Re: [EXT.MSG] Re: [PATCH RFC bpf-next 0/2] bpf: Do not let preemption cause false recursion drops in trampoline progs

[Martin Kelly]

On Wed, Nov 4, 2026 at 9:21 PM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
> >      run_cnt 38040015, recursion_misses 4063
>
> hmm. you really cannot tolerate that small drop? it's tracing at the end.

We realize tracing programs don't have a full guarantee, but prior to
7.0 and HRTICK, they were still reliable in many practical cases. It's
less about the absolute rate and more that the same workload had a
~35-300x increased drop rate (depending on hook) in one release, and
that behavior is now the default. Since the long-term direction of the
kernel is for more preemption, the recursion_misses issue looks like it
will keep getting worse until we can move away from per-CPU BPF
recursion counters.

> Also it needs to be per-task + per-prog. How will it look like?

We're going to look into options for per-task + per-prog tracking and
let you know how it goes.

Thanks,
Martin