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/compiler.h> 38 #include <linux/atomic.h> 39 #include <linux/irqflags.h> 40 #include <linux/preempt.h> 41 #include <linux/bottom_half.h> 42 #include <linux/lockdep.h> 43 #include <asm/processor.h> 44 #include <linux/cpumask.h> 45 46 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) 47 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) 48 #define ulong2long(a) (*(long *)(&(a))) 49 50 /* Exported common interfaces */ 51 52 #ifdef CONFIG_PREEMPT_RCU 53 void call_rcu(struct rcu_head *head, rcu_callback_t func); 54 #else /* #ifdef CONFIG_PREEMPT_RCU */ 55 #define call_rcu call_rcu_sched 56 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 57 58 void call_rcu_bh(struct rcu_head *head, rcu_callback_t func); 59 void call_rcu_sched(struct rcu_head *head, rcu_callback_t func); 60 void synchronize_sched(void); 61 void rcu_barrier_tasks(void); 62 63 #ifdef CONFIG_PREEMPT_RCU 64 65 void __rcu_read_lock(void); 66 void __rcu_read_unlock(void); 67 void rcu_read_unlock_special(struct task_struct *t); 68 void synchronize_rcu(void); 69 70 /* 71 * Defined as a macro as it is a very low level header included from 72 * areas that don't even know about current. This gives the rcu_read_lock() 73 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other 74 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. 75 */ 76 #define rcu_preempt_depth() (current->rcu_read_lock_nesting) 77 78 #else /* #ifdef CONFIG_PREEMPT_RCU */ 79 80 static inline void __rcu_read_lock(void) 81 { 82 if (IS_ENABLED(CONFIG_PREEMPT_COUNT)) 83 preempt_disable(); 84 } 85 86 static inline void __rcu_read_unlock(void) 87 { 88 if (IS_ENABLED(CONFIG_PREEMPT_COUNT)) 89 preempt_enable(); 90 } 91 92 static inline void synchronize_rcu(void) 93 { 94 synchronize_sched(); 95 } 96 97 static inline int rcu_preempt_depth(void) 98 { 99 return 0; 100 } 101 102 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 103 104 /* Internal to kernel */ 105 void rcu_init(void); 106 extern int rcu_scheduler_active __read_mostly; 107 void rcu_sched_qs(void); 108 void rcu_bh_qs(void); 109 void rcu_check_callbacks(int user); 110 void rcu_report_dead(unsigned int cpu); 111 void rcutree_migrate_callbacks(int cpu); 112 113 #ifdef CONFIG_RCU_STALL_COMMON 114 void rcu_sysrq_start(void); 115 void rcu_sysrq_end(void); 116 #else /* #ifdef CONFIG_RCU_STALL_COMMON */ 117 static inline void rcu_sysrq_start(void) { } 118 static inline void rcu_sysrq_end(void) { } 119 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ 120 121 #ifdef CONFIG_NO_HZ_FULL 122 void rcu_user_enter(void); 123 void rcu_user_exit(void); 124 #else 125 static inline void rcu_user_enter(void) { } 126 static inline void rcu_user_exit(void) { } 127 #endif /* CONFIG_NO_HZ_FULL */ 128 129 #ifdef CONFIG_RCU_NOCB_CPU 130 void rcu_init_nohz(void); 131 #else /* #ifdef CONFIG_RCU_NOCB_CPU */ 132 static inline void rcu_init_nohz(void) { } 133 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 134 135 /** 136 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers 137 * @a: Code that RCU needs to pay attention to. 138 * 139 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden 140 * in the inner idle loop, that is, between the rcu_idle_enter() and 141 * the rcu_idle_exit() -- RCU will happily ignore any such read-side 142 * critical sections. However, things like powertop need tracepoints 143 * in the inner idle loop. 144 * 145 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) 146 * will tell RCU that it needs to pay attention, invoke its argument 147 * (in this example, calling the do_something_with_RCU() function), 148 * and then tell RCU to go back to ignoring this CPU. It is permissible 149 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is 150 * on the order of a million or so, even on 32-bit systems). It is 151 * not legal to block within RCU_NONIDLE(), nor is it permissible to 152 * transfer control either into or out of RCU_NONIDLE()'s statement. 