| 1 | /* |
| 2 | * test-psrwlock.c |
| 3 | */ |
| 4 | |
| 5 | #include <linux/module.h> |
| 6 | #include <linux/proc_fs.h> |
| 7 | #include <linux/sched.h> |
| 8 | #include <linux/timex.h> |
| 9 | #include <linux/kthread.h> |
| 10 | #include <linux/delay.h> |
| 11 | #include <linux/hardirq.h> |
| 12 | #include <linux/module.h> |
| 13 | #include <linux/percpu.h> |
| 14 | #include <linux/spinlock.h> |
| 15 | #include <asm/ptrace.h> |
| 16 | #include <linux/psrwlock.h> |
| 17 | |
| 18 | /* Test with no contention duration, in seconds */ |
| 19 | #define SINGLE_WRITER_TEST_DURATION 10 |
| 20 | #define SINGLE_READER_TEST_DURATION 10 |
| 21 | #define MULTIPLE_READERS_TEST_DURATION 10 |
| 22 | |
| 23 | /* Test duration, in seconds */ |
| 24 | #define TEST_DURATION 60 |
| 25 | |
| 26 | #define NR_VARS 100 |
| 27 | #define NR_WRITERS 2 |
| 28 | #define NR_TRYLOCK_WRITERS 1 |
| 29 | #define NR_NPREADERS 2 |
| 30 | #define NR_TRYLOCK_READERS 1 |
| 31 | |
| 32 | /* |
| 33 | * 1 : test standard rwlock |
| 34 | * 0 : test psrwlock |
| 35 | */ |
| 36 | #define TEST_STD_RWLOCK 0 |
| 37 | |
| 38 | /* |
| 39 | * 1 : test with thread and interrupt readers. |
| 40 | * 0 : test only with thread readers. |
| 41 | */ |
| 42 | #define TEST_INTERRUPTS 1 |
| 43 | |
| 44 | #if (TEST_INTERRUPTS) |
| 45 | #define NR_INTERRUPT_READERS 1 |
| 46 | #define NR_TRYLOCK_INTERRUPT_READERS 1 |
| 47 | #else |
| 48 | #define NR_INTERRUPT_READERS 0 |
| 49 | #define NR_TRYLOCK_INTERRUPT_READERS 0 |
| 50 | #endif |
| 51 | |
| 52 | /* |
| 53 | * 1 : test with thread preemption readers. |
| 54 | * 0 : test only with non-preemptable thread readers. |
| 55 | */ |
| 56 | #define TEST_PREEMPT 1 |
| 57 | |
| 58 | #if (TEST_PREEMPT) |
| 59 | #define NR_PREADERS 2 |
| 60 | #else |
| 61 | #define NR_PREADERS 0 |
| 62 | #endif |
| 63 | |
| 64 | |
| 65 | /* |
| 66 | * Writer iteration delay, in us. 0 for busy loop. Caution : writers can |
| 67 | * starve readers. |
| 68 | */ |
| 69 | #define WRITER_DELAY 100 |
| 70 | #define TRYLOCK_WRITER_DELAY 1000 |
| 71 | |
| 72 | /* |
| 73 | * Number of iterations after which a trylock writer fails. |
| 74 | * -1 for infinite loop. |
| 75 | */ |
| 76 | #define TRYLOCK_WRITERS_FAIL_ITER 100 |
| 77 | |
| 78 | /* Thread and interrupt reader delay, in ms */ |
| 79 | #define THREAD_READER_DELAY 0 /* busy loop */ |
| 80 | #define INTERRUPT_READER_DELAY 100 |
| 81 | |
| 82 | #ifdef CONFIG_PREEMPT |
| 83 | #define yield_in_non_preempt() |
| 84 | #else |
| 85 | #define yield_in_non_preempt() yield() |
| 86 | #endif |
| 87 | |
| 88 | static int var[NR_VARS]; |
| 89 | static struct task_struct *preader_threads[NR_PREADERS]; |
| 90 | static struct task_struct *npreader_threads[NR_NPREADERS]; |
| 91 | static struct task_struct *trylock_reader_threads[NR_TRYLOCK_READERS]; |
| 92 | static struct task_struct *writer_threads[NR_WRITERS]; |
| 93 | static struct task_struct *trylock_writer_threads[NR_TRYLOCK_WRITERS]; |
| 94 | static struct task_struct *interrupt_reader[NR_INTERRUPT_READERS]; |
| 95 | static struct task_struct *trylock_interrupt_reader[NR_TRYLOCK_INTERRUPT_READERS]; |
| 96 | |
| 97 | #if (TEST_STD_RWLOCK) |
| 98 | |
| 99 | static DEFINE_RWLOCK(std_rw_lock); |
| 100 | |
| 101 | #define wrap_read_lock() read_lock(&std_rw_lock) |
| 102 | #define wrap_read_trylock() read_trylock(&std_rw_lock) |
| 103 | #define wrap_read_unlock() read_unlock(&std_rw_lock) |
| 104 | |
| 105 | #define wrap_read_lock_inatomic() read_lock(&std_rw_lock) |
| 106 | #define wrap_read_trylock_inatomic() read_trylock(&std_rw_lock) |
| 107 | |
| 108 | #define wrap_read_lock_irq() read_lock(&std_rw_lock) |
| 109 | #define wrap_read_trylock_irq() read_trylock(&std_rw_lock) |
| 110 | |
| 111 | #if (TEST_INTERRUPTS) |
| 112 | #define wrap_write_lock() write_lock_irq(&std_rw_lock) |
