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/types.h> 37#include <linux/cache.h> 38#include <linux/spinlock.h> 39#include <linux/threads.h> 40#include <linux/cpumask.h> 41#include <linux/seqlock.h> 42#include <linux/lockdep.h> 43#include <linux/completion.h> 44#include <linux/debugobjects.h> 45#include <linux/bug.h> 46#include <linux/compiler.h> 47 48#ifdef CONFIG_RCU_TORTURE_TEST 49extern int rcutorture_runnable; /* for sysctl */ 50#endif /* #ifdef CONFIG_RCU_TORTURE_TEST */ 51 52#if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) 53extern void rcutorture_record_test_transition(void); 54extern void rcutorture_record_progress(unsigned long vernum); 55extern void do_trace_rcu_torture_read(const char *rcutorturename, 56 struct rcu_head *rhp, 57 unsigned long secs, 58 unsigned long c_old, 59 unsigned long c); 60#else 61static inline void rcutorture_record_test_transition(void) 62{ 63} 64static inline void rcutorture_record_progress(unsigned long vernum) 65{ 66} 67#ifdef CONFIG_RCU_TRACE 68extern void do_trace_rcu_torture_read(const char *rcutorturename, 69 struct rcu_head *rhp, 70 unsigned long secs, 71 unsigned long c_old, 72 unsigned long c); 73#else 74#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ 75 do { } while (0) 76#endif 77#endif 78 79#define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b)) 80#define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b)) 81#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) 82#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) 83#define ulong2long(a) (*(long *)(&(a))) 84 85/* Exported common interfaces */ 86 87#ifdef CONFIG_PREEMPT_RCU 88 89/** 90 * call_rcu() - Queue an RCU callback for invocation after a grace period. 91 * @head: structure to be used for queueing the RCU updates. 92 * @func: actual callback function to be invoked after the grace period 93 * 94 * The callback function will be invoked some time after a full grace 95 * period elapses, in other words after all pre-existing RCU read-side 96 * critical sections have completed. However, the callback function 97 * might well execute concurrently with RCU read-side critical sections 98 * that started after call_rcu() was invoked. RCU read-side critical 99 * sections are delimited by rcu_read_lock() and rcu_read_unlock(), 100 * and may be nested. 101 * 102 * Note that all CPUs must agree that the grace period extended beyond 103 * all pre-existing RCU read-side critical section. On systems with more 104 * than one CPU, this means that when "func()" is invoked, each CPU is 105 * guaranteed to have executed a full memory barrier since the end of its 106 * last RCU read-side critical section whose beginning preceded the call 107 * to call_rcu(). It also means that each CPU executing an RCU read-side 108 * critical section that continues beyond the start of "func()" must have 109 * executed a memory barrier after the call_rcu() but before the beginning 110 * of that RCU read-side critical section. Note that these guarantees 111 * include CPUs that are offline, idle, or executing in user mode, as 112 * well as CPUs that are executing in the kernel. 113 * 114 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the 115 * resulting RCU callback function "func()", then both CPU A and CPU B are 116 * guaranteed to execute a full memory barrier during the time interval 117 * between the call to call_rcu() and the invocation of "func()" -- even 118 * if CPU A and CPU B are the same CPU (but again only if the system has 119 * more than one CPU). 120 */ 121extern void call_rcu(struct rcu_head *head, 122 void (*func)(struct rcu_head *head)); 123 124#else /* #ifdef CONFIG_PREEMPT_RCU */ 125 126/* In classic RCU, call_rcu() is just call_rcu_sched(). */ 127#define call_rcu call_rcu_sched 128 129#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 130 131/** 132 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period. 133 * @head: structure to be used for queueing the RCU updates. 134 * @func: actual callback function to be invoked after the grace period 135 * 136 * The callback function will be invoked some time after a full grace 137 * period elapses, in other words after all currently executing RCU 138 * read-side critical sections have completed. call_rcu_bh() assumes 139 * that the read-side critical sections end on completion of a softirq 140 * handler. This means that read-side critical sections in process 141 * context must not be interrupted by softirqs. This interface is to be 142 * used when most of the read-side critical sections are in softirq context. 143 * RCU read-side critical sections are delimited by : 144 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context. 145 * OR 146 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context. 147 * These may be nested. 