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, you can access it online at 16 * http://www.gnu.org/licenses/gpl-2.0.html. 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/compiler.h> 38#include <linux/atomic.h> 39#include <linux/irqflags.h> 40#include <linux/preempt.h> 41#include <linux/bottom_half.h> 42#include <linux/lockdep.h> 43#include <asm/processor.h> 44#include <linux/cpumask.h> 45 46#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) 47#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) 48#define ulong2long(a) (*(long *)(&(a))) 49 50/* Exported common interfaces */ 51 52#ifdef CONFIG_PREEMPT_RCU 53void call_rcu(struct rcu_head *head, rcu_callback_t func); 54#else /* #ifdef CONFIG_PREEMPT_RCU */ 55#define call_rcu call_rcu_sched 56#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 57 58void call_rcu_bh(struct rcu_head *head, rcu_callback_t func); 59void call_rcu_sched(struct rcu_head *head, rcu_callback_t func); 60void synchronize_sched(void); 61void rcu_barrier_tasks(void); 62 63#ifdef CONFIG_PREEMPT_RCU 64 65void __rcu_read_lock(void); 66void __rcu_read_unlock(void); 67void rcu_read_unlock_special(struct task_struct *t); 68void synchronize_rcu(void); 69 70/* 71 * Defined as a macro as it is a very low level header included from 72 * areas that don't even know about current. This gives the rcu_read_lock() 73 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other 74 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. 75 */ 76#define rcu_preempt_depth() (current->rcu_read_lock_nesting) 77 78#else /* #ifdef CONFIG_PREEMPT_RCU */ 79 80static inline void __rcu_read_lock(void) 81{ 82 if (IS_ENABLED(CONFIG_PREEMPT_COUNT)) 83 preempt_disable(); 84} 85 86static inline void __rcu_read_unlock(void) 87{ 88 if (IS_ENABLED(CONFIG_PREEMPT_COUNT)) 89 preempt_enable(); 90} 91 92static inline void synchronize_rcu(void) 93{ 94 synchronize_sched(); 95} 96 97static inline int rcu_preempt_depth(void) 98{ 99 return 0; 100} 101 102#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 103 104/* Internal to kernel */ 105void rcu_init(void); 106extern int rcu_scheduler_active __read_mostly; 107void rcu_sched_qs(void); 108void rcu_bh_qs(void); 109void rcu_check_callbacks(int user); 110void rcu_report_dead(unsigned int cpu); 111void rcu_cpu_starting(unsigned int cpu); 112void rcutree_migrate_callbacks(int cpu); 113 114#ifdef CONFIG_RCU_STALL_COMMON 115void rcu_sysrq_start(void); 116void rcu_sysrq_end(void); 117#else /* #ifdef CONFIG_RCU_STALL_COMMON */ 118static inline void rcu_sysrq_start(void) { } 119static inline void rcu_sysrq_end(void) { } 120#endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ 121 122#ifdef CONFIG_NO_HZ_FULL 123void rcu_user_enter(void); 124void rcu_user_exit(void); 125#else 126static inline void rcu_user_enter(void) { } 127static inline void rcu_user_exit(void) { } 128#endif /* CONFIG_NO_HZ_FULL */ 129 130#ifdef CONFIG_RCU_NOCB_CPU 131void rcu_init_nohz(void); 132#else /* #ifdef CONFIG_RCU_NOCB_CPU */ 133static inline void rcu_init_nohz(void) { } 134#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 135 136/** 137 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers 138 * @a: Code that RCU needs to pay attention to. 139 * 140 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden 141 * in the inner idle loop, that is, between the rcu_idle_enter() and 142 * the rcu_idle_exit() -- RCU will happily ignore any such read-side 143 * critical sections. However, things like powertop need tracepoints 144 * in the inner idle loop. 145 * 146 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) 147 * will tell RCU that it needs to pay attention, invoke its argument 148 * (in this example, calling the do_something_with_RCU() function), 149 * and then tell RCU to go back to ignoring this CPU. It is permissible 150 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is 151 * on the order of a million or so, even on 32-bit systems). It is 152 * not legal to block within RCU_NONIDLE(), nor is it permissible to 153 * transfer control either into or out of RCU_NONIDLE()'s statement. 