Remove "legacy" triggering benchmarks which rely on syscalls (and thus syscall
overhead is a noticeable part of benchmark, unfortunately). Replace them with
faster versions that rely on triggering BPF programs in-kernel through another
simple "driver" BPF program. See patch #2 with comparison results.
raw_tp/tp/fmodret benchmarks required adding a simple kfunc in kernel to be
able to trigger a simple tracepoint from BPF program (plus we also mark that
function as allowable for fmod_ret programs). This limits raw_tp/tp/fmodret
benchmarks to new kernels only, but it keeps bench tool itself very portable
and most of other benchmarks will still work on wide variety of kernels
without the need to worry about building and deploying custom kernel module.
See patches #5 and #6 for details.
Andrii Nakryiko (6):
selftests/bpf: rename and clean up userspace-triggered benchmarks
selftests/bpf: add batched, mostly in-kernel BPF triggering benchmarks
selftests/bpf: remove syscall-driven benchs, keep syscall-count only
selftests/bpf: lazy-load trigger bench BPF programs
bpf: add bpf_test_tp() kfunc triggering tp and allowing error
injection
selftests/bpf: add batched tp/raw_tp/fmodret tests
kernel/bpf/bpf_test.h | 34 ++
kernel/bpf/helpers.c | 13 +
tools/testing/selftests/bpf/bench.c | 33 +-
.../selftests/bpf/benchs/bench_trigger.c | 388 +++++++++---------
.../selftests/bpf/benchs/run_bench_trigger.sh | 22 +-
.../selftests/bpf/benchs/run_bench_uprobes.sh | 2 +-
.../selftests/bpf/progs/trigger_bench.c | 68 ++-
7 files changed, 332 insertions(+), 228 deletions(-)
create mode 100644 kernel/bpf/bpf_test.h
--
2.43.0
