Hello. We process big traffic flow (about 16GB/s) on 2x8 CPU server, it required optimizations. We use ebtables broute table, and saw in perf what CPU spend a lot of time in read_lock_bh (in ebt_do_table). Because ebtables rules change rarely, we accept memory leak and remove all locks to table. It seems not difficult to replace read and write locks in ebtables to rcu, like in iptables. So, good chance to do first commit into Linux :) But why nobody do it before, maybe task contains hidden hard issues? Or ebtables no often use and nobody got performance issues with locks in ebtables before? I try to find information or open issues, but got only old read-hat presentation «why ebtables is bad» with paragraph about RCU, or something. It makes sense, to try fix this by merge request?
