1 /*
   2  * Copyright (c) 2008, 2017, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #ifndef OS_CPU_LINUX_ARM_VM_ATOMIC_LINUX_ARM_HPP
  26 #define OS_CPU_LINUX_ARM_VM_ATOMIC_LINUX_ARM_HPP
  27 
  28 #include "runtime/os.hpp"
  29 #include "vm_version_arm.hpp"
  30 
  31 // Implementation of class atomic
  32 
  33 /*
  34  * Atomic long operations on 32-bit ARM
  35  * ARM v7 supports LDREXD/STREXD synchronization instructions so no problem.
  36  * ARM < v7 does not have explicit 64 atomic load/store capability.
  37  * However, gcc emits LDRD/STRD instructions on v5te and LDM/STM on v5t
  38  * when loading/storing 64 bits.
  39  * For non-MP machines (which is all we support for ARM < v7)
  40  * under current Linux distros these instructions appear atomic.
  41  * See section A3.5.3 of ARM Architecture Reference Manual for ARM v7.
  42  * Also, for cmpxchg64, if ARM < v7 we check for cmpxchg64 support in the
  43  * Linux kernel using _kuser_helper_version. See entry-armv.S in the Linux
  44  * kernel source or kernel_user_helpers.txt in Linux Doc.
  45  */
  46 
  47 inline void Atomic::store    (jbyte    store_value, jbyte*    dest) { *dest = store_value; }
  48 inline void Atomic::store    (jshort   store_value, jshort*   dest) { *dest = store_value; }
  49 inline void Atomic::store    (jint     store_value, jint*     dest) { *dest = store_value; }
  50 inline void Atomic::store_ptr(intptr_t store_value, intptr_t* dest) { *dest = store_value; }
  51 inline void Atomic::store_ptr(void*    store_value, void*     dest) { *(void**)dest = store_value; }
  52 
  53 inline void Atomic::store    (jbyte    store_value, volatile jbyte*    dest) { *dest = store_value; }
  54 inline void Atomic::store    (jshort   store_value, volatile jshort*   dest) { *dest = store_value; }
  55 inline void Atomic::store    (jint     store_value, volatile jint*     dest) { *dest = store_value; }
  56 inline void Atomic::store_ptr(intptr_t store_value, volatile intptr_t* dest) { *dest = store_value; }
  57 inline void Atomic::store_ptr(void*    store_value, volatile void*     dest) { *(void* volatile *)dest = store_value; }
  58 
  59 inline jlong Atomic::load (const volatile jlong* src) {
  60   assert(((intx)src & (sizeof(jlong)-1)) == 0, "Atomic load jlong mis-aligned");
  61 #ifdef AARCH64
  62   return *src;
  63 #else
  64   return (*os::atomic_load_long_func)(src);
  65 #endif
  66 }
  67 
  68 inline void Atomic::store (jlong value, volatile jlong* dest) {
  69   assert(((intx)dest & (sizeof(jlong)-1)) == 0, "Atomic store jlong mis-aligned");
  70 #ifdef AARCH64
  71   *dest = value;
  72 #else
  73   (*os::atomic_store_long_func)(value, dest);
  74 #endif
  75 }
  76 
  77 inline void Atomic::store (jlong value, jlong* dest) {
  78   store(value, (volatile jlong*)dest);
  79 }
  80 
  81 // As per atomic.hpp all read-modify-write operations have to provide two-way
  82 // barriers semantics. For AARCH64 we are using load-acquire-with-reservation and
  83 // store-release-with-reservation. While load-acquire combined with store-release
  84 // do not generally form two-way barriers, their use with reservations does - the
  85 // ARMv8 architecture manual Section F "Barrier Litmus Tests" indicates they
  86 // provide sequentially consistent semantics. All we need to add is an explicit
  87 // barrier in the failure path of the cmpxchg operations (as these don't execute
  88 // the store) - arguably this may be overly cautious as there is a very low
  89 // likelihood that the hardware would pull loads/stores into the region guarded
  90 // by the reservation.
  91 //
  92 // For ARMv7 we add explicit barriers in the stubs.