| 113 | #define wrap_write_unlock() write_unlock_irq(&std_rw_lock) |
| 114 | #else |
| 115 | #define wrap_write_lock() write_lock(&std_rw_lock) |
| 116 | #define wrap_write_unlock() write_unlock(&std_rw_lock) |
| 117 | #endif |
| 118 | |
| 119 | #define wrap_write_trylock() write_trylock(&std_rw_lock) |
| 120 | |
| 121 | #else |
| 122 | |
| 123 | #if (TEST_INTERRUPTS) |
| 124 | #if (TEST_PREEMPT) |
| 125 | #define PSRWLOCKWCTX PSRW_PRIO_P |
| 126 | #define PSRWLOCKRCTX (PSR_IRQ | PSR_NPTHREAD | PSR_PTHREAD) |
| 127 | #else |
| 128 | #define PSRWLOCKWCTX PSRW_PRIO_NP |
| 129 | #define PSRWLOCKRCTX (PSR_IRQ | PSR_NPTHREAD) |
| 130 | #endif |
| 131 | #else |
| 132 | #if (TEST_PREEMPT) |
| 133 | #define PSRWLOCKWCTX PSRW_PRIO_P |
| 134 | #define PSRWLOCKRCTX (PSR_NPTHREAD | PSR_PTHREAD) |
| 135 | #else |
| 136 | #define PSRWLOCKWCTX PSRW_PRIO_NP |
| 137 | #define PSRWLOCKRCTX (PSR_NPTHREAD) |
| 138 | #endif |
| 139 | #endif |
| 140 | |
| 141 | static DEFINE_PSRWLOCK(psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX); |
| 142 | CHECK_PSRWLOCK_MAP(psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX); |
| 143 | |
| 144 | |
| 145 | #if (TEST_PREEMPT) |
| 146 | #define wrap_read_lock() \ |
| 147 | psread_lock(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 148 | #define wrap_read_trylock() \ |
| 149 | psread_trylock(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 150 | #else |
| 151 | #define wrap_read_lock() \ |
| 152 | psread_lock_inatomic(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 153 | #define wrap_read_trylock() \ |
| 154 | psread_trylock_inatomic(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 155 | #endif |
| 156 | #define wrap_read_unlock() \ |
| 157 | psread_unlock(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 158 | |
| 159 | #define wrap_read_lock_inatomic() \ |
| 160 | psread_lock_inatomic(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 161 | #define wrap_read_trylock_inatomic() \ |
| 162 | psread_trylock_inatomic(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 163 | |
| 164 | #define wrap_read_lock_irq() \ |
| 165 | psread_lock_irq(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 166 | #define wrap_read_trylock_irq() \ |
| 167 | psread_trylock_irq(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 168 | |
| 169 | #define wrap_write_lock() \ |
| 170 | pswrite_lock(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 171 | #define wrap_write_unlock() \ |
| 172 | pswrite_unlock(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 173 | #define wrap_write_trylock() \ |
| 174 | pswrite_trylock(&psrwlock, PSRWLOCKWCTX, PSRWLOCKRCTX) |
| 175 | |
| 176 | #endif |
| 177 | |
| 178 | static cycles_t cycles_calibration_min, |
| 179 | cycles_calibration_avg, |
| 180 | cycles_calibration_max; |
| 181 | |
| 182 | static inline cycles_t calibrate_cycles(cycles_t cycles) |
| 183 | { |
| 184 | return cycles - cycles_calibration_avg; |
| 185 | } |
| 186 | |
| 187 | struct proc_dir_entry *pentry = NULL; |
| 188 | |
| 189 | static int p_or_np_reader_thread(const char *typename, |
| 190 | void *data, int preemptable) |
| 191 | { |
| 192 | int i; |
| 193 | int prev, cur; |
| 194 | unsigned long iter = 0; |
| 195 | cycles_t time1, time2, delay; |
| 196 | cycles_t ldelaymax = 0, ldelaymin = ULLONG_MAX, ldelayavg = 0; |
| 197 | cycles_t udelaymax = 0, udelaymin = ULLONG_MAX, udelayavg = 0; |
| 198 | |
| 199 | printk("%s/%lu runnning\n", typename, (unsigned long)data); |
| 200 | do { |
| 201 | iter++; |
| 202 | if (!preemptable) |
| 203 | preempt_disable(); |
| 204 | rdtsc_barrier(); |
| 205 | time1 = get_cycles(); |
| 206 | rdtsc_barrier(); |
| 207 | |
| 208 | if (!preemptable) |
| 209 | wrap_read_lock_inatomic(); |
| 210 | else |
| 211 | wrap_read_lock(); |
| 212 | |
| 213 | rdtsc_barrier(); |
| 214 | time2 = get_cycles(); |
| 215 | rdtsc_barrier(); |
