sysdeps/x86_64/bits/atomic.h

   1 /* Copyright (C) 2002-2014 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 #include <libc-internal.h>
  22
  23
  24 typedef int8_t atomic8_t;
  25 typedef uint8_t uatomic8_t;
  26 typedef int_fast8_t atomic_fast8_t;
  27 typedef uint_fast8_t uatomic_fast8_t;
  28
  29 typedef int16_t atomic16_t;
  30 typedef uint16_t uatomic16_t;
  31 typedef int_fast16_t atomic_fast16_t;
  32 typedef uint_fast16_t uatomic_fast16_t;
  33
  34 typedef int32_t atomic32_t;
  35 typedef uint32_t uatomic32_t;
  36 typedef int_fast32_t atomic_fast32_t;
  37 typedef uint_fast32_t uatomic_fast32_t;
  38
  39 typedef int64_t atomic64_t;
  40 typedef uint64_t uatomic64_t;
  41 typedef int_fast64_t atomic_fast64_t;
  42 typedef uint_fast64_t uatomic_fast64_t;
  43
  44 typedef intptr_t atomicptr_t;
  45 typedef uintptr_t uatomicptr_t;
  46 typedef intmax_t atomic_max_t;
  47 typedef uintmax_t uatomic_max_t;
  48
  49
  50 #ifndef LOCK_PREFIX
  51 # ifdef UP
  52 #  define LOCK_PREFIX   /* nothing */
  53 # else
  54 #  define LOCK_PREFIX "lock;"
  55 # endif
  56 #endif
  57
  58 #define __HAVE_64B_ATOMICS 1
  59 #if __GNUC_PREREQ (4, 7)
  60 #define USE_ATOMIC_COMPILER_BUILTINS 1
  61 #else
  62 #define USE_ATOMIC_COMPILER_BUILTINS 0
  63 #endif
  64
  65 #define atomic_compare_and_exchange_val_acq(mem, newval, oldval) \
  66   __sync_val_compare_and_swap (mem, oldval, newval)
  67 #define atomic_compare_and_exchange_bool_acq(mem, newval, oldval) \
  68   (! __sync_bool_compare_and_swap (mem, oldval, newval))
  69
  70
  71 #define __arch_c_compare_and_exchange_val_8_acq(mem, newval, oldval) \
  72   ({ __typeof (*mem) ret;                                                     \
  73     __asm __volatile ("cmpl $0, %%fs:%P5\n\t"                                 \
  74                       "je 0f\n\t"                                             \
  75                       "lock\n"                                                \
  76                        "0:\tcmpxchgb %b2, %1"                                 \
  77                        : "=a" (ret), "=m" (*mem)                              \
  78                        : "q" (newval), "m" (*mem), "0" (oldval),              \
  79                          "i" (offsetof (tcbhead_t, multiple_threads)));       \
  80      ret; })
  81
  82 #define __arch_c_compare_and_exchange_val_16_acq(mem, newval, oldval) \
  83   ({ __typeof (*mem) ret;                                                     \
  84     __asm __volatile ("cmpl $0, %%fs:%P5\n\t"                                 \
  85                       "je 0f\n\t"                                             \
  86                       "lock\n"                                                \
  87                        "0:\tcmpxchgw %w2, %1"                                 \
  88                        : "=a" (ret), "=m" (*mem)                              \
  89                        : "q" (newval), "m" (*mem), "0" (oldval),              \
  90                          "i" (offsetof (tcbhead_t, multiple_threads)));       \
  91      ret; })
  92
  93 #define __arch_c_compare_and_exchange_val_32_acq(mem, newval, oldval) \
  94   ({ __typeof (*mem) ret;                                                     \
  95     __asm __volatile ("cmpl $0, %%fs:%P5\n\t"                                 \
  96                       "je 0f\n\t"                                             \
  97                       "lock\n"                                                \
  98                        "0:\tcmpxchgl %2, %1"                                  \
  99                        : "=a" (ret), "=m" (*mem)                              \
 100                        : "q" (newval), "m" (*mem), "0" (oldval),              \
 101                          "i" (offsetof (tcbhead_t, multiple_threads)));       \
 102      ret; })
 103
 104 #define __arch_c_compare_and_exchange_val_64_acq(mem, newval, oldval) \
 105   ({ __typeof (*mem) ret;                                                     \
 106      __asm __volatile ("cmpl $0, %%fs:%P5\n\t"                                \
 107                        "je 0f\n\t"                                            \
 108                        "lock\n"                                               \
 109                        "0:\tcmpxchgq %q2, %1"                                 \
 110                        : "=a" (ret), "=m" (*mem)                              \
 111                        : "q" ((atomic64_t) cast_to_integer (newval)),         \
 112                          "m" (*mem),                                          \
 113                          "0" ((atomic64_t) cast_to_integer (oldval)),         \
 114                          "i" (offsetof (tcbhead_t, multiple_threads)));       \
 115      ret; })
 116
 117
 118 /* Note that we need no lock prefix.  */
