1 /* SPDX-License-Identifier: GPL-2.0+ */
3 * Read-Copy Update mechanism for mutual exclusion
5 * Copyright IBM Corporation, 2001
7 * Author: Dipankar Sarma <dipankar@in.ibm.com>
9 * Based on the original work by Paul McKenney <paulmck@vnet.ibm.com>
10 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
12 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
13 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
15 * For detailed explanation of Read-Copy Update mechanism see -
16 * http://lse.sourceforge.net/locking/rcupdate.html
20 #ifndef __LINUX_RCUPDATE_H
21 #define __LINUX_RCUPDATE_H
23 #include <linux/types.h>
24 #include <linux/compiler.h>
25 #include <linux/atomic.h>
26 #include <linux/irqflags.h>
27 #include <linux/preempt.h>
28 #include <linux/bottom_half.h>
29 #include <linux/lockdep.h>
30 #include <asm/processor.h>
31 #include <linux/cpumask.h>
33 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
34 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
35 #define ulong2long(a) (*(long *)(&(a)))
36 #define USHORT_CMP_GE(a, b) (USHRT_MAX / 2 >= (unsigned short)((a) - (b)))
37 #define USHORT_CMP_LT(a, b) (USHRT_MAX / 2 < (unsigned short)((a) - (b)))
39 /* Exported common interfaces */
40 void call_rcu(struct rcu_head *head, rcu_callback_t func);
41 void rcu_barrier_tasks(void);
42 void rcu_barrier_tasks_rude(void);
43 void synchronize_rcu(void);
45 #ifdef CONFIG_PREEMPT_RCU
47 void __rcu_read_lock(void);
48 void __rcu_read_unlock(void);
51 * Defined as a macro as it is a very low level header included from
52 * areas that don't even know about current. This gives the rcu_read_lock()
53 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
54 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
56 #define rcu_preempt_depth() READ_ONCE(current->rcu_read_lock_nesting)
58 #else /* #ifdef CONFIG_PREEMPT_RCU */
60 #ifdef CONFIG_TINY_RCU
61 #define rcu_read_unlock_strict() do { } while (0)
63 void rcu_read_unlock_strict(void);
66 static inline void __rcu_read_lock(void)
71 static inline void __rcu_read_unlock(void)
74 if (IS_ENABLED(CONFIG_RCU_STRICT_GRACE_PERIOD))
75 rcu_read_unlock_strict();
78 static inline int rcu_preempt_depth(void)
83 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
85 /* Internal to kernel */
87 extern int rcu_scheduler_active __read_mostly;
88 void rcu_sched_clock_irq(int user);
89 void rcu_report_dead(unsigned int cpu);
90 void rcutree_migrate_callbacks(int cpu);
92 #ifdef CONFIG_TASKS_RCU_GENERIC
93 void rcu_init_tasks_generic(void);
95 static inline void rcu_init_tasks_generic(void) { }
98 #ifdef CONFIG_RCU_STALL_COMMON
99 void rcu_sysrq_start(void);
100 void rcu_sysrq_end(void);
101 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
102 static inline void rcu_sysrq_start(void) { }
103 static inline void rcu_sysrq_end(void) { }
104 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
106 #ifdef CONFIG_NO_HZ_FULL
107 void rcu_user_enter(void);
108 void rcu_user_exit(void);
110 static inline void rcu_user_enter(void) { }
111 static inline void rcu_user_exit(void) { }
112 #endif /* CONFIG_NO_HZ_FULL */
114 #ifdef CONFIG_RCU_NOCB_CPU
115 void rcu_init_nohz(void);
116 int rcu_nocb_cpu_offload(int cpu);
117 int rcu_nocb_cpu_deoffload(int cpu);
118 void rcu_nocb_flush_deferred_wakeup(void);
119 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
120 static inline void rcu_init_nohz(void) { }
121 static inline int rcu_nocb_cpu_offload(int cpu) { return -EINVAL; }
122 static inline int rcu_nocb_cpu_deoffload(int cpu) { return 0; }
123 static inline void rcu_nocb_flush_deferred_wakeup(void) { }
124 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
127 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
128 * @a: Code that RCU needs to pay attention to.
130 * RCU read-side critical sections are forbidden in the inner idle loop,
131 * that is, between the rcu_idle_enter() and the rcu_idle_exit() -- RCU
132 * will happily ignore any such read-side critical sections. However,
133 * things like powertop need tracepoints in the inner idle loop.
