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 PREEMPTION 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) 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, _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 int rcu_read_lock_any_held(void); 225 226 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 227 228 # define rcu_lock_acquire(a) do { } while (0) 229 # define rcu_lock_release(a) do { } while (0) 230 231 static inline int rcu_read_lock_held(void) 232 { 233 return 1; 234 } 235 236 static inline int rcu_read_lock_bh_held(void) 237 { 238 return 1; 239 } 240 241 static inline int rcu_read_lock_sched_held(void) 242 { 243 return !preemptible(); 244 } 245 246 static inline int rcu_read_lock_any_held(void) 247 { 248 return !preemptible(); 249 } 250 251 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 252 253 #ifdef CONFIG_PROVE_RCU 254 255 /** 256 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met 257 * @c: condition to check 258 * @s: informative message 259 */ 260 #define RCU_LOCKDEP_WARN(c, s) \ 261 do { \ 262 static bool __section(.data.unlikely) __warned; \ 263 if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \ 264 __warned = true; \ 265 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 266 } \ 267 } while (0) 268 269 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 270 static inline void rcu_preempt_sleep_check(void) 271 { 272 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), 273 "Illegal context switch in RCU read-side critical section"); 274 } 275 #else /* #ifdef CONFIG_PROVE_RCU */ 276 static inline void rcu_preempt_sleep_check(void) { } 277 #endif /* #else #ifdef CONFIG_PROVE_RCU */ 278 279 #define rcu_sleep_check() \ 280 do { \ 281 rcu_preempt_sleep_check(); \ 282 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ 283 "Illegal context switch in RCU-bh read-side critical section"); \ 284 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ 285 "Illegal context switch in RCU-sched read-side critical section"); \ 286 } while (0) 287 288 #else /* #ifdef CONFIG_PROVE_RCU */ 289 290 #define RCU_LOCKDEP_WARN(c, s) do { } while (0) 291 #define rcu_sleep_check() do { } while (0) 292 293 #endif /* #else #ifdef CONFIG_PROVE_RCU */ 294 295 /* 296 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 297 * and rcu_assign_pointer(). Some of these could be folded into their 298 * callers, but they are left separate in order to ease introduction of 299 * multiple pointers markings to match different RCU implementations 300 * (e.g., __srcu), should this make sense in the future. 301 */ 302 303 #ifdef __CHECKER__ 304 #define rcu_check_sparse(p, space) \ 305 ((void)(((typeof(*p) space *)p) == p)) 306 #else /* #ifdef __CHECKER__ */ 307 #define rcu_check_sparse(p, space) 308 #endif /* #else #ifdef __CHECKER__ */ 309 310 #define __rcu_access_pointer(p, space) \ 311 ({ \ 312 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 313 rcu_check_sparse(p, space); \ 314 ((typeof(*p) __force __kernel *)(_________p1)); \ 315 }) 316 #define __rcu_dereference_check(p, c, space) \ 317 ({ \ 318 /* Dependency order vs. p above. */ \ 319 typeof(*p) *________p1 = (typeof(*p) *__force)READ_ONCE(p); \ 320 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ 321 rcu_check_sparse(p, space); \ 322 ((typeof(*p) __force __kernel *)(________p1)); \ 323 }) 324 #define __rcu_dereference_protected(p, c, space) \ 325 ({ \ 326 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ 327 rcu_check_sparse(p, space); \ 328 ((typeof(*p) __force __kernel *)(p)); \ 329 }) 330 #define rcu_dereference_raw(p) \ 331 ({ \ 332 /* Dependency order vs. p above. */ \ 333 typeof(p) ________p1 = READ_ONCE(p); \ 334 ((typeof(*p) __force __kernel *)(________p1)); \ 335 }) 336 337 /** 338 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 339 * @v: The value to statically initialize with. 340 */ 341 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 342 343 /** 344 * rcu_assign_pointer() - assign to RCU-protected pointer 345 * @p: pointer to assign to 346 * @v: value to assign (publish) 347 * 348 * Assigns the specified value to the specified RCU-protected 349 * pointer, ensuring that any concurrent RCU readers will see 350 * any prior initialization. 351 * 352 * Inserts memory barriers on architectures that require them 353 * (which is most of them), and also prevents the compiler from 354 * reordering the code that initializes the structure after the pointer 355 * assignment. More importantly, this call documents which pointers 356 * will be dereferenced by RCU read-side code. 357 * 358 * In some special cases, you may use RCU_INIT_POINTER() instead 359 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 360 * to the fact that it does not constrain either the CPU or the compiler. 