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 */
51 #ifdef CONFIG_TINY_RCU
52 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
53 static inline bool rcu_gp_is_expedited(void) /* Internal RCU use. */
58 static inline void rcu_expedite_gp(void)
62 static inline void rcu_unexpedite_gp(void)
65 #else /* #ifdef CONFIG_TINY_RCU */
66 bool rcu_gp_is_expedited(void); /* Internal RCU use. */
67 void rcu_expedite_gp(void);
68 void rcu_unexpedite_gp(void);
69 #endif /* #else #ifdef CONFIG_TINY_RCU */
71 enum rcutorture_type {
80 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
81 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
82 unsigned long *gpnum, unsigned long *completed);
83 void rcutorture_record_test_transition(void);
84 void rcutorture_record_progress(unsigned long vernum);
85 void do_trace_rcu_torture_read(const char *rcutorturename,
91 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
94 unsigned long *completed)
100 static inline void rcutorture_record_test_transition(void)
103 static inline void rcutorture_record_progress(unsigned long vernum)
106 #ifdef CONFIG_RCU_TRACE
107 void do_trace_rcu_torture_read(const char *rcutorturename,
108 struct rcu_head *rhp,
113 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
118 #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b))
119 #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b))
120 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
121 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
122 #define ulong2long(a) (*(long *)(&(a)))
124 /* Exported common interfaces */
126 #ifdef CONFIG_PREEMPT_RCU
129 * call_rcu() - Queue an RCU callback for invocation after a grace period.
130 * @head: structure to be used for queueing the RCU updates.
131 * @func: actual callback function to be invoked after the grace period
133 * The callback function will be invoked some time after a full grace
134 * period elapses, in other words after all pre-existing RCU read-side
135 * critical sections have completed. However, the callback function
136 * might well execute concurrently with RCU read-side critical sections
137 * that started after call_rcu() was invoked. RCU read-side critical
138 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
141 * Note that all CPUs must agree that the grace period extended beyond
142 * all pre-existing RCU read-side critical section. On systems with more
143 * than one CPU, this means that when "func()" is invoked, each CPU is
144 * guaranteed to have executed a full memory barrier since the end of its
145 * last RCU read-side critical section whose beginning preceded the call
146 * to call_rcu(). It also means that each CPU executing an RCU read-side
147 * critical section that continues beyond the start of "func()" must have
148 * executed a memory barrier after the call_rcu() but before the beginning
149 * of that RCU read-side critical section. Note that these guarantees
150 * include CPUs that are offline, idle, or executing in user mode, as
151 * well as CPUs that are executing in the kernel.
153 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
154 * resulting RCU callback function "func()", then both CPU A and CPU B are
155 * guaranteed to execute a full memory barrier during the time interval
156 * between the call to call_rcu() and the invocation of "func()" -- even
157 * if CPU A and CPU B are the same CPU (but again only if the system has
158 * more than one CPU).
160 void call_rcu(struct rcu_head *head,
161 void (*func)(struct rcu_head *head));
163 #else /* #ifdef CONFIG_PREEMPT_RCU */
165 /* In classic RCU, call_rcu() is just call_rcu_sched(). */
166 #define call_rcu call_rcu_sched
168 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
170 #ifdef CONFIG_PREEMPT_RT_FULL
171 #define call_rcu_bh call_rcu
174 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
175 * @head: structure to be used for queueing the RCU updates.
176 * @func: actual callback function to be invoked after the grace period
178 * The callback function will be invoked some time after a full grace
179 * period elapses, in other words after all currently executing RCU
180 * read-side critical sections have completed. call_rcu_bh() assumes
181 * that the read-side critical sections end on completion of a softirq
182 * handler. This means that read-side critical sections in process
183 * context must not be interrupted by softirqs. This interface is to be
184 * used when most of the read-side critical sections are in softirq context.
185 * RCU read-side critical sections are delimited by :
186 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context.
188 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
189 * These may be nested.
191 * See the description of call_rcu() for more detailed information on
192 * memory ordering guarantees.
194 void call_rcu_bh(struct rcu_head *head,
195 void (*func)(struct rcu_head *head));
199 * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
200 * @head: structure to be used for queueing the RCU updates.
201 * @func: actual callback function to be invoked after the grace period
203 * The callback function will be invoked some time after a full grace
204 * period elapses, in other words after all currently executing RCU
205 * read-side critical sections have completed. call_rcu_sched() assumes
206 * that the read-side critical sections end on enabling of preemption
207 * or on voluntary preemption.
