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_stop(void); 242 void exit_tasks_rcu_finish(void); 243 #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */ 244 #define rcu_tasks_classic_qs(t, preempt) do { } while (0) 245 #define rcu_tasks_qs(t, preempt) do { } while (0) 246 #define rcu_note_voluntary_context_switch(t) do { } while (0) 247 #define call_rcu_tasks call_rcu 248 #define synchronize_rcu_tasks synchronize_rcu 249 static inline void exit_tasks_rcu_start(void) { } 250 static inline void exit_tasks_rcu_stop(void) { } 251 static inline void exit_tasks_rcu_finish(void) { } 252 #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */ 253 254 /** 255 * rcu_trace_implies_rcu_gp - does an RCU Tasks Trace grace period imply an RCU grace period? 256 * 257 * As an accident of implementation, an RCU Tasks Trace grace period also 258 * acts as an RCU grace period. However, this could change at any time. 259 * Code relying on this accident must call this function to verify that 260 * this accident is still happening. 261 * 262 * You have been warned! 263 */ 264 static inline bool rcu_trace_implies_rcu_gp(void) { return true; } 265 266 /** 267 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU 268 * 269 * This macro resembles cond_resched(), except that it is defined to 270 * report potential quiescent states to RCU-tasks even if the cond_resched() 271 * machinery were to be shut off, as some advocate for PREEMPTION kernels. 272 */ 273 #define cond_resched_tasks_rcu_qs() \ 274 do { \ 275 rcu_tasks_qs(current, false); \ 276 cond_resched(); \ 277 } while (0) 278 279 /* 280 * Infrastructure to implement the synchronize_() primitives in 281 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU. 282 */ 283 284 #if defined(CONFIG_TREE_RCU) 285 #include <linux/rcutree.h> 286 #elif defined(CONFIG_TINY_RCU) 287 #include <linux/rcutiny.h> 288 #else 289 #error "Unknown RCU implementation specified to kernel configuration" 290 #endif 291 292 /* 293 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls 294 * are needed for dynamic initialization and destruction of rcu_head 295 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for 296 * dynamic initialization and destruction of statically allocated rcu_head 297 * structures. However, rcu_head structures allocated dynamically in the 298 * heap don't need any initialization. 299 */ 300 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 301 void init_rcu_head(struct rcu_head *head); 302 void destroy_rcu_head(struct rcu_head *head); 303 void init_rcu_head_on_stack(struct rcu_head *head); 304 void destroy_rcu_head_on_stack(struct rcu_head *head); 305 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 306 static inline void init_rcu_head(struct rcu_head *head) { } 307 static inline void destroy_rcu_head(struct rcu_head *head) { } 308 static inline void init_rcu_head_on_stack(struct rcu_head *head) { } 309 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { } 310 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 311 312 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) 313 bool rcu_lockdep_current_cpu_online(void); 314 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 315 static inline bool rcu_lockdep_current_cpu_online(void) { return true; } 316 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */ 317 318 extern struct lockdep_map rcu_lock_map; 319 extern struct lockdep_map rcu_bh_lock_map; 320 extern struct lockdep_map rcu_sched_lock_map; 321 extern struct lockdep_map rcu_callback_map; 322 323 #ifdef CONFIG_DEBUG_LOCK_ALLOC 324 325 static inline void rcu_lock_acquire(struct lockdep_map *map) 326 { 327 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_); 328 } 329 330 static inline void rcu_lock_release(struct lockdep_map *map) 331 { 332 lock_release(map, _THIS_IP_); 333 } 334 335 int debug_lockdep_rcu_enabled(void); 336 int rcu_read_lock_held(void); 337 int rcu_read_lock_bh_held(void); 338 int rcu_read_lock_sched_held(void); 339 int rcu_read_lock_any_held(void); 340 341 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 342 343 # define rcu_lock_acquire(a) do { } while (0) 344 # define rcu_lock_release(a) do { } while (0) 345 346 static inline int rcu_read_lock_held(void) 347 { 348 return 1; 349 } 350 351 static inline int rcu_read_lock_bh_held(void) 352 { 353 return 1; 354 } 355 356 static inline int