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#define USHORT_CMP_GE(a, b) (USHRT_MAX / 2 >= (unsigned short)((a) - (b))) 37#define USHORT_CMP_LT(a, b) (USHRT_MAX / 2 < (unsigned short)((a) - (b))) 38 39/* Exported common interfaces */ 40void call_rcu(struct rcu_head *head, rcu_callback_t func); 41void rcu_barrier_tasks(void); 42void rcu_barrier_tasks_rude(void); 43void synchronize_rcu(void); 44 45#ifdef CONFIG_PREEMPT_RCU 46 47void __rcu_read_lock(void); 48void __rcu_read_unlock(void); 49 50/* 51 * Defined as a macro as it is a very low level header included from 52 * areas that don't even know about current. This gives the rcu_read_lock() 53 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other 54 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. 55 */ 56#define rcu_preempt_depth() READ_ONCE(current->rcu_read_lock_nesting) 57 58#else /* #ifdef CONFIG_PREEMPT_RCU */ 59 60#ifdef CONFIG_TINY_RCU 61#define rcu_read_unlock_strict() do { } while (0) 62#else 63void rcu_read_unlock_strict(void); 64#endif 65 66static inline void __rcu_read_lock(void) 67{ 68 preempt_disable(); 69} 70 71static inline void __rcu_read_unlock(void) 72{ 73 preempt_enable(); 74 rcu_read_unlock_strict(); 75} 76 77static inline int rcu_preempt_depth(void) 78{ 79 return 0; 80} 81 82#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 83 84/* Internal to kernel */ 85void rcu_init(void); 86extern int rcu_scheduler_active __read_mostly; 87void rcu_sched_clock_irq(int user); 88void rcu_report_dead(unsigned int cpu); 89void rcutree_migrate_callbacks(int cpu); 90 91#ifdef CONFIG_TASKS_RCU_GENERIC 92void rcu_init_tasks_generic(void); 93#else 94static inline void rcu_init_tasks_generic(void) { } 95#endif 96 97#ifdef CONFIG_RCU_STALL_COMMON 98void rcu_sysrq_start(void); 99void rcu_sysrq_end(void); 100#else /* #ifdef CONFIG_RCU_STALL_COMMON */ 101static inline void rcu_sysrq_start(void) { } 102static inline void rcu_sysrq_end(void) { } 103#endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ 104 105#ifdef CONFIG_NO_HZ_FULL 106void rcu_user_enter(void); 107void rcu_user_exit(void); 108#else 109static inline void rcu_user_enter(void) { } 110static inline void rcu_user_exit(void) { } 111#endif /* CONFIG_NO_HZ_FULL */ 112 113#ifdef CONFIG_RCU_NOCB_CPU 114void rcu_init_nohz(void); 115int rcu_nocb_cpu_offload(int cpu); 116int rcu_nocb_cpu_deoffload(int cpu); 117void rcu_nocb_flush_deferred_wakeup(void); 118#else /* #ifdef CONFIG_RCU_NOCB_CPU */ 119static inline void rcu_init_nohz(void) { } 120static inline int rcu_nocb_cpu_offload(int cpu) { return -EINVAL; } 121static inline int rcu_nocb_cpu_deoffload(int cpu) { return 0; } 122static inline void rcu_nocb_flush_deferred_wakeup(void) { } 123#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 124 125/** 126 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers 127 * @a: Code that RCU needs to pay attention to. 128 * 129 * RCU read-side critical sections are forbidden in the inner idle loop, 130 * that is, between the rcu_idle_enter() and the rcu_idle_exit() -- RCU 131 * will happily ignore any such read-side critical sections. However, 132 * things like powertop need tracepoints in the inner idle loop. 133 * 134 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) 135 * will tell RCU that it needs to pay attention, invoke its argument 136 * (in this example, calling the do_something_with_RCU() function), 137 * and then tell RCU to go back to ignoring this CPU. It is permissible 138 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is 139 * on the order of a million or so, even on 32-bit systems). It is 140 * not legal to block within RCU_NONIDLE(), nor is it permissible to 141 * transfer control either into or out of RCU_NONIDLE()'s statement. 142 */ 143#define RCU_NONIDLE(a) \ 144 do { \ 145 rcu_irq_enter_irqson(); \ 146 do { a; } while (0); \ 147 rcu_irq_exit_irqson(); \ 148 } while (0) 149 150/* 151 * Note a quasi-voluntary context switch for RCU-tasks's benefit. 152 * This is a macro rather than an inline function to avoid #include hell. 153 */ 154#ifdef CONFIG_TASKS_RCU_GENERIC 155 156# ifdef CONFIG_TASKS_RCU 157# define rcu_tasks_classic_qs(t, preempt) \ 158 do { \ 159 if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout)) \ 160 WRITE_ONCE((t)->rcu_tasks_holdout, false); \ 161 } while (0) 162void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func); 163void synchronize_rcu_tasks(void); 164# else 165# define rcu_tasks_classic_qs(t, preempt) do { } while (0) 166# define call_rcu_tasks call_rcu 167# define synchronize_rcu_tasks synchronize_rcu 168# endif 169 170# ifdef CONFIG_TASKS_TRACE_RCU 171# define rcu_tasks_trace_qs(t) \ 172 do { \ 173 if (!likely(READ_ONCE((t)->trc_reader_checked)) && \ 