1 //===-- atomic.c - Implement support functions for atomic operations.------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // atomic.c defines a set of functions for performing atomic accesses on
10 // arbitrary-sized memory locations. This design uses locks that should
11 // be fast in the uncontended case, for two reasons:
12 //
13 // 1) This code must work with C programs that do not link to anything
14 // (including pthreads) and so it should not depend on any pthread
15 // functions. If the user wishes to opt into using pthreads, they may do so.
16 // 2) Atomic operations, rather than explicit mutexes, are most commonly used
17 // on code where contended operations are rate.
18 //
19 // To avoid needing a per-object lock, this code allocates an array of
20 // locks and hashes the object pointers to find the one that it should use.
21 // For operations that must be atomic on two locations, the lower lock is
22 // always acquired first, to avoid deadlock.
23 //
24 //===----------------------------------------------------------------------===//
25
26 #include <stdbool.h>
27 #include <stddef.h>
28 #include <stdint.h>
29
30 #include "assembly.h"
31
32 // We use __builtin_mem* here to avoid dependencies on libc-provided headers.
33 #define memcpy __builtin_memcpy
34 #define memcmp __builtin_memcmp
35
36 // Clang objects if you redefine a builtin. This little hack allows us to
37 // define a function with the same name as an intrinsic.
38 #pragma redefine_extname __atomic_load_c SYMBOL_NAME(__atomic_load)
39 #pragma redefine_extname __atomic_store_c SYMBOL_NAME(__atomic_store)
40 #pragma redefine_extname __atomic_exchange_c SYMBOL_NAME(__atomic_exchange)
41 #pragma redefine_extname __atomic_compare_exchange_c SYMBOL_NAME( \
42 __atomic_compare_exchange)
43 #pragma redefine_extname __atomic_is_lock_free_c SYMBOL_NAME( \
44 __atomic_is_lock_free)
45
46 /// Number of locks. This allocates one page on 32-bit platforms, two on
47 /// 64-bit. This can be specified externally if a different trade between
48 /// memory usage and contention probability is required for a given platform.
49 #ifndef SPINLOCK_COUNT
50 #define SPINLOCK_COUNT (1 << 10)
51 #endif
52 static const long SPINLOCK_MASK = SPINLOCK_COUNT - 1;
53
54 ////////////////////////////////////////////////////////////////////////////////
55 // Platform-specific lock implementation. Falls back to spinlocks if none is
56 // defined. Each platform should define the Lock type, and corresponding
57 // lock() and unlock() functions.
58 ////////////////////////////////////////////////////////////////////////////////
59 #if defined(_LIBATOMIC_USE_PTHREAD)
60 #include <pthread.h>
61 typedef pthread_mutex_t Lock;
62 /// Unlock a lock. This is a release operation.
unlock(Lock * l)63 __inline static void unlock(Lock *l) { pthread_mutex_unlock(l); }
64 /// Locks a lock.
lock(Lock * l)65 __inline static void lock(Lock *l) { pthread_mutex_lock(l); }
66 /// locks for atomic operations
67 static Lock locks[SPINLOCK_COUNT];
68
69 #elif defined(__FreeBSD__) || defined(__DragonFly__)
70 #include <errno.h>
71 // clang-format off
72 #include <sys/types.h>
73 #include <machine/atomic.h>
74 #include <sys/umtx.h>
75 // clang-format on
76 typedef struct _usem Lock;
unlock(Lock * l)77 __inline static void unlock(Lock *l) {
78 __c11_atomic_store((_Atomic(uint32_t) *)&l->_count, 1, __ATOMIC_RELEASE);
79 __c11_atomic_thread_fence(__ATOMIC_SEQ_CST);
80 if (l->_has_waiters)
81 _umtx_op(l, UMTX_OP_SEM_WAKE, 1, 0, 0);
82 }
lock(Lock * l)83 __inline static void lock(Lock *l) {
84 uint32_t old = 1;
85 while (!__c11_atomic_compare_exchange_weak((_Atomic(uint32_t) *)&l->_count,
86 &old, 0, __ATOMIC_ACQUIRE,
87 __ATOMIC_RELAXED)) {
88 _umtx_op(l, UMTX_OP_SEM_WAIT, 0, 0, 0);
89 old = 1;
90 }
91 }
92 /// locks for atomic operations
93 static Lock locks[SPINLOCK_COUNT] = {[0 ... SPINLOCK_COUNT - 1] = {0, 1, 0}};
94
95 #elif defined(__APPLE__)
96 #include <libkern/OSAtomic.h>
97 typedef OSSpinLock Lock;
unlock(Lock * l)98 __inline static void unlock(Lock *l) { OSSpinLockUnlock(l); }
99 /// Locks a lock. In the current implementation, this is potentially
100 /// unbounded in the contended case.
