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 <asm/barrier.h> 48 49extern int rcu_expedited; /* for sysctl */ 50 51enum rcutorture_type { 52 RCU_FLAVOR, 53 RCU_BH_FLAVOR, 54 RCU_SCHED_FLAVOR, 55 RCU_TASKS_FLAVOR, 56 SRCU_FLAVOR, 57 INVALID_RCU_FLAVOR 58}; 59 60#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) 61void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags, 62 unsigned long *gpnum, unsigned long *completed); 63void rcutorture_record_test_transition(void); 64void rcutorture_record_progress(unsigned long vernum); 65void do_trace_rcu_torture_read(const char *rcutorturename, 66 struct rcu_head *rhp, 67 unsigned long secs, 68 unsigned long c_old, 69 unsigned long c); 70#else 71static inline void rcutorture_get_gp_data(enum rcutorture_type test_type, 72 int *flags, 73 unsigned long *gpnum, 74 unsigned long *completed) 75{ 76 *flags = 0; 77 *gpnum = 0; 78 *completed = 0; 79} 80static inline void rcutorture_record_test_transition(void) 81{ 82} 83static inline void rcutorture_record_progress(unsigned long vernum) 84{ 85} 86#ifdef CONFIG_RCU_TRACE 87void do_trace_rcu_torture_read(const char *rcutorturename, 88 struct rcu_head *rhp, 89 unsigned long secs, 90 unsigned long c_old, 91 unsigned long c); 92#else 93#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ 94 do { } while (0) 95#endif 96#endif 97 98#define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b)) 99#define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b)) 100#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) 101#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) 102#define ulong2long(a) (*(long *)(&(a))) 103 104/* Exported common interfaces */ 105 106#ifdef CONFIG_PREEMPT_RCU 107 108/** 109 * call_rcu() - Queue an RCU callback for invocation after a grace period. 110 * @head: structure to be used for queueing the RCU updates. 111 * @func: actual callback function to be invoked after the grace period 112 * 113 * The callback function will be invoked some time after a full grace 114 * period elapses, in other words after all pre-existing RCU read-side 115 * critical sections have completed. However, the callback function 116 * might well execute concurrently with RCU read-side critical sections 117 * that started after call_rcu() was invoked. RCU read-side critical 118 * sections are delimited by rcu_read_lock() and rcu_read_unlock(), 119 * and may be nested. 120 * 121 * Note that all CPUs must agree that the grace period extended beyond 122 * all pre-existing RCU read-side critical section. On systems with more 123 * than one CPU, this means that when "func()" is invoked, each CPU is 124 * guaranteed to have executed a full memory barrier since the end of its 125 * last RCU read-side critical section whose beginning preceded the call 126 * to call_rcu(). It also means that each CPU executing an RCU read-side 127 * critical section that continues beyond the start of "func()" must have 128 * executed a memory barrier after the call_rcu() but before the beginning 129 * of that RCU read-side critical section. Note that these guarantees 130 * include CPUs that are offline, idle, or executing in user mode, as 131 * well as CPUs that are executing in the kernel. 132 * 133 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the 134 * resulting RCU callback function "func()", then both CPU A and CPU B are 135 * guaranteed to execute a full memory barrier during the time interval 136 * between the call to call_rcu() and the invocation of "func()" -- even 137 * if CPU A and CPU B are the same CPU (but again only if the system has 138 * more than one CPU). 139 */ 140void call_rcu(struct rcu_head *head, 141 void (*func)(struct rcu_head *head)); 142 143#else /* #ifdef CONFIG_PREEMPT_RCU */ 144 145/* In classic RCU, call_rcu() is just call_rcu_sched(). */ 146#define call_rcu call_rcu_sched 147 148#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 149 150/** 151 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period. 152 * @head: structure to be used for queueing the RCU updates. 153 * @func: actual callback function to be invoked after the grace period 154 * 155 * The callback function will be invoked some time after a full grace 156 * period elapses, in other words after all currently executing RCU 157 * read-side critical sections have completed. call_rcu_bh() assumes 158 * that the read-side critical sections end on completion of a softirq 159 * handler. This means that read-side critical sections in process 160 * context must not be interrupted by softirqs. This interface is to be 161 * used when most of the read-side critical sections are in softirq context. 162 * RCU read-side critical sections are delimited by : 163 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context. 164 * OR 165 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context. 166 * These may be nested. 167 * 168 * See the description of call_rcu() for more detailed information on 169 * memory ordering guarantees. 170 */ 171void call_rcu_bh(struct rcu_head *head, 172 void (*func)(struct rcu_head *head)); 173 174/** 175 * call_rcu_sched() - Queue an RCU for invocation after sched grace period. 176 * @head: structure to be used for queueing the RCU updates. 