148 * 149 * See the description of call_rcu() for more detailed information on 150 * memory ordering guarantees. 151 */ 152extern void call_rcu_bh(struct rcu_head *head, 153 void (*func)(struct rcu_head *head)); 154 155/** 156 * call_rcu_sched() - Queue an RCU for invocation after sched grace period. 157 * @head: structure to be used for queueing the RCU updates. 158 * @func: actual callback function to be invoked after the grace period 159 * 160 * The callback function will be invoked some time after a full grace 161 * period elapses, in other words after all currently executing RCU 162 * read-side critical sections have completed. call_rcu_sched() assumes 163 * that the read-side critical sections end on enabling of preemption 164 * or on voluntary preemption. 165 * RCU read-side critical sections are delimited by : 166 * - rcu_read_lock_sched() and rcu_read_unlock_sched(), 167 * OR 168 * anything that disables preemption. 169 * These may be nested. 170 * 171 * See the description of call_rcu() for more detailed information on 172 * memory ordering guarantees. 173 */ 174extern void call_rcu_sched(struct rcu_head *head, 175 void (*func)(struct rcu_head *rcu)); 176 177extern void synchronize_sched(void); 178 179#ifdef CONFIG_PREEMPT_RCU 180 181extern void __rcu_read_lock(void); 182extern void __rcu_read_unlock(void); 183extern void rcu_read_unlock_special(struct task_struct *t); 184void synchronize_rcu(void); 185 186/* 187 * Defined as a macro as it is a very low level header included from 188 * areas that don't even know about current. This gives the rcu_read_lock() 189 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other 190 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. 191 */ 192#define rcu_preempt_depth() (current->rcu_read_lock_nesting) 193 194#else /* #ifdef CONFIG_PREEMPT_RCU */ 195 196static inline void __rcu_read_lock(void) 197{ 198 preempt_disable(); 199} 200 201static inline void __rcu_read_unlock(void) 202{ 203 preempt_enable(); 204} 205 206static inline void synchronize_rcu(void) 207{ 208 synchronize_sched(); 209} 210 211static inline int rcu_preempt_depth(void) 212{ 213 return 0; 214} 215 216#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 217 218/* Internal to kernel */ 219extern void rcu_init(void); 220extern void rcu_sched_qs(int cpu); 221extern void rcu_bh_qs(int cpu); 222extern void rcu_check_callbacks(int cpu, int user); 223struct notifier_block; 224extern void rcu_idle_enter(void); 225extern void rcu_idle_exit(void); 226extern void rcu_irq_enter(void); 227extern void rcu_irq_exit(void); 228 229#ifdef CONFIG_RCU_USER_QS 230extern void rcu_user_enter(void); 231extern void rcu_user_exit(void); 232#else 233static inline void rcu_user_enter(void) { } 234static inline void rcu_user_exit(void) { } 235static inline void rcu_user_hooks_switch(struct task_struct *prev, 236 struct task_struct *next) { } 237#endif /* CONFIG_RCU_USER_QS */ 238 239/** 240 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers 241 * @a: Code that RCU needs to pay attention to. 242 * 243 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden 244 * in the inner idle loop, that is, between the rcu_idle_enter() and 245 * the rcu_idle_exit() -- RCU will happily ignore any such read-side 246 * critical sections. However, things like powertop need tracepoints 247 * in the inner idle loop. 248 * 249 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) 250 * will tell RCU that it needs to pay attending, invoke its argument 251 * (in this example, a call to the do_something_with_RCU() function), 252 * and then tell RCU to go back to ignoring this CPU. It is permissible 253 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently 254 * quite limited. If deeper nesting is required, it will be necessary 255 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly. 256 */ 257#define RCU_NONIDLE(a) \ 258 do { \ 259 rcu_irq_enter(); \ 260 do { a; } while (0); \ 261 rcu_irq_exit(); \ 262 } while (0) 263 264/* 265 * Infrastructure to implement the synchronize_() primitives in 266 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 267 */ 268 269typedef void call_rcu_func_t(struct rcu_head *head, 270 void (*func)(struct rcu_head *head)); 271void wait_rcu_gp(call_rcu_func_t crf); 272 273#if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) 274#include <linux/rcutree.h> 275#elif defined(CONFIG_TINY_RCU) 276#include <linux/rcutiny.h> 277#else 278#error "Unknown RCU implementation specified to kernel configuration" 279#endif 280 281/* 282 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic 283 * initialization and destruction of rcu_head on the stack. rcu_head structures 284 * allocated dynamically in the heap or defined statically don't need any 285 * initialization. 