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 {
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);
202 * call_rcu_tasks() - Queue an RCU for invocation task-based grace period
203 * @head: structure to be used for queueing the RCU updates.
204 * @func: actual callback function to be invoked after the grace period
206 * The callback function will be invoked some time after a full grace
207 * period elapses, in other words after all currently executing RCU
208 * read-side critical sections have completed. call_rcu_tasks() assumes
209 * that the read-side critical sections end at a voluntary context
210 * switch (not a preemption!), entry into idle, or transition to usermode
211 * execution. As such, there are no read-side primitives analogous to
212 * rcu_read_lock() and rcu_read_unlock() because this primitive is intended
213 * to determine that all tasks have passed through a safe state, not so
214 * much for data-strcuture synchronization.
216 * See the description of call_rcu() for more detailed information on
217 * memory ordering guarantees.
219 void call_rcu_tasks(struct rcu_head *head, void (*func)(struct rcu_head *head));
220 void synchronize_rcu_tasks(void);
221 void rcu_barrier_tasks(void);
223 #ifdef CONFIG_PREEMPT_RCU
225 void __rcu_read_lock(void);
226 void __rcu_read_unlock(void);
227 void rcu_read_unlock_special(struct task_struct *t);
228 void synchronize_rcu(void);
231 * Defined as a macro as it is a very low level header included from
232 * areas that don't even know about current. This gives the rcu_read_lock()
233 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
234 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
236 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
238 #else /* #ifdef CONFIG_PREEMPT_RCU */
240 static inline void __rcu_read_lock(void)
245 static inline void __rcu_read_unlock(void)
250 static inline void synchronize_rcu(void)
255 static inline int rcu_preempt_depth(void)
260 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
262 /* Internal to kernel */
264 void rcu_sched_qs(int cpu);
265 void rcu_bh_qs(int cpu);
266 void rcu_check_callbacks(int cpu, int user);
267 struct notifier_block;
268 void rcu_idle_enter(void);
269 void rcu_idle_exit(void);
270 void rcu_irq_enter(void);
271 void rcu_irq_exit(void);
273 #ifdef CONFIG_RCU_STALL_COMMON
274 void rcu_sysrq_start(void);
275 void rcu_sysrq_end(void);
276 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
277 static inline void rcu_sysrq_start(void)
280 static inline void rcu_sysrq_end(void)
283 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
285 #ifdef CONFIG_RCU_USER_QS
286 void rcu_user_enter(void);
287 void rcu_user_exit(void);
289 static inline void rcu_user_enter(void) { }
290 static inline void rcu_user_exit(void) { }
291 static inline void rcu_user_hooks_switch(struct task_struct *prev,
292 struct task_struct *next) { }
293 #endif /* CONFIG_RCU_USER_QS */
296 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
297 * @a: Code that RCU needs to pay attention to.
299 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
300 * in the inner idle loop, that is, between the rcu_idle_enter() and
301 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
302 * critical sections. However, things like powertop need tracepoints
303 * in the inner idle loop.
305 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
306 * will tell RCU that it needs to pay attending, invoke its argument
307 * (in this example, a call to the do_something_with_RCU() function),
308 * and then tell RCU to go back to ignoring this CPU. It is permissible
309 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently
310 * quite limited. If deeper nesting is required, it will be necessary
311 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly.
313 #define RCU_NONIDLE(a) \
316 do { a; } while (0); \
321 * Note a voluntary context switch for RCU-tasks benefit. This is a
322 * macro rather than an inline function to avoid #include hell.
324 #ifdef CONFIG_TASKS_RCU
325 #define TASKS_RCU(x) x
326 extern struct srcu_struct tasks_rcu_exit_srcu;
327 #define rcu_note_voluntary_context_switch(t) \
329 preempt_disable(); /* Exclude synchronize_sched(); */ \
330 if (ACCESS_ONCE((t)->rcu_tasks_holdout)) \
331 ACCESS_ONCE((t)->rcu_tasks_holdout) = false; \
334 #else /* #ifdef CONFIG_TASKS_RCU */
335 #define TASKS_RCU(x) do { } while (0)
336 #define rcu_note_voluntary_context_switch(t) do { } while (0)
337 #endif /* #else #ifdef CONFIG_TASKS_RCU */
340 * cond_resched_rcu_qs - Report potential quiescent states to RCU
342 * This macro resembles cond_resched(), except that it is defined to
343 * report potential quiescent states to RCU-tasks even if the cond_resched()
344 * machinery were to be shut off, as some advocate for PREEMPT kernels.