208 * RCU read-side critical sections are delimited by :
209 * - rcu_read_lock_sched() and rcu_read_unlock_sched(),
211 * anything that disables preemption.
212 * These may be nested.
214 * See the description of call_rcu() for more detailed information on
215 * memory ordering guarantees.
217 void call_rcu_sched(struct rcu_head *head,
218 void (*func)(struct rcu_head *rcu));
220 void synchronize_sched(void);
223 * Structure allowing asynchronous waiting on RCU.
225 struct rcu_synchronize {
226 struct rcu_head head;
227 struct completion completion;
229 void wakeme_after_rcu(struct rcu_head *head);
232 * call_rcu_tasks() - Queue an RCU for invocation task-based grace period
233 * @head: structure to be used for queueing the RCU updates.
234 * @func: actual callback function to be invoked after the grace period
236 * The callback function will be invoked some time after a full grace
237 * period elapses, in other words after all currently executing RCU
238 * read-side critical sections have completed. call_rcu_tasks() assumes
239 * that the read-side critical sections end at a voluntary context
240 * switch (not a preemption!), entry into idle, or transition to usermode
241 * execution. As such, there are no read-side primitives analogous to
242 * rcu_read_lock() and rcu_read_unlock() because this primitive is intended
243 * to determine that all tasks have passed through a safe state, not so
244 * much for data-strcuture synchronization.
246 * See the description of call_rcu() for more detailed information on
247 * memory ordering guarantees.
249 void call_rcu_tasks(struct rcu_head *head, void (*func)(struct rcu_head *head));
250 void synchronize_rcu_tasks(void);
251 void rcu_barrier_tasks(void);
253 #ifdef CONFIG_PREEMPT_RCU
255 void __rcu_read_lock(void);
256 void __rcu_read_unlock(void);
257 void rcu_read_unlock_special(struct task_struct *t);
258 void synchronize_rcu(void);
261 * Defined as a macro as it is a very low level header included from
262 * areas that don't even know about current. This gives the rcu_read_lock()
263 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
264 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
266 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
267 #ifndef CONFIG_PREEMPT_RT_FULL
268 #define sched_rcu_preempt_depth() rcu_preempt_depth()
270 static inline int sched_rcu_preempt_depth(void) { return 0; }
273 #else /* #ifdef CONFIG_PREEMPT_RCU */
275 static inline void __rcu_read_lock(void)
280 static inline void __rcu_read_unlock(void)
285 static inline void synchronize_rcu(void)
290 static inline int rcu_preempt_depth(void)
295 #define sched_rcu_preempt_depth() rcu_preempt_depth()
297 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
299 /* Internal to kernel */
301 void rcu_end_inkernel_boot(void);
302 void rcu_sched_qs(void);
303 void rcu_bh_qs(void);
304 void rcu_check_callbacks(int user);
305 struct notifier_block;
306 void rcu_idle_enter(void);
307 void rcu_idle_exit(void);
308 void rcu_irq_enter(void);
309 void rcu_irq_exit(void);
310 int rcu_cpu_notify(struct notifier_block *self,
311 unsigned long action, void *hcpu);
313 #ifdef CONFIG_RCU_STALL_COMMON
314 void rcu_sysrq_start(void);
315 void rcu_sysrq_end(void);
316 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
317 static inline void rcu_sysrq_start(void)
320 static inline void rcu_sysrq_end(void)
323 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
325 #ifdef CONFIG_RCU_USER_QS
326 void rcu_user_enter(void);
327 void rcu_user_exit(void);
329 static inline void rcu_user_enter(void) { }
330 static inline void rcu_user_exit(void) { }
331 static inline void rcu_user_hooks_switch(struct task_struct *prev,
332 struct task_struct *next) { }
333 #endif /* CONFIG_RCU_USER_QS */
335 #ifdef CONFIG_RCU_NOCB_CPU
336 void rcu_init_nohz(void);
337 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
338 static inline void rcu_init_nohz(void)
341 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
344 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
345 * @a: Code that RCU needs to pay attention to.
347 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
348 * in the inner idle loop, that is, between the rcu_idle_enter() and
349 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
350 * critical sections. However, things like powertop need tracepoints
351 * in the inner idle loop.
353 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
354 * will tell RCU that it needs to pay attending, invoke its argument
355 * (in this example, a call to the do_something_with_RCU() function),
356 * and then tell RCU to go back to ignoring this CPU. It is permissible
357 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently
358 * quite limited. If deeper nesting is required, it will be necessary
359 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly.
