2 * Read-Copy Update mechanism for mutual exclusion
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, you can access it online at
16 * http://www.gnu.org/licenses/gpl-2.0.html.
18 * Copyright IBM Corporation, 2001
20 * Author: Dipankar Sarma <dipankar@in.ibm.com>
22 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
23 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
25 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
26 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
28 * For detailed explanation of Read-Copy Update mechanism see -
29 * http://lse.sourceforge.net/locking/rcupdate.html
33 #ifndef __LINUX_RCUPDATE_H
34 #define __LINUX_RCUPDATE_H
36 #include <linux/types.h>
37 #include <linux/cache.h>
38 #include <linux/spinlock.h>
39 #include <linux/threads.h>
40 #include <linux/cpumask.h>
41 #include <linux/seqlock.h>
42 #include <linux/lockdep.h>
43 #include <linux/completion.h>
44 #include <linux/debugobjects.h>
45 #include <linux/bug.h>
46 #include <linux/compiler.h>
47 #include <linux/ktime.h>
48 #include <linux/irqflags.h>
50 #include <asm/barrier.h>
52 #ifndef CONFIG_TINY_RCU
53 extern int rcu_expedited; /* for sysctl */
54 extern int rcu_normal; /* also for sysctl */
55 #endif /* #ifndef CONFIG_TINY_RCU */
57 #ifdef CONFIG_TINY_RCU
58 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
59 static inline bool rcu_gp_is_normal(void) /* Internal RCU use. */
63 static inline bool rcu_gp_is_expedited(void) /* Internal RCU use. */
68 static inline void rcu_expedite_gp(void)
72 static inline void rcu_unexpedite_gp(void)
75 #else /* #ifdef CONFIG_TINY_RCU */
76 bool rcu_gp_is_normal(void); /* Internal RCU use. */
77 bool rcu_gp_is_expedited(void); /* Internal RCU use. */
78 void rcu_expedite_gp(void);
79 void rcu_unexpedite_gp(void);
80 #endif /* #else #ifdef CONFIG_TINY_RCU */
82 enum rcutorture_type {
91 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
92 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
93 unsigned long *gpnum, unsigned long *completed);
94 void rcutorture_record_test_transition(void);
95 void rcutorture_record_progress(unsigned long vernum);
96 void do_trace_rcu_torture_read(const char *rcutorturename,
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 #ifdef CONFIG_RCU_TRACE
118 void do_trace_rcu_torture_read(const char *rcutorturename,
119 struct rcu_head *rhp,
124 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
129 #define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b))
130 #define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b))
131 #define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
132 #define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
133 #define ulong2long(a) (*(long *)(&(a)))
135 /* Exported common interfaces */
137 #ifdef CONFIG_PREEMPT_RCU
140 * call_rcu() - Queue an RCU callback for invocation after a grace period.
141 * @head: structure to be used for queueing the RCU updates.
142 * @func: actual callback function to be invoked after the grace period
144 * The callback function will be invoked some time after a full grace
145 * period elapses, in other words after all pre-existing RCU read-side
146 * critical sections have completed. However, the callback function
147 * might well execute concurrently with RCU read-side critical sections
148 * that started after call_rcu() was invoked. RCU read-side critical
149 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
152 * Note that all CPUs must agree that the grace period extended beyond
153 * all pre-existing RCU read-side critical section. On systems with more
154 * than one CPU, this means that when "func()" is invoked, each CPU is
155 * guaranteed to have executed a full memory barrier since the end of its
156 * last RCU read-side critical section whose beginning preceded the call
157 * to call_rcu(). It also means that each CPU executing an RCU read-side
158 * critical section that continues beyond the start of "func()" must have
159 * executed a memory barrier after the call_rcu() but before the beginning
160 * of that RCU read-side critical section. Note that these guarantees
161 * include CPUs that are offline, idle, or executing in user mode, as
162 * well as CPUs that are executing in the kernel.
164 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
165 * resulting RCU callback function "func()", then both CPU A and CPU B are
166 * guaranteed to execute a full memory barrier during the time interval
167 * between the call to call_rcu() and the invocation of "func()" -- even
168 * if CPU A and CPU B are the same CPU (but again only if the system has
169 * more than one CPU).
171 void call_rcu(struct rcu_head *head,
172 rcu_callback_t func);
174 #else /* #ifdef CONFIG_PREEMPT_RCU */
176 /* In classic RCU, call_rcu() is just call_rcu_sched(). */
177 #define call_rcu call_rcu_sched
179 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
182 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
183 * @head: structure to be used for queueing the RCU updates.
