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 <linux/cleanup.h> 31#include <asm/processor.h> 32#include <linux/cpumask.h> 33#include <linux/context_tracking_irq.h> 34 35#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) 36#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) 37 38/* Exported common interfaces */ 39void call_rcu(struct rcu_head *head, rcu_callback_t func); 40void rcu_barrier_tasks(void); 41void rcu_barrier_tasks_rude(void); 42void synchronize_rcu(void); 43 44struct rcu_gp_oldstate; 45unsigned long get_completed_synchronize_rcu(void); 46void get_completed_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp); 47 48// Maximum number of unsigned long values corresponding to 49// not-yet-completed RCU grace periods. 50#define NUM_ACTIVE_RCU_POLL_OLDSTATE 2 51 52/** 53 * same_state_synchronize_rcu - Are two old-state values identical? 54 * @oldstate1: First old-state value. 55 * @oldstate2: Second old-state value. 56 * 57 * The two old-state values must have been obtained from either 58 * get_state_synchronize_rcu(), start_poll_synchronize_rcu(), or 59 * get_completed_synchronize_rcu(). Returns @true if the two values are 60 * identical and @false otherwise. This allows structures whose lifetimes 61 * are tracked by old-state values to push these values to a list header, 62 * allowing those structures to be slightly smaller. 63 */ 64static inline bool same_state_synchronize_rcu(unsigned long oldstate1, unsigned long oldstate2) 65{ 66 return oldstate1 == oldstate2; 67} 68 69#ifdef CONFIG_PREEMPT_RCU 70 71void __rcu_read_lock(void); 72void __rcu_read_unlock(void); 73 74/* 75 * Defined as a macro as it is a very low level header included from 76 * areas that don't even know about current. This gives the rcu_read_lock() 77 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other 78 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. 79 */ 80#define rcu_preempt_depth() READ_ONCE(current->rcu_read_lock_nesting) 81 82#else /* #ifdef CONFIG_PREEMPT_RCU */ 83 84#ifdef CONFIG_TINY_RCU 85#define rcu_read_unlock_strict() do { } while (0) 86#else 87void rcu_read_unlock_strict(void); 88#endif 89 90static inline void __rcu_read_lock(void) 91{ 92 preempt_disable(); 93} 94 95static inline void __rcu_read_unlock(void) 96{ 97 preempt_enable(); 98 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) 99 rcu_read_unlock_strict(); 100} 101 102static inline int rcu_preempt_depth(void) 103{ 104 return 0; 105} 106 107#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 108 109#ifdef CONFIG_RCU_LAZY 110void call_rcu_hurry(struct rcu_head *head, rcu_callback_t func); 111#else 112static inline void call_rcu_hurry(struct rcu_head *head, rcu_callback_t func) 113{ 114 call_rcu(head, func); 115} 116#endif 117 118/* Internal to kernel */ 119void rcu_init(void); 120extern int rcu_scheduler_active; 121void rcu_sched_clock_irq(int user); 122 123#ifdef CONFIG_TASKS_RCU_GENERIC 124void rcu_init_tasks_generic(void); 125#else 126static inline void rcu_init_tasks_generic(void) { } 127#endif 128 129#ifdef CONFIG_RCU_STALL_COMMON 130void rcu_sysrq_start(void); 131void rcu_sysrq_end(void); 132#else /* #ifdef CONFIG_RCU_STALL_COMMON */ 133static inline void rcu_sysrq_start(void) { } 134static inline void rcu_sysrq_end(void) { } 135#endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ 136 137#if defined(CONFIG_NO_HZ_FULL) && (!defined(CONFIG_GENERIC_ENTRY) || !defined(CONFIG_KVM_XFER_TO_GUEST_WORK)) 138void rcu_irq_work_resched(void); 139#else 140static inline void rcu_irq_work_resched(void) { } 141#endif 142 143#ifdef CONFIG_RCU_NOCB_CPU 144void rcu_init_nohz(void); 145int rcu_nocb_cpu_offload(int cpu); 146int rcu_nocb_cpu_deoffload(int cpu); 147void rcu_nocb_flush_deferred_wakeup(void); 148#else /* #ifdef CONFIG_RCU_NOCB_CPU */ 149static inline void rcu_init_nohz(void) { } 150static inline int rcu_nocb_cpu_offload(int cpu) { return -EINVAL; } 151static inline int rcu_nocb_cpu_deoffload(int cpu) { return 0; } 152static inline void rcu_nocb_flush_deferred_wakeup(void) { } 153#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 154 155/* 156 * Note a quasi-voluntary context switch for RCU-tasks's benefit. 157 * This is a macro rather than an inline function to avoid #include hell. 158 */ 159#ifdef CONFIG_TASKS_RCU_GENERIC 160 161# ifdef CONFIG_TASKS_RCU 162# define rcu_tasks_classic_qs(t, preempt) \ 163 do { \ 164 if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout)) \ 165 WRITE_ONCE((t)->rcu_tasks_holdout, false); \ 166 } while (0) 167void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func); 168void synchronize_rcu_tasks(void); 169# else 170# define rcu_tasks_classic_qs(t, preempt) do { } while (0) 171# define call_rcu_tasks call_rcu 172# define synchronize_rcu_tasks synchronize_rcu 173# endif 174 175# ifdef CONFIG_TASKS_TRACE_RCU 176// Bits for ->trc_reader_special.b.need_qs field. 177#define TRC_NEED_QS 0x1 // Task needs a quiescent state. 178#define TRC_NEED_QS_CHECKED 0x2 // Task has been checked for needing quiescent state. 179 180u8 rcu_trc_cmpxchg_need_qs(struct task_struct *t, u8 old, u8 new); 181void rcu_tasks_trace_qs_blkd(struct task_struct *t); 182 183# define rcu_tasks_trace_qs(t) \ 184 do { \ 185 int ___rttq_nesting = READ_ONCE((t)->trc_reader_nesting); \ 186 \ 187 if (unlikely(READ_ONCE((t)->trc_reader_special.b.need_qs) == TRC_NEED_QS) && \ 188 likely(!