blob: 33ec16b9c2ee9b8b3bd4cad7cb387e82f671c3b2 [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
Paul E. McKenney87de1cf2013-12-03 10:02:52 -080015 * along with this program; if not, you can access it online at
16 * http://www.gnu.org/licenses/gpl-2.0.html.
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 *
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>
Thomas Gleixnerc1ad3482015-04-14 21:08:58 +000047#include <linux/ktime.h>
48
Paul E. McKenney88c18632013-12-16 13:24:32 -080049#include <asm/barrier.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Paul Gortmaker7a754742014-02-11 16:10:12 -050051extern int rcu_expedited; /* for sysctl */
Dave Younge5ab6772010-03-10 15:24:05 -080052
Paul E. McKenney0d394822015-02-18 12:24:30 -080053#ifdef CONFIG_TINY_RCU
54/* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
55static inline bool rcu_gp_is_expedited(void) /* Internal RCU use. */
56{
57 return false;
58}
59
60static inline void rcu_expedite_gp(void)
61{
62}
63
64static inline void rcu_unexpedite_gp(void)
65{
66}
67#else /* #ifdef CONFIG_TINY_RCU */
68bool rcu_gp_is_expedited(void); /* Internal RCU use. */
69void rcu_expedite_gp(void);
70void rcu_unexpedite_gp(void);
71#endif /* #else #ifdef CONFIG_TINY_RCU */
72
Paul E. McKenneyad0dc7f2014-02-19 10:51:42 -080073enum rcutorture_type {
74 RCU_FLAVOR,
75 RCU_BH_FLAVOR,
76 RCU_SCHED_FLAVOR,
Paul E. McKenney69c60452014-07-01 11:59:36 -070077 RCU_TASKS_FLAVOR,
Paul E. McKenneyad0dc7f2014-02-19 10:51:42 -080078 SRCU_FLAVOR,
79 INVALID_RCU_FLAVOR
80};
81
Pranith Kumar28f65692014-09-22 14:00:48 -040082#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
Paul E. McKenneyad0dc7f2014-02-19 10:51:42 -080083void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
84 unsigned long *gpnum, unsigned long *completed);
Teodora Baluta584dc4c2013-11-11 17:11:23 +020085void rcutorture_record_test_transition(void);
86void rcutorture_record_progress(unsigned long vernum);
87void do_trace_rcu_torture_read(const char *rcutorturename,
88 struct rcu_head *rhp,
89 unsigned long secs,
90 unsigned long c_old,
91 unsigned long c);
Paul E. McKenney4a298652011-04-03 21:33:51 -070092#else
Paul E. McKenneyad0dc7f2014-02-19 10:51:42 -080093static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
94 int *flags,
95 unsigned long *gpnum,
96 unsigned long *completed)
97{
98 *flags = 0;
99 *gpnum = 0;
100 *completed = 0;
101}
Paul E. McKenney4a298652011-04-03 21:33:51 -0700102static inline void rcutorture_record_test_transition(void)
103{
104}
105static inline void rcutorture_record_progress(unsigned long vernum)
106{
107}
Paul E. McKenney91afaf32011-10-02 07:44:32 -0700108#ifdef CONFIG_RCU_TRACE
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200109void do_trace_rcu_torture_read(const char *rcutorturename,
110 struct rcu_head *rhp,
111 unsigned long secs,
112 unsigned long c_old,
113 unsigned long c);
Paul E. McKenney91afaf32011-10-02 07:44:32 -0700114#else
Paul E. McKenney52494532012-11-14 16:26:40 -0800115#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
116 do { } while (0)
Paul E. McKenney91afaf32011-10-02 07:44:32 -0700117#endif
Paul E. McKenney4a298652011-04-03 21:33:51 -0700118#endif
119
Tejun Heoe27fc962010-11-22 21:36:11 -0800120#define UINT_CMP_GE(a, b) (UINT_MAX / 2 >= (a) - (b))
121#define UINT_CMP_LT(a, b) (UINT_MAX / 2 < (a) - (b))
Paul E. McKenneya3dc3fb2010-08-13 16:16:25 -0700122#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b))
123#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
Paul E. McKenneyc0f4dfd2012-12-28 11:30:36 -0800124#define ulong2long(a) (*(long *)(&(a)))
Paul E. McKenneya3dc3fb2010-08-13 16:16:25 -0700125
Paul E. McKenney03b042b2009-06-25 09:08:16 -0700126/* Exported common interfaces */
Paul E. McKenney2c428182011-05-26 22:14:36 -0700127
128#ifdef CONFIG_PREEMPT_RCU
129
130/**
131 * call_rcu() - Queue an RCU callback for invocation after a grace period.
132 * @head: structure to be used for queueing the RCU updates.
133 * @func: actual callback function to be invoked after the grace period
134 *
135 * The callback function will be invoked some time after a full grace
136 * period elapses, in other words after all pre-existing RCU read-side
137 * critical sections have completed. However, the callback function
138 * might well execute concurrently with RCU read-side critical sections
139 * that started after call_rcu() was invoked. RCU read-side critical
140 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
141 * and may be nested.
Paul E. McKenneyf0a0e6f2012-10-23 13:47:01 -0700142 *
143 * Note that all CPUs must agree that the grace period extended beyond
144 * all pre-existing RCU read-side critical section. On systems with more
145 * than one CPU, this means that when "func()" is invoked, each CPU is
146 * guaranteed to have executed a full memory barrier since the end of its
147 * last RCU read-side critical section whose beginning preceded the call
148 * to call_rcu(). It also means that each CPU executing an RCU read-side
149 * critical section that continues beyond the start of "func()" must have
150 * executed a memory barrier after the call_rcu() but before the beginning
151 * of that RCU read-side critical section. Note that these guarantees
152 * include CPUs that are offline, idle, or executing in user mode, as
153 * well as CPUs that are executing in the kernel.
154 *
155 * Furthermore, if CPU A invoked call_rcu() and CPU B invoked the
156 * resulting RCU callback function "func()", then both CPU A and CPU B are
157 * guaranteed to execute a full memory barrier during the time interval
158 * between the call to call_rcu() and the invocation of "func()" -- even
159 * if CPU A and CPU B are the same CPU (but again only if the system has
160 * more than one CPU).
Paul E. McKenney2c428182011-05-26 22:14:36 -0700161 */
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200162void call_rcu(struct rcu_head *head,
163 void (*func)(struct rcu_head *head));
Paul E. McKenney2c428182011-05-26 22:14:36 -0700164
165#else /* #ifdef CONFIG_PREEMPT_RCU */
166
167/* In classic RCU, call_rcu() is just call_rcu_sched(). */
168#define call_rcu call_rcu_sched
169
170#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
171
172/**
173 * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
174 * @head: structure to be used for queueing the RCU updates.
175 * @func: actual callback function to be invoked after the grace period
176 *
177 * The callback function will be invoked some time after a full grace
178 * period elapses, in other words after all currently executing RCU
179 * read-side critical sections have completed. call_rcu_bh() assumes
180 * that the read-side critical sections end on completion of a softirq
181 * handler. This means that read-side critical sections in process
182 * context must not be interrupted by softirqs. This interface is to be
183 * used when most of the read-side critical sections are in softirq context.
