blob: 72bf3a01a4ee67ac8908212c3de3897d383a2161 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Paul E. McKenneya71fca52009-09-18 10:28:19 -07002 * Read-Copy Update mechanism for mutual exclusion
Linus Torvalds1da177e2005-04-16 15:20:36 -07003 *
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.
8 *
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.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
Paul E. McKenney01c1c662008-01-25 21:08:24 +010018 * Copyright IBM Corporation, 2001
Linus Torvalds1da177e2005-04-16 15:20:36 -070019 *
20 * Author: Dipankar Sarma <dipankar@in.ibm.com>
Paul E. McKenneya71fca52009-09-18 10:28:19 -070021 *
Josh Triplett595182b2006-10-04 02:17:21 -070022 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
24 * Papers:
25 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
26 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
27 *
28 * For detailed explanation of Read-Copy Update mechanism see -
Paul E. McKenneya71fca52009-09-18 10:28:19 -070029 * http://lse.sourceforge.net/locking/rcupdate.html
Linus Torvalds1da177e2005-04-16 15:20:36 -070030 *
31 */
32
33#ifndef __LINUX_RCUPDATE_H
34#define __LINUX_RCUPDATE_H
35
Paul E. McKenney99098752011-05-31 21:03:55 -070036#include <linux/types.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/cache.h>
38#include <linux/spinlock.h>
39#include <linux/threads.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <linux/cpumask.h>
41#include <linux/seqlock.h>
Peter Zijlstra851a67b2007-10-11 22:11:12 +020042#include <linux/lockdep.h>
Paul E. McKenney4446a362008-05-12 21:21:05 +020043#include <linux/completion.h>
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -040044#include <linux/debugobjects.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050045#include <linux/bug.h>
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -070046#include <linux/compiler.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
Dave Younge5ab6772010-03-10 15:24:05 -080048#ifdef CONFIG_RCU_TORTURE_TEST
49extern int rcutorture_runnable; /* for sysctl */
50#endif /* #ifdef CONFIG_RCU_TORTURE_TEST */
51
Paul E. McKenney4a298652011-04-03 21:33:51 -070052#if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
Teodora Baluta584dc4c2013-11-11 17:11:23 +020053void rcutorture_record_test_transition(void);
54void rcutorture_record_progress(unsigned long vernum);
55void do_trace_rcu_torture_read(const char *rcutorturename,
56 struct rcu_head *rhp,
57 unsigned long secs,
58 unsigned long c_old,
59 unsigned long c);
Paul E. McKenney4a298652011-04-03 21:33:51 -070060#else
61static inline void rcutorture_record_test_transition(void)
62{
63}
64static inline void rcutorture_record_progress(unsigned long vernum)
65{
66}
Paul E. McKenney91afaf32011-10-02 07:44:32 -070067#ifdef CONFIG_RCU_TRACE
Teodora Baluta584dc4c2013-11-11 17:11:23 +020068void do_trace_rcu_torture_read(const char *rcutorturename,
69 struct rcu_head *rhp,
70 unsigned long secs,
71 unsigned long c_old,
72 unsigned long c);
Paul E. McKenney91afaf32011-10-02 07:44:32 -070073#else
Paul E. McKenney52494532012-11-14 16:26:40 -080074#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
75 do { } while (0)
Paul E. McKenney91afaf32011-10-02 07:44:32 -070076#endif
Paul E. McKenney4a298652011-04-03 21:33:51 -070077#endif
78
Tejun Heoe27fc962010-11-22 21:36:11 -080079#define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b))
80#define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b))
Paul E. McKenneya3dc3fb2010-08-13 16:16:25 -070081#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
82#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
Paul E. McKenneyc0f4dfd2012-12-28 11:30:36 -080083#define ulong2long(a) (*(long *)(&(a)))
Paul E. McKenneya3dc3fb2010-08-13 16:16:25 -070084
Paul E. McKenney03b042b2009-06-25 09:08:16 -070085/* Exported common interfaces */
Paul E. McKenney2c428182011-05-26 22:14:36 -070086
87#ifdef CONFIG_PREEMPT_RCU
88
89/**
90 * call_rcu() - Queue an RCU callback for invocation after a grace period.
91 * @head: structure to be used for queueing the RCU updates.
92 * @func: actual callback function to be invoked after the grace period
93 *
94 * The callback function will be invoked some time after a full grace
95 * period elapses, in other words after all pre-existing RCU read-side
96 * critical sections have completed. However, the callback function
97 * might well execute concurrently with RCU read-side critical sections
98 * that started after call_rcu() was invoked. RCU read-side critical
99 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
100 * and may be nested.
Paul E. McKenneyf0a0e6f2012-10-23 13:47:01 -0700101 *
102 * Note that all CPUs must agree that the grace period extended beyond
103 * all pre-existing RCU read-side critical section. On systems with more
104 * than one CPU, this means that when "func()" is invoked, each CPU is
105 * guaranteed to have executed a full memory barrier since the end of its
106 * last RCU read-side critical section whose beginning preceded the call
107 * to call_rcu(). It also means that each CPU executing an RCU read-side
108 * critical section that continues beyond the start of "func()" must have
109 * executed a memory barrier after the call_rcu() but before the beginning
110 * of that RCU read-side critical section. Note that these guarantees
111 * include CPUs that are offline, idle, or executing in user mode, as
112 * well as CPUs that are executing in the kernel.
113 *
114 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
115 * resulting RCU callback function "func()", then both CPU A and CPU B are
116 * guaranteed to execute a full memory barrier during the time interval
117 * between the call to call_rcu() and the invocation of "func()" -- even
118 * if CPU A and CPU B are the same CPU (but again only if the system has
119 * more than one CPU).
Paul E. McKenney2c428182011-05-26 22:14:36 -0700120 */
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200121void call_rcu(struct rcu_head *head,
122 void (*func)(struct rcu_head *head));
Paul E. McKenney2c428182011-05-26 22:14:36 -0700123
124#else /* #ifdef CONFIG_PREEMPT_RCU */
125
126/* In classic RCU, call_rcu() is just call_rcu_sched(). */
127#define call_rcu call_rcu_sched
128
129#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
130
131/**
132 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
133 * @head: structure to be used for queueing the RCU updates.