174 !unlikely(READ_ONCE((t)->trc_reader_nesting))) { \ 175 smp_store_release(&(t)->trc_reader_checked, true); \ 176 smp_mb(); /* Readers partitioned by store. */ \ 177 } \ 178 } while (0) 179# else 180# define rcu_tasks_trace_qs(t) do { } while (0) 181# endif 182 183#define rcu_tasks_qs(t, preempt) \ 184do { \ 185 rcu_tasks_classic_qs((t), (preempt)); \ 186 rcu_tasks_trace_qs((t)); \ 187} while (0) 188 189# ifdef CONFIG_TASKS_RUDE_RCU 190void call_rcu_tasks_rude(struct rcu_head *head, rcu_callback_t func); 191void synchronize_rcu_tasks_rude(void); 192# endif 193 194#define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false) 195void exit_tasks_rcu_start(void); 196void exit_tasks_rcu_finish(void); 197#else /* #ifdef CONFIG_TASKS_RCU_GENERIC */ 198#define rcu_tasks_qs(t, preempt) do { } while (0) 199#define rcu_note_voluntary_context_switch(t) do { } while (0) 200#define call_rcu_tasks call_rcu 201#define synchronize_rcu_tasks synchronize_rcu 202static inline void exit_tasks_rcu_start(void) { } 203static inline void exit_tasks_rcu_finish(void) { } 204#endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */ 205 206/** 207 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU 208 * 209 * This macro resembles cond_resched(), except that it is defined to 210 * report potential quiescent states to RCU-tasks even if the cond_resched() 211 * machinery were to be shut off, as some advocate for PREEMPTION kernels. 212 */ 213#define cond_resched_tasks_rcu_qs() \ 214do { \ 215 rcu_tasks_qs(current, false); \ 216 cond_resched(); \ 217} while (0) 218 219/* 220 * Infrastructure to implement the synchronize_() primitives in 221 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 222 */ 223 224#if defined(CONFIG_TREE_RCU) 225#include <linux/rcutree.h> 226#elif defined(CONFIG_TINY_RCU) 227#include <linux/rcutiny.h> 228#else 229#error "Unknown RCU implementation specified to kernel configuration" 230#endif 231 232/* 233 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls 234 * are needed for dynamic initialization and destruction of rcu_head 235 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for 236 * dynamic initialization and destruction of statically allocated rcu_head 237 * structures. However, rcu_head structures allocated dynamically in the 238 * heap don't need any initialization. 239 */ 240#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 241void init_rcu_head(struct rcu_head *head); 242void destroy_rcu_head(struct rcu_head *head); 243void init_rcu_head_on_stack(struct rcu_head *head); 244void destroy_rcu_head_on_stack(struct rcu_head *head); 245#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 246static inline void init_rcu_head(struct rcu_head *head) { } 247static inline void destroy_rcu_head(struct rcu_head *head) { } 248static inline void init_rcu_head_on_stack(struct rcu_head *head) { } 249static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { } 250#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 251 252#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 253bool rcu_lockdep_current_cpu_online(void); 254#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 255static inline bool rcu_lockdep_current_cpu_online(void) { return true; } 256#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 257 258extern struct lockdep_map rcu_lock_map; 259extern struct lockdep_map rcu_bh_lock_map; 260extern struct lockdep_map rcu_sched_lock_map; 261extern struct lockdep_map rcu_callback_map; 262 263#ifdef CONFIG_DEBUG_LOCK_ALLOC 264 265static inline void rcu_lock_acquire(struct lockdep_map *map) 266{ 267 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); 268} 269 270static inline void rcu_lock_release(struct lockdep_map *map) 271{ 272 lock_release(map, _THIS_IP_); 273} 274 275int debug_lockdep_rcu_enabled(void); 276int rcu_read_lock_held(void); 277int rcu_read_lock_bh_held(void); 278int rcu_read_lock_sched_held(void); 279int rcu_read_lock_any_held(void); 280 281#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 282 283# define rcu_lock_acquire(a) do { } while (0) 284# define rcu_lock_release(a) do { } while (0) 285 286static inline int rcu_read_lock_held(void) 287{ 288 return 1; 289} 290 291static inline int rcu_read_lock_bh_held(void) 292{ 293 return 1; 294} 295 296static inline int rcu_read_lock_sched_held(void) 297{ 298 return !preemptible(); 299} 300 301static inline int rcu_read_lock_any_held(void) 302{ 303 return !preemptible(); 304} 305 306#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 307 308#ifdef CONFIG_PROVE_RCU 309 310/** 311 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met 312 * @c: condition to check 313 * @s: informative message 314 */ 315#define RCU_LOCKDEP_WARN(c, s) \ 316 do { \ 317 static bool __section(".data.unlikely") __warned; \ 318 if ((c) && debug_lockdep_rcu_enabled() && !