184 * RCU read-side critical sections are delimited by :
185 * - rcu_read_lock() and rcu_read_unlock(), if in interrupt context.
186 * OR
187 * - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
188 * These may be nested.
Paul E. McKenneyf0a0e6f2012-10-23 13:47:01 -0700189 *
190 * See the description of call_rcu() for more detailed information on
191 * memory ordering guarantees.
Paul E. McKenney2c428182011-05-26 22:14:36 -0700192 */
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200193void call_rcu_bh(struct rcu_head *head,
194 void (*func)(struct rcu_head *head));
Paul E. McKenney2c428182011-05-26 22:14:36 -0700195
196/**
197 * call_rcu_sched() - Queue an RCU for invocation after sched grace period.
198 * @head: structure to be used for queueing the RCU updates.
199 * @func: actual callback function to be invoked after the grace period
200 *
201 * The callback function will be invoked some time after a full grace
202 * period elapses, in other words after all currently executing RCU
203 * read-side critical sections have completed. call_rcu_sched() assumes
204 * that the read-side critical sections end on enabling of preemption
205 * or on voluntary preemption.
206 * RCU read-side critical sections are delimited by :
207 * - rcu_read_lock_sched() and rcu_read_unlock_sched(),
208 * OR
209 * anything that disables preemption.
210 * These may be nested.
Paul E. McKenneyf0a0e6f2012-10-23 13:47:01 -0700211 *
212 * See the description of call_rcu() for more detailed information on
213 * memory ordering guarantees.
Paul E. McKenney2c428182011-05-26 22:14:36 -0700214 */
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200215void call_rcu_sched(struct rcu_head *head,
216 void (*func)(struct rcu_head *rcu));
Paul E. McKenney2c428182011-05-26 22:14:36 -0700217
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200218void synchronize_sched(void);
Paul E. McKenney03b042b2009-06-25 09:08:16 -0700219
Paul E. McKenneyee376db2015-01-10 19:47:10 -0800220/*
221 * Structure allowing asynchronous waiting on RCU.
222 */
223struct rcu_synchronize {
224 struct rcu_head head;
225 struct completion completion;
226};
227void wakeme_after_rcu(struct rcu_head *head);
228
Paul E. McKenneyec90a192015-06-10 12:53:06 -0700229void __wait_rcu_gp(bool checktiny, int n, call_rcu_func_t *crcu_array,
230 struct rcu_synchronize *rs_array);
231
232#define _wait_rcu_gp(checktiny, ...) \
233do { \
234 call_rcu_func_t __crcu_array[] = { __VA_ARGS__ }; \
235 const int __n = ARRAY_SIZE(__crcu_array); \
236 struct rcu_synchronize __rs_array[__n]; \
237 \
238 __wait_rcu_gp(checktiny, __n, __crcu_array, __rs_array); \
239} while (0)
240
241#define wait_rcu_gp(...) _wait_rcu_gp(false, __VA_ARGS__)
242
243/**
244 * synchronize_rcu_mult - Wait concurrently for multiple grace periods
245 * @...: List of call_rcu() functions for the flavors to wait on.
246 *
247 * This macro waits concurrently for multiple flavors of RCU grace periods.
248 * For example, synchronize_rcu_mult(call_rcu, call_rcu_bh) would wait
249 * on concurrent RCU and RCU-bh grace periods. Waiting on a give SRCU
250 * domain requires you to write a wrapper function for that SRCU domain's
251 * call_srcu() function, supplying the corresponding srcu_struct.
252 *
253 * If Tiny RCU, tell _wait_rcu_gp() not to bother waiting for RCU
254 * or RCU-bh, given that anywhere synchronize_rcu_mult() can be called
255 * is automatically a grace period.
256 */
257#define synchronize_rcu_mult(...) \
258 _wait_rcu_gp(IS_ENABLED(CONFIG_TINY_RCU), __VA_ARGS__)
259
Paul E. McKenney8315f422014-06-27 13:42:20 -0700260/**
261 * call_rcu_tasks() - Queue an RCU for invocation task-based grace period
262 * @head: structure to be used for queueing the RCU updates.
263 * @func: actual callback function to be invoked after the grace period
264 *
265 * The callback function will be invoked some time after a full grace
266 * period elapses, in other words after all currently executing RCU
267 * read-side critical sections have completed. call_rcu_tasks() assumes
268 * that the read-side critical sections end at a voluntary context
269 * switch (not a preemption!), entry into idle, or transition to usermode
270 * execution. As such, there are no read-side primitives analogous to
271 * rcu_read_lock() and rcu_read_unlock() because this primitive is intended
272 * to determine that all tasks have passed through a safe state, not so
273 * much for data-strcuture synchronization.
274 *
275 * See the description of call_rcu() for more detailed information on
276 * memory ordering guarantees.
277 */
278void call_rcu_tasks(struct rcu_head *head, void (*func)(struct rcu_head *head));
Paul E. McKenney53c6d4e2014-07-01 12:22:23 -0700279void synchronize_rcu_tasks(void);
280void rcu_barrier_tasks(void);
Paul E. McKenney8315f422014-06-27 13:42:20 -0700281
Paul E. McKenneya3dc3fb2010-08-13 16:16:25 -0700282#ifdef CONFIG_PREEMPT_RCU
283
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200284void __rcu_read_lock(void);
285void __rcu_read_unlock(void);
286void rcu_read_unlock_special(struct task_struct *t);
Paul E. McKenney7b0b7592010-08-17 14:18:46 -0700287void synchronize_rcu(void);
288
Paul E. McKenneya3dc3fb2010-08-13 16:16:25 -0700289/*
290 * Defined as a macro as it is a very low level header included from
291 * areas that don't even know about current. This gives the rcu_read_lock()
292 * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
293 * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
294 */
295#define rcu_preempt_depth() (current->rcu_read_lock_nesting)
296
Paul E. McKenney7b0b7592010-08-17 14:18:46 -0700297#else /* #ifdef CONFIG_PREEMPT_RCU */
298
299static inline void __rcu_read_lock(void)
300{
301 preempt_disable();
302}
303
304static inline void __rcu_read_unlock(void)
305{
306 preempt_enable();
307}
308
309static inline void synchronize_rcu(void)
310{
311 synchronize_sched();
312}
313
314static inline int rcu_preempt_depth(void)
315{
316 return 0;
317}
318
319#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
320
321/* Internal to kernel */
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200322void rcu_init(void);
Paul E. McKenneyee425712015-02-19 10:51:32 -0800323void rcu_end_inkernel_boot(void);
Paul E. McKenney284a8c92014-08-14 16:38:46 -0700324void rcu_sched_qs(void);
325void rcu_bh_qs(void);
Paul E. McKenneyc3377c2d2014-10-21 07:53:02 -0700326void rcu_check_callbacks(int user);
Paul E. McKenney7b0b7592010-08-17 14:18:46 -0700327struct notifier_block;
Paul E. McKenney88428cc2015-01-28 14:42:09 -0800328int rcu_cpu_notify(struct notifier_block *self,
329 unsigned long action, void *hcpu);
Frederic Weisbecker2b1d5022012-07-11 20:26:30 +0200330
Rik van Riel61f38db2014-04-26 23:15:35 -0700331#ifdef CONFIG_RCU_STALL_COMMON
332void rcu_sysrq_start(void);
333void rcu_sysrq_end(void);
334#else /* #ifdef CONFIG_RCU_STALL_COMMON */
335static inline void rcu_sysrq_start(void)
336{
337}
338static inline void rcu_sysrq_end(void)
339{
340}
341#endif /* #else #ifdef CONFIG_RCU_STALL_COMMON */
342
Paul E. McKenneyd1ec4c32015-05-13 10:41:58 -0700343#ifdef CONFIG_NO_HZ_FULL
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200344void rcu_user_enter(void);
345void rcu_user_exit(void);
Frederic Weisbecker2b1d5022012-07-11 20:26:30 +0200346#else
347static inline void rcu_user_enter(void) { }
348static inline void rcu_user_exit(void) { }
Frederic Weisbecker4d9a5d42012-10-11 01:47:16 +0200349static inline void rcu_user_hooks_switch(struct task_struct *prev,
350 struct task_struct *next) { }
Paul E. McKenneyd1ec4c32015-05-13 10:41:58 -0700351#endif /* CONFIG_NO_HZ_FULL */
Frederic Weisbecker2b1d5022012-07-11 20:26:30 +0200352
Paul E. McKenneyf4579fc2014-07-25 11:21:47 -0700353#ifdef CONFIG_RCU_NOCB_CPU
354void rcu_init_nohz(void);
355#else /* #ifdef CONFIG_RCU_NOCB_CPU */
356static inline void rcu_init_nohz(void)
357{
358}
359#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
360
Paul E. McKenney8a2ecf42012-02-02 15:42:04 -0800361/**
362 * RCU_NONIDLE - Indicate idle-loop code that needs RCU readers
363 * @a: Code that RCU needs to pay attention to.
