2 * Read-Copy Update mechanism for mutual exclusion
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.
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.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright IBM Corporation, 2001
20 * Author: Dipankar Sarma <dipankar@in.ibm.com>
22 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
23 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
25 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
26 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
28 * For detailed explanation of Read-Copy Update mechanism see -
29 * http://lse.sourceforge.net/locking/rcupdate.html
33 #ifndef __LINUX_RCUPDATE_H
34 #define __LINUX_RCUPDATE_H
36 #include <linux/types.h>
37 #include <linux/cache.h>
38 #include <linux/spinlock.h>
39 #include <linux/threads.h>
40 #include <linux/cpumask.h>
41 #include <linux/seqlock.h>
42 #include <linux/lockdep.h>
43 #include <linux/completion.h>
44 #include <linux/debugobjects.h>
45 #include <linux/bug.h>
46 #include <linux/compiler.h>
48 #ifdef CONFIG_RCU_TORTURE_TEST
49 extern int rcutorture_runnable; /* for sysctl */
50 #endif /* #ifdef CONFIG_RCU_TORTURE_TEST */
52 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
53 extern void rcutorture_record_test_transition(void);
54 extern void rcutorture_record_progress(unsigned long vernum);
55 extern void do_trace_rcu_torture_read(char *rcutorturename,
56 struct rcu_head *rhp);
58 static inline void rcutorture_record_test_transition(void)
61 static inline void rcutorture_record_progress(unsigned long vernum)
64 #ifdef CONFIG_RCU_TRACE
65 extern void do_trace_rcu_torture_read(char *rcutorturename,
66 struct rcu_head *rhp);
68 #define do_trace_rcu_torture_read(rcutorturename, rhp) do { } while (0)
72 #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b))
73 #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b))
74 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
75 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
77 /* Exported common interfaces */
79 #ifdef CONFIG_PREEMPT_RCU
82 * call_rcu() - Queue an RCU callback for invocation after a grace period.
83 * @head: structure to be used for queueing the RCU updates.
84 * @func: actual callback function to be invoked after the grace period
86 * The callback function will be invoked some time after a full grace
87 * period elapses, in other words after all pre-existing RCU read-side
88 * critical sections have completed. However, the callback function
89 * might well execute concurrently with RCU read-side critical sections
90 * that started after call_rcu() was invoked. RCU read-side critical
91 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
94 * Note that all CPUs must agree that the grace period extended beyond
95 * all pre-existing RCU read-side critical section. On systems with more
96 * than one CPU, this means that when "func()" is invoked, each CPU is
97 * guaranteed to have executed a full memory barrier since the end of its
98 * last RCU read-side critical section whose beginning preceded the call
99 * to call_rcu(). It also means that each CPU executing an RCU read-side
100 * critical section that continues beyond the start of "func()" must have
101 * executed a memory barrier after the call_rcu() but before the beginning
102 * of that RCU read-side critical section. Note that these guarantees
103 * include CPUs that are offline, idle, or executing in user mode, as
104 * well as CPUs that are executing in the kernel.
106 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
107 * resulting RCU callback function "func()", then both CPU A and CPU B are
108 * guaranteed to execute a full memory barrier during the time interval
109 * between the call to call_rcu() and the invocation of "func()" -- even
110 * if CPU A and CPU B are the same CPU (but again only if the system has
111 * more than one CPU).
113 extern void call_rcu(struct rcu_head *head,
114 void (*func)(struct rcu_head *head));
116 #else /* #ifdef CONFIG_PREEMPT_RCU */
118 /* In classic RCU, call_rcu() is just call_rcu_sched(). */
119 #define call_rcu call_rcu_sched
121 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
124 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
125 * @head: structure to be used for queueing the RCU updates.
126 * @func: actual callback function to be invoked after the grace period
128 * The callback function will be invoked some time after a full grace
129 * period elapses, in other words after all currently executing RCU
130 * read-side critical sections have completed. call_rcu_bh() assumes
131 * that the read-side critical sections end on completion of a softirq
132 * handler. This means that read-side critical sections in process
133 * context must not be interrupted by softirqs. This interface is to be
134 * used when most of the read-side critical sections are in softirq context.
135 * RCU read-side critical sections are delimited by :
136 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context.
138 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
139 * These may be nested.
141 * See the description of call_rcu() for more detailed information on
142 * memory ordering guarantees.