rcu_read_lock_sched_held(void) 357 { 358 return !preemptible(); 359 } 360 361 static inline int rcu_read_lock_any_held(void) 362 { 363 return !preemptible(); 364 } 365 366 static inline int debug_lockdep_rcu_enabled(void) 367 { 368 return 0; 369 } 370 371 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 372 373 #ifdef CONFIG_PROVE_RCU 374 375 /** 376 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met 377 * @c: condition to check 378 * @s: informative message 379 * 380 * This checks debug_lockdep_rcu_enabled() before checking (c) to 381 * prevent early boot splats due to lockdep not yet being initialized, 382 * and rechecks it after checking (c) to prevent false-positive splats 383 * due to races with lockdep being disabled. See commit 3066820034b5dd 384 * ("rcu: Reject RCU_LOCKDEP_WARN() false positives") for more detail. 385 */ 386 #define RCU_LOCKDEP_WARN(c, s) \ 387 do { \ 388 static bool __section(".data.unlikely") __warned; \ 389 if (debug_lockdep_rcu_enabled() && (c) && \ 390 debug_lockdep_rcu_enabled() && !__warned) { \ 391 __warned = true; \ 392 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \ 393 } \ 394 } while (0) 395 396 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU) 397 static inline void rcu_preempt_sleep_check(void) 398 { 399 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map), 400 "Illegal context switch in RCU read-side critical section"); 401 } 402 #else /* #ifdef CONFIG_PROVE_RCU */ 403 static inline void rcu_preempt_sleep_check(void) { } 404 #endif /* #else #ifdef CONFIG_PROVE_RCU */ 405 406 #define rcu_sleep_check() \ 407 do { \ 408 rcu_preempt_sleep_check(); \ 409 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) \ 410 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \ 411 "Illegal context switch in RCU-bh read-side critical section"); \ 412 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \ 413 "Illegal context switch in RCU-sched read-side critical section"); \ 414 } while (0) 415 416 #else /* #ifdef CONFIG_PROVE_RCU */ 417 418 #define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c)) 419 #define rcu_sleep_check() do { } while (0) 420 421 #endif /* #else #ifdef CONFIG_PROVE_RCU */ 422 423 /* 424 * Helper functions for rcu_dereference_check(), rcu_dereference_protected() 425 * and rcu_assign_pointer(). Some of these could be folded into their 426 * callers, but they are left separate in order to ease introduction of 427 * multiple pointers markings to match different RCU implementations 428 * (e.g., __srcu), should this make sense in the future. 429 */ 430 431 #ifdef __CHECKER__ 432 #define rcu_check_sparse(p, space) \ 433 ((void)(((typeof(*p) space *)p) == p)) 434 #else /* #ifdef __CHECKER__ */ 435 #define rcu_check_sparse(p, space) 436 #endif /* #else #ifdef __CHECKER__ */ 437 438 #define __unrcu_pointer(p, local) \ 439 ({ \ 440 typeof(*p) *local = (typeof(*p) *__force)(p); \ 441 rcu_check_sparse(p, __rcu); \ 442 ((typeof(*p) __force __kernel *)(local)); \ 443 }) 444 /** 445 * unrcu_pointer - mark a pointer as not being RCU protected 446 * @p: pointer needing to lose its __rcu property 447 * 448 * Converts @p from an __rcu pointer to a __kernel pointer. 449 * This allows an __rcu pointer to be used with xchg() and friends. 450 */ 451 #define unrcu_pointer(p) __unrcu_pointer(p, __UNIQUE_ID(rcu)) 452 453 #define __rcu_access_pointer(p, local, space) \ 454 ({ \ 455 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \ 456 rcu_check_sparse(p, space); \ 457 ((typeof(*p) __force __kernel *)(local)); \ 458 }) 459 #define __rcu_dereference_check(p, local, c, space) \ 460 ({ \ 461 /* Dependency order vs. p above. */ \ 462 typeof(*p) *local = (typeof(*p) *__force)READ_ONCE(p); \ 463 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \ 464 rcu_check_sparse(p, space); \ 465 ((typeof(*p) __force __kernel *)(local)); \ 466 }) 467 #define __rcu_dereference_protected(p, local, c, space) \ 468 ({ \ 469 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \ 470 rcu_check_sparse(p, space); \ 471 ((typeof(*p) __force __kernel *)(p)); \ 472 }) 473 #define __rcu_dereference_raw(p, local) \ 474 ({ \ 475 /* Dependency order vs. p above. */ \ 476 typeof(p) local = READ_ONCE(p); \ 477 ((typeof(*p) __force __kernel *)(local)); \ 478 }) 479 #define rcu_dereference_raw(p) __rcu_dereference_raw(p, __UNIQUE_ID(rcu)) 480 481 /** 482 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable 483 * @v: The value to statically initialize with. 484 */ 485 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v) 486 487 /** 488 * rcu_assign_pointer() - assign to RCU-protected pointer 489 * @p: pointer to assign to 490 * @v: value to assign (publish) 491 * 492 * Assigns the specified value to the specified RCU-protected 493 * pointer, ensuring that any concurrent RCU readers will see 494 * any prior initialization. 