286 */ 287#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 288extern void init_rcu_head_on_stack(struct rcu_head *head); 289extern void destroy_rcu_head_on_stack(struct rcu_head *head); 290#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 291static inline void init_rcu_head_on_stack(struct rcu_head *head) 292{ 293} 294 295static inline void destroy_rcu_head_on_stack(struct rcu_head *head) 296{ 297} 298#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 299 300#if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SMP) 301extern int rcu_is_cpu_idle(void); 302#endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SMP) */ 303 304#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 305bool rcu_lockdep_current_cpu_online(void); 306#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 307static inline bool rcu_lockdep_current_cpu_online(void) 308{ 309 return 1; 310} 311#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 312 313#ifdef CONFIG_DEBUG_LOCK_ALLOC 314 315static inline void rcu_lock_acquire(struct lockdep_map *map) 316{ 317 lock_acquire(map, 0, 0, 2, 1, NULL, _THIS_IP_); 318} 319 320static inline void rcu_lock_release(struct lockdep_map *map) 321{ 322 lock_release(map, 1, _THIS_IP_); 323} 324 325extern struct lockdep_map rcu_lock_map; 326extern struct lockdep_map rcu_bh_lock_map; 327extern struct lockdep_map rcu_sched_lock_map; 328extern int debug_lockdep_rcu_enabled(void); 329 330/** 331 * rcu_read_lock_held() - might we be in RCU read-side critical section? 332 * 333 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU 334 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC, 335 * this assumes we are in an RCU read-side critical section unless it can 336 * prove otherwise. This is useful for debug checks in functions that 337 * require that they be called within an RCU read-side critical section. 338 * 339 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot 340 * and while lockdep is disabled. 341 * 342 * Note that rcu_read_lock() and the matching rcu_read_unlock() must 343 * occur in the same context, for example, it is illegal to invoke 344 * rcu_read_unlock() in process context if the matching rcu_read_lock() 345 * was invoked from within an irq handler. 346 * 347 * Note that rcu_read_lock() is disallowed if the CPU is either idle or 348 * offline from an RCU perspective, so check for those as well. 349 */ 350static inline int rcu_read_lock_held(void) 351{ 352 if (!debug_lockdep_rcu_enabled()) 353 return 1; 354 if (rcu_is_cpu_idle()) 355 return 0; 356 if (!rcu_lockdep_current_cpu_online()) 357 return 0; 358 return lock_is_held(&rcu_lock_map); 359} 360 361/* 362 * rcu_read_lock_bh_held() is defined out of line to avoid #include-file 363 * hell. 364 */ 365extern int rcu_read_lock_bh_held(void); 366 367/** 368 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section? 369 * 370 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an 371 * RCU-sched read-side critical section. In absence of 372 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side 373 * critical section unless it can prove otherwise. Note that disabling 374 * of preemption (including disabling irqs) counts as an RCU-sched 375 * read-side critical section. This is useful for debug checks in functions 376 * that required that they be called within an RCU-sched read-side 377 * critical section. 378 * 379 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot 380 * and while lockdep is disabled. 381 * 382 * Note that if the CPU is in the idle loop from an RCU point of 383 * view (ie: that we are in the section between rcu_idle_enter() and 384 * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU 385 * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs 386 * that are in such a section, considering these as in extended quiescent 387 * state, so such a CPU is effectively never in an RCU read-side critical 388 * section regardless of what RCU primitives it invokes. This state of 389 * affairs is required --- we need to keep an RCU-free window in idle 390 * where the CPU may possibly enter into low power mode. This way we can 391 * notice an extended quiescent state to other CPUs that started a grace 392 * period. Otherwise we would delay any grace period as long as we run in 393 * the idle task. 394 * 395 * Similarly, we avoid claiming an SRCU read lock held if the current 396 * CPU is offline. 