153 */ 154 #define RCU_NONIDLE(a) \ 155 do { \ 156 rcu_irq_enter_irqson(); \ 157 do { a; } while (0); \ 158 rcu_irq_exit_irqson(); \ 159 } while (0) 160 161 /* 162 * Note a voluntary context switch for RCU-tasks benefit. This is a 163 * macro rather than an inline function to avoid #include hell. 164 */ 165 #ifdef CONFIG_TASKS_RCU 166 #define rcu_note_voluntary_context_switch_lite(t) \ 167 do { \ 168 if (READ_ONCE((t)->rcu_tasks_holdout)) \ 169 WRITE_ONCE((t)->rcu_tasks_holdout, false); \ 170 } while (0) 171 #define rcu_note_voluntary_context_switch(t) \ 172 do { \ 173 rcu_all_qs(); \ 174 rcu_note_voluntary_context_switch_lite(t); \ 175 } while (0) 176 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func); 177 void synchronize_rcu_tasks(void); 178 void exit_tasks_rcu_start(void); 179 void exit_tasks_rcu_finish(void); 180 #else /* #ifdef CONFIG_TASKS_RCU */ 181 #define rcu_note_voluntary_context_switch_lite(t) do { } while (0) 182 #define rcu_note_voluntary_context_switch(t) rcu_all_qs() 183 #define call_rcu_tasks call_rcu_sched 184 #define synchronize_rcu_tasks synchronize_sched 185 static inline void exit_tasks_rcu_start(void) { } 186 static inline void exit_tasks_rcu_finish(void) { } 187 #endif /* #else #ifdef CONFIG_TASKS_RCU */ 188 189 /** 190 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU 191 * 192 * This macro resembles cond_resched(), except that it is defined to 193 * report potential quiescent states to RCU-tasks even if the cond_resched() 194 * machinery were to be shut off, as some advocate for PREEMPT kernels. 195 */ 196 #define cond_resched_tasks_rcu_qs() \ 197 do { \ 198 if (!cond_resched()) \ 199 rcu_note_voluntary_context_switch_lite(current); \ 200 } while (0) 201 202 /* 203 * Infrastructure to implement the synchronize_() primitives in 204 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 205 */ 206 207 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) 208 #include <linux/rcutree.h> 209 #elif defined(CONFIG_TINY_RCU) 210 #include <linux/rcutiny.h> 211 #else 212 #error "Unknown RCU implementation specified to kernel configuration" 213 #endif 214 215 /* 216 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls 217 * are needed for dynamic initialization and destruction of rcu_head 218 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for 219 * dynamic initialization and destruction of statically allocated rcu_head 220 * structures. However, rcu_head structures allocated dynamically in the 221 * heap don't need any initialization. 222 */ 223 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 224 void init_rcu_head(struct rcu_head *head); 225 void destroy_rcu_head(struct rcu_head *head); 226 void init_rcu_head_on_stack(struct rcu_head *head); 227 void destroy_rcu_head_on_stack(struct rcu_head *head); 228 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 229 static inline void init_rcu_head(struct rcu_head *head) { } 230 static inline void destroy_rcu_head(struct rcu_head *head) { } 231 static inline void init_rcu_head_on_stack(struct rcu_head *head) { } 232 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { } 233 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 234 235 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 236 bool rcu_lockdep_current_cpu_online(void); 237 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 238 static inline bool rcu_lockdep_current_cpu_online(void) { return true; } 239 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 240 241 #ifdef CONFIG_DEBUG_LOCK_ALLOC 242 243 static inline void rcu_lock_acquire(struct lockdep_map *map) 244 { 245 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); 246 } 247 248 static inline void rcu_lock_release(struct lockdep_map *map) 249 { 250 lock_release(map, 1, _THIS_IP_); 251 } 252 253 extern struct lockdep_map rcu_lock_map; 254 extern struct lockdep_map rcu_bh_lock_map; 255 extern struct lockdep_map rcu_sched_lock_map; 256 extern struct lockdep_map rcu_callback_map; 257 int debug_lockdep_rcu_enabled(void); 258 int rcu_read_lock_held(void); 259 int rcu_read_lock_bh_held(void); 260 int rcu_read_lock_sched_held(void); 261 262 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 263 264 # define rcu_lock_acquire(a) do { } while (0) 265 # define rcu_lock_release(a) do { } while (0) 266 267 static inline int rcu_read_lock_held(void) 268 { 269 return 1; 270 } 271 272 static inline int rcu_read_lock_bh_held(void) 