177 * @func: actual callback function to be invoked after the grace period 178 * 179 * The callback function will be invoked some time after a full grace 180 * period elapses, in other words after all currently executing RCU 181 * read-side critical sections have completed. call_rcu_sched() assumes 182 * that the read-side critical sections end on enabling of preemption 183 * or on voluntary preemption. 184 * RCU read-side critical sections are delimited by : 185 * - rcu_read_lock_sched() and rcu_read_unlock_sched(), 186 * OR 187 * anything that disables preemption. 188 * These may be nested. 189 * 190 * See the description of call_rcu() for more detailed information on 191 * memory ordering guarantees. 192 */ 193void call_rcu_sched(struct rcu_head *head, 194 void (*func)(struct rcu_head *rcu)); 195 196void synchronize_sched(void); 197 198/** 199 * call_rcu_tasks() - Queue an RCU for invocation task-based grace period 200 * @head: structure to be used for queueing the RCU updates. 201 * @func: actual callback function to be invoked after the grace period 202 * 203 * The callback function will be invoked some time after a full grace 204 * period elapses, in other words after all currently executing RCU 205 * read-side critical sections have completed. call_rcu_tasks() assumes 206 * that the read-side critical sections end at a voluntary context 207 * switch (not a preemption!), entry into idle, or transition to usermode 208 * execution. As such, there are no read-side primitives analogous to 209 * rcu_read_lock() and rcu_read_unlock() because this primitive is intended 210 * to determine that all tasks have passed through a safe state, not so 211 * much for data-strcuture synchronization. 212 * 213 * See the description of call_rcu() for more detailed information on 214 * memory ordering guarantees. 215 */ 216void call_rcu_tasks(struct rcu_head *head, void (*func)(struct rcu_head *head)); 217void synchronize_rcu_tasks(void); 218void rcu_barrier_tasks(void); 219 220#ifdef CONFIG_PREEMPT_RCU 221 222void __rcu_read_lock(void); 223void __rcu_read_unlock(void); 224void rcu_read_unlock_special(struct task_struct *t); 225void synchronize_rcu(void); 226 227/* 228 * Defined as a macro as it is a very low level header included from 229 * areas that don't even know about current. This gives the rcu_read_lock() 230 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other 231 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. 232 */ 233#define rcu_preempt_depth() (current->rcu_read_lock_nesting) 234 235#else /* #ifdef CONFIG_PREEMPT_RCU */ 236 237static inline void __rcu_read_lock(void) 238{ 239 preempt_disable(); 240} 241 242static inline void __rcu_read_unlock(void) 243{ 244 preempt_enable(); 245} 246 247static inline void synchronize_rcu(void) 248{ 249 synchronize_sched(); 250} 251 252static inline int rcu_preempt_depth(void) 253{ 254 return 0; 255} 256 257#endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 258 259/* Internal to kernel */ 260void rcu_init(void); 261void rcu_sched_qs(void); 262void rcu_bh_qs(void); 263void rcu_check_callbacks(int user); 264struct notifier_block; 265void rcu_idle_enter(void); 266void rcu_idle_exit(void); 267void rcu_irq_enter(void); 268void rcu_irq_exit(void); 269 270#ifdef CONFIG_RCU_STALL_COMMON 271void rcu_sysrq_start(void); 272void rcu_sysrq_end(void); 273#else /* #ifdef CONFIG_RCU_STALL_COMMON */ 274static inline void rcu_sysrq_start(void) 275{ 276} 277static inline void rcu_sysrq_end(void) 278{ 279} 280#endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ 281 282#ifdef CONFIG_RCU_USER_QS 283void rcu_user_enter(void); 284void rcu_user_exit(void); 285#else 286static inline void rcu_user_enter(void) { } 287static inline void rcu_user_exit(void) { } 288static inline void rcu_user_hooks_switch(struct task_struct *prev, 289 struct task_struct *next) { } 290#endif /* CONFIG_RCU_USER_QS */ 291 292#ifdef CONFIG_RCU_NOCB_CPU 293void rcu_init_nohz(void); 294#else /* #ifdef CONFIG_RCU_NOCB_CPU */ 295static inline void rcu_init_nohz(void) 296{ 297} 298#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 299 300/** 301 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers 302 * @a: Code that RCU needs to pay attention to. 303 * 304 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden 305 * in the inner idle loop, that is, between the rcu_idle_enter() and 306 * the rcu_idle_exit() -- RCU will happily ignore any such read-side 307 * critical sections. However, things like powertop need tracepoints 308 * in the inner idle loop. 309 * 310 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) 311 * will tell RCU that it needs to pay attending, invoke its argument 312 * (in this example, a call to the do_something_with_RCU() function), 313 * and then tell RCU to go back to ignoring this CPU. It is permissible 314 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently 315 * quite limited. If deeper nesting is required, it will be necessary 316 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly. 317 */ 318#define RCU_NONIDLE(a) \ 319 do { \ 320 rcu_irq_enter(); \ 321 do { a; } while (0); \ 322 rcu_irq_exit(); \ 323 } while (0) 324 325/* 326 * Note a voluntary context switch for RCU-tasks benefit. This is a 327 * macro rather than an inline function to avoid #include hell. 328 */ 329#ifdef CONFIG_TASKS_RCU 330#define TASKS_RCU(x) x 331extern struct srcu_struct tasks_rcu_exit_srcu; 332#define rcu_note_voluntary_context_switch(t) \ 333 do { \ 334 rcu_all_qs(); \ 335 if (ACCESS_ONCE((t)->rcu_tasks_holdout)) \ 336 ACCESS_ONCE((t)->rcu_tasks_holdout) = false; \ 337 } while (0) 338#else /* #ifdef CONFIG_TASKS_RCU */ 339#define TASKS_RCU(x) do { } while (0) 340#define rcu_note_voluntary_context_switch(t) rcu_all_qs() 341#endif /* #else #ifdef CONFIG_TASKS_RCU */ 342 343/** 344 * cond_resched_rcu_qs - Report potential quiescent states to RCU 345 * 346 * This macro resembles cond_resched(), except that it is defined to 347 * report potential quiescent states to RCU-tasks even if the cond_resched() 348 * machinery were to be shut off, as some advocate for PREEMPT kernels. 