Gitiles
Code Review Sign In
192.168.1.100 / T106_DC / bdd93d52cf43d2991561d70e238bf11e422b794b / . / ap / libc / glibc / glibc-2.23 / sysdeps / i386 / atomic-machine.h
blob: ce62b335afeaa2adab51b08fe6ebbcc3e678c2a2 [file] [log] [blame]
xf.libdd93d52023-05-12 07:10:14 -0700[diff] [blame^]1/* Copyright (C) 2002-2016 Free Software Foundation, Inc.
2 This file is part of the GNU C Library.
3 Contributed by Ulrich Drepper <drepper@redhat.com>, 2002.
4
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
9
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
14
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library; if not, see
17 <http://www.gnu.org/licenses/>. */
18
19#include <stdint.h>
20#include <tls.h> /* For tcbhead_t. */
21
22
23typedef int8_t atomic8_t;
24typedef uint8_t uatomic8_t;
25typedef int_fast8_t atomic_fast8_t;
26typedef uint_fast8_t uatomic_fast8_t;
27
28typedef int16_t atomic16_t;
29typedef uint16_t uatomic16_t;
30typedef int_fast16_t atomic_fast16_t;
31typedef uint_fast16_t uatomic_fast16_t;
32
33typedef int32_t atomic32_t;
34typedef uint32_t uatomic32_t;
35typedef int_fast32_t atomic_fast32_t;
36typedef uint_fast32_t uatomic_fast32_t;
37
38typedef int64_t atomic64_t;
39typedef uint64_t uatomic64_t;
40typedef int_fast64_t atomic_fast64_t;
41typedef uint_fast64_t uatomic_fast64_t;
42
43typedef intptr_t atomicptr_t;
44typedef uintptr_t uatomicptr_t;
45typedef intmax_t atomic_max_t;
46typedef uintmax_t uatomic_max_t;
47
48
49#ifndef LOCK_PREFIX
50# ifdef UP
51# define LOCK_PREFIX /* nothing */
52# else
53# define LOCK_PREFIX "lock;"
54# endif
55#endif
56
57#define __HAVE_64B_ATOMICS 0
58#define USE_ATOMIC_COMPILER_BUILTINS 0
59
60
61#define atomic_compare_and_exchange_val_acq(mem, newval, oldval) \
62 __sync_val_compare_and_swap (mem, oldval, newval)
63#define atomic_compare_and_exchange_bool_acq(mem, newval, oldval) \
64 (! __sync_bool_compare_and_swap (mem, oldval, newval))
65
66
67#define __arch_c_compare_and_exchange_val_8_acq(mem, newval, oldval) \
68 ({ __typeof (*mem) ret; \
69 __asm __volatile ("cmpl $0, %%gs:%P5\n\t" \
70 "je 0f\n\t" \
71 "lock\n" \
72 "0:\tcmpxchgb %b2, %1" \
73 : "=a" (ret), "=m" (*mem) \
74 : "q" (newval), "m" (*mem), "0" (oldval), \
75 "i" (offsetof (tcbhead_t, multiple_threads))); \
76 ret; })
77
78#define __arch_c_compare_and_exchange_val_16_acq(mem, newval, oldval) \
79 ({ __typeof (*mem) ret; \
80 __asm __volatile ("cmpl $0, %%gs:%P5\n\t" \
81 "je 0f\n\t" \
82 "lock\n" \
83 "0:\tcmpxchgw %w2, %1" \
84 : "=a" (ret), "=m" (*mem) \
85 : "r" (newval), "m" (*mem), "0" (oldval), \
86 "i" (offsetof (tcbhead_t, multiple_threads))); \
87 ret; })
88
89#define __arch_c_compare_and_exchange_val_32_acq(mem, newval, oldval) \
90 ({ __typeof (*mem) ret; \
91 __asm __volatile ("cmpl $0, %%gs:%P5\n\t" \
92 "je 0f\n\t" \
93 "lock\n" \
94 "0:\tcmpxchgl %2, %1" \
95 : "=a" (ret), "=m" (*mem) \
96 : "r" (newval), "m" (*mem), "0" (oldval), \
