1 /*
2   Simple DirectMedia Layer
3   Copyright (C) 1997-2020 Sam Lantinga <slouken@libsdl.org>
4 
5   This software is provided 'as-is', without any express or implied
6   warranty.  In no event will the authors be held liable for any damages
7   arising from the use of this software.
8 
9   Permission is granted to anyone to use this software for any purpose,
10   including commercial applications, and to alter it and redistribute it
11   freely, subject to the following restrictions:
12 
13   1. The origin of this software must not be misrepresented; you must not
14      claim that you wrote the original software. If you use this software
15      in a product, an acknowledgment in the product documentation would be
16      appreciated but is not required.
17   2. Altered source versions must be plainly marked as such, and must not be
18      misrepresented as being the original software.
19   3. This notice may not be removed or altered from any source distribution.
20 */
21 #include "../SDL_internal.h"
22 
23 #include "SDL_atomic.h"
24 
25 #if defined(_MSC_VER) && (_MSC_VER >= 1500)
26 #include <intrin.h>
27 #define HAVE_MSC_ATOMICS 1
28 #endif
29 
30 #if defined(__MACOSX__)  /* !!! FIXME: should we favor gcc atomics? */
31 #include <libkern/OSAtomic.h>
32 #endif
33 
34 #if !defined(HAVE_GCC_ATOMICS) && defined(__SOLARIS__)
35 #include <atomic.h>
36 #endif
37 
38 /* The __atomic_load_n() intrinsic showed up in different times for different compilers. */
39 #if defined(HAVE_GCC_ATOMICS)
40 # if defined(__clang__)
41 #   if __has_builtin(__atomic_load_n)
42       /* !!! FIXME: this advertises as available in the NDK but uses an external symbol we don't have.
43          It might be in a later NDK or we might need an extra library? --ryan. */
44 #     if !defined(__ANDROID__)
45 #       define HAVE_ATOMIC_LOAD_N 1
46 #     endif
47 #   endif
48 # elif defined(__GNUC__)
49 #   if (__GNUC__ >= 5)
50 #     define HAVE_ATOMIC_LOAD_N 1
51 #   endif
52 # endif
53 #endif
54 
55 #if defined(__WATCOMC__) && defined(__386__)
56 SDL_COMPILE_TIME_ASSERT(intsize, 4==sizeof(int));
57 #define HAVE_WATCOM_ATOMICS
58 extern _inline int _SDL_xchg_watcom(volatile int *a, int v);
59 #pragma aux _SDL_xchg_watcom = \
60   "lock xchg [ecx], eax" \
61   parm [ecx] [eax] \
62   value [eax] \
63   modify exact [eax];
64 
65 extern _inline unsigned char _SDL_cmpxchg_watcom(volatile int *a, int newval, int oldval);
66 #pragma aux _SDL_cmpxchg_watcom = \
67   "lock cmpxchg [edx], ecx" \
68   "setz al" \
69   parm [edx] [ecx] [eax] \
70   value [al] \
71   modify exact [eax];
72 
73 extern _inline int _SDL_xadd_watcom(volatile int *a, int v);
74 #pragma aux _SDL_xadd_watcom = \
75   "lock xadd [ecx], eax" \
76   parm [ecx] [eax] \
77   value [eax] \
78   modify exact [eax];
79 #endif /* __WATCOMC__ && __386__ */
80 
81 /*
82   If any of the operations are not provided then we must emulate some
83   of them. That means we need a nice implementation of spin locks
84   that avoids the "one big lock" problem. We use a vector of spin
85   locks and pick which one to use based on the address of the operand
86   of the function.
87 
88   To generate the index of the lock we first shift by 3 bits to get
89   rid on the zero bits that result from 32 and 64 bit allignment of
90   data. We then mask off all but 5 bits and use those 5 bits as an
91   index into the table.
92 
93   Picking the lock this way insures that accesses to the same data at
94   the same time will go to the same lock. OTOH, accesses to different
95   data have only a 1/32 chance of hitting the same lock. That should
96   pretty much eliminate the chances of several atomic operations on
97   different data from waiting on the same "big lock". If it isn't
98   then the table of locks can be expanded to a new size so long as
99   the new size is a power of two.
