blob: a967c9feb90a045f4f7f3631dbc93ad5aebf67f9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Read-Copy Update mechanism for mutual exclusion
3 *
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 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
21 * Manfred Spraul <manfred@colorfullife.com>
22 *
23 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
24 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
25 * Papers:
26 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
27 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
28 *
29 * For detailed explanation of Read-Copy Update mechanism see -
30 * http://lse.sourceforge.net/locking/rcupdate.html
31 *
32 */
33#include <linux/types.h>
34#include <linux/kernel.h>
35#include <linux/init.h>
36#include <linux/spinlock.h>
37#include <linux/smp.h>
38#include <linux/interrupt.h>
39#include <linux/sched.h>
40#include <asm/atomic.h>
41#include <linux/bitops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/percpu.h>
43#include <linux/notifier.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <linux/cpu.h>
Ingo Molnar9331b312006-03-23 03:00:19 -080045#include <linux/mutex.h>
Paul E. McKenney01c1c662008-01-25 21:08:24 +010046#include <linux/module.h>
Paul E. McKenneya6826042009-02-25 18:03:42 -080047#include <linux/kernel_stat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Paul E. McKenney70f12f82008-05-12 21:21:05 +020049enum rcu_barrier {
50 RCU_BARRIER_STD,
51 RCU_BARRIER_BH,
52 RCU_BARRIER_SCHED,
53};
54
Paul E. McKenney01c1c662008-01-25 21:08:24 +010055static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
Dipankar Sarma21a1ea92006-03-07 21:55:33 -080056static atomic_t rcu_barrier_cpu_count;
Ingo Molnar9331b312006-03-23 03:00:19 -080057static DEFINE_MUTEX(rcu_barrier_mutex);
Dipankar Sarma21a1ea92006-03-07 21:55:33 -080058static struct completion rcu_barrier_completion;
Paul E. McKenneya6826042009-02-25 18:03:42 -080059int rcu_scheduler_active __read_mostly;
Dipankar Sarma21a1ea92006-03-07 21:55:33 -080060
David Howells5b1d07e2009-04-15 19:35:01 +010061static atomic_t rcu_migrate_type_count = ATOMIC_INIT(0);
62static struct rcu_head rcu_migrate_head[3];
63static DECLARE_WAIT_QUEUE_HEAD(rcu_migrate_wq);
64
Paul E. McKenneyfbf6bfc2008-02-13 15:03:15 -080065/*
66 * Awaken the corresponding synchronize_rcu() instance now that a
67 * grace period has elapsed.
68 */
Paul E. McKenney4446a362008-05-12 21:21:05 +020069void wakeme_after_rcu(struct rcu_head *head)
Dipankar Sarma21a1ea92006-03-07 21:55:33 -080070{
Paul E. McKenney01c1c662008-01-25 21:08:24 +010071 struct rcu_synchronize *rcu;
72
73 rcu = container_of(head, struct rcu_synchronize, head);
74 complete(&rcu->completion);
Dipankar Sarma21a1ea92006-03-07 21:55:33 -080075}
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77/**
Paul E. McKenney01c1c662008-01-25 21:08:24 +010078 * synchronize_rcu - wait until a grace period has elapsed.
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 *
Paul E. McKenney01c1c662008-01-25 21:08:24 +010080 * Control will return to the caller some time after a full grace
81 * period has elapsed, in other words after all currently executing RCU
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 * read-side critical sections have completed. RCU read-side critical
83 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
84 * and may be nested.
85 */
Paul E. McKenneyea7d3fe2009-01-04 13:03:02 -080086void synchronize_rcu(void)
87{
88 struct rcu_synchronize rcu;
Paul E. McKenneya6826042009-02-25 18:03:42 -080089
90 if (rcu_blocking_is_gp())
91 return;
92
Paul E. McKenneyea7d3fe2009-01-04 13:03:02 -080093 init_completion(&rcu.completion);
94 /* Will wake me after RCU finished. */
95 call_rcu(&rcu.head, wakeme_after_rcu);
96 /* Wait for it. */
97 wait_for_completion(&rcu.completion);
98}
Paul E. McKenney01c1c662008-01-25 21:08:24 +010099EXPORT_SYMBOL_GPL(synchronize_rcu);
Paul E. McKenneyc32e0662006-06-27 02:54:04 -0700100
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800101static void rcu_barrier_callback(struct rcu_head *notused)
102{
103 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
104 complete(&rcu_barrier_completion);
105}
106
107/*
108 * Called with preemption disabled, and from cross-cpu IRQ context.
