On Mon, May 16, 2022 at 05:06:28PM +0200, Claudio Jeker wrote: > > In veb configuration we are holding the netlock and sleep in > > smr_barrier() and refcnt_finalize(). An additional sleep in malloc() > > is fine here. > > Are you sure about this? smr_barrier() on busy systems with many cpus can > take 100ms and more. Isn't all processing stopped during that time (IIRC > the code uses the write exclusive netlock).
Interface ioctls are called with netlock. veb_ioctl() calls veb_add_port() and veb_del_port() which call smr_barrier(). I did not notice 100ms network sleep during ifconfig. I would not consider it an urgent problem. If you optimize for one thing, you pay the price somewhere else. > Sleeping with a global rwlock held is not a good design. And here we are back to the net lock design discussion we had with mpi@ years ago. More important is progress. We should not spend too much time on design discussion. Better solve problems. My plan is, in more or less this order: - use exclusive netlock for configuration - use shared netlock for parts that are MP safe - kernel lock together with net lock allows quick fixes to make progress without making everything MP perfect - replace kernel lock with mutex - use something smarter than mutex to make things fast Others have other plans, so we have different maturity of MP support in the kernel. bluhm