273 { 274 return 1; 275 } 276 277 static inline int rcu_read_lock_sched_held(void) 278 { 279 return !preemptible(); 280 } 281 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 282 283 #ifdef CONFIG_PROVE_RCU 284 285 /** 286 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met 287 * @c: condition to check 288 * @s: informative message 289 */ 290 #define RCU_LOCKDEP_WARN(c, s) \ 291 do { \ 292 static bool __section(.data.unlikely) __warned; \ 293 if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \ 294 __warned = true; \ 295 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 296 } \ 297 } while (0) 298 299 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 300 static inline void rcu_preempt_sleep_check(void) 301 { 302 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), 303 "Illegal context switch in RCU read-side critical section"); 304 } 305 #else /* #ifdef CONFIG_PROVE_RCU */ 306 static inline void rcu_preempt_sleep_check(void) { } 307 #endif /* #else #ifdef CONFIG_PROVE_RCU */ 308 309 #define rcu_sleep_check() \ 310 do { \ 311 rcu_preempt_sleep_check(); \ 312 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ 313 "Illegal context switch in RCU-bh read-side critical section"); \ 314 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ 315 "Illegal context switch in RCU-sched read-side critical section"); \ 316 } while (0) 317 318 #else /* #ifdef CONFIG_PROVE_RCU */ 319 320 #define RCU_LOCKDEP_WARN(c, s) do { } while (0) 321 #define rcu_sleep_check() do { } while (0) 322 323 #endif /* #else #ifdef CONFIG_PROVE_RCU */ 324 325 /* 326 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 327 * and rcu_assign_pointer(). Some of these could be folded into their 328 * callers, but they are left separate in order to ease introduction of 329 * multiple flavors of pointers to match the multiple flavors of RCU 330 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in 331 * the future. 332 */ 333 334 #ifdef __CHECKER__ 335 #define rcu_dereference_sparse(p, space) \ 336 ((void)(((typeof(*p) space *)p) == p)) 337 #else /* #ifdef __CHECKER__ */ 338 #define rcu_dereference_sparse(p, space) 339 #endif /* #else #ifdef __CHECKER__ */ 340 341 #define __rcu_access_pointer(p, space) \ 342 ({ \ 343 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 344 rcu_dereference_sparse(p, space); \ 345 ((typeof(*p) __force __kernel *)(_________p1)); \ 346 }) 347 #define __rcu_dereference_check(p, c, space) \ 348 ({ \ 349 /* Dependency order vs. p above. */ \ 350 typeof(*p) *________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 351 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ 352 rcu_dereference_sparse(p, space); \ 353 ((typeof(*p) __force __kernel *)(________p1)); \ 354 }) 355 #define __rcu_dereference_protected(p, c, space) \ 356 ({ \ 357 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ 358 rcu_dereference_sparse(p, space); \ 359 ((typeof(*p) __force __kernel *)(p)); \ 360 }) 361 #define rcu_dereference_raw(p) \ 362 ({ \ 363 /* Dependency order vs. p above. */ \ 364 typeof(p) ________p1 = READ_ONCE(p); \ 365 ((typeof(*p) __force __kernel *)(________p1)); \ 366 }) 367 368 /** 369 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 370 * @v: The value to statically initialize with. 371 */ 372 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 373 374 /** 375 * rcu_assign_pointer() - assign to RCU-protected pointer 376 * @p: pointer to assign to 377 * @v: value to assign (publish) 378 * 379 * Assigns the specified value to the specified RCU-protected 380 * pointer, ensuring that any concurrent RCU readers will see 381 * any prior initialization. 382 * 383 * Inserts memory barriers on architectures that require them 384 * (which is most of them), and also prevents the compiler from 385 * reordering the code that initializes the structure after the pointer 386 * assignment. More importantly, this call documents which pointers 387 * will be dereferenced by RCU read-side code. 388 * 389 * In some special cases, you may use RCU_INIT_POINTER() instead 390 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 391 * to the fact that it does not constrain either the CPU or the compiler. 