Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | * |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 18 | * Copyright IBM Corporation, 2001 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 19 | * |
| 20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> |
| 21 | * Manfred Spraul <manfred@colorfullife.com> |
Paul E. McKenney | a71fca5 | 2009-09-18 10:28:19 -0700 | [diff] [blame] | 22 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 - |
Paul E. McKenney | a71fca5 | 2009-09-18 10:28:19 -0700 | [diff] [blame] | 30 | * http://lse.sourceforge.net/locking/rcupdate.html |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | #include <linux/percpu.h> |
| 43 | #include <linux/notifier.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 44 | #include <linux/cpu.h> |
Ingo Molnar | 9331b31 | 2006-03-23 03:00:19 -0800 | [diff] [blame] | 45 | #include <linux/mutex.h> |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 46 | #include <linux/module.h> |
Paul E. McKenney | a682604 | 2009-02-25 18:03:42 -0800 | [diff] [blame] | 47 | #include <linux/kernel_stat.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 48 | |
Paul E. McKenney | 70f12f8 | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 49 | enum rcu_barrier { |
| 50 | RCU_BARRIER_STD, |
| 51 | RCU_BARRIER_BH, |
| 52 | RCU_BARRIER_SCHED, |
| 53 | }; |
| 54 | |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 55 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; |
Dipankar Sarma | 21a1ea9 | 2006-03-07 21:55:33 -0800 | [diff] [blame] | 56 | static atomic_t rcu_barrier_cpu_count; |
Ingo Molnar | 9331b31 | 2006-03-23 03:00:19 -0800 | [diff] [blame] | 57 | static DEFINE_MUTEX(rcu_barrier_mutex); |
Dipankar Sarma | 21a1ea9 | 2006-03-07 21:55:33 -0800 | [diff] [blame] | 58 | static struct completion rcu_barrier_completion; |
Paul E. McKenney | a682604 | 2009-02-25 18:03:42 -0800 | [diff] [blame] | 59 | int rcu_scheduler_active __read_mostly; |
Dipankar Sarma | 21a1ea9 | 2006-03-07 21:55:33 -0800 | [diff] [blame] | 60 | |
David Howells | 5b1d07e | 2009-04-15 19:35:01 +0100 | [diff] [blame] | 61 | static atomic_t rcu_migrate_type_count = ATOMIC_INIT(0); |
| 62 | static struct rcu_head rcu_migrate_head[3]; |
| 63 | static DECLARE_WAIT_QUEUE_HEAD(rcu_migrate_wq); |
| 64 | |
Paul E. McKenney | fbf6bfc | 2008-02-13 15:03:15 -0800 | [diff] [blame] | 65 | /* |
| 66 | * Awaken the corresponding synchronize_rcu() instance now that a |
| 67 | * grace period has elapsed. |
| 68 | */ |
Paul E. McKenney | 4446a36 | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 69 | void wakeme_after_rcu(struct rcu_head *head) |
Dipankar Sarma | 21a1ea9 | 2006-03-07 21:55:33 -0800 | [diff] [blame] | 70 | { |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 71 | struct rcu_synchronize *rcu; |
| 72 | |
| 73 | rcu = container_of(head, struct rcu_synchronize, head); |
| 74 | complete(&rcu->completion); |
Dipankar Sarma | 21a1ea9 | 2006-03-07 21:55:33 -0800 | [diff] [blame] | 75 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | |
Paul E. McKenney | 16e3081 | 2009-09-13 09:15:11 -0700 | [diff] [blame] | 77 | #ifdef CONFIG_TREE_PREEMPT_RCU |
| 78 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 79 | /** |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 80 | * synchronize_rcu - wait until a grace period has elapsed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 81 | * |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 82 | * Control will return to the caller some time after a full grace |
| 83 | * period has elapsed, in other words after all currently executing RCU |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | * read-side critical sections have completed. RCU read-side critical |
| 85 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), |
| 86 | * and may be nested. |
| 87 | */ |
Paul E. McKenney | ea7d3fe | 2009-01-04 13:03:02 -0800 | [diff] [blame] | 88 | void synchronize_rcu(void) |
| 89 | { |
| 90 | struct rcu_synchronize rcu; |
