1 /* SPDX-License-Identifier: GPL-2.0+ */
2 /*
3 * Read-Copy Update mechanism for mutual exclusion
4 *
5 * Copyright IBM Corporation, 2001
6 *
7 * Author: Dipankar Sarma <dipankar@in.ibm.com>
8 *
9 * Based on the original work by Paul McKenney <paulmck@vnet.ibm.com>
10 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
11 * Papers:
12 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
13 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
14 *
15 * For detailed explanation of Read-Copy Update mechanism see -
16 * http://lse.sourceforge.net/locking/rcupdate.html
17 *
18 */
19
20 #ifndef __LINUX_RCUPDATE_H
21 #define __LINUX_RCUPDATE_H
22
23 #include <linux/types.h>
24 #include <linux/compiler.h>
25 #include <linux/atomic.h>
26 #include <linux/irqflags.h>
27 #include <linux/preempt.h>
28 #include <linux/bottom_half.h>
29 #include <linux/lockdep.h>
30 #include <asm/processor.h>
31 #include <linux/cpumask.h>
32
33 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
34 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
35 #define ulong2long(a) (*(long *)(&(a)))
36
37 /* Exported common interfaces */
38 void call_rcu(struct rcu_head *head, rcu_callback_t func);
39 void rcu_barrier_tasks(void);
40 void synchronize_rcu(void);
41
42 #ifdef CONFIG_PREEMPT_RCU
43
44 void __rcu_read_lock(void);
45 void __rcu_read_unlock(void);
46
47 /*
48 * Defined as a macro as it is a very low level header included from
49 * areas that don't even know about current. This gives the rcu_read_lock()
50 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
51 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
52 */
53 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
54
55 #else /* #ifdef CONFIG_PREEMPT_RCU */
56
57 static inline void __rcu_read_lock(void)
58 {
59 preempt_disable();
60 }
61
62 static inline void __rcu_read_unlock(void)
63 {
64 preempt_enable();
65 }
66
67 static inline int rcu_preempt_depth(void)
68 {
69 return 0;
70 }
71
72 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
73
74 /* Internal to kernel */
75 void rcu_init(void);
76 extern int rcu_scheduler_active __read_mostly;
77 void rcu_sched_clock_irq(int user);
78 void rcu_report_dead(unsigned int cpu);
79 void rcutree_migrate_callbacks(int cpu);
80
81 #ifdef CONFIG_RCU_STALL_COMMON
82 void rcu_sysrq_start(void);
83 void rcu_sysrq_end(void);
84 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
85 static inline void rcu_sysrq_start(void) { }
86 static inline void rcu_sysrq_end(void) { }
87 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
88
89 #ifdef CONFIG_NO_HZ_FULL
90 void rcu_user_enter(void);
91 void rcu_user_exit(void);
92 #else
93 static inline void rcu_user_enter(void) { }
94 static inline void rcu_user_exit(void) { }
95 #endif /* CONFIG_NO_HZ_FULL */
96
97 #ifdef CONFIG_RCU_NOCB_CPU
98 void rcu_init_nohz(void);
99 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
100 static inline void rcu_init_nohz(void) { }
101 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
102
103 /**
104 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
105 * @a: Code that RCU needs to pay attention to.
106 *
107 * RCU read-side critical sections are forbidden in the inner idle loop,
108 * that is, between the rcu_idle_enter() and the rcu_idle_exit() -- RCU
109 * will happily ignore any such read-side critical sections. However,
110 * things like powertop need tracepoints in the inner idle loop.
111 *
112 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
113 * will tell RCU that it needs to pay attention, invoke its argument
114 * (in this example, calling the do_something_with_RCU() function),
115 * and then tell RCU to go back to ignoring this CPU. It is permissible
116 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is
117 * on the order of a million or so, even on 32-bit systems). It is
118 * not legal to block within RCU_NONIDLE(), nor is it permissible to
119 * transfer control either into or out of RCU_NONIDLE()'s statement.
120 */
121 #define RCU_NONIDLE(a) \
122 do { \
123 rcu_irq_enter_irqson(); \
124 do { a; } while (0); \
125 rcu_irq_exit_irqson(); \
126 } while (0)
127
128 /*
129 * Note a quasi-voluntary context switch for RCU-tasks's benefit.
130 * This is a macro rather than an inline function to avoid #include hell.
