1 /* SPDX-License-Identifier: GPL-2.0+ */ 2 /* 3 * Read-Copy Update mechanism for mutual exclusion 4 * 5 * Copyright IBM Corporation, 2001 6 * 7 * Author: Dipankar Sarma <dipankar@in.ibm.com> 8 * 9 * Based on the original work by Paul McKenney <paulmck@vnet.ibm.com> 10 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. 11 * Papers: 12 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf 13 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001) 14 * 15 * For detailed explanation of Read-Copy Update mechanism see - 16 * http://lse.sourceforge.net/locking/rcupdate.html 17 * 18 */ 19 20 #ifndef __LINUX_RCUPDATE_H 21 #define __LINUX_RCUPDATE_H 22 23 #include <linux/types.h> 24 #include <linux/compiler.h> 25 #include <linux/atomic.h> 26 #include <linux/irqflags.h> 27 #include <linux/preempt.h> 28 #include <linux/bottom_half.h> 29 #include <linux/lockdep.h> 30 #include <asm/processor.h> 31 #include <linux/cpumask.h> 32 33 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) 34 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) 35 #define ulong2long(a) (*(long *)(&(a))) 36 37 /* Exported common interfaces */ 38 void call_rcu(struct rcu_head *head, rcu_callback_t func); 39 void rcu_barrier_tasks(void); 40 void synchronize_rcu(void); 41 42 #ifdef CONFIG_PREEMPT_RCU 43 44 void __rcu_read_lock(void); 45 void __rcu_read_unlock(void); 46 47 /* 48 * Defined as a macro as it is a very low level header included from 49 * areas that don't even know about current. This gives the rcu_read_lock() 50 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other 51 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. 52 */ 53 #define rcu_preempt_depth() (current->rcu_read_lock_nesting) 54 55 #else /* #ifdef CONFIG_PREEMPT_RCU */ 56 57 static inline void __rcu_read_lock(void) 58 { 59 preempt_disable(); 60 } 61 62 static inline void __rcu_read_unlock(void) 63 { 64 preempt_enable(); 65 } 66 67 static inline int rcu_preempt_depth(void) 68 { 69 return 0; 70 } 71 72 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 73 74 /* Internal to kernel */ 75 void rcu_init(void); 76 extern int rcu_scheduler_active __read_mostly; 77 void rcu_sched_clock_irq(int user); 78 void rcu_report_dead(unsigned int cpu); 79 void rcutree_migrate_callbacks(int cpu); 80 81 #ifdef CONFIG_RCU_STALL_COMMON 82 void rcu_sysrq_start(void); 83 void rcu_sysrq_end(void); 84 #else /* #ifdef CONFIG_RCU_STALL_COMMON */ 85 static inline void rcu_sysrq_start(void) { } 86 static inline void rcu_sysrq_end(void) { } 87 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ 88 89 #ifdef CONFIG_NO_HZ_FULL 90 void rcu_user_enter(void); 91 void rcu_user_exit(void); 92 #else 93 static inline void rcu_user_enter(void) { } 94 static inline void rcu_user_exit(void) { } 95 #endif /* CONFIG_NO_HZ_FULL */ 96 97 #ifdef CONFIG_RCU_NOCB_CPU 98 void rcu_init_nohz(void); 99 #else /* #ifdef CONFIG_RCU_NOCB_CPU */ 100 static inline void rcu_init_nohz(void) { } 101 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 102 103 /** 104 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers 105 * @a: Code that RCU needs to pay attention to. 106 * 107 * RCU read-side critical sections are forbidden in the inner idle loop, 108 * that is, between the rcu_idle_enter() and the rcu_idle_exit() -- RCU 109 * will happily ignore any such read-side critical sections. However, 110 * things like powertop need tracepoints in the inner idle loop. 111 * 112 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) 113 * will tell RCU that it needs to pay attention, invoke its argument 114 * (in this example, calling the do_something_with_RCU() function), 115 * and then tell RCU to go back to ignoring this CPU. It is permissible 116 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is 117 * on the order of a million or so, even on 32-bit systems). It is 118 * not legal to block within RCU_NONIDLE(), nor is it permissible to 119 * transfer control either into or out of RCU_NONIDLE()'s statement. 120 */ 121 #define RCU_NONIDLE(a) \ 122 do { \ 123 rcu_irq_enter_irqson(); \ 124 do { a; } while (0); \ 125 rcu_irq_exit_irqson(); \ 126 } while (0) 127 128 /* 129 * Note a quasi-voluntary context switch for RCU-tasks's benefit. 