 119 #define atomic_exchange_acq(mem, newvalue) \
 120   ({ __typeof (*mem) result;                                                  \
 121      if (sizeof (*mem) == 1)                                                  \
 122        __asm __volatile ("xchgb %b0, %1"                                      \
 123                          : "=q" (result), "=m" (*mem)                         \
 124                          : "0" (newvalue), "m" (*mem));                       \
 125      else if (sizeof (*mem) == 2)                                             \
 126        __asm __volatile ("xchgw %w0, %1"                                      \
 127                          : "=r" (result), "=m" (*mem)                         \
 128                          : "0" (newvalue), "m" (*mem));                       \
 129      else if (sizeof (*mem) == 4)                                             \
 130        __asm __volatile ("xchgl %0, %1"                                       \
 131                          : "=r" (result), "=m" (*mem)                         \
 132                          : "0" (newvalue), "m" (*mem));                       \
 133      else                                                                     \
 134        __asm __volatile ("xchgq %q0, %1"                                      \
 135                          : "=r" (result), "=m" (*mem)                         \
 136                          : "0" ((atomic64_t) cast_to_integer (newvalue)),     \
 137                            "m" (*mem));                                       \
 138      result; })
 139
 140
 141 #define __arch_exchange_and_add_body(lock, mem, value)                        \
 142   ({ __typeof (*mem) result;                                                  \
 143      if (sizeof (*mem) == 1)                                                  \
 144        __asm __volatile (lock "xaddb %b0, %1"                                 \
 145                          : "=q" (result), "=m" (*mem)                         \
 146                          : "0" (value), "m" (*mem),                           \
 147                            "i" (offsetof (tcbhead_t, multiple_threads)));     \
 148      else if (sizeof (*mem) == 2)                                             \
 149        __asm __volatile (lock "xaddw %w0, %1"                                 \
 150                          : "=r" (result), "=m" (*mem)                         \
 151                          : "0" (value), "m" (*mem),                           \
 152                            "i" (offsetof (tcbhead_t, multiple_threads)));     \
 153      else if (sizeof (*mem) == 4)                                             \
 154        __asm __volatile (lock "xaddl %0, %1"                                  \
 155                          : "=r" (result), "=m" (*mem)                         \
 156                          : "0" (value), "m" (*mem),                           \
 157                            "i" (offsetof (tcbhead_t, multiple_threads)));     \
 158      else                                                                     \
 159        __asm __volatile (lock "xaddq %q0, %1"                                 \
 160                          : "=r" (result), "=m" (*mem)                         \
 161                          : "0" ((atomic64_t) cast_to_integer (value)),        \
 162                            "m" (*mem),                                        \
 163                            "i" (offsetof (tcbhead_t, multiple_threads)));     \
 164      result; })
 165
 166 #define atomic_exchange_and_add(mem, value) \
 167   __sync_fetch_and_add (mem, value)
 168
 169 #define __arch_exchange_and_add_cprefix \
 170   "cmpl $0, %%fs:%P4\n\tje 0f\n\tlock\n0:\t"
 171
 172 #define catomic_exchange_and_add(mem, value) \
 173   __arch_exchange_and_add_body (__arch_exchange_and_add_cprefix, mem, value)
 174
 175
 176 #define __arch_add_body(lock, pfx, mem, value)                                \
 177   do {                                                                        \
 178     if (__builtin_constant_p (value) && (value) == 1)                         \
 179       pfx##_increment (mem);                                                  \
 180     else if (__builtin_constant_p (value) && (value) == -1)                   \
 181       pfx##_decrement (mem);                                                  \
 182     else if (sizeof (*mem) == 1)                                              \
 183       __asm __volatile (lock "addb %b1, %0"                                   \
 184                         : "=m" (*mem)                                         \
 185                         : "iq" (value), "m" (*mem),                           \
 186                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 187     else if (sizeof (*mem) == 2)                                              \
