kern_rmlock.c revision 215399
1/*- 2 * Copyright (c) 2007 Stephan Uphoff <ups@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the author nor the names of any co-contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30/* 31 * Machine independent bits of reader/writer lock implementation. 32 */ 33 34#include <sys/cdefs.h> 35__FBSDID("$FreeBSD: head/sys/kern/kern_rmlock.c 215399 2010-11-16 14:08:21Z cognet $"); 36 37#include "opt_ddb.h" 38#include "opt_kdtrace.h" 39 40#include <sys/param.h> 41#include <sys/systm.h> 42 43#include <sys/kernel.h> 44#include <sys/ktr.h> 45#include <sys/lock.h> 46#include <sys/mutex.h> 47#include <sys/proc.h> 48#include <sys/rmlock.h> 49#include <sys/sched.h> 50#include <sys/smp.h> 51#include <sys/turnstile.h> 52#include <sys/lock_profile.h> 53#include <machine/cpu.h> 54 55#ifdef DDB 56#include <ddb/ddb.h> 57#endif 58 59#define RMPF_ONQUEUE 1 60#define RMPF_SIGNAL 2 61 62/* 63 * To support usage of rmlock in CVs and msleep yet another list for the 64 * priority tracker would be needed. Using this lock for cv and msleep also 65 * does not seem very useful 66 */ 67 68static __inline void compiler_memory_barrier(void) { 69 __asm __volatile("":::"memory"); 70} 71 72static void assert_rm(struct lock_object *lock, int what); 73static void lock_rm(struct lock_object *lock, int how); 74#ifdef KDTRACE_HOOKS 75static int owner_rm(struct lock_object *lock, struct thread **owner); 76#endif 77static int unlock_rm(struct lock_object *lock); 78 79struct lock_class lock_class_rm = { 80 .lc_name = "rm", 81 .lc_flags = LC_SLEEPLOCK | LC_RECURSABLE, 82 .lc_assert = assert_rm, 83#if 0 84#ifdef DDB 85 .lc_ddb_show = db_show_rwlock, 86#endif 87#endif 88 .lc_lock = lock_rm, 89 .lc_unlock = unlock_rm, 90#ifdef KDTRACE_HOOKS 91 .lc_owner = owner_rm, 92#endif 93}; 94 95static void 96assert_rm(struct lock_object *lock, int what) 97{ 98 99 panic("assert_rm called"); 100} 101 102static void 103lock_rm(struct lock_object *lock, int how) 104{ 105 106 panic("lock_rm called"); 107} 108 109static int 110unlock_rm(struct lock_object *lock) 111{ 112 113 panic("unlock_rm called"); 114} 115 116#ifdef KDTRACE_HOOKS 117static int 118owner_rm(struct lock_object *lock, struct thread **owner) 119{ 120 121 panic("owner_rm called"); 122} 123#endif 124 125static struct mtx rm_spinlock; 126 127MTX_SYSINIT(rm_spinlock, &rm_spinlock, "rm_spinlock", MTX_SPIN); 128 129/* 130 * Add or remove tracker from per-cpu list. 131 * 132 * The per-cpu list can be traversed at any time in forward direction from an 133 * interrupt on the *local* cpu. 134 */ 135static void inline 136rm_tracker_add(struct pcpu *pc, struct rm_priotracker *tracker) 137{ 138 struct rm_queue *next; 139 140 /* Initialize all tracker pointers */ 141 tracker->rmp_cpuQueue.rmq_prev = &pc->pc_rm_queue; 142 next = pc->pc_rm_queue.rmq_next; 143 tracker->rmp_cpuQueue.rmq_next = next; 144 145 /* rmq_prev is not used during froward traversal. */ 146 next->rmq_prev = &tracker->rmp_cpuQueue; 147 148 /* Update pointer to first element. */ 149 pc->pc_rm_queue.rmq_next = &tracker->rmp_cpuQueue; 150} 151 152static void inline 153rm_tracker_remove(struct pcpu *pc, struct rm_priotracker *tracker) 154{ 155 struct rm_queue *next, *prev; 156 157 next = tracker->rmp_cpuQueue.rmq_next; 158 prev = tracker->rmp_cpuQueue.rmq_prev; 159 160 /* Not used during forward traversal. */ 161 next->rmq_prev = prev; 162 163 /* Remove from list. */ 164 prev->rmq_next = next; 165} 166 167static void 168rm_cleanIPI(void *arg) 169{ 170 struct pcpu *pc; 171 struct rmlock *rm = arg; 172 struct rm_priotracker *tracker; 173 struct rm_queue *queue; 174 pc = pcpu_find(curcpu); 175 176 for (queue = pc->pc_rm_queue.rmq_next; queue != &pc->pc_rm_queue; 177 queue = queue->rmq_next) { 178 tracker = (struct rm_priotracker *)queue; 179 if (tracker->rmp_rmlock == rm && tracker->rmp_flags == 0) { 180 tracker->rmp_flags = RMPF_ONQUEUE; 181 mtx_lock_spin(&rm_spinlock); 182 LIST_INSERT_HEAD(&rm->rm_activeReaders, tracker, 183 rmp_qentry); 184 mtx_unlock_spin(&rm_spinlock); 185 } 186 } 187} 188 189CTASSERT((RM_SLEEPABLE & LO_CLASSFLAGS) == RM_SLEEPABLE); 190 191void 192rm_init_flags(struct rmlock *rm, const char *name, int opts) 193{ 194 int liflags; 195 196 liflags = 0; 197 if (!(opts & RM_NOWITNESS)) 198 liflags |= LO_WITNESS; 199 if (opts & RM_RECURSE) 200 liflags |= LO_RECURSABLE; 201 rm->rm_writecpus = all_cpus; 202 LIST_INIT(&rm->rm_activeReaders); 203 if (opts & RM_SLEEPABLE) { 204 liflags |= RM_SLEEPABLE; 205 sx_init_flags(&rm->rm_lock_sx, "rmlock_sx", SX_RECURSE); 206 } else 207 mtx_init(&rm->rm_lock_mtx, name, "rmlock_mtx", MTX_NOWITNESS); 208 lock_init(&rm->lock_object, &lock_class_rm, name, NULL, liflags); 209} 210 211void 212rm_init(struct rmlock *rm, const char *name) 213{ 214 215 rm_init_flags(rm, name, 0); 216} 217 218void 219rm_destroy(struct rmlock *rm) 220{ 221 222 if (rm->lock_object.lo_flags & RM_SLEEPABLE) 223 sx_destroy(&rm->rm_lock_sx); 224 else 225 mtx_destroy(&rm->rm_lock_mtx); 226 lock_destroy(&rm->lock_object); 227} 228 229int 230rm_wowned(struct rmlock *rm) 231{ 232 233 if (rm->lock_object.lo_flags & RM_SLEEPABLE) 234 return (sx_xlocked(&rm->rm_lock_sx)); 235 else 236 return (mtx_owned(&rm->rm_lock_mtx)); 237} 238 239void 240rm_sysinit(void *arg) 241{ 242 struct rm_args *args = arg; 243 244 rm_init(args->ra_rm, args->ra_desc); 245} 246 247void 248rm_sysinit_flags(void *arg) 249{ 250 struct rm_args_flags *args = arg; 251 252 rm_init_flags(args->ra_rm, args->ra_desc, args->ra_opts); 253} 254 255static int 256_rm_rlock_hard(struct rmlock *rm, struct rm_priotracker *tracker, int trylock) 257{ 258 struct pcpu *pc; 259 struct rm_queue *queue; 260 struct rm_priotracker *atracker; 261 262 critical_enter(); 263 pc = pcpu_find(curcpu); 264 265 /* Check if we just need to do a proper critical_exit. */ 266 if (!(pc->pc_cpumask & rm->rm_writecpus)) { 267 critical_exit(); 268 return (1); 269 } 270 271 /* Remove our tracker from the per-cpu list. */ 272 rm_tracker_remove(pc, tracker); 273 274 /* Check to see if the IPI granted us the lock after all. */ 275 if (tracker->rmp_flags) { 276 /* Just add back tracker - we hold the lock. */ 277 rm_tracker_add(pc, tracker); 278 critical_exit(); 279 return (1); 280 } 281 282 /* 283 * We allow readers to aquire a lock even if a writer is blocked if 284 * the lock is recursive and the reader already holds the lock. 285 */ 286 if ((rm->lock_object.lo_flags & LO_RECURSABLE) != 0) { 287 /* 288 * Just grant the lock if this thread already has a tracker 289 * for this lock on the per-cpu queue. 