include/linux/rcupdate.h

   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  * Author: Dipankar Sarma <dipankar@in.ibm.com>
  21  *
  22  * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
  23  * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
  24  * Papers:
  25  * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
  26  * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
  27  *
  28  * For detailed explanation of Read-Copy Update mechanism see -
  29  *              http://lse.sourceforge.net/locking/rcupdate.html
  30  *
  31  */
  32
  33 #ifndef __LINUX_RCUPDATE_H
  34 #define __LINUX_RCUPDATE_H
  35
  36 #include <linux/cache.h>
  37 #include <linux/spinlock.h>
  38 #include <linux/threads.h>
  39 #include <linux/cpumask.h>
  40 #include <linux/seqlock.h>
  41 #include <linux/lockdep.h>
  42 #include <linux/completion.h>
  43 #include <linux/debugobjects.h>
  44 #include <linux/compiler.h>
  45
  46 #ifdef CONFIG_RCU_TORTURE_TEST
  47 extern int rcutorture_runnable; /* for sysctl */
  48 #endif /* #ifdef CONFIG_RCU_TORTURE_TEST */
  49
  50 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
  51 extern void rcutorture_record_test_transition(void);
  52 extern void rcutorture_record_progress(unsigned long vernum);
  53 #else
  54 static inline void rcutorture_record_test_transition(void)
  55 {
  56 }
  57 static inline void rcutorture_record_progress(unsigned long vernum)
  58 {
  59 }
  60 #endif
  61
  62 #define UINT_CMP_GE(a, b)       (UINT_MAX / 2 >= (a) - (b))
  63 #define UINT_CMP_LT(a, b)       (UINT_MAX / 2 < (a) - (b))
  64 #define ULONG_CMP_GE(a, b)      (ULONG_MAX / 2 >= (a) - (b))
  65 #define ULONG_CMP_LT(a, b)      (ULONG_MAX / 2 < (a) - (b))
  66
  67 /**
  68  * struct rcu_head - callback structure for use with RCU
  69  * @next: next update requests in a list
  70  * @func: actual update function to call after the grace period.
  71  */
  72 struct rcu_head {
  73         struct rcu_head *next;
  74         void (*func)(struct rcu_head *head);
  75 };
  76
  77 /* Exported common interfaces */
  78 extern void call_rcu_sched(struct rcu_head *head,
  79                            void (*func)(struct rcu_head *rcu));
  80 extern void synchronize_sched(void);
  81 extern void rcu_barrier_bh(void);
  82 extern void rcu_barrier_sched(void);
  83
  84 static inline void __rcu_read_lock_bh(void)
  85 {
  86         local_bh_disable();
  87 }
  88
  89 static inline void __rcu_read_unlock_bh(void)
  90 {
  91         local_bh_enable();
  92 }
  93
  94 #ifdef CONFIG_PREEMPT_RCU
  95
  96 extern void __rcu_read_lock(void);
  97 extern void __rcu_read_unlock(void);
  98 void synchronize_rcu(void);
  99
 100 /*
 101  * Defined as a macro as it is a very low level header included from
 102  * areas that don't even know about current.  This gives the rcu_read_lock()
 103  * nesting depth, but makes sense only if CONFIG_PREEMPT_RCU -- in other
 104  * types of kernel builds, the rcu_read_lock() nesting depth is unknowable.