144 extern void call_rcu_bh(struct rcu_head *head,
145 void (*func)(struct rcu_head *head));
148 * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
149 * @head: structure to be used for queueing the RCU updates.
150 * @func: actual callback function to be invoked after the grace period
152 * The callback function will be invoked some time after a full grace
153 * period elapses, in other words after all currently executing RCU
154 * read-side critical sections have completed. call_rcu_sched() assumes
155 * that the read-side critical sections end on enabling of preemption
156 * or on voluntary preemption.
157 * RCU read-side critical sections are delimited by :
158 * - rcu_read_lock_sched() and rcu_read_unlock_sched(),
160 * anything that disables preemption.
161 * These may be nested.
163 * See the description of call_rcu() for more detailed information on
164 * memory ordering guarantees.
166 extern void call_rcu_sched(struct rcu_head *head,
167 void (*func)(struct rcu_head *rcu));
169 extern void synchronize_sched(void);
171 #ifdef CONFIG_PREEMPT_RCU
173 extern void __rcu_read_lock(void);
174 extern void __rcu_read_unlock(void);
175 extern void rcu_read_unlock_special(struct task_struct *t);
176 void synchronize_rcu(void);
179 * Defined as a macro as it is a very low level header included from
180 * areas that don't even know about current. This gives the rcu_read_lock()
181 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
182 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
184 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
186 #else /* #ifdef CONFIG_PREEMPT_RCU */
188 static inline void __rcu_read_lock(void)
193 static inline void __rcu_read_unlock(void)
198 static inline void synchronize_rcu(void)
203 static inline int rcu_preempt_depth(void)
208 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
210 /* Internal to kernel */
211 extern void rcu_sched_qs(int cpu);
212 extern void rcu_bh_qs(int cpu);
213 extern void rcu_check_callbacks(int cpu, int user);
214 struct notifier_block;
215 extern void rcu_idle_enter(void);
216 extern void rcu_idle_exit(void);
217 extern void rcu_irq_enter(void);
218 extern void rcu_irq_exit(void);
220 #ifdef CONFIG_RCU_USER_QS
221 extern void rcu_user_enter(void);
222 extern void rcu_user_exit(void);
223 extern void rcu_user_enter_after_irq(void);
224 extern void rcu_user_exit_after_irq(void);
225 extern void rcu_user_hooks_switch(struct task_struct *prev,
226 struct task_struct *next);
228 static inline void rcu_user_enter(void) { }
229 static inline void rcu_user_exit(void) { }
230 static inline void rcu_user_enter_after_irq(void) { }
231 static inline void rcu_user_exit_after_irq(void) { }
232 static inline void rcu_user_hooks_switch(struct task_struct *prev,
233 struct task_struct *next) { }
234 #endif /* CONFIG_RCU_USER_QS */
236 extern void exit_rcu(void);
239 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
240 * @a: Code that RCU needs to pay attention to.
242 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
243 * in the inner idle loop, that is, between the rcu_idle_enter() and
244 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
245 * critical sections. However, things like powertop need tracepoints
246 * in the inner idle loop.
248 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
249 * will tell RCU that it needs to pay attending, invoke its argument
250 * (in this example, a call to the do_something_with_RCU() function),
251 * and then tell RCU to go back to ignoring this CPU. It is permissible
252 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently
253 * quite limited. If deeper nesting is required, it will be necessary
254 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly.