109 */
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200110static void rcu_barrier_func(void *type)
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800111{
112 int cpu = smp_processor_id();
Paul E. McKenney01c1c662008-01-25 21:08:24 +0100113 struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800114
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800115 atomic_inc(&rcu_barrier_cpu_count);
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200116 switch ((enum rcu_barrier)type) {
117 case RCU_BARRIER_STD:
118 call_rcu(head, rcu_barrier_callback);
119 break;
120 case RCU_BARRIER_BH:
121 call_rcu_bh(head, rcu_barrier_callback);
122 break;
123 case RCU_BARRIER_SCHED:
124 call_rcu_sched(head, rcu_barrier_callback);
125 break;
126 }
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800127}
128
David Howells5b1d07e2009-04-15 19:35:01 +0100129static inline void wait_migrated_callbacks(void)
130{
131 wait_event(rcu_migrate_wq, !atomic_read(&rcu_migrate_type_count));
132}
Lai Jiangshanf69b17d2009-03-20 17:40:06 +0800133
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200134/*
135 * Orchestrate the specified type of RCU barrier, waiting for all
136 * RCU callbacks of the specified type to complete.
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800137 */
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200138static void _rcu_barrier(enum rcu_barrier type)
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800139{
140 BUG_ON(in_interrupt());
Ingo Molnar9331b312006-03-23 03:00:19 -0800141 /* Take cpucontrol mutex to protect against CPU hotplug */
142 mutex_lock(&rcu_barrier_mutex);
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800143 init_completion(&rcu_barrier_completion);
Paul E. McKenneye0ecfa72008-01-25 21:08:24 +0100144 /*
Lai Jiangshan5f865152008-10-17 14:40:30 +0800145 * Initialize rcu_barrier_cpu_count to 1, then invoke
146 * rcu_barrier_func() on each CPU, so that each CPU also has
147 * incremented rcu_barrier_cpu_count. Only then is it safe to
148 * decrement rcu_barrier_cpu_count -- otherwise the first CPU
149 * might complete its grace period before all of the other CPUs
150 * did their increment, causing this function to return too
151 * early.
Paul E. McKenneye0ecfa72008-01-25 21:08:24 +0100152 */
Lai Jiangshan5f865152008-10-17 14:40:30 +0800153 atomic_set(&rcu_barrier_cpu_count, 1);
Linus Torvalds59190f42008-07-15 14:02:33 -0700154 on_each_cpu(rcu_barrier_func, (void *)type, 1);
Lai Jiangshan5f865152008-10-17 14:40:30 +0800155 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
156 complete(&rcu_barrier_completion);
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800157 wait_for_completion(&rcu_barrier_completion);
Ingo Molnar9331b312006-03-23 03:00:19 -0800158 mutex_unlock(&rcu_barrier_mutex);
Lai Jiangshanf69b17d2009-03-20 17:40:06 +0800159 wait_migrated_callbacks();
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800160}
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200161
162/**
163 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
164 */
165void rcu_barrier(void)
166{
167 _rcu_barrier(RCU_BARRIER_STD);
168}
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800169EXPORT_SYMBOL_GPL(rcu_barrier);
170
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200171/**
172 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
173 */
174void rcu_barrier_bh(void)
175{
176 _rcu_barrier(RCU_BARRIER_BH);
177}
178EXPORT_SYMBOL_GPL(rcu_barrier_bh);
179
180/**
181 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
182 */
183void rcu_barrier_sched(void)
184{
185 _rcu_barrier(RCU_BARRIER_SCHED);
186}
187EXPORT_SYMBOL_GPL(rcu_barrier_sched);
188
Lai Jiangshanf69b17d2009-03-20 17:40:06 +0800189static void rcu_migrate_callback(struct rcu_head *notused)
190{
191 if (atomic_dec_and_test(&rcu_migrate_type_count))
192 wake_up(&rcu_migrate_wq);
193}
194
Lai Jiangshanf69b17d2009-03-20 17:40:06 +0800195static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self,
196 unsigned long action, void *hcpu)
197{
198 if (action == CPU_DYING) {
199 /*
200 * preempt_disable() in on_each_cpu() prevents stop_machine(),
201 * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
202 * returns, all online cpus have queued rcu_barrier_func(),
203 * and the dead cpu(if it exist) queues rcu_migrate_callback()s.
204 *
205 * These callbacks ensure _rcu_barrier() waits for all
206 * RCU callbacks of the specified type to complete.
207 */
208 atomic_set(&rcu_migrate_type_count, 3);
209 call_rcu_bh(rcu_migrate_head, rcu_migrate_callback);
210 call_rcu_sched(rcu_migrate_head + 1, rcu_migrate_callback);
211 call_rcu(rcu_migrate_head + 2, rcu_migrate_callback);
212 } else if (action == CPU_POST_DEAD) {
213 /* rcu_migrate_head is protected by cpu_add_remove_lock */
214 wait_migrated_callbacks();
215 }
216
217 return NOTIFY_OK;
218}
219
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220void __init rcu_init(void)
221{
Paul E. McKenney01c1c662008-01-25 21:08:24 +0100222 __rcu_init();
Lai Jiangshanf69b17d2009-03-20 17:40:06 +0800223 hotcpu_notifier(rcu_barrier_cpu_hotplug, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
Paul E. McKenneya6826042009-02-25 18:03:42 -0800226void rcu_scheduler_starting(void)
227{
228 WARN_ON(num_online_cpus() != 1);
229 WARN_ON(nr_context_switches() > 0);
230 rcu_scheduler_active = 1;
231}