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/debugobjects.h>
44 #include <linux/bug.h>
45 #include <linux/compiler.h>
46 #include <linux/ktime.h>
47 #include <linux/irqflags.h>
49 #include <asm/barrier.h>
51 #ifndef CONFIG_TINY_RCU
52 extern int rcu_expedited
; /* for sysctl */
53 extern int rcu_normal
; /* also for sysctl */
54 #endif /* #ifndef CONFIG_TINY_RCU */
56 #ifdef CONFIG_TINY_RCU
57 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
58 static inline bool rcu_gp_is_normal(void) /* Internal RCU use. */
62 static inline bool rcu_gp_is_expedited(void) /* Internal RCU use. */
67 static inline void rcu_expedite_gp(void)
71 static inline void rcu_unexpedite_gp(void)
74 #else /* #ifdef CONFIG_TINY_RCU */
75 bool rcu_gp_is_normal(void); /* Internal RCU use. */
76 bool rcu_gp_is_expedited(void); /* Internal RCU use. */
77 void rcu_expedite_gp(void);
78 void rcu_unexpedite_gp(void);
79 #endif /* #else #ifdef CONFIG_TINY_RCU */
81 enum rcutorture_type
{
90 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
91 void rcutorture_get_gp_data(enum rcutorture_type test_type
, int *flags
,
92 unsigned long *gpnum
, unsigned long *completed
);
93 void rcutorture_record_test_transition(void);
94 void rcutorture_record_progress(unsigned long vernum
);
95 void do_trace_rcu_torture_read(const char *rcutorturename
,
100 bool rcu_irq_enter_disabled(void);
102 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type
,
104 unsigned long *gpnum
,
105 unsigned long *completed
)
111 static inline void rcutorture_record_test_transition(void)
114 static inline void rcutorture_record_progress(unsigned long vernum
)
117 static inline bool rcu_irq_enter_disabled(void)
121 #ifdef CONFIG_RCU_TRACE
122 void do_trace_rcu_torture_read(const char *rcutorturename
,
123 struct rcu_head
*rhp
,
128 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
133 #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b))
134 #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b))
135 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
136 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
137 #define ulong2long(a) (*(long *)(&(a)))
139 /* Exported common interfaces */
141 #ifdef CONFIG_PREEMPT_RCU
144 * call_rcu() - Queue an RCU callback for invocation after a grace period.
145 * @head: structure to be used for queueing the RCU updates.
146 * @func: actual callback function to be invoked after the grace period
148 * The callback function will be invoked some time after a full grace
149 * period elapses, in other words after all pre-existing RCU read-side
150 * critical sections have completed. However, the callback function
151 * might well execute concurrently with RCU read-side critical sections
152 * that started after call_rcu() was invoked. RCU read-side critical
153 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
156 * Note that all CPUs must agree that the grace period extended beyond
157 * all pre-existing RCU read-side critical section. On systems with more
158 * than one CPU, this means that when "func()" is invoked, each CPU is
159 * guaranteed to have executed a full memory barrier since the end of its
160 * last RCU read-side critical section whose beginning preceded the call
161 * to call_rcu(). It also means that each CPU executing an RCU read-side
162 * critical section that continues beyond the start of "func()" must have
163 * executed a memory barrier after the call_rcu() but before the beginning
164 * of that RCU read-side critical section. Note that these guarantees
165 * include CPUs that are offline, idle, or executing in user mode, as
166 * well as CPUs that are executing in the kernel.
168 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
169 * resulting RCU callback function "func()", then both CPU A and CPU B are
170 * guaranteed to execute a full memory barrier during the time interval
171 * between the call to call_rcu() and the invocation of "func()" -- even
172 * if CPU A and CPU B are the same CPU (but again only if the system has
173 * more than one CPU).
175 void call_rcu(struct rcu_head
*head
,
176 rcu_callback_t func
);
178 #else /* #ifdef CONFIG_PREEMPT_RCU */
180 /* In classic RCU, call_rcu() is just call_rcu_sched(). */
181 #define call_rcu call_rcu_sched
183 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
186 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
187 * @head: structure to be used for queueing the RCU updates.