256 #define RCU_NONIDLE(a) \
259 do { a; } while (0); \
264 * Infrastructure to implement the synchronize_() primitives in
265 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
268 typedef void call_rcu_func_t(struct rcu_head *head,
269 void (*func)(struct rcu_head *head));
270 void wait_rcu_gp(call_rcu_func_t crf);
272 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
273 #include <linux/rcutree.h>
274 #elif defined(CONFIG_TINY_RCU) || defined(CONFIG_TINY_PREEMPT_RCU)
275 #include <linux/rcutiny.h>
277 #error "Unknown RCU implementation specified to kernel configuration"
281 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
282 * initialization and destruction of rcu_head on the stack. rcu_head structures
283 * allocated dynamically in the heap or defined statically don't need any
286 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
287 extern void init_rcu_head_on_stack(struct rcu_head *head);
288 extern void destroy_rcu_head_on_stack(struct rcu_head *head);
289 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
290 static inline void init_rcu_head_on_stack(struct rcu_head *head)
294 static inline void destroy_rcu_head_on_stack(struct rcu_head *head)
297 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
299 #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SMP)
300 extern int rcu_is_cpu_idle(void);
301 #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_SMP) */
303 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
304 bool rcu_lockdep_current_cpu_online(void);
305 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
306 static inline bool rcu_lockdep_current_cpu_online(void)
310 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
312 #ifdef CONFIG_DEBUG_LOCK_ALLOC
314 static inline void rcu_lock_acquire(struct lockdep_map *map)
316 lock_acquire(map, 0, 0, 2, 1, NULL, _THIS_IP_);
319 static inline void rcu_lock_release(struct lockdep_map *map)
321 lock_release(map, 1, _THIS_IP_);
324 extern struct lockdep_map rcu_lock_map;
325 extern struct lockdep_map rcu_bh_lock_map;
326 extern struct lockdep_map rcu_sched_lock_map;
327 extern int debug_lockdep_rcu_enabled(void);
330 * rcu_read_lock_held() - might we be in RCU read-side critical section?
332 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
333 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
334 * this assumes we are in an RCU read-side critical section unless it can
335 * prove otherwise. This is useful for debug checks in functions that
336 * require that they be called within an RCU read-side critical section.
338 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
339 * and while lockdep is disabled.
341 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
342 * occur in the same context, for example, it is illegal to invoke
343 * rcu_read_unlock() in process context if the matching rcu_read_lock()
344 * was invoked from within an irq handler.
346 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
347 * offline from an RCU perspective, so check for those as well.
349 static inline int rcu_read_lock_held(void)
351 if (!debug_lockdep_rcu_enabled())
353 if (rcu_is_cpu_idle())
355 if (!rcu_lockdep_current_cpu_online())
357 return lock_is_held(&rcu_lock_map);
361 * rcu_read_lock_bh_held() is defined out of line to avoid #include-file
364 extern int rcu_read_lock_bh_held(void);
367 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
369 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
370 * RCU-sched read-side critical section. In absence of
371 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
372 * critical section unless it can prove otherwise. Note that disabling
373 * of preemption (including disabling irqs) counts as an RCU-sched
374 * read-side critical section. This is useful for debug checks in functions
375 * that required that they be called within an RCU-sched read-side
378 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
379 * and while lockdep is disabled.
381 * Note that if the CPU is in the idle loop from an RCU point of
382 * view (ie: that we are in the section between rcu_idle_enter() and
383 * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU
384 * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs
385 * that are in such a section, considering these as in extended quiescent
386 * state, so such a CPU is effectively never in an RCU read-side critical
387 * section regardless of what RCU primitives it invokes. This state of
388 * affairs is required --- we need to keep an RCU-free window in idle
389 * where the CPU may possibly enter into low power mode. This way we can
390 * notice an extended quiescent state to other CPUs that started a grace
391 * period. Otherwise we would delay any grace period as long as we run in
394 * Similarly, we avoid claiming an SRCU read lock held if the current
397 #ifdef CONFIG_PREEMPT_COUNT
398 static inline int rcu_read_lock_sched_held(void)
400 int lockdep_opinion = 0;
402 if (!debug_lockdep_rcu_enabled())
404 if (rcu_is_cpu_idle())
406 if (!rcu_lockdep_current_cpu_online())
409 lockdep_opinion = lock_is_held(&rcu_sched_lock_map);
410 return lockdep_opinion || preempt_count() != 0 || irqs_disabled();
412 #else /* #ifdef CONFIG_PREEMPT_COUNT */
413 static inline int rcu_read_lock_sched_held(void)
417 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
419 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