346 #define cond_resched_rcu_qs() \
348 rcu_note_voluntary_context_switch(current); \
352 #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP)
353 bool __rcu_is_watching(void);
354 #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */
357 * Infrastructure to implement the synchronize_() primitives in
358 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
361 typedef void call_rcu_func_t(struct rcu_head *head,
362 void (*func)(struct rcu_head *head));
363 void wait_rcu_gp(call_rcu_func_t crf);
365 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
366 #include <linux/rcutree.h>
367 #elif defined(CONFIG_TINY_RCU)
368 #include <linux/rcutiny.h>
370 #error "Unknown RCU implementation specified to kernel configuration"
374 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
375 * initialization and destruction of rcu_head on the stack. rcu_head structures
376 * allocated dynamically in the heap or defined statically don't need any
379 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
380 void init_rcu_head(struct rcu_head *head);
381 void destroy_rcu_head(struct rcu_head *head);
382 void init_rcu_head_on_stack(struct rcu_head *head);
383 void destroy_rcu_head_on_stack(struct rcu_head *head);
384 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
385 static inline void init_rcu_head(struct rcu_head *head)
389 static inline void destroy_rcu_head(struct rcu_head *head)
393 static inline void init_rcu_head_on_stack(struct rcu_head *head)
397 static inline void destroy_rcu_head_on_stack(struct rcu_head *head)
400 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
402 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
403 bool rcu_lockdep_current_cpu_online(void);
404 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
405 static inline bool rcu_lockdep_current_cpu_online(void)
409 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
411 #ifdef CONFIG_DEBUG_LOCK_ALLOC
413 static inline void rcu_lock_acquire(struct lockdep_map *map)
415 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
418 static inline void rcu_lock_release(struct lockdep_map *map)
420 lock_release(map, 1, _THIS_IP_);
423 extern struct lockdep_map rcu_lock_map;
424 extern struct lockdep_map rcu_bh_lock_map;
425 extern struct lockdep_map rcu_sched_lock_map;
426 extern struct lockdep_map rcu_callback_map;
427 int debug_lockdep_rcu_enabled(void);
430 * rcu_read_lock_held() - might we be in RCU read-side critical section?
432 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
433 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
434 * this assumes we are in an RCU read-side critical section unless it can
435 * prove otherwise. This is useful for debug checks in functions that
436 * require that they be called within an RCU read-side critical section.
438 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
439 * and while lockdep is disabled.
441 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
442 * occur in the same context, for example, it is illegal to invoke
443 * rcu_read_unlock() in process context if the matching rcu_read_lock()
444 * was invoked from within an irq handler.
446 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
447 * offline from an RCU perspective, so check for those as well.
449 static inline int rcu_read_lock_held(void)
451 if (!debug_lockdep_rcu_enabled())
453 if (!rcu_is_watching())
455 if (!rcu_lockdep_current_cpu_online())
457 return lock_is_held(&rcu_lock_map);
461 * rcu_read_lock_bh_held() is defined out of line to avoid #include-file
464 int rcu_read_lock_bh_held(void);
467 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
469 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
470 * RCU-sched read-side critical section. In absence of
471 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
472 * critical section unless it can prove otherwise. Note that disabling
473 * of preemption (including disabling irqs) counts as an RCU-sched
474 * read-side critical section. This is useful for debug checks in functions
475 * that required that they be called within an RCU-sched read-side
478 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
479 * and while lockdep is disabled.