361 * That said, using RCU_INIT_POINTER() when you should have used 362 * rcu_assign_pointer() is a very bad thing that results in 363 * impossible-to-diagnose memory corruption. So please be careful. 364 * See the RCU_INIT_POINTER() comment header for details. 365 * 366 * Note that rcu_assign_pointer() evaluates each of its arguments only 367 * once, appearances notwithstanding. One of the "extra" evaluations 368 * is in typeof() and the other visible only to sparse (__CHECKER__), 369 * neither of which actually execute the argument. As with most cpp 370 * macros, this execute-arguments-only-once property is important, so 371 * please be careful when making changes to rcu_assign_pointer() and the 372 * other macros that it invokes. 373 */ 374 #define rcu_assign_pointer(p, v) \ 375 do { \ 376 uintptr_t _r_a_p__v = (uintptr_t)(v); \ 377 rcu_check_sparse(p, __rcu); \ 378 \ 379 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \ 380 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \ 381 else \ 382 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \ 383 } while (0) 384 385 /** 386 * rcu_replace_pointer() - replace an RCU pointer, returning its old value 387 * @rcu_ptr: RCU pointer, whose old value is returned 388 * @ptr: regular pointer 389 * @c: the lockdep conditions under which the dereference will take place 390 * 391 * Perform a replacement, where @rcu_ptr is an RCU-annotated 392 * pointer and @c is the lockdep argument that is passed to the 393 * rcu_dereference_protected() call used to read that pointer. The old 394 * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr. 395 */ 396 #define rcu_replace_pointer(rcu_ptr, ptr, c) \ 397 ({ \ 398 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \ 399 rcu_assign_pointer((rcu_ptr), (ptr)); \ 400 __tmp; \ 401 }) 402 403 /** 404 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 405 * @p: The pointer to read 406 * 407 * Return the value of the specified RCU-protected pointer, but omit the 408 * lockdep checks for being in an RCU read-side critical section. This is 409 * useful when the value of this pointer is accessed, but the pointer is 410 * not dereferenced, for example, when testing an RCU-protected pointer 411 * against NULL. Although rcu_access_pointer() may also be used in cases 412 * where update-side locks prevent the value of the pointer from changing, 413 * you should instead use rcu_dereference_protected() for this use case. 414 * 415 * It is also permissible to use rcu_access_pointer() when read-side 416 * access to the pointer was removed at least one grace period ago, as 417 * is the case in the context of the RCU callback that is freeing up 418 * the data, or after a synchronize_rcu() returns. This can be useful 419 * when tearing down multi-linked structures after a grace period 420 * has elapsed. 421 */ 422 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu) 423 424 /** 425 * rcu_dereference_check() - rcu_dereference with debug checking 426 * @p: The pointer to read, prior to dereferencing 427 * @c: The conditions under which the dereference will take place 428 * 429 * Do an rcu_dereference(), but check that the conditions under which the 430 * dereference will take place are correct. Typically the conditions 431 * indicate the various locking conditions that should be held at that 432 * point. The check should return true if the conditions are satisfied. 433 * An implicit check for being in an RCU read-side critical section 434 * (rcu_read_lock()) is included. 435 * 436 * For example: 437 * 438 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 439 * 440 * could be used to indicate to lockdep that foo->bar may only be dereferenced 441 * if either rcu_read_lock() is held, or that the lock required to replace 442 * the bar struct at foo->bar is held. 443 * 444 * Note that the list of conditions may also include indications of when a lock 445 * need not be held, for example during initialisation or destruction of the 446 * target struct: 447 * 448 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 449 * atomic_read(&foo->usage) == 0); 450 * 451 * Inserts memory barriers on architectures that require them 452 * (currently only the Alpha), prevents the compiler from refetching 453 * (and from merging fetches), and, more importantly, documents exactly 454 * which pointers are protected by RCU and checks that the pointer is 455 * annotated as __rcu. 456 */ 457 #define rcu_dereference_check(p, c) \ 458 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu) 459 460 /** 461 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 462 * @p: The pointer to read, prior to dereferencing 463 * @c: The conditions under which the dereference will take place 464 * 465 * This is the RCU-bh counterpart to rcu_dereference_check(). 