421 # define rcu_lock_acquire(a) do { } while (0)
422 # define rcu_lock_release(a) do { } while (0)
424 static inline int rcu_read_lock_held(void)
429 static inline int rcu_read_lock_bh_held(void)
434 #ifdef CONFIG_PREEMPT_COUNT
435 static inline int rcu_read_lock_sched_held(void)
437 return preempt_count() != 0 || irqs_disabled();
439 #else /* #ifdef CONFIG_PREEMPT_COUNT */
440 static inline int rcu_read_lock_sched_held(void)
444 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
446 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
448 #ifdef CONFIG_PROVE_RCU
450 extern int rcu_my_thread_group_empty(void);
453 * rcu_lockdep_assert - emit lockdep splat if specified condition not met
454 * @c: condition to check
455 * @s: informative message
457 #define rcu_lockdep_assert(c, s) \
459 static bool __section(.data.unlikely) __warned; \
460 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \
462 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
466 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
467 static inline void rcu_preempt_sleep_check(void)
469 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
470 "Illegal context switch in RCU read-side critical section");
472 #else /* #ifdef CONFIG_PROVE_RCU */
473 static inline void rcu_preempt_sleep_check(void)
476 #endif /* #else #ifdef CONFIG_PROVE_RCU */
478 #define rcu_sleep_check() \
480 rcu_preempt_sleep_check(); \
481 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \
482 "Illegal context switch in RCU-bh" \
483 " read-side critical section"); \
484 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \
485 "Illegal context switch in RCU-sched"\
486 " read-side critical section"); \
489 #else /* #ifdef CONFIG_PROVE_RCU */
491 #define rcu_lockdep_assert(c, s) do { } while (0)
492 #define rcu_sleep_check() do { } while (0)
494 #endif /* #else #ifdef CONFIG_PROVE_RCU */
497 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
498 * and rcu_assign_pointer(). Some of these could be folded into their
499 * callers, but they are left separate in order to ease introduction of
500 * multiple flavors of pointers to match the multiple flavors of RCU
501 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
506 #define rcu_dereference_sparse(p, space) \
507 ((void)(((typeof(*p) space *)p) == p))
508 #else /* #ifdef __CHECKER__ */
509 #define rcu_dereference_sparse(p, space)
510 #endif /* #else #ifdef __CHECKER__ */
512 #define __rcu_access_pointer(p, space) \
514 typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \
515 rcu_dereference_sparse(p, space); \
516 ((typeof(*p) __force __kernel *)(_________p1)); \
518 #define __rcu_dereference_check(p, c, space) \
520 typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \
521 rcu_lockdep_assert(c, "suspicious rcu_dereference_check()" \
523 rcu_dereference_sparse(p, space); \
524 smp_read_barrier_depends(); \
525 ((typeof(*p) __force __kernel *)(_________p1)); \
527 #define __rcu_dereference_protected(p, c, space) \
529 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected()" \
531 rcu_dereference_sparse(p, space); \
532 ((typeof(*p) __force __kernel *)(p)); \
535 #define __rcu_access_index(p, space) \
537 typeof(p) _________p1 = ACCESS_ONCE(p); \
538 rcu_dereference_sparse(p, space); \
541 #define __rcu_dereference_index_check(p, c) \
543 typeof(p) _________p1 = ACCESS_ONCE(p); \
544 rcu_lockdep_assert(c, \
545 "suspicious rcu_dereference_index_check()" \
547 smp_read_barrier_depends(); \
550 #define __rcu_assign_pointer(p, v, space) \
553 (p) = (typeof(*v) __force space *)(v); \
558 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
559 * @p: The pointer to read
561 * Return the value of the specified RCU-protected pointer, but omit the
562 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
563 * when the value of this pointer is accessed, but the pointer is not
564 * dereferenced, for example, when testing an RCU-protected pointer against
565 * NULL. Although rcu_access_pointer() may also be used in cases where
566 * update-side locks prevent the value of the pointer from changing, you
567 * should instead use rcu_dereference_protected() for this use case.
569 * It is also permissible to use rcu_access_pointer() when read-side
570 * access to the pointer was removed at least one grace period ago, as
571 * is the case in the context of the RCU callback that is freeing up
572 * the data, or after a synchronize_rcu() returns. This can be useful
573 * when tearing down multi-linked structures after a grace period
576 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
579 * rcu_dereference_check() - rcu_dereference with debug checking
580 * @p: The pointer to read, prior to dereferencing
581 * @c: The conditions under which the dereference will take place
583 * Do an rcu_dereference(), but check that the conditions under which the
584 * dereference will take place are correct. Typically the conditions
585 * indicate the various locking conditions that should be held at that
586 * point. The check should return true if the conditions are satisfied.
587 * An implicit check for being in an RCU read-side critical section
588 * (rcu_read_lock()) is included.
592 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
594 * could be used to indicate to lockdep that foo->bar may only be dereferenced
595 * if either rcu_read_lock() is held, or that the lock required to replace
596 * the bar struct at foo->bar is held.
