Re: [PATCH mptcp-next v6 4/9] selftests/bpf: Add MPTCP_SCHED_TEST macro

[Matthieu Baerts <matttbe@kernel.org>]

Hi Geliang,

On 04/04/2024 15:03, Geliang Tang wrote:
> From: Geliang Tang <tanggeliang@kylinos.cn>
> 
> This patch defines MPTCP_SCHED_TEST macro, a template for all scheduler
> tests. Every scheduler is identified by argument name, and use sysctl
> to set net.mptcp.scheduler as "bpf_name" to use this sched. Add two
> veth net devices to simulate the multiple addresses case. Use 'ip mptcp
> endpoint' command to add the new endpoint ADDR2 to PM netlink. Arguments
> addr1/add2 means whether the data has been sent on the first/second subflow
> or not. Send data and check bytes_sent of 'ss' output after it using
> send_data_and_verify().

Should we not squash this in "selftests/bpf: Add bpf scheduler test" as
well? We already have a lot of different commits.

> 
> Signed-off-by: Geliang Tang <tanggeliang@kylinos.cn>
> ---
>  .../testing/selftests/bpf/prog_tests/mptcp.c  | 30 +++++++++++++++++++
>  1 file changed, 30 insertions(+)
> 
> diff --git a/tools/testing/selftests/bpf/prog_tests/mptcp.c b/tools/testing/selftests/bpf/prog_tests/mptcp.c
> index ffcd5ebe38b9..8a6b09a97698 100644
> --- a/tools/testing/selftests/bpf/prog_tests/mptcp.c
> +++ b/tools/testing/selftests/bpf/prog_tests/mptcp.c
> @@ -498,6 +498,36 @@ static void test_default(void)
>  	cleanup_netns(nstoken);
>  }
>  
> +#define MPTCP_SCHED_TEST(name, addr1, addr2)			\
> +static void test_##name(void)					\
> +{								\
> +	struct mptcp_bpf_##name *skel;				\
> +	struct nstoken *nstoken;				\
> +	struct bpf_link *link;					\
> +	struct bpf_map *map;					\
> +								\
> +	skel = mptcp_bpf_##name##__open_and_load();		\
> +	if (!ASSERT_OK_PTR(skel, "open_and_load " #name))	\
> +		return;						\
> +								\
> +	map = bpf_object__find_map_by_name(skel->obj, #name);	\
> +	link = bpf_map__attach_struct_ops(map);			\
> +	if (!ASSERT_OK_PTR(link, "attach_struct_ops " #name))	\
> +		goto skel_destroy;				\
> +								\
> +	nstoken = sched_init("subflow", "bpf_" #name);		\
> +	if (!ASSERT_OK_PTR(nstoken, "sched_init " #name))	\
> +		goto link_destroy;				\
> +								\
> +	send_data_and_verify(#name, atoi(#addr1), atoi(#addr2));\

Can you not use the values of addr1 and addr2 directly as number instead
of string + atoi()?

> +								\
> +	cleanup_netns(nstoken);					\
> +link_destroy:							\
> +	bpf_link__destroy(link);				\
> +skel_destroy:							\
> +	mptcp_bpf_##name##__destroy(skel);			\
> +}

I don't mind having functions defined in macros, but I think we should
try to reduce their size to the minimum (if possible) because they are
hard to read and debug in case of error.

Here, do you think we could have a smaller macro passing the name as a
string (for the error messages) and the different functions pointers you
need to a new helper? (+ using generic skel structure?)

Maybe it will not work or will not be easier to read, but worth a try I
think. If it is not possible, please explain why.


> +
>  static void test_first(void)
>  {
>  	struct mptcp_bpf_first *first_skel;

Cheers,
Matt
-- 
Sponsored by the NGI0 Core fund.