| 216 | delay = time2 - time1; |
| 217 | ldelaymax = max(ldelaymax, delay); |
| 218 | ldelaymin = min(ldelaymin, delay); |
| 219 | ldelayavg += delay; |
| 220 | prev = var[0]; |
| 221 | for (i = 1; i < NR_VARS; i++) { |
| 222 | cur = var[i]; |
| 223 | if (cur != prev) { |
| 224 | printk(KERN_ALERT |
| 225 | "Unequal cur %d/prev %d at i %d, iter %lu " |
| 226 | "in reader thread\n", |
| 227 | cur, prev, i, iter); |
| 228 | } |
| 229 | } |
| 230 | |
| 231 | rdtsc_barrier(); |
| 232 | time1 = get_cycles(); |
| 233 | rdtsc_barrier(); |
| 234 | |
| 235 | wrap_read_unlock(); |
| 236 | |
| 237 | rdtsc_barrier(); |
| 238 | time2 = get_cycles(); |
| 239 | rdtsc_barrier(); |
| 240 | delay = time2 - time1; |
| 241 | udelaymax = max(udelaymax, delay); |
| 242 | udelaymin = min(udelaymin, delay); |
| 243 | udelayavg += delay; |
| 244 | |
| 245 | if (!preemptable) |
| 246 | preempt_enable(); |
| 247 | |
| 248 | if (THREAD_READER_DELAY) |
| 249 | msleep(THREAD_READER_DELAY); |
| 250 | yield_in_non_preempt(); |
| 251 | } while (!kthread_should_stop()); |
| 252 | if (!iter) { |
| 253 | printk("%s/%lu iterations : %lu", typename, |
| 254 | (unsigned long)data, iter); |
| 255 | } else { |
| 256 | ldelayavg /= iter; |
| 257 | udelayavg /= iter; |
| 258 | printk("%s/%lu iterations : %lu, " |
| 259 | "lock delay [min,avg,max] %llu,%llu,%llu cycles\n", |
| 260 | typename, |
| 261 | (unsigned long)data, iter, |
| 262 | calibrate_cycles(ldelaymin), |
| 263 | calibrate_cycles(ldelayavg), |
| 264 | calibrate_cycles(ldelaymax)); |
| 265 | printk("%s/%lu iterations : %lu, " |
| 266 | "unlock delay [min,avg,max] %llu,%llu,%llu cycles\n", |
| 267 | typename, |
| 268 | (unsigned long)data, iter, |
| 269 | calibrate_cycles(udelaymin), |
| 270 | calibrate_cycles(udelayavg), |
| 271 | calibrate_cycles(udelaymax)); |
| 272 | } |
| 273 | return 0; |
| 274 | } |
| 275 | |
| 276 | static int preader_thread(void *data) |
| 277 | { |
| 278 | return p_or_np_reader_thread("preader_thread", data, 1); |
| 279 | } |
| 280 | |
| 281 | static int npreader_thread(void *data) |
| 282 | { |
| 283 | return p_or_np_reader_thread("npreader_thread", data, 0); |
| 284 | } |
| 285 | |
| 286 | static int trylock_reader_thread(void *data) |
| 287 | { |
| 288 | int i; |
| 289 | int prev, cur; |
| 290 | unsigned long iter = 0, success_iter = 0; |
| 291 | |
| 292 | printk("trylock_reader_thread/%lu runnning\n", (unsigned long)data); |
| 293 | do { |
| 294 | #if (!TEST_PREEMPT) |
| 295 | preempt_disable(); |
| 296 | #endif |
| 297 | while (!wrap_read_trylock()) { |
| 298 | cpu_relax(); |
| 299 | iter++; |
| 300 | } |
| 301 | success_iter++; |
| 302 | prev = var[0]; |
| 303 | for (i = 1; i < NR_VARS; i++) { |
| 304 | cur = var[i]; |
| 305 | if (cur != prev) { |
| 306 | printk(KERN_ALERT |
| 307 | "Unequal cur %d/prev %d at i %d, iter %lu " |
| 308 | "in trylock reader thread\n", |
| 309 | cur, prev, i, iter); |
| 310 | } |
| 311 | } |
| 312 | wrap_read_unlock(); |
| 313 | #if (!TEST_PREEMPT) |
| 314 | preempt_enable(); |
| 315 | #endif |
| 316 | if (THREAD_READER_DELAY) |
| 317 | msleep(THREAD_READER_DELAY); |
| 318 | yield_in_non_preempt(); |
| 319 | } while (!kthread_should_stop()); |
| 320 | printk("trylock_reader_thread/%lu iterations : %lu, " |
| 321 | "successful iterations : %lu\n", |
| 322 | (unsigned long)data, iter + success_iter, success_iter); |
| 323 | return 0; |
| 324 | } |
| 325 | |
| 326 | DEFINE_PER_CPU(cycles_t, int_ldelaymin); |
| 327 | DEFINE_PER_CPU(cycles_t, int_ldelayavg); |
| 328 | DEFINE_PER_CPU(cycles_t, int_ldelaymax); |
| 329 | DEFINE_PER_CPU(cycles_t, int_udelaymin); |
| 330 | DEFINE_PER_CPU(cycles_t, int_udelayavg); |
| 331 | DEFINE_PER_CPU(cycles_t, int_udelaymax); |
| 332 | DEFINE_PER_CPU(cycles_t, int_ipi_nr); |
| 333 | |
| 334 | static void interrupt_reader_ipi(void *data) |
| 335 | { |
