kernel/rcupdate.c

   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  *
  18  * Copyright IBM Corporation, 2001
  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>
  42 #include <linux/percpu.h>
  43 #include <linux/notifier.h>
  44 #include <linux/cpu.h>
  45 #include <linux/mutex.h>
  46 #include <linux/module.h>
  47 #include <linux/kernel_stat.h>
  48
  49 enum rcu_barrier {
  50         RCU_BARRIER_STD,
  51         RCU_BARRIER_BH,
  52         RCU_BARRIER_SCHED,
  53 };
  54
  55 static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
  56 static atomic_t rcu_barrier_cpu_count;
  57 static DEFINE_MUTEX(rcu_barrier_mutex);
  58 static struct completion rcu_barrier_completion;
  59 int rcu_scheduler_active __read_mostly;
  60
  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
  65 /*
  66  * Awaken the corresponding synchronize_rcu() instance now that a
  67  * grace period has elapsed.
  68  */
  69 void wakeme_after_rcu(struct rcu_head  *head)
  70 {
  71         struct rcu_synchronize *rcu;
  72
  73         rcu = container_of(head, struct rcu_synchronize, head);
  74         complete(&rcu->completion);
  75 }
  76
  77 /**
  78  * synchronize_rcu - wait until a grace period has elapsed.
  79  *
  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
  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  */
  86 void synchronize_rcu(void)
  87 {
  88         struct rcu_synchronize rcu;
  89
  90         if (rcu_blocking_is_gp())
  91                 return;
  92
  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 }
  99 EXPORT_SYMBOL_GPL(synchronize_rcu);
 100
 101 /**
 102  * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
 103  *
 104  * Control will return to the caller some time after a full rcu_bh grace
 105  * period has elapsed, in other words after all currently executing rcu_bh
 106  * read-side critical sections have completed.  RCU read-side critical
 107  * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(),
 108  * and may be nested.
 109  */
 110 void synchronize_rcu_bh(void)
 111 {
 112         struct rcu_synchronize rcu;
 113
 114         if (rcu_blocking_is_gp())
 115                 return;
 116
 117         init_completion(&rcu.completion);
 118         /* Will wake me after RCU finished. */
 119         call_rcu_bh(&rcu.head, wakeme_after_rcu);
 120         /* Wait for it. */
 121         wait_for_completion(&rcu.completion);
 122 }
 123 EXPORT_SYMBOL_GPL(synchronize_rcu_bh);
 124
 125 static void rcu_barrier_callback(struct rcu_head *notused)
 126 {
 127         if (atomic_dec_and_test(&rcu_barrier_cpu_count))
 128                 complete(&rcu_barrier_completion);
 129 }
 130
 131 /*
 132  * Called with preemption disabled, and from cross-cpu IRQ context.
 133  */
 134 static void rcu_barrier_func(void *type)
 135 {
 136         int cpu = smp_processor_id();
 137         struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
 138
 139         atomic_inc(&rcu_barrier_cpu_count);
 140         switch ((enum rcu_barrier)type) {
 141         case RCU_BARRIER_STD:
 142                 call_rcu(head, rcu_barrier_callback);
 143                 break;
 144         case RCU_BARRIER_BH:
 145                 call_rcu_bh(head, rcu_barrier_callback);
 146                 break;
 147         case RCU_BARRIER_SCHED:
 148                 call_rcu_sched(head, rcu_barrier_callback);
 149                 break;
 150         }
 151 }
 152
 153 static inline void wait_migrated_callbacks(void)
 154 {
 155         wait_event(rcu_migrate_wq, !atomic_read(&rcu_migrate_type_count));
 156         smp_mb(); /* In case we didn't sleep. */
 157 }
 158
 159 /*
 160  * Orchestrate the specified type of RCU barrier, waiting for all
 161  * RCU callbacks of the specified type to complete.