___rttq_nesting)) { \ 189 rcu_trc_cmpxchg_need_qs((t), TRC_NEED_QS, TRC_NEED_QS_CHECKED); \ 190 } else if (___rttq_nesting && ___rttq_nesting != INT_MIN && \ 191 !READ_ONCE((t)->trc_reader_special.b.blocked)) { \ 192 rcu_tasks_trace_qs_blkd(t); \ 193 } \ 194 } while (0) 195# else 196# define rcu_tasks_trace_qs(t) do { } while (0) 197# endif 198 199#define rcu_tasks_qs(t, preempt) \ 200do { \ 201 rcu_tasks_classic_qs((t), (preempt)); \ 202 rcu_tasks_trace_qs(t); \ 203} while (0) 204 205# ifdef CONFIG_TASKS_RUDE_RCU 206void call_rcu_tasks_rude(struct rcu_head *head, rcu_callback_t func); 207void synchronize_rcu_tasks_rude(void); 208# endif 209 210#define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false) 211void exit_tasks_rcu_start(void); 212void exit_tasks_rcu_stop(void); 213void exit_tasks_rcu_finish(void); 214#else /* #ifdef CONFIG_TASKS_RCU_GENERIC */ 215#define rcu_tasks_classic_qs(t, preempt) do { } while (0) 216#define rcu_tasks_qs(t, preempt) do { } while (0) 217#define rcu_note_voluntary_context_switch(t) do { } while (0) 218#define call_rcu_tasks call_rcu 219#define synchronize_rcu_tasks synchronize_rcu 220static inline void exit_tasks_rcu_start(void) { } 221static inline void exit_tasks_rcu_stop(void) { } 222static inline void exit_tasks_rcu_finish(void) { } 223#endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */ 224 225/** 226 * rcu_trace_implies_rcu_gp - does an RCU Tasks Trace grace period imply an RCU grace period? 227 * 228 * As an accident of implementation, an RCU Tasks Trace grace period also 229 * acts as an RCU grace period. However, this could change at any time. 230 * Code relying on this accident must call this function to verify that 231 * this accident is still happening. 232 * 233 * You have been warned! 234 */ 235static inline bool rcu_trace_implies_rcu_gp(void) { return true; } 236 237/** 238 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU 239 * 240 * This macro resembles cond_resched(), except that it is defined to 241 * report potential quiescent states to RCU-tasks even if the cond_resched() 242 * machinery were to be shut off, as some advocate for PREEMPTION kernels. 243 */ 244#define cond_resched_tasks_rcu_qs() \ 245do { \ 246 rcu_tasks_qs(current, false); \ 247 cond_resched(); \ 248} while (0) 249 250/** 251 * rcu_softirq_qs_periodic - Report RCU and RCU-Tasks quiescent states 252 * @old_ts: jiffies at start of processing. 253 * 254 * This helper is for long-running softirq handlers, such as NAPI threads in 255 * networking. The caller should initialize the variable passed in as @old_ts 256 * at the beginning of the softirq handler. When invoked frequently, this macro 257 * will invoke rcu_softirq_qs() every 100 milliseconds thereafter, which will 258 * provide both RCU and RCU-Tasks quiescent states. Note that this macro 259 * modifies its old_ts argument. 260 * 261 * Because regions of code that have disabled softirq act as RCU read-side 262 * critical sections, this macro should be invoked with softirq (and 263 * preemption) enabled. 264 * 265 * The macro is not needed when CONFIG_PREEMPT_RT is defined. RT kernels would 266 * have more chance to invoke schedule() calls and provide necessary quiescent 267 * states. As a contrast, calling cond_resched() only won't achieve the same 268 * effect because cond_resched() does not provide RCU-Tasks quiescent states. 269 */ 270#define rcu_softirq_qs_periodic(old_ts) \ 271do { \ 272 if (!IS_ENABLED(CONFIG_PREEMPT_RT) && \ 273 time_after(jiffies, (old_ts) + HZ / 10)) { \ 274 preempt_disable(); \ 275 rcu_softirq_qs(); \ 276 preempt_enable(); \ 277 (old_ts) = jiffies; \ 278 } \ 279} while (0) 280 281/* 282 * Infrastructure to implement the synchronize_() primitives in 283 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 284 */ 285 286#if defined(CONFIG_TREE_RCU) 287#include <linux/rcutree.h> 288#elif defined(CONFIG_TINY_RCU) 289#include <linux/rcutiny.h> 290#else 291#error "Unknown RCU implementation specified to kernel configuration" 292#endif 293 294/* 295 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls 296 * are needed for dynamic initialization and destruction of rcu_head 297 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for 298 * dynamic initialization and destruction of statically allocated rcu_head 299 * structures. However, rcu_head structures allocated dynamically in the 300 * heap don't need any initialization. 