349 */ 350#define cond_resched_rcu_qs() \ 351do { \ 352 if (!cond_resched()) \ 353 rcu_note_voluntary_context_switch(current); \ 354} while (0) 355 356#if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) 357bool __rcu_is_watching(void); 358#endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */ 359 360/* 361 * Infrastructure to implement the synchronize_() primitives in 362 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 363 */ 364 365typedef void call_rcu_func_t(struct rcu_head *head, 366 void (*func)(struct rcu_head *head)); 367void wait_rcu_gp(call_rcu_func_t crf); 368 369#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) 370#include <linux/rcutree.h> 371#elif defined(CONFIG_TINY_RCU) 372#include <linux/rcutiny.h> 373#else 374#error "Unknown RCU implementation specified to kernel configuration" 375#endif 376 377/* 378 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic 379 * initialization and destruction of rcu_head on the stack. rcu_head structures 380 * allocated dynamically in the heap or defined statically don't need any 381 * initialization. 382 */ 383#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 384void init_rcu_head(struct rcu_head *head); 385void destroy_rcu_head(struct rcu_head *head); 386void init_rcu_head_on_stack(struct rcu_head *head); 387void destroy_rcu_head_on_stack(struct rcu_head *head); 388#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 389static inline void init_rcu_head(struct rcu_head *head) 390{ 391} 392 393static inline void destroy_rcu_head(struct rcu_head *head) 394{ 395} 396 397static inline void init_rcu_head_on_stack(struct rcu_head *head) 398{ 399} 400 401static inline void destroy_rcu_head_on_stack(struct rcu_head *head) 402{ 403} 404#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 405 406#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 407bool rcu_lockdep_current_cpu_online(void); 408#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 409static inline bool rcu_lockdep_current_cpu_online(void) 410{ 411 return true; 412} 413#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 414 415#ifdef CONFIG_DEBUG_LOCK_ALLOC 416 417static inline void rcu_lock_acquire(struct lockdep_map *map) 418{ 419 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); 420} 421 422static inline void rcu_lock_release(struct lockdep_map *map) 423{ 424 lock_release(map, 1, _THIS_IP_); 425} 426 427extern struct lockdep_map rcu_lock_map; 428extern struct lockdep_map rcu_bh_lock_map; 429extern struct lockdep_map rcu_sched_lock_map; 430extern struct lockdep_map rcu_callback_map; 431int debug_lockdep_rcu_enabled(void); 432 433int rcu_read_lock_held(void); 434int rcu_read_lock_bh_held(void); 435 436/** 437 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section? 438 * 439 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an 440 * RCU-sched read-side critical section. In absence of 441 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side 442 * critical section unless it can prove otherwise. Note that disabling 443 * of preemption (including disabling irqs) counts as an RCU-sched 444 * read-side critical section. This is useful for debug checks in functions 445 * that required that they be called within an RCU-sched read-side 446 * critical section. 447 * 448 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot 449 * and while lockdep is disabled. 450 * 451 * Note that if the CPU is in the idle loop from an RCU point of 452 * view (ie: that we are in the section between rcu_idle_enter() and 453 * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU 454 * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs 455 * that are in such a section, considering these as in extended quiescent 456 * state, so such a CPU is effectively never in an RCU read-side critical 457 * section regardless of what RCU primitives it invokes. This state of 458 * affairs is required --- we need to keep an RCU-free window in idle 459 * where the CPU may possibly enter into low power mode. This way we can 460 * notice an extended quiescent state to other CPUs that started a grace 461 * period. Otherwise we would delay any grace period as long as we run in 462 * the idle task. 463 * 464 * Similarly, we avoid claiming an SRCU read lock held if the current 465 * CPU is offline. 