135 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
136 * will tell RCU that it needs to pay attention, invoke its argument
137 * (in this example, calling the do_something_with_RCU() function),
138 * and then tell RCU to go back to ignoring this CPU. It is permissible
139 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is
140 * on the order of a million or so, even on 32-bit systems). It is
141 * not legal to block within RCU_NONIDLE(), nor is it permissible to
142 * transfer control either into or out of RCU_NONIDLE()'s statement.
144 #define RCU_NONIDLE(a) \
146 rcu_irq_enter_irqson(); \
147 do { a; } while (0); \
148 rcu_irq_exit_irqson(); \
152 * Note a quasi-voluntary context switch for RCU-tasks's benefit.
153 * This is a macro rather than an inline function to avoid #include hell.
155 #ifdef CONFIG_TASKS_RCU_GENERIC
157 # ifdef CONFIG_TASKS_RCU
158 # define rcu_tasks_classic_qs(t, preempt) \
160 if (!(preempt) && READ_ONCE((t)->rcu_tasks_holdout)) \
161 WRITE_ONCE((t)->rcu_tasks_holdout, false); \
163 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func);
164 void synchronize_rcu_tasks(void);
166 # define rcu_tasks_classic_qs(t, preempt) do { } while (0)
167 # define call_rcu_tasks call_rcu
168 # define synchronize_rcu_tasks synchronize_rcu
171 # ifdef CONFIG_TASKS_TRACE_RCU
172 # define rcu_tasks_trace_qs(t) \
174 if (!likely(READ_ONCE((t)->trc_reader_checked)) && \
175 !unlikely(READ_ONCE((t)->trc_reader_nesting))) { \
176 smp_store_release(&(t)->trc_reader_checked, true); \
177 smp_mb(); /* Readers partitioned by store. */ \
181 # define rcu_tasks_trace_qs(t) do { } while (0)
184 #define rcu_tasks_qs(t, preempt) \
186 rcu_tasks_classic_qs((t), (preempt)); \
187 rcu_tasks_trace_qs((t)); \
190 # ifdef CONFIG_TASKS_RUDE_RCU
191 void call_rcu_tasks_rude(struct rcu_head *head, rcu_callback_t func);
192 void synchronize_rcu_tasks_rude(void);
195 #define rcu_note_voluntary_context_switch(t) rcu_tasks_qs(t, false)
196 void exit_tasks_rcu_start(void);
197 void exit_tasks_rcu_finish(void);
198 #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */
199 #define rcu_tasks_qs(t, preempt) do { } while (0)
200 #define rcu_note_voluntary_context_switch(t) do { } while (0)
201 #define call_rcu_tasks call_rcu
202 #define synchronize_rcu_tasks synchronize_rcu
203 static inline void exit_tasks_rcu_start(void) { }
204 static inline void exit_tasks_rcu_finish(void) { }
205 #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */
208 * cond_resched_tasks_rcu_qs - Report potential quiescent states to RCU
210 * This macro resembles cond_resched(), except that it is defined to
211 * report potential quiescent states to RCU-tasks even if the cond_resched()
212 * machinery were to be shut off, as some advocate for PREEMPTION kernels.
214 #define cond_resched_tasks_rcu_qs() \
216 rcu_tasks_qs(current, false); \
221 * Infrastructure to implement the synchronize_() primitives in
222 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
225 #if defined(CONFIG_TREE_RCU)
226 #include <linux/rcutree.h>
227 #elif defined(CONFIG_TINY_RCU)
228 #include <linux/rcutiny.h>
230 #error "Unknown RCU implementation specified to kernel configuration"
234 * The init_rcu_head_on_stack() and destroy_rcu_head_on_stack() calls
235 * are needed for dynamic initialization and destruction of rcu_head
236 * on the stack, and init_rcu_head()/destroy_rcu_head() are needed for
237 * dynamic initialization and destruction of statically allocated rcu_head
238 * structures. However, rcu_head structures allocated dynamically in the
239 * heap don't need any initialization.
