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, you can access it online at
16 * http://www.gnu.org/licenses/gpl-2.0.html.
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>
47 #include <asm/barrier.h>
49 extern int rcu_expedited; /* for sysctl */
50 #ifdef CONFIG_RCU_TORTURE_TEST
51 extern int rcutorture_runnable; /* for sysctl */
52 #endif /* #ifdef CONFIG_RCU_TORTURE_TEST */
54 enum rcutorture_type {
62 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
63 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
64 unsigned long *gpnum, unsigned long *completed);
65 void rcutorture_record_test_transition(void);
66 void rcutorture_record_progress(unsigned long vernum);
67 void do_trace_rcu_torture_read(const char *rcutorturename,
73 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
76 unsigned long *completed)
82 static inline void rcutorture_record_test_transition(void)
85 static inline void rcutorture_record_progress(unsigned long vernum)
88 #ifdef CONFIG_RCU_TRACE
89 void do_trace_rcu_torture_read(const char *rcutorturename,
95 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
100 #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b))
101 #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b))
102 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
103 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
104 #define ulong2long(a) (*(long *)(&(a)))
106 /* Exported common interfaces */
108 #ifdef CONFIG_PREEMPT_RCU
111 * call_rcu() - Queue an RCU callback for invocation after a grace period.
112 * @head: structure to be used for queueing the RCU updates.
113 * @func: actual callback function to be invoked after the grace period
115 * The callback function will be invoked some time after a full grace
116 * period elapses, in other words after all pre-existing RCU read-side
117 * critical sections have completed. However, the callback function
118 * might well execute concurrently with RCU read-side critical sections
119 * that started after call_rcu() was invoked. RCU read-side critical
120 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
123 * Note that all CPUs must agree that the grace period extended beyond
124 * all pre-existing RCU read-side critical section. On systems with more
125 * than one CPU, this means that when "func()" is invoked, each CPU is
126 * guaranteed to have executed a full memory barrier since the end of its
127 * last RCU read-side critical section whose beginning preceded the call
128 * to call_rcu(). It also means that each CPU executing an RCU read-side
129 * critical section that continues beyond the start of "func()" must have
130 * executed a memory barrier after the call_rcu() but before the beginning
131 * of that RCU read-side critical section. Note that these guarantees
132 * include CPUs that are offline, idle, or executing in user mode, as
133 * well as CPUs that are executing in the kernel.
135 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
136 * resulting RCU callback function "func()", then both CPU A and CPU B are
137 * guaranteed to execute a full memory barrier during the time interval
138 * between the call to call_rcu() and the invocation of "func()" -- even
139 * if CPU A and CPU B are the same CPU (but again only if the system has
140 * more than one CPU).
142 void call_rcu(struct rcu_head *head,
143 void (*func)(struct rcu_head *head));
145 #else /* #ifdef CONFIG_PREEMPT_RCU */
147 /* In classic RCU, call_rcu() is just call_rcu_sched(). */
148 #define call_rcu call_rcu_sched
150 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
153 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
154 * @head: structure to be used for queueing the RCU updates.
155 * @func: actual callback function to be invoked after the grace period
157 * The callback function will be invoked some time after a full grace
158 * period elapses, in other words after all currently executing RCU
159 * read-side critical sections have completed. call_rcu_bh() assumes
160 * that the read-side critical sections end on completion of a softirq
161 * handler. This means that read-side critical sections in process
162 * context must not be interrupted by softirqs. This interface is to be
163 * used when most of the read-side critical sections are in softirq context.
164 * RCU read-side critical sections are delimited by :
165 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context.
167 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
168 * These may be nested.
170 * See the description of call_rcu() for more detailed information on
171 * memory ordering guarantees.
173 void call_rcu_bh(struct rcu_head *head,
174 void (*func)(struct rcu_head *head));
177 * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
178 * @head: structure to be used for queueing the RCU updates.