466 */ 467 #define rcu_dereference_bh_check(p, c) \ 468 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu) 469 470 /** 471 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 472 * @p: The pointer to read, prior to dereferencing 473 * @c: The conditions under which the dereference will take place 474 * 475 * This is the RCU-sched counterpart to rcu_dereference_check(). 476 */ 477 #define rcu_dereference_sched_check(p, c) \ 478 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \ 479 __rcu) 480 481 /* 482 * The tracing infrastructure traces RCU (we want that), but unfortunately 483 * some of the RCU checks causes tracing to lock up the system. 484 * 485 * The no-tracing version of rcu_dereference_raw() must not call 486 * rcu_read_lock_held(). 487 */ 488 #define rcu_dereference_raw_check(p) __rcu_dereference_check((p), 1, __rcu) 489 490 /** 491 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 492 * @p: The pointer to read, prior to dereferencing 493 * @c: The conditions under which the dereference will take place 494 * 495 * Return the value of the specified RCU-protected pointer, but omit 496 * the READ_ONCE(). This is useful in cases where update-side locks 497 * prevent the value of the pointer from changing. Please note that this 498 * primitive does *not* prevent the compiler from repeating this reference 499 * or combining it with other references, so it should not be used without 500 * protection of appropriate locks. 501 * 502 * This function is only for update-side use. Using this function 503 * when protected only by rcu_read_lock() will result in infrequent 504 * but very ugly failures. 505 */ 506 #define rcu_dereference_protected(p, c) \ 507 __rcu_dereference_protected((p), (c), __rcu) 508 509 510 /** 511 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 512 * @p: The pointer to read, prior to dereferencing 513 * 514 * This is a simple wrapper around rcu_dereference_check(). 515 */ 516 #define rcu_dereference(p) rcu_dereference_check(p, 0) 517 518 /** 519 * rcu_dereference_bh() - fetch an RCU-bh-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_bh(p) rcu_dereference_bh_check(p, 0) 525 526 /** 527 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 528 * @p: The pointer to read, prior to dereferencing 529 * 530 * Makes rcu_dereference_check() do the dirty work. 531 */ 532 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 533 534 /** 535 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism 536 * @p: The pointer to hand off 537 * 538 * This is simply an identity function, but it documents where a pointer 539 * is handed off from RCU to some other synchronization mechanism, for 540 * example, reference counting or locking. In C11, it would map to 541 * kill_dependency(). It could be used as follows:: 542 * 543 * rcu_read_lock(); 544 * p = rcu_dereference(gp); 545 * long_lived = is_long_lived(p); 546 * if (long_lived) { 547 * if (!atomic_inc_not_zero(p->refcnt)) 548 * long_lived = false; 549 * else 550 * p = rcu_pointer_handoff(p); 551 * } 552 * rcu_read_unlock(); 553 */ 554 #define rcu_pointer_handoff(p) (p) 555 556 /** 557 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 558 * 559 * When synchronize_rcu() is invoked on one CPU while other CPUs 560 * are within RCU read-side critical sections, then the 561 * synchronize_rcu() is guaranteed to block until after all the other 562 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 563 * on one CPU while other CPUs are within RCU read-side critical 564 * sections, invocation of the corresponding RCU callback is deferred 565 * until after the all the other CPUs exit their critical sections. 566 * 567 * Note, however, that RCU callbacks are permitted to run concurrently 568 * with new RCU read-side critical sections. One way that this can happen 569 * is via the following sequence of events: (1) CPU 0 enters an RCU 570 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 571 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 572 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 573 * callback is invoked. This is legal, because the RCU read-side critical 574 * section that was running concurrently with the call_rcu() (and which 575 * therefore might be referencing something that the corresponding RCU 576 * callback would free up) has completed before the corresponding 577 * RCU callback is invoked. 