290 */ 291 for (queue = pc->pc_rm_queue.rmq_next; 292 queue != &pc->pc_rm_queue; queue = queue->rmq_next) { 293 atracker = (struct rm_priotracker *)queue; 294 if ((atracker->rmp_rmlock == rm) && 295 (atracker->rmp_thread == tracker->rmp_thread)) { 296 mtx_lock_spin(&rm_spinlock); 297 LIST_INSERT_HEAD(&rm->rm_activeReaders, 298 tracker, rmp_qentry); 299 tracker->rmp_flags = RMPF_ONQUEUE; 300 mtx_unlock_spin(&rm_spinlock); 301 rm_tracker_add(pc, tracker); 302 critical_exit(); 303 return (1); 304 } 305 } 306 } 307 308 sched_unpin(); 309 critical_exit(); 310 311 if (trylock) { 312 if (rm->lock_object.lo_flags & RM_SLEEPABLE) { 313 if (!sx_try_xlock(&rm->rm_lock_sx)) 314 return (0); 315 } else { 316 if (!mtx_trylock(&rm->rm_lock_mtx)) 317 return (0); 318 } 319 } else { 320 if (rm->lock_object.lo_flags & RM_SLEEPABLE) 321 sx_xlock(&rm->rm_lock_sx); 322 else 323 mtx_lock(&rm->rm_lock_mtx); 324 } 325 326 critical_enter(); 327 pc = pcpu_find(curcpu); 328 rm->rm_writecpus &= ~pc->pc_cpumask; 329 rm_tracker_add(pc, tracker); 330 sched_pin(); 331 critical_exit(); 332 333 if (rm->lock_object.lo_flags & RM_SLEEPABLE) 334 sx_xunlock(&rm->rm_lock_sx); 335 else 336 mtx_unlock(&rm->rm_lock_mtx); 337 338 return (1); 339} 340 341int 342_rm_rlock(struct rmlock *rm, struct rm_priotracker *tracker, int trylock) 343{ 344 struct thread *td = curthread; 345 struct pcpu *pc; 346 347 tracker->rmp_flags = 0; 348 tracker->rmp_thread = td; 349 tracker->rmp_rmlock = rm; 350 351 td->td_critnest++; /* critical_enter(); */ 352 353 compiler_memory_barrier(); 354 355 pc = cpuid_to_pcpu[td->td_oncpu]; /* pcpu_find(td->td_oncpu); */ 356 357 rm_tracker_add(pc, tracker); 358 359 sched_pin(); 360 361 compiler_memory_barrier(); 362 363 td->td_critnest--; 364 365 /* 366 * Fast path to combine two common conditions into a single 367 * conditional jump. 368 */ 369 if (0 == (td->td_owepreempt | (rm->rm_writecpus & pc->pc_cpumask))) 370 return (1); 371 372 /* We do not have a read token and need to acquire one. */ 373 return _rm_rlock_hard(rm, tracker, trylock); 374} 375 376static void 377_rm_unlock_hard(struct thread *td,struct rm_priotracker *tracker) 378{ 379 380 if (td->td_owepreempt) { 381 td->td_critnest++; 382 critical_exit(); 383 } 384 385 if (!tracker->rmp_flags) 386 return; 387 388 mtx_lock_spin(&rm_spinlock); 389 LIST_REMOVE(tracker, rmp_qentry); 390 391 if (tracker->rmp_flags & RMPF_SIGNAL) { 392 struct rmlock *rm; 393 struct turnstile *ts; 394 395 rm = tracker->rmp_rmlock; 396 397 turnstile_chain_lock(&rm->lock_object); 398 mtx_unlock_spin(&rm_spinlock); 399 400 ts = turnstile_lookup(&rm->lock_object); 401 402 turnstile_signal(ts, TS_EXCLUSIVE_QUEUE); 403 turnstile_unpend(ts, TS_EXCLUSIVE_LOCK); 404 turnstile_chain_unlock(&rm->lock_object); 405 } else 406 mtx_unlock_spin(&rm_spinlock); 407} 408 409void 410_rm_runlock(struct rmlock *rm, struct rm_priotracker *tracker) 411{ 412 struct pcpu *pc; 413 struct thread *td = tracker->rmp_thread; 414 415 td->td_critnest++; /* critical_enter(); */ 416 pc = cpuid_to_pcpu[td->td_oncpu]; /* pcpu_find(td->td_oncpu); */ 417 rm_tracker_remove(pc, tracker); 418 td->td_critnest--; 419 sched_unpin(); 420 421 if (0 == (td->td_owepreempt | tracker->rmp_flags)) 422 return; 423 424 _rm_unlock_hard(td, tracker); 425} 426 427void 428_rm_wlock(struct rmlock *rm) 429{ 430 struct rm_priotracker *prio; 431 struct turnstile *ts; 432 cpumask_t readcpus; 433 434 if (rm->lock_object.lo_flags & RM_SLEEPABLE) 435 sx_xlock(&rm->rm_lock_sx); 436 else 437 mtx_lock(&rm->rm_lock_mtx); 438 439 if (rm->rm_writecpus != all_cpus) { 440 /* Get all read tokens back */ 441 442 readcpus = all_cpus & (all_cpus & ~rm->rm_writecpus); 443 rm->rm_writecpus = all_cpus; 444 445 /* 446 * Assumes rm->rm_writecpus update is visible on other CPUs 447 * before rm_cleanIPI is called. 