301 */ 302#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 303void init_rcu_head(struct rcu_head *head); 304void destroy_rcu_head(struct rcu_head *head); 305void init_rcu_head_on_stack(struct rcu_head *head); 306void destroy_rcu_head_on_stack(struct rcu_head *head); 307#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 308static inline void init_rcu_head(struct rcu_head *head) { } 309static inline void destroy_rcu_head(struct rcu_head *head) { } 310static inline void init_rcu_head_on_stack(struct rcu_head *head) { } 311static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { } 312#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 313 314#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 315bool rcu_lockdep_current_cpu_online(void); 316#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 317static inline bool rcu_lockdep_current_cpu_online(void) { return true; } 318#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 319 320extern struct lockdep_map rcu_lock_map; 321extern struct lockdep_map rcu_bh_lock_map; 322extern struct lockdep_map rcu_sched_lock_map; 323extern struct lockdep_map rcu_callback_map; 324 325#ifdef CONFIG_DEBUG_LOCK_ALLOC 326 327static inline void rcu_lock_acquire(struct lockdep_map *map) 328{ 329 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); 330} 331 332static inline void rcu_try_lock_acquire(struct lockdep_map *map) 333{ 334 lock_acquire(map, 0, 1, 2, 0, NULL, _THIS_IP_); 335} 336 337static inline void rcu_lock_release(struct lockdep_map *map) 338{ 339 lock_release(map, _THIS_IP_); 340} 341 342int debug_lockdep_rcu_enabled(void); 343int rcu_read_lock_held(void); 344int rcu_read_lock_bh_held(void); 345int rcu_read_lock_sched_held(void); 346int rcu_read_lock_any_held(void); 347 348#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 349 350# define rcu_lock_acquire(a) do { } while (0) 351# define rcu_try_lock_acquire(a) do { } while (0) 352# define rcu_lock_release(a) do { } while (0) 353 354static inline int rcu_read_lock_held(void) 355{ 356 return 1; 357} 358 359static inline int rcu_read_lock_bh_held(void) 360{ 361 return 1; 362} 363 364static inline int rcu_read_lock_sched_held(void) 365{ 366 return !preemptible(); 367} 368 369static inline int rcu_read_lock_any_held(void) 370{ 371 return !preemptible(); 372} 373 374static inline int debug_lockdep_rcu_enabled(void) 375{ 376 return 0; 377} 378 379#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 380 381#ifdef CONFIG_PROVE_RCU 382 383/** 384 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met 385 * @c: condition to check 386 * @s: informative message 387 * 388 * This checks debug_lockdep_rcu_enabled() before checking (c) to 389 * prevent early boot splats due to lockdep not yet being initialized, 390 * and rechecks it after checking (c) to prevent false-positive splats 391 * due to races with lockdep being disabled. See commit 3066820034b5dd 392 * ("rcu: Reject RCU_LOCKDEP_WARN() false positives") for more detail. 393 */ 394#define RCU_LOCKDEP_WARN(c, s) \ 395 do { \ 396 static bool __section(".data.unlikely") __warned; \ 397 if (debug_lockdep_rcu_enabled() && (c) && \ 398 debug_lockdep_rcu_enabled() && !__warned) { \ 399 __warned = true; \ 400 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 401 } \ 402 } while (0) 403 404#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 405static inline void rcu_preempt_sleep_check(void) 406{ 407 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), 408 "Illegal context switch in RCU read-side critical section"); 409} 410#else /* #ifdef CONFIG_PROVE_RCU */ 411static inline void rcu_preempt_sleep_check(void) { } 412#endif /* #else #ifdef CONFIG_PROVE_RCU */ 413 414#define rcu_sleep_check() \ 415 do { \ 416 rcu_preempt_sleep_check(); \ 417 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) \ 418 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ 419 "Illegal context switch in RCU-bh read-side critical section"); \ 420 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ 421 "Illegal context switch in RCU-sched read-side critical section"); \ 422 } while (0) 423 424#else /* #ifdef CONFIG_PROVE_RCU */ 425 426#define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c)) 427#define rcu_sleep_check() do { } while (0) 428 429#endif /* #else #ifdef CONFIG_PROVE_RCU */ 430 431/* 432 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 433 * and rcu_assign_pointer(). Some of these could be folded into their 434 * callers, but they are left separate in order to ease introduction of 435 * multiple pointers markings to match different RCU implementations 436 * (e.g., __srcu), should this make sense in the future. 437 */ 438 439#ifdef __CHECKER__ 440#define rcu_check_sparse(p, space) \ 441 ((void)(((typeof(*p) space *)p) == p)) 442#else /* #ifdef __CHECKER__ */ 443#define rcu_check_sparse(p, space) 444#endif /* #else #ifdef __CHECKER__ */ 445 446#define __unrcu_pointer(p, local) \ 447({ \ 448 typeof(*p) *local = (typeof(*p) *__force)(p); \ 449 rcu_check_sparse(p, __rcu); \ 450 ((typeof(*p) __force __kernel *)(local)); \ 451}) 452/** 453 * unrcu_pointer - mark a pointer as not being RCU protected 454 * @p: pointer needing to lose its __rcu property 455 * 456 * Converts @p from an __rcu pointer to a __kernel pointer. 457 * This allows an __rcu pointer to be used with xchg() and friends. 