481 * Note that if the CPU is in the idle loop from an RCU point of
482 * view (ie: that we are in the section between rcu_idle_enter() and
483 * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU
484 * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs
485 * that are in such a section, considering these as in extended quiescent
486 * state, so such a CPU is effectively never in an RCU read-side critical
487 * section regardless of what RCU primitives it invokes. This state of
488 * affairs is required --- we need to keep an RCU-free window in idle
489 * where the CPU may possibly enter into low power mode. This way we can
490 * notice an extended quiescent state to other CPUs that started a grace
491 * period. Otherwise we would delay any grace period as long as we run in
494 * Similarly, we avoid claiming an SRCU read lock held if the current
497 #ifdef CONFIG_PREEMPT_COUNT
498 static inline int rcu_read_lock_sched_held(void)
500 int lockdep_opinion = 0;
502 if (!debug_lockdep_rcu_enabled())
504 if (!rcu_is_watching())
506 if (!rcu_lockdep_current_cpu_online())
509 lockdep_opinion = lock_is_held(&rcu_sched_lock_map);
510 return lockdep_opinion || preempt_count() != 0 || irqs_disabled();
512 #else /* #ifdef CONFIG_PREEMPT_COUNT */
513 static inline int rcu_read_lock_sched_held(void)
517 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
519 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
521 # define rcu_lock_acquire(a) do { } while (0)
522 # define rcu_lock_release(a) do { } while (0)
524 static inline int rcu_read_lock_held(void)
529 static inline int rcu_read_lock_bh_held(void)
534 #ifdef CONFIG_PREEMPT_COUNT
535 static inline int rcu_read_lock_sched_held(void)
537 return preempt_count() != 0 || irqs_disabled();
539 #else /* #ifdef CONFIG_PREEMPT_COUNT */
540 static inline int rcu_read_lock_sched_held(void)
544 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
546 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
548 #ifdef CONFIG_PROVE_RCU
551 * rcu_lockdep_assert - emit lockdep splat if specified condition not met
552 * @c: condition to check
553 * @s: informative message
555 #define rcu_lockdep_assert(c, s) \
557 static bool __section(.data.unlikely) __warned; \
558 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \
560 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
564 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
565 static inline void rcu_preempt_sleep_check(void)
567 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
568 "Illegal context switch in RCU read-side critical section");
570 #else /* #ifdef CONFIG_PROVE_RCU */
571 static inline void rcu_preempt_sleep_check(void)
574 #endif /* #else #ifdef CONFIG_PROVE_RCU */
576 #define rcu_sleep_check() \
578 rcu_preempt_sleep_check(); \
579 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \
580 "Illegal context switch in RCU-bh read-side critical section"); \
581 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \
582 "Illegal context switch in RCU-sched read-side critical section"); \
585 #else /* #ifdef CONFIG_PROVE_RCU */
587 #define rcu_lockdep_assert(c, s) do { } while (0)
588 #define rcu_sleep_check() do { } while (0)
590 #endif /* #else #ifdef CONFIG_PROVE_RCU */
593 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
594 * and rcu_assign_pointer(). Some of these could be folded into their
595 * callers, but they are left separate in order to ease introduction of
596 * multiple flavors of pointers to match the multiple flavors of RCU
597 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
602 #define rcu_dereference_sparse(p, space) \
603 ((void)(((typeof(*p) space *)p) == p))
604 #else /* #ifdef __CHECKER__ */
605 #define rcu_dereference_sparse(p, space)
606 #endif /* #else #ifdef __CHECKER__ */
608 #define __rcu_access_pointer(p, space) \
610 typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \
611 rcu_dereference_sparse(p, space); \
612 ((typeof(*p) __force __kernel *)(_________p1)); \
614 #define __rcu_dereference_check(p, c, space) \
616 typeof(*p) *_________p1 = (typeof(*p) *__force)ACCESS_ONCE(p); \
617 rcu_lockdep_assert(c, "suspicious rcu_dereference_check() usage"); \
618 rcu_dereference_sparse(p, space); \
619 smp_read_barrier_depends(); /* Dependency order vs. p above. */ \
620 ((typeof(*p) __force __kernel *)(_________p1)); \
622 #define __rcu_dereference_protected(p, c, space) \
624 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected() usage"); \
625 rcu_dereference_sparse(p, space); \
626 ((typeof(*p) __force __kernel *)(p)); \
629 #define __rcu_access_index(p, space) \
631 typeof(p) _________p1 = ACCESS_ONCE(p); \
632 rcu_dereference_sparse(p, space); \
635 #define __rcu_dereference_index_check(p, c) \
637 typeof(p) _________p1 = ACCESS_ONCE(p); \
638 rcu_lockdep_assert(c, \
639 "suspicious rcu_dereference_index_check() usage"); \
640 smp_read_barrier_depends(); /* Dependency order vs. p above. */ \
645 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
646 * @v: The value to statically initialize with.
648 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
651 * rcu_assign_pointer() - assign to RCU-protected pointer
652 * @p: pointer to assign to
653 * @v: value to assign (publish)
655 * Assigns the specified value to the specified RCU-protected
656 * pointer, ensuring that any concurrent RCU readers will see
657 * any prior initialization.
