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. 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 <stdint.h> 28 #include <string.h> 29 30 #include "assembly.h" 31 32 // Clang objects if you redefine a builtin. This little hack allows us to 33 // define a function with the same name as an intrinsic. 34 #pragma redefine_extname __atomic_load_c SYMBOL_NAME(__atomic_load) 35 #pragma redefine_extname __atomic_store_c SYMBOL_NAME(__atomic_store) 36 #pragma redefine_extname __atomic_exchange_c SYMBOL_NAME(__atomic_exchange) 37 #pragma redefine_extname __atomic_compare_exchange_c SYMBOL_NAME( \ 38 __atomic_compare_exchange) 39 40 /// Number of locks. This allocates one page on 32-bit platforms, two on 41 /// 64-bit. This can be specified externally if a different trade between 42 /// memory usage and contention probability is required for a given platform. 43 #ifndef SPINLOCK_COUNT 44 #define SPINLOCK_COUNT (1 << 10) 45 #endif 46 static const long SPINLOCK_MASK = SPINLOCK_COUNT - 1; 47 48 //////////////////////////////////////////////////////////////////////////////// 49 // Platform-specific lock implementation. Falls back to spinlocks if none is 50 // defined. Each platform should define the Lock type, and corresponding 51 // lock() and unlock() functions. 52 //////////////////////////////////////////////////////////////////////////////// 53 #ifdef __FreeBSD__ 54 #include <errno.h> 55 // clang-format off 56 #include <sys/types.h> 57 #include <machine/atomic.h> 58 #include <sys/umtx.h> 59 // clang-format on 60 typedef struct _usem Lock; 61 __inline static void unlock(Lock *l) { 62 __c11_atomic_store((_Atomic(uint32_t) *)&l->_count, 1, __ATOMIC_RELEASE); 63 __c11_atomic_thread_fence(__ATOMIC_SEQ_CST); 64 if (l->_has_waiters) 65 _umtx_op(l, UMTX_OP_SEM_WAKE, 1, 0, 0); 66 } 67 __inline static void lock(Lock *l) { 68 uint32_t old = 1; 69 while (!__c11_atomic_compare_exchange_weak((_Atomic(uint32_t) *)&l->_count, 70 &old, 0, __ATOMIC_ACQUIRE, 71 __ATOMIC_RELAXED)) { 72 _umtx_op(l, UMTX_OP_SEM_WAIT, 0, 0, 0); 73 old = 1; 74 } 75 } 76 /// locks for atomic operations 77 static Lock locks[SPINLOCK_COUNT] = {[0 ... SPINLOCK_COUNT - 1] = {0, 1, 0}}; 78 79 #elif defined(__APPLE__) 80 #include <libkern/OSAtomic.h> 81 typedef OSSpinLock Lock; 82 __inline static void unlock(Lock *l) { OSSpinLockUnlock(l); } 83 /// Locks a lock. In the current implementation, this is potentially 84 /// unbounded in the contended case. 85 __inline static void lock(Lock *l) { OSSpinLockLock(l); } 86 static Lock locks[SPINLOCK_COUNT]; // initialized to OS_SPINLOCK_INIT which is 0 87 88 #else 89 typedef _Atomic(uintptr_t) Lock; 90 /// Unlock a lock. This is a release operation. 91 __inline static void unlock(Lock *l) { 92 __c11_atomic_store(l, 0, __ATOMIC_RELEASE); 93 } 94 /// Locks a lock. In the current implementation, this is potentially 95 /// unbounded in the contended case. 96 __inline static void lock(Lock *l) { 97 uintptr_t old = 0; 98 while (!__c11_atomic_compare_exchange_weak(l, &old, 1, __ATOMIC_ACQUIRE, 99 __ATOMIC_RELAXED)) 100 old = 0; 101 } 102 /// locks for atomic operations 103 static Lock locks[SPINLOCK_COUNT]; 104 #endif 105 106 /// Returns a lock to use for a given pointer. 