On 12/19/22 12:04 AM, Song Liu wrote:
While max time cost might be useful in some debugging, I don't think
we should add it with kernel.bpf_stats_enabled. Otherwise, we can
argue p50/p90/p99 are also useful in some cases, and some other
metrics are useful in some other cases. These metrics together will
make kernel.bpf_stats_enabled too expensive for the use case above.
Since the use case is for debugging, have you considered using
some other BPF programs to profile the target BPF program?
Please refer to "bpftool prog profile" or "perf stat -b " for
examples of similar solutions.
+1. Please consider other suggested solutions that the same can be done in a
more flexible way.