659 * Inserts memory barriers on architectures that require them
660 * (which is most of them), and also prevents the compiler from
661 * reordering the code that initializes the structure after the pointer
662 * assignment. More importantly, this call documents which pointers
663 * will be dereferenced by RCU read-side code.
665 * In some special cases, you may use RCU_INIT_POINTER() instead
666 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
667 * to the fact that it does not constrain either the CPU or the compiler.
668 * That said, using RCU_INIT_POINTER() when you should have used
669 * rcu_assign_pointer() is a very bad thing that results in
670 * impossible-to-diagnose memory corruption. So please be careful.
671 * See the RCU_INIT_POINTER() comment header for details.
673 * Note that rcu_assign_pointer() evaluates each of its arguments only
674 * once, appearances notwithstanding. One of the "extra" evaluations
675 * is in typeof() and the other visible only to sparse (__CHECKER__),
676 * neither of which actually execute the argument. As with most cpp
677 * macros, this execute-arguments-only-once property is important, so
678 * please be careful when making changes to rcu_assign_pointer() and the
679 * other macros that it invokes.
681 #define rcu_assign_pointer(p, v) smp_store_release(&p, RCU_INITIALIZER(v))
684 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
685 * @p: The pointer to read
687 * Return the value of the specified RCU-protected pointer, but omit the
688 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
689 * when the value of this pointer is accessed, but the pointer is not
690 * dereferenced, for example, when testing an RCU-protected pointer against
691 * NULL. Although rcu_access_pointer() may also be used in cases where
692 * update-side locks prevent the value of the pointer from changing, you
693 * should instead use rcu_dereference_protected() for this use case.
695 * It is also permissible to use rcu_access_pointer() when read-side
696 * access to the pointer was removed at least one grace period ago, as
697 * is the case in the context of the RCU callback that is freeing up
698 * the data, or after a synchronize_rcu() returns. This can be useful
699 * when tearing down multi-linked structures after a grace period
702 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
705 * rcu_dereference_check() - rcu_dereference with debug checking
706 * @p: The pointer to read, prior to dereferencing
707 * @c: The conditions under which the dereference will take place
709 * Do an rcu_dereference(), but check that the conditions under which the
710 * dereference will take place are correct. Typically the conditions
711 * indicate the various locking conditions that should be held at that
712 * point. The check should return true if the conditions are satisfied.
713 * An implicit check for being in an RCU read-side critical section
714 * (rcu_read_lock()) is included.
718 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
720 * could be used to indicate to lockdep that foo->bar may only be dereferenced
721 * if either rcu_read_lock() is held, or that the lock required to replace
722 * the bar struct at foo->bar is held.
724 * Note that the list of conditions may also include indications of when a lock
725 * need not be held, for example during initialisation or destruction of the
728 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
729 * atomic_read(&foo->usage) == 0);
731 * Inserts memory barriers on architectures that require them
732 * (currently only the Alpha), prevents the compiler from refetching
733 * (and from merging fetches), and, more importantly, documents exactly
734 * which pointers are protected by RCU and checks that the pointer is
735 * annotated as __rcu.
737 #define rcu_dereference_check(p, c) \
738 __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu)
741 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
742 * @p: The pointer to read, prior to dereferencing
743 * @c: The conditions under which the dereference will take place
745 * This is the RCU-bh counterpart to rcu_dereference_check().
747 #define rcu_dereference_bh_check(p, c) \
748 __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu)
751 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
752 * @p: The pointer to read, prior to dereferencing
753 * @c: The conditions under which the dereference will take place
755 * This is the RCU-sched counterpart to rcu_dereference_check().
757 #define rcu_dereference_sched_check(p, c) \
758 __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \
761 #define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/
764 * The tracing infrastructure traces RCU (we want that), but unfortunately
765 * some of the RCU checks causes tracing to lock up the system.
767 * The tracing version of rcu_dereference_raw() must not call
768 * rcu_read_lock_held().
770 #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
773 * rcu_access_index() - fetch RCU index with no dereferencing
774 * @p: The index to read
776 * Return the value of the specified RCU-protected index, but omit the
777 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
778 * when the value of this index is accessed, but the index is not
779 * dereferenced, for example, when testing an RCU-protected index against
780 * -1. Although rcu_access_index() may also be used in cases where
781 * update-side locks prevent the value of the index from changing, you
782 * should instead use rcu_dereference_index_protected() for this use case.