| 336 | int i; |
| 337 | int prev, cur; |
| 338 | cycles_t time1, time2; |
| 339 | cycles_t *ldelaymax, *ldelaymin, *ldelayavg, *ipi_nr, delay; |
| 340 | cycles_t *udelaymax, *udelaymin, *udelayavg; |
| 341 | |
| 342 | /* |
| 343 | * Skip the ipi caller, not in irq context. |
| 344 | */ |
| 345 | if (!in_irq()) |
| 346 | return; |
| 347 | |
| 348 | ldelaymax = &per_cpu(int_ldelaymax, smp_processor_id()); |
| 349 | ldelaymin = &per_cpu(int_ldelaymin, smp_processor_id()); |
| 350 | ldelayavg = &per_cpu(int_ldelayavg, smp_processor_id()); |
| 351 | udelaymax = &per_cpu(int_udelaymax, smp_processor_id()); |
| 352 | udelaymin = &per_cpu(int_udelaymin, smp_processor_id()); |
| 353 | udelayavg = &per_cpu(int_udelayavg, smp_processor_id()); |
| 354 | ipi_nr = &per_cpu(int_ipi_nr, smp_processor_id()); |
| 355 | |
| 356 | rdtsc_barrier(); |
| 357 | time1 = get_cycles(); |
| 358 | rdtsc_barrier(); |
| 359 | |
| 360 | wrap_read_lock_irq(); |
| 361 | |
| 362 | rdtsc_barrier(); |
| 363 | time2 = get_cycles(); |
| 364 | rdtsc_barrier(); |
| 365 | delay = time2 - time1; |
| 366 | *ldelaymax = max(*ldelaymax, delay); |
| 367 | *ldelaymin = min(*ldelaymin, delay); |
| 368 | *ldelayavg += delay; |
| 369 | (*ipi_nr)++; |
| 370 | prev = var[0]; |
| 371 | for (i = 1; i < NR_VARS; i++) { |
| 372 | cur = var[i]; |
| 373 | if (cur != prev) |
| 374 | printk(KERN_ALERT |
| 375 | "Unequal cur %d/prev %d at i %d in interrupt\n", |
| 376 | cur, prev, i); |
| 377 | } |
| 378 | rdtsc_barrier(); |
| 379 | time1 = get_cycles(); |
| 380 | rdtsc_barrier(); |
| 381 | wrap_read_unlock(); |
| 382 | time2 = get_cycles(); |
| 383 | rdtsc_barrier(); |
| 384 | delay = time2 - time1; |
| 385 | *udelaymax = max(*udelaymax, delay); |
| 386 | *udelaymin = min(*udelaymin, delay); |
| 387 | *udelayavg += delay; |
| 388 | } |
| 389 | |
| 390 | DEFINE_PER_CPU(unsigned long, trylock_int_iter); |
| 391 | DEFINE_PER_CPU(unsigned long, trylock_int_success); |
| 392 | |
| 393 | static void trylock_interrupt_reader_ipi(void *data) |
| 394 | { |
| 395 | int i; |
| 396 | int prev, cur; |
| 397 | |
| 398 | /* |
| 399 | * Skip the ipi caller, not in irq context. |
| 400 | */ |
| 401 | if (!in_irq()) |
| 402 | return; |
| 403 | |
| 404 | per_cpu(trylock_int_iter, smp_processor_id())++; |
| 405 | while (!wrap_read_trylock_irq()) |
| 406 | per_cpu(trylock_int_iter, smp_processor_id())++; |
| 407 | per_cpu(trylock_int_success, smp_processor_id())++; |
| 408 | prev = var[0]; |
| 409 | for (i = 1; i < NR_VARS; i++) { |
| 410 | cur = var[i]; |
| 411 | if (cur != prev) |
| 412 | printk(KERN_ALERT |
| 413 | "Unequal cur %d/prev %d at i %d in interrupt\n", |
| 414 | cur, prev, i); |
| 415 | } |
| 416 | wrap_read_unlock(); |
| 417 | } |
| 418 | |
| 419 | |
| 420 | static int interrupt_reader_thread(void *data) |
| 421 | { |
| 422 | unsigned long iter = 0; |
| 423 | int i; |
| 424 | |
| 425 | for_each_online_cpu(i) { |
| 426 | per_cpu(int_ldelaymax, i) = 0; |
| 427 | per_cpu(int_ldelaymin, i) = ULLONG_MAX; |
| 428 | per_cpu(int_ldelayavg, i) = 0; |
| 429 | per_cpu(int_udelaymax, i) = 0; |
| 430 | per_cpu(int_udelaymin, i) = ULLONG_MAX; |
| 431 | per_cpu(int_udelayavg, i) = 0; |
| 432 | per_cpu(int_ipi_nr, i) = 0; |
| 433 | } |
| 434 | do { |
| 435 | iter++; |
| 436 | on_each_cpu(interrupt_reader_ipi, NULL, 0); |
| 437 | if (INTERRUPT_READER_DELAY) |
| 438 | msleep(INTERRUPT_READER_DELAY); |
| 439 | yield_in_non_preempt(); |
| 440 | } while (!kthread_should_stop()); |
| 441 | printk("interrupt_reader_thread/%lu iterations : %lu\n", |
| 442 | (unsigned long)data, iter); |
| 443 | for_each_online_cpu(i) { |
| 444 | if (!per_cpu(int_ipi_nr, i)) |
| 445 | continue; |
| 446 | per_cpu(int_ldelayavg, i) /= per_cpu(int_ipi_nr, i); |
| 447 | per_cpu(int_udelayavg, i) /= per_cpu(int_ipi_nr, i); |
| 448 | printk("interrupt readers on CPU %i, " |
