Re: [External] Re: [PATCH bpf-next] selftests/bpf: fix some bugs in map_lookup_percpu_elem testcase
From: Feng Zhou
Date: Thu May 19 2022 - 03:31:52 EST
在 2022/5/19 下午12:38, Andrii Nakryiko 写道:
On Wed, May 18, 2022 at 8:27 PM Feng Zhou <zhoufeng.zf@xxxxxxxxxxxxx> wrote:
在 2022/5/19 上午8:17, Andrii Nakryiko 写道:
On Sun, May 15, 2022 at 7:25 PM Feng zhou <zhoufeng.zf@xxxxxxxxxxxxx> wrote:
From: Feng Zhou <zhoufeng.zf@xxxxxxxxxxxxx>
comments from Andrii Nakryiko, details in here:
https://lore.kernel.org/lkml/20220511093854.411-1-zhoufeng.zf@xxxxxxxxxxxxx/T/
use /* */ instead of //
use libbpf_num_possible_cpus() instead of sysconf(_SC_NPROCESSORS_ONLN)
use 8 bytes for value size
fix memory leak
use ASSERT_EQ instead of ASSERT_OK
add bpf_loop to fetch values on each possible CPU
Fixes: ed7c13776e20c74486b0939a3c1de984c5efb6aa ("selftests/bpf: add test case for bpf_map_lookup_percpu_elem")
Signed-off-by: Feng Zhou <zhoufeng.zf@xxxxxxxxxxxxx>
---
.../bpf/prog_tests/map_lookup_percpu_elem.c | 49 +++++++++------
.../bpf/progs/test_map_lookup_percpu_elem.c | 61 ++++++++++++-------
2 files changed, 70 insertions(+), 40 deletions(-)
diff --git a/tools/testing/selftests/bpf/prog_tests/map_lookup_percpu_elem.c b/tools/testing/selftests/bpf/prog_tests/map_lookup_percpu_elem.c
index 58b24c2112b0..89ca170f1c25 100644
--- a/tools/testing/selftests/bpf/prog_tests/map_lookup_percpu_elem.c
+++ b/tools/testing/selftests/bpf/prog_tests/map_lookup_percpu_elem.c
@@ -1,30 +1,39 @@
-// SPDX-License-Identifier: GPL-2.0
-// Copyright (c) 2022 Bytedance
+/* SPDX-License-Identifier: GPL-2.0 */
heh, so for SPDX license comment the rule is to use // in .c files :)
so keep SPDX as // and all others as /* */
will do. Thanks.
+/* Copyright (c) 2022 Bytedance */
#include <test_progs.h>
#include "test_map_lookup_percpu_elem.skel.h"
-#define TEST_VALUE 1
-
void test_map_lookup_percpu_elem(void)
{
struct test_map_lookup_percpu_elem *skel;
- int key = 0, ret;
- int nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
- int *buf;
+ __u64 key = 0, sum;
+ int ret, i;
+ int nr_cpus = libbpf_num_possible_cpus();
+ __u64 *buf;
- buf = (int *)malloc(nr_cpus*sizeof(int));
+ buf = (__u64 *)malloc(nr_cpus*sizeof(__u64));
no need for casting
casting means no '(__u64 *)'?
just like this:
'buf = malloc(nr_cpus * sizeof(__u64));'
yes, in C you don't need to explicitly cast void * to other pointer types
Ok, Thanks.
if (!ASSERT_OK_PTR(buf, "malloc"))
return;
- memset(buf, 0, nr_cpus*sizeof(int));
- buf[0] = TEST_VALUE;
- skel = test_map_lookup_percpu_elem__open_and_load();
- if (!ASSERT_OK_PTR(skel, "test_map_lookup_percpu_elem__open_and_load"))
- return;
+ for (i=0; i<nr_cpus; i++)
spaces between operators
will do. Thanks.
+ buf[i] = i;
+ sum = (nr_cpus-1)*nr_cpus/2;
same, please follow kernel code style
will do. Thanks.
+
+ skel = test_map_lookup_percpu_elem__open();
+ if (!ASSERT_OK_PTR(skel, "test_map_lookup_percpu_elem__open"))
+ goto exit;
+
nit: keep it simple, init skel to NULL and use single cleanup goto
label that will destroy skel unconditionally (it deals with NULL just
fine)
will do. Thanks.
+ skel->rodata->nr_cpus = nr_cpus;
+
+ ret = test_map_lookup_percpu_elem__load(skel);
+ if (!ASSERT_OK(ret, "test_map_lookup_percpu_elem__load"))
+ goto cleanup;
+
ret = test_map_lookup_percpu_elem__attach(skel);
- ASSERT_OK(ret, "test_map_lookup_percpu_elem__attach");
+ if (!ASSERT_OK(ret, "test_map_lookup_percpu_elem__attach"))
+ goto cleanup;
ret = bpf_map_update_elem(bpf_map__fd(skel->maps.percpu_array_map), &key, buf, 0);
ASSERT_OK(ret, "percpu_array_map update");
[...]