97 "i" (offsetof (tcbhead_t, multiple_threads))); \
98 ret; })
99
100/* XXX We do not really need 64-bit compare-and-exchange. At least
101 not in the moment. Using it would mean causing portability
102 problems since not many other 32-bit architectures have support for
103 such an operation. So don't define any code for now. If it is
104 really going to be used the code below can be used on Intel Pentium
105 and later, but NOT on i486. */
106#if 1
107# define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
108 ({ __typeof (*mem) ret = *(mem); \
109 abort (); \
110 ret = (newval); \
111 ret = (oldval); \
112 ret; })
113# define __arch_c_compare_and_exchange_val_64_acq(mem, newval, oldval) \
114 ({ __typeof (*mem) ret = *(mem); \
115 abort (); \
116 ret = (newval); \
117 ret = (oldval); \
118 ret; })
119#else
120# ifdef __PIC__
121# define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
122 ({ __typeof (*mem) ret; \
123 __asm __volatile ("xchgl %2, %%ebx\n\t" \
124 LOCK_PREFIX "cmpxchg8b %1\n\t" \
125 "xchgl %2, %%ebx" \
126 : "=A" (ret), "=m" (*mem) \
127 : "DS" (((unsigned long long int) (newval)) \
128 & 0xffffffff), \
129 "c" (((unsigned long long int) (newval)) >> 32), \
130 "m" (*mem), "a" (((unsigned long long int) (oldval)) \
131 & 0xffffffff), \
132 "d" (((unsigned long long int) (oldval)) >> 32)); \
133 ret; })
134
135# define __arch_c_compare_and_exchange_val_64_acq(mem, newval, oldval) \
136 ({ __typeof (*mem) ret; \
137 __asm __volatile ("xchgl %2, %%ebx\n\t" \
138 "cmpl $0, %%gs:%P7\n\t" \
139 "je 0f\n\t" \
140 "lock\n" \
141 "0:\tcmpxchg8b %1\n\t" \
142 "xchgl %2, %%ebx" \
143 : "=A" (ret), "=m" (*mem) \
144 : "DS" (((unsigned long long int) (newval)) \
145 & 0xffffffff), \
146 "c" (((unsigned long long int) (newval)) >> 32), \
147 "m" (*mem), "a" (((unsigned long long int) (oldval)) \
148 & 0xffffffff), \
149 "d" (((unsigned long long int) (oldval)) >> 32), \
150 "i" (offsetof (tcbhead_t, multiple_threads))); \
151 ret; })
152# else
153# define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \
154 ({ __typeof (*mem) ret; \
155 __asm __volatile (LOCK_PREFIX "cmpxchg8b %1" \
156 : "=A" (ret), "=m" (*mem) \
157 : "b" (((unsigned long long int) (newval)) \
158 & 0xffffffff), \
159 "c" (((unsigned long long int) (newval)) >> 32), \
160 "m" (*mem), "a" (((unsigned long long int) (oldval)) \
161 & 0xffffffff), \
162 "d" (((unsigned long long int) (oldval)) >> 32)); \
163 ret; })
164
165# define __arch_c_compare_and_exchange_val_64_acq(mem, newval, oldval) \
166 ({ __typeof (*mem) ret; \
167 __asm __volatile ("cmpl $0, %%gs:%P7\n\t" \
168 "je 0f\n\t" \
169 "lock\n" \
170 "0:\tcmpxchg8b %1" \
171 : "=A" (ret), "=m" (*mem) \
172 : "b" (((unsigned long long int) (newval)) \
173 & 0xffffffff), \
174 "c" (((unsigned long long int) (newval)) >> 32), \
175 "m" (*mem), "a" (((unsigned long long int) (oldval)) \
