1/*
2 * Read-Copy Update mechanism for mutual exclusion
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, you can access it online at
16 * http://www.gnu.org/licenses/gpl-2.0.html.
17 *
18 * Copyright IBM Corporation, 2001
19 *
20 * Author: Dipankar Sarma <dipankar@in.ibm.com>
21 *
22 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
23 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
24 * Papers:
25 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
26 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
27 *
28 * For detailed explanation of Read-Copy Update mechanism see -
29 * http://lse.sourceforge.net/locking/rcupdate.html
30 *
31 */
32
33#ifndef __LINUX_RCUPDATE_H
34#define __LINUX_RCUPDATE_H
35
36#include <linux/types.h>
37#include <linux/compiler.h>
38#include <linux/atomic.h>
39#include <linux/irqflags.h>
40#include <linux/preempt.h>
41#include <linux/bottom_half.h>
42#include <linux/lockdep.h>
43#include <asm/processor.h>
44#include <linux/cpumask.h>
45
46#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
47#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
48#define ulong2long(a) (*(long *)(&(a)))
49
50/* Exported common interfaces */
51
52#ifdef CONFIG_PREEMPT_RCU
53void call_rcu(struct rcu_head *head, rcu_callback_t func);
54#else /* #ifdef CONFIG_PREEMPT_RCU */
55#define call_rcu call_rcu_sched
56#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
57
58void call_rcu_bh(struct rcu_head *head, rcu_callback_t func);
59void call_rcu_sched(struct rcu_head *head, rcu_callback_t func);
60void synchronize_sched(void);
61void rcu_barrier_tasks(void);
62
63#ifdef CONFIG_PREEMPT_RCU
64
65void __rcu_read_lock(void);
66void __rcu_read_unlock(void);
67void synchronize_rcu(void);
68
69/*
70 * Defined as a macro as it is a very low level header included from
71 * areas that don't even know about current. This gives the rcu_read_lock()
72 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
73 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
74 */
75#define rcu_preempt_depth() (current->rcu_read_lock_nesting)
76
77#else /* #ifdef CONFIG_PREEMPT_RCU */
78
79static inline void __rcu_read_lock(void)
80{
81 if (IS_ENABLED(CONFIG_PREEMPT_COUNT))
82 preempt_disable();
83}
84
85static inline void __rcu_read_unlock(void)
86{
87 if (IS_ENABLED(CONFIG_PREEMPT_COUNT))
88 preempt_enable();
89}
90
91static inline void synchronize_rcu(void)
92{
93 synchronize_sched();
94}
95
96static inline int rcu_preempt_depth(void)
97{
98 return 0;
99}
100
101#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
102
103/* Internal to kernel */
104void rcu_init(void);
105extern int rcu_scheduler_active __read_mostly;
106void rcu_sched_qs(void);
107void rcu_bh_qs(void);
108void rcu_check_callbacks(int user);
109void rcu_report_dead(unsigned int cpu);
110void rcutree_migrate_callbacks(int cpu);
111
112#ifdef CONFIG_RCU_STALL_COMMON
113void rcu_sysrq_start(void);
114void rcu_sysrq_end(void);
115#else /* #ifdef CONFIG_RCU_STALL_COMMON */
116static inline void rcu_sysrq_start(void) { }
117static inline void rcu_sysrq_end(void) { }
118#endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
119
120#ifdef CONFIG_NO_HZ_FULL
121void rcu_user_enter(void);
122void rcu_user_exit(void);
123#else
124static inline void rcu_user_enter(void) { }
125static inline void rcu_user_exit(void) { }
126#endif /* CONFIG_NO_HZ_FULL */
127
128#ifdef CONFIG_RCU_NOCB_CPU
129void rcu_init_nohz(void);
130#else /* #ifdef CONFIG_RCU_NOCB_CPU */
131static inline void rcu_init_nohz(void) { }
132#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
133
134/**
135 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
136 * @a: Code that RCU needs to pay attention to.
137 *
138 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
139 * in the inner idle loop, that is, between the rcu_idle_enter() and
140 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
141 * critical sections. However, things like powertop need tracepoints
142 * in the inner idle loop.
143 *
144 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
145 * will tell RCU that it needs to pay attention, invoke its argument
146 * (in this example, calling the do_something_with_RCU() function),
147 * and then tell RCU to go back to ignoring this CPU. It is permissible
148 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is
149 * on the order of a million or so, even on 32-bit systems). It is
150 * not legal to block within RCU_NONIDLE(), nor is it permissible to
151 * transfer control either into or out of RCU_NONIDLE()'s statement.