154 */ 155#define RCU_NONIDLE(a) \ 156 do { \ 157 rcu_irq_enter_irqson(); \ 158 do { a; } while (0); \ 159 rcu_irq_exit_irqson(); \ 160 } while (0) 161 162/* 163 * Note a voluntary context switch for RCU-tasks benefit. This is a 164 * macro rather than an inline function to avoid #include hell. 165 */ 166#ifdef CONFIG_TASKS_RCU 167#define rcu_note_voluntary_context_switch_lite(t) \ 168 do { \ 169 if (READ_ONCE((t)->rcu_tasks_holdout)) \ 170 WRITE_ONCE((t)->rcu_tasks_holdout, false); \ 171 } while (0) 172#define rcu_note_voluntary_context_switch(t) \ 173 do { \ 174 rcu_all_qs(); \ 175 rcu_note_voluntary_context_switch_lite(t); \ 176 } while (0) 177void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func); 178void synchronize_rcu_tasks(void); 179void exit_tasks_rcu_start(void); 180void exit_tasks_rcu_finish(void); 181#else /* #ifdef CONFIG_TASKS_RCU */ 182#define rcu_note_voluntary_context_switch_lite(t) do { } while (0) 183#define rcu_note_voluntary_context_switch(t) rcu_all_qs() 184#define call_rcu_tasks call_rcu_sched 185#define synchronize_rcu_tasks synchronize_sched 186static inline void exit_tasks_rcu_start(void) { } 187static inline void exit_tasks_rcu_finish(void) { } 188#endif /* #else #ifdef CONFIG_TASKS_RCU */ 189 190/** 191 * cond_resched_rcu_qs - Report potential quiescent states to RCU 192 * 193 * This macro resembles cond_resched(), except that it is defined to 194 * report potential quiescent states to RCU-tasks even if the cond_resched() 195 * machinery were to be shut off, as some advocate for PREEMPT kernels. 196 */ 197#define cond_resched_rcu_qs() \ 198do { \ 199 if (!cond_resched()) \ 200 rcu_note_voluntary_context_switch_lite(current); \ 201} while (0) 202 203/* 204 * Infrastructure to implement the synchronize_() primitives in 205 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 206 */ 207 208#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) 209#include <linux/rcutree.h> 210#elif defined(CONFIG_TINY_RCU) 211#include <linux/rcutiny.h> 212#else 213#error "Unknown RCU implementation specified to kernel configuration" 214#endif 215 216/* 217 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls 218 * are needed for dynamic initialization and destruction of rcu_head 219 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for 220 * dynamic initialization and destruction of statically allocated rcu_head 221 * structures. However, rcu_head structures allocated dynamically in the 222 * heap don't need any initialization. 223 */ 224#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 225void init_rcu_head(struct rcu_head *head); 226void destroy_rcu_head(struct rcu_head *head); 227void init_rcu_head_on_stack(struct rcu_head *head); 228void destroy_rcu_head_on_stack(struct rcu_head *head); 229#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 230static inline void init_rcu_head(struct rcu_head *head) { } 231static inline void destroy_rcu_head(struct rcu_head *head) { } 232static inline void init_rcu_head_on_stack(struct rcu_head *head) { } 233static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { } 234#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 235 236#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 237bool rcu_lockdep_current_cpu_online(void); 238#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 239static inline bool rcu_lockdep_current_cpu_online(void) { return true; } 240#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 241 242#ifdef CONFIG_DEBUG_LOCK_ALLOC 243 244static inline void rcu_lock_acquire(struct lockdep_map *map) 245{ 246 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); 247} 248 249static inline void rcu_lock_release(struct lockdep_map *map) 250{ 251 lock_release(map, 1, _THIS_IP_); 252} 253 254extern struct lockdep_map rcu_lock_map; 255extern struct lockdep_map rcu_bh_lock_map; 256extern struct lockdep_map rcu_sched_lock_map; 257extern struct lockdep_map rcu_callback_map; 258int debug_lockdep_rcu_enabled(void); 259int rcu_read_lock_held(void); 260int rcu_read_lock_bh_held(void); 261int rcu_read_lock_sched_held(void); 262 263#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 264 265# define rcu_lock_acquire(a) do { } while (0) 266# define rcu_lock_release(a) do { } while (0) 267 268static inline int rcu_read_lock_held(void) 269{ 270 return 1; 271} 272 273static