1/* SPDX-License-Identifier: GPL-2.0+ */ 2/* 3 * Read-Copy Update mechanism for mutual exclusion 4 * 5 * Copyright IBM Corporation, 2001 6 * 7 * Author: Dipankar Sarma <dipankar@in.ibm.com> 8 * 9 * Based on the original work by Paul McKenney <paulmck@vnet.ibm.com> 10 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. 11 * Papers: 12 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf 13 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001) 14 * 15 * For detailed explanation of Read-Copy Update mechanism see - 16 * http://lse.sourceforge.net/locking/rcupdate.html 17 * 18 */ 19 20#ifndef __LINUX_RCUPDATE_H 21#define __LINUX_RCUPDATE_H 22 23#include <linux/types.h> 24#include <linux/compiler.h> 25#include <linux/atomic.h> 26#include <linux/irqflags.h> 27#include <linux/preempt.h> 28#include <linux/bottom_half.h> 29#include <linux/lockdep.h> 30#include <asm/processor.h> 31#include <linux/cpumask.h> 32 33#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) 34#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) 35#define ulong2long(a) (*(long *)(&(a))) 36 37/* Exported common interfaces */ 38void call_rcu(struct rcu_head *head, rcu_callback_t func); 39void rcu_barrier_tasks(void); 40void rcu_barrier_tasks_rude(void); 41void synchronize_rcu(void); 42 43#ifdef CONFIG_PREEMPT_RCU 44 45void __rcu_read_lock(void); 46void __rcu_read_unlock(void); 47 48/* 49 * Defined as a macro as it is a very low level header included from 50 * areas that don't even know about current. This gives the rcu_read_lock() 51 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other 52 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. 53 */ 54#define rcu_preempt_depth() (current->rcu_read_lock_nesting) 55 56#else /* #ifdef CONFIG_PREEMPT_RCU */ 57 58#ifdef CONFIG_TINY_RCU 59#define rcu_read_unlock_strict() do { } while (0) 60#else 61void rcu_read_unlock_strict(void); 62#endif 63 64static inline void __rcu_read_lock(void) 65{ 66 preempt_disable(); 67} 68 69static inline void __rcu_read_unlock(void) 70{ 71 preempt_enable(); 72 rcu_read_unlock_strict(); 73} 74 75static inline int rcu_preempt_depth(void) 76{ 77 return 0; 78} 79 80#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 81 82/* Internal to kernel */ 83void rcu_init(void); 84extern int rcu_scheduler_active __read_mostly; 85void rcu_sched_clock_irq(int user); 86void rcu_report_dead(unsigned int cpu); 87void rcutree_migrate_callbacks(int cpu); 88 89#ifdef CONFIG_RCU_STALL_COMMON 90void rcu_sysrq_start(void); 91void rcu_sysrq_end(void); 92#else /* #ifdef CONFIG_RCU_STALL_COMMON */ 93static inline void rcu_sysrq_start(void) { } 94static inline void rcu_sysrq_end(void) { } 95#endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ 96 97#ifdef CONFIG_NO_HZ_FULL 98void rcu_user_enter(void); 99void rcu_user_exit(void); 100#else 101static inline void rcu_user_enter(void) { } 102static inline void rcu_user_exit(void) { } 103#endif /* CONFIG_NO_HZ_FULL */ 104 105#ifdef CONFIG_RCU_NOCB_CPU 106void rcu_init_nohz(void); 107#else /* #ifdef CONFIG_RCU_NOCB_CPU */ 108static inline void rcu_init_nohz(void) { } 109#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 110 111/** 112 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers 113 * @a: Code that RCU needs to pay attention to. 114 * 115 * RCU read-side critical sections are forbidden in the inner idle loop, 116 * that is, between the rcu_idle_enter() and the rcu_idle_exit() -- RCU 117 * will happily ignore any such read-side critical sections. However, 118 * things like powertop need tracepoints in the inner idle loop. 119 * 120 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) 121 * will tell RCU that it needs to pay attention, invoke its argument 122 * (in this example, calling the do_something_with_RCU() function), 123 * and then tell RCU to go back to ignoring this CPU. It is permissible 124 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is 125 * on the order of a million or so, even on 32-bit systems). It is 126 * not legal to block within RCU_NONIDLE(), nor is it permissible to 127 * transfer control either into or out of RCU_NONIDLE()'s statement. 128 */ 129#define RCU_NONIDLE(a) \ 130 do { \ 131 rcu_irq_enter_irqson(); \ 132 do { a; } while (0); \ 133 rcu_irq_exit_irqson(); \ 134 } while (0) 135 136/* 137 * Note a quasi-voluntary context switch for RCU-tasks's benefit. 