130 * This is a macro rather than an inline function to avoid #include hell. 131 */ 132 #ifdef CONFIG_TASKS_RCU 133 #define rcu_tasks_qs(t) \ 134 do { \ 135 if (READ_ONCE((t)->rcu_tasks_holdout)) \ 136 WRITE_ONCE((t)->rcu_tasks_holdout, false); \ 137 } while (0) 138 #define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t) 139 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func); 140 void synchronize_rcu_tasks(void); 141 void exit_tasks_rcu_start(void); 142 void exit_tasks_rcu_finish(void); 143 #else /* #ifdef CONFIG_TASKS_RCU */ 144 #define rcu_tasks_qs(t) do { } while (0) 145 #define rcu_note_voluntary_context_switch(t) do { } while (0) 146 #define call_rcu_tasks call_rcu 147 #define synchronize_rcu_tasks synchronize_rcu 148 static inline void exit_tasks_rcu_start(void) { } 149 static inline void exit_tasks_rcu_finish(void) { } 150 #endif /* #else #ifdef CONFIG_TASKS_RCU */ 151 152 /** 153 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU 154 * 155 * This macro resembles cond_resched(), except that it is defined to 156 * report potential quiescent states to RCU-tasks even if the cond_resched() 157 * machinery were to be shut off, as some advocate for PREEMPT kernels. 158 */ 159 #define cond_resched_tasks_rcu_qs() \ 160 do { \ 161 rcu_tasks_qs(current); \ 162 cond_resched(); \ 163 } while (0) 164 165 /* 166 * Infrastructure to implement the synchronize_() primitives in 167 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 168 */ 169 170 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU) 171 #include <linux/rcutree.h> 172 #elif defined(CONFIG_TINY_RCU) 173 #include <linux/rcutiny.h> 174 #else 175 #error "Unknown RCU implementation specified to kernel configuration" 176 #endif 177 178 /* 179 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls 180 * are needed for dynamic initialization and destruction of rcu_head 181 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for 182 * dynamic initialization and destruction of statically allocated rcu_head 183 * structures. However, rcu_head structures allocated dynamically in the 184 * heap don't need any initialization. 185 */ 186 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 187 void init_rcu_head(struct rcu_head *head); 188 void destroy_rcu_head(struct rcu_head *head); 189 void init_rcu_head_on_stack(struct rcu_head *head); 190 void destroy_rcu_head_on_stack(struct rcu_head *head); 191 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 192 static inline void init_rcu_head(struct rcu_head *head) { } 193 static inline void destroy_rcu_head(struct rcu_head *head) { } 194 static inline void init_rcu_head_on_stack(struct rcu_head *head) { } 195 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { } 196 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 197 198 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 199 bool rcu_lockdep_current_cpu_online(void); 200 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 201 static inline bool rcu_lockdep_current_cpu_online(void) { return true; } 202 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 203 204 #ifdef CONFIG_DEBUG_LOCK_ALLOC 205 206 static inline void rcu_lock_acquire(struct lockdep_map *map) 207 { 208 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); 209 } 210 211 static inline void rcu_lock_release(struct lockdep_map *map) 212 { 213 lock_release(map, 1, _THIS_IP_); 214 } 215 216 extern struct lockdep_map rcu_lock_map; 217 extern struct lockdep_map rcu_bh_lock_map; 218 extern struct lockdep_map rcu_sched_lock_map; 219 extern struct lockdep_map rcu_callback_map; 220 int debug_lockdep_rcu_enabled(void); 221 int rcu_read_lock_held(void); 222 int rcu_read_lock_bh_held(void); 223 int rcu_read_lock_sched_held(void); 224 225 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 226 227 # define rcu_lock_acquire(a) do { } while (0) 228 # define rcu_lock_release(a) do { } while (0) 229 230 static inline int rcu_read_lock_held(void) 231 { 232 return 1; 233 } 234 235 static inline int rcu_read_lock_bh_held(void) 236 { 237 return 1; 238 } 239 240 static inline int rcu_read_lock_sched_held(void) 241 { 242 return !preemptible(); 243 } 244 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 245 246 #ifdef CONFIG_PROVE_RCU 247 248 /** 249 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met 250 * @c: condition to check 251 * @s: informative message 252 */ 253 #define RCU_LOCKDEP_WARN(c, s) \ 254 do { \ 255 static bool __section(.data.unlikely) __warned; \ 256 if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \ 257 __warned = true; \ 258 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 259 } \ 260 } while (0) 261 262 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 263 static inline void rcu_preempt_sleep_check(void) 264 { 265 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), 266 "Illegal context switch in RCU read-side critical section"); 267 } 268 #else /* #ifdef CONFIG_PROVE_RCU */ 269 static inline void rcu_preempt_sleep_check(void) { } 270 #endif /* #else #ifdef CONFIG_PROVE_RCU */ 271 272 #define rcu_sleep_check() \ 273 do { \ 274 rcu_preempt_sleep_check(); \ 275 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ 276 "Illegal context switch in RCU-bh read-side critical section"); \ 277 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ 278 "Illegal context switch in RCU-sched read-side critical section"); \ 279 } while (0) 280 281 #else /* #ifdef CONFIG_PROVE_RCU */ 282 283 #define RCU_LOCKDEP_WARN(c, s) do { } while (0) 284 #define rcu_sleep_check() do { } while (0) 285 286 #endif /* #else #ifdef CONFIG_PROVE_RCU */ 287 288 /* 289 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 290 * and rcu_assign_pointer(). Some of these could be folded into their 291 * callers, but they are left separate in order to ease introduction of 292 * multiple pointers markings to match different RCU implementations 293 * (e.g., __srcu), should this make sense in the future. 294 */ 295 296 #ifdef __CHECKER__ 297 #define rcu_check_sparse(p, space) \ 298 ((void)(((typeof(*p) space *)p) == p)) 299 #else /* #ifdef __CHECKER__ */ 300 #define rcu_check_sparse(p, space) 301 #endif /* #else #ifdef __CHECKER__ */ 302 303 #define __rcu_access_pointer(p, space) \ 304 ({ \ 305 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 306 rcu_check_sparse(p, space); \ 307 ((typeof(*p) __force __kernel *)(_________p1)); \ 308 }) 309 #define __rcu_dereference_check(p, c, space) \ 310 ({ \ 311 /* Dependency order vs. p above. */ \ 312 typeof(*p) *________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 313 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ 314 rcu_check_sparse(p, space); \ 315 ((typeof(*p) __force __kernel *)(________p1)); \ 316 }) 317 #define __rcu_dereference_protected(p, c, space) \ 318 ({ \ 319 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ 320 rcu_check_sparse(p, space); \ 321 ((typeof(*p) __force __kernel *)(p)); \ 322 }) 323 #define rcu_dereference_raw(p) \ 324 ({ \ 325 /* Dependency order vs. p above. */ \ 326 typeof(p) ________p1 = READ_ONCE(p); \ 327 ((typeof(*p) __force __kernel *)(________p1)); \ 328 }) 329 330 /** 331 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 332 * @v: The value to statically initialize with. 333 */ 334 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 335 336 /** 337 * rcu_assign_pointer() - assign to RCU-protected pointer 338 * @p: pointer to assign to 339 * @v: value to assign (publish) 340 * 341 * Assigns the specified value to the specified RCU-protected 342 * pointer, ensuring that any concurrent RCU readers will see 343 * any prior initialization. 344 * 345 * Inserts memory barriers on architectures that require them 346 * (which is most of them), and also prevents the compiler from 347 * reordering the code that initializes the structure after the pointer 348 * assignment. More importantly, this call documents which pointers 349 * will be dereferenced by RCU read-side code. 350 * 351 * In some special cases, you may use RCU_INIT_POINTER() instead 352 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 353 * to the fact that it does not constrain either the CPU or the compiler. 354 * That said, using RCU_INIT_POINTER() when you should have used 355 * rcu_assign_pointer() is a very bad thing that results in 356 * impossible-to-diagnose memory corruption. So please be careful. 357 * See the RCU_INIT_POINTER() comment header for details. 358 * 359 * Note that rcu_assign_pointer() evaluates each of its arguments only 360 * once, appearances notwithstanding. One of the "extra" evaluations 361 * is in typeof() and the other visible only to sparse (__CHECKER__), 362 * neither of which actually execute the argument. As with most cpp 363 * macros, this execute-arguments-only-once property is important, so 364 * please be careful when making changes to rcu_assign_pointer() and the 365 * other macros that it invokes. 