361 #define RCU_NONIDLE(a) \
364 do { a; } while (0); \
369 * Note a voluntary context switch for RCU-tasks benefit. This is a
370 * macro rather than an inline function to avoid #include hell.
372 #ifdef CONFIG_TASKS_RCU
373 #define TASKS_RCU(x) x
374 extern struct srcu_struct tasks_rcu_exit_srcu;
375 #define rcu_note_voluntary_context_switch(t) \
378 if (ACCESS_ONCE((t)->rcu_tasks_holdout)) \
379 ACCESS_ONCE((t)->rcu_tasks_holdout) = false; \
381 #else /* #ifdef CONFIG_TASKS_RCU */
382 #define TASKS_RCU(x) do { } while (0)
383 #define rcu_note_voluntary_context_switch(t) rcu_all_qs()
384 #endif /* #else #ifdef CONFIG_TASKS_RCU */
387 * cond_resched_rcu_qs - Report potential quiescent states to RCU
389 * This macro resembles cond_resched(), except that it is defined to
390 * report potential quiescent states to RCU-tasks even if the cond_resched()
391 * machinery were to be shut off, as some advocate for PREEMPT kernels.
393 #define cond_resched_rcu_qs() \
395 if (!cond_resched()) \
396 rcu_note_voluntary_context_switch(current); \
399 #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP)
400 bool __rcu_is_watching(void);
401 #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */
404 * Infrastructure to implement the synchronize_() primitives in
405 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
408 typedef void call_rcu_func_t(struct rcu_head *head,
409 void (*func)(struct rcu_head *head));
410 void wait_rcu_gp(call_rcu_func_t crf);
412 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
413 #include <linux/rcutree.h>
414 #elif defined(CONFIG_TINY_RCU)
415 #include <linux/rcutiny.h>
417 #error "Unknown RCU implementation specified to kernel configuration"
421 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
422 * initialization and destruction of rcu_head on the stack. rcu_head structures
423 * allocated dynamically in the heap or defined statically don't need any
426 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
427 void init_rcu_head(struct rcu_head *head);
428 void destroy_rcu_head(struct rcu_head *head);
429 void init_rcu_head_on_stack(struct rcu_head *head);
430 void destroy_rcu_head_on_stack(struct rcu_head *head);
431 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
432 static inline void init_rcu_head(struct rcu_head *head)
436 static inline void destroy_rcu_head(struct rcu_head *head)
440 static inline void init_rcu_head_on_stack(struct rcu_head *head)
444 static inline void destroy_rcu_head_on_stack(struct rcu_head *head)
447 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
449 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
450 bool rcu_lockdep_current_cpu_online(void);
451 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
452 static inline bool rcu_lockdep_current_cpu_online(void)
456 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
458 #ifdef CONFIG_DEBUG_LOCK_ALLOC
460 static inline void rcu_lock_acquire(struct lockdep_map *map)
462 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
465 static inline void rcu_lock_release(struct lockdep_map *map)
467 lock_release(map, 1, _THIS_IP_);
470 extern struct lockdep_map rcu_lock_map;
471 extern struct lockdep_map rcu_bh_lock_map;
472 extern struct lockdep_map rcu_sched_lock_map;
473 extern struct lockdep_map rcu_callback_map;
474 int debug_lockdep_rcu_enabled(void);
476 int rcu_read_lock_held(void);
477 #ifdef CONFIG_PREEMPT_RT_FULL
478 static inline int rcu_read_lock_bh_held(void)
480 return rcu_read_lock_held();
483 int rcu_read_lock_bh_held(void);
487 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
489 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
490 * RCU-sched read-side critical section. In absence of
491 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
492 * critical section unless it can prove otherwise. Note that disabling
493 * of preemption (including disabling irqs) counts as an RCU-sched
494 * read-side critical section. This is useful for debug checks in functions
495 * that required that they be called within an RCU-sched read-side
498 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
499 * and while lockdep is disabled.