138 * This is a macro rather than an inline function to avoid #include hell. 139 */ 140#ifdef CONFIG_TASKS_RCU_GENERIC 141 142# ifdef CONFIG_TASKS_RCU 143# define rcu_tasks_classic_qs(t, preempt) \ 144 do { \ 145 if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout)) \ 146 WRITE_ONCE((t)->rcu_tasks_holdout, false); \ 147 } while (0) 148void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func); 149void synchronize_rcu_tasks(void); 150# else 151# define rcu_tasks_classic_qs(t, preempt) do { } while (0) 152# define call_rcu_tasks call_rcu 153# define synchronize_rcu_tasks synchronize_rcu 154# endif 155 156# ifdef CONFIG_TASKS_RCU_TRACE 157# define rcu_tasks_trace_qs(t) \ 158 do { \ 159 if (!likely(READ_ONCE((t)->trc_reader_checked)) && \ 160 !unlikely(READ_ONCE((t)->trc_reader_nesting))) { \ 161 smp_store_release(&(t)->trc_reader_checked, true); \ 162 smp_mb(); /* Readers partitioned by store. */ \ 163 } \ 164 } while (0) 165# else 166# define rcu_tasks_trace_qs(t) do { } while (0) 167# endif 168 169#define rcu_tasks_qs(t, preempt) \ 170do { \ 171 rcu_tasks_classic_qs((t), (preempt)); \ 172 rcu_tasks_trace_qs((t)); \ 173} while (0) 174 175# ifdef CONFIG_TASKS_RUDE_RCU 176void call_rcu_tasks_rude(struct rcu_head *head, rcu_callback_t func); 177void synchronize_rcu_tasks_rude(void); 178# endif 179 180#define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false) 181void exit_tasks_rcu_start(void); 182void exit_tasks_rcu_finish(void); 183#else /* #ifdef CONFIG_TASKS_RCU_GENERIC */ 184#define rcu_tasks_qs(t, preempt) do { } while (0) 185#define rcu_note_voluntary_context_switch(t) do { } while (0) 186#define call_rcu_tasks call_rcu 187#define synchronize_rcu_tasks synchronize_rcu 188static inline void exit_tasks_rcu_start(void) { } 189static inline void exit_tasks_rcu_finish(void) { } 190#endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */ 191 192/** 193 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU 194 * 195 * This macro resembles cond_resched(), except that it is defined to 196 * report potential quiescent states to RCU-tasks even if the cond_resched() 197 * machinery were to be shut off, as some advocate for PREEMPTION kernels. 198 */ 199#define cond_resched_tasks_rcu_qs() \ 200do { \ 201 rcu_tasks_qs(current, false); \ 202 cond_resched(); \ 203} while (0) 204 205/* 206 * Infrastructure to implement the synchronize_() primitives in 207 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 208 */ 209 210#if defined(CONFIG_TREE_RCU) 211#include <linux/rcutree.h> 212#elif defined(CONFIG_TINY_RCU) 213#include <linux/rcutiny.h> 214#else 215#error "Unknown RCU implementation specified to kernel configuration" 216#endif 217 218/* 219 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls 220 * are needed for dynamic initialization and destruction of rcu_head 221 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for 222 * dynamic initialization and destruction of statically allocated rcu_head 223 * structures. However, rcu_head structures allocated dynamically in the 224 * heap don't need any initialization. 225 */ 226#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 227void init_rcu_head(struct rcu_head *head); 228void destroy_rcu_head(struct rcu_head *head); 229void init_rcu_head_on_stack(struct rcu_head *head); 230void destroy_rcu_head_on_stack(struct rcu_head *head); 231#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 232static inline void init_rcu_head(struct rcu_head *head) { } 233static inline void destroy_rcu_head(struct rcu_head *head) { } 234static inline void init_rcu_head_on_stack(struct rcu_head *head) { } 235static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { } 236#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 237 238#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 239bool rcu_lockdep_current_cpu_online(void); 240#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 241static inline bool rcu_lockdep_current_cpu_online(void) { return true; } 242#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 243 244#ifdef CONFIG_DEBUG_LOCK_ALLOC 245 246static inline void rcu_lock_acquire(struct lockdep_map *map) 247{ 248 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); 249} 250 251static inline void rcu_lock_release(struct lockdep_map *map) 252{ 253 lock_release(map, _THIS_IP_); 254} 255 256extern struct lockdep_map rcu_lock_map; 257extern struct lockdep_map rcu_bh_lock_map; 258extern struct lockdep_map rcu_sched_lock_map; 259extern struct lockdep_map rcu_callback_map; 260int debug_lockdep_rcu_enabled(void); 261int rcu_read_lock_held(void); 262int rcu_read_lock_bh_held(void); 263int rcu_read_lock_sched_held(void); 264int rcu_read_lock_any_held(void); 265 266#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 267 268# define rcu_lock_acquire(a) do { } while (0) 269# define rcu_lock_release(a) do { } while (0) 270 271static inline int rcu_read_lock_held(void) 272{ 273 return 1; 274} 275 276static