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/bug.h>
44 #include <linux/compiler.h>
45 #include <linux/ktime.h>
46 #include <linux/irqflags.h>
48 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
49 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
50 #define ulong2long(a) (*(long *)(&(a)))
52 /* Exported common interfaces */
54 #ifdef CONFIG_PREEMPT_RCU
55 void call_rcu(struct rcu_head *head, rcu_callback_t func);
56 #else /* #ifdef CONFIG_PREEMPT_RCU */
57 #define call_rcu call_rcu_sched
58 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
60 void call_rcu_bh(struct rcu_head *head, rcu_callback_t func);
61 void call_rcu_sched(struct rcu_head *head, rcu_callback_t func);
62 void synchronize_sched(void);
63 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func);
64 void synchronize_rcu_tasks(void);
65 void rcu_barrier_tasks(void);
67 #ifdef CONFIG_PREEMPT_RCU
69 void __rcu_read_lock(void);
70 void __rcu_read_unlock(void);
71 void rcu_read_unlock_special(struct task_struct *t);
72 void synchronize_rcu(void);
75 * Defined as a macro as it is a very low level header included from
76 * areas that don't even know about current. This gives the rcu_read_lock()
77 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
78 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
80 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
82 #else /* #ifdef CONFIG_PREEMPT_RCU */
84 static inline void __rcu_read_lock(void)
86 if (IS_ENABLED(CONFIG_PREEMPT_COUNT))
90 static inline void __rcu_read_unlock(void)
92 if (IS_ENABLED(CONFIG_PREEMPT_COUNT))
96 static inline void synchronize_rcu(void)
101 static inline int rcu_preempt_depth(void)
106 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
108 /* Internal to kernel */
110 void rcu_sched_qs(void);
111 void rcu_bh_qs(void);
112 void rcu_check_callbacks(int user);
113 void rcu_report_dead(unsigned int cpu);
114 void rcu_cpu_starting(unsigned int cpu);
116 #ifndef CONFIG_TINY_RCU
117 void rcu_end_inkernel_boot(void);
118 #else /* #ifndef CONFIG_TINY_RCU */
119 static inline void rcu_end_inkernel_boot(void) { }
120 #endif /* #ifndef CONFIG_TINY_RCU */
122 #ifdef CONFIG_RCU_STALL_COMMON
123 void rcu_sysrq_start(void);
124 void rcu_sysrq_end(void);
125 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
126 static inline void rcu_sysrq_start(void) { }
127 static inline void rcu_sysrq_end(void) { }
128 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
130 #ifdef CONFIG_NO_HZ_FULL
131 void rcu_user_enter(void);
132 void rcu_user_exit(void);
134 static inline void rcu_user_enter(void) { }
135 static inline void rcu_user_exit(void) { }
136 #endif /* CONFIG_NO_HZ_FULL */
138 #ifdef CONFIG_RCU_NOCB_CPU
139 void rcu_init_nohz(void);
140 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
141 static inline void rcu_init_nohz(void) { }
142 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
145 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
146 * @a: Code that RCU needs to pay attention to.
148 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
149 * in the inner idle loop, that is, between the rcu_idle_enter() and
150 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
151 * critical sections. However, things like powertop need tracepoints
152 * in the inner idle loop.
154 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
155 * will tell RCU that it needs to pay attention, invoke its argument
156 * (in this example, calling the do_something_with_RCU() function),
157 * and then tell RCU to go back to ignoring this CPU. It is permissible
158 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is
159 * on the order of a million or so, even on 32-bit systems). It is
160 * not legal to block within RCU_NONIDLE(), nor is it permissible to
161 * transfer control either into or out of RCU_NONIDLE()'s statement.
163 #define RCU_NONIDLE(a) \
165 rcu_irq_enter_irqson(); \
166 do { a; } while (0); \
167 rcu_irq_exit_irqson(); \
171 * Note a voluntary context switch for RCU-tasks benefit. This is a
172 * macro rather than an inline function to avoid #include hell.