inline int rcu_read_lock_bh_held(void) 274{ 275 return 1; 276} 277 278static inline int rcu_read_lock_sched_held(void) 279{ 280 return !preemptible(); 281} 282#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 283 284#ifdef CONFIG_PROVE_RCU 285 286/** 287 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met 288 * @c: condition to check 289 * @s: informative message 290 */ 291#define RCU_LOCKDEP_WARN(c, s) \ 292 do { \ 293 static bool __section(.data.unlikely) __warned; \ 294 if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \ 295 __warned = true; \ 296 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 297 } \ 298 } while (0) 299 300#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 301static inline void rcu_preempt_sleep_check(void) 302{ 303 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), 304 "Illegal context switch in RCU read-side critical section"); 305} 306#else /* #ifdef CONFIG_PROVE_RCU */ 307static inline void rcu_preempt_sleep_check(void) { } 308#endif /* #else #ifdef CONFIG_PROVE_RCU */ 309 310#define rcu_sleep_check() \ 311 do { \ 312 rcu_preempt_sleep_check(); \ 313 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ 314 "Illegal context switch in RCU-bh read-side critical section"); \ 315 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ 316 "Illegal context switch in RCU-sched read-side critical section"); \ 317 } while (0) 318 319#else /* #ifdef CONFIG_PROVE_RCU */ 320 321#define RCU_LOCKDEP_WARN(c, s) do { } while (0) 322#define rcu_sleep_check() do { } while (0) 323 324#endif /* #else #ifdef CONFIG_PROVE_RCU */ 325 326/* 327 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 328 * and rcu_assign_pointer(). Some of these could be folded into their 329 * callers, but they are left separate in order to ease introduction of 330 * multiple flavors of pointers to match the multiple flavors of RCU 331 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in 332 * the future. 333 */ 334 335#ifdef __CHECKER__ 336#define rcu_dereference_sparse(p, space) \ 337 ((void)(((typeof(*p) space *)p) == p)) 338#else /* #ifdef __CHECKER__ */ 339#define rcu_dereference_sparse(p, space) 340#endif /* #else #ifdef __CHECKER__ */ 341 342#define __rcu_access_pointer(p, space) \ 343({ \ 344 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 345 rcu_dereference_sparse(p, space); \ 346 ((typeof(*p) __force __kernel *)(_________p1)); \ 347}) 348#define __rcu_dereference_check(p, c, space) \ 349({ \ 350 /* Dependency order vs. p above. */ \ 351 typeof(*p) *________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 352 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ 353 rcu_dereference_sparse(p, space); \ 354 ((typeof(*p) __force __kernel *)(________p1)); \ 355}) 356#define __rcu_dereference_protected(p, c, space) \ 357({ \ 358 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ 359 rcu_dereference_sparse(p, space); \ 360 ((typeof(*p) __force __kernel *)(p)); \ 361}) 362#define rcu_dereference_raw(p) \ 363({ \ 364 /* Dependency order vs. p above. */ \ 365 typeof(p) ________p1 = READ_ONCE(p); \ 366 ((typeof(*p) __force __kernel *)(________p1)); \ 367}) 368 369/** 370 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 371 * @v: The value to statically initialize with. 372 */ 373#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 374 375/** 376 * rcu_assign_pointer() - assign to RCU-protected pointer 377 * @p: pointer to assign to 378 * @v: value to assign (publish) 379 * 380 * Assigns the specified value to the specified RCU-protected 381 * pointer, ensuring that any concurrent RCU readers will see 382 * any prior initialization. 383 * 384 * Inserts memory barriers on architectures that require them 385 * (which is most of them), and also prevents the compiler from 386 * reordering the code that initializes the structure after the pointer 387 * assignment. More importantly, this call documents which pointers 388 * will be dereferenced by RCU read-side code. 389 * 390 * In some special cases, you may use RCU_INIT_POINTER() instead 391 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 392 * to the fact that it does not constrain either the CPU or the compiler. 