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 <linux/percpu.h>
48 #include <asm/barrier.h>
50 extern int rcu_expedited; /* for sysctl */
51 #ifdef CONFIG_RCU_TORTURE_TEST
52 extern int rcutorture_runnable; /* for sysctl */
53 #endif /* #ifdef CONFIG_RCU_TORTURE_TEST */
55 enum rcutorture_type {
63 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
64 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
65 unsigned long *gpnum, unsigned long *completed);
66 void rcutorture_record_test_transition(void);
67 void rcutorture_record_progress(unsigned long vernum);
68 void do_trace_rcu_torture_read(const char *rcutorturename,
74 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
77 unsigned long *completed)
83 static inline void rcutorture_record_test_transition(void)
86 static inline void rcutorture_record_progress(unsigned long vernum)
89 #ifdef CONFIG_RCU_TRACE
90 void do_trace_rcu_torture_read(const char *rcutorturename,
96 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
101 #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b))
102 #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b))
103 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
104 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
105 #define ulong2long(a) (*(long *)(&(a)))
107 /* Exported common interfaces */
109 #ifdef CONFIG_PREEMPT_RCU
112 * call_rcu() - Queue an RCU callback for invocation after a grace period.
113 * @head: structure to be used for queueing the RCU updates.
114 * @func: actual callback function to be invoked after the grace period
116 * The callback function will be invoked some time after a full grace
117 * period elapses, in other words after all pre-existing RCU read-side
118 * critical sections have completed. However, the callback function
119 * might well execute concurrently with RCU read-side critical sections
120 * that started after call_rcu() was invoked. RCU read-side critical
121 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
124 * Note that all CPUs must agree that the grace period extended beyond
125 * all pre-existing RCU read-side critical section. On systems with more
126 * than one CPU, this means that when "func()" is invoked, each CPU is
127 * guaranteed to have executed a full memory barrier since the end of its
128 * last RCU read-side critical section whose beginning preceded the call
129 * to call_rcu(). It also means that each CPU executing an RCU read-side
130 * critical section that continues beyond the start of "func()" must have
131 * executed a memory barrier after the call_rcu() but before the beginning
132 * of that RCU read-side critical section. Note that these guarantees
133 * include CPUs that are offline, idle, or executing in user mode, as
134 * well as CPUs that are executing in the kernel.
136 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
137 * resulting RCU callback function "func()", then both CPU A and CPU B are
138 * guaranteed to execute a full memory barrier during the time interval
139 * between the call to call_rcu() and the invocation of "func()" -- even
140 * if CPU A and CPU B are the same CPU (but again only if the system has
141 * more than one CPU).
143 void call_rcu(struct rcu_head *head,
144 void (*func)(struct rcu_head *head));
146 #else /* #ifdef CONFIG_PREEMPT_RCU */
148 /* In classic RCU, call_rcu() is just call_rcu_sched(). */
149 #define call_rcu call_rcu_sched
151 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
154 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
155 * @head: structure to be used for queueing the RCU updates.
156 * @func: actual callback function to be invoked after the grace period
158 * The callback function will be invoked some time after a full grace
159 * period elapses, in other words after all currently executing RCU
160 * read-side critical sections have completed. call_rcu_bh() assumes
161 * that the read-side critical sections end on completion of a softirq
162 * handler. This means that read-side critical sections in process
163 * context must not be interrupted by softirqs. This interface is to be
164 * used when most of the read-side critical sections are in softirq context.
165 * RCU read-side critical sections are delimited by :
166 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context.
168 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
169 * These may be nested.
171 * See the description of call_rcu() for more detailed information on
172 * memory ordering guarantees.
174 void call_rcu_bh(struct rcu_head *head,
175 void (*func)(struct rcu_head *head));
178 * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
179 * @head: structure to be used for queueing the RCU updates.
180 * @func: actual callback function to be invoked after the grace period
182 * The callback function will be invoked some time after a full grace
183 * period elapses, in other words after all currently executing RCU
184 * read-side critical sections have completed. call_rcu_sched() assumes
185 * that the read-side critical sections end on enabling of preemption
186 * or on voluntary preemption.