501 * Note that if the CPU is in the idle loop from an RCU point of
502 * view (ie: that we are in the section between rcu_idle_enter() and
503 * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU
504 * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs
505 * that are in such a section, considering these as in extended quiescent
506 * state, so such a CPU is effectively never in an RCU read-side critical
507 * section regardless of what RCU primitives it invokes. This state of
508 * affairs is required --- we need to keep an RCU-free window in idle
509 * where the CPU may possibly enter into low power mode. This way we can
510 * notice an extended quiescent state to other CPUs that started a grace
511 * period. Otherwise we would delay any grace period as long as we run in
514 * Similarly, we avoid claiming an SRCU read lock held if the current
517 #ifdef CONFIG_PREEMPT_COUNT
518 static inline int rcu_read_lock_sched_held(void)
520 int lockdep_opinion = 0;
522 if (!debug_lockdep_rcu_enabled())
524 if (!rcu_is_watching())
526 if (!rcu_lockdep_current_cpu_online())
529 lockdep_opinion = lock_is_held(&rcu_sched_lock_map);
530 return lockdep_opinion || preempt_count() != 0 || irqs_disabled();
532 #else /* #ifdef CONFIG_PREEMPT_COUNT */
533 static inline int rcu_read_lock_sched_held(void)
537 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
539 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
541 # define rcu_lock_acquire(a) do { } while (0)
542 # define rcu_lock_release(a) do { } while (0)
544 static inline int rcu_read_lock_held(void)
549 static inline int rcu_read_lock_bh_held(void)
554 #ifdef CONFIG_PREEMPT_COUNT
555 static inline int rcu_read_lock_sched_held(void)
557 return preempt_count() != 0 || irqs_disabled();
559 #else /* #ifdef CONFIG_PREEMPT_COUNT */
560 static inline int rcu_read_lock_sched_held(void)
564 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
566 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
568 #ifdef CONFIG_PROVE_RCU
571 * rcu_lockdep_assert - emit lockdep splat if specified condition not met
572 * @c: condition to check
573 * @s: informative message
575 #define rcu_lockdep_assert(c, s) \
577 static bool __section(.data.unlikely) __warned; \
578 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \
580 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
584 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
585 static inline void rcu_preempt_sleep_check(void)
587 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
588 "Illegal context switch in RCU read-side critical section");
590 #else /* #ifdef CONFIG_PROVE_RCU */
591 static inline void rcu_preempt_sleep_check(void)
594 #endif /* #else #ifdef CONFIG_PROVE_RCU */
596 #define rcu_sleep_check() \
598 rcu_preempt_sleep_check(); \
599 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \
600 "Illegal context switch in RCU-bh read-side critical section"); \
601 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \
602 "Illegal context switch in RCU-sched read-side critical section"); \
605 #else /* #ifdef CONFIG_PROVE_RCU */
607 #define rcu_lockdep_assert(c, s) do { } while (0)
608 #define rcu_sleep_check() do { } while (0)
610 #endif /* #else #ifdef CONFIG_PROVE_RCU */
613 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
614 * and rcu_assign_pointer(). Some of these could be folded into their
615 * callers, but they are left separate in order to ease introduction of
616 * multiple flavors of pointers to match the multiple flavors of RCU
617 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
622 #define rcu_dereference_sparse(p, space) \
623 ((void)(((typeof(*p) space *)p) == p))
624 #else /* #ifdef __CHECKER__ */
625 #define rcu_dereference_sparse(p, space)
626 #endif /* #else #ifdef __CHECKER__ */
628 #define __rcu_access_pointer(p, space) \
630 typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \
631 rcu_dereference_sparse(p, space); \
632 ((typeof(*p) __force __kernel *)(_________p1)); \
634 #define __rcu_dereference_check(p, c, space) \
636 /* Dependency order vs. p above. */ \
637 typeof(*p) *________p1 = (typeof(*p) *__force)lockless_dereference(p); \
638 rcu_lockdep_assert(c, "suspicious rcu_dereference_check() usage"); \
639 rcu_dereference_sparse(p, space); \
640 ((typeof(*p) __force __kernel *)(________p1)); \
642 #define __rcu_dereference_protected(p, c, space) \
644 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected() usage"); \
645 rcu_dereference_sparse(p, space); \
646 ((typeof(*p) __force __kernel *)(p)); \
649 #define __rcu_access_index(p, space) \
651 typeof(p) _________p1 = ACCESS_ONCE(p); \
652 rcu_dereference_sparse(p, space); \
655 #define __rcu_dereference_index_check(p, c) \
657 /* Dependency order vs. p above. */ \
658 typeof(p) _________p1 = lockless_dereference(p); \
659 rcu_lockdep_assert(c, \
660 "suspicious rcu_dereference_index_check() usage"); \
665 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
666 * @v: The value to statically initialize with.
668 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
671 * lockless_dereference() - safely load a pointer for later dereference
672 * @p: The pointer to load
674 * Similar to rcu_dereference(), but for situations where the pointed-to
675 * object's lifetime is managed by something other than RCU. That
676 * "something other" might be reference counting or simple immortality.
