1/* 2 * Read-Copy Update mechanism for mutual exclusion 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU General Public License as published by 6 * the Free Software Foundation; either version 2 of the License, or 7 * (at your option) any later version. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, you can access it online at 16 * http://www.gnu.org/licenses/gpl-2.0.html. 17 * 18 * Copyright IBM Corporation, 2001 19 * 20 * Author: Dipankar Sarma <dipankar@in.ibm.com> 21 * 22 * Based on the original work by Paul McKenney <paulmck@us.ibm.com> 23 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. 24 * Papers: 25 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf 26 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001) 27 * 28 * For detailed explanation of Read-Copy Update mechanism see - 29 * http://lse.sourceforge.net/locking/rcupdate.html 30 * 31 */ 32 33#ifndef __LINUX_RCUPDATE_H 34#define __LINUX_RCUPDATE_H 35 36#include <linux/types.h> 37#include <linux/cache.h> 38#include <linux/spinlock.h> 39#include <linux/threads.h> 40#include <linux/cpumask.h> 41#include <linux/seqlock.h> 42#include <linux/lockdep.h> 43#include <linux/completion.h> 44#include <linux/debugobjects.h> 45#include <linux/bug.h> 46#include <linux/compiler.h> 47#include <linux/ktime.h> 48 49#include <asm/barrier.h> 50 51#ifndef CONFIG_TINY_RCU 52extern int rcu_expedited; /* for sysctl */ 53extern int rcu_normal; /* also for sysctl */ 54#endif /* #ifndef CONFIG_TINY_RCU */ 55 56#ifdef CONFIG_TINY_RCU 57/* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */ 58static inline bool rcu_gp_is_normal(void) /* Internal RCU use. */ 59{ 60 return true; 61} 62static inline bool rcu_gp_is_expedited(void) /* Internal RCU use. */ 63{ 64 return false; 65} 66 67static inline void rcu_expedite_gp(void) 68{ 69} 70 71static inline void rcu_unexpedite_gp(void) 72{ 73} 74#else /* #ifdef CONFIG_TINY_RCU */ 75bool rcu_gp_is_normal(void); /* Internal RCU use. */ 76bool rcu_gp_is_expedited(void); /* Internal RCU use. */ 77void rcu_expedite_gp(void); 78void rcu_unexpedite_gp(void); 79#endif /* #else #ifdef CONFIG_TINY_RCU */ 80 81enum rcutorture_type { 82 RCU_FLAVOR, 83 RCU_BH_FLAVOR, 84 RCU_SCHED_FLAVOR, 85 RCU_TASKS_FLAVOR, 86 SRCU_FLAVOR, 87 INVALID_RCU_FLAVOR 88}; 89 90#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) 91void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, 92 unsigned long *gpnum, unsigned long *completed); 93void rcutorture_record_test_transition(void); 94void rcutorture_record_progress(unsigned long vernum); 95void do_trace_rcu_torture_read(const char *rcutorturename, 96 struct rcu_head *rhp, 97 unsigned long secs, 98 unsigned long c_old, 99 unsigned long c); 100#else 101static inline void rcutorture_get_gp_data(enum rcutorture_type test_type, 102 int *flags, 103 unsigned long *gpnum, 104 unsigned long *completed) 105{ 106 *flags = 0; 107 *gpnum = 0; 108 *completed = 0; 109} 110static inline void rcutorture_record_test_transition(void) 111{ 112} 113static inline void rcutorture_record_progress(unsigned long vernum) 114{ 115} 116#ifdef CONFIG_RCU_TRACE 117void do_trace_rcu_torture_read(const char *rcutorturename, 118 struct rcu_head *rhp, 119 unsigned long secs, 120 unsigned long c_old, 121 unsigned long c); 122#else 123#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ 124 do { } while (0) 125#endif 126#endif 127 128#define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b)) 129#define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b)) 130#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) 131#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) 132#define ulong2long(a) (*(long *)(&(a))) 133 134/* Exported common interfaces */ 135 136#ifdef CONFIG_PREEMPT_RCU 137 138/** 139 * call_rcu() - Queue an RCU callback for invocation after a grace period. 140 * @head: structure to be used for queueing the RCU updates. 141 * @func: actual callback function to be invoked after the grace period 142 * 143 * The callback function will be invoked some time after a full grace 144 * period elapses, in other words after all pre-existing RCU read-side 145 * critical sections have completed. However, the callback function 146 * might well execute concurrently with RCU read-side critical sections 147 * that started after call_rcu() was invoked. RCU read-side critical 148 * sections are delimited by rcu_read_lock() and rcu_read_unlock(), 149 * and may be nested. 150 * 151 * Note that all CPUs must agree that the grace period extended beyond 152 * all pre-existing RCU read-side critical section. On systems with more 153 * than one CPU, this means that when "func()" is invoked, each CPU is 154 * guaranteed to have executed a full memory barrier since the end of its 155 * last RCU read-side critical section whose beginning preceded the call 156 * to call_rcu(). It also means that each CPU executing an RCU read-side 157 * critical section that continues beyond the start of "func()" must have 158 * executed a memory barrier after the call_rcu() but before the beginning 159 * of that RCU read-side critical section. Note that these guarantees 160 * include CPUs that are offline, idle, or executing in user mode, as 161 * well as CPUs that are executing in the kernel. 