 105  */
 106 #define rcu_preempt_depth() (current->rcu_read_lock_nesting)
 107
 108 #else /* #ifdef CONFIG_PREEMPT_RCU */
 109
 110 static inline void __rcu_read_lock(void)
 111 {
 112         preempt_disable();
 113 }
 114
 115 static inline void __rcu_read_unlock(void)
 116 {
 117         preempt_enable();
 118 }
 119
 120 static inline void synchronize_rcu(void)
 121 {
 122         synchronize_sched();
 123 }
 124
 125 static inline int rcu_preempt_depth(void)
 126 {
 127         return 0;
 128 }
 129
 130 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
 131
 132 /* Internal to kernel */
 133 extern void rcu_sched_qs(int cpu);
 134 extern void rcu_bh_qs(int cpu);
 135 extern void rcu_check_callbacks(int cpu, int user);
 136 struct notifier_block;
 137
 138 #ifdef CONFIG_NO_HZ
 139
 140 extern void rcu_enter_nohz(void);
 141 extern void rcu_exit_nohz(void);
 142
 143 #else /* #ifdef CONFIG_NO_HZ */
 144
 145 static inline void rcu_enter_nohz(void)
 146 {
 147 }
 148
 149 static inline void rcu_exit_nohz(void)
 150 {
 151 }
 152
 153 #endif /* #else #ifdef CONFIG_NO_HZ */
 154
 155 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU)
 156 #include <linux/rcutree.h>
 157 #elif defined(CONFIG_TINY_RCU) || defined(CONFIG_TINY_PREEMPT_RCU)
 158 #include <linux/rcutiny.h>
 159 #else
 160 #error "Unknown RCU implementation specified to kernel configuration"
 161 #endif
 162
 163 /*
 164  * init_rcu_head_on_stack()/destroy_rcu_head_on_stack() are needed for dynamic
 165  * initialization and destruction of rcu_head on the stack. rcu_head structures
 166  * allocated dynamically in the heap or defined statically don't need any
 167  * initialization.
 168  */
 169 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
 170 extern void init_rcu_head_on_stack(struct rcu_head *head);
 171 extern void destroy_rcu_head_on_stack(struct rcu_head *head);
 172 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 173 static inline void init_rcu_head_on_stack(struct rcu_head *head)
 174 {
 175 }
 176
 177 static inline void destroy_rcu_head_on_stack(struct rcu_head *head)
 178 {
 179 }
 180 #endif  /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 181
 182 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 183
 184 extern struct lockdep_map rcu_lock_map;
 185 # define rcu_read_acquire() \
 186                 lock_acquire(&rcu_lock_map, 0, 0, 2, 1, NULL, _THIS_IP_)
 187 # define rcu_read_release()     lock_release(&rcu_lock_map, 1, _THIS_IP_)
 188
 189 extern struct lockdep_map rcu_bh_lock_map;
 190 # define rcu_read_acquire_bh() \
 191                 lock_acquire(&rcu_bh_lock_map, 0, 0, 2, 1, NULL, _THIS_IP_)
 192 # define rcu_read_release_bh()  lock_release(&rcu_bh_lock_map, 1, _THIS_IP_)
 193
 194 extern struct lockdep_map rcu_sched_lock_map;
 195 # define rcu_read_acquire_sched() \
 196                 lock_acquire(&rcu_sched_lock_map, 0, 0, 2, 1, NULL, _THIS_IP_)
 197 # define rcu_read_release_sched() \
 198                 lock_release(&rcu_sched_lock_map, 1, _THIS_IP_)
 199
 200 extern int debug_lockdep_rcu_enabled(void);
 201
 202 /**
 203  * rcu_read_lock_held() - might we be in RCU read-side critical section?
 204  *
 205  * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU
 206  * read-side critical section.  In absence of CONFIG_DEBUG_LOCK_ALLOC,
 207  * this assumes we are in an RCU read-side critical section unless it can
 208  * prove otherwise.  This is useful for debug checks in functions that
 209  * require that they be called within an RCU read-side critical section.
 210  *
 211  * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot
 212  * and while lockdep is disabled.
 213  */
 214 static inline int rcu_read_lock_held(void)
 215 {
 216         if (!debug_lockdep_rcu_enabled())
 217                 return 1;
 218         return lock_is_held(&rcu_lock_map);
 219 }
 220
 221 /*
 222  * rcu_read_lock_bh_held() is defined out of line to avoid #include-file
 223  * hell.
 224  */
 225 extern int rcu_read_lock_bh_held(void);
 226
 227 /**
 228  * rcu_read_lock_sched_held() - might we be in RCU-sched read-side critical section?
 229  *
 230  * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an
 231  * RCU-sched read-side critical section.  In absence of
 232  * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side
 233  * critical section unless it can prove otherwise.  Note that disabling
 234  * of preemption (including disabling irqs) counts as an RCU-sched
 235  * read-side critical section.  This is useful for debug checks in functions
 236  * that required that they be called within an RCU-sched read-side
 237  * critical section.
