xref: /freebsd/contrib/llvm-project/compiler-rt/lib/builtins/atomic.c (revision 95eb4b873b6a8b527c5bd78d7191975dfca38998)
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 <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(__FreeBSD__) || defined(__DragonFly__)
60 #include <errno.h>
61 // clang-format off
62 #include <sys/types.h>
63 #include <machine/atomic.h>
64 #include <sys/umtx.h>
65 // clang-format on
66 typedef struct _usem Lock;
67 __inline static void unlock(Lock *l) {
68   __c11_atomic_store((_Atomic(uint32_t) *)&l->_count, 1, __ATOMIC_RELEASE);
69   __c11_atomic_thread_fence(__ATOMIC_SEQ_CST);
70   if (l->_has_waiters)
71     _umtx_op(l, UMTX_OP_SEM_WAKE, 1, 0, 0);
72 }
73 __inline static void lock(Lock *l) {
74   uint32_t old = 1;
75   while (!__c11_atomic_compare_exchange_weak((_Atomic(uint32_t) *)&l->_count,
76                                              &old, 0, __ATOMIC_ACQUIRE,
77                                              __ATOMIC_RELAXED)) {
78     _umtx_op(l, UMTX_OP_SEM_WAIT, 0, 0, 0);
79     old = 1;
80   }
81 }
82 /// locks for atomic operations
83 static Lock locks[SPINLOCK_COUNT] = {[0 ... SPINLOCK_COUNT - 1] = {0, 1, 0}};
84 
85 #elif defined(__APPLE__)
86 #include <libkern/OSAtomic.h>
87 typedef OSSpinLock Lock;
88 __inline static void unlock(Lock *l) { OSSpinLockUnlock(l); }
89 /// Locks a lock.  In the current implementation, this is potentially
90 /// unbounded in the contended case.
91 __inline static void lock(Lock *l) { OSSpinLockLock(l); }
92 static Lock locks[SPINLOCK_COUNT]; // initialized to OS_SPINLOCK_INIT which is 0
93 
94 #else
95 _Static_assert(__atomic_always_lock_free(sizeof(uintptr_t), 0),
96                "Implementation assumes lock-free pointer-size cmpxchg");
97 typedef _Atomic(uintptr_t) Lock;
98 /// Unlock a lock.  This is a release operation.
99 __inline static void unlock(Lock *l) {
100   __c11_atomic_store(l, 0, __ATOMIC_RELEASE);
101 }
102 /// Locks a lock.  In the current implementation, this is potentially
103 /// unbounded in the contended case.
104 __inline static void lock(Lock *l) {
105   uintptr_t old = 0;
106   while (!__c11_atomic_compare_exchange_weak(l, &old, 1, __ATOMIC_ACQUIRE,
107                                              __ATOMIC_RELAXED))
108     old = 0;
109 }
110 /// locks for atomic operations
111 static Lock locks[SPINLOCK_COUNT];
112 #endif
113 
114 /// Returns a lock to use for a given pointer.
115 static __inline Lock *lock_for_pointer(void *ptr) {
116   intptr_t hash = (intptr_t)ptr;
117   // Disregard the lowest 4 bits.  We want all values that may be part of the
118   // same memory operation to hash to the same value and therefore use the same
119   // lock.
120   hash >>= 4;
121   // Use the next bits as the basis for the hash
122   intptr_t low = hash & SPINLOCK_MASK;
123   // Now use the high(er) set of bits to perturb the hash, so that we don't
124   // get collisions from atomic fields in a single object
125   hash >>= 16;
126   hash ^= low;
127   // Return a pointer to the word to use
128   return locks + (hash & SPINLOCK_MASK);
129 }
130 
131 /// Macros for determining whether a size is lock free.
132 #define ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(size, p)                  \
133   (__atomic_always_lock_free(size, p) ||                                       \
134    (__atomic_always_lock_free(size, 0) && ((uintptr_t)p % size) == 0))
135 #define IS_LOCK_FREE_1(p) ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(1, p)
136 #define IS_LOCK_FREE_2(p) ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(2, p)
137 #define IS_LOCK_FREE_4(p) ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(4, p)
138 #define IS_LOCK_FREE_8(p) ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(8, p)
139 #define IS_LOCK_FREE_16(p) ATOMIC_ALWAYS_LOCK_FREE_OR_ALIGNED_LOCK_FREE(16, p)
140 
141 /// Macro that calls the compiler-generated lock-free versions of functions
142 /// when they exist.
