2 * Copyright (C) 2008 The Android Open Source Project
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * * Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * * Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in
12 * the documentation and/or other materials provided with the
15 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
18 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
19 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
21 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
22 * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
23 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
24 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
25 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 // Memory order requirements for POSIX semaphores appear unclear and are
30 // currently interpreted inconsistently.
31 // We conservatively prefer sequentially consistent operations for now.
32 // CAUTION: This is more conservative than some other major implementations,
33 // and may change if and when the issue is resolved.
35 #include <semaphore.h>
38 #include <stdatomic.h>
42 #include "private/bionic_constants.h"
43 #include "private/bionic_futex.h"
44 #include "private/bionic_time_conversions.h"
46 // In this implementation, a semaphore contains a
47 // 31-bit signed value and a 1-bit 'shared' flag
48 // (for process-sharing purpose).
50 // We use the value -1 to indicate contention on the
51 // semaphore, 0 or more to indicate uncontended state,
52 // any value lower than -2 is invalid at runtime.
58 // post(-1) ==> 1, then wake all waiters
62 // wait(0) ==> -1 then wait for a wake up + loop
63 // wait(-1) ==> -1 then wait for a wake up + loop
65 // Use the upper 31-bits for the counter, and the lower one
66 // for the shared flag.
67 #define SEMCOUNT_SHARED_MASK 0x00000001
68 #define SEMCOUNT_VALUE_MASK 0xfffffffe
69 #define SEMCOUNT_VALUE_SHIFT 1
71 // Convert a value into the corresponding sem->count bit pattern.
72 #define SEMCOUNT_FROM_VALUE(val) (((val) << SEMCOUNT_VALUE_SHIFT) & SEMCOUNT_VALUE_MASK)
74 // Convert a sem->count bit pattern into the corresponding signed value.
75 static inline int SEMCOUNT_TO_VALUE(unsigned int sval) {
76 return (static_cast<int>(sval) >> SEMCOUNT_VALUE_SHIFT);
79 // The value +1 as a sem->count bit-pattern.
80 #define SEMCOUNT_ONE SEMCOUNT_FROM_VALUE(1)
82 // The value -1 as a sem->count bit-pattern.
83 #define SEMCOUNT_MINUS_ONE SEMCOUNT_FROM_VALUE(-1)
85 #define SEMCOUNT_DECREMENT(sval) (((sval) - (1U << SEMCOUNT_VALUE_SHIFT)) & SEMCOUNT_VALUE_MASK)
86 #define SEMCOUNT_INCREMENT(sval) (((sval) + (1U << SEMCOUNT_VALUE_SHIFT)) & SEMCOUNT_VALUE_MASK)
88 static inline atomic_uint* SEM_TO_ATOMIC_POINTER(sem_t* sem) {
89 static_assert(sizeof(atomic_uint) == sizeof(sem->count),
90 "sem->count should actually be atomic_uint in implementation.");
92 // We prefer casting to atomic_uint instead of declaring sem->count to be atomic_uint directly.
93 // Because using the second method pollutes semaphore.h.
94 return reinterpret_cast<atomic_uint*>(&sem->count);
97 // Return the shared bitflag from a semaphore counter.
98 static inline unsigned int SEM_GET_SHARED(atomic_uint* sem_count_ptr) {
99 // memory_order_relaxed is used as SHARED flag will not be changed after init.
100 return (atomic_load_explicit(sem_count_ptr, memory_order_relaxed) & SEMCOUNT_SHARED_MASK);
103 int sem_init(sem_t* sem, int pshared, unsigned int value) {
104 // Ensure that 'value' can be stored in the semaphore.
105 if (value > SEM_VALUE_MAX) {
110 unsigned int count = SEMCOUNT_FROM_VALUE(value);
112 count |= SEMCOUNT_SHARED_MASK;
115 atomic_uint* sem_count_ptr = SEM_TO_ATOMIC_POINTER(sem);
116 atomic_init(sem_count_ptr, count);
120 int sem_destroy(sem_t*) {
124 sem_t* sem_open(const char*, int, ...) {
129 int sem_close(sem_t*) {
134 int sem_unlink(const char*) {
139 // Decrement a semaphore's value atomically,
140 // and return the old one. As a special case,
141 // this returns immediately if the value is
142 // negative (i.e. -1)
143 static int __sem_dec(atomic_uint* sem_count_ptr) {
144 unsigned int old_value = atomic_load_explicit(sem_count_ptr, memory_order_relaxed);
145 unsigned int shared = old_value & SEMCOUNT_SHARED_MASK;
147 // Use memory_order_seq_cst in atomic_compare_exchange operation to ensure all
148 // memory access made by other threads can be seen in current thread.
149 // An acquire fence may be sufficient, but it is still in discussion whether
150 // POSIX semaphores should provide sequential consistency.
152 if (SEMCOUNT_TO_VALUE(old_value) < 0) {
155 } while (!atomic_compare_exchange_weak(sem_count_ptr, &old_value,
156 SEMCOUNT_DECREMENT(old_value) | shared));
158 return SEMCOUNT_TO_VALUE(old_value);
161 // Same as __sem_dec, but will not touch anything if the
162 // value is already negative *or* 0. Returns the old value.