107 static __inline Lock *lock_for_pointer(void *ptr) { 108 intptr_t hash = (intptr_t)ptr; 109 // Disregard the lowest 4 bits. We want all values that may be part of the 110 // same memory operation to hash to the same value and therefore use the same 111 // lock. 112 hash >>= 4; 113 // Use the next bits as the basis for the hash 114 intptr_t low = hash & SPINLOCK_MASK; 115 // Now use the high(er) set of bits to perturb the hash, so that we don't 116 // get collisions from atomic fields in a single object 117 hash >>= 16; 118 hash ^= low; 119 // Return a pointer to the word to use 120 return locks + (hash & SPINLOCK_MASK); 121 } 122 123 /// Macros for determining whether a size is lock free. 124 #define IS_LOCK_FREE_1 __c11_atomic_is_lock_free(1) 125 #define IS_LOCK_FREE_2 __c11_atomic_is_lock_free(2) 126 #define IS_LOCK_FREE_4 __c11_atomic_is_lock_free(4) 127 128 /// 32 bit MIPS and PowerPC don't support 8-byte lock_free atomics 129 #if defined(__mips__) || (!defined(__powerpc64__) && defined(__powerpc__)) 130 #define IS_LOCK_FREE_8 0 131 #else 132 #define IS_LOCK_FREE_8 __c11_atomic_is_lock_free(8) 133 #endif 134 135 /// Clang can not yet codegen __atomic_is_lock_free(16), so for now we assume 136 /// 16-byte values are not lock free. 137 #define IS_LOCK_FREE_16 0 138 139 /// Macro that calls the compiler-generated lock-free versions of functions 140 /// when they exist. 141 #define LOCK_FREE_CASES() \ 142 do { \ 143 switch (size) { \ 144 case 1: \ 145 if (IS_LOCK_FREE_1) { \ 146 LOCK_FREE_ACTION(uint8_t); \ 147 } \ 148 break; \ 149 case 2: \ 150 if (IS_LOCK_FREE_2) { \ 151 LOCK_FREE_ACTION(uint16_t); \ 152 } \ 153 break; \ 154 case 4: \ 155 if (IS_LOCK_FREE_4) { \ 156 LOCK_FREE_ACTION(uint32_t); \ 157 } \ 158 break; \ 159 case 8: \ 160 if (IS_LOCK_FREE_8) { \ 161 LOCK_FREE_ACTION(uint64_t); \ 162 } \ 163 break; \ 164 case 16: \ 165 if (IS_LOCK_FREE_16) { \ 166 /* FIXME: __uint128_t isn't available on 32 bit platforms. \ 167 LOCK_FREE_ACTION(__uint128_t);*/ \ 168 } \ 169 break; \ 170 } \ 171 } while (0) 172 173 /// An atomic load operation. This is atomic with respect to the source 174 /// pointer only. 175 void __atomic_load_c(int size, void *src, void *dest, int model) { 176 #define LOCK_FREE_ACTION(type) \ 177 *((type *)dest) = __c11_atomic_load((_Atomic(type) *)src, model); \ 178 return; 179 LOCK_FREE_CASES(); 180 #undef LOCK_FREE_ACTION 181 Lock *l = lock_for_pointer(src); 182 lock(l); 183 memcpy(dest, src, size); 184 unlock(l); 185 } 186 187 /// An atomic store operation. This is atomic with respect to the destination 188 /// pointer only. 189 void __atomic_store_c(int size, void *dest, void *src, int model) { 190 #define LOCK_FREE_ACTION(type) \ 191 __c11_atomic_store((_Atomic(type) *)dest, *(type *)src, model); \ 192 return; 193 LOCK_FREE_CASES(); 194 #undef LOCK_FREE_ACTION 195 Lock *l = lock_for_pointer(dest); 196 lock(l); 197 memcpy(dest, src, size); 198 unlock(l); 199 } 200 201 /// Atomic compare and exchange operation. If the value at *ptr is identical 202 /// to the value at *expected, then this copies value at *desired to *ptr. If 203 /// they are not, then this stores the current value from *ptr in *expected. 204 /// 205 /// This function returns 1 if the exchange takes place or 0 if it fails. 206 int __atomic_compare_exchange_c(int size, void *ptr, void *expected, 207 void *desired, int success, int failure) { 208 #define LOCK_FREE_ACTION(type) \ 209 return __c11_atomic_compare_exchange_strong( \ 210 (_Atomic(type) *)ptr, (type *)expected, *(type *)desired, success, \ 211 failure) 212 LOCK_FREE_CASES(); 213 #undef LOCK_FREE_ACTION 214 Lock *l = lock_for_pointer(ptr); 215 lock(l); 216 if (memcmp(ptr, expected, size) == 0) { 217 memcpy(ptr, desired, size); 218 unlock(l); 219 return 1; 220 } 221 memcpy(expected, ptr, size); 222 unlock(l); 223 return 0; 224 } 225 226 /// Performs an atomic exchange operation between two pointers. This is atomic 227 /// with respect to the target address. 228 void __atomic_exchange_c(int size, void *ptr, void *val, void *old, int model) { 229 #define LOCK_FREE_ACTION(type) \ 230 *(type *)old = \ 231 __c11_atomic_exchange((_Atomic(type) *)ptr, *(type *)val, model); \ 232 return; 233 LOCK_FREE_CASES(); 234 #undef LOCK_FREE_ACTION 235 Lock *l = lock_for_pointer(ptr); 236 lock(l); 237 memcpy(old, ptr, size); 238 memcpy(ptr, val, size); 239 unlock(l); 240 } 241 242 //////////////////////////////////////////////////////////////////////////////// 243 // Where the size is known at compile time, the compiler may emit calls to 244 // specialised versions of the above functions. 