Re: [PATCH bpf-next v2] selftests/bpf: Avoid flaky resize value test for percpu data

On Tue, Aug 18, 2026 at 6:44 PM Leon Hwang <[email protected]> wrote:
>
> On 19/8/26 00:15, [email protected] wrote:
> >> diff --git a/tools/testing/selftests/bpf/prog_tests/global_data_init.c b/tools/testing/selftests/bpf/prog_tests/global_data_init.c
> >> index 5671c31085cd..2b69290bcc3e 100644
> >> --- a/tools/testing/selftests/bpf/prog_tests/global_data_init.c
> >> +++ b/tools/testing/selftests/bpf/prog_tests/global_data_init.c
> >
> > [ ... ]
> >
> >> @@ -236,6 +223,57 @@ static void test_global_percpu_data_lskel(void)
> >>      test_global_percpu_data_lskel__destroy(lskel);
> >>  }
> >>
> >> +#define PAGE_SIZE 4096
> >> +
> >> +static void test_global_percpu_data_map_resize(void)
> >> +{
> >> +    const __u32 desired_sz = sysconf(_SC_PAGE_SIZE) * 2;
> >> +    struct test_global_percpu_data *skel;
> >> +    const int arr_value = -1;
> >> +    struct bpf_map *map;
> >> +    int err, prog_fd;
> >> +    size_t data_sz;
> >> +
> >> +    if (sysconf(_SC_PAGE_SIZE) != PAGE_SIZE) {
> >> +            test__skip();
> >> +            return;
> >> +    }
> >> +
> >> +    skel = test_global_percpu_data__open();
> >> +    if (!ASSERT_OK_PTR(skel, "test_global_percpu_data__open"))
> >> +            return;
> >> +
> >> +    map = skel->maps.percpu_arr;
> >> +    skel->percpu_arr->arr[0] = arr_value;
> >> +    err = bpf_map__set_value_size(map, desired_sz);
> >
> > This isn't a bug, but since the guard above already pins the page size to
> > 4096, would `const __u32 desired_sz = PAGE_SIZE * 2;` read more directly
>
> Will change to 'PAGE_SIZE * 2' if the next revision is needed.
>
> > here - and is the unguarded `#define PAGE_SIZE 4096` preferable to the
> > `#ifndef PAGE_SIZE` form used in sockopt.c?
>
> No. PAGE_SIZE must be 4096 here.
>

can you please elaborate why do we care about specific PAGE_SIZE
value? it's a bit weird to guard test based on page size...

> >
> >> +    if (!ASSERT_OK(err, "bpf_map__set_value_size"))
> >> +            goto out;
> >> +    if (!ASSERT_EQ(bpf_map__value_size(map), desired_sz, "bpf_map__value_size"))
> >> +            goto out;
> >> +    if (!ASSERT_NEQ(bpf_map__btf_value_type_id(map), 0, "bpf_map__btf_value_type_id"))
> >> +            goto out;
> >> +
> >> +    skel->percpu_arr = bpf_map__initial_value(map, &data_sz);
> >> +    if (!ASSERT_EQ(data_sz, desired_sz, "data_sz"))
> >> +            goto out;
> >> +    if (!ASSERT_OK_PTR(skel->percpu_arr, "skel->percpu_arr"))
> >> +            goto out;
> >
> > This isn't a bug, but would it read more consistently with
> > test_global_percpu_data_init() to check ASSERT_OK_PTR(skel->percpu_arr)
> > before ASSERT_EQ(data_sz, desired_sz)? The block this replaces checked
> > pointer-then-size, and the surviving test_global_percpu_data_init() at the
> > same file still does pointer-then-size.
>
> Unnecessary to keep consistent for their order. This is a really minor
> readability concern.
>
> Thanks,
> Leon
>