lock(Lock * l)101 __inline static void lock(Lock *l) { OSSpinLockLock(l); }
102 static Lock locks[SPINLOCK_COUNT]; // initialized to OS_SPINLOCK_INIT which is 0
103
104 #else
105 _Static_assert(__atomic_always_lock_free(sizeof(uintptr_t), 0),
106 "Implementation assumes lock-free pointer-size cmpxchg");
107 typedef _Atomic(uintptr_t) Lock;
108 /// Unlock a lock. This is a release operation.
unlock(Lock * l)109 __inline static void unlock(Lock *l) {
110 __c11_atomic_store(l, 0, __ATOMIC_RELEASE);
111 }
112 /// Locks a lock. In the current implementation, this is potentially
113 /// unbounded in the contended case.
lock(Lock * l)114 __inline static void lock(Lock *l) {
115 uintptr_t old = 0;
116 while (!__c11_atomic_compare_exchange_weak(l, &old, 1, __ATOMIC_ACQUIRE,
117 __ATOMIC_RELAXED))
118 old = 0;
119 }
120 /// locks for atomic operations
121 static Lock locks[SPINLOCK_COUNT];
122 #endif
123
124 /// Returns a lock to use for a given pointer.
lock_for_pointer(void * ptr)125 static __inline Lock *lock_for_pointer(void *ptr) {
126 intptr_t hash = (intptr_t)ptr;
127 // Disregard the lowest 4 bits. We want all values that may be part of the
128 // same memory operation to hash to the same value and therefore use the same
129 // lock.
130 hash >>= 4;
131 // Use the next bits as the basis for the hash
132 intptr_t low = hash & SPINLOCK_MASK;
133 // Now use the high(er) set of bits to perturb the hash, so that we don't
134 // get collisions from atomic fields in a single object
135 hash >>= 16;
136 hash ^= low;
137 // Return a pointer to the word to use
138 return locks + (hash & SPINLOCK_MASK);
139 }
140
141 /// Macros for determining whether a size is lock free.
142 #define ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(size, p) \
143 (__atomic_always_lock_free(size, p) || \
144 (__atomic_always_lock_free(size, 0) && ((uintptr_t)p % size) == 0))
145 #define IS_LOCK_FREE_1(p) ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(1, p)
146 #define IS_LOCK_FREE_2(p) ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(2, p)
147 #define IS_LOCK_FREE_4(p) ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(4, p)
148 #define IS_LOCK_FREE_8(p) ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(8, p)
149 #define IS_LOCK_FREE_16(p) ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(16, p)
150
151 /// Macro that calls the compiler-generated lock-free versions of functions
152 /// when they exist.
153 #define TRY_LOCK_FREE_CASE(n, type, ptr) \
154 case n: \
155 if (IS_LOCK_FREE_##n(ptr)) { \
156 LOCK_FREE_ACTION(type); \
157 } \
158 break;
159 #ifdef __SIZEOF_INT128__
160 #define TRY_LOCK_FREE_CASE_16(p) TRY_LOCK_FREE_CASE(16, __uint128_t, p)
161 #else
162 #define TRY_LOCK_FREE_CASE_16(p) /* __uint128_t not available */
163 #endif
164
165 #define LOCK_FREE_CASES(ptr) \
166 do { \
167 switch (size) { \
168 TRY_LOCK_FREE_CASE(1, uint8_t, ptr) \
169 TRY_LOCK_FREE_CASE(2, uint16_t, ptr) \
170 TRY_LOCK_FREE_CASE(4, uint32_t, ptr) \
171 TRY_LOCK_FREE_CASE(8, uint64_t, ptr) \
172 TRY_LOCK_FREE_CASE_16(ptr) /* __uint128_t may not be supported */ \
173 default: \
174 break; \
175 } \
176 } while (0)
177
178 /// Whether atomic operations for the given size (and alignment) are lock-free.