188 * @func: actual callback function to be invoked after the grace period
190 * The callback function will be invoked some time after a full grace
191 * period elapses, in other words after all currently executing RCU
192 * read-side critical sections have completed. call_rcu_bh() assumes
193 * that the read-side critical sections end on completion of a softirq
194 * handler. This means that read-side critical sections in process
195 * context must not be interrupted by softirqs. This interface is to be
196 * used when most of the read-side critical sections are in softirq context.
197 * RCU read-side critical sections are delimited by :
198 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context.
200 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
201 * These may be nested.
203 * See the description of call_rcu() for more detailed information on
204 * memory ordering guarantees.
206 void call_rcu_bh(struct rcu_head
*head
,
207 rcu_callback_t func
);
210 * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
211 * @head: structure to be used for queueing the RCU updates.
212 * @func: actual callback function to be invoked after the grace period
214 * The callback function will be invoked some time after a full grace
215 * period elapses, in other words after all currently executing RCU
216 * read-side critical sections have completed. call_rcu_sched() assumes
217 * that the read-side critical sections end on enabling of preemption
218 * or on voluntary preemption.
219 * RCU read-side critical sections are delimited by :
220 * - rcu_read_lock_sched() and rcu_read_unlock_sched(),
222 * anything that disables preemption.
223 * These may be nested.
225 * See the description of call_rcu() for more detailed information on
226 * memory ordering guarantees.
228 void call_rcu_sched(struct rcu_head
*head
,
229 rcu_callback_t func
);
231 void synchronize_sched(void);
234 * call_rcu_tasks() - Queue an RCU for invocation task-based grace period
235 * @head: structure to be used for queueing the RCU updates.
236 * @func: actual callback function to be invoked after the grace period
238 * The callback function will be invoked some time after a full grace
239 * period elapses, in other words after all currently executing RCU
240 * read-side critical sections have completed. call_rcu_tasks() assumes
241 * that the read-side critical sections end at a voluntary context
242 * switch (not a preemption!), entry into idle, or transition to usermode
243 * execution. As such, there are no read-side primitives analogous to
244 * rcu_read_lock() and rcu_read_unlock() because this primitive is intended
245 * to determine that all tasks have passed through a safe state, not so
246 * much for data-strcuture synchronization.
248 * See the description of call_rcu() for more detailed information on
249 * memory ordering guarantees.
251 void call_rcu_tasks(struct rcu_head
*head
, rcu_callback_t func
);
252 void synchronize_rcu_tasks(void);
253 void rcu_barrier_tasks(void);
255 #ifdef CONFIG_PREEMPT_RCU
257 void __rcu_read_lock(void);
258 void __rcu_read_unlock(void);
259 void rcu_read_unlock_special(struct task_struct
*t
);
260 void synchronize_rcu(void);
263 * Defined as a macro as it is a very low level header included from
264 * areas that don't even know about current. This gives the rcu_read_lock()
265 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
266 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
268 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
270 #else /* #ifdef CONFIG_PREEMPT_RCU */
272 static inline void __rcu_read_lock(void)
274 if (IS_ENABLED(CONFIG_PREEMPT_COUNT
))
278 static inline void __rcu_read_unlock(void)
280 if (IS_ENABLED(CONFIG_PREEMPT_COUNT
))
284 static inline void synchronize_rcu(void)
289 static inline int rcu_preempt_depth(void)
294 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
296 /* Internal to kernel */
298 void rcu_sched_qs(void);
299 void rcu_bh_qs(void);
300 void rcu_check_callbacks(int user
);
301 void rcu_report_dead(unsigned int cpu
);
302 void rcu_cpu_starting(unsigned int cpu
);
304 #ifndef CONFIG_TINY_RCU
305 void rcu_end_inkernel_boot(void);
306 #else /* #ifndef CONFIG_TINY_RCU */
307 static inline void rcu_end_inkernel_boot(void) { }
308 #endif /* #ifndef CONFIG_TINY_RCU */
310 #ifdef CONFIG_RCU_STALL_COMMON
311 void rcu_sysrq_start(void);
312 void rcu_sysrq_end(void);
313 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
314 static inline void rcu_sysrq_start(void)
317 static inline void rcu_sysrq_end(void)
320 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
322 #ifdef CONFIG_NO_HZ_FULL
323 void rcu_user_enter(void);
324 void rcu_user_exit(void);
326 static inline void rcu_user_enter(void) { }
327 static inline void rcu_user_exit(void) { }
328 #endif /* CONFIG_NO_HZ_FULL */
330 #ifdef CONFIG_RCU_NOCB_CPU
331 void rcu_init_nohz(void);
332 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
333 static inline void rcu_init_nohz(void)
336 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
339 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
340 * @a: Code that RCU needs to pay attention to.