134 * @func: actual callback function to be invoked after the grace period
135 *
136 * The callback function will be invoked some time after a full grace
137 * period elapses, in other words after all currently executing RCU
138 * read-side critical sections have completed. call_rcu_bh() assumes
139 * that the read-side critical sections end on completion of a softirq
140 * handler. This means that read-side critical sections in process
141 * context must not be interrupted by softirqs. This interface is to be
142 * used when most of the read-side critical sections are in softirq context.
143 * RCU read-side critical sections are delimited by :
144 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context.
145 * OR
146 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
147 * These may be nested.
Paul E. McKenneyf0a0e6f2012-10-23 13:47:01 -0700148 *
149 * See the description of call_rcu() for more detailed information on
150 * memory ordering guarantees.
Paul E. McKenney2c428182011-05-26 22:14:36 -0700151 */
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200152void call_rcu_bh(struct rcu_head *head,
153 void (*func)(struct rcu_head *head));
Paul E. McKenney2c428182011-05-26 22:14:36 -0700154
155/**
156 * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
157 * @head: structure to be used for queueing the RCU updates.
158 * @func: actual callback function to be invoked after the grace period
159 *
160 * The callback function will be invoked some time after a full grace
161 * period elapses, in other words after all currently executing RCU
162 * read-side critical sections have completed. call_rcu_sched() assumes
163 * that the read-side critical sections end on enabling of preemption
164 * or on voluntary preemption.
165 * RCU read-side critical sections are delimited by :
166 * - rcu_read_lock_sched() and rcu_read_unlock_sched(),
167 * OR
168 * anything that disables preemption.
169 * These may be nested.
Paul E. McKenneyf0a0e6f2012-10-23 13:47:01 -0700170 *
171 * See the description of call_rcu() for more detailed information on
172 * memory ordering guarantees.
Paul E. McKenney2c428182011-05-26 22:14:36 -0700173 */
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200174void call_rcu_sched(struct rcu_head *head,
175 void (*func)(struct rcu_head *rcu));
Paul E. McKenney2c428182011-05-26 22:14:36 -0700176
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200177void synchronize_sched(void);
Paul E. McKenney03b042b2009-06-25 09:08:16 -0700178
Paul E. McKenneya3dc3fb2010-08-13 16:16:25 -0700179#ifdef CONFIG_PREEMPT_RCU
180
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200181void __rcu_read_lock(void);
182void __rcu_read_unlock(void);
183void rcu_read_unlock_special(struct task_struct *t);
Paul E. McKenney7b0b7592010-08-17 14:18:46 -0700184void synchronize_rcu(void);
185
Paul E. McKenneya3dc3fb2010-08-13 16:16:25 -0700186/*
187 * Defined as a macro as it is a very low level header included from
188 * areas that don't even know about current. This gives the rcu_read_lock()
189 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
190 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
191 */
192#define rcu_preempt_depth() (current->rcu_read_lock_nesting)
193
Paul E. McKenney7b0b7592010-08-17 14:18:46 -0700194#else /* #ifdef CONFIG_PREEMPT_RCU */
195
196static inline void __rcu_read_lock(void)
197{
198 preempt_disable();
199}
200
201static inline void __rcu_read_unlock(void)
202{
203 preempt_enable();
204}
205
206static inline void synchronize_rcu(void)
207{
208 synchronize_sched();
209}
210
211static inline int rcu_preempt_depth(void)
212{
213 return 0;
214}
215
216#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
217
218/* Internal to kernel */
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200219void rcu_init(void);
220void rcu_sched_qs(int cpu);
221void rcu_bh_qs(int cpu);
222void rcu_check_callbacks(int cpu, int user);
Paul E. McKenney7b0b7592010-08-17 14:18:46 -0700223struct notifier_block;
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200224void rcu_idle_enter(void);
225void rcu_idle_exit(void);
226void rcu_irq_enter(void);
227void rcu_irq_exit(void);
Frederic Weisbecker2b1d5022012-07-11 20:26:30 +0200228
229#ifdef CONFIG_RCU_USER_QS
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200230void rcu_user_enter(void);
231void rcu_user_exit(void);
Frederic Weisbecker2b1d5022012-07-11 20:26:30 +0200232#else
233static inline void rcu_user_enter(void) { }
234static inline void rcu_user_exit(void) { }
Frederic Weisbecker4d9a5d42012-10-11 01:47:16 +0200235static inline void rcu_user_hooks_switch(struct task_struct *prev,
236 struct task_struct *next) { }
Frederic Weisbecker2b1d5022012-07-11 20:26:30 +0200237#endif /* CONFIG_RCU_USER_QS */
238
Paul E. McKenney8a2ecf42012-02-02 15:42:04 -0800239/**
240 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
241 * @a: Code that RCU needs to pay attention to.
242 *
243 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
244 * in the inner idle loop, that is, between the rcu_idle_enter() and
245 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
246 * critical sections. However, things like powertop need tracepoints
247 * in the inner idle loop.
248 *
249 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
250 * will tell RCU that it needs to pay attending, invoke its argument
251 * (in this example, a call to the do_something_with_RCU() function),
252 * and then tell RCU to go back to ignoring this CPU. It is permissible
253 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently
254 * quite limited. If deeper nesting is required, it will be necessary
255 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly.