458 */ 459#define unrcu_pointer(p) __unrcu_pointer(p, __UNIQUE_ID(rcu)) 460 461#define __rcu_access_pointer(p, local, space) \ 462({ \ 463 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \ 464 rcu_check_sparse(p, space); \ 465 ((typeof(*p) __force __kernel *)(local)); \ 466}) 467#define __rcu_dereference_check(p, local, c, space) \ 468({ \ 469 /* Dependency order vs. p above. */ \ 470 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \ 471 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ 472 rcu_check_sparse(p, space); \ 473 ((typeof(*p) __force __kernel *)(local)); \ 474}) 475#define __rcu_dereference_protected(p, local, c, space) \ 476({ \ 477 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ 478 rcu_check_sparse(p, space); \ 479 ((typeof(*p) __force __kernel *)(p)); \ 480}) 481#define __rcu_dereference_raw(p, local) \ 482({ \ 483 /* Dependency order vs. p above. */ \ 484 typeof(p) local = READ_ONCE(p); \ 485 ((typeof(*p) __force __kernel *)(local)); \ 486}) 487#define rcu_dereference_raw(p) __rcu_dereference_raw(p, __UNIQUE_ID(rcu)) 488 489/** 490 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 491 * @v: The value to statically initialize with. 492 */ 493#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 494 495/** 496 * rcu_assign_pointer() - assign to RCU-protected pointer 497 * @p: pointer to assign to 498 * @v: value to assign (publish) 499 * 500 * Assigns the specified value to the specified RCU-protected 501 * pointer, ensuring that any concurrent RCU readers will see 502 * any prior initialization. 503 * 504 * Inserts memory barriers on architectures that require them 505 * (which is most of them), and also prevents the compiler from 506 * reordering the code that initializes the structure after the pointer 507 * assignment. More importantly, this call documents which pointers 508 * will be dereferenced by RCU read-side code. 509 * 510 * In some special cases, you may use RCU_INIT_POINTER() instead 511 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 512 * to the fact that it does not constrain either the CPU or the compiler. 513 * That said, using RCU_INIT_POINTER() when you should have used 514 * rcu_assign_pointer() is a very bad thing that results in 515 * impossible-to-diagnose memory corruption. So please be careful. 516 * See the RCU_INIT_POINTER() comment header for details. 517 * 518 * Note that rcu_assign_pointer() evaluates each of its arguments only 519 * once, appearances notwithstanding. One of the "extra" evaluations 520 * is in typeof() and the other visible only to sparse (__CHECKER__), 521 * neither of which actually execute the argument. As with most cpp 522 * macros, this execute-arguments-only-once property is important, so 523 * please be careful when making changes to rcu_assign_pointer() and the 524 * other macros that it invokes. 525 */ 526#define rcu_assign_pointer(p, v) \ 527do { \ 528 uintptr_t _r_a_p__v = (uintptr_t)(v); \ 529 rcu_check_sparse(p, __rcu); \ 530 \ 531 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \ 532 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \ 533 else \ 534 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \ 535} while (0) 536 537/** 538 * rcu_replace_pointer() - replace an RCU pointer, returning its old value 539 * @rcu_ptr: RCU pointer, whose old value is returned 540 * @ptr: regular pointer 541 * @c: the lockdep conditions under which the dereference will take place 542 * 543 * Perform a replacement, where @rcu_ptr is an RCU-annotated 544 * pointer and @c is the lockdep argument that is passed to the 545 * rcu_dereference_protected() call used to read that pointer. The old 546 * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr. 547 */ 548#define rcu_replace_pointer(rcu_ptr, ptr, c) \ 549({ \ 550 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \ 551 rcu_assign_pointer((rcu_ptr), (ptr)); \ 552 __tmp; \ 553}) 554 555/** 556 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 557 * @p: The pointer to read 558 * 559 * Return the value of the specified RCU-protected pointer, but omit the 560 * lockdep checks for being in an RCU read-side critical section. This is 561 * useful when the value of this pointer is accessed, but the pointer is 562 * not dereferenced, for example, when testing an RCU-protected pointer 563 * against NULL. Although rcu_access_pointer() may also be used in cases 564 * where update-side locks prevent the value of the pointer from changing, 565 * you should instead use rcu_dereference_protected() for this use case. 566 * Within an RCU read-side critical section, there is little reason to 567 * use rcu_access_pointer(). 568 * 569 * It is usually best to test the rcu_access_pointer() return value 570 * directly in order to avoid accidental dereferences being introduced 571 * by later inattentive changes. In other words, assigning the 572 * rcu_access_pointer() return value to a local variable results in an 573 * accident waiting to happen. 574 * 575 * It is also permissible to use rcu_access_pointer() when read-side 576 * access to the pointer was removed at least one grace period ago, as is 577 * the case in the context of the RCU callback that is freeing up the data, 578 * or after a synchronize_rcu() returns. This can be useful when tearing 579 * down multi-linked structures after a grace period has elapsed. However, 580 * rcu_dereference_protected() is normally preferred for this use case. 581 */ 582#define rcu_access_pointer(p) __rcu_access_pointer((p), __UNIQUE_ID(rcu), __rcu) 583 584/** 585 * rcu_dereference_check() - rcu_dereference with debug checking 586 * @p: The pointer to read, prior to dereferencing 587 * @c: The conditions under which the dereference will take place 588 * 589 * Do an rcu_dereference(), but check that the conditions under which the 590 * dereference will take place are correct. Typically the conditions 591 * indicate the various locking conditions that should be held at that 592 * point. The check should return true if the conditions are satisfied. 