495 * 496 * Inserts memory barriers on architectures that require them 497 * (which is most of them), and also prevents the compiler from 498 * reordering the code that initializes the structure after the pointer 499 * assignment. More importantly, this call documents which pointers 500 * will be dereferenced by RCU read-side code. 501 * 502 * In some special cases, you may use RCU_INIT_POINTER() instead 503 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due 504 * to the fact that it does not constrain either the CPU or the compiler. 505 * That said, using RCU_INIT_POINTER() when you should have used 506 * rcu_assign_pointer() is a very bad thing that results in 507 * impossible-to-diagnose memory corruption. So please be careful. 508 * See the RCU_INIT_POINTER() comment header for details. 509 * 510 * Note that rcu_assign_pointer() evaluates each of its arguments only 511 * once, appearances notwithstanding. One of the "extra" evaluations 512 * is in typeof() and the other visible only to sparse (__CHECKER__), 513 * neither of which actually execute the argument. As with most cpp 514 * macros, this execute-arguments-only-once property is important, so 515 * please be careful when making changes to rcu_assign_pointer() and the 516 * other macros that it invokes. 517 */ 518 #define rcu_assign_pointer(p, v) \ 519 do { \ 520 uintptr_t _r_a_p__v = (uintptr_t)(v); \ 521 rcu_check_sparse(p, __rcu); \ 522 \ 523 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \ 524 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \ 525 else \ 526 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \ 527 } while (0) 528 529 /** 530 * rcu_replace_pointer() - replace an RCU pointer, returning its old value 531 * @rcu_ptr: RCU pointer, whose old value is returned 532 * @ptr: regular pointer 533 * @c: the lockdep conditions under which the dereference will take place 534 * 535 * Perform a replacement, where @rcu_ptr is an RCU-annotated 536 * pointer and @c is the lockdep argument that is passed to the 537 * rcu_dereference_protected() call used to read that pointer. The old 538 * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr. 539 */ 540 #define rcu_replace_pointer(rcu_ptr, ptr, c) \ 541 ({ \ 542 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \ 543 rcu_assign_pointer((rcu_ptr), (ptr)); \ 544 __tmp; \ 545 }) 546 547 /** 548 * rcu_access_pointer() - fetch RCU pointer with no dereferencing 549 * @p: The pointer to read 550 * 551 * Return the value of the specified RCU-protected pointer, but omit the 552 * lockdep checks for being in an RCU read-side critical section. This is 553 * useful when the value of this pointer is accessed, but the pointer is 554 * not dereferenced, for example, when testing an RCU-protected pointer 555 * against NULL. Although rcu_access_pointer() may also be used in cases 556 * where update-side locks prevent the value of the pointer from changing, 557 * you should instead use rcu_dereference_protected() for this use case. 558 * Within an RCU read-side critical section, there is little reason to 559 * use rcu_access_pointer(). 560 * 561 * It is usually best to test the rcu_access_pointer() return value 562 * directly in order to avoid accidental dereferences being introduced 563 * by later inattentive changes. In other words, assigning the 564 * rcu_access_pointer() return value to a local variable results in an 565 * accident waiting to happen. 