162 * 163 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the 164 * resulting RCU callback function "func()", then both CPU A and CPU B are 165 * guaranteed to execute a full memory barrier during the time interval 166 * between the call to call_rcu() and the invocation of "func()" -- even 167 * if CPU A and CPU B are the same CPU (but again only if the system has 168 * more than one CPU). 169 */ 170void call_rcu(struct rcu_head *head, 171 rcu_callback_t func); 172 173#else /* #ifdef CONFIG_PREEMPT_RCU */ 174 175/* In classic RCU, call_rcu() is just call_rcu_sched(). */ 176#define call_rcu call_rcu_sched 177 178#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 179 180/** 181 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period. 182 * @head: structure to be used for queueing the RCU updates. 183 * @func: actual callback function to be invoked after the grace period 184 * 185 * The callback function will be invoked some time after a full grace 186 * period elapses, in other words after all currently executing RCU 187 * read-side critical sections have completed. call_rcu_bh() assumes 188 * that the read-side critical sections end on completion of a softirq 189 * handler. This means that read-side critical sections in process 190 * context must not be interrupted by softirqs. This interface is to be 191 * used when most of the read-side critical sections are in softirq context. 192 * RCU read-side critical sections are delimited by : 193 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context. 194 * OR 195 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context. 196 * These may be nested. 197 * 198 * See the description of call_rcu() for more detailed information on 199 * memory ordering guarantees. 200 */ 201void call_rcu_bh(struct rcu_head *head, 202 rcu_callback_t func); 203 204/** 205 * call_rcu_sched() - Queue an RCU for invocation after sched grace period. 206 * @head: structure to be used for queueing the RCU updates. 207 * @func: actual callback function to be invoked after the grace period 208 * 209 * The callback function will be invoked some time after a full grace 210 * period elapses, in other words after all currently executing RCU 211 * read-side critical sections have completed. call_rcu_sched() assumes 212 * that the read-side critical sections end on enabling of preemption 213 * or on voluntary preemption. 214 * RCU read-side critical sections are delimited by : 215 * - rcu_read_lock_sched() and rcu_read_unlock_sched(), 216 * OR 217 * anything that disables preemption. 218 * These may be nested. 219 * 220 * See the description of call_rcu() for more detailed information on 221 * memory ordering guarantees. 222 */ 223void call_rcu_sched(struct rcu_head *head, 224 rcu_callback_t func); 225 226void synchronize_sched(void); 227 228/* 229 * Structure allowing asynchronous waiting on RCU. 230 */ 231struct rcu_synchronize { 232 struct rcu_head head; 233 struct completion completion; 234}; 235void wakeme_after_rcu(struct rcu_head *head); 236 237void __wait_rcu_gp(bool checktiny, int n, call_rcu_func_t *crcu_array, 238 struct rcu_synchronize *rs_array); 239 240#define _wait_rcu_gp(checktiny, ...) \ 241do { \ 242 call_rcu_func_t __crcu_array[] = { __VA_ARGS__ }; \ 243 struct rcu_synchronize __rs_array[ARRAY_SIZE(__crcu_array)]; \ 244 __wait_rcu_gp(checktiny, ARRAY_SIZE(__crcu_array), \ 245 __crcu_array, __rs_array); \ 246} while (0) 247 248#define wait_rcu_gp(...) _wait_rcu_gp(false, __VA_ARGS__) 249 250/** 251 * synchronize_rcu_mult - Wait concurrently for multiple grace periods 252 * @...: List of call_rcu() functions for the flavors to wait on. 253 * 254 * This macro waits concurrently for multiple flavors of RCU grace periods. 255 * For example, synchronize_rcu_mult(call_rcu, call_rcu_bh) would wait 256 * on concurrent RCU and RCU-bh grace periods. Waiting on a give SRCU 257 * domain requires you to write a wrapper function for that SRCU domain's 258 * call_srcu() function, supplying the corresponding srcu_struct. 259 * 260 * If Tiny RCU, tell _wait_rcu_gp() not to bother waiting for RCU 261 * or RCU-bh, given that anywhere synchronize_rcu_mult() can be called 262 * is automatically a grace period. 263 */ 264#define synchronize_rcu_mult(...) \ 265 _wait_rcu_gp(IS_ENABLED(CONFIG_TINY_RCU), __VA_ARGS__) 266 267/** 268 * call_rcu_tasks() - Queue an RCU for invocation task-based grace period 269 * @head: structure to be used for queueing the RCU updates. 270 * @func: actual callback function to be invoked after the grace period 271 * 272 * The callback function will be invoked some time after a full grace 273 * period elapses, in other words after all currently executing RCU 274 * read-side critical sections have completed. call_rcu_tasks() assumes 275 * that the read-side critical sections end at a voluntary context 276 * switch (not a preemption!), entry into idle, or transition to usermode 277 * execution. As such, there are no read-side primitives analogous to 278 * rcu_read_lock() and rcu_read_unlock() because this primitive is intended 279 * to determine that all tasks have passed through a safe state, not so 280 * much for data-strcuture synchronization. 281 * 282 * See the description of call_rcu() for more detailed information on 283 * memory ordering guarantees. 