184 * @func: actual callback function to be invoked after the grace period
186 * The callback function will be invoked some time after a full grace
187 * period elapses, in other words after all currently executing RCU
188 * read-side critical sections have completed. call_rcu_bh() assumes
189 * that the read-side critical sections end on completion of a softirq
190 * handler. This means that read-side critical sections in process
191 * context must not be interrupted by softirqs. This interface is to be
192 * used when most of the read-side critical sections are in softirq context.
193 * RCU read-side critical sections are delimited by :
194 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context.
196 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
197 * These may be nested.
199 * See the description of call_rcu() for more detailed information on
200 * memory ordering guarantees.
202 void call_rcu_bh(struct rcu_head *head,
203 rcu_callback_t func);
206 * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
207 * @head: structure to be used for queueing the RCU updates.
208 * @func: actual callback function to be invoked after the grace period
210 * The callback function will be invoked some time after a full grace
211 * period elapses, in other words after all currently executing RCU
212 * read-side critical sections have completed. call_rcu_sched() assumes
213 * that the read-side critical sections end on enabling of preemption
214 * or on voluntary preemption.
215 * RCU read-side critical sections are delimited by :
216 * - rcu_read_lock_sched() and rcu_read_unlock_sched(),
218 * anything that disables preemption.
219 * These may be nested.
221 * See the description of call_rcu() for more detailed information on
222 * memory ordering guarantees.
224 void call_rcu_sched(struct rcu_head *head,
225 rcu_callback_t func);
227 void synchronize_sched(void);
230 * Structure allowing asynchronous waiting on RCU.
232 struct rcu_synchronize {
233 struct rcu_head head;
234 struct completion completion;
236 void wakeme_after_rcu(struct rcu_head *head);
238 void __wait_rcu_gp(bool checktiny, int n, call_rcu_func_t *crcu_array,
239 struct rcu_synchronize *rs_array);
241 #define _wait_rcu_gp(checktiny, ...) \
243 call_rcu_func_t __crcu_array[] = { __VA_ARGS__ }; \
244 struct rcu_synchronize __rs_array[ARRAY_SIZE(__crcu_array)]; \
245 __wait_rcu_gp(checktiny, ARRAY_SIZE(__crcu_array), \
246 __crcu_array, __rs_array); \
249 #define wait_rcu_gp(...) _wait_rcu_gp(false, __VA_ARGS__)
252 * synchronize_rcu_mult - Wait concurrently for multiple grace periods
253 * @...: List of call_rcu() functions for the flavors to wait on.
255 * This macro waits concurrently for multiple flavors of RCU grace periods.
256 * For example, synchronize_rcu_mult(call_rcu, call_rcu_bh) would wait
257 * on concurrent RCU and RCU-bh grace periods. Waiting on a give SRCU
258 * domain requires you to write a wrapper function for that SRCU domain's
259 * call_srcu() function, supplying the corresponding srcu_struct.
261 * If Tiny RCU, tell _wait_rcu_gp() not to bother waiting for RCU
262 * or RCU-bh, given that anywhere synchronize_rcu_mult() can be called
263 * is automatically a grace period.
265 #define synchronize_rcu_mult(...) \
266 _wait_rcu_gp(IS_ENABLED(CONFIG_TINY_RCU), __VA_ARGS__)
269 * call_rcu_tasks() - Queue an RCU for invocation task-based grace period
270 * @head: structure to be used for queueing the RCU updates.
271 * @func: actual callback function to be invoked after the grace period
273 * The callback function will be invoked some time after a full grace
274 * period elapses, in other words after all currently executing RCU
275 * read-side critical sections have completed. call_rcu_tasks() assumes
276 * that the read-side critical sections end at a voluntary context
277 * switch (not a preemption!), entry into idle, or transition to usermode
278 * execution. As such, there are no read-side primitives analogous to
279 * rcu_read_lock() and rcu_read_unlock() because this primitive is intended
280 * to determine that all tasks have passed through a safe state, not so
281 * much for data-strcuture synchronization.
283 * See the description of call_rcu() for more detailed information on
284 * memory ordering guarantees.