593 * An implicit check for being in an RCU read-side critical section 594 * (rcu_read_lock()) is included. 595 * 596 * For example: 597 * 598 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 599 * 600 * could be used to indicate to lockdep that foo->bar may only be dereferenced 601 * if either rcu_read_lock() is held, or that the lock required to replace 602 * the bar struct at foo->bar is held. 603 * 604 * Note that the list of conditions may also include indications of when a lock 605 * need not be held, for example during initialisation or destruction of the 606 * target struct: 607 * 608 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 609 * atomic_read(&foo->usage) == 0); 610 * 611 * Inserts memory barriers on architectures that require them 612 * (currently only the Alpha), prevents the compiler from refetching 613 * (and from merging fetches), and, more importantly, documents exactly 614 * which pointers are protected by RCU and checks that the pointer is 615 * annotated as __rcu. 616 */ 617#define rcu_dereference_check(p, c) \ 618 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \ 619 (c) || rcu_read_lock_held(), __rcu) 620 621/** 622 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 623 * @p: The pointer to read, prior to dereferencing 624 * @c: The conditions under which the dereference will take place 625 * 626 * This is the RCU-bh counterpart to rcu_dereference_check(). However, 627 * please note that starting in v5.0 kernels, vanilla RCU grace periods 628 * wait for local_bh_disable() regions of code in addition to regions of 629 * code demarked by rcu_read_lock() and rcu_read_unlock(). This means 630 * that synchronize_rcu(), call_rcu, and friends all take not only 631 * rcu_read_lock() but also rcu_read_lock_bh() into account. 632 */ 633#define rcu_dereference_bh_check(p, c) \ 634 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \ 635 (c) || rcu_read_lock_bh_held(), __rcu) 636 637/** 638 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 639 * @p: The pointer to read, prior to dereferencing 640 * @c: The conditions under which the dereference will take place 641 * 642 * This is the RCU-sched counterpart to rcu_dereference_check(). 643 * However, please note that starting in v5.0 kernels, vanilla RCU grace 644 * periods wait for preempt_disable() regions of code in addition to 645 * regions of code demarked by rcu_read_lock() and rcu_read_unlock(). 646 * This means that synchronize_rcu(), call_rcu, and friends all take not 647 * only rcu_read_lock() but also rcu_read_lock_sched() into account. 648 */ 649#define rcu_dereference_sched_check(p, c) \ 650 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \ 651 (c) || rcu_read_lock_sched_held(), \ 652 __rcu) 653 654/* 655 * The tracing infrastructure traces RCU (we want that), but unfortunately 656 * some of the RCU checks causes tracing to lock up the system. 657 * 658 * The no-tracing version of rcu_dereference_raw() must not call 659 * rcu_read_lock_held(). 660 */ 661#define rcu_dereference_raw_check(p) \ 662 __rcu_dereference_check((p), __UNIQUE_ID(rcu), 1, __rcu) 663 664/** 665 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 666 * @p: The pointer to read, prior to dereferencing 667 * @c: The conditions under which the dereference will take place 668 * 669 * Return the value of the specified RCU-protected pointer, but omit 670 * the READ_ONCE(). This is useful in cases where update-side locks 671 * prevent the value of the pointer from changing. Please note that this 672 * primitive does *not* prevent the compiler from repeating this reference 673 * or combining it with other references, so it should not be used without 674 * protection of appropriate locks. 675 * 676 * This function is only for update-side use. Using this function 677 * when protected only by rcu_read_lock() will result in infrequent 678 * but very ugly failures. 679 */ 680#define rcu_dereference_protected(p, c) \ 681 __rcu_dereference_protected((p), __UNIQUE_ID(rcu), (c), __rcu) 682 683 684/** 685 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 686 * @p: The pointer to read, prior to dereferencing 687 * 688 * This is a simple wrapper around rcu_dereference_check(). 