241 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
242 void init_rcu_head(struct rcu_head *head);
243 void destroy_rcu_head(struct rcu_head *head);
244 void init_rcu_head_on_stack(struct rcu_head *head);
245 void destroy_rcu_head_on_stack(struct rcu_head *head);
246 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
247 static inline void init_rcu_head(struct rcu_head *head) { }
248 static inline void destroy_rcu_head(struct rcu_head *head) { }
249 static inline void init_rcu_head_on_stack(struct rcu_head *head) { }
250 static inline void destroy_rcu_head_on_stack(struct rcu_head *head) { }
251 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
253 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
254 bool rcu_lockdep_current_cpu_online(void);
255 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
256 static inline bool rcu_lockdep_current_cpu_online(void) { return true; }
257 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
259 extern struct lockdep_map rcu_lock_map;
260 extern struct lockdep_map rcu_bh_lock_map;
261 extern struct lockdep_map rcu_sched_lock_map;
262 extern struct lockdep_map rcu_callback_map;
264 #ifdef CONFIG_DEBUG_LOCK_ALLOC
266 static inline void rcu_lock_acquire(struct lockdep_map *map)
268 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
271 static inline void rcu_lock_release(struct lockdep_map *map)
273 lock_release(map, _THIS_IP_);
276 int debug_lockdep_rcu_enabled(void);
277 int rcu_read_lock_held(void);
278 int rcu_read_lock_bh_held(void);
279 int rcu_read_lock_sched_held(void);
280 int rcu_read_lock_any_held(void);
282 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
284 # define rcu_lock_acquire(a) do { } while (0)
285 # define rcu_lock_release(a) do { } while (0)
287 static inline int rcu_read_lock_held(void)
292 static inline int rcu_read_lock_bh_held(void)
297 static inline int rcu_read_lock_sched_held(void)
299 return !preemptible();
302 static inline int rcu_read_lock_any_held(void)
304 return !preemptible();
307 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
309 #ifdef CONFIG_PROVE_RCU
312 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
313 * @c: condition to check
314 * @s: informative message
316 #define RCU_LOCKDEP_WARN(c, s) \
318 static bool __section(".data.unlikely") __warned; \
319 if ((c) && debug_lockdep_rcu_enabled() && !__warned) { \
321 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
325 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
326 static inline void rcu_preempt_sleep_check(void)
328 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map),
329 "Illegal context switch in RCU read-side critical section");
331 #else /* #ifdef CONFIG_PROVE_RCU */
332 static inline void rcu_preempt_sleep_check(void) { }
333 #endif /* #else #ifdef CONFIG_PROVE_RCU */
335 #define rcu_sleep_check() \
337 rcu_preempt_sleep_check(); \
338 if (!IS_ENABLED(CONFIG_PREEMPT_RT)) \
339 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \
340 "Illegal context switch in RCU-bh read-side critical section"); \
341 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \
342 "Illegal context switch in RCU-sched read-side critical section"); \
345 #else /* #ifdef CONFIG_PROVE_RCU */
347 #define RCU_LOCKDEP_WARN(c, s) do { } while (0 && (c))
348 #define rcu_sleep_check() do { } while (0)
350 #endif /* #else #ifdef CONFIG_PROVE_RCU */
353 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
354 * and rcu_assign_pointer(). Some of these could be folded into their
355 * callers, but they are left separate in order to ease introduction of
356 * multiple pointers markings to match different RCU implementations
357 * (e.g., __srcu), should this make sense in the future.
361 #define rcu_check_sparse(p, space) \
362 ((void)(((typeof(*p) space *)p) == p))
363 #else /* #ifdef __CHECKER__ */
364 #define rcu_check_sparse(p, space)
365 #endif /* #else #ifdef __CHECKER__ */
368 * unrcu_pointer - mark a pointer as not being RCU protected
369 * @p: pointer needing to lose its __rcu property
371 * Converts @p from an __rcu pointer to a __kernel pointer.
372 * This allows an __rcu pointer to be used with xchg() and friends.
374 #define unrcu_pointer(p) \
376 typeof(*p) *_________p1 = (typeof(*p) *__force)(p); \
377 rcu_check_sparse(p, __rcu); \
378 ((typeof(*p) __force __kernel *)(_________p1)); \
381 #define __rcu_access_pointer(p, space) \
383 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \
384 rcu_check_sparse(p, space); \
385 ((typeof(*p) __force __kernel *)(_________p1)); \
387 #define __rcu_dereference_check(p, c, space) \
389 /* Dependency order vs. p above. */ \
390 typeof(*p) *________p1 = (typeof(*p) *__force)READ_ONCE(p); \
391 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
392 rcu_check_sparse(p, space); \
393 ((typeof(*p) __force __kernel *)(________p1)); \
395 #define __rcu_dereference_protected(p, c, space) \
397 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
398 rcu_check_sparse(p, space); \
399 ((typeof(*p) __force __kernel *)(p)); \
401 #define rcu_dereference_raw(p) \
403 /* Dependency order vs. p above. */ \
404 typeof(p) ________p1 = READ_ONCE(p); \
405 ((typeof(*p) __force __kernel *)(________p1)); \
409 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
410 * @v: The value to statically initialize with.