284 */ 285void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func); 286void synchronize_rcu_tasks(void); 287void rcu_barrier_tasks(void); 288 289#ifdef CONFIG_PREEMPT_RCU 290 291void __rcu_read_lock(void); 292void __rcu_read_unlock(void); 293void rcu_read_unlock_special(struct task_struct *t); 294void synchronize_rcu(void); 295 296/* 297 * Defined as a macro as it is a very low level header included from 298 * areas that don't even know about current. This gives the rcu_read_lock() 299 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other 300 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. 301 */ 302#define rcu_preempt_depth() (current->rcu_read_lock_nesting) 303 304#else /* #ifdef CONFIG_PREEMPT_RCU */ 305 306static inline void __rcu_read_lock(void) 307{ 308 if (IS_ENABLED(CONFIG_PREEMPT_COUNT)) 309 preempt_disable(); 310} 311 312static inline void __rcu_read_unlock(void) 313{ 314 if (IS_ENABLED(CONFIG_PREEMPT_COUNT)) 315 preempt_enable(); 316} 317 318static inline void synchronize_rcu(void) 319{ 320 synchronize_sched(); 321} 322 323static inline int rcu_preempt_depth(void) 324{ 325 return 0; 326} 327 328#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 329 330/* Internal to kernel */ 331void rcu_init(void); 332void rcu_sched_qs(void); 333void rcu_bh_qs(void); 334void rcu_check_callbacks(int user); 335void rcu_report_dead(unsigned int cpu); 336 337#ifndef CONFIG_TINY_RCU 338void rcu_end_inkernel_boot(void); 339#else /* #ifndef CONFIG_TINY_RCU */ 340static inline void rcu_end_inkernel_boot(void) { } 341#endif /* #ifndef CONFIG_TINY_RCU */ 342 343#ifdef CONFIG_RCU_STALL_COMMON 344void rcu_sysrq_start(void); 345void rcu_sysrq_end(void); 346#else /* #ifdef CONFIG_RCU_STALL_COMMON */ 347static inline void rcu_sysrq_start(void) 348{ 349} 350static inline void rcu_sysrq_end(void) 351{ 352} 353#endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ 354 355#ifdef CONFIG_NO_HZ_FULL 356void rcu_user_enter(void); 357void rcu_user_exit(void); 358#else 359static inline void rcu_user_enter(void) { } 360static inline void rcu_user_exit(void) { } 361#endif /* CONFIG_NO_HZ_FULL */ 362 363#ifdef CONFIG_RCU_NOCB_CPU 364void rcu_init_nohz(void); 365#else /* #ifdef CONFIG_RCU_NOCB_CPU */ 366static inline void rcu_init_nohz(void) 367{ 368} 369#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 370 371/** 372 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers 373 * @a: Code that RCU needs to pay attention to. 374 * 375 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden 376 * in the inner idle loop, that is, between the rcu_idle_enter() and 377 * the rcu_idle_exit() -- RCU will happily ignore any such read-side 378 * critical sections. However, things like powertop need tracepoints 379 * in the inner idle loop. 380 * 381 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) 382 * will tell RCU that it needs to pay attending, invoke its argument 383 * (in this example, a call to the do_something_with_RCU() function), 384 * and then tell RCU to go back to ignoring this CPU. It is permissible 385 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently 386 * quite limited. If deeper nesting is required, it will be necessary 387 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly. 388 */ 389#define RCU_NONIDLE(a) \ 390 do { \ 391 rcu_irq_enter_irqson(); \ 392 do { a; } while (0); \ 393 rcu_irq_exit_irqson(); \ 394 } while (0) 395 396/* 397 * Note a voluntary context switch for RCU-tasks benefit. This is a 398 * macro rather than an inline function to avoid #include hell. 399 */ 400#ifdef CONFIG_TASKS_RCU 401#define TASKS_RCU(x) x 402extern struct srcu_struct tasks_rcu_exit_srcu; 403#define rcu_note_voluntary_context_switch(t) \ 404 do { \ 405 rcu_all_qs(); \ 406 if (READ_ONCE((t)->rcu_tasks_holdout)) \ 407 WRITE_ONCE((t)->rcu_tasks_holdout, false); \ 408 } while (0) 409#else /* #ifdef CONFIG_TASKS_RCU */ 410#define TASKS_RCU(x) do { } while (0) 411#define rcu_note_voluntary_context_switch(t) rcu_all_qs() 412#endif /* #else #ifdef CONFIG_TASKS_RCU */ 413 414/** 415 * cond_resched_rcu_qs - Report potential quiescent states to RCU 416 * 417 * This macro resembles cond_resched(), except that it is defined to 418 * report potential quiescent states to RCU-tasks even if the cond_resched() 419 * machinery were to be shut off, as some advocate for PREEMPT kernels. 420 */ 421#define cond_resched_rcu_qs() \ 422do { \ 423 if (!cond_resched()) \ 424 rcu_note_voluntary_context_switch(current); \ 425} while (0) 426 427#if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) 428bool __rcu_is_watching(void); 429#endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */ 430 431/* 432 * Infrastructure to implement the synchronize_() primitives in 433 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 434 */ 435 436#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) 437#include <linux/rcutree.h> 438#elif defined(CONFIG_TINY_RCU) 439#include <linux/rcutiny.h> 440#else 441#error "Unknown RCU implementation specified to kernel configuration" 442#endif 443 444/* 445 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic 446 * initialization and destruction of rcu_head on the stack. rcu_head structures 447 * allocated dynamically in the heap or defined statically don't need any 448 * initialization. 