689 */ 690#define rcu_dereference(p) rcu_dereference_check(p, 0) 691 692/** 693 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 694 * @p: The pointer to read, prior to dereferencing 695 * 696 * Makes rcu_dereference_check() do the dirty work. 697 */ 698#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 699 700/** 701 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 702 * @p: The pointer to read, prior to dereferencing 703 * 704 * Makes rcu_dereference_check() do the dirty work. 705 */ 706#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 707 708/** 709 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism 710 * @p: The pointer to hand off 711 * 712 * This is simply an identity function, but it documents where a pointer 713 * is handed off from RCU to some other synchronization mechanism, for 714 * example, reference counting or locking. In C11, it would map to 715 * kill_dependency(). It could be used as follows:: 716 * 717 * rcu_read_lock(); 718 * p = rcu_dereference(gp); 719 * long_lived = is_long_lived(p); 720 * if (long_lived) { 721 * if (!atomic_inc_not_zero(p->refcnt)) 722 * long_lived = false; 723 * else 724 * p = rcu_pointer_handoff(p); 725 * } 726 * rcu_read_unlock(); 727 */ 728#define rcu_pointer_handoff(p) (p) 729 730/** 731 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 732 * 733 * When synchronize_rcu() is invoked on one CPU while other CPUs 734 * are within RCU read-side critical sections, then the 735 * synchronize_rcu() is guaranteed to block until after all the other 736 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 737 * on one CPU while other CPUs are within RCU read-side critical 738 * sections, invocation of the corresponding RCU callback is deferred 739 * until after the all the other CPUs exit their critical sections. 740 * 741 * In v5.0 and later kernels, synchronize_rcu() and call_rcu() also 742 * wait for regions of code with preemption disabled, including regions of 743 * code with interrupts or softirqs disabled. In pre-v5.0 kernels, which 744 * define synchronize_sched(), only code enclosed within rcu_read_lock() 745 * and rcu_read_unlock() are guaranteed to be waited for. 746 * 747 * Note, however, that RCU callbacks are permitted to run concurrently 748 * with new RCU read-side critical sections. One way that this can happen 749 * is via the following sequence of events: (1) CPU 0 enters an RCU 750 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 751 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 752 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 753 * callback is invoked. This is legal, because the RCU read-side critical 754 * section that was running concurrently with the call_rcu() (and which 755 * therefore might be referencing something that the corresponding RCU 756 * callback would free up) has completed before the corresponding 757 * RCU callback is invoked. 758 * 759 * RCU read-side critical sections may be nested. Any deferred actions 760 * will be deferred until the outermost RCU read-side critical section 761 * completes. 762 * 763 * You can avoid reading and understanding the next paragraph by 764 * following this rule: don't put anything in an rcu_read_lock() RCU 765 * read-side critical section that would block in a !PREEMPTION kernel. 766 * But if you want the full story, read on! 767 * 768 * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU), 769 * it is illegal to block while in an RCU read-side critical section. 770 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION 771 * kernel builds, RCU read-side critical sections may be preempted, 772 * but explicit blocking is illegal. Finally, in preemptible RCU 773 * implementations in real-time (with -rt patchset) kernel builds, RCU 774 * read-side critical sections may be preempted and they may also block, but 775 * only when acquiring spinlocks that are subject to priority inheritance. 776 */ 777static __always_inline void rcu_read_lock(void) 778{ 779 __rcu_read_lock(); 780 __acquire(RCU); 781 rcu_lock_acquire(&rcu_lock_map); 782 RCU_LOCKDEP_WARN(!rcu_is_watching(), 783 "rcu_read_lock() used illegally while idle"); 784} 785 786/* 787 * So where is rcu_write_lock()? It does not exist, as there is no 788 * way for writers to lock out RCU readers. This is a feature, not 789 * a bug -- this property is what provides RCU's performance benefits. 790 * Of course, writers must coordinate with each other. The normal 791 * spinlock primitives work well for this, but any other technique may be 792 * used as well. RCU does not care how the writers keep out of each 793 * others' way, as long as they do so. 794 */ 795 796/** 797 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 798 * 799 * In almost all situations, rcu_read_unlock() is immune from deadlock. 800 * In recent kernels that have consolidated synchronize_sched() and 801 * synchronize_rcu_bh() into synchronize_rcu(), this deadlock immunity 802 * also extends to the scheduler's runqueue and priority-inheritance 803 * spinlocks, courtesy of the quiescent-state deferral that is carried 804 * out when rcu_read_unlock() is invoked with interrupts disabled. 805 * 806 * See rcu_read_lock() for more information. 807 */ 808static inline void rcu_read_unlock(void) 809{ 810 RCU_LOCKDEP_WARN(!rcu_is_watching(), 811 "rcu_read_unlock() used illegally while idle"); 812 __release(RCU); 813 __rcu_read_unlock(); 814 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ 815} 816 817/** 818 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 819 * 820 * This is equivalent to rcu_read_lock(), but also disables softirqs. 821 * Note that anything else that disables softirqs can also serve as an RCU 822 * read-side critical section. However, please note that this equivalence 823 * applies only to v5.0 and later. Before v5.0, rcu_read_lock() and 824 * rcu_read_lock_bh() were unrelated. 