598 * Note that the list of conditions may also include indications of when a lock
599 * need not be held, for example during initialisation or destruction of the
602 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
603 * atomic_read(&foo->usage) == 0);
605 * Inserts memory barriers on architectures that require them
606 * (currently only the Alpha), prevents the compiler from refetching
607 * (and from merging fetches), and, more importantly, documents exactly
608 * which pointers are protected by RCU and checks that the pointer is
609 * annotated as __rcu.
611 #define rcu_dereference_check(p, c) \
612 __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu)
615 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
616 * @p: The pointer to read, prior to dereferencing
617 * @c: The conditions under which the dereference will take place
619 * This is the RCU-bh counterpart to rcu_dereference_check().
621 #define rcu_dereference_bh_check(p, c) \
622 __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu)
625 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
626 * @p: The pointer to read, prior to dereferencing
627 * @c: The conditions under which the dereference will take place
629 * This is the RCU-sched counterpart to rcu_dereference_check().
631 #define rcu_dereference_sched_check(p, c) \
632 __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \
635 #define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/
638 * rcu_access_index() - fetch RCU index with no dereferencing
639 * @p: The index to read
641 * Return the value of the specified RCU-protected index, but omit the
642 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
643 * when the value of this index is accessed, but the index is not
644 * dereferenced, for example, when testing an RCU-protected index against
645 * -1. Although rcu_access_index() may also be used in cases where
646 * update-side locks prevent the value of the index from changing, you
647 * should instead use rcu_dereference_index_protected() for this use case.
649 #define rcu_access_index(p) __rcu_access_index((p), __rcu)
652 * rcu_dereference_index_check() - rcu_dereference for indices with debug checking
653 * @p: The pointer to read, prior to dereferencing
654 * @c: The conditions under which the dereference will take place
656 * Similar to rcu_dereference_check(), but omits the sparse checking.
657 * This allows rcu_dereference_index_check() to be used on integers,
658 * which can then be used as array indices. Attempting to use
659 * rcu_dereference_check() on an integer will give compiler warnings
660 * because the sparse address-space mechanism relies on dereferencing
661 * the RCU-protected pointer. Dereferencing integers is not something
662 * that even gcc will put up with.
664 * Note that this function does not implicitly check for RCU read-side
665 * critical sections. If this function gains lots of uses, it might
666 * make sense to provide versions for each flavor of RCU, but it does
667 * not make sense as of early 2010.
669 #define rcu_dereference_index_check(p, c) \
670 __rcu_dereference_index_check((p), (c))
673 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
674 * @p: The pointer to read, prior to dereferencing
675 * @c: The conditions under which the dereference will take place
677 * Return the value of the specified RCU-protected pointer, but omit
678 * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This
679 * is useful in cases where update-side locks prevent the value of the
680 * pointer from changing. Please note that this primitive does -not-
681 * prevent the compiler from repeating this reference or combining it
682 * with other references, so it should not be used without protection
683 * of appropriate locks.
685 * This function is only for update-side use. Using this function
686 * when protected only by rcu_read_lock() will result in infrequent
687 * but very ugly failures.
689 #define rcu_dereference_protected(p, c) \
690 __rcu_dereference_protected((p), (c), __rcu)
694 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
695 * @p: The pointer to read, prior to dereferencing
697 * This is a simple wrapper around rcu_dereference_check().
699 #define rcu_dereference(p) rcu_dereference_check(p, 0)
702 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
703 * @p: The pointer to read, prior to dereferencing
705 * Makes rcu_dereference_check() do the dirty work.
707 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
710 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
711 * @p: The pointer to read, prior to dereferencing
713 * Makes rcu_dereference_check() do the dirty work.
715 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
718 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
720 * When synchronize_rcu() is invoked on one CPU while other CPUs
721 * are within RCU read-side critical sections, then the
722 * synchronize_rcu() is guaranteed to block until after all the other
723 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
724 * on one CPU while other CPUs are within RCU read-side critical
725 * sections, invocation of the corresponding RCU callback is deferred
726 * until after the all the other CPUs exit their critical sections.
728 * Note, however, that RCU callbacks are permitted to run concurrently
729 * with new RCU read-side critical sections. One way that this can happen
730 * is via the following sequence of events: (1) CPU 0 enters an RCU
731 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
732 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
733 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
734 * callback is invoked. This is legal, because the RCU read-side critical
735 * section that was running concurrently with the call_rcu() (and which
736 * therefore might be referencing something that the corresponding RCU
737 * callback would free up) has completed before the corresponding
738 * RCU callback is invoked.