152 */
153#define RCU_NONIDLE(a) \
154 do { \
155 rcu_irq_enter_irqson(); \
156 do { a; } while (0); \
157 rcu_irq_exit_irqson(); \
158 } while (0)
159
160/*
161 * Note a quasi-voluntary context switch for RCU-tasks's benefit.
162 * This is a macro rather than an inline function to avoid #include hell.
163 */
164#ifdef CONFIG_TASKS_RCU
165#define rcu_tasks_qs(t) \
166 do { \
167 if (READ_ONCE((t)->rcu_tasks_holdout)) \
168 WRITE_ONCE((t)->rcu_tasks_holdout, false); \
169 } while (0)
170#define rcu_note_voluntary_context_switch(t) \
171 do { \
172 rcu_all_qs(); \
173 rcu_tasks_qs(t); \
174 } while (0)
175void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func);
176void synchronize_rcu_tasks(void);
177void exit_tasks_rcu_start(void);
178void exit_tasks_rcu_finish(void);
179#else /* #ifdef CONFIG_TASKS_RCU */
180#define rcu_tasks_qs(t) do { } while (0)
181#define rcu_note_voluntary_context_switch(t) rcu_all_qs()
182#define call_rcu_tasks call_rcu_sched
183#define synchronize_rcu_tasks synchronize_sched
184static inline void exit_tasks_rcu_start(void) { }
185static inline void exit_tasks_rcu_finish(void) { }
186#endif /* #else #ifdef CONFIG_TASKS_RCU */
187
188/**
189 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU
190 *
191 * This macro resembles cond_resched(), except that it is defined to
192 * report potential quiescent states to RCU-tasks even if the cond_resched()
193 * machinery were to be shut off, as some advocate for PREEMPT kernels.
194 */
195#define cond_resched_tasks_rcu_qs() \
196do { \
197 rcu_tasks_qs(current); \
198 cond_resched(); \
199} while (0)
200
201/*
202 * Infrastructure to implement the synchronize_() primitives in
203 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
204 */
205
206#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
207#include <linux/rcutree.h>
208#elif defined(CONFIG_TINY_RCU)
209#include <linux/rcutiny.h>
210#else
211#error "Unknown RCU implementation specified to kernel configuration"
212#endif
213
214/*
215 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls
216 * are needed for dynamic initialization and destruction of rcu_head
217 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for
218 * dynamic initialization and destruction of statically allocated rcu_head
219 * structures. However, rcu_head structures allocated dynamically in the
220 * heap don't need any initialization.
221 */
222#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
223void init_rcu_head(struct rcu_head *head);
224void destroy_rcu_head(struct rcu_head *head);
225void init_rcu_head_on_stack(struct rcu_head *head);
226void destroy_rcu_head_on_stack(struct rcu_head *head);
227#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
228static inline void init_rcu_head(struct rcu_head *head) { }
229static inline void destroy_rcu_head(struct rcu_head *head) { }
230static inline void init_rcu_head_on_stack(struct rcu_head *head) { }
231static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { }
232#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
233
234#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
235bool rcu_lockdep_current_cpu_online(void);
236#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
237static inline bool rcu_lockdep_current_cpu_online(void) { return true; }
238#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
239
240#ifdef CONFIG_DEBUG_LOCK_ALLOC
241
242static inline void rcu_lock_acquire(struct lockdep_map *map)
243{
244 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
245}
246
247static inline void rcu_lock_release(struct lockdep_map *map)
248{
249 lock_release(map, 1, _THIS_IP_);
250}
251
252extern struct lockdep_map rcu_lock_map;
253extern struct lockdep_map rcu_bh_lock_map;
254extern struct lockdep_map rcu_sched_lock_map;
255extern struct lockdep_map rcu_callback_map;
256int debug_lockdep_rcu_enabled(void);
257int rcu_read_lock_held(void);
258int rcu_read_lock_bh_held(void);
259int rcu_read_lock_sched_held(void);
260
261#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
262
263# define rcu_lock_acquire(a) do { } while (0)
264# define rcu_lock_release(a) do { } while (0)
265
266static inline int rcu_read_lock_held(void)
267{
268 return 1;
269}
270
271static inline int rcu_read_lock_bh_held(void)
272{
273 return 1;
274}
275
276static inline int rcu_read_lock_sched_held(void)
277{
278 return !preemptible();
279}
280#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
281
282#ifdef CONFIG_PROVE_RCU
283
284/**
285 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
286 * @c: condition to check
