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