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 #include <linux/context_tracking_irq.h> 33 34 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) 35 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b)) 36 #define ulong2long(a) (*(long *)(&(a))) 37 #define USHORT_CMP_GE(a, b) (USHRT_MAX / 2 >= (unsigned short)((a) - (b))) 38 #define USHORT_CMP_LT(a, b) (USHRT_MAX / 2 < (unsigned short)((a) - (b))) 39 40 /* Exported common interfaces */ 41 void call_rcu(struct rcu_head *head, rcu_callback_t func); 42 void rcu_barrier_tasks(void); 43 void rcu_barrier_tasks_rude(void); 44 void synchronize_rcu(void); 45 46 struct rcu_gp_oldstate; 47 unsigned long get_completed_synchronize_rcu(void); 48 void get_completed_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp); 49 50 // Maximum number of unsigned long values corresponding to 51 // not-yet-completed RCU grace periods. 52 #define NUM_ACTIVE_RCU_POLL_OLDSTATE 2 53 54 /** 55 * same_state_synchronize_rcu - Are two old-state values identical? 56 * @oldstate1: First old-state value. 57 * @oldstate2: Second old-state value. 58 * 59 * The two old-state values must have been obtained from either 60 * get_state_synchronize_rcu(), start_poll_synchronize_rcu(), or 61 * get_completed_synchronize_rcu(). Returns @true if the two values are 62 * identical and @false otherwise. This allows structures whose lifetimes 63 * are tracked by old-state values to push these values to a list header, 64 * allowing those structures to be slightly smaller. 65 */ 66 static inline bool same_state_synchronize_rcu(unsigned long oldstate1, unsigned long oldstate2) 67 { 68 return oldstate1 == oldstate2; 69 } 70 71 #ifdef CONFIG_PREEMPT_RCU 72 73 void __rcu_read_lock(void); 74 void __rcu_read_unlock(void); 75 76 /* 77 * Defined as a macro as it is a very low level header included from 78 * areas that don't even know about current. This gives the rcu_read_lock() 79 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other 80 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable. 81 */ 82 #define rcu_preempt_depth() READ_ONCE(current->rcu_read_lock_nesting) 83 84 #else /* #ifdef CONFIG_PREEMPT_RCU */ 85 86 #ifdef CONFIG_TINY_RCU 87 #define rcu_read_unlock_strict() do { } while (0) 88 #else 89 void rcu_read_unlock_strict(void); 90 #endif 91 92 static inline void __rcu_read_lock(void) 93 { 94 preempt_disable(); 95 } 96 97 static inline void __rcu_read_unlock(void) 98 { 99 preempt_enable(); 100 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD)) 101 rcu_read_unlock_strict(); 102 } 103 104 static inline int rcu_preempt_depth(void) 105 { 106 return 0; 107 } 108 109 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 110 111 #ifdef CONFIG_RCU_LAZY 112 void call_rcu_hurry(struct rcu_head *head, rcu_callback_t func); 113 #else 114 static inline void call_rcu_hurry(struct rcu_head *head, rcu_callback_t func) 115 { 116 call_rcu(head, func); 117 } 118 #endif 119 120 /* Internal to kernel */ 121 void rcu_init(void); 122 extern int rcu_scheduler_active; 123 void rcu_sched_clock_irq(int user); 124 void rcu_report_dead(unsigned int cpu); 125 void rcutree_migrate_callbacks(int cpu); 126 127 #ifdef CONFIG_TASKS_RCU_GENERIC 128 void rcu_init_tasks_generic(void); 129 #else 130 static inline void rcu_init_tasks_generic(void) { } 131 #endif 132 133 #ifdef CONFIG_RCU_STALL_COMMON 134 void rcu_sysrq_start(void); 135 void rcu_sysrq_end(void); 136 #else /* #ifdef CONFIG_RCU_STALL_COMMON */ 137 static inline void rcu_sysrq_start(void) { } 138 static inline void rcu_sysrq_end(void) { } 139 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */ 140 141 #if defined(CONFIG_NO_HZ_FULL) && (!defined(CONFIG_GENERIC_ENTRY) || !defined(CONFIG_KVM_XFER_TO_GUEST_WORK)) 142 void rcu_irq_work_resched(void); 143 #else 144 static inline void rcu_irq_work_resched(void) { } 145 #endif 146 147 #ifdef CONFIG_RCU_NOCB_CPU 148 void rcu_init_nohz(void); 149 int rcu_nocb_cpu_offload(int cpu); 150 int rcu_nocb_cpu_deoffload(int cpu); 151 void rcu_nocb_flush_deferred_wakeup(void); 152 #else /* #ifdef CONFIG_RCU_NOCB_CPU */ 153 static inline void rcu_init_nohz(void) { } 154 static inline int rcu_nocb_cpu_offload(int cpu) { return -EINVAL; } 155 static inline int rcu_nocb_cpu_deoffload(int cpu) { return 0; } 156 static inline void rcu_nocb_flush_deferred_wakeup(void) { } 157 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ 158 159 /** 160 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers 161 * @a: Code that RCU needs to pay attention to. 162 * 163 * RCU read-side critical sections are forbidden in the inner idle loop, 164 * that is, between the ct_idle_enter() and the ct_idle_exit() -- RCU 165 * will happily ignore any such read-side critical sections. However, 166 * things like powertop need tracepoints in the inner idle loop. 167 * 168 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU()) 169 * will tell RCU that it needs to pay attention, invoke its argument 170 * (in this example, calling the do_something_with_RCU() function), 171 * and then tell RCU to go back to ignoring this CPU. It is permissible 172 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is 173 * on the order of a million or so, even on 32-bit systems). It is 174 * not legal to block within RCU_NONIDLE(), nor is it permissible to 175 * transfer control either into or out of RCU_NONIDLE()'s statement. 