187 * RCU read-side critical sections are delimited by :
188 * - rcu_read_lock_sched() and rcu_read_unlock_sched(),
190 * anything that disables preemption.
191 * These may be nested.
193 * See the description of call_rcu() for more detailed information on
194 * memory ordering guarantees.
196 void call_rcu_sched(struct rcu_head *head,
197 void (*func)(struct rcu_head *rcu));
199 void synchronize_sched(void);
201 #ifdef CONFIG_PREEMPT_RCU
203 void __rcu_read_lock(void);
204 void __rcu_read_unlock(void);
205 void rcu_read_unlock_special(struct task_struct *t);
206 void synchronize_rcu(void);
209 * Defined as a macro as it is a very low level header included from
210 * areas that don't even know about current. This gives the rcu_read_lock()
211 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
212 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
214 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
216 #else /* #ifdef CONFIG_PREEMPT_RCU */
218 static inline void __rcu_read_lock(void)
223 static inline void __rcu_read_unlock(void)
228 static inline void synchronize_rcu(void)
233 static inline int rcu_preempt_depth(void)
238 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
240 /* Internal to kernel */
242 void rcu_sched_qs(int cpu);
243 void rcu_bh_qs(int cpu);
244 void rcu_check_callbacks(int cpu, int user);
245 struct notifier_block;
246 void rcu_idle_enter(void);
247 void rcu_idle_exit(void);
248 void rcu_irq_enter(void);
249 void rcu_irq_exit(void);
251 #ifdef CONFIG_RCU_STALL_COMMON
252 void rcu_sysrq_start(void);
253 void rcu_sysrq_end(void);
254 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
255 static inline void rcu_sysrq_start(void)
258 static inline void rcu_sysrq_end(void)
261 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
263 #ifdef CONFIG_RCU_USER_QS
264 void rcu_user_enter(void);
265 void rcu_user_exit(void);
267 static inline void rcu_user_enter(void) { }
268 static inline void rcu_user_exit(void) { }
269 static inline void rcu_user_hooks_switch(struct task_struct *prev,
270 struct task_struct *next) { }
271 #endif /* CONFIG_RCU_USER_QS */
274 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
275 * @a: Code that RCU needs to pay attention to.
277 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
278 * in the inner idle loop, that is, between the rcu_idle_enter() and
279 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
280 * critical sections. However, things like powertop need tracepoints
281 * in the inner idle loop.
283 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
284 * will tell RCU that it needs to pay attending, invoke its argument
285 * (in this example, a call to the do_something_with_RCU() function),
286 * and then tell RCU to go back to ignoring this CPU. It is permissible
287 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently
288 * quite limited. If deeper nesting is required, it will be necessary
289 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly.
291 #define RCU_NONIDLE(a) \
294 do { a; } while (0); \
298 #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP)
299 bool __rcu_is_watching(void);
300 #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */
303 * Hooks for cond_resched() and friends to avoid RCU CPU stall warnings.
306 #define RCU_COND_RESCHED_LIM 256 /* ms vs. 100s of ms. */
307 DECLARE_PER_CPU(int, rcu_cond_resched_count);
308 void rcu_resched(void);
311 * Is it time to report RCU quiescent states?
313 * Note unsynchronized access to rcu_cond_resched_count. Yes, we might
314 * increment some random CPU's count, and possibly also load the result from
315 * yet another CPU's count. We might even clobber some other CPU's attempt
316 * to zero its counter. This is all OK because the goal is not precision,
317 * but rather reasonable amortization of rcu_note_context_switch() overhead
318 * and extremely high probability of avoiding RCU CPU stall warnings.
319 * Note that this function has to be preempted in just the wrong place,
320 * many thousands of times in a row, for anything bad to happen.
322 static inline bool rcu_should_resched(void)
324 return raw_cpu_inc_return(rcu_cond_resched_count) >=
325 RCU_COND_RESCHED_LIM;
329 * Report quiscent states to RCU if it is time to do so.