364 *
365 * RCU, RCU-bh, and RCU-sched read-side critical sections are forbidden
366 * in the inner idle loop, that is, between the rcu_idle_enter() and
367 * the rcu_idle_exit() -- RCU will happily ignore any such read-side
368 * critical sections. However, things like powertop need tracepoints
369 * in the inner idle loop.
370 *
371 * This macro provides the way out: RCU_NONIDLE(do_something_with_RCU())
372 * will tell RCU that it needs to pay attending, invoke its argument
373 * (in this example, a call to the do_something_with_RCU() function),
374 * and then tell RCU to go back to ignoring this CPU. It is permissible
375 * to nest RCU_NONIDLE() wrappers, but the nesting level is currently
376 * quite limited. If deeper nesting is required, it will be necessary
377 * to adjust DYNTICK_TASK_NESTING_VALUE accordingly.
Paul E. McKenney8a2ecf42012-02-02 15:42:04 -0800378 */
379#define RCU_NONIDLE(a) \
380 do { \
Paul E. McKenneyb4270ee2012-07-31 10:12:48 -0700381 rcu_irq_enter(); \
Paul E. McKenney8a2ecf42012-02-02 15:42:04 -0800382 do { a; } while (0); \
Paul E. McKenneyb4270ee2012-07-31 10:12:48 -0700383 rcu_irq_exit(); \
Paul E. McKenney8a2ecf42012-02-02 15:42:04 -0800384 } while (0)
385
Paul E. McKenney8315f422014-06-27 13:42:20 -0700386/*
387 * Note a voluntary context switch for RCU-tasks benefit. This is a
388 * macro rather than an inline function to avoid #include hell.
389 */
390#ifdef CONFIG_TASKS_RCU
Paul E. McKenney3f95aa82014-08-04 06:10:23 -0700391#define TASKS_RCU(x) x
392extern struct srcu_struct tasks_rcu_exit_srcu;
Paul E. McKenney8315f422014-06-27 13:42:20 -0700393#define rcu_note_voluntary_context_switch(t) \
394 do { \
Paul E. McKenney5cd37192014-12-13 20:32:04 -0800395 rcu_all_qs(); \
Paul E. McKenney7d0ae802015-03-03 14:57:58 -0800396 if (READ_ONCE((t)->rcu_tasks_holdout)) \
397 WRITE_ONCE((t)->rcu_tasks_holdout, false); \
Paul E. McKenney8315f422014-06-27 13:42:20 -0700398 } while (0)
399#else /* #ifdef CONFIG_TASKS_RCU */
Paul E. McKenney3f95aa82014-08-04 06:10:23 -0700400#define TASKS_RCU(x) do { } while (0)
Paul E. McKenney5cd37192014-12-13 20:32:04 -0800401#define rcu_note_voluntary_context_switch(t) rcu_all_qs()
Paul E. McKenney8315f422014-06-27 13:42:20 -0700402#endif /* #else #ifdef CONFIG_TASKS_RCU */
403
Paul E. McKenneybde6c3a2014-07-01 11:26:57 -0700404/**
405 * cond_resched_rcu_qs - Report potential quiescent states to RCU
406 *
407 * This macro resembles cond_resched(), except that it is defined to
408 * report potential quiescent states to RCU-tasks even if the cond_resched()
409 * machinery were to be shut off, as some advocate for PREEMPT kernels.
410 */
411#define cond_resched_rcu_qs() \
412do { \
Paul E. McKenneyb6331ae2014-10-04 03:43:41 -0700413 if (!cond_resched()) \
414 rcu_note_voluntary_context_switch(current); \
Paul E. McKenneybde6c3a2014-07-01 11:26:57 -0700415} while (0)
416
Paul E. McKenneycc6783f2013-09-06 17:39:49 -0700417#if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP)
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200418bool __rcu_is_watching(void);
Paul E. McKenneycc6783f2013-09-06 17:39:49 -0700419#endif /* #if defined(CONFIG_DEBUG_LOCK_ALLOC) || defined(CONFIG_RCU_TRACE) || defined(CONFIG_SMP) */
420
Paul E. McKenney2c428182011-05-26 22:14:36 -0700421/*
422 * Infrastructure to implement the synchronize_() primitives in
423 * TREE_RCU and rcu_barrier_() primitives in TINY_RCU.
424 */
425
Pranith Kumar28f65692014-09-22 14:00:48 -0400426#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100427#include <linux/rcutree.h>
Paul E. McKenney127781d2013-03-27 08:44:00 -0700428#elif defined(CONFIG_TINY_RCU)
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700429#include <linux/rcutiny.h>
Paul E. McKenney64db4cf2008-12-18 21:55:32 +0100430#else
431#error "Unknown RCU implementation specified to kernel configuration"
Paul E. McKenney6b3ef482009-08-22 13:56:53 -0700432#endif
Paul E. McKenney01c1c662008-01-25 21:08:24 +0100433
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400434/*
435 * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
436 * initialization and destruction of rcu_head on the stack. rcu_head structures
437 * allocated dynamically in the heap or defined statically don't need any
438 * initialization.
