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