131 */
132 #ifdef CONFIG_TASKS_RCU
133 #define rcu_tasks_qs(t) \
134 do { \
135 if (READ_ONCE((t)->rcu_tasks_holdout)) \
136 WRITE_ONCE((t)->rcu_tasks_holdout, false); \
137 } while (0)
138 #define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t)
139 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func);
140 void synchronize_rcu_tasks(void);
141 void exit_tasks_rcu_start(void);
142 void exit_tasks_rcu_finish(void);
143 #else /* #ifdef CONFIG_TASKS_RCU */
144 #define rcu_tasks_qs(t) do { } while (0)
145 #define rcu_note_voluntary_context_switch(t) do { } while (0)
146 #define call_rcu_tasks call_rcu
147 #define synchronize_rcu_tasks synchronize_rcu
148 static inline void exit_tasks_rcu_start(void) { }
149 static inline void exit_tasks_rcu_finish(void) { }
150 #endif /* #else #ifdef CONFIG_TASKS_RCU */
151
152 /**
153 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU
154 *
155 * This macro resembles cond_resched(), except that it is defined to
156 * report potential quiescent states to RCU-tasks even if the cond_resched()
157 * machinery were to be shut off, as some advocate for PREEMPT kernels.
158 */
159 #define cond_resched_tasks_rcu_qs() \
160 do { \
161 rcu_tasks_qs(current); \
162 cond_resched(); \
163 } while (0)
164
165 /*
166 * Infrastructure to implement the synchronize_() primitives in
167 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
168 */
169
170 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
171 #include <linux/rcutree.h>
172 #elif defined(CONFIG_TINY_RCU)
173 #include <linux/rcutiny.h>
174 #else
175 #error "Unknown RCU implementation specified to kernel configuration"
176 #endif
177
178 /*
179 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls
180 * are needed for dynamic initialization and destruction of rcu_head
181 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for
182 * dynamic initialization and destruction of statically allocated rcu_head
183 * structures. However, rcu_head structures allocated dynamically in the
184 * heap don't need any initialization.
185 */
186 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
187 void init_rcu_head(struct rcu_head *head);
188 void destroy_rcu_head(struct rcu_head *head);
189 void init_rcu_head_on_stack(struct rcu_head *head);
190 void destroy_rcu_head_on_stack(struct rcu_head *head);
191 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
192 static inline void init_rcu_head(struct rcu_head *head) { }
193 static inline void destroy_rcu_head(struct rcu_head *head) { }
194 static inline void init_rcu_head_on_stack(struct rcu_head *head) { }
195 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { }
196 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
197
198 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
199 bool rcu_lockdep_current_cpu_online(void);
200 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
201 static inline bool rcu_lockdep_current_cpu_online(void) { return true; }
202 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
203
204 #ifdef CONFIG_DEBUG_LOCK_ALLOC
205
206 static inline void rcu_lock_acquire(struct lockdep_map *map)
207 {
208 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
209 }
210
211 static inline void rcu_lock_release(struct lockdep_map *map)
212 {
213 lock_release(map, 1, _THIS_IP_);
214 }
215
216 extern struct lockdep_map rcu_lock_map;
217 extern struct lockdep_map rcu_bh_lock_map;
218 extern struct lockdep_map rcu_sched_lock_map;
219 extern struct lockdep_map rcu_callback_map;
220 int debug_lockdep_rcu_enabled(void);
221 int rcu_read_lock_held(void);
222 int rcu_read_lock_bh_held(void);
223 int rcu_read_lock_sched_held(void);
224 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_swap_protected() - swap an RCU and a regular pointer
387 * @rcu_ptr: RCU pointer
388 * @ptr: regular pointer
389 * @c: the conditions under which the dereference will take place
390 *
391 * Perform swap(@rcu_ptr, @ptr) where @rcu_ptr is an RCU-annotated pointer and
392 * @c is the argument that is passed to the rcu_dereference_protected() call
393 * used to read that pointer.
394 */
395 #define rcu_swap_protected(rcu_ptr, ptr, c) do { \
396 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \
397 rcu_assign_pointer((rcu_ptr), (ptr)); \
398 (ptr) = __tmp; \
399 } while (0)
400
401 /**
402 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
403 * @p: The pointer to read
404 *
405 * Return the value of the specified RCU-protected pointer, but omit the
406 * lockdep checks for being in an RCU read-side critical section. This is
407 * useful when the value of this pointer is accessed, but the pointer is
408 * not dereferenced, for example, when testing an RCU-protected pointer
409 * against NULL. Although rcu_access_pointer() may also be used in cases
410 * where update-side locks prevent the value of the pointer from changing,
411 * you should instead use rcu_dereference_protected() for this use case.