331 static inline void rcu_cond_resched(void)
333 if (unlikely(rcu_should_resched()))
338 * Infrastructure to implement the synchronize_() primitives in
339 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
342 typedef void call_rcu_func_t(struct rcu_head *head,
343 void (*func)(struct rcu_head *head));
344 void wait_rcu_gp(call_rcu_func_t crf);
346 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
347 #include <linux/rcutree.h>
348 #elif defined(CONFIG_TINY_RCU)
349 #include <linux/rcutiny.h>
351 #error "Unknown RCU implementation specified to kernel configuration"
355 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
356 * initialization and destruction of rcu_head on the stack. rcu_head structures
357 * allocated dynamically in the heap or defined statically don't need any
360 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
361 void init_rcu_head(struct rcu_head *head);
362 void destroy_rcu_head(struct rcu_head *head);
363 void init_rcu_head_on_stack(struct rcu_head *head);
364 void destroy_rcu_head_on_stack(struct rcu_head *head);
365 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
366 static inline void init_rcu_head(struct rcu_head *head)
370 static inline void destroy_rcu_head(struct rcu_head *head)
374 static inline void init_rcu_head_on_stack(struct rcu_head *head)
378 static inline void destroy_rcu_head_on_stack(struct rcu_head *head)
381 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
383 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
384 bool rcu_lockdep_current_cpu_online(void);
385 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
386 static inline bool rcu_lockdep_current_cpu_online(void)
390 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
392 #ifdef CONFIG_DEBUG_LOCK_ALLOC
394 static inline void rcu_lock_acquire(struct lockdep_map *map)
396 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
399 static inline void rcu_lock_release(struct lockdep_map *map)
401 lock_release(map, 1, _THIS_IP_);
404 extern struct lockdep_map rcu_lock_map;
405 extern struct lockdep_map rcu_bh_lock_map;
406 extern struct lockdep_map rcu_sched_lock_map;
407 extern struct lockdep_map rcu_callback_map;
408 int debug_lockdep_rcu_enabled(void);
411 * rcu_read_lock_held() - might we be in RCU read-side critical section?
413 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
414 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
415 * this assumes we are in an RCU read-side critical section unless it can
416 * prove otherwise. This is useful for debug checks in functions that
417 * require that they be called within an RCU read-side critical section.
419 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
420 * and while lockdep is disabled.
422 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
423 * occur in the same context, for example, it is illegal to invoke
424 * rcu_read_unlock() in process context if the matching rcu_read_lock()
425 * was invoked from within an irq handler.
427 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
428 * offline from an RCU perspective, so check for those as well.
430 static inline int rcu_read_lock_held(void)
432 if (!debug_lockdep_rcu_enabled())
434 if (!rcu_is_watching())
436 if (!rcu_lockdep_current_cpu_online())
438 return lock_is_held(&rcu_lock_map);
442 * rcu_read_lock_bh_held() is defined out of line to avoid #include-file
445 int rcu_read_lock_bh_held(void);
448 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
450 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
451 * RCU-sched read-side critical section. In absence of
452 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
453 * critical section unless it can prove otherwise. Note that disabling
454 * of preemption (including disabling irqs) counts as an RCU-sched
455 * read-side critical section. This is useful for debug checks in functions
456 * that required that they be called within an RCU-sched read-side
459 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
460 * and while lockdep is disabled.