 238  *
 239  * Check debug_lockdep_rcu_enabled() to prevent false positives during boot
 240  * and while lockdep is disabled.
 241  */
 242 #ifdef CONFIG_PREEMPT_COUNT
 243 static inline int rcu_read_lock_sched_held(void)
 244 {
 245         int lockdep_opinion = 0;
 246
 247         if (!debug_lockdep_rcu_enabled())
 248                 return 1;
 249         if (debug_locks)
 250                 lockdep_opinion = lock_is_held(&rcu_sched_lock_map);
 251         return lockdep_opinion || preempt_count() != 0 || irqs_disabled();
 252 }
 253 #else /* #ifdef CONFIG_PREEMPT_COUNT */
 254 static inline int rcu_read_lock_sched_held(void)
 255 {
 256         return 1;
 257 }
 258 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
 259
 260 #else /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
 261
 262 # define rcu_read_acquire()             do { } while (0)
 263 # define rcu_read_release()             do { } while (0)
 264 # define rcu_read_acquire_bh()          do { } while (0)
 265 # define rcu_read_release_bh()          do { } while (0)
 266 # define rcu_read_acquire_sched()       do { } while (0)
 267 # define rcu_read_release_sched()       do { } while (0)
 268
 269 static inline int rcu_read_lock_held(void)
 270 {
 271         return 1;
 272 }
 273
 274 static inline int rcu_read_lock_bh_held(void)
 275 {
 276         return 1;
 277 }
 278
 279 #ifdef CONFIG_PREEMPT_COUNT
 280 static inline int rcu_read_lock_sched_held(void)
 281 {
 282         return preempt_count() != 0 || irqs_disabled();
 283 }
 284 #else /* #ifdef CONFIG_PREEMPT_COUNT */
 285 static inline int rcu_read_lock_sched_held(void)
 286 {
 287         return 1;
 288 }
 289 #endif /* #else #ifdef CONFIG_PREEMPT_COUNT */
 290
 291 #endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */
 292
 293 #ifdef CONFIG_PROVE_RCU
 294
 295 extern int rcu_my_thread_group_empty(void);
 296
 297 /**
 298  * rcu_lockdep_assert - emit lockdep splat if specified condition not met
 299  * @c: condition to check
 300  * @s: informative message
 301  */
 302 #define rcu_lockdep_assert(c, s)                                        \
 303         do {                                                            \
 304                 static bool __warned;                                   \
 305                 if (debug_lockdep_rcu_enabled() && !__warned && !(c)) { \
 306                         __warned = true;                                \
 307                         lockdep_rcu_suspicious(__FILE__, __LINE__, s);  \
 308                 }                                                       \
 309         } while (0)
 310
 311 #define rcu_sleep_check()                                               \
 312         do {                                                            \
 313                 rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map),     \
 314                                    "Illegal context switch in RCU-bh"   \
 315                                    " read-side critical section");      \
 316                 rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map),  \
 317                                    "Illegal context switch in RCU-sched"\
 318                                    " read-side critical section");      \
 319         } while (0)
 320
 321 #else /* #ifdef CONFIG_PROVE_RCU */
 322
 323 #define rcu_lockdep_assert(c, s) do { } while (0)
 324 #define rcu_sleep_check() do { } while (0)
 325
 326 #endif /* #else #ifdef CONFIG_PROVE_RCU */
 327
 328 /*
 329  * Helper functions for rcu_dereference_check(), rcu_dereference_protected()
 330  * and rcu_assign_pointer().  Some of these could be folded into their
 331  * callers, but they are left separate in order to ease introduction of
 332  * multiple flavors of pointers to match the multiple flavors of RCU
 333  * (e.g., __rcu_bh, * __rcu_sched, and __srcu), should this make sense in
 334  * the future.