174 #ifdef CONFIG_TASKS_RCU
175 #define TASKS_RCU(x) x
176 extern struct srcu_struct tasks_rcu_exit_srcu;
177 #define rcu_note_voluntary_context_switch_lite(t) \
179 if (READ_ONCE((t)->rcu_tasks_holdout)) \
180 WRITE_ONCE((t)->rcu_tasks_holdout, false); \
182 #define rcu_note_voluntary_context_switch(t) \
185 rcu_note_voluntary_context_switch_lite(t); \
187 #else /* #ifdef CONFIG_TASKS_RCU */
188 #define TASKS_RCU(x) do { } while (0)
189 #define rcu_note_voluntary_context_switch_lite(t) do { } while (0)
190 #define rcu_note_voluntary_context_switch(t) rcu_all_qs()
191 #endif /* #else #ifdef CONFIG_TASKS_RCU */
194 * cond_resched_rcu_qs - Report potential quiescent states to RCU
196 * This macro resembles cond_resched(), except that it is defined to
197 * report potential quiescent states to RCU-tasks even if the cond_resched()
198 * machinery were to be shut off, as some advocate for PREEMPT kernels.
200 #define cond_resched_rcu_qs() \
202 if (!cond_resched()) \
203 rcu_note_voluntary_context_switch(current); \
207 * Infrastructure to implement the synchronize_() primitives in
208 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
211 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
212 #include <linux/rcutree.h>
213 #elif defined(CONFIG_TINY_RCU)
214 #include <linux/rcutiny.h>
216 #error "Unknown RCU implementation specified to kernel configuration"
220 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
221 * initialization and destruction of rcu_head on the stack. rcu_head structures
222 * allocated dynamically in the heap or defined statically don't need any
225 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
226 void init_rcu_head(struct rcu_head *head);
227 void destroy_rcu_head(struct rcu_head *head);
228 void init_rcu_head_on_stack(struct rcu_head *head);
229 void destroy_rcu_head_on_stack(struct rcu_head *head);
230 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
231 static inline void init_rcu_head(struct rcu_head *head) { }
232 static inline void destroy_rcu_head(struct rcu_head *head) { }
233 static inline void init_rcu_head_on_stack(struct rcu_head *head) { }
234 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { }
235 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
237 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
238 bool rcu_lockdep_current_cpu_online(void);
239 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
240 static inline bool rcu_lockdep_current_cpu_online(void)
244 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
246 #ifdef CONFIG_DEBUG_LOCK_ALLOC
248 static inline void rcu_lock_acquire(struct lockdep_map *map)
250 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
253 static inline void rcu_lock_release(struct lockdep_map *map)
255 lock_release(map, 1, _THIS_IP_);
258 extern struct lockdep_map rcu_lock_map;
259 extern struct lockdep_map rcu_bh_lock_map;
260 extern struct lockdep_map rcu_sched_lock_map;
261 extern struct lockdep_map rcu_callback_map;
262 int debug_lockdep_rcu_enabled(void);
263 int rcu_read_lock_held(void);
264 int rcu_read_lock_bh_held(void);
265 int rcu_read_lock_sched_held(void);
267 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
269 # define rcu_lock_acquire(a) do { } while (0)
270 # define rcu_lock_release(a) do { } while (0)
272 static inline int rcu_read_lock_held(void)
277 static inline int rcu_read_lock_bh_held(void)
282 static inline int rcu_read_lock_sched_held(void)
284 return !preemptible();
286 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
288 #ifdef CONFIG_PROVE_RCU
291 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
292 * @c: condition to check
293 * @s: informative message
295 #define RCU_LOCKDEP_WARN(c, s) \
297 static bool __section(.data.unlikely) __warned; \
298 if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \
300 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
304 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
305 static inline void rcu_preempt_sleep_check(void)
307 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map),
308 "Illegal context switch in RCU read-side critical section");
310 #else /* #ifdef CONFIG_PROVE_RCU */
311 static inline void rcu_preempt_sleep_check(void) { }
312 #endif /* #else #ifdef CONFIG_PROVE_RCU */
314 #define rcu_sleep_check() \
316 rcu_preempt_sleep_check(); \
317 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \
318 "Illegal context switch in RCU-bh read-side critical section"); \
319 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \
320 "Illegal context switch in RCU-sched read-side critical section"); \
323 #else /* #ifdef CONFIG_PROVE_RCU */
325 #define RCU_LOCKDEP_WARN(c, s) do { } while (0)
326 #define rcu_sleep_check() do { } while (0)
328 #endif /* #else #ifdef CONFIG_PROVE_RCU */
331 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
332 * and rcu_assign_pointer(). Some of these could be folded into their
333 * callers, but they are left separate in order to ease introduction of
334 * multiple flavors of pointers to match the multiple flavors of RCU
335 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
340 #define rcu_dereference_sparse(p, space) \
341 ((void)(((typeof(*p) space *)p) == p))
342 #else /* #ifdef __CHECKER__ */
343 #define rcu_dereference_sparse(p, space)
344 #endif /* #else #ifdef __CHECKER__ */
346 #define __rcu_access_pointer(p, space) \
348 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \
349 rcu_dereference_sparse(p, space); \
350 ((typeof(*p) __force __kernel *)(_________p1)); \
352 #define __rcu_dereference_check(p, c, space) \
354 /* Dependency order vs. p above. */ \
355 typeof(*p) *________p1 = (typeof(*p) *__force)lockless_dereference(p); \
356 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
357 rcu_dereference_sparse(p, space); \
358 ((typeof(*p) __force __kernel *)(________p1)); \
360 #define __rcu_dereference_protected(p, c, space) \
362 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
363 rcu_dereference_sparse(p, space); \
364 ((typeof(*p) __force __kernel *)(p)); \
366 #define rcu_dereference_raw(p) \
368 /* Dependency order vs. p above. */ \
369 typeof(p) ________p1 = lockless_dereference(p); \
370 ((typeof(*p) __force __kernel *)(________p1)); \
374 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
375 * @v: The value to statically initialize with.
377 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
380 * rcu_assign_pointer() - assign to RCU-protected pointer
381 * @p: pointer to assign to
382 * @v: value to assign (publish)
384 * Assigns the specified value to the specified RCU-protected
385 * pointer, ensuring that any concurrent RCU readers will see
386 * any prior initialization.
388 * Inserts memory barriers on architectures that require them
389 * (which is most of them), and also prevents the compiler from
390 * reordering the code that initializes the structure after the pointer
391 * assignment. More importantly, this call documents which pointers
392 * will be dereferenced by RCU read-side code.
394 * In some special cases, you may use RCU_INIT_POINTER() instead
395 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
396 * to the fact that it does not constrain either the CPU or the compiler.
397 * That said, using RCU_INIT_POINTER() when you should have used
398 * rcu_assign_pointer() is a very bad thing that results in
399 * impossible-to-diagnose memory corruption. So please be careful.
400 * See the RCU_INIT_POINTER() comment header for details.
402 * Note that rcu_assign_pointer() evaluates each of its arguments only
403 * once, appearances notwithstanding. One of the "extra" evaluations
404 * is in typeof() and the other visible only to sparse (__CHECKER__),
405 * neither of which actually execute the argument. As with most cpp
406 * macros, this execute-arguments-only-once property is important, so
407 * please be careful when making changes to rcu_assign_pointer() and the
408 * other macros that it invokes.
410 #define rcu_assign_pointer(p, v) \
412 uintptr_t _r_a_p__v = (uintptr_t)(v); \
414 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \
415 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \
417 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
422 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
423 * @p: The pointer to read
425 * Return the value of the specified RCU-protected pointer, but omit the
426 * smp_read_barrier_depends() and keep the READ_ONCE(). This is useful
427 * when the value of this pointer is accessed, but the pointer is not
428 * dereferenced, for example, when testing an RCU-protected pointer against
429 * NULL. Although rcu_access_pointer() may also be used in cases where
430 * update-side locks prevent the value of the pointer from changing, you
431 * should instead use rcu_dereference_protected() for this use case.