176 & 0xffffffff), \
177 "d" (((unsigned long long int) (oldval)) >> 32), \
178 "i" (offsetof (tcbhead_t, multiple_threads))); \
179 ret; })
180# endif
181#endif
182
183
184/* Note that we need no lock prefix. */
185#define atomic_exchange_acq(mem, newvalue) \
186 ({ __typeof (*mem) result; \
187 if (sizeof (*mem) == 1) \
188 __asm __volatile ("xchgb %b0, %1" \
189 : "=q" (result), "=m" (*mem) \
190 : "0" (newvalue), "m" (*mem)); \
191 else if (sizeof (*mem) == 2) \
192 __asm __volatile ("xchgw %w0, %1" \
193 : "=r" (result), "=m" (*mem) \
194 : "0" (newvalue), "m" (*mem)); \
195 else if (sizeof (*mem) == 4) \
196 __asm __volatile ("xchgl %0, %1" \
197 : "=r" (result), "=m" (*mem) \
198 : "0" (newvalue), "m" (*mem)); \
199 else \
200 { \
201 result = 0; \
202 abort (); \
203 } \
204 result; })
205
206
207#define __arch_exchange_and_add_body(lock, pfx, mem, value) \
208 ({ __typeof (*mem) __result; \
209 __typeof (value) __addval = (value); \
210 if (sizeof (*mem) == 1) \
211 __asm __volatile (lock "xaddb %b0, %1" \
212 : "=q" (__result), "=m" (*mem) \
213 : "0" (__addval), "m" (*mem), \
214 "i" (offsetof (tcbhead_t, multiple_threads))); \
215 else if (sizeof (*mem) == 2) \
216 __asm __volatile (lock "xaddw %w0, %1" \
217 : "=r" (__result), "=m" (*mem) \
218 : "0" (__addval), "m" (*mem), \
219 "i" (offsetof (tcbhead_t, multiple_threads))); \
220 else if (sizeof (*mem) == 4) \
221 __asm __volatile (lock "xaddl %0, %1" \
222 : "=r" (__result), "=m" (*mem) \
223 : "0" (__addval), "m" (*mem), \
224 "i" (offsetof (tcbhead_t, multiple_threads))); \
225 else \
226 { \
227 __typeof (mem) __memp = (mem); \
228 __typeof (*mem) __tmpval; \
229 __result = *__memp; \
230 do \
231 __tmpval = __result; \
232 while ((__result = pfx##_compare_and_exchange_val_64_acq \
233 (__memp, __result + __addval, __result)) == __tmpval); \
234 } \
235 __result; })
236
237#define atomic_exchange_and_add(mem, value) \
238 __sync_fetch_and_add (mem, value)
239
240#define __arch_exchange_and_add_cprefix \
241 "cmpl $0, %%gs:%P4\n\tje 0f\n\tlock\n0:\t"
242
243#define catomic_exchange_and_add(mem, value) \
244 __arch_exchange_and_add_body (__arch_exchange_and_add_cprefix, __arch_c, \
245 mem, value)
246
247
248#define __arch_add_body(lock, pfx, mem, value) \
249 do { \
250 if (__builtin_constant_p (value) && (value) == 1) \
251 atomic_increment (mem); \
252 else if (__builtin_constant_p (value) && (value) == -1) \
253 atomic_decrement (mem); \
254 else if (sizeof (*mem) == 1) \
255 __asm __volatile (lock "addb %b1, %0" \
256 : "=m" (*mem) \
257 : "iq" (value), "m" (*mem), \
258 "i" (offsetof (tcbhead_t, multiple_threads))); \
259 else if (sizeof (*mem) == 2) \
260 __asm __volatile (lock "addw %w1, %0" \
261 : "=m" (*mem) \
262 : "ir" (value), "m" (*mem), \
263 "i" (offsetof (tcbhead_t, multiple_threads))); \
264 else if (sizeof (*mem) == 4) \
265 __asm __volatile (lock "addl %1, %0" \
266 : "=m" (*mem) \