392 * That said, using RCU_INIT_POINTER() when you should have used 393 * rcu_assign_pointer() is a very bad thing that results in 394 * impossible-to-diagnose memory corruption. So please be careful. 395 * See the RCU_INIT_POINTER() comment header for details. 396 * 397 * Note that rcu_assign_pointer() evaluates each of its arguments only 398 * once, appearances notwithstanding. One of the "extra" evaluations 399 * is in typeof() and the other visible only to sparse (__CHECKER__), 400 * neither of which actually execute the argument. As with most cpp 401 * macros, this execute-arguments-only-once property is important, so 402 * please be careful when making changes to rcu_assign_pointer() and the 403 * other macros that it invokes. 404 */ 405 #define rcu_assign_pointer(p, v) \ 406 ({ \ 407 uintptr_t _r_a_p__v = (uintptr_t)(v); \ 408 \ 409 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \ 410 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \ 411 else \ 412 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \ 413 _r_a_p__v; \ 414 }) 415 416 /** 417 * rcu_swap_protected() - swap an RCU and a regular pointer 418 * @rcu_ptr: RCU pointer 419 * @ptr: regular pointer 420 * @c: the conditions under which the dereference will take place 421 * 422 * Perform swap(@rcu_ptr, @ptr) where @rcu_ptr is an RCU-annotated pointer and 423 * @c is the argument that is passed to the rcu_dereference_protected() call 424 * used to read that pointer. 425 */ 426 #define rcu_swap_protected(rcu_ptr, ptr, c) do { \ 427 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \ 428 rcu_assign_pointer((rcu_ptr), (ptr)); \ 429 (ptr) = __tmp; \ 430 } while (0) 431 432 /** 433 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 434 * @p: The pointer to read 435 * 436 * Return the value of the specified RCU-protected pointer, but omit the 437 * lockdep checks for being in an RCU read-side critical section. This is 438 * useful when the value of this pointer is accessed, but the pointer is 439 * not dereferenced, for example, when testing an RCU-protected pointer 440 * against NULL. Although rcu_access_pointer() may also be used in cases 441 * where update-side locks prevent the value of the pointer from changing, 442 * you should instead use rcu_dereference_protected() for this use case. 443 * 444 * It is also permissible to use rcu_access_pointer() when read-side 445 * access to the pointer was removed at least one grace period ago, as 446 * is the case in the context of the RCU callback that is freeing up 447 * the data, or after a synchronize_rcu() returns. This can be useful 448 * when tearing down multi-linked structures after a grace period 449 * has elapsed. 450 */ 451 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) 452 453 /** 454 * rcu_dereference_check() - rcu_dereference with debug checking 455 * @p: The pointer to read, prior to dereferencing 456 * @c: The conditions under which the dereference will take place 457 * 458 * Do an rcu_dereference(), but check that the conditions under which the 459 * dereference will take place are correct. Typically the conditions 460 * indicate the various locking conditions that should be held at that 461 * point. The check should return true if the conditions are satisfied. 462 * An implicit check for being in an RCU read-side critical section 463 * (rcu_read_lock()) is included. 464 * 465 * For example: 466 * 467 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 468 * 469 * could be used to indicate to lockdep that foo->bar may only be dereferenced 470 * if either rcu_read_lock() is held, or that the lock required to replace 471 * the bar struct at foo->bar is held. 472 * 473 * Note that the list of conditions may also include indications of when a lock 474 * need not be held, for example during initialisation or destruction of the 475 * target struct: 476 * 477 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 478 * atomic_read(&foo->usage) == 0); 479 * 480 * Inserts memory barriers on architectures that require them 481 * (currently only the Alpha), prevents the compiler from refetching 482 * (and from merging fetches), and, more importantly, documents exactly 483 * which pointers are protected by RCU and checks that the pointer is 484 * annotated as __rcu. 485 */ 486 #define rcu_dereference_check(p, c) \ 487 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu) 488 489 /** 490 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 491 * @p: The pointer to read, prior to dereferencing 492 * @c: The conditions under which the dereference will take place 493 * 494 * This is the RCU-bh counterpart to rcu_dereference_check(). 