342 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
343 * in the inner idle loop, that is, between the rcu_idle_enter() and
344 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
345 * critical sections. However, things like powertop need tracepoints
346 * in the inner idle loop.
348 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
349 * will tell RCU that it needs to pay attention, invoke its argument
350 * (in this example, calling the do_something_with_RCU() function),
351 * and then tell RCU to go back to ignoring this CPU. It is permissible
352 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is
353 * on the order of a million or so, even on 32-bit systems). It is
354 * not legal to block within RCU_NONIDLE(), nor is it permissible to
355 * transfer control either into or out of RCU_NONIDLE()'s statement.
357 #define RCU_NONIDLE(a) \
359 rcu_irq_enter_irqson(); \
360 do { a; } while (0); \
361 rcu_irq_exit_irqson(); \
365 * Note a voluntary context switch for RCU-tasks benefit. This is a
366 * macro rather than an inline function to avoid #include hell.
368 #ifdef CONFIG_TASKS_RCU
369 #define TASKS_RCU(x) x
370 extern struct srcu_struct tasks_rcu_exit_srcu
;
371 #define rcu_note_voluntary_context_switch_lite(t) \
373 if (READ_ONCE((t)->rcu_tasks_holdout)) \
374 WRITE_ONCE((t)->rcu_tasks_holdout, false); \
376 #define rcu_note_voluntary_context_switch(t) \
379 rcu_note_voluntary_context_switch_lite(t); \
381 #else /* #ifdef CONFIG_TASKS_RCU */
382 #define TASKS_RCU(x) do { } while (0)
383 #define rcu_note_voluntary_context_switch_lite(t) do { } while (0)
384 #define rcu_note_voluntary_context_switch(t) rcu_all_qs()
385 #endif /* #else #ifdef CONFIG_TASKS_RCU */
388 * cond_resched_rcu_qs - Report potential quiescent states to RCU
390 * This macro resembles cond_resched(), except that it is defined to
391 * report potential quiescent states to RCU-tasks even if the cond_resched()
392 * machinery were to be shut off, as some advocate for PREEMPT kernels.
394 #define cond_resched_rcu_qs() \
396 if (!cond_resched()) \
397 rcu_note_voluntary_context_switch(current); \
400 #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP)
401 bool __rcu_is_watching(void);
402 #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */
405 * Infrastructure to implement the synchronize_() primitives in
406 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
409 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
410 #include <linux/rcutree.h>
411 #elif defined(CONFIG_TINY_RCU)
412 #include <linux/rcutiny.h>
414 #error "Unknown RCU implementation specified to kernel configuration"
417 #define RCU_SCHEDULER_INACTIVE 0
418 #define RCU_SCHEDULER_INIT 1
419 #define RCU_SCHEDULER_RUNNING 2
422 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
423 * initialization and destruction of rcu_head on the stack. rcu_head structures
424 * allocated dynamically in the heap or defined statically don't need any
427 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
428 void init_rcu_head(struct rcu_head
*head
);
429 void destroy_rcu_head(struct rcu_head
*head
);
430 void init_rcu_head_on_stack(struct rcu_head
*head
);
431 void destroy_rcu_head_on_stack(struct rcu_head
*head
);
432 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
433 static inline void init_rcu_head(struct rcu_head
*head
)
437 static inline void destroy_rcu_head(struct rcu_head
*head
)
441 static inline void init_rcu_head_on_stack(struct rcu_head
*head
)
445 static inline void destroy_rcu_head_on_stack(struct rcu_head
*head
)
448 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
450 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
451 bool rcu_lockdep_current_cpu_online(void);
452 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
453 static inline bool rcu_lockdep_current_cpu_online(void)
457 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
459 #ifdef CONFIG_DEBUG_LOCK_ALLOC
461 static inline void rcu_lock_acquire(struct lockdep_map
*map
)
463 lock_acquire(map
, 0, 0, 2, 0, NULL
, _THIS_IP_
);
466 static inline void rcu_lock_release(struct lockdep_map
*map
)
468 lock_release(map
, 1, _THIS_IP_
);
471 extern struct lockdep_map rcu_lock_map
;
472 extern struct lockdep_map rcu_bh_lock_map
;
473 extern struct lockdep_map rcu_sched_lock_map
;
474 extern struct lockdep_map rcu_callback_map
;
475 int debug_lockdep_rcu_enabled(void);
477 int rcu_read_lock_held(void);
478 int rcu_read_lock_bh_held(void);
481 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
483 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
484 * RCU-sched read-side critical section. In absence of
485 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
486 * critical section unless it can prove otherwise.