Paul E. McKenney8a2ecf42012-02-02 15:42:04 -0800256 */
257#define RCU_NONIDLE(a) \
258 do { \
Paul E. McKenneyb4270ee2012-07-31 10:12:48 -0700259 rcu_irq_enter(); \
Paul E. McKenney8a2ecf42012-02-02 15:42:04 -0800260 do { a; } while (0); \
Paul E. McKenneyb4270ee2012-07-31 10:12:48 -0700261 rcu_irq_exit(); \
Paul E. McKenney8a2ecf42012-02-02 15:42:04 -0800262 } while (0)
263
Paul E. McKenneycc6783f2013-09-06 17:39:49 -0700264#if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP)
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200265bool __rcu_is_watching(void);
Paul E. McKenneycc6783f2013-09-06 17:39:49 -0700266#endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */
267
Paul E. McKenney2c428182011-05-26 22:14:36 -0700268/*
269 * Infrastructure to implement the synchronize_() primitives in
270 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
271 */
272
273typedef void call_rcu_func_t(struct rcu_head *head,
274 void (*func)(struct rcu_head *head));
275void wait_rcu_gp(call_rcu_func_t crf);
276
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700277#if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100278#include <linux/rcutree.h>
Paul E. McKenney127781d2013-03-27 08:44:00 -0700279#elif defined(CONFIG_TINY_RCU)
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700280#include <linux/rcutiny.h>
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100281#else
282#error "Unknown RCU implementation specified to kernel configuration"
Paul E. McKenney6b3ef482009-08-22 13:56:53 -0700283#endif
Paul E. McKenney01c1c662008-01-25 21:08:24 +0100284
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400285/*
286 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
287 * initialization and destruction of rcu_head on the stack. rcu_head structures
288 * allocated dynamically in the heap or defined statically don't need any
289 * initialization.
290 */
291#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200292void init_rcu_head_on_stack(struct rcu_head *head);
293void destroy_rcu_head_on_stack(struct rcu_head *head);
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400294#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
Mathieu Desnoyers43760302010-04-17 08:48:39 -0400295static inline void init_rcu_head_on_stack(struct rcu_head *head)
296{
297}
298
299static inline void destroy_rcu_head_on_stack(struct rcu_head *head)
300{
301}
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400302#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
Mathieu Desnoyers43760302010-04-17 08:48:39 -0400303
Paul E. McKenneyc0d6d012012-01-23 12:41:26 -0800304#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
305bool rcu_lockdep_current_cpu_online(void);
306#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
307static inline bool rcu_lockdep_current_cpu_online(void)
308{
309 return 1;
310}
311#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
312
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700313#ifdef CONFIG_DEBUG_LOCK_ALLOC
Paul E. McKenney632ee202010-02-22 17:04:45 -0800314
Frederic Weisbecker00f49e52011-10-07 18:22:02 +0200315static inline void rcu_lock_acquire(struct lockdep_map *map)
316{
Frederic Weisbecker00f49e52011-10-07 18:22:02 +0200317 lock_acquire(map, 0, 0, 2, 1, NULL, _THIS_IP_);
318}
319
320static inline void rcu_lock_release(struct lockdep_map *map)
321{
Frederic Weisbecker00f49e52011-10-07 18:22:02 +0200322 lock_release(map, 1, _THIS_IP_);
323}
324
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700325extern struct lockdep_map rcu_lock_map;
Paul E. McKenney632ee202010-02-22 17:04:45 -0800326extern struct lockdep_map rcu_bh_lock_map;
Paul E. McKenney632ee202010-02-22 17:04:45 -0800327extern struct lockdep_map rcu_sched_lock_map;
Paul E. McKenney24ef6592013-10-28 09:22:24 -0700328extern struct lockdep_map rcu_callback_map;
Paul E. McKenneybc293d62010-04-15 12:50:39 -0700329extern int debug_lockdep_rcu_enabled(void);
Paul E. McKenney54dbf962010-03-03 07:46:57 -0800330
Paul E. McKenney632ee202010-02-22 17:04:45 -0800331/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700332 * rcu_read_lock_held() - might we be in RCU read-side critical section?
Paul E. McKenney632ee202010-02-22 17:04:45 -0800333 *
Paul E. McKenneyd20200b2010-03-30 10:52:21 -0700334 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
335 * read-side critical section. In absence of CONFIG_DEBUG_LOCK_ALLOC,
Paul E. McKenney632ee202010-02-22 17:04:45 -0800336 * this assumes we are in an RCU read-side critical section unless it can
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700337 * prove otherwise. This is useful for debug checks in functions that
338 * require that they be called within an RCU read-side critical section.
Paul E. McKenney54dbf962010-03-03 07:46:57 -0800339 *
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700340 * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
Paul E. McKenney32c141a2010-03-30 10:59:28 -0700341 * and while lockdep is disabled.
Paul E. McKenney3842a082011-11-28 10:42:42 -0800342 *
343 * Note that rcu_read_lock() and the matching rcu_read_unlock() must
344 * occur in the same context, for example, it is illegal to invoke
345 * rcu_read_unlock() in process context if the matching rcu_read_lock()
346 * was invoked from within an irq handler.
Paul E. McKenneyc0d6d012012-01-23 12:41:26 -0800347 *
348 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
349 * offline from an RCU perspective, so check for those as well.
Paul E. McKenney632ee202010-02-22 17:04:45 -0800350 */
351static inline int rcu_read_lock_held(void)
352{
Paul E. McKenney54dbf962010-03-03 07:46:57 -0800353 if (!debug_lockdep_rcu_enabled())
354 return 1;
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700355 if (!rcu_is_watching())
Frederic Weisbeckere6b80a32011-10-07 16:25:18 -0700356 return 0;
Paul E. McKenneyc0d6d012012-01-23 12:41:26 -0800357 if (!rcu_lockdep_current_cpu_online())
358 return 0;
Paul E. McKenney54dbf962010-03-03 07:46:57 -0800359 return lock_is_held(&rcu_lock_map);
Paul E. McKenney632ee202010-02-22 17:04:45 -0800360}
361
Paul E. McKenneye3818b82010-03-15 17:03:43 -0700362/*
363 * rcu_read_lock_bh_held() is defined out of line to avoid #include-file
364 * hell.
Paul E. McKenney632ee202010-02-22 17:04:45 -0800365 */
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200366int rcu_read_lock_bh_held(void);
Paul E. McKenney632ee202010-02-22 17:04:45 -0800367
368/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700369 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
Paul E. McKenney632ee202010-02-22 17:04:45 -0800370 *
Paul E. McKenneyd20200b2010-03-30 10:52:21 -0700371 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
372 * RCU-sched read-side critical section. In absence of
373 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
374 * critical section unless it can prove otherwise. Note that disabling
375 * of preemption (including disabling irqs) counts as an RCU-sched
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700376 * read-side critical section. This is useful for debug checks in functions
377 * that required that they be called within an RCU-sched read-side
378 * critical section.