Paul E. McKenney | a682604 | 2009-02-25 18:03:42 -0800 | [diff] [blame] | 91 | |
Paul E. McKenney | 16e3081 | 2009-09-13 09:15:11 -0700 | [diff] [blame] | 92 | if (!rcu_scheduler_active) |
Paul E. McKenney | a682604 | 2009-02-25 18:03:42 -0800 | [diff] [blame] | 93 | return; |
| 94 | |
Paul E. McKenney | ea7d3fe | 2009-01-04 13:03:02 -0800 | [diff] [blame] | 95 | init_completion(&rcu.completion); |
| 96 | /* Will wake me after RCU finished. */ |
| 97 | call_rcu(&rcu.head, wakeme_after_rcu); |
| 98 | /* Wait for it. */ |
| 99 | wait_for_completion(&rcu.completion); |
| 100 | } |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 101 | EXPORT_SYMBOL_GPL(synchronize_rcu); |
Paul E. McKenney | c32e066 | 2006-06-27 02:54:04 -0700 | [diff] [blame] | 102 | |
Paul E. McKenney | 16e3081 | 2009-09-13 09:15:11 -0700 | [diff] [blame] | 103 | #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */ |
| 104 | |
| 105 | /** |
| 106 | * synchronize_sched - wait until an rcu-sched grace period has elapsed. |
| 107 | * |
| 108 | * Control will return to the caller some time after a full rcu-sched |
| 109 | * grace period has elapsed, in other words after all currently executing |
| 110 | * rcu-sched read-side critical sections have completed. These read-side |
| 111 | * critical sections are delimited by rcu_read_lock_sched() and |
| 112 | * rcu_read_unlock_sched(), and may be nested. Note that preempt_disable(), |
| 113 | * local_irq_disable(), and so on may be used in place of |
| 114 | * rcu_read_lock_sched(). |
| 115 | * |
| 116 | * This means that all preempt_disable code sequences, including NMI and |
| 117 | * hardware-interrupt handlers, in progress on entry will have completed |
| 118 | * before this primitive returns. However, this does not guarantee that |
| 119 | * softirq handlers will have completed, since in some kernels, these |
| 120 | * handlers can run in process context, and can block. |
| 121 | * |
| 122 | * This primitive provides the guarantees made by the (now removed) |
| 123 | * synchronize_kernel() API. In contrast, synchronize_rcu() only |
| 124 | * guarantees that rcu_read_lock() sections will have completed. |
| 125 | * In "classic RCU", these two guarantees happen to be one and |
| 126 | * the same, but can differ in realtime RCU implementations. |
| 127 | */ |
| 128 | void synchronize_sched(void) |
| 129 | { |
| 130 | struct rcu_synchronize rcu; |
| 131 | |
| 132 | if (rcu_blocking_is_gp()) |
| 133 | return; |
| 134 | |
| 135 | init_completion(&rcu.completion); |
| 136 | /* Will wake me after RCU finished. */ |
| 137 | call_rcu_sched(&rcu.head, wakeme_after_rcu); |
| 138 | /* Wait for it. */ |
| 139 | wait_for_completion(&rcu.completion); |
| 140 | } |
| 141 | EXPORT_SYMBOL_GPL(synchronize_sched); |
| 142 | |
Paul E. McKenney | 03b042b | 2009-06-25 09:08:16 -0700 | [diff] [blame] | 143 | /** |
| 144 | * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed. |
| 145 | * |
| 146 | * Control will return to the caller some time after a full rcu_bh grace |
| 147 | * period has elapsed, in other words after all currently executing rcu_bh |
| 148 | * read-side critical sections have completed. RCU read-side critical |
| 149 | * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(), |
| 150 | * and may be nested. |
| 151 | */ |
| 152 | void synchronize_rcu_bh(void) |
| 153 | { |
| 154 | struct rcu_synchronize rcu; |
| 155 | |
| 156 | if (rcu_blocking_is_gp()) |
| 157 | return; |
| 158 | |
| 159 | init_completion(&rcu.completion); |
| 160 | /* Will wake me after RCU finished. */ |
| 161 | call_rcu_bh(&rcu.head, wakeme_after_rcu); |
| 162 | /* Wait for it. */ |
| 163 | wait_for_completion(&rcu.completion); |
| 164 | } |
| 165 | EXPORT_SYMBOL_GPL(synchronize_rcu_bh); |
| 166 | |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 167 | static void rcu_barrier_callback(struct rcu_head *notused) |
| 168 | { |
| 169 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) |
| 170 | complete(&rcu_barrier_completion); |
| 171 | } |
| 172 | |