462 * Note that if the CPU is in the idle loop from an RCU point of
463 * view (ie: that we are in the section between rcu_idle_enter() and
464 * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU
465 * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs
466 * that are in such a section, considering these as in extended quiescent
467 * state, so such a CPU is effectively never in an RCU read-side critical
468 * section regardless of what RCU primitives it invokes. This state of
469 * affairs is required --- we need to keep an RCU-free window in idle
470 * where the CPU may possibly enter into low power mode. This way we can
471 * notice an extended quiescent state to other CPUs that started a grace
472 * period. Otherwise we would delay any grace period as long as we run in
475 * Similarly, we avoid claiming an SRCU read lock held if the current
478 #ifdef CONFIG_PREEMPT_COUNT
479 static inline int rcu_read_lock_sched_held(void)
481 int lockdep_opinion = 0;
483 if (!debug_lockdep_rcu_enabled())
485 if (!rcu_is_watching())
487 if (!rcu_lockdep_current_cpu_online())
490 lockdep_opinion = lock_is_held(&rcu_sched_lock_map);
491 return lockdep_opinion || preempt_count() != 0 || irqs_disabled();
493 #else /* #ifdef CONFIG_PREEMPT_COUNT */
494 static inline int rcu_read_lock_sched_held(void)
498 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
500 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
502 # define rcu_lock_acquire(a) do { } while (0)
503 # define rcu_lock_release(a) do { } while (0)
505 static inline int rcu_read_lock_held(void)
510 static inline int rcu_read_lock_bh_held(void)
515 #ifdef CONFIG_PREEMPT_COUNT
516 static inline int rcu_read_lock_sched_held(void)
518 return preempt_count() != 0 || irqs_disabled();
520 #else /* #ifdef CONFIG_PREEMPT_COUNT */
521 static inline int rcu_read_lock_sched_held(void)
525 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
527 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
529 #ifdef CONFIG_PROVE_RCU
532 * rcu_lockdep_assert - emit lockdep splat if specified condition not met
533 * @c: condition to check
534 * @s: informative message
536 #define rcu_lockdep_assert(c, s) \
538 static bool __section(.data.unlikely) __warned; \
539 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \
541 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
545 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
546 static inline void rcu_preempt_sleep_check(void)
548 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
549 "Illegal context switch in RCU read-side critical section");
551 #else /* #ifdef CONFIG_PROVE_RCU */
552 static inline void rcu_preempt_sleep_check(void)
555 #endif /* #else #ifdef CONFIG_PROVE_RCU */
557 #define rcu_sleep_check() \
559 rcu_preempt_sleep_check(); \
560 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \
561 "Illegal context switch in RCU-bh read-side critical section"); \
562 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \
563 "Illegal context switch in RCU-sched read-side critical section"); \
566 #else /* #ifdef CONFIG_PROVE_RCU */
568 #define rcu_lockdep_assert(c, s) do { } while (0)
569 #define rcu_sleep_check() do { } while (0)
571 #endif /* #else #ifdef CONFIG_PROVE_RCU */
574 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
575 * and rcu_assign_pointer(). Some of these could be folded into their
576 * callers, but they are left separate in order to ease introduction of
577 * multiple flavors of pointers to match the multiple flavors of RCU
578 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
583 #define rcu_dereference_sparse(p, space) \
584 ((void)(((typeof(*p) space *)p) == p))
585 #else /* #ifdef __CHECKER__ */
586 #define rcu_dereference_sparse(p, space)
587 #endif /* #else #ifdef __CHECKER__ */
589 #define __rcu_access_pointer(p, space) \
591 typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \
592 rcu_dereference_sparse(p, space); \
593 ((typeof(*p) __force __kernel *)(_________p1)); \
595 #define __rcu_dereference_check(p, c, space) \
597 typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \
598 rcu_lockdep_assert(c, "suspicious rcu_dereference_check() usage"); \
599 rcu_dereference_sparse(p, space); \
600 smp_read_barrier_depends(); /* Dependency order vs. p above. */ \
601 ((typeof(*p) __force __kernel *)(_________p1)); \
603 #define __rcu_dereference_protected(p, c, space) \
605 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected() usage"); \
606 rcu_dereference_sparse(p, space); \
607 ((typeof(*p) __force __kernel *)(p)); \
610 #define __rcu_access_index(p, space) \
612 typeof(p) _________p1 = ACCESS_ONCE(p); \
613 rcu_dereference_sparse(p, space); \
616 #define __rcu_dereference_index_check(p, c) \
618 typeof(p) _________p1 = ACCESS_ONCE(p); \
619 rcu_lockdep_assert(c, \
620 "suspicious rcu_dereference_index_check() usage"); \
621 smp_read_barrier_depends(); /* Dependency order vs. p above. */ \
626 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
627 * @v: The value to statically initialize with.