287 * @s: informative message
288 */
289#define RCU_LOCKDEP_WARN(c, s) \
290 do { \
291 static bool __section(.data.unlikely) __warned; \
292 if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \
293 __warned = true; \
294 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
295 } \
296 } while (0)
297
298#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
299static inline void rcu_preempt_sleep_check(void)
300{
301 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map),
302 "Illegal context switch in RCU read-side critical section");
303}
304#else /* #ifdef CONFIG_PROVE_RCU */
305static inline void rcu_preempt_sleep_check(void) { }
306#endif /* #else #ifdef CONFIG_PROVE_RCU */
307
308#define rcu_sleep_check() \
309 do { \
310 rcu_preempt_sleep_check(); \
311 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \
312 "Illegal context switch in RCU-bh read-side critical section"); \
313 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \
314 "Illegal context switch in RCU-sched read-side critical section"); \
315 } while (0)
316
317#else /* #ifdef CONFIG_PROVE_RCU */
318
319#define RCU_LOCKDEP_WARN(c, s) do { } while (0)
320#define rcu_sleep_check() do { } while (0)
321
322#endif /* #else #ifdef CONFIG_PROVE_RCU */
323
324/*
325 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
326 * and rcu_assign_pointer(). Some of these could be folded into their
327 * callers, but they are left separate in order to ease introduction of
328 * multiple flavors of pointers to match the multiple flavors of RCU
329 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
330 * the future.
331 */
332
333#ifdef __CHECKER__
334#define rcu_dereference_sparse(p, space) \
335 ((void)(((typeof(*p) space *)p) == p))
336#else /* #ifdef __CHECKER__ */
337#define rcu_dereference_sparse(p, space)
338#endif /* #else #ifdef __CHECKER__ */
339
340#define __rcu_access_pointer(p, space) \
341({ \
342 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \
343 rcu_dereference_sparse(p, space); \
344 ((typeof(*p) __force __kernel *)(_________p1)); \
345})
346#define __rcu_dereference_check(p, c, space) \
347({ \
348 /* Dependency order vs. p above. */ \
349 typeof(*p) *________p1 = (typeof(*p) *__force)READ_ONCE(p); \
350 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
351 rcu_dereference_sparse(p, space); \
352 ((typeof(*p) __force __kernel *)(________p1)); \
353})
354#define __rcu_dereference_protected(p, c, space) \
355({ \
356 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
357 rcu_dereference_sparse(p, space); \
358 ((typeof(*p) __force __kernel *)(p)); \
359})
360#define rcu_dereference_raw(p) \
361({ \
362 /* Dependency order vs. p above. */ \
363 typeof(p) ________p1 = READ_ONCE(p); \
364 ((typeof(*p) __force __kernel *)(________p1)); \
365})
366
367/**
368 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
369 * @v: The value to statically initialize with.
370 */
371#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
372
373/**
374 * rcu_assign_pointer() - assign to RCU-protected pointer
375 * @p: pointer to assign to
376 * @v: value to assign (publish)
377 *
378 * Assigns the specified value to the specified RCU-protected
379 * pointer, ensuring that any concurrent RCU readers will see
380 * any prior initialization.
381 *
382 * Inserts memory barriers on architectures that require them
383 * (which is most of them), and also prevents the compiler from
384 * reordering the code that initializes the structure after the pointer
385 * assignment. More importantly, this call documents which pointers
386 * will be dereferenced by RCU read-side code.
387 *
388 * In some special cases, you may use RCU_INIT_POINTER() instead
389 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
390 * to the fact that it does not constrain either the CPU or the compiler.
391 * That said, using RCU_INIT_POINTER() when you should have used
392 * rcu_assign_pointer() is a very bad thing that results in
393 * impossible-to-diagnose memory corruption. So please be careful.
394 * See the RCU_INIT_POINTER() comment header for details.
395 *
396 * Note that rcu_assign_pointer() evaluates each of its arguments only
397 * once, appearances notwithstanding. One of the "extra" evaluations
398 * is in typeof() and the other visible only to sparse (__CHECKER__),
399 * neither of which actually execute the argument. As with most cpp
400 * macros, this execute-arguments-only-once property is important, so
401 * please be careful when making changes to rcu_assign_pointer() and the
402 * other macros that it invokes.