143 #define TRY_LOCK_FREE_CASE(n, type, ptr)                                       \
144   case n:                                                                      \
145     if (IS_LOCK_FREE_##n(ptr)) {                                               \
146       LOCK_FREE_ACTION(type);                                                  \
147     }                                                                          \
148     break;
149 #ifdef __SIZEOF_INT128__
150 #define TRY_LOCK_FREE_CASE_16(p) TRY_LOCK_FREE_CASE(16, __uint128_t, p)
151 #else
152 #define TRY_LOCK_FREE_CASE_16(p) /* __uint128_t not available */
153 #endif
154 
155 #define LOCK_FREE_CASES(ptr)                                                   \
156   do {                                                                         \
157     switch (size) {                                                            \
158       TRY_LOCK_FREE_CASE(1, uint8_t, ptr)                                      \
159       TRY_LOCK_FREE_CASE(2, uint16_t, ptr)                                     \
160       TRY_LOCK_FREE_CASE(4, uint32_t, ptr)                                     \
161       TRY_LOCK_FREE_CASE(8, uint64_t, ptr)                                     \
162       TRY_LOCK_FREE_CASE_16(ptr) /* __uint128_t may not be supported */        \
163     default:                                                                   \
164       break;                                                                   \
165     }                                                                          \
166   } while (0)
167 
168 /// Whether atomic operations for the given size (and alignment) are lock-free.
169 bool __atomic_is_lock_free_c(size_t size, void *ptr) {
170 #define LOCK_FREE_ACTION(type) return true;
171   LOCK_FREE_CASES(ptr);
172 #undef LOCK_FREE_ACTION
173   return false;
174 }
175 
176 /// An atomic load operation.  This is atomic with respect to the source
177 /// pointer only.
178 void __atomic_load_c(int size, void *src, void *dest, int model) {
179 #define LOCK_FREE_ACTION(type)                                                 \
180   *((type *)dest) = __c11_atomic_load((_Atomic(type) *)src, model);            \
181   return;
182   LOCK_FREE_CASES(src);
183 #undef LOCK_FREE_ACTION
184   Lock *l = lock_for_pointer(src);
185   lock(l);
186   memcpy(dest, src, size);
187   unlock(l);
188 }
189 
190 /// An atomic store operation.  This is atomic with respect to the destination
191 /// pointer only.
192 void __atomic_store_c(int size, void *dest, void *src, int model) {
193 #define LOCK_FREE_ACTION(type)                                                 \
194   __c11_atomic_store((_Atomic(type) *)dest, *(type *)src, model);              \
195   return;
196   LOCK_FREE_CASES(dest);
197 #undef LOCK_FREE_ACTION
198   Lock *l = lock_for_pointer(dest);
199   lock(l);
200   memcpy(dest, src, size);
201   unlock(l);
202 }
203 
204 /// Atomic compare and exchange operation.  If the value at *ptr is identical
205 /// to the value at *expected, then this copies value at *desired to *ptr.  If
206 /// they  are not, then this stores the current value from *ptr in *expected.
207 ///
208 /// This function returns 1 if the exchange takes place or 0 if it fails.
209 int __atomic_compare_exchange_c(int size, void *ptr, void *expected,
210                                 void *desired, int success, int failure) {
211 #define LOCK_FREE_ACTION(type)                                                 \
212   return __c11_atomic_compare_exchange_strong(                                 \
213       (_Atomic(type) *)ptr, (type *)expected, *(type *)desired, success,       \
214       failure)
215   LOCK_FREE_CASES(ptr);
216 #undef LOCK_FREE_ACTION
217   Lock *l = lock_for_pointer(ptr);
218   lock(l);
219   if (memcmp(ptr, expected, size) == 0) {
220     memcpy(ptr, desired, size);
221     unlock(l);
222     return 1;
223   }
224   memcpy(expected, ptr, size);
225   unlock(l);
226   return 0;
227 }
228 
229 /// Performs an atomic exchange operation between two pointers.  This is atomic
230 /// with respect to the target address.
231 void __atomic_exchange_c(int size, void *ptr, void *val, void *old, int model) {
232 #define LOCK_FREE_ACTION(type)                                                 \
233   *(type *)old =                                                               \
234       __c11_atomic_exchange((_Atomic(type) *)ptr, *(type *)val, model);        \
235   return;
236   LOCK_FREE_CASES(ptr);
237 #undef LOCK_FREE_ACTION
238   Lock *l = lock_for_pointer(ptr);
239   lock(l);
240   memcpy(old, ptr, size);
241   memcpy(ptr, val, size);
242   unlock(l);
243 }
244 
245 ////////////////////////////////////////////////////////////////////////////////
246 // Where the size is known at compile time, the compiler may emit calls to
247 // specialised versions of the above functions.
