On Sat, Nov 15, 2025 at 4:46 AM Andrii Nakryiko andrii.nakryiko@gmail.com wrote:

On Mon, Nov 10, 2025 at 5:49 AM Menglong Dong menglong8.dong@gmail.com wrote:

...

Remove the "trigger_count" in trigger_bench.c and reuse trigger_driver() instead for trigger_kernel_count_setup().

With the calling to bpf_get_numa_node_id(), the result for "kernel_count" will become a little more accurate.

"more accurate" is a bit misleading here. I think you meant that it will do same amount of helper calls as fentry and other benchmarks, and in that sense will be closer as a baseline comparison, is that right? Can you clarify that in the next revision, please?

Yeah, this is what I mean. The call to "bpf_get_numa_node_id" should be considered as the baseline comparison.

...

It will also easier if we want to test the performance of livepatch, just hook the bpf_get_numa_node_id() and run the "kernel_count" bench trigger.

Signed-off-by: Menglong Dong dongml2@chinatelecom.cn

.../selftests/bpf/benchs/bench_trigger.c | 5 +---- .../testing/selftests/bpf/progs/trigger_bench.c | 17 +++++------------ 2 files changed, 6 insertions(+), 16 deletions(-)

diff --git a/tools/testing/selftests/bpf/benchs/bench_trigger.c b/tools/testing/selftests/bpf/benchs/bench_trigger.c index 1e2aff007c2a..34fd8fa3b803 100644 --- a/tools/testing/selftests/bpf/benchs/bench_trigger.c +++ b/tools/testing/selftests/bpf/benchs/bench_trigger.c @@ -179,11 +179,8 @@ static void trigger_syscall_count_setup(void) static void trigger_kernel_count_setup(void) { setup_ctx();

•   bpf_program__set_autoload(ctx.skel->progs.trigger_driver, false);
  

•   bpf_program__set_autoload(ctx.skel->progs.trigger_count, true);
  

•   ctx.skel->rodata->kernel_count = 1;
    load_ctx();
  

•   /* override driver program */
  

•   ctx.driver_prog_fd = bpf_program__fd(ctx.skel->progs.trigger_count);
  

}

static void trigger_kprobe_setup(void) diff --git a/tools/testing/selftests/bpf/progs/trigger_bench.c b/tools/testing/selftests/bpf/progs/trigger_bench.c index 3d5f30c29ae3..6564d1909c7b 100644 --- a/tools/testing/selftests/bpf/progs/trigger_bench.c +++ b/tools/testing/selftests/bpf/progs/trigger_bench.c @@ -39,26 +39,19 @@ int bench_trigger_uprobe_multi(void *ctx) return 0; }

+const volatile int kernel_count = 0;

nit: use bool? it's not a counter, no need to use int here

...

const volatile int batch_iters = 0;

-SEC("?raw_tp") -int trigger_count(void *ctx) -{

•   int i;
  

•   for (i = 0; i < batch_iters; i++)
  

•           inc_counter();
  

•   return 0;
  

-}

SEC("?raw_tp") int trigger_driver(void *ctx) { int i;

•   for (i = 0; i < batch_iters; i++)
  

•   for (i = 0; i < batch_iters; i++) {
            (void)bpf_get_numa_node_id(); /* attach point for benchmarking */
  

•           if (kernel_count)
  

•                   inc_counter();
  

•   }
  

tbh, I wouldn't touch trigger_driver() adding unnecessary if conditions to it. It's fine, IMO, to have bpf_get_numa_node_id() call in trigger_count() for being closer in terms of actual work being done, but I'd keep trigger_driver and trigger_count separate (maybe renaming trigger_count to trigger_kernel_count would help, I don't know)

Ah, OK! I'll add the call to bpf_get_numa_node_id() in trigger_count() instead. I think the "trigger_kernel_count" makes more sense to me.

Thanks! Menglong Dong

pw-bot: cr

...

    return 0;

}

2.51.2