176 */ 177 #define RCU_NONIDLE(a) \ 178 do { \ 179 ct_irq_enter_irqson(); \ 180 do { a; } while (0); \ 181 ct_irq_exit_irqson(); \ 182 } while (0) 183 184 /* 185 * Note a quasi-voluntary context switch for RCU-tasks's benefit. 186 * This is a macro rather than an inline function to avoid #include hell. 187 */ 188 #ifdef CONFIG_TASKS_RCU_GENERIC 189 190 # ifdef CONFIG_TASKS_RCU 191 # define rcu_tasks_classic_qs(t, preempt) \ 192 do { \ 193 if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout)) \ 194 WRITE_ONCE((t)->rcu_tasks_holdout, false); \ 195 } while (0) 196 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func); 197 void synchronize_rcu_tasks(void); 198 # else 199 # define rcu_tasks_classic_qs(t, preempt) do { } while (0) 200 # define call_rcu_tasks call_rcu 201 # define synchronize_rcu_tasks synchronize_rcu 202 # endif 203 204 # ifdef CONFIG_TASKS_TRACE_RCU 205 // Bits for ->trc_reader_special.b.need_qs field. 206 #define TRC_NEED_QS 0x1 // Task needs a quiescent state. 207 #define TRC_NEED_QS_CHECKED 0x2 // Task has been checked for needing quiescent state. 208 209 u8 rcu_trc_cmpxchg_need_qs(struct task_struct *t, u8 old, u8 new); 210 void rcu_tasks_trace_qs_blkd(struct task_struct *t); 211 212 # define rcu_tasks_trace_qs(t) \ 213 do { \ 214 int ___rttq_nesting = READ_ONCE((t)->trc_reader_nesting); \ 215 \ 216 if (likely(!READ_ONCE((t)->trc_reader_special.b.need_qs)) && \ 217 likely(!___rttq_nesting)) { \ 218 rcu_trc_cmpxchg_need_qs((t), 0, TRC_NEED_QS_CHECKED); \ 219 } else if (___rttq_nesting && ___rttq_nesting != INT_MIN && \ 220 !READ_ONCE((t)->trc_reader_special.b.blocked)) { \ 221 rcu_tasks_trace_qs_blkd(t); \ 222 } \ 223 } while (0) 224 # else 225 # define rcu_tasks_trace_qs(t) do { } while (0) 226 # endif 227 228 #define rcu_tasks_qs(t, preempt) \ 229 do { \ 230 rcu_tasks_classic_qs((t), (preempt)); \ 231 rcu_tasks_trace_qs(t); \ 232 } while (0) 233 234 # ifdef CONFIG_TASKS_RUDE_RCU 235 void call_rcu_tasks_rude(struct rcu_head *head, rcu_callback_t func); 236 void synchronize_rcu_tasks_rude(void); 237 # endif 238 239 #define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false) 240 void exit_tasks_rcu_start(void); 241 void exit_tasks_rcu_finish(void); 242 #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */ 243 #define rcu_tasks_classic_qs(t, preempt) do { } while (0) 244 #define rcu_tasks_qs(t, preempt) do { } while (0) 245 #define rcu_note_voluntary_context_switch(t) do { } while (0) 246 #define call_rcu_tasks call_rcu 247 #define synchronize_rcu_tasks synchronize_rcu 248 static inline void exit_tasks_rcu_start(void) { } 249 static inline void exit_tasks_rcu_finish(void) { } 250 #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */ 251 252 /** 253 * rcu_trace_implies_rcu_gp - does an RCU Tasks Trace grace period imply an RCU grace period? 254 * 255 * As an accident of implementation, an RCU Tasks Trace grace period also 256 * acts as an RCU grace period. However, this could change at any time. 257 * Code relying on this accident must call this function to verify that 258 * this accident is still happening. 259 * 260 * You have been warned! 261 */ 262 static inline bool rcu_trace_implies_rcu_gp(void) { return true; } 263 264 /** 265 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU 266 * 267 * This macro resembles cond_resched(), except that it is defined to 268 * report potential quiescent states to RCU-tasks even if the cond_resched() 269 * machinery were to be shut off, as some advocate for PREEMPTION kernels. 270 */ 271 #define cond_resched_tasks_rcu_qs() \ 272 do { \ 273 rcu_tasks_qs(current, false); \ 274 cond_resched(); \ 275 } while (0) 276 277 /* 278 * Infrastructure to implement the synchronize_() primitives in 279 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 280 */ 281 282 #if defined(CONFIG_TREE_RCU) 283 #include <linux/rcutree.h> 284 #elif defined(CONFIG_TINY_RCU) 285 #include <linux/rcutiny.h> 286 #else 287 #error "Unknown RCU implementation specified to kernel configuration" 288 #endif 289 290 /* 291 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls 292 * are needed for dynamic initialization and destruction of rcu_head 293 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for 294 * dynamic initialization and destruction of statically allocated rcu_head 295 * structures. However, rcu_head structures allocated dynamically in the 296 * heap don't need any initialization. 