Download as PDF, PPTX
![[OpenStack Days Korea 2016] Track1 - All flash CEPH 구성 및 최적화](https://cdn.slidesharecdn.com/ss_thumbnails/12skt-160226171513-thumbnail.jpg?width=640&height=640&fit=bounds)
Locking in Linux Traffic Control subsystem
- 1.
- 2. Agenda Current locking RCU completeness Locklessqdisc TX queue mapping @ c0ngwang
- 3. Current locking RTNL lockon slow path RCU read lock on fast path Spinlock to sync enqueue with dequeue Lock root qdisc Lockless on ingress @ c0ngwang
- 4.
- 5.
- 6. @ c0ngwang qdisc filter and embeddedaction standalone action add dev->qdisc_sleeping dev->qdisc idr delete dev->qdisc_sleeping dev->qdisc idr dump dev->qdisc dev->qdisc idr sync with fast path synchronize_net() call_rcu() Filter’s call_rcu()
- 7.
- 8. RCU completeness Qdisc layerstill has spinlocks on fast path Filter layer is already RCU-complete Action layer, copy is still missing Completely get rid of TC action spinlocks @ c0ngwang
- 9. Lockless Qdisc? It isa ring buffer! No tree lock, but consumer and producer lock Decouple enqueue and dequeue FIFO is easy, it is class-less and filter-less @ c0ngwang
- 10.
- 11. Really lockless? skbs enqueuedand dequeued on fast path Qdisc can be classful and hierarchical BH context vs process context Different internal data structures @ c0ngwang
- 12. Lockless list? NULL terminatedsingle linked list, LIFO No counting or size limit Still have to lock del_first() Totally batch enqueue and dequeue? @ c0ngwang
- 13. Locking ideas dev->qdisc_sleeping canbe eliminated? Break down RTNL to per device locking? Break down tree lock to each impl./layer? Break down tree lock down to each tx queue? @ c0ngwang
- 14. Tx queue mapping Aqdisc is shared by all TX queues and CPU’s Queue selection is done before enqueue Except mq qdisc Moving TX queue under Qdisc control? XPS @ c0ngwang
- 15.
- 16.
- 17.