393 * That said, using RCU_INIT_POINTER() when you should have used 394 * rcu_assign_pointer() is a very bad thing that results in 395 * impossible-to-diagnose memory corruption. So please be careful. 396 * See the RCU_INIT_POINTER() comment header for details. 397 * 398 * Note that rcu_assign_pointer() evaluates each of its arguments only 399 * once, appearances notwithstanding. One of the "extra" evaluations 400 * is in typeof() and the other visible only to sparse (__CHECKER__), 401 * neither of which actually execute the argument. As with most cpp 402 * macros, this execute-arguments-only-once property is important, so 403 * please be careful when making changes to rcu_assign_pointer() and the 404 * other macros that it invokes. 405 */ 406#define rcu_assign_pointer(p, v) \ 407({ \ 408 uintptr_t _r_a_p__v = (uintptr_t)(v); \ 409 \ 410 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \ 411 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \ 412 else \ 413 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \ 414 _r_a_p__v; \ 415}) 416 417/** 418 * rcu_swap_protected() - swap an RCU and a regular pointer 419 * @rcu_ptr: RCU pointer 420 * @ptr: regular pointer 421 * @c: the conditions under which the dereference will take place 422 * 423 * Perform swap(@rcu_ptr, @ptr) where @rcu_ptr is an RCU-annotated pointer and 424 * @c is the argument that is passed to the rcu_dereference_protected() call 425 * used to read that pointer. 426 */ 427#define rcu_swap_protected(rcu_ptr, ptr, c) do { \ 428 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \ 429 rcu_assign_pointer((rcu_ptr), (ptr)); \ 430 (ptr) = __tmp; \ 431} while (0) 432 433/** 434 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 435 * @p: The pointer to read 436 * 437 * Return the value of the specified RCU-protected pointer, but omit the 438 * lockdep checks for being in an RCU read-side critical section. This is 439 * useful when the value of this pointer is accessed, but the pointer is 440 * not dereferenced, for example, when testing an RCU-protected pointer 441 * against NULL. Although rcu_access_pointer() may also be used in cases 442 * where update-side locks prevent the value of the pointer from changing, 443 * you should instead use rcu_dereference_protected() for this use case. 444 * 445 * It is also permissible to use rcu_access_pointer() when read-side 446 * access to the pointer was removed at least one grace period ago, as 447 * is the case in the context of the RCU callback that is freeing up 448 * the data, or after a synchronize_rcu() returns. This can be useful 449 * when tearing down multi-linked structures after a grace period 450 * has elapsed. 451 */ 452#define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) 453 454/** 455 * rcu_dereference_check() - rcu_dereference with debug checking 456 * @p: The pointer to read, prior to dereferencing 457 * @c: The conditions under which the dereference will take place 458 * 459 * Do an rcu_dereference(), but check that the conditions under which the 460 * dereference will take place are correct. Typically the conditions 461 * indicate the various locking conditions that should be held at that 462 * point. The check should return true if the conditions are satisfied. 463 * An implicit check for being in an RCU read-side critical section 464 * (rcu_read_lock()) is included. 465 * 466 * For example: 467 * 468 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 469 * 470 * could be used to indicate to lockdep that foo->bar may only be dereferenced 471 * if either rcu_read_lock() is held, or that the lock required to replace 472 * the bar struct at foo->bar is held. 473 * 474 * Note that the list of conditions may also include indications of when a lock 475 * need not be held, for example during initialisation or destruction of the 476 * target struct: 477 * 478 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 479 * atomic_read(&foo->usage) == 0); 480 * 481 * Inserts memory barriers on architectures that require them 482 * (currently only the Alpha), prevents the compiler from refetching 483 * (and from merging fetches), and, more importantly, documents exactly 484 * which pointers are protected by RCU and checks that the pointer is 485 * annotated as __rcu. 486 */ 487#define rcu_dereference_check(p, c) \ 488 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu) 489 490/** 491 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 492 * @p: The pointer to read, prior to dereferencing 493 * @c: The conditions under which the dereference will take place 494 * 495 * This is the RCU-bh counterpart to rcu_dereference_check(). 