| 1 | #define RCU_GP_CTR_BIT (1 << 7) |
| 2 | #define RCU_GP_CTR_NEST_MASK (RCU_GP_CTR_BIT - 1) |
| 3 | |
| 4 | #define read_free_race (read_generation == last_free_gen) |
| 5 | #define read_free (free_done && data_access) |
| 6 | |
| 7 | #ifndef READER_NEST_LEVEL |
| 8 | #define READER_NEST_LEVEL 2 |
| 9 | #endif |
| 10 | |
| 11 | #define REMOTE_BARRIERS |
| 12 | /* |
| 13 | * mem.spin: Promela code to validate memory barriers with OOO memory. |
| 14 | * |
| 15 | * This program is free software; you can redistribute it and/or modify |
| 16 | * it under the terms of the GNU General Public License as published by |
| 17 | * the Free Software Foundation; either version 2 of the License, or |
| 18 | * (at your option) any later version. |
| 19 | * |
| 20 | * This program is distributed in the hope that it will be useful, |
| 21 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 22 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 23 | * GNU General Public License for more details. |
| 24 | * |
| 25 | * You should have received a copy of the GNU General Public License |
| 26 | * along with this program; if not, write to the Free Software |
| 27 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. |
| 28 | * |
| 29 | * Copyright (c) 2009 Mathieu Desnoyers |
| 30 | */ |
| 31 | |
| 32 | /* Promela validation variables. */ |
| 33 | |
| 34 | #define NR_READERS 1 |
| 35 | #define NR_WRITERS 1 |
| 36 | |
| 37 | #define NR_PROCS 2 |
| 38 | |
| 39 | #define get_pid() (_pid) |
| 40 | |
| 41 | /* |
| 42 | * Each process have its own data in cache. Caches are randomly updated. |
| 43 | * smp_wmb and smp_rmb forces cache updates (write and read), wmb_mb forces |
| 44 | * both. |
| 45 | */ |
| 46 | |
| 47 | #define DECLARE_CACHED_VAR(type, x, v) \ |
| 48 | type mem_##x = v; \ |
| 49 | type cached_##x[NR_PROCS] = v; \ |
| 50 | bit cache_dirty_##x[NR_PROCS] = 0 |
| 51 | |
| 52 | #define IS_CACHE_DIRTY(x, id) (cache_dirty_##x[id]) |
| 53 | |
| 54 | #define READ_CACHED_VAR(x) (cached_##x[get_pid()]) |
| 55 | |
| 56 | #define WRITE_CACHED_VAR(x, v) \ |
| 57 | atomic { \ |
| 58 | cached_##x[get_pid()] = v; \ |
| 59 | cache_dirty_##x[get_pid()] = 1; \ |
| 60 | } |
| 61 | |
| 62 | #define CACHE_WRITE_TO_MEM(x, id) \ |
| 63 | if \ |
| 64 | :: IS_CACHE_DIRTY(x, id) -> \ |
| 65 | mem_##x = cached_##x[id]; \ |
| 66 | cache_dirty_##x[id] = 0; \ |
| 67 | :: else -> \ |
| 68 | skip \ |
| 69 | fi; |
| 70 | |
| 71 | #define CACHE_READ_FROM_MEM(x, id) \ |
| 72 | if \ |
| 73 | :: !IS_CACHE_DIRTY(x, id) -> \ |
| 74 | cached_##x[id] = mem_##x;\ |
| 75 | :: else -> \ |
| 76 | skip \ |
| 77 | fi; |
| 78 | |
| 79 | /* |
| 80 | * May update other caches if cache is dirty, or not. |
| 81 | */ |
| 82 | #define RANDOM_CACHE_WRITE_TO_MEM(x, id)\ |
| 83 | if \ |
| 84 | :: 1 -> CACHE_WRITE_TO_MEM(x, id); \ |
| 85 | :: 1 -> skip \ |
| 86 | fi; |
| 87 | |
| 88 | #define RANDOM_CACHE_READ_FROM_MEM(x, id)\ |
| 89 | if \ |
| 90 | :: 1 -> CACHE_READ_FROM_MEM(x, id); \ |