179 * @func: actual callback function to be invoked after the grace period
181 * The callback function will be invoked some time after a full grace
182 * period elapses, in other words after all currently executing RCU
183 * read-side critical sections have completed. call_rcu_sched() assumes
184 * that the read-side critical sections end on enabling of preemption
185 * or on voluntary preemption.
186 * RCU read-side critical sections are delimited by :
187 * - rcu_read_lock_sched() and rcu_read_unlock_sched(),
189 * anything that disables preemption.
190 * These may be nested.
192 * See the description of call_rcu() for more detailed information on
193 * memory ordering guarantees.
195 void call_rcu_sched(struct rcu_head *head,
196 void (*func)(struct rcu_head *rcu));
198 void synchronize_sched(void);
200 #ifdef CONFIG_PREEMPT_RCU
202 void __rcu_read_lock(void);
203 void __rcu_read_unlock(void);
204 void rcu_read_unlock_special(struct task_struct *t);
205 void synchronize_rcu(void);
208 * Defined as a macro as it is a very low level header included from
209 * areas that don't even know about current. This gives the rcu_read_lock()
210 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
211 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
213 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
215 #else /* #ifdef CONFIG_PREEMPT_RCU */
217 static inline void __rcu_read_lock(void)
222 static inline void __rcu_read_unlock(void)
227 static inline void synchronize_rcu(void)
232 static inline int rcu_preempt_depth(void)
237 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
239 /* Internal to kernel */
241 void rcu_sched_qs(int cpu);
242 void rcu_bh_qs(int cpu);
243 void rcu_check_callbacks(int cpu, int user);
244 struct notifier_block;
245 void rcu_idle_enter(void);
246 void rcu_idle_exit(void);
247 void rcu_irq_enter(void);
248 void rcu_irq_exit(void);
250 #ifdef CONFIG_RCU_STALL_COMMON
251 void rcu_sysrq_start(void);
252 void rcu_sysrq_end(void);
253 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
254 static inline void rcu_sysrq_start(void)
257 static inline void rcu_sysrq_end(void)
260 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
262 #ifdef CONFIG_RCU_USER_QS
263 void rcu_user_enter(void);
264 void rcu_user_exit(void);
266 static inline void rcu_user_enter(void) { }
267 static inline void rcu_user_exit(void) { }
268 static inline void rcu_user_hooks_switch(struct task_struct *prev,
269 struct task_struct *next) { }
270 #endif /* CONFIG_RCU_USER_QS */
272 #ifdef CONFIG_RCU_NOCB_CPU
273 void rcu_init_nohz(void);
274 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
275 static inline void rcu_init_nohz(void)
278 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
281 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
282 * @a: Code that RCU needs to pay attention to.
284 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
285 * in the inner idle loop, that is, between the rcu_idle_enter() and
286 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
287 * critical sections. However, things like powertop need tracepoints
288 * in the inner idle loop.
290 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
291 * will tell RCU that it needs to pay attending, invoke its argument
292 * (in this example, a call to the do_something_with_RCU() function),
293 * and then tell RCU to go back to ignoring this CPU. It is permissible
294 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently
295 * quite limited. If deeper nesting is required, it will be necessary
296 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly.
