1 /* 2 * Atomic helper templates 3 * Included from tcg-runtime.c and cputlb.c. 4 * 5 * Copyright (c) 2016 Red Hat, Inc 6 * 7 * This library is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU Lesser General Public 9 * License as published by the Free Software Foundation; either 10 * version 2.1 of the License, or (at your option) any later version. 11 * 12 * This library is distributed in the hope that it will be useful, 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 15 * Lesser General Public License for more details. 16 * 17 * You should have received a copy of the GNU Lesser General Public 18 * License along with this library; if not, see <http://www.gnu.org/licenses/>. 19 */ 20 21 #include "qemu/plugin.h" 22 #include "trace/mem.h" 23 24 #if DATA_SIZE == 16 25 # define SUFFIX o 26 # define DATA_TYPE Int128 27 # define BSWAP bswap128 28 # define SHIFT 4 29 #elif DATA_SIZE == 8 30 # define SUFFIX q 31 # define DATA_TYPE aligned_uint64_t 32 # define SDATA_TYPE aligned_int64_t 33 # define BSWAP bswap64 34 # define SHIFT 3 35 #elif DATA_SIZE == 4 36 # define SUFFIX l 37 # define DATA_TYPE uint32_t 38 # define SDATA_TYPE int32_t 39 # define BSWAP bswap32 40 # define SHIFT 2 41 #elif DATA_SIZE == 2 42 # define SUFFIX w 43 # define DATA_TYPE uint16_t 44 # define SDATA_TYPE int16_t 45 # define BSWAP bswap16 46 # define SHIFT 1 47 #elif DATA_SIZE == 1 48 # define SUFFIX b 49 # define DATA_TYPE uint8_t 50 # define SDATA_TYPE int8_t 51 # define BSWAP 52 # define SHIFT 0 53 #else 54 # error unsupported data size 55 #endif 56 57 #if DATA_SIZE >= 4 58 # define ABI_TYPE DATA_TYPE 59 #else 60 # define ABI_TYPE uint32_t 61 #endif 62 63 /* Define host-endian atomic operations. Note that END is used within 64 the ATOMIC_NAME macro, and redefined below. */ 65 #if DATA_SIZE == 1 66 # define END 67 #elif defined(HOST_WORDS_BIGENDIAN) 68 # define END _be 69 #else 70 # define END _le 71 #endif 72 73 ABI_TYPE ATOMIC_NAME(cmpxchg)(CPUArchState *env, target_ulong addr, 74 ABI_TYPE cmpv, ABI_TYPE newv, 75 MemOpIdx oi, uintptr_t retaddr) 76 { 77 DATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, 78 PAGE_READ | PAGE_WRITE, retaddr); 79 DATA_TYPE ret; 80 uint16_t info = atomic_trace_rmw_pre(env, addr, oi); 81 82 #if DATA_SIZE == 16 83 ret = atomic16_cmpxchg(haddr, cmpv, newv); 84 #else 85 ret = qatomic_cmpxchg__nocheck(haddr, cmpv, newv); 86 #endif 87 ATOMIC_MMU_CLEANUP; 88 atomic_trace_rmw_post(env, addr, info); 89 return ret; 90 } 91 92 #if DATA_SIZE >= 16 93 #if HAVE_ATOMIC128 94 ABI_TYPE ATOMIC_NAME(ld)(CPUArchState *env, target_ulong addr, 95 MemOpIdx oi, uintptr_t retaddr) 96 { 97 DATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, 98 PAGE_READ, retaddr); 99 DATA_TYPE val; 100 uint16_t info = atomic_trace_ld_pre(env, addr, oi); 101 102 val = atomic16_read(haddr); 103 ATOMIC_MMU_CLEANUP; 104 atomic_trace_ld_post(env, addr, info); 105 return val; 106 } 107 108 void ATOMIC_NAME(st)(CPUArchState *env, target_ulong addr, ABI_TYPE val, 109 MemOpIdx oi, uintptr_t retaddr) 110 { 111 DATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, 112 PAGE_WRITE, retaddr); 113 uint16_t info = atomic_trace_st_pre(env, addr, oi); 114 115 atomic16_set(haddr, val); 116 ATOMIC_MMU_CLEANUP; 117 atomic_trace_st_post(env, addr, info); 118 } 119 #endif 120 #else 121 ABI_TYPE ATOMIC_NAME(xchg)(CPUArchState *env, target_ulong addr, ABI_TYPE val, 122 MemOpIdx oi, uintptr_t retaddr) 123 { 124 DATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, 125 PAGE_READ | PAGE_WRITE, retaddr); 126 DATA_TYPE ret; 127 uint16_t info = atomic_trace_rmw_pre(env, addr, oi); 128 129 ret = qatomic_xchg__nocheck(haddr, val); 130 ATOMIC_MMU_CLEANUP; 131 atomic_trace_rmw_post(env, addr, info); 132 return ret; 133 } 134 135 #define GEN_ATOMIC_HELPER(X) \ 136 ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \ 137 ABI_TYPE val, MemOpIdx oi, uintptr_t retaddr) \ 138 { \ 139 DATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, \ 140 PAGE_READ | PAGE_WRITE, retaddr); \ 141 DATA_TYPE ret; \ 142 uint16_t info = atomic_trace_rmw_pre(env, addr, oi); \ 143 ret = qatomic_##X(haddr, val); \ 144 ATOMIC_MMU_CLEANUP; \ 145 atomic_trace_rmw_post(env, addr, info); \ 146 return ret; \ 147 } 148 149 GEN_ATOMIC_HELPER(fetch_add) 150 GEN_ATOMIC_HELPER(fetch_and) 151 GEN_ATOMIC_HELPER(fetch_or) 152 GEN_ATOMIC_HELPER(fetch_xor) 153 GEN_ATOMIC_HELPER(add_fetch) 154 GEN_ATOMIC_HELPER(and_fetch) 155 GEN_ATOMIC_HELPER(or_fetch) 156 GEN_ATOMIC_HELPER(xor_fetch) 157 158 #undef GEN_ATOMIC_HELPER 159 160 /* 161 * These helpers are, as a whole, full barriers. Within the helper, 162 * the leading barrier is explicit and the trailing barrier is within 163 * cmpxchg primitive. 164 * 165 * Trace this load + RMW loop as a single RMW op. This way, regardless 166 * of CF_PARALLEL's value, we'll trace just a read and a write. 167 */ 168 #define GEN_ATOMIC_HELPER_FN(X, FN, XDATA_TYPE, RET) \ 169 ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \ 170 ABI_TYPE xval, MemOpIdx oi, uintptr_t retaddr) \ 171 { \ 172 XDATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, \ 173 PAGE_READ | PAGE_WRITE, retaddr); \ 174 XDATA_TYPE cmp, old, new, val = xval; \ 175 uint16_t info = atomic_trace_rmw_pre(env, addr, oi); \ 176 smp_mb(); \ 177 cmp = qatomic_read__nocheck(haddr); \ 178 do { \ 179 old = cmp; new = FN(old, val); \ 180 cmp = qatomic_cmpxchg__nocheck(haddr, old, new); \ 181 } while (cmp != old); \ 182 ATOMIC_MMU_CLEANUP; \ 183 atomic_trace_rmw_post(env, addr, info); \ 184 return RET; \ 185 } 