433 * It is also permissible to use rcu_access_pointer() when read-side
434 * access to the pointer was removed at least one grace period ago, as
435 * is the case in the context of the RCU callback that is freeing up
436 * the data, or after a synchronize_rcu() returns. This can be useful
437 * when tearing down multi-linked structures after a grace period
440 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
443 * rcu_dereference_check() - rcu_dereference with debug checking
444 * @p: The pointer to read, prior to dereferencing
445 * @c: The conditions under which the dereference will take place
447 * Do an rcu_dereference(), but check that the conditions under which the
448 * dereference will take place are correct. Typically the conditions
449 * indicate the various locking conditions that should be held at that
450 * point. The check should return true if the conditions are satisfied.
451 * An implicit check for being in an RCU read-side critical section
452 * (rcu_read_lock()) is included.
456 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
458 * could be used to indicate to lockdep that foo->bar may only be dereferenced
459 * if either rcu_read_lock() is held, or that the lock required to replace
460 * the bar struct at foo->bar is held.
462 * Note that the list of conditions may also include indications of when a lock
463 * need not be held, for example during initialisation or destruction of the
466 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
467 * atomic_read(&foo->usage) == 0);
469 * Inserts memory barriers on architectures that require them
470 * (currently only the Alpha), prevents the compiler from refetching
471 * (and from merging fetches), and, more importantly, documents exactly
472 * which pointers are protected by RCU and checks that the pointer is
473 * annotated as __rcu.
475 #define rcu_dereference_check(p, c) \
476 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu)
479 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
480 * @p: The pointer to read, prior to dereferencing
481 * @c: The conditions under which the dereference will take place
483 * This is the RCU-bh counterpart to rcu_dereference_check().
485 #define rcu_dereference_bh_check(p, c) \
486 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu)
489 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
490 * @p: The pointer to read, prior to dereferencing
491 * @c: The conditions under which the dereference will take place
493 * This is the RCU-sched counterpart to rcu_dereference_check().
495 #define rcu_dereference_sched_check(p, c) \
496 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \
500 * The tracing infrastructure traces RCU (we want that), but unfortunately
501 * some of the RCU checks causes tracing to lock up the system.
503 * The no-tracing version of rcu_dereference_raw() must not call
504 * rcu_read_lock_held().
506 #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
509 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
510 * @p: The pointer to read, prior to dereferencing
511 * @c: The conditions under which the dereference will take place
513 * Return the value of the specified RCU-protected pointer, but omit
514 * both the smp_read_barrier_depends() and the READ_ONCE(). This
515 * is useful in cases where update-side locks prevent the value of the
516 * pointer from changing. Please note that this primitive does -not-
517 * prevent the compiler from repeating this reference or combining it
518 * with other references, so it should not be used without protection
519 * of appropriate locks.
521 * This function is only for update-side use. Using this function
522 * when protected only by rcu_read_lock() will result in infrequent
523 * but very ugly failures.
525 #define rcu_dereference_protected(p, c) \
526 __rcu_dereference_protected((p), (c), __rcu)
530 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
531 * @p: The pointer to read, prior to dereferencing
533 * This is a simple wrapper around rcu_dereference_check().
535 #define rcu_dereference(p) rcu_dereference_check(p, 0)
538 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
539 * @p: The pointer to read, prior to dereferencing
541 * Makes rcu_dereference_check() do the dirty work.
543 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
546 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
547 * @p: The pointer to read, prior to dereferencing
549 * Makes rcu_dereference_check() do the dirty work.
