at v2.6.35 520 lines 17 kB view raw
1/* 2 * Read-Copy Update mechanism for mutual exclusion 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU General Public License as published by 6 * the Free Software Foundation; either version 2 of the License, or 7 * (at your option) any later version. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write to the Free Software 16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. 17 * 18 * Copyright IBM Corporation, 2001 19 * 20 * Author: Dipankar Sarma <dipankar@in.ibm.com> 21 * 22 * Based on the original work by Paul McKenney <paulmck@us.ibm.com> 23 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. 24 * Papers: 25 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf 26 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001) 27 * 28 * For detailed explanation of Read-Copy Update mechanism see - 29 * http://lse.sourceforge.net/locking/rcupdate.html 30 * 31 */ 32 33#ifndef __LINUX_RCUPDATE_H 34#define __LINUX_RCUPDATE_H 35 36#include <linux/cache.h> 37#include <linux/spinlock.h> 38#include <linux/threads.h> 39#include <linux/cpumask.h> 40#include <linux/seqlock.h> 41#include <linux/lockdep.h> 42#include <linux/completion.h> 43 44#ifdef CONFIG_RCU_TORTURE_TEST 45extern int rcutorture_runnable; /* for sysctl */ 46#endif /* #ifdef CONFIG_RCU_TORTURE_TEST */ 47 48/** 49 * struct rcu_head - callback structure for use with RCU 50 * @next: next update requests in a list 51 * @func: actual update function to call after the grace period. 52 */ 53struct rcu_head { 54 struct rcu_head *next; 55 void (*func)(struct rcu_head *head); 56}; 57 58/* Exported common interfaces */ 59extern void rcu_barrier(void); 60extern void rcu_barrier_bh(void); 61extern void rcu_barrier_sched(void); 62extern void synchronize_sched_expedited(void); 63extern int sched_expedited_torture_stats(char *page); 64 65/* Internal to kernel */ 66extern void rcu_init(void); 67 68#if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) 69#include <linux/rcutree.h> 70#elif defined(CONFIG_TINY_RCU) 71#include <linux/rcutiny.h> 72#else 73#error "Unknown RCU implementation specified to kernel configuration" 74#endif 75 76#define RCU_HEAD_INIT { .next = NULL, .func = NULL } 77#define RCU_HEAD(head) struct rcu_head head = RCU_HEAD_INIT 78#define INIT_RCU_HEAD(ptr) do { \ 79 (ptr)->next = NULL; (ptr)->func = NULL; \ 80} while (0) 81 82static inline void init_rcu_head_on_stack(struct rcu_head *head) 83{ 84} 85 86static inline void destroy_rcu_head_on_stack(struct rcu_head *head) 87{ 88} 89 90#ifdef CONFIG_DEBUG_LOCK_ALLOC 91 92extern struct lockdep_map rcu_lock_map; 93# define rcu_read_acquire() \ 94 lock_acquire(&rcu_lock_map, 0, 0, 2, 1, NULL, _THIS_IP_) 95# define rcu_read_release() lock_release(&rcu_lock_map, 1, _THIS_IP_) 96 97extern struct lockdep_map rcu_bh_lock_map; 98# define rcu_read_acquire_bh() \ 99 lock_acquire(&rcu_bh_lock_map, 0, 0, 2, 1, NULL, _THIS_IP_) 100# define rcu_read_release_bh() lock_release(&rcu_bh_lock_map, 1, _THIS_IP_) 101 102extern struct lockdep_map rcu_sched_lock_map; 103# define rcu_read_acquire_sched() \ 104 lock_acquire(&rcu_sched_lock_map, 0, 0, 2, 1, NULL, _THIS_IP_) 105# define rcu_read_release_sched() \ 106 lock_release(&rcu_sched_lock_map, 1, _THIS_IP_) 107 108extern int debug_lockdep_rcu_enabled(void); 109 110/** 111 * rcu_read_lock_held - might we be in RCU read-side critical section? 112 * 113 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU 114 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC, 115 * this assumes we are in an RCU read-side critical section unless it can 116 * prove otherwise. 117 * 118 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot 119 * and while lockdep is disabled. 120 */ 121static inline int rcu_read_lock_held(void) 122{ 123 if (!debug_lockdep_rcu_enabled()) 124 return 1; 125 return lock_is_held(&rcu_lock_map); 126} 127 128/* 129 * rcu_read_lock_bh_held() is defined out of line to avoid #include-file 130 * hell. 131 */ 132extern int rcu_read_lock_bh_held(void); 133 134/** 135 * rcu_read_lock_sched_held - might we be in RCU-sched read-side critical section? 136 * 137 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an 138 * RCU-sched read-side critical section. In absence of 139 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side 140 * critical section unless it can prove otherwise. Note that disabling 141 * of preemption (including disabling irqs) counts as an RCU-sched 142 * read-side critical section. 