412 *
413 * It is also permissible to use rcu_access_pointer() when read-side
414 * access to the pointer was removed at least one grace period ago, as
415 * is the case in the context of the RCU callback that is freeing up
416 * the data, or after a synchronize_rcu() returns. This can be useful
417 * when tearing down multi-linked structures after a grace period
418 * has elapsed.
419 */
420 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
421
422 /**
423 * rcu_dereference_check() - rcu_dereference with debug checking
424 * @p: The pointer to read, prior to dereferencing
425 * @c: The conditions under which the dereference will take place
426 *
427 * Do an rcu_dereference(), but check that the conditions under which the
428 * dereference will take place are correct. Typically the conditions
429 * indicate the various locking conditions that should be held at that
430 * point. The check should return true if the conditions are satisfied.
431 * An implicit check for being in an RCU read-side critical section
432 * (rcu_read_lock()) is included.
433 *
434 * For example:
435 *
436 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
437 *
438 * could be used to indicate to lockdep that foo->bar may only be dereferenced
439 * if either rcu_read_lock() is held, or that the lock required to replace
440 * the bar struct at foo->bar is held.
441 *
442 * Note that the list of conditions may also include indications of when a lock
443 * need not be held, for example during initialisation or destruction of the
444 * target struct:
445 *
446 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
447 * atomic_read(&foo->usage) == 0);
448 *
449 * Inserts memory barriers on architectures that require them
450 * (currently only the Alpha), prevents the compiler from refetching
451 * (and from merging fetches), and, more importantly, documents exactly
452 * which pointers are protected by RCU and checks that the pointer is
453 * annotated as __rcu.
454 */
455 #define rcu_dereference_check(p, c) \
456 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu)
457
458 /**
459 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
460 * @p: The pointer to read, prior to dereferencing
461 * @c: The conditions under which the dereference will take place
462 *
463 * This is the RCU-bh counterpart to rcu_dereference_check().
464 */
465 #define rcu_dereference_bh_check(p, c) \
466 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu)
467
468 /**
469 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
470 * @p: The pointer to read, prior to dereferencing
471 * @c: The conditions under which the dereference will take place
472 *
473 * This is the RCU-sched counterpart to rcu_dereference_check().
474 */
475 #define rcu_dereference_sched_check(p, c) \
476 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \
477 __rcu)
478
479 /*
480 * The tracing infrastructure traces RCU (we want that), but unfortunately
481 * some of the RCU checks causes tracing to lock up the system.
482 *
483 * The no-tracing version of rcu_dereference_raw() must not call
484 * rcu_read_lock_held().
485 */
486 #define rcu_dereference_raw_check(p) __rcu_dereference_check((p), 1, __rcu)
487
488 /**
489 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
490 * @p: The pointer to read, prior to dereferencing
491 * @c: The conditions under which the dereference will take place
492 *
493 * Return the value of the specified RCU-protected pointer, but omit
494 * the READ_ONCE(). This is useful in cases where update-side locks
495 * prevent the value of the pointer from changing. Please note that this
496 * primitive does *not* prevent the compiler from repeating this reference
497 * or combining it with other references, so it should not be used without
498 * protection of appropriate locks.
499 *
500 * This function is only for update-side use. Using this function
501 * when protected only by rcu_read_lock() will result in infrequent
502 * but very ugly failures.
503 */
504 #define rcu_dereference_protected(p, c) \
505 __rcu_dereference_protected((p), (c), __rcu)
506
507
508 /**
509 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
510 * @p: The pointer to read, prior to dereferencing
511 *
512 * This is a simple wrapper around rcu_dereference_check().
513 */
514 #define rcu_dereference(p) rcu_dereference_check(p, 0)
515
516 /**
517 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
518 * @p: The pointer to read, prior to dereferencing
519 *
520 * Makes rcu_dereference_check() do the dirty work.
521 */
522 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
523
524 /**
525 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
526 * @p: The pointer to read, prior to dereferencing
527 *
528 * Makes rcu_dereference_check() do the dirty work.
