blob: 2c7b8457d0d234203517f7aa6a1ddc6b724ef447 [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
Paul E. McKenneyfbf6bfc2008-02-13 15:03:15 -080061/*
62 * Awaken the corresponding synchronize_rcu() instance now that a
63 * grace period has elapsed.
64 */
Paul E. McKenney4446a362008-05-12 21:21:05 +020065void wakeme_after_rcu(struct rcu_head *head)
Dipankar Sarma21a1ea92006-03-07 21:55:33 -080066{
Paul E. McKenney01c1c662008-01-25 21:08:24 +010067 struct rcu_synchronize *rcu;
68
69 rcu = container_of(head, struct rcu_synchronize, head);
70 complete(&rcu->completion);
Dipankar Sarma21a1ea92006-03-07 21:55:33 -080071}
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73/**
Paul E. McKenney01c1c662008-01-25 21:08:24 +010074 * synchronize_rcu - wait until a grace period has elapsed.
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 *
Paul E. McKenney01c1c662008-01-25 21:08:24 +010076 * Control will return to the caller some time after a full grace
77 * period has elapsed, in other words after all currently executing RCU
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 * read-side critical sections have completed. RCU read-side critical
79 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
80 * and may be nested.
81 */
Paul E. McKenneyea7d3fe2009-01-04 13:03:02 -080082void synchronize_rcu(void)
83{
84 struct rcu_synchronize rcu;
Paul E. McKenneya6826042009-02-25 18:03:42 -080085
86 if (rcu_blocking_is_gp())
87 return;
88
Paul E. McKenneyea7d3fe2009-01-04 13:03:02 -080089 init_completion(&rcu.completion);
90 /* Will wake me after RCU finished. */
91 call_rcu(&rcu.head, wakeme_after_rcu);
92 /* Wait for it. */
93 wait_for_completion(&rcu.completion);
94}
Paul E. McKenney01c1c662008-01-25 21:08:24 +010095EXPORT_SYMBOL_GPL(synchronize_rcu);
Paul E. McKenneyc32e0662006-06-27 02:54:04 -070096
Dipankar Sarmaab4720e2005-12-12 00:37:05 -080097static void rcu_barrier_callback(struct rcu_head *notused)
98{
99 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
100 complete(&rcu_barrier_completion);
101}
102
103/*
104 * Called with preemption disabled, and from cross-cpu IRQ context.
105 */
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200106static void rcu_barrier_func(void *type)
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800107{
108 int cpu = smp_processor_id();
Paul E. McKenney01c1c662008-01-25 21:08:24 +0100109 struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800110
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800111 atomic_inc(&rcu_barrier_cpu_count);
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200112 switch ((enum rcu_barrier)type) {
113 case RCU_BARRIER_STD:
114 call_rcu(head, rcu_barrier_callback);
115 break;
116 case RCU_BARRIER_BH:
117 call_rcu_bh(head, rcu_barrier_callback);
118 break;
119 case RCU_BARRIER_SCHED:
120 call_rcu_sched(head, rcu_barrier_callback);
121 break;
122 }
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800123}
124
Lai Jiangshanf69b17d2009-03-20 17:40:06 +0800125static inline void wait_migrated_callbacks(void);
126
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200127/*
128 * Orchestrate the specified type of RCU barrier, waiting for all
129 * RCU callbacks of the specified type to complete.
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800130 */
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200131static void _rcu_barrier(enum rcu_barrier type)
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800132{
133 BUG_ON(in_interrupt());
Ingo Molnar9331b312006-03-23 03:00:19 -0800134 /* Take cpucontrol mutex to protect against CPU hotplug */
135 mutex_lock(&rcu_barrier_mutex);
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800136 init_completion(&rcu_barrier_completion);
Paul E. McKenneye0ecfa72008-01-25 21:08:24 +0100137 /*
Lai Jiangshan5f865152008-10-17 14:40:30 +0800138 * Initialize rcu_barrier_cpu_count to 1, then invoke
139 * rcu_barrier_func() on each CPU, so that each CPU also has
140 * incremented rcu_barrier_cpu_count. Only then is it safe to
141 * decrement rcu_barrier_cpu_count -- otherwise the first CPU
142 * might complete its grace period before all of the other CPUs
143 * did their increment, causing this function to return too
144 * early.