  93 
  94 inline jint Atomic::add(jint add_value, volatile jint* dest) {
  95 #ifdef AARCH64
  96   jint val;
  97   int tmp;
  98   __asm__ volatile(
  99     "1:\n\t"
 100     " ldaxr %w[val], [%[dest]]\n\t"
 101     " add %w[val], %w[val], %w[add_val]\n\t"
 102     " stlxr %w[tmp], %w[val], [%[dest]]\n\t"
 103     " cbnz %w[tmp], 1b\n\t"
 104     : [val] "=&r" (val), [tmp] "=&r" (tmp)
 105     : [add_val] "r" (add_value), [dest] "r" (dest)
 106     : "memory");
 107   return val;
 108 #else
 109   return (*os::atomic_add_func)(add_value, dest);
 110 #endif
 111 }
 112 
 113 inline void Atomic::inc(volatile jint* dest) {
 114   Atomic::add(1, (volatile jint *)dest);
 115 }
 116 
 117 inline void Atomic::dec(volatile jint* dest) {
 118   Atomic::add(-1, (volatile jint *)dest);
 119 }
 120 
 121 inline intptr_t Atomic::add_ptr(intptr_t add_value, volatile intptr_t* dest) {
 122 #ifdef AARCH64
 123   intptr_t val;
 124   int tmp;
 125   __asm__ volatile(
 126     "1:\n\t"
 127     " ldaxr %[val], [%[dest]]\n\t"
 128     " add %[val], %[val], %[add_val]\n\t"
 129     " stlxr %w[tmp], %[val], [%[dest]]\n\t"
 130     " cbnz %w[tmp], 1b\n\t"
 131     : [val] "=&r" (val), [tmp] "=&r" (tmp)
 132     : [add_val] "r" (add_value), [dest] "r" (dest)
 133     : "memory");
 134   return val;
 135 #else
 136   return (intptr_t)Atomic::add((jint)add_value, (volatile jint*)dest);
 137 #endif
 138 }
 139 
 140 inline void* Atomic::add_ptr(intptr_t add_value, volatile void* dest) {
 141   return (void*)add_ptr(add_value, (volatile intptr_t*)dest);
 142 }
 143 
 144 inline void Atomic::inc_ptr(volatile intptr_t* dest) {
 145   Atomic::add_ptr(1, dest);
 146 }
 147 
 148 inline void Atomic::dec_ptr(volatile intptr_t* dest) {
 149   Atomic::add_ptr(-1, dest);
 150 }
 151 
 152 inline void Atomic::inc_ptr(volatile void* dest) {
 153   inc_ptr((volatile intptr_t*)dest);
 154 }
 155 
 156 inline void Atomic::dec_ptr(volatile void* dest) {
 157   dec_ptr((volatile intptr_t*)dest);
 158 }
 159 
 160 
 161 inline jint Atomic::xchg(jint exchange_value, volatile jint* dest) {
 162 #ifdef AARCH64
 163   jint old_val;
 164   int tmp;
 165   __asm__ volatile(
 166     "1:\n\t"
 167     " ldaxr %w[old_val], [%[dest]]\n\t"
 168     " stlxr %w[tmp], %w[new_val], [%[dest]]\n\t"
 169     " cbnz %w[tmp], 1b\n\t"
 170     : [old_val] "=&r" (old_val), [tmp] "=&r" (tmp)
 171     : [new_val] "r" (exchange_value), [dest] "r" (dest)
 172     : "memory");
 173   return old_val;
 174 #else
 175   return (*os::atomic_xchg_func)(exchange_value, dest);
 176 #endif
 177 }
 178 
 179 inline intptr_t Atomic::xchg_ptr(intptr_t exchange_value, volatile intptr_t* dest) {
 180 #ifdef AARCH64
 181   intptr_t old_val;
 182   int tmp;
 183   __asm__ volatile(
 184     "1:\n\t"
 185     " ldaxr %[old_val], [%[dest]]\n\t"
 186     " stlxr %w[tmp], %[new_val], [%[dest]]\n\t"
 187     " cbnz %w[tmp], 1b\n\t"
 188     : [old_val] "=&r" (old_val), [tmp] "=&r" (tmp)
 189     : [new_val] "r" (exchange_value), [dest] "r" (dest)
 190     : "memory");
 191   return old_val;
 192 #else
 193   return (intptr_t)xchg((jint)exchange_value, (volatile jint*)dest);
 194 #endif
 195 }
 196 
 197 inline void* Atomic::xchg_ptr(void* exchange_value, volatile void* dest) {