784 #define rcu_access_index(p) __rcu_access_index((p), __rcu)
787 * rcu_dereference_index_check() - rcu_dereference for indices with debug checking
788 * @p: The pointer to read, prior to dereferencing
789 * @c: The conditions under which the dereference will take place
791 * Similar to rcu_dereference_check(), but omits the sparse checking.
792 * This allows rcu_dereference_index_check() to be used on integers,
793 * which can then be used as array indices. Attempting to use
794 * rcu_dereference_check() on an integer will give compiler warnings
795 * because the sparse address-space mechanism relies on dereferencing
796 * the RCU-protected pointer. Dereferencing integers is not something
797 * that even gcc will put up with.
799 * Note that this function does not implicitly check for RCU read-side
800 * critical sections. If this function gains lots of uses, it might
801 * make sense to provide versions for each flavor of RCU, but it does
802 * not make sense as of early 2010.
804 #define rcu_dereference_index_check(p, c) \
805 __rcu_dereference_index_check((p), (c))
808 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
809 * @p: The pointer to read, prior to dereferencing
810 * @c: The conditions under which the dereference will take place
812 * Return the value of the specified RCU-protected pointer, but omit
813 * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This
814 * is useful in cases where update-side locks prevent the value of the
815 * pointer from changing. Please note that this primitive does -not-
816 * prevent the compiler from repeating this reference or combining it
817 * with other references, so it should not be used without protection
818 * of appropriate locks.
820 * This function is only for update-side use. Using this function
821 * when protected only by rcu_read_lock() will result in infrequent
822 * but very ugly failures.
824 #define rcu_dereference_protected(p, c) \
825 __rcu_dereference_protected((p), (c), __rcu)
829 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
830 * @p: The pointer to read, prior to dereferencing
832 * This is a simple wrapper around rcu_dereference_check().
834 #define rcu_dereference(p) rcu_dereference_check(p, 0)
837 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
838 * @p: The pointer to read, prior to dereferencing
840 * Makes rcu_dereference_check() do the dirty work.
842 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
845 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
846 * @p: The pointer to read, prior to dereferencing
848 * Makes rcu_dereference_check() do the dirty work.
850 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
853 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
855 * When synchronize_rcu() is invoked on one CPU while other CPUs
856 * are within RCU read-side critical sections, then the
857 * synchronize_rcu() is guaranteed to block until after all the other
858 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
859 * on one CPU while other CPUs are within RCU read-side critical
860 * sections, invocation of the corresponding RCU callback is deferred
861 * until after the all the other CPUs exit their critical sections.
863 * Note, however, that RCU callbacks are permitted to run concurrently
864 * with new RCU read-side critical sections. One way that this can happen
865 * is via the following sequence of events: (1) CPU 0 enters an RCU
866 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
867 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
868 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
869 * callback is invoked. This is legal, because the RCU read-side critical
870 * section that was running concurrently with the call_rcu() (and which
871 * therefore might be referencing something that the corresponding RCU
872 * callback would free up) has completed before the corresponding
873 * RCU callback is invoked.
875 * RCU read-side critical sections may be nested. Any deferred actions
876 * will be deferred until the outermost RCU read-side critical section
879 * You can avoid reading and understanding the next paragraph by
880 * following this rule: don't put anything in an rcu_read_lock() RCU
881 * read-side critical section that would block in a !PREEMPT kernel.
882 * But if you want the full story, read on!
884 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU),
885 * it is illegal to block while in an RCU read-side critical section.
886 * In preemptible RCU implementations (TREE_PREEMPT_RCU) in CONFIG_PREEMPT
887 * kernel builds, RCU read-side critical sections may be preempted,
888 * but explicit blocking is illegal. Finally, in preemptible RCU
889 * implementations in real-time (with -rt patchset) kernel builds, RCU
890 * read-side critical sections may be preempted and they may also block, but
891 * only when acquiring spinlocks that are subject to priority inheritance.
893 static inline void rcu_read_lock(void)
897 rcu_lock_acquire(&rcu_lock_map);
898 rcu_lockdep_assert(rcu_is_watching(),
899 "rcu_read_lock() used illegally while idle");
903 * So where is rcu_write_lock()? It does not exist, as there is no
904 * way for writers to lock out RCU readers. This is a feature, not
905 * a bug -- this property is what provides RCU's performance benefits.
