Re: [PATCH bpf 2/2] selftests/bpf: add variable-length data concatenation pattern test

From: Andrii Nakryiko <andrii.nakryiko@gmail.com>
To: Daniel Borkmann <daniel@iogearbox.net>
Cc: Andrii Nakryiko <andriin@fb.com>, bpf <bpf@vger.kernel.org>,
	Networking <netdev@vger.kernel.org>,
	Alexei Starovoitov <ast@fb.com>, Kernel Team <kernel-team@fb.com>,
	Christoph Hellwig <hch@lst.de>
Subject: Re: [PATCH bpf 2/2] selftests/bpf: add variable-length data concatenation pattern test
Date: Tue, 16 Jun 2020 16:14:39 -0700	[thread overview]
Message-ID: <CAEf4BzYB+gqGEOfuOpJZHP7-e76Y=gp8SQ7rSZ3EGpwQjF6hLA@mail.gmail.com> (raw)
In-Reply-To: <dd14f356-44bc-0ff0-a089-ce9fb9936c62@iogearbox.net>

On Tue, Jun 16, 2020 at 3:23 PM Daniel Borkmann <daniel@iogearbox.net> wrote:
>
> On 6/16/20 11:27 PM, Andrii Nakryiko wrote:
> > On Tue, Jun 16, 2020 at 1:21 PM Daniel Borkmann <daniel@iogearbox.net> wrote:
> >> On 6/16/20 7:04 AM, Andrii Nakryiko wrote:
> >>> Add selftest that validates variable-length data reading and concatentation
> >>> with one big shared data array. This is a common pattern in production use for
> >>> monitoring and tracing applications, that potentially can read a lot of data,
> >>> but usually reads much less. Such pattern allows to determine precisely what
> >>> amount of data needs to be sent over perfbuf/ringbuf and maximize efficiency.
> >>>
> >>> This is the first BPF selftest that at all looks at and tests
> >>> bpf_probe_read_str()-like helper's return value, closing a major gap in BPF
> >>> testing. It surfaced the problem with bpf_probe_read_kernel_str() returning
> >>> 0 on success, instead of amount of bytes successfully read.
> >>>
> >>> Signed-off-by: Andrii Nakryiko <andriin@fb.com>
> >>
> >> Fix looks good, but I'm seeing an issue in the selftest on my side. With latest
> >> Clang/LLVM I'm getting:
> >>
> >> # ./test_progs -t varlen
> >> #86 varlen:OK
> >> Summary: 1/0 PASSED, 0 SKIPPED, 0 FAILED
> >>
> >> All good, however, the test_progs-no_alu32 fails for me with:
> >
> > Yeah, same here. It's due to Clang emitting unnecessary bit shifts
> > because bpf_probe_read_kernel_str() is defined as returning 32-bit
> > int. I have a patch ready locally, just waiting for bpf-next to open,
> > which switches those helpers to return long, which auto-matically
> > fixes this test.
> >
> > If it's not a problem, I'd just wait for that patch to go into
> > bpf-next. If not, I can sprinkle bits of assembly magic around to
> > force the kernel to do those bitshifts earlier. But I figured having
> > test_progs-no_alu32 failing one selftest temporarily wasn't too bad.
>
> Given {net,bpf}-next will open up soon, another option could be to take in the fix
> itself to bpf and selftest would be submitted together with your other improvement;
> any objections?
>

Yeah, no objections.

> Thanks,
> Daniel