4 * This file contains the main data structure and API definitions of MCS lock.
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.
12 #ifndef __LINUX_MCS_SPINLOCK_H
13 #define __LINUX_MCS_SPINLOCK_H
15 #include <asm/mcs_spinlock.h>
18 struct mcs_spinlock *next;
19 int locked; /* 1 if lock acquired */
22 #ifndef arch_mcs_spin_lock_contended
24 * Using smp_load_acquire() provides a memory barrier that ensures
25 * subsequent operations happen after the lock is acquired.
27 #define arch_mcs_spin_lock_contended(l) \
29 while (!(smp_load_acquire(l))) \
30 arch_mutex_cpu_relax(); \
34 #ifndef arch_mcs_spin_unlock_contended
36 * smp_store_release() provides a memory barrier to ensure all
37 * operations in the critical section has been completed before
40 #define arch_mcs_spin_unlock_contended(l) \
41 smp_store_release((l), 1)
45 * Note: the smp_load_acquire/smp_store_release pair is not
46 * sufficient to form a full memory barrier across
47 * cpus for many architectures (except x86) for mcs_unlock and mcs_lock.
48 * For applications that need a full barrier across multiple cpus
49 * with mcs_unlock and mcs_lock pair, smp_mb__after_unlock_lock() should be
50 * used after mcs_lock.
54 * In order to acquire the lock, the caller should declare a local node and
55 * pass a reference of the node to this function in addition to the lock.
56 * If the lock has already been acquired, then this will proceed to spin
57 * on this node->locked until the previous lock holder sets the node->locked
58 * in mcs_spin_unlock().
60 * We don't inline mcs_spin_lock() so that perf can correctly account for the
61 * time spent in this lock function.
64 void mcs_spin_lock(struct mcs_spinlock **lock, struct mcs_spinlock *node)
66 struct mcs_spinlock *prev;
72 prev = xchg(lock, node);
73 if (likely(prev == NULL)) {
75 * Lock acquired, don't need to set node->locked to 1. Threads
76 * only spin on its own node->locked value for lock acquisition.
77 * However, since this thread can immediately acquire the lock
78 * and does not proceed to spin on its own node->locked, this
79 * value won't be used. If a debug mode is needed to
80 * audit lock status, then set node->locked value here.
84 ACCESS_ONCE(prev->next) = node;
86 /* Wait until the lock holder passes the lock down. */
87 arch_mcs_spin_lock_contended(&node->locked);
91 * Releases the lock. The caller should pass in the corresponding node that
92 * was used to acquire the lock.
95 void mcs_spin_unlock(struct mcs_spinlock **lock, struct mcs_spinlock *node)
97 struct mcs_spinlock *next = ACCESS_ONCE(node->next);
101 * Release the lock by setting it to NULL
103 if (likely(cmpxchg(lock, node, NULL) == node))
105 /* Wait until the next pointer is set */
106 while (!(next = ACCESS_ONCE(node->next)))
107 arch_mutex_cpu_relax();
110 /* Pass lock to next waiter. */
111 arch_mcs_spin_unlock_contended(&next->locked);
115 * Cancellable version of the MCS lock above.
117 * Intended for adaptive spinning of sleeping locks:
118 * mutex_lock()/rwsem_down_{read,write}() etc.
121 struct optimistic_spin_queue {
122 struct optimistic_spin_queue *next, *prev;
123 int locked; /* 1 if lock acquired */
126 extern bool osq_lock(struct optimistic_spin_queue **lock);
127 extern void osq_unlock(struct optimistic_spin_queue **lock);
129 #endif /* __LINUX_MCS_SPINLOCK_H */