297 */ 298 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 299 void init_rcu_head(struct rcu_head *head); 300 void destroy_rcu_head(struct rcu_head *head); 301 void init_rcu_head_on_stack(struct rcu_head *head); 302 void destroy_rcu_head_on_stack(struct rcu_head *head); 303 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 304 static inline void init_rcu_head(struct rcu_head *head) { } 305 static inline void destroy_rcu_head(struct rcu_head *head) { } 306 static inline void init_rcu_head_on_stack(struct rcu_head *head) { } 307 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { } 308 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 309 310 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 311 bool rcu_lockdep_current_cpu_online(void); 312 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 313 static inline bool rcu_lockdep_current_cpu_online(void) { return true; } 314 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 315 316 extern struct lockdep_map rcu_lock_map; 317 extern struct lockdep_map rcu_bh_lock_map; 318 extern struct lockdep_map rcu_sched_lock_map; 319 extern struct lockdep_map rcu_callback_map; 320 321 #ifdef CONFIG_DEBUG_LOCK_ALLOC 322 323 static inline void rcu_lock_acquire(struct lockdep_map *map) 324 { 325 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); 326 } 327 328 static inline void rcu_lock_release(struct lockdep_map *map) 329 { 330 lock_release(map, _THIS_IP_); 331 } 332 333 int debug_lockdep_rcu_enabled(void); 334 int rcu_read_lock_held(void); 335 int rcu_read_lock_bh_held(void); 336 int rcu_read_lock_sched_held(void); 337 int rcu_read_lock_any_held(void); 338 339 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 340 341 # define rcu_lock_acquire(a) do { } while (0) 342 # define rcu_lock_release(a) do { } while (0) 343 344 static inline int rcu_read_lock_held(void) 345 { 346 return 1; 347 } 348 349 static inline int rcu_read_lock_bh_held(void) 350 { 351 return 1; 352 } 353 354 static inline int rcu_read_lock_sched_held(void) 355 { 356 return !preemptible(); 357 } 358 359 static inline int rcu_read_lock_any_held(void) 360 { 361 return !preemptible(); 362 } 363 364 static inline int debug_lockdep_rcu_enabled(void) 365 { 366 return 0; 367 } 368 369 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 370 371 #ifdef CONFIG_PROVE_RCU 372 373 /** 374 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met 375 * @c: condition to check 376 * @s: informative message 377 */ 378 #define RCU_LOCKDEP_WARN(c, s) \ 379 do { \ 380 static bool __section(".data.unlikely") __warned; \ 381 if ((c) && debug_lockdep_rcu_enabled() && !__warned) { \ 382 __warned = true; \ 383 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 384 } \ 385 } while (0) 386 387 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 388 static inline void rcu_preempt_sleep_check(void) 389 { 390 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), 391 "Illegal context switch in RCU read-side critical section"); 392 } 393 #else /* #ifdef CONFIG_PROVE_RCU */ 394 static inline void rcu_preempt_sleep_check(void) { } 395 #endif /* #else #ifdef CONFIG_PROVE_RCU */ 396 397 #define rcu_sleep_check() \ 398 do { \ 399 rcu_preempt_sleep_check(); \ 400 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) \ 401 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ 402 "Illegal context switch in RCU-bh read-side critical section"); \ 403 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ 404 "Illegal context switch in RCU-sched read-side critical section"); \ 405 } while (0) 406 407 #else /* #ifdef CONFIG_PROVE_RCU */ 408 409 #define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c)) 410 #define rcu_sleep_check() do { } while (0) 411 412 #endif /* #else #ifdef CONFIG_PROVE_RCU */ 413 414 /* 415 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 416 * and rcu_assign_pointer(). Some of these could be folded into their 417 * callers, but they are left separate in order to ease introduction of 418 * multiple pointers markings to match different RCU implementations 419 * (e.g., __srcu), should this make sense in the future. 420 */ 421 422 #ifdef __CHECKER__ 423 #define rcu_check_sparse(p, space) \ 424 ((void)(((typeof(*p) space *)p) == p)) 425 #else /* #ifdef __CHECKER__ */ 426 #define rcu_check_sparse(p, space) 427 #endif /* #else #ifdef __CHECKER__ */ 428 429 #define __unrcu_pointer(p, local) \ 430 ({ \ 431 typeof(*p) *local = (typeof(*p) *__force)(p); \ 432 rcu_check_sparse(p, __rcu); \ 433 ((typeof(*p) __force __kernel *)(local)); \ 434 }) 435 /** 436 * unrcu_pointer - mark a pointer as not being RCU protected 437 * @p: pointer needing to lose its __rcu property 438 * 439 * Converts @p from an __rcu pointer to a __kernel pointer. 440 * This allows an __rcu pointer to be used with xchg() and friends. 441 */ 442 #define unrcu_pointer(p) __unrcu_pointer(p, __UNIQUE_ID(rcu)) 443 444 #define __rcu_access_pointer(p, local, space) \ 445 ({ \ 446 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \ 447 rcu_check_sparse(p, space); \ 448 ((typeof(*p) __force __kernel *)(local)); \ 449 }) 450 #define __rcu_dereference_check(p, local, c, space) \ 451 ({ \ 452 /* Dependency order vs. p above. */ \ 453 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \ 454 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ 455 rcu_check_sparse(p, space); \ 456 ((typeof(*p) __force __kernel *)(local)); \ 457 }) 458 #define __rcu_dereference_protected(p, local, c, space) \ 459 ({ \ 460 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ 461 rcu_check_sparse(p, space); \ 462 ((typeof(*p) __force __kernel *)(p)); \ 463 }) 464 #define __rcu_dereference_raw(p, local) \ 465 ({ \ 466 /* Dependency order vs. p above. */ \ 467 typeof(p) local = READ_ONCE(p); \ 468 ((typeof(*p) __force __kernel *)(local)); \ 469 }) 470 #define rcu_dereference_raw(p) __rcu_dereference_raw(p, __UNIQUE_ID(rcu)) 471 472 /** 473 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 474 * @v: The value to statically initialize with. 475 */ 476 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 477 478 /** 479 * rcu_assign_pointer() - assign to RCU-protected pointer 480 * @p: pointer to assign to 481 * @v: value to assign (publish) 482 * 483 * Assigns the specified value to the specified RCU-protected 484 * pointer, ensuring that any concurrent RCU readers will see 485 * any prior initialization. 