449 */ 450#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 451void init_rcu_head(struct rcu_head *head); 452void destroy_rcu_head(struct rcu_head *head); 453void init_rcu_head_on_stack(struct rcu_head *head); 454void destroy_rcu_head_on_stack(struct rcu_head *head); 455#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 456static inline void init_rcu_head(struct rcu_head *head) 457{ 458} 459 460static inline void destroy_rcu_head(struct rcu_head *head) 461{ 462} 463 464static inline void init_rcu_head_on_stack(struct rcu_head *head) 465{ 466} 467 468static inline void destroy_rcu_head_on_stack(struct rcu_head *head) 469{ 470} 471#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 472 473#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 474bool rcu_lockdep_current_cpu_online(void); 475#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 476static inline bool rcu_lockdep_current_cpu_online(void) 477{ 478 return true; 479} 480#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 481 482#ifdef CONFIG_DEBUG_LOCK_ALLOC 483 484static inline void rcu_lock_acquire(struct lockdep_map *map) 485{ 486 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); 487} 488 489static inline void rcu_lock_release(struct lockdep_map *map) 490{ 491 lock_release(map, 1, _THIS_IP_); 492} 493 494extern struct lockdep_map rcu_lock_map; 495extern struct lockdep_map rcu_bh_lock_map; 496extern struct lockdep_map rcu_sched_lock_map; 497extern struct lockdep_map rcu_callback_map; 498int debug_lockdep_rcu_enabled(void); 499 500int rcu_read_lock_held(void); 501int rcu_read_lock_bh_held(void); 502 503/** 504 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section? 505 * 506 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an 507 * RCU-sched read-side critical section. In absence of 508 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side 509 * critical section unless it can prove otherwise. 510 */ 511#ifdef CONFIG_PREEMPT_COUNT 512int rcu_read_lock_sched_held(void); 513#else /* #ifdef CONFIG_PREEMPT_COUNT */ 514static inline int rcu_read_lock_sched_held(void) 515{ 516 return 1; 517} 518#endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ 519 520#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 521 522# define rcu_lock_acquire(a) do { } while (0) 523# define rcu_lock_release(a) do { } while (0) 524 525static inline int rcu_read_lock_held(void) 526{ 527 return 1; 528} 529 530static inline int rcu_read_lock_bh_held(void) 531{ 532 return 1; 533} 534 535#ifdef CONFIG_PREEMPT_COUNT 536static inline int rcu_read_lock_sched_held(void) 537{ 538 return preempt_count() != 0 || irqs_disabled(); 539} 540#else /* #ifdef CONFIG_PREEMPT_COUNT */ 541static inline int rcu_read_lock_sched_held(void) 542{ 543 return 1; 544} 545#endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ 546 547#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 548 549#ifdef CONFIG_PROVE_RCU 550 551/** 552 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met 553 * @c: condition to check 554 * @s: informative message 555 */ 556#define RCU_LOCKDEP_WARN(c, s) \ 557 do { \ 558 static bool __section(.data.unlikely) __warned; \ 559 if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \ 560 __warned = true; \ 561 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 562 } \ 563 } while (0) 564 565#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 566static inline void rcu_preempt_sleep_check(void) 567{ 568 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), 569 "Illegal context switch in RCU read-side critical section"); 570} 571#else /* #ifdef CONFIG_PROVE_RCU */ 572static inline void rcu_preempt_sleep_check(void) 573{ 574} 575#endif /* #else #ifdef CONFIG_PROVE_RCU */ 576 577#define rcu_sleep_check() \ 578 do { \ 579 rcu_preempt_sleep_check(); \ 580 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ 581 "Illegal context switch in RCU-bh read-side critical section"); \ 582 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ 583 "Illegal context switch in RCU-sched read-side critical section"); \ 584 } while (0) 585 586#else /* #ifdef CONFIG_PROVE_RCU */ 587 588#define RCU_LOCKDEP_WARN(c, s) do { } while (0) 589#define rcu_sleep_check() do { } while (0) 590 591#endif /* #else #ifdef CONFIG_PROVE_RCU */ 592 593/* 594 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 595 * and rcu_assign_pointer(). Some of these could be folded into their 596 * callers, but they are left separate in order to ease introduction of 597 * multiple flavors of pointers to match the multiple flavors of RCU 598 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in 599 * the future. 600 */ 601 602#ifdef __CHECKER__ 603#define rcu_dereference_sparse(p, space) \ 604 ((void)(((typeof(*p) space *)p) == p)) 605#else /* #ifdef __CHECKER__ */ 606#define rcu_dereference_sparse(p, space) 607#endif /* #else #ifdef __CHECKER__ */ 608 609#define __rcu_access_pointer(p, space) \ 610({ \ 611 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 612 rcu_dereference_sparse(p, space); \ 613 ((typeof(*p) __force __kernel *)(_________p1)); \ 614}) 615#define __rcu_dereference_check(p, c, space) \ 616({ \ 617 /* Dependency order vs. p above. */ \ 618 typeof(*p) *________p1 = (typeof(*p) *__force)lockless_dereference(p); \ 619 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ 620 rcu_dereference_sparse(p, space); \ 621 ((typeof(*p) __force __kernel *)(________p1)); \ 622}) 623#define __rcu_dereference_protected(p, c, space) \ 624({ \ 625 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ 626 rcu_dereference_sparse(p, space); \ 627 ((typeof(*p) __force __kernel *)(p)); \ 628}) 629 630/** 631 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 632 * @v: The value to statically initialize with. 633 */ 634#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 635 636/** 637 * rcu_assign_pointer() - assign to RCU-protected pointer 638 * @p: pointer to assign to 639 * @v: value to assign (publish) 640 * 641 * Assigns the specified value to the specified RCU-protected 642 * pointer, ensuring that any concurrent RCU readers will see 643 * any prior initialization. 