286 void call_rcu_tasks(struct rcu_head *head, rcu_callback_t func);
287 void synchronize_rcu_tasks(void);
288 void rcu_barrier_tasks(void);
290 #ifdef CONFIG_PREEMPT_RCU
292 void __rcu_read_lock(void);
293 void __rcu_read_unlock(void);
294 void rcu_read_unlock_special(struct task_struct *t);
295 void synchronize_rcu(void);
298 * Defined as a macro as it is a very low level header included from
299 * areas that don't even know about current. This gives the rcu_read_lock()
300 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
301 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
303 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
305 #else /* #ifdef CONFIG_PREEMPT_RCU */
307 static inline void __rcu_read_lock(void)
312 static inline void __rcu_read_unlock(void)
317 static inline void synchronize_rcu(void)
322 static inline int rcu_preempt_depth(void)
327 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
329 /* Internal to kernel */
331 void rcu_sched_qs(void);
332 void rcu_bh_qs(void);
333 void rcu_check_callbacks(int user);
334 void rcu_report_dead(unsigned int cpu);
335 void rcu_cpu_starting(unsigned int cpu);
337 #ifndef CONFIG_TINY_RCU
338 void rcu_end_inkernel_boot(void);
339 #else /* #ifndef CONFIG_TINY_RCU */
340 static inline void rcu_end_inkernel_boot(void) { }
341 #endif /* #ifndef CONFIG_TINY_RCU */
343 #ifdef CONFIG_RCU_STALL_COMMON
344 void rcu_sysrq_start(void);
345 void rcu_sysrq_end(void);
346 #else /* #ifdef CONFIG_RCU_STALL_COMMON */
347 static inline void rcu_sysrq_start(void)
350 static inline void rcu_sysrq_end(void)
353 #endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
355 #ifdef CONFIG_NO_HZ_FULL
356 void rcu_user_enter(void);
357 void rcu_user_exit(void);
359 static inline void rcu_user_enter(void) { }
360 static inline void rcu_user_exit(void) { }
361 #endif /* CONFIG_NO_HZ_FULL */
363 #ifdef CONFIG_RCU_NOCB_CPU
364 void rcu_init_nohz(void);
365 #else /* #ifdef CONFIG_RCU_NOCB_CPU */
366 static inline void rcu_init_nohz(void)
369 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
372 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
373 * @a: Code that RCU needs to pay attention to.
375 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
376 * in the inner idle loop, that is, between the rcu_idle_enter() and
377 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
378 * critical sections. However, things like powertop need tracepoints
379 * in the inner idle loop.
381 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
382 * will tell RCU that it needs to pay attention, invoke its argument
383 * (in this example, calling the do_something_with_RCU() function),
384 * and then tell RCU to go back to ignoring this CPU. It is permissible
385 * to nest RCU_NONIDLE() wrappers, but not indefinitely (but the limit is
386 * on the order of a million or so, even on 32-bit systems). It is
387 * not legal to block within RCU_NONIDLE(), nor is it permissible to
388 * transfer control either into or out of RCU_NONIDLE()'s statement.
390 #define RCU_NONIDLE(a) \
392 rcu_irq_enter_irqson(); \
393 do { a; } while (0); \
394 rcu_irq_exit_irqson(); \
398 * Note a voluntary context switch for RCU-tasks benefit. This is a
399 * macro rather than an inline function to avoid #include hell.
401 #ifdef CONFIG_TASKS_RCU
402 #define TASKS_RCU(x) x
403 extern struct srcu_struct tasks_rcu_exit_srcu;
404 #define rcu_note_voluntary_context_switch(t) \
407 if (READ_ONCE((t)->rcu_tasks_holdout)) \
408 WRITE_ONCE((t)->rcu_tasks_holdout, false); \
410 #else /* #ifdef CONFIG_TASKS_RCU */
411 #define TASKS_RCU(x) do { } while (0)
412 #define rcu_note_voluntary_context_switch(t) rcu_all_qs()
413 #endif /* #else #ifdef CONFIG_TASKS_RCU */
416 * cond_resched_rcu_qs - Report potential quiescent states to RCU
418 * This macro resembles cond_resched(), except that it is defined to
419 * report potential quiescent states to RCU-tasks even if the cond_resched()
420 * machinery were to be shut off, as some advocate for PREEMPT kernels.