486 * 487 * Inserts memory barriers on architectures that require them 488 * (which is most of them), and also prevents the compiler from 489 * reordering the code that initializes the structure after the pointer 490 * assignment. More importantly, this call documents which pointers 491 * will be dereferenced by RCU read-side code. 492 * 493 * In some special cases, you may use RCU_INIT_POINTER() instead 494 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 495 * to the fact that it does not constrain either the CPU or the compiler. 496 * That said, using RCU_INIT_POINTER() when you should have used 497 * rcu_assign_pointer() is a very bad thing that results in 498 * impossible-to-diagnose memory corruption. So please be careful. 499 * See the RCU_INIT_POINTER() comment header for details. 500 * 501 * Note that rcu_assign_pointer() evaluates each of its arguments only 502 * once, appearances notwithstanding. One of the "extra" evaluations 503 * is in typeof() and the other visible only to sparse (__CHECKER__), 504 * neither of which actually execute the argument. As with most cpp 505 * macros, this execute-arguments-only-once property is important, so 506 * please be careful when making changes to rcu_assign_pointer() and the 507 * other macros that it invokes. 508 */ 509 #define rcu_assign_pointer(p, v) \ 510 do { \ 511 uintptr_t _r_a_p__v = (uintptr_t)(v); \ 512 rcu_check_sparse(p, __rcu); \ 513 \ 514 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \ 515 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \ 516 else \ 517 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \ 518 } while (0) 519 520 /** 521 * rcu_replace_pointer() - replace an RCU pointer, returning its old value 522 * @rcu_ptr: RCU pointer, whose old value is returned 523 * @ptr: regular pointer 524 * @c: the lockdep conditions under which the dereference will take place 525 * 526 * Perform a replacement, where @rcu_ptr is an RCU-annotated 527 * pointer and @c is the lockdep argument that is passed to the 528 * rcu_dereference_protected() call used to read that pointer. The old 529 * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr. 530 */ 531 #define rcu_replace_pointer(rcu_ptr, ptr, c) \ 532 ({ \ 533 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \ 534 rcu_assign_pointer((rcu_ptr), (ptr)); \ 535 __tmp; \ 536 }) 537 538 /** 539 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 540 * @p: The pointer to read 541 * 542 * Return the value of the specified RCU-protected pointer, but omit the 543 * lockdep checks for being in an RCU read-side critical section. This is 544 * useful when the value of this pointer is accessed, but the pointer is 545 * not dereferenced, for example, when testing an RCU-protected pointer 546 * against NULL. Although rcu_access_pointer() may also be used in cases 547 * where update-side locks prevent the value of the pointer from changing, 548 * you should instead use rcu_dereference_protected() for this use case. 549 * Within an RCU read-side critical section, there is little reason to 550 * use rcu_access_pointer(). 551 * 552 * It is usually best to test the rcu_access_pointer() return value 553 * directly in order to avoid accidental dereferences being introduced 554 * by later inattentive changes. In other words, assigning the 555 * rcu_access_pointer() return value to a local variable results in an 556 * accident waiting to happen. 557 * 558 * It is also permissible to use rcu_access_pointer() when read-side 559 * access to the pointer was removed at least one grace period ago, as is 560 * the case in the context of the RCU callback that is freeing up the data, 561 * or after a synchronize_rcu() returns. This can be useful when tearing 562 * down multi-linked structures after a grace period has elapsed. However, 563 * rcu_dereference_protected() is normally preferred for this use case. 564 */ 565 #define rcu_access_pointer(p) __rcu_access_pointer((p), __UNIQUE_ID(rcu), __rcu) 566 567 /** 568 * rcu_dereference_check() - rcu_dereference with debug checking 569 * @p: The pointer to read, prior to dereferencing 570 * @c: The conditions under which the dereference will take place 571 * 572 * Do an rcu_dereference(), but check that the conditions under which the 573 * dereference will take place are correct. Typically the conditions 574 * indicate the various locking conditions that should be held at that 575 * point. The check should return true if the conditions are satisfied. 576 * An implicit check for being in an RCU read-side critical section 577 * (rcu_read_lock()) is included. 