439 */
440#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
Paul E. McKenney546a9d82014-06-19 14:57:10 -0700441void init_rcu_head(struct rcu_head *head);
442void destroy_rcu_head(struct rcu_head *head);
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200443void init_rcu_head_on_stack(struct rcu_head *head);
444void destroy_rcu_head_on_stack(struct rcu_head *head);
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400445#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
Paul E. McKenney546a9d82014-06-19 14:57:10 -0700446static inline void init_rcu_head(struct rcu_head *head)
447{
448}
449
450static inline void destroy_rcu_head(struct rcu_head *head)
451{
452}
453
Mathieu Desnoyers43760302010-04-17 08:48:39 -0400454static inline void init_rcu_head_on_stack(struct rcu_head *head)
455{
456}
457
458static inline void destroy_rcu_head_on_stack(struct rcu_head *head)
459{
460}
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400461#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
Mathieu Desnoyers43760302010-04-17 08:48:39 -0400462
Paul E. McKenneyc0d6d012012-01-23 12:41:26 -0800463#if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU)
464bool rcu_lockdep_current_cpu_online(void);
465#else /* #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
466static inline bool rcu_lockdep_current_cpu_online(void)
467{
Pranith Kumar521d24e2014-07-08 18:26:18 -0400468 return true;
Paul E. McKenneyc0d6d012012-01-23 12:41:26 -0800469}
470#endif /* #else #if defined(CONFIG_HOTPLUG_CPU) && defined(CONFIG_PROVE_RCU) */
471
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700472#ifdef CONFIG_DEBUG_LOCK_ALLOC
Paul E. McKenney632ee202010-02-22 17:04:45 -0800473
Frederic Weisbecker00f49e52011-10-07 18:22:02 +0200474static inline void rcu_lock_acquire(struct lockdep_map *map)
475{
Oleg Nesterovfb9edbe2014-01-20 19:20:06 +0100476 lock_acquire(map, 0, 0, 2, 0, NULL, _THIS_IP_);
Frederic Weisbecker00f49e52011-10-07 18:22:02 +0200477}
478
479static inline void rcu_lock_release(struct lockdep_map *map)
480{
Frederic Weisbecker00f49e52011-10-07 18:22:02 +0200481 lock_release(map, 1, _THIS_IP_);
482}
483
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700484extern struct lockdep_map rcu_lock_map;
Paul E. McKenney632ee202010-02-22 17:04:45 -0800485extern struct lockdep_map rcu_bh_lock_map;
Paul E. McKenney632ee202010-02-22 17:04:45 -0800486extern struct lockdep_map rcu_sched_lock_map;
Paul E. McKenney24ef6592013-10-28 09:22:24 -0700487extern struct lockdep_map rcu_callback_map;
Iulia Mandaa235c092014-03-12 18:37:24 +0200488int debug_lockdep_rcu_enabled(void);
Paul E. McKenney54dbf962010-03-03 07:46:57 -0800489
Oleg Nesterov85b39d32014-07-08 15:17:59 -0700490int rcu_read_lock_held(void);
Teodora Baluta584dc4c2013-11-11 17:11:23 +0200491int rcu_read_lock_bh_held(void);
Paul E. McKenney632ee202010-02-22 17:04:45 -0800492
493/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700494 * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
Paul E. McKenney632ee202010-02-22 17:04:45 -0800495 *
Paul E. McKenneyd20200b2010-03-30 10:52:21 -0700496 * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
497 * RCU-sched read-side critical section. In absence of
498 * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
Denys Vlasenkod5671f62015-05-26 17:48:34 +0200499 * critical section unless it can prove otherwise.
Paul E. McKenney632ee202010-02-22 17:04:45 -0800500 */
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200501#ifdef CONFIG_PREEMPT_COUNT
Denys Vlasenkod5671f62015-05-26 17:48:34 +0200502int rcu_read_lock_sched_held(void);
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200503#else /* #ifdef CONFIG_PREEMPT_COUNT */
Paul E. McKenneye6033e32010-03-03 17:50:16 -0800504static inline int rcu_read_lock_sched_held(void)
505{
506 return 1;
507}
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200508#endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
Paul E. McKenney632ee202010-02-22 17:04:45 -0800509
510#else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
511
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200512# define rcu_lock_acquire(a) do { } while (0)
513# define rcu_lock_release(a) do { } while (0)
Paul E. McKenney632ee202010-02-22 17:04:45 -0800514
515static inline int rcu_read_lock_held(void)
516{
517 return 1;
518}
519
520static inline int rcu_read_lock_bh_held(void)
521{
522 return 1;
523}
524
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200525#ifdef CONFIG_PREEMPT_COUNT
Paul E. McKenney632ee202010-02-22 17:04:45 -0800526static inline int rcu_read_lock_sched_held(void)
527{
Paul E. McKenneybbad9372010-04-02 16:17:17 -0700528 return preempt_count() != 0 || irqs_disabled();
Paul E. McKenney632ee202010-02-22 17:04:45 -0800529}
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200530#else /* #ifdef CONFIG_PREEMPT_COUNT */
Paul E. McKenneye6033e32010-03-03 17:50:16 -0800531static inline int rcu_read_lock_sched_held(void)
532{
533 return 1;
534}
Frederic Weisbeckerbdd4e852011-06-08 01:13:27 +0200535#endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
Paul E. McKenney632ee202010-02-22 17:04:45 -0800536
537#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
538
539#ifdef CONFIG_PROVE_RCU
540
Tetsuo Handa4221a992010-06-26 01:08:19 +0900541/**
542 * rcu_lockdep_assert - emit lockdep splat if specified condition not met
543 * @c: condition to check
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700544 * @s: informative message
Tetsuo Handa4221a992010-06-26 01:08:19 +0900545 */
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700546#define rcu_lockdep_assert(c, s) \
Lai Jiangshan2b3fc352010-04-20 16:23:07 +0800547 do { \
Jan Beulich7ccaba52012-03-23 15:01:52 -0700548 static bool __section(.data.unlikely) __warned; \
Lai Jiangshan2b3fc352010-04-20 16:23:07 +0800549 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \
550 __warned = true; \
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700551 lockdep_rcu_suspicious(__FILE__, __LINE__, s); \
Lai Jiangshan2b3fc352010-04-20 16:23:07 +0800552 } \
553 } while (0)
554
Paul E. McKenney50406b92012-01-12 13:49:19 -0800555#if defined(CONFIG_PROVE_RCU) && !defined(CONFIG_PREEMPT_RCU)
556static inline void rcu_preempt_sleep_check(void)
557{
558 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
Paul E. McKenney5cf05ad2012-05-17 15:12:45 -0700559 "Illegal context switch in RCU read-side critical section");
Paul E. McKenney50406b92012-01-12 13:49:19 -0800560}
561#else /* #ifdef CONFIG_PROVE_RCU */
562static inline void rcu_preempt_sleep_check(void)
563{
564}
565#endif /* #else #ifdef CONFIG_PROVE_RCU */
566
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700567#define rcu_sleep_check() \
568 do { \
Paul E. McKenney50406b92012-01-12 13:49:19 -0800569 rcu_preempt_sleep_check(); \
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700570 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map), \
Joe Perches41f4abd2013-12-05 15:10:23 -0800571 "Illegal context switch in RCU-bh read-side critical section"); \
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700572 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map), \
Joe Perches41f4abd2013-12-05 15:10:23 -0800573 "Illegal context switch in RCU-sched read-side critical section"); \
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700574 } while (0)
575
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700576#else /* #ifdef CONFIG_PROVE_RCU */
577
Paul E. McKenneyb3fbab02011-05-24 08:31:09 -0700578#define rcu_lockdep_assert(c, s) do { } while (0)
579#define rcu_sleep_check() do { } while (0)
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700580
581#endif /* #else #ifdef CONFIG_PROVE_RCU */
582
583/*
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
589 * the future.