403 */
404#define rcu_assign_pointer(p, v) \
405({ \
406 uintptr_t _r_a_p__v = (uintptr_t)(v); \
407 \
408 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \
409 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \
410 else \
411 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
412 _r_a_p__v; \
413})
414
415/**
416 * rcu_swap_protected() - swap an RCU and a regular pointer
417 * @rcu_ptr: RCU pointer
418 * @ptr: regular pointer
419 * @c: the conditions under which the dereference will take place
420 *
421 * Perform swap(@rcu_ptr, @ptr) where @rcu_ptr is an RCU-annotated pointer and
422 * @c is the argument that is passed to the rcu_dereference_protected() call
423 * used to read that pointer.
424 */
425#define rcu_swap_protected(rcu_ptr, ptr, c) do { \
426 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \
427 rcu_assign_pointer((rcu_ptr), (ptr)); \
428 (ptr) = __tmp; \
429} while (0)
430
431/**
432 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
433 * @p: The pointer to read
434 *
435 * Return the value of the specified RCU-protected pointer, but omit the
436 * lockdep checks for being in an RCU read-side critical section. This is
437 * useful when the value of this pointer is accessed, but the pointer is
438 * not dereferenced, for example, when testing an RCU-protected pointer
439 * against NULL. Although rcu_access_pointer() may also be used in cases
440 * where update-side locks prevent the value of the pointer from changing,
441 * you should instead use rcu_dereference_protected() for this use case.
442 *
443 * It is also permissible to use rcu_access_pointer() when read-side
444 * access to the pointer was removed at least one grace period ago, as
445 * is the case in the context of the RCU callback that is freeing up
446 * the data, or after a synchronize_rcu() returns. This can be useful
447 * when tearing down multi-linked structures after a grace period
448 * has elapsed.
449 */
450#define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
451
452/**
453 * rcu_dereference_check() - rcu_dereference with debug checking
454 * @p: The pointer to read, prior to dereferencing
455 * @c: The conditions under which the dereference will take place
456 *
457 * Do an rcu_dereference(), but check that the conditions under which the
458 * dereference will take place are correct. Typically the conditions
459 * indicate the various locking conditions that should be held at that
460 * point. The check should return true if the conditions are satisfied.
461 * An implicit check for being in an RCU read-side critical section
462 * (rcu_read_lock()) is included.
463 *
464 * For example:
465 *
466 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
467 *
468 * could be used to indicate to lockdep that foo->bar may only be dereferenced
469 * if either rcu_read_lock() is held, or that the lock required to replace
470 * the bar struct at foo->bar is held.
471 *
472 * Note that the list of conditions may also include indications of when a lock
473 * need not be held, for example during initialisation or destruction of the
474 * target struct:
475 *
476 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
477 * atomic_read(&foo->usage) == 0);
478 *
479 * Inserts memory barriers on architectures that require them
480 * (currently only the Alpha), prevents the compiler from refetching
481 * (and from merging fetches), and, more importantly, documents exactly
482 * which pointers are protected by RCU and checks that the pointer is
483 * annotated as __rcu.
484 */
485#define rcu_dereference_check(p, c) \
486 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu)
487
488/**
489 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
490 * @p: The pointer to read, prior to dereferencing
491 * @c: The conditions under which the dereference will take place
492 *
493 * This is the RCU-bh counterpart to rcu_dereference_check().
494 */
495#define rcu_dereference_bh_check(p, c) \
496 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu)
497
498/**
499 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
500 * @p: The pointer to read, prior to dereferencing
501 * @c: The conditions under which the dereference will take place
502 *
503 * This is the RCU-sched counterpart to rcu_dereference_check().
504 */
505#define rcu_dereference_sched_check(p, c) \
506 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \
507 __rcu)
508
509/*
510 * The tracing infrastructure traces RCU (we want that), but unfortunately
511 * some of the RCU checks causes tracing to lock up the system.
512 *
513 * The no-tracing version of rcu_dereference_raw() must not call
514 * rcu_read_lock_held().
