libatomic/gcas.c

   1 /* Copyright (C) 2012-2024 Free Software Foundation, Inc.
   2    Contributed by Richard Henderson <rth@redhat.com>.
   3
   4    This file is part of the GNU Atomic Library (libatomic).
   5
   6    Libatomic is free software; you can redistribute it and/or modify it
   7    under the terms of the GNU General Public License as published by
   8    the Free Software Foundation; either version 3 of the License, or
   9    (at your option) any later version.
  10
  11    Libatomic is distributed in the hope that it will be useful, but WITHOUT ANY
  12    WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
  13    FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  14    more details.
  15
  16    Under Section 7 of GPL version 3, you are granted additional
  17    permissions described in the GCC Runtime Library Exception, version
  18    3.1, as published by the Free Software Foundation.
  19
  20    You should have received a copy of the GNU General Public License and
  21    a copy of the GCC Runtime Library Exception along with this program;
  22    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
  23    <http://www.gnu.org/licenses/>.  */
  24
  25 #define LAT_GCAS
  26 #include "libatomic_i.h"
  27
  28
  29 /* If we natively support the cas, and if we're unconcerned with extra
  30    barriers (e.g. fully in-order cpu for which barriers are a nop), then
  31    go ahead and expand the operation inline.  */
  32 #if !defined(WANT_SPECIALCASE_RELAXED) && !defined(__OPTIMIZE_SIZE__)
  33 # define EXACT_INLINE(N)                                        \
  34   if (C2(HAVE_ATOMIC_CAS_,N))                                   \
  35     return __atomic_compare_exchange_n                          \
  36       (PTR(N,mptr), PTR(N,eptr), *PTR(N,dptr), false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)
  37 #else
  38 # define EXACT_INLINE(N)
  39 #endif
  40
  41 /* ... and if all that fails, invoke the function we generated elsewhere.
  42    Worst case, this will *also* use locks.  */
  43 #define EXACT(N)                                                \
  44   do {                                                          \
  45     if (!C2(HAVE_INT,N)) break;                                 \
  46     if ((uintptr_t)mptr & (N - 1)) break;                       \
  47     EXACT_INLINE (N);                                           \
  48     return C3(local_,compare_exchange_,N)                       \
  49       (PTR(N,mptr), PTR(N,eptr), *PTR(N,dptr), smodel, fmodel); \
  50   } while (0)
  51
  52 #define LARGER(N)                                               \
  53   do {                                                          \
  54     if (!C2(HAVE_INT,N)) break;                                 \
  55     if (!C2(HAVE_ATOMIC_LDST_,N)) break;                        \
  56     if (!C2(MAYBE_HAVE_ATOMIC_CAS_,N)) break;                   \
  57     r = (uintptr_t)mptr & (N - 1);                              \
  58     a = (uintptr_t)mptr & -N;                                   \
  59     if (r + n <= N)                                             \
  60       {                                                         \
  61         pre_barrier (smodel);                                   \
  62         u.C2(i,N) = __atomic_load_n (PTR(N,a), __ATOMIC_RELAXED); \
  63         do {                                                    \
  64           if (memcmp (u.b + r, eptr, n) != 0) goto Lfail;       \
  65           v = u; memcpy (v.b + r, dptr, n);                     \
  66         } while (!(C2(HAVE_ATOMIC_CAS_,N)                       \
  67                    ? __atomic_compare_exchange_n (PTR(N,a),     \
  68                         &u.C2(i,N), v.C2(i,N), true,            \
  69                         __ATOMIC_RELAXED, __ATOMIC_RELAXED)     \
  70                    : C3(local_,compare_exchange_,N) (PTR(N,a),  \
  71                         &u.C2(i,N), v.C2(i,N),                  \
  72                         __ATOMIC_RELAXED, __ATOMIC_RELAXED)));  \
  73         goto Lsucc;                                             \
  74       }                                                         \
  75   } while (0)
  76
  77
  78
  79 bool
  80 libat_compare_exchange (size_t n, void *mptr, void *eptr, void *dptr,
  81                         int smodel, int fmodel)
  82 {
  83   union max_size_u u, v;
  84   uintptr_t r, a;
  85   bool ret;
  86
  87   switch (n)
  88     {
  89     case 0:                             return true;
  90     case 1:             EXACT(1);       goto L4;
  91     case 2:             EXACT(2);       goto L4;
  92     case 4:             EXACT(4);       goto L8;
  93     case 8:             EXACT(8);       goto L16;
  94     case 16:            EXACT(16);      break;
  95
  96     case 3: L4:         LARGER(4);      /* FALLTHRU */
  97     case 5 ... 7: L8:   LARGER(8);      /* FALLTHRU */
  98     case 9 ... 15: L16: LARGER(16);     break;
  99
 100     Lsucc:
 101       post_barrier (smodel);
 102       return true;
 103     Lfail:
 104       post_barrier (fmodel);
 105       memcpy (eptr, u.b + r, n);
 106       return false;
 107     }
 108
 109   pre_seq_barrier (smodel);
 110   libat_lock_n (mptr, n);
 111
 112   ret = memcmp (mptr, eptr, n) == 0;
 113   memcpy ((ret ? mptr : eptr), (ret ? dptr : mptr), n);
 114
 115   libat_unlock_n (mptr, n);
 116   post_seq_barrier (ret ? smodel : fmodel);
 117
 118   return ret;
 119 }
 120
 121 EXPORT_ALIAS (compare_exchange);
 122 #undef LAT_GCAS