Paul E. McKenney54dbf962010-03-03 07:46:57 -0800379 *
Paul E. McKenney32c141a2010-03-30 10:59:28 -0700380 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
381 * and while lockdep is disabled.
Frederic Weisbeckere6b80a32011-10-07 16:25:18 -0700382 *
383 * Note that if the CPU is in the idle loop from an RCU point of
384 * view (ie: that we are in the section between rcu_idle_enter() and
385 * rcu_idle_exit()) then rcu_read_lock_held() returns false even if the CPU
386 * did an rcu_read_lock(). The reason for this is that RCU ignores CPUs
387 * that are in such a section, considering these as in extended quiescent
388 * state, so such a CPU is effectively never in an RCU read-side critical
389 * section regardless of what RCU primitives it invokes. This state of
390 * affairs is required --- we need to keep an RCU-free window in idle
391 * where the CPU may possibly enter into low power mode. This way we can
392 * notice an extended quiescent state to other CPUs that started a grace
393 * period. Otherwise we would delay any grace period as long as we run in
394 * the idle task.
Paul E. McKenneyc0d6d012012-01-23 12:41:26 -0800395 *
396 * Similarly, we avoid claiming an SRCU read lock held if the current
397 * CPU is offline.
Paul E. McKenney632ee202010-02-22 17:04:45 -0800398 */
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200399#ifdef CONFIG_PREEMPT_COUNT
Paul E. McKenney632ee202010-02-22 17:04:45 -0800400static inline int rcu_read_lock_sched_held(void)
401{
402 int lockdep_opinion = 0;
403
Paul E. McKenney54dbf962010-03-03 07:46:57 -0800404 if (!debug_lockdep_rcu_enabled())
405 return 1;
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700406 if (!rcu_is_watching())
Frederic Weisbeckere6b80a32011-10-07 16:25:18 -0700407 return 0;
Paul E. McKenneyc0d6d012012-01-23 12:41:26 -0800408 if (!rcu_lockdep_current_cpu_online())
409 return 0;
Paul E. McKenney632ee202010-02-22 17:04:45 -0800410 if (debug_locks)
411 lockdep_opinion = lock_is_held(&rcu_sched_lock_map);
Lai Jiangshan0cff8102010-03-18 12:25:33 -0700412 return lockdep_opinion || preempt_count() != 0 || irqs_disabled();
Paul E. McKenney632ee202010-02-22 17:04:45 -0800413}
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200414#else /* #ifdef CONFIG_PREEMPT_COUNT */
Paul E. McKenneye6033e32010-03-03 17:50:16 -0800415static inline int rcu_read_lock_sched_held(void)
416{
417 return 1;
418}
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200419#endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
Paul E. McKenney632ee202010-02-22 17:04:45 -0800420
421#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
422
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200423# define rcu_lock_acquire(a) do { } while (0)
424# define rcu_lock_release(a) do { } while (0)
Paul E. McKenney632ee202010-02-22 17:04:45 -0800425
426static inline int rcu_read_lock_held(void)
427{
428 return 1;
429}
430
431static inline int rcu_read_lock_bh_held(void)
432{
433 return 1;
434}
435
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200436#ifdef CONFIG_PREEMPT_COUNT
Paul E. McKenney632ee202010-02-22 17:04:45 -0800437static inline int rcu_read_lock_sched_held(void)
438{
Paul E. McKenneybbad9372010-04-02 16:17:17 -0700439 return preempt_count() != 0 || irqs_disabled();
Paul E. McKenney632ee202010-02-22 17:04:45 -0800440}
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200441#else /* #ifdef CONFIG_PREEMPT_COUNT */
Paul E. McKenneye6033e32010-03-03 17:50:16 -0800442static inline int rcu_read_lock_sched_held(void)
443{
444 return 1;
445}
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200446#endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
Paul E. McKenney632ee202010-02-22 17:04:45 -0800447
448#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
449
450#ifdef CONFIG_PROVE_RCU
451
Tetsuo Handa4221a992010-06-26 01:08:19 +0900452/**
453 * rcu_lockdep_assert - emit lockdep splat if specified condition not met
454 * @c: condition to check
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700455 * @s: informative message
Tetsuo Handa4221a992010-06-26 01:08:19 +0900456 */
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700457#define rcu_lockdep_assert(c, s) \
Lai Jiangshan2b3fc352010-04-20 16:23:07 +0800458 do { \
Jan Beulich7ccaba52012-03-23 15:01:52 -0700459 static bool __section(.data.unlikely) __warned; \
Lai Jiangshan2b3fc352010-04-20 16:23:07 +0800460 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \
461 __warned = true; \
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700462 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
Lai Jiangshan2b3fc352010-04-20 16:23:07 +0800463 } \
464 } while (0)
465
Paul E. McKenney50406b92012-01-12 13:49:19 -0800466#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
467static inline void rcu_preempt_sleep_check(void)
468{
469 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
Paul E. McKenney5cf05ad2012-05-17 15:12:45 -0700470 "Illegal context switch in RCU read-side critical section");
Paul E. McKenney50406b92012-01-12 13:49:19 -0800471}
472#else /* #ifdef CONFIG_PROVE_RCU */
473static inline void rcu_preempt_sleep_check(void)
474{
475}
476#endif /* #else #ifdef CONFIG_PROVE_RCU */
477
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700478#define rcu_sleep_check() \
479 do { \
Paul E. McKenney50406b92012-01-12 13:49:19 -0800480 rcu_preempt_sleep_check(); \
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700481 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \
482 "Illegal context switch in RCU-bh" \
483 " read-side critical section"); \
484 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \
485 "Illegal context switch in RCU-sched"\
486 " read-side critical section"); \
487 } while (0)
488
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700489#else /* #ifdef CONFIG_PROVE_RCU */
490
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700491#define rcu_lockdep_assert(c, s) do { } while (0)
492#define rcu_sleep_check() do { } while (0)
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700493
494#endif /* #else #ifdef CONFIG_PROVE_RCU */
495
496/*
497 * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
498 * and rcu_assign_pointer(). Some of these could be folded into their
499 * callers, but they are left separate in order to ease introduction of
500 * multiple flavors of pointers to match the multiple flavors of RCU
501 * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
502 * the future.
