1 /* 2 * Copyright (c) 2008, 2018, 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 #ifndef AARCH64 48 template<> 49 template<typename T> 50 inline T Atomic::PlatformLoad<8>::operator()(T const volatile* src) const { 51 STATIC_ASSERT(8 == sizeof(T)); 52 return PrimitiveConversions::cast<T>( 53 (*os::atomic_load_long_func)(reinterpret_cast<const volatile int64_t*>(src))); 54 } 55 56 template<> 57 template<typename T> 58 inline void Atomic::PlatformStore<8>::operator()(T store_value, 59 T volatile* dest) const { 60 STATIC_ASSERT(8 == sizeof(T)); 61 (*os::atomic_store_long_func)( 62 PrimitiveConversions::cast<int64_t>(store_value), reinterpret_cast<volatile int64_t*>(dest)); 63 } 64 #endif 65 66 // As per atomic.hpp all read-modify-write operations have to provide two-way 67 // barriers semantics. For AARCH64 we are using load-acquire-with-reservation and 68 // store-release-with-reservation. While load-acquire combined with store-release 69 // do not generally form two-way barriers, their use with reservations does - the 70 // ARMv8 architecture manual Section F "Barrier Litmus Tests" indicates they 71 // provide sequentially consistent semantics. All we need to add is an explicit 72 // barrier in the failure path of the cmpxchg operations (as these don't execute 73 // the store) - arguably this may be overly cautious as there is a very low 74 // likelihood that the hardware would pull loads/stores into the region guarded 75 // by the reservation. 76 // 77 // For ARMv7 we add explicit barriers in the stubs. 78 79 template<size_t byte_size> 80 struct Atomic::PlatformAdd 81 : Atomic::AddAndFetch<Atomic::PlatformAdd<byte_size> > 82 { 83 template<typename I, typename D> 84 D add_and_fetch(I add_value, D volatile* dest, atomic_memory_order order) const; 85 }; 86 87 template<> 88 template<typename I, typename D> 89 inline D Atomic::PlatformAdd<4>::add_and_fetch(I add_value, D volatile* dest, 90 atomic_memory_order order) const { 91 STATIC_ASSERT(4 == sizeof(I)); 92 STATIC_ASSERT(4 == sizeof(D)); 93 #ifdef AARCH64 94 D val; 95 int tmp; 96 __asm__ volatile( 97 "1:\n\t" 98 " ldaxr %w[val], [%[dest]]\n\t" 99 " add %w[val], %w[val], %w[add_val]\n\t" 100 " stlxr %w[tmp], %w[val], [%[dest]]\n\t" 101 " cbnz %w[tmp], 1b\n\t" 102 : [val] "=&r" (val), [tmp] "=&r" (tmp) 103 : [add_val] "r" (add_value), [dest] "r" (dest) 104 : "memory"); 105 return val; 106 #else 107 return add_using_helper<int32_t>(os::atomic_add_func, add_value, dest); 108 #endif 109 } 110 111 #ifdef AARCH64 112 template<> 113 template<typename I, typename D> 114 inline D Atomic::PlatformAdd<8>::add_and_fetch(I add_value, D volatile* dest, 115 atomic_memory_order order) const { 116 STATIC_ASSERT(8 == sizeof(I)); 117 STATIC_ASSERT(8 == sizeof(D)); 118 D val; 119 int tmp; 120 __asm__ volatile( 121 "1:\n\t" 122 " ldaxr %[val], [%[dest]]\n\t" 123 " add %[val], %[val], %[add_val]\n\t" 124 " stlxr %w[tmp], %[val], [%[dest]]\n\t" 125 " cbnz %w[tmp], 1b\n\t" 126 : [val] "=&r" (val), [tmp] "=&r" (tmp) 127 : [add_val] "r" (add_value), [dest] "r" (dest) 128 : "memory"); 129 return val; 130 } 131 #endif 132 133 template<> 134 template<typename T> 135 inline T Atomic::PlatformXchg<4>::operator()(T exchange_value, 136 T volatile* dest, 137 atomic_memory_order order) const { 138 STATIC_ASSERT(4 == sizeof(T)); 139 #ifdef AARCH64 140 T old_val; 141 int tmp; 142 __asm__ volatile( 143 "1:\n\t" 144 " ldaxr %w[old_val], [%[dest]]\n\t" 