 188       __asm __volatile (lock "addw %w1, %0"                                   \
 189                         : "=m" (*mem)                                         \
 190                         : "ir" (value), "m" (*mem),                           \
 191                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 192     else if (sizeof (*mem) == 4)                                              \
 193       __asm __volatile (lock "addl %1, %0"                                    \
 194                         : "=m" (*mem)                                         \
 195                         : "ir" (value), "m" (*mem),                           \
 196                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 197     else                                                                      \
 198       __asm __volatile (lock "addq %q1, %0"                                   \
 199                         : "=m" (*mem)                                         \
 200                         : "ir" ((atomic64_t) cast_to_integer (value)),        \
 201                           "m" (*mem),                                         \
 202                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 203   } while (0)
 204
 205 #define atomic_add(mem, value) \
 206   __arch_add_body (LOCK_PREFIX, atomic, mem, value)
 207
 208 #define __arch_add_cprefix \
 209   "cmpl $0, %%fs:%P3\n\tje 0f\n\tlock\n0:\t"
 210
 211 #define catomic_add(mem, value) \
 212   __arch_add_body (__arch_add_cprefix, catomic, mem, value)
 213
 214
 215 #define atomic_add_negative(mem, value) \
 216   ({ unsigned char __result;                                                  \
 217      if (sizeof (*mem) == 1)                                                  \
 218        __asm __volatile (LOCK_PREFIX "addb %b2, %0; sets %1"                  \
 219                          : "=m" (*mem), "=qm" (__result)                      \
 220                          : "iq" (value), "m" (*mem));                         \
 221      else if (sizeof (*mem) == 2)                                             \
 222        __asm __volatile (LOCK_PREFIX "addw %w2, %0; sets %1"                  \
 223                          : "=m" (*mem), "=qm" (__result)                      \
 224                          : "ir" (value), "m" (*mem));                         \
 225      else if (sizeof (*mem) == 4)                                             \
 226        __asm __volatile (LOCK_PREFIX "addl %2, %0; sets %1"                   \
 227                          : "=m" (*mem), "=qm" (__result)                      \
 228                          : "ir" (value), "m" (*mem));                         \
 229      else                                                                     \
 230        __asm __volatile (LOCK_PREFIX "addq %q2, %0; sets %1"                  \
 231                          : "=m" (*mem), "=qm" (__result)                      \
 232                          : "ir" ((atomic64_t) cast_to_integer (value)),       \
 233                            "m" (*mem));                                       \
 234      __result; })
 235
 236
 237 #define atomic_add_zero(mem, value) \
 238   ({ unsigned char __result;                                                  \
 239      if (sizeof (*mem) == 1)                                                  \
 240        __asm __volatile (LOCK_PREFIX "addb %b2, %0; setz %1"                  \
 241                          : "=m" (*mem), "=qm" (__result)                      \
 242                          : "iq" (value), "m" (*mem));                         \
 243      else if (sizeof (*mem) == 2)                                             \
 244        __asm __volatile (LOCK_PREFIX "addw %w2, %0; setz %1"                  \
 245                          : "=m" (*mem), "=qm" (__result)                      \
 246                          : "ir" (value), "m" (*mem));                         \
 247      else if (sizeof (*mem) == 4)                                             \
 248        __asm __volatile (LOCK_PREFIX "addl %2, %0; setz %1"                   \
 249                          : "=m" (*mem), "=qm" (__result)                      \
 250                          : "ir" (value), "m" (*mem));                         \
 251      else                                                                     \
 252        __asm __volatile (LOCK_PREFIX "addq %q2, %0; setz %1"                  \
 253                          : "=m" (*mem), "=qm" (__result)                      \
 254                          : "ir" ((atomic64_t) cast_to_integer (value)),       \
 255                            "m" (*mem));                                       \
 256      __result; })
 257
 258
 259 #define __arch_increment_body(lock, mem) \