298 #define RCU_NONIDLE(a) \
301 do { a; } while (0); \
305 #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP)
306 bool __rcu_is_watching(void);
307 #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */
310 * Infrastructure to implement the synchronize_() primitives in
311 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
314 typedef void call_rcu_func_t(struct rcu_head *head,
315 void (*func)(struct rcu_head *head));
316 void wait_rcu_gp(call_rcu_func_t crf);
318 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
319 #include <linux/rcutree.h>
320 #elif defined(CONFIG_TINY_RCU)
321 #include <linux/rcutiny.h>
323 #error "Unknown RCU implementation specified to kernel configuration"
327 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
328 * initialization and destruction of rcu_head on the stack. rcu_head structures
329 * allocated dynamically in the heap or defined statically don't need any
332 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
333 void init_rcu_head(struct rcu_head *head);
334 void destroy_rcu_head(struct rcu_head *head);
335 void init_rcu_head_on_stack(struct rcu_head *head);
336 void destroy_rcu_head_on_stack(struct rcu_head *head);
337 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
338 static inline void init_rcu_head(struct rcu_head *head)
342 static inline void destroy_rcu_head(struct rcu_head *head)
346 static inline void init_rcu_head_on_stack(struct rcu_head *head)
350 static inline void destroy_rcu_head_on_stack(struct rcu_head *head)
353 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
355 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
356 bool rcu_lockdep_current_cpu_online(void);
357 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
358 static inline bool rcu_lockdep_current_cpu_online(void)
362 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
364 #ifdef CONFIG_DEBUG_LOCK_ALLOC
366 static inline void rcu_lock_acquire(struct lockdep_map *map)
368 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
371 static inline void rcu_lock_release(struct lockdep_map *map)
373 lock_release(map, 1, _THIS_IP_);
376 extern struct lockdep_map rcu_lock_map;
377 extern struct lockdep_map rcu_bh_lock_map;
378 extern struct lockdep_map rcu_sched_lock_map;
379 extern struct lockdep_map rcu_callback_map;
380 int debug_lockdep_rcu_enabled(void);
382 int rcu_read_lock_held(void);
383 int rcu_read_lock_bh_held(void);
386 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
388 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
389 * RCU-sched read-side critical section. In absence of
390 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
391 * critical section unless it can prove otherwise. Note that disabling
392 * of preemption (including disabling irqs) counts as an RCU-sched
393 * read-side critical section. This is useful for debug checks in functions
394 * that required that they be called within an RCU-sched read-side
397 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
398 * and while lockdep is disabled.
400 * Note that if the CPU is in the idle loop from an RCU point of
401 * view (ie: that we are in the section between rcu_idle_enter() and
402 * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU
403 * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs
404 * that are in such a section, considering these as in extended quiescent
405 * state, so such a CPU is effectively never in an RCU read-side critical
406 * section regardless of what RCU primitives it invokes. This state of
407 * affairs is required --- we need to keep an RCU-free window in idle
408 * where the CPU may possibly enter into low power mode. This way we can
409 * notice an extended quiescent state to other CPUs that started a grace
410 * period. Otherwise we would delay any grace period as long as we run in
413 * Similarly, we avoid claiming an SRCU read lock held if the current
416 #ifdef CONFIG_PREEMPT_COUNT
417 static inline int rcu_read_lock_sched_held(void)
419 int lockdep_opinion = 0;
421 if (!debug_lockdep_rcu_enabled())
423 if (!rcu_is_watching())
425 if (!rcu_lockdep_current_cpu_online())
428 lockdep_opinion = lock_is_held(&rcu_sched_lock_map);
429 return lockdep_opinion || preempt_count() != 0 || irqs_disabled();
431 #else /* #ifdef CONFIG_PREEMPT_COUNT */
432 static inline int rcu_read_lock_sched_held(void)
436 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
438 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
440 # define rcu_lock_acquire(a) do { } while (0)
441 # define rcu_lock_release(a) do { } while (0)
443 static inline int rcu_read_lock_held(void)
448 static inline int rcu_read_lock_bh_held(void)
453 #ifdef CONFIG_PREEMPT_COUNT
454 static inline int rcu_read_lock_sched_held(void)
456 return preempt_count() != 0 || irqs_disabled();
458 #else /* #ifdef CONFIG_PREEMPT_COUNT */