515 */
516#define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
517
518/**
519 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
520 * @p: The pointer to read, prior to dereferencing
521 * @c: The conditions under which the dereference will take place
522 *
523 * Return the value of the specified RCU-protected pointer, but omit
524 * the READ_ONCE(). This is useful in cases where update-side locks
525 * prevent the value of the pointer from changing. Please note that this
526 * primitive does *not* prevent the compiler from repeating this reference
527 * or combining it with other references, so it should not be used without
528 * protection of appropriate locks.
529 *
530 * This function is only for update-side use. Using this function
531 * when protected only by rcu_read_lock() will result in infrequent
532 * but very ugly failures.
533 */
534#define rcu_dereference_protected(p, c) \
535 __rcu_dereference_protected((p), (c), __rcu)
536
537
538/**
539 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
540 * @p: The pointer to read, prior to dereferencing
541 *
542 * This is a simple wrapper around rcu_dereference_check().
543 */
544#define rcu_dereference(p) rcu_dereference_check(p, 0)
545
546/**
547 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
548 * @p: The pointer to read, prior to dereferencing
549 *
550 * Makes rcu_dereference_check() do the dirty work.
551 */
552#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
553
554/**
555 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
556 * @p: The pointer to read, prior to dereferencing
557 *
558 * Makes rcu_dereference_check() do the dirty work.
559 */
560#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
561
562/**
563 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
564 * @p: The pointer to hand off
565 *
566 * This is simply an identity function, but it documents where a pointer
567 * is handed off from RCU to some other synchronization mechanism, for
568 * example, reference counting or locking. In C11, it would map to
569 * kill_dependency(). It could be used as follows::
570 *
571 * rcu_read_lock();
572 * p = rcu_dereference(gp);
573 * long_lived = is_long_lived(p);
574 * if (long_lived) {
575 * if (!atomic_inc_not_zero(p->refcnt))
576 * long_lived = false;
577 * else
578 * p = rcu_pointer_handoff(p);
579 * }
580 * rcu_read_unlock();
581 */
582#define rcu_pointer_handoff(p) (p)
583
584/**
585 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
586 *
587 * When synchronize_rcu() is invoked on one CPU while other CPUs
588 * are within RCU read-side critical sections, then the
589 * synchronize_rcu() is guaranteed to block until after all the other
590 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
591 * on one CPU while other CPUs are within RCU read-side critical
592 * sections, invocation of the corresponding RCU callback is deferred
593 * until after the all the other CPUs exit their critical sections.
594 *
595 * Note, however, that RCU callbacks are permitted to run concurrently
596 * with new RCU read-side critical sections. One way that this can happen
597 * is via the following sequence of events: (1) CPU 0 enters an RCU
598 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
599 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
600 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
601 * callback is invoked. This is legal, because the RCU read-side critical
602 * section that was running concurrently with the call_rcu() (and which
603 * therefore might be referencing something that the corresponding RCU
604 * callback would free up) has completed before the corresponding
605 * RCU callback is invoked.
606 *
607 * RCU read-side critical sections may be nested. Any deferred actions
608 * will be deferred until the outermost RCU read-side critical section
609 * completes.
610 *
611 * You can avoid reading and understanding the next paragraph by
612 * following this rule: don't put anything in an rcu_read_lock() RCU
613 * read-side critical section that would block in a !PREEMPT kernel.
614 * But if you want the full story, read on!
615 *
616 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU),
617 * it is illegal to block while in an RCU read-side critical section.
618 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT
619 * kernel builds, RCU read-side critical sections may be preempted,
620 * but explicit blocking is illegal. Finally, in preemptible RCU
621 * implementations in real-time (with -rt patchset) kernel builds, RCU
622 * read-side critical sections may be preempted and they may also block, but
623 * only when acquiring spinlocks that are subject to priority inheritance.
624 */
625static inline void rcu_read_lock(void)
626{
627 __rcu_read_lock();
628 __acquire(RCU);
629 rcu_lock_acquire(&rcu_lock_map);
630 RCU_LOCKDEP_WARN(!rcu_is_watching(),
631 "rcu_read_lock() used illegally while idle");
632}
633
634/*
635 * So where is rcu_write_lock()? It does not exist, as there is no
636 * way for writers to lock out RCU readers. This is a feature, not
637 * a bug -- this property is what provides RCU's performance benefits.
638 * Of course, writers must coordinate with each other. The normal
639 * spinlock primitives work well for this, but any other technique may be
640 * used as well. RCU does not care how the writers keep out of each
641 * others' way, as long as they do so.
642 */
643
644/**
645 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
646 *
647 * In most situations, rcu_read_unlock() is immune from deadlock.