629 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
632 * rcu_assign_pointer() - assign to RCU-protected pointer
633 * @p: pointer to assign to
634 * @v: value to assign (publish)
636 * Assigns the specified value to the specified RCU-protected
637 * pointer, ensuring that any concurrent RCU readers will see
638 * any prior initialization.
640 * Inserts memory barriers on architectures that require them
641 * (which is most of them), and also prevents the compiler from
642 * reordering the code that initializes the structure after the pointer
643 * assignment. More importantly, this call documents which pointers
644 * will be dereferenced by RCU read-side code.
646 * In some special cases, you may use RCU_INIT_POINTER() instead
647 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
648 * to the fact that it does not constrain either the CPU or the compiler.
649 * That said, using RCU_INIT_POINTER() when you should have used
650 * rcu_assign_pointer() is a very bad thing that results in
651 * impossible-to-diagnose memory corruption. So please be careful.
652 * See the RCU_INIT_POINTER() comment header for details.
654 * Note that rcu_assign_pointer() evaluates each of its arguments only
655 * once, appearances notwithstanding. One of the "extra" evaluations
656 * is in typeof() and the other visible only to sparse (__CHECKER__),
657 * neither of which actually execute the argument. As with most cpp
658 * macros, this execute-arguments-only-once property is important, so
659 * please be careful when making changes to rcu_assign_pointer() and the
660 * other macros that it invokes.
662 #define rcu_assign_pointer(p, v) smp_store_release(&p, RCU_INITIALIZER(v))
665 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
666 * @p: The pointer to read
668 * Return the value of the specified RCU-protected pointer, but omit the
669 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
670 * when the value of this pointer is accessed, but the pointer is not
671 * dereferenced, for example, when testing an RCU-protected pointer against
672 * NULL. Although rcu_access_pointer() may also be used in cases where
673 * update-side locks prevent the value of the pointer from changing, you
674 * should instead use rcu_dereference_protected() for this use case.
676 * It is also permissible to use rcu_access_pointer() when read-side
677 * access to the pointer was removed at least one grace period ago, as
678 * is the case in the context of the RCU callback that is freeing up
679 * the data, or after a synchronize_rcu() returns. This can be useful
680 * when tearing down multi-linked structures after a grace period
683 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
686 * rcu_dereference_check() - rcu_dereference with debug checking
687 * @p: The pointer to read, prior to dereferencing
688 * @c: The conditions under which the dereference will take place
690 * Do an rcu_dereference(), but check that the conditions under which the
691 * dereference will take place are correct. Typically the conditions
692 * indicate the various locking conditions that should be held at that
693 * point. The check should return true if the conditions are satisfied.
694 * An implicit check for being in an RCU read-side critical section
695 * (rcu_read_lock()) is included.
699 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
701 * could be used to indicate to lockdep that foo->bar may only be dereferenced
702 * if either rcu_read_lock() is held, or that the lock required to replace
703 * the bar struct at foo->bar is held.
705 * Note that the list of conditions may also include indications of when a lock
706 * need not be held, for example during initialisation or destruction of the
709 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
710 * atomic_read(&foo->usage) == 0);
712 * Inserts memory barriers on architectures that require them
713 * (currently only the Alpha), prevents the compiler from refetching
714 * (and from merging fetches), and, more importantly, documents exactly
715 * which pointers are protected by RCU and checks that the pointer is
716 * annotated as __rcu.
718 #define rcu_dereference_check(p, c) \
719 __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu)
722 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
723 * @p: The pointer to read, prior to dereferencing
724 * @c: The conditions under which the dereference will take place
726 * This is the RCU-bh counterpart to rcu_dereference_check().
728 #define rcu_dereference_bh_check(p, c) \
729 __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu)
732 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
733 * @p: The pointer to read, prior to dereferencing
734 * @c: The conditions under which the dereference will take place
736 * This is the RCU-sched counterpart to rcu_dereference_check().