Paul E. McKenneye0ecfa72008-01-25 21:08:24 +0100145 */
Lai Jiangshan5f865152008-10-17 14:40:30 +0800146 atomic_set(&rcu_barrier_cpu_count, 1);
Linus Torvalds59190f42008-07-15 14:02:33 -0700147 on_each_cpu(rcu_barrier_func, (void *)type, 1);
Lai Jiangshan5f865152008-10-17 14:40:30 +0800148 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
149 complete(&rcu_barrier_completion);
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800150 wait_for_completion(&rcu_barrier_completion);
Ingo Molnar9331b312006-03-23 03:00:19 -0800151 mutex_unlock(&rcu_barrier_mutex);
Lai Jiangshanf69b17d2009-03-20 17:40:06 +0800152 wait_migrated_callbacks();
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800153}
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200154
155/**
156 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
157 */
158void rcu_barrier(void)
159{
160 _rcu_barrier(RCU_BARRIER_STD);
161}
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800162EXPORT_SYMBOL_GPL(rcu_barrier);
163
Paul E. McKenney70f12f82008-05-12 21:21:05 +0200164/**
165 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
166 */
167void rcu_barrier_bh(void)
168{
169 _rcu_barrier(RCU_BARRIER_BH);
170}
171EXPORT_SYMBOL_GPL(rcu_barrier_bh);
172
173/**
174 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
175 */
176void rcu_barrier_sched(void)
177{
178 _rcu_barrier(RCU_BARRIER_SCHED);
179}
180EXPORT_SYMBOL_GPL(rcu_barrier_sched);
181
Lai Jiangshanf69b17d2009-03-20 17:40:06 +0800182static atomic_t rcu_migrate_type_count = ATOMIC_INIT(0);
183static struct rcu_head rcu_migrate_head[3];
184static DECLARE_WAIT_QUEUE_HEAD(rcu_migrate_wq);
185
186static void rcu_migrate_callback(struct rcu_head *notused)
187{
188 if (atomic_dec_and_test(&rcu_migrate_type_count))
189 wake_up(&rcu_migrate_wq);
190}
191
192static inline void wait_migrated_callbacks(void)
193{
194 wait_event(rcu_migrate_wq, !atomic_read(&rcu_migrate_type_count));
195}
196
197static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self,
198 unsigned long action, void *hcpu)
199{
200 if (action == CPU_DYING) {
201 /*
202 * preempt_disable() in on_each_cpu() prevents stop_machine(),
203 * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
204 * returns, all online cpus have queued rcu_barrier_func(),
205 * and the dead cpu(if it exist) queues rcu_migrate_callback()s.
206 *
207 * These callbacks ensure _rcu_barrier() waits for all
208 * RCU callbacks of the specified type to complete.
209 */
210 atomic_set(&rcu_migrate_type_count, 3);
211 call_rcu_bh(rcu_migrate_head, rcu_migrate_callback);
212 call_rcu_sched(rcu_migrate_head + 1, rcu_migrate_callback);
213 call_rcu(rcu_migrate_head + 2, rcu_migrate_callback);
214 } else if (action == CPU_POST_DEAD) {
215 /* rcu_migrate_head is protected by cpu_add_remove_lock */
216 wait_migrated_callbacks();
217 }
218
219 return NOTIFY_OK;
220}
221
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222void __init rcu_init(void)
223{
Paul E. McKenney01c1c662008-01-25 21:08:24 +0100224 __rcu_init();
Lai Jiangshanf69b17d2009-03-20 17:40:06 +0800225 hotcpu_notifier(rcu_barrier_cpu_hotplug, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
227
Paul E. McKenneya6826042009-02-25 18:03:42 -0800228void rcu_scheduler_starting(void)
229{
230 WARN_ON(num_online_cpus() != 1);
231 WARN_ON(nr_context_switches() > 0);
232 rcu_scheduler_active = 1;
233}