422 #define cond_resched_rcu_qs() \
424 if (!cond_resched()) \
425 rcu_note_voluntary_context_switch(current); \
428 #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP)
429 bool __rcu_is_watching(void);
430 #endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */
433 * Infrastructure to implement the synchronize_() primitives in
434 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
437 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
438 #include <linux/rcutree.h>
439 #elif defined(CONFIG_TINY_RCU)
440 #include <linux/rcutiny.h>
442 #error "Unknown RCU implementation specified to kernel configuration"
445 #define RCU_SCHEDULER_INACTIVE 0
446 #define RCU_SCHEDULER_INIT 1
447 #define RCU_SCHEDULER_RUNNING 2
450 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
451 * initialization and destruction of rcu_head on the stack. rcu_head structures
452 * allocated dynamically in the heap or defined statically don't need any
455 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
456 void init_rcu_head(struct rcu_head *head);
457 void destroy_rcu_head(struct rcu_head *head);
458 void init_rcu_head_on_stack(struct rcu_head *head);
459 void destroy_rcu_head_on_stack(struct rcu_head *head);
460 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
461 static inline void init_rcu_head(struct rcu_head *head)
465 static inline void destroy_rcu_head(struct rcu_head *head)
469 static inline void init_rcu_head_on_stack(struct rcu_head *head)
473 static inline void destroy_rcu_head_on_stack(struct rcu_head *head)
476 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
478 #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
479 bool rcu_lockdep_current_cpu_online(void);
480 #else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
481 static inline bool rcu_lockdep_current_cpu_online(void)
485 #endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
487 #ifdef CONFIG_DEBUG_LOCK_ALLOC
489 static inline void rcu_lock_acquire(struct lockdep_map *map)
491 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
494 static inline void rcu_lock_release(struct lockdep_map *map)
496 lock_release(map, 1, _THIS_IP_);
499 extern struct lockdep_map rcu_lock_map;
500 extern struct lockdep_map rcu_bh_lock_map;
501 extern struct lockdep_map rcu_sched_lock_map;
502 extern struct lockdep_map rcu_callback_map;
503 int debug_lockdep_rcu_enabled(void);
505 int rcu_read_lock_held(void);
506 int rcu_read_lock_bh_held(void);
509 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
511 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
512 * RCU-sched read-side critical section. In absence of
513 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
514 * critical section unless it can prove otherwise.
516 int rcu_read_lock_sched_held(void);
518 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
520 # define rcu_lock_acquire(a) do { } while (0)
521 # define rcu_lock_release(a) do { } while (0)
523 static inline int rcu_read_lock_held(void)
528 static inline int rcu_read_lock_bh_held(void)
533 static inline int rcu_read_lock_sched_held(void)
535 return !preemptible();
537 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
539 #ifdef CONFIG_PROVE_RCU
542 * RCU_LOCKDEP_WARN - emit lockdep splat if specified condition is met
543 * @c: condition to check
544 * @s: informative message
546 #define RCU_LOCKDEP_WARN(c, s) \
548 static bool __section(.data.unlikely) __warned; \
549 if (debug_lockdep_rcu_enabled() && !__warned && (c)) { \
551 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
555 #if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
556 static inline void rcu_preempt_sleep_check(void)
558 RCU_LOCKDEP_WARN(lock_is_held(&rcu_lock_map),
559 "Illegal context switch in RCU read-side critical section");
561 #else /* #ifdef CONFIG_PROVE_RCU */
562 static inline void rcu_preempt_sleep_check(void)
565 #endif /* #else #ifdef CONFIG_PROVE_RCU */
567 #define rcu_sleep_check() \
569 rcu_preempt_sleep_check(); \
570 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map), \
571 "Illegal context switch in RCU-bh read-side critical section"); \
572 RCU_LOCKDEP_WARN(lock_is_held(&rcu_sched_lock_map), \
573 "Illegal context switch in RCU-sched read-side critical section"); \
576 #else /* #ifdef CONFIG_PROVE_RCU */
578 #define RCU_LOCKDEP_WARN(c, s) do { } while (0)
579 #define rcu_sleep_check() do { } while (0)
581 #endif /* #else #ifdef CONFIG_PROVE_RCU */
584 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
585 * and rcu_assign_pointer(). Some of these could be folded into their
586 * callers, but they are left separate in order to ease introduction of
587 * multiple flavors of pointers to match the multiple flavors of RCU
588 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
593 #define rcu_dereference_sparse(p, space) \
594 ((void)(((typeof(*p) space *)p) == p))
595 #else /* #ifdef __CHECKER__ */
596 #define rcu_dereference_sparse(p, space)
597 #endif /* #else #ifdef __CHECKER__ */
599 #define __rcu_access_pointer(p, space) \
601 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \
602 rcu_dereference_sparse(p, space); \
603 ((typeof(*p) __force __kernel *)(_________p1)); \
605 #define __rcu_dereference_check(p, c, space) \
607 /* Dependency order vs. p above. */ \
608 typeof(*p) *________p1 = (typeof(*p) *__force)lockless_dereference(p); \
609 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_check() usage"); \
610 rcu_dereference_sparse(p, space); \
611 ((typeof(*p) __force __kernel *)(________p1)); \
613 #define __rcu_dereference_protected(p, c, space) \
615 RCU_LOCKDEP_WARN(!(c), "suspicious rcu_dereference_protected() usage"); \
616 rcu_dereference_sparse(p, space); \
617 ((typeof(*p) __force __kernel *)(p)); \
619 #define rcu_dereference_raw(p) \
621 /* Dependency order vs. p above. */ \
622 typeof(p) ________p1 = lockless_dereference(p); \
623 ((typeof(*p) __force __kernel *)(________p1)); \
627 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
628 * @v: The value to statically initialize with.