551 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
554 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
555 * @p: The pointer to hand off
557 * This is simply an identity function, but it documents where a pointer
558 * is handed off from RCU to some other synchronization mechanism, for
559 * example, reference counting or locking. In C11, it would map to
560 * kill_dependency(). It could be used as follows:
563 * p = rcu_dereference(gp);
564 * long_lived = is_long_lived(p);
566 * if (!atomic_inc_not_zero(p->refcnt))
567 * long_lived = false;
569 * p = rcu_pointer_handoff(p);
573 #define rcu_pointer_handoff(p) (p)
576 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
578 * When synchronize_rcu() is invoked on one CPU while other CPUs
579 * are within RCU read-side critical sections, then the
580 * synchronize_rcu() is guaranteed to block until after all the other
581 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
582 * on one CPU while other CPUs are within RCU read-side critical
583 * sections, invocation of the corresponding RCU callback is deferred
584 * until after the all the other CPUs exit their critical sections.
586 * Note, however, that RCU callbacks are permitted to run concurrently
587 * with new RCU read-side critical sections. One way that this can happen
588 * is via the following sequence of events: (1) CPU 0 enters an RCU
589 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
590 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
591 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
592 * callback is invoked. This is legal, because the RCU read-side critical
593 * section that was running concurrently with the call_rcu() (and which
594 * therefore might be referencing something that the corresponding RCU
595 * callback would free up) has completed before the corresponding
596 * RCU callback is invoked.
598 * RCU read-side critical sections may be nested. Any deferred actions
599 * will be deferred until the outermost RCU read-side critical section
602 * You can avoid reading and understanding the next paragraph by
603 * following this rule: don't put anything in an rcu_read_lock() RCU
604 * read-side critical section that would block in a !PREEMPT kernel.
605 * But if you want the full story, read on!
607 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU),
608 * it is illegal to block while in an RCU read-side critical section.
609 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT
610 * kernel builds, RCU read-side critical sections may be preempted,
611 * but explicit blocking is illegal. Finally, in preemptible RCU
612 * implementations in real-time (with -rt patchset) kernel builds, RCU
613 * read-side critical sections may be preempted and they may also block, but
614 * only when acquiring spinlocks that are subject to priority inheritance.
616 static inline void rcu_read_lock(void)
620 rcu_lock_acquire(&rcu_lock_map);
621 RCU_LOCKDEP_WARN(!rcu_is_watching(),
622 "rcu_read_lock() used illegally while idle");
626 * So where is rcu_write_lock()? It does not exist, as there is no
627 * way for writers to lock out RCU readers. This is a feature, not
628 * a bug -- this property is what provides RCU's performance benefits.
629 * Of course, writers must coordinate with each other. The normal
630 * spinlock primitives work well for this, but any other technique may be
631 * used as well. RCU does not care how the writers keep out of each
632 * others' way, as long as they do so.
636 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
638 * In most situations, rcu_read_unlock() is immune from deadlock.
639 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock()
640 * is responsible for deboosting, which it does via rt_mutex_unlock().
641 * Unfortunately, this function acquires the scheduler's runqueue and
642 * priority-inheritance spinlocks. This means that deadlock could result
643 * if the caller of rcu_read_unlock() already holds one of these locks or
644 * any lock that is ever acquired while holding them; or any lock which
645 * can be taken from interrupt context because rcu_boost()->rt_mutex_lock()
646 * does not disable irqs while taking ->wait_lock.
648 * That said, RCU readers are never priority boosted unless they were
649 * preempted. Therefore, one way to avoid deadlock is to make sure
650 * that preemption never happens within any RCU read-side critical
651 * section whose outermost rcu_read_unlock() is called with one of
652 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in
653 * a number of ways, for example, by invoking preempt_disable() before
654 * critical section's outermost rcu_read_lock().
656 * Given that the set of locks acquired by rt_mutex_unlock() might change
657 * at any time, a somewhat more future-proofed approach is to make sure
658 * that that preemption never happens within any RCU read-side critical
659 * section whose outermost rcu_read_unlock() is called with irqs disabled.