143 * 144 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot 145 * and while lockdep is disabled. 146 */ 147#ifdef CONFIG_PREEMPT 148static inline int rcu_read_lock_sched_held(void) 149{ 150 int lockdep_opinion = 0; 151 152 if (!debug_lockdep_rcu_enabled()) 153 return 1; 154 if (debug_locks) 155 lockdep_opinion = lock_is_held(&rcu_sched_lock_map); 156 return lockdep_opinion || preempt_count() != 0 || irqs_disabled(); 157} 158#else /* #ifdef CONFIG_PREEMPT */ 159static inline int rcu_read_lock_sched_held(void) 160{ 161 return 1; 162} 163#endif /* #else #ifdef CONFIG_PREEMPT */ 164 165#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 166 167# define rcu_read_acquire() do { } while (0) 168# define rcu_read_release() do { } while (0) 169# define rcu_read_acquire_bh() do { } while (0) 170# define rcu_read_release_bh() do { } while (0) 171# define rcu_read_acquire_sched() do { } while (0) 172# define rcu_read_release_sched() do { } while (0) 173 174static inline int rcu_read_lock_held(void) 175{ 176 return 1; 177} 178 179static inline int rcu_read_lock_bh_held(void) 180{ 181 return 1; 182} 183 184#ifdef CONFIG_PREEMPT 185static inline int rcu_read_lock_sched_held(void) 186{ 187 return preempt_count() != 0 || irqs_disabled(); 188} 189#else /* #ifdef CONFIG_PREEMPT */ 190static inline int rcu_read_lock_sched_held(void) 191{ 192 return 1; 193} 194#endif /* #else #ifdef CONFIG_PREEMPT */ 195 196#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 197 198#ifdef CONFIG_PROVE_RCU 199 200extern int rcu_my_thread_group_empty(void); 201 202#define __do_rcu_dereference_check(c) \ 203 do { \ 204 static bool __warned; \ 205 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \ 206 __warned = true; \ 207 lockdep_rcu_dereference(__FILE__, __LINE__); \ 208 } \ 209 } while (0) 210 211/** 212 * rcu_dereference_check - rcu_dereference with debug checking 213 * @p: The pointer to read, prior to dereferencing 214 * @c: The conditions under which the dereference will take place 215 * 216 * Do an rcu_dereference(), but check that the conditions under which the 217 * dereference will take place are correct. Typically the conditions indicate 218 * the various locking conditions that should be held at that point. The check 219 * should return true if the conditions are satisfied. 220 * 221 * For example: 222 * 223 * bar = rcu_dereference_check(foo->bar, rcu_read_lock_held() || 224 * lockdep_is_held(&foo->lock)); 225 * 226 * could be used to indicate to lockdep that foo->bar may only be dereferenced 227 * if either the RCU read lock is held, or that the lock required to replace 228 * the bar struct at foo->bar is held. 229 * 230 * Note that the list of conditions may also include indications of when a lock 231 * need not be held, for example during initialisation or destruction of the 232 * target struct: 233 * 234 * bar = rcu_dereference_check(foo->bar, rcu_read_lock_held() || 235 * lockdep_is_held(&foo->lock) || 236 * atomic_read(&foo->usage) == 0); 237 */ 238#define rcu_dereference_check(p, c) \ 239 ({ \ 240 __do_rcu_dereference_check(c); \ 241 rcu_dereference_raw(p); \ 242 }) 243 244/** 245 * rcu_dereference_protected - fetch RCU pointer when updates prevented 246 * 247 * Return the value of the specified RCU-protected pointer, but omit 248 * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This 249 * is useful in cases where update-side locks prevent the value of the 250 * pointer from changing. Please note that this primitive does -not- 251 * prevent the compiler from repeating this reference or combining it 252 * with other references, so it should not be used without protection 253 * of appropriate locks. 254 */ 255#define rcu_dereference_protected(p, c) \ 256 ({ \ 257 __do_rcu_dereference_check(c); \ 258 (p); \ 259 }) 260 261#else /* #ifdef CONFIG_PROVE_RCU */ 262 263#define rcu_dereference_check(p, c) rcu_dereference_raw(p) 264#define rcu_dereference_protected(p, c) (p) 265 266#endif /* #else #ifdef CONFIG_PROVE_RCU */ 267 268/** 269 * rcu_access_pointer - fetch RCU pointer with no dereferencing 270 * 271 * Return the value of the specified RCU-protected pointer, but omit the 272 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful 273 * when the value of this pointer is accessed, but the pointer is not 274 * dereferenced, for example, when testing an RCU-protected pointer against 275 * NULL. This may also be used in cases where update-side locks prevent 276 * the value of the pointer from changing, but rcu_dereference_protected() 277 * is a lighter-weight primitive for this use case. 278 */ 279#define rcu_access_pointer(p) ACCESS_ONCE(p) 280 281/** 282 * rcu_read_lock - mark the beginning of an RCU read-side critical section. 