267 : "ir" (value), "m" (*mem), \
268 "i" (offsetof (tcbhead_t, multiple_threads))); \
269 else \
270 { \
271 __typeof (value) __addval = (value); \
272 __typeof (mem) __memp = (mem); \
273 __typeof (*mem) __oldval = *__memp; \
274 __typeof (*mem) __tmpval; \
275 do \
276 __tmpval = __oldval; \
277 while ((__oldval = pfx##_compare_and_exchange_val_64_acq \
278 (__memp, __oldval + __addval, __oldval)) == __tmpval); \
279 } \
280 } while (0)
281
282#define atomic_add(mem, value) \
283 __arch_add_body (LOCK_PREFIX, __arch, mem, value)
284
285#define __arch_add_cprefix \
286 "cmpl $0, %%gs:%P3\n\tje 0f\n\tlock\n0:\t"
287
288#define catomic_add(mem, value) \
289 __arch_add_body (__arch_add_cprefix, __arch_c, mem, value)
290
291
292#define atomic_add_negative(mem, value) \
293 ({ unsigned char __result; \
294 if (sizeof (*mem) == 1) \
295 __asm __volatile (LOCK_PREFIX "addb %b2, %0; sets %1" \
296 : "=m" (*mem), "=qm" (__result) \
297 : "iq" (value), "m" (*mem)); \
298 else if (sizeof (*mem) == 2) \
299 __asm __volatile (LOCK_PREFIX "addw %w2, %0; sets %1" \
300 : "=m" (*mem), "=qm" (__result) \
301 : "ir" (value), "m" (*mem)); \
302 else if (sizeof (*mem) == 4) \
303 __asm __volatile (LOCK_PREFIX "addl %2, %0; sets %1" \
304 : "=m" (*mem), "=qm" (__result) \
305 : "ir" (value), "m" (*mem)); \
306 else \
307 abort (); \
308 __result; })
309
310
311#define atomic_add_zero(mem, value) \
312 ({ unsigned char __result; \
313 if (sizeof (*mem) == 1) \
314 __asm __volatile (LOCK_PREFIX "addb %b2, %0; setz %1" \
315 : "=m" (*mem), "=qm" (__result) \
316 : "iq" (value), "m" (*mem)); \
317 else if (sizeof (*mem) == 2) \
318 __asm __volatile (LOCK_PREFIX "addw %w2, %0; setz %1" \
319 : "=m" (*mem), "=qm" (__result) \
320 : "ir" (value), "m" (*mem)); \
321 else if (sizeof (*mem) == 4) \
322 __asm __volatile (LOCK_PREFIX "addl %2, %0; setz %1" \
323 : "=m" (*mem), "=qm" (__result) \
324 : "ir" (value), "m" (*mem)); \
325 else \
326 abort (); \
327 __result; })
328
329
330#define __arch_increment_body(lock, pfx, mem) \
331 do { \
332 if (sizeof (*mem) == 1) \
333 __asm __volatile (lock "incb %b0" \
334 : "=m" (*mem) \
335 : "m" (*mem), \
336 "i" (offsetof (tcbhead_t, multiple_threads))); \
337 else if (sizeof (*mem) == 2) \
338 __asm __volatile (lock "incw %w0" \
339 : "=m" (*mem) \
340 : "m" (*mem), \
341 "i" (offsetof (tcbhead_t, multiple_threads))); \
342 else if (sizeof (*mem) == 4) \
343 __asm __volatile (lock "incl %0" \
344 : "=m" (*mem) \
345 : "m" (*mem), \
346 "i" (offsetof (tcbhead_t, multiple_threads))); \
347 else \
348 { \
349 __typeof (mem) __memp = (mem); \
350 __typeof (*mem) __oldval = *__memp; \
351 __typeof (*mem) __tmpval; \
352 do \
353 __tmpval = __oldval; \
354 while ((__oldval = pfx##_compare_and_exchange_val_64_acq \
355 (__memp, __oldval + 1, __oldval)) == __tmpval); \
356 } \
357 } while (0)
358
359#define atomic_increment(mem) __arch_increment_body (LOCK_PREFIX, __arch, mem)