__atomic_is_lock_free_c(size_t size,void * ptr)179 bool __atomic_is_lock_free_c(size_t size, void *ptr) {
180 #define LOCK_FREE_ACTION(type) return true;
181 LOCK_FREE_CASES(ptr);
182 #undef LOCK_FREE_ACTION
183 return false;
184 }
185
186 /// An atomic load operation. This is atomic with respect to the source
187 /// pointer only.
__atomic_load_c(int size,void * src,void * dest,int model)188 void __atomic_load_c(int size, void *src, void *dest, int model) {
189 #define LOCK_FREE_ACTION(type) \
190 *((type *)dest) = __c11_atomic_load((_Atomic(type) *)src, model); \
191 return;
192 LOCK_FREE_CASES(src);
193 #undef LOCK_FREE_ACTION
194 Lock *l = lock_for_pointer(src);
195 lock(l);
196 memcpy(dest, src, size);
197 unlock(l);
198 }
199
200 /// An atomic store operation. This is atomic with respect to the destination
201 /// pointer only.
__atomic_store_c(int size,void * dest,void * src,int model)202 void __atomic_store_c(int size, void *dest, void *src, int model) {
203 #define LOCK_FREE_ACTION(type) \
204 __c11_atomic_store((_Atomic(type) *)dest, *(type *)src, model); \
205 return;
206 LOCK_FREE_CASES(dest);
207 #undef LOCK_FREE_ACTION
208 Lock *l = lock_for_pointer(dest);
209 lock(l);
210 memcpy(dest, src, size);
211 unlock(l);
212 }
213
214 /// Atomic compare and exchange operation. If the value at *ptr is identical
215 /// to the value at *expected, then this copies value at *desired to *ptr. If
216 /// they are not, then this stores the current value from *ptr in *expected.
217 ///
218 /// This function returns 1 if the exchange takes place or 0 if it fails.
__atomic_compare_exchange_c(int size,void * ptr,void * expected,void * desired,int success,int failure)219 int __atomic_compare_exchange_c(int size, void *ptr, void *expected,
220 void *desired, int success, int failure) {
221 #define LOCK_FREE_ACTION(type) \
222 return __c11_atomic_compare_exchange_strong( \
223 (_Atomic(type) *)ptr, (type *)expected, *(type *)desired, success, \
224 failure)
225 LOCK_FREE_CASES(ptr);
226 #undef LOCK_FREE_ACTION
227 Lock *l = lock_for_pointer(ptr);
228 lock(l);
229 if (memcmp(ptr, expected, size) == 0) {
230 memcpy(ptr, desired, size);
231 unlock(l);
232 return 1;
233 }
234 memcpy(expected, ptr, size);
235 unlock(l);
236 return 0;
237 }
238
239 /// Performs an atomic exchange operation between two pointers. This is atomic
240 /// with respect to the target address.
__atomic_exchange_c(int size,void * ptr,void * val,void * old,int model)241 void __atomic_exchange_c(int size, void *ptr, void *val, void *old, int model) {
242 #define LOCK_FREE_ACTION(type) \
243 *(type *)old = \
244 __c11_atomic_exchange((_Atomic(type) *)ptr, *(type *)val, model); \
245 return;
246 LOCK_FREE_CASES(ptr);
247 #undef LOCK_FREE_ACTION
248 Lock *l = lock_for_pointer(ptr);
249 lock(l);
250 memcpy(old, ptr, size);
251 memcpy(ptr, val, size);
252 unlock(l);
253 }
254
255 ////////////////////////////////////////////////////////////////////////////////
256 // Where the size is known at compile time, the compiler may emit calls to
257 // specialised versions of the above functions.