566 * 567 * It is also permissible to use rcu_access_pointer() when read-side 568 * access to the pointer was removed at least one grace period ago, as is 569 * the case in the context of the RCU callback that is freeing up the data, 570 * or after a synchronize_rcu() returns. This can be useful when tearing 571 * down multi-linked structures after a grace period has elapsed. However, 572 * rcu_dereference_protected() is normally preferred for this use case. 573 */ 574 #define rcu_access_pointer(p) __rcu_access_pointer((p), __UNIQUE_ID(rcu), __rcu) 575 576 /** 577 * rcu_dereference_check() - rcu_dereference with debug checking 578 * @p: The pointer to read, prior to dereferencing 579 * @c: The conditions under which the dereference will take place 580 * 581 * Do an rcu_dereference(), but check that the conditions under which the 582 * dereference will take place are correct. Typically the conditions 583 * indicate the various locking conditions that should be held at that 584 * point. The check should return true if the conditions are satisfied. 585 * An implicit check for being in an RCU read-side critical section 586 * (rcu_read_lock()) is included. 587 * 588 * For example: 589 * 590 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock)); 591 * 592 * could be used to indicate to lockdep that foo->bar may only be dereferenced 593 * if either rcu_read_lock() is held, or that the lock required to replace 594 * the bar struct at foo->bar is held. 595 * 596 * Note that the list of conditions may also include indications of when a lock 597 * need not be held, for example during initialisation or destruction of the 598 * target struct: 599 * 600 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) || 601 * atomic_read(&foo->usage) == 0); 602 * 603 * Inserts memory barriers on architectures that require them 604 * (currently only the Alpha), prevents the compiler from refetching 605 * (and from merging fetches), and, more importantly, documents exactly 606 * which pointers are protected by RCU and checks that the pointer is 607 * annotated as __rcu. 608 */ 609 #define rcu_dereference_check(p, c) \ 610 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \ 611 (c) || rcu_read_lock_held(), __rcu) 612 613 /** 614 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking 615 * @p: The pointer to read, prior to dereferencing 616 * @c: The conditions under which the dereference will take place 617 * 618 * This is the RCU-bh counterpart to rcu_dereference_check(). However, 619 * please note that starting in v5.0 kernels, vanilla RCU grace periods 620 * wait for local_bh_disable() regions of code in addition to regions of 621 * code demarked by rcu_read_lock() and rcu_read_unlock(). This means 622 * that synchronize_rcu(), call_rcu, and friends all take not only 623 * rcu_read_lock() but also rcu_read_lock_bh() into account. 624 */ 625 #define rcu_dereference_bh_check(p, c) \ 626 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \ 627 (c) || rcu_read_lock_bh_held(), __rcu) 628 629 /** 630 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking 631 * @p: The pointer to read, prior to dereferencing 632 * @c: The conditions under which the dereference will take place 633 * 634 * This is the RCU-sched counterpart to rcu_dereference_check(). 635 * However, please note that starting in v5.0 kernels, vanilla RCU grace 636 * periods wait for preempt_disable() regions of code in addition to 637 * regions of code demarked by rcu_read_lock() and rcu_read_unlock(). 638 * This means that synchronize_rcu(), call_rcu, and friends all take not 639 * only rcu_read_lock() but also rcu_read_lock_sched() into account. 640 */ 641 #define rcu_dereference_sched_check(p, c) \ 642 __rcu_dereference_check((p), __UNIQUE_ID(rcu), \ 643 (c) || rcu_read_lock_sched_held(), \ 644 __rcu) 645 646 /* 647 * The tracing infrastructure traces RCU (we want that), but unfortunately 648 * some of the RCU checks causes tracing to lock up the system. 649 * 650 * The no-tracing version of rcu_dereference_raw() must not call 651 * rcu_read_lock_held(). 652 */ 653 #define rcu_dereference_raw_check(p) \ 654 __rcu_dereference_check((p), __UNIQUE_ID(rcu), 1, __rcu) 655 656 /** 657 * rcu_dereference_protected() - fetch RCU pointer when updates prevented 658 * @p: The pointer to read, prior to dereferencing 659 * @c: The conditions under which the dereference will take place 660 * 661 * Return the value of the specified RCU-protected pointer, but omit 662 * the READ_ONCE(). This is useful in cases where update-side locks 663 * prevent the value of the pointer from changing. Please note that this 664 * primitive does *not* prevent the compiler from repeating this reference 665 * or combining it with other references, so it should not be used without 666 * protection of appropriate locks. 