906 * Of course, writers must coordinate with each other. The normal
907 * spinlock primitives work well for this, but any other technique may be
908 * used as well. RCU does not care how the writers keep out of each
909 * others' way, as long as they do so.
913 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
915 * In most situations, rcu_read_unlock() is immune from deadlock.
916 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock()
917 * is responsible for deboosting, which it does via rt_mutex_unlock().
918 * Unfortunately, this function acquires the scheduler's runqueue and
919 * priority-inheritance spinlocks. This means that deadlock could result
920 * if the caller of rcu_read_unlock() already holds one of these locks or
921 * any lock that is ever acquired while holding them.
923 * That said, RCU readers are never priority boosted unless they were
924 * preempted. Therefore, one way to avoid deadlock is to make sure
925 * that preemption never happens within any RCU read-side critical
926 * section whose outermost rcu_read_unlock() is called with one of
927 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in
928 * a number of ways, for example, by invoking preempt_disable() before
929 * critical section's outermost rcu_read_lock().
931 * Given that the set of locks acquired by rt_mutex_unlock() might change
932 * at any time, a somewhat more future-proofed approach is to make sure
933 * that that preemption never happens within any RCU read-side critical
934 * section whose outermost rcu_read_unlock() is called with irqs disabled.
935 * This approach relies on the fact that rt_mutex_unlock() currently only
936 * acquires irq-disabled locks.
938 * The second of these two approaches is best in most situations,
939 * however, the first approach can also be useful, at least to those
940 * developers willing to keep abreast of the set of locks acquired by
943 * See rcu_read_lock() for more information.
945 static inline void rcu_read_unlock(void)
947 rcu_lockdep_assert(rcu_is_watching(),
948 "rcu_read_unlock() used illegally while idle");
949 rcu_lock_release(&rcu_lock_map);
955 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
957 * This is equivalent of rcu_read_lock(), but to be used when updates
958 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
959 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
960 * softirq handler to be a quiescent state, a process in RCU read-side
961 * critical section must be protected by disabling softirqs. Read-side
962 * critical sections in interrupt context can use just rcu_read_lock(),
963 * though this should at least be commented to avoid confusing people
966 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
967 * must occur in the same context, for example, it is illegal to invoke
968 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
969 * was invoked from some other task.
971 static inline void rcu_read_lock_bh(void)
975 rcu_lock_acquire(&rcu_bh_lock_map);
976 rcu_lockdep_assert(rcu_is_watching(),
977 "rcu_read_lock_bh() used illegally while idle");
981 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
983 * See rcu_read_lock_bh() for more information.
985 static inline void rcu_read_unlock_bh(void)
987 rcu_lockdep_assert(rcu_is_watching(),
988 "rcu_read_unlock_bh() used illegally while idle");
989 rcu_lock_release(&rcu_bh_lock_map);
995 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
997 * This is equivalent of rcu_read_lock(), but to be used when updates
998 * are being done using call_rcu_sched() or synchronize_rcu_sched().
999 * Read-side critical sections can also be introduced by anything that
1000 * disables preemption, including local_irq_disable() and friends.
1002 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
1003 * must occur in the same context, for example, it is illegal to invoke
1004 * rcu_read_unlock_sched() from process context if the matching
1005 * rcu_read_lock_sched() was invoked from an NMI handler.
1007 static inline void rcu_read_lock_sched(void)
1010 __acquire(RCU_SCHED);
1011 rcu_lock_acquire(&rcu_sched_lock_map);
1012 rcu_lockdep_assert(rcu_is_watching(),
1013 "rcu_read_lock_sched() used illegally while idle");
1016 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
1017 static inline notrace void rcu_read_lock_sched_notrace(void)
1019 preempt_disable_notrace();
1020 __acquire(RCU_SCHED);
1024 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
1026 * See rcu_read_lock_sched for more information.