145 " stlxr %w[tmp], %w[new_val], [%[dest]]\n\t" 146 " cbnz %w[tmp], 1b\n\t" 147 : [old_val] "=&r" (old_val), [tmp] "=&r" (tmp) 148 : [new_val] "r" (exchange_value), [dest] "r" (dest) 149 : "memory"); 150 return old_val; 151 #else 152 return xchg_using_helper<int32_t>(os::atomic_xchg_func, exchange_value, dest); 153 #endif 154 } 155 156 #ifdef AARCH64 157 template<> 158 template<typename T> 159 inline T Atomic::PlatformXchg<8>::operator()(T exchange_value, 160 T volatile* dest, 161 atomic_memory_order order) const { 162 STATIC_ASSERT(8 == sizeof(T)); 163 T old_val; 164 int tmp; 165 __asm__ volatile( 166 "1:\n\t" 167 " ldaxr %[old_val], [%[dest]]\n\t" 168 " stlxr %w[tmp], %[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 } 175 #endif // AARCH64 176 177 // The memory_order parameter is ignored - we always provide the strongest/most-conservative ordering 178 179 // No direct support for cmpxchg of bytes; emulate using int. 180 template<> 181 struct Atomic::PlatformCmpxchg<1> : Atomic::CmpxchgByteUsingInt {}; 182 183 #ifndef AARCH64 184 185 inline int32_t reorder_cmpxchg_func(int32_t exchange_value, 186 int32_t volatile* dest, 187 int32_t compare_value) { 188 // Warning: Arguments are swapped to avoid moving them for kernel call 189 return (*os::atomic_cmpxchg_func)(compare_value, exchange_value, dest); 190 } 191 192 inline int64_t reorder_cmpxchg_long_func(int64_t exchange_value, 193 int64_t volatile* dest, 194 int64_t compare_value) { 195 assert(VM_Version::supports_cx8(), "Atomic compare and exchange int64_t not supported on this architecture!"); 196 // Warning: Arguments are swapped to avoid moving them for kernel call 197 return (*os::atomic_cmpxchg_long_func)(compare_value, exchange_value, dest); 198 } 199 200 #endif // !AARCH64 201 202 template<> 203 template<typename T> 204 inline T Atomic::PlatformCmpxchg<4>::operator()(T exchange_value, 205 T volatile* dest, 206 T compare_value, 207 atomic_memory_order order) const { 208 STATIC_ASSERT(4 == sizeof(T)); 209 #ifdef AARCH64 210 T rv; 211 int tmp; 212 __asm__ volatile( 213 "1:\n\t" 214 " ldaxr %w[rv], [%[dest]]\n\t" 215 " cmp %w[rv], %w[cv]\n\t" 216 " b.ne 2f\n\t" 217 " stlxr %w[tmp], %w[ev], [%[dest]]\n\t" 218 " cbnz %w[tmp], 1b\n\t" 219 " b 3f\n\t" 220 "2:\n\t" 221 " dmb sy\n\t" 222 "3:\n\t" 223 : [rv] "=&r" (rv), [tmp] "=&r" (tmp) 224 : [ev] "r" (exchange_value), [dest] "r" (dest), [cv] "r" (compare_value) 225 : "memory"); 226 return rv; 227 #else 228 return cmpxchg_using_helper<int32_t>(reorder_cmpxchg_func, exchange_value, dest, compare_value); 229 #endif 230 } 231 232 template<> 233 template<typename T> 234 inline T Atomic::PlatformCmpxchg<8>::operator()(T exchange_value, 235 T volatile* dest, 236 T compare_value, 237 atomic_memory_order order) const { 238 STATIC_ASSERT(8 == sizeof(T)); 239 #ifdef AARCH64 240 T rv; 241 int tmp; 242 __asm__ volatile( 243 "1:\n\t" 244 " ldaxr %[rv], [%[dest]]\n\t" 245 " cmp %[rv], %[cv]\n\t" 246 " b.ne 2f\n\t" 247 " stlxr %w[tmp], %[ev], [%[dest]]\n\t" 248 " cbnz %w[tmp], 1b\n\t" 249 " b 3f\n\t" 250 "2:\n\t" 251 " dmb sy\n\t" 252 "3:\n\t" 253 : [rv] "=&r" (rv), [tmp] "=&r" (tmp) 254 : [ev] "r" (exchange_value), [dest] "r" (dest), [cv] "r" (compare_value) 255 : "memory"); 256 return rv; 257 #else 258 return cmpxchg_using_helper<int64_t>(reorder_cmpxchg_long_func, exchange_value, dest, compare_value); 259 #endif 260 } 261 262 #endif // OS_CPU_LINUX_ARM_VM_ATOMIC_LINUX_ARM_HPP