245 //////////////////////////////////////////////////////////////////////////////// 246 #ifdef __SIZEOF_INT128__ 247 #define OPTIMISED_CASES \ 248 OPTIMISED_CASE(1, IS_LOCK_FREE_1, uint8_t) \ 249 OPTIMISED_CASE(2, IS_LOCK_FREE_2, uint16_t) \ 250 OPTIMISED_CASE(4, IS_LOCK_FREE_4, uint32_t) \ 251 OPTIMISED_CASE(8, IS_LOCK_FREE_8, uint64_t) \ 252 OPTIMISED_CASE(16, IS_LOCK_FREE_16, __uint128_t) 253 #else 254 #define OPTIMISED_CASES \ 255 OPTIMISED_CASE(1, IS_LOCK_FREE_1, uint8_t) \ 256 OPTIMISED_CASE(2, IS_LOCK_FREE_2, uint16_t) \ 257 OPTIMISED_CASE(4, IS_LOCK_FREE_4, uint32_t) \ 258 OPTIMISED_CASE(8, IS_LOCK_FREE_8, uint64_t) 259 #endif 260 261 #define OPTIMISED_CASE(n, lockfree, type) \ 262 type __atomic_load_##n(type *src, int model) { \ 263 if (lockfree) \ 264 return __c11_atomic_load((_Atomic(type) *)src, model); \ 265 Lock *l = lock_for_pointer(src); \ 266 lock(l); \ 267 type val = *src; \ 268 unlock(l); \ 269 return val; \ 270 } 271 OPTIMISED_CASES 272 #undef OPTIMISED_CASE 273 274 #define OPTIMISED_CASE(n, lockfree, type) \ 275 void __atomic_store_##n(type *dest, type val, int model) { \ 276 if (lockfree) { \ 277 __c11_atomic_store((_Atomic(type) *)dest, val, model); \ 278 return; \ 279 } \ 280 Lock *l = lock_for_pointer(dest); \ 281 lock(l); \ 282 *dest = val; \ 283 unlock(l); \ 284 return; \ 285 } 286 OPTIMISED_CASES 287 #undef OPTIMISED_CASE 288 289 #define OPTIMISED_CASE(n, lockfree, type) \ 290 type __atomic_exchange_##n(type *dest, type val, int model) { \ 291 if (lockfree) \ 292 return __c11_atomic_exchange((_Atomic(type) *)dest, val, model); \ 293 Lock *l = lock_for_pointer(dest); \ 294 lock(l); \ 295 type tmp = *dest; \ 296 *dest = val; \ 297 unlock(l); \ 298 return tmp; \ 299 } 300 OPTIMISED_CASES 301 #undef OPTIMISED_CASE 302 303 #define OPTIMISED_CASE(n, lockfree, type) \ 304 bool __atomic_compare_exchange_##n(type *ptr, type *expected, type desired, \ 305 int success, int failure) { \ 306 if (lockfree) \ 307 return __c11_atomic_compare_exchange_strong( \ 308 (_Atomic(type) *)ptr, expected, desired, success, failure); \ 309 Lock *l = lock_for_pointer(ptr); \ 310 lock(l); \ 311 if (*ptr == *expected) { \ 312 *ptr = desired; \ 313 unlock(l); \ 314 return true; \ 315 } \ 316 *expected = *ptr; \ 317 unlock(l); \ 318 return false; \ 319 } 320 OPTIMISED_CASES 321 #undef OPTIMISED_CASE 322 323 //////////////////////////////////////////////////////////////////////////////// 324 // Atomic read-modify-write operations for integers of various sizes. 325 //////////////////////////////////////////////////////////////////////////////// 326 #define ATOMIC_RMW(n, lockfree, type, opname, op) \ 327 type __atomic_fetch_##opname##_##n(type *ptr, type val, int model) { \ 328 if (lockfree) \ 329 return __c11_atomic_fetch_##opname((_Atomic(type) *)ptr, val, model); \ 330 Lock *l = lock_for_pointer(ptr); \ 331 lock(l); \ 332 type tmp = *ptr; \ 333 *ptr = tmp op val; \ 334 unlock(l); \ 335 return tmp; \ 336 } 337 338 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, add, +) 339 OPTIMISED_CASES 340 #undef OPTIMISED_CASE 341 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, sub, -) 342 OPTIMISED_CASES 343 #undef OPTIMISED_CASE 344 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, and, &) 345 OPTIMISED_CASES 346 #undef OPTIMISED_CASE 347 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, or, |) 348 OPTIMISED_CASES 349 #undef OPTIMISED_CASE 350 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, xor, ^) 351 OPTIMISED_CASES 352 #undef OPTIMISED_CASE 353