644 * 645 * Inserts memory barriers on architectures that require them 646 * (which is most of them), and also prevents the compiler from 647 * reordering the code that initializes the structure after the pointer 648 * assignment. More importantly, this call documents which pointers 649 * will be dereferenced by RCU read-side code. 650 * 651 * In some special cases, you may use RCU_INIT_POINTER() instead 652 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 653 * to the fact that it does not constrain either the CPU or the compiler. 654 * That said, using RCU_INIT_POINTER() when you should have used 655 * rcu_assign_pointer() is a very bad thing that results in 656 * impossible-to-diagnose memory corruption. So please be careful. 657 * See the RCU_INIT_POINTER() comment header for details. 658 * 659 * Note that rcu_assign_pointer() evaluates each of its arguments only 660 * once, appearances notwithstanding. One of the "extra" evaluations 661 * is in typeof() and the other visible only to sparse (__CHECKER__), 662 * neither of which actually execute the argument. As with most cpp 663 * macros, this execute-arguments-only-once property is important, so 664 * please be careful when making changes to rcu_assign_pointer() and the 665 * other macros that it invokes. 666 */ 667#define rcu_assign_pointer(p, v) smp_store_release(&p, RCU_INITIALIZER(v)) 668 669/** 670 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 671 * @p: The pointer to read 672 * 673 * Return the value of the specified RCU-protected pointer, but omit the 674 * smp_read_barrier_depends() and keep the READ_ONCE(). This is useful 675 * when the value of this pointer is accessed, but the pointer is not 676 * dereferenced, for example, when testing an RCU-protected pointer against 677 * NULL. Although rcu_access_pointer() may also be used in cases where 678 * update-side locks prevent the value of the pointer from changing, you 679 * should instead use rcu_dereference_protected() for this use case. 680 * 681 * It is also permissible to use rcu_access_pointer() when read-side 682 * access to the pointer was removed at least one grace period ago, as 683 * is the case in the context of the RCU callback that is freeing up 684 * the data, or after a synchronize_rcu() returns. This can be useful 685 * when tearing down multi-linked structures after a grace period 686 * has elapsed. 687 */ 688#define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) 689 690/** 691 * rcu_dereference_check() - rcu_dereference with debug checking 692 * @p: The pointer to read, prior to dereferencing 693 * @c: The conditions under which the dereference will take place 694 * 695 * Do an rcu_dereference(), but check that the conditions under which the 696 * dereference will take place are correct. Typically the conditions 697 * indicate the various locking conditions that should be held at that 698 * point. The check should return true if the conditions are satisfied. 699 * An implicit check for being in an RCU read-side critical section 700 * (rcu_read_lock()) is included. 701 * 702 * For example: 703 * 704 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 705 * 706 * could be used to indicate to lockdep that foo->bar may only be dereferenced 707 * if either rcu_read_lock() is held, or that the lock required to replace 708 * the bar struct at foo->bar is held. 709 * 710 * Note that the list of conditions may also include indications of when a lock 711 * need not be held, for example during initialisation or destruction of the 712 * target struct: 713 * 714 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 715 * atomic_read(&foo->usage) == 0); 716 * 717 * Inserts memory barriers on architectures that require them 718 * (currently only the Alpha), prevents the compiler from refetching 719 * (and from merging fetches), and, more importantly, documents exactly 720 * which pointers are protected by RCU and checks that the pointer is 721 * annotated as __rcu. 722 */ 723#define rcu_dereference_check(p, c) \ 724 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu) 725 726/** 727 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 728 * @p: The pointer to read, prior to dereferencing 729 * @c: The conditions under which the dereference will take place 730 * 731 * This is the RCU-bh counterpart to rcu_dereference_check(). 732 */ 733#define rcu_dereference_bh_check(p, c) \ 734 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu) 735 736/** 737 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 738 * @p: The pointer to read, prior to dereferencing 739 * @c: The conditions under which the dereference will take place 740 * 741 * This is the RCU-sched counterpart to rcu_dereference_check(). 742 */ 743#define rcu_dereference_sched_check(p, c) \ 744 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \ 745 __rcu) 746 747#define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/ 748 749/* 750 * The tracing infrastructure traces RCU (we want that), but unfortunately 751 * some of the RCU checks causes tracing to lock up the system. 