496 */ 497#define rcu_dereference_bh_check(p, c) \ 498 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu) 499 500/** 501 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 502 * @p: The pointer to read, prior to dereferencing 503 * @c: The conditions under which the dereference will take place 504 * 505 * This is the RCU-sched counterpart to rcu_dereference_check(). 506 */ 507#define rcu_dereference_sched_check(p, c) \ 508 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \ 509 __rcu) 510 511/* 512 * The tracing infrastructure traces RCU (we want that), but unfortunately 513 * some of the RCU checks causes tracing to lock up the system. 514 * 515 * The no-tracing version of rcu_dereference_raw() must not call 516 * rcu_read_lock_held(). 517 */ 518#define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu) 519 520/** 521 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 522 * @p: The pointer to read, prior to dereferencing 523 * @c: The conditions under which the dereference will take place 524 * 525 * Return the value of the specified RCU-protected pointer, but omit 526 * the READ_ONCE(). This is useful in cases where update-side locks 527 * prevent the value of the pointer from changing. Please note that this 528 * primitive does *not* prevent the compiler from repeating this reference 529 * or combining it with other references, so it should not be used without 530 * protection of appropriate locks. 531 * 532 * This function is only for update-side use. Using this function 533 * when protected only by rcu_read_lock() will result in infrequent 534 * but very ugly failures. 535 */ 536#define rcu_dereference_protected(p, c) \ 537 __rcu_dereference_protected((p), (c), __rcu) 538 539 540/** 541 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 542 * @p: The pointer to read, prior to dereferencing 543 * 544 * This is a simple wrapper around rcu_dereference_check(). 545 */ 546#define rcu_dereference(p) rcu_dereference_check(p, 0) 547 548/** 549 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 550 * @p: The pointer to read, prior to dereferencing 551 * 552 * Makes rcu_dereference_check() do the dirty work. 553 */ 554#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 555 556/** 557 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 558 * @p: The pointer to read, prior to dereferencing 559 * 560 * Makes rcu_dereference_check() do the dirty work. 561 */ 562#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 563 564/** 565 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism 566 * @p: The pointer to hand off 567 * 568 * This is simply an identity function, but it documents where a pointer 569 * is handed off from RCU to some other synchronization mechanism, for 570 * example, reference counting or locking. In C11, it would map to 571 * kill_dependency(). It could be used as follows: 572 * `` 573 * rcu_read_lock(); 574 * p = rcu_dereference(gp); 575 * long_lived = is_long_lived(p); 576 * if (long_lived) { 577 * if (!atomic_inc_not_zero(p->refcnt)) 578 * long_lived = false; 579 * else 580 * p = rcu_pointer_handoff(p); 581 * } 582 * rcu_read_unlock(); 583 *`` 584 */ 585#define rcu_pointer_handoff(p) (p) 586 587/** 588 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 589 * 590 * When synchronize_rcu() is invoked on one CPU while other CPUs 591 * are within RCU read-side critical sections, then the 592 * synchronize_rcu() is guaranteed to block until after all the other 593 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 594 * on one CPU while other CPUs are within RCU read-side critical 595 * sections, invocation of the corresponding RCU callback is deferred 596 * until after the all the other CPUs exit their critical sections. 