660 * This approach relies on the fact that rt_mutex_unlock() currently only
661 * acquires irq-disabled locks.
663 * The second of these two approaches is best in most situations,
664 * however, the first approach can also be useful, at least to those
665 * developers willing to keep abreast of the set of locks acquired by
668 * See rcu_read_lock() for more information.
670 static inline void rcu_read_unlock(void)
672 RCU_LOCKDEP_WARN(!rcu_is_watching(),
673 "rcu_read_unlock() used illegally while idle");
676 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
680 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
682 * This is equivalent of rcu_read_lock(), but to be used when updates
683 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
684 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
685 * softirq handler to be a quiescent state, a process in RCU read-side
686 * critical section must be protected by disabling softirqs. Read-side
687 * critical sections in interrupt context can use just rcu_read_lock(),
688 * though this should at least be commented to avoid confusing people
691 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
692 * must occur in the same context, for example, it is illegal to invoke
693 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
694 * was invoked from some other task.
696 static inline void rcu_read_lock_bh(void)
700 rcu_lock_acquire(&rcu_bh_lock_map);
701 RCU_LOCKDEP_WARN(!rcu_is_watching(),
702 "rcu_read_lock_bh() used illegally while idle");
706 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
708 * See rcu_read_lock_bh() for more information.
710 static inline void rcu_read_unlock_bh(void)
712 RCU_LOCKDEP_WARN(!rcu_is_watching(),
713 "rcu_read_unlock_bh() used illegally while idle");
714 rcu_lock_release(&rcu_bh_lock_map);
720 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
722 * This is equivalent of rcu_read_lock(), but to be used when updates
723 * are being done using call_rcu_sched() or synchronize_rcu_sched().
724 * Read-side critical sections can also be introduced by anything that
725 * disables preemption, including local_irq_disable() and friends.
727 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
728 * must occur in the same context, for example, it is illegal to invoke
729 * rcu_read_unlock_sched() from process context if the matching
730 * rcu_read_lock_sched() was invoked from an NMI handler.
732 static inline void rcu_read_lock_sched(void)
735 __acquire(RCU_SCHED);
736 rcu_lock_acquire(&rcu_sched_lock_map);
737 RCU_LOCKDEP_WARN(!rcu_is_watching(),
738 "rcu_read_lock_sched() used illegally while idle");
741 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
742 static inline notrace void rcu_read_lock_sched_notrace(void)
744 preempt_disable_notrace();
745 __acquire(RCU_SCHED);
749 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
751 * See rcu_read_lock_sched for more information.
753 static inline void rcu_read_unlock_sched(void)
755 RCU_LOCKDEP_WARN(!rcu_is_watching(),
756 "rcu_read_unlock_sched() used illegally while idle");
757 rcu_lock_release(&rcu_sched_lock_map);
758 __release(RCU_SCHED);
762 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
763 static inline notrace void rcu_read_unlock_sched_notrace(void)
765 __release(RCU_SCHED);
766 preempt_enable_notrace();
770 * RCU_INIT_POINTER() - initialize an RCU protected pointer
772 * Initialize an RCU-protected pointer in special cases where readers
773 * do not need ordering constraints on the CPU or the compiler. These
776 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
777 * 2. The caller has taken whatever steps are required to prevent
778 * RCU readers from concurrently accessing this pointer -or-
779 * 3. The referenced data structure has already been exposed to
780 * readers either at compile time or via rcu_assign_pointer() -and-
781 * a. You have not made -any- reader-visible changes to
782 * this structure since then -or-
783 * b. It is OK for readers accessing this structure from its
784 * new location to see the old state of the structure. (For
785 * example, the changes were to statistical counters or to
786 * other state where exact synchronization is not required.)
788 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
789 * result in impossible-to-diagnose memory corruption. As in the structures
790 * will look OK in crash dumps, but any concurrent RCU readers might
791 * see pre-initialized values of the referenced data structure. So
792 * please be very careful how you use RCU_INIT_POINTER()!!!