 260   do {                                                                        \
 261     if (sizeof (*mem) == 1)                                                   \
 262       __asm __volatile (lock "incb %b0"                                       \
 263                         : "=m" (*mem)                                         \
 264                         : "m" (*mem),                                         \
 265                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 266     else if (sizeof (*mem) == 2)                                              \
 267       __asm __volatile (lock "incw %w0"                                       \
 268                         : "=m" (*mem)                                         \
 269                         : "m" (*mem),                                         \
 270                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 271     else if (sizeof (*mem) == 4)                                              \
 272       __asm __volatile (lock "incl %0"                                        \
 273                         : "=m" (*mem)                                         \
 274                         : "m" (*mem),                                         \
 275                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 276     else                                                                      \
 277       __asm __volatile (lock "incq %q0"                                       \
 278                         : "=m" (*mem)                                         \
 279                         : "m" (*mem),                                         \
 280                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 281   } while (0)
 282
 283 #define atomic_increment(mem) __arch_increment_body (LOCK_PREFIX, mem)
 284
 285 #define __arch_increment_cprefix \
 286   "cmpl $0, %%fs:%P2\n\tje 0f\n\tlock\n0:\t"
 287
 288 #define catomic_increment(mem) \
 289   __arch_increment_body (__arch_increment_cprefix, mem)
 290
 291
 292 #define atomic_increment_and_test(mem) \
 293   ({ unsigned char __result;                                                  \
 294      if (sizeof (*mem) == 1)                                                  \
 295        __asm __volatile (LOCK_PREFIX "incb %b0; sete %1"                      \
 296                          : "=m" (*mem), "=qm" (__result)                      \
 297                          : "m" (*mem));                                       \
 298      else if (sizeof (*mem) == 2)                                             \
 299        __asm __volatile (LOCK_PREFIX "incw %w0; sete %1"                      \
 300                          : "=m" (*mem), "=qm" (__result)                      \
 301                          : "m" (*mem));                                       \
 302      else if (sizeof (*mem) == 4)                                             \
 303        __asm __volatile (LOCK_PREFIX "incl %0; sete %1"                       \
 304                          : "=m" (*mem), "=qm" (__result)                      \
 305                          : "m" (*mem));                                       \
 306      else                                                                     \
 307        __asm __volatile (LOCK_PREFIX "incq %q0; sete %1"                      \
 308                          : "=m" (*mem), "=qm" (__result)                      \
 309                          : "m" (*mem));                                       \
 310      __result; })
 311
 312
 313 #define __arch_decrement_body(lock, mem) \
 314   do {                                                                        \
 315     if (sizeof (*mem) == 1)                                                   \
 316       __asm __volatile (lock "decb %b0"                                       \
 317                         : "=m" (*mem)                                         \
 318                         : "m" (*mem),                                         \
 319                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 320     else if (sizeof (*mem) == 2)                                              \
 321       __asm __volatile (lock "decw %w0"                                       \
 322                         : "=m" (*mem)                                         \
 323                         : "m" (*mem),                                         \
 324                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 325     else if (sizeof (*mem) == 4)                                              \
 326       __asm __volatile (lock "decl %0"                                        \
 327                         : "=m" (*mem)                                         \
 328                         : "m" (*mem),                                         \