752 * 753 * The no-tracing version of rcu_dereference_raw() must not call 754 * rcu_read_lock_held(). 755 */ 756#define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu) 757 758/** 759 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 760 * @p: The pointer to read, prior to dereferencing 761 * @c: The conditions under which the dereference will take place 762 * 763 * Return the value of the specified RCU-protected pointer, but omit 764 * both the smp_read_barrier_depends() and the READ_ONCE(). This 765 * is useful in cases where update-side locks prevent the value of the 766 * pointer from changing. Please note that this primitive does -not- 767 * prevent the compiler from repeating this reference or combining it 768 * with other references, so it should not be used without protection 769 * of appropriate locks. 770 * 771 * This function is only for update-side use. Using this function 772 * when protected only by rcu_read_lock() will result in infrequent 773 * but very ugly failures. 774 */ 775#define rcu_dereference_protected(p, c) \ 776 __rcu_dereference_protected((p), (c), __rcu) 777 778 779/** 780 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 781 * @p: The pointer to read, prior to dereferencing 782 * 783 * This is a simple wrapper around rcu_dereference_check(). 784 */ 785#define rcu_dereference(p) rcu_dereference_check(p, 0) 786 787/** 788 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 789 * @p: The pointer to read, prior to dereferencing 790 * 791 * Makes rcu_dereference_check() do the dirty work. 792 */ 793#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 794 795/** 796 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 797 * @p: The pointer to read, prior to dereferencing 798 * 799 * Makes rcu_dereference_check() do the dirty work. 800 */ 801#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 802 803/** 804 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism 805 * @p: The pointer to hand off 806 * 807 * This is simply an identity function, but it documents where a pointer 808 * is handed off from RCU to some other synchronization mechanism, for 809 * example, reference counting or locking. In C11, it would map to 810 * kill_dependency(). It could be used as follows: 811 * 812 * rcu_read_lock(); 813 * p = rcu_dereference(gp); 814 * long_lived = is_long_lived(p); 815 * if (long_lived) { 816 * if (!atomic_inc_not_zero(p->refcnt)) 817 * long_lived = false; 818 * else 819 * p = rcu_pointer_handoff(p); 820 * } 821 * rcu_read_unlock(); 822 */ 823#define rcu_pointer_handoff(p) (p) 824 825/** 826 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 827 * 828 * When synchronize_rcu() is invoked on one CPU while other CPUs 829 * are within RCU read-side critical sections, then the 830 * synchronize_rcu() is guaranteed to block until after all the other 831 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 832 * on one CPU while other CPUs are within RCU read-side critical 833 * sections, invocation of the corresponding RCU callback is deferred 834 * until after the all the other CPUs exit their critical sections. 835 * 836 * Note, however, that RCU callbacks are permitted to run concurrently 837 * with new RCU read-side critical sections. One way that this can happen 838 * is via the following sequence of events: (1) CPU 0 enters an RCU 839 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 840 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 841 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 842 * callback is invoked. This is legal, because the RCU read-side critical 843 * section that was running concurrently with the call_rcu() (and which 844 * therefore might be referencing something that the corresponding RCU 845 * callback would free up) has completed before the corresponding 846 * RCU callback is invoked. 847 * 848 * RCU read-side critical sections may be nested. Any deferred actions 849 * will be deferred until the outermost RCU read-side critical section 850 * completes. 851 * 852 * You can avoid reading and understanding the next paragraph by 853 * following this rule: don't put anything in an rcu_read_lock() RCU 854 * read-side critical section that would block in a !PREEMPT kernel. 855 * But if you want the full story, read on! 856 * 857 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), 858 * it is illegal to block while in an RCU read-side critical section. 859 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT 860 * kernel builds, RCU read-side critical sections may be preempted, 861 * but explicit blocking is illegal. Finally, in preemptible RCU 862 * implementations in real-time (with -rt patchset) kernel builds, RCU 863 * read-side critical sections may be preempted and they may also block, but 864 * only when acquiring spinlocks that are subject to priority inheritance. 865 */ 866static inline void rcu_read_lock(void) 867{ 868 __rcu_read_lock(); 869 __acquire(RCU); 870 rcu_lock_acquire(&rcu_lock_map); 871 RCU_LOCKDEP_WARN(!rcu_is_watching(), 872 "rcu_read_lock() used illegally while idle"); 873} 874 875/* 876 * So where is rcu_write_lock()? It does not exist, as there is no 877 * way for writers to lock out RCU readers. This is a feature, not 878 * a bug -- this property is what provides RCU's performance benefits. 