597 * 598 * Note, however, that RCU callbacks are permitted to run concurrently 599 * with new RCU read-side critical sections. One way that this can happen 600 * is via the following sequence of events: (1) CPU 0 enters an RCU 601 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 602 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 603 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 604 * callback is invoked. This is legal, because the RCU read-side critical 605 * section that was running concurrently with the call_rcu() (and which 606 * therefore might be referencing something that the corresponding RCU 607 * callback would free up) has completed before the corresponding 608 * RCU callback is invoked. 609 * 610 * RCU read-side critical sections may be nested. Any deferred actions 611 * will be deferred until the outermost RCU read-side critical section 612 * completes. 613 * 614 * You can avoid reading and understanding the next paragraph by 615 * following this rule: don't put anything in an rcu_read_lock() RCU 616 * read-side critical section that would block in a !PREEMPT kernel. 617 * But if you want the full story, read on! 618 * 619 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), 620 * it is illegal to block while in an RCU read-side critical section. 621 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT 622 * kernel builds, RCU read-side critical sections may be preempted, 623 * but explicit blocking is illegal. Finally, in preemptible RCU 624 * implementations in real-time (with -rt patchset) kernel builds, RCU 625 * read-side critical sections may be preempted and they may also block, but 626 * only when acquiring spinlocks that are subject to priority inheritance. 627 */ 628static inline void rcu_read_lock(void) 629{ 630 __rcu_read_lock(); 631 __acquire(RCU); 632 rcu_lock_acquire(&rcu_lock_map); 633 RCU_LOCKDEP_WARN(!rcu_is_watching(), 634 "rcu_read_lock() used illegally while idle"); 635} 636 637/* 638 * So where is rcu_write_lock()? It does not exist, as there is no 639 * way for writers to lock out RCU readers. This is a feature, not 640 * a bug -- this property is what provides RCU's performance benefits. 641 * Of course, writers must coordinate with each other. The normal 642 * spinlock primitives work well for this, but any other technique may be 643 * used as well. RCU does not care how the writers keep out of each 644 * others' way, as long as they do so. 645 */ 646 647/** 648 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 649 * 650 * In most situations, rcu_read_unlock() is immune from deadlock. 651 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() 652 * is responsible for deboosting, which it does via rt_mutex_unlock(). 653 * Unfortunately, this function acquires the scheduler's runqueue and 654 * priority-inheritance spinlocks. This means that deadlock could result 655 * if the caller of rcu_read_unlock() already holds one of these locks or 656 * any lock that is ever acquired while holding them; or any lock which 657 * can be taken from interrupt context because rcu_boost()->rt_mutex_lock() 658 * does not disable irqs while taking ->wait_lock. 659 * 660 * That said, RCU readers are never priority boosted unless they were 661 * preempted. Therefore, one way to avoid deadlock is to make sure 662 * that preemption never happens within any RCU read-side critical 663 * section whose outermost rcu_read_unlock() is called with one of 664 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in 665 * a number of ways, for example, by invoking preempt_disable() before 666 * critical section's outermost rcu_read_lock(). 667 * 668 * Given that the set of locks acquired by rt_mutex_unlock() might change 669 * at any time, a somewhat more future-proofed approach is to make sure 670 * that that preemption never happens within any RCU read-side critical 671 * section whose outermost rcu_read_unlock() is called with irqs disabled. 672 * This approach relies on the fact that rt_mutex_unlock() currently only 673 * acquires irq-disabled locks. 674 * 675 * The second of these two approaches is best in most situations, 676 * however, the first approach can also be useful, at least to those 677 * developers willing to keep abreast of the set of locks acquired by 678 * rt_mutex_unlock(). 