397 */ 398#ifdef CONFIG_PREEMPT_COUNT 399static inline int rcu_read_lock_sched_held(void) 400{ 401 int lockdep_opinion = 0; 402 403 if (!debug_lockdep_rcu_enabled()) 404 return 1; 405 if (rcu_is_cpu_idle()) 406 return 0; 407 if (!rcu_lockdep_current_cpu_online()) 408 return 0; 409 if (debug_locks) 410 lockdep_opinion = lock_is_held(&rcu_sched_lock_map); 411 return lockdep_opinion || preempt_count() != 0 || irqs_disabled(); 412} 413#else /* #ifdef CONFIG_PREEMPT_COUNT */ 414static inline int rcu_read_lock_sched_held(void) 415{ 416 return 1; 417} 418#endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ 419 420#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 421 422# define rcu_lock_acquire(a) do { } while (0) 423# define rcu_lock_release(a) do { } while (0) 424 425static inline int rcu_read_lock_held(void) 426{ 427 return 1; 428} 429 430static inline int rcu_read_lock_bh_held(void) 431{ 432 return 1; 433} 434 435#ifdef CONFIG_PREEMPT_COUNT 436static inline int rcu_read_lock_sched_held(void) 437{ 438 return preempt_count() != 0 || irqs_disabled(); 439} 440#else /* #ifdef CONFIG_PREEMPT_COUNT */ 441static inline int rcu_read_lock_sched_held(void) 442{ 443 return 1; 444} 445#endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ 446 447#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 448 449#ifdef CONFIG_PROVE_RCU 450 451extern int rcu_my_thread_group_empty(void); 452 453/** 454 * rcu_lockdep_assert - emit lockdep splat if specified condition not met 455 * @c: condition to check 456 * @s: informative message 457 */ 458#define rcu_lockdep_assert(c, s) \ 459 do { \ 460 static bool __section(.data.unlikely) __warned; \ 461 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \ 462 __warned = true; \ 463 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 464 } \ 465 } while (0) 466 467#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 468static inline void rcu_preempt_sleep_check(void) 469{ 470 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map), 471 "Illegal context switch in RCU read-side critical section"); 472} 473#else /* #ifdef CONFIG_PROVE_RCU */ 474static inline void rcu_preempt_sleep_check(void) 475{ 476} 477#endif /* #else #ifdef CONFIG_PROVE_RCU */ 478 479#define rcu_sleep_check() \ 480 do { \ 481 rcu_preempt_sleep_check(); \ 482 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \ 483 "Illegal context switch in RCU-bh" \ 484 " read-side critical section"); \ 485 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \ 486 "Illegal context switch in RCU-sched"\ 487 " read-side critical section"); \ 488 } while (0) 489 490#else /* #ifdef CONFIG_PROVE_RCU */ 491 492#define rcu_lockdep_assert(c, s) do { } while (0) 493#define rcu_sleep_check() do { } while (0) 494 495#endif /* #else #ifdef CONFIG_PROVE_RCU */ 496 497/* 498 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 499 * and rcu_assign_pointer(). Some of these could be folded into their 500 * callers, but they are left separate in order to ease introduction of 501 * multiple flavors of pointers to match the multiple flavors of RCU 502 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in 503 * the future. 504 */ 505 506#ifdef __CHECKER__ 507#define rcu_dereference_sparse(p, space) \ 508 ((void)(((typeof(*p) space *)p) == p)) 509#else /* #ifdef __CHECKER__ */ 510#define rcu_dereference_sparse(p, space) 511#endif /* #else #ifdef __CHECKER__ */ 512 513#define __rcu_access_pointer(p, space) \ 514 ({ \ 515 typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \ 516 rcu_dereference_sparse(p, space); \ 517 ((typeof(*p) __force __kernel *)(_________p1)); \ 518 }) 519#define __rcu_dereference_check(p, c, space) \ 520 ({ \ 521 typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \ 522 rcu_lockdep_assert(c, "suspicious rcu_dereference_check()" \ 523 " usage"); \ 524 rcu_dereference_sparse(p, space); \ 525 smp_read_barrier_depends(); \ 526 ((typeof(*p) __force __kernel *)(_________p1)); \ 527 }) 528#define __rcu_dereference_protected(p, c, space) \ 529 ({ \ 530 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected()" \ 531 " usage"); \ 532 rcu_dereference_sparse(p, space); \ 533 ((typeof(*p) __force __kernel *)(p)); \ 534 }) 535 536#define __rcu_access_index(p, space) \ 537 ({ \ 538 typeof(p) _________p1 = ACCESS_ONCE(p); \ 539 rcu_dereference_sparse(p, space); \ 540 (_________p1); \ 541 }) 542#define __rcu_dereference_index_check(p, c) \ 543 ({ \ 544 typeof(p) _________p1 = ACCESS_ONCE(p); \ 545 rcu_lockdep_assert(c, \ 546 "suspicious rcu_dereference_index_check()" \ 547 " usage"); \ 548 smp_read_barrier_depends(); \ 549 (_________p1); \ 550 }) 551#define __rcu_assign_pointer(p, v, space) \ 552 do { \ 553 smp_wmb(); \ 554 (p) = (typeof(*v) __force space *)(v); \ 555 } while (0) 556 557 558/** 559 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 560 * @p: The pointer to read 561 * 562 * Return the value of the specified RCU-protected pointer, but omit the 563 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful 564 * when the value of this pointer is accessed, but the pointer is not 565 * dereferenced, for example, when testing an RCU-protected pointer against 566 * NULL. Although rcu_access_pointer() may also be used in cases where 567 * update-side locks prevent the value of the pointer from changing, you 568 * should instead use rcu_dereference_protected() for this use case. 569 * 570 * It is also permissible to use rcu_access_pointer() when read-side 571 * access to the pointer was removed at least one grace period ago, as 572 * is the case in the context of the RCU callback that is freeing up 573 * the data, or after a synchronize_rcu() returns. This can be useful 574 * when tearing down multi-linked structures after a grace period 575 * has elapsed. 