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 <linux/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/export.h>
  47 #include <linux/hardirq.h>
  48
  49 #define CREATE_TRACE_POINTS
  50 #include <trace/events/rcu.h>
  51
  52 #include "rcu.h"
  53
  54 #ifdef CONFIG_PREEMPT_RCU
  55
  56 /*
  57  * Preemptible RCU implementation for rcu_read_lock().
  58  * Just increment ->rcu_read_lock_nesting, shared state will be updated
  59  * if we block.
  60  */
  61 void __rcu_read_lock(void)
  62 {
  63         current->rcu_read_lock_nesting++;
  64         barrier();  /* critical section after entry code. */
  65 }
  66 EXPORT_SYMBOL_GPL(__rcu_read_lock);
  67
  68 /*
  69  * Preemptible RCU implementation for rcu_read_unlock().
  70  * Decrement ->rcu_read_lock_nesting.  If the result is zero (outermost
  71  * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
  72  * invoke rcu_read_unlock_special() to clean up after a context switch
  73  * in an RCU read-side critical section and other special cases.
  74  */
  75 void __rcu_read_unlock(void)
  76 {
  77         struct task_struct *t = current;
  78
  79         if (t->rcu_read_lock_nesting != 1) {
  80                 --t->rcu_read_lock_nesting;
  81         } else {
  82                 barrier();  /* critical section before exit code. */
  83                 t->rcu_read_lock_nesting = INT_MIN;
  84                 barrier();  /* assign before ->rcu_read_unlock_special load */
  85                 if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
  86                         rcu_read_unlock_special(t);
  87                 barrier();  /* ->rcu_read_unlock_special load before assign */
  88                 t->rcu_read_lock_nesting = 0;
  89         }
  90 #ifdef CONFIG_PROVE_LOCKING
  91         {
  92                 int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting);
  93
  94                 WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2);
  95         }
  96 #endif /* #ifdef CONFIG_PROVE_LOCKING */
  97 }
  98 EXPORT_SYMBOL_GPL(__rcu_read_unlock);
  99
 100 /*
 101  * Check for a task exiting while in a preemptible-RCU read-side
 102  * critical section, clean up if so.  No need to issue warnings,
 103  * as debug_check_no_locks_held() already does this if lockdep
 104  * is enabled.
 105  */
 106 void exit_rcu(void)
 107 {
 108         struct task_struct *t = current;
 109
 110         if (likely(list_empty(&current->rcu_node_entry)))
 111                 return;
 112         t->rcu_read_lock_nesting = 1;
 113         barrier();
 114         t->rcu_read_unlock_special = RCU_READ_UNLOCK_BLOCKED;
 115         __rcu_read_unlock();
 116 }
 117
 118 #else /* #ifdef CONFIG_PREEMPT_RCU */
 119
 120 void exit_rcu(void)
 121 {
 122 }
 123
 124 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
 125
 126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 127 static struct lock_class_key rcu_lock_key;
 128 struct lockdep_map rcu_lock_map =
 129         STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
 130 EXPORT_SYMBOL_GPL(rcu_lock_map);
 131
 132 static struct lock_class_key rcu_bh_lock_key;
 133 struct lockdep_map rcu_bh_lock_map =
 134         STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_bh", &rcu_bh_lock_key);
 135 EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
 136
 137 static struct lock_class_key rcu_sched_lock_key;
 138 struct lockdep_map rcu_sched_lock_map =
 139         STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key);
 140 EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
 141 #endif
 142
 143 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 144
 145 int debug_lockdep_rcu_enabled(void)
 146 {
 147         return rcu_scheduler_active && debug_locks &&
 148                current->lockdep_recursion == 0;
 149 }
 150 EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
 151
 152 /**
 153  * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
 154  *
 155  * Check for bottom half being disabled, which covers both the
 156  * CONFIG_PROVE_RCU and not cases.  Note that if someone uses
 157  * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
 158  * will show the situation.  This is useful for debug checks in functions
 159  * that require that they be called within an RCU read-side critical
 160  * section.
 161  *
 162  * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
 163  *
 164  * Note that rcu_read_lock() is disallowed if the CPU is either idle or
 165  * offline from an RCU perspective, so check for those as well.