488 int rcu_read_lock_sched_held(void);
490 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
492 # define rcu_lock_acquire(a) do { } while (0)
493 # define rcu_lock_release(a) do { } while (0)
495 static inline int rcu_read_lock_held(void)
500 static inline int rcu_read_lock_bh_held(void)
505 static inline int rcu_read_lock_sched_held(void)
507 return !preemptible();
509 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
511 #ifdef CONFIG_PROVE_RCU
514 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
515 * @c: condition to check
516 * @s: informative message
518 #define RCU_LOCKDEP_WARN(c, s) \
520 static bool __section(.data.unlikely) __warned; \
521 if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \
523 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
527 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
528 static inline void rcu_preempt_sleep_check(void)
530 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map
),
531 "Illegal context switch in RCU read-side critical section");
533 #else /* #ifdef CONFIG_PROVE_RCU */
534 static inline void rcu_preempt_sleep_check(void)
537 #endif /* #else #ifdef CONFIG_PROVE_RCU */
539 #define rcu_sleep_check() \
541 rcu_preempt_sleep_check(); \
542 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \
543 "Illegal context switch in RCU-bh read-side critical section"); \
544 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \
545 "Illegal context switch in RCU-sched read-side critical section"); \
548 #else /* #ifdef CONFIG_PROVE_RCU */
550 #define RCU_LOCKDEP_WARN(c, s) do { } while (0)
551 #define rcu_sleep_check() do { } while (0)
553 #endif /* #else #ifdef CONFIG_PROVE_RCU */
556 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
557 * and rcu_assign_pointer(). Some of these could be folded into their
558 * callers, but they are left separate in order to ease introduction of
559 * multiple flavors of pointers to match the multiple flavors of RCU
560 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
565 #define rcu_dereference_sparse(p, space) \
566 ((void)(((typeof(*p) space *)p) == p))
567 #else /* #ifdef __CHECKER__ */
568 #define rcu_dereference_sparse(p, space)
569 #endif /* #else #ifdef __CHECKER__ */
571 #define __rcu_access_pointer(p, space) \
573 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \
574 rcu_dereference_sparse(p, space); \
575 ((typeof(*p) __force __kernel *)(_________p1)); \
577 #define __rcu_dereference_check(p, c, space) \
579 /* Dependency order vs. p above. */ \
580 typeof(*p) *________p1 = (typeof(*p) *__force)lockless_dereference(p); \
581 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
582 rcu_dereference_sparse(p, space); \
583 ((typeof(*p) __force __kernel *)(________p1)); \
585 #define __rcu_dereference_protected(p, c, space) \
587 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
588 rcu_dereference_sparse(p, space); \
589 ((typeof(*p) __force __kernel *)(p)); \
591 #define rcu_dereference_raw(p) \
593 /* Dependency order vs. p above. */ \
594 typeof(p) ________p1 = lockless_dereference(p); \
595 ((typeof(*p) __force __kernel *)(________p1)); \
599 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
600 * @v: The value to statically initialize with.
602 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
605 * rcu_assign_pointer() - assign to RCU-protected pointer
606 * @p: pointer to assign to
607 * @v: value to assign (publish)
609 * Assigns the specified value to the specified RCU-protected
610 * pointer, ensuring that any concurrent RCU readers will see
611 * any prior initialization.
613 * Inserts memory barriers on architectures that require them
614 * (which is most of them), and also prevents the compiler from
615 * reordering the code that initializes the structure after the pointer
616 * assignment. More importantly, this call documents which pointers
617 * will be dereferenced by RCU read-side code.
619 * In some special cases, you may use RCU_INIT_POINTER() instead
620 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
621 * to the fact that it does not constrain either the CPU or the compiler.
622 * That said, using RCU_INIT_POINTER() when you should have used
623 * rcu_assign_pointer() is a very bad thing that results in
624 * impossible-to-diagnose memory corruption. So please be careful.