163 static int __sem_trydec(atomic_uint* sem_count_ptr) {
164 unsigned int old_value = atomic_load_explicit(sem_count_ptr, memory_order_relaxed);
165 unsigned int shared = old_value & SEMCOUNT_SHARED_MASK;
167 // Use memory_order_seq_cst in atomic_compare_exchange operation to ensure all
168 // memory access made by other threads can be seen in current thread.
169 // An acquire fence may be sufficient, but it is still in discussion whether
170 // POSIX semaphores should provide sequential consistency.
172 if (SEMCOUNT_TO_VALUE(old_value) <= 0) {
175 } while (!atomic_compare_exchange_weak(sem_count_ptr, &old_value,
176 SEMCOUNT_DECREMENT(old_value) | shared));
178 return SEMCOUNT_TO_VALUE(old_value);
181 // "Increment" the value of a semaphore atomically and
182 // return its old value. Note that this implements
183 // the special case of "incrementing" any negative
184 // value to +1 directly.
186 // NOTE: The value will _not_ wrap above SEM_VALUE_MAX
187 static int __sem_inc(atomic_uint* sem_count_ptr) {
188 unsigned int old_value = atomic_load_explicit(sem_count_ptr, memory_order_relaxed);
189 unsigned int shared = old_value & SEMCOUNT_SHARED_MASK;
190 unsigned int new_value;
192 // Use memory_order_seq_cst in atomic_compare_exchange operation to ensure all
193 // memory access made before can be seen in other threads.
194 // A release fence may be sufficient, but it is still in discussion whether
195 // POSIX semaphores should provide sequential consistency.
197 // Can't go higher than SEM_VALUE_MAX.
198 if (SEMCOUNT_TO_VALUE(old_value) == SEM_VALUE_MAX) {
202 // If the counter is negative, go directly to one, otherwise just increment.
203 if (SEMCOUNT_TO_VALUE(old_value) < 0) {
204 new_value = SEMCOUNT_ONE | shared;
206 new_value = SEMCOUNT_INCREMENT(old_value) | shared;
208 } while (!atomic_compare_exchange_weak(sem_count_ptr, &old_value,
211 return SEMCOUNT_TO_VALUE(old_value);
214 int sem_wait(sem_t* sem) {
215 atomic_uint* sem_count_ptr = SEM_TO_ATOMIC_POINTER(sem);
216 unsigned int shared = SEM_GET_SHARED(sem_count_ptr);
219 if (__sem_dec(sem_count_ptr) > 0) {
223 __futex_wait_ex(sem_count_ptr, shared, shared | SEMCOUNT_MINUS_ONE, NULL);
227 int sem_timedwait(sem_t* sem, const timespec* abs_timeout) {
228 atomic_uint* sem_count_ptr = SEM_TO_ATOMIC_POINTER(sem);
230 // POSIX says we need to try to decrement the semaphore
231 // before checking the timeout value. Note that if the
232 // value is currently 0, __sem_trydec() does nothing.
233 if (__sem_trydec(sem_count_ptr) > 0) {
237 // Check it as per POSIX.
238 if (abs_timeout == NULL || abs_timeout->tv_sec < 0 || abs_timeout->tv_nsec < 0 || abs_timeout->tv_nsec >= NS_PER_S) {
243 unsigned int shared = SEM_GET_SHARED(sem_count_ptr);
246 // POSIX mandates CLOCK_REALTIME here.
248 if (!timespec_from_absolute_timespec(ts, *abs_timeout, CLOCK_REALTIME)) {
253 // Try to grab the semaphore. If the value was 0, this will also change it to -1.
254 if (__sem_dec(sem_count_ptr) > 0) {
258 // Contention detected. Wait for a wakeup event.
259 int ret = __futex_wait_ex(sem_count_ptr, shared, shared | SEMCOUNT_MINUS_ONE, &ts);
261 // Return in case of timeout or interrupt.
262 if (ret == -ETIMEDOUT || ret == -EINTR) {
270 int sem_post(sem_t* sem) {
271 atomic_uint* sem_count_ptr = SEM_TO_ATOMIC_POINTER(sem);
272 unsigned int shared = SEM_GET_SHARED(sem_count_ptr);
274 int old_value = __sem_inc(sem_count_ptr);
276 // Contention on the semaphore. Wake up all waiters.
277 __futex_wake_ex(sem_count_ptr, shared, INT_MAX);
278 } else if (old_value == SEM_VALUE_MAX) {
279 // Overflow detected.
287 int sem_trywait(sem_t* sem) {
288 atomic_uint* sem_count_ptr = SEM_TO_ATOMIC_POINTER(sem);
289 if (__sem_trydec(sem_count_ptr) > 0) {
297 int sem_getvalue(sem_t* sem, int* sval) {
298 atomic_uint* sem_count_ptr = SEM_TO_ATOMIC_POINTER(sem);
300 // Use memory_order_seq_cst in atomic_load operation.
301 // memory_order_relaxed may be fine here, but it is still in discussion
302 // whether POSIX semaphores should provide sequential consistency.
303 int val = SEMCOUNT_TO_VALUE(atomic_load(sem_count_ptr));