578 * 579 * For example: 580 * 581 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 582 * 583 * could be used to indicate to lockdep that foo->bar may only be dereferenced 584 * if either rcu_read_lock() is held, or that the lock required to replace 585 * the bar struct at foo->bar is held. 586 * 587 * Note that the list of conditions may also include indications of when a lock 588 * need not be held, for example during initialisation or destruction of the 589 * target struct: 590 * 591 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 592 * atomic_read(&foo->usage) == 0); 593 * 594 * Inserts memory barriers on architectures that require them 595 * (currently only the Alpha), prevents the compiler from refetching 596 * (and from merging fetches), and, more importantly, documents exactly 597 * which pointers are protected by RCU and checks that the pointer is 598 * annotated as __rcu. 599 */ 600 #define rcu_dereference_check(p, c) \ 601 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \ 602 (c) || rcu_read_lock_held(), __rcu) 603 604 /** 605 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 606 * @p: The pointer to read, prior to dereferencing 607 * @c: The conditions under which the dereference will take place 608 * 609 * This is the RCU-bh counterpart to rcu_dereference_check(). However, 610 * please note that starting in v5.0 kernels, vanilla RCU grace periods 611 * wait for local_bh_disable() regions of code in addition to regions of 612 * code demarked by rcu_read_lock() and rcu_read_unlock(). This means 613 * that synchronize_rcu(), call_rcu, and friends all take not only 614 * rcu_read_lock() but also rcu_read_lock_bh() into account. 615 */ 616 #define rcu_dereference_bh_check(p, c) \ 617 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \ 618 (c) || rcu_read_lock_bh_held(), __rcu) 619 620 /** 621 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 622 * @p: The pointer to read, prior to dereferencing 623 * @c: The conditions under which the dereference will take place 624 * 625 * This is the RCU-sched counterpart to rcu_dereference_check(). 626 * However, please note that starting in v5.0 kernels, vanilla RCU grace 627 * periods wait for preempt_disable() regions of code in addition to 628 * regions of code demarked by rcu_read_lock() and rcu_read_unlock(). 629 * This means that synchronize_rcu(), call_rcu, and friends all take not 630 * only rcu_read_lock() but also rcu_read_lock_sched() into account. 631 */ 632 #define rcu_dereference_sched_check(p, c) \ 633 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \ 634 (c) || rcu_read_lock_sched_held(), \ 635 __rcu) 636 637 /* 638 * The tracing infrastructure traces RCU (we want that), but unfortunately 639 * some of the RCU checks causes tracing to lock up the system. 640 * 641 * The no-tracing version of rcu_dereference_raw() must not call 642 * rcu_read_lock_held(). 643 */ 644 #define rcu_dereference_raw_check(p) \ 645 __rcu_dereference_check((p), __UNIQUE_ID(rcu), 1, __rcu) 646 647 /** 648 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 649 * @p: The pointer to read, prior to dereferencing 650 * @c: The conditions under which the dereference will take place 651 * 652 * Return the value of the specified RCU-protected pointer, but omit 653 * the READ_ONCE(). This is useful in cases where update-side locks 654 * prevent the value of the pointer from changing. Please note that this 655 * primitive does *not* prevent the compiler from repeating this reference 656 * or combining it with other references, so it should not be used without 657 * protection of appropriate locks. 658 * 659 * This function is only for update-side use. Using this function 660 * when protected only by rcu_read_lock() will result in infrequent 661 * but very ugly failures. 662 */ 663 #define rcu_dereference_protected(p, c) \ 664 __rcu_dereference_protected((p), __UNIQUE_ID(rcu), (c), __rcu) 665 666 667 /** 668 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 669 * @p: The pointer to read, prior to dereferencing 670 * 671 * This is a simple wrapper around rcu_dereference_check(). 672 */ 673 #define rcu_dereference(p) rcu_dereference_check(p, 0) 674 675 /** 676 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 677 * @p: The pointer to read, prior to dereferencing 678 * 679 * Makes rcu_dereference_check() do the dirty work. 680 */ 681 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 682 683 /** 684 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 685 * @p: The pointer to read, prior to dereferencing 686 * 687 * Makes rcu_dereference_check() do the dirty work. 688 */ 689 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 690 691 /** 692 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism 693 * @p: The pointer to hand off 694 * 695 * This is simply an identity function, but it documents where a pointer 696 * is handed off from RCU to some other synchronization mechanism, for 697 * example, reference counting or locking. In C11, it would map to 698 * kill_dependency(). It could be used as follows:: 699 * 700 * rcu_read_lock(); 701 * p = rcu_dereference(gp); 702 * long_lived = is_long_lived(p); 703 * if (long_lived) { 704 * if (!atomic_inc_not_zero(p->refcnt)) 705 * long_lived = false; 706 * else 707 * p = rcu_pointer_handoff(p); 708 * } 709 * rcu_read_unlock(); 710 */ 711 #define rcu_pointer_handoff(p) (p) 712 713 /** 714 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 715 * 716 * When synchronize_rcu() is invoked on one CPU while other CPUs 717 * are within RCU read-side critical sections, then the 718 * synchronize_rcu() is guaranteed to block until after all the other 719 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 720 * on one CPU while other CPUs are within RCU read-side critical 721 * sections, invocation of the corresponding RCU callback is deferred 722 * until after the all the other CPUs exit their critical sections. 