738 #define rcu_dereference_sched_check(p, c) \
739 __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \
742 #define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/
745 * The tracing infrastructure traces RCU (we want that), but unfortunately
746 * some of the RCU checks causes tracing to lock up the system.
748 * The tracing version of rcu_dereference_raw() must not call
749 * rcu_read_lock_held().
751 #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
754 * rcu_access_index() - fetch RCU index with no dereferencing
755 * @p: The index to read
757 * Return the value of the specified RCU-protected index, but omit the
758 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
759 * when the value of this index is accessed, but the index is not
760 * dereferenced, for example, when testing an RCU-protected index against
761 * -1. Although rcu_access_index() may also be used in cases where
762 * update-side locks prevent the value of the index from changing, you
763 * should instead use rcu_dereference_index_protected() for this use case.
765 #define rcu_access_index(p) __rcu_access_index((p), __rcu)
768 * rcu_dereference_index_check() - rcu_dereference for indices with debug checking
769 * @p: The pointer to read, prior to dereferencing
770 * @c: The conditions under which the dereference will take place
772 * Similar to rcu_dereference_check(), but omits the sparse checking.
773 * This allows rcu_dereference_index_check() to be used on integers,
774 * which can then be used as array indices. Attempting to use
775 * rcu_dereference_check() on an integer will give compiler warnings
776 * because the sparse address-space mechanism relies on dereferencing
777 * the RCU-protected pointer. Dereferencing integers is not something
778 * that even gcc will put up with.
780 * Note that this function does not implicitly check for RCU read-side
781 * critical sections. If this function gains lots of uses, it might
782 * make sense to provide versions for each flavor of RCU, but it does
783 * not make sense as of early 2010.
785 #define rcu_dereference_index_check(p, c) \
786 __rcu_dereference_index_check((p), (c))
789 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
790 * @p: The pointer to read, prior to dereferencing
791 * @c: The conditions under which the dereference will take place
793 * Return the value of the specified RCU-protected pointer, but omit
794 * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This
795 * is useful in cases where update-side locks prevent the value of the
796 * pointer from changing. Please note that this primitive does -not-
797 * prevent the compiler from repeating this reference or combining it
798 * with other references, so it should not be used without protection
799 * of appropriate locks.
801 * This function is only for update-side use. Using this function
802 * when protected only by rcu_read_lock() will result in infrequent
803 * but very ugly failures.
805 #define rcu_dereference_protected(p, c) \
806 __rcu_dereference_protected((p), (c), __rcu)
810 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
811 * @p: The pointer to read, prior to dereferencing
813 * This is a simple wrapper around rcu_dereference_check().
815 #define rcu_dereference(p) rcu_dereference_check(p, 0)
818 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
819 * @p: The pointer to read, prior to dereferencing
821 * Makes rcu_dereference_check() do the dirty work.
823 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
826 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
827 * @p: The pointer to read, prior to dereferencing
829 * Makes rcu_dereference_check() do the dirty work.
831 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
834 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
836 * When synchronize_rcu() is invoked on one CPU while other CPUs
837 * are within RCU read-side critical sections, then the
838 * synchronize_rcu() is guaranteed to block until after all the other
839 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
840 * on one CPU while other CPUs are within RCU read-side critical
841 * sections, invocation of the corresponding RCU callback is deferred
842 * until after the all the other CPUs exit their critical sections.
844 * Note, however, that RCU callbacks are permitted to run concurrently
845 * with new RCU read-side critical sections. One way that this can happen
846 * is via the following sequence of events: (1) CPU 0 enters an RCU
847 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
848 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
849 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
850 * callback is invoked. This is legal, because the RCU read-side critical
851 * section that was running concurrently with the call_rcu() (and which
852 * therefore might be referencing something that the corresponding RCU
853 * callback would free up) has completed before the corresponding
854 * RCU callback is invoked.
856 * RCU read-side critical sections may be nested. Any deferred actions
857 * will be deferred until the outermost RCU read-side critical section
860 * You can avoid reading and understanding the next paragraph by
861 * following this rule: don't put anything in an rcu_read_lock() RCU
862 * read-side critical section that would block in a !PREEMPT kernel.