1028 static inline void rcu_read_unlock_sched(void)
1030 rcu_lockdep_assert(rcu_is_watching(),
1031 "rcu_read_unlock_sched() used illegally while idle");
1032 rcu_lock_release(&rcu_sched_lock_map);
1033 __release(RCU_SCHED);
1037 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
1038 static inline notrace void rcu_read_unlock_sched_notrace(void)
1040 __release(RCU_SCHED);
1041 preempt_enable_notrace();
1045 * RCU_INIT_POINTER() - initialize an RCU protected pointer
1047 * Initialize an RCU-protected pointer in special cases where readers
1048 * do not need ordering constraints on the CPU or the compiler. These
1049 * special cases are:
1051 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
1052 * 2. The caller has taken whatever steps are required to prevent
1053 * RCU readers from concurrently accessing this pointer -or-
1054 * 3. The referenced data structure has already been exposed to
1055 * readers either at compile time or via rcu_assign_pointer() -and-
1056 * a. You have not made -any- reader-visible changes to
1057 * this structure since then -or-
1058 * b. It is OK for readers accessing this structure from its
1059 * new location to see the old state of the structure. (For
1060 * example, the changes were to statistical counters or to
1061 * other state where exact synchronization is not required.)
1063 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
1064 * result in impossible-to-diagnose memory corruption. As in the structures
1065 * will look OK in crash dumps, but any concurrent RCU readers might
1066 * see pre-initialized values of the referenced data structure. So
1067 * please be very careful how you use RCU_INIT_POINTER()!!!
1069 * If you are creating an RCU-protected linked structure that is accessed
1070 * by a single external-to-structure RCU-protected pointer, then you may
1071 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
1072 * pointers, but you must use rcu_assign_pointer() to initialize the
1073 * external-to-structure pointer -after- you have completely initialized
1074 * the reader-accessible portions of the linked structure.
1076 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
1077 * ordering guarantees for either the CPU or the compiler.
1079 #define RCU_INIT_POINTER(p, v) \
1081 p = RCU_INITIALIZER(v); \
1085 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
1087 * GCC-style initialization for an RCU-protected pointer in a structure field.
1089 #define RCU_POINTER_INITIALIZER(p, v) \
1090 .p = RCU_INITIALIZER(v)
1093 * Does the specified offset indicate that the corresponding rcu_head
1094 * structure can be handled by kfree_rcu()?
1096 #define __is_kfree_rcu_offset(offset) ((offset) < 4096)
1099 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
1101 #define __kfree_rcu(head, offset) \
1103 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
1104 kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \
1108 * kfree_rcu() - kfree an object after a grace period.
1109 * @ptr: pointer to kfree
1110 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
1112 * Many rcu callbacks functions just call kfree() on the base structure.
1113 * These functions are trivial, but their size adds up, and furthermore
1114 * when they are used in a kernel module, that module must invoke the
1115 * high-latency rcu_barrier() function at module-unload time.
1117 * The kfree_rcu() function handles this issue. Rather than encoding a
1118 * function address in the embedded rcu_head structure, kfree_rcu() instead
1119 * encodes the offset of the rcu_head structure within the base structure.
1120 * Because the functions are not allowed in the low-order 4096 bytes of
1121 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
1122 * If the offset is larger than 4095 bytes, a compile-time error will
1123 * be generated in __kfree_rcu(). If this error is triggered, you can
1124 * either fall back to use of call_rcu() or rearrange the structure to
1125 * position the rcu_head structure into the first 4096 bytes.
1127 * Note that the allowable offset might decrease in the future, for example,
1128 * to allow something like kmem_cache_free_rcu().
1130 * The BUILD_BUG_ON check must not involve any function calls, hence the
1131 * checks are done in macros here.
1133 #define kfree_rcu(ptr, rcu_head) \
1134 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
1136 #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL)
1137 static inline int rcu_needs_cpu(int cpu, unsigned long *delta_jiffies)
1139 *delta_jiffies = ULONG_MAX;
1142 #endif /* #if defined(CONFIG_TINY_RCU) || defined(CONFIG_RCU_NOCB_CPU_ALL) */
1144 #if defined(CONFIG_RCU_NOCB_CPU_ALL)
1145 static inline bool rcu_is_nocb_cpu(int cpu) { return true; }
1146 #elif defined(CONFIG_RCU_NOCB_CPU)
1147 bool rcu_is_nocb_cpu(int cpu);
1149 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
1153 /* Only for use by adaptive-ticks code. */
1154 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
1155 bool rcu_sys_is_idle(void);
1156 void rcu_sysidle_force_exit(void);
1157 #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1159 static inline bool rcu_sys_is_idle(void)
1164 static inline void rcu_sysidle_force_exit(void)
1168 #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1171 #endif /* __LINUX_RCUPDATE_H */