625 * See the RCU_INIT_POINTER() comment header for details.
627 * Note that rcu_assign_pointer() evaluates each of its arguments only
628 * once, appearances notwithstanding. One of the "extra" evaluations
629 * is in typeof() and the other visible only to sparse (__CHECKER__),
630 * neither of which actually execute the argument. As with most cpp
631 * macros, this execute-arguments-only-once property is important, so
632 * please be careful when making changes to rcu_assign_pointer() and the
633 * other macros that it invokes.
635 #define rcu_assign_pointer(p, v) \
637 uintptr_t _r_a_p__v = (uintptr_t)(v); \
639 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \
640 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \
642 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
647 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
648 * @p: The pointer to read
650 * Return the value of the specified RCU-protected pointer, but omit the
651 * smp_read_barrier_depends() and keep the READ_ONCE(). This is useful
652 * when the value of this pointer is accessed, but the pointer is not
653 * dereferenced, for example, when testing an RCU-protected pointer against
654 * NULL. Although rcu_access_pointer() may also be used in cases where
655 * update-side locks prevent the value of the pointer from changing, you
656 * should instead use rcu_dereference_protected() for this use case.
658 * It is also permissible to use rcu_access_pointer() when read-side
659 * access to the pointer was removed at least one grace period ago, as
660 * is the case in the context of the RCU callback that is freeing up
661 * the data, or after a synchronize_rcu() returns. This can be useful
662 * when tearing down multi-linked structures after a grace period
665 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
668 * rcu_dereference_check() - rcu_dereference with debug checking
669 * @p: The pointer to read, prior to dereferencing
670 * @c: The conditions under which the dereference will take place
672 * Do an rcu_dereference(), but check that the conditions under which the
673 * dereference will take place are correct. Typically the conditions
674 * indicate the various locking conditions that should be held at that
675 * point. The check should return true if the conditions are satisfied.
676 * An implicit check for being in an RCU read-side critical section
677 * (rcu_read_lock()) is included.
681 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
683 * could be used to indicate to lockdep that foo->bar may only be dereferenced
684 * if either rcu_read_lock() is held, or that the lock required to replace
685 * the bar struct at foo->bar is held.
687 * Note that the list of conditions may also include indications of when a lock
688 * need not be held, for example during initialisation or destruction of the
691 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
692 * atomic_read(&foo->usage) == 0);
694 * Inserts memory barriers on architectures that require them
695 * (currently only the Alpha), prevents the compiler from refetching
696 * (and from merging fetches), and, more importantly, documents exactly
697 * which pointers are protected by RCU and checks that the pointer is
698 * annotated as __rcu.
700 #define rcu_dereference_check(p, c) \
701 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu)
704 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
705 * @p: The pointer to read, prior to dereferencing
706 * @c: The conditions under which the dereference will take place
708 * This is the RCU-bh counterpart to rcu_dereference_check().
710 #define rcu_dereference_bh_check(p, c) \
711 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu)
714 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
715 * @p: The pointer to read, prior to dereferencing
716 * @c: The conditions under which the dereference will take place
718 * This is the RCU-sched counterpart to rcu_dereference_check().
720 #define rcu_dereference_sched_check(p, c) \
721 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \
725 * The tracing infrastructure traces RCU (we want that), but unfortunately
726 * some of the RCU checks causes tracing to lock up the system.
728 * The no-tracing version of rcu_dereference_raw() must not call
729 * rcu_read_lock_held().
731 #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
734 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
735 * @p: The pointer to read, prior to dereferencing
736 * @c: The conditions under which the dereference will take place
738 * Return the value of the specified RCU-protected pointer, but omit
739 * both the smp_read_barrier_depends() and the READ_ONCE(). This
740 * is useful in cases where update-side locks prevent the value of the
741 * pointer from changing. Please note that this primitive does -not-
742 * prevent the compiler from repeating this reference or combining it
743 * with other references, so it should not be used without protection
744 * of appropriate locks.
746 * This function is only for update-side use. Using this function
747 * when protected only by rcu_read_lock() will result in infrequent
748 * but very ugly failures.
750 #define rcu_dereference_protected(p, c) \
751 __rcu_dereference_protected((p), (c), __rcu)
755 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
756 * @p: The pointer to read, prior to dereferencing
758 * This is a simple wrapper around rcu_dereference_check().