879 * Of course, writers must coordinate with each other. The normal 880 * spinlock primitives work well for this, but any other technique may be 881 * used as well. RCU does not care how the writers keep out of each 882 * others' way, as long as they do so. 883 */ 884 885/** 886 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 887 * 888 * In most situations, rcu_read_unlock() is immune from deadlock. 889 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() 890 * is responsible for deboosting, which it does via rt_mutex_unlock(). 891 * Unfortunately, this function acquires the scheduler's runqueue and 892 * priority-inheritance spinlocks. This means that deadlock could result 893 * if the caller of rcu_read_unlock() already holds one of these locks or 894 * any lock that is ever acquired while holding them; or any lock which 895 * can be taken from interrupt context because rcu_boost()->rt_mutex_lock() 896 * does not disable irqs while taking ->wait_lock. 897 * 898 * That said, RCU readers are never priority boosted unless they were 899 * preempted. Therefore, one way to avoid deadlock is to make sure 900 * that preemption never happens within any RCU read-side critical 901 * section whose outermost rcu_read_unlock() is called with one of 902 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in 903 * a number of ways, for example, by invoking preempt_disable() before 904 * critical section's outermost rcu_read_lock(). 905 * 906 * Given that the set of locks acquired by rt_mutex_unlock() might change 907 * at any time, a somewhat more future-proofed approach is to make sure 908 * that that preemption never happens within any RCU read-side critical 909 * section whose outermost rcu_read_unlock() is called with irqs disabled. 910 * This approach relies on the fact that rt_mutex_unlock() currently only 911 * acquires irq-disabled locks. 912 * 913 * The second of these two approaches is best in most situations, 914 * however, the first approach can also be useful, at least to those 915 * developers willing to keep abreast of the set of locks acquired by 916 * rt_mutex_unlock(). 917 * 918 * See rcu_read_lock() for more information. 919 */ 920static inline void rcu_read_unlock(void) 921{ 922 RCU_LOCKDEP_WARN(!rcu_is_watching(), 923 "rcu_read_unlock() used illegally while idle"); 924 __release(RCU); 925 __rcu_read_unlock(); 926 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ 927} 928 929/** 930 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 931 * 932 * This is equivalent of rcu_read_lock(), but to be used when updates 933 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since 934 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a 935 * softirq handler to be a quiescent state, a process in RCU read-side 936 * critical section must be protected by disabling softirqs. Read-side 937 * critical sections in interrupt context can use just rcu_read_lock(), 938 * though this should at least be commented to avoid confusing people 939 * reading the code. 940 * 941 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 942 * must occur in the same context, for example, it is illegal to invoke 943 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 944 * was invoked from some other task. 945 */ 946static inline void rcu_read_lock_bh(void) 947{ 948 local_bh_disable(); 949 __acquire(RCU_BH); 950 rcu_lock_acquire(&rcu_bh_lock_map); 951 RCU_LOCKDEP_WARN(!rcu_is_watching(), 952 "rcu_read_lock_bh() used illegally while idle"); 953} 954 955/* 956 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section 957 * 958 * See rcu_read_lock_bh() for more information. 959 */ 960static inline void rcu_read_unlock_bh(void) 961{ 962 RCU_LOCKDEP_WARN(!rcu_is_watching(), 963 "rcu_read_unlock_bh() used illegally while idle"); 964 rcu_lock_release(&rcu_bh_lock_map); 965 __release(RCU_BH); 966 local_bh_enable(); 967} 968 969/** 970 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 971 * 972 * This is equivalent of rcu_read_lock(), but to be used when updates 973 * are being done using call_rcu_sched() or synchronize_rcu_sched(). 974 * Read-side critical sections can also be introduced by anything that 975 * disables preemption, including local_irq_disable() and friends. 976 * 977 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 978 * must occur in the same context, for example, it is illegal to invoke 979 * rcu_read_unlock_sched() from process context if the matching 980 * rcu_read_lock_sched() was invoked from an NMI handler. 981 */ 982static inline void rcu_read_lock_sched(void) 983{ 984 preempt_disable(); 985 __acquire(RCU_SCHED); 986 rcu_lock_acquire(&rcu_sched_lock_map); 987 RCU_LOCKDEP_WARN(!rcu_is_watching(), 988 "rcu_read_lock_sched() used illegally while idle"); 989} 990 991/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 992static inline notrace void rcu_read_lock_sched_notrace(void) 993{ 994 preempt_disable_notrace(); 995 __acquire(RCU_SCHED); 996} 997 998/* 999 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section 1000 *