825 * 826 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 827 * must occur in the same context, for example, it is illegal to invoke 828 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 829 * was invoked from some other task. 830 */ 831static inline void rcu_read_lock_bh(void) 832{ 833 local_bh_disable(); 834 __acquire(RCU_BH); 835 rcu_lock_acquire(&rcu_bh_lock_map); 836 RCU_LOCKDEP_WARN(!rcu_is_watching(), 837 "rcu_read_lock_bh() used illegally while idle"); 838} 839 840/** 841 * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section 842 * 843 * See rcu_read_lock_bh() for more information. 844 */ 845static inline void rcu_read_unlock_bh(void) 846{ 847 RCU_LOCKDEP_WARN(!rcu_is_watching(), 848 "rcu_read_unlock_bh() used illegally while idle"); 849 rcu_lock_release(&rcu_bh_lock_map); 850 __release(RCU_BH); 851 local_bh_enable(); 852} 853 854/** 855 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 856 * 857 * This is equivalent to rcu_read_lock(), but also disables preemption. 858 * Read-side critical sections can also be introduced by anything else that 859 * disables preemption, including local_irq_disable() and friends. However, 860 * please note that the equivalence to rcu_read_lock() applies only to 861 * v5.0 and later. Before v5.0, rcu_read_lock() and rcu_read_lock_sched() 862 * were unrelated. 863 * 864 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 865 * must occur in the same context, for example, it is illegal to invoke 866 * rcu_read_unlock_sched() from process context if the matching 867 * rcu_read_lock_sched() was invoked from an NMI handler. 868 */ 869static inline void rcu_read_lock_sched(void) 870{ 871 preempt_disable(); 872 __acquire(RCU_SCHED); 873 rcu_lock_acquire(&rcu_sched_lock_map); 874 RCU_LOCKDEP_WARN(!rcu_is_watching(), 875 "rcu_read_lock_sched() used illegally while idle"); 876} 877 878/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 879static inline notrace void rcu_read_lock_sched_notrace(void) 880{ 881 preempt_disable_notrace(); 882 __acquire(RCU_SCHED); 883} 884 885/** 886 * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section 887 * 888 * See rcu_read_lock_sched() for more information. 889 */ 890static inline void rcu_read_unlock_sched(void) 891{ 892 RCU_LOCKDEP_WARN(!rcu_is_watching(), 893 "rcu_read_unlock_sched() used illegally while idle"); 894 rcu_lock_release(&rcu_sched_lock_map); 895 __release(RCU_SCHED); 896 preempt_enable(); 897} 898 899/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 900static inline notrace void rcu_read_unlock_sched_notrace(void) 901{ 902 __release(RCU_SCHED); 903 preempt_enable_notrace(); 904} 905 906/** 907 * RCU_INIT_POINTER() - initialize an RCU protected pointer 908 * @p: The pointer to be initialized. 909 * @v: The value to initialized the pointer to. 910 * 911 * Initialize an RCU-protected pointer in special cases where readers 912 * do not need ordering constraints on the CPU or the compiler. These 913 * special cases are: 914 * 915 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or* 916 * 2. The caller has taken whatever steps are required to prevent 917 * RCU readers from concurrently accessing this pointer *or* 918 * 3. The referenced data structure has already been exposed to 919 * readers either at compile time or via rcu_assign_pointer() *and* 920 * 921 * a. You have not made *any* reader-visible changes to 922 * this structure since then *or* 923 * b. It is OK for readers accessing this structure from its 924 * new location to see the old state of the structure. (For 925 * example, the changes were to statistical counters or to 926 * other state where exact synchronization is not required.) 927 * 928 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 929 * result in impossible-to-diagnose memory corruption. As in the structures 930 * will look OK in crash dumps, but any concurrent RCU readers might 931 * see pre-initialized values of the referenced data structure. So 932 * please be very careful how you use RCU_INIT_POINTER()!!! 933 * 934 * If you are creating an RCU-protected linked structure that is accessed 935 * by a single external-to-structure RCU-protected pointer, then you may 936 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 937 * pointers, but you must use rcu_assign_pointer() to initialize the 938 * external-to-structure pointer *after* you have completely initialized 939 * the reader-accessible portions of the linked structure. 940 * 941 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 942 * ordering guarantees for either the CPU or the compiler. 943 */ 944#define RCU_INIT_POINTER(p, v) \ 945 do { \ 946 rcu_check_sparse(p, __rcu); \ 947 WRITE_ONCE(p, RCU_INITIALIZER(v)); \ 948 } while (0) 949 950/** 951 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 952 * @p: The pointer to be initialized. 953 * @v: The value to initialized the pointer to. 954 * 955 * GCC-style initialization for an RCU-protected pointer in a structure field. 956 */ 957#define RCU_POINTER_INITIALIZER(p, v) \ 958 .p = RCU_INITIALIZER(v) 959 960/* 961 * Does the specified offset indicate that the corresponding rcu_head 962 * structure can be handled by kvfree_rcu()? 963 */ 964#define __is_kvfree_rcu_offset(offset) ((offset) < 4096) 965 966/** 967 * kfree_rcu() - kfree an object after a grace period. 