On Sat, Jul 19, 2025 at 2:15 PM Jesper Dangaard Brouer <hawk@xxxxxxxxxx> wrote:
>
> On 18/07/2025 17.05, Matt Fleming wrote:
>
> > [...]
> > diff --git a/tools/testing/selftests/bpf/progs/lpm_trie_bench.c b/tools/testing/selftests/bpf/progs/lpm_trie_bench.c
> > new file mode 100644
> > index 000000000000..c335718cc240
> > --- /dev/null
> > +++ b/tools/testing/selftests/bpf/progs/lpm_trie_bench.c
> > @@ -0,0 +1,175 @@
> [...]
> > +
> > +static __always_inline void atomic_inc(long *cnt)
> > +{
> > + __atomic_add_fetch(cnt, 1, __ATOMIC_SEQ_CST);
> > +}
> > +
> > +static __always_inline long atomic_swap(long *cnt, long val)
> > +{
> > + return __atomic_exchange_n(cnt, val, __ATOMIC_SEQ_CST);
> > +}
>
> For userspace includes we have similar defines in bench.h.
> Except they use __ATOMIC_RELAXED and here __ATOMIC_SEQ_CST.
> Which is the correct to use?
>
> For BPF kernel-side do selftests have another header file that define
> these `atomic_inc` and `atomic_swap` ?
Actually, we can side step this problem completely by consistently
using __sync_fetch_and_add() for duration_ns and hits and removing the
atomic operations for DELETE, which doesn't need atomicity anyway
since only a single producer can run.
I'll send a v2.