466 */ 467#ifdef CONFIG_PREEMPT_COUNT 468static inline int rcu_read_lock_sched_held(void) 469{ 470 int lockdep_opinion = 0; 471 472 if (!debug_lockdep_rcu_enabled()) 473 return 1; 474 if (!rcu_is_watching()) 475 return 0; 476 if (!rcu_lockdep_current_cpu_online()) 477 return 0; 478 if (debug_locks) 479 lockdep_opinion = lock_is_held(&rcu_sched_lock_map); 480 return lockdep_opinion || preempt_count() != 0 || irqs_disabled(); 481} 482#else /* #ifdef CONFIG_PREEMPT_COUNT */ 483static inline int rcu_read_lock_sched_held(void) 484{ 485 return 1; 486} 487#endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ 488 489#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 490 491# define rcu_lock_acquire(a) do { } while (0) 492# define rcu_lock_release(a) do { } while (0) 493 494static inline int rcu_read_lock_held(void) 495{ 496 return 1; 497} 498 499static inline int rcu_read_lock_bh_held(void) 500{ 501 return 1; 502} 503 504#ifdef CONFIG_PREEMPT_COUNT 505static inline int rcu_read_lock_sched_held(void) 506{ 507 return preempt_count() != 0 || irqs_disabled(); 508} 509#else /* #ifdef CONFIG_PREEMPT_COUNT */ 510static inline int rcu_read_lock_sched_held(void) 511{ 512 return 1; 513} 514#endif /* #else #ifdef CONFIG_PREEMPT_COUNT */ 515 516#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 517 518#ifdef CONFIG_PROVE_RCU 519 520/** 521 * rcu_lockdep_assert - emit lockdep splat if specified condition not met 522 * @c: condition to check 523 * @s: informative message 524 */ 525#define rcu_lockdep_assert(c, s) \ 526 do { \ 527 static bool __section(.data.unlikely) __warned; \ 528 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \ 529 __warned = true; \ 530 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 531 } \ 532 } while (0) 533 534#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 535static inline void rcu_preempt_sleep_check(void) 536{ 537 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map), 538 "Illegal context switch in RCU read-side critical section"); 539} 540#else /* #ifdef CONFIG_PROVE_RCU */ 541static inline void rcu_preempt_sleep_check(void) 542{ 543} 544#endif /* #else #ifdef CONFIG_PROVE_RCU */ 545 546#define rcu_sleep_check() \ 547 do { \ 548 rcu_preempt_sleep_check(); \ 549 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \ 550 "Illegal context switch in RCU-bh read-side critical section"); \ 551 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \ 552 "Illegal context switch in RCU-sched read-side critical section"); \ 553 } while (0) 554 555#else /* #ifdef CONFIG_PROVE_RCU */ 556 557#define rcu_lockdep_assert(c, s) do { } while (0) 558#define rcu_sleep_check() do { } while (0) 559 560#endif /* #else #ifdef CONFIG_PROVE_RCU */ 561 562/* 563 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 564 * and rcu_assign_pointer(). Some of these could be folded into their 565 * callers, but they are left separate in order to ease introduction of 566 * multiple flavors of pointers to match the multiple flavors of RCU 567 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in 568 * the future. 569 */ 570 571#ifdef __CHECKER__ 572#define rcu_dereference_sparse(p, space) \ 573 ((void)(((typeof(*p) space *)p) == p)) 574#else /* #ifdef __CHECKER__ */ 575#define rcu_dereference_sparse(p, space) 576#endif /* #else #ifdef __CHECKER__ */ 577 578#define __rcu_access_pointer(p, space) \ 579({ \ 580 typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \ 581 rcu_dereference_sparse(p, space); \ 582 ((typeof(*p) __force __kernel *)(_________p1)); \ 583}) 584#define __rcu_dereference_check(p, c, space) \ 585({ \ 586 /* Dependency order vs. p above. */ \ 587 typeof(*p) *________p1 = (typeof(*p) *__force)lockless_dereference(p); \ 588 rcu_lockdep_assert(c, "suspicious rcu_dereference_check() usage"); \ 589 rcu_dereference_sparse(p, space); \ 590 ((typeof(*p) __force __kernel *)(________p1)); \ 591}) 592#define __rcu_dereference_protected(p, c, space) \ 593({ \ 594 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected() usage"); \ 595 rcu_dereference_sparse(p, space); \ 596 ((typeof(*p) __force __kernel *)(p)); \ 597}) 598 599#define __rcu_access_index(p, space) \ 600({ \ 601 typeof(p) _________p1 = ACCESS_ONCE(p); \ 602 rcu_dereference_sparse(p, space); \ 603 (_________p1); \ 604}) 605#define __rcu_dereference_index_check(p, c) \ 606({ \ 607 /* Dependency order vs. p above. */ \ 608 typeof(p) _________p1 = lockless_dereference(p); \ 609 rcu_lockdep_assert(c, \ 610 "suspicious rcu_dereference_index_check() usage"); \ 611 (_________p1); \ 612}) 613 614/** 615 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 616 * @v: The value to statically initialize with. 617 */ 618#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 619 620/** 621 * lockless_dereference() - safely load a pointer for later dereference 622 * @p: The pointer to load 623 * 624 * Similar to rcu_dereference(), but for situations where the pointed-to 625 * object's lifetime is managed by something other than RCU. That 626 * "something other" might be reference counting or simple immortality. 627 */ 628#define lockless_dereference(p) \ 629({ \ 630 typeof(p) _________p1 = ACCESS_ONCE(p); \ 631 smp_read_barrier_depends(); /* Dependency order vs. p above. */ \ 632 (_________p1); \ 633}) 634 635/** 636 * rcu_assign_pointer() - assign to RCU-protected pointer 637 * @p: pointer to assign to 638 * @v: value to assign (publish) 639 * 640 * Assigns the specified value to the specified RCU-protected 641 * pointer, ensuring that any concurrent RCU readers will see 642 * any prior initialization. 643 * 644 * Inserts memory barriers on architectures that require them 645 * (which is most of them), and also prevents the compiler from 646 * reordering the code that initializes the structure after the pointer 647 * assignment. More importantly, this call documents which pointers 648 * will be dereferenced by RCU read-side code. 649 * 650 * In some special cases, you may use RCU_INIT_POINTER() instead 651 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 652 * to the fact that it does not constrain either the CPU or the compiler. 