__warned) { \ 319 __warned = true; \ 320 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 321 } \ 322 } while (0) 323 324#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 325static inline void rcu_preempt_sleep_check(void) 326{ 327 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), 328 "Illegal context switch in RCU read-side critical section"); 329} 330#else /* #ifdef CONFIG_PROVE_RCU */ 331static inline void rcu_preempt_sleep_check(void) { } 332#endif /* #else #ifdef CONFIG_PROVE_RCU */ 333 334#define rcu_sleep_check() \ 335 do { \ 336 rcu_preempt_sleep_check(); \ 337 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) \ 338 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ 339 "Illegal context switch in RCU-bh read-side critical section"); \ 340 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ 341 "Illegal context switch in RCU-sched read-side critical section"); \ 342 } while (0) 343 344#else /* #ifdef CONFIG_PROVE_RCU */ 345 346#define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c)) 347#define rcu_sleep_check() do { } while (0) 348 349#endif /* #else #ifdef CONFIG_PROVE_RCU */ 350 351/* 352 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 353 * and rcu_assign_pointer(). Some of these could be folded into their 354 * callers, but they are left separate in order to ease introduction of 355 * multiple pointers markings to match different RCU implementations 356 * (e.g., __srcu), should this make sense in the future. 357 */ 358 359#ifdef __CHECKER__ 360#define rcu_check_sparse(p, space) \ 361 ((void)(((typeof(*p) space *)p) == p)) 362#else /* #ifdef __CHECKER__ */ 363#define rcu_check_sparse(p, space) 364#endif /* #else #ifdef __CHECKER__ */ 365 366/** 367 * unrcu_pointer - mark a pointer as not being RCU protected 368 * @p: pointer needing to lose its __rcu property 369 * 370 * Converts @p from an __rcu pointer to a __kernel pointer. 371 * This allows an __rcu pointer to be used with xchg() and friends. 372 */ 373#define unrcu_pointer(p) \ 374({ \ 375 typeof(*p) *_________p1 = (typeof(*p) *__force)(p); \ 376 rcu_check_sparse(p, __rcu); \ 377 ((typeof(*p) __force __kernel *)(_________p1)); \ 378}) 379 380#define __rcu_access_pointer(p, space) \ 381({ \ 382 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 383 rcu_check_sparse(p, space); \ 384 ((typeof(*p) __force __kernel *)(_________p1)); \ 385}) 386#define __rcu_dereference_check(p, c, space) \ 387({ \ 388 /* Dependency order vs. p above. */ \ 389 typeof(*p) *________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 390 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ 391 rcu_check_sparse(p, space); \ 392 ((typeof(*p) __force __kernel *)(________p1)); \ 393}) 394#define __rcu_dereference_protected(p, c, space) \ 395({ \ 396 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ 397 rcu_check_sparse(p, space); \ 398 ((typeof(*p) __force __kernel *)(p)); \ 399}) 400#define rcu_dereference_raw(p) \ 401({ \ 402 /* Dependency order vs. p above. */ \ 403 typeof(p) ________p1 = READ_ONCE(p); \ 404 ((typeof(*p) __force __kernel *)(________p1)); \ 405}) 406 407/** 408 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 409 * @v: The value to statically initialize with. 410 */ 411#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 412 413/** 414 * rcu_assign_pointer() - assign to RCU-protected pointer 415 * @p: pointer to assign to 416 * @v: value to assign (publish) 417 * 418 * Assigns the specified value to the specified RCU-protected 419 * pointer, ensuring that any concurrent RCU readers will see 420 * any prior initialization. 421 * 422 * Inserts memory barriers on architectures that require them 423 * (which is most of them), and also prevents the compiler from 424 * reordering the code that initializes the structure after the pointer 425 * assignment. More importantly, this call documents which pointers 426 * will be dereferenced by RCU read-side code. 427 * 428 * In some special cases, you may use RCU_INIT_POINTER() instead 429 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 430 * to the fact that it does not constrain either the CPU or the compiler. 431 * That said, using RCU_INIT_POINTER() when you should have used 432 * rcu_assign_pointer() is a very bad thing that results in 433 * impossible-to-diagnose memory corruption. So please be careful. 434 * See the RCU_INIT_POINTER() comment header for details. 