412 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
415 * rcu_assign_pointer() - assign to RCU-protected pointer
416 * @p: pointer to assign to
417 * @v: value to assign (publish)
419 * Assigns the specified value to the specified RCU-protected
420 * pointer, ensuring that any concurrent RCU readers will see
421 * any prior initialization.
423 * Inserts memory barriers on architectures that require them
424 * (which is most of them), and also prevents the compiler from
425 * reordering the code that initializes the structure after the pointer
426 * assignment. More importantly, this call documents which pointers
427 * will be dereferenced by RCU read-side code.
429 * In some special cases, you may use RCU_INIT_POINTER() instead
430 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
431 * to the fact that it does not constrain either the CPU or the compiler.
432 * That said, using RCU_INIT_POINTER() when you should have used
433 * rcu_assign_pointer() is a very bad thing that results in
434 * impossible-to-diagnose memory corruption. So please be careful.
435 * See the RCU_INIT_POINTER() comment header for details.
437 * Note that rcu_assign_pointer() evaluates each of its arguments only
438 * once, appearances notwithstanding. One of the "extra" evaluations
439 * is in typeof() and the other visible only to sparse (__CHECKER__),
440 * neither of which actually execute the argument. As with most cpp
441 * macros, this execute-arguments-only-once property is important, so
442 * please be careful when making changes to rcu_assign_pointer() and the
443 * other macros that it invokes.
445 #define rcu_assign_pointer(p, v) \
447 uintptr_t _r_a_p__v = (uintptr_t)(v); \
448 rcu_check_sparse(p, __rcu); \
450 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \
451 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \
453 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
457 * rcu_replace_pointer() - replace an RCU pointer, returning its old value
458 * @rcu_ptr: RCU pointer, whose old value is returned
459 * @ptr: regular pointer
460 * @c: the lockdep conditions under which the dereference will take place
462 * Perform a replacement, where @rcu_ptr is an RCU-annotated
463 * pointer and @c is the lockdep argument that is passed to the
464 * rcu_dereference_protected() call used to read that pointer. The old
465 * value of @rcu_ptr is returned, and @rcu_ptr is set to @ptr.
467 #define rcu_replace_pointer(rcu_ptr, ptr, c) \
469 typeof(ptr) __tmp = rcu_dereference_protected((rcu_ptr), (c)); \
470 rcu_assign_pointer((rcu_ptr), (ptr)); \
475 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
476 * @p: The pointer to read
478 * Return the value of the specified RCU-protected pointer, but omit the
479 * lockdep checks for being in an RCU read-side critical section. This is
480 * useful when the value of this pointer is accessed, but the pointer is
481 * not dereferenced, for example, when testing an RCU-protected pointer
482 * against NULL. Although rcu_access_pointer() may also be used in cases
483 * where update-side locks prevent the value of the pointer from changing,
484 * you should instead use rcu_dereference_protected() for this use case.
486 * It is also permissible to use rcu_access_pointer() when read-side
487 * access to the pointer was removed at least one grace period ago, as
488 * is the case in the context of the RCU callback that is freeing up
489 * the data, or after a synchronize_rcu() returns. This can be useful
490 * when tearing down multi-linked structures after a grace period
493 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
496 * rcu_dereference_check() - rcu_dereference with debug checking
497 * @p: The pointer to read, prior to dereferencing
498 * @c: The conditions under which the dereference will take place
500 * Do an rcu_dereference(), but check that the conditions under which the
501 * dereference will take place are correct. Typically the conditions
502 * indicate the various locking conditions that should be held at that
503 * point. The check should return true if the conditions are satisfied.
504 * An implicit check for being in an RCU read-side critical section
505 * (rcu_read_lock()) is included.
509 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
511 * could be used to indicate to lockdep that foo->bar may only be dereferenced
512 * if either rcu_read_lock() is held, or that the lock required to replace
513 * the bar struct at foo->bar is held.