760 #define rcu_dereference(p) rcu_dereference_check(p, 0)
763 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
764 * @p: The pointer to read, prior to dereferencing
766 * Makes rcu_dereference_check() do the dirty work.
768 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
771 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
772 * @p: The pointer to read, prior to dereferencing
774 * Makes rcu_dereference_check() do the dirty work.
776 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
779 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
780 * @p: The pointer to hand off
782 * This is simply an identity function, but it documents where a pointer
783 * is handed off from RCU to some other synchronization mechanism, for
784 * example, reference counting or locking. In C11, it would map to
785 * kill_dependency(). It could be used as follows:
788 * p = rcu_dereference(gp);
789 * long_lived = is_long_lived(p);
791 * if (!atomic_inc_not_zero(p->refcnt))
792 * long_lived = false;
794 * p = rcu_pointer_handoff(p);
798 #define rcu_pointer_handoff(p) (p)
801 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
803 * When synchronize_rcu() is invoked on one CPU while other CPUs
804 * are within RCU read-side critical sections, then the
805 * synchronize_rcu() is guaranteed to block until after all the other
806 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
807 * on one CPU while other CPUs are within RCU read-side critical
808 * sections, invocation of the corresponding RCU callback is deferred
809 * until after the all the other CPUs exit their critical sections.
811 * Note, however, that RCU callbacks are permitted to run concurrently
812 * with new RCU read-side critical sections. One way that this can happen
813 * is via the following sequence of events: (1) CPU 0 enters an RCU
814 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
815 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
816 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
817 * callback is invoked. This is legal, because the RCU read-side critical
818 * section that was running concurrently with the call_rcu() (and which
819 * therefore might be referencing something that the corresponding RCU
820 * callback would free up) has completed before the corresponding
821 * RCU callback is invoked.
823 * RCU read-side critical sections may be nested. Any deferred actions
824 * will be deferred until the outermost RCU read-side critical section
827 * You can avoid reading and understanding the next paragraph by
828 * following this rule: don't put anything in an rcu_read_lock() RCU
829 * read-side critical section that would block in a !PREEMPT kernel.
830 * But if you want the full story, read on!
832 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU),
833 * it is illegal to block while in an RCU read-side critical section.
834 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT
835 * kernel builds, RCU read-side critical sections may be preempted,
836 * but explicit blocking is illegal. Finally, in preemptible RCU
837 * implementations in real-time (with -rt patchset) kernel builds, RCU
838 * read-side critical sections may be preempted and they may also block, but
839 * only when acquiring spinlocks that are subject to priority inheritance.
841 static inline void rcu_read_lock(void)
845 rcu_lock_acquire(&rcu_lock_map
);
846 RCU_LOCKDEP_WARN(!rcu_is_watching(),
847 "rcu_read_lock() used illegally while idle");
851 * So where is rcu_write_lock()? It does not exist, as there is no
852 * way for writers to lock out RCU readers. This is a feature, not
853 * a bug -- this property is what provides RCU's performance benefits.
854 * Of course, writers must coordinate with each other. The normal
855 * spinlock primitives work well for this, but any other technique may be
856 * used as well. RCU does not care how the writers keep out of each
857 * others' way, as long as they do so.
861 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
863 * In most situations, rcu_read_unlock() is immune from deadlock.
864 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock()
865 * is responsible for deboosting, which it does via rt_mutex_unlock().
866 * Unfortunately, this function acquires the scheduler's runqueue and
867 * priority-inheritance spinlocks. This means that deadlock could result
868 * if the caller of rcu_read_unlock() already holds one of these locks or
869 * any lock that is ever acquired while holding them; or any lock which
870 * can be taken from interrupt context because rcu_boost()->rt_mutex_lock()
871 * does not disable irqs while taking ->wait_lock.
873 * That said, RCU readers are never priority boosted unless they were
874 * preempted. Therefore, one way to avoid deadlock is to make sure
875 * that preemption never happens within any RCU read-side critical
876 * section whose outermost rcu_read_unlock() is called with one of
877 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in
878 * a number of ways, for example, by invoking preempt_disable() before
879 * critical section's outermost rcu_read_lock().