590 */
Paul E. McKenney53ecfba2010-09-13 17:24:21 -0700591
592#ifdef __CHECKER__
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__ */
598
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700599#define __rcu_access_pointer(p, space) \
Joe Perches0adab9b2013-12-05 16:19:15 -0800600({ \
Paul E. McKenney7d0ae802015-03-03 14:57:58 -0800601 typeof(*p) *_________p1 = (typeof(*p) *__force)READ_ONCE(p); \
Joe Perches0adab9b2013-12-05 16:19:15 -0800602 rcu_dereference_sparse(p, space); \
603 ((typeof(*p) __force __kernel *)(_________p1)); \
604})
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700605#define __rcu_dereference_check(p, c, space) \
Joe Perches0adab9b2013-12-05 16:19:15 -0800606({ \
Pranith Kumarac598532014-11-13 14:24:14 -0500607 /* Dependency order vs. p above. */ \
608 typeof(*p) *________p1 = (typeof(*p) *__force)lockless_dereference(p); \
Joe Perches0adab9b2013-12-05 16:19:15 -0800609 rcu_lockdep_assert(c, "suspicious rcu_dereference_check() usage"); \
610 rcu_dereference_sparse(p, space); \
Pranith Kumarac598532014-11-13 14:24:14 -0500611 ((typeof(*p) __force __kernel *)(________p1)); \
Joe Perches0adab9b2013-12-05 16:19:15 -0800612})
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700613#define __rcu_dereference_protected(p, c, space) \
Joe Perches0adab9b2013-12-05 16:19:15 -0800614({ \
615 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected() usage"); \
616 rcu_dereference_sparse(p, space); \
617 ((typeof(*p) __force __kernel *)(p)); \
618})
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700619
Paul E. McKenney462225ae2013-11-11 09:59:34 -0800620/**
621 * RCU_INITIALIZER() - statically initialize an RCU-protected global variable
622 * @v: The value to statically initialize with.
623 */
624#define RCU_INITIALIZER(v) (typeof(*(v)) __force __rcu *)(v)
625
626/**
627 * rcu_assign_pointer() - assign to RCU-protected pointer
628 * @p: pointer to assign to
629 * @v: value to assign (publish)
630 *
631 * Assigns the specified value to the specified RCU-protected
632 * pointer, ensuring that any concurrent RCU readers will see
633 * any prior initialization.
634 *
635 * Inserts memory barriers on architectures that require them
636 * (which is most of them), and also prevents the compiler from
637 * reordering the code that initializes the structure after the pointer
638 * assignment. More importantly, this call documents which pointers
639 * will be dereferenced by RCU read-side code.
640 *
641 * In some special cases, you may use RCU_INIT_POINTER() instead
642 * of rcu_assign_pointer(). RCU_INIT_POINTER() is a bit faster due
643 * to the fact that it does not constrain either the CPU or the compiler.
644 * That said, using RCU_INIT_POINTER() when you should have used
645 * rcu_assign_pointer() is a very bad thing that results in
646 * impossible-to-diagnose memory corruption. So please be careful.
647 * See the RCU_INIT_POINTER() comment header for details.
648 *
649 * Note that rcu_assign_pointer() evaluates each of its arguments only
650 * once, appearances notwithstanding. One of the "extra" evaluations
651 * is in typeof() and the other visible only to sparse (__CHECKER__),
652 * neither of which actually execute the argument. As with most cpp
653 * macros, this execute-arguments-only-once property is important, so
654 * please be careful when making changes to rcu_assign_pointer() and the
655 * other macros that it invokes.
656 */
Paul E. McKenney88c18632013-12-16 13:24:32 -0800657#define rcu_assign_pointer(p, v) smp_store_release(&p, RCU_INITIALIZER(v))
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700658
Paul E. McKenney632ee202010-02-22 17:04:45 -0800659/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700660 * rcu_access_pointer() - fetch RCU pointer with no dereferencing
661 * @p: The pointer to read
662 *
663 * Return the value of the specified RCU-protected pointer, but omit the
Paul E. McKenney7d0ae802015-03-03 14:57:58 -0800664 * smp_read_barrier_depends() and keep the READ_ONCE(). This is useful
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700665 * when the value of this pointer is accessed, but the pointer is not
666 * dereferenced, for example, when testing an RCU-protected pointer against
667 * NULL. Although rcu_access_pointer() may also be used in cases where
668 * update-side locks prevent the value of the pointer from changing, you
669 * should instead use rcu_dereference_protected() for this use case.
Paul E. McKenney5e1ee6e2012-01-12 17:21:20 -0800670 *
671 * It is also permissible to use rcu_access_pointer() when read-side
672 * access to the pointer was removed at least one grace period ago, as
673 * is the case in the context of the RCU callback that is freeing up
674 * the data, or after a synchronize_rcu() returns. This can be useful
675 * when tearing down multi-linked structures after a grace period
676 * has elapsed.
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700677 */
678#define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
679
680/**
681 * rcu_dereference_check() - rcu_dereference with debug checking
David Howellsc08c68d2010-04-09 15:39:11 -0700682 * @p: The pointer to read, prior to dereferencing
683 * @c: The conditions under which the dereference will take place
Paul E. McKenney632ee202010-02-22 17:04:45 -0800684 *
David Howellsc08c68d2010-04-09 15:39:11 -0700685 * Do an rcu_dereference(), but check that the conditions under which the
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700686 * dereference will take place are correct. Typically the conditions
687 * indicate the various locking conditions that should be held at that
688 * point. The check should return true if the conditions are satisfied.
689 * An implicit check for being in an RCU read-side critical section
690 * (rcu_read_lock()) is included.
David Howellsc08c68d2010-04-09 15:39:11 -0700691 *
692 * For example:
693 *
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700694 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
David Howellsc08c68d2010-04-09 15:39:11 -0700695 *
696 * could be used to indicate to lockdep that foo->bar may only be dereferenced
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700697 * if either rcu_read_lock() is held, or that the lock required to replace
David Howellsc08c68d2010-04-09 15:39:11 -0700698 * the bar struct at foo->bar is held.