435 * 436 * Note that rcu_assign_pointer() evaluates each of its arguments only 437 * once, appearances notwithstanding. One of the "extra" evaluations 438 * is in typeof() and the other visible only to sparse (__CHECKER__), 439 * neither of which actually execute the argument. As with most cpp 440 * macros, this execute-arguments-only-once property is important, so 441 * please be careful when making changes to rcu_assign_pointer() and the 442 * other macros that it invokes. 443 */ 444#define rcu_assign_pointer(p, v) \ 445do { \ 446 uintptr_t _r_a_p__v = (uintptr_t)(v); \ 447 rcu_check_sparse(p, __rcu); \ 448 \ 449 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \ 450 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \ 451 else \ 452 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \ 453} while (0) 454 455/** 456 * rcu_replace_pointer() - replace an RCU pointer, returning its old value 457 * @rcu_ptr: RCU pointer, whose old value is returned 458 * @ptr: regular pointer 459 * @c: the lockdep conditions under which the dereference will take place 460 * 461 * Perform a replacement, where @rcu_ptr is an RCU-annotated 462 * pointer and @c is the lockdep argument that is passed to the 463 * rcu_dereference_protected() call used to read that pointer. The old 464 * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr. 465 */ 466#define rcu_replace_pointer(rcu_ptr, ptr, c) \ 467({ \ 468 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \ 469 rcu_assign_pointer((rcu_ptr), (ptr)); \ 470 __tmp; \ 471}) 472 473/** 474 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 475 * @p: The pointer to read 476 * 477 * Return the value of the specified RCU-protected pointer, but omit the 478 * lockdep checks for being in an RCU read-side critical section. This is 479 * useful when the value of this pointer is accessed, but the pointer is 480 * not dereferenced, for example, when testing an RCU-protected pointer 481 * against NULL. Although rcu_access_pointer() may also be used in cases 482 * where update-side locks prevent the value of the pointer from changing, 483 * you should instead use rcu_dereference_protected() for this use case. 484 * 485 * It is also permissible to use rcu_access_pointer() when read-side 486 * access to the pointer was removed at least one grace period ago, as 487 * is the case in the context of the RCU callback that is freeing up 488 * the data, or after a synchronize_rcu() returns. This can be useful 489 * when tearing down multi-linked structures after a grace period 490 * has elapsed. 491 */ 492#define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) 493 494/** 495 * rcu_dereference_check() - rcu_dereference with debug checking 496 * @p: The pointer to read, prior to dereferencing 497 * @c: The conditions under which the dereference will take place 498 * 499 * Do an rcu_dereference(), but check that the conditions under which the 500 * dereference will take place are correct. Typically the conditions 501 * indicate the various locking conditions that should be held at that 502 * point. The check should return true if the conditions are satisfied. 503 * An implicit check for being in an RCU read-side critical section 504 * (rcu_read_lock()) is included. 505 * 506 * For example: 507 * 508 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 509 * 510 * could be used to indicate to lockdep that foo->bar may only be dereferenced 511 * if either rcu_read_lock() is held, or that the lock required to replace 512 * the bar struct at foo->bar is held. 513 * 514 * Note that the list of conditions may also include indications of when a lock 515 * need not be held, for example during initialisation or destruction of the 516 * target struct: 517 * 518 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 519 * atomic_read(&foo->usage) == 0); 520 * 521 * Inserts memory barriers on architectures that require them 522 * (currently only the Alpha), prevents the compiler from refetching 523 * (and from merging fetches), and, more importantly, documents exactly 524 * which pointers are protected by RCU and checks that the pointer is 525 * annotated as __rcu. 526 */ 527#define rcu_dereference_check(p, c) \ 528 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu) 529 530/** 531 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 532 * @p: The pointer to read, prior to dereferencing 533 * @c: The conditions under which the dereference will take place 534 * 535 * This is the RCU-bh counterpart to rcu_dereference_check(). However, 536 * please note that starting in v5.0 kernels, vanilla RCU grace periods 537 * wait for local_bh_disable() regions of code in addition to regions of 538 * code demarked by rcu_read_lock() and rcu_read_unlock(). This means 539 * that synchronize_rcu(), call_rcu, and friends all take not only 540 * rcu_read_lock() but also rcu_read_lock_bh() into account. 