630 #define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
633 * rcu_assign_pointer() - assign to RCU-protected pointer
634 * @p: pointer to assign to
635 * @v: value to assign (publish)
637 * Assigns the specified value to the specified RCU-protected
638 * pointer, ensuring that any concurrent RCU readers will see
639 * any prior initialization.
641 * Inserts memory barriers on architectures that require them
642 * (which is most of them), and also prevents the compiler from
643 * reordering the code that initializes the structure after the pointer
644 * assignment. More importantly, this call documents which pointers
645 * will be dereferenced by RCU read-side code.
647 * In some special cases, you may use RCU_INIT_POINTER() instead
648 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
649 * to the fact that it does not constrain either the CPU or the compiler.
650 * That said, using RCU_INIT_POINTER() when you should have used
651 * rcu_assign_pointer() is a very bad thing that results in
652 * impossible-to-diagnose memory corruption. So please be careful.
653 * See the RCU_INIT_POINTER() comment header for details.
655 * Note that rcu_assign_pointer() evaluates each of its arguments only
656 * once, appearances notwithstanding. One of the "extra" evaluations
657 * is in typeof() and the other visible only to sparse (__CHECKER__),
658 * neither of which actually execute the argument. As with most cpp
659 * macros, this execute-arguments-only-once property is important, so
660 * please be careful when making changes to rcu_assign_pointer() and the
661 * other macros that it invokes.
663 #define rcu_assign_pointer(p, v) \
665 uintptr_t _r_a_p__v = (uintptr_t)(v); \
667 if (__builtin_constant_p(v) && (_r_a_p__v) == (uintptr_t)NULL) \
668 WRITE_ONCE((p), (typeof(p))(_r_a_p__v)); \
670 smp_store_release(&p, RCU_INITIALIZER((typeof(p))_r_a_p__v)); \
675 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
676 * @p: The pointer to read
678 * Return the value of the specified RCU-protected pointer, but omit the
679 * smp_read_barrier_depends() and keep the READ_ONCE(). This is useful
680 * when the value of this pointer is accessed, but the pointer is not
681 * dereferenced, for example, when testing an RCU-protected pointer against
682 * NULL. Although rcu_access_pointer() may also be used in cases where
683 * update-side locks prevent the value of the pointer from changing, you
684 * should instead use rcu_dereference_protected() for this use case.
686 * It is also permissible to use rcu_access_pointer() when read-side
687 * access to the pointer was removed at least one grace period ago, as
688 * is the case in the context of the RCU callback that is freeing up
689 * the data, or after a synchronize_rcu() returns. This can be useful
690 * when tearing down multi-linked structures after a grace period
693 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
696 * rcu_dereference_check() - rcu_dereference with debug checking
697 * @p: The pointer to read, prior to dereferencing
698 * @c: The conditions under which the dereference will take place
700 * Do an rcu_dereference(), but check that the conditions under which the
701 * dereference will take place are correct. Typically the conditions
702 * indicate the various locking conditions that should be held at that
703 * point. The check should return true if the conditions are satisfied.
704 * An implicit check for being in an RCU read-side critical section
705 * (rcu_read_lock()) is included.
709 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
711 * could be used to indicate to lockdep that foo->bar may only be dereferenced
712 * if either rcu_read_lock() is held, or that the lock required to replace
713 * the bar struct at foo->bar is held.
715 * Note that the list of conditions may also include indications of when a lock
716 * need not be held, for example during initialisation or destruction of the
719 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
720 * atomic_read(&foo->usage) == 0);
722 * Inserts memory barriers on architectures that require them
723 * (currently only the Alpha), prevents the compiler from refetching
724 * (and from merging fetches), and, more importantly, documents exactly
725 * which pointers are protected by RCU and checks that the pointer is
726 * annotated as __rcu.
728 #define rcu_dereference_check(p, c) \
729 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu)
732 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
733 * @p: The pointer to read, prior to dereferencing
734 * @c: The conditions under which the dereference will take place
736 * This is the RCU-bh counterpart to rcu_dereference_check().
