[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20241018204109.713820-14-jolsa@kernel.org>
Date: Fri, 18 Oct 2024 22:41:09 +0200
From: Jiri Olsa <jolsa@...nel.org>
To: Oleg Nesterov <oleg@...hat.com>,
Peter Zijlstra <peterz@...radead.org>,
Alexei Starovoitov <ast@...nel.org>,
Daniel Borkmann <daniel@...earbox.net>,
Andrii Nakryiko <andrii@...nel.org>
Cc: bpf@...r.kernel.org,
Martin KaFai Lau <kafai@...com>,
Song Liu <songliubraving@...com>,
Yonghong Song <yhs@...com>,
John Fastabend <john.fastabend@...il.com>,
KP Singh <kpsingh@...omium.org>,
Stanislav Fomichev <sdf@...ichev.me>,
Hao Luo <haoluo@...gle.com>,
Steven Rostedt <rostedt@...dmis.org>,
Masami Hiramatsu <mhiramat@...nel.org>,
linux-kernel@...r.kernel.org,
linux-trace-kernel@...r.kernel.org
Subject: [PATCHv8 bpf-next 13/13] selftests/bpf: Add threads to consumer test
With recent uprobe fix [1] the sync time after unregistering uprobe is
much longer and prolongs the consumer test which creates and destroys
hundreds of uprobes.
This change adds 16 threads (which fits the test logic) and speeds up
the test.
Before the change:
# perf stat --null ./test_progs -t uprobe_multi_test/consumers
#421/9 uprobe_multi_test/consumers:OK
#421 uprobe_multi_test:OK
Summary: 1/1 PASSED, 0 SKIPPED, 0 FAILED
Performance counter stats for './test_progs -t uprobe_multi_test/consumers':
28.818778973 seconds time elapsed
0.745518000 seconds user
0.919186000 seconds sys
After the change:
# perf stat --null ./test_progs -t uprobe_multi_test/consumers 2>&1
#421/9 uprobe_multi_test/consumers:OK
#421 uprobe_multi_test:OK
Summary: 1/1 PASSED, 0 SKIPPED, 0 FAILED
Performance counter stats for './test_progs -t uprobe_multi_test/consumers':
3.504790814 seconds time elapsed
0.012141000 seconds user
0.751760000 seconds sys
[1] commit 87195a1ee332 ("uprobes: switch to RCU Tasks Trace flavor for better performance")
Acked-by: Andrii Nakryiko <andrii@...nel.org>
Signed-off-by: Jiri Olsa <jolsa@...nel.org>
---
.../bpf/prog_tests/uprobe_multi_test.c | 98 +++++++++++++++----
1 file changed, 80 insertions(+), 18 deletions(-)
diff --git a/tools/testing/selftests/bpf/prog_tests/uprobe_multi_test.c b/tools/testing/selftests/bpf/prog_tests/uprobe_multi_test.c
index e96b153a0f5d..cdb986871835 100644
--- a/tools/testing/selftests/bpf/prog_tests/uprobe_multi_test.c
+++ b/tools/testing/selftests/bpf/prog_tests/uprobe_multi_test.c
@@ -789,7 +789,7 @@ get_link(struct uprobe_multi_consumers *skel, int link)
}
}
-static int uprobe_attach(struct uprobe_multi_consumers *skel, int idx)
+static int uprobe_attach(struct uprobe_multi_consumers *skel, int idx, unsigned long offset)
{
struct bpf_program *prog = get_program(skel, idx);
struct bpf_link **link = get_link(skel, idx);
@@ -798,6 +798,9 @@ static int uprobe_attach(struct uprobe_multi_consumers *skel, int idx)
if (!prog || !link)
return -1;
+ opts.offsets = &offset;
+ opts.cnt = 1;
+
/*
* bit/prog: 0 uprobe entry
* bit/prog: 1 uprobe return
@@ -807,9 +810,7 @@ static int uprobe_attach(struct uprobe_multi_consumers *skel, int idx)
opts.retprobe = idx == 1;
opts.session = idx == 2 || idx == 3;
- *link = bpf_program__attach_uprobe_multi(prog, 0, "/proc/self/exe",
- "uprobe_consumer_test",
- &opts);
+ *link = bpf_program__attach_uprobe_multi(prog, 0, "/proc/self/exe", NULL, &opts);
if (!ASSERT_OK_PTR(*link, "bpf_program__attach_uprobe_multi"))
return -1;
return 0;
@@ -830,7 +831,8 @@ static bool test_bit(int bit, unsigned long val)
noinline int
uprobe_consumer_test(struct uprobe_multi_consumers *skel,
- unsigned long before, unsigned long after)
+ unsigned long before, unsigned long after,
+ unsigned long offset)
{
int idx;
@@ -843,15 +845,43 @@ uprobe_consumer_test(struct uprobe_multi_consumers *skel,
/* ... and attach all new programs in 'after' state */
for (idx = 0; idx < 4; idx++) {
if (!test_bit(idx, before) && test_bit(idx, after)) {
- if (!ASSERT_OK(uprobe_attach(skel, idx), "uprobe_attach_after"))
+ if (!ASSERT_OK(uprobe_attach(skel, idx, offset), "uprobe_attach_after"))
return -1;
}
}
return 0;
}
+/*
+ * We generate 16 consumer_testX functions that will have uprobe installed on
+ * and will be called in separate threads. All function pointer are stored in
+ * "consumers" section and each thread will pick one function based on index.