248 ////////////////////////////////////////////////////////////////////////////////
249 #ifdef __SIZEOF_INT128__
250 #define OPTIMISED_CASES                                                        \
251   OPTIMISED_CASE(1, IS_LOCK_FREE_1, uint8_t)                                   \
252   OPTIMISED_CASE(2, IS_LOCK_FREE_2, uint16_t)                                  \
253   OPTIMISED_CASE(4, IS_LOCK_FREE_4, uint32_t)                                  \
254   OPTIMISED_CASE(8, IS_LOCK_FREE_8, uint64_t)                                  \
255   OPTIMISED_CASE(16, IS_LOCK_FREE_16, __uint128_t)
256 #else
257 #define OPTIMISED_CASES                                                        \
258   OPTIMISED_CASE(1, IS_LOCK_FREE_1, uint8_t)                                   \
259   OPTIMISED_CASE(2, IS_LOCK_FREE_2, uint16_t)                                  \
260   OPTIMISED_CASE(4, IS_LOCK_FREE_4, uint32_t)                                  \
261   OPTIMISED_CASE(8, IS_LOCK_FREE_8, uint64_t)
262 #endif
263 
264 #define OPTIMISED_CASE(n, lockfree, type)                                      \
265   type __atomic_load_##n(type *src, int model) {                               \
266     if (lockfree(src))                                                         \
267       return __c11_atomic_load((_Atomic(type) *)src, model);                   \
268     Lock *l = lock_for_pointer(src);                                           \
269     lock(l);                                                                   \
270     type val = *src;                                                           \
271     unlock(l);                                                                 \
272     return val;                                                                \
273   }
274 OPTIMISED_CASES
275 #undef OPTIMISED_CASE
276 
277 #define OPTIMISED_CASE(n, lockfree, type)                                      \
278   void __atomic_store_##n(type *dest, type val, int model) {                   \
279     if (lockfree(dest)) {                                                      \
280       __c11_atomic_store((_Atomic(type) *)dest, val, model);                   \
281       return;                                                                  \
282     }                                                                          \
283     Lock *l = lock_for_pointer(dest);                                          \
284     lock(l);                                                                   \
285     *dest = val;                                                               \
286     unlock(l);                                                                 \
287     return;                                                                    \
288   }
289 OPTIMISED_CASES
290 #undef OPTIMISED_CASE
291 
292 #define OPTIMISED_CASE(n, lockfree, type)                                      \
293   type __atomic_exchange_##n(type *dest, type val, int model) {                \
294     if (lockfree(dest))                                                        \
295       return __c11_atomic_exchange((_Atomic(type) *)dest, val, model);         \
296     Lock *l = lock_for_pointer(dest);                                          \
297     lock(l);                                                                   \
298     type tmp = *dest;                                                          \
299     *dest = val;                                                               \
300     unlock(l);                                                                 \
301     return tmp;                                                                \
302   }
303 OPTIMISED_CASES
304 #undef OPTIMISED_CASE
305 
306 #define OPTIMISED_CASE(n, lockfree, type)                                      \
307   bool __atomic_compare_exchange_##n(type *ptr, type *expected, type desired,  \
308                                      int success, int failure) {               \
309     if (lockfree(ptr))                                                         \
310       return __c11_atomic_compare_exchange_strong(                             \
311           (_Atomic(type) *)ptr, expected, desired, success, failure);          \
312     Lock *l = lock_for_pointer(ptr);                                           \
313     lock(l);                                                                   \
314     if (*ptr == *expected) {                                                   \
315       *ptr = desired;                                                          \
316       unlock(l);                                                               \
317       return true;                                                             \
318     }                                                                          \
319     *expected = *ptr;                                                          \
320     unlock(l);                                                                 \
321     return false;                                                              \
322   }
323 OPTIMISED_CASES
324 #undef OPTIMISED_CASE
325 
326 ////////////////////////////////////////////////////////////////////////////////
327 // Atomic read-modify-write operations for integers of various sizes.
328 ////////////////////////////////////////////////////////////////////////////////
329 #define ATOMIC_RMW(n, lockfree, type, opname, op)                              \
330   type __atomic_fetch_##opname##_##n(type *ptr, type val, int model) {         \
331     if (lockfree(ptr))                                                         \
332       return __c11_atomic_fetch_##opname((_Atomic(type) *)ptr, val, model);    \
333     Lock *l = lock_for_pointer(ptr);                                           \
334     lock(l);                                                                   \
335     type tmp = *ptr;                                                           \
336     *ptr = tmp op val;                                                         \
337     unlock(l);                                                                 \
338     return tmp;                                                                \
339   }
340 
341 #define ATOMIC_RMW_NAND(n, lockfree, type)                                     \
342   type __atomic_fetch_nand_##n(type *ptr, type val, int model) {               \
343     if (lockfree(ptr))                                                         \
344       return __c11_atomic_fetch_nand((_Atomic(type) *)ptr, val, model);        \
345     Lock *l = lock_for_pointer(ptr);                                           \
346     lock(l);                                                                   \
347     type tmp = *ptr;                                                           \
348     *ptr = ~(tmp & val);                                                       \
349     unlock(l);                                                                 \
350     return tmp;                                                                \
351   }
352 
353 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, add, +)
354 OPTIMISED_CASES
355 #undef OPTIMISED_CASE
356 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, sub, -)
357 OPTIMISED_CASES
358 #undef OPTIMISED_CASE
359 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, and, &)
360 OPTIMISED_CASES
361 #undef OPTIMISED_CASE
362 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, or, |)
363 OPTIMISED_CASES
364 #undef OPTIMISED_CASE
365 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW(n, lockfree, type, xor, ^)
366 OPTIMISED_CASES
367 #undef OPTIMISED_CASE
368 // Allow build with clang without __c11_atomic_fetch_nand builtin (pre-14)
369 #if __has_builtin(__c11_atomic_fetch_nand)
370 #define OPTIMISED_CASE(n, lockfree, type) ATOMIC_RMW_NAND(n, lockfree, type)
371 OPTIMISED_CASES
372 #undef OPTIMISED_CASE
373 #endif
374