740 * RCU read-side critical sections may be nested. Any deferred actions
741 * will be deferred until the outermost RCU read-side critical section
744 * You can avoid reading and understanding the next paragraph by
745 * following this rule: don't put anything in an rcu_read_lock() RCU
746 * read-side critical section that would block in a !PREEMPT kernel.
747 * But if you want the full story, read on!
749 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), it
750 * is illegal to block while in an RCU read-side critical section. In
751 * preemptible RCU implementations (TREE_PREEMPT_RCU and TINY_PREEMPT_RCU)
752 * in CONFIG_PREEMPT kernel builds, RCU read-side critical sections may
753 * be preempted, but explicit blocking is illegal. Finally, in preemptible
754 * RCU implementations in real-time (CONFIG_PREEMPT_RT) kernel builds,
755 * RCU read-side critical sections may be preempted and they may also
756 * block, but only when acquiring spinlocks that are subject to priority
759 static inline void rcu_read_lock(void)
763 rcu_lock_acquire(&rcu_lock_map);
764 rcu_lockdep_assert(!rcu_is_cpu_idle(),
765 "rcu_read_lock() used illegally while idle");
769 * So where is rcu_write_lock()? It does not exist, as there is no
770 * way for writers to lock out RCU readers. This is a feature, not
771 * a bug -- this property is what provides RCU's performance benefits.
772 * Of course, writers must coordinate with each other. The normal
773 * spinlock primitives work well for this, but any other technique may be
774 * used as well. RCU does not care how the writers keep out of each
775 * others' way, as long as they do so.
779 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
781 * See rcu_read_lock() for more information.
783 static inline void rcu_read_unlock(void)
785 rcu_lockdep_assert(!rcu_is_cpu_idle(),
786 "rcu_read_unlock() used illegally while idle");
787 rcu_lock_release(&rcu_lock_map);
793 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
795 * This is equivalent of rcu_read_lock(), but to be used when updates
796 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
797 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
798 * softirq handler to be a quiescent state, a process in RCU read-side
799 * critical section must be protected by disabling softirqs. Read-side
800 * critical sections in interrupt context can use just rcu_read_lock(),
801 * though this should at least be commented to avoid confusing people
804 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
805 * must occur in the same context, for example, it is illegal to invoke
806 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
807 * was invoked from some other task.
809 static inline void rcu_read_lock_bh(void)
813 rcu_lock_acquire(&rcu_bh_lock_map);
814 rcu_lockdep_assert(!rcu_is_cpu_idle(),
815 "rcu_read_lock_bh() used illegally while idle");
819 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
821 * See rcu_read_lock_bh() for more information.
823 static inline void rcu_read_unlock_bh(void)
825 rcu_lockdep_assert(!rcu_is_cpu_idle(),
826 "rcu_read_unlock_bh() used illegally while idle");
827 rcu_lock_release(&rcu_bh_lock_map);
833 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
835 * This is equivalent of rcu_read_lock(), but to be used when updates
836 * are being done using call_rcu_sched() or synchronize_rcu_sched().
837 * Read-side critical sections can also be introduced by anything that
838 * disables preemption, including local_irq_disable() and friends.
840 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
841 * must occur in the same context, for example, it is illegal to invoke
842 * rcu_read_unlock_sched() from process context if the matching
843 * rcu_read_lock_sched() was invoked from an NMI handler.
845 static inline void rcu_read_lock_sched(void)
848 __acquire(RCU_SCHED);
849 rcu_lock_acquire(&rcu_sched_lock_map);
850 rcu_lockdep_assert(!rcu_is_cpu_idle(),
851 "rcu_read_lock_sched() used illegally while idle");
854 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
855 static inline notrace void rcu_read_lock_sched_notrace(void)
857 preempt_disable_notrace();
858 __acquire(RCU_SCHED);
862 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
864 * See rcu_read_lock_sched for more information.