| 173 | /* |
| 174 | * Called with preemption disabled, and from cross-cpu IRQ context. |
| 175 | */ |
Paul E. McKenney | 70f12f8 | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 176 | static void rcu_barrier_func(void *type) |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 177 | { |
| 178 | int cpu = smp_processor_id(); |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 179 | struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu); |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 180 | |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 181 | atomic_inc(&rcu_barrier_cpu_count); |
Paul E. McKenney | 70f12f8 | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 182 | switch ((enum rcu_barrier)type) { |
| 183 | case RCU_BARRIER_STD: |
| 184 | call_rcu(head, rcu_barrier_callback); |
| 185 | break; |
| 186 | case RCU_BARRIER_BH: |
| 187 | call_rcu_bh(head, rcu_barrier_callback); |
| 188 | break; |
| 189 | case RCU_BARRIER_SCHED: |
| 190 | call_rcu_sched(head, rcu_barrier_callback); |
| 191 | break; |
| 192 | } |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 193 | } |
| 194 | |
David Howells | 5b1d07e | 2009-04-15 19:35:01 +0100 | [diff] [blame] | 195 | static inline void wait_migrated_callbacks(void) |
| 196 | { |
| 197 | wait_event(rcu_migrate_wq, !atomic_read(&rcu_migrate_type_count)); |
Paul E. McKenney | 03b042b | 2009-06-25 09:08:16 -0700 | [diff] [blame] | 198 | smp_mb(); /* In case we didn't sleep. */ |
David Howells | 5b1d07e | 2009-04-15 19:35:01 +0100 | [diff] [blame] | 199 | } |
Lai Jiangshan | f69b17d | 2009-03-20 17:40:06 +0800 | [diff] [blame] | 200 | |
Paul E. McKenney | 70f12f8 | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 201 | /* |
| 202 | * Orchestrate the specified type of RCU barrier, waiting for all |
| 203 | * RCU callbacks of the specified type to complete. |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 204 | */ |
Paul E. McKenney | 70f12f8 | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 205 | static void _rcu_barrier(enum rcu_barrier type) |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 206 | { |
| 207 | BUG_ON(in_interrupt()); |
Ingo Molnar | 9331b31 | 2006-03-23 03:00:19 -0800 | [diff] [blame] | 208 | /* Take cpucontrol mutex to protect against CPU hotplug */ |
| 209 | mutex_lock(&rcu_barrier_mutex); |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 210 | init_completion(&rcu_barrier_completion); |
Paul E. McKenney | e0ecfa7 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 211 | /* |
Lai Jiangshan | 5f86515 | 2008-10-17 14:40:30 +0800 | [diff] [blame] | 212 | * Initialize rcu_barrier_cpu_count to 1, then invoke |
| 213 | * rcu_barrier_func() on each CPU, so that each CPU also has |
| 214 | * incremented rcu_barrier_cpu_count. Only then is it safe to |
| 215 | * decrement rcu_barrier_cpu_count -- otherwise the first CPU |
| 216 | * might complete its grace period before all of the other CPUs |
| 217 | * did their increment, causing this function to return too |
| 218 | * early. |
Paul E. McKenney | e0ecfa7 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 219 | */ |
Lai Jiangshan | 5f86515 | 2008-10-17 14:40:30 +0800 | [diff] [blame] | 220 | atomic_set(&rcu_barrier_cpu_count, 1); |
Linus Torvalds | 59190f4 | 2008-07-15 14:02:33 -0700 | [diff] [blame] | 221 | on_each_cpu(rcu_barrier_func, (void *)type, 1); |
Lai Jiangshan | 5f86515 | 2008-10-17 14:40:30 +0800 | [diff] [blame] | 222 | if (atomic_dec_and_test(&rcu_barrier_cpu_count)) |
| 223 | complete(&rcu_barrier_completion); |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 224 | wait_for_completion(&rcu_barrier_completion); |
Ingo Molnar | 9331b31 | 2006-03-23 03:00:19 -0800 | [diff] [blame] | 225 | mutex_unlock(&rcu_barrier_mutex); |
Lai Jiangshan | f69b17d | 2009-03-20 17:40:06 +0800 | [diff] [blame] | 226 | wait_migrated_callbacks(); |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 227 | } |
Paul E. McKenney | 70f12f8 | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 228 | |