699 *
700 * Note that the list of conditions may also include indications of when a lock
701 * need not be held, for example during initialisation or destruction of the
702 * target struct:
703 *
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700704 * bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
David Howellsc08c68d2010-04-09 15:39:11 -0700705 * atomic_read(&foo->usage) == 0);
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700706 *
707 * Inserts memory barriers on architectures that require them
708 * (currently only the Alpha), prevents the compiler from refetching
709 * (and from merging fetches), and, more importantly, documents exactly
710 * which pointers are protected by RCU and checks that the pointer is
711 * annotated as __rcu.
Paul E. McKenney632ee202010-02-22 17:04:45 -0800712 */
713#define rcu_dereference_check(p, c) \
Paul E. McKenneyb826565a2015-02-02 11:46:33 -0800714 __rcu_dereference_check((p), (c) || rcu_read_lock_held(), __rcu)
Paul E. McKenney632ee202010-02-22 17:04:45 -0800715
Paul E. McKenneyb62730b2010-04-09 15:39:10 -0700716/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700717 * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
718 * @p: The pointer to read, prior to dereferencing
719 * @c: The conditions under which the dereference will take place
720 *
721 * This is the RCU-bh counterpart to rcu_dereference_check().
722 */
723#define rcu_dereference_bh_check(p, c) \
Paul E. McKenneyb826565a2015-02-02 11:46:33 -0800724 __rcu_dereference_check((p), (c) || rcu_read_lock_bh_held(), __rcu)
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700725
726/**
727 * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
728 * @p: The pointer to read, prior to dereferencing
729 * @c: The conditions under which the dereference will take place
730 *
731 * This is the RCU-sched counterpart to rcu_dereference_check().
732 */
733#define rcu_dereference_sched_check(p, c) \
Paul E. McKenneyb826565a2015-02-02 11:46:33 -0800734 __rcu_dereference_check((p), (c) || rcu_read_lock_sched_held(), \
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700735 __rcu)
736
737#define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/
738
Steven Rostedt12bcbe62013-05-28 14:38:42 -0400739/*
740 * The tracing infrastructure traces RCU (we want that), but unfortunately
741 * some of the RCU checks causes tracing to lock up the system.
742 *
743 * The tracing version of rcu_dereference_raw() must not call
744 * rcu_read_lock_held().
745 */
746#define rcu_dereference_raw_notrace(p) __rcu_dereference_check((p), 1, __rcu)
747
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700748/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700749 * rcu_dereference_protected() - fetch RCU pointer when updates prevented
750 * @p: The pointer to read, prior to dereferencing
751 * @c: The conditions under which the dereference will take place
Paul E. McKenneyb62730b2010-04-09 15:39:10 -0700752 *
753 * Return the value of the specified RCU-protected pointer, but omit
Paul E. McKenney7d0ae802015-03-03 14:57:58 -0800754 * both the smp_read_barrier_depends() and the READ_ONCE(). This
Paul E. McKenneyb62730b2010-04-09 15:39:10 -0700755 * is useful in cases where update-side locks prevent the value of the
756 * pointer from changing. Please note that this primitive does -not-
757 * prevent the compiler from repeating this reference or combining it
758 * with other references, so it should not be used without protection
759 * of appropriate locks.
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700760 *
761 * This function is only for update-side use. Using this function
762 * when protected only by rcu_read_lock() will result in infrequent
763 * but very ugly failures.
Paul E. McKenneyb62730b2010-04-09 15:39:10 -0700764 */
765#define rcu_dereference_protected(p, c) \
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700766 __rcu_dereference_protected((p), (c), __rcu)
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700767
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700768
769/**
770 * rcu_dereference() - fetch RCU-protected pointer for dereferencing
771 * @p: The pointer to read, prior to dereferencing
772 *
773 * This is a simple wrapper around rcu_dereference_check().
774 */
775#define rcu_dereference(p) rcu_dereference_check(p, 0)
776
777/**
778 * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
779 * @p: The pointer to read, prior to dereferencing
780 *
781 * Makes rcu_dereference_check() do the dirty work.
782 */
783#define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
784
785/**
786 * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
787 * @p: The pointer to read, prior to dereferencing
788 *
789 * Makes rcu_dereference_check() do the dirty work.
790 */
791#define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
792
793/**
794 * rcu_read_lock() - mark the beginning of an RCU read-side critical section
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 *
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700796 * When synchronize_rcu() is invoked on one CPU while other CPUs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 * are within RCU read-side critical sections, then the
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700798 * synchronize_rcu() is guaranteed to block until after all the other
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
800 * on one CPU while other CPUs are within RCU read-side critical
801 * sections, invocation of the corresponding RCU callback is deferred
802 * until after the all the other CPUs exit their critical sections.
803 *
804 * Note, however, that RCU callbacks are permitted to run concurrently
Paul E. McKenney77d84852010-07-08 17:38:59 -0700805 * with new RCU read-side critical sections. One way that this can happen
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 * is via the following sequence of events: (1) CPU 0 enters an RCU
807 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
808 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
809 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
810 * callback is invoked. This is legal, because the RCU read-side critical
811 * section that was running concurrently with the call_rcu() (and which
812 * therefore might be referencing something that the corresponding RCU
813 * callback would free up) has completed before the corresponding
814 * RCU callback is invoked.
815 *
816 * RCU read-side critical sections may be nested. Any deferred actions
817 * will be deferred until the outermost RCU read-side critical section
818 * completes.
819 *
Paul E. McKenney9079fd72010-08-07 21:59:54 -0700820 * You can avoid reading and understanding the next paragraph by
821 * following this rule: don't put anything in an rcu_read_lock() RCU
822 * read-side critical section that would block in a !PREEMPT kernel.
823 * But if you want the full story, read on!
824 *
Paul E. McKenneyab74fdf2014-05-04 15:41:21 -0700825 * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU),
826 * it is illegal to block while in an RCU read-side critical section.
Pranith Kumar28f65692014-09-22 14:00:48 -0400827 * In preemptible RCU implementations (PREEMPT_RCU) in CONFIG_PREEMPT
Paul E. McKenneyab74fdf2014-05-04 15:41:21 -0700828 * kernel builds, RCU read-side critical sections may be preempted,
829 * but explicit blocking is illegal. Finally, in preemptible RCU
830 * implementations in real-time (with -rt patchset) kernel builds, RCU
831 * read-side critical sections may be preempted and they may also block, but
832 * only when acquiring spinlocks that are subject to priority inheritance.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 */
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700834static inline void rcu_read_lock(void)
835{
836 __rcu_read_lock();
837 __acquire(RCU);
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200838 rcu_lock_acquire(&rcu_lock_map);
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700839 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800840 "rcu_read_lock() used illegally while idle");
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700841}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843/*
844 * So where is rcu_write_lock()? It does not exist, as there is no
845 * way for writers to lock out RCU readers. This is a feature, not
846 * a bug -- this property is what provides RCU's performance benefits.
847 * Of course, writers must coordinate with each other. The normal
848 * spinlock primitives work well for this, but any other technique may be
849 * used as well. RCU does not care how the writers keep out of each
850 * others' way, as long as they do so.
851 */
Paul E. McKenney3d76c082009-09-28 07:46:32 -0700852
853/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700854 * rcu_read_unlock() - marks the end of an RCU read-side critical section.