578 * 579 * RCU read-side critical sections may be nested. Any deferred actions 580 * will be deferred until the outermost RCU read-side critical section 581 * completes. 582 * 583 * You can avoid reading and understanding the next paragraph by 584 * following this rule: don't put anything in an rcu_read_lock() RCU 585 * read-side critical section that would block in a !PREEMPTION kernel. 586 * But if you want the full story, read on! 587 * 588 * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU), 589 * it is illegal to block while in an RCU read-side critical section. 590 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION 591 * kernel builds, RCU read-side critical sections may be preempted, 592 * but explicit blocking is illegal. Finally, in preemptible RCU 593 * implementations in real-time (with -rt patchset) kernel builds, RCU 594 * read-side critical sections may be preempted and they may also block, but 595 * only when acquiring spinlocks that are subject to priority inheritance. 596 */ 597 static __always_inline void rcu_read_lock(void) 598 { 599 __rcu_read_lock(); 600 __acquire(RCU); 601 rcu_lock_acquire(&rcu_lock_map); 602 RCU_LOCKDEP_WARN(!rcu_is_watching(), 603 "rcu_read_lock() used illegally while idle"); 604 } 605 606 /* 607 * So where is rcu_write_lock()? It does not exist, as there is no 608 * way for writers to lock out RCU readers. This is a feature, not 609 * a bug -- this property is what provides RCU's performance benefits. 610 * Of course, writers must coordinate with each other. The normal 611 * spinlock primitives work well for this, but any other technique may be 612 * used as well. RCU does not care how the writers keep out of each 613 * others' way, as long as they do so. 614 */ 615 616 /** 617 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 618 * 619 * In most situations, rcu_read_unlock() is immune from deadlock. 620 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() 621 * is responsible for deboosting, which it does via rt_mutex_unlock(). 622 * Unfortunately, this function acquires the scheduler's runqueue and 623 * priority-inheritance spinlocks. This means that deadlock could result 624 * if the caller of rcu_read_unlock() already holds one of these locks or 625 * any lock that is ever acquired while holding them. 626 * 627 * That said, RCU readers are never priority boosted unless they were 628 * preempted. Therefore, one way to avoid deadlock is to make sure 629 * that preemption never happens within any RCU read-side critical 630 * section whose outermost rcu_read_unlock() is called with one of 631 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in 632 * a number of ways, for example, by invoking preempt_disable() before 633 * critical section's outermost rcu_read_lock(). 634 * 635 * Given that the set of locks acquired by rt_mutex_unlock() might change 636 * at any time, a somewhat more future-proofed approach is to make sure 637 * that that preemption never happens within any RCU read-side critical 638 * section whose outermost rcu_read_unlock() is called with irqs disabled. 639 * This approach relies on the fact that rt_mutex_unlock() currently only 640 * acquires irq-disabled locks. 641 * 642 * The second of these two approaches is best in most situations, 643 * however, the first approach can also be useful, at least to those 644 * developers willing to keep abreast of the set of locks acquired by 645 * rt_mutex_unlock(). 646 * 647 * See rcu_read_lock() for more information. 648 */ 649 static inline void rcu_read_unlock(void) 650 { 651 RCU_LOCKDEP_WARN(!rcu_is_watching(), 652 "rcu_read_unlock() used illegally while idle"); 653 __release(RCU); 654 __rcu_read_unlock(); 655 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ 656 } 657 658 /** 659 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 660 * 661 * This is equivalent of rcu_read_lock(), but also disables softirqs. 662 * Note that anything else that disables softirqs can also serve as 663 * an RCU read-side critical section. 