653 * That said, using RCU_INIT_POINTER() when you should have used 654 * rcu_assign_pointer() is a very bad thing that results in 655 * impossible-to-diagnose memory corruption. So please be careful. 656 * See the RCU_INIT_POINTER() comment header for details. 657 * 658 * Note that rcu_assign_pointer() evaluates each of its arguments only 659 * once, appearances notwithstanding. One of the "extra" evaluations 660 * is in typeof() and the other visible only to sparse (__CHECKER__), 661 * neither of which actually execute the argument. As with most cpp 662 * macros, this execute-arguments-only-once property is important, so 663 * please be careful when making changes to rcu_assign_pointer() and the 664 * other macros that it invokes. 665 */ 666#define rcu_assign_pointer(p, v) smp_store_release(&p, RCU_INITIALIZER(v)) 667 668/** 669 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 670 * @p: The pointer to read 671 * 672 * Return the value of the specified RCU-protected pointer, but omit the 673 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful 674 * when the value of this pointer is accessed, but the pointer is not 675 * dereferenced, for example, when testing an RCU-protected pointer against 676 * NULL. Although rcu_access_pointer() may also be used in cases where 677 * update-side locks prevent the value of the pointer from changing, you 678 * should instead use rcu_dereference_protected() for this use case. 679 * 680 * It is also permissible to use rcu_access_pointer() when read-side 681 * access to the pointer was removed at least one grace period ago, as 682 * is the case in the context of the RCU callback that is freeing up 683 * the data, or after a synchronize_rcu() returns. This can be useful 684 * when tearing down multi-linked structures after a grace period 685 * has elapsed. 686 */ 687#define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) 688 689/** 690 * rcu_dereference_check() - rcu_dereference with debug checking 691 * @p: The pointer to read, prior to dereferencing 692 * @c: The conditions under which the dereference will take place 693 * 694 * Do an rcu_dereference(), but check that the conditions under which the 695 * dereference will take place are correct. Typically the conditions 696 * indicate the various locking conditions that should be held at that 697 * point. The check should return true if the conditions are satisfied. 698 * An implicit check for being in an RCU read-side critical section 699 * (rcu_read_lock()) is included. 700 * 701 * For example: 702 * 703 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 704 * 705 * could be used to indicate to lockdep that foo->bar may only be dereferenced 706 * if either rcu_read_lock() is held, or that the lock required to replace 707 * the bar struct at foo->bar is held. 708 * 709 * Note that the list of conditions may also include indications of when a lock 710 * need not be held, for example during initialisation or destruction of the 711 * target struct: 712 * 713 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 714 * atomic_read(&foo->usage) == 0); 715 * 716 * Inserts memory barriers on architectures that require them 717 * (currently only the Alpha), prevents the compiler from refetching 718 * (and from merging fetches), and, more importantly, documents exactly 719 * which pointers are protected by RCU and checks that the pointer is 720 * annotated as __rcu. 721 */ 722#define rcu_dereference_check(p, c) \ 723 __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu) 724 725/** 726 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 727 * @p: The pointer to read, prior to dereferencing 728 * @c: The conditions under which the dereference will take place 729 * 730 * This is the RCU-bh counterpart to rcu_dereference_check(). 731 */ 732#define rcu_dereference_bh_check(p, c) \ 733 __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu) 734 735/** 736 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 737 * @p: The pointer to read, prior to dereferencing 738 * @c: The conditions under which the dereference will take place 739 * 740 * This is the RCU-sched counterpart to rcu_dereference_check(). 741 */ 742#define rcu_dereference_sched_check(p, c) \ 743 __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \ 744 __rcu) 745 746#define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/ 747 748/* 749 * The tracing infrastructure traces RCU (we want that), but unfortunately 750 * some of the RCU checks causes tracing to lock up the system. 751 * 752 * The tracing version of rcu_dereference_raw() must not call 753 * rcu_read_lock_held(). 754 */ 755#define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu) 756 757/** 758 * rcu_access_index() - fetch RCU index with no dereferencing 759 * @p: The index to read 760 * 761 * Return the value of the specified RCU-protected index, but omit the 762 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful 763 * when the value of this index is accessed, but the index is not 764 * dereferenced, for example, when testing an RCU-protected index against 765 * -1. Although rcu_access_index() may also be used in cases where 766 * update-side locks prevent the value of the index from changing, you 767 * should instead use rcu_dereference_index_protected() for this use case. 768 */ 769#define rcu_access_index(p) __rcu_access_index((p), __rcu) 770 771/** 772 * rcu_dereference_index_check() - rcu_dereference for indices with debug checking 773 * @p: The pointer to read, prior to dereferencing 774 * @c: The conditions under which the dereference will take place 775 * 776 * Similar to rcu_dereference_check(), but omits the sparse checking. 