minmax: fix up min3() and max3() too
[linux-block.git] / kernel / locking / osq_lock.c
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/percpu.h>
3 #include <linux/sched.h>
4 #include <linux/osq_lock.h>
5
6 /*
7  * An MCS like lock especially tailored for optimistic spinning for sleeping
8  * lock implementations (mutex, rwsem, etc).
9  *
10  * Using a single mcs node per CPU is safe because sleeping locks should not be
11  * called from interrupt context and we have preemption disabled while
12  * spinning.
13  */
14
15 struct optimistic_spin_node {
16         struct optimistic_spin_node *next, *prev;
17         int locked; /* 1 if lock acquired */
18         int cpu; /* encoded CPU # + 1 value */
19 };
20
21 static DEFINE_PER_CPU_SHARED_ALIGNED(struct optimistic_spin_node, osq_node);
22
23 /*
24  * We use the value 0 to represent "no CPU", thus the encoded value
25  * will be the CPU number incremented by 1.
26  */
27 static inline int encode_cpu(int cpu_nr)
28 {
29         return cpu_nr + 1;
30 }
31
32 static inline int node_cpu(struct optimistic_spin_node *node)
33 {
34         return node->cpu - 1;
35 }
36
37 static inline struct optimistic_spin_node *decode_cpu(int encoded_cpu_val)
38 {
39         int cpu_nr = encoded_cpu_val - 1;
40
41         return per_cpu_ptr(&osq_node, cpu_nr);
42 }
43
44 /*
45  * Get a stable @node->next pointer, either for unlock() or unqueue() purposes.
46  * Can return NULL in case we were the last queued and we updated @lock instead.
47  *
48  * If osq_lock() is being cancelled there must be a previous node
49  * and 'old_cpu' is its CPU #.
50  * For osq_unlock() there is never a previous node and old_cpu is
51  * set to OSQ_UNLOCKED_VAL.
52  */
53 static inline struct optimistic_spin_node *
54 osq_wait_next(struct optimistic_spin_queue *lock,
55               struct optimistic_spin_node *node,
56               int old_cpu)
57 {
58         int curr = encode_cpu(smp_processor_id());
59
60         for (;;) {
61                 if (atomic_read(&lock->tail) == curr &&
62                     atomic_cmpxchg_acquire(&lock->tail, curr, old_cpu) == curr) {
63                         /*
64                          * We were the last queued, we moved @lock back. @prev
65                          * will now observe @lock and will complete its
66                          * unlock()/unqueue().
67                          */
68                         return NULL;
69                 }
70
71                 /*
72                  * We must xchg() the @node->next value, because if we were to
73                  * leave it in, a concurrent unlock()/unqueue() from
74                  * @node->next might complete Step-A and think its @prev is
75                  * still valid.
76                  *
77                  * If the concurrent unlock()/unqueue() wins the race, we'll
78                  * wait for either @lock to point to us, through its Step-B, or
79                  * wait for a new @node->next from its Step-C.
80                  */
81                 if (node->next) {
82                         struct optimistic_spin_node *next;
83
84                         next = xchg(&node->next, NULL);
85                         if (next)
86                                 return next;
87                 }
88
89                 cpu_relax();
90         }
91 }
92
93 bool osq_lock(struct optimistic_spin_queue *lock)
94 {
95         struct optimistic_spin_node *node = this_cpu_ptr(&osq_node);
96         struct optimistic_spin_node *prev, *next;
97         int curr = encode_cpu(smp_processor_id());
98         int old;
99
100         node->locked = 0;
101         node->next = NULL;
102         node->cpu = curr;
103
104         /*
105          * We need both ACQUIRE (pairs with corresponding RELEASE in
106          * unlock() uncontended, or fastpath) and RELEASE (to publish
107          * the node fields we just initialised) semantics when updating
108          * the lock tail.