667 * 668 * This function is only for update-side use. Using this function 669 * when protected only by rcu_read_lock() will result in infrequent 670 * but very ugly failures. 671 */ 672 #define rcu_dereference_protected(p, c) \ 673 __rcu_dereference_protected((p), __UNIQUE_ID(rcu), (c), __rcu) 674 675 676 /** 677 * rcu_dereference() - fetch RCU-protected pointer for dereferencing 678 * @p: The pointer to read, prior to dereferencing 679 * 680 * This is a simple wrapper around rcu_dereference_check(). 681 */ 682 #define rcu_dereference(p) rcu_dereference_check(p, 0) 683 684 /** 685 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing 686 * @p: The pointer to read, prior to dereferencing 687 * 688 * Makes rcu_dereference_check() do the dirty work. 689 */ 690 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0) 691 692 /** 693 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing 694 * @p: The pointer to read, prior to dereferencing 695 * 696 * Makes rcu_dereference_check() do the dirty work. 697 */ 698 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0) 699 700 /** 701 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism 702 * @p: The pointer to hand off 703 * 704 * This is simply an identity function, but it documents where a pointer 705 * is handed off from RCU to some other synchronization mechanism, for 706 * example, reference counting or locking. In C11, it would map to 707 * kill_dependency(). It could be used as follows:: 708 * 709 * rcu_read_lock(); 710 * p = rcu_dereference(gp); 711 * long_lived = is_long_lived(p); 712 * if (long_lived) { 713 * if (!atomic_inc_not_zero(p->refcnt)) 714 * long_lived = false; 715 * else 716 * p = rcu_pointer_handoff(p); 717 * } 718 * rcu_read_unlock(); 719 */ 720 #define rcu_pointer_handoff(p) (p) 721 722 /** 723 * rcu_read_lock() - mark the beginning of an RCU read-side critical section 724 * 725 * When synchronize_rcu() is invoked on one CPU while other CPUs 726 * are within RCU read-side critical sections, then the 727 * synchronize_rcu() is guaranteed to block until after all the other 728 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked 729 * on one CPU while other CPUs are within RCU read-side critical 730 * sections, invocation of the corresponding RCU callback is deferred 731 * until after the all the other CPUs exit their critical sections. 732 * 733 * In v5.0 and later kernels, synchronize_rcu() and call_rcu() also 734 * wait for regions of code with preemption disabled, including regions of 735 * code with interrupts or softirqs disabled. In pre-v5.0 kernels, which 736 * define synchronize_sched(), only code enclosed within rcu_read_lock() 737 * and rcu_read_unlock() are guaranteed to be waited for. 738 * 739 * Note, however, that RCU callbacks are permitted to run concurrently 740 * with new RCU read-side critical sections. One way that this can happen 741 * is via the following sequence of events: (1) CPU 0 enters an RCU 742 * read-side critical section, (2) CPU 1 invokes call_rcu() to register 743 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, 744 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU 745 * callback is invoked. This is legal, because the RCU read-side critical 746 * section that was running concurrently with the call_rcu() (and which 747 * therefore might be referencing something that the corresponding RCU 748 * callback would free up) has completed before the corresponding 749 * RCU callback is invoked. 750 * 751 * RCU read-side critical sections may be nested. Any deferred actions 752 * will be deferred until the outermost RCU read-side critical section 753 * completes. 754 * 755 * You can avoid reading and understanding the next paragraph by 756 * following this rule: don't put anything in an rcu_read_lock() RCU 757 * read-side critical section that would block in a !PREEMPTION kernel. 758 * But if you want the full story, read on! 759 * 760 * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU), 761 * it is illegal to block while in an RCU read-side critical section. 762 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION 763 * kernel builds, RCU read-side critical sections may be preempted, 764 * but explicit blocking is illegal. Finally, in preemptible RCU 765 * implementations in real-time (with -rt patchset) kernel builds, RCU 766 * read-side critical sections may be preempted and they may also block, but 767 * only when acquiring spinlocks that are subject to priority inheritance. 