863 * But if you want the full story, read on!
865 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), it
866 * is illegal to block while in an RCU read-side critical section. In
867 * preemptible RCU implementations (TREE_PREEMPT_RCU and TINY_PREEMPT_RCU)
868 * in CONFIG_PREEMPT kernel builds, RCU read-side critical sections may
869 * be preempted, but explicit blocking is illegal. Finally, in preemptible
870 * RCU implementations in real-time (with -rt patchset) kernel builds,
871 * RCU read-side critical sections may be preempted and they may also
872 * block, but only when acquiring spinlocks that are subject to priority
875 static inline void rcu_read_lock(void)
879 rcu_lock_acquire(&rcu_lock_map);
880 rcu_lockdep_assert(rcu_is_watching(),
881 "rcu_read_lock() used illegally while idle");
885 * So where is rcu_write_lock()? It does not exist, as there is no
886 * way for writers to lock out RCU readers. This is a feature, not
887 * a bug -- this property is what provides RCU's performance benefits.
888 * Of course, writers must coordinate with each other. The normal
889 * spinlock primitives work well for this, but any other technique may be
890 * used as well. RCU does not care how the writers keep out of each
891 * others' way, as long as they do so.
895 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
897 * See rcu_read_lock() for more information.
899 static inline void rcu_read_unlock(void)
901 rcu_lockdep_assert(rcu_is_watching(),
902 "rcu_read_unlock() used illegally while idle");
903 rcu_lock_release(&rcu_lock_map);
909 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
911 * This is equivalent of rcu_read_lock(), but to be used when updates
912 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
913 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
914 * softirq handler to be a quiescent state, a process in RCU read-side
915 * critical section must be protected by disabling softirqs. Read-side
916 * critical sections in interrupt context can use just rcu_read_lock(),
917 * though this should at least be commented to avoid confusing people
920 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
921 * must occur in the same context, for example, it is illegal to invoke
922 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
923 * was invoked from some other task.
925 static inline void rcu_read_lock_bh(void)
929 rcu_lock_acquire(&rcu_bh_lock_map);
930 rcu_lockdep_assert(rcu_is_watching(),
931 "rcu_read_lock_bh() used illegally while idle");
935 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
937 * See rcu_read_lock_bh() for more information.
939 static inline void rcu_read_unlock_bh(void)
941 rcu_lockdep_assert(rcu_is_watching(),
942 "rcu_read_unlock_bh() used illegally while idle");
943 rcu_lock_release(&rcu_bh_lock_map);
949 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
951 * This is equivalent of rcu_read_lock(), but to be used when updates
952 * are being done using call_rcu_sched() or synchronize_rcu_sched().
953 * Read-side critical sections can also be introduced by anything that
954 * disables preemption, including local_irq_disable() and friends.
956 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
957 * must occur in the same context, for example, it is illegal to invoke
958 * rcu_read_unlock_sched() from process context if the matching
959 * rcu_read_lock_sched() was invoked from an NMI handler.
961 static inline void rcu_read_lock_sched(void)
964 __acquire(RCU_SCHED);
965 rcu_lock_acquire(&rcu_sched_lock_map);
966 rcu_lockdep_assert(rcu_is_watching(),
967 "rcu_read_lock_sched() used illegally while idle");
970 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
971 static inline notrace void rcu_read_lock_sched_notrace(void)
973 preempt_disable_notrace();
974 __acquire(RCU_SCHED);
978 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
980 * See rcu_read_lock_sched for more information.