| 449 | "lock delay [min,avg,max] %llu,%llu,%llu cycles\n", |
| 450 | i, |
| 451 | calibrate_cycles(per_cpu(int_ldelaymin, i)), |
| 452 | calibrate_cycles(per_cpu(int_ldelayavg, i)), |
| 453 | calibrate_cycles(per_cpu(int_ldelaymax, i))); |
| 454 | printk("interrupt readers on CPU %i, " |
| 455 | "unlock delay [min,avg,max] %llu,%llu,%llu cycles\n", |
| 456 | i, |
| 457 | calibrate_cycles(per_cpu(int_udelaymin, i)), |
| 458 | calibrate_cycles(per_cpu(int_udelayavg, i)), |
| 459 | calibrate_cycles(per_cpu(int_udelaymax, i))); |
| 460 | } |
| 461 | return 0; |
| 462 | } |
| 463 | |
| 464 | static int trylock_interrupt_reader_thread(void *data) |
| 465 | { |
| 466 | unsigned long iter = 0; |
| 467 | int i; |
| 468 | |
| 469 | do { |
| 470 | iter++; |
| 471 | on_each_cpu(trylock_interrupt_reader_ipi, NULL, 0); |
| 472 | if (INTERRUPT_READER_DELAY) |
| 473 | msleep(INTERRUPT_READER_DELAY); |
| 474 | yield_in_non_preempt(); |
| 475 | } while (!kthread_should_stop()); |
| 476 | printk("trylock_interrupt_reader_thread/%lu iterations : %lu\n", |
| 477 | (unsigned long)data, iter); |
| 478 | for_each_online_cpu(i) { |
| 479 | printk("trylock interrupt readers on CPU %i, " |
| 480 | "iterations %lu, " |
| 481 | "successful iterations : %lu\n", |
| 482 | i, per_cpu(trylock_int_iter, i), |
| 483 | per_cpu(trylock_int_success, i)); |
| 484 | per_cpu(trylock_int_iter, i) = 0; |
| 485 | per_cpu(trylock_int_success, i) = 0; |
| 486 | } |
| 487 | return 0; |
| 488 | } |
| 489 | |
| 490 | static int writer_thread(void *data) |
| 491 | { |
| 492 | int i; |
| 493 | int new, prev, cur; |
| 494 | unsigned long iter = 0; |
| 495 | cycles_t time1, time2, delay; |
| 496 | cycles_t ldelaymax = 0, ldelaymin = ULLONG_MAX, ldelayavg = 0; |
| 497 | cycles_t udelaymax = 0, udelaymin = ULLONG_MAX, udelayavg = 0; |
| 498 | |
| 499 | printk("writer_thread/%lu runnning\n", (unsigned long)data); |
| 500 | do { |
| 501 | iter++; |
| 502 | #if (!TEST_PREEMPT) |
| 503 | preempt_disable(); |
| 504 | #endif |
| 505 | rdtsc_barrier(); |
| 506 | time1 = get_cycles(); |
| 507 | rdtsc_barrier(); |
| 508 | |
| 509 | wrap_write_lock(); |
| 510 | |
| 511 | rdtsc_barrier(); |
| 512 | time2 = get_cycles(); |
| 513 | rdtsc_barrier(); |
| 514 | delay = time2 - time1; |
| 515 | ldelaymax = max(ldelaymax, delay); |
| 516 | ldelaymin = min(ldelaymin, delay); |
| 517 | ldelayavg += delay; |
| 518 | /* |
| 519 | * Read the previous values, check that they are coherent. |
| 520 | */ |
| 521 | prev = var[0]; |
| 522 | for (i = 1; i < NR_VARS; i++) { |
| 523 | cur = var[i]; |
| 524 | if (cur != prev) |
| 525 | printk(KERN_ALERT |
| 526 | "Unequal cur %d/prev %d at i %d, iter %lu " |
| 527 | "in writer thread\n", |
| 528 | cur, prev, i, iter); |
| 529 | } |
| 530 | new = (int)get_cycles(); |
| 531 | for (i = 0; i < NR_VARS; i++) { |
| 532 | var[i] = new; |
| 533 | } |
| 534 | |
| 535 | rdtsc_barrier(); |
| 536 | time1 = get_cycles(); |
| 537 | rdtsc_barrier(); |
| 538 | |
| 539 | wrap_write_unlock(); |
| 540 | |
| 541 | rdtsc_barrier(); |
| 542 | time2 = get_cycles(); |
| 543 | rdtsc_barrier(); |
| 544 | delay = time2 - time1; |
| 545 | udelaymax = max(udelaymax, delay); |
| 546 | udelaymin = min(udelaymin, delay); |
| 547 | udelayavg += delay; |
| 548 | |
| 549 | #if (!TEST_PREEMPT) |
| 550 | preempt_enable(); |
| 551 | #endif |
| 552 | if (WRITER_DELAY > 0) |
| 553 | udelay(WRITER_DELAY); |
| 554 | cpu_relax(); /* |
| 555 | * make sure we don't busy-loop faster than |
| 556 | * the lock busy-loop, it would cause reader and |
| 557 | * writer starvation. |
| 558 | */ |
| 559 | yield_in_non_preempt(); |
| 560 | } while (!kthread_should_stop()); |
| 561 | ldelayavg /= iter; |
| 562 | udelayavg /= iter; |
| 563 | printk("writer_thread/%lu iterations : %lu, " |
| 564 | "lock delay [min,avg,max] %llu,%llu,%llu cycles\n", |
