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 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  50 static struct lock_class_key rcu_lock_key;
  51 struct lockdep_map rcu_lock_map =
  52         STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
  53 EXPORT_SYMBOL_GPL(rcu_lock_map);
  54 #endif
  55
  56 enum rcu_barrier {
  57         RCU_BARRIER_STD,
  58         RCU_BARRIER_BH,
  59         RCU_BARRIER_SCHED,
  60 };
  61
  62 static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
  63 static atomic_t rcu_barrier_cpu_count;
  64 static DEFINE_MUTEX(rcu_barrier_mutex);
  65 static struct completion rcu_barrier_completion;
  66 int rcu_scheduler_active __read_mostly;
  67
  68 static atomic_t rcu_migrate_type_count = ATOMIC_INIT(0);
  69 static struct rcu_head rcu_migrate_head[3];
  70 static DECLARE_WAIT_QUEUE_HEAD(rcu_migrate_wq);
  71
  72 /*
  73  * Awaken the corresponding synchronize_rcu() instance now that a
  74  * grace period has elapsed.
  75  */
  76 void wakeme_after_rcu(struct rcu_head  *head)
  77 {
  78         struct rcu_synchronize *rcu;
  79
  80         rcu = container_of(head, struct rcu_synchronize, head);
  81         complete(&rcu->completion);
  82 }
  83
  84 #ifdef CONFIG_TREE_PREEMPT_RCU
  85
  86 /**
  87  * synchronize_rcu - wait until a grace period has elapsed.
  88  *
  89  * Control will return to the caller some time after a full grace
  90  * period has elapsed, in other words after all currently executing RCU
  91  * read-side critical sections have completed.  RCU read-side critical
  92  * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
  93  * and may be nested.
  94  */
  95 void synchronize_rcu(void)
  96 {
  97         struct rcu_synchronize rcu;
  98
  99         if (!rcu_scheduler_active)
 100                 return;
 101
 102         init_completion(&rcu.completion);
 103         /* Will wake me after RCU finished. */
 104         call_rcu(&rcu.head, wakeme_after_rcu);
 105         /* Wait for it. */
 106         wait_for_completion(&rcu.completion);
 107 }
 108 EXPORT_SYMBOL_GPL(synchronize_rcu);
 109
 110 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
 111
 112 /**
 113  * synchronize_sched - wait until an rcu-sched grace period has elapsed.
 114  *
 115  * Control will return to the caller some time after a full rcu-sched
 116  * grace period has elapsed, in other words after all currently executing
 117  * rcu-sched read-side critical sections have completed.   These read-side
 118  * critical sections are delimited by rcu_read_lock_sched() and
 119  * rcu_read_unlock_sched(), and may be nested.  Note that preempt_disable(),
 120  * local_irq_disable(), and so on may be used in place of
 121  * rcu_read_lock_sched().
 122  *
 123  * This means that all preempt_disable code sequences, including NMI and
 124  * hardware-interrupt handlers, in progress on entry will have completed
 125  * before this primitive returns.  However, this does not guarantee that
 126  * softirq handlers will have completed, since in some kernels, these
 127  * handlers can run in process context, and can block.
 128  *
 129  * This primitive provides the guarantees made by the (now removed)
 130  * synchronize_kernel() API.  In contrast, synchronize_rcu() only
 131  * guarantees that rcu_read_lock() sections will have completed.
 132  * In "classic RCU", these two guarantees happen to be one and
 133  * the same, but can differ in realtime RCU implementations.
 134  */
 135 void synchronize_sched(void)
 136 {
 137         struct rcu_synchronize rcu;
 138
 139         if (rcu_blocking_is_gp())
 140                 return;
 141
 142         init_completion(&rcu.completion);
 143         /* Will wake me after RCU finished. */
 144         call_rcu_sched(&rcu.head, wakeme_after_rcu);
 145         /* Wait for it. */
 146         wait_for_completion(&rcu.completion);
 147 }
 148 EXPORT_SYMBOL_GPL(synchronize_sched);
 149
 150 /**
 151  * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
 152  *
 153  * Control will return to the caller some time after a full rcu_bh grace
 154  * period has elapsed, in other words after all currently executing rcu_bh
 155  * read-side critical sections have completed.  RCU read-side critical
 156  * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(),
 157  * and may be nested.
 158  */
 159 void synchronize_rcu_bh(void)
 160 {
 161         struct rcu_synchronize rcu;
 162
 163         if (rcu_blocking_is_gp())
 164                 return;
 165
 166         init_completion(&rcu.completion);
 167         /* Will wake me after RCU finished. */
 168         call_rcu_bh(&rcu.head, wakeme_after_rcu);
 169         /* Wait for it. */
 170         wait_for_completion(&rcu.completion);
 171 }
 172 EXPORT_SYMBOL_GPL(synchronize_rcu_bh);
 173
 174 static void rcu_barrier_callback(struct rcu_head *notused)
 175 {
 176         if (atomic_dec_and_test(&rcu_barrier_cpu_count))
 177                 complete(&rcu_barrier_completion);
 178 }
 179
 180 /*
 181  * Called with preemption disabled, and from cross-cpu IRQ context.