366 */ 367 #define rcu_assign_pointer(p, v) \ 368 ({ \ 369 uintptr_t _r_a_p__v = (uintptr_t)(v); \ 370 rcu_check_sparse(p, __rcu); \ 371 \ 372 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \ 373 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \ 374 else \ 375 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \ 376 _r_a_p__v; \ 377 }) 378 379 /** 380 * rcu_swap_protected() - swap an RCU and a regular pointer 381 * @rcu_ptr: RCU pointer 382 * @ptr: regular pointer 383 * @c: the conditions under which the dereference will take place 384 * 385 * Perform swap(@rcu_ptr, @ptr) where @rcu_ptr is an RCU-annotated pointer and 386 * @c is the argument that is passed to the rcu_dereference_protected() call 387 * used to read that pointer. 388 */ 389 #define rcu_swap_protected(rcu_ptr, ptr, c) do { \ 390 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \ 391 rcu_assign_pointer((rcu_ptr), (ptr)); \ 392 (ptr) = __tmp; \ 393 } while (0) 394 395 /** 396 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 397 * @p: The pointer to read 398 * 399 * Return the value of the specified RCU-protected pointer, but omit the 400 * lockdep checks for being in an RCU read-side critical section. This is 401 * useful when the value of this pointer is accessed, but the pointer is 402 * not dereferenced, for example, when testing an RCU-protected pointer 403 * against NULL. Although rcu_access_pointer() may also be used in cases 404 * where update-side locks prevent the value of the pointer from changing, 405 * you should instead use rcu_dereference_protected() for this use case. 406 * 407 * It is also permissible to use rcu_access_pointer() when read-side 408 * access to the pointer was removed at least one grace period ago, as 409 * is the case in the context of the RCU callback that is freeing up 410 * the data, or after a synchronize_rcu() returns. This can be useful 411 * when tearing down multi-linked structures after a grace period 412 * has elapsed. 413 */ 414 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) 415 416 /** 417 * rcu_dereference_check() - rcu_dereference with debug checking 418 * @p: The pointer to read, prior to dereferencing 419 * @c: The conditions under which the dereference will take place 420 * 421 * Do an rcu_dereference(), but check that the conditions under which the 422 * dereference will take place are correct. Typically the conditions 423 * indicate the various locking conditions that should be held at that 424 * point. The check should return true if the conditions are satisfied. 425 * An implicit check for being in an RCU read-side critical section 426 * (rcu_read_lock()) is included. 427 * 428 * For example: 429 * 430 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 431 * 432 * could be used to indicate to lockdep that foo->bar may only be dereferenced 433 * if either rcu_read_lock() is held, or that the lock required to replace 434 * the bar struct at foo->bar is held. 435 * 436 * Note that the list of conditions may also include indications of when a lock 437 * need not be held, for example during initialisation or destruction of the 438 * target struct: 439 * 440 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 441 * atomic_read(&foo->usage) == 0); 442 * 443 * Inserts memory barriers on architectures that require them 444 * (currently only the Alpha), prevents the compiler from refetching 445 * (and from merging fetches), and, more importantly, documents exactly 446 * which pointers are protected by RCU and checks that the pointer is 447 * annotated as __rcu. 448 */ 449 #define rcu_dereference_check(p, c) \ 450 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu) 451 452 /** 453 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 454 * @p: The pointer to read, prior to dereferencing 455 * @c: The conditions under which the dereference will take place 456 * 457 * This is the RCU-bh counterpart to rcu_dereference_check(). 458 */ 459 #define rcu_dereference_bh_check(p, c) \ 460 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu) 461 462 /** 463 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 464 * @p: The pointer to read, prior to dereferencing 465 * @c: The conditions under which the dereference will take place 466 * 467 * This is the RCU-sched counterpart to rcu_dereference_check(). 