678 #define lockless_dereference(p) \
680 typeof(p) _________p1 = ACCESS_ONCE(p); \
681 smp_read_barrier_depends(); /* Dependency order vs. p above. */ \
686 * rcu_assign_pointer() - assign to RCU-protected pointer
687 * @p: pointer to assign to
688 * @v: value to assign (publish)
690 * Assigns the specified value to the specified RCU-protected
691 * pointer, ensuring that any concurrent RCU readers will see
692 * any prior initialization.
694 * Inserts memory barriers on architectures that require them
695 * (which is most of them), and also prevents the compiler from
696 * reordering the code that initializes the structure after the pointer
697 * assignment. More importantly, this call documents which pointers
698 * will be dereferenced by RCU read-side code.
700 * In some special cases, you may use RCU_INIT_POINTER() instead
701 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
702 * to the fact that it does not constrain either the CPU or the compiler.
703 * That said, using RCU_INIT_POINTER() when you should have used
704 * rcu_assign_pointer() is a very bad thing that results in
705 * impossible-to-diagnose memory corruption. So please be careful.
706 * See the RCU_INIT_POINTER() comment header for details.
708 * Note that rcu_assign_pointer() evaluates each of its arguments only
709 * once, appearances notwithstanding. One of the "extra" evaluations
710 * is in typeof() and the other visible only to sparse (__CHECKER__),
711 * neither of which actually execute the argument. As with most cpp
712 * macros, this execute-arguments-only-once property is important, so
713 * please be careful when making changes to rcu_assign_pointer() and the
714 * other macros that it invokes.
716 #define rcu_assign_pointer(p, v) smp_store_release(&p, RCU_INITIALIZER(v))
719 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
720 * @p: The pointer to read
722 * Return the value of the specified RCU-protected pointer, but omit the
723 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
724 * when the value of this pointer is accessed, but the pointer is not
725 * dereferenced, for example, when testing an RCU-protected pointer against
726 * NULL. Although rcu_access_pointer() may also be used in cases where
727 * update-side locks prevent the value of the pointer from changing, you
728 * should instead use rcu_dereference_protected() for this use case.
730 * It is also permissible to use rcu_access_pointer() when read-side
731 * access to the pointer was removed at least one grace period ago, as
732 * is the case in the context of the RCU callback that is freeing up
733 * the data, or after a synchronize_rcu() returns. This can be useful
734 * when tearing down multi-linked structures after a grace period
737 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
740 * rcu_dereference_check() - rcu_dereference with debug checking
741 * @p: The pointer to read, prior to dereferencing
742 * @c: The conditions under which the dereference will take place
744 * Do an rcu_dereference(), but check that the conditions under which the
745 * dereference will take place are correct. Typically the conditions
746 * indicate the various locking conditions that should be held at that
747 * point. The check should return true if the conditions are satisfied.
748 * An implicit check for being in an RCU read-side critical section
749 * (rcu_read_lock()) is included.
753 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
755 * could be used to indicate to lockdep that foo->bar may only be dereferenced
756 * if either rcu_read_lock() is held, or that the lock required to replace
757 * the bar struct at foo->bar is held.
759 * Note that the list of conditions may also include indications of when a lock
760 * need not be held, for example during initialisation or destruction of the
763 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
764 * atomic_read(&foo->usage) == 0);
766 * Inserts memory barriers on architectures that require them
767 * (currently only the Alpha), prevents the compiler from refetching
768 * (and from merging fetches), and, more importantly, documents exactly
769 * which pointers are protected by RCU and checks that the pointer is
770 * annotated as __rcu.
772 #define rcu_dereference_check(p, c) \
773 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu)
776 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
777 * @p: The pointer to read, prior to dereferencing
778 * @c: The conditions under which the dereference will take place
780 * This is the RCU-bh counterpart to rcu_dereference_check().
782 #define rcu_dereference_bh_check(p, c) \
783 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu)
786 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
787 * @p: The pointer to read, prior to dereferencing
788 * @c: The conditions under which the dereference will take place
790 * This is the RCU-sched counterpart to rcu_dereference_check().
792 #define rcu_dereference_sched_check(p, c) \
793 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \
796 #define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/
799 * The tracing infrastructure traces RCU (we want that), but unfortunately
800 * some of the RCU checks causes tracing to lock up the system.
802 * The tracing version of rcu_dereference_raw() must not call
803 * rcu_read_lock_held().