541 */ 542#define rcu_dereference_bh_check(p, c) \ 543 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu) 544 545/** 546 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 547 * @p: The pointer to read, prior to dereferencing 548 * @c: The conditions under which the dereference will take place 549 * 550 * This is the RCU-sched counterpart to rcu_dereference_check(). 551 * However, please note that starting in v5.0 kernels, vanilla RCU grace 552 * periods wait for preempt_disable() regions of code in addition to 553 * regions of code demarked by rcu_read_lock() and rcu_read_unlock(). 554 * This means that synchronize_rcu(), call_rcu, and friends all take not 555 * only rcu_read_lock() but also rcu_read_lock_sched() into account. 556 */ 557#define rcu_dereference_sched_check(p, c) \ 558 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \ 559 __rcu) 560 561/* 562 * The tracing infrastructure traces RCU (we want that), but unfortunately 563 * some of the RCU checks causes tracing to lock up the system. 564 * 565 * The no-tracing version of rcu_dereference_raw() must not call 566 * rcu_read_lock_held(). 567 */ 568#define rcu_dereference_raw_check(p) __rcu_dereference_check((p), 1, __rcu) 569 570/** 571 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 572 * @p: The pointer to read, prior to dereferencing 573 * @c: The conditions under which the dereference will take place 574 * 575 * Return the value of the specified RCU-protected pointer, but omit 576 * the READ_ONCE(). This is useful in cases where update-side locks 577 * prevent the value of the pointer from changing. Please note that this 578 * primitive does *not* prevent the compiler from repeating this reference 579 * or combining it with other references, so it should not be used without 580 * protection of appropriate locks. 581 * 582 * This function is only for update-side use. Using this function 583 * when protected only by rcu_read_lock() will result in infrequent 584 * but very ugly failures. 585 */ 586#define rcu_dereference_protected(p, c) \ 587 __rcu_dereference_protected((p), (c), __rcu) 588 589 590/** 591 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 592 * @p: The pointer to read, prior to dereferencing 593 * 594 * This is a simple wrapper around rcu_dereference_check(). 595 */ 596#define rcu_dereference(p) rcu_dereference_check(p, 0) 597 598/** 599 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 600 * @p: The pointer to read, prior to dereferencing 601 * 602 * Makes rcu_dereference_check() do the dirty work. 603 */ 604#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 605 606/** 607 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 608 * @p: The pointer to read, prior to dereferencing 609 * 610 * Makes rcu_dereference_check() do the dirty work. 611 */ 612#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 613 614/** 615 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism 616 * @p: The pointer to hand off 617 * 618 * This is simply an identity function, but it documents where a pointer 619 * is handed off from RCU to some other synchronization mechanism, for 620 * example, reference counting or locking. In C11, it would map to 621 * kill_dependency(). It could be used as follows:: 622 * 623 * rcu_read_lock(); 624 * p = rcu_dereference(gp); 625 * long_lived = is_long_lived(p); 626 * if (long_lived) { 627 * if (!atomic_inc_not_zero(p->refcnt)) 628 * long_lived = false; 629 * else 630 * p = rcu_pointer_handoff(p); 631 * } 632 * rcu_read_unlock(); 633 */ 634#define rcu_pointer_handoff(p) (p) 635 636/** 637 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 638 * 639 * When synchronize_rcu() is invoked on one CPU while other CPUs 640 * are within RCU read-side critical sections, then the 641 * synchronize_rcu() is guaranteed to block until after all the other 642 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 643 * on one CPU while other CPUs are within RCU read-side critical 644 * sections, invocation of the corresponding RCU callback is deferred 645 * until after the all the other CPUs exit their critical sections. 646 * 647 * In v5.0 and later kernels, synchronize_rcu() and call_rcu() also 648 * wait for regions of code with preemption disabled, including regions of 649 * code with interrupts or softirqs disabled. In pre-v5.0 kernels, which 650 * define synchronize_sched(), only code enclosed within rcu_read_lock() 651 * and rcu_read_unlock() are guaranteed to be waited for. 652 * 653 * Note, however, that RCU callbacks are permitted to run concurrently 654 * with new RCU read-side critical sections. One way that this can happen 655 * is via the following sequence of events: (1) CPU 0 enters an RCU 656 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 657 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 658 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 659 * callback is invoked. This is legal, because the RCU read-side critical 660 * section that was running concurrently with the call_rcu() (and which 661 * therefore might be referencing something that the corresponding RCU 662 * callback would free up) has completed before the corresponding 663 * RCU callback is invoked. 