495 */ 496 #define rcu_dereference_bh_check(p, c) \ 497 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu) 498 499 /** 500 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 501 * @p: The pointer to read, prior to dereferencing 502 * @c: The conditions under which the dereference will take place 503 * 504 * This is the RCU-sched counterpart to rcu_dereference_check(). 505 */ 506 #define rcu_dereference_sched_check(p, c) \ 507 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \ 508 __rcu) 509 510 /* 511 * The tracing infrastructure traces RCU (we want that), but unfortunately 512 * some of the RCU checks causes tracing to lock up the system. 513 * 514 * The no-tracing version of rcu_dereference_raw() must not call 515 * rcu_read_lock_held(). 516 */ 517 #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu) 518 519 /** 520 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 521 * @p: The pointer to read, prior to dereferencing 522 * @c: The conditions under which the dereference will take place 523 * 524 * Return the value of the specified RCU-protected pointer, but omit 525 * the READ_ONCE(). This is useful in cases where update-side locks 526 * prevent the value of the pointer from changing. Please note that this 527 * primitive does *not* prevent the compiler from repeating this reference 528 * or combining it with other references, so it should not be used without 529 * protection of appropriate locks. 530 * 531 * This function is only for update-side use. Using this function 532 * when protected only by rcu_read_lock() will result in infrequent 533 * but very ugly failures. 534 */ 535 #define rcu_dereference_protected(p, c) \ 536 __rcu_dereference_protected((p), (c), __rcu) 537 538 539 /** 540 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 541 * @p: The pointer to read, prior to dereferencing 542 * 543 * This is a simple wrapper around rcu_dereference_check(). 544 */ 545 #define rcu_dereference(p) rcu_dereference_check(p, 0) 546 547 /** 548 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 549 * @p: The pointer to read, prior to dereferencing 550 * 551 * Makes rcu_dereference_check() do the dirty work. 552 */ 553 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 554 555 /** 556 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 557 * @p: The pointer to read, prior to dereferencing 558 * 559 * Makes rcu_dereference_check() do the dirty work. 560 */ 561 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 562 563 /** 564 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism 565 * @p: The pointer to hand off 566 * 567 * This is simply an identity function, but it documents where a pointer 568 * is handed off from RCU to some other synchronization mechanism, for 569 * example, reference counting or locking. In C11, it would map to 570 * kill_dependency(). It could be used as follows: 571 * `` 572 * rcu_read_lock(); 573 * p = rcu_dereference(gp); 574 * long_lived = is_long_lived(p); 575 * if (long_lived) { 576 * if (!atomic_inc_not_zero(p->refcnt)) 577 * long_lived = false; 578 * else 579 * p = rcu_pointer_handoff(p); 580 * } 581 * rcu_read_unlock(); 582 *`` 583 */ 584 #define rcu_pointer_handoff(p) (p) 585 586 /** 587 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 588 * 589 * When synchronize_rcu() is invoked on one CPU while other CPUs 590 * are within RCU read-side critical sections, then the 591 * synchronize_rcu() is guaranteed to block until after all the other 592 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 593 * on one CPU while other CPUs are within RCU read-side critical 594 * sections, invocation of the corresponding RCU callback is deferred 595 * until after the all the other CPUs exit their critical sections. 