805 #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
808 * rcu_access_index() - fetch RCU index with no dereferencing
809 * @p: The index to read
811 * Return the value of the specified RCU-protected index, but omit the
812 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
813 * when the value of this index is accessed, but the index is not
814 * dereferenced, for example, when testing an RCU-protected index against
815 * -1. Although rcu_access_index() may also be used in cases where
816 * update-side locks prevent the value of the index from changing, you
817 * should instead use rcu_dereference_index_protected() for this use case.
819 #define rcu_access_index(p) __rcu_access_index((p), __rcu)
822 * rcu_dereference_index_check() - rcu_dereference for indices with debug checking
823 * @p: The pointer to read, prior to dereferencing
824 * @c: The conditions under which the dereference will take place
826 * Similar to rcu_dereference_check(), but omits the sparse checking.
827 * This allows rcu_dereference_index_check() to be used on integers,
828 * which can then be used as array indices. Attempting to use
829 * rcu_dereference_check() on an integer will give compiler warnings
830 * because the sparse address-space mechanism relies on dereferencing
831 * the RCU-protected pointer. Dereferencing integers is not something
832 * that even gcc will put up with.
834 * Note that this function does not implicitly check for RCU read-side
835 * critical sections. If this function gains lots of uses, it might
836 * make sense to provide versions for each flavor of RCU, but it does
837 * not make sense as of early 2010.
839 #define rcu_dereference_index_check(p, c) \
840 __rcu_dereference_index_check((p), (c))
843 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
844 * @p: The pointer to read, prior to dereferencing
845 * @c: The conditions under which the dereference will take place
847 * Return the value of the specified RCU-protected pointer, but omit
848 * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This
849 * is useful in cases where update-side locks prevent the value of the
850 * pointer from changing. Please note that this primitive does -not-
851 * prevent the compiler from repeating this reference or combining it
852 * with other references, so it should not be used without protection
853 * of appropriate locks.
855 * This function is only for update-side use. Using this function
856 * when protected only by rcu_read_lock() will result in infrequent
857 * but very ugly failures.
859 #define rcu_dereference_protected(p, c) \
860 __rcu_dereference_protected((p), (c), __rcu)
864 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
865 * @p: The pointer to read, prior to dereferencing
867 * This is a simple wrapper around rcu_dereference_check().
869 #define rcu_dereference(p) rcu_dereference_check(p, 0)
872 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
873 * @p: The pointer to read, prior to dereferencing
875 * Makes rcu_dereference_check() do the dirty work.
877 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
880 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
881 * @p: The pointer to read, prior to dereferencing
883 * Makes rcu_dereference_check() do the dirty work.
885 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
888 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
890 * When synchronize_rcu() is invoked on one CPU while other CPUs
891 * are within RCU read-side critical sections, then the
892 * synchronize_rcu() is guaranteed to block until after all the other
893 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
894 * on one CPU while other CPUs are within RCU read-side critical
895 * sections, invocation of the corresponding RCU callback is deferred
896 * until after the all the other CPUs exit their critical sections.
898 * Note, however, that RCU callbacks are permitted to run concurrently
899 * with new RCU read-side critical sections. One way that this can happen
900 * is via the following sequence of events: (1) CPU 0 enters an RCU
901 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
902 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
903 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
904 * callback is invoked. This is legal, because the RCU read-side critical
905 * section that was running concurrently with the call_rcu() (and which
906 * therefore might be referencing something that the corresponding RCU
907 * callback would free up) has completed before the corresponding
908 * RCU callback is invoked.
910 * RCU read-side critical sections may be nested. Any deferred actions
911 * will be deferred until the outermost RCU read-side critical section
914 * You can avoid reading and understanding the next paragraph by
915 * following this rule: don't put anything in an rcu_read_lock() RCU
916 * read-side critical section that would block in a !PREEMPT kernel.
917 * But if you want the full story, read on!
919 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU),
920 * it is illegal to block while in an RCU read-side critical section.
921 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT
922 * kernel builds, RCU read-side critical sections may be preempted,
923 * but explicit blocking is illegal. Finally, in preemptible RCU
924 * implementations in real-time (with -rt patchset) kernel builds, RCU
925 * read-side critical sections may be preempted and they may also block, but
926 * only when acquiring spinlocks that are subject to priority inheritance.
928 static inline void rcu_read_lock(void)
932 rcu_lock_acquire(&rcu_lock_map);
933 rcu_lockdep_assert(rcu_is_watching(),
934 "rcu_read_lock() used illegally while idle");
938 * So where is rcu_write_lock()? It does not exist, as there is no
939 * way for writers to lock out RCU readers. This is a feature, not
940 * a bug -- this property is what provides RCU's performance benefits.