738 #define rcu_dereference_bh_check(p, c) \
739 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu)
742 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
743 * @p: The pointer to read, prior to dereferencing
744 * @c: The conditions under which the dereference will take place
746 * This is the RCU-sched counterpart to rcu_dereference_check().
748 #define rcu_dereference_sched_check(p, c) \
749 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \
753 * The tracing infrastructure traces RCU (we want that), but unfortunately
754 * some of the RCU checks causes tracing to lock up the system.
756 * The no-tracing version of rcu_dereference_raw() must not call
757 * rcu_read_lock_held().
759 #define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
762 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
763 * @p: The pointer to read, prior to dereferencing
764 * @c: The conditions under which the dereference will take place
766 * Return the value of the specified RCU-protected pointer, but omit
767 * both the smp_read_barrier_depends() and the READ_ONCE(). This
768 * is useful in cases where update-side locks prevent the value of the
769 * pointer from changing. Please note that this primitive does -not-
770 * prevent the compiler from repeating this reference or combining it
771 * with other references, so it should not be used without protection
772 * of appropriate locks.
774 * This function is only for update-side use. Using this function
775 * when protected only by rcu_read_lock() will result in infrequent
776 * but very ugly failures.
778 #define rcu_dereference_protected(p, c) \
779 __rcu_dereference_protected((p), (c), __rcu)
783 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
784 * @p: The pointer to read, prior to dereferencing
786 * This is a simple wrapper around rcu_dereference_check().
788 #define rcu_dereference(p) rcu_dereference_check(p, 0)
791 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
792 * @p: The pointer to read, prior to dereferencing
794 * Makes rcu_dereference_check() do the dirty work.
796 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
799 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
800 * @p: The pointer to read, prior to dereferencing
802 * Makes rcu_dereference_check() do the dirty work.
804 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
807 * rcu_pointer_handoff() - Hand off a pointer from RCU to other mechanism
808 * @p: The pointer to hand off
810 * This is simply an identity function, but it documents where a pointer
811 * is handed off from RCU to some other synchronization mechanism, for
812 * example, reference counting or locking. In C11, it would map to
813 * kill_dependency(). It could be used as follows:
816 * p = rcu_dereference(gp);
817 * long_lived = is_long_lived(p);
819 * if (!atomic_inc_not_zero(p->refcnt))
820 * long_lived = false;
822 * p = rcu_pointer_handoff(p);
826 #define rcu_pointer_handoff(p) (p)
829 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
831 * When synchronize_rcu() is invoked on one CPU while other CPUs
832 * are within RCU read-side critical sections, then the
833 * synchronize_rcu() is guaranteed to block until after all the other
834 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
835 * on one CPU while other CPUs are within RCU read-side critical
836 * sections, invocation of the corresponding RCU callback is deferred
837 * until after the all the other CPUs exit their critical sections.
839 * Note, however, that RCU callbacks are permitted to run concurrently
840 * with new RCU read-side critical sections. One way that this can happen
841 * is via the following sequence of events: (1) CPU 0 enters an RCU
842 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
843 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
844 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
845 * callback is invoked. This is legal, because the RCU read-side critical
846 * section that was running concurrently with the call_rcu() (and which
847 * therefore might be referencing something that the corresponding RCU
848 * callback would free up) has completed before the corresponding
849 * RCU callback is invoked.
851 * RCU read-side critical sections may be nested. Any deferred actions
852 * will be deferred until the outermost RCU read-side critical section
855 * You can avoid reading and understanding the next paragraph by
856 * following this rule: don't put anything in an rcu_read_lock() RCU
857 * read-side critical section that would block in a !PREEMPT kernel.
858 * But if you want the full story, read on!
860 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU),
861 * it is illegal to block while in an RCU read-side critical section.
862 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT
863 * kernel builds, RCU read-side critical sections may be preempted,
864 * but explicit blocking is illegal. Finally, in preemptible RCU
865 * implementations in real-time (with -rt patchset) kernel builds, RCU
866 * read-side critical sections may be preempted and they may also block, but
867 * only when acquiring spinlocks that are subject to priority inheritance.
869 static __always_inline void rcu_read_lock(void)
873 rcu_lock_acquire(&rcu_lock_map);
874 RCU_LOCKDEP_WARN(!rcu_is_watching(),
875 "rcu_read_lock() used illegally while idle");
879 * So where is rcu_write_lock()? It does not exist, as there is no
880 * way for writers to lock out RCU readers. This is a feature, not
881 * a bug -- this property is what provides RCU's performance benefits.
882 * Of course, writers must coordinate with each other. The normal
883 * spinlock primitives work well for this, but any other technique may be
884 * used as well. RCU does not care how the writers keep out of each
885 * others' way, as long as they do so.