inline int rcu_read_lock_bh_held(void) 277{ 278 return 1; 279} 280 281static inline int rcu_read_lock_sched_held(void) 282{ 283 return !preemptible(); 284} 285 286static inline int rcu_read_lock_any_held(void) 287{ 288 return !preemptible(); 289} 290 291#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 292 293#ifdef CONFIG_PROVE_RCU 294 295/** 296 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met 297 * @c: condition to check 298 * @s: informative message 299 */ 300#define RCU_LOCKDEP_WARN(c, s) \ 301 do { \ 302 static bool __section(".data.unlikely") __warned; \ 303 if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \ 304 __warned = true; \ 305 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 306 } \ 307 } while (0) 308 309#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 310static inline void rcu_preempt_sleep_check(void) 311{ 312 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), 313 "Illegal context switch in RCU read-side critical section"); 314} 315#else /* #ifdef CONFIG_PROVE_RCU */ 316static inline void rcu_preempt_sleep_check(void) { } 317#endif /* #else #ifdef CONFIG_PROVE_RCU */ 318 319#define rcu_sleep_check() \ 320 do { \ 321 rcu_preempt_sleep_check(); \ 322 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ 323 "Illegal context switch in RCU-bh read-side critical section"); \ 324 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ 325 "Illegal context switch in RCU-sched read-side critical section"); \ 326 } while (0) 327 328#else /* #ifdef CONFIG_PROVE_RCU */ 329 330#define RCU_LOCKDEP_WARN(c, s) do { } while (0) 331#define rcu_sleep_check() do { } while (0) 332 333#endif /* #else #ifdef CONFIG_PROVE_RCU */ 334 335/* 336 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 337 * and rcu_assign_pointer(). Some of these could be folded into their 338 * callers, but they are left separate in order to ease introduction of 339 * multiple pointers markings to match different RCU implementations 340 * (e.g., __srcu), should this make sense in the future. 341 */ 342 343#ifdef __CHECKER__ 344#define rcu_check_sparse(p, space) \ 345 ((void)(((typeof(*p) space *)p) == p)) 346#else /* #ifdef __CHECKER__ */ 347#define rcu_check_sparse(p, space) 348#endif /* #else #ifdef __CHECKER__ */ 349 350#define __rcu_access_pointer(p, space) \ 351({ \ 352 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 353 rcu_check_sparse(p, space); \ 354 ((typeof(*p) __force __kernel *)(_________p1)); \ 355}) 356#define __rcu_dereference_check(p, c, space) \ 357({ \ 358 /* Dependency order vs. p above. */ \ 359 typeof(*p) *________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 360 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ 361 rcu_check_sparse(p, space); \ 362 ((typeof(*p) __force __kernel *)(________p1)); \ 363}) 364#define __rcu_dereference_protected(p, c, space) \ 365({ \ 366 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ 367 rcu_check_sparse(p, space); \ 368 ((typeof(*p) __force __kernel *)(p)); \ 369}) 370#define rcu_dereference_raw(p) \ 371({ \ 372 /* Dependency order vs. p above. */ \ 373 typeof(p) ________p1 = READ_ONCE(p); \ 374 ((typeof(*p) __force __kernel *)(________p1)); \ 375}) 376 377/** 378 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 379 * @v: The value to statically initialize with. 380 */ 381#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 382 383/** 384 * rcu_assign_pointer() - assign to RCU-protected pointer 385 * @p: pointer to assign to 386 * @v: value to assign (publish) 387 * 388 * Assigns the specified value to the specified RCU-protected 389 * pointer, ensuring that any concurrent RCU readers will see 390 * any prior initialization. 391 * 392 * Inserts memory barriers on architectures that require them 393 * (which is most of them), and also prevents the compiler from 394 * reordering the code that initializes the structure after the pointer 395 * assignment. More importantly, this call documents which pointers 396 * will be dereferenced by RCU read-side code. 397 * 398 * In some special cases, you may use RCU_INIT_POINTER() instead 399 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 400 * to the fact that it does not constrain either the CPU or the compiler. 401 * That said, using RCU_INIT_POINTER() when you should have used 402 * rcu_assign_pointer() is a very bad thing that results in 403 * impossible-to-diagnose memory corruption. So please be careful. 404 * See the RCU_INIT_POINTER() comment header for details. 