 335  */
 336
 337 #ifdef __CHECKER__
 338 #define rcu_dereference_sparse(p, space) \
 339         ((void)(((typeof(*p) space *)p) == p))
 340 #else /* #ifdef __CHECKER__ */
 341 #define rcu_dereference_sparse(p, space)
 342 #endif /* #else #ifdef __CHECKER__ */
 343
 344 #define __rcu_access_pointer(p, space) \
 345         ({ \
 346                 typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \
 347                 rcu_dereference_sparse(p, space); \
 348                 ((typeof(*p) __force __kernel *)(_________p1)); \
 349         })
 350 #define __rcu_dereference_check(p, c, space) \
 351         ({ \
 352                 typeof(*p) *_________p1 = (typeof(*p)*__force )ACCESS_ONCE(p); \
 353                 rcu_lockdep_assert(c, "suspicious rcu_dereference_check()" \
 354                                       " usage"); \
 355                 rcu_dereference_sparse(p, space); \
 356                 smp_read_barrier_depends(); \
 357                 ((typeof(*p) __force __kernel *)(_________p1)); \
 358         })
 359 #define __rcu_dereference_protected(p, c, space) \
 360         ({ \
 361                 rcu_lockdep_assert(c, "suspicious rcu_dereference_protected()" \
 362                                       " usage"); \
 363                 rcu_dereference_sparse(p, space); \
 364                 ((typeof(*p) __force __kernel *)(p)); \
 365         })
 366
 367 #define __rcu_access_index(p, space) \
 368         ({ \
 369                 typeof(p) _________p1 = ACCESS_ONCE(p); \
 370                 rcu_dereference_sparse(p, space); \
 371                 (_________p1); \
 372         })
 373 #define __rcu_dereference_index_check(p, c) \
 374         ({ \
 375                 typeof(p) _________p1 = ACCESS_ONCE(p); \
 376                 rcu_lockdep_assert(c, \
 377                                    "suspicious rcu_dereference_index_check()" \
 378                                    " usage"); \
 379                 smp_read_barrier_depends(); \
 380                 (_________p1); \
 381         })
 382 #define __rcu_assign_pointer(p, v, space) \
 383         ({ \
 384                 if (!__builtin_constant_p(v) || \
 385                     ((v) != NULL)) \
 386                         smp_wmb(); \
 387                 (p) = (typeof(*v) __force space *)(v); \
 388         })
 389
 390
 391 /**
 392  * rcu_access_pointer() - fetch RCU pointer with no dereferencing
 393  * @p: The pointer to read
 394  *
 395  * Return the value of the specified RCU-protected pointer, but omit the
 396  * smp_read_barrier_depends() and keep the ACCESS_ONCE().  This is useful
 397  * when the value of this pointer is accessed, but the pointer is not
 398  * dereferenced, for example, when testing an RCU-protected pointer against
 399  * NULL.  Although rcu_access_pointer() may also be used in cases where
 400  * update-side locks prevent the value of the pointer from changing, you
 401  * should instead use rcu_dereference_protected() for this use case.
 402  */
 403 #define rcu_access_pointer(p) __rcu_access_pointer((p), __rcu)
 404
 405 /**
 406  * rcu_dereference_check() - rcu_dereference with debug checking
 407  * @p: The pointer to read, prior to dereferencing
 408  * @c: The conditions under which the dereference will take place
 409  *
 410  * Do an rcu_dereference(), but check that the conditions under which the
 411  * dereference will take place are correct.  Typically the conditions
 412  * indicate the various locking conditions that should be held at that
 413  * point.  The check should return true if the conditions are satisfied.
 414  * An implicit check for being in an RCU read-side critical section
 415  * (rcu_read_lock()) is included.
 416  *
 417  * For example:
 418  *
 419  *      bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock));
 420  *
 421  * could be used to indicate to lockdep that foo->bar may only be dereferenced
 422  * if either rcu_read_lock() is held, or that the lock required to replace
 423  * the bar struct at foo->bar is held.