503 */
Paul E. McKenney53ecfba2010-09-13 17:24:21 -0700504
505#ifdef __CHECKER__
506#define rcu_dereference_sparse(p, space) \
507 ((void)(((typeof(*p) space *)p) == p))
508#else /* #ifdef __CHECKER__ */
509#define rcu_dereference_sparse(p, space)
510#endif /* #else #ifdef __CHECKER__ */
511
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700512#define __rcu_access_pointer(p, space) \
513 ({ \
514 typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \
Paul E. McKenney53ecfba2010-09-13 17:24:21 -0700515 rcu_dereference_sparse(p, space); \
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700516 ((typeof(*p) __force __kernel *)(_________p1)); \
517 })
518#define __rcu_dereference_check(p, c, space) \
519 ({ \
520 typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700521 rcu_lockdep_assert(c, "suspicious rcu_dereference_check()" \
522 " usage"); \
Paul E. McKenney53ecfba2010-09-13 17:24:21 -0700523 rcu_dereference_sparse(p, space); \
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700524 smp_read_barrier_depends(); \
525 ((typeof(*p) __force __kernel *)(_________p1)); \
526 })
527#define __rcu_dereference_protected(p, c, space) \
528 ({ \
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700529 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected()" \
530 " usage"); \
Paul E. McKenney53ecfba2010-09-13 17:24:21 -0700531 rcu_dereference_sparse(p, space); \
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700532 ((typeof(*p) __force __kernel *)(p)); \
533 })
534
Paul E. McKenneya4dd9922011-04-01 07:15:14 -0700535#define __rcu_access_index(p, space) \
536 ({ \
537 typeof(p) _________p1 = ACCESS_ONCE(p); \
538 rcu_dereference_sparse(p, space); \
539 (_________p1); \
540 })
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700541#define __rcu_dereference_index_check(p, c) \
542 ({ \
543 typeof(p) _________p1 = ACCESS_ONCE(p); \
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700544 rcu_lockdep_assert(c, \
545 "suspicious rcu_dereference_index_check()" \
546 " usage"); \
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700547 smp_read_barrier_depends(); \
548 (_________p1); \
549 })
Paul E. McKenney462225ae2013-11-11 09:59:34 -0800550
551/**
552 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
553 * @v: The value to statically initialize with.
554 */
555#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
556
557/**
558 * rcu_assign_pointer() - assign to RCU-protected pointer
559 * @p: pointer to assign to
560 * @v: value to assign (publish)
561 *
562 * Assigns the specified value to the specified RCU-protected
563 * pointer, ensuring that any concurrent RCU readers will see
564 * any prior initialization.
565 *
566 * Inserts memory barriers on architectures that require them
567 * (which is most of them), and also prevents the compiler from
568 * reordering the code that initializes the structure after the pointer
569 * assignment. More importantly, this call documents which pointers
570 * will be dereferenced by RCU read-side code.
571 *
572 * In some special cases, you may use RCU_INIT_POINTER() instead
573 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
574 * to the fact that it does not constrain either the CPU or the compiler.
575 * That said, using RCU_INIT_POINTER() when you should have used
576 * rcu_assign_pointer() is a very bad thing that results in
577 * impossible-to-diagnose memory corruption. So please be careful.
578 * See the RCU_INIT_POINTER() comment header for details.
579 *
580 * Note that rcu_assign_pointer() evaluates each of its arguments only
581 * once, appearances notwithstanding. One of the "extra" evaluations
582 * is in typeof() and the other visible only to sparse (__CHECKER__),
583 * neither of which actually execute the argument. As with most cpp
584 * macros, this execute-arguments-only-once property is important, so
585 * please be careful when making changes to rcu_assign_pointer() and the
586 * other macros that it invokes.
587 */
588#define rcu_assign_pointer(p, v) \
Paul E. McKenneye9023c42012-05-16 15:51:08 -0700589 do { \
Eric Dumazetd322f452011-07-31 22:09:25 -0700590 smp_wmb(); \
Paul E. McKenney462225ae2013-11-11 09:59:34 -0800591 ACCESS_ONCE(p) = RCU_INITIALIZER(v); \
Paul E. McKenneye9023c42012-05-16 15:51:08 -0700592 } while (0)
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700593
594
Paul E. McKenney632ee202010-02-22 17:04:45 -0800595/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700596 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
597 * @p: The pointer to read
598 *
599 * Return the value of the specified RCU-protected pointer, but omit the
600 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
601 * when the value of this pointer is accessed, but the pointer is not
602 * dereferenced, for example, when testing an RCU-protected pointer against
603 * NULL. Although rcu_access_pointer() may also be used in cases where
604 * update-side locks prevent the value of the pointer from changing, you
605 * should instead use rcu_dereference_protected() for this use case.
Paul E. McKenney5e1ee6e2012-01-12 17:21:20 -0800606 *
607 * It is also permissible to use rcu_access_pointer() when read-side
608 * access to the pointer was removed at least one grace period ago, as
609 * is the case in the context of the RCU callback that is freeing up
610 * the data, or after a synchronize_rcu() returns. This can be useful
611 * when tearing down multi-linked structures after a grace period
612 * has elapsed.
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700613 */
614#define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
615
616/**
617 * rcu_dereference_check() - rcu_dereference with debug checking
David Howellsc08c68d2010-04-09 15:39:11 -0700618 * @p: The pointer to read, prior to dereferencing
619 * @c: The conditions under which the dereference will take place
Paul E. McKenney632ee202010-02-22 17:04:45 -0800620 *
David Howellsc08c68d2010-04-09 15:39:11 -0700621 * Do an rcu_dereference(), but check that the conditions under which the
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700622 * dereference will take place are correct. Typically the conditions
623 * indicate the various locking conditions that should be held at that
624 * point. The check should return true if the conditions are satisfied.