 182  */
 183 static void rcu_barrier_func(void *type)
 184 {
 185         int cpu = smp_processor_id();
 186         struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
 187
 188         atomic_inc(&rcu_barrier_cpu_count);
 189         switch ((enum rcu_barrier)type) {
 190         case RCU_BARRIER_STD:
 191                 call_rcu(head, rcu_barrier_callback);
 192                 break;
 193         case RCU_BARRIER_BH:
 194                 call_rcu_bh(head, rcu_barrier_callback);
 195                 break;
 196         case RCU_BARRIER_SCHED:
 197                 call_rcu_sched(head, rcu_barrier_callback);
 198                 break;
 199         }
 200 }
 201
 202 static inline void wait_migrated_callbacks(void)
 203 {
 204         wait_event(rcu_migrate_wq, !atomic_read(&rcu_migrate_type_count));
 205         smp_mb(); /* In case we didn't sleep. */
 206 }
 207
 208 /*
 209  * Orchestrate the specified type of RCU barrier, waiting for all
 210  * RCU callbacks of the specified type to complete.
 211  */
 212 static void _rcu_barrier(enum rcu_barrier type)
 213 {
 214         BUG_ON(in_interrupt());
 215         /* Take cpucontrol mutex to protect against CPU hotplug */
 216         mutex_lock(&rcu_barrier_mutex);
 217         init_completion(&rcu_barrier_completion);
 218         /*
 219          * Initialize rcu_barrier_cpu_count to 1, then invoke
 220          * rcu_barrier_func() on each CPU, so that each CPU also has
 221          * incremented rcu_barrier_cpu_count.  Only then is it safe to
 222          * decrement rcu_barrier_cpu_count -- otherwise the first CPU
 223          * might complete its grace period before all of the other CPUs
 224          * did their increment, causing this function to return too
 225          * early.
 226          */
 227         atomic_set(&rcu_barrier_cpu_count, 1);
 228         on_each_cpu(rcu_barrier_func, (void *)type, 1);
 229         if (atomic_dec_and_test(&rcu_barrier_cpu_count))
 230                 complete(&rcu_barrier_completion);
 231         wait_for_completion(&rcu_barrier_completion);
 232         mutex_unlock(&rcu_barrier_mutex);
 233         wait_migrated_callbacks();
 234 }
 235
 236 /**
 237  * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
 238  */
 239 void rcu_barrier(void)
 240 {
 241         _rcu_barrier(RCU_BARRIER_STD);
 242 }
 243 EXPORT_SYMBOL_GPL(rcu_barrier);
 244
 245 /**
 246  * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
 247  */
 248 void rcu_barrier_bh(void)
 249 {
 250         _rcu_barrier(RCU_BARRIER_BH);
 251 }
 252 EXPORT_SYMBOL_GPL(rcu_barrier_bh);
 253
 254 /**
 255  * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
 256  */
 257 void rcu_barrier_sched(void)
 258 {
 259         _rcu_barrier(RCU_BARRIER_SCHED);
 260 }
 261 EXPORT_SYMBOL_GPL(rcu_barrier_sched);
 262
 263 static void rcu_migrate_callback(struct rcu_head *notused)
 264 {
 265         if (atomic_dec_and_test(&rcu_migrate_type_count))
 266                 wake_up(&rcu_migrate_wq);
 267 }
 268
 269 static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self,
 270                 unsigned long action, void *hcpu)
 271 {
 272         rcu_cpu_notify(self, action, hcpu);
 273         if (action == CPU_DYING) {
 274                 /*
 275                  * preempt_disable() in on_each_cpu() prevents stop_machine(),
 276                  * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
 277                  * returns, all online cpus have queued rcu_barrier_func(),
 278                  * and the dead cpu(if it exist) queues rcu_migrate_callback()s.
 279                  *
 280                  * These callbacks ensure _rcu_barrier() waits for all
 281                  * RCU callbacks of the specified type to complete.
 282                  */
 283                 atomic_set(&rcu_migrate_type_count, 3);
 284                 call_rcu_bh(rcu_migrate_head, rcu_migrate_callback);
 285                 call_rcu_sched(rcu_migrate_head + 1, rcu_migrate_callback);
 286                 call_rcu(rcu_migrate_head + 2, rcu_migrate_callback);
 287         } else if (action == CPU_DOWN_PREPARE) {
 288                 /* Don't need to wait until next removal operation. */
 289                 /* rcu_migrate_head is protected by cpu_add_remove_lock */
 290                 wait_migrated_callbacks();
 291         }
 292
 293         return NOTIFY_OK;
 294 }
 295
 296 void __init rcu_init(void)
 297 {
 298         int i;
 299
 300         __rcu_init();
 301         cpu_notifier(rcu_barrier_cpu_hotplug, 0);
 302
 303         /*
 304          * We don't need protection against CPU-hotplug here because
 305          * this is called early in boot, before either interrupts
 306          * or the scheduler are operational.
 307          */
 308         for_each_online_cpu(i)
 309                 rcu_barrier_cpu_hotplug(NULL, CPU_UP_PREPARE, (void *)(long)i);
 310 }
 311
 312 void rcu_scheduler_starting(void)
 313 {
 314         WARN_ON(num_online_cpus() != 1);
 315         WARN_ON(nr_context_switches() > 0);
 316         rcu_scheduler_active = 1;
 317 }