576 */ 577#define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) 578 579/** 580 * rcu_dereference_check() - rcu_dereference with debug checking 581 * @p: The pointer to read, prior to dereferencing 582 * @c: The conditions under which the dereference will take place 583 * 584 * Do an rcu_dereference(), but check that the conditions under which the 585 * dereference will take place are correct. Typically the conditions 586 * indicate the various locking conditions that should be held at that 587 * point. The check should return true if the conditions are satisfied. 588 * An implicit check for being in an RCU read-side critical section 589 * (rcu_read_lock()) is included. 590 * 591 * For example: 592 * 593 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 594 * 595 * could be used to indicate to lockdep that foo->bar may only be dereferenced 596 * if either rcu_read_lock() is held, or that the lock required to replace 597 * the bar struct at foo->bar is held. 598 * 599 * Note that the list of conditions may also include indications of when a lock 600 * need not be held, for example during initialisation or destruction of the 601 * target struct: 602 * 603 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 604 * atomic_read(&foo->usage) == 0); 605 * 606 * Inserts memory barriers on architectures that require them 607 * (currently only the Alpha), prevents the compiler from refetching 608 * (and from merging fetches), and, more importantly, documents exactly 609 * which pointers are protected by RCU and checks that the pointer is 610 * annotated as __rcu. 611 */ 612#define rcu_dereference_check(p, c) \ 613 __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu) 614 615/** 616 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 617 * @p: The pointer to read, prior to dereferencing 618 * @c: The conditions under which the dereference will take place 619 * 620 * This is the RCU-bh counterpart to rcu_dereference_check(). 621 */ 622#define rcu_dereference_bh_check(p, c) \ 623 __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu) 624 625/** 626 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 627 * @p: The pointer to read, prior to dereferencing 628 * @c: The conditions under which the dereference will take place 629 * 630 * This is the RCU-sched counterpart to rcu_dereference_check(). 631 */ 632#define rcu_dereference_sched_check(p, c) \ 633 __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \ 634 __rcu) 635 636#define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/ 637 638/* 639 * The tracing infrastructure traces RCU (we want that), but unfortunately 640 * some of the RCU checks causes tracing to lock up the system. 641 * 642 * The tracing version of rcu_dereference_raw() must not call 643 * rcu_read_lock_held(). 644 */ 645#define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu) 646 647/** 648 * rcu_access_index() - fetch RCU index with no dereferencing 649 * @p: The index to read 650 * 651 * Return the value of the specified RCU-protected index, but omit the 652 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful 653 * when the value of this index is accessed, but the index is not 654 * dereferenced, for example, when testing an RCU-protected index against 655 * -1. Although rcu_access_index() may also be used in cases where 656 * update-side locks prevent the value of the index from changing, you 657 * should instead use rcu_dereference_index_protected() for this use case. 658 */ 659#define rcu_access_index(p) __rcu_access_index((p), __rcu) 660 661/** 662 * rcu_dereference_index_check() - rcu_dereference for indices with debug checking 663 * @p: The pointer to read, prior to dereferencing 664 * @c: The conditions under which the dereference will take place 665 * 666 * Similar to rcu_dereference_check(), but omits the sparse checking. 667 * This allows rcu_dereference_index_check() to be used on integers, 668 * which can then be used as array indices. Attempting to use 669 * rcu_dereference_check() on an integer will give compiler warnings 670 * because the sparse address-space mechanism relies on dereferencing 671 * the RCU-protected pointer. Dereferencing integers is not something 672 * that even gcc will put up with. 673 * 674 * Note that this function does not implicitly check for RCU read-side 675 * critical sections. If this function gains lots of uses, it might 676 * make sense to provide versions for each flavor of RCU, but it does 677 * not make sense as of early 2010. 