 329                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 330     else                                                                      \
 331       __asm __volatile (lock "decq %q0"                                       \
 332                         : "=m" (*mem)                                         \
 333                         : "m" (*mem),                                         \
 334                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 335   } while (0)
 336
 337 #define atomic_decrement(mem) __arch_decrement_body (LOCK_PREFIX, mem)
 338
 339 #define __arch_decrement_cprefix \
 340   "cmpl $0, %%fs:%P2\n\tje 0f\n\tlock\n0:\t"
 341
 342 #define catomic_decrement(mem) \
 343   __arch_decrement_body (__arch_decrement_cprefix, mem)
 344
 345
 346 #define atomic_decrement_and_test(mem) \
 347   ({ unsigned char __result;                                                  \
 348      if (sizeof (*mem) == 1)                                                  \
 349        __asm __volatile (LOCK_PREFIX "decb %b0; sete %1"                      \
 350                          : "=m" (*mem), "=qm" (__result)                      \
 351                          : "m" (*mem));                                       \
 352      else if (sizeof (*mem) == 2)                                             \
 353        __asm __volatile (LOCK_PREFIX "decw %w0; sete %1"                      \
 354                          : "=m" (*mem), "=qm" (__result)                      \
 355                          : "m" (*mem));                                       \
 356      else if (sizeof (*mem) == 4)                                             \
 357        __asm __volatile (LOCK_PREFIX "decl %0; sete %1"                       \
 358                          : "=m" (*mem), "=qm" (__result)                      \
 359                          : "m" (*mem));                                       \
 360      else                                                                     \
 361        __asm __volatile (LOCK_PREFIX "decq %q0; sete %1"                      \
 362                          : "=m" (*mem), "=qm" (__result)                      \
 363                          : "m" (*mem));                                       \
 364      __result; })
 365
 366
 367 #define atomic_bit_set(mem, bit) \
 368   do {                                                                        \
 369     if (sizeof (*mem) == 1)                                                   \
 370       __asm __volatile (LOCK_PREFIX "orb %b2, %0"                             \
 371                         : "=m" (*mem)                                         \
 372                         : "m" (*mem), "iq" (1L << (bit)));                    \
 373     else if (sizeof (*mem) == 2)                                              \
 374       __asm __volatile (LOCK_PREFIX "orw %w2, %0"                             \
 375                         : "=m" (*mem)                                         \
 376                         : "m" (*mem), "ir" (1L << (bit)));                    \
 377     else if (sizeof (*mem) == 4)                                              \
 378       __asm __volatile (LOCK_PREFIX "orl %2, %0"                              \
 379                         : "=m" (*mem)                                         \
 380                         : "m" (*mem), "ir" (1L << (bit)));                    \
 381     else if (__builtin_constant_p (bit) && (bit) < 32)                        \
 382       __asm __volatile (LOCK_PREFIX "orq %2, %0"                              \
 383                         : "=m" (*mem)                                         \
 384                         : "m" (*mem), "i" (1L << (bit)));                     \
 385     else                                                                      \
 386       __asm __volatile (LOCK_PREFIX "orq %q2, %0"                             \
 387                         : "=m" (*mem)                                         \
 388                         : "m" (*mem), "r" (1UL << (bit)));                    \
 389   } while (0)
 390
 391
 392 #define atomic_bit_test_set(mem, bit) \
 393   ({ unsigned char __result;                                                  \
 394      if (sizeof (*mem) == 1)                                                  \
 395        __asm __volatile (LOCK_PREFIX "btsb %3, %1; setc %0"                   \
 396                          : "=q" (__result), "=m" (*mem)                       \
 397                          : "m" (*mem), "iq" (bit));                           \
 398      else if (sizeof (*mem) == 2)                                             \
 399        __asm __volatile (LOCK_PREFIX "btsw %3, %1; setc %0"                   \
 400                          : "=q" (__result), "=m" (*mem)                       \
