> But you will need to keep the local_bh_disable/enable around > for counter sycnhonization, so not sure if there is a big win.
CPU spend many time on _spin_lock inside of reead_lock, because where 2 NUMA nodes use same ebtable's table. spin_lock use about 3% of cpu, by perf top data, most calls was from ebtables. Anyway, patch require some proofs of optimizations?