596 * 597 * Note, however, that RCU callbacks are permitted to run concurrently 598 * with new RCU read-side critical sections. One way that this can happen 599 * is via the following sequence of events: (1) CPU 0 enters an RCU 600 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 601 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 602 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 603 * callback is invoked. This is legal, because the RCU read-side critical 604 * section that was running concurrently with the call_rcu() (and which 605 * therefore might be referencing something that the corresponding RCU 606 * callback would free up) has completed before the corresponding 607 * RCU callback is invoked. 608 * 609 * RCU read-side critical sections may be nested. Any deferred actions 610 * will be deferred until the outermost RCU read-side critical section 611 * completes. 612 * 613 * You can avoid reading and understanding the next paragraph by 614 * following this rule: don't put anything in an rcu_read_lock() RCU 615 * read-side critical section that would block in a !PREEMPT kernel. 616 * But if you want the full story, read on! 617 * 618 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), 619 * it is illegal to block while in an RCU read-side critical section. 620 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT 621 * kernel builds, RCU read-side critical sections may be preempted, 622 * but explicit blocking is illegal. Finally, in preemptible RCU 623 * implementations in real-time (with -rt patchset) kernel builds, RCU 624 * read-side critical sections may be preempted and they may also block, but 625 * only when acquiring spinlocks that are subject to priority inheritance. 626 */ 627 static inline void rcu_read_lock(void) 628 { 629 __rcu_read_lock(); 630 __acquire(RCU); 631 rcu_lock_acquire(&rcu_lock_map); 632 RCU_LOCKDEP_WARN(!rcu_is_watching(), 633 "rcu_read_lock() used illegally while idle"); 634 } 635 636 /* 637 * So where is rcu_write_lock()? It does not exist, as there is no 638 * way for writers to lock out RCU readers. This is a feature, not 639 * a bug -- this property is what provides RCU's performance benefits. 640 * Of course, writers must coordinate with each other. The normal 641 * spinlock primitives work well for this, but any other technique may be 642 * used as well. RCU does not care how the writers keep out of each 643 * others' way, as long as they do so. 644 */ 645 646 /** 647 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 648 * 649 * In most situations, rcu_read_unlock() is immune from deadlock. 650 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() 651 * is responsible for deboosting, which it does via rt_mutex_unlock(). 652 * Unfortunately, this function acquires the scheduler's runqueue and 653 * priority-inheritance spinlocks. This means that deadlock could result 654 * if the caller of rcu_read_unlock() already holds one of these locks or 655 * any lock that is ever acquired while holding them. 656 * 657 * That said, RCU readers are never priority boosted unless they were 658 * preempted. Therefore, one way to avoid deadlock is to make sure 659 * that preemption never happens within any RCU read-side critical 660 * section whose outermost rcu_read_unlock() is called with one of 661 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in 662 * a number of ways, for example, by invoking preempt_disable() before 663 * critical section's outermost rcu_read_lock(). 664 * 665 * Given that the set of locks acquired by rt_mutex_unlock() might change 666 * at any time, a somewhat more future-proofed approach is to make sure 667 * that that preemption never happens within any RCU read-side critical 668 * section whose outermost rcu_read_unlock() is called with irqs disabled. 669 * This approach relies on the fact that rt_mutex_unlock() currently only 670 * acquires irq-disabled locks. 671 * 672 * The second of these two approaches is best in most situations, 673 * however, the first approach can also be useful, at least to those 674 * developers willing to keep abreast of the set of locks acquired by 675 * rt_mutex_unlock(). 676 * 677 * See rcu_read_lock() for more information. 