Looking for a thread running on the same node (pre-scan). After acquiring the MCS lock andīefore acquiring the spinlock, the lock holder scans the main queue Threads store the ID of the node on which Node as the current lock holder, and a secondary queue for threads Organized in two queues, a main queue for threads running on the same It isĮnabled through a configuration option (NUMA_AWARE_SPINLOCKS).ĬNA is a NUMA-aware version of the MCS lock. (compact NUMA-aware lock) as the slow path for qspinlock. Starvation of waiters on other NUMA nodes. Same NUMA node as the lock holder, provided care is taken to avoid Lock throughput can be increased by handing a lock to a waiter on the Change the type of intra_node_handoff_threshold to unsigned int. Fix the comment for cna_scan_main_queue(). Add _init to cna_configure_spin_lock_slowpath(). Minor changes from v8 based on feedback from Longman:
Linux-arm-kernel, linux-kernel, tglx, bp, hpa, x86, guohanjun,Ĭc: steven.sistare, daniel.m.jordan, alex.kogan, dave.dice To: linux, peterz, mingo, acon, arnd, longman, linux-arch, ` (6 more replies) 0 siblings, 7 replies 40+ messages in thread
3:59 ` locking/qspinlock: Rename mcs lock/unlock macros and make them more generic Alex Kogan Add NUMA-awareness to qspinlock LKML Archive on help / color / mirror / Atom feed * Add NUMA-awareness to qspinlock 3:59 Alex Kogan
0 kommentar(er)