723 * 724 * In v5.0 and later kernels, synchronize_rcu() and call_rcu() also 725 * wait for regions of code with preemption disabled, including regions of 726 * code with interrupts or softirqs disabled. In pre-v5.0 kernels, which 727 * define synchronize_sched(), only code enclosed within rcu_read_lock() 728 * and rcu_read_unlock() are guaranteed to be waited for. 729 * 730 * Note, however, that RCU callbacks are permitted to run concurrently 731 * with new RCU read-side critical sections. One way that this can happen 732 * is via the following sequence of events: (1) CPU 0 enters an RCU 733 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 734 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 735 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 736 * callback is invoked. This is legal, because the RCU read-side critical 737 * section that was running concurrently with the call_rcu() (and which 738 * therefore might be referencing something that the corresponding RCU 739 * callback would free up) has completed before the corresponding 740 * RCU callback is invoked. 741 * 742 * RCU read-side critical sections may be nested. Any deferred actions 743 * will be deferred until the outermost RCU read-side critical section 744 * completes. 745 * 746 * You can avoid reading and understanding the next paragraph by 747 * following this rule: don't put anything in an rcu_read_lock() RCU 748 * read-side critical section that would block in a !PREEMPTION kernel. 749 * But if you want the full story, read on! 750 * 751 * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU), 752 * it is illegal to block while in an RCU read-side critical section. 753 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION 754 * kernel builds, RCU read-side critical sections may be preempted, 755 * but explicit blocking is illegal. Finally, in preemptible RCU 756 * implementations in real-time (with -rt patchset) kernel builds, RCU 757 * read-side critical sections may be preempted and they may also block, but 758 * only when acquiring spinlocks that are subject to priority inheritance. 759 */ 760 static __always_inline void rcu_read_lock(void) 761 { 762 __rcu_read_lock(); 763 __acquire(RCU); 764 rcu_lock_acquire(&rcu_lock_map); 765 RCU_LOCKDEP_WARN(!rcu_is_watching(), 766 "rcu_read_lock() used illegally while idle"); 767 } 768 769 /* 770 * So where is rcu_write_lock()? It does not exist, as there is no 771 * way for writers to lock out RCU readers. This is a feature, not 772 * a bug -- this property is what provides RCU's performance benefits. 773 * Of course, writers must coordinate with each other. The normal 774 * spinlock primitives work well for this, but any other technique may be 775 * used as well. RCU does not care how the writers keep out of each 776 * others' way, as long as they do so. 777 */ 778 779 /** 780 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 781 * 782 * In almost all situations, rcu_read_unlock() is immune from deadlock. 783 * In recent kernels that have consolidated synchronize_sched() and 784 * synchronize_rcu_bh() into synchronize_rcu(), this deadlock immunity 785 * also extends to the scheduler's runqueue and priority-inheritance 786 * spinlocks, courtesy of the quiescent-state deferral that is carried 787 * out when rcu_read_unlock() is invoked with interrupts disabled. 788 * 789 * See rcu_read_lock() for more information. 790 */ 791 static inline void rcu_read_unlock(void) 792 { 793 RCU_LOCKDEP_WARN(!rcu_is_watching(), 794 "rcu_read_unlock() used illegally while idle"); 795 __release(RCU); 796 __rcu_read_unlock(); 797 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ 798 } 799 800 /** 801 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 802 * 803 * This is equivalent to rcu_read_lock(), but also disables softirqs. 804 * Note that anything else that disables softirqs can also serve as an RCU 805 * read-side critical section. However, please note that this equivalence 806 * applies only to v5.0 and later. Before v5.0, rcu_read_lock() and 807 * rcu_read_lock_bh() were unrelated. 808 * 809 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 810 * must occur in the same context, for example, it is illegal to invoke 811 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 812 * was invoked from some other task. 813 */ 814 static inline void rcu_read_lock_bh(void) 815 { 816 local_bh_disable(); 817 __acquire(RCU_BH); 818 rcu_lock_acquire(&rcu_bh_lock_map); 819 RCU_LOCKDEP_WARN(!rcu_is_watching(), 820 "rcu_read_lock_bh() used illegally while idle"); 821 } 822 823 /** 824 * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section 825 * 826 * See rcu_read_lock_bh() for more information. 