515 * Note that the list of conditions may also include indications of when a lock
516 * need not be held, for example during initialisation or destruction of the
519 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
520 * atomic_read(&foo->usage) == 0);
522 * Inserts memory barriers on architectures that require them
523 * (currently only the Alpha), prevents the compiler from refetching
524 * (and from merging fetches), and, more importantly, documents exactly
525 * which pointers are protected by RCU and checks that the pointer is
526 * annotated as __rcu.
528 #define rcu_dereference_check(p, c) \
529 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu)
532 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
533 * @p: The pointer to read, prior to dereferencing
534 * @c: The conditions under which the dereference will take place
536 * This is the RCU-bh counterpart to rcu_dereference_check(). However,
537 * please note that starting in v5.0 kernels, vanilla RCU grace periods
538 * wait for local_bh_disable() regions of code in addition to regions of
539 * code demarked by rcu_read_lock() and rcu_read_unlock(). This means
540 * that synchronize_rcu(), call_rcu, and friends all take not only
541 * rcu_read_lock() but also rcu_read_lock_bh() into account.
543 #define rcu_dereference_bh_check(p, c) \
544 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu)
547 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
548 * @p: The pointer to read, prior to dereferencing
549 * @c: The conditions under which the dereference will take place
551 * This is the RCU-sched counterpart to rcu_dereference_check().
552 * However, please note that starting in v5.0 kernels, vanilla RCU grace
553 * periods wait for preempt_disable() regions of code in addition to
554 * regions of code demarked by rcu_read_lock() and rcu_read_unlock().
555 * This means that synchronize_rcu(), call_rcu, and friends all take not
556 * only rcu_read_lock() but also rcu_read_lock_sched() into account.
558 #define rcu_dereference_sched_check(p, c) \
559 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \
563 * The tracing infrastructure traces RCU (we want that), but unfortunately
564 * some of the RCU checks causes tracing to lock up the system.
566 * The no-tracing version of rcu_dereference_raw() must not call
567 * rcu_read_lock_held().
569 #define rcu_dereference_raw_check(p) __rcu_dereference_check((p), 1, __rcu)
572 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
573 * @p: The pointer to read, prior to dereferencing
574 * @c: The conditions under which the dereference will take place
576 * Return the value of the specified RCU-protected pointer, but omit
577 * the READ_ONCE(). This is useful in cases where update-side locks
578 * prevent the value of the pointer from changing. Please note that this
579 * primitive does *not* prevent the compiler from repeating this reference
580 * or combining it with other references, so it should not be used without
581 * protection of appropriate locks.
583 * This function is only for update-side use. Using this function
584 * when protected only by rcu_read_lock() will result in infrequent
585 * but very ugly failures.
587 #define rcu_dereference_protected(p, c) \
588 __rcu_dereference_protected((p), (c), __rcu)
592 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
593 * @p: The pointer to read, prior to dereferencing
595 * This is a simple wrapper around rcu_dereference_check().
597 #define rcu_dereference(p) rcu_dereference_check(p, 0)
600 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
601 * @p: The pointer to read, prior to dereferencing
603 * Makes rcu_dereference_check() do the dirty work.
605 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
608 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
609 * @p: The pointer to read, prior to dereferencing
611 * Makes rcu_dereference_check() do the dirty work.
613 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
616 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
617 * @p: The pointer to hand off
619 * This is simply an identity function, but it documents where a pointer
620 * is handed off from RCU to some other synchronization mechanism, for
621 * example, reference counting or locking. In C11, it would map to
622 * kill_dependency(). It could be used as follows::
625 * p = rcu_dereference(gp);
626 * long_lived = is_long_lived(p);
628 * if (!atomic_inc_not_zero(p->refcnt))
629 * long_lived = false;
631 * p = rcu_pointer_handoff(p);
635 #define rcu_pointer_handoff(p) (p)
638 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
640 * When synchronize_rcu() is invoked on one CPU while other CPUs
641 * are within RCU read-side critical sections, then the
642 * synchronize_rcu() is guaranteed to block until after all the other
643 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
644 * on one CPU while other CPUs are within RCU read-side critical
645 * sections, invocation of the corresponding RCU callback is deferred
646 * until after the all the other CPUs exit their critical sections.