968 * @ptr: pointer to kfree for double-argument invocations. 969 * @rhf: the name of the struct rcu_head within the type of @ptr. 970 * 971 * Many rcu callbacks functions just call kfree() on the base structure. 972 * These functions are trivial, but their size adds up, and furthermore 973 * when they are used in a kernel module, that module must invoke the 974 * high-latency rcu_barrier() function at module-unload time. 975 * 976 * The kfree_rcu() function handles this issue. Rather than encoding a 977 * function address in the embedded rcu_head structure, kfree_rcu() instead 978 * encodes the offset of the rcu_head structure within the base structure. 979 * Because the functions are not allowed in the low-order 4096 bytes of 980 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 981 * If the offset is larger than 4095 bytes, a compile-time error will 982 * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can 983 * either fall back to use of call_rcu() or rearrange the structure to 984 * position the rcu_head structure into the first 4096 bytes. 985 * 986 * The object to be freed can be allocated either by kmalloc() or 987 * kmem_cache_alloc(). 988 * 989 * Note that the allowable offset might decrease in the future. 990 * 991 * The BUILD_BUG_ON check must not involve any function calls, hence the 992 * checks are done in macros here. 993 */ 994#define kfree_rcu(ptr, rhf) kvfree_rcu_arg_2(ptr, rhf) 995#define kvfree_rcu(ptr, rhf) kvfree_rcu_arg_2(ptr, rhf) 996 997/** 998 * kfree_rcu_mightsleep() - kfree an object after a grace period. 999 * @ptr: pointer to kfree for single-argument invocations. 1000 * 1001 * When it comes to head-less variant, only one argument 1002 * is passed and that is just a pointer which has to be 1003 * freed after a grace period. Therefore the semantic is 1004 * 1005 * kfree_rcu_mightsleep(ptr); 1006 * 1007 * where @ptr is the pointer to be freed by kvfree(). 1008 * 1009 * Please note, head-less way of freeing is permitted to 1010 * use from a context that has to follow might_sleep() 1011 * annotation. Otherwise, please switch and embed the 1012 * rcu_head structure within the type of @ptr. 1013 */ 1014#define kfree_rcu_mightsleep(ptr) kvfree_rcu_arg_1(ptr) 1015#define kvfree_rcu_mightsleep(ptr) kvfree_rcu_arg_1(ptr) 1016 1017#define kvfree_rcu_arg_2(ptr, rhf) \ 1018do { \ 1019 typeof (ptr) ___p = (ptr); \ 1020 \ 1021 if (___p) { \ 1022 BUILD_BUG_ON(!__is_kvfree_rcu_offset(offsetof(typeof(*(ptr)), rhf))); \ 1023 kvfree_call_rcu(&((___p)->rhf), (void *) (___p)); \ 1024 } \ 1025} while (0) 1026 1027#define kvfree_rcu_arg_1(ptr) \ 1028do { \ 1029 typeof(ptr) ___p = (ptr); \ 1030 \ 1031 if (___p) \ 1032 kvfree_call_rcu(NULL, (void *) (___p)); \ 1033} while (0) 1034 1035/* 1036 * Place this after a lock-acquisition primitive to guarantee that 1037 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies 1038 * if the UNLOCK and LOCK are executed by the same CPU or if the 1039 * UNLOCK and LOCK operate on the same lock variable. 1040 */ 1041#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE 1042#define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */ 1043#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 1044#define smp_mb__after_unlock_lock() do { } while (0) 1045#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 1046 1047 1048/* Has the specified rcu_head structure been handed to call_rcu()? */ 1049 1050/** 1051 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu() 1052 * @rhp: The rcu_head structure to initialize. 1053 * 1054 * If you intend to invoke rcu_head_after_call_rcu() to test whether a 1055 * given rcu_head structure has already been passed to call_rcu(), then 1056 * you must also invoke this rcu_head_init() function on it just after 1057 * allocating that structure. Calls to this function must not race with 1058 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation. 1059 */ 1060static inline void rcu_head_init(struct rcu_head *rhp) 1061{ 1062 rhp->func = (rcu_callback_t)~0L; 1063} 1064 1065/** 1066 * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()? 1067 * @rhp: The rcu_head structure to test. 1068 * @f: The function passed to call_rcu() along with @rhp. 1069 * 1070 * Returns @true if the @rhp has been passed to call_rcu() with @func, 1071 * and @false otherwise. Emits a warning in any other case, including 1072 * the case where @rhp has already been invoked after a grace period. 1073 * Calls to this function must not race with callback invocation. One way 1074 * to avoid such races is to enclose the call to rcu_head_after_call_rcu() 1075 * in an RCU read-side critical section that includes a read-side fetch 1076 * of the pointer to the structure containing @rhp. 1077 */ 1078static inline bool 1079rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f) 1080{ 1081 rcu_callback_t func = READ_ONCE(rhp->func); 1082 1083 if (func == f) 1084 return true; 1085 WARN_ON_ONCE(func != (rcu_callback_t)~0L); 1086 return false; 1087} 1088 1089/* kernel/ksysfs.c definitions */ 1090extern int rcu_expedited; 1091extern int rcu_normal; 1092 1093DEFINE_LOCK_GUARD_0(rcu, rcu_read_lock(), rcu_read_unlock()) 1094 1095#endif /* __LINUX_RCUPDATE_H */ 1096