941 * Of course, writers must coordinate with each other. The normal
942 * spinlock primitives work well for this, but any other technique may be
943 * used as well. RCU does not care how the writers keep out of each
944 * others' way, as long as they do so.
948 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
950 * In most situations, rcu_read_unlock() is immune from deadlock.
951 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock()
952 * is responsible for deboosting, which it does via rt_mutex_unlock().
953 * Unfortunately, this function acquires the scheduler's runqueue and
954 * priority-inheritance spinlocks. This means that deadlock could result
955 * if the caller of rcu_read_unlock() already holds one of these locks or
956 * any lock that is ever acquired while holding them; or any lock which
957 * can be taken from interrupt context because rcu_boost()->rt_mutex_lock()
958 * does not disable irqs while taking ->wait_lock.
960 * That said, RCU readers are never priority boosted unless they were
961 * preempted. Therefore, one way to avoid deadlock is to make sure
962 * that preemption never happens within any RCU read-side critical
963 * section whose outermost rcu_read_unlock() is called with one of
964 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in
965 * a number of ways, for example, by invoking preempt_disable() before
966 * critical section's outermost rcu_read_lock().
968 * Given that the set of locks acquired by rt_mutex_unlock() might change
969 * at any time, a somewhat more future-proofed approach is to make sure
970 * that that preemption never happens within any RCU read-side critical
971 * section whose outermost rcu_read_unlock() is called with irqs disabled.
972 * This approach relies on the fact that rt_mutex_unlock() currently only
973 * acquires irq-disabled locks.
975 * The second of these two approaches is best in most situations,
976 * however, the first approach can also be useful, at least to those
977 * developers willing to keep abreast of the set of locks acquired by
980 * See rcu_read_lock() for more information.
982 static inline void rcu_read_unlock(void)
984 rcu_lockdep_assert(rcu_is_watching(),
985 "rcu_read_unlock() used illegally while idle");
988 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
992 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
994 * This is equivalent of rcu_read_lock(), but to be used when updates
995 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
996 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
997 * softirq handler to be a quiescent state, a process in RCU read-side
998 * critical section must be protected by disabling softirqs. Read-side
999 * critical sections in interrupt context can use just rcu_read_lock(),
1000 * though this should at least be commented to avoid confusing people
1003 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
1004 * must occur in the same context, for example, it is illegal to invoke
1005 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
1006 * was invoked from some other task.
1008 static inline void rcu_read_lock_bh(void)
1011 #ifdef CONFIG_PREEMPT_RT_FULL
1015 rcu_lock_acquire(&rcu_bh_lock_map);
1016 rcu_lockdep_assert(rcu_is_watching(),
1017 "rcu_read_lock_bh() used illegally while idle");
1022 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
1024 * See rcu_read_lock_bh() for more information.
1026 static inline void rcu_read_unlock_bh(void)
1028 #ifdef CONFIG_PREEMPT_RT_FULL
1031 rcu_lockdep_assert(rcu_is_watching(),
1032 "rcu_read_unlock_bh() used illegally while idle");
1033 rcu_lock_release(&rcu_bh_lock_map);
1040 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
1042 * This is equivalent of rcu_read_lock(), but to be used when updates
1043 * are being done using call_rcu_sched() or synchronize_rcu_sched().
1044 * Read-side critical sections can also be introduced by anything that
1045 * disables preemption, including local_irq_disable() and friends.
1047 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
1048 * must occur in the same context, for example, it is illegal to invoke
1049 * rcu_read_unlock_sched() from process context if the matching
1050 * rcu_read_lock_sched() was invoked from an NMI handler.
1052 static inline void rcu_read_lock_sched(void)
1055 __acquire(RCU_SCHED);
1056 rcu_lock_acquire(&rcu_sched_lock_map);
1057 rcu_lockdep_assert(rcu_is_watching(),
1058 "rcu_read_lock_sched() used illegally while idle");
1061 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
1062 static inline notrace void rcu_read_lock_sched_notrace(void)
1064 preempt_disable_notrace();
1065 __acquire(RCU_SCHED);
1069 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
1071 * See rcu_read_lock_sched for more information.