459 static inline int rcu_read_lock_sched_held(void)
463 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
465 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
467 #ifdef CONFIG_PROVE_RCU
470 * rcu_lockdep_assert - emit lockdep splat if specified condition not met
471 * @c: condition to check
472 * @s: informative message
474 #define rcu_lockdep_assert(c, s) \
476 static bool __section(.data.unlikely) __warned; \
477 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \
479 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
483 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
484 static inline void rcu_preempt_sleep_check(void)
486 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
487 "Illegal context switch in RCU read-side critical section");
489 #else /* #ifdef CONFIG_PROVE_RCU */
490 static inline void rcu_preempt_sleep_check(void)
493 #endif /* #else #ifdef CONFIG_PROVE_RCU */
495 #define rcu_sleep_check() \
497 rcu_preempt_sleep_check(); \
498 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \
499 "Illegal context switch in RCU-bh read-side critical section"); \
500 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \
501 "Illegal context switch in RCU-sched read-side critical section"); \
504 #else /* #ifdef CONFIG_PROVE_RCU */
506 #define rcu_lockdep_assert(c, s) do { } while (0)
507 #define rcu_sleep_check() do { } while (0)
509 #endif /* #else #ifdef CONFIG_PROVE_RCU */
512 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
513 * and rcu_assign_pointer(). Some of these could be folded into their
514 * callers, but they are left separate in order to ease introduction of
515 * multiple flavors of pointers to match the multiple flavors of RCU
516 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
521 #define rcu_dereference_sparse(p, space) \
522 ((void)(((typeof(*p) space *)p) == p))
523 #else /* #ifdef __CHECKER__ */
524 #define rcu_dereference_sparse(p, space)
525 #endif /* #else #ifdef __CHECKER__ */
527 #define __rcu_access_pointer(p, space) \
529 typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \
530 rcu_dereference_sparse(p, space); \
531 ((typeof(*p) __force __kernel *)(_________p1)); \
533 #define __rcu_dereference_check(p, c, space) \
535 typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \
536 rcu_lockdep_assert(c, "suspicious rcu_dereference_check() usage"); \
537 rcu_dereference_sparse(p, space); \
538 smp_read_barrier_depends(); /* Dependency order vs. p above. */ \
539 ((typeof(*p) __force __kernel *)(_________p1)); \
541 #define __rcu_dereference_protected(p, c, space) \
543 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected() usage"); \
544 rcu_dereference_sparse(p, space); \
545 ((typeof(*p) __force __kernel *)(p)); \
548 #define __rcu_access_index(p, space) \
550 typeof(p) _________p1 = ACCESS_ONCE(p); \
551 rcu_dereference_sparse(p, space); \
554 #define __rcu_dereference_index_check(p, c) \
556 typeof(p) _________p1 = ACCESS_ONCE(p); \
557 rcu_lockdep_assert(c, \
558 "suspicious rcu_dereference_index_check() usage"); \
559 smp_read_barrier_depends(); /* Dependency order vs. p above. */ \
564 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
565 * @v: The value to statically initialize with.
567 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
570 * rcu_assign_pointer() - assign to RCU-protected pointer
571 * @p: pointer to assign to
572 * @v: value to assign (publish)
574 * Assigns the specified value to the specified RCU-protected
575 * pointer, ensuring that any concurrent RCU readers will see
576 * any prior initialization.
578 * Inserts memory barriers on architectures that require them
579 * (which is most of them), and also prevents the compiler from
580 * reordering the code that initializes the structure after the pointer
581 * assignment. More importantly, this call documents which pointers
582 * will be dereferenced by RCU read-side code.
584 * In some special cases, you may use RCU_INIT_POINTER() instead
585 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
586 * to the fact that it does not constrain either the CPU or the compiler.
587 * That said, using RCU_INIT_POINTER() when you should have used
588 * rcu_assign_pointer() is a very bad thing that results in
589 * impossible-to-diagnose memory corruption. So please be careful.
590 * See the RCU_INIT_POINTER() comment header for details.
592 * Note that rcu_assign_pointer() evaluates each of its arguments only
593 * once, appearances notwithstanding. One of the "extra" evaluations
594 * is in typeof() and the other visible only to sparse (__CHECKER__),
595 * neither of which actually execute the argument. As with most cpp
596 * macros, this execute-arguments-only-once property is important, so
597 * please be careful when making changes to rcu_assign_pointer() and the
598 * other macros that it invokes.