768 */ 769 static __always_inline void rcu_read_lock(void) 770 { 771 __rcu_read_lock(); 772 __acquire(RCU); 773 rcu_lock_acquire(&rcu_lock_map); 774 RCU_LOCKDEP_WARN(!rcu_is_watching(), 775 "rcu_read_lock() used illegally while idle"); 776 } 777 778 /* 779 * So where is rcu_write_lock()? It does not exist, as there is no 780 * way for writers to lock out RCU readers. This is a feature, not 781 * a bug -- this property is what provides RCU's performance benefits. 782 * Of course, writers must coordinate with each other. The normal 783 * spinlock primitives work well for this, but any other technique may be 784 * used as well. RCU does not care how the writers keep out of each 785 * others' way, as long as they do so. 786 */ 787 788 /** 789 * rcu_read_unlock() - marks the end of an RCU read-side critical section. 790 * 791 * In almost all situations, rcu_read_unlock() is immune from deadlock. 792 * In recent kernels that have consolidated synchronize_sched() and 793 * synchronize_rcu_bh() into synchronize_rcu(), this deadlock immunity 794 * also extends to the scheduler's runqueue and priority-inheritance 795 * spinlocks, courtesy of the quiescent-state deferral that is carried 796 * out when rcu_read_unlock() is invoked with interrupts disabled. 797 * 798 * See rcu_read_lock() for more information. 799 */ 800 static inline void rcu_read_unlock(void) 801 { 802 RCU_LOCKDEP_WARN(!rcu_is_watching(), 803 "rcu_read_unlock() used illegally while idle"); 804 __release(RCU); 805 __rcu_read_unlock(); 806 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ 807 } 808 809 /** 810 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section 811 * 812 * This is equivalent to rcu_read_lock(), but also disables softirqs. 813 * Note that anything else that disables softirqs can also serve as an RCU 814 * read-side critical section. However, please note that this equivalence 815 * applies only to v5.0 and later. Before v5.0, rcu_read_lock() and 816 * rcu_read_lock_bh() were unrelated. 817 * 818 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh() 819 * must occur in the same context, for example, it is illegal to invoke 820 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh() 821 * was invoked from some other task. 822 */ 823 static inline void rcu_read_lock_bh(void) 824 { 825 local_bh_disable(); 826 __acquire(RCU_BH); 827 rcu_lock_acquire(&rcu_bh_lock_map); 828 RCU_LOCKDEP_WARN(!rcu_is_watching(), 829 "rcu_read_lock_bh() used illegally while idle"); 830 } 831 832 /** 833 * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section 834 * 835 * See rcu_read_lock_bh() for more information. 836 */ 837 static inline void rcu_read_unlock_bh(void) 838 { 839 RCU_LOCKDEP_WARN(!rcu_is_watching(), 840 "rcu_read_unlock_bh() used illegally while idle"); 841 rcu_lock_release(&rcu_bh_lock_map); 842 __release(RCU_BH); 843 local_bh_enable(); 844 } 845 846 /** 847 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section 848 * 849 * This is equivalent to rcu_read_lock(), but also disables preemption. 850 * Read-side critical sections can also be introduced by anything else that 851 * disables preemption, including local_irq_disable() and friends. However, 852 * please note that the equivalence to rcu_read_lock() applies only to 853 * v5.0 and later. Before v5.0, rcu_read_lock() and rcu_read_lock_sched() 854 * were unrelated. 855 * 856 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched() 857 * must occur in the same context, for example, it is illegal to invoke 858 * rcu_read_unlock_sched() from process context if the matching 859 * rcu_read_lock_sched() was invoked from an NMI handler. 860 */ 861 static inline void rcu_read_lock_sched(void) 862 { 863 preempt_disable(); 864 __acquire(RCU_SCHED); 865 rcu_lock_acquire(&rcu_sched_lock_map); 866 RCU_LOCKDEP_WARN(!rcu_is_watching(), 867 "rcu_read_lock_sched() used illegally while idle"); 868 } 869 870 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 871 static inline notrace void rcu_read_lock_sched_notrace(void) 872 { 873 preempt_disable_notrace(); 874 __acquire(RCU_SCHED); 875 } 876 877 /** 878 * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section 879 * 880 * See rcu_read_lock_sched() for more information. 