648 * In v5.0 and later kernels, synchronize_rcu() and call_rcu() also
649 * wait for regions of code with preemption disabled, including regions of
650 * code with interrupts or softirqs disabled. In pre-v5.0 kernels, which
651 * define synchronize_sched(), only code enclosed within rcu_read_lock()
652 * and rcu_read_unlock() are guaranteed to be waited for.
654 * Note, however, that RCU callbacks are permitted to run concurrently
655 * with new RCU read-side critical sections. One way that this can happen
656 * is via the following sequence of events: (1) CPU 0 enters an RCU
657 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
658 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
659 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
660 * callback is invoked. This is legal, because the RCU read-side critical
661 * section that was running concurrently with the call_rcu() (and which
662 * therefore might be referencing something that the corresponding RCU
663 * callback would free up) has completed before the corresponding
664 * RCU callback is invoked.
666 * RCU read-side critical sections may be nested. Any deferred actions
667 * will be deferred until the outermost RCU read-side critical section
670 * You can avoid reading and understanding the next paragraph by
671 * following this rule: don't put anything in an rcu_read_lock() RCU
672 * read-side critical section that would block in a !PREEMPTION kernel.
673 * But if you want the full story, read on!
675 * In non-preemptible RCU implementations (pure TREE_RCU and TINY_RCU),
676 * it is illegal to block while in an RCU read-side critical section.
677 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPTION
678 * kernel builds, RCU read-side critical sections may be preempted,
679 * but explicit blocking is illegal. Finally, in preemptible RCU
680 * implementations in real-time (with -rt patchset) kernel builds, RCU
681 * read-side critical sections may be preempted and they may also block, but
682 * only when acquiring spinlocks that are subject to priority inheritance.
684 static __always_inline void rcu_read_lock(void)
688 rcu_lock_acquire(&rcu_lock_map);
689 RCU_LOCKDEP_WARN(!rcu_is_watching(),
690 "rcu_read_lock() used illegally while idle");
694 * So where is rcu_write_lock()? It does not exist, as there is no
695 * way for writers to lock out RCU readers. This is a feature, not
696 * a bug -- this property is what provides RCU's performance benefits.
697 * Of course, writers must coordinate with each other. The normal
698 * spinlock primitives work well for this, but any other technique may be
699 * used as well. RCU does not care how the writers keep out of each
700 * others' way, as long as they do so.
704 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
706 * In almost all situations, rcu_read_unlock() is immune from deadlock.
707 * In recent kernels that have consolidated synchronize_sched() and
708 * synchronize_rcu_bh() into synchronize_rcu(), this deadlock immunity
709 * also extends to the scheduler's runqueue and priority-inheritance
710 * spinlocks, courtesy of the quiescent-state deferral that is carried
711 * out when rcu_read_unlock() is invoked with interrupts disabled.
713 * See rcu_read_lock() for more information.
715 static inline void rcu_read_unlock(void)
717 RCU_LOCKDEP_WARN(!rcu_is_watching(),
718 "rcu_read_unlock() used illegally while idle");
721 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
725 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
727 * This is equivalent to rcu_read_lock(), but also disables softirqs.
728 * Note that anything else that disables softirqs can also serve as an RCU
729 * read-side critical section. However, please note that this equivalence
730 * applies only to v5.0 and later. Before v5.0, rcu_read_lock() and
731 * rcu_read_lock_bh() were unrelated.
733 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
734 * must occur in the same context, for example, it is illegal to invoke
735 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
736 * was invoked from some other task.
738 static inline void rcu_read_lock_bh(void)
742 rcu_lock_acquire(&rcu_bh_lock_map);
743 RCU_LOCKDEP_WARN(!rcu_is_watching(),
744 "rcu_read_lock_bh() used illegally while idle");
748 * rcu_read_unlock_bh() - marks the end of a softirq-only RCU critical section
750 * See rcu_read_lock_bh() for more information.
752 static inline void rcu_read_unlock_bh(void)
754 RCU_LOCKDEP_WARN(!rcu_is_watching(),
755 "rcu_read_unlock_bh() used illegally while idle");
756 rcu_lock_release(&rcu_bh_lock_map);
762 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
764 * This is equivalent to rcu_read_lock(), but also disables preemption.