600 #define rcu_assign_pointer(p, v) smp_store_release(&p, RCU_INITIALIZER(v))
603 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
604 * @p: The pointer to read
606 * Return the value of the specified RCU-protected pointer, but omit the
607 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
608 * when the value of this pointer is accessed, but the pointer is not
609 * dereferenced, for example, when testing an RCU-protected pointer against
610 * NULL. Although rcu_access_pointer() may also be used in cases where
611 * update-side locks prevent the value of the pointer from changing, you
612 * should instead use rcu_dereference_protected() for this use case.
614 * It is also permissible to use rcu_access_pointer() when read-side
615 * access to the pointer was removed at least one grace period ago, as
616 * is the case in the context of the RCU callback that is freeing up
617 * the data, or after a synchronize_rcu() returns. This can be useful
618 * when tearing down multi-linked structures after a grace period
621 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
624 * rcu_dereference_check() - rcu_dereference with debug checking
625 * @p: The pointer to read, prior to dereferencing
626 * @c: The conditions under which the dereference will take place
628 * Do an rcu_dereference(), but check that the conditions under which the
629 * dereference will take place are correct. Typically the conditions
630 * indicate the various locking conditions that should be held at that
631 * point. The check should return true if the conditions are satisfied.
632 * An implicit check for being in an RCU read-side critical section
633 * (rcu_read_lock()) is included.
637 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
639 * could be used to indicate to lockdep that foo->bar may only be dereferenced
640 * if either rcu_read_lock() is held, or that the lock required to replace
641 * the bar struct at foo->bar is held.
643 * Note that the list of conditions may also include indications of when a lock
644 * need not be held, for example during initialisation or destruction of the
647 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
648 * atomic_read(&foo->usage) == 0);
650 * Inserts memory barriers on architectures that require them
651 * (currently only the Alpha), prevents the compiler from refetching
652 * (and from merging fetches), and, more importantly, documents exactly
653 * which pointers are protected by RCU and checks that the pointer is
654 * annotated as __rcu.
656 #define rcu_dereference_check(p, c) \
657 __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu)
660 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
661 * @p: The pointer to read, prior to dereferencing
662 * @c: The conditions under which the dereference will take place
664 * This is the RCU-bh counterpart to rcu_dereference_check().
666 #define rcu_dereference_bh_check(p, c) \
667 __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu)
670 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
671 * @p: The pointer to read, prior to dereferencing
672 * @c: The conditions under which the dereference will take place
674 * This is the RCU-sched counterpart to rcu_dereference_check().
676 #define rcu_dereference_sched_check(p, c) \
677 __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \
680 #define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/
683 * The tracing infrastructure traces RCU (we want that), but unfortunately
684 * some of the RCU checks causes tracing to lock up the system.
686 * The tracing version of rcu_dereference_raw() must not call
687 * rcu_read_lock_held().
689 #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
692 * rcu_access_index() - fetch RCU index with no dereferencing
693 * @p: The index to read
695 * Return the value of the specified RCU-protected index, but omit the
696 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
697 * when the value of this index is accessed, but the index is not
698 * dereferenced, for example, when testing an RCU-protected index against
699 * -1. Although rcu_access_index() may also be used in cases where
700 * update-side locks prevent the value of the index from changing, you
701 * should instead use rcu_dereference_index_protected() for this use case.
703 #define rcu_access_index(p) __rcu_access_index((p), __rcu)
706 * rcu_dereference_index_check() - rcu_dereference for indices with debug checking
707 * @p: The pointer to read, prior to dereferencing
708 * @c: The conditions under which the dereference will take place
710 * Similar to rcu_dereference_check(), but omits the sparse checking.
711 * This allows rcu_dereference_index_check() to be used on integers,
712 * which can then be used as array indices. Attempting to use
713 * rcu_dereference_check() on an integer will give compiler warnings
714 * because the sparse address-space mechanism relies on dereferencing
715 * the RCU-protected pointer. Dereferencing integers is not something
716 * that even gcc will put up with.