100 
101   Contributed by Bob Pendleton, bob@pendleton.com
102 */
103 
104 #if !defined(HAVE_MSC_ATOMICS) && !defined(HAVE_GCC_ATOMICS) && !defined(__MACOSX__) && !defined(__SOLARIS__) && !defined(HAVE_WATCOM_ATOMICS)
105 #define EMULATE_CAS 1
106 #endif
107 
108 #if EMULATE_CAS
109 static SDL_SpinLock locks[32];
110 
111 static SDL_INLINE void
enterLock(void * a)112 enterLock(void *a)
113 {
114     uintptr_t index = ((((uintptr_t)a) >> 3) & 0x1f);
115 
116     SDL_AtomicLock(&locks[index]);
117 }
118 
119 static SDL_INLINE void
leaveLock(void * a)120 leaveLock(void *a)
121 {
122     uintptr_t index = ((((uintptr_t)a) >> 3) & 0x1f);
123 
124     SDL_AtomicUnlock(&locks[index]);
125 }
126 #endif
127 
128 
129 SDL_bool
SDL_AtomicCAS(SDL_atomic_t * a,int oldval,int newval)130 SDL_AtomicCAS(SDL_atomic_t *a, int oldval, int newval)
131 {
132 #ifdef HAVE_MSC_ATOMICS
133     return (_InterlockedCompareExchange((long*)&a->value, (long)newval, (long)oldval) == (long)oldval);
134 #elif defined(HAVE_WATCOM_ATOMICS)
135     return (SDL_bool) _SDL_cmpxchg_watcom(&a->value, newval, oldval);
136 #elif defined(HAVE_GCC_ATOMICS)
137     return (SDL_bool) __sync_bool_compare_and_swap(&a->value, oldval, newval);
138 #elif defined(__MACOSX__)  /* this is deprecated in 10.12 sdk; favor gcc atomics. */
139     return (SDL_bool) OSAtomicCompareAndSwap32Barrier(oldval, newval, &a->value);
140 #elif defined(__SOLARIS__) && defined(_LP64)
141     return (SDL_bool) ((int) atomic_cas_64((volatile uint64_t*)&a->value, (uint64_t)oldval, (uint64_t)newval) == oldval);
142 #elif defined(__SOLARIS__) && !defined(_LP64)
143     return (SDL_bool) ((int) atomic_cas_32((volatile uint32_t*)&a->value, (uint32_t)oldval, (uint32_t)newval) == oldval);
144 #elif EMULATE_CAS
145     SDL_bool retval = SDL_FALSE;
146 
147     enterLock(a);
148     if (a->value == oldval) {
149         a->value = newval;
150         retval = SDL_TRUE;
151     }
152     leaveLock(a);
153 
154     return retval;
155 #else
156     #error Please define your platform.
157 #endif
158 }
159 
160 SDL_bool
SDL_AtomicCASPtr(void ** a,void * oldval,void * newval)161 SDL_AtomicCASPtr(void **a, void *oldval, void *newval)
162 {
163 #if defined(HAVE_MSC_ATOMICS) && (_M_IX86)
164     return (_InterlockedCompareExchange((long*)a, (long)newval, (long)oldval) == (long)oldval);
165 #elif defined(HAVE_MSC_ATOMICS) && (!_M_IX86)
166     return (_InterlockedCompareExchangePointer(a, newval, oldval) == oldval);
167 #elif defined(HAVE_WATCOM_ATOMICS)
168     return (SDL_bool) _SDL_cmpxchg_watcom((int *)a, (long)newval, (long)oldval);
169 #elif defined(HAVE_GCC_ATOMICS)
170     return __sync_bool_compare_and_swap(a, oldval, newval);
171 #elif defined(__MACOSX__) && defined(__LP64__)  /* this is deprecated in 10.12 sdk; favor gcc atomics. */
172     return (SDL_bool) OSAtomicCompareAndSwap64Barrier((int64_t)oldval, (int64_t)newval, (int64_t*) a);
173 #elif defined(__MACOSX__) && !defined(__LP64__)  /* this is deprecated in 10.12 sdk; favor gcc atomics. */
174     return (SDL_bool) OSAtomicCompareAndSwap32Barrier((int32_t)oldval, (int32_t)newval, (int32_t*) a);
175 #elif defined(__SOLARIS__)
176     return (SDL_bool) (atomic_cas_ptr(a, oldval, newval) == oldval);
177 #elif EMULATE_CAS
178     SDL_bool retval = SDL_FALSE;
179 
180     enterLock(a);
181     if (*a == oldval) {
182         *a = newval;
183         retval = SDL_TRUE;
184     }
185     leaveLock(a);
186 
187     return retval;
188 #else
189     #error Please define your platform.