794 * If you are creating an RCU-protected linked structure that is accessed
795 * by a single external-to-structure RCU-protected pointer, then you may
796 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
797 * pointers, but you must use rcu_assign_pointer() to initialize the
798 * external-to-structure pointer -after- you have completely initialized
799 * the reader-accessible portions of the linked structure.
801 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
802 * ordering guarantees for either the CPU or the compiler.
804 #define RCU_INIT_POINTER(p, v) \
806 rcu_dereference_sparse(p, __rcu); \
807 WRITE_ONCE(p, RCU_INITIALIZER(v)); \
811 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
813 * GCC-style initialization for an RCU-protected pointer in a structure field.
815 #define RCU_POINTER_INITIALIZER(p, v) \
816 .p = RCU_INITIALIZER(v)
819 * Does the specified offset indicate that the corresponding rcu_head
820 * structure can be handled by kfree_rcu()?
822 #define __is_kfree_rcu_offset(offset) ((offset) < 4096)
825 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
827 #define __kfree_rcu(head, offset) \
829 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
830 kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \
834 * kfree_rcu() - kfree an object after a grace period.
835 * @ptr: pointer to kfree
836 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
838 * Many rcu callbacks functions just call kfree() on the base structure.
839 * These functions are trivial, but their size adds up, and furthermore
840 * when they are used in a kernel module, that module must invoke the
841 * high-latency rcu_barrier() function at module-unload time.
843 * The kfree_rcu() function handles this issue. Rather than encoding a
844 * function address in the embedded rcu_head structure, kfree_rcu() instead
845 * encodes the offset of the rcu_head structure within the base structure.
846 * Because the functions are not allowed in the low-order 4096 bytes of
847 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
848 * If the offset is larger than 4095 bytes, a compile-time error will
849 * be generated in __kfree_rcu(). If this error is triggered, you can
850 * either fall back to use of call_rcu() or rearrange the structure to
851 * position the rcu_head structure into the first 4096 bytes.
853 * Note that the allowable offset might decrease in the future, for example,
854 * to allow something like kmem_cache_free_rcu().
856 * The BUILD_BUG_ON check must not involve any function calls, hence the
857 * checks are done in macros here.
859 #define kfree_rcu(ptr, rcu_head) \
860 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
862 #ifdef CONFIG_TINY_RCU
863 static inline int rcu_needs_cpu(u64 basemono, u64 *nextevt)
865 *nextevt = KTIME_MAX;
868 #endif /* #ifdef CONFIG_TINY_RCU */
870 #if defined(CONFIG_RCU_NOCB_CPU_ALL)
871 static inline bool rcu_is_nocb_cpu(int cpu) { return true; }
872 #elif defined(CONFIG_RCU_NOCB_CPU)
873 bool rcu_is_nocb_cpu(int cpu);
875 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
879 /* Only for use by adaptive-ticks code. */
880 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
881 bool rcu_sys_is_idle(void);
882 void rcu_sysidle_force_exit(void);
883 #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
885 static inline bool rcu_sys_is_idle(void)
890 static inline void rcu_sysidle_force_exit(void) { }
892 #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
896 * Dump the ftrace buffer, but only one time per callsite per boot.
898 #define rcu_ftrace_dump(oops_dump_mode) \
900 static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
902 if (!atomic_read(&___rfd_beenhere) && \
903 !atomic_xchg(&___rfd_beenhere, 1)) \
904 ftrace_dump(oops_dump_mode); \
908 * Place this after a lock-acquisition primitive to guarantee that
909 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies
910 * if the UNLOCK and LOCK are executed by the same CPU or if the
911 * UNLOCK and LOCK operate on the same lock variable.
913 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
914 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */
915 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
916 #define smp_mb__after_unlock_lock() do { } while (0)
917 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
920 #endif /* __LINUX_RCUPDATE_H */