679 * 680 * See rcu_read_lock() for more information. 681 */ 682static inline void rcu_read_unlock(void) 683{ 684 RCU_LOCKDEP_WARN(!rcu_is_watching(), 685 "rcu_read_unlock() used illegally while idle"); 686 __release(RCU); 687 __rcu_read_unlock(); 688 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ 689} 690 691/** 692 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 693 * 694 * This is equivalent of rcu_read_lock(), but to be used when updates 695 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since 696 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a 697 * softirq handler to be a quiescent state, a process in RCU read-side 698 * critical section must be protected by disabling softirqs. Read-side 699 * critical sections in interrupt context can use just rcu_read_lock(), 700 * though this should at least be commented to avoid confusing people 701 * reading the code. 702 * 703 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 704 * must occur in the same context, for example, it is illegal to invoke 705 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 706 * was invoked from some other task. 707 */ 708static inline void rcu_read_lock_bh(void) 709{ 710 local_bh_disable(); 711 __acquire(RCU_BH); 712 rcu_lock_acquire(&rcu_bh_lock_map); 713 RCU_LOCKDEP_WARN(!rcu_is_watching(), 714 "rcu_read_lock_bh() used illegally while idle"); 715} 716 717/* 718 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section 719 * 720 * See rcu_read_lock_bh() for more information. 721 */ 722static inline void rcu_read_unlock_bh(void) 723{ 724 RCU_LOCKDEP_WARN(!rcu_is_watching(), 725 "rcu_read_unlock_bh() used illegally while idle"); 726 rcu_lock_release(&rcu_bh_lock_map); 727 __release(RCU_BH); 728 local_bh_enable(); 729} 730 731/** 732 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 733 * 734 * This is equivalent of rcu_read_lock(), but to be used when updates 735 * are being done using call_rcu_sched() or synchronize_rcu_sched(). 736 * Read-side critical sections can also be introduced by anything that 737 * disables preemption, including local_irq_disable() and friends. 738 * 739 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 740 * must occur in the same context, for example, it is illegal to invoke 741 * rcu_read_unlock_sched() from process context if the matching 742 * rcu_read_lock_sched() was invoked from an NMI handler. 743 */ 744static inline void rcu_read_lock_sched(void) 745{ 746 preempt_disable(); 747 __acquire(RCU_SCHED); 748 rcu_lock_acquire(&rcu_sched_lock_map); 749 RCU_LOCKDEP_WARN(!rcu_is_watching(), 750 "rcu_read_lock_sched() used illegally while idle"); 751} 752 753/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 754static inline notrace void rcu_read_lock_sched_notrace(void) 755{ 756 preempt_disable_notrace(); 757 __acquire(RCU_SCHED); 758} 759 760/* 761 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section 762 * 763 * See rcu_read_lock_sched for more information. 764 */ 765static inline void rcu_read_unlock_sched(void) 766{ 767 RCU_LOCKDEP_WARN(!rcu_is_watching(), 768 "rcu_read_unlock_sched() used illegally while idle"); 769 rcu_lock_release(&rcu_sched_lock_map); 770 __release(RCU_SCHED); 771 preempt_enable(); 772} 773 774/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 775static inline notrace void rcu_read_unlock_sched_notrace(void) 776{ 777 __release(RCU_SCHED); 778 preempt_enable_notrace(); 779} 780 781/** 782 * RCU_INIT_POINTER() - initialize an RCU protected pointer 783 * @p: The pointer to be initialized. 784 * @v: The value to initialized the pointer to. 785 * 786 * Initialize an RCU-protected pointer in special cases where readers 787 * do not need ordering constraints on the CPU or the compiler. These 788 * special cases are: 789 * 790 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or* 791 * 2. The caller has taken whatever steps are required to prevent 792 * RCU readers from concurrently accessing this pointer *or* 793 * 3. The referenced data structure has already been exposed to 794 * readers either at compile time or via rcu_assign_pointer() *and* 795 * 796 * a. You have not made *any* reader-visible changes to 797 * this structure since then *or* 798 * b. It is OK for readers accessing this structure from its 799 * new location to see the old state of the structure. (For 800 * example, the changes were to statistical counters or to 801 * other state where exact synchronization is not required.) 