881 * Given that the set of locks acquired by rt_mutex_unlock() might change
882 * at any time, a somewhat more future-proofed approach is to make sure
883 * that that preemption never happens within any RCU read-side critical
884 * section whose outermost rcu_read_unlock() is called with irqs disabled.
885 * This approach relies on the fact that rt_mutex_unlock() currently only
886 * acquires irq-disabled locks.
888 * The second of these two approaches is best in most situations,
889 * however, the first approach can also be useful, at least to those
890 * developers willing to keep abreast of the set of locks acquired by
893 * See rcu_read_lock() for more information.
895 static inline void rcu_read_unlock(void)
897 RCU_LOCKDEP_WARN(!rcu_is_watching(),
898 "rcu_read_unlock() used illegally while idle");
901 rcu_lock_release(&rcu_lock_map
); /* Keep acq info for rls diags. */
905 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
907 * This is equivalent of rcu_read_lock(), but to be used when updates
908 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
909 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
910 * softirq handler to be a quiescent state, a process in RCU read-side
911 * critical section must be protected by disabling softirqs. Read-side
912 * critical sections in interrupt context can use just rcu_read_lock(),
913 * though this should at least be commented to avoid confusing people
916 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
917 * must occur in the same context, for example, it is illegal to invoke
918 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
919 * was invoked from some other task.
921 static inline void rcu_read_lock_bh(void)
925 rcu_lock_acquire(&rcu_bh_lock_map
);
926 RCU_LOCKDEP_WARN(!rcu_is_watching(),
927 "rcu_read_lock_bh() used illegally while idle");
931 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
933 * See rcu_read_lock_bh() for more information.
935 static inline void rcu_read_unlock_bh(void)
937 RCU_LOCKDEP_WARN(!rcu_is_watching(),
938 "rcu_read_unlock_bh() used illegally while idle");
939 rcu_lock_release(&rcu_bh_lock_map
);
945 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
947 * This is equivalent of rcu_read_lock(), but to be used when updates
948 * are being done using call_rcu_sched() or synchronize_rcu_sched().
949 * Read-side critical sections can also be introduced by anything that
950 * disables preemption, including local_irq_disable() and friends.
952 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
953 * must occur in the same context, for example, it is illegal to invoke
954 * rcu_read_unlock_sched() from process context if the matching
955 * rcu_read_lock_sched() was invoked from an NMI handler.
957 static inline void rcu_read_lock_sched(void)
960 __acquire(RCU_SCHED
);
961 rcu_lock_acquire(&rcu_sched_lock_map
);
962 RCU_LOCKDEP_WARN(!rcu_is_watching(),
963 "rcu_read_lock_sched() used illegally while idle");
966 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
967 static inline notrace
void rcu_read_lock_sched_notrace(void)
969 preempt_disable_notrace();
970 __acquire(RCU_SCHED
);
974 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
976 * See rcu_read_lock_sched for more information.
978 static inline void rcu_read_unlock_sched(void)
980 RCU_LOCKDEP_WARN(!rcu_is_watching(),
981 "rcu_read_unlock_sched() used illegally while idle");
982 rcu_lock_release(&rcu_sched_lock_map
);
983 __release(RCU_SCHED
);
987 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
988 static inline notrace
void rcu_read_unlock_sched_notrace(void)
990 __release(RCU_SCHED
);
991 preempt_enable_notrace();
995 * RCU_INIT_POINTER() - initialize an RCU protected pointer
997 * Initialize an RCU-protected pointer in special cases where readers
998 * do not need ordering constraints on the CPU or the compiler. These
1001 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
1002 * 2. The caller has taken whatever steps are required to prevent
1003 * RCU readers from concurrently accessing this pointer -or-
1004 * 3. The referenced data structure has already been exposed to
1005 * readers either at compile time or via rcu_assign_pointer() -and-
1006 * a. You have not made -any- reader-visible changes to
1007 * this structure since then -or-
1008 * b. It is OK for readers accessing this structure from its
1009 * new location to see the old state of the structure. (For
1010 * example, the changes were to statistical counters or to
1011 * other state where exact synchronization is not required.)
1013 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
1014 * result in impossible-to-diagnose memory corruption. As in the structures
1015 * will look OK in crash dumps, but any concurrent RCU readers might
1016 * see pre-initialized values of the referenced data structure. So
1017 * please be very careful how you use RCU_INIT_POINTER()!!!