360
361#define __arch_increment_cprefix \
362 "cmpl $0, %%gs:%P2\n\tje 0f\n\tlock\n0:\t"
363
364#define catomic_increment(mem) \
365 __arch_increment_body (__arch_increment_cprefix, __arch_c, mem)
366
367
368#define atomic_increment_and_test(mem) \
369 ({ unsigned char __result; \
370 if (sizeof (*mem) == 1) \
371 __asm __volatile (LOCK_PREFIX "incb %0; sete %b1" \
372 : "=m" (*mem), "=qm" (__result) \
373 : "m" (*mem)); \
374 else if (sizeof (*mem) == 2) \
375 __asm __volatile (LOCK_PREFIX "incw %0; sete %w1" \
376 : "=m" (*mem), "=qm" (__result) \
377 : "m" (*mem)); \
378 else if (sizeof (*mem) == 4) \
379 __asm __volatile (LOCK_PREFIX "incl %0; sete %1" \
380 : "=m" (*mem), "=qm" (__result) \
381 : "m" (*mem)); \
382 else \
383 abort (); \
384 __result; })
385
386
387#define __arch_decrement_body(lock, pfx, mem) \
388 do { \
389 if (sizeof (*mem) == 1) \
390 __asm __volatile (lock "decb %b0" \
391 : "=m" (*mem) \
392 : "m" (*mem), \
393 "i" (offsetof (tcbhead_t, multiple_threads))); \
394 else if (sizeof (*mem) == 2) \
395 __asm __volatile (lock "decw %w0" \
396 : "=m" (*mem) \
397 : "m" (*mem), \
398 "i" (offsetof (tcbhead_t, multiple_threads))); \
399 else if (sizeof (*mem) == 4) \
400 __asm __volatile (lock "decl %0" \
401 : "=m" (*mem) \
402 : "m" (*mem), \
403 "i" (offsetof (tcbhead_t, multiple_threads))); \
404 else \
405 { \
406 __typeof (mem) __memp = (mem); \
407 __typeof (*mem) __oldval = *__memp; \
408 __typeof (*mem) __tmpval; \
409 do \
410 __tmpval = __oldval; \
411 while ((__oldval = pfx##_compare_and_exchange_val_64_acq \
412 (__memp, __oldval - 1, __oldval)) == __tmpval); \
413 } \
414 } while (0)
415
416#define atomic_decrement(mem) __arch_decrement_body (LOCK_PREFIX, __arch, mem)
417
418#define __arch_decrement_cprefix \
419 "cmpl $0, %%gs:%P2\n\tje 0f\n\tlock\n0:\t"
420
421#define catomic_decrement(mem) \
422 __arch_decrement_body (__arch_decrement_cprefix, __arch_c, mem)
423
424
425#define atomic_decrement_and_test(mem) \
426 ({ unsigned char __result; \
427 if (sizeof (*mem) == 1) \
428 __asm __volatile (LOCK_PREFIX "decb %b0; sete %1" \
429 : "=m" (*mem), "=qm" (__result) \
430 : "m" (*mem)); \
431 else if (sizeof (*mem) == 2) \
432 __asm __volatile (LOCK_PREFIX "decw %w0; sete %1" \
433 : "=m" (*mem), "=qm" (__result) \
434 : "m" (*mem)); \
435 else if (sizeof (*mem) == 4) \
436 __asm __volatile (LOCK_PREFIX "decl %0; sete %1" \
437 : "=m" (*mem), "=qm" (__result) \
438 : "m" (*mem)); \
439 else \
440 abort (); \
441 __result; })
442
443
444#define atomic_bit_set(mem, bit) \
445 do { \
446 if (sizeof (*mem) == 1) \
447 __asm __volatile (LOCK_PREFIX "orb %b2, %0" \
448 : "=m" (*mem) \
449 : "m" (*mem), "iq" (1 << (bit))); \
450 else if (sizeof (*mem) == 2) \
451 __asm __volatile (LOCK_PREFIX "orw %w2, %0" \
452 : "=m" (*mem) \
453 : "m" (*mem), "ir" (1 << (bit))); \
454 else if (sizeof (*mem) == 4) \