190 #endif
191 }
192 
193 int
SDL_AtomicSet(SDL_atomic_t * a,int v)194 SDL_AtomicSet(SDL_atomic_t *a, int v)
195 {
196 #ifdef HAVE_MSC_ATOMICS
197     return _InterlockedExchange((long*)&a->value, v);
198 #elif defined(HAVE_WATCOM_ATOMICS)
199     return _SDL_xchg_watcom(&a->value, v);
200 #elif defined(HAVE_GCC_ATOMICS)
201     return __sync_lock_test_and_set(&a->value, v);
202 #elif defined(__SOLARIS__) && defined(_LP64)
203     return (int) atomic_swap_64((volatile uint64_t*)&a->value, (uint64_t)v);
204 #elif defined(__SOLARIS__) && !defined(_LP64)
205     return (int) atomic_swap_32((volatile uint32_t*)&a->value, (uint32_t)v);
206 #else
207     int value;
208     do {
209         value = a->value;
210     } while (!SDL_AtomicCAS(a, value, v));
211     return value;
212 #endif
213 }
214 
215 void*
SDL_AtomicSetPtr(void ** a,void * v)216 SDL_AtomicSetPtr(void **a, void *v)
217 {
218 #if defined(HAVE_MSC_ATOMICS) && (_M_IX86)
219     return (void *) _InterlockedExchange((long *)a, (long) v);
220 #elif defined(HAVE_MSC_ATOMICS) && (!_M_IX86)
221     return _InterlockedExchangePointer(a, v);
222 #elif defined(HAVE_WATCOM_ATOMICS)
223     return (void *) _SDL_xchg_watcom((int *)a, (long)v);
224 #elif defined(HAVE_GCC_ATOMICS)
225     return __sync_lock_test_and_set(a, v);
226 #elif defined(__SOLARIS__)
227     return atomic_swap_ptr(a, v);
228 #else
229     void *value;
230     do {
231         value = *a;
232     } while (!SDL_AtomicCASPtr(a, value, v));
233     return value;
234 #endif
235 }
236 
237 int
SDL_AtomicAdd(SDL_atomic_t * a,int v)238 SDL_AtomicAdd(SDL_atomic_t *a, int v)
239 {
240 #ifdef HAVE_MSC_ATOMICS
241     return _InterlockedExchangeAdd((long*)&a->value, v);
242 #elif defined(HAVE_WATCOM_ATOMICS)
243     return _SDL_xadd_watcom(&a->value, v);
244 #elif defined(HAVE_GCC_ATOMICS)
245     return __sync_fetch_and_add(&a->value, v);
246 #elif defined(__SOLARIS__)
247     int pv = a->value;
248     membar_consumer();
249 #if defined(_LP64)
250     atomic_add_64((volatile uint64_t*)&a->value, v);
251 #elif !defined(_LP64)
252     atomic_add_32((volatile uint32_t*)&a->value, v);
253 #endif
254     return pv;
255 #else
256     int value;
257     do {
258         value = a->value;
259     } while (!SDL_AtomicCAS(a, value, (value + v)));
260     return value;
261 #endif
262 }
263 
264 int
SDL_AtomicGet(SDL_atomic_t * a)265 SDL_AtomicGet(SDL_atomic_t *a)
266 {
267 #ifdef HAVE_ATOMIC_LOAD_N
268     return __atomic_load_n(&a->value, __ATOMIC_SEQ_CST);
269 #else
270     int value;
271     do {
272         value = a->value;
273     } while (!SDL_AtomicCAS(a, value, value));
274     return value;
275 #endif
276 }
277 
278 void *
SDL_AtomicGetPtr(void ** a)279 SDL_AtomicGetPtr(void **a)
280 {
281 #ifdef HAVE_ATOMIC_LOAD_N
282     return __atomic_load_n(a, __ATOMIC_SEQ_CST);
283 #else
284     void *value;
285     do {
286         value = *a;
287     } while (!SDL_AtomicCASPtr(a, value, value));
288     return value;
289 #endif
290 }
291 
292 #ifdef SDL_MEMORY_BARRIER_USES_FUNCTION
293 #error This file should be built in arm mode so the mcr instruction is available for memory barriers
294 #endif
295 
296 void
SDL_MemoryBarrierReleaseFunction(void)297 SDL_MemoryBarrierReleaseFunction(void)
298 {
299     SDL_MemoryBarrierRelease();
300 }
301 
302 void
SDL_MemoryBarrierAcquireFunction(void)303 SDL_MemoryBarrierAcquireFunction(void)
304 {
305     SDL_MemoryBarrierAcquire();
306 }
307 
308 /* vi: set ts=4 sw=4 expandtab: */
309