827 */ 828 static inline void rcu_read_unlock_bh(void) 829 { 830 RCU_LOCKDEP_WARN(!rcu_is_watching(), 831 "rcu_read_unlock_bh() used illegally while idle"); 832 rcu_lock_release(&rcu_bh_lock_map); 833 __release(RCU_BH); 834 local_bh_enable(); 835 } 836 837 /** 838 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 839 * 840 * This is equivalent to rcu_read_lock(), but also disables preemption. 841 * Read-side critical sections can also be introduced by anything else that 842 * disables preemption, including local_irq_disable() and friends. However, 843 * please note that the equivalence to rcu_read_lock() applies only to 844 * v5.0 and later. Before v5.0, rcu_read_lock() and rcu_read_lock_sched() 845 * were unrelated. 846 * 847 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 848 * must occur in the same context, for example, it is illegal to invoke 849 * rcu_read_unlock_sched() from process context if the matching 850 * rcu_read_lock_sched() was invoked from an NMI handler. 851 */ 852 static inline void rcu_read_lock_sched(void) 853 { 854 preempt_disable(); 855 __acquire(RCU_SCHED); 856 rcu_lock_acquire(&rcu_sched_lock_map); 857 RCU_LOCKDEP_WARN(!rcu_is_watching(), 858 "rcu_read_lock_sched() used illegally while idle"); 859 } 860 861 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 862 static inline notrace void rcu_read_lock_sched_notrace(void) 863 { 864 preempt_disable_notrace(); 865 __acquire(RCU_SCHED); 866 } 867 868 /** 869 * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section 870 * 871 * See rcu_read_lock_sched() for more information. 872 */ 873 static inline void rcu_read_unlock_sched(void) 874 { 875 RCU_LOCKDEP_WARN(!rcu_is_watching(), 876 "rcu_read_unlock_sched() used illegally while idle"); 877 rcu_lock_release(&rcu_sched_lock_map); 878 __release(RCU_SCHED); 879 preempt_enable(); 880 } 881 882 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 883 static inline notrace void rcu_read_unlock_sched_notrace(void) 884 { 885 __release(RCU_SCHED); 886 preempt_enable_notrace(); 887 } 888 889 /** 890 * RCU_INIT_POINTER() - initialize an RCU protected pointer 891 * @p: The pointer to be initialized. 892 * @v: The value to initialized the pointer to. 893 * 894 * Initialize an RCU-protected pointer in special cases where readers 895 * do not need ordering constraints on the CPU or the compiler. These 896 * special cases are: 897 * 898 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or* 899 * 2. The caller has taken whatever steps are required to prevent 900 * RCU readers from concurrently accessing this pointer *or* 901 * 3. The referenced data structure has already been exposed to 902 * readers either at compile time or via rcu_assign_pointer() *and* 903 * 904 * a. You have not made *any* reader-visible changes to 905 * this structure since then *or* 906 * b. It is OK for readers accessing this structure from its 907 * new location to see the old state of the structure. (For 908 * example, the changes were to statistical counters or to 909 * other state where exact synchronization is not required.) 910 * 911 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 912 * result in impossible-to-diagnose memory corruption. As in the structures 913 * will look OK in crash dumps, but any concurrent RCU readers might 914 * see pre-initialized values of the referenced data structure. So 915 * please be very careful how you use RCU_INIT_POINTER()!!! 916 * 917 * If you are creating an RCU-protected linked structure that is accessed 918 * by a single external-to-structure RCU-protected pointer, then you may 919 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 920 * pointers, but you must use rcu_assign_pointer() to initialize the 921 * external-to-structure pointer *after* you have completely initialized 922 * the reader-accessible portions of the linked structure. 923 * 924 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 925 * ordering guarantees for either the CPU or the compiler. 926 */ 927 #define RCU_INIT_POINTER(p, v) \ 928 do { \ 929 rcu_check_sparse(p, __rcu); \ 930 WRITE_ONCE(p, RCU_INITIALIZER(v)); \ 931 } while (0) 932 933 /** 934 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 935 * @p: The pointer to be initialized. 936 * @v: The value to initialized the pointer to. 937 * 938 * GCC-style initialization for an RCU-protected pointer in a structure field. 939 */ 940 #define RCU_POINTER_INITIALIZER(p, v) \ 941 .p = RCU_INITIALIZER(v) 942 943 /* 944 * Does the specified offset indicate that the corresponding rcu_head 945 * structure can be handled by kvfree_rcu()? 946 */ 947 #define __is_kvfree_rcu_offset(offset) ((offset) < 4096) 948 949 /** 950 * kfree_rcu() - kfree an object after a grace period. 951 * @ptr: pointer to kfree for both single- and double-argument invocations. 952 * @rhf: the name of the struct rcu_head within the type of @ptr, 953 * but only for double-argument invocations. 954 * 955 * Many rcu callbacks functions just call kfree() on the base structure. 