Hello Stanislas, thanks for the reviews ! On 11/13/24 18:42, Stanislav Fomichev wrote: > On 11/13, Alexis Lothoré (eBPF Foundation) wrote: >> The flow_dissector runs plenty of tests over diffent kind of packets, >> grouped into three categories: skb mode, non-skb mode with direct >> attach, and non-skb with indirect attach. >> >> Re-split the main function into dedicated tests. Each test now must have >> its own setup/teardown, but for the advantage of being able to run them >> separately. >> >> Signed-off-by: Alexis Lothoré (eBPF Foundation) <alexis.lothore@xxxxxxxxxxx> >> --- >> .../selftests/bpf/prog_tests/flow_dissector.c | 92 ++++++++++++++-------- >> 1 file changed, 57 insertions(+), 35 deletions(-) >> >> diff --git a/tools/testing/selftests/bpf/prog_tests/flow_dissector.c b/tools/testing/selftests/bpf/prog_tests/flow_dissector.c >> index 6fbe8b6dad561aec02db552caea02517ac1e2109..c5dfff333fe31dd55ac152fe9b107828227c8177 100644 >> --- a/tools/testing/selftests/bpf/prog_tests/flow_dissector.c >> +++ b/tools/testing/selftests/bpf/prog_tests/flow_dissector.c >> @@ -549,63 +549,101 @@ static void run_tests_skb_less(int tap_fd, struct bpf_map *keys) >> } >> } >> >> -static void test_skb_less_prog_attach(struct bpf_flow *skel, int tap_fd) > > [..] > >> +void serial_test_flow_dissector_skb_less_direct_attach(void) > > Any specific reason you keep these as serial? Seems like one of the benefits > of splitting them up is to be able to run them in parallel? I guess there is no reason (I added those while investigating the namespace exclusivity test issue, and forgot to remove them). I'll remove the serial prefix. By the way I realize that each of those new tests could likely benefit from running in an isolated net namespace (especially if they can run in parallel), I'll add that too in v2. Thanks, Alexis -- Alexis Lothoré, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