 166  */
 167 int rcu_read_lock_bh_held(void)
 168 {
 169         if (!debug_lockdep_rcu_enabled())
 170                 return 1;
 171         if (rcu_is_cpu_idle())
 172                 return 0;
 173         if (!rcu_lockdep_current_cpu_online())
 174                 return 0;
 175         return in_softirq() || irqs_disabled();
 176 }
 177 EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
 178
 179 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
 180
 181 struct rcu_synchronize {
 182         struct rcu_head head;
 183         struct completion completion;
 184 };
 185
 186 /*
 187  * Awaken the corresponding synchronize_rcu() instance now that a
 188  * grace period has elapsed.
 189  */
 190 static void wakeme_after_rcu(struct rcu_head  *head)
 191 {
 192         struct rcu_synchronize *rcu;
 193
 194         rcu = container_of(head, struct rcu_synchronize, head);
 195         complete(&rcu->completion);
 196 }
 197
 198 void wait_rcu_gp(call_rcu_func_t crf)
 199 {
 200         struct rcu_synchronize rcu;
 201
 202         init_rcu_head_on_stack(&rcu.head);
 203         init_completion(&rcu.completion);
 204         /* Will wake me after RCU finished. */
 205         crf(&rcu.head, wakeme_after_rcu);
 206         /* Wait for it. */
 207         wait_for_completion(&rcu.completion);
 208         destroy_rcu_head_on_stack(&rcu.head);
 209 }
 210 EXPORT_SYMBOL_GPL(wait_rcu_gp);
 211
 212 #ifdef CONFIG_PROVE_RCU
 213 /*
 214  * wrapper function to avoid #include problems.
 215  */
 216 int rcu_my_thread_group_empty(void)
 217 {
 218         return thread_group_empty(current);
 219 }
 220 EXPORT_SYMBOL_GPL(rcu_my_thread_group_empty);
 221 #endif /* #ifdef CONFIG_PROVE_RCU */
 222
 223 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
 224 static inline void debug_init_rcu_head(struct rcu_head *head)
 225 {
 226         debug_object_init(head, &rcuhead_debug_descr);
 227 }
 228
 229 static inline void debug_rcu_head_free(struct rcu_head *head)
 230 {
 231         debug_object_free(head, &rcuhead_debug_descr);
 232 }
 233
 234 /*
 235  * fixup_init is called when:
 236  * - an active object is initialized
 237  */
 238 static int rcuhead_fixup_init(void *addr, enum debug_obj_state state)
 239 {
 240         struct rcu_head *head = addr;
 241
 242         switch (state) {
 243         case ODEBUG_STATE_ACTIVE:
 244                 /*
 245                  * Ensure that queued callbacks are all executed.
 246                  * If we detect that we are nested in a RCU read-side critical
 247                  * section, we should simply fail, otherwise we would deadlock.
 248                  * In !PREEMPT configurations, there is no way to tell if we are
 249                  * in a RCU read-side critical section or not, so we never
 250                  * attempt any fixup and just print a warning.
 251                  */
 252 #ifndef CONFIG_PREEMPT
 253                 WARN_ON_ONCE(1);
 254                 return 0;
 255 #endif
 256                 if (rcu_preempt_depth() != 0 || preempt_count() != 0 ||
 257                     irqs_disabled()) {
 258                         WARN_ON_ONCE(1);
 259                         return 0;
 260                 }
 261                 rcu_barrier();
 262                 rcu_barrier_sched();
 263                 rcu_barrier_bh();
 264                 debug_object_init(head, &rcuhead_debug_descr);
 265                 return 1;
 266         default:
 267                 return 0;
 268         }
 269 }
 270
 271 /*
 272  * fixup_activate is called when:
 273  * - an active object is activated
 274  * - an unknown object is activated (might be a statically initialized object)
 275  * Activation is performed internally by call_rcu().
 276  */
 277 static int rcuhead_fixup_activate(void *addr, enum debug_obj_state state)
 278 {
 279         struct rcu_head *head = addr;
 280
 281         switch (state) {
 282
 283         case ODEBUG_STATE_NOTAVAILABLE:
 284                 /*
 285                  * This is not really a fixup. We just make sure that it is
 286                  * tracked in the object tracker.
 287                  */
 288                 debug_object_init(head, &rcuhead_debug_descr);
 289                 debug_object_activate(head, &rcuhead_debug_descr);
 290                 return 0;
 291
 292         case ODEBUG_STATE_ACTIVE:
 293                 /*
 294                  * Ensure that queued callbacks are all executed.