| 91 | :: 1 -> skip \ |
| 92 | fi; |
| 93 | |
| 94 | /* |
| 95 | * Remote barriers tests the scheme where a signal (or IPI) is sent to all |
| 96 | * reader threads to promote their compiler barrier to a smp_mb(). |
| 97 | */ |
| 98 | #ifdef REMOTE_BARRIERS |
| 99 | |
| 100 | inline smp_rmb_pid(i) |
| 101 | { |
| 102 | atomic { |
| 103 | CACHE_READ_FROM_MEM(urcu_gp_ctr, i); |
| 104 | CACHE_READ_FROM_MEM(urcu_active_readers_one, i); |
| 105 | CACHE_READ_FROM_MEM(generation_ptr, i); |
| 106 | } |
| 107 | } |
| 108 | |
| 109 | inline smp_wmb_pid(i) |
| 110 | { |
| 111 | atomic { |
| 112 | CACHE_WRITE_TO_MEM(urcu_gp_ctr, i); |
| 113 | CACHE_WRITE_TO_MEM(urcu_active_readers_one, i); |
| 114 | CACHE_WRITE_TO_MEM(generation_ptr, i); |
| 115 | } |
| 116 | } |
| 117 | |
| 118 | inline smp_mb_pid(i) |
| 119 | { |
| 120 | atomic { |
| 121 | #ifndef NO_WMB |
| 122 | smp_wmb_pid(i); |
| 123 | #endif |
| 124 | #ifndef NO_RMB |
| 125 | smp_rmb_pid(i); |
| 126 | #endif |
| 127 | #ifdef NO_WMB |
| 128 | #ifdef NO_RMB |
| 129 | ooo_mem(i); |
| 130 | #endif |
| 131 | #endif |
| 132 | } |
| 133 | } |
| 134 | |
| 135 | /* |
| 136 | * Readers do a simple barrier(), writers are doing a smp_mb() _and_ sending a |
| 137 | * signal or IPI to have all readers execute a smp_mb. |
| 138 | * We are not modeling the whole rendez-vous between readers and writers here, |
| 139 | * we just let the writer update each reader's caches remotely. |
| 140 | */ |
| 141 | inline smp_mb(i) |
| 142 | { |
| 143 | if |
| 144 | :: get_pid() >= NR_READERS -> |
| 145 | smp_mb_pid(get_pid()); |
| 146 | i = 0; |
| 147 | do |
| 148 | :: i < NR_READERS -> |
| 149 | smp_mb_pid(i); |
| 150 | i++; |
| 151 | :: i >= NR_READERS -> break |
| 152 | od; |
| 153 | smp_mb_pid(get_pid()); |
| 154 | :: else -> skip; |
| 155 | fi; |
| 156 | } |
| 157 | |
| 158 | #else |
| 159 | |
| 160 | inline smp_rmb(i) |
| 161 | { |
| 162 | atomic { |
| 163 | CACHE_READ_FROM_MEM(urcu_gp_ctr, get_pid()); |
| 164 | CACHE_READ_FROM_MEM(urcu_active_readers_one, get_pid()); |
| 165 | CACHE_READ_FROM_MEM(generation_ptr, get_pid()); |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | inline smp_wmb(i) |
| 170 | { |
| 171 | atomic { |
| 172 | CACHE_WRITE_TO_MEM(urcu_gp_ctr, get_pid()); |
| 173 | CACHE_WRITE_TO_MEM(urcu_active_readers_one, get_pid()); |
| 174 | CACHE_WRITE_TO_MEM(generation_ptr, get_pid()); |
| 175 | } |
| 176 | } |
| 177 | |
| 178 | inline smp_mb(i) |
| 179 | { |
| 180 | atomic { |
| 181 | #ifndef NO_WMB |
| 182 | smp_wmb(i); |
| 183 | #endif |
| 184 | #ifndef NO_RMB |
| 185 | smp_rmb(i); |
| 186 | #endif |
| 187 | #ifdef NO_WMB |
| 188 | #ifdef NO_RMB |
| 189 | ooo_mem(i); |
| 190 | #endif |
| 191 | #endif |
| 192 | } |
| 193 | } |
| 194 | |
| 195 | #endif |
| 196 | |
| 197 | /* Keep in sync manually with smp_rmb, wmp_wmb and ooo_mem */ |
| 198 | DECLARE_CACHED_VAR(byte, urcu_gp_ctr, 1); |
| 199 | /* Note ! currently only one reader */ |