468 */ 469 #define rcu_dereference_sched_check(p, c) \ 470 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \ 471 __rcu) 472 473 /* 474 * The tracing infrastructure traces RCU (we want that), but unfortunately 475 * some of the RCU checks causes tracing to lock up the system. 476 * 477 * The no-tracing version of rcu_dereference_raw() must not call 478 * rcu_read_lock_held(). 479 */ 480 #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu) 481 482 /** 483 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 484 * @p: The pointer to read, prior to dereferencing 485 * @c: The conditions under which the dereference will take place 486 * 487 * Return the value of the specified RCU-protected pointer, but omit 488 * the READ_ONCE(). This is useful in cases where update-side locks 489 * prevent the value of the pointer from changing. Please note that this 490 * primitive does *not* prevent the compiler from repeating this reference 491 * or combining it with other references, so it should not be used without 492 * protection of appropriate locks. 493 * 494 * This function is only for update-side use. Using this function 495 * when protected only by rcu_read_lock() will result in infrequent 496 * but very ugly failures. 497 */ 498 #define rcu_dereference_protected(p, c) \ 499 __rcu_dereference_protected((p), (c), __rcu) 500 501 502 /** 503 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 504 * @p: The pointer to read, prior to dereferencing 505 * 506 * This is a simple wrapper around rcu_dereference_check(). 507 */ 508 #define rcu_dereference(p) rcu_dereference_check(p, 0) 509 510 /** 511 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 512 * @p: The pointer to read, prior to dereferencing 513 * 514 * Makes rcu_dereference_check() do the dirty work. 515 */ 516 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 517 518 /** 519 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 520 * @p: The pointer to read, prior to dereferencing 521 * 522 * Makes rcu_dereference_check() do the dirty work. 523 */ 524 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 525 526 /** 527 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism 528 * @p: The pointer to hand off 529 * 530 * This is simply an identity function, but it documents where a pointer 531 * is handed off from RCU to some other synchronization mechanism, for 532 * example, reference counting or locking. In C11, it would map to 533 * kill_dependency(). It could be used as follows:: 534 * 535 * rcu_read_lock(); 536 * p = rcu_dereference(gp); 537 * long_lived = is_long_lived(p); 538 * if (long_lived) { 539 * if (!atomic_inc_not_zero(p->refcnt)) 540 * long_lived = false; 541 * else 542 * p = rcu_pointer_handoff(p); 543 * } 544 * rcu_read_unlock(); 545 */ 546 #define rcu_pointer_handoff(p) (p) 547 548 /** 549 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 550 * 551 * When synchronize_rcu() is invoked on one CPU while other CPUs 552 * are within RCU read-side critical sections, then the 553 * synchronize_rcu() is guaranteed to block until after all the other 554 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 555 * on one CPU while other CPUs are within RCU read-side critical 556 * sections, invocation of the corresponding RCU callback is deferred 557 * until after the all the other CPUs exit their critical sections. 558 * 559 * Note, however, that RCU callbacks are permitted to run concurrently 560 * with new RCU read-side critical sections. One way that this can happen 561 * is via the following sequence of events: (1) CPU 0 enters an RCU 562 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 563 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 564 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 565 * callback is invoked. This is legal, because the RCU read-side critical 566 * section that was running concurrently with the call_rcu() (and which 567 * therefore might be referencing something that the corresponding RCU 568 * callback would free up) has completed before the corresponding 569 * RCU callback is invoked. 570 * 571 * RCU read-side critical sections may be nested. Any deferred actions 572 * will be deferred until the outermost RCU read-side critical section 573 * completes. 