Paul E. McKenney3d76c082009-09-28 07:46:32 -0700855 *
Paul E. McKenneyf27bc482014-05-04 15:38:38 -0700856 * In most situations, rcu_read_unlock() is immune from deadlock.
857 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock()
858 * is responsible for deboosting, which it does via rt_mutex_unlock().
859 * Unfortunately, this function acquires the scheduler's runqueue and
860 * priority-inheritance spinlocks. This means that deadlock could result
861 * if the caller of rcu_read_unlock() already holds one of these locks or
Oleg Nesterovce36f2f2014-09-28 23:44:21 +0200862 * any lock that is ever acquired while holding them; or any lock which
863 * can be taken from interrupt context because rcu_boost()->rt_mutex_lock()
864 * does not disable irqs while taking ->wait_lock.
Paul E. McKenneyf27bc482014-05-04 15:38:38 -0700865 *
866 * That said, RCU readers are never priority boosted unless they were
867 * preempted. Therefore, one way to avoid deadlock is to make sure
868 * that preemption never happens within any RCU read-side critical
869 * section whose outermost rcu_read_unlock() is called with one of
870 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in
871 * a number of ways, for example, by invoking preempt_disable() before
872 * critical section's outermost rcu_read_lock().
873 *
874 * Given that the set of locks acquired by rt_mutex_unlock() might change
875 * at any time, a somewhat more future-proofed approach is to make sure
876 * that that preemption never happens within any RCU read-side critical
877 * section whose outermost rcu_read_unlock() is called with irqs disabled.
878 * This approach relies on the fact that rt_mutex_unlock() currently only
879 * acquires irq-disabled locks.
880 *
881 * The second of these two approaches is best in most situations,
882 * however, the first approach can also be useful, at least to those
883 * developers willing to keep abreast of the set of locks acquired by
884 * rt_mutex_unlock().
885 *
Paul E. McKenney3d76c082009-09-28 07:46:32 -0700886 * See rcu_read_lock() for more information.
887 */
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700888static inline void rcu_read_unlock(void)
889{
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700890 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800891 "rcu_read_unlock() used illegally while idle");
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700892 __release(RCU);
893 __rcu_read_unlock();
Paul E. McKenneyd24209bb2015-01-21 15:26:03 -0800894 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700895}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700898 * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 *
900 * This is equivalent of rcu_read_lock(), but to be used when updates
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700901 * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
902 * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
903 * softirq handler to be a quiescent state, a process in RCU read-side
904 * critical section must be protected by disabling softirqs. Read-side
905 * critical sections in interrupt context can use just rcu_read_lock(),
906 * though this should at least be commented to avoid confusing people
907 * reading the code.
Paul E. McKenney3842a082011-11-28 10:42:42 -0800908 *
909 * Note that rcu_read_lock_bh() and the matching rcu_read_unlock_bh()
910 * must occur in the same context, for example, it is illegal to invoke
911 * rcu_read_unlock_bh() from one task if the matching rcu_read_lock_bh()
912 * was invoked from some other task.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 */
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700914static inline void rcu_read_lock_bh(void)
915{
Paul E. McKenney6206ab92011-08-01 06:22:11 -0700916 local_bh_disable();
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700917 __acquire(RCU_BH);
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200918 rcu_lock_acquire(&rcu_bh_lock_map);
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700919 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800920 "rcu_read_lock_bh() used illegally while idle");
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700921}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
923/*
924 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
925 *
926 * See rcu_read_lock_bh() for more information.
927 */
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700928static inline void rcu_read_unlock_bh(void)
929{
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700930 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800931 "rcu_read_unlock_bh() used illegally while idle");
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200932 rcu_lock_release(&rcu_bh_lock_map);
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700933 __release(RCU_BH);
Paul E. McKenney6206ab92011-08-01 06:22:11 -0700934 local_bh_enable();
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700935}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
937/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700938 * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
Mathieu Desnoyers1c50b722008-09-29 11:06:46 -0400939 *
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700940 * This is equivalent of rcu_read_lock(), but to be used when updates
941 * are being done using call_rcu_sched() or synchronize_rcu_sched().
942 * Read-side critical sections can also be introduced by anything that
943 * disables preemption, including local_irq_disable() and friends.
Paul E. McKenney3842a082011-11-28 10:42:42 -0800944 *
945 * Note that rcu_read_lock_sched() and the matching rcu_read_unlock_sched()
946 * must occur in the same context, for example, it is illegal to invoke
947 * rcu_read_unlock_sched() from process context if the matching
948 * rcu_read_lock_sched() was invoked from an NMI handler.
Mathieu Desnoyers1c50b722008-09-29 11:06:46 -0400949 */
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700950static inline void rcu_read_lock_sched(void)
951{
952 preempt_disable();
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700953 __acquire(RCU_SCHED);
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200954 rcu_lock_acquire(&rcu_sched_lock_map);
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700955 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800956 "rcu_read_lock_sched() used illegally while idle");
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700957}
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700958
959/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
Paul E. McKenney7c614d62009-08-24 09:42:00 -0700960static inline notrace void rcu_read_lock_sched_notrace(void)
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700961{
962 preempt_disable_notrace();
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700963 __acquire(RCU_SCHED);
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700964}
Mathieu Desnoyers1c50b722008-09-29 11:06:46 -0400965
966/*
967 * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
968 *
969 * See rcu_read_lock_sched for more information.
970 */
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700971static inline void rcu_read_unlock_sched(void)
972{
Paul E. McKenney5c173eb2013-09-13 17:20:11 -0700973 rcu_lockdep_assert(rcu_is_watching(),
Heiko Carstensbde23c62012-02-01 10:30:46 -0800974 "rcu_read_unlock_sched() used illegally while idle");
Paul E. McKenneyd8ab29f2011-10-07 18:22:03 +0200975 rcu_lock_release(&rcu_sched_lock_map);
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700976 __release(RCU_SCHED);
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700977 preempt_enable();
978}
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700979
980/* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
Paul E. McKenney7c614d62009-08-24 09:42:00 -0700981static inline notrace void rcu_read_unlock_sched_notrace(void)
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700982{
Paul E. McKenneybc33f242009-08-22 13:56:47 -0700983 __release(RCU_SCHED);
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700984 preempt_enable_notrace();
985}
Mathieu Desnoyers1c50b722008-09-29 11:06:46 -0400986
Mathieu Desnoyers1c50b722008-09-29 11:06:46 -0400987/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700988 * RCU_INIT_POINTER() - initialize an RCU protected pointer
989 *
Paul E. McKenney6846c0c2011-07-31 22:33:02 -0700990 * Initialize an RCU-protected pointer in special cases where readers
991 * do not need ordering constraints on the CPU or the compiler. These
992 * special cases are:
993 *
994 * 1. This use of RCU_INIT_POINTER() is NULLing out the pointer -or-
995 * 2. The caller has taken whatever steps are required to prevent
996 * RCU readers from concurrently accessing this pointer -or-
997 * 3. The referenced data structure has already been exposed to
998 * readers either at compile time or via rcu_assign_pointer() -and-
999 * a. You have not made -any- reader-visible changes to
1000 * this structure since then -or-
1001 * b. It is OK for readers accessing this structure from its
1002 * new location to see the old state of the structure. (For
1003 * example, the changes were to statistical counters or to
1004 * other state where exact synchronization is not required.)