625 * An implicit check for being in an RCU read-side critical section
626 * (rcu_read_lock()) is included.
David Howellsc08c68d2010-04-09 15:39:11 -0700627 *
628 * For example:
629 *
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700630 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
David Howellsc08c68d2010-04-09 15:39:11 -0700631 *
632 * could be used to indicate to lockdep that foo->bar may only be dereferenced
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700633 * if either rcu_read_lock() is held, or that the lock required to replace
David Howellsc08c68d2010-04-09 15:39:11 -0700634 * the bar struct at foo->bar is held.
635 *
636 * Note that the list of conditions may also include indications of when a lock
637 * need not be held, for example during initialisation or destruction of the
638 * target struct:
639 *
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700640 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
David Howellsc08c68d2010-04-09 15:39:11 -0700641 * atomic_read(&foo->usage) == 0);
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700642 *
643 * Inserts memory barriers on architectures that require them
644 * (currently only the Alpha), prevents the compiler from refetching
645 * (and from merging fetches), and, more importantly, documents exactly
646 * which pointers are protected by RCU and checks that the pointer is
647 * annotated as __rcu.
Paul E. McKenney632ee202010-02-22 17:04:45 -0800648 */
649#define rcu_dereference_check(p, c) \
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700650 __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu)
Paul E. McKenney632ee202010-02-22 17:04:45 -0800651
Paul E. McKenneyb62730b2010-04-09 15:39:10 -0700652/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700653 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
654 * @p: The pointer to read, prior to dereferencing
655 * @c: The conditions under which the dereference will take place
656 *
657 * This is the RCU-bh counterpart to rcu_dereference_check().
658 */
659#define rcu_dereference_bh_check(p, c) \
660 __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu)
661
662/**
663 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
664 * @p: The pointer to read, prior to dereferencing
665 * @c: The conditions under which the dereference will take place
666 *
667 * This is the RCU-sched counterpart to rcu_dereference_check().
668 */
669#define rcu_dereference_sched_check(p, c) \
670 __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \
671 __rcu)
672
673#define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/
674
Steven Rostedt12bcbe62013-05-28 14:38:42 -0400675/*
676 * The tracing infrastructure traces RCU (we want that), but unfortunately
677 * some of the RCU checks causes tracing to lock up the system.
678 *
679 * The tracing version of rcu_dereference_raw() must not call
680 * rcu_read_lock_held().
681 */
682#define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
683
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700684/**
Paul E. McKenneya4dd9922011-04-01 07:15:14 -0700685 * rcu_access_index() - fetch RCU index with no dereferencing
686 * @p: The index to read
687 *
688 * Return the value of the specified RCU-protected index, but omit the
689 * smp_read_barrier_depends() and keep the ACCESS_ONCE(). This is useful
690 * when the value of this index is accessed, but the index is not
691 * dereferenced, for example, when testing an RCU-protected index against
692 * -1. Although rcu_access_index() may also be used in cases where
693 * update-side locks prevent the value of the index from changing, you
694 * should instead use rcu_dereference_index_protected() for this use case.
695 */
696#define rcu_access_index(p) __rcu_access_index((p), __rcu)
697
698/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700699 * rcu_dereference_index_check() - rcu_dereference for indices with debug checking
700 * @p: The pointer to read, prior to dereferencing
701 * @c: The conditions under which the dereference will take place
702 *
703 * Similar to rcu_dereference_check(), but omits the sparse checking.
704 * This allows rcu_dereference_index_check() to be used on integers,
705 * which can then be used as array indices. Attempting to use
706 * rcu_dereference_check() on an integer will give compiler warnings
707 * because the sparse address-space mechanism relies on dereferencing
708 * the RCU-protected pointer. Dereferencing integers is not something
709 * that even gcc will put up with.
710 *
711 * Note that this function does not implicitly check for RCU read-side
712 * critical sections. If this function gains lots of uses, it might
713 * make sense to provide versions for each flavor of RCU, but it does
714 * not make sense as of early 2010.
715 */
716#define rcu_dereference_index_check(p, c) \
717 __rcu_dereference_index_check((p), (c))
718
719/**
720 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
721 * @p: The pointer to read, prior to dereferencing
722 * @c: The conditions under which the dereference will take place
Paul E. McKenneyb62730b2010-04-09 15:39:10 -0700723 *
724 * Return the value of the specified RCU-protected pointer, but omit
725 * both the smp_read_barrier_depends() and the ACCESS_ONCE(). This
726 * is useful in cases where update-side locks prevent the value of the
727 * pointer from changing. Please note that this primitive does -not-
728 * prevent the compiler from repeating this reference or combining it
729 * with other references, so it should not be used without protection
730 * of appropriate locks.
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700731 *
732 * This function is only for update-side use. Using this function
733 * when protected only by rcu_read_lock() will result in infrequent
734 * but very ugly failures.
Paul E. McKenneyb62730b2010-04-09 15:39:10 -0700735 */
736#define rcu_dereference_protected(p, c) \
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700737 __rcu_dereference_protected((p), (c), __rcu)
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700738
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700739
740/**
741 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
742 * @p: The pointer to read, prior to dereferencing
743 *
744 * This is a simple wrapper around rcu_dereference_check().
745 */
746#define rcu_dereference(p) rcu_dereference_check(p, 0)
747
748/**
749 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
750 * @p: The pointer to read, prior to dereferencing
751 *
752 * Makes rcu_dereference_check() do the dirty work.
753 */
754#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
755
756/**
757 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
758 * @p: The pointer to read, prior to dereferencing
759 *
760 * Makes rcu_dereference_check() do the dirty work.
761 */
762#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
763
764/**
765 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 *
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700767 * When synchronize_rcu() is invoked on one CPU while other CPUs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 * are within RCU read-side critical sections, then the
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700769 * synchronize_rcu() is guaranteed to block until after all the other
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
771 * on one CPU while other CPUs are within RCU read-side critical
772 * sections, invocation of the corresponding RCU callback is deferred
773 * until after the all the other CPUs exit their critical sections.