| 200 | DECLARE_CACHED_VAR(byte, urcu_active_readers_one, 0); |
| 201 | /* pointer generation */ |
| 202 | DECLARE_CACHED_VAR(byte, generation_ptr, 0); |
| 203 | |
| 204 | byte last_free_gen = 0; |
| 205 | bit free_done = 0; |
| 206 | byte read_generation = 1; |
| 207 | bit data_access = 0; |
| 208 | |
| 209 | bit write_lock = 0; |
| 210 | |
| 211 | inline ooo_mem(i) |
| 212 | { |
| 213 | atomic { |
| 214 | RANDOM_CACHE_WRITE_TO_MEM(urcu_gp_ctr, get_pid()); |
| 215 | RANDOM_CACHE_WRITE_TO_MEM(urcu_active_readers_one, |
| 216 | get_pid()); |
| 217 | RANDOM_CACHE_WRITE_TO_MEM(generation_ptr, get_pid()); |
| 218 | RANDOM_CACHE_READ_FROM_MEM(urcu_gp_ctr, get_pid()); |
| 219 | RANDOM_CACHE_READ_FROM_MEM(urcu_active_readers_one, |
| 220 | get_pid()); |
| 221 | RANDOM_CACHE_READ_FROM_MEM(generation_ptr, get_pid()); |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | #define get_readerid() (get_pid()) |
| 226 | #define get_writerid() (get_readerid() + NR_READERS) |
| 227 | |
| 228 | inline wait_for_reader(tmp, id, i) |
| 229 | { |
| 230 | do |
| 231 | :: 1 -> |
| 232 | tmp = READ_CACHED_VAR(urcu_active_readers_one); |
| 233 | ooo_mem(i); |
| 234 | if |
| 235 | :: (tmp & RCU_GP_CTR_NEST_MASK) |
| 236 | && ((tmp ^ READ_CACHED_VAR(urcu_gp_ctr)) |
| 237 | & RCU_GP_CTR_BIT) -> |
| 238 | #ifndef GEN_ERROR_WRITER_PROGRESS |
| 239 | smp_mb(i); |
| 240 | #else |
| 241 | ooo_mem(i); |
| 242 | #endif |
| 243 | :: else -> |
| 244 | break; |
| 245 | fi; |
| 246 | od; |
| 247 | } |
| 248 | |
| 249 | inline wait_for_quiescent_state(tmp, i, j) |
| 250 | { |
| 251 | i = 0; |
| 252 | do |
| 253 | :: i < NR_READERS -> |
| 254 | wait_for_reader(tmp, i, j); |
| 255 | if |
| 256 | :: (NR_READERS > 1) && (i < NR_READERS - 1) |
| 257 | -> ooo_mem(j); |
| 258 | :: else |
| 259 | -> skip; |
| 260 | fi; |
| 261 | i++ |
| 262 | :: i >= NR_READERS -> break |
| 263 | od; |
| 264 | } |
| 265 | |
| 266 | /* Model the RCU read-side critical section. */ |
| 267 | |
| 268 | inline urcu_one_read(i, nest_i, tmp, tmp2) |
| 269 | { |
| 270 | nest_i = 0; |
| 271 | do |
| 272 | :: nest_i < READER_NEST_LEVEL -> |
| 273 | ooo_mem(i); |
| 274 | tmp = READ_CACHED_VAR(urcu_active_readers_one); |
| 275 | ooo_mem(i); |
| 276 | if |
| 277 | :: (!(tmp & RCU_GP_CTR_NEST_MASK)) |
| 278 | -> |
| 279 | tmp2 = READ_CACHED_VAR(urcu_gp_ctr); |
| 280 | ooo_mem(i); |
| 281 | WRITE_CACHED_VAR(urcu_active_readers_one, tmp2); |
| 282 | :: else -> |
| 283 | WRITE_CACHED_VAR(urcu_active_readers_one, |
| 284 | tmp + 1); |
| 285 | fi; |
| 286 | smp_mb(i); |
| 287 | nest_i++; |
| 288 | :: nest_i >= READER_NEST_LEVEL -> break; |
| 289 | od; |
| 290 | |
| 291 | ooo_mem(i); |
| 292 | read_generation = READ_CACHED_VAR(generation_ptr); |
| 293 | ooo_mem(i); |
| 294 | data_access = 1; |
| 295 | ooo_mem(i); |
| 296 | data_access = 0; |
| 297 | |
| 298 | nest_i = 0; |
| 299 | do |
| 300 | :: nest_i < READER_NEST_LEVEL -> |
| 301 | smp_mb(i); |
| 302 | tmp2 = READ_CACHED_VAR(urcu_active_readers_one); |