648 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock()
649 * is responsible for deboosting, which it does via rt_mutex_unlock().
650 * Unfortunately, this function acquires the scheduler's runqueue and
651 * priority-inheritance spinlocks. This means that deadlock could result
652 * if the caller of rcu_read_unlock() already holds one of these locks or
653 * any lock that is ever acquired while holding them.
654 *
655 * That said, RCU readers are never priority boosted unless they were
656 * preempted. Therefore, one way to avoid deadlock is to make sure
657 * that preemption never happens within any RCU read-side critical
658 * section whose outermost rcu_read_unlock() is called with one of
659 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in
660 * a number of ways, for example, by invoking preempt_disable() before
661 * critical section's outermost rcu_read_lock().
662 *
663 * Given that the set of locks acquired by rt_mutex_unlock() might change
664 * at any time, a somewhat more future-proofed approach is to make sure
665 * that that preemption never happens within any RCU read-side critical
666 * section whose outermost rcu_read_unlock() is called with irqs disabled.
667 * This approach relies on the fact that rt_mutex_unlock() currently only
668 * acquires irq-disabled locks.
669 *
670 * The second of these two approaches is best in most situations,
671 * however, the first approach can also be useful, at least to those
672 * developers willing to keep abreast of the set of locks acquired by
673 * rt_mutex_unlock().
674 *
675 * See rcu_read_lock() for more information.
676 */
677static inline void rcu_read_unlock(void)
678{
679 RCU_LOCKDEP_WARN(!rcu_is_watching(),
680 "rcu_read_unlock() used illegally while idle");
681 __release(RCU);
682 __rcu_read_unlock();
683 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
684}
685
686/**
687 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
688 *
689 * This is equivalent of rcu_read_lock(), but to be used when updates
690 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
691 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
692 * softirq handler to be a quiescent state, a process in RCU read-side
693 * critical section must be protected by disabling softirqs. Read-side
694 * critical sections in interrupt context can use just rcu_read_lock(),
695 * though this should at least be commented to avoid confusing people
696 * reading the code.
697 *
698 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
699 * must occur in the same context, for example, it is illegal to invoke
700 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
701 * was invoked from some other task.
702 */
703static inline void rcu_read_lock_bh(void)
704{
705 local_bh_disable();
706 __acquire(RCU_BH);
707 rcu_lock_acquire(&rcu_bh_lock_map);
708 RCU_LOCKDEP_WARN(!rcu_is_watching(),
709 "rcu_read_lock_bh() used illegally while idle");
710}
711
712/*
713 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
714 *
715 * See rcu_read_lock_bh() for more information.
716 */
717static inline void rcu_read_unlock_bh(void)
718{
719 RCU_LOCKDEP_WARN(!rcu_is_watching(),
720 "rcu_read_unlock_bh() used illegally while idle");
721 rcu_lock_release(&rcu_bh_lock_map);
722 __release(RCU_BH);
723 local_bh_enable();
724}
725
726/**
727 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
728 *
729 * This is equivalent of rcu_read_lock(), but to be used when updates
730 * are being done using call_rcu_sched() or synchronize_rcu_sched().
731 * Read-side critical sections can also be introduced by anything that
732 * disables preemption, including local_irq_disable() and friends.
733 *
734 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
735 * must occur in the same context, for example, it is illegal to invoke
736 * rcu_read_unlock_sched() from process context if the matching
737 * rcu_read_lock_sched() was invoked from an NMI handler.
738 */
739static inline void rcu_read_lock_sched(void)
740{
741 preempt_disable();
742 __acquire(RCU_SCHED);
743 rcu_lock_acquire(&rcu_sched_lock_map);
744 RCU_LOCKDEP_WARN(!rcu_is_watching(),
745 "rcu_read_lock_sched() used illegally while idle");
746}
747
748/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
749static inline notrace void rcu_read_lock_sched_notrace(void)
750{
751 preempt_disable_notrace();
752 __acquire(RCU_SCHED);
753}
754
755/*
756 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
757 *
758 * See rcu_read_lock_sched for more information.