664 * 665 * RCU read-side critical sections may be nested. Any deferred actions 666 * will be deferred until the outermost RCU read-side critical section 667 * completes. 668 * 669 * You can avoid reading and understanding the next paragraph by 670 * following this rule: don't put anything in an rcu_read_lock() RCU 671 * read-side critical section that would block in a !PREEMPTION kernel. 672 * But if you want the full story, read on! 673 * 674 * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU), 675 * it is illegal to block while in an RCU read-side critical section. 676 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION 677 * kernel builds, RCU read-side critical sections may be preempted, 678 * but explicit blocking is illegal. Finally, in preemptible RCU 679 * implementations in real-time (with -rt patchset) kernel builds, RCU 680 * read-side critical sections may be preempted and they may also block, but 681 * only when acquiring spinlocks that are subject to priority inheritance. 682 */ 683static __always_inline void rcu_read_lock(void) 684{ 685 __rcu_read_lock(); 686 __acquire(RCU); 687 rcu_lock_acquire(&rcu_lock_map); 688 RCU_LOCKDEP_WARN(!rcu_is_watching(), 689 "rcu_read_lock() used illegally while idle"); 690} 691 692/* 693 * So where is rcu_write_lock()? It does not exist, as there is no 694 * way for writers to lock out RCU readers. This is a feature, not 695 * a bug -- this property is what provides RCU's performance benefits. 696 * Of course, writers must coordinate with each other. The normal 697 * spinlock primitives work well for this, but any other technique may be 698 * used as well. RCU does not care how the writers keep out of each 699 * others' way, as long as they do so. 700 */ 701 702/** 703 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 704 * 705 * In almost all situations, rcu_read_unlock() is immune from deadlock. 706 * In recent kernels that have consolidated synchronize_sched() and 707 * synchronize_rcu_bh() into synchronize_rcu(), this deadlock immunity 708 * also extends to the scheduler's runqueue and priority-inheritance 709 * spinlocks, courtesy of the quiescent-state deferral that is carried 710 * out when rcu_read_unlock() is invoked with interrupts disabled. 711 * 712 * See rcu_read_lock() for more information. 713 */ 714static inline void rcu_read_unlock(void) 715{ 716 RCU_LOCKDEP_WARN(!rcu_is_watching(), 717 "rcu_read_unlock() used illegally while idle"); 718 __release(RCU); 719 __rcu_read_unlock(); 720 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ 721} 722 723/** 724 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 725 * 726 * This is equivalent to rcu_read_lock(), but also disables softirqs. 727 * Note that anything else that disables softirqs can also serve as an RCU 728 * read-side critical section. However, please note that this equivalence 729 * applies only to v5.0 and later. Before v5.0, rcu_read_lock() and 730 * rcu_read_lock_bh() were unrelated. 731 * 732 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 733 * must occur in the same context, for example, it is illegal to invoke 734 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 735 * was invoked from some other task. 736 */ 737static inline void rcu_read_lock_bh(void) 738{ 739 local_bh_disable(); 740 __acquire(RCU_BH); 741 rcu_lock_acquire(&rcu_bh_lock_map); 742 RCU_LOCKDEP_WARN(!rcu_is_watching(), 743 "rcu_read_lock_bh() used illegally while idle"); 744} 745 746/** 747 * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section 748 * 749 * See rcu_read_lock_bh() for more information. 750 */ 751static inline void rcu_read_unlock_bh(void) 752{ 753 RCU_LOCKDEP_WARN(!rcu_is_watching(), 754 "rcu_read_unlock_bh() used illegally while idle"); 755 rcu_lock_release(&rcu_bh_lock_map); 756 __release(RCU_BH); 757 local_bh_enable(); 758} 759 760/** 761 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 762 * 763 * This is equivalent to rcu_read_lock(), but also disables preemption. 764 * Read-side critical sections can also be introduced by anything else that 765 * disables preemption, including local_irq_disable() and friends. However, 766 * please note that the equivalence to rcu_read_lock() applies only to 767 * v5.0 and later. Before v5.0, rcu_read_lock() and rcu_read_lock_sched() 768 * were unrelated. 