1073 static inline void rcu_read_unlock_sched(void)
1075 rcu_lockdep_assert(rcu_is_watching(),
1076 "rcu_read_unlock_sched() used illegally while idle");
1077 rcu_lock_release(&rcu_sched_lock_map);
1078 __release(RCU_SCHED);
1082 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
1083 static inline notrace void rcu_read_unlock_sched_notrace(void)
1085 __release(RCU_SCHED);
1086 preempt_enable_notrace();
1090 * RCU_INIT_POINTER() - initialize an RCU protected pointer
1092 * Initialize an RCU-protected pointer in special cases where readers
1093 * do not need ordering constraints on the CPU or the compiler. These
1094 * special cases are:
1096 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
1097 * 2. The caller has taken whatever steps are required to prevent
1098 * RCU readers from concurrently accessing this pointer -or-
1099 * 3. The referenced data structure has already been exposed to
1100 * readers either at compile time or via rcu_assign_pointer() -and-
1101 * a. You have not made -any- reader-visible changes to
1102 * this structure since then -or-
1103 * b. It is OK for readers accessing this structure from its
1104 * new location to see the old state of the structure. (For
1105 * example, the changes were to statistical counters or to
1106 * other state where exact synchronization is not required.)
1108 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
1109 * result in impossible-to-diagnose memory corruption. As in the structures
1110 * will look OK in crash dumps, but any concurrent RCU readers might
1111 * see pre-initialized values of the referenced data structure. So
1112 * please be very careful how you use RCU_INIT_POINTER()!!!
1114 * If you are creating an RCU-protected linked structure that is accessed
1115 * by a single external-to-structure RCU-protected pointer, then you may
1116 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
1117 * pointers, but you must use rcu_assign_pointer() to initialize the
1118 * external-to-structure pointer -after- you have completely initialized
1119 * the reader-accessible portions of the linked structure.
1121 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
1122 * ordering guarantees for either the CPU or the compiler.
1124 #define RCU_INIT_POINTER(p, v) \
1126 rcu_dereference_sparse(p, __rcu); \
1127 p = RCU_INITIALIZER(v); \
1131 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
1133 * GCC-style initialization for an RCU-protected pointer in a structure field.
1135 #define RCU_POINTER_INITIALIZER(p, v) \
1136 .p = RCU_INITIALIZER(v)
1139 * Does the specified offset indicate that the corresponding rcu_head
1140 * structure can be handled by kfree_rcu()?
1142 #define __is_kfree_rcu_offset(offset) ((offset) < 4096)
1145 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
1147 #define __kfree_rcu(head, offset) \
1149 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
1150 kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \
1154 * kfree_rcu() - kfree an object after a grace period.
1155 * @ptr: pointer to kfree
1156 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
1158 * Many rcu callbacks functions just call kfree() on the base structure.
1159 * These functions are trivial, but their size adds up, and furthermore
1160 * when they are used in a kernel module, that module must invoke the
1161 * high-latency rcu_barrier() function at module-unload time.
1163 * The kfree_rcu() function handles this issue. Rather than encoding a
1164 * function address in the embedded rcu_head structure, kfree_rcu() instead
1165 * encodes the offset of the rcu_head structure within the base structure.
1166 * Because the functions are not allowed in the low-order 4096 bytes of
1167 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
1168 * If the offset is larger than 4095 bytes, a compile-time error will
1169 * be generated in __kfree_rcu(). If this error is triggered, you can
1170 * either fall back to use of call_rcu() or rearrange the structure to
1171 * position the rcu_head structure into the first 4096 bytes.
1173 * Note that the allowable offset might decrease in the future, for example,
1174 * to allow something like kmem_cache_free_rcu().
1176 * The BUILD_BUG_ON check must not involve any function calls, hence the
1177 * checks are done in macros here.
1179 #define kfree_rcu(ptr, rcu_head) \
1180 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
1182 #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL)
1183 static inline int rcu_needs_cpu(unsigned long *delta_jiffies)
1185 *delta_jiffies = ULONG_MAX;
1188 #endif /* #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL) */
1190 #if defined(CONFIG_RCU_NOCB_CPU_ALL)
1191 static inline bool rcu_is_nocb_cpu(int cpu) { return true; }
1192 #elif defined(CONFIG_RCU_NOCB_CPU)
1193 bool rcu_is_nocb_cpu(int cpu);
1195 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
1199 /* Only for use by adaptive-ticks code. */
1200 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
1201 bool rcu_sys_is_idle(void);
1202 void rcu_sysidle_force_exit(void);
1203 #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1205 static inline bool rcu_sys_is_idle(void)
1210 static inline void rcu_sysidle_force_exit(void)
1214 #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1217 #endif /* __LINUX_RCUPDATE_H */