mbox series

[v9,bpf-next,0/3] bpf: sharing bpf runtime stats with

Message ID 20200430071506.1408910-1-songliubraving@fb.com
Headers show
Series bpf: sharing bpf runtime stats with | expand

Message

Song Liu April 30, 2020, 7:15 a.m. UTC 
run_time_ns is a useful stats for BPF programs. However, it is gated by
sysctl kernel.bpf_stats_enabled. When multiple user space tools are
toggling kernl.bpf_stats_enabled at the same time, they may confuse each
other.

Solve this problem with a new BPF command BPF_ENABLE_STATS.

Changes v8 => v9:
  1. Clean up in selftest (Andrii).
  2. Not using static variable in test program (Andrii).

Changes v7 => v8:
  1. Change name BPF_STATS_RUNTIME_CNT => BPF_STATS_RUN_TIME (Alexei).
  2. Add CHECK_ATTR to bpf_enable_stats() (Alexei).
  3. Rebase (Andrii).
  4. Simplfy the selftest (Alexei).

Changes v6 => v7:
  1. Add test to verify run_cnt matches count measured by the program.

Changes v5 => v6:
  1. Simplify test program (Yonghong).
  2. Rebase (with some conflicts).

Changes v4 => v5:
  1. Use memset to zero bpf_attr in bpf_enable_stats() (Andrii).

Changes v3 => v4:
  1. Add libbpf support and selftest;
  2. Avoid cleaning trailing space.

Changes v2 => v3:
  1. Rename the command to BPF_ENABLE_STATS, and make it extendible.
  2. fix commit log;
  3. remove unnecessary headers.

Song Liu (3):
  bpf: sharing bpf runtime stats with BPF_ENABLE_STATS
  libbpf: add support for command BPF_ENABLE_STATS
  bpf: add selftest for BPF_ENABLE_STATS

 include/linux/bpf.h                           |  1 +
 include/uapi/linux/bpf.h                      | 11 ++++
 kernel/bpf/syscall.c                          | 57 +++++++++++++++++++
 kernel/sysctl.c                               | 36 +++++++++++-
 tools/include/uapi/linux/bpf.h                | 11 ++++
 tools/lib/bpf/bpf.c                           | 10 ++++
 tools/lib/bpf/bpf.h                           |  1 +
 tools/lib/bpf/libbpf.map                      |  1 +
 .../selftests/bpf/prog_tests/enable_stats.c   | 45 +++++++++++++++
 .../selftests/bpf/progs/test_enable_stats.c   | 18 ++++++
 10 files changed, 190 insertions(+), 1 deletion(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/enable_stats.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_enable_stats.c

--
2.24.1

Comments

Alexei Starovoitov May 1, 2020, 5:56 p.m. UTC | #1 
On Thu, Apr 30, 2020 at 12:15 AM Song Liu <songliubraving@fb.com> wrote:
>
> run_time_ns is a useful stats for BPF programs. However, it is gated by
> sysctl kernel.bpf_stats_enabled. When multiple user space tools are
> toggling kernl.bpf_stats_enabled at the same time, they may confuse each
> other.
>
> Solve this problem with a new BPF command BPF_ENABLE_STATS.
>
> Changes v8 => v9:
>   1. Clean up in selftest (Andrii).
>   2. Not using static variable in test program (Andrii).

Applied. Thanks