769 * 770 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 771 * must occur in the same context, for example, it is illegal to invoke 772 * rcu_read_unlock_sched() from process context if the matching 773 * rcu_read_lock_sched() was invoked from an NMI handler. 774 */ 775static inline void rcu_read_lock_sched(void) 776{ 777 preempt_disable(); 778 __acquire(RCU_SCHED); 779 rcu_lock_acquire(&rcu_sched_lock_map); 780 RCU_LOCKDEP_WARN(!rcu_is_watching(), 781 "rcu_read_lock_sched() used illegally while idle"); 782} 783 784/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 785static inline notrace void rcu_read_lock_sched_notrace(void) 786{ 787 preempt_disable_notrace(); 788 __acquire(RCU_SCHED); 789} 790 791/** 792 * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section 793 * 794 * See rcu_read_lock_sched() for more information. 795 */ 796static inline void rcu_read_unlock_sched(void) 797{ 798 RCU_LOCKDEP_WARN(!rcu_is_watching(), 799 "rcu_read_unlock_sched() used illegally while idle"); 800 rcu_lock_release(&rcu_sched_lock_map); 801 __release(RCU_SCHED); 802 preempt_enable(); 803} 804 805/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 806static inline notrace void rcu_read_unlock_sched_notrace(void) 807{ 808 __release(RCU_SCHED); 809 preempt_enable_notrace(); 810} 811 812/** 813 * RCU_INIT_POINTER() - initialize an RCU protected pointer 814 * @p: The pointer to be initialized. 815 * @v: The value to initialized the pointer to. 816 * 817 * Initialize an RCU-protected pointer in special cases where readers 818 * do not need ordering constraints on the CPU or the compiler. These 819 * special cases are: 820 * 821 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or* 822 * 2. The caller has taken whatever steps are required to prevent 823 * RCU readers from concurrently accessing this pointer *or* 824 * 3. The referenced data structure has already been exposed to 825 * readers either at compile time or via rcu_assign_pointer() *and* 826 * 827 * a. You have not made *any* reader-visible changes to 828 * this structure since then *or* 829 * b. It is OK for readers accessing this structure from its 830 * new location to see the old state of the structure. (For 831 * example, the changes were to statistical counters or to 832 * other state where exact synchronization is not required.) 833 * 834 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 835 * result in impossible-to-diagnose memory corruption. As in the structures 836 * will look OK in crash dumps, but any concurrent RCU readers might 837 * see pre-initialized values of the referenced data structure. So 838 * please be very careful how you use RCU_INIT_POINTER()!!! 839 * 840 * If you are creating an RCU-protected linked structure that is accessed 841 * by a single external-to-structure RCU-protected pointer, then you may 842 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 843 * pointers, but you must use rcu_assign_pointer() to initialize the 844 * external-to-structure pointer *after* you have completely initialized 845 * the reader-accessible portions of the linked structure. 846 * 847 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 848 * ordering guarantees for either the CPU or the compiler. 849 */ 850#define RCU_INIT_POINTER(p, v) \ 851 do { \ 852 rcu_check_sparse(p, __rcu); \ 853 WRITE_ONCE(p, RCU_INITIALIZER(v)); \ 854 } while (0) 855 856/** 857 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 858 * @p: The pointer to be initialized. 859 * @v: The value to initialized the pointer to. 860 * 861 * GCC-style initialization for an RCU-protected pointer in a structure field. 862 */ 863#define RCU_POINTER_INITIALIZER(p, v) \ 864 .p = RCU_INITIALIZER(v) 865 866/* 867 * Does the specified offset indicate that the corresponding rcu_head 868 * structure can be handled by kvfree_rcu()? 869 */ 870#define __is_kvfree_rcu_offset(offset) ((offset) < 4096) 871 872/** 873 * kfree_rcu() - kfree an object after a grace period. 874 * @ptr: pointer to kfree for both single- and double-argument invocations. 875 * @rhf: the name of the struct rcu_head within the type of @ptr, 876 * but only for double-argument invocations. 877 * 878 * Many rcu callbacks functions just call kfree() on the base structure. 879 * These functions are trivial, but their size adds up, and furthermore 880 * when they are used in a kernel module, that module must invoke the 881 * high-latency rcu_barrier() function at module-unload time. 882 * 883 * The kfree_rcu() function handles this issue. Rather than encoding a 884 * function address in the embedded rcu_head structure, kfree_rcu() instead 885 * encodes the offset of the rcu_head structure within the base structure. 