802 * 803 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 804 * result in impossible-to-diagnose memory corruption. As in the structures 805 * will look OK in crash dumps, but any concurrent RCU readers might 806 * see pre-initialized values of the referenced data structure. So 807 * please be very careful how you use RCU_INIT_POINTER()!!! 808 * 809 * If you are creating an RCU-protected linked structure that is accessed 810 * by a single external-to-structure RCU-protected pointer, then you may 811 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 812 * pointers, but you must use rcu_assign_pointer() to initialize the 813 * external-to-structure pointer *after* you have completely initialized 814 * the reader-accessible portions of the linked structure. 815 * 816 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 817 * ordering guarantees for either the CPU or the compiler. 818 */ 819#define RCU_INIT_POINTER(p, v) \ 820 do { \ 821 rcu_dereference_sparse(p, __rcu); \ 822 WRITE_ONCE(p, RCU_INITIALIZER(v)); \ 823 } while (0) 824 825/** 826 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 827 * @p: The pointer to be initialized. 828 * @v: The value to initialized the pointer to. 829 * 830 * GCC-style initialization for an RCU-protected pointer in a structure field. 831 */ 832#define RCU_POINTER_INITIALIZER(p, v) \ 833 .p = RCU_INITIALIZER(v) 834 835/* 836 * Does the specified offset indicate that the corresponding rcu_head 837 * structure can be handled by kfree_rcu()? 838 */ 839#define __is_kfree_rcu_offset(offset) ((offset) < 4096) 840 841/* 842 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. 843 */ 844#define __kfree_rcu(head, offset) \ 845 do { \ 846 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \ 847 kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \ 848 } while (0) 849 850/** 851 * kfree_rcu() - kfree an object after a grace period. 852 * @ptr: pointer to kfree 853 * @rcu_head: the name of the struct rcu_head within the type of @ptr. 854 * 855 * Many rcu callbacks functions just call kfree() on the base structure. 856 * These functions are trivial, but their size adds up, and furthermore 857 * when they are used in a kernel module, that module must invoke the 858 * high-latency rcu_barrier() function at module-unload time. 859 * 860 * The kfree_rcu() function handles this issue. Rather than encoding a 861 * function address in the embedded rcu_head structure, kfree_rcu() instead 862 * encodes the offset of the rcu_head structure within the base structure. 863 * Because the functions are not allowed in the low-order 4096 bytes of 864 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 865 * If the offset is larger than 4095 bytes, a compile-time error will 866 * be generated in __kfree_rcu(). If this error is triggered, you can 867 * either fall back to use of call_rcu() or rearrange the structure to 868 * position the rcu_head structure into the first 4096 bytes. 869 * 870 * Note that the allowable offset might decrease in the future, for example, 871 * to allow something like kmem_cache_free_rcu(). 872 * 873 * The BUILD_BUG_ON check must not involve any function calls, hence the 874 * checks are done in macros here. 875 */ 876#define kfree_rcu(ptr, rcu_head) \ 877 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head)) 878 879 880/* 881 * Place this after a lock-acquisition primitive to guarantee that 882 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies 883 * if the UNLOCK and LOCK are executed by the same CPU or if the 884 * UNLOCK and LOCK operate on the same lock variable. 885 */ 886#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE 887#define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */ 888#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 889#define smp_mb__after_unlock_lock() do { } while (0) 890#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 891 892 893#endif /* __LINUX_RCUPDATE_H */ 894