1001 * See rcu_read_lock_sched for more information. 1002 */ 1003static inline void rcu_read_unlock_sched(void) 1004{ 1005 RCU_LOCKDEP_WARN(!rcu_is_watching(), 1006 "rcu_read_unlock_sched() used illegally while idle"); 1007 rcu_lock_release(&rcu_sched_lock_map); 1008 __release(RCU_SCHED); 1009 preempt_enable(); 1010} 1011 1012/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 1013static inline notrace void rcu_read_unlock_sched_notrace(void) 1014{ 1015 __release(RCU_SCHED); 1016 preempt_enable_notrace(); 1017} 1018 1019/** 1020 * RCU_INIT_POINTER() - initialize an RCU protected pointer 1021 * 1022 * Initialize an RCU-protected pointer in special cases where readers 1023 * do not need ordering constraints on the CPU or the compiler. These 1024 * special cases are: 1025 * 1026 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or- 1027 * 2. The caller has taken whatever steps are required to prevent 1028 * RCU readers from concurrently accessing this pointer -or- 1029 * 3. The referenced data structure has already been exposed to 1030 * readers either at compile time or via rcu_assign_pointer() -and- 1031 * a. You have not made -any- reader-visible changes to 1032 * this structure since then -or- 1033 * b. It is OK for readers accessing this structure from its 1034 * new location to see the old state of the structure. (For 1035 * example, the changes were to statistical counters or to 1036 * other state where exact synchronization is not required.) 1037 * 1038 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 1039 * result in impossible-to-diagnose memory corruption. As in the structures 1040 * will look OK in crash dumps, but any concurrent RCU readers might 1041 * see pre-initialized values of the referenced data structure. So 1042 * please be very careful how you use RCU_INIT_POINTER()!!! 1043 * 1044 * If you are creating an RCU-protected linked structure that is accessed 1045 * by a single external-to-structure RCU-protected pointer, then you may 1046 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 1047 * pointers, but you must use rcu_assign_pointer() to initialize the 1048 * external-to-structure pointer -after- you have completely initialized 1049 * the reader-accessible portions of the linked structure. 1050 * 1051 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 1052 * ordering guarantees for either the CPU or the compiler. 1053 */ 1054#define RCU_INIT_POINTER(p, v) \ 1055 do { \ 1056 rcu_dereference_sparse(p, __rcu); \ 1057 WRITE_ONCE(p, RCU_INITIALIZER(v)); \ 1058 } while (0) 1059 1060/** 1061 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 1062 * 1063 * GCC-style initialization for an RCU-protected pointer in a structure field. 1064 */ 1065#define RCU_POINTER_INITIALIZER(p, v) \ 1066 .p = RCU_INITIALIZER(v) 1067 1068/* 1069 * Does the specified offset indicate that the corresponding rcu_head 1070 * structure can be handled by kfree_rcu()? 1071 */ 1072#define __is_kfree_rcu_offset(offset) ((offset) < 4096) 1073 1074/* 1075 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. 1076 */ 1077#define __kfree_rcu(head, offset) \ 1078 do { \ 1079 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \ 1080 kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \ 1081 } while (0) 1082 1083/** 1084 * kfree_rcu() - kfree an object after a grace period. 1085 * @ptr: pointer to kfree 1086 * @rcu_head: the name of the struct rcu_head within the type of @ptr. 1087 * 1088 * Many rcu callbacks functions just call kfree() on the base structure. 1089 * These functions are trivial, but their size adds up, and furthermore 1090 * when they are used in a kernel module, that module must invoke the 1091 * high-latency rcu_barrier() function at module-unload time. 1092 * 1093 * The kfree_rcu() function handles this issue. Rather than encoding a 1094 * function address in the embedded rcu_head structure, kfree_rcu() instead 1095 * encodes the offset of the rcu_head structure within the base structure. 1096 * Because the functions are not allowed in the low-order 4096 bytes of 1097 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 1098 * If the offset is larger than 4095 bytes, a compile-time error will 1099 * be generated in __kfree_rcu(). If this error is triggered, you can 1100 * either fall back to use of call_rcu() or rearrange the structure to 1101 * position the rcu_head structure into the first 4096 bytes. 1102 * 1103 * Note that the allowable offset might decrease in the future, for example, 1104 * to allow something like kmem_cache_free_rcu(). 1105 * 1106 * The BUILD_BUG_ON check must not involve any function calls, hence the 1107 * checks are done in macros here. 1108 */ 1109#define kfree_rcu(ptr, rcu_head) \ 1110 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head)) 1111 1112#ifdef CONFIG_TINY_RCU 1113static inline int rcu_needs_cpu(u64 basemono, u64 *nextevt) 1114{ 1115 *nextevt = KTIME_MAX; 1116 return 0; 1117} 1118#endif /* #ifdef CONFIG_TINY_RCU */ 1119 1120#if defined(CONFIG_RCU_NOCB_CPU_ALL) 1121static inline bool rcu_is_nocb_cpu(int cpu) { return true; } 1122#elif defined(CONFIG_RCU_NOCB_CPU) 1123bool rcu_is_nocb_cpu(int cpu); 1124#else 1125static inline bool rcu_is_nocb_cpu(int cpu) { return false; } 1126#endif 1127 1128 1129/* Only for use by adaptive-ticks code. */ 1130#ifdef CONFIG_NO_HZ_FULL_SYSIDLE 1131bool rcu_sys_is_idle(void); 1132void rcu_sysidle_force_exit(void); 1133#else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ 1134 1135static inline bool rcu_sys_is_idle(void) 1136{ 1137 return false; 1138} 1139 1140static inline void rcu_sysidle_force_exit(void) 1141{ 1142} 1143 1144#endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ 1145 1146 1147#endif /* __LINUX_RCUPDATE_H */ 1148