| 229 | /** |
| 230 | * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete. |
| 231 | */ |
| 232 | void rcu_barrier(void) |
| 233 | { |
| 234 | _rcu_barrier(RCU_BARRIER_STD); |
| 235 | } |
Dipankar Sarma | ab4720e | 2005-12-12 00:37:05 -0800 | [diff] [blame] | 236 | EXPORT_SYMBOL_GPL(rcu_barrier); |
| 237 | |
Paul E. McKenney | 70f12f8 | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 238 | /** |
| 239 | * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete. |
| 240 | */ |
| 241 | void rcu_barrier_bh(void) |
| 242 | { |
| 243 | _rcu_barrier(RCU_BARRIER_BH); |
| 244 | } |
| 245 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); |
| 246 | |
| 247 | /** |
| 248 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. |
| 249 | */ |
| 250 | void rcu_barrier_sched(void) |
| 251 | { |
| 252 | _rcu_barrier(RCU_BARRIER_SCHED); |
| 253 | } |
| 254 | EXPORT_SYMBOL_GPL(rcu_barrier_sched); |
| 255 | |
Lai Jiangshan | f69b17d | 2009-03-20 17:40:06 +0800 | [diff] [blame] | 256 | static void rcu_migrate_callback(struct rcu_head *notused) |
| 257 | { |
| 258 | if (atomic_dec_and_test(&rcu_migrate_type_count)) |
| 259 | wake_up(&rcu_migrate_wq); |
| 260 | } |
| 261 | |
Paul E. McKenney | 2e59755 | 2009-08-15 09:53:48 -0700 | [diff] [blame] | 262 | extern int rcu_cpu_notify(struct notifier_block *self, |
| 263 | unsigned long action, void *hcpu); |
| 264 | |
Lai Jiangshan | f69b17d | 2009-03-20 17:40:06 +0800 | [diff] [blame] | 265 | static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self, |
| 266 | unsigned long action, void *hcpu) |
| 267 | { |
Paul E. McKenney | 2e59755 | 2009-08-15 09:53:48 -0700 | [diff] [blame] | 268 | rcu_cpu_notify(self, action, hcpu); |
Lai Jiangshan | f69b17d | 2009-03-20 17:40:06 +0800 | [diff] [blame] | 269 | if (action == CPU_DYING) { |
| 270 | /* |
| 271 | * preempt_disable() in on_each_cpu() prevents stop_machine(), |
| 272 | * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);" |
| 273 | * returns, all online cpus have queued rcu_barrier_func(), |
| 274 | * and the dead cpu(if it exist) queues rcu_migrate_callback()s. |
| 275 | * |
| 276 | * These callbacks ensure _rcu_barrier() waits for all |
| 277 | * RCU callbacks of the specified type to complete. |
| 278 | */ |
| 279 | atomic_set(&rcu_migrate_type_count, 3); |
| 280 | call_rcu_bh(rcu_migrate_head, rcu_migrate_callback); |
| 281 | call_rcu_sched(rcu_migrate_head + 1, rcu_migrate_callback); |
| 282 | call_rcu(rcu_migrate_head + 2, rcu_migrate_callback); |
Paul E. McKenney | 1423cc0 | 2009-08-18 23:06:14 -0700 | [diff] [blame] | 283 | } else if (action == CPU_DOWN_PREPARE) { |
| 284 | /* Don't need to wait until next removal operation. */ |
Lai Jiangshan | f69b17d | 2009-03-20 17:40:06 +0800 | [diff] [blame] | 285 | /* rcu_migrate_head is protected by cpu_add_remove_lock */ |
| 286 | wait_migrated_callbacks(); |
| 287 | } |
| 288 | |
| 289 | return NOTIFY_OK; |
| 290 | } |
| 291 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | void __init rcu_init(void) |
| 293 | { |
Paul E. McKenney | 2e59755 | 2009-08-15 09:53:48 -0700 | [diff] [blame] | 294 | int i; |
| 295 | |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 296 | __rcu_init(); |
Paul E. McKenney | 2e59755 | 2009-08-15 09:53:48 -0700 | [diff] [blame] | 297 | cpu_notifier(rcu_barrier_cpu_hotplug, 0); |
| 298 | |
| 299 | /* |
| 300 | * We don't need protection against CPU-hotplug here because |
| 301 | * this is called early in boot, before either interrupts |
| 302 | * or the scheduler are operational. |
| 303 | */ |
| 304 | for_each_online_cpu(i) |
| 305 | rcu_barrier_cpu_hotplug(NULL, CPU_UP_PREPARE, (void *)(long)i); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 306 | } |
| 307 | |
Paul E. McKenney | a682604 | 2009-02-25 18:03:42 -0800 | [diff] [blame] | 308 | void rcu_scheduler_starting(void) |
| 309 | { |
| 310 | WARN_ON(num_online_cpus() != 1); |
| 311 | WARN_ON(nr_context_switches() > 0); |
| 312 | rcu_scheduler_active = 1; |
| 313 | } |