
 	ret |= test_send_signal_tracepoint();
+	ret |= test_send_signal_perf();
 	ret |= test_send_signal_nmi();
 	if (!ret)
 		printf("test_send_signal:OK\n");
-- 
2.21.0

Re: [PATCH bpf v2] selftests/bpf: skip nmi test when perf hw events are disabled

[Andrii Nakryiko]

On Tue, Jul 16, 2019 at 3:56 AM Ilya Leoshkevich <iii@linux.ibm.com> wrote:
>
> Some setups (e.g. virtual machines) might run with hardware perf events
> disabled. If this is the case, skip the test_send_signal_nmi test.
>
> Add a separate test involving a software perf event. This allows testing
> the perf event path regardless of hardware perf event support.
>
> Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
> ---

LGTM!

Acked-by: Andrii Nakryiko <andriin@fb.com>

>
> v1->v2: Skip the test instead of using a software event.
> Add a separate test with a software event.
>
>  .../selftests/bpf/prog_tests/send_signal.c    | 33 ++++++++++++++++++-
>  1 file changed, 32 insertions(+), 1 deletion(-)
>
> diff --git a/tools/testing/selftests/bpf/prog_tests/send_signal.c b/tools/testing/selftests/bpf/prog_tests/send_signal.c
> index 67cea1686305..54218ee3c004 100644
> --- a/tools/testing/selftests/bpf/prog_tests/send_signal.c
> +++ b/tools/testing/selftests/bpf/prog_tests/send_signal.c
> @@ -173,6 +173,18 @@ static int test_send_signal_tracepoint(void)
>         return test_send_signal_common(&attr, BPF_PROG_TYPE_TRACEPOINT, "tracepoint");
>  }
>

[...]

Re: [PATCH bpf v2] selftests/bpf: skip nmi test when perf hw events are disabled

[Alexei Starovoitov]

On Tue, Jul 16, 2019 at 07:57:04AM -0700, Andrii Nakryiko wrote:
> On Tue, Jul 16, 2019 at 3:56 AM Ilya Leoshkevich <iii@linux.ibm.com> wrote:
> >
> > Some setups (e.g. virtual machines) might run with hardware perf events
> > disabled. If this is the case, skip the test_send_signal_nmi test.
> >
> > Add a separate test involving a software perf event. This allows testing
> > the perf event path regardless of hardware perf event support.
> >
> > Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
> > ---
> 
> LGTM!
> 
> Acked-by: Andrii Nakryiko <andriin@fb.com>

Applied, Thanks