]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - kernel/rcupdate.c
Merge branch 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[mirror_ubuntu-bionic-kernel.git] / kernel / rcupdate.c
CommitLineData
1da177e4
LT
1/*
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 *
01c1c660 18 * Copyright IBM Corporation, 2001
1da177e4
LT
19 *
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>
1da177e4
LT
42#include <linux/percpu.h>
43#include <linux/notifier.h>
1da177e4 44#include <linux/cpu.h>
9331b315 45#include <linux/mutex.h>
01c1c660 46#include <linux/module.h>
a6826048 47#include <linux/kernel_stat.h>
1da177e4 48
70f12f84
PM
49enum rcu_barrier {
50 RCU_BARRIER_STD,
51 RCU_BARRIER_BH,
52 RCU_BARRIER_SCHED,
53};
54
01c1c660 55static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
21a1ea9e 56static atomic_t rcu_barrier_cpu_count;
9331b315 57static DEFINE_MUTEX(rcu_barrier_mutex);
21a1ea9e 58static struct completion rcu_barrier_completion;
a6826048 59int rcu_scheduler_active __read_mostly;
21a1ea9e 60
5b1d07ed
DH
61static 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
fbf6bfca
PM
65/*
66 * Awaken the corresponding synchronize_rcu() instance now that a
67 * grace period has elapsed.
68 */
4446a36f 69void wakeme_after_rcu(struct rcu_head *head)
21a1ea9e 70{
01c1c660
PM
71 struct rcu_synchronize *rcu;
72
73 rcu = container_of(head, struct rcu_synchronize, head);
74 complete(&rcu->completion);
21a1ea9e 75}
1da177e4
LT
76
77/**
01c1c660 78 * synchronize_rcu - wait until a grace period has elapsed.
1da177e4 79 *
01c1c660
PM
80 * Control will return to the caller some time after a full grace
81 * period has elapsed, in other words after all currently executing RCU
1da177e4
LT
82 * 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 */
ea7d3fef
PM
86void synchronize_rcu(void)
87{
88 struct rcu_synchronize rcu;
a6826048
PM
89
90 if (rcu_blocking_is_gp())
91 return;
92
ea7d3fef
PM
93 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}
01c1c660 99EXPORT_SYMBOL_GPL(synchronize_rcu);
c32e0660 100
ab4720ec
DS
101static 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 */
70f12f84 110static void rcu_barrier_func(void *type)
ab4720ec
DS
111{
112 int cpu = smp_processor_id();
01c1c660 113 struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
ab4720ec 114
ab4720ec 115 atomic_inc(&rcu_barrier_cpu_count);
70f12f84
PM
116 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 }
ab4720ec
DS
127}
128
5b1d07ed
DH
129static inline void wait_migrated_callbacks(void)
130{
131 wait_event(rcu_migrate_wq, !atomic_read(&rcu_migrate_type_count));
132}
f69b17d7 133
70f12f84
PM
134/*
135 * Orchestrate the specified type of RCU barrier, waiting for all
136 * RCU callbacks of the specified type to complete.
ab4720ec 137 */
70f12f84 138static void _rcu_barrier(enum rcu_barrier type)
ab4720ec
DS
139{
140 BUG_ON(in_interrupt());
9331b315
IM
141 /* Take cpucontrol mutex to protect against CPU hotplug */
142 mutex_lock(&rcu_barrier_mutex);
ab4720ec 143 init_completion(&rcu_barrier_completion);
e0ecfa79 144 /*
5f865151
LJ
145 * 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.
e0ecfa79 152 */
5f865151 153 atomic_set(&rcu_barrier_cpu_count, 1);
59190f42 154 on_each_cpu(rcu_barrier_func, (void *)type, 1);
5f865151
LJ
155 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
156 complete(&rcu_barrier_completion);
ab4720ec 157 wait_for_completion(&rcu_barrier_completion);
9331b315 158 mutex_unlock(&rcu_barrier_mutex);
f69b17d7 159 wait_migrated_callbacks();
ab4720ec 160}
70f12f84
PM
161
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}
ab4720ec
DS
169EXPORT_SYMBOL_GPL(rcu_barrier);
170
70f12f84
PM
171/**
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
f69b17d7
LJ
189static 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
f69b17d7
LJ
195static 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
1da177e4
LT
220void __init rcu_init(void)
221{
01c1c660 222 __rcu_init();
f69b17d7 223 hotcpu_notifier(rcu_barrier_cpu_hotplug, 0);
1da177e4
LT
224}
225
a6826048
PM
226void 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}