866 static inline void rcu_read_unlock_sched(void)
868 rcu_lockdep_assert(!rcu_is_cpu_idle(),
869 "rcu_read_unlock_sched() used illegally while idle");
870 rcu_lock_release(&rcu_sched_lock_map);
871 __release(RCU_SCHED);
875 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
876 static inline notrace void rcu_read_unlock_sched_notrace(void)
878 __release(RCU_SCHED);
879 preempt_enable_notrace();
883 * rcu_assign_pointer() - assign to RCU-protected pointer
884 * @p: pointer to assign to
885 * @v: value to assign (publish)
887 * Assigns the specified value to the specified RCU-protected
888 * pointer, ensuring that any concurrent RCU readers will see
889 * any prior initialization.
891 * Inserts memory barriers on architectures that require them
892 * (which is most of them), and also prevents the compiler from
893 * reordering the code that initializes the structure after the pointer
894 * assignment. More importantly, this call documents which pointers
895 * will be dereferenced by RCU read-side code.
897 * In some special cases, you may use RCU_INIT_POINTER() instead
898 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
899 * to the fact that it does not constrain either the CPU or the compiler.
900 * That said, using RCU_INIT_POINTER() when you should have used
901 * rcu_assign_pointer() is a very bad thing that results in
902 * impossible-to-diagnose memory corruption. So please be careful.
903 * See the RCU_INIT_POINTER() comment header for details.
905 #define rcu_assign_pointer(p, v) \
906 __rcu_assign_pointer((p), (v), __rcu)
909 * RCU_INIT_POINTER() - initialize an RCU protected pointer
911 * Initialize an RCU-protected pointer in special cases where readers
912 * do not need ordering constraints on the CPU or the compiler. These
915 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
916 * 2. The caller has taken whatever steps are required to prevent
917 * RCU readers from concurrently accessing this pointer -or-
918 * 3. The referenced data structure has already been exposed to
919 * readers either at compile time or via rcu_assign_pointer() -and-
920 * a. You have not made -any- reader-visible changes to
921 * this structure since then -or-
922 * b. It is OK for readers accessing this structure from its
923 * new location to see the old state of the structure. (For
924 * example, the changes were to statistical counters or to
925 * other state where exact synchronization is not required.)
927 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
928 * result in impossible-to-diagnose memory corruption. As in the structures
929 * will look OK in crash dumps, but any concurrent RCU readers might
930 * see pre-initialized values of the referenced data structure. So
931 * please be very careful how you use RCU_INIT_POINTER()!!!
933 * If you are creating an RCU-protected linked structure that is accessed
934 * by a single external-to-structure RCU-protected pointer, then you may
935 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
936 * pointers, but you must use rcu_assign_pointer() to initialize the
937 * external-to-structure pointer -after- you have completely initialized
938 * the reader-accessible portions of the linked structure.
940 #define RCU_INIT_POINTER(p, v) \
942 p = (typeof(*v) __force __rcu *)(v); \
946 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
948 * GCC-style initialization for an RCU-protected pointer in a structure field.
950 #define RCU_POINTER_INITIALIZER(p, v) \
951 .p = (typeof(*v) __force __rcu *)(v)
954 * Does the specified offset indicate that the corresponding rcu_head
955 * structure can be handled by kfree_rcu()?
957 #define __is_kfree_rcu_offset(offset) ((offset) < 4096)
960 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
962 #define __kfree_rcu(head, offset) \
964 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
965 kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \
969 * kfree_rcu() - kfree an object after a grace period.
970 * @ptr: pointer to kfree
971 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
973 * Many rcu callbacks functions just call kfree() on the base structure.
974 * These functions are trivial, but their size adds up, and furthermore
975 * when they are used in a kernel module, that module must invoke the
976 * high-latency rcu_barrier() function at module-unload time.
978 * The kfree_rcu() function handles this issue. Rather than encoding a
979 * function address in the embedded rcu_head structure, kfree_rcu() instead
980 * encodes the offset of the rcu_head structure within the base structure.
981 * Because the functions are not allowed in the low-order 4096 bytes of
982 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
983 * If the offset is larger than 4095 bytes, a compile-time error will
984 * be generated in __kfree_rcu(). If this error is triggered, you can
985 * either fall back to use of call_rcu() or rearrange the structure to
986 * position the rcu_head structure into the first 4096 bytes.
988 * Note that the allowable offset might decrease in the future, for example,
989 * to allow something like kmem_cache_free_rcu().
991 * The BUILD_BUG_ON check must not involve any function calls, hence the
992 * checks are done in macros here.
994 #define kfree_rcu(ptr, rcu_head) \
995 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
997 #endif /* __LINUX_RCUPDATE_H */