718 * Note that this function does not implicitly check for RCU read-side
719 * critical sections. If this function gains lots of uses, it might
720 * make sense to provide versions for each flavor of RCU, but it does
721 * not make sense as of early 2010.
723 #define rcu_dereference_index_check(p, c) \
724 __rcu_dereference_index_check((p), (c))
727 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
728 * @p: The pointer to read, prior to dereferencing
729 * @c: The conditions under which the dereference will take place
731 * Return the value of the specified RCU-protected pointer, but omit
732 * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This
733 * is useful in cases where update-side locks prevent the value of the
734 * pointer from changing. Please note that this primitive does -not-
735 * prevent the compiler from repeating this reference or combining it
736 * with other references, so it should not be used without protection
737 * of appropriate locks.
739 * This function is only for update-side use. Using this function
740 * when protected only by rcu_read_lock() will result in infrequent
741 * but very ugly failures.
743 #define rcu_dereference_protected(p, c) \
744 __rcu_dereference_protected((p), (c), __rcu)
748 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
749 * @p: The pointer to read, prior to dereferencing
751 * This is a simple wrapper around rcu_dereference_check().
753 #define rcu_dereference(p) rcu_dereference_check(p, 0)
756 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
757 * @p: The pointer to read, prior to dereferencing
759 * Makes rcu_dereference_check() do the dirty work.
761 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
764 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
765 * @p: The pointer to read, prior to dereferencing
767 * Makes rcu_dereference_check() do the dirty work.
769 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
772 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
774 * When synchronize_rcu() is invoked on one CPU while other CPUs
775 * are within RCU read-side critical sections, then the
776 * synchronize_rcu() is guaranteed to block until after all the other
777 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
778 * on one CPU while other CPUs are within RCU read-side critical
779 * sections, invocation of the corresponding RCU callback is deferred
780 * until after the all the other CPUs exit their critical sections.
782 * Note, however, that RCU callbacks are permitted to run concurrently
783 * with new RCU read-side critical sections. One way that this can happen
784 * is via the following sequence of events: (1) CPU 0 enters an RCU
785 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
786 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
787 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
788 * callback is invoked. This is legal, because the RCU read-side critical
789 * section that was running concurrently with the call_rcu() (and which
790 * therefore might be referencing something that the corresponding RCU
791 * callback would free up) has completed before the corresponding
792 * RCU callback is invoked.
794 * RCU read-side critical sections may be nested. Any deferred actions
795 * will be deferred until the outermost RCU read-side critical section
798 * You can avoid reading and understanding the next paragraph by
799 * following this rule: don't put anything in an rcu_read_lock() RCU
800 * read-side critical section that would block in a !PREEMPT kernel.
801 * But if you want the full story, read on!
803 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU),
804 * it is illegal to block while in an RCU read-side critical section.
805 * In preemptible RCU implementations (TREE_PREEMPT_RCU) in CONFIG_PREEMPT
806 * kernel builds, RCU read-side critical sections may be preempted,
807 * but explicit blocking is illegal. Finally, in preemptible RCU
808 * implementations in real-time (with -rt patchset) kernel builds, RCU
809 * read-side critical sections may be preempted and they may also block, but
810 * only when acquiring spinlocks that are subject to priority inheritance.
812 static inline void rcu_read_lock(void)
816 rcu_lock_acquire(&rcu_lock_map);
817 rcu_lockdep_assert(rcu_is_watching(),
818 "rcu_read_lock() used illegally while idle");
822 * So where is rcu_write_lock()? It does not exist, as there is no
823 * way for writers to lock out RCU readers. This is a feature, not
824 * a bug -- this property is what provides RCU's performance benefits.
825 * Of course, writers must coordinate with each other. The normal
826 * spinlock primitives work well for this, but any other technique may be
827 * used as well. RCU does not care how the writers keep out of each
828 * others' way, as long as they do so.
832 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
834 * In most situations, rcu_read_unlock() is immune from deadlock.
835 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock()
836 * is responsible for deboosting, which it does via rt_mutex_unlock().