678 */ 679 static inline void rcu_read_unlock(void) 680 { 681 RCU_LOCKDEP_WARN(!rcu_is_watching(), 682 "rcu_read_unlock() used illegally while idle"); 683 __release(RCU); 684 __rcu_read_unlock(); 685 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ 686 } 687 688 /** 689 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 690 * 691 * This is equivalent of rcu_read_lock(), but to be used when updates 692 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since 693 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a 694 * softirq handler to be a quiescent state, a process in RCU read-side 695 * critical section must be protected by disabling softirqs. Read-side 696 * critical sections in interrupt context can use just rcu_read_lock(), 697 * though this should at least be commented to avoid confusing people 698 * reading the code. 699 * 700 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 701 * must occur in the same context, for example, it is illegal to invoke 702 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 703 * was invoked from some other task. 704 */ 705 static inline void rcu_read_lock_bh(void) 706 { 707 local_bh_disable(); 708 __acquire(RCU_BH); 709 rcu_lock_acquire(&rcu_bh_lock_map); 710 RCU_LOCKDEP_WARN(!rcu_is_watching(), 711 "rcu_read_lock_bh() used illegally while idle"); 712 } 713 714 /* 715 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section 716 * 717 * See rcu_read_lock_bh() for more information. 718 */ 719 static inline void rcu_read_unlock_bh(void) 720 { 721 RCU_LOCKDEP_WARN(!rcu_is_watching(), 722 "rcu_read_unlock_bh() used illegally while idle"); 723 rcu_lock_release(&rcu_bh_lock_map); 724 __release(RCU_BH); 725 local_bh_enable(); 726 } 727 728 /** 729 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 730 * 731 * This is equivalent of rcu_read_lock(), but to be used when updates 732 * are being done using call_rcu_sched() or synchronize_rcu_sched(). 733 * Read-side critical sections can also be introduced by anything that 734 * disables preemption, including local_irq_disable() and friends. 735 * 736 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 737 * must occur in the same context, for example, it is illegal to invoke 738 * rcu_read_unlock_sched() from process context if the matching 739 * rcu_read_lock_sched() was invoked from an NMI handler. 740 */ 741 static inline void rcu_read_lock_sched(void) 742 { 743 preempt_disable(); 744 __acquire(RCU_SCHED); 745 rcu_lock_acquire(&rcu_sched_lock_map); 746 RCU_LOCKDEP_WARN(!rcu_is_watching(), 747 "rcu_read_lock_sched() used illegally while idle"); 748 } 749 750 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 751 static inline notrace void rcu_read_lock_sched_notrace(void) 752 { 753 preempt_disable_notrace(); 754 __acquire(RCU_SCHED); 755 } 756 757 /* 758 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section 759 * 760 * See rcu_read_lock_sched for more information. 761 */ 762 static inline void rcu_read_unlock_sched(void) 763 { 764 RCU_LOCKDEP_WARN(!rcu_is_watching(), 765 "rcu_read_unlock_sched() used illegally while idle"); 766 rcu_lock_release(&rcu_sched_lock_map); 767 __release(RCU_SCHED); 768 preempt_enable(); 769 } 770 771 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 772 static inline notrace void rcu_read_unlock_sched_notrace(void) 773 { 774 __release(RCU_SCHED); 775 preempt_enable_notrace(); 776 } 777 778 /** 779 * RCU_INIT_POINTER() - initialize an RCU protected pointer 780 * @p: The pointer to be initialized. 781 * @v: The value to initialized the pointer to. 782 * 783 * Initialize an RCU-protected pointer in special cases where readers 784 * do not need ordering constraints on the CPU or the compiler. These 785 * special cases are: 786 * 787 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or* 788 * 2. The caller has taken whatever steps are required to prevent 789 * RCU readers from concurrently accessing this pointer *or* 790 * 3. The referenced data structure has already been exposed to 791 * readers either at compile time or via rcu_assign_pointer() *and* 792 * 793 * a. You have not made *any* reader-visible changes to 794 * this structure since then *or* 795 * b. It is OK for readers accessing this structure from its 796 * new location to see the old state of the structure. (For 797 * example, the changes were to statistical counters or to 798 * other state where exact synchronization is not required.) 