678 */ 679#define rcu_dereference_index_check(p, c) \ 680 __rcu_dereference_index_check((p), (c)) 681 682/** 683 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 684 * @p: The pointer to read, prior to dereferencing 685 * @c: The conditions under which the dereference will take place 686 * 687 * Return the value of the specified RCU-protected pointer, but omit 688 * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This 689 * is useful in cases where update-side locks prevent the value of the 690 * pointer from changing. Please note that this primitive does -not- 691 * prevent the compiler from repeating this reference or combining it 692 * with other references, so it should not be used without protection 693 * of appropriate locks. 694 * 695 * This function is only for update-side use. Using this function 696 * when protected only by rcu_read_lock() will result in infrequent 697 * but very ugly failures. 698 */ 699#define rcu_dereference_protected(p, c) \ 700 __rcu_dereference_protected((p), (c), __rcu) 701 702 703/** 704 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 705 * @p: The pointer to read, prior to dereferencing 706 * 707 * This is a simple wrapper around rcu_dereference_check(). 708 */ 709#define rcu_dereference(p) rcu_dereference_check(p, 0) 710 711/** 712 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 713 * @p: The pointer to read, prior to dereferencing 714 * 715 * Makes rcu_dereference_check() do the dirty work. 716 */ 717#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 718 719/** 720 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 721 * @p: The pointer to read, prior to dereferencing 722 * 723 * Makes rcu_dereference_check() do the dirty work. 724 */ 725#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 726 727/** 728 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 729 * 730 * When synchronize_rcu() is invoked on one CPU while other CPUs 731 * are within RCU read-side critical sections, then the 732 * synchronize_rcu() is guaranteed to block until after all the other 733 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 734 * on one CPU while other CPUs are within RCU read-side critical 735 * sections, invocation of the corresponding RCU callback is deferred 736 * until after the all the other CPUs exit their critical sections. 737 * 738 * Note, however, that RCU callbacks are permitted to run concurrently 739 * with new RCU read-side critical sections. One way that this can happen 740 * is via the following sequence of events: (1) CPU 0 enters an RCU 741 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 742 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 743 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 744 * callback is invoked. This is legal, because the RCU read-side critical 745 * section that was running concurrently with the call_rcu() (and which 746 * therefore might be referencing something that the corresponding RCU 747 * callback would free up) has completed before the corresponding 748 * RCU callback is invoked. 749 * 750 * RCU read-side critical sections may be nested. Any deferred actions 751 * will be deferred until the outermost RCU read-side critical section 752 * completes. 753 * 754 * You can avoid reading and understanding the next paragraph by 755 * following this rule: don't put anything in an rcu_read_lock() RCU 756 * read-side critical section that would block in a !PREEMPT kernel. 757 * But if you want the full story, read on! 758 * 759 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), it 760 * is illegal to block while in an RCU read-side critical section. In 761 * preemptible RCU implementations (TREE_PREEMPT_RCU and TINY_PREEMPT_RCU) 762 * in CONFIG_PREEMPT kernel builds, RCU read-side critical sections may 763 * be preempted, but explicit blocking is illegal. Finally, in preemptible 764 * RCU implementations in real-time (with -rt patchset) kernel builds, 765 * RCU read-side critical sections may be preempted and they may also 766 * block, but only when acquiring spinlocks that are subject to priority 767 * inheritance. 768 */ 769static inline void rcu_read_lock(void) 770{ 771 __rcu_read_lock(); 772 __acquire(RCU); 773 rcu_lock_acquire(&rcu_lock_map); 774 rcu_lockdep_assert(!rcu_is_cpu_idle(), 775 "rcu_read_lock() used illegally while idle"); 776} 777 778/* 779 * So where is rcu_write_lock()? It does not exist, as there is no 780 * way for writers to lock out RCU readers. This is a feature, not 781 * a bug -- this property is what provides RCU's performance benefits. 