 162  */
 163 static void _rcu_barrier(enum rcu_barrier type)
 164 {
 165         BUG_ON(in_interrupt());
 166         /* Take cpucontrol mutex to protect against CPU hotplug */
 167         mutex_lock(&rcu_barrier_mutex);
 168         init_completion(&rcu_barrier_completion);
 169         /*
 170          * Initialize rcu_barrier_cpu_count to 1, then invoke
 171          * rcu_barrier_func() on each CPU, so that each CPU also has
 172          * incremented rcu_barrier_cpu_count.  Only then is it safe to
 173          * decrement rcu_barrier_cpu_count -- otherwise the first CPU
 174          * might complete its grace period before all of the other CPUs
 175          * did their increment, causing this function to return too
 176          * early.
 177          */
 178         atomic_set(&rcu_barrier_cpu_count, 1);
 179         on_each_cpu(rcu_barrier_func, (void *)type, 1);
 180         if (atomic_dec_and_test(&rcu_barrier_cpu_count))
 181                 complete(&rcu_barrier_completion);
 182         wait_for_completion(&rcu_barrier_completion);
 183         mutex_unlock(&rcu_barrier_mutex);
 184         wait_migrated_callbacks();
 185 }
 186
 187 /**
 188  * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
 189  */
 190 void rcu_barrier(void)
 191 {
 192         _rcu_barrier(RCU_BARRIER_STD);
 193 }
 194 EXPORT_SYMBOL_GPL(rcu_barrier);
 195
 196 /**
 197  * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
 198  */
 199 void rcu_barrier_bh(void)
 200 {
 201         _rcu_barrier(RCU_BARRIER_BH);
 202 }
 203 EXPORT_SYMBOL_GPL(rcu_barrier_bh);
 204
 205 /**
 206  * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
 207  */
 208 void rcu_barrier_sched(void)
 209 {
 210         _rcu_barrier(RCU_BARRIER_SCHED);
 211 }
 212 EXPORT_SYMBOL_GPL(rcu_barrier_sched);
 213
 214 static void rcu_migrate_callback(struct rcu_head *notused)
 215 {
 216         if (atomic_dec_and_test(&rcu_migrate_type_count))
 217                 wake_up(&rcu_migrate_wq);
 218 }
 219
 220 extern int rcu_cpu_notify(struct notifier_block *self,
 221                           unsigned long action, void *hcpu);
 222
 223 static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self,
 224                 unsigned long action, void *hcpu)
 225 {
 226         rcu_cpu_notify(self, action, hcpu);
 227         if (action == CPU_DYING) {
 228                 /*
 229                  * preempt_disable() in on_each_cpu() prevents stop_machine(),
 230                  * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
 231                  * returns, all online cpus have queued rcu_barrier_func(),
 232                  * and the dead cpu(if it exist) queues rcu_migrate_callback()s.
 233                  *
 234                  * These callbacks ensure _rcu_barrier() waits for all
 235                  * RCU callbacks of the specified type to complete.
 236                  */
 237                 atomic_set(&rcu_migrate_type_count, 3);
 238                 call_rcu_bh(rcu_migrate_head, rcu_migrate_callback);
 239                 call_rcu_sched(rcu_migrate_head + 1, rcu_migrate_callback);
 240                 call_rcu(rcu_migrate_head + 2, rcu_migrate_callback);
 241         } else if (action == CPU_DOWN_PREPARE) {
 242                 /* Don't need to wait until next removal operation. */
 243                 /* rcu_migrate_head is protected by cpu_add_remove_lock */
 244                 wait_migrated_callbacks();
 245         }
 246
 247         return NOTIFY_OK;
 248 }
 249
 250 void __init rcu_init(void)
 251 {
 252         int i;
 253
 254         __rcu_init();
 255         cpu_notifier(rcu_barrier_cpu_hotplug, 0);
 256
 257         /*
 258          * We don't need protection against CPU-hotplug here because
 259          * this is called early in boot, before either interrupts
 260          * or the scheduler are operational.
 261          */
 262         for_each_online_cpu(i)
 263                 rcu_barrier_cpu_hotplug(NULL, CPU_UP_PREPARE, (void *)(long)i);
 264 }
 265
 266 void rcu_scheduler_starting(void)
 267 {
 268         WARN_ON(num_online_cpus() != 1);
 269         WARN_ON(nr_context_switches() > 0);
 270         rcu_scheduler_active = 1;
 271 }