889 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
891 * In most situations, rcu_read_unlock() is immune from deadlock.
892 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock()
893 * is responsible for deboosting, which it does via rt_mutex_unlock().
894 * Unfortunately, this function acquires the scheduler's runqueue and
895 * priority-inheritance spinlocks. This means that deadlock could result
896 * if the caller of rcu_read_unlock() already holds one of these locks or
897 * any lock that is ever acquired while holding them; or any lock which
898 * can be taken from interrupt context because rcu_boost()->rt_mutex_lock()
899 * does not disable irqs while taking ->wait_lock.
901 * That said, RCU readers are never priority boosted unless they were
902 * preempted. Therefore, one way to avoid deadlock is to make sure
903 * that preemption never happens within any RCU read-side critical
904 * section whose outermost rcu_read_unlock() is called with one of
905 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in
906 * a number of ways, for example, by invoking preempt_disable() before
907 * critical section's outermost rcu_read_lock().
909 * Given that the set of locks acquired by rt_mutex_unlock() might change
910 * at any time, a somewhat more future-proofed approach is to make sure
911 * that that preemption never happens within any RCU read-side critical
912 * section whose outermost rcu_read_unlock() is called with irqs disabled.
913 * This approach relies on the fact that rt_mutex_unlock() currently only
914 * acquires irq-disabled locks.
916 * The second of these two approaches is best in most situations,
917 * however, the first approach can also be useful, at least to those
918 * developers willing to keep abreast of the set of locks acquired by
921 * See rcu_read_lock() for more information.
923 static inline void rcu_read_unlock(void)
925 RCU_LOCKDEP_WARN(!rcu_is_watching(),
926 "rcu_read_unlock() used illegally while idle");
929 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
933 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
935 * This is equivalent of rcu_read_lock(), but to be used when updates
936 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
937 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
938 * softirq handler to be a quiescent state, a process in RCU read-side
939 * critical section must be protected by disabling softirqs. Read-side
940 * critical sections in interrupt context can use just rcu_read_lock(),
941 * though this should at least be commented to avoid confusing people
944 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
945 * must occur in the same context, for example, it is illegal to invoke
946 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
947 * was invoked from some other task.
949 static inline void rcu_read_lock_bh(void)
953 rcu_lock_acquire(&rcu_bh_lock_map);
954 RCU_LOCKDEP_WARN(!rcu_is_watching(),
955 "rcu_read_lock_bh() used illegally while idle");
959 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
961 * See rcu_read_lock_bh() for more information.
963 static inline void rcu_read_unlock_bh(void)
965 RCU_LOCKDEP_WARN(!rcu_is_watching(),
966 "rcu_read_unlock_bh() used illegally while idle");
967 rcu_lock_release(&rcu_bh_lock_map);
973 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
975 * This is equivalent of rcu_read_lock(), but to be used when updates
976 * are being done using call_rcu_sched() or synchronize_rcu_sched().
977 * Read-side critical sections can also be introduced by anything that
978 * disables preemption, including local_irq_disable() and friends.
980 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
981 * must occur in the same context, for example, it is illegal to invoke
982 * rcu_read_unlock_sched() from process context if the matching
983 * rcu_read_lock_sched() was invoked from an NMI handler.
985 static inline void rcu_read_lock_sched(void)
988 __acquire(RCU_SCHED);
989 rcu_lock_acquire(&rcu_sched_lock_map);
990 RCU_LOCKDEP_WARN(!rcu_is_watching(),
991 "rcu_read_lock_sched() used illegally while idle");
994 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
995 static inline notrace void rcu_read_lock_sched_notrace(void)
997 preempt_disable_notrace();
998 __acquire(RCU_SCHED);
1002 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
1004 * See rcu_read_lock_sched for more information.
1006 static inline void rcu_read_unlock_sched(void)
1008 RCU_LOCKDEP_WARN(!rcu_is_watching(),
1009 "rcu_read_unlock_sched() used illegally while idle");
1010 rcu_lock_release(&rcu_sched_lock_map);
1011 __release(RCU_SCHED);
1015 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
1016 static inline notrace void rcu_read_unlock_sched_notrace(void)
1018 __release(RCU_SCHED);
1019 preempt_enable_notrace();
1023 * RCU_INIT_POINTER() - initialize an RCU protected pointer
1025 * Initialize an RCU-protected pointer in special cases where readers
1026 * do not need ordering constraints on the CPU or the compiler. These
1027 * special cases are:
1029 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
1030 * 2. The caller has taken whatever steps are required to prevent
1031 * RCU readers from concurrently accessing this pointer -or-
1032 * 3. The referenced data structure has already been exposed to
1033 * readers either at compile time or via rcu_assign_pointer() -and-
1034 * a. You have not made -any- reader-visible changes to
1035 * this structure since then -or-
1036 * b. It is OK for readers accessing this structure from its
1037 * new location to see the old state of the structure. (For
1038 * example, the changes were to statistical counters or to
1039 * other state where exact synchronization is not required.)