186 187 GEN_ATOMIC_HELPER_FN(fetch_smin, MIN, SDATA_TYPE, old) 188 GEN_ATOMIC_HELPER_FN(fetch_umin, MIN, DATA_TYPE, old) 189 GEN_ATOMIC_HELPER_FN(fetch_smax, MAX, SDATA_TYPE, old) 190 GEN_ATOMIC_HELPER_FN(fetch_umax, MAX, DATA_TYPE, old) 191 192 GEN_ATOMIC_HELPER_FN(smin_fetch, MIN, SDATA_TYPE, new) 193 GEN_ATOMIC_HELPER_FN(umin_fetch, MIN, DATA_TYPE, new) 194 GEN_ATOMIC_HELPER_FN(smax_fetch, MAX, SDATA_TYPE, new) 195 GEN_ATOMIC_HELPER_FN(umax_fetch, MAX, DATA_TYPE, new) 196 197 #undef GEN_ATOMIC_HELPER_FN 198 #endif /* DATA SIZE >= 16 */ 199 200 #undef END 201 202 #if DATA_SIZE > 1 203 204 /* Define reverse-host-endian atomic operations. Note that END is used 205 within the ATOMIC_NAME macro. */ 206 #ifdef HOST_WORDS_BIGENDIAN 207 # define END _le 208 #else 209 # define END _be 210 #endif 211 212 ABI_TYPE ATOMIC_NAME(cmpxchg)(CPUArchState *env, target_ulong addr, 213 ABI_TYPE cmpv, ABI_TYPE newv, 214 MemOpIdx oi, uintptr_t retaddr) 215 { 216 DATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, 217 PAGE_READ | PAGE_WRITE, retaddr); 218 DATA_TYPE ret; 219 uint16_t info = atomic_trace_rmw_pre(env, addr, oi); 220 221 #if DATA_SIZE == 16 222 ret = atomic16_cmpxchg(haddr, BSWAP(cmpv), BSWAP(newv)); 223 #else 224 ret = qatomic_cmpxchg__nocheck(haddr, BSWAP(cmpv), BSWAP(newv)); 225 #endif 226 ATOMIC_MMU_CLEANUP; 227 atomic_trace_rmw_post(env, addr, info); 228 return BSWAP(ret); 229 } 230 231 #if DATA_SIZE >= 16 232 #if HAVE_ATOMIC128 233 ABI_TYPE ATOMIC_NAME(ld)(CPUArchState *env, target_ulong addr, 234 MemOpIdx oi, uintptr_t retaddr) 235 { 236 DATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, 237 PAGE_READ, retaddr); 238 DATA_TYPE val; 239 uint16_t info = atomic_trace_ld_pre(env, addr, oi); 240 241 val = atomic16_read(haddr); 242 ATOMIC_MMU_CLEANUP; 243 atomic_trace_ld_post(env, addr, info); 244 return BSWAP(val); 245 } 246 247 void ATOMIC_NAME(st)(CPUArchState *env, target_ulong addr, ABI_TYPE val, 248 MemOpIdx oi, uintptr_t retaddr) 249 { 250 DATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, 251 PAGE_WRITE, retaddr); 252 uint16_t info = atomic_trace_st_pre(env, addr, oi); 253 254 val = BSWAP(val); 255 atomic16_set(haddr, val); 256 ATOMIC_MMU_CLEANUP; 257 atomic_trace_st_post(env, addr, info); 258 } 259 #endif 260 #else 261 ABI_TYPE ATOMIC_NAME(xchg)(CPUArchState *env, target_ulong addr, ABI_TYPE val, 262 MemOpIdx oi, uintptr_t retaddr) 263 { 264 DATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, 265 PAGE_READ | PAGE_WRITE, retaddr); 266 ABI_TYPE ret; 267 uint16_t info = atomic_trace_rmw_pre(env, addr, oi); 268 269 ret = qatomic_xchg__nocheck(haddr, BSWAP(val)); 270 ATOMIC_MMU_CLEANUP; 271 atomic_trace_rmw_post(env, addr, info); 272 return BSWAP(ret); 273 } 274 275 #define GEN_ATOMIC_HELPER(X) \ 