 198   return (void*)xchg_ptr((intptr_t)exchange_value, (volatile intptr_t*)dest);
 199 }
 200 
 201 // The memory_order parameter is ignored - we always provide the strongest/most-conservative ordering
 202 
 203 template<>
 204 struct Atomic::PlatformCmpxchg<1> : Atomic::CmpxchgByteUsingInt {};
 205 
 206 #ifndef AARCH64
 207 
 208 inline jint reorder_cmpxchg_func(jint exchange_value,
 209                                  jint volatile* dest,
 210                                  jint compare_value) {
 211   // Warning:  Arguments are swapped to avoid moving them for kernel call
 212   return (*os::atomic_cmpxchg_func)(compare_value, exchange_value, dest);
 213 }
 214 
 215 inline jlong reorder_cmpxchg_long_func(jlong exchange_value,
 216                                        jlong volatile* dest,
 217                                        jlong compare_value) {
 218   assert(VM_Version::supports_cx8(), "Atomic compare and exchange jlong not supported on this architecture!");
 219   // Warning:  Arguments are swapped to avoid moving them for kernel call
 220   return (*os::atomic_cmpxchg_long_func)(compare_value, exchange_value, dest);
 221 }
 222 
 223 #endif // !AARCH64
 224 
 225 template<>
 226 template<typename T>
 227 inline T Atomic::PlatformCmpxchg<4>::operator()(T exchange_value,
 228                                                 T volatile* dest,
 229                                                 T compare_value,
 230                                                 cmpxchg_memory_order order) const {
 231   STATIC_ASSERT(4 == sizeof(T));
 232 #ifdef AARCH64
 233   T rv;
 234   int tmp;
 235   __asm__ volatile(
 236     "1:\n\t"
 237     " ldaxr %w[rv], [%[dest]]\n\t"
 238     " cmp %w[rv], %w[cv]\n\t"
 239     " b.ne 2f\n\t"
 240     " stlxr %w[tmp], %w[ev], [%[dest]]\n\t"
 241     " cbnz %w[tmp], 1b\n\t"
 242     " b 3f\n\t"
 243     "2:\n\t"
 244     " dmb sy\n\t"
 245     "3:\n\t"
 246     : [rv] "=&r" (rv), [tmp] "=&r" (tmp)
 247     : [ev] "r" (exchange_value), [dest] "r" (dest), [cv] "r" (compare_value)
 248     : "memory");
 249   return rv;
 250 #else
 251   return cmpxchg_using_stub<jint>(reorder_cmpxchg_func, exchange_value, dest, compare_value);
 252 #endif
 253 }
 254 
 255 template<>
 256 template<typename T>
 257 inline T Atomic::PlatformCmpxchg<8>::operator()(T exchange_value,
 258                                                 T volatile* dest,
 259                                                 T compare_value,
 260                                                 cmpxchg_memory_order order) const {
 261   STATIC_ASSERT(8 == sizeof(T));
 262 #ifdef AARCH64
 263   T rv;
 264   int tmp;
 265   __asm__ volatile(
 266     "1:\n\t"
 267     " ldaxr %[rv], [%[dest]]\n\t"
 268     " cmp %[rv], %[cv]\n\t"
 269     " b.ne 2f\n\t"
 270     " stlxr %w[tmp], %[ev], [%[dest]]\n\t"
 271     " cbnz %w[tmp], 1b\n\t"
 272     " b 3f\n\t"
 273     "2:\n\t"
 274     " dmb sy\n\t"
 275     "3:\n\t"
 276     : [rv] "=&r" (rv), [tmp] "=&r" (tmp)
 277     : [ev] "r" (exchange_value), [dest] "r" (dest), [cv] "r" (compare_value)
 278     : "memory");
 279   return rv;
 280 #else
 281   return cmpxchg_using_stub<jlong>(reorder_cmpxchg_long_func, exchange_value, dest, compare_value);
 282 #endif
 283 }
 284 
 285 #endif // OS_CPU_LINUX_ARM_VM_ATOMIC_LINUX_ARM_HPP