283 * 284 * When synchronize_rcu() is invoked on one CPU while other CPUs 285 * are within RCU read-side critical sections, then the 286 * synchronize_rcu() is guaranteed to block until after all the other 287 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 288 * on one CPU while other CPUs are within RCU read-side critical 289 * sections, invocation of the corresponding RCU callback is deferred 290 * until after the all the other CPUs exit their critical sections. 291 * 292 * Note, however, that RCU callbacks are permitted to run concurrently 293 * with RCU read-side critical sections. One way that this can happen 294 * is via the following sequence of events: (1) CPU 0 enters an RCU 295 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 296 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 297 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 298 * callback is invoked. This is legal, because the RCU read-side critical 299 * section that was running concurrently with the call_rcu() (and which 300 * therefore might be referencing something that the corresponding RCU 301 * callback would free up) has completed before the corresponding 302 * RCU callback is invoked. 303 * 304 * RCU read-side critical sections may be nested. Any deferred actions 305 * will be deferred until the outermost RCU read-side critical section 306 * completes. 307 * 308 * It is illegal to block while in an RCU read-side critical section. 309 */ 310static inline void rcu_read_lock(void) 311{ 312 __rcu_read_lock(); 313 __acquire(RCU); 314 rcu_read_acquire(); 315} 316 317/* 318 * So where is rcu_write_lock()? It does not exist, as there is no 319 * way for writers to lock out RCU readers. This is a feature, not 320 * a bug -- this property is what provides RCU's performance benefits. 321 * Of course, writers must coordinate with each other. The normal 322 * spinlock primitives work well for this, but any other technique may be 323 * used as well. RCU does not care how the writers keep out of each 324 * others' way, as long as they do so. 325 */ 326 327/** 328 * rcu_read_unlock - marks the end of an RCU read-side critical section. 329 * 330 * See rcu_read_lock() for more information. 331 */ 332static inline void rcu_read_unlock(void) 333{ 334 rcu_read_release(); 335 __release(RCU); 336 __rcu_read_unlock(); 337} 338 339/** 340 * rcu_read_lock_bh - mark the beginning of a softirq-only RCU critical section 341 * 342 * This is equivalent of rcu_read_lock(), but to be used when updates 343 * are being done using call_rcu_bh(). Since call_rcu_bh() callbacks 344 * consider completion of a softirq handler to be a quiescent state, 345 * a process in RCU read-side critical section must be protected by 346 * disabling softirqs. Read-side critical sections in interrupt context 347 * can use just rcu_read_lock(). 348 * 349 */ 350static inline void rcu_read_lock_bh(void) 351{ 352 __rcu_read_lock_bh(); 353 __acquire(RCU_BH); 354 rcu_read_acquire_bh(); 355} 356 357/* 358 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section 359 * 360 * See rcu_read_lock_bh() for more information. 361 */ 362static inline void rcu_read_unlock_bh(void) 363{ 364 rcu_read_release_bh(); 365 __release(RCU_BH); 366 __rcu_read_unlock_bh(); 367} 368 369/** 370 * rcu_read_lock_sched - mark the beginning of a RCU-classic critical section 371 * 372 * Should be used with either 373 * - synchronize_sched() 374 * or 375 * - call_rcu_sched() and rcu_barrier_sched() 376 * on the write-side to insure proper synchronization. 377 */ 378static inline void rcu_read_lock_sched(void) 379{ 380 preempt_disable(); 381 __acquire(RCU_SCHED); 382 rcu_read_acquire_sched(); 383} 384 385/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 386static inline notrace void rcu_read_lock_sched_notrace(void) 387{ 388 preempt_disable_notrace(); 389 __acquire(RCU_SCHED); 390} 391 392/* 393 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section 394 * 395 * See rcu_read_lock_sched for more information. 