886 * Because the functions are not allowed in the low-order 4096 bytes of 887 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 888 * If the offset is larger than 4095 bytes, a compile-time error will 889 * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can 890 * either fall back to use of call_rcu() or rearrange the structure to 891 * position the rcu_head structure into the first 4096 bytes. 892 * 893 * Note that the allowable offset might decrease in the future, for example, 894 * to allow something like kmem_cache_free_rcu(). 895 * 896 * The BUILD_BUG_ON check must not involve any function calls, hence the 897 * checks are done in macros here. 898 */ 899#define kfree_rcu(ptr, rhf...) kvfree_rcu(ptr, ## rhf) 900 901/** 902 * kvfree_rcu() - kvfree an object after a grace period. 903 * 904 * This macro consists of one or two arguments and it is 905 * based on whether an object is head-less or not. If it 906 * has a head then a semantic stays the same as it used 907 * to be before: 908 * 909 * kvfree_rcu(ptr, rhf); 910 * 911 * where @ptr is a pointer to kvfree(), @rhf is the name 912 * of the rcu_head structure within the type of @ptr. 913 * 914 * When it comes to head-less variant, only one argument 915 * is passed and that is just a pointer which has to be 916 * freed after a grace period. Therefore the semantic is 917 * 918 * kvfree_rcu(ptr); 919 * 920 * where @ptr is a pointer to kvfree(). 921 * 922 * Please note, head-less way of freeing is permitted to 923 * use from a context that has to follow might_sleep() 924 * annotation. Otherwise, please switch and embed the 925 * rcu_head structure within the type of @ptr. 926 */ 927#define kvfree_rcu(...) KVFREE_GET_MACRO(__VA_ARGS__, \ 928 kvfree_rcu_arg_2, kvfree_rcu_arg_1)(__VA_ARGS__) 929 930#define KVFREE_GET_MACRO(_1, _2, NAME, ...) NAME 931#define kvfree_rcu_arg_2(ptr, rhf) \ 932do { \ 933 typeof (ptr) ___p = (ptr); \ 934 \ 935 if (___p) { \ 936 BUILD_BUG_ON(!__is_kvfree_rcu_offset(offsetof(typeof(*(ptr)), rhf))); \ 937 kvfree_call_rcu(&((___p)->rhf), (rcu_callback_t)(unsigned long) \ 938 (offsetof(typeof(*(ptr)), rhf))); \ 939 } \ 940} while (0) 941 942#define kvfree_rcu_arg_1(ptr) \ 943do { \ 944 typeof(ptr) ___p = (ptr); \ 945 \ 946 if (___p) \ 947 kvfree_call_rcu(NULL, (rcu_callback_t) (___p)); \ 948} while (0) 949 950/* 951 * Place this after a lock-acquisition primitive to guarantee that 952 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies 953 * if the UNLOCK and LOCK are executed by the same CPU or if the 954 * UNLOCK and LOCK operate on the same lock variable. 955 */ 956#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE 957#define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */ 958#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 959#define smp_mb__after_unlock_lock() do { } while (0) 960#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 961 962 963/* Has the specified rcu_head structure been handed to call_rcu()? */ 964 965/** 966 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu() 967 * @rhp: The rcu_head structure to initialize. 968 * 969 * If you intend to invoke rcu_head_after_call_rcu() to test whether a 970 * given rcu_head structure has already been passed to call_rcu(), then 971 * you must also invoke this rcu_head_init() function on it just after 972 * allocating that structure. Calls to this function must not race with 973 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation. 974 */ 975static inline void rcu_head_init(struct rcu_head *rhp) 976{ 977 rhp->func = (rcu_callback_t)~0L; 978} 979 980/** 981 * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()? 982 * @rhp: The rcu_head structure to test. 983 * @f: The function passed to call_rcu() along with @rhp. 984 * 985 * Returns @true if the @rhp has been passed to call_rcu() with @func, 986 * and @false otherwise. Emits a warning in any other case, including 987 * the case where @rhp has already been invoked after a grace period. 988 * Calls to this function must not race with callback invocation. One way 989 * to avoid such races is to enclose the call to rcu_head_after_call_rcu() 990 * in an RCU read-side critical section that includes a read-side fetch 991 * of the pointer to the structure containing @rhp. 992 */ 993static inline bool 994rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f) 995{ 996 rcu_callback_t func = READ_ONCE(rhp->func); 997 998 if (func == f) 999 return true; 1000 WARN_ON_ONCE(func != (rcu_callback_t)~0L);
1001 return false; 1002} 1003 1004/* kernel/ksysfs.c definitions */ 1005extern int rcu_expedited; 1006extern int rcu_normal; 1007 1008#endif /* __LINUX_RCUPDATE_H */ 1009