258 ////////////////////////////////////////////////////////////////////////////////
259 #ifdef __SIZEOF_INT128__
260 #define OPTIMISED_CASES \
261 OPTIMISED_CASE(1, IS_LOCK_FREE_1, uint8_t) \
262 OPTIMISED_CASE(2, IS_LOCK_FREE_2, uint16_t) \
263 OPTIMISED_CASE(4, IS_LOCK_FREE_4, uint32_t) \
264 OPTIMISED_CASE(8, IS_LOCK_FREE_8, uint64_t) \
265 OPTIMISED_CASE(16, IS_LOCK_FREE_16, __uint128_t)
266 #else
267 #define OPTIMISED_CASES \
268 OPTIMISED_CASE(1, IS_LOCK_FREE_1, uint8_t) \
269 OPTIMISED_CASE(2, IS_LOCK_FREE_2, uint16_t) \
270 OPTIMISED_CASE(4, IS_LOCK_FREE_4, uint32_t) \
271 OPTIMISED_CASE(8, IS_LOCK_FREE_8, uint64_t)
272 #endif
273
274 #define OPTIMISED_CASE(n, lockfree, type) \
275 type __atomic_load_##n(type *src, int model) { \
276 if (lockfree(src)) \
277 return __c11_atomic_load((_Atomic(type) *)src, model); \
278 Lock *l = lock_for_pointer(src); \
279 lock(l); \
280 type val = *src; \
281 unlock(l); \
282 return val; \
283 }
284 OPTIMISED_CASES
285 #undef OPTIMISED_CASE
286
287 #define OPTIMISED_CASE(n, lockfree, type) \
288 void __atomic_store_##n(type *dest, type val, int model) { \
289 if (lockfree(dest)) { \
290 __c11_atomic_store((_Atomic(type) *)dest, val, model); \
291 return; \
292 } \
293 Lock *l = lock_for_pointer(dest); \
294 lock(l); \
295 *dest = val; \
296 unlock(l); \
297 return; \
298 }
299 OPTIMISED_CASES
300 #undef OPTIMISED_CASE
301
302 #define OPTIMISED_CASE(n, lockfree, type) \
303 type __atomic_exchange_##n(type *dest, type val, int model) { \
304 if (lockfree(dest)) \
305 return __c11_atomic_exchange((_Atomic(type) *)dest, val, model); \
306 Lock *l = lock_for_pointer(dest); \
307 lock(l); \
308 type tmp = *dest; \
309 *dest = val; \
310 unlock(l); \
311 return tmp; \
312 }
313 OPTIMISED_CASES
314 #undef OPTIMISED_CASE
315
316 #define OPTIMISED_CASE(n, lockfree, type) \
317 bool __atomic_compare_exchange_##n(type *ptr, type *expected, type desired, \
318 int success, int failure) { \
319 if (lockfree(ptr)) \
320 return __c11_atomic_compare_exchange_strong( \
321 (_Atomic(type) *)ptr, expected, desired, success, failure); \
322 Lock *l = lock_for_pointer(ptr); \
323 lock(l); \
324 if (*ptr == *expected) { \
325 *ptr = desired; \
326 unlock(l); \
327 return true; \
328 } \
329 *expected = *ptr; \
330 unlock(l); \
331 return false; \
332 }
333 OPTIMISED_CASES
334 #undef OPTIMISED_CASE
335
336 ////////////////////////////////////////////////////////////////////////////////
337 // Atomic read-modify-write operations for integers of various sizes.
338 ////////////////////////////////////////////////////////////////////////////////
339 #define ATOMIC_RMW(n, lockfree, type, opname, op) \
340 type __atomic_fetch_##opname##_##n(type *ptr, type val, int model) { \
341 if (lockfree(ptr)) \
342 return __c11_atomic_fetch_##opname((_Atomic(type) *)ptr, val, model); \
343 Lock *l = lock_for_pointer(ptr); \
344 lock(l); \
345 type tmp = *ptr; \
346 *ptr = tmp op val; \
347 unlock(l); \
348 return tmp; \
349 }
350
351 #define ATOMIC_RMW_NAND(n, lockfree, type) \
352 type __atomic_fetch_nand_##n(type *ptr, type val, int model) { \
353 if (lockfree(ptr)) \
354 return __c11_atomic_fetch_nand((_Atomic(type) *)ptr, val, model); \
355 Lock *l = lock_for_pointer(ptr); \
356 lock(l); \
357 type tmp = *ptr; \
358 *ptr = ~(tmp & val); \
359 unlock(l); \
360 return tmp; \
361 }
362
363 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, add, +)
364 OPTIMISED_CASES
365 #undef OPTIMISED_CASE
366 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, sub, -)
367 OPTIMISED_CASES
368 #undef OPTIMISED_CASE
369 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, and, &)
370 OPTIMISED_CASES
371 #undef OPTIMISED_CASE
372 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, or, |)
373 OPTIMISED_CASES
374 #undef OPTIMISED_CASE
375 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, xor, ^)
376 OPTIMISED_CASES
377 #undef OPTIMISED_CASE
378 // Allow build with clang without __c11_atomic_fetch_nand builtin (pre-14)
379 #if __has_builtin(__c11_atomic_fetch_nand)
380 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW_NAND(n, lockfree, type)
381 OPTIMISED_CASES
382 #undef OPTIMISED_CASE
383 #endif
384