On Tue, Jul 30, 2013 at 05:50:02PM -0700, Ethan Jackson wrote: > > There are definitely ways around the thread safety analysis. My > > feeling is that we should be pretty strict about it except in modules > > where contention is a real problem. > > > >> Why does stp need a recursive mutex? I don't see any natural > >> recursion here. > > So looking back at this, I'd like to keep the recursive mutex. > stp_send_bpdu() uses a call back function to send a packet through > ofproto-dpif. If this packet is going through a patch port it could > loop back into the stp module causing a dead lock. I don't think > there's a clean thread safe way to do this without the recursive mutex > short of ditching the call back function. I'd be all for doing that > in future, but for now I'd prefer to keep the patch as-is (with the > possible addition of a comment explaining why the recursive mutex is > necessary).
Thanks for the explanation, I'll look at the reposted version now. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
