1 /* 2 * MCS lock defines 3 * 4 * This file contains the main data structure and API definitions of MCS lock. 5 * 6 * The MCS lock (proposed by Mellor-Crummey and Scott) is a simple spin-lock 7 * with the desirable properties of being fair, and with each cpu trying 8 * to acquire the lock spinning on a local variable. 9 * It avoids expensive cache bouncings that common test-and-set spin-lock 10 * implementations incur. 11 */ 12 #ifndef __LINUX_MCS_SPINLOCK_H 13 #define __LINUX_MCS_SPINLOCK_H 14 15 #include <asm/mcs_spinlock.h> 16 17 struct mcs_spinlock { 18 struct mcs_spinlock *next; 19 int locked; /* 1 if lock acquired */ 20 int count; /* nesting count, see qspinlock.c */ 21 }; 22 23 #ifndef arch_mcs_spin_lock_contended 24 /* 25 * Using smp_load_acquire() provides a memory barrier that ensures 26 * subsequent operations happen after the lock is acquired. 27 */ 28 #define arch_mcs_spin_lock_contended(l) \ 29 do { \ 30 while (!(smp_load_acquire(l))) \ 31 cpu_relax_lowlatency(); \ 32 } while (0) 33 #endif 34 35 #ifndef arch_mcs_spin_unlock_contended 36 /* 37 * smp_store_release() provides a memory barrier to ensure all 38 * operations in the critical section has been completed before 39 * unlocking. 40 */ 41 #define arch_mcs_spin_unlock_contended(l) \ 42 smp_store_release((l), 1) 43 #endif 44 45 /* 46 * Note: the smp_load_acquire/smp_store_release pair is not 47 * sufficient to form a full memory barrier across 48 * cpus for many architectures (except x86) for mcs_unlock and mcs_lock. 49 * For applications that need a full barrier across multiple cpus 50 * with mcs_unlock and mcs_lock pair, smp_mb__after_unlock_lock() should be 51 * used after mcs_lock. 52 */ 53 54 /* 55 * In order to acquire the lock, the caller should declare a local node and 56 * pass a reference of the node to this function in addition to the lock. 57 * If the lock has already been acquired, then this will proceed to spin 58 * on this node->locked until the previous lock holder sets the node->locked 59 * in mcs_spin_unlock(). 60 */ 61 static inline 62 void mcs_spin_lock(struct mcs_spinlock **lock, struct mcs_spinlock *node) 63 { 64 struct mcs_spinlock *prev; 65 66 /* Init node */ 67 node->locked = 0; 68 node->next = NULL; 69 70 /* 71 * We rely on the full barrier with global transitivity implied by the 72 * below xchg() to order the initialization stores above against any 73 * observation of @node. And to provide the ACQUIRE ordering associated 74 * with a LOCK primitive. 75 */ 76 prev = xchg(lock, node); 77 if (likely(prev == NULL)) { 78 /* 79 * Lock acquired, don't need to set node->locked to 1. Threads 80 * only spin on its own node->locked value for lock acquisition. 81 * However, since this thread can immediately acquire the lock 82 * and does not proceed to spin on its own node->locked, this 83 * value won't be used. If a debug mode is needed to 84 * audit lock status, then set node->locked value here. 85 */ 86 return; 87 } 88 WRITE_ONCE(prev->next, node); 89 90 /* Wait until the lock holder passes the lock down. */ 91 arch_mcs_spin_lock_contended(&node->locked); 92 } 93 94 /* 95 * Releases the lock. The caller should pass in the corresponding node that 96 * was used to acquire the lock. 97 */ 98 static inline 99 void mcs_spin_unlock(struct mcs_spinlock **lock, struct mcs_spinlock *node) 100 { 101 struct mcs_spinlock *next = READ_ONCE(node->next); 102 103 if (likely(!next)) { 104 /* 105 * Release the lock by setting it to NULL 106 */ 107 if (likely(cmpxchg_release(lock, node, NULL) == node)) 108 return; 109 /* Wait until the next pointer is set */ 110 while (!(next = READ_ONCE(node->next))) 111 cpu_relax_lowlatency(); 112 } 113 114 /* Pass lock to next waiter. */ 115 arch_mcs_spin_unlock_contended(&next->locked); 116 } 117 118 #endif /* __LINUX_MCS_SPINLOCK_H */ 119