574 * 575 * You can avoid reading and understanding the next paragraph by 576 * following this rule: don't put anything in an rcu_read_lock() RCU 577 * read-side critical section that would block in a !PREEMPT kernel. 578 * But if you want the full story, read on! 579 * 580 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), 581 * it is illegal to block while in an RCU read-side critical section. 582 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT 583 * kernel builds, RCU read-side critical sections may be preempted, 584 * but explicit blocking is illegal. Finally, in preemptible RCU 585 * implementations in real-time (with -rt patchset) kernel builds, RCU 586 * read-side critical sections may be preempted and they may also block, but 587 * only when acquiring spinlocks that are subject to priority inheritance. 588 */ 589 static inline void rcu_read_lock(void) 590 { 591 __rcu_read_lock(); 592 __acquire(RCU); 593 rcu_lock_acquire(&rcu_lock_map); 594 RCU_LOCKDEP_WARN(!rcu_is_watching(), 595 "rcu_read_lock() used illegally while idle"); 596 } 597 598 /* 599 * So where is rcu_write_lock()? It does not exist, as there is no 600 * way for writers to lock out RCU readers. This is a feature, not 601 * a bug -- this property is what provides RCU's performance benefits. 602 * Of course, writers must coordinate with each other. The normal 603 * spinlock primitives work well for this, but any other technique may be 604 * used as well. RCU does not care how the writers keep out of each 605 * others' way, as long as they do so. 606 */ 607 608 /** 609 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 610 * 611 * In most situations, rcu_read_unlock() is immune from deadlock. 612 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() 613 * is responsible for deboosting, which it does via rt_mutex_unlock(). 614 * Unfortunately, this function acquires the scheduler's runqueue and 615 * priority-inheritance spinlocks. This means that deadlock could result 616 * if the caller of rcu_read_unlock() already holds one of these locks or 617 * any lock that is ever acquired while holding them. 618 * 619 * That said, RCU readers are never priority boosted unless they were 620 * preempted. Therefore, one way to avoid deadlock is to make sure 621 * that preemption never happens within any RCU read-side critical 622 * section whose outermost rcu_read_unlock() is called with one of 623 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in 624 * a number of ways, for example, by invoking preempt_disable() before 625 * critical section's outermost rcu_read_lock(). 626 * 627 * Given that the set of locks acquired by rt_mutex_unlock() might change 628 * at any time, a somewhat more future-proofed approach is to make sure 629 * that that preemption never happens within any RCU read-side critical 630 * section whose outermost rcu_read_unlock() is called with irqs disabled. 631 * This approach relies on the fact that rt_mutex_unlock() currently only 632 * acquires irq-disabled locks. 633 * 634 * The second of these two approaches is best in most situations, 635 * however, the first approach can also be useful, at least to those 636 * developers willing to keep abreast of the set of locks acquired by 637 * rt_mutex_unlock(). 638 * 639 * See rcu_read_lock() for more information. 640 */ 641 static inline void rcu_read_unlock(void) 642 { 643 RCU_LOCKDEP_WARN(!rcu_is_watching(), 644 "rcu_read_unlock() used illegally while idle"); 645 __release(RCU); 646 __rcu_read_unlock(); 647 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ 648 } 649 650 /** 651 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 652 * 653 * This is equivalent of rcu_read_lock(), but also disables softirqs. 654 * Note that anything else that disables softirqs can also serve as 655 * an RCU read-side critical section. 656 * 657 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 658 * must occur in the same context, for example, it is illegal to invoke 659 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 660 * was invoked from some other task. 661 */ 662 static inline void rcu_read_lock_bh(void) 663 { 664 local_bh_disable(); 665 __acquire(RCU_BH); 666 rcu_lock_acquire(&rcu_bh_lock_map); 667 RCU_LOCKDEP_WARN(!rcu_is_watching(), 668 "rcu_read_lock_bh() used illegally while idle"); 669 } 670 671 /* 672 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section 673 * 674 * See rcu_read_lock_bh() for more information. 