782 * Of course, writers must coordinate with each other. The normal 783 * spinlock primitives work well for this, but any other technique may be 784 * used as well. RCU does not care how the writers keep out of each 785 * others' way, as long as they do so. 786 */ 787 788/** 789 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 790 * 791 * See rcu_read_lock() for more information. 792 */ 793static inline void rcu_read_unlock(void) 794{ 795 rcu_lockdep_assert(!rcu_is_cpu_idle(), 796 "rcu_read_unlock() used illegally while idle"); 797 rcu_lock_release(&rcu_lock_map); 798 __release(RCU); 799 __rcu_read_unlock(); 800} 801 802/** 803 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 804 * 805 * This is equivalent of rcu_read_lock(), but to be used when updates 806 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since 807 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a 808 * softirq handler to be a quiescent state, a process in RCU read-side 809 * critical section must be protected by disabling softirqs. Read-side 810 * critical sections in interrupt context can use just rcu_read_lock(), 811 * though this should at least be commented to avoid confusing people 812 * reading the code. 813 * 814 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 815 * must occur in the same context, for example, it is illegal to invoke 816 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 817 * was invoked from some other task. 818 */ 819static inline void rcu_read_lock_bh(void) 820{ 821 local_bh_disable(); 822 __acquire(RCU_BH); 823 rcu_lock_acquire(&rcu_bh_lock_map); 824 rcu_lockdep_assert(!rcu_is_cpu_idle(), 825 "rcu_read_lock_bh() used illegally while idle"); 826} 827 828/* 829 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section 830 * 831 * See rcu_read_lock_bh() for more information. 832 */ 833static inline void rcu_read_unlock_bh(void) 834{ 835 rcu_lockdep_assert(!rcu_is_cpu_idle(), 836 "rcu_read_unlock_bh() used illegally while idle"); 837 rcu_lock_release(&rcu_bh_lock_map); 838 __release(RCU_BH); 839 local_bh_enable(); 840} 841 842/** 843 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 844 * 845 * This is equivalent of rcu_read_lock(), but to be used when updates 846 * are being done using call_rcu_sched() or synchronize_rcu_sched(). 847 * Read-side critical sections can also be introduced by anything that 848 * disables preemption, including local_irq_disable() and friends. 849 * 850 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 851 * must occur in the same context, for example, it is illegal to invoke 852 * rcu_read_unlock_sched() from process context if the matching 853 * rcu_read_lock_sched() was invoked from an NMI handler. 854 */ 855static inline void rcu_read_lock_sched(void) 856{ 857 preempt_disable(); 858 __acquire(RCU_SCHED); 859 rcu_lock_acquire(&rcu_sched_lock_map); 860 rcu_lockdep_assert(!rcu_is_cpu_idle(), 861 "rcu_read_lock_sched() used illegally while idle"); 862} 863 864/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 865static inline notrace void rcu_read_lock_sched_notrace(void) 866{ 867 preempt_disable_notrace(); 868 __acquire(RCU_SCHED); 869} 870 871/* 872 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section 873 * 874 * See rcu_read_lock_sched for more information. 875 */ 876static inline void rcu_read_unlock_sched(void) 877{ 878 rcu_lockdep_assert(!rcu_is_cpu_idle(), 879 "rcu_read_unlock_sched() used illegally while idle"); 880 rcu_lock_release(&rcu_sched_lock_map); 881 __release(RCU_SCHED); 882 preempt_enable(); 883} 884 885/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 886static inline notrace void rcu_read_unlock_sched_notrace(void) 887{ 888 __release(RCU_SCHED); 889 preempt_enable_notrace(); 890} 891 892/** 893 * rcu_assign_pointer() - assign to RCU-protected pointer 894 * @p: pointer to assign to 895 * @v: value to assign (publish) 896 * 897 * Assigns the specified value to the specified RCU-protected 898 * pointer, ensuring that any concurrent RCU readers will see 899 * any prior initialization. 900 * 901 * Inserts memory barriers on architectures that require them 902 * (which is most of them), and also prevents the compiler from 903 * reordering the code that initializes the structure after the pointer 904 * assignment. More importantly, this call documents which pointers 905 * will be dereferenced by RCU read-side code. 906 * 907 * In some special cases, you may use RCU_INIT_POINTER() instead 908 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 909 * to the fact that it does not constrain either the CPU or the compiler. 910 * That said, using RCU_INIT_POINTER() when you should have used 911 * rcu_assign_pointer() is a very bad thing that results in 912 * impossible-to-diagnose memory corruption. So please be careful. 913 * See the RCU_INIT_POINTER() comment header for details. 