405 * 406 * Note that rcu_assign_pointer() evaluates each of its arguments only 407 * once, appearances notwithstanding. One of the "extra" evaluations 408 * is in typeof() and the other visible only to sparse (__CHECKER__), 409 * neither of which actually execute the argument. As with most cpp 410 * macros, this execute-arguments-only-once property is important, so 411 * please be careful when making changes to rcu_assign_pointer() and the 412 * other macros that it invokes. 413 */ 414#define rcu_assign_pointer(p, v) \ 415do { \ 416 uintptr_t _r_a_p__v = (uintptr_t)(v); \ 417 rcu_check_sparse(p, __rcu); \ 418 \ 419 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \ 420 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \ 421 else \ 422 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \ 423} while (0) 424 425/** 426 * rcu_replace_pointer() - replace an RCU pointer, returning its old value 427 * @rcu_ptr: RCU pointer, whose old value is returned 428 * @ptr: regular pointer 429 * @c: the lockdep conditions under which the dereference will take place 430 * 431 * Perform a replacement, where @rcu_ptr is an RCU-annotated 432 * pointer and @c is the lockdep argument that is passed to the 433 * rcu_dereference_protected() call used to read that pointer. The old 434 * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr. 435 */ 436#define rcu_replace_pointer(rcu_ptr, ptr, c) \ 437({ \ 438 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \ 439 rcu_assign_pointer((rcu_ptr), (ptr)); \ 440 __tmp; \ 441}) 442 443/** 444 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 445 * @p: The pointer to read 446 * 447 * Return the value of the specified RCU-protected pointer, but omit the 448 * lockdep checks for being in an RCU read-side critical section. This is 449 * useful when the value of this pointer is accessed, but the pointer is 450 * not dereferenced, for example, when testing an RCU-protected pointer 451 * against NULL. Although rcu_access_pointer() may also be used in cases 452 * where update-side locks prevent the value of the pointer from changing, 453 * you should instead use rcu_dereference_protected() for this use case. 454 * 455 * It is also permissible to use rcu_access_pointer() when read-side 456 * access to the pointer was removed at least one grace period ago, as 457 * is the case in the context of the RCU callback that is freeing up 458 * the data, or after a synchronize_rcu() returns. This can be useful 459 * when tearing down multi-linked structures after a grace period 460 * has elapsed. 461 */ 462#define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) 463 464/** 465 * rcu_dereference_check() - rcu_dereference with debug checking 466 * @p: The pointer to read, prior to dereferencing 467 * @c: The conditions under which the dereference will take place 468 * 469 * Do an rcu_dereference(), but check that the conditions under which the 470 * dereference will take place are correct. Typically the conditions 471 * indicate the various locking conditions that should be held at that 472 * point. The check should return true if the conditions are satisfied. 473 * An implicit check for being in an RCU read-side critical section 474 * (rcu_read_lock()) is included. 475 * 476 * For example: 477 * 478 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 479 * 480 * could be used to indicate to lockdep that foo->bar may only be dereferenced 481 * if either rcu_read_lock() is held, or that the lock required to replace 482 * the bar struct at foo->bar is held. 483 * 484 * Note that the list of conditions may also include indications of when a lock 485 * need not be held, for example during initialisation or destruction of the 486 * target struct: 487 * 488 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 489 * atomic_read(&foo->usage) == 0); 490 * 491 * Inserts memory barriers on architectures that require them 492 * (currently only the Alpha), prevents the compiler from refetching 493 * (and from merging fetches), and, more importantly, documents exactly 494 * which pointers are protected by RCU and checks that the pointer is 495 * annotated as __rcu. 496 */ 497#define rcu_dereference_check(p, c) \ 498 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu) 499 500/** 501 * rcu_dereference_bh_check() - rcu_dereference_bh 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-bh counterpart to rcu_dereference_check(). 506 */ 507#define rcu_dereference_bh_check(p, c) \ 508 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu) 509 510/** 511 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 512 * @p: The pointer to read, prior to dereferencing 513 * @c: The conditions under which the dereference will take place 514 * 515 * This is the RCU-sched counterpart to rcu_dereference_check(). 