881 */ 882 static inline void rcu_read_unlock_sched(void) 883 { 884 RCU_LOCKDEP_WARN(!rcu_is_watching(), 885 "rcu_read_unlock_sched() used illegally while idle"); 886 rcu_lock_release(&rcu_sched_lock_map); 887 __release(RCU_SCHED); 888 preempt_enable(); 889 } 890 891 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */ 892 static inline notrace void rcu_read_unlock_sched_notrace(void) 893 { 894 __release(RCU_SCHED); 895 preempt_enable_notrace(); 896 } 897 898 /** 899 * RCU_INIT_POINTER() - initialize an RCU protected pointer 900 * @p: The pointer to be initialized. 901 * @v: The value to initialized the pointer to. 902 * 903 * Initialize an RCU-protected pointer in special cases where readers 904 * do not need ordering constraints on the CPU or the compiler. These 905 * special cases are: 906 * 907 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or* 908 * 2. The caller has taken whatever steps are required to prevent 909 * RCU readers from concurrently accessing this pointer *or* 910 * 3. The referenced data structure has already been exposed to 911 * readers either at compile time or via rcu_assign_pointer() *and* 912 * 913 * a. You have not made *any* reader-visible changes to 914 * this structure since then *or* 915 * b. It is OK for readers accessing this structure from its 916 * new location to see the old state of the structure. (For 917 * example, the changes were to statistical counters or to 918 * other state where exact synchronization is not required.) 919 * 920 * Failure to follow these rules governing use of RCU_INIT_POINTER() will 921 * result in impossible-to-diagnose memory corruption. As in the structures 922 * will look OK in crash dumps, but any concurrent RCU readers might 923 * see pre-initialized values of the referenced data structure. So 924 * please be very careful how you use RCU_INIT_POINTER()!!! 925 * 926 * If you are creating an RCU-protected linked structure that is accessed 927 * by a single external-to-structure RCU-protected pointer, then you may 928 * use RCU_INIT_POINTER() to initialize the internal RCU-protected 929 * pointers, but you must use rcu_assign_pointer() to initialize the 930 * external-to-structure pointer *after* you have completely initialized 931 * the reader-accessible portions of the linked structure. 932 * 933 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no 934 * ordering guarantees for either the CPU or the compiler. 935 */ 936 #define RCU_INIT_POINTER(p, v) \ 937 do { \ 938 rcu_check_sparse(p, __rcu); \ 939 WRITE_ONCE(p, RCU_INITIALIZER(v)); \ 940 } while (0) 941 942 /** 943 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer 944 * @p: The pointer to be initialized. 945 * @v: The value to initialized the pointer to. 946 * 947 * GCC-style initialization for an RCU-protected pointer in a structure field. 948 */ 949 #define RCU_POINTER_INITIALIZER(p, v) \ 950 .p = RCU_INITIALIZER(v) 951 952 /* 953 * Does the specified offset indicate that the corresponding rcu_head 954 * structure can be handled by kvfree_rcu()? 955 */ 956 #define __is_kvfree_rcu_offset(offset) ((offset) < 4096) 957 958 /** 959 * kfree_rcu() - kfree an object after a grace period. 960 * @ptr: pointer to kfree for both single- and double-argument invocations. 961 * @rhf: the name of the struct rcu_head within the type of @ptr, 962 * but only for double-argument invocations. 963 * 964 * Many rcu callbacks functions just call kfree() on the base structure. 965 * These functions are trivial, but their size adds up, and furthermore 966 * when they are used in a kernel module, that module must invoke the 967 * high-latency rcu_barrier() function at module-unload time. 968 * 969 * The kfree_rcu() function handles this issue. Rather than encoding a 970 * function address in the embedded rcu_head structure, kfree_rcu() instead 971 * encodes the offset of the rcu_head structure within the base structure. 972 * Because the functions are not allowed in the low-order 4096 bytes of 973 * kernel virtual memory, offsets up to 4095 bytes can be accommodated. 974 * If the offset is larger than 4095 bytes, a compile-time error will 975 * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can 976 * either fall back to use of call_rcu() or rearrange the structure to 977 * position the rcu_head structure into the first 4096 bytes. 978 * 979 * Note that the allowable offset might decrease in the future, for example, 980 * to allow something like kmem_cache_free_rcu(). 