837 * Unfortunately, this function acquires the scheduler's runqueue and
838 * priority-inheritance spinlocks. This means that deadlock could result
839 * if the caller of rcu_read_unlock() already holds one of these locks or
840 * any lock that is ever acquired while holding them.
842 * That said, RCU readers are never priority boosted unless they were
843 * preempted. Therefore, one way to avoid deadlock is to make sure
844 * that preemption never happens within any RCU read-side critical
845 * section whose outermost rcu_read_unlock() is called with one of
846 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in
847 * a number of ways, for example, by invoking preempt_disable() before
848 * critical section's outermost rcu_read_lock().
850 * Given that the set of locks acquired by rt_mutex_unlock() might change
851 * at any time, a somewhat more future-proofed approach is to make sure
852 * that that preemption never happens within any RCU read-side critical
853 * section whose outermost rcu_read_unlock() is called with irqs disabled.
854 * This approach relies on the fact that rt_mutex_unlock() currently only
855 * acquires irq-disabled locks.
857 * The second of these two approaches is best in most situations,
858 * however, the first approach can also be useful, at least to those
859 * developers willing to keep abreast of the set of locks acquired by
862 * See rcu_read_lock() for more information.
864 static inline void rcu_read_unlock(void)
866 rcu_lockdep_assert(rcu_is_watching(),
867 "rcu_read_unlock() used illegally while idle");
868 rcu_lock_release(&rcu_lock_map);
874 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
876 * This is equivalent of rcu_read_lock(), but to be used when updates
877 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
878 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
879 * softirq handler to be a quiescent state, a process in RCU read-side
880 * critical section must be protected by disabling softirqs. Read-side
881 * critical sections in interrupt context can use just rcu_read_lock(),
882 * though this should at least be commented to avoid confusing people
885 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
886 * must occur in the same context, for example, it is illegal to invoke
887 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
888 * was invoked from some other task.
890 static inline void rcu_read_lock_bh(void)
894 rcu_lock_acquire(&rcu_bh_lock_map);
895 rcu_lockdep_assert(rcu_is_watching(),
896 "rcu_read_lock_bh() used illegally while idle");
900 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
902 * See rcu_read_lock_bh() for more information.
904 static inline void rcu_read_unlock_bh(void)
906 rcu_lockdep_assert(rcu_is_watching(),
907 "rcu_read_unlock_bh() used illegally while idle");
908 rcu_lock_release(&rcu_bh_lock_map);
914 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
916 * This is equivalent of rcu_read_lock(), but to be used when updates
917 * are being done using call_rcu_sched() or synchronize_rcu_sched().
918 * Read-side critical sections can also be introduced by anything that
919 * disables preemption, including local_irq_disable() and friends.
921 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
922 * must occur in the same context, for example, it is illegal to invoke
923 * rcu_read_unlock_sched() from process context if the matching
924 * rcu_read_lock_sched() was invoked from an NMI handler.
926 static inline void rcu_read_lock_sched(void)
929 __acquire(RCU_SCHED);
930 rcu_lock_acquire(&rcu_sched_lock_map);
931 rcu_lockdep_assert(rcu_is_watching(),
932 "rcu_read_lock_sched() used illegally while idle");
935 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
936 static inline notrace void rcu_read_lock_sched_notrace(void)
938 preempt_disable_notrace();
939 __acquire(RCU_SCHED);
943 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
945 * See rcu_read_lock_sched for more information.