516 */ 517#define rcu_dereference_sched_check(p, c) \ 518 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \ 519 __rcu) 520 521/* 522 * The tracing infrastructure traces RCU (we want that), but unfortunately 523 * some of the RCU checks causes tracing to lock up the system. 524 * 525 * The no-tracing version of rcu_dereference_raw() must not call 526 * rcu_read_lock_held(). 527 */ 528#define rcu_dereference_raw_check(p) __rcu_dereference_check((p), 1, __rcu) 529 530/** 531 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 532 * @p: The pointer to read, prior to dereferencing 533 * @c: The conditions under which the dereference will take place 534 * 535 * Return the value of the specified RCU-protected pointer, but omit 536 * the READ_ONCE(). This is useful in cases where update-side locks 537 * prevent the value of the pointer from changing. Please note that this 538 * primitive does *not* prevent the compiler from repeating this reference 539 * or combining it with other references, so it should not be used without 540 * protection of appropriate locks. 541 * 542 * This function is only for update-side use. Using this function 543 * when protected only by rcu_read_lock() will result in infrequent 544 * but very ugly failures. 545 */ 546#define rcu_dereference_protected(p, c) \ 547 __rcu_dereference_protected((p), (c), __rcu) 548 549 550/** 551 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 552 * @p: The pointer to read, prior to dereferencing 553 * 554 * This is a simple wrapper around rcu_dereference_check(). 555 */ 556#define rcu_dereference(p) rcu_dereference_check(p, 0) 557 558/** 559 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 560 * @p: The pointer to read, prior to dereferencing 561 * 562 * Makes rcu_dereference_check() do the dirty work. 563 */ 564#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 565 566/** 567 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 568 * @p: The pointer to read, prior to dereferencing 569 * 570 * Makes rcu_dereference_check() do the dirty work. 571 */ 572#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 573 574/** 575 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism 576 * @p: The pointer to hand off 577 * 578 * This is simply an identity function, but it documents where a pointer 579 * is handed off from RCU to some other synchronization mechanism, for 580 * example, reference counting or locking. In C11, it would map to 581 * kill_dependency(). It could be used as follows:: 582 * 583 * rcu_read_lock(); 584 * p = rcu_dereference(gp); 585 * long_lived = is_long_lived(p); 586 * if (long_lived) { 587 * if (!atomic_inc_not_zero(p->refcnt)) 588 * long_lived = false; 589 * else 590 * p = rcu_pointer_handoff(p); 591 * } 592 * rcu_read_unlock(); 593 */ 594#define rcu_pointer_handoff(p) (p) 595 596/** 597 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 598 * 599 * When synchronize_rcu() is invoked on one CPU while other CPUs 600 * are within RCU read-side critical sections, then the 601 * synchronize_rcu() is guaranteed to block until after all the other 602 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 603 * on one CPU while other CPUs are within RCU read-side critical 604 * sections, invocation of the corresponding RCU callback is deferred 605 * until after the all the other CPUs exit their critical sections. 606 * 607 * Note, however, that RCU callbacks are permitted to run concurrently 608 * with new RCU read-side critical sections. One way that this can happen 609 * is via the following sequence of events: (1) CPU 0 enters an RCU 610 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 611 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 612 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 613 * callback is invoked. This is legal, because the RCU read-side critical 614 * section that was running concurrently with the call_rcu() (and which 615 * therefore might be referencing something that the corresponding RCU 616 * callback would free up) has completed before the corresponding 617 * RCU callback is invoked. 618 * 619 * RCU read-side critical sections may be nested. Any deferred actions 620 * will be deferred until the outermost RCU read-side critical section 621 * completes. 622 * 623 * You can avoid reading and understanding the next paragraph by 624 * following this rule: don't put anything in an rcu_read_lock() RCU 625 * read-side critical section that would block in a !PREEMPTION kernel. 626 * But if you want the full story, read on! 627 * 628 * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU), 629 * it is illegal to block while in an RCU read-side critical section. 