981 * 982 * The BUILD_BUG_ON check must not involve any function calls, hence the 983 * checks are done in macros here. 984 */ 985 #define kfree_rcu(ptr, rhf...) kvfree_rcu(ptr, ## rhf) 986 987 /** 988 * kvfree_rcu() - kvfree an object after a grace period. 989 * 990 * This macro consists of one or two arguments and it is 991 * based on whether an object is head-less or not. If it 992 * has a head then a semantic stays the same as it used 993 * to be before: 994 * 995 * kvfree_rcu(ptr, rhf); 996 * 997 * where @ptr is a pointer to kvfree(), @rhf is the name 998 * of the rcu_head structure within the type of @ptr. 999 * 1000 * When it comes to head-less variant, only one argument 1001 * is passed and that is just a pointer which has to be 1002 * freed after a grace period. Therefore the semantic is 1003 * 1004 * kvfree_rcu(ptr); 1005 * 1006 * where @ptr is the pointer to be freed by kvfree(). 1007 * 1008 * Please note, head-less way of freeing is permitted to 1009 * use from a context that has to follow might_sleep() 1010 * annotation. Otherwise, please switch and embed the 1011 * rcu_head structure within the type of @ptr. 1012 */ 1013 #define kvfree_rcu(...) KVFREE_GET_MACRO(__VA_ARGS__, \ 1014 kvfree_rcu_arg_2, kvfree_rcu_arg_1)(__VA_ARGS__) 1015 1016 #define kvfree_rcu_mightsleep(ptr) kvfree_rcu_arg_1(ptr) 1017 #define kfree_rcu_mightsleep(ptr) kvfree_rcu_mightsleep(ptr) 1018 1019 #define KVFREE_GET_MACRO(_1, _2, NAME, ...) NAME 1020 #define kvfree_rcu_arg_2(ptr, rhf) \ 1021 do { \ 1022 typeof (ptr) ___p = (ptr); \ 1023 \ 1024 if (___p) { \ 1025 BUILD_BUG_ON(!__is_kvfree_rcu_offset(offsetof(typeof(*(ptr)), rhf))); \ 1026 kvfree_call_rcu(&((___p)->rhf), (void *) (___p)); \ 1027 } \ 1028 } while (0) 1029 1030 #define kvfree_rcu_arg_1(ptr) \ 1031 do { \ 1032 typeof(ptr) ___p = (ptr); \ 1033 \ 1034 if (___p) \ 1035 kvfree_call_rcu(NULL, (void *) (___p)); \ 1036 } while (0) 1037 1038 /* 1039 * Place this after a lock-acquisition primitive to guarantee that 1040 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies 1041 * if the UNLOCK and LOCK are executed by the same CPU or if the 1042 * UNLOCK and LOCK operate on the same lock variable. 1043 */ 1044 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE 1045 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */ 1046 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 1047 #define smp_mb__after_unlock_lock() do { } while (0) 1048 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */ 1049 1050 1051 /* Has the specified rcu_head structure been handed to call_rcu()? */ 1052 1053 /** 1054 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu() 1055 * @rhp: The rcu_head structure to initialize. 1056 * 1057 * If you intend to invoke rcu_head_after_call_rcu() to test whether a 1058 * given rcu_head structure has already been passed to call_rcu(), then 1059 * you must also invoke this rcu_head_init() function on it just after 1060 * allocating that structure. Calls to this function must not race with 1061 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation. 1062 */ 1063 static inline void rcu_head_init(struct rcu_head *rhp) 1064 { 1065 rhp->func = (rcu_callback_t)~0L; 1066 } 1067 1068 /** 1069 * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()? 1070 * @rhp: The rcu_head structure to test. 1071 * @f: The function passed to call_rcu() along with @rhp. 1072 * 1073 * Returns @true if the @rhp has been passed to call_rcu() with @func, 1074 * and @false otherwise. Emits a warning in any other case, including 1075 * the case where @rhp has already been invoked after a grace period. 1076 * Calls to this function must not race with callback invocation. One way 1077 * to avoid such races is to enclose the call to rcu_head_after_call_rcu() 1078 * in an RCU read-side critical section that includes a read-side fetch 1079 * of the pointer to the structure containing @rhp. 1080 */ 1081 static inline bool 1082 rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f) 1083 { 1084 rcu_callback_t func = READ_ONCE(rhp->func); 1085 1086 if (func == f) 1087 return true; 1088 WARN_ON_ONCE(func != (rcu_callback_t)~0L); 1089 return false; 1090 } 1091 1092 /* kernel/ksysfs.c definitions */ 1093 extern int rcu_expedited; 1094 extern int rcu_normal; 1095 1096 #endif /* __LINUX_RCUPDATE_H */ 1097