675 */ 676 static inline void rcu_read_unlock_bh(void) 677 { 678 RCU_LOCKDEP_WARN(!rcu_is_watching(), 679 "rcu_read_unlock_bh() used illegally while idle"); 680 rcu_lock_release(&rcu_bh_lock_map); 681 __release(RCU_BH); 682 local_bh_enable(); 683 } 684 685 /** 686 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 687 * 688 * This is equivalent of rcu_read_lock(), but disables preemption. 689 * Read-side critical sections can also be introduced by anything else 690 * that disables preemption, including local_irq_disable() and friends. 691 * 692 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 693 * must occur in the same context, for example, it is illegal to invoke 694 * rcu_read_unlock_sched() from process context if the matching 695 * rcu_read_lock_sched() was invoked from an NMI handler. 696 */ 697 static inline void rcu_read_lock_sched(void) 698 { 699 preempt_disable(); 700 __acquire(RCU_SCHED); 701 rcu_lock_acquire(&rcu_sched_lock_map); 702 RCU_LOCKDEP_WARN(!rcu_is_watching(), 703 "rcu_read_lock_sched() used illegally while idle"); 704 } 705 706 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 707 static inline notrace void rcu_read_lock_sched_notrace(void) 708 { 709 preempt_disable_notrace(); 710 __acquire(RCU_SCHED); 711 } 712 713 /* 714 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section 715 * 716 * See rcu_read_lock_sched for more information. 717 */ 718 static inline void rcu_read_unlock_sched(void) 719 { 720 RCU_LOCKDEP_WARN(!rcu_is_watching(), 721 "rcu_read_unlock_sched() used illegally while idle"); 722 rcu_lock_release(&rcu_sched_lock_map); 723 __release(RCU_SCHED); 724 preempt_enable(); 725 } 726 727 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 728 static inline notrace void rcu_read_unlock_sched_notrace(void) 729 { 730 __release(RCU_SCHED); 731 preempt_enable_notrace(); 732 } 733 734 /** 735 * RCU_INIT_POINTER() - initialize an RCU protected pointer 736 * @p: The pointer to be initialized. 737 * @v: The value to initialized the pointer to. 738 * 739 * Initialize an RCU-protected pointer in special cases where readers 740 * do not need ordering constraints on the CPU or the compiler. These 741 * special cases are: 742 * 743 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or* 744 * 2. The caller has taken whatever steps are required to prevent 745 * RCU readers from concurrently accessing this pointer *or* 746 * 3. The referenced data structure has already been exposed to 747 * readers either at compile time or via rcu_assign_pointer() *and* 748 * 749 * a. You have not made *any* reader-visible changes to 750 * this structure since then *or* 751 * b. It is OK for readers accessing this structure from its 752 * new location to see the old state of the structure. (For 753 * example, the changes were to statistical counters or to 754 * other state where exact synchronization is not required.) 755 * 756 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 757 * result in impossible-to-diagnose memory corruption. As in the structures 758 * will look OK in crash dumps, but any concurrent RCU readers might 759 * see pre-initialized values of the referenced data structure. So 760 * please be very careful how you use RCU_INIT_POINTER()!!! 761 * 762 * If you are creating an RCU-protected linked structure that is accessed 763 * by a single external-to-structure RCU-protected pointer, then you may 764 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 765 * pointers, but you must use rcu_assign_pointer() to initialize the 766 * external-to-structure pointer *after* you have completely initialized 767 * the reader-accessible portions of the linked structure. 768 * 769 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 770 * ordering guarantees for either the CPU or the compiler. 771 */ 772 #define RCU_INIT_POINTER(p, v) \ 773 do { \ 774 rcu_check_sparse(p, __rcu); \ 775 WRITE_ONCE(p, RCU_INITIALIZER(v)); \ 776 } while (0) 777 778 /** 779 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 780 * @p: The pointer to be initialized. 781 * @v: The value to initialized the pointer to. 782 * 783 * GCC-style initialization for an RCU-protected pointer in a structure field. 