448 */ 449#ifdef SMP 450 smp_rendezvous_cpus(readcpus, 451 smp_no_rendevous_barrier, 452 rm_cleanIPI, 453 smp_no_rendevous_barrier, 454 rm); 455 456#else 457 rm_cleanIPI(rm); 458#endif 459 460 mtx_lock_spin(&rm_spinlock); 461 while ((prio = LIST_FIRST(&rm->rm_activeReaders)) != NULL) { 462 ts = turnstile_trywait(&rm->lock_object); 463 prio->rmp_flags = RMPF_ONQUEUE | RMPF_SIGNAL; 464 mtx_unlock_spin(&rm_spinlock); 465 turnstile_wait(ts, prio->rmp_thread, 466 TS_EXCLUSIVE_QUEUE); 467 mtx_lock_spin(&rm_spinlock); 468 } 469 mtx_unlock_spin(&rm_spinlock); 470 } 471} 472 473void 474_rm_wunlock(struct rmlock *rm) 475{ 476 477 if (rm->lock_object.lo_flags & RM_SLEEPABLE) 478 sx_xunlock(&rm->rm_lock_sx); 479 else 480 mtx_unlock(&rm->rm_lock_mtx); 481} 482 483#ifdef LOCK_DEBUG 484 485void _rm_wlock_debug(struct rmlock *rm, const char *file, int line) 486{ 487 488 WITNESS_CHECKORDER(&rm->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, 489 file, line, NULL); 490 491 _rm_wlock(rm); 492 493 LOCK_LOG_LOCK("RMWLOCK", &rm->lock_object, 0, 0, file, line); 494 495 if (rm->lock_object.lo_flags & RM_SLEEPABLE) 496 WITNESS_LOCK(&rm->rm_lock_sx.lock_object, LOP_EXCLUSIVE, 497 file, line); 498 else 499 WITNESS_LOCK(&rm->lock_object, LOP_EXCLUSIVE, file, line); 500 501 curthread->td_locks++; 502 503} 504 505void 506_rm_wunlock_debug(struct rmlock *rm, const char *file, int line) 507{ 508 509 curthread->td_locks--; 510 if (rm->lock_object.lo_flags & RM_SLEEPABLE) 511 WITNESS_UNLOCK(&rm->rm_lock_sx.lock_object, LOP_EXCLUSIVE, 512 file, line); 513 else 514 WITNESS_UNLOCK(&rm->lock_object, LOP_EXCLUSIVE, file, line); 515 LOCK_LOG_LOCK("RMWUNLOCK", &rm->lock_object, 0, 0, file, line); 516 _rm_wunlock(rm); 517} 518 519int 520_rm_rlock_debug(struct rmlock *rm, struct rm_priotracker *tracker, 521 int trylock, const char *file, int line) 522{ 523 if (!trylock && (rm->lock_object.lo_flags & RM_SLEEPABLE)) 524 WITNESS_CHECKORDER(&rm->rm_lock_sx.lock_object, LOP_NEWORDER, 525 file, line, NULL); 526 WITNESS_CHECKORDER(&rm->lock_object, LOP_NEWORDER, file, line, NULL); 527 528 if (_rm_rlock(rm, tracker, trylock)) { 529 LOCK_LOG_LOCK("RMRLOCK", &rm->lock_object, 0, 0, file, line); 530 531 WITNESS_LOCK(&rm->lock_object, 0, file, line); 532 533 curthread->td_locks++; 534 535 return (1); 536 } 537 538 return (0); 539} 540 541void 542_rm_runlock_debug(struct rmlock *rm, struct rm_priotracker *tracker, 543 const char *file, int line) 544{ 545 546 curthread->td_locks--; 547 WITNESS_UNLOCK(&rm->lock_object, 0, file, line); 548 LOCK_LOG_LOCK("RMRUNLOCK", &rm->lock_object, 0, 0, file, line); 549 _rm_runlock(rm, tracker); 550} 551 552#else 553 554/* 555 * Just strip out file and line arguments if no lock debugging is enabled in 556 * the kernel - we are called from a kernel module. 557 */ 558void 559_rm_wlock_debug(struct rmlock *rm, const char *file, int line) 560{ 561 562 _rm_wlock(rm); 563} 564 565void 566_rm_wunlock_debug(struct rmlock *rm, const char *file, int line) 567{ 568 569 _rm_wunlock(rm); 570} 571 572int 573_rm_rlock_debug(struct rmlock *rm, struct rm_priotracker *tracker, 574 int trylock, const char *file, int line) 575{ 576 577 return _rm_rlock(rm, tracker, trylock); 578} 579 580void 581_rm_runlock_debug(struct rmlock *rm, struct rm_priotracker *tracker, 582 const char *file, int line) 583{ 584 585 _rm_runlock(rm, tracker); 586} 587 588#endif 589