630 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION 631 * kernel builds, RCU read-side critical sections may be preempted, 632 * but explicit blocking is illegal. Finally, in preemptible RCU 633 * implementations in real-time (with -rt patchset) kernel builds, RCU 634 * read-side critical sections may be preempted and they may also block, but 635 * only when acquiring spinlocks that are subject to priority inheritance. 636 */ 637static __always_inline void rcu_read_lock(void) 638{ 639 __rcu_read_lock(); 640 __acquire(RCU); 641 rcu_lock_acquire(&rcu_lock_map); 642 RCU_LOCKDEP_WARN(!rcu_is_watching(), 643 "rcu_read_lock() used illegally while idle"); 644} 645 646/* 647 * So where is rcu_write_lock()? It does not exist, as there is no 648 * way for writers to lock out RCU readers. This is a feature, not 649 * a bug -- this property is what provides RCU's performance benefits. 650 * Of course, writers must coordinate with each other. The normal 651 * spinlock primitives work well for this, but any other technique may be 652 * used as well. RCU does not care how the writers keep out of each 653 * others' way, as long as they do so. 654 */ 655 656/** 657 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 658 * 659 * In most situations, rcu_read_unlock() is immune from deadlock. 660 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() 661 * is responsible for deboosting, which it does via rt_mutex_unlock(). 662 * Unfortunately, this function acquires the scheduler's runqueue and 663 * priority-inheritance spinlocks. This means that deadlock could result 664 * if the caller of rcu_read_unlock() already holds one of these locks or 665 * any lock that is ever acquired while holding them. 666 * 667 * That said, RCU readers are never priority boosted unless they were 668 * preempted. Therefore, one way to avoid deadlock is to make sure 669 * that preemption never happens within any RCU read-side critical 670 * section whose outermost rcu_read_unlock() is called with one of 671 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in 672 * a number of ways, for example, by invoking preempt_disable() before 673 * critical section's outermost rcu_read_lock(). 674 * 675 * Given that the set of locks acquired by rt_mutex_unlock() might change 676 * at any time, a somewhat more future-proofed approach is to make sure 677 * that that preemption never happens within any RCU read-side critical 678 * section whose outermost rcu_read_unlock() is called with irqs disabled. 679 * This approach relies on the fact that rt_mutex_unlock() currently only 680 * acquires irq-disabled locks. 681 * 682 * The second of these two approaches is best in most situations, 683 * however, the first approach can also be useful, at least to those 684 * developers willing to keep abreast of the set of locks acquired by 685 * rt_mutex_unlock(). 686 * 687 * See rcu_read_lock() for more information. 688 */ 689static inline void rcu_read_unlock(void) 690{ 691 RCU_LOCKDEP_WARN(!rcu_is_watching(), 692 "rcu_read_unlock() used illegally while idle"); 693 __release(RCU); 694 __rcu_read_unlock(); 695 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ 696} 697 698/** 699 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 700 * 701 * This is equivalent of rcu_read_lock(), but also disables softirqs. 702 * Note that anything else that disables softirqs can also serve as 703 * an RCU read-side critical section. 704 * 705 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 706 * must occur in the same context, for example, it is illegal to invoke 707 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 708 * was invoked from some other task. 709 */ 710static inline void rcu_read_lock_bh(void) 711{ 712 local_bh_disable(); 713 __acquire(RCU_BH); 714 rcu_lock_acquire(&rcu_bh_lock_map); 715 RCU_LOCKDEP_WARN(!rcu_is_watching(), 716 "rcu_read_lock_bh() used illegally while idle"); 717} 718 719/** 720 * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section 721 * 722 * See rcu_read_lock_bh() for more information. 723 */ 724static inline void rcu_read_unlock_bh(void) 725{ 726 RCU_LOCKDEP_WARN(!rcu_is_watching(), 727 "rcu_read_unlock_bh() used illegally while idle"); 728 rcu_lock_release(&rcu_bh_lock_map); 729 __release(RCU_BH); 730 local_bh_enable(); 731} 732 733/** 734 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 735 * 736 * This is equivalent of rcu_read_lock(), but disables preemption. 737 * Read-side critical sections can also be introduced by anything else 738 * that disables preemption, including local_irq_disable() and friends. 