455 __asm __volatile (LOCK_PREFIX "orl %2, %0" \
456 : "=m" (*mem) \
457 : "m" (*mem), "ir" (1 << (bit))); \
458 else \
459 abort (); \
460 } while (0)
461
462
463#define atomic_bit_test_set(mem, bit) \
464 ({ unsigned char __result; \
465 if (sizeof (*mem) == 1) \
466 __asm __volatile (LOCK_PREFIX "btsb %3, %1; setc %0" \
467 : "=q" (__result), "=m" (*mem) \
468 : "m" (*mem), "ir" (bit)); \
469 else if (sizeof (*mem) == 2) \
470 __asm __volatile (LOCK_PREFIX "btsw %3, %1; setc %0" \
471 : "=q" (__result), "=m" (*mem) \
472 : "m" (*mem), "ir" (bit)); \
473 else if (sizeof (*mem) == 4) \
474 __asm __volatile (LOCK_PREFIX "btsl %3, %1; setc %0" \
475 : "=q" (__result), "=m" (*mem) \
476 : "m" (*mem), "ir" (bit)); \
477 else \
478 abort (); \
479 __result; })
480
481
482#define atomic_spin_nop() asm ("rep; nop")
483
484
485#define __arch_and_body(lock, mem, mask) \
486 do { \
487 if (sizeof (*mem) == 1) \
488 __asm __volatile (lock "andb %b1, %0" \
489 : "=m" (*mem) \
490 : "iq" (mask), "m" (*mem), \
491 "i" (offsetof (tcbhead_t, multiple_threads))); \
492 else if (sizeof (*mem) == 2) \
493 __asm __volatile (lock "andw %w1, %0" \
494 : "=m" (*mem) \
495 : "ir" (mask), "m" (*mem), \
496 "i" (offsetof (tcbhead_t, multiple_threads))); \
497 else if (sizeof (*mem) == 4) \
498 __asm __volatile (lock "andl %1, %0" \
499 : "=m" (*mem) \
500 : "ir" (mask), "m" (*mem), \
501 "i" (offsetof (tcbhead_t, multiple_threads))); \
502 else \
503 abort (); \
504 } while (0)
505
506#define __arch_cprefix \
507 "cmpl $0, %%gs:%P3\n\tje 0f\n\tlock\n0:\t"
508
509#define atomic_and(mem, mask) __arch_and_body (LOCK_PREFIX, mem, mask)
510
511#define catomic_and(mem, mask) __arch_and_body (__arch_cprefix, mem, mask)
512
513
514#define __arch_or_body(lock, mem, mask) \
515 do { \
516 if (sizeof (*mem) == 1) \
517 __asm __volatile (lock "orb %b1, %0" \
518 : "=m" (*mem) \
519 : "iq" (mask), "m" (*mem), \
520 "i" (offsetof (tcbhead_t, multiple_threads))); \
521 else if (sizeof (*mem) == 2) \
522 __asm __volatile (lock "orw %w1, %0" \
523 : "=m" (*mem) \
524 : "ir" (mask), "m" (*mem), \
525 "i" (offsetof (tcbhead_t, multiple_threads))); \
526 else if (sizeof (*mem) == 4) \
527 __asm __volatile (lock "orl %1, %0" \
528 : "=m" (*mem) \
529 : "ir" (mask), "m" (*mem), \
530 "i" (offsetof (tcbhead_t, multiple_threads))); \
531 else \
532 abort (); \
533 } while (0)
534
535#define atomic_or(mem, mask) __arch_or_body (LOCK_PREFIX, mem, mask)
536
537#define catomic_or(mem, mask) __arch_or_body (__arch_cprefix, mem, mask)
538
539/* We don't use mfence because it is supposedly slower due to having to
540 provide stronger guarantees (e.g., regarding self-modifying code). */
541#define atomic_full_barrier() \
542 __asm __volatile (LOCK_PREFIX "orl $0, (%%esp)" ::: "memory")
543#define atomic_read_barrier() __asm ("" ::: "memory")
544#define atomic_write_barrier() __asm ("" ::: "memory")