109          */
110         old = atomic_xchg(&lock->tail, curr);
111         if (old == OSQ_UNLOCKED_VAL)
112                 return true;
113
114         prev = decode_cpu(old);
115         node->prev = prev;
116
117         /*
118          * osq_lock()                   unqueue
119          *
120          * node->prev = prev            osq_wait_next()
121          * WMB                          MB
122          * prev->next = node            next->prev = prev // unqueue-C
123          *
124          * Here 'node->prev' and 'next->prev' are the same variable and we need
125          * to ensure these stores happen in-order to avoid corrupting the list.
126          */
127         smp_wmb();
128
129         WRITE_ONCE(prev->next, node);
130
131         /*
132          * Normally @prev is untouchable after the above store; because at that
133          * moment unlock can proceed and wipe the node element from stack.
134          *
135          * However, since our nodes are static per-cpu storage, we're
136          * guaranteed their existence -- this allows us to apply
137          * cmpxchg in an attempt to undo our queueing.
138          */
139
140         /*
141          * Wait to acquire the lock or cancellation. Note that need_resched()
142          * will come with an IPI, which will wake smp_cond_load_relaxed() if it
143          * is implemented with a monitor-wait. vcpu_is_preempted() relies on
144          * polling, be careful.
145          */
146         if (smp_cond_load_relaxed(&node->locked, VAL || need_resched() ||
147                                   vcpu_is_preempted(node_cpu(node->prev))))
148                 return true;
149
150         /* unqueue */
151         /*
152          * Step - A  -- stabilize @prev
153          *
154          * Undo our @prev->next assignment; this will make @prev's
155          * unlock()/unqueue() wait for a next pointer since @lock points to us
156          * (or later).
157          */
158
159         for (;;) {
160                 /*
161                  * cpu_relax() below implies a compiler barrier which would
162                  * prevent this comparison being optimized away.
163                  */
164                 if (data_race(prev->next) == node &&
165                     cmpxchg(&prev->next, node, NULL) == node)
166                         break;
167
168                 /*
169                  * We can only fail the cmpxchg() racing against an unlock(),
170                  * in which case we should observe @node->locked becoming
171                  * true.
172                  */
173                 if (smp_load_acquire(&node->locked))
174                         return true;
175
176                 cpu_relax();
177
178                 /*
179                  * Or we race against a concurrent unqueue()'s step-B, in which
180                  * case its step-C will write us a new @node->prev pointer.
181                  */
182                 prev = READ_ONCE(node->prev);
183         }
184
185         /*
186          * Step - B -- stabilize @next
187          *
188          * Similar to unlock(), wait for @node->next or move @lock from @node
189          * back to @prev.
190          */
191
192         next = osq_wait_next(lock, node, prev->cpu);
193         if (!next)
194                 return false;
195
196         /*
197          * Step - C -- unlink
198          *
199          * @prev is stable because its still waiting for a new @prev->next
200          * pointer, @next is stable because our @node->next pointer is NULL and
201          * it will wait in Step-A.
202          */
203
204         WRITE_ONCE(next->prev, prev);
205         WRITE_ONCE(prev->next, next);
206
207         return false;
208 }
209
210 void osq_unlock(struct optimistic_spin_queue *lock)
211 {
212         struct optimistic_spin_node *node, *next;
213         int curr = encode_cpu(smp_processor_id());
214
215         /*
216          * Fast path for the uncontended case.
217          */
218         if (likely(atomic_cmpxchg_release(&lock->tail, curr,
219                                           OSQ_UNLOCKED_VAL) == curr))
220                 return;
221
222         /*
223          * Second most likely case.
224          */
225         node = this_cpu_ptr(&osq_node);
226         next = xchg(&node->next, NULL);
227         if (next) {
228                 WRITE_ONCE(next->locked, 1);
229                 return;
230         }
231
232         next = osq_wait_next(lock, node, OSQ_UNLOCKED_VAL);
233         if (next)
234                 WRITE_ONCE(next->locked, 1);
235 }