| 303 | ooo_mem(i); |
| 304 | WRITE_CACHED_VAR(urcu_active_readers_one, tmp2 - 1); |
| 305 | nest_i++; |
| 306 | :: nest_i >= READER_NEST_LEVEL -> break; |
| 307 | od; |
| 308 | ooo_mem(i); |
| 309 | //smp_mc(i); /* added */ |
| 310 | } |
| 311 | |
| 312 | active [NR_READERS] proctype urcu_reader() |
| 313 | { |
| 314 | byte i, nest_i; |
| 315 | byte tmp, tmp2; |
| 316 | |
| 317 | assert(get_pid() < NR_PROCS); |
| 318 | |
| 319 | end_reader: |
| 320 | do |
| 321 | :: 1 -> |
| 322 | /* |
| 323 | * We do not test reader's progress here, because we are mainly |
| 324 | * interested in writer's progress. The reader never blocks |
| 325 | * anyway. We have to test for reader/writer's progress |
| 326 | * separately, otherwise we could think the writer is doing |
| 327 | * progress when it's blocked by an always progressing reader. |
| 328 | */ |
| 329 | #ifdef READER_PROGRESS |
| 330 | progress_reader: |
| 331 | #endif |
| 332 | urcu_one_read(i, nest_i, tmp, tmp2); |
| 333 | od; |
| 334 | } |
| 335 | |
| 336 | /* Model the RCU update process. */ |
| 337 | |
| 338 | active [NR_WRITERS] proctype urcu_writer() |
| 339 | { |
| 340 | byte i, j; |
| 341 | byte tmp; |
| 342 | byte old_gen; |
| 343 | |
| 344 | assert(get_pid() < NR_PROCS); |
| 345 | |
| 346 | do |
| 347 | :: (READ_CACHED_VAR(generation_ptr) < 5) -> |
| 348 | #ifdef WRITER_PROGRESS |
| 349 | progress_writer1: |
| 350 | #endif |
| 351 | ooo_mem(i); |
| 352 | atomic { |
| 353 | old_gen = READ_CACHED_VAR(generation_ptr); |
| 354 | WRITE_CACHED_VAR(generation_ptr, old_gen + 1); |
| 355 | } |
| 356 | ooo_mem(i); |
| 357 | |
| 358 | do |
| 359 | :: 1 -> |
| 360 | atomic { |
| 361 | if |
| 362 | :: write_lock == 0 -> |
| 363 | write_lock = 1; |
| 364 | break; |
| 365 | :: else -> |
| 366 | skip; |
| 367 | fi; |
| 368 | } |
| 369 | od; |
| 370 | smp_mb(i); |
| 371 | tmp = READ_CACHED_VAR(urcu_gp_ctr); |
| 372 | ooo_mem(i); |
| 373 | WRITE_CACHED_VAR(urcu_gp_ctr, tmp ^ RCU_GP_CTR_BIT); |
| 374 | ooo_mem(i); |
| 375 | //smp_mc(i); |
| 376 | wait_for_quiescent_state(tmp, i, j); |
| 377 | //smp_mc(i); |
| 378 | #ifndef SINGLE_FLIP |
| 379 | ooo_mem(i); |
| 380 | tmp = READ_CACHED_VAR(urcu_gp_ctr); |
| 381 | ooo_mem(i); |
| 382 | WRITE_CACHED_VAR(urcu_gp_ctr, tmp ^ RCU_GP_CTR_BIT); |
| 383 | //smp_mc(i); |
| 384 | ooo_mem(i); |
| 385 | wait_for_quiescent_state(tmp, i, j); |
| 386 | #endif |
| 387 | smp_mb(i); |
| 388 | write_lock = 0; |
| 389 | /* free-up step, e.g., kfree(). */ |
| 390 | atomic { |
| 391 | last_free_gen = old_gen; |
| 392 | free_done = 1; |
| 393 | } |
| 394 | :: else -> break; |
| 395 | od; |
| 396 | /* |
| 397 | * Given the reader loops infinitely, let the writer also busy-loop |
| 398 | * with progress here so, with weak fairness, we can test the |
| 399 | * writer's progress. |
| 400 | */ |
| 401 | end_writer: |
| 402 | do |
| 403 | :: 1 -> |
| 404 | #ifdef WRITER_PROGRESS |
| 405 | progress_writer2: |
| 406 | #endif |
| 407 | skip; |
| 408 | od; |
| 409 | } |