On Tue, Aug 2, 2022 at 11:29 PM Kurt Kanzenbach <kurt@xxxxxxxxxxxxx> wrote:
>
> On Tue Aug 02 2022, Alexei Starovoitov wrote:
> > On Tue, Aug 2, 2022 at 12:06 AM Kurt Kanzenbach <kurt@xxxxxxxxxxxxx> wrote:
> >>
> >> Hi Alexei,
> >>
> >> On Tue Jun 07 2022, Alexei Starovoitov wrote:
> >> > Anyway I guess new helper bpf_ktime_get_tai_ns() is ok, since
> >> > it's so trivial, but selftest is necessary.
> >>
> >> So, I did write a selftest [1] for testing bpf_ktime_get_tai_ns() and
> >> verifying that the access to the clock works. It uses AF_XDP sockets and
> >> timestamps the incoming packets. The timestamps are then validated in
> >> user space.
> >>
> >> Since AF_XDP related code is migrating from libbpf to libxdp, I'm
> >> wondering if that sample fits into the kernel's selftests or not. What
> >> kind of selftest are you looking for?
> >
> > Please use selftests/bpf framework.
> > There are plenty of networking tests in there.
> > bpf_ktime_get_tai_ns() doesn't have to rely on af_xdp.
>
> OK.
>
> > It can be skb based.
>
> Something like this?
>
> +++ b/tools/testing/selftests/bpf/prog_tests/check_tai.c
> @@ -0,0 +1,57 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/* Copyright (C) 2022 Linutronix GmbH */
> +
> +#include <test_progs.h>
> +#include <network_helpers.h>
> +
> +#include <time.h>
> +#include <stdint.h>
> +
> +#define TAI_THRESHOLD 1000000000ULL /* 1s */
> +#define NSEC_PER_SEC 1000000000ULL
> +
> +static __u64 ts_to_ns(const struct timespec *ts)
> +{
> + return ts->tv_sec * NSEC_PER_SEC + ts->tv_nsec;
> +}
> +
> +void test_tai(void)
> +{
> + struct __sk_buff skb = {
> + .tstamp = 0,
> + .hwtstamp = 0,
> + };
> + LIBBPF_OPTS(bpf_test_run_opts, topts,
> + .data_in = &pkt_v4,
> + .data_size_in = sizeof(pkt_v4),
> + .ctx_in = &skb,
> + .ctx_size_in = sizeof(skb),
> + .ctx_out = &skb,
> + .ctx_size_out = sizeof(skb),
> + );
> + struct timespec now_tai;
> + struct bpf_object *obj;
> + int ret, prog_fd;
> +
> + ret = bpf_prog_test_load("./test_tai.o",
> + BPF_PROG_TYPE_SCHED_CLS, &obj, &prog_fd);
it would be best to rely on BPF skeleton, please see other tests
including *.skel.h, thanks
> + if (!ASSERT_OK(ret, "load"))
> + return;
> + ret = bpf_prog_test_run_opts(prog_fd, &topts);
> + ASSERT_OK(ret, "test_run");
> +
> + /* TAI != 0 */
> + ASSERT_NEQ(skb.tstamp, 0, "tai_ts0_0");
> + ASSERT_NEQ(skb.hwtstamp, 0, "tai_ts0_1");
> +
> + /* TAI is moving forward only */
> + ASSERT_GT(skb.hwtstamp, skb.tstamp, "tai_forward");
> +
> + /* Check for reasoneable range */
> + ret = clock_gettime(CLOCK_TAI, &now_tai);
> + ASSERT_EQ(ret, 0, "tai_gettime");
> + ASSERT_TRUE((ts_to_ns(&now_tai) - skb.hwtstamp) < TAI_THRESHOLD,
> + "tai_range");
> +
> + bpf_object__close(obj);
> +}
> diff --git a/tools/testing/selftests/bpf/progs/test_tai.c b/tools/testing/selftests/bpf/progs/test_tai.c
> new file mode 100644
> index 000000000000..34ac4175e29d
> --- /dev/null
> +++ b/tools/testing/selftests/bpf/progs/test_tai.c
> @@ -0,0 +1,17 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/* Copyright (C) 2022 Linutronix GmbH */
> +
> +#include <linux/bpf.h>
> +#include <bpf/bpf_helpers.h>
> +
> +char _license[] SEC("license") = "GPL";
> +
> +SEC("tc")
> +int save_tai(struct __sk_buff *skb)
> +{
> + /* Save TAI timestamps */
> + skb->tstamp = bpf_ktime_get_tai_ns();
> + skb->hwtstamp = bpf_ktime_get_tai_ns();
> +
> + return 0;
> +}