774 *
775 * Note, however, that RCU callbacks are permitted to run concurrently
Paul E. McKenney77d84852010-07-08 17:38:59 -0700776 * with new RCU read-side critical sections. One way that this can happen
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 * is via the following sequence of events: (1) CPU 0 enters an RCU
778 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
779 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
780 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
781 * callback is invoked. This is legal, because the RCU read-side critical
782 * section that was running concurrently with the call_rcu() (and which
783 * therefore might be referencing something that the corresponding RCU
784 * callback would free up) has completed before the corresponding
785 * RCU callback is invoked.
786 *
787 * RCU read-side critical sections may be nested. Any deferred actions
788 * will be deferred until the outermost RCU read-side critical section
789 * completes.
790 *
Paul E. McKenney9079fd72010-08-07 21:59:54 -0700791 * You can avoid reading and understanding the next paragraph by
792 * following this rule: don't put anything in an rcu_read_lock() RCU
793 * read-side critical section that would block in a !PREEMPT kernel.
794 * But if you want the full story, read on!
795 *
796 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), it
797 * is illegal to block while in an RCU read-side critical section. In
798 * preemptible RCU implementations (TREE_PREEMPT_RCU and TINY_PREEMPT_RCU)
799 * in CONFIG_PREEMPT kernel builds, RCU read-side critical sections may
800 * be preempted, but explicit blocking is illegal. Finally, in preemptible
Paul E. McKenney90f45e42013-01-10 05:24:49 -0800801 * RCU implementations in real-time (with -rt patchset) kernel builds,
Paul E. McKenney9079fd72010-08-07 21:59:54 -0700802 * RCU read-side critical sections may be preempted and they may also
803 * block, but only when acquiring spinlocks that are subject to priority
804 * inheritance.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 */
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700806static inline void rcu_read_lock(void)
807{
808 __rcu_read_lock();
809 __acquire(RCU);
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200810 rcu_lock_acquire(&rcu_lock_map);
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700811 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800812 "rcu_read_lock() used illegally while idle");
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700813}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815/*
816 * So where is rcu_write_lock()? It does not exist, as there is no
817 * way for writers to lock out RCU readers. This is a feature, not
818 * a bug -- this property is what provides RCU's performance benefits.
819 * Of course, writers must coordinate with each other. The normal
820 * spinlock primitives work well for this, but any other technique may be
821 * used as well. RCU does not care how the writers keep out of each
822 * others' way, as long as they do so.
823 */
Paul E. McKenney3d76c082009-09-28 07:46:32 -0700824
825/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700826 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
Paul E. McKenney3d76c082009-09-28 07:46:32 -0700827 *
828 * See rcu_read_lock() for more information.
829 */
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700830static inline void rcu_read_unlock(void)
831{
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700832 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800833 "rcu_read_unlock() used illegally while idle");
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200834 rcu_lock_release(&rcu_lock_map);
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700835 __release(RCU);
836 __rcu_read_unlock();
837}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838
839/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700840 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 *
842 * This is equivalent of rcu_read_lock(), but to be used when updates
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700843 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
844 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
845 * softirq handler to be a quiescent state, a process in RCU read-side
846 * critical section must be protected by disabling softirqs. Read-side
847 * critical sections in interrupt context can use just rcu_read_lock(),
848 * though this should at least be commented to avoid confusing people
849 * reading the code.
Paul E. McKenney3842a082011-11-28 10:42:42 -0800850 *
851 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
852 * must occur in the same context, for example, it is illegal to invoke
853 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
854 * was invoked from some other task.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 */
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700856static inline void rcu_read_lock_bh(void)
857{
Paul E. McKenney6206ab92011-08-01 06:22:11 -0700858 local_bh_disable();
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700859 __acquire(RCU_BH);
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200860 rcu_lock_acquire(&rcu_bh_lock_map);
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700861 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800862 "rcu_read_lock_bh() used illegally while idle");
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700863}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
865/*
866 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
867 *
868 * See rcu_read_lock_bh() for more information.
869 */
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700870static inline void rcu_read_unlock_bh(void)
871{
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700872 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800873 "rcu_read_unlock_bh() used illegally while idle");
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200874 rcu_lock_release(&rcu_bh_lock_map);
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700875 __release(RCU_BH);
Paul E. McKenney6206ab92011-08-01 06:22:11 -0700876 local_bh_enable();
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700877}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878
879/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700880 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
Mathieu Desnoyers1c50b722008-09-29 11:06:46 -0400881 *
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700882 * This is equivalent of rcu_read_lock(), but to be used when updates
883 * are being done using call_rcu_sched() or synchronize_rcu_sched().
884 * Read-side critical sections can also be introduced by anything that
885 * disables preemption, including local_irq_disable() and friends.
Paul E. McKenney3842a082011-11-28 10:42:42 -0800886 *
887 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
888 * must occur in the same context, for example, it is illegal to invoke
889 * rcu_read_unlock_sched() from process context if the matching
890 * rcu_read_lock_sched() was invoked from an NMI handler.
Mathieu Desnoyers1c50b722008-09-29 11:06:46 -0400891 */
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700892static inline void rcu_read_lock_sched(void)
893{
894 preempt_disable();
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700895 __acquire(RCU_SCHED);
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200896 rcu_lock_acquire(&rcu_sched_lock_map);
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700897 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800898 "rcu_read_lock_sched() used illegally while idle");
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700899}
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700900
901/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
Paul E. McKenney7c614d62009-08-24 09:42:00 -0700902static inline notrace void rcu_read_lock_sched_notrace(void)
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700903{
904 preempt_disable_notrace();
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700905 __acquire(RCU_SCHED);
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700906}
Mathieu Desnoyers1c50b722008-09-29 11:06:46 -0400907
908/*
909 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
910 *
911 * See rcu_read_lock_sched for more information.