1005 *
1006 * Failure to follow these rules governing use of RCU_INIT_POINTER() will
1007 * result in impossible-to-diagnose memory corruption. As in the structures
1008 * will look OK in crash dumps, but any concurrent RCU readers might
1009 * see pre-initialized values of the referenced data structure. So
1010 * please be very careful how you use RCU_INIT_POINTER()!!!
1011 *
1012 * If you are creating an RCU-protected linked structure that is accessed
1013 * by a single external-to-structure RCU-protected pointer, then you may
1014 * use RCU_INIT_POINTER() to initialize the internal RCU-protected
1015 * pointers, but you must use rcu_assign_pointer() to initialize the
1016 * external-to-structure pointer -after- you have completely initialized
1017 * the reader-accessible portions of the linked structure.
Paul E. McKenney71a9b262014-03-31 13:13:02 -07001018 *
1019 * Note that unlike rcu_assign_pointer(), RCU_INIT_POINTER() provides no
1020 * ordering guarantees for either the CPU or the compiler.
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -07001021 */
1022#define RCU_INIT_POINTER(p, v) \
Paul E. McKenneyd1b88eb2012-05-16 15:42:30 -07001023 do { \
Pranith Kumar1a6c9b22014-09-25 14:03:34 -04001024 rcu_dereference_sparse(p, __rcu); \
Peter Zijlstra155d1d12015-06-02 17:26:48 +02001025 WRITE_ONCE(p, RCU_INITIALIZER(v)); \
Paul E. McKenneyd1b88eb2012-05-16 15:42:30 -07001026 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
Paul E. McKenney172708d2012-05-16 15:23:45 -07001028/**
1029 * RCU_POINTER_INITIALIZER() - statically initialize an RCU protected pointer
1030 *
1031 * GCC-style initialization for an RCU-protected pointer in a structure field.
1032 */
1033#define RCU_POINTER_INITIALIZER(p, v) \
Paul E. McKenney462225ae2013-11-11 09:59:34 -08001034 .p = RCU_INITIALIZER(v)
Lai Jiangshan9ab15442011-03-18 11:15:47 +08001035
Jan Engelhardtd8169d42012-04-19 11:44:39 -07001036/*
1037 * Does the specified offset indicate that the corresponding rcu_head
1038 * structure can be handled by kfree_rcu()?
1039 */
1040#define __is_kfree_rcu_offset(offset) ((offset) < 4096)
1041
1042/*
1043 * Helper macro for kfree_rcu() to prevent argument-expansion eyestrain.
1044 */
1045#define __kfree_rcu(head, offset) \
1046 do { \
1047 BUILD_BUG_ON(!__is_kfree_rcu_offset(offset)); \
Paul E. McKenney4fa3b6c2012-06-05 15:53:53 -07001048 kfree_call_rcu(head, (void (*)(struct rcu_head *))(unsigned long)(offset)); \
Jan Engelhardtd8169d42012-04-19 11:44:39 -07001049 } while (0)
1050
Lai Jiangshan9ab15442011-03-18 11:15:47 +08001051/**
1052 * kfree_rcu() - kfree an object after a grace period.
1053 * @ptr: pointer to kfree
1054 * @rcu_head: the name of the struct rcu_head within the type of @ptr.
1055 *
1056 * Many rcu callbacks functions just call kfree() on the base structure.
1057 * These functions are trivial, but their size adds up, and furthermore
1058 * when they are used in a kernel module, that module must invoke the
1059 * high-latency rcu_barrier() function at module-unload time.
1060 *
1061 * The kfree_rcu() function handles this issue. Rather than encoding a
1062 * function address in the embedded rcu_head structure, kfree_rcu() instead
1063 * encodes the offset of the rcu_head structure within the base structure.
1064 * Because the functions are not allowed in the low-order 4096 bytes of
1065 * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
1066 * If the offset is larger than 4095 bytes, a compile-time error will
1067 * be generated in __kfree_rcu(). If this error is triggered, you can
1068 * either fall back to use of call_rcu() or rearrange the structure to
1069 * position the rcu_head structure into the first 4096 bytes.
1070 *
1071 * Note that the allowable offset might decrease in the future, for example,
1072 * to allow something like kmem_cache_free_rcu().
Jan Engelhardtd8169d42012-04-19 11:44:39 -07001073 *
1074 * The BUILD_BUG_ON check must not involve any function calls, hence the
1075 * checks are done in macros here.
Lai Jiangshan9ab15442011-03-18 11:15:47 +08001076 */
1077#define kfree_rcu(ptr, rcu_head) \
1078 __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
1079
Paul E. McKenney3382adb2015-03-04 15:41:24 -08001080#ifdef CONFIG_TINY_RCU
Thomas Gleixnerc1ad3482015-04-14 21:08:58 +00001081static inline int rcu_needs_cpu(u64 basemono, u64 *nextevt)
Paul E. McKenneyffa83fb2013-11-17 19:27:16 -08001082{
Thomas Gleixnerc1ad3482015-04-14 21:08:58 +00001083 *nextevt = KTIME_MAX;
Paul E. McKenneyffa83fb2013-11-17 19:27:16 -08001084 return 0;
1085}
Paul E. McKenney3382adb2015-03-04 15:41:24 -08001086#endif /* #ifdef CONFIG_TINY_RCU */
Paul E. McKenneyffa83fb2013-11-17 19:27:16 -08001087
Paul E. McKenney2f33b512013-11-17 18:25:48 -08001088#if defined(CONFIG_RCU_NOCB_CPU_ALL)
1089static inline bool rcu_is_nocb_cpu(int cpu) { return true; }
1090#elif defined(CONFIG_RCU_NOCB_CPU)
Teodora Baluta584dc4c2013-11-11 17:11:23 +02001091bool rcu_is_nocb_cpu(int cpu);
Frederic Weisbeckerd1e43fa2013-03-26 23:47:24 +01001092#else
1093static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
Paul E. McKenney2f33b512013-11-17 18:25:48 -08001094#endif
Frederic Weisbeckerd1e43fa2013-03-26 23:47:24 +01001095
1096
Paul E. McKenney0edd1b12013-06-21 16:37:22 -07001097/* Only for use by adaptive-ticks code. */
1098#ifdef CONFIG_NO_HZ_FULL_SYSIDLE
Teodora Baluta584dc4c2013-11-11 17:11:23 +02001099bool rcu_sys_is_idle(void);
1100void rcu_sysidle_force_exit(void);
Paul E. McKenney0edd1b12013-06-21 16:37:22 -07001101#else /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1102
1103static inline bool rcu_sys_is_idle(void)
1104{
1105 return false;
1106}
1107
1108static inline void rcu_sysidle_force_exit(void)
1109{
1110}
1111
1112#endif /* #else #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
1113
1114
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115#endif /* __LINUX_RCUPDATE_H */