765 * Read-side critical sections can also be introduced by anything else that
766 * disables preemption, including local_irq_disable() and friends. However,
767 * please note that the equivalence to rcu_read_lock() applies only to
768 * v5.0 and later. Before v5.0, rcu_read_lock() and rcu_read_lock_sched()
771 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
772 * must occur in the same context, for example, it is illegal to invoke
773 * rcu_read_unlock_sched() from process context if the matching
774 * rcu_read_lock_sched() was invoked from an NMI handler.
776 static inline void rcu_read_lock_sched(void)
779 __acquire(RCU_SCHED);
780 rcu_lock_acquire(&rcu_sched_lock_map);
781 RCU_LOCKDEP_WARN(!rcu_is_watching(),
782 "rcu_read_lock_sched() used illegally while idle");
785 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
786 static inline notrace void rcu_read_lock_sched_notrace(void)
788 preempt_disable_notrace();
789 __acquire(RCU_SCHED);
793 * rcu_read_unlock_sched() - marks the end of a RCU-classic critical section
795 * See rcu_read_lock_sched() for more information.
797 static inline void rcu_read_unlock_sched(void)
799 RCU_LOCKDEP_WARN(!rcu_is_watching(),
800 "rcu_read_unlock_sched() used illegally while idle");
801 rcu_lock_release(&rcu_sched_lock_map);
802 __release(RCU_SCHED);
806 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
807 static inline notrace void rcu_read_unlock_sched_notrace(void)
809 __release(RCU_SCHED);
810 preempt_enable_notrace();
814 * RCU_INIT_POINTER() - initialize an RCU protected pointer
815 * @p: The pointer to be initialized.
816 * @v: The value to initialized the pointer to.
818 * Initialize an RCU-protected pointer in special cases where readers
819 * do not need ordering constraints on the CPU or the compiler. These
822 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer *or*
823 * 2. The caller has taken whatever steps are required to prevent
824 * RCU readers from concurrently accessing this pointer *or*
825 * 3. The referenced data structure has already been exposed to
826 * readers either at compile time or via rcu_assign_pointer() *and*
828 * a. You have not made *any* reader-visible changes to
829 * this structure since then *or*
830 * b. It is OK for readers accessing this structure from its
831 * new location to see the old state of the structure. (For
832 * example, the changes were to statistical counters or to
833 * other state where exact synchronization is not required.)
835 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
836 * result in impossible-to-diagnose memory corruption. As in the structures
837 * will look OK in crash dumps, but any concurrent RCU readers might
838 * see pre-initialized values of the referenced data structure. So
839 * please be very careful how you use RCU_INIT_POINTER()!!!
841 * If you are creating an RCU-protected linked structure that is accessed
842 * by a single external-to-structure RCU-protected pointer, then you may
843 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
844 * pointers, but you must use rcu_assign_pointer() to initialize the
845 * external-to-structure pointer *after* you have completely initialized
846 * the reader-accessible portions of the linked structure.
848 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
849 * ordering guarantees for either the CPU or the compiler.
851 #define RCU_INIT_POINTER(p, v) \
853 rcu_check_sparse(p, __rcu); \
854 WRITE_ONCE(p, RCU_INITIALIZER(v)); \
858 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
859 * @p: The pointer to be initialized.
860 * @v: The value to initialized the pointer to.
862 * GCC-style initialization for an RCU-protected pointer in a structure field.
864 #define RCU_POINTER_INITIALIZER(p, v) \
865 .p = RCU_INITIALIZER(v)
868 * Does the specified offset indicate that the corresponding rcu_head
869 * structure can be handled by kvfree_rcu()?
871 #define __is_kvfree_rcu_offset(offset) ((offset) < 4096)
874 * kfree_rcu() - kfree an object after a grace period.
875 * @ptr: pointer to kfree for both single- and double-argument invocations.
876 * @rhf: the name of the struct rcu_head within the type of @ptr,
877 * but only for double-argument invocations.
879 * Many rcu callbacks functions just call kfree() on the base structure.
880 * These functions are trivial, but their size adds up, and furthermore
881 * when they are used in a kernel module, that module must invoke the
882 * high-latency rcu_barrier() function at module-unload time.
884 * The kfree_rcu() function handles this issue. Rather than encoding a
885 * function address in the embedded rcu_head structure, kfree_rcu() instead
886 * encodes the offset of the rcu_head structure within the base structure.