276 ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \ 277 ABI_TYPE val, MemOpIdx oi, uintptr_t retaddr) \ 278 { \ 279 DATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, \ 280 PAGE_READ | PAGE_WRITE, retaddr); \ 281 DATA_TYPE ret; \ 282 uint16_t info = atomic_trace_rmw_pre(env, addr, oi); \ 283 ret = qatomic_##X(haddr, BSWAP(val)); \ 284 ATOMIC_MMU_CLEANUP; \ 285 atomic_trace_rmw_post(env, addr, info); \ 286 return BSWAP(ret); \ 287 } 288 289 GEN_ATOMIC_HELPER(fetch_and) 290 GEN_ATOMIC_HELPER(fetch_or) 291 GEN_ATOMIC_HELPER(fetch_xor) 292 GEN_ATOMIC_HELPER(and_fetch) 293 GEN_ATOMIC_HELPER(or_fetch) 294 GEN_ATOMIC_HELPER(xor_fetch) 295 296 #undef GEN_ATOMIC_HELPER 297 298 /* These helpers are, as a whole, full barriers. Within the helper, 299 * the leading barrier is explicit and the trailing barrier is within 300 * cmpxchg primitive. 301 * 302 * Trace this load + RMW loop as a single RMW op. This way, regardless 303 * of CF_PARALLEL's value, we'll trace just a read and a write. 304 */ 305 #define GEN_ATOMIC_HELPER_FN(X, FN, XDATA_TYPE, RET) \ 306 ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \ 307 ABI_TYPE xval, MemOpIdx oi, uintptr_t retaddr) \ 308 { \ 309 XDATA_TYPE *haddr = atomic_mmu_lookup(env, addr, oi, DATA_SIZE, \ 310 PAGE_READ | PAGE_WRITE, retaddr); \ 311 XDATA_TYPE ldo, ldn, old, new, val = xval; \ 312 uint16_t info = atomic_trace_rmw_pre(env, addr, oi); \ 313 smp_mb(); \ 314 ldn = qatomic_read__nocheck(haddr); \ 315 do { \ 316 ldo = ldn; old = BSWAP(ldo); new = FN(old, val); \ 317 ldn = qatomic_cmpxchg__nocheck(haddr, ldo, BSWAP(new)); \ 318 } while (ldo != ldn); \ 319 ATOMIC_MMU_CLEANUP; \ 320 atomic_trace_rmw_post(env, addr, info); \ 321 return RET; \ 322 } 323 324 GEN_ATOMIC_HELPER_FN(fetch_smin, MIN, SDATA_TYPE, old) 325 GEN_ATOMIC_HELPER_FN(fetch_umin, MIN, DATA_TYPE, old) 326 GEN_ATOMIC_HELPER_FN(fetch_smax, MAX, SDATA_TYPE, old) 327 GEN_ATOMIC_HELPER_FN(fetch_umax, MAX, DATA_TYPE, old) 328 329 GEN_ATOMIC_HELPER_FN(smin_fetch, MIN, SDATA_TYPE, new) 330 GEN_ATOMIC_HELPER_FN(umin_fetch, MIN, DATA_TYPE, new) 331 GEN_ATOMIC_HELPER_FN(smax_fetch, MAX, SDATA_TYPE, new) 332 GEN_ATOMIC_HELPER_FN(umax_fetch, MAX, DATA_TYPE, new) 333 334 /* Note that for addition, we need to use a separate cmpxchg loop instead 335 of bswaps for the reverse-host-endian helpers. */ 336 #define ADD(X, Y) (X + Y) 337 GEN_ATOMIC_HELPER_FN(fetch_add, ADD, DATA_TYPE, old) 338 GEN_ATOMIC_HELPER_FN(add_fetch, ADD, DATA_TYPE, new) 339 #undef ADD 340 341 #undef GEN_ATOMIC_HELPER_FN 342 #endif /* DATA_SIZE >= 16 */ 343 344 #undef END 345 #endif /* DATA_SIZE > 1 */ 346 347 #undef BSWAP 348 #undef ABI_TYPE 349 #undef DATA_TYPE 350 #undef SDATA_TYPE 351 #undef SUFFIX 352 #undef DATA_SIZE 353 #undef SHIFT 354