| 565 | (unsigned long)data, iter, |
| 566 | calibrate_cycles(ldelaymin), |
| 567 | calibrate_cycles(ldelayavg), |
| 568 | calibrate_cycles(ldelaymax)); |
| 569 | printk("writer_thread/%lu iterations : %lu, " |
| 570 | "unlock delay [min,avg,max] %llu,%llu,%llu cycles\n", |
| 571 | (unsigned long)data, iter, |
| 572 | calibrate_cycles(udelaymin), |
| 573 | calibrate_cycles(udelayavg), |
| 574 | calibrate_cycles(udelaymax)); |
| 575 | return 0; |
| 576 | } |
| 577 | |
| 578 | static int trylock_writer_thread(void *data) |
| 579 | { |
| 580 | int i; |
| 581 | int new; |
| 582 | unsigned long iter = 0, success = 0, fail = 0; |
| 583 | |
| 584 | printk("trylock_writer_thread/%lu runnning\n", (unsigned long)data); |
| 585 | do { |
| 586 | #if ((!TEST_PREEMPT) && (!TEST_STD_RWLOCK)) |
| 587 | preempt_disable(); |
| 588 | #endif |
| 589 | |
| 590 | #if (TEST_STD_RWLOCK && TEST_INTERRUPTS) |
| 591 | /* std write trylock cannot disable interrupts. */ |
| 592 | local_irq_disable(); |
| 593 | #endif |
| 594 | |
| 595 | #if (TRYLOCK_WRITERS_FAIL_ITER == -1) |
| 596 | for (;;) { |
| 597 | iter++; |
| 598 | if (wrap_write_trylock()) |
| 599 | goto locked; |
| 600 | cpu_relax(); |
| 601 | } |
| 602 | #else |
| 603 | for (i = 0; i < TRYLOCK_WRITERS_FAIL_ITER; i++) { |
| 604 | iter++; |
| 605 | if (wrap_write_trylock()) |
| 606 | goto locked; |
| 607 | cpu_relax(); |
| 608 | } |
| 609 | #endif |
| 610 | fail++; |
| 611 | |
| 612 | #if (TEST_STD_RWLOCK && TEST_INTERRUPTS) |
| 613 | local_irq_enable(); |
| 614 | #endif |
| 615 | |
| 616 | #if ((!TEST_PREEMPT) && (!TEST_STD_RWLOCK)) |
| 617 | preempt_enable(); |
| 618 | #endif |
| 619 | goto loop; |
| 620 | locked: |
| 621 | success++; |
| 622 | new = (int)get_cycles(); |
| 623 | for (i = 0; i < NR_VARS; i++) { |
| 624 | var[i] = new; |
| 625 | } |
| 626 | wrap_write_unlock(); |
| 627 | #if ((!TEST_PREEMPT) && (!TEST_STD_RWLOCK)) |
| 628 | preempt_enable(); |
| 629 | #endif |
| 630 | loop: |
| 631 | if (TRYLOCK_WRITER_DELAY > 0) |
| 632 | udelay(TRYLOCK_WRITER_DELAY); |
| 633 | cpu_relax(); /* |
| 634 | * make sure we don't busy-loop faster than |
| 635 | * the lock busy-loop, it would cause reader and |
| 636 | * writer starvation. |
| 637 | */ |
| 638 | yield_in_non_preempt(); |
| 639 | } while (!kthread_should_stop()); |
| 640 | printk("trylock_writer_thread/%lu iterations : " |
| 641 | "[try,success,fail after %d try], " |
| 642 | "%lu,%lu,%lu\n", |
| 643 | (unsigned long)data, TRYLOCK_WRITERS_FAIL_ITER, |
| 644 | iter, success, fail); |
| 645 | return 0; |
| 646 | } |
| 647 | |
| 648 | static void psrwlock_create(void) |
| 649 | { |
| 650 | unsigned long i; |
| 651 | |
| 652 | for (i = 0; i < NR_PREADERS; i++) { |
| 653 | printk("starting preemptable reader thread %lu\n", i); |
| 654 | preader_threads[i] = kthread_run(preader_thread, (void *)i, |
| 655 | "psrwlock_preader"); |
| 656 | BUG_ON(!preader_threads[i]); |
| 657 | } |
| 658 | |
| 659 | for (i = 0; i < NR_NPREADERS; i++) { |
| 660 | printk("starting non-preemptable reader thread %lu\n", i); |
| 661 | npreader_threads[i] = kthread_run(npreader_thread, (void *)i, |
| 662 | "psrwlock_npreader"); |
| 663 | BUG_ON(!npreader_threads[i]); |
| 664 | } |
| 665 | |
| 666 | for (i = 0; i < NR_TRYLOCK_READERS; i++) { |
| 667 | printk("starting trylock reader thread %lu\n", i); |
| 668 | trylock_reader_threads[i] = kthread_run(trylock_reader_thread, |
| 669 | (void *)i, "psrwlock_trylock_reader"); |
| 670 | BUG_ON(!trylock_reader_threads[i]); |
| 671 | } |
| 672 | for (i = 0; i < NR_INTERRUPT_READERS; i++) { |
| 673 | printk("starting interrupt reader %lu\n", i); |
| 674 | interrupt_reader[i] = kthread_run(interrupt_reader_thread, |
| 675 | (void *)i, |
| 676 | "psrwlock_interrupt_reader"); |
| 677 | } |
| 678 | for (i = 0; i < NR_TRYLOCK_INTERRUPT_READERS; i++) { |