956 * These functions are trivial, but their size adds up, and furthermore 957 * when they are used in a kernel module, that module must invoke the 958 * high-latency rcu_barrier() function at module-unload time. 959 * 960 * The kfree_rcu() function handles this issue. Rather than encoding a 961 * function address in the embedded rcu_head structure, kfree_rcu() instead 962 * encodes the offset of the rcu_head structure within the base structure. 963 * Because the functions are not allowed in the low-order 4096 bytes of 964 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 965 * If the offset is larger than 4095 bytes, a compile-time error will 966 * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can 967 * either fall back to use of call_rcu() or rearrange the structure to 968 * position the rcu_head structure into the first 4096 bytes. 969 * 970 * Note that the allowable offset might decrease in the future, for example, 971 * to allow something like kmem_cache_free_rcu(). 972 * 973 * The BUILD_BUG_ON check must not involve any function calls, hence the 974 * checks are done in macros here. 975 */ 976 #define kfree_rcu(ptr, rhf...) kvfree_rcu(ptr, ## rhf) 977 978 /** 979 * kvfree_rcu() - kvfree an object after a grace period. 980 * 981 * This macro consists of one or two arguments and it is 982 * based on whether an object is head-less or not. If it 983 * has a head then a semantic stays the same as it used 984 * to be before: 985 * 986 * kvfree_rcu(ptr, rhf); 987 * 988 * where @ptr is a pointer to kvfree(), @rhf is the name 989 * of the rcu_head structure within the type of @ptr. 990 * 991 * When it comes to head-less variant, only one argument 992 * is passed and that is just a pointer which has to be 993 * freed after a grace period. Therefore the semantic is 994 * 995 * kvfree_rcu(ptr); 996 * 997 * where @ptr is the pointer to be freed by kvfree(). 998 * 999 * Please note, head-less way of freeing is permitted to 1000 * use from a context that has to follow might_sleep() 1001 * annotation. Otherwise, please switch and embed the 1002 * rcu_head structure within the type of @ptr. 1003 */ 1004 #define kvfree_rcu(...) KVFREE_GET_MACRO(__VA_ARGS__, \ 1005 kvfree_rcu_arg_2, kvfree_rcu_arg_1)(__VA_ARGS__) 1006 1007 #define KVFREE_GET_MACRO(_1, _2, NAME, ...) NAME 1008 #define kvfree_rcu_arg_2(ptr, rhf) \ 1009 do { \ 1010 typeof (ptr) ___p = (ptr); \ 1011 \ 1012 if (___p) { \ 1013 BUILD_BUG_ON(!__is_kvfree_rcu_offset(offsetof(typeof(*(ptr)), rhf))); \ 1014 kvfree_call_rcu(&((___p)->rhf), (rcu_callback_t)(unsigned long) \ 1015 (offsetof(typeof(*(ptr)), rhf))); \ 1016 } \ 1017 } while (0) 1018 1019 #define kvfree_rcu_arg_1(ptr) \ 1020 do { \ 1021 typeof(ptr) ___p = (ptr); \ 1022 \ 1023 if (___p) \ 1024 kvfree_call_rcu(NULL, (rcu_callback_t) (___p)); \ 1025 } while (0) 1026 1027 /* 1028 * Place this after a lock-acquisition primitive to guarantee that 1029 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies 1030 * if the UNLOCK and LOCK are executed by the same CPU or if the 1031 * UNLOCK and LOCK operate on the same lock variable. 1032 */ 1033 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE 1034 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */ 1035 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 1036 #define smp_mb__after_unlock_lock() do { } while (0) 1037 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 1038 1039 1040 /* Has the specified rcu_head structure been handed to call_rcu()? */ 1041 1042 /** 1043 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu() 1044 * @rhp: The rcu_head structure to initialize. 1045 * 1046 * If you intend to invoke rcu_head_after_call_rcu() to test whether a 1047 * given rcu_head structure has already been passed to call_rcu(), then 1048 * you must also invoke this rcu_head_init() function on it just after 1049 * allocating that structure. Calls to this function must not race with 1050 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation. 1051 */ 1052 static inline void rcu_head_init(struct rcu_head *rhp) 1053 { 1054 rhp->func = (rcu_callback_t)~0L; 1055 } 1056 1057 /** 1058 * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()? 1059 * @rhp: The rcu_head structure to test. 1060 * @f: The function passed to call_rcu() along with @rhp. 1061 * 1062 * Returns @true if the @rhp has been passed to call_rcu() with @func, 1063 * and @false otherwise. Emits a warning in any other case, including 1064 * the case where @rhp has already been invoked after a grace period. 1065 * Calls to this function must not race with callback invocation. One way 1066 * to avoid such races is to enclose the call to rcu_head_after_call_rcu() 1067 * in an RCU read-side critical section that includes a read-side fetch 1068 * of the pointer to the structure containing @rhp. 1069 */ 1070 static inline bool 1071 rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f) 1072 { 1073 rcu_callback_t func = READ_ONCE(rhp->func); 1074 1075 if (func == f) 1076 return true; 1077 WARN_ON_ONCE(func != (rcu_callback_t)~0L); 1078 return false; 1079 } 1080 1081 /* kernel/ksysfs.c definitions */ 1082 extern int rcu_expedited; 1083 extern int rcu_normal; 1084 1085 #endif /* __LINUX_RCUPDATE_H */ 1086