 401                          : "m" (*mem), "ir" (bit));                           \
 402      else if (sizeof (*mem) == 4)                                             \
 403        __asm __volatile (LOCK_PREFIX "btsl %3, %1; setc %0"                   \
 404                          : "=q" (__result), "=m" (*mem)                       \
 405                          : "m" (*mem), "ir" (bit));                           \
 406      else                                                                     \
 407        __asm __volatile (LOCK_PREFIX "btsq %3, %1; setc %0"                   \
 408                          : "=q" (__result), "=m" (*mem)                       \
 409                          : "m" (*mem), "ir" (bit));                           \
 410      __result; })
 411
 412
 413 #define atomic_delay() asm ("rep; nop")
 414
 415
 416 #define __arch_and_body(lock, mem, mask) \
 417   do {                                                                        \
 418     if (sizeof (*mem) == 1)                                                   \
 419       __asm __volatile (lock "andb %b1, %0"                                   \
 420                         : "=m" (*mem)                                         \
 421                         : "iq" (mask), "m" (*mem),                            \
 422                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 423     else if (sizeof (*mem) == 2)                                              \
 424       __asm __volatile (lock "andw %w1, %0"                                   \
 425                         : "=m" (*mem)                                         \
 426                         : "ir" (mask), "m" (*mem),                            \
 427                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 428     else if (sizeof (*mem) == 4)                                              \
 429       __asm __volatile (lock "andl %1, %0"                                    \
 430                         : "=m" (*mem)                                         \
 431                         : "ir" (mask), "m" (*mem),                            \
 432                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 433     else                                                                      \
 434       __asm __volatile (lock "andq %q1, %0"                                   \
 435                         : "=m" (*mem)                                         \
 436                         : "ir" (mask), "m" (*mem),                            \
 437                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 438   } while (0)
 439
 440 #define __arch_cprefix \
 441   "cmpl $0, %%fs:%P3\n\tje 0f\n\tlock\n0:\t"
 442
 443 #define atomic_and(mem, mask) __arch_and_body (LOCK_PREFIX, mem, mask)
 444
 445 #define catomic_and(mem, mask) __arch_and_body (__arch_cprefix, mem, mask)
 446
 447
 448 #define __arch_or_body(lock, mem, mask)                                       \
 449   do {                                                                        \
 450     if (sizeof (*mem) == 1)                                                   \
 451       __asm __volatile (lock "orb %b1, %0"                                    \
 452                         : "=m" (*mem)                                         \
 453                         : "iq" (mask), "m" (*mem),                            \
 454                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 455     else if (sizeof (*mem) == 2)                                              \
 456       __asm __volatile (lock "orw %w1, %0"                                    \
 457                         : "=m" (*mem)                                         \
 458                         : "ir" (mask), "m" (*mem),                            \
 459                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 460     else if (sizeof (*mem) == 4)                                              \
 461       __asm __volatile (lock "orl %1, %0"                                     \
 462                         : "=m" (*mem)                                         \
 463                         : "ir" (mask), "m" (*mem),                            \
 464                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 465     else                                                                      \
 466       __asm __volatile (lock "orq %q1, %0"                                    \
 467                         : "=m" (*mem)                                         \
 468                         : "ir" (mask), "m" (*mem),                            \
 469                           "i" (offsetof (tcbhead_t, multiple_threads)));      \
 470   } while (0)
 471
 472 #define atomic_or(mem, mask) __arch_or_body (LOCK_PREFIX, mem, mask)
 473
 474 #define catomic_or(mem, mask) __arch_or_body (__arch_cprefix, mem, mask)