| 679 | printk("starting trylock interrupt reader %lu\n", i); |
| 680 | trylock_interrupt_reader[i] = |
| 681 | kthread_run(trylock_interrupt_reader_thread, |
| 682 | (void *)i, "psrwlock_trylock_interrupt_reader"); |
| 683 | } |
| 684 | for (i = 0; i < NR_WRITERS; i++) { |
| 685 | printk("starting writer thread %lu\n", i); |
| 686 | writer_threads[i] = kthread_run(writer_thread, (void *)i, |
| 687 | "psrwlock_writer"); |
| 688 | BUG_ON(!writer_threads[i]); |
| 689 | } |
| 690 | for (i = 0; i < NR_TRYLOCK_WRITERS; i++) { |
| 691 | printk("starting trylock writer thread %lu\n", i); |
| 692 | trylock_writer_threads[i] = kthread_run(trylock_writer_thread, |
| 693 | (void *)i, "psrwlock_trylock_writer"); |
| 694 | BUG_ON(!trylock_writer_threads[i]); |
| 695 | } |
| 696 | } |
| 697 | |
| 698 | static void psrwlock_stop(void) |
| 699 | { |
| 700 | unsigned long i; |
| 701 | |
| 702 | for (i = 0; i < NR_WRITERS; i++) |
| 703 | kthread_stop(writer_threads[i]); |
| 704 | for (i = 0; i < NR_TRYLOCK_WRITERS; i++) |
| 705 | kthread_stop(trylock_writer_threads[i]); |
| 706 | for (i = 0; i < NR_NPREADERS; i++) |
| 707 | kthread_stop(npreader_threads[i]); |
| 708 | for (i = 0; i < NR_PREADERS; i++) |
| 709 | kthread_stop(preader_threads[i]); |
| 710 | for (i = 0; i < NR_TRYLOCK_READERS; i++) |
| 711 | kthread_stop(trylock_reader_threads[i]); |
| 712 | for (i = 0; i < NR_INTERRUPT_READERS; i++) |
| 713 | kthread_stop(interrupt_reader[i]); |
| 714 | for (i = 0; i < NR_TRYLOCK_INTERRUPT_READERS; i++) |
| 715 | kthread_stop(trylock_interrupt_reader[i]); |
| 716 | } |
| 717 | |
| 718 | |
| 719 | static void perform_test(const char *name, void (*callback)(void)) |
| 720 | { |
| 721 | printk("%s\n", name); |
| 722 | callback(); |
| 723 | } |
| 724 | |
| 725 | static int my_open(struct inode *inode, struct file *file) |
| 726 | { |
| 727 | unsigned long i; |
| 728 | cycles_t time1, time2, delay; |
| 729 | |
| 730 | printk("** get_cycles calibration **\n"); |
| 731 | cycles_calibration_min = ULLONG_MAX; |
| 732 | cycles_calibration_avg = 0; |
| 733 | cycles_calibration_max = 0; |
| 734 | |
| 735 | local_irq_disable(); |
| 736 | for (i = 0; i < 10; i++) { |
| 737 | rdtsc_barrier(); |
| 738 | time1 = get_cycles(); |
| 739 | rdtsc_barrier(); |
| 740 | rdtsc_barrier(); |
| 741 | time2 = get_cycles(); |
| 742 | rdtsc_barrier(); |
| 743 | delay = time2 - time1; |
| 744 | cycles_calibration_min = min(cycles_calibration_min, delay); |
| 745 | cycles_calibration_avg += delay; |
| 746 | cycles_calibration_max = max(cycles_calibration_max, delay); |
| 747 | } |
| 748 | cycles_calibration_avg /= 10; |
| 749 | local_irq_enable(); |
| 750 | |
| 751 | printk("get_cycles takes [min,avg,max] %llu,%llu,%llu cycles, " |
| 752 | "results calibrated on avg\n", |
| 753 | cycles_calibration_min, |
| 754 | cycles_calibration_avg, |
| 755 | cycles_calibration_max); |
| 756 | printk("\n"); |
| 757 | |
| 758 | #if (NR_WRITERS) |
| 759 | printk("** Single writer test, no contention **\n"); |
| 760 | psrwlock_profile_latency_reset(); |
| 761 | writer_threads[0] = kthread_run(writer_thread, (void *)0, |
| 762 | "psrwlock_writer"); |
| 763 | BUG_ON(!writer_threads[0]); |
| 764 | ssleep(SINGLE_WRITER_TEST_DURATION); |
| 765 | kthread_stop(writer_threads[0]); |
| 766 | printk("\n"); |
| 767 | |
| 768 | psrwlock_profile_latency_print(); |
| 769 | #endif |
| 770 | |
| 771 | #if (NR_TRYLOCK_WRITERS) |
| 772 | printk("** Single trylock writer test, no contention **\n"); |
| 773 | psrwlock_profile_latency_reset(); |
| 774 | trylock_writer_threads[0] = kthread_run(trylock_writer_thread, |
| 775 | (void *)0, |
| 776 | "trylock_psrwlock_writer"); |
| 777 | BUG_ON(!trylock_writer_threads[0]); |
| 778 | ssleep(SINGLE_WRITER_TEST_DURATION); |
| 779 | kthread_stop(trylock_writer_threads[0]); |
| 780 | printk("\n"); |
| 781 | |