529 */
530 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
531
532 /**
533 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
534 * @p: The pointer to hand off
535 *
536 * This is simply an identity function, but it documents where a pointer
537 * is handed off from RCU to some other synchronization mechanism, for
538 * example, reference counting or locking. In C11, it would map to
539 * kill_dependency(). It could be used as follows::
540 *
541 * rcu_read_lock();
542 * p = rcu_dereference(gp);
543 * long_lived = is_long_lived(p);
544 * if (long_lived) {
545 * if (!atomic_inc_not_zero(p->refcnt))
546 * long_lived = false;
547 * else
548 * p = rcu_pointer_handoff(p);
549 * }
550 * rcu_read_unlock();
551 */
552 #define rcu_pointer_handoff(p) (p)
553
554 /**
555 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
556 *
557 * When synchronize_rcu() is invoked on one CPU while other CPUs
558 * are within RCU read-side critical sections, then the
559 * synchronize_rcu() is guaranteed to block until after all the other
560 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
561 * on one CPU while other CPUs are within RCU read-side critical
562 * sections, invocation of the corresponding RCU callback is deferred
563 * until after the all the other CPUs exit their critical sections.
564 *
565 * Note, however, that RCU callbacks are permitted to run concurrently
566 * with new RCU read-side critical sections. One way that this can happen
567 * is via the following sequence of events: (1) CPU 0 enters an RCU
568 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
569 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
570 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
571 * callback is invoked. This is legal, because the RCU read-side critical
572 * section that was running concurrently with the call_rcu() (and which
573 * therefore might be referencing something that the corresponding RCU
574 * callback would free up) has completed before the corresponding
575 * RCU callback is invoked.
576 *
577 * RCU read-side critical sections may be nested. Any deferred actions
578 * will be deferred until the outermost RCU read-side critical section
579 * completes.
580 *
581 * You can avoid reading and understanding the next paragraph by
582 * following this rule: don't put anything in an rcu_read_lock() RCU
583 * read-side critical section that would block in a !PREEMPT kernel.
584 * But if you want the full story, read on!
585 *
586 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU),
587 * it is illegal to block while in an RCU read-side critical section.
588 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION
589 * kernel builds, RCU read-side critical sections may be preempted,
590 * but explicit blocking is illegal. Finally, in preemptible RCU
591 * implementations in real-time (with -rt patchset) kernel builds, RCU
592 * read-side critical sections may be preempted and they may also block, but
593 * only when acquiring spinlocks that are subject to priority inheritance.
594 */
595 static __always_inline void rcu_read_lock(void)
596 {
597 __rcu_read_lock();
598 __acquire(RCU);
599 rcu_lock_acquire(&rcu_lock_map);
600 RCU_LOCKDEP_WARN(!rcu_is_watching(),
601 "rcu_read_lock() used illegally while idle");
602 }
603
604 /*
605 * So where is rcu_write_lock()? It does not exist, as there is no
606 * way for writers to lock out RCU readers. This is a feature, not
607 * a bug -- this property is what provides RCU's performance benefits.
608 * Of course, writers must coordinate with each other. The normal
609 * spinlock primitives work well for this, but any other technique may be
610 * used as well. RCU does not care how the writers keep out of each
611 * others' way, as long as they do so.
612 */
613
614 /**
615 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
616 *
617 * In most situations, rcu_read_unlock() is immune from deadlock.
618 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock()
619 * is responsible for deboosting, which it does via rt_mutex_unlock().
620 * Unfortunately, this function acquires the scheduler's runqueue and
621 * priority-inheritance spinlocks. This means that deadlock could result
622 * if the caller of rcu_read_unlock() already holds one of these locks or
623 * any lock that is ever acquired while holding them.
624 *
625 * That said, RCU readers are never priority boosted unless they were
626 * preempted. Therefore, one way to avoid deadlock is to make sure
627 * that preemption never happens within any RCU read-side critical
628 * section whose outermost rcu_read_unlock() is called with one of
629 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in
630 * a number of ways, for example, by invoking preempt_disable() before
631 * critical section's outermost rcu_read_lock().
632 *
633 * Given that the set of locks acquired by rt_mutex_unlock() might change
634 * at any time, a somewhat more future-proofed approach is to make sure
635 * that that preemption never happens within any RCU read-side critical
636 * section whose outermost rcu_read_unlock() is called with irqs disabled.
637 * This approach relies on the fact that rt_mutex_unlock() currently only
638 * acquires irq-disabled locks.
639 *
640 * The second of these two approaches is best in most situations,
641 * however, the first approach can also be useful, at least to those
642 * developers willing to keep abreast of the set of locks acquired by
643 * rt_mutex_unlock().
644 *
645 * See rcu_read_lock() for more information.