777 * This allows rcu_dereference_index_check() to be used on integers, 778 * which can then be used as array indices. Attempting to use 779 * rcu_dereference_check() on an integer will give compiler warnings 780 * because the sparse address-space mechanism relies on dereferencing 781 * the RCU-protected pointer. Dereferencing integers is not something 782 * that even gcc will put up with. 783 * 784 * Note that this function does not implicitly check for RCU read-side 785 * critical sections. If this function gains lots of uses, it might 786 * make sense to provide versions for each flavor of RCU, but it does 787 * not make sense as of early 2010. 788 */ 789#define rcu_dereference_index_check(p, c) \ 790 __rcu_dereference_index_check((p), (c)) 791 792/** 793 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 794 * @p: The pointer to read, prior to dereferencing 795 * @c: The conditions under which the dereference will take place 796 * 797 * Return the value of the specified RCU-protected pointer, but omit 798 * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This 799 * is useful in cases where update-side locks prevent the value of the 800 * pointer from changing. Please note that this primitive does -not- 801 * prevent the compiler from repeating this reference or combining it 802 * with other references, so it should not be used without protection 803 * of appropriate locks. 804 * 805 * This function is only for update-side use. Using this function 806 * when protected only by rcu_read_lock() will result in infrequent 807 * but very ugly failures. 808 */ 809#define rcu_dereference_protected(p, c) \ 810 __rcu_dereference_protected((p), (c), __rcu) 811 812 813/** 814 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 815 * @p: The pointer to read, prior to dereferencing 816 * 817 * This is a simple wrapper around rcu_dereference_check(). 818 */ 819#define rcu_dereference(p) rcu_dereference_check(p, 0) 820 821/** 822 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 823 * @p: The pointer to read, prior to dereferencing 824 * 825 * Makes rcu_dereference_check() do the dirty work. 826 */ 827#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 828 829/** 830 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 831 * @p: The pointer to read, prior to dereferencing 832 * 833 * Makes rcu_dereference_check() do the dirty work. 834 */ 835#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 836 837/** 838 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 839 * 840 * When synchronize_rcu() is invoked on one CPU while other CPUs 841 * are within RCU read-side critical sections, then the 842 * synchronize_rcu() is guaranteed to block until after all the other 843 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 844 * on one CPU while other CPUs are within RCU read-side critical 845 * sections, invocation of the corresponding RCU callback is deferred 846 * until after the all the other CPUs exit their critical sections. 847 * 848 * Note, however, that RCU callbacks are permitted to run concurrently 849 * with new RCU read-side critical sections. One way that this can happen 850 * is via the following sequence of events: (1) CPU 0 enters an RCU 851 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 852 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 853 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 854 * callback is invoked. This is legal, because the RCU read-side critical 855 * section that was running concurrently with the call_rcu() (and which 856 * therefore might be referencing something that the corresponding RCU 857 * callback would free up) has completed before the corresponding 858 * RCU callback is invoked. 859 * 860 * RCU read-side critical sections may be nested. Any deferred actions 861 * will be deferred until the outermost RCU read-side critical section 862 * completes. 863 * 864 * You can avoid reading and understanding the next paragraph by 865 * following this rule: don't put anything in an rcu_read_lock() RCU 866 * read-side critical section that would block in a !PREEMPT kernel. 867 * But if you want the full story, read on! 868 * 869 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), 870 * it is illegal to block while in an RCU read-side critical section. 871 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT 872 * kernel builds, RCU read-side critical sections may be preempted, 873 * but explicit blocking is illegal. Finally, in preemptible RCU 874 * implementations in real-time (with -rt patchset) kernel builds, RCU 875 * read-side critical sections may be preempted and they may also block, but 876 * only when acquiring spinlocks that are subject to priority inheritance. 877 */ 878static inline void rcu_read_lock(void) 879{ 880 __rcu_read_lock(); 881 __acquire(RCU); 882 rcu_lock_acquire(&rcu_lock_map); 883 rcu_lockdep_assert(rcu_is_watching(), 884 "rcu_read_lock() used illegally while idle"); 885} 886 887/* 888 * So where is rcu_write_lock()? It does not exist, as there is no 889 * way for writers to lock out RCU readers. This is a feature, not 890 * a bug -- this property is what provides RCU's performance benefits. 