| 782 | psrwlock_profile_latency_print(); |
| 783 | #endif |
| 784 | |
| 785 | #if (TEST_PREEMPT) |
| 786 | printk("** Single preemptable reader test, no contention **\n"); |
| 787 | psrwlock_profile_latency_reset(); |
| 788 | preader_threads[0] = kthread_run(preader_thread, (void *)0, |
| 789 | "psrwlock_preader"); |
| 790 | BUG_ON(!preader_threads[0]); |
| 791 | ssleep(SINGLE_READER_TEST_DURATION); |
| 792 | kthread_stop(preader_threads[0]); |
| 793 | printk("\n"); |
| 794 | |
| 795 | psrwlock_profile_latency_print(); |
| 796 | #endif |
| 797 | |
| 798 | printk("** Single non-preemptable reader test, no contention **\n"); |
| 799 | psrwlock_profile_latency_reset(); |
| 800 | npreader_threads[0] = kthread_run(npreader_thread, (void *)0, |
| 801 | "psrwlock_npreader"); |
| 802 | BUG_ON(!npreader_threads[0]); |
| 803 | ssleep(SINGLE_READER_TEST_DURATION); |
| 804 | kthread_stop(npreader_threads[0]); |
| 805 | printk("\n"); |
| 806 | |
| 807 | psrwlock_profile_latency_print(); |
| 808 | |
| 809 | printk("** Multiple p/non-p readers test, no contention **\n"); |
| 810 | psrwlock_profile_latency_reset(); |
| 811 | #if (TEST_PREEMPT) |
| 812 | for (i = 0; i < NR_PREADERS; i++) { |
| 813 | printk("starting preader thread %lu\n", i); |
| 814 | preader_threads[i] = kthread_run(preader_thread, (void *)i, |
| 815 | "psrwlock_preader"); |
| 816 | BUG_ON(!preader_threads[i]); |
| 817 | } |
| 818 | #endif |
| 819 | for (i = 0; i < NR_NPREADERS; i++) { |
| 820 | printk("starting npreader thread %lu\n", i); |
| 821 | npreader_threads[i] = kthread_run(npreader_thread, (void *)i, |
| 822 | "psrwlock_npreader"); |
| 823 | BUG_ON(!npreader_threads[i]); |
| 824 | } |
| 825 | ssleep(SINGLE_READER_TEST_DURATION); |
| 826 | for (i = 0; i < NR_NPREADERS; i++) |
| 827 | kthread_stop(npreader_threads[i]); |
| 828 | #if (TEST_PREEMPT) |
| 829 | for (i = 0; i < NR_PREADERS; i++) |
| 830 | kthread_stop(preader_threads[i]); |
| 831 | #endif |
| 832 | printk("\n"); |
| 833 | |
| 834 | psrwlock_profile_latency_print(); |
| 835 | |
| 836 | printk("** High contention test **\n"); |
| 837 | psrwlock_profile_latency_reset(); |
| 838 | perform_test("psrwlock-create", psrwlock_create); |
| 839 | ssleep(TEST_DURATION); |
| 840 | perform_test("psrwlock-stop", psrwlock_stop); |
| 841 | printk("\n"); |
| 842 | psrwlock_profile_latency_print(); |
| 843 | |
| 844 | return -EPERM; |
| 845 | } |
| 846 | |
| 847 | |
| 848 | static struct file_operations my_operations = { |
| 849 | .open = my_open, |
| 850 | }; |
| 851 | |
| 852 | int init_module(void) |
| 853 | { |
| 854 | pentry = create_proc_entry("testpsrwlock", 0444, NULL); |
| 855 | if (pentry) |
| 856 | pentry->proc_fops = &my_operations; |
| 857 | |
| 858 | printk("UC_READER_MASK : %08X\n", UC_READER_MASK); |
| 859 | printk("UC_HARDIRQ_R_MASK: %08X\n", UC_HARDIRQ_READER_MASK); |
| 860 | printk("UC_SOFTIRQ_R_MASK: %08X\n", UC_SOFTIRQ_READER_MASK); |
| 861 | printk("UC_NPTHREA_R_MASK: %08X\n", UC_NPTHREAD_READER_MASK); |
| 862 | printk("UC_PTHREAD_R_MASK: %08X\n", UC_PTHREAD_READER_MASK); |
| 863 | printk("UC_WRITER : %08X\n", UC_WRITER); |
| 864 | printk("UC_SLOW_WRITER : %08X\n", UC_SLOW_WRITER); |
| 865 | printk("UC_WQ_ACTIVE : %08X\n", UC_WQ_ACTIVE); |
| 866 | printk("WS_MASK : %08X\n", WS_MASK); |
| 867 | printk("WS_WQ_MUTEX : %08X\n", WS_WQ_MUTEX); |
| 868 | printk("WS_COUNT_MUTEX : %08X\n", WS_COUNT_MUTEX); |
| 869 | printk("WS_LOCK_MUTEX : %08X\n", WS_LOCK_MUTEX); |
| 870 | printk("CTX_RMASK : %016lX\n", CTX_RMASK); |
| 871 | printk("CTX_WMASK : %016lX\n", CTX_WMASK); |
| 872 | |
| 873 | return 0; |
| 874 | } |
| 875 | |
| 876 | void cleanup_module(void) |
| 877 | { |
| 878 | remove_proc_entry("testpsrwlock", NULL); |
| 879 | } |
| 880 | |
| 881 | MODULE_LICENSE("GPL"); |
| 882 | MODULE_AUTHOR("Mathieu Desnoyers"); |
| 883 | MODULE_DESCRIPTION("psrwlock test"); |