396 */ 397static inline void rcu_read_unlock_sched(void) 398{ 399 rcu_read_release_sched(); 400 __release(RCU_SCHED); 401 preempt_enable(); 402} 403 404/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 405static inline notrace void rcu_read_unlock_sched_notrace(void) 406{ 407 __release(RCU_SCHED); 408 preempt_enable_notrace(); 409} 410 411 412/** 413 * rcu_dereference_raw - fetch an RCU-protected pointer 414 * 415 * The caller must be within some flavor of RCU read-side critical 416 * section, or must be otherwise preventing the pointer from changing, 417 * for example, by holding an appropriate lock. This pointer may later 418 * be safely dereferenced. It is the caller's responsibility to have 419 * done the right thing, as this primitive does no checking of any kind. 420 * 421 * Inserts memory barriers on architectures that require them 422 * (currently only the Alpha), and, more importantly, documents 423 * exactly which pointers are protected by RCU. 424 */ 425#define rcu_dereference_raw(p) ({ \ 426 typeof(p) _________p1 = ACCESS_ONCE(p); \ 427 smp_read_barrier_depends(); \ 428 (_________p1); \ 429 }) 430 431/** 432 * rcu_dereference - fetch an RCU-protected pointer, checking for RCU 433 * 434 * Makes rcu_dereference_check() do the dirty work. 435 */ 436#define rcu_dereference(p) \ 437 rcu_dereference_check(p, rcu_read_lock_held()) 438 439/** 440 * rcu_dereference_bh - fetch an RCU-protected pointer, checking for RCU-bh 441 * 442 * Makes rcu_dereference_check() do the dirty work. 443 */ 444#define rcu_dereference_bh(p) \ 445 rcu_dereference_check(p, rcu_read_lock_bh_held()) 446 447/** 448 * rcu_dereference_sched - fetch RCU-protected pointer, checking for RCU-sched 449 * 450 * Makes rcu_dereference_check() do the dirty work. 451 */ 452#define rcu_dereference_sched(p) \ 453 rcu_dereference_check(p, rcu_read_lock_sched_held()) 454 455/** 456 * rcu_assign_pointer - assign (publicize) a pointer to a newly 457 * initialized structure that will be dereferenced by RCU read-side 458 * critical sections. Returns the value assigned. 459 * 460 * Inserts memory barriers on architectures that require them 461 * (pretty much all of them other than x86), and also prevents 462 * the compiler from reordering the code that initializes the 463 * structure after the pointer assignment. More importantly, this 464 * call documents which pointers will be dereferenced by RCU read-side 465 * code. 466 */ 467 468#define rcu_assign_pointer(p, v) \ 469 ({ \ 470 if (!__builtin_constant_p(v) || \ 471 ((v) != NULL)) \ 472 smp_wmb(); \ 473 (p) = (v); \ 474 }) 475 476/* Infrastructure to implement the synchronize_() primitives. */ 477 478struct rcu_synchronize { 479 struct rcu_head head; 480 struct completion completion; 481}; 482 483extern void wakeme_after_rcu(struct rcu_head *head); 484 485/** 486 * call_rcu - Queue an RCU callback for invocation after a grace period. 487 * @head: structure to be used for queueing the RCU updates. 488 * @func: actual update function to be invoked after the grace period 489 * 490 * The update function will be invoked some time after a full grace 491 * period elapses, in other words after all currently executing RCU 492 * read-side critical sections have completed. RCU read-side critical 493 * sections are delimited by rcu_read_lock() and rcu_read_unlock(), 494 * and may be nested. 495 */ 496extern void call_rcu(struct rcu_head *head, 497 void (*func)(struct rcu_head *head)); 498 499/** 500 * call_rcu_bh - Queue an RCU for invocation after a quicker grace period. 501 * @head: structure to be used for queueing the RCU updates. 502 * @func: actual update function to be invoked after the grace period 503 * 504 * The update function will be invoked some time after a full grace 505 * period elapses, in other words after all currently executing RCU 506 * read-side critical sections have completed. call_rcu_bh() assumes 507 * that the read-side critical sections end on completion of a softirq 508 * handler. This means that read-side critical sections in process 509 * context must not be interrupted by softirqs. This interface is to be 510 * used when most of the read-side critical sections are in softirq context. 511 * RCU read-side critical sections are delimited by : 512 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context. 513 * OR 514 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context. 515 * These may be nested. 516 */ 517extern void call_rcu_bh(struct rcu_head *head, 518 void (*func)(struct rcu_head *head)); 519 520#endif /* __LINUX_RCUPDATE_H */