891 * Of course, writers must coordinate with each other. The normal 892 * spinlock primitives work well for this, but any other technique may be 893 * used as well. RCU does not care how the writers keep out of each 894 * others' way, as long as they do so. 895 */ 896 897/** 898 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 899 * 900 * In most situations, rcu_read_unlock() is immune from deadlock. 901 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() 902 * is responsible for deboosting, which it does via rt_mutex_unlock(). 903 * Unfortunately, this function acquires the scheduler's runqueue and 904 * priority-inheritance spinlocks. This means that deadlock could result 905 * if the caller of rcu_read_unlock() already holds one of these locks or 906 * any lock that is ever acquired while holding them; or any lock which 907 * can be taken from interrupt context because rcu_boost()->rt_mutex_lock() 908 * does not disable irqs while taking ->wait_lock. 909 * 910 * That said, RCU readers are never priority boosted unless they were 911 * preempted. Therefore, one way to avoid deadlock is to make sure 912 * that preemption never happens within any RCU read-side critical 913 * section whose outermost rcu_read_unlock() is called with one of 914 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in 915 * a number of ways, for example, by invoking preempt_disable() before 916 * critical section's outermost rcu_read_lock(). 917 * 918 * Given that the set of locks acquired by rt_mutex_unlock() might change 919 * at any time, a somewhat more future-proofed approach is to make sure 920 * that that preemption never happens within any RCU read-side critical 921 * section whose outermost rcu_read_unlock() is called with irqs disabled. 922 * This approach relies on the fact that rt_mutex_unlock() currently only 923 * acquires irq-disabled locks. 924 * 925 * The second of these two approaches is best in most situations, 926 * however, the first approach can also be useful, at least to those 927 * developers willing to keep abreast of the set of locks acquired by 928 * rt_mutex_unlock(). 929 * 930 * See rcu_read_lock() for more information. 931 */ 932static inline void rcu_read_unlock(void) 933{ 934 rcu_lockdep_assert(rcu_is_watching(), 935 "rcu_read_unlock() used illegally while idle"); 936 rcu_lock_release(&rcu_lock_map); 937 __release(RCU); 938 __rcu_read_unlock(); 939} 940 941/** 942 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 943 * 944 * This is equivalent of rcu_read_lock(), but to be used when updates 945 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since 946 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a 947 * softirq handler to be a quiescent state, a process in RCU read-side 948 * critical section must be protected by disabling softirqs. Read-side 949 * critical sections in interrupt context can use just rcu_read_lock(), 950 * though this should at least be commented to avoid confusing people 951 * reading the code. 952 * 953 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 954 * must occur in the same context, for example, it is illegal to invoke 955 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 956 * was invoked from some other task. 957 */ 958static inline void rcu_read_lock_bh(void) 959{ 960 local_bh_disable(); 961 __acquire(RCU_BH); 962 rcu_lock_acquire(&rcu_bh_lock_map); 963 rcu_lockdep_assert(rcu_is_watching(), 964 "rcu_read_lock_bh() used illegally while idle"); 965} 966 967/* 968 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section 969 * 970 * See rcu_read_lock_bh() for more information. 971 */ 972static inline void rcu_read_unlock_bh(void) 973{ 974 rcu_lockdep_assert(rcu_is_watching(), 975 "rcu_read_unlock_bh() used illegally while idle"); 976 rcu_lock_release(&rcu_bh_lock_map); 977 __release(RCU_BH); 978 local_bh_enable(); 979} 980 981/** 982 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 983 * 984 * This is equivalent of rcu_read_lock(), but to be used when updates 985 * are being done using call_rcu_sched() or synchronize_rcu_sched(). 986 * Read-side critical sections can also be introduced by anything that 987 * disables preemption, including local_irq_disable() and friends. 988 * 989 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 990 * must occur in the same context, for example, it is illegal to invoke 991 * rcu_read_unlock_sched() from process context if the matching 992 * rcu_read_lock_sched() was invoked from an NMI handler. 993 */ 994static inline void rcu_read_lock_sched(void) 995{ 996 preempt_disable(); 997 __acquire(RCU_SCHED); 998 rcu_lock_acquire(&rcu_sched_lock_map); 999 rcu_lockdep_assert(rcu_is_watching(), 1000 "rcu_read_lock_sched() used illegally while idle");
