Re: [PATCH v4 4/4] selftests/bpf: Add tests for ring__consume_n and ring_buffer__consume_n

[Andrea Righi <andrea.righi@canonical.com>]

On Sat, Apr 06, 2024 at 10:52:10AM -0700, Andrii Nakryiko wrote:
> On Sat, Apr 6, 2024 at 10:39 AM Andrii Nakryiko
> <andrii.nakryiko@gmail.com> wrote:
> >
> > On Sat, Apr 6, 2024 at 2:20 AM Andrea Righi <andrea.righi@canonical.com> wrote:
> > >
> > > Add tests for new API ring__consume_n() and ring_buffer__consume_n().
> > >
> > > Signed-off-by: Andrea Righi <andrea.righi@canonical.com>
> > > ---
> > >  tools/testing/selftests/bpf/prog_tests/ringbuf.c | 8 ++++++++
> > >  1 file changed, 8 insertions(+)
> > >
> > > diff --git a/tools/testing/selftests/bpf/prog_tests/ringbuf.c b/tools/testing/selftests/bpf/prog_tests/ringbuf.c
> > > index 48c5695b7abf..33aba7684ab9 100644
> > > --- a/tools/testing/selftests/bpf/prog_tests/ringbuf.c
> > > +++ b/tools/testing/selftests/bpf/prog_tests/ringbuf.c
> > > @@ -304,10 +304,18 @@ static void ringbuf_subtest(void)
> > >         err = ring_buffer__consume(ringbuf);
> > >         CHECK(err < 0, "rb_consume", "failed: %d\b", err);
> > >
> > > +       /* try to consume up to one item */
> > > +       err = ring_buffer__consume_n(ringbuf, 1);
> > > +       CHECK(err < 0 || err > 1, "rb_consume_n", "failed: %d\b", err);
> > > +
> > >         /* also consume using ring__consume to make sure it works the same */
> > >         err = ring__consume(ring);
> > >         ASSERT_GE(err, 0, "ring_consume");
> > >
> > > +       /* try to consume up to one item */
> > > +       err = ring__consume_n(ring, 1);
> > > +       CHECK(err < 0 || err > 1, "ring_consume_n", "failed: %d\b", err);
> > > +
> >
> > Did you actually run this test? There is ring_buffer__consume() and
> > ring__consume() calls right before your added calls, so consume_n will
> > return zero.
> >
> > I dropped this broken patch. Please send a proper test as a follow up.
> 
> Sorry, technically, it's not broken, it just doesn't test much (CHECK
> conditions confused me, I didn't realize you allow zero initially). We
> will never consume anything and the result will be zero, which isn't
> very meaningful.
> 
> "Interesting" test would set up things so that we have >1 item in
> ringbuf and we consume exactly one at a time, because that's the new
> logic you added.
> 
> I think it will be simpler to add a dedicated and simpler ringbuf test
> for this, where you can specify how many items to submit, and then do
> a bunch of consume/consume_n invocations, checking exact results.
> 
> Plus, please don't add new CHECK() uses, use ASSERT_XXX() ones instead.
> 
> I've applied first three patches because they look correct and it's
> good to setup libbpf 1.5 dev cycle, but please do follow up with a
> better test. Thanks.

Yeah, sorry, I tried to add a minimal test to the existing one, but I
agree that it not very meaningful.

I already have a better dedicated test case for this
(https://github.com/arighi/ebpf-maps/blob/libbpf-consume-n/src/main.c#L118),
I just need to integrate it in the kselftest properly (and maybe
pre-generate more than N records in the ring buffer, so that we can
better test if the limit works as expected).

I'll send another patch to add a proper test case.

Thanks for applying the other patches!
-Andrea

> 
> >
> > >         /* 3 rounds, 2 samples each */
> > >         cnt = atomic_xchg(&sample_cnt, 0);
> > >         CHECK(cnt != 6, "cnt", "exp %d samples, got %d\n", 6, cnt);
> > > --
> > > 2.43.0
> > >