646 */
647 static inline void rcu_read_unlock(void)
648 {
649 RCU_LOCKDEP_WARN(!rcu_is_watching(),
650 "rcu_read_unlock() used illegally while idle");
651 __release(RCU);
652 __rcu_read_unlock();
653 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
654 }
655
656 /**
657 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
658 *
659 * This is equivalent of rcu_read_lock(), but also disables softirqs.
660 * Note that anything else that disables softirqs can also serve as
661 * an RCU read-side critical section.
662 *
663 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
664 * must occur in the same context, for example, it is illegal to invoke
665 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
666 * was invoked from some other task.
667 */
668 static inline void rcu_read_lock_bh(void)
669 {
670 local_bh_disable();
671 __acquire(RCU_BH);
672 rcu_lock_acquire(&rcu_bh_lock_map);
673 RCU_LOCKDEP_WARN(!rcu_is_watching(),
674 "rcu_read_lock_bh() used illegally while idle");
675 }
676
677 /*
678 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
679 *
680 * See rcu_read_lock_bh() for more information.
681 */
682 static inline void rcu_read_unlock_bh(void)
683 {
684 RCU_LOCKDEP_WARN(!rcu_is_watching(),
685 "rcu_read_unlock_bh() used illegally while idle");
686 rcu_lock_release(&rcu_bh_lock_map);
687 __release(RCU_BH);
688 local_bh_enable();
689 }
690
691 /**
692 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
693 *
694 * This is equivalent of rcu_read_lock(), but disables preemption.
695 * Read-side critical sections can also be introduced by anything else
696 * that disables preemption, including local_irq_disable() and friends.
697 *
698 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
699 * must occur in the same context, for example, it is illegal to invoke
700 * rcu_read_unlock_sched() from process context if the matching
701 * rcu_read_lock_sched() was invoked from an NMI handler.
702 */
703 static inline void rcu_read_lock_sched(void)
704 {
705 preempt_disable();
706 __acquire(RCU_SCHED);
707 rcu_lock_acquire(&rcu_sched_lock_map);
708 RCU_LOCKDEP_WARN(!rcu_is_watching(),
709 "rcu_read_lock_sched() used illegally while idle");
710 }
711
712 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
713 static inline notrace void rcu_read_lock_sched_notrace(void)
714 {
715 preempt_disable_notrace();
716 __acquire(RCU_SCHED);
717 }
718
719 /*
720 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
721 *
722 * See rcu_read_lock_sched for more information.
723 */
724 static inline void rcu_read_unlock_sched(void)
725 {
726 RCU_LOCKDEP_WARN(!rcu_is_watching(),
727 "rcu_read_unlock_sched() used illegally while idle");
728 rcu_lock_release(&rcu_sched_lock_map);
729 __release(RCU_SCHED);
730 preempt_enable();
731 }
732
733 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
734 static inline notrace void rcu_read_unlock_sched_notrace(void)
735 {
736 __release(RCU_SCHED);
737 preempt_enable_notrace();
738 }
739
740 /**
741 * RCU_INIT_POINTER() - initialize an RCU protected pointer
742 * @p: The pointer to be initialized.
743 * @v: The value to initialized the pointer to.
744 *
745 * Initialize an RCU-protected pointer in special cases where readers
746 * do not need ordering constraints on the CPU or the compiler. These
747 * special cases are:
748 *
749 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or*
750 * 2. The caller has taken whatever steps are required to prevent
751 * RCU readers from concurrently accessing this pointer *or*
752 * 3. The referenced data structure has already been exposed to
753 * readers either at compile time or via rcu_assign_pointer() *and*
754 *
755 * a. You have not made *any* reader-visible changes to
756 * this structure since then *or*
757 * b. It is OK for readers accessing this structure from its
758 * new location to see the old state of the structure. (For
759 * example, the changes were to statistical counters or to
760 * other state where exact synchronization is not required.)
761 *
762 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
763 * result in impossible-to-diagnose memory corruption. As in the structures
764 * will look OK in crash dumps, but any concurrent RCU readers might
765 * see pre-initialized values of the referenced data structure. So
766 * please be very careful how you use RCU_INIT_POINTER()!!!
767 *
768 * If you are creating an RCU-protected linked structure that is accessed
769 * by a single external-to-structure RCU-protected pointer, then you may
770 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
771 * pointers, but you must use rcu_assign_pointer() to initialize the
772 * external-to-structure pointer *after* you have completely initialized
773 * the reader-accessible portions of the linked structure.
774 *
775 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
776 * ordering guarantees for either the CPU or the compiler.