664 * 665 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 666 * must occur in the same context, for example, it is illegal to invoke 667 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 668 * was invoked from some other task. 669 */ 670 static inline void rcu_read_lock_bh(void) 671 { 672 local_bh_disable(); 673 __acquire(RCU_BH); 674 rcu_lock_acquire(&rcu_bh_lock_map); 675 RCU_LOCKDEP_WARN(!rcu_is_watching(), 676 "rcu_read_lock_bh() used illegally while idle"); 677 } 678 679 /* 680 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section 681 * 682 * See rcu_read_lock_bh() for more information. 683 */ 684 static inline void rcu_read_unlock_bh(void) 685 { 686 RCU_LOCKDEP_WARN(!rcu_is_watching(), 687 "rcu_read_unlock_bh() used illegally while idle"); 688 rcu_lock_release(&rcu_bh_lock_map); 689 __release(RCU_BH); 690 local_bh_enable(); 691 } 692 693 /** 694 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 695 * 696 * This is equivalent of rcu_read_lock(), but disables preemption. 697 * Read-side critical sections can also be introduced by anything else 698 * that disables preemption, including local_irq_disable() and friends. 699 * 700 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 701 * must occur in the same context, for example, it is illegal to invoke 702 * rcu_read_unlock_sched() from process context if the matching 703 * rcu_read_lock_sched() was invoked from an NMI handler. 704 */ 705 static inline void rcu_read_lock_sched(void) 706 { 707 preempt_disable(); 708 __acquire(RCU_SCHED); 709 rcu_lock_acquire(&rcu_sched_lock_map); 710 RCU_LOCKDEP_WARN(!rcu_is_watching(), 711 "rcu_read_lock_sched() used illegally while idle"); 712 } 713 714 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 715 static inline notrace void rcu_read_lock_sched_notrace(void) 716 { 717 preempt_disable_notrace(); 718 __acquire(RCU_SCHED); 719 } 720 721 /* 722 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section 723 * 724 * See rcu_read_lock_sched for more information. 725 */ 726 static inline void rcu_read_unlock_sched(void) 727 { 728 RCU_LOCKDEP_WARN(!rcu_is_watching(), 729 "rcu_read_unlock_sched() used illegally while idle"); 730 rcu_lock_release(&rcu_sched_lock_map); 731 __release(RCU_SCHED); 732 preempt_enable(); 733 } 734 735 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 736 static inline notrace void rcu_read_unlock_sched_notrace(void) 737 { 738 __release(RCU_SCHED); 739 preempt_enable_notrace(); 740 } 741 742 /** 743 * RCU_INIT_POINTER() - initialize an RCU protected pointer 744 * @p: The pointer to be initialized. 745 * @v: The value to initialized the pointer to. 746 * 747 * Initialize an RCU-protected pointer in special cases where readers 748 * do not need ordering constraints on the CPU or the compiler. These 749 * special cases are: 750 * 751 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or* 752 * 2. The caller has taken whatever steps are required to prevent 753 * RCU readers from concurrently accessing this pointer *or* 754 * 3. The referenced data structure has already been exposed to 755 * readers either at compile time or via rcu_assign_pointer() *and* 756 * 757 * a. You have not made *any* reader-visible changes to 758 * this structure since then *or* 759 * b. It is OK for readers accessing this structure from its 760 * new location to see the old state of the structure. (For 761 * example, the changes were to statistical counters or to 762 * other state where exact synchronization is not required.) 763 * 764 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 765 * result in impossible-to-diagnose memory corruption. As in the structures 766 * will look OK in crash dumps, but any concurrent RCU readers might 767 * see pre-initialized values of the referenced data structure. So 768 * please be very careful how you use RCU_INIT_POINTER()!!! 769 * 770 * If you are creating an RCU-protected linked structure that is accessed 771 * by a single external-to-structure RCU-protected pointer, then you may 772 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 773 * pointers, but you must use rcu_assign_pointer() to initialize the 774 * external-to-structure pointer *after* you have completely initialized 775 * the reader-accessible portions of the linked structure. 776 * 777 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 778 * ordering guarantees for either the CPU or the compiler. 779 */ 780 #define RCU_INIT_POINTER(p, v) \ 781 do { \ 782 rcu_check_sparse(p, __rcu); \ 783 WRITE_ONCE(p, RCU_INITIALIZER(v)); \ 784 } while (0) 785 786 /** 787 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 788 * @p: The pointer to be initialized. 