1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2019 Arm Limited
3 */
4
5 #ifndef _RTE_MCSLOCK_H_
6 #define _RTE_MCSLOCK_H_
7
8 /**
9 * @file
10 *
11 * RTE MCS lock
12 *
13 * This file defines the main data structure and APIs for MCS queued lock.
14 *
15 * The MCS lock (proposed by John M. Mellor-Crummey and Michael L. Scott)
16 * provides scalability by spinning on a CPU/thread local variable which
17 * avoids expensive cache bouncings. It provides fairness by maintaining
18 * a list of acquirers and passing the lock to each CPU/thread in the order
19 * they acquired the lock.
20 */
21
22 #include <rte_lcore.h>
23 #include <rte_common.h>
24 #include <rte_pause.h>
25
26 /**
27 * The rte_mcslock_t type.
28 */
29 typedef struct rte_mcslock {
30 struct rte_mcslock *next;
31 int locked; /* 1 if the queue locked, 0 otherwise */
32 } rte_mcslock_t;
33
34 /**
35 * Take the MCS lock.
36 *
37 * @param msl
38 * A pointer to the pointer of a MCS lock.
39 * When the lock is initialized or declared, the msl pointer should be
40 * set to NULL.
41 * @param me
42 * A pointer to a new node of MCS lock. Each CPU/thread acquiring the
43 * lock should use its 'own node'.
44 */
45 static inline void
rte_mcslock_lock(rte_mcslock_t ** msl,rte_mcslock_t * me)46 rte_mcslock_lock(rte_mcslock_t **msl, rte_mcslock_t *me)
47 {
48 rte_mcslock_t *prev;
49
50 /* Init me node */
51 __atomic_store_n(&me->locked, 1, __ATOMIC_RELAXED);
52 __atomic_store_n(&me->next, NULL, __ATOMIC_RELAXED);
53
54 /* If the queue is empty, the exchange operation is enough to acquire
55 * the lock. Hence, the exchange operation requires acquire semantics.
56 * The store to me->next above should complete before the node is
57 * visible to other CPUs/threads. Hence, the exchange operation requires
58 * release semantics as well.
59 */
60 prev = __atomic_exchange_n(msl, me, __ATOMIC_ACQ_REL);
61 if (likely(prev == NULL)) {
62 /* Queue was empty, no further action required,
63 * proceed with lock taken.
64 */
65 return;
66 }
67 /* The store to me->next above should also complete before the node is
68 * visible to predecessor thread releasing the lock. Hence, the store
69 * prev->next also requires release semantics. Note that, for example,
70 * on ARM, the release semantics in the exchange operation is not
71 * strong as a release fence and is not sufficient to enforce the
72 * desired order here.
73 */
74 __atomic_store_n(&prev->next, me, __ATOMIC_RELEASE);
75
76 /* The while-load of me->locked should not move above the previous
77 * store to prev->next. Otherwise it will cause a deadlock. Need a
78 * store-load barrier.
79 */
80 __atomic_thread_fence(__ATOMIC_ACQ_REL);
81 /* If the lock has already been acquired, it first atomically
82 * places the node at the end of the queue and then proceeds
83 * to spin on me->locked until the previous lock holder resets
84 * the me->locked using mcslock_unlock().
85 */
86 while (__atomic_load_n(&me->locked, __ATOMIC_ACQUIRE))
87 rte_pause();
88 }
89
90 /**
91 * Release the MCS lock.
92 *
93 * @param msl
94 * A pointer to the pointer of a MCS lock.
95 * @param me
96 * A pointer to the node of MCS lock passed in rte_mcslock_lock.
97 */
98 static inline void
rte_mcslock_unlock(rte_mcslock_t ** msl,rte_mcslock_t * me)99 rte_mcslock_unlock(rte_mcslock_t **msl, rte_mcslock_t *me)
100 {
101 /* Check if there are more nodes in the queue. */
102 if (likely(__atomic_load_n(&me->next, __ATOMIC_RELAXED) == NULL)) {
103 /* No, last member in the queue. */
104 rte_mcslock_t *save_me = __atomic_load_n(&me, __ATOMIC_RELAXED);
105
106 /* Release the lock by setting it to NULL */
107 if (likely(__atomic_compare_exchange_n(msl, &save_me, NULL, 0,
108 __ATOMIC_RELEASE, __ATOMIC_RELAXED)))
109 return;
110
111 /* Speculative execution would be allowed to read in the
112 * while-loop first. This has the potential to cause a
113 * deadlock. Need a load barrier.
114 */
115 __atomic_thread_fence(__ATOMIC_ACQUIRE);
116 /* More nodes added to the queue by other CPUs.
117 * Wait until the next pointer is set.
118 */
119 while (__atomic_load_n(&me->next, __ATOMIC_RELAXED) == NULL)
120 rte_pause();
121 }
122
123 /* Pass lock to next waiter. */
124 __atomic_store_n(&me->next->locked, 0, __ATOMIC_RELEASE);
125 }
126
127 /**
128 * Try to take the lock.
129 *
130 * @param msl
131 * A pointer to the pointer of a MCS lock.
132 * @param me
133 * A pointer to a new node of MCS lock.
134 * @return
135 * 1 if the lock is successfully taken; 0 otherwise.
136 */
137 static inline int
rte_mcslock_trylock(rte_mcslock_t ** msl,rte_mcslock_t * me)138 rte_mcslock_trylock(rte_mcslock_t **msl, rte_mcslock_t *me)
139 {
140 /* Init me node */
141 __atomic_store_n(&me->next, NULL, __ATOMIC_RELAXED);
142
143 /* Try to lock */
144 rte_mcslock_t *expected = NULL;
145
146 /* The lock can be taken only when the queue is empty. Hence,
147 * the compare-exchange operation requires acquire semantics.
148 * The store to me->next above should complete before the node
149 * is visible to other CPUs/threads. Hence, the compare-exchange
150 * operation requires release semantics as well.
151 */
152 return __atomic_compare_exchange_n(msl, &expected, me, 0,
153 __ATOMIC_ACQ_REL, __ATOMIC_RELAXED);
154 }
155
156 /**
157 * Test if the lock is taken.
158 *
159 * @param msl
160 * A pointer to a MCS lock node.
161 * @return
162 * 1 if the lock is currently taken; 0 otherwise.
163 */
164 static inline int
rte_mcslock_is_locked(rte_mcslock_t * msl)165 rte_mcslock_is_locked(rte_mcslock_t *msl)
166 {
167 return (__atomic_load_n(&msl, __ATOMIC_RELAXED) != NULL);
168 }
169
170 #endif /* _RTE_MCSLOCK_H_ */
171