739 * 740 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 741 * must occur in the same context, for example, it is illegal to invoke 742 * rcu_read_unlock_sched() from process context if the matching 743 * rcu_read_lock_sched() was invoked from an NMI handler. 744 */ 745static inline void rcu_read_lock_sched(void) 746{ 747 preempt_disable(); 748 __acquire(RCU_SCHED); 749 rcu_lock_acquire(&rcu_sched_lock_map); 750 RCU_LOCKDEP_WARN(!rcu_is_watching(), 751 "rcu_read_lock_sched() used illegally while idle"); 752} 753 754/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 755static inline notrace void rcu_read_lock_sched_notrace(void) 756{ 757 preempt_disable_notrace(); 758 __acquire(RCU_SCHED); 759} 760 761/** 762 * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section 763 * 764 * See rcu_read_lock_sched() for more information. 765 */ 766static inline void rcu_read_unlock_sched(void) 767{ 768 RCU_LOCKDEP_WARN(!rcu_is_watching(), 769 "rcu_read_unlock_sched() used illegally while idle"); 770 rcu_lock_release(&rcu_sched_lock_map); 771 __release(RCU_SCHED); 772 preempt_enable(); 773} 774 775/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 776static inline notrace void rcu_read_unlock_sched_notrace(void) 777{ 778 __release(RCU_SCHED); 779 preempt_enable_notrace(); 780} 781 782/** 783 * RCU_INIT_POINTER() - initialize an RCU protected pointer 784 * @p: The pointer to be initialized. 785 * @v: The value to initialized the pointer to. 786 * 787 * Initialize an RCU-protected pointer in special cases where readers 788 * do not need ordering constraints on the CPU or the compiler. These 789 * special cases are: 790 * 791 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or* 792 * 2. The caller has taken whatever steps are required to prevent 793 * RCU readers from concurrently accessing this pointer *or* 794 * 3. The referenced data structure has already been exposed to 795 * readers either at compile time or via rcu_assign_pointer() *and* 796 * 797 * a. You have not made *any* reader-visible changes to 798 * this structure since then *or* 799 * b. It is OK for readers accessing this structure from its 800 * new location to see the old state of the structure. (For 801 * example, the changes were to statistical counters or to 802 * other state where exact synchronization is not required.) 803 * 804 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 805 * result in impossible-to-diagnose memory corruption. As in the structures 806 * will look OK in crash dumps, but any concurrent RCU readers might 807 * see pre-initialized values of the referenced data structure. So 808 * please be very careful how you use RCU_INIT_POINTER()!!! 809 * 810 * If you are creating an RCU-protected linked structure that is accessed 811 * by a single external-to-structure RCU-protected pointer, then you may 812 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 813 * pointers, but you must use rcu_assign_pointer() to initialize the 814 * external-to-structure pointer *after* you have completely initialized 815 * the reader-accessible portions of the linked structure. 816 * 817 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 818 * ordering guarantees for either the CPU or the compiler. 819 */ 820#define RCU_INIT_POINTER(p, v) \ 821 do { \ 822 rcu_check_sparse(p, __rcu); \ 823 WRITE_ONCE(p, RCU_INITIALIZER(v)); \ 824 } while (0) 825 826/** 827 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 828 * @p: The pointer to be initialized. 829 * @v: The value to initialized the pointer to. 830 * 831 * GCC-style initialization for an RCU-protected pointer in a structure field. 832 */ 833#define RCU_POINTER_INITIALIZER(p, v) \ 834 .p = RCU_INITIALIZER(v) 835 836/* 837 * Does the specified offset indicate that the corresponding rcu_head 838 * structure can be handled by kvfree_rcu()? 839 */ 840#define __is_kvfree_rcu_offset(offset) ((offset) < 4096) 841 842/* 843 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. 844 */ 845#define __kvfree_rcu(head, offset) \ 846 do { \ 847 BUILD_BUG_ON(!__is_kvfree_rcu_offset(offset)); \ 848 kvfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \ 849 } while (0) 850 851/** 852 * kfree_rcu() - kfree an object after a grace period. 853 * @ptr: pointer to kfree 854 * @rhf: the name of the struct rcu_head within the type of @ptr. 855 * 856 * Many rcu callbacks functions just call kfree() on the base structure. 857 * These functions are trivial, but their size adds up, and furthermore 858 * when they are used in a kernel module, that module must invoke the 859 * high-latency rcu_barrier() function at module-unload time. 860 * 861 * The kfree_rcu() function handles this issue. Rather than encoding a 862 * function address in the embedded rcu_head structure, kfree_rcu() instead 863 * encodes the offset of the rcu_head structure within the base structure. 