+ */
+
+extern const void *__start_consumers;
+
+#define __CONSUMER_TEST(func) \
+noinline int func(struct uprobe_multi_consumers *skel, unsigned long before, \
+ unsigned long after, unsigned long offset) \
+{ \
+ return uprobe_consumer_test(skel, before, after, offset); \
+} \
+static void *__ ## func __used __attribute__((section("consumers"))) = (void *) func;
+
+#define CONSUMER_TEST(func) __CONSUMER_TEST(func)
+
+#define C1 CONSUMER_TEST(__PASTE(consumer_test, __COUNTER__))
+#define C4 C1 C1 C1 C1
+#define C16 C4 C4 C4 C4
+
+C16
+
+typedef int (*test_t)(struct uprobe_multi_consumers *, unsigned long,
+ unsigned long, unsigned long);
+
static int consumer_test(struct uprobe_multi_consumers *skel,
- unsigned long before, unsigned long after)
+ unsigned long before, unsigned long after,
+ test_t test, unsigned long offset)
{
int err, idx, ret = -1;
@@ -860,12 +890,12 @@ static int consumer_test(struct uprobe_multi_consumers *skel,
/* 'before' is each, we attach uprobe for every set idx */
for (idx = 0; idx < 4; idx++) {
if (test_bit(idx, before)) {
- if (!ASSERT_OK(uprobe_attach(skel, idx), "uprobe_attach_before"))
+ if (!ASSERT_OK(uprobe_attach(skel, idx, offset), "uprobe_attach_before"))
goto cleanup;
}
}
- err = uprobe_consumer_test(skel, before, after);
+ err = test(skel, before, after, offset);
if (!ASSERT_EQ(err, 0, "uprobe_consumer_test"))
goto cleanup;
@@ -934,14 +964,46 @@ static int consumer_test(struct uprobe_multi_consumers *skel,
return ret;
}
-static void test_consumers(void)
+#define CONSUMER_MAX 16
+
+/*
+ * Each thread runs 1/16 of the load by running test for single
+ * 'before' number (based on thread index) and full scale of
+ * 'after' numbers.
+ */
+static void *consumer_thread(void *arg)
{
+ unsigned long idx = (unsigned long) arg;
struct uprobe_multi_consumers *skel;
- int before, after;
+ unsigned long offset;
+ const void *func;
+ int after;
skel = uprobe_multi_consumers__open_and_load();
if (!ASSERT_OK_PTR(skel, "uprobe_multi_consumers__open_and_load"))
- return;
+ return NULL;
+
+ func = *((&__start_consumers) + idx);
+
+ offset = get_uprobe_offset(func);
+ if (!ASSERT_GE(offset, 0, "uprobe_offset"))
+ goto out;
+
+ for (after = 0; after < CONSUMER_MAX; after++)
+ if (consumer_test(skel, idx, after, func, offset))
+ goto out;
+
+out:
+ uprobe_multi_consumers__destroy(skel);
+ return NULL;
+}
+
+
+static void test_consumers(void)
+{
+ pthread_t pt[CONSUMER_MAX];
+ unsigned long idx;
+ int err;
/*
* The idea of this test is to try all possible combinations of
@@ -982,14 +1044,14 @@ static void test_consumers(void)
* before/after bits.
*/
- for (before = 0; before < 16; before++) {
- for (after = 0; after < 16; after++)
- if (consumer_test(skel, before, after))
- goto out;
+ for (idx = 0; idx < CONSUMER_MAX; idx++) {
+ err = pthread_create(&pt[idx], NULL, consumer_thread, (void *) idx);
+ if (!ASSERT_OK(err, "pthread_create"))
+ break;
}
-out:
- uprobe_multi_consumers__destroy(skel);
+ while (idx)
+ pthread_join(pt[--idx], NULL);
}
static struct bpf_program *uprobe_multi_program(struct uprobe_multi_pid_filter *skel, int idx)
--
2.46.2
Powered by blists - more mailing lists