1001} 1002 1003/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 1004static inline notrace void rcu_read_lock_sched_notrace(void) 1005{ 1006 preempt_disable_notrace(); 1007 __acquire(RCU_SCHED); 1008} 1009 1010/* 1011 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section 1012 * 1013 * See rcu_read_lock_sched for more information. 1014 */ 1015static inline void rcu_read_unlock_sched(void) 1016{ 1017 rcu_lockdep_assert(rcu_is_watching(), 1018 "rcu_read_unlock_sched() used illegally while idle"); 1019 rcu_lock_release(&rcu_sched_lock_map); 1020 __release(RCU_SCHED); 1021 preempt_enable(); 1022} 1023 1024/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 1025static inline notrace void rcu_read_unlock_sched_notrace(void) 1026{ 1027 __release(RCU_SCHED); 1028 preempt_enable_notrace(); 1029} 1030 1031/** 1032 * RCU_INIT_POINTER() - initialize an RCU protected pointer 1033 * 1034 * Initialize an RCU-protected pointer in special cases where readers 1035 * do not need ordering constraints on the CPU or the compiler. These 1036 * special cases are: 1037 * 1038 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or- 1039 * 2. The caller has taken whatever steps are required to prevent 1040 * RCU readers from concurrently accessing this pointer -or- 1041 * 3. The referenced data structure has already been exposed to 1042 * readers either at compile time or via rcu_assign_pointer() -and- 1043 * a. You have not made -any- reader-visible changes to 1044 * this structure since then -or- 1045 * b. It is OK for readers accessing this structure from its 1046 * new location to see the old state of the structure. (For 1047 * example, the changes were to statistical counters or to 1048 * other state where exact synchronization is not required.) 1049 * 1050 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 1051 * result in impossible-to-diagnose memory corruption. As in the structures 1052 * will look OK in crash dumps, but any concurrent RCU readers might 1053 * see pre-initialized values of the referenced data structure. So 1054 * please be very careful how you use RCU_INIT_POINTER()!!! 1055 * 1056 * If you are creating an RCU-protected linked structure that is accessed 1057 * by a single external-to-structure RCU-protected pointer, then you may 1058 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 1059 * pointers, but you must use rcu_assign_pointer() to initialize the 1060 * external-to-structure pointer -after- you have completely initialized 1061 * the reader-accessible portions of the linked structure. 1062 * 1063 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 1064 * ordering guarantees for either the CPU or the compiler. 1065 */ 1066#define RCU_INIT_POINTER(p, v) \ 1067 do { \ 1068 rcu_dereference_sparse(p, __rcu); \ 1069 p = RCU_INITIALIZER(v); \ 1070 } while (0) 1071 1072/** 1073 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 1074 * 1075 * GCC-style initialization for an RCU-protected pointer in a structure field. 1076 */ 1077#define RCU_POINTER_INITIALIZER(p, v) \ 1078 .p = RCU_INITIALIZER(v) 1079 1080/* 1081 * Does the specified offset indicate that the corresponding rcu_head 1082 * structure can be handled by kfree_rcu()? 1083 */ 1084#define __is_kfree_rcu_offset(offset) ((offset) < 4096) 1085 1086/* 1087 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. 1088 */ 1089#define __kfree_rcu(head, offset) \ 1090 do { \ 1091 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \ 1092 kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \ 1093 } while (0) 1094 1095/** 1096 * kfree_rcu() - kfree an object after a grace period. 1097 * @ptr: pointer to kfree 1098 * @rcu_head: the name of the struct rcu_head within the type of @ptr. 1099 * 1100 * Many rcu callbacks functions just call kfree() on the base structure. 1101 * These functions are trivial, but their size adds up, and furthermore 1102 * when they are used in a kernel module, that module must invoke the 1103 * high-latency rcu_barrier() function at module-unload time. 1104 * 1105 * The kfree_rcu() function handles this issue. Rather than encoding a 1106 * function address in the embedded rcu_head structure, kfree_rcu() instead 1107 * encodes the offset of the rcu_head structure within the base structure. 1108 * Because the functions are not allowed in the low-order 4096 bytes of 1109 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 1110 * If the offset is larger than 4095 bytes, a compile-time error will 1111 * be generated in __kfree_rcu(). If this error is triggered, you can 1112 * either fall back to use of call_rcu() or rearrange the structure to 1113 * position the rcu_head structure into the first 4096 bytes. 1114 * 1115 * Note that the allowable offset might decrease in the future, for example, 1116 * to allow something like kmem_cache_free_rcu(). 1117 * 1118 * The BUILD_BUG_ON check must not involve any function calls, hence the 1119 * checks are done in macros here. 1120 */ 1121#define kfree_rcu(ptr, rcu_head) \ 1122 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head)) 1123 1124#if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL) 1125static inline int rcu_needs_cpu(unsigned long *delta_jiffies) 1126{ 1127 *delta_jiffies = ULONG_MAX; 1128 return 0; 1129} 1130#endif /* #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL) */ 1131 1132#if defined(CONFIG_RCU_NOCB_CPU_ALL) 1133static inline bool rcu_is_nocb_cpu(int cpu) { return true; } 1134#elif defined(CONFIG_RCU_NOCB_CPU) 1135bool rcu_is_nocb_cpu(int cpu); 1136#else 1137static inline bool rcu_is_nocb_cpu(int cpu) { return false; } 1138#endif 1139 1140 1141/* Only for use by adaptive-ticks code. */ 1142#ifdef CONFIG_NO_HZ_FULL_SYSIDLE 1143bool rcu_sys_is_idle(void); 1144void rcu_sysidle_force_exit(void); 1145#else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ 1146 1147static inline bool rcu_sys_is_idle(void) 1148{ 1149 return false; 1150} 1151 1152static inline void rcu_sysidle_force_exit(void) 1153{ 1154} 1155 1156#endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */ 1157 1158 1159#endif /* __LINUX_RCUPDATE_H */ 1160