 295                  * If we detect that we are nested in a RCU read-side critical
 296                  * section, we should simply fail, otherwise we would deadlock.
 297                  * In !PREEMPT configurations, there is no way to tell if we are
 298                  * in a RCU read-side critical section or not, so we never
 299                  * attempt any fixup and just print a warning.
 300                  */
 301 #ifndef CONFIG_PREEMPT
 302                 WARN_ON_ONCE(1);
 303                 return 0;
 304 #endif
 305                 if (rcu_preempt_depth() != 0 || preempt_count() != 0 ||
 306                     irqs_disabled()) {
 307                         WARN_ON_ONCE(1);
 308                         return 0;
 309                 }
 310                 rcu_barrier();
 311                 rcu_barrier_sched();
 312                 rcu_barrier_bh();
 313                 debug_object_activate(head, &rcuhead_debug_descr);
 314                 return 1;
 315         default:
 316                 return 0;
 317         }
 318 }
 319
 320 /*
 321  * fixup_free is called when:
 322  * - an active object is freed
 323  */
 324 static int rcuhead_fixup_free(void *addr, enum debug_obj_state state)
 325 {
 326         struct rcu_head *head = addr;
 327
 328         switch (state) {
 329         case ODEBUG_STATE_ACTIVE:
 330                 /*
 331                  * Ensure that queued callbacks are all executed.
 332                  * If we detect that we are nested in a RCU read-side critical
 333                  * section, we should simply fail, otherwise we would deadlock.
 334                  * In !PREEMPT configurations, there is no way to tell if we are
 335                  * in a RCU read-side critical section or not, so we never
 336                  * attempt any fixup and just print a warning.
 337                  */
 338 #ifndef CONFIG_PREEMPT
 339                 WARN_ON_ONCE(1);
 340                 return 0;
 341 #endif
 342                 if (rcu_preempt_depth() != 0 || preempt_count() != 0 ||
 343                     irqs_disabled()) {
 344                         WARN_ON_ONCE(1);
 345                         return 0;
 346                 }
 347                 rcu_barrier();
 348                 rcu_barrier_sched();
 349                 rcu_barrier_bh();
 350                 debug_object_free(head, &rcuhead_debug_descr);
 351                 return 1;
 352         default:
 353                 return 0;
 354         }
 355 }
 356
 357 /**
 358  * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
 359  * @head: pointer to rcu_head structure to be initialized
 360  *
 361  * This function informs debugobjects of a new rcu_head structure that
 362  * has been allocated as an auto variable on the stack.  This function
 363  * is not required for rcu_head structures that are statically defined or
 364  * that are dynamically allocated on the heap.  This function has no
 365  * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
 366  */
 367 void init_rcu_head_on_stack(struct rcu_head *head)
 368 {
 369         debug_object_init_on_stack(head, &rcuhead_debug_descr);
 370 }
 371 EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
 372
 373 /**
 374  * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
 375  * @head: pointer to rcu_head structure to be initialized
 376  *
 377  * This function informs debugobjects that an on-stack rcu_head structure
 378  * is about to go out of scope.  As with init_rcu_head_on_stack(), this
 379  * function is not required for rcu_head structures that are statically
 380  * defined or that are dynamically allocated on the heap.  Also as with
 381  * init_rcu_head_on_stack(), this function has no effect for
 382  * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
 383  */
 384 void destroy_rcu_head_on_stack(struct rcu_head *head)
 385 {
 386         debug_object_free(head, &rcuhead_debug_descr);
 387 }
 388 EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
 389
 390 struct debug_obj_descr rcuhead_debug_descr = {
 391         .name = "rcu_head",
 392         .fixup_init = rcuhead_fixup_init,
 393         .fixup_activate = rcuhead_fixup_activate,
 394         .fixup_free = rcuhead_fixup_free,
 395 };
 396 EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
 397 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 398
 399 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) || defined(CONFIG_RCU_TRACE)
 400 void do_trace_rcu_torture_read(char *rcutorturename, struct rcu_head *rhp)
 401 {
 402         trace_rcu_torture_read(rcutorturename, rhp);
 403 }
 404 EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
 405 #else
 406 #define do_trace_rcu_torture_read(rcutorturename, rhp) do { } while (0)
 407 #endif