Chuck Lever writes via Kernel.org Bugzilla: No progress. It looks like a long term structural problem that we won't be able to address quickly. It would be good to confirm that the issue is indeed contention in the IOVA allocator. Compiling with LOCK_STAT enabled and looking in /proc/lock_stat would give some indication that Robin's theory in on the right track. Meanwhile I plan to ask the RDMA gurus if there is an improvement that can be made in the svcrdma implementation to help address the problem. If you need immediately relief, reducing the NFSD thread count might help. View: https://bugzilla.kernel.org/show_bug.cgi?id=219865#c6 You can reply to this message to join the discussion. -- Deet-doot-dot, I am a bot. Kernel.org Bugzilla (bugspray 0.1-dev)