864 * Because the functions are not allowed in the low-order 4096 bytes of 865 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 866 * If the offset is larger than 4095 bytes, a compile-time error will 867 * be generated in __kvfree_rcu(). If this error is triggered, you can 868 * either fall back to use of call_rcu() or rearrange the structure to 869 * position the rcu_head structure into the first 4096 bytes. 870 * 871 * Note that the allowable offset might decrease in the future, for example, 872 * to allow something like kmem_cache_free_rcu(). 873 * 874 * The BUILD_BUG_ON check must not involve any function calls, hence the 875 * checks are done in macros here. 876 */ 877#define kfree_rcu(ptr, rhf) \ 878do { \ 879 typeof (ptr) ___p = (ptr); \ 880 \ 881 if (___p) \ 882 __kvfree_rcu(&((___p)->rhf), offsetof(typeof(*(ptr)), rhf)); \ 883} while (0) 884 885/** 886 * kvfree_rcu() - kvfree an object after a grace period. 887 * 888 * This macro consists of one or two arguments and it is 889 * based on whether an object is head-less or not. If it 890 * has a head then a semantic stays the same as it used 891 * to be before: 892 * 893 * kvfree_rcu(ptr, rhf); 894 * 895 * where @ptr is a pointer to kvfree(), @rhf is the name 896 * of the rcu_head structure within the type of @ptr. 897 * 898 * When it comes to head-less variant, only one argument 899 * is passed and that is just a pointer which has to be 900 * freed after a grace period. Therefore the semantic is 901 * 902 * kvfree_rcu(ptr); 903 * 904 * where @ptr is a pointer to kvfree(). 905 * 906 * Please note, head-less way of freeing is permitted to 907 * use from a context that has to follow might_sleep() 908 * annotation. Otherwise, please switch and embed the 909 * rcu_head structure within the type of @ptr. 910 */ 911#define kvfree_rcu(...) KVFREE_GET_MACRO(__VA_ARGS__, \ 912 kvfree_rcu_arg_2, kvfree_rcu_arg_1)(__VA_ARGS__) 913 914#define KVFREE_GET_MACRO(_1, _2, NAME, ...) NAME 915#define kvfree_rcu_arg_2(ptr, rhf) kfree_rcu(ptr, rhf) 916#define kvfree_rcu_arg_1(ptr) \ 917do { \ 918 typeof(ptr) ___p = (ptr); \ 919 \ 920 if (___p) \ 921 kvfree_call_rcu(NULL, (rcu_callback_t) (___p)); \ 922} while (0) 923 924/* 925 * Place this after a lock-acquisition primitive to guarantee that 926 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies 927 * if the UNLOCK and LOCK are executed by the same CPU or if the 928 * UNLOCK and LOCK operate on the same lock variable. 929 */ 930#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE 931#define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */ 932#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 933#define smp_mb__after_unlock_lock() do { } while (0) 934#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 935 936 937/* Has the specified rcu_head structure been handed to call_rcu()? */ 938 939/** 940 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu() 941 * @rhp: The rcu_head structure to initialize. 942 * 943 * If you intend to invoke rcu_head_after_call_rcu() to test whether a 944 * given rcu_head structure has already been passed to call_rcu(), then 945 * you must also invoke this rcu_head_init() function on it just after 946 * allocating that structure. Calls to this function must not race with 947 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation. 948 */ 949static inline void rcu_head_init(struct rcu_head *rhp) 950{ 951 rhp->func = (rcu_callback_t)~0L; 952} 953 954/** 955 * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()? 956 * @rhp: The rcu_head structure to test. 957 * @f: The function passed to call_rcu() along with @rhp. 958 * 959 * Returns @true if the @rhp has been passed to call_rcu() with @func, 960 * and @false otherwise. Emits a warning in any other case, including 961 * the case where @rhp has already been invoked after a grace period. 962 * Calls to this function must not race with callback invocation. One way 963 * to avoid such races is to enclose the call to rcu_head_after_call_rcu() 964 * in an RCU read-side critical section that includes a read-side fetch 965 * of the pointer to the structure containing @rhp. 966 */ 967static inline bool 968rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f) 969{ 970 rcu_callback_t func = READ_ONCE(rhp->func); 971 972 if (func == f) 973 return true; 974 WARN_ON_ONCE(func != (rcu_callback_t)~0L); 975 return false; 976} 977 978/* kernel/ksysfs.c definitions */ 979extern int rcu_expedited; 980extern int rcu_normal; 981 982#endif /* __LINUX_RCUPDATE_H */ 983