789 * @v: The value to initialized the pointer to. 790 * 791 * GCC-style initialization for an RCU-protected pointer in a structure field. 792 */ 793 #define RCU_POINTER_INITIALIZER(p, v) \ 794 .p = RCU_INITIALIZER(v) 795 796 /* 797 * Does the specified offset indicate that the corresponding rcu_head 798 * structure can be handled by kfree_rcu()? 799 */ 800 #define __is_kfree_rcu_offset(offset) ((offset) < 4096) 801 802 /* 803 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain. 804 */ 805 #define __kfree_rcu(head, offset) \ 806 do { \ 807 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \ 808 kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \ 809 } while (0) 810 811 /** 812 * kfree_rcu() - kfree an object after a grace period. 813 * @ptr: pointer to kfree 814 * @rhf: the name of the struct rcu_head within the type of @ptr. 815 * 816 * Many rcu callbacks functions just call kfree() on the base structure. 817 * These functions are trivial, but their size adds up, and furthermore 818 * when they are used in a kernel module, that module must invoke the 819 * high-latency rcu_barrier() function at module-unload time. 820 * 821 * The kfree_rcu() function handles this issue. Rather than encoding a 822 * function address in the embedded rcu_head structure, kfree_rcu() instead 823 * encodes the offset of the rcu_head structure within the base structure. 824 * Because the functions are not allowed in the low-order 4096 bytes of 825 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 826 * If the offset is larger than 4095 bytes, a compile-time error will 827 * be generated in __kfree_rcu(). If this error is triggered, you can 828 * either fall back to use of call_rcu() or rearrange the structure to 829 * position the rcu_head structure into the first 4096 bytes. 830 * 831 * Note that the allowable offset might decrease in the future, for example, 832 * to allow something like kmem_cache_free_rcu(). 833 * 834 * The BUILD_BUG_ON check must not involve any function calls, hence the 835 * checks are done in macros here. 836 */ 837 #define kfree_rcu(ptr, rhf) \ 838 do { \ 839 typeof (ptr) ___p = (ptr); \ 840 \ 841 if (___p) \ 842 __kfree_rcu(&((___p)->rhf), offsetof(typeof(*(ptr)), rhf)); \ 843 } while (0) 844 845 /* 846 * Place this after a lock-acquisition primitive to guarantee that 847 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies 848 * if the UNLOCK and LOCK are executed by the same CPU or if the 849 * UNLOCK and LOCK operate on the same lock variable. 850 */ 851 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE 852 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */ 853 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 854 #define smp_mb__after_unlock_lock() do { } while (0) 855 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 856 857 858 /* Has the specified rcu_head structure been handed to call_rcu()? */ 859 860 /** 861 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu() 862 * @rhp: The rcu_head structure to initialize. 863 * 864 * If you intend to invoke rcu_head_after_call_rcu() to test whether a 865 * given rcu_head structure has already been passed to call_rcu(), then 866 * you must also invoke this rcu_head_init() function on it just after 867 * allocating that structure. Calls to this function must not race with 868 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation. 869 */ 870 static inline void rcu_head_init(struct rcu_head *rhp) 871 { 872 rhp->func = (rcu_callback_t)~0L; 873 } 874 875 /** 876 * rcu_head_after_call_rcu - Has this rcu_head been passed to call_rcu()? 877 * @rhp: The rcu_head structure to test. 878 * @f: The function passed to call_rcu() along with @rhp. 879 * 880 * Returns @true if the @rhp has been passed to call_rcu() with @func, 881 * and @false otherwise. Emits a warning in any other case, including 882 * the case where @rhp has already been invoked after a grace period. 883 * Calls to this function must not race with callback invocation. One way 884 * to avoid such races is to enclose the call to rcu_head_after_call_rcu() 885 * in an RCU read-side critical section that includes a read-side fetch 886 * of the pointer to the structure containing @rhp. 887 */ 888 static inline bool 889 rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f) 890 { 891 rcu_callback_t func = READ_ONCE(rhp->func); 892 893 if (func == f) 894 return true; 895 WARN_ON_ONCE(func != (rcu_callback_t)~0L); 896 return false; 897 } 898 899 /* kernel/ksysfs.c definitions */ 900 extern int rcu_expedited; 901 extern int rcu_normal; 902 903 #endif /* __LINUX_RCUPDATE_H */ 904