912 */
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700913static inline void rcu_read_unlock_sched(void)
914{
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700915 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800916 "rcu_read_unlock_sched() used illegally while idle");
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200917 rcu_lock_release(&rcu_sched_lock_map);
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700918 __release(RCU_SCHED);
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700919 preempt_enable();
920}
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700921
922/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
Paul E. McKenney7c614d62009-08-24 09:42:00 -0700923static inline notrace void rcu_read_unlock_sched_notrace(void)
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700924{
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700925 __release(RCU_SCHED);
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700926 preempt_enable_notrace();
927}
Mathieu Desnoyers1c50b722008-09-29 11:06:46 -0400928
Mathieu Desnoyers1c50b722008-09-29 11:06:46 -0400929/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700930 * RCU_INIT_POINTER() - initialize an RCU protected pointer
931 *
Paul E. McKenney6846c0c2011-07-31 22:33:02 -0700932 * Initialize an RCU-protected pointer in special cases where readers
933 * do not need ordering constraints on the CPU or the compiler. These
934 * special cases are:
935 *
936 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
937 * 2. The caller has taken whatever steps are required to prevent
938 * RCU readers from concurrently accessing this pointer -or-
939 * 3. The referenced data structure has already been exposed to
940 * readers either at compile time or via rcu_assign_pointer() -and-
941 * a. You have not made -any- reader-visible changes to
942 * this structure since then -or-
943 * b. It is OK for readers accessing this structure from its
944 * new location to see the old state of the structure. (For
945 * example, the changes were to statistical counters or to
946 * other state where exact synchronization is not required.)
947 *
948 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
949 * result in impossible-to-diagnose memory corruption. As in the structures
950 * will look OK in crash dumps, but any concurrent RCU readers might
951 * see pre-initialized values of the referenced data structure. So
952 * please be very careful how you use RCU_INIT_POINTER()!!!
953 *
954 * If you are creating an RCU-protected linked structure that is accessed
955 * by a single external-to-structure RCU-protected pointer, then you may
956 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
957 * pointers, but you must use rcu_assign_pointer() to initialize the
958 * external-to-structure pointer -after- you have completely initialized
959 * the reader-accessible portions of the linked structure.
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700960 */
961#define RCU_INIT_POINTER(p, v) \
Paul E. McKenneyd1b88eb2012-05-16 15:42:30 -0700962 do { \
Paul E. McKenney462225ae2013-11-11 09:59:34 -0800963 p = RCU_INITIALIZER(v); \
Paul E. McKenneyd1b88eb2012-05-16 15:42:30 -0700964 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Paul E. McKenney172708d2012-05-16 15:23:45 -0700966/**
967 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
968 *
969 * GCC-style initialization for an RCU-protected pointer in a structure field.
970 */
971#define RCU_POINTER_INITIALIZER(p, v) \
Paul E. McKenney462225ae2013-11-11 09:59:34 -0800972 .p = RCU_INITIALIZER(v)
Lai Jiangshan9ab15442011-03-18 11:15:47 +0800973
Jan Engelhardtd8169d42012-04-19 11:44:39 -0700974/*
975 * Does the specified offset indicate that the corresponding rcu_head
976 * structure can be handled by kfree_rcu()?
977 */
978#define __is_kfree_rcu_offset(offset) ((offset) < 4096)
979
980/*
981 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
982 */
983#define __kfree_rcu(head, offset) \
984 do { \
985 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
Paul E. McKenney4fa3b6c2012-06-05 15:53:53 -0700986 kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \
Jan Engelhardtd8169d42012-04-19 11:44:39 -0700987 } while (0)
988
Lai Jiangshan9ab15442011-03-18 11:15:47 +0800989/**
990 * kfree_rcu() - kfree an object after a grace period.
991 * @ptr: pointer to kfree
992 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
993 *
994 * Many rcu callbacks functions just call kfree() on the base structure.
995 * These functions are trivial, but their size adds up, and furthermore
996 * when they are used in a kernel module, that module must invoke the
997 * high-latency rcu_barrier() function at module-unload time.
998 *
999 * The kfree_rcu() function handles this issue. Rather than encoding a
1000 * function address in the embedded rcu_head structure, kfree_rcu() instead
1001 * encodes the offset of the rcu_head structure within the base structure.
1002 * Because the functions are not allowed in the low-order 4096 bytes of
1003 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
1004 * If the offset is larger than 4095 bytes, a compile-time error will
1005 * be generated in __kfree_rcu(). If this error is triggered, you can
1006 * either fall back to use of call_rcu() or rearrange the structure to
1007 * position the rcu_head structure into the first 4096 bytes.
1008 *
1009 * Note that the allowable offset might decrease in the future, for example,
1010 * to allow something like kmem_cache_free_rcu().
Jan Engelhardtd8169d42012-04-19 11:44:39 -07001011 *
1012 * The BUILD_BUG_ON check must not involve any function calls, hence the
1013 * checks are done in macros here.
Lai Jiangshan9ab15442011-03-18 11:15:47 +08001014 */
1015#define kfree_rcu(ptr, rcu_head) \
1016 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
1017
Frederic Weisbeckerd1e43fa2013-03-26 23:47:24 +01001018#ifdef CONFIG_RCU_NOCB_CPU
Teodora Baluta584dc4c2013-11-11 17:11:23 +02001019bool rcu_is_nocb_cpu(int cpu);
Frederic Weisbeckerd1e43fa2013-03-26 23:47:24 +01001020#else
1021static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
1022#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
1023
1024
Paul E. McKenney0edd1b12013-06-21 16:37:22 -07001025/* Only for use by adaptive-ticks code. */
1026#ifdef CONFIG_NO_HZ_FULL_SYSIDLE
Teodora Baluta584dc4c2013-11-11 17:11:23 +02001027bool rcu_sys_is_idle(void);
1028void rcu_sysidle_force_exit(void);
Paul E. McKenney0edd1b12013-06-21 16:37:22 -07001029#else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1030
1031static inline bool rcu_sys_is_idle(void)
1032{
1033 return false;
1034}
1035
1036static inline void rcu_sysidle_force_exit(void)
1037{
1038}
1039
1040#endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1041
1042
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043#endif /* __LINUX_RCUPDATE_H */