1019 * If you are creating an RCU-protected linked structure that is accessed
1020 * by a single external-to-structure RCU-protected pointer, then you may
1021 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
1022 * pointers, but you must use rcu_assign_pointer() to initialize the
1023 * external-to-structure pointer -after- you have completely initialized
1024 * the reader-accessible portions of the linked structure.
1026 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
1027 * ordering guarantees for either the CPU or the compiler.
1029 #define RCU_INIT_POINTER(p, v) \
1031 rcu_dereference_sparse(p, __rcu); \
1032 WRITE_ONCE(p, RCU_INITIALIZER(v)); \
1036 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
1038 * GCC-style initialization for an RCU-protected pointer in a structure field.
1040 #define RCU_POINTER_INITIALIZER(p, v) \
1041 .p = RCU_INITIALIZER(v)
1044 * Does the specified offset indicate that the corresponding rcu_head
1045 * structure can be handled by kfree_rcu()?
1047 #define __is_kfree_rcu_offset(offset) ((offset) < 4096)
1050 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
1052 #define __kfree_rcu(head, offset) \
1054 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
1055 kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \
1059 * kfree_rcu() - kfree an object after a grace period.
1060 * @ptr: pointer to kfree
1061 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
1063 * Many rcu callbacks functions just call kfree() on the base structure.
1064 * These functions are trivial, but their size adds up, and furthermore
1065 * when they are used in a kernel module, that module must invoke the
1066 * high-latency rcu_barrier() function at module-unload time.
1068 * The kfree_rcu() function handles this issue. Rather than encoding a
1069 * function address in the embedded rcu_head structure, kfree_rcu() instead
1070 * encodes the offset of the rcu_head structure within the base structure.
1071 * Because the functions are not allowed in the low-order 4096 bytes of
1072 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
1073 * If the offset is larger than 4095 bytes, a compile-time error will
1074 * be generated in __kfree_rcu(). If this error is triggered, you can
1075 * either fall back to use of call_rcu() or rearrange the structure to
1076 * position the rcu_head structure into the first 4096 bytes.
1078 * Note that the allowable offset might decrease in the future, for example,
1079 * to allow something like kmem_cache_free_rcu().
1081 * The BUILD_BUG_ON check must not involve any function calls, hence the
1082 * checks are done in macros here.
1084 #define kfree_rcu(ptr, rcu_head) \
1085 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
1087 #ifdef CONFIG_TINY_RCU
1088 static inline int rcu_needs_cpu(u64 basemono
, u64
*nextevt
)
1090 *nextevt
= KTIME_MAX
;
1093 #endif /* #ifdef CONFIG_TINY_RCU */
1095 #if defined(CONFIG_RCU_NOCB_CPU_ALL)
1096 static inline bool rcu_is_nocb_cpu(int cpu
) { return true; }
1097 #elif defined(CONFIG_RCU_NOCB_CPU)
1098 bool rcu_is_nocb_cpu(int cpu
);
1100 static inline bool rcu_is_nocb_cpu(int cpu
) { return false; }
1104 /* Only for use by adaptive-ticks code. */
1105 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
1106 bool rcu_sys_is_idle(void);
1107 void rcu_sysidle_force_exit(void);
1108 #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1110 static inline bool rcu_sys_is_idle(void)
1115 static inline void rcu_sysidle_force_exit(void)
1119 #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1123 * Dump the ftrace buffer, but only one time per callsite per boot.
1125 #define rcu_ftrace_dump(oops_dump_mode) \
1127 static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
1129 if (!atomic_read(&___rfd_beenhere) && \
1130 !atomic_xchg(&___rfd_beenhere, 1)) \
1131 ftrace_dump(oops_dump_mode); \
1135 * Place this after a lock-acquisition primitive to guarantee that
1136 * an UNLOCK+LOCK pair acts as a full barrier. This guarantee applies
1137 * if the UNLOCK and LOCK are executed by the same CPU or if the
1138 * UNLOCK and LOCK operate on the same lock variable.
1140 #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE
1141 #define smp_mb__after_unlock_lock() smp_mb() /* Full ordering for lock. */
1142 #else /* #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
1143 #define smp_mb__after_unlock_lock() do { } while (0)
1144 #endif /* #else #ifdef CONFIG_ARCH_WEAK_RELEASE_ACQUIRE */
1147 #endif /* __LINUX_RCUPDATE_H */