947 static inline void rcu_read_unlock_sched(void)
949 rcu_lockdep_assert(rcu_is_watching(),
950 "rcu_read_unlock_sched() used illegally while idle");
951 rcu_lock_release(&rcu_sched_lock_map);
952 __release(RCU_SCHED);
956 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
957 static inline notrace void rcu_read_unlock_sched_notrace(void)
959 __release(RCU_SCHED);
960 preempt_enable_notrace();
964 * RCU_INIT_POINTER() - initialize an RCU protected pointer
966 * Initialize an RCU-protected pointer in special cases where readers
967 * do not need ordering constraints on the CPU or the compiler. These
970 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
971 * 2. The caller has taken whatever steps are required to prevent
972 * RCU readers from concurrently accessing this pointer -or-
973 * 3. The referenced data structure has already been exposed to
974 * readers either at compile time or via rcu_assign_pointer() -and-
975 * a. You have not made -any- reader-visible changes to
976 * this structure since then -or-
977 * b. It is OK for readers accessing this structure from its
978 * new location to see the old state of the structure. (For
979 * example, the changes were to statistical counters or to
980 * other state where exact synchronization is not required.)
982 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
983 * result in impossible-to-diagnose memory corruption. As in the structures
984 * will look OK in crash dumps, but any concurrent RCU readers might
985 * see pre-initialized values of the referenced data structure. So
986 * please be very careful how you use RCU_INIT_POINTER()!!!
988 * If you are creating an RCU-protected linked structure that is accessed
989 * by a single external-to-structure RCU-protected pointer, then you may
990 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
991 * pointers, but you must use rcu_assign_pointer() to initialize the
992 * external-to-structure pointer -after- you have completely initialized
993 * the reader-accessible portions of the linked structure.
995 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
996 * ordering guarantees for either the CPU or the compiler.
998 #define RCU_INIT_POINTER(p, v) \
1000 p = RCU_INITIALIZER(v); \
1004 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
1006 * GCC-style initialization for an RCU-protected pointer in a structure field.
1008 #define RCU_POINTER_INITIALIZER(p, v) \
1009 .p = RCU_INITIALIZER(v)
1012 * Does the specified offset indicate that the corresponding rcu_head
1013 * structure can be handled by kfree_rcu()?
1015 #define __is_kfree_rcu_offset(offset) ((offset) < 4096)
1018 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
1020 #define __kfree_rcu(head, offset) \
1022 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
1023 kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \
1027 * kfree_rcu() - kfree an object after a grace period.
1028 * @ptr: pointer to kfree
1029 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
1031 * Many rcu callbacks functions just call kfree() on the base structure.
1032 * These functions are trivial, but their size adds up, and furthermore
1033 * when they are used in a kernel module, that module must invoke the
1034 * high-latency rcu_barrier() function at module-unload time.
1036 * The kfree_rcu() function handles this issue. Rather than encoding a
1037 * function address in the embedded rcu_head structure, kfree_rcu() instead
1038 * encodes the offset of the rcu_head structure within the base structure.
1039 * Because the functions are not allowed in the low-order 4096 bytes of
1040 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
1041 * If the offset is larger than 4095 bytes, a compile-time error will
1042 * be generated in __kfree_rcu(). If this error is triggered, you can
1043 * either fall back to use of call_rcu() or rearrange the structure to
1044 * position the rcu_head structure into the first 4096 bytes.
1046 * Note that the allowable offset might decrease in the future, for example,
1047 * to allow something like kmem_cache_free_rcu().
1049 * The BUILD_BUG_ON check must not involve any function calls, hence the
1050 * checks are done in macros here.
1052 #define kfree_rcu(ptr, rcu_head) \
1053 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
1055 #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL)
1056 static inline int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1058 *delta_jiffies = ULONG_MAX;
1061 #endif /* #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL) */
1063 #if defined(CONFIG_RCU_NOCB_CPU_ALL)
1064 static inline bool rcu_is_nocb_cpu(int cpu) { return true; }
1065 #elif defined(CONFIG_RCU_NOCB_CPU)
1066 bool rcu_is_nocb_cpu(int cpu);
1068 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
1072 /* Only for use by adaptive-ticks code. */
1073 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
1074 bool rcu_sys_is_idle(void);
1075 void rcu_sysidle_force_exit(void);
1076 #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1078 static inline bool rcu_sys_is_idle(void)
1083 static inline void rcu_sysidle_force_exit(void)
1087 #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1090 #endif /* __LINUX_RCUPDATE_H */