759 */
760static inline void rcu_read_unlock_sched(void)
761{
762 RCU_LOCKDEP_WARN(!rcu_is_watching(),
763 "rcu_read_unlock_sched() used illegally while idle");
764 rcu_lock_release(&rcu_sched_lock_map);
765 __release(RCU_SCHED);
766 preempt_enable();
767}
768
769/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
770static inline notrace void rcu_read_unlock_sched_notrace(void)
771{
772 __release(RCU_SCHED);
773 preempt_enable_notrace();
774}
775
776/**
777 * RCU_INIT_POINTER() - initialize an RCU protected pointer
778 * @p: The pointer to be initialized.
779 * @v: The value to initialized the pointer to.
780 *
781 * Initialize an RCU-protected pointer in special cases where readers
782 * do not need ordering constraints on the CPU or the compiler. These
783 * special cases are:
784 *
785 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or*
786 * 2. The caller has taken whatever steps are required to prevent
787 * RCU readers from concurrently accessing this pointer *or*
788 * 3. The referenced data structure has already been exposed to
789 * readers either at compile time or via rcu_assign_pointer() *and*
790 *
791 * a. You have not made *any* reader-visible changes to
792 * this structure since then *or*
793 * b. It is OK for readers accessing this structure from its
794 * new location to see the old state of the structure. (For
795 * example, the changes were to statistical counters or to
796 * other state where exact synchronization is not required.)
797 *
798 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
799 * result in impossible-to-diagnose memory corruption. As in the structures
800 * will look OK in crash dumps, but any concurrent RCU readers might
801 * see pre-initialized values of the referenced data structure. So
802 * please be very careful how you use RCU_INIT_POINTER()!!!
803 *
804 * If you are creating an RCU-protected linked structure that is accessed
805 * by a single external-to-structure RCU-protected pointer, then you may
806 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
807 * pointers, but you must use rcu_assign_pointer() to initialize the
808 * external-to-structure pointer *after* you have completely initialized
809 * the reader-accessible portions of the linked structure.
810 *
811 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
812 * ordering guarantees for either the CPU or the compiler.
813 */
814#define RCU_INIT_POINTER(p, v) \
815 do { \
816 rcu_dereference_sparse(p, __rcu); \
817 WRITE_ONCE(p, RCU_INITIALIZER(v)); \
818 } while (0)
819
820/**
821 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
822 * @p: The pointer to be initialized.
823 * @v: The value to initialized the pointer to.
824 *
825 * GCC-style initialization for an RCU-protected pointer in a structure field.
826 */
827#define RCU_POINTER_INITIALIZER(p, v) \
828 .p = RCU_INITIALIZER(v)
829
830/*
831 * Does the specified offset indicate that the corresponding rcu_head
832 * structure can be handled by kfree_rcu()?
833 */
834#define __is_kfree_rcu_offset(offset) ((offset) < 4096)
835
836/*
837 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
838 */
839#define __kfree_rcu(head, offset) \
840 do { \
841 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
842 kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \
843 } while (0)
844
845/**
846 * kfree_rcu() - kfree an object after a grace period.
847 * @ptr: pointer to kfree
848 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
849 *
850 * Many rcu callbacks functions just call kfree() on the base structure.
851 * These functions are trivial, but their size adds up, and furthermore
852 * when they are used in a kernel module, that module must invoke the
853 * high-latency rcu_barrier() function at module-unload time.
854 *
855 * The kfree_rcu() function handles this issue. Rather than encoding a
856 * function address in the embedded rcu_head structure, kfree_rcu() instead
857 * encodes the offset of the rcu_head structure within the base structure.
858 * Because the functions are not allowed in the low-order 4096 bytes of
859 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
860 * If the offset is larger than 4095 bytes, a compile-time error will
861 * be generated in __kfree_rcu(). If this error is triggered, you can
862 * either fall back to use of call_rcu() or rearrange the structure to
863 * position the rcu_head structure into the first 4096 bytes.
864 *
865 * Note that the allowable offset might decrease in the future, for example,
866 * to allow something like kmem_cache_free_rcu().
867 *
868 * The BUILD_BUG_ON check must not involve any function calls, hence the
869 * checks are done in macros here.
870 */
871#define kfree_rcu(ptr, rcu_head) \
872 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
873
874
875/*
876 * Place this after a lock-acquisition primitive to guarantee that
877 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies
878 * if the UNLOCK and LOCK are executed by the same CPU or if the
879 * UNLOCK and LOCK operate on the same lock variable.
880 */
881#ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
882#define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */
883#else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
884#define smp_mb__after_unlock_lock() do { } while (0)
885#endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
886
887
888#endif /* __LINUX_RCUPDATE_H */
889