914 */ 915#define rcu_assign_pointer(p, v) \ 916 __rcu_assign_pointer((p), (v), __rcu) 917 918/** 919 * RCU_INIT_POINTER() - initialize an RCU protected pointer 920 * 921 * Initialize an RCU-protected pointer in special cases where readers 922 * do not need ordering constraints on the CPU or the compiler. These 923 * special cases are: 924 * 925 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or- 926 * 2. The caller has taken whatever steps are required to prevent 927 * RCU readers from concurrently accessing this pointer -or- 928 * 3. The referenced data structure has already been exposed to 929 * readers either at compile time or via rcu_assign_pointer() -and- 930 * a. You have not made -any- reader-visible changes to 931 * this structure since then -or- 932 * b. It is OK for readers accessing this structure from its 933 * new location to see the old state of the structure. (For 934 * example, the changes were to statistical counters or to 935 * other state where exact synchronization is not required.) 936 * 937 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 938 * result in impossible-to-diagnose memory corruption. As in the structures 939 * will look OK in crash dumps, but any concurrent RCU readers might 940 * see pre-initialized values of the referenced data structure. So 941 * please be very careful how you use RCU_INIT_POINTER()!!! 942 * 943 * If you are creating an RCU-protected linked structure that is accessed 944 * by a single external-to-structure RCU-protected pointer, then you may 945 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 946 * pointers, but you must use rcu_assign_pointer() to initialize the 947 * external-to-structure pointer -after- you have completely initialized 948 * the reader-accessible portions of the linked structure. 949 */ 950#define RCU_INIT_POINTER(p, v) \ 951 do { \ 952 p = (typeof(*v) __force __rcu *)(v); \ 953 } while (0) 954 955/** 956 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 957 * 958 * GCC-style initialization for an RCU-protected pointer in a structure field. 959 */ 960#define RCU_POINTER_INITIALIZER(p, v) \ 961 .p = (typeof(*v) __force __rcu *)(v) 962 963/* 964 * Does the specified offset indicate that the corresponding rcu_head 965 * structure can be handled by kfree_rcu()? 966 */ 967#define __is_kfree_rcu_offset(offset) ((offset) < 4096) 968 969/* 970 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. 971 */ 972#define __kfree_rcu(head, offset) \ 973 do { \ 974 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \ 975 kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \ 976 } while (0) 977 978/** 979 * kfree_rcu() - kfree an object after a grace period. 980 * @ptr: pointer to kfree 981 * @rcu_head: the name of the struct rcu_head within the type of @ptr. 982 * 983 * Many rcu callbacks functions just call kfree() on the base structure. 984 * These functions are trivial, but their size adds up, and furthermore 985 * when they are used in a kernel module, that module must invoke the 986 * high-latency rcu_barrier() function at module-unload time. 987 * 988 * The kfree_rcu() function handles this issue. Rather than encoding a 989 * function address in the embedded rcu_head structure, kfree_rcu() instead 990 * encodes the offset of the rcu_head structure within the base structure. 991 * Because the functions are not allowed in the low-order 4096 bytes of 992 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 993 * If the offset is larger than 4095 bytes, a compile-time error will 994 * be generated in __kfree_rcu(). If this error is triggered, you can 995 * either fall back to use of call_rcu() or rearrange the structure to 996 * position the rcu_head structure into the first 4096 bytes. 997 * 998 * Note that the allowable offset might decrease in the future, for example, 999 * to allow something like kmem_cache_free_rcu(). 1000 *
1001 * The BUILD_BUG_ON check must not involve any function calls, hence the 1002 * checks are done in macros here. 1003 */ 1004#define kfree_rcu(ptr, rcu_head) \ 1005 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head)) 1006 1007#ifdef CONFIG_RCU_NOCB_CPU 1008extern bool rcu_is_nocb_cpu(int cpu); 1009#else 1010static inline bool rcu_is_nocb_cpu(int cpu) { return false; } 1011#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 1012 1013 1014/* Only for use by adaptive-ticks code. */ 1015#ifdef CONFIG_NO_HZ_FULL_SYSIDLE 1016extern bool rcu_sys_is_idle(void); 1017extern void rcu_sysidle_force_exit(void); 1018#else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ 1019 1020static inline bool rcu_sys_is_idle(void) 1021{ 1022 return false; 1023} 1024 1025static inline void rcu_sysidle_force_exit(void) 1026{ 1027} 1028 1029#endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ 1030 1031 1032#endif /* __LINUX_RCUPDATE_H */ 1033