887 * Because the functions are not allowed in the low-order 4096 bytes of
888 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
889 * If the offset is larger than 4095 bytes, a compile-time error will
890 * be generated in kvfree_rcu_arg_2(). If this error is triggered, you can
891 * either fall back to use of call_rcu() or rearrange the structure to
892 * position the rcu_head structure into the first 4096 bytes.
894 * Note that the allowable offset might decrease in the future, for example,
895 * to allow something like kmem_cache_free_rcu().
897 * The BUILD_BUG_ON check must not involve any function calls, hence the
898 * checks are done in macros here.
900 #define kfree_rcu(ptr, rhf...) kvfree_rcu(ptr, ## rhf)
903 * kvfree_rcu() - kvfree an object after a grace period.
905 * This macro consists of one or two arguments and it is
906 * based on whether an object is head-less or not. If it
907 * has a head then a semantic stays the same as it used
910 * kvfree_rcu(ptr, rhf);
912 * where @ptr is a pointer to kvfree(), @rhf is the name
913 * of the rcu_head structure within the type of @ptr.
915 * When it comes to head-less variant, only one argument
916 * is passed and that is just a pointer which has to be
917 * freed after a grace period. Therefore the semantic is
921 * where @ptr is a pointer to kvfree().
923 * Please note, head-less way of freeing is permitted to
924 * use from a context that has to follow might_sleep()
925 * annotation. Otherwise, please switch and embed the
926 * rcu_head structure within the type of @ptr.
928 #define kvfree_rcu(...) KVFREE_GET_MACRO(__VA_ARGS__, \
929 kvfree_rcu_arg_2, kvfree_rcu_arg_1)(__VA_ARGS__)
931 #define KVFREE_GET_MACRO(_1, _2, NAME, ...) NAME
932 #define kvfree_rcu_arg_2(ptr, rhf) \
934 typeof (ptr) ___p = (ptr); \
937 BUILD_BUG_ON(!__is_kvfree_rcu_offset(offsetof(typeof(*(ptr)), rhf))); \
938 kvfree_call_rcu(&((___p)->rhf), (rcu_callback_t)(unsigned long) \
939 (offsetof(typeof(*(ptr)), rhf))); \
943 #define kvfree_rcu_arg_1(ptr) \
945 typeof(ptr) ___p = (ptr); \
948 kvfree_call_rcu(NULL, (rcu_callback_t) (___p)); \
952 * Place this after a lock-acquisition primitive to guarantee that
953 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies
954 * if the UNLOCK and LOCK are executed by the same CPU or if the
955 * UNLOCK and LOCK operate on the same lock variable.
957 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
958 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */
959 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
960 #define smp_mb__after_unlock_lock() do { } while (0)
961 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
964 /* Has the specified rcu_head structure been handed to call_rcu()? */
967 * rcu_head_init - Initialize rcu_head for rcu_head_after_call_rcu()
968 * @rhp: The rcu_head structure to initialize.
970 * If you intend to invoke rcu_head_after_call_rcu() to test whether a
971 * given rcu_head structure has already been passed to call_rcu(), then
972 * you must also invoke this rcu_head_init() function on it just after
973 * allocating that structure. Calls to this function must not race with
974 * calls to call_rcu(), rcu_head_after_call_rcu(), or callback invocation.
976 static inline void rcu_head_init(struct rcu_head *rhp)
978 rhp->func = (rcu_callback_t)~0L;
982 * rcu_head_after_call_rcu() - Has this rcu_head been passed to call_rcu()?
983 * @rhp: The rcu_head structure to test.
984 * @f: The function passed to call_rcu() along with @rhp.
986 * Returns @true if the @rhp has been passed to call_rcu() with @func,
987 * and @false otherwise. Emits a warning in any other case, including
988 * the case where @rhp has already been invoked after a grace period.
989 * Calls to this function must not race with callback invocation. One way
990 * to avoid such races is to enclose the call to rcu_head_after_call_rcu()
991 * in an RCU read-side critical section that includes a read-side fetch
992 * of the pointer to the structure containing @rhp.
995 rcu_head_after_call_rcu(struct rcu_head *rhp, rcu_callback_t f)
997 rcu_callback_t func = READ_ONCE(rhp->func);
1001 WARN_ON_ONCE(func != (rcu_callback_t)~0L);
1005 /* kernel/ksysfs.c definitions */
1006 extern int rcu_expedited;
1007 extern int rcu_normal;
1009 #endif /* __LINUX_RCUPDATE_H */