1041 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
1042 * result in impossible-to-diagnose memory corruption. As in the structures
1043 * will look OK in crash dumps, but any concurrent RCU readers might
1044 * see pre-initialized values of the referenced data structure. So
1045 * please be very careful how you use RCU_INIT_POINTER()!!!
1047 * If you are creating an RCU-protected linked structure that is accessed
1048 * by a single external-to-structure RCU-protected pointer, then you may
1049 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
1050 * pointers, but you must use rcu_assign_pointer() to initialize the
1051 * external-to-structure pointer -after- you have completely initialized
1052 * the reader-accessible portions of the linked structure.
1054 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
1055 * ordering guarantees for either the CPU or the compiler.
1057 #define RCU_INIT_POINTER(p, v) \
1059 rcu_dereference_sparse(p, __rcu); \
1060 WRITE_ONCE(p, RCU_INITIALIZER(v)); \
1064 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
1066 * GCC-style initialization for an RCU-protected pointer in a structure field.
1068 #define RCU_POINTER_INITIALIZER(p, v) \
1069 .p = RCU_INITIALIZER(v)
1072 * Does the specified offset indicate that the corresponding rcu_head
1073 * structure can be handled by kfree_rcu()?
1075 #define __is_kfree_rcu_offset(offset) ((offset) < 4096)
1078 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
1080 #define __kfree_rcu(head, offset) \
1082 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
1083 kfree_call_rcu(head, (rcu_callback_t)(unsigned long)(offset)); \
1087 * kfree_rcu() - kfree an object after a grace period.
1088 * @ptr: pointer to kfree
1089 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
1091 * Many rcu callbacks functions just call kfree() on the base structure.
1092 * These functions are trivial, but their size adds up, and furthermore
1093 * when they are used in a kernel module, that module must invoke the
1094 * high-latency rcu_barrier() function at module-unload time.
1096 * The kfree_rcu() function handles this issue. Rather than encoding a
1097 * function address in the embedded rcu_head structure, kfree_rcu() instead
1098 * encodes the offset of the rcu_head structure within the base structure.
1099 * Because the functions are not allowed in the low-order 4096 bytes of
1100 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
1101 * If the offset is larger than 4095 bytes, a compile-time error will
1102 * be generated in __kfree_rcu(). If this error is triggered, you can
1103 * either fall back to use of call_rcu() or rearrange the structure to
1104 * position the rcu_head structure into the first 4096 bytes.
1106 * Note that the allowable offset might decrease in the future, for example,
1107 * to allow something like kmem_cache_free_rcu().
1109 * The BUILD_BUG_ON check must not involve any function calls, hence the
1110 * checks are done in macros here.
1112 #define kfree_rcu(ptr, rcu_head) \
1113 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
1115 #ifdef CONFIG_TINY_RCU
1116 static inline int rcu_needs_cpu(u64 basemono, u64 *nextevt)
1118 *nextevt = KTIME_MAX;
1121 #endif /* #ifdef CONFIG_TINY_RCU */
1123 #if defined(CONFIG_RCU_NOCB_CPU_ALL)
1124 static inline bool rcu_is_nocb_cpu(int cpu) { return true; }
1125 #elif defined(CONFIG_RCU_NOCB_CPU)
1126 bool rcu_is_nocb_cpu(int cpu);
1128 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
1132 /* Only for use by adaptive-ticks code. */
1133 #ifdef CONFIG_NO_HZ_FULL_SYSIDLE
1134 bool rcu_sys_is_idle(void);
1135 void rcu_sysidle_force_exit(void);
1136 #else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1138 static inline bool rcu_sys_is_idle(void)
1143 static inline void rcu_sysidle_force_exit(void)
1147 #endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1151 * Dump the ftrace buffer, but only one time per callsite per boot.
1153 #define rcu_ftrace_dump(oops_dump_mode) \
1155 static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
1157 if (!atomic_read(&___rfd_beenhere) && \
1158 !atomic_xchg(&___rfd_beenhere, 1)) \
1159 ftrace_dump(oops_dump_mode); \
1163 #endif /* __LINUX_RCUPDATE_H */