784 */ 785 #define RCU_POINTER_INITIALIZER(p, v) \ 786 .p = RCU_INITIALIZER(v) 787 788 /* 789 * Does the specified offset indicate that the corresponding rcu_head 790 * structure can be handled by kfree_rcu()? 791 */ 792 #define __is_kfree_rcu_offset(offset) ((offset) < 4096) 793 794 /* 795 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. 796 */ 797 #define __kfree_rcu(head, offset) \ 798 do { \ 799 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \ 800 kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \ 801 } while (0) 802 803 /** 804 * kfree_rcu() - kfree an object after a grace period. 805 * @ptr: pointer to kfree 806 * @rcu_head: the name of the struct rcu_head within the type of @ptr. 807 * 808 * Many rcu callbacks functions just call kfree() on the base structure. 809 * These functions are trivial, but their size adds up, and furthermore 810 * when they are used in a kernel module, that module must invoke the 811 * high-latency rcu_barrier() function at module-unload time. 812 * 813 * The kfree_rcu() function handles this issue. Rather than encoding a 814 * function address in the embedded rcu_head structure, kfree_rcu() instead 815 * encodes the offset of the rcu_head structure within the base structure. 816 * Because the functions are not allowed in the low-order 4096 bytes of 817 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 818 * If the offset is larger than 4095 bytes, a compile-time error will 819 * be generated in __kfree_rcu(). If this error is triggered, you can 820 * either fall back to use of call_rcu() or rearrange the structure to 821 * position the rcu_head structure into the first 4096 bytes. 822 * 823 * Note that the allowable offset might decrease in the future, for example, 824 * to allow something like kmem_cache_free_rcu(). 825 * 826 * The BUILD_BUG_ON check must not involve any function calls, hence the 827 * checks are done in macros here. 828 */ 829 #define kfree_rcu(ptr, rcu_head) \ 830 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head)) 831 832 833 /* 834 * Place this after a lock-acquisition primitive to guarantee that 835 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies 836 * if the UNLOCK and LOCK are executed by the same CPU or if the 837 * UNLOCK and LOCK operate on the same lock variable. 838 */ 839 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE 840 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */ 841 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 842 #define smp_mb__after_unlock_lock() do { } while (0) 843 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 844 845 846 /* Has the specified rcu_head structure been handed to call_rcu()? */ 847 848 /** 849 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu() 850 * @rhp: The rcu_head structure to initialize. 851 * 852 * If you intend to invoke rcu_head_after_call_rcu() to test whether a 853 * given rcu_head structure has already been passed to call_rcu(), then 854 * you must also invoke this rcu_head_init() function on it just after 855 * allocating that structure. Calls to this function must not race with 856 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation. 857 */ 858 static inline void rcu_head_init(struct rcu_head *rhp) 859 { 860 rhp->func = (rcu_callback_t)~0L; 861 } 862 863 /** 864 * rcu_head_after_call_rcu - Has this rcu_head been passed to call_rcu()? 865 * @rhp: The rcu_head structure to test. 866 * @f: The function passed to call_rcu() along with @rhp. 867 * 868 * Returns @true if the @rhp has been passed to call_rcu() with @func, 869 * and @false otherwise. Emits a warning in any other case, including 870 * the case where @rhp has already been invoked after a grace period. 871 * Calls to this function must not race with callback invocation. One way 872 * to avoid such races is to enclose the call to rcu_head_after_call_rcu() 873 * in an RCU read-side critical section that includes a read-side fetch 874 * of the pointer to the structure containing @rhp. 875 */ 876 static inline bool 877 rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f) 878 { 879 rcu_callback_t func = READ_ONCE(rhp->func); 880 881 if (func == f) 882 return true; 883 WARN_ON_ONCE(func != (rcu_callback_t)~0L); 884 return false; 885 } 886 887 #endif /* __LINUX_RCUPDATE_H */ 888