799 * 800 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 801 * result in impossible-to-diagnose memory corruption. As in the structures 802 * will look OK in crash dumps, but any concurrent RCU readers might 803 * see pre-initialized values of the referenced data structure. So 804 * please be very careful how you use RCU_INIT_POINTER()!!! 805 * 806 * If you are creating an RCU-protected linked structure that is accessed 807 * by a single external-to-structure RCU-protected pointer, then you may 808 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 809 * pointers, but you must use rcu_assign_pointer() to initialize the 810 * external-to-structure pointer *after* you have completely initialized 811 * the reader-accessible portions of the linked structure. 812 * 813 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 814 * ordering guarantees for either the CPU or the compiler. 815 */ 816 #define RCU_INIT_POINTER(p, v) \ 817 do { \ 818 rcu_dereference_sparse(p, __rcu); \ 819 WRITE_ONCE(p, RCU_INITIALIZER(v)); \ 820 } while (0) 821 822 /** 823 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 824 * @p: The pointer to be initialized. 825 * @v: The value to initialized the pointer to. 826 * 827 * GCC-style initialization for an RCU-protected pointer in a structure field. 828 */ 829 #define RCU_POINTER_INITIALIZER(p, v) \ 830 .p = RCU_INITIALIZER(v) 831 832 /* 833 * Does the specified offset indicate that the corresponding rcu_head 834 * structure can be handled by kfree_rcu()? 835 */ 836 #define __is_kfree_rcu_offset(offset) ((offset) < 4096) 837 838 /* 839 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. 840 */ 841 #define __kfree_rcu(head, offset) \ 842 do { \ 843 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \ 844 kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \ 845 } while (0) 846 847 /** 848 * kfree_rcu() - kfree an object after a grace period. 849 * @ptr: pointer to kfree 850 * @rcu_head: the name of the struct rcu_head within the type of @ptr. 851 * 852 * Many rcu callbacks functions just call kfree() on the base structure. 853 * These functions are trivial, but their size adds up, and furthermore 854 * when they are used in a kernel module, that module must invoke the 855 * high-latency rcu_barrier() function at module-unload time. 856 * 857 * The kfree_rcu() function handles this issue. Rather than encoding a 858 * function address in the embedded rcu_head structure, kfree_rcu() instead 859 * encodes the offset of the rcu_head structure within the base structure. 860 * Because the functions are not allowed in the low-order 4096 bytes of 861 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 862 * If the offset is larger than 4095 bytes, a compile-time error will 863 * be generated in __kfree_rcu(). If this error is triggered, you can 864 * either fall back to use of call_rcu() or rearrange the structure to 865 * position the rcu_head structure into the first 4096 bytes. 866 * 867 * Note that the allowable offset might decrease in the future, for example, 868 * to allow something like kmem_cache_free_rcu(). 869 * 870 * The BUILD_BUG_ON check must not involve any function calls, hence the 871 * checks are done in macros here. 872 */ 873 #define kfree_rcu(ptr, rcu_head) \ 874 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head)) 875 876 877 /* 878 * Place this after a lock-acquisition primitive to guarantee that 879 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies 880 * if the UNLOCK and LOCK are executed by the same CPU or if the 881 * UNLOCK and LOCK operate on the same lock variable. 882 */ 883 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE 884 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */ 885 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 886 #define smp_mb__after_unlock_lock() do { } while (0) 887 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 888 889 890 #endif /* __LINUX_RCUPDATE_H */ 891