777 */
778 #define RCU_INIT_POINTER(p, v) \
779 do { \
780 rcu_check_sparse(p, __rcu); \
781 WRITE_ONCE(p, RCU_INITIALIZER(v)); \
782 } while (0)
783
784 /**
785 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
786 * @p: The pointer to be initialized.
787 * @v: The value to initialized the pointer to.
788 *
789 * GCC-style initialization for an RCU-protected pointer in a structure field.
790 */
791 #define RCU_POINTER_INITIALIZER(p, v) \
792 .p = RCU_INITIALIZER(v)
793
794 /*
795 * Does the specified offset indicate that the corresponding rcu_head
796 * structure can be handled by kfree_rcu()?
797 */
798 #define __is_kfree_rcu_offset(offset) ((offset) < 4096)
799
800 /*
801 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
802 */
803 #define __kfree_rcu(head, offset) \
804 do { \
805 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
806 kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \
807 } while (0)
808
809 /**
810 * kfree_rcu() - kfree an object after a grace period.
811 * @ptr: pointer to kfree
812 * @rhf: the name of the struct rcu_head within the type of @ptr.
813 *
814 * Many rcu callbacks functions just call kfree() on the base structure.
815 * These functions are trivial, but their size adds up, and furthermore
816 * when they are used in a kernel module, that module must invoke the
817 * high-latency rcu_barrier() function at module-unload time.
818 *
819 * The kfree_rcu() function handles this issue. Rather than encoding a
820 * function address in the embedded rcu_head structure, kfree_rcu() instead
821 * encodes the offset of the rcu_head structure within the base structure.
822 * Because the functions are not allowed in the low-order 4096 bytes of
823 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
824 * If the offset is larger than 4095 bytes, a compile-time error will
825 * be generated in __kfree_rcu(). If this error is triggered, you can
826 * either fall back to use of call_rcu() or rearrange the structure to
827 * position the rcu_head structure into the first 4096 bytes.
828 *
829 * Note that the allowable offset might decrease in the future, for example,
830 * to allow something like kmem_cache_free_rcu().
831 *
832 * The BUILD_BUG_ON check must not involve any function calls, hence the
833 * checks are done in macros here.
834 */
835 #define kfree_rcu(ptr, rhf) \
836 do { \
837 typeof (ptr) ___p = (ptr); \
838 \
839 if (___p) \
840 __kfree_rcu(&((___p)->rhf), offsetof(typeof(*(ptr)), rhf)); \
841 } while (0)
842
843 /*
844 * Place this after a lock-acquisition primitive to guarantee that
845 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies
846 * if the UNLOCK and LOCK are executed by the same CPU or if the
847 * UNLOCK and LOCK operate on the same lock variable.
848 */
849 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
850 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */
851 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
852 #define smp_mb__after_unlock_lock() do { } while (0)
853 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
854
855
856 /* Has the specified rcu_head structure been handed to call_rcu()? */
857
858 /**
859 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu()
860 * @rhp: The rcu_head structure to initialize.
861 *
862 * If you intend to invoke rcu_head_after_call_rcu() to test whether a
863 * given rcu_head structure has already been passed to call_rcu(), then
864 * you must also invoke this rcu_head_init() function on it just after
865 * allocating that structure. Calls to this function must not race with
866 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation.
867 */
868 static inline void rcu_head_init(struct rcu_head *rhp)
869 {
870 rhp->func = (rcu_callback_t)~0L;
871 }
872
873 /**
874 * rcu_head_after_call_rcu - Has this rcu_head been passed to call_rcu()?
875 * @rhp: The rcu_head structure to test.
876 * @f: The function passed to call_rcu() along with @rhp.
877 *
878 * Returns @true if the @rhp has been passed to call_rcu() with @func,
879 * and @false otherwise. Emits a warning in any other case, including
880 * the case where @rhp has already been invoked after a grace period.
881 * Calls to this function must not race with callback invocation. One way
882 * to avoid such races is to enclose the call to rcu_head_after_call_rcu()
883 * in an RCU read-side critical section that includes a read-side fetch
884 * of the pointer to the structure containing @rhp.
885 */
886 static inline bool
887 rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f)
888 {
889 rcu_callback_t func = READ_ONCE(rhp->func);
890
891 if (func == f)
892 return true;
893 WARN_ON_ONCE(func != (rcu_callback_t)~0L);
894 return false;
895 }
896
897 #endif /* __LINUX_RCUPDATE_H */