982 static inline void rcu_read_unlock_sched(void)
984 rcu_lockdep_assert(rcu_is_watching(),
985 "rcu_read_unlock_sched() used illegally while idle");
986 rcu_lock_release(&rcu_sched_lock_map);
987 __release(RCU_SCHED);
991 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
992 static inline notrace void rcu_read_unlock_sched_notrace(void)
994 __release(RCU_SCHED);
995 preempt_enable_notrace();
999 * RCU_INIT_POINTER() - initialize an RCU protected pointer
1001 * Initialize an RCU-protected pointer in special cases where readers
1002 * do not need ordering constraints on the CPU or the compiler. These
1003 * special cases are:
1005 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
1006 * 2. The caller has taken whatever steps are required to prevent
1007 * RCU readers from concurrently accessing this pointer -or-
1008 * 3. The referenced data structure has already been exposed to
1009 * readers either at compile time or via rcu_assign_pointer() -and-
1010 * a. You have not made -any- reader-visible changes to
1011 * this structure since then -or-
1012 * b. It is OK for readers accessing this structure from its
1013 * new location to see the old state of the structure. (For
1014 * example, the changes were to statistical counters or to
1015 * other state where exact synchronization is not required.)
1017 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
1018 * result in impossible-to-diagnose memory corruption. As in the structures
1019 * will look OK in crash dumps, but any concurrent RCU readers might
1020 * see pre-initialized values of the referenced data structure. So
1021 * please be very careful how you use RCU_INIT_POINTER()!!!
1023 * If you are creating an RCU-protected linked structure that is accessed
1024 * by a single external-to-structure RCU-protected pointer, then you may
1025 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
1026 * pointers, but you must use rcu_assign_pointer() to initialize the
1027 * external-to-structure pointer -after- you have completely initialized
1028 * the reader-accessible portions of the linked structure.
1030 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
1031 * ordering guarantees for either the CPU or the compiler.
1033 #define RCU_INIT_POINTER(p, v) \
1035 p = RCU_INITIALIZER(v); \
1039 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
1041 * GCC-style initialization for an RCU-protected pointer in a structure field.
1043 #define RCU_POINTER_INITIALIZER(p, v) \
1044 .p = RCU_INITIALIZER(v)
1047 * Does the specified offset indicate that the corresponding rcu_head
1048 * structure can be handled by kfree_rcu()?
1050 #define __is_kfree_rcu_offset(offset) ((offset) < 4096)
1053 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
1055 #define __kfree_rcu(head, offset) \
1057 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
1058 kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \
1062 * kfree_rcu() - kfree an object after a grace period.
1063 * @ptr: pointer to kfree
1064 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
1066 * Many rcu callbacks functions just call kfree() on the base structure.
1067 * These functions are trivial, but their size adds up, and furthermore
1068 * when they are used in a kernel module, that module must invoke the
1069 * high-latency rcu_barrier() function at module-unload time.
1071 * The kfree_rcu() function handles this issue. Rather than encoding a
1072 * function address in the embedded rcu_head structure, kfree_rcu() instead
1073 * encodes the offset of the rcu_head structure within the base structure.
1074 * Because the functions are not allowed in the low-order 4096 bytes of
1075 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
1076 * If the offset is larger than 4095 bytes, a compile-time error will
1077 * be generated in __kfree_rcu(). If this error is triggered, you can
1078 * either fall back to use of call_rcu() or rearrange the structure to
1079 * position the rcu_head structure into the first 4096 bytes.
1081 * Note that the allowable offset might decrease in the future, for example,
1082 * to allow something like kmem_cache_free_rcu().
1084 * The BUILD_BUG_ON check must not involve any function calls, hence the
1085 * checks are done in macros here.
1087 #define kfree_rcu(ptr, rcu_head) \
1088 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
1090 #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL)
1091 static inline int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1093 *delta_jiffies = ULONG_MAX;
1096 #endif /* #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL) */
1098 #if defined(CONFIG_RCU_NOCB_CPU_ALL)
1099 static inline bool rcu_is_nocb_cpu(int cpu) { return true; }
1100 #elif defined(CONFIG_RCU_NOCB_CPU)
1101 bool rcu_is_nocb_cpu(int cpu);
1103 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
1107 /* Only for use by adaptive-ticks code. */
1108 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
1109 bool rcu_sys_is_idle(void);
1110 void rcu_sysidle_force_exit(void);
1111 #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1113 static inline bool rcu_sys_is_idle(void)
1118 static inline void rcu_sysidle_force_exit(void)
1122 #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1125 #endif /* __LINUX_RCUPDATE_H */