 424  *
 425  * Note that the list of conditions may also include indications of when a lock
 426  * need not be held, for example during initialisation or destruction of the
 427  * target struct:
 428  *
 429  *      bar = rcu_dereference_check(foo->bar, lockdep_is_held(&foo->lock) ||
 430  *                                            atomic_read(&foo->usage) == 0);
 431  *
 432  * Inserts memory barriers on architectures that require them
 433  * (currently only the Alpha), prevents the compiler from refetching
 434  * (and from merging fetches), and, more importantly, documents exactly
 435  * which pointers are protected by RCU and checks that the pointer is
 436  * annotated as __rcu.
 437  */
 438 #define rcu_dereference_check(p, c) \
 439         __rcu_dereference_check((p), rcu_read_lock_held() || (c), __rcu)
 440
 441 /**
 442  * rcu_dereference_bh_check() - rcu_dereference_bh with debug checking
 443  * @p: The pointer to read, prior to dereferencing
 444  * @c: The conditions under which the dereference will take place
 445  *
 446  * This is the RCU-bh counterpart to rcu_dereference_check().
 447  */
 448 #define rcu_dereference_bh_check(p, c) \
 449         __rcu_dereference_check((p), rcu_read_lock_bh_held() || (c), __rcu)
 450
 451 /**
 452  * rcu_dereference_sched_check() - rcu_dereference_sched with debug checking
 453  * @p: The pointer to read, prior to dereferencing
 454  * @c: The conditions under which the dereference will take place
 455  *
 456  * This is the RCU-sched counterpart to rcu_dereference_check().
 457  */
 458 #define rcu_dereference_sched_check(p, c) \
 459         __rcu_dereference_check((p), rcu_read_lock_sched_held() || (c), \
 460                                 __rcu)
 461
 462 #define rcu_dereference_raw(p) rcu_dereference_check(p, 1) /*@@@ needed? @@@*/
 463
 464 /**
 465  * rcu_access_index() - fetch RCU index with no dereferencing
 466  * @p: The index to read
 467  *
 468  * Return the value of the specified RCU-protected index, but omit the
 469  * smp_read_barrier_depends() and keep the ACCESS_ONCE().  This is useful
 470  * when the value of this index is accessed, but the index is not
 471  * dereferenced, for example, when testing an RCU-protected index against
 472  * -1.  Although rcu_access_index() may also be used in cases where
 473  * update-side locks prevent the value of the index from changing, you
 474  * should instead use rcu_dereference_index_protected() for this use case.
 475  */
 476 #define rcu_access_index(p) __rcu_access_index((p), __rcu)
 477
 478 /**
 479  * rcu_dereference_index_check() - rcu_dereference for indices with debug checking
 480  * @p: The pointer to read, prior to dereferencing
 481  * @c: The conditions under which the dereference will take place
 482  *
 483  * Similar to rcu_dereference_check(), but omits the sparse checking.
 484  * This allows rcu_dereference_index_check() to be used on integers,
 485  * which can then be used as array indices.  Attempting to use
 486  * rcu_dereference_check() on an integer will give compiler warnings
 487  * because the sparse address-space mechanism relies on dereferencing
 488  * the RCU-protected pointer.  Dereferencing integers is not something
 489  * that even gcc will put up with.
 490  *
 491  * Note that this function does not implicitly check for RCU read-side
 492  * critical sections.  If this function gains lots of uses, it might
 493  * make sense to provide versions for each flavor of RCU, but it does
 494  * not make sense as of early 2010.
 495  */
 496 #define rcu_dereference_index_check(p, c) \
 497         __rcu_dereference_index_check((p), (c))
 498
 499 /**
 500  * rcu_dereference_protected() - fetch RCU pointer when updates prevented
 501  * @p: The pointer to read, prior to dereferencing
 502  * @c: The conditions under which the dereference will take place
 503  *
 504  * Return the value of the specified RCU-protected pointer, but omit
 505  * both the smp_read_barrier_depends() and the ACCESS_ONCE().  This
 506  * is useful in cases where update-side locks prevent the value of the
 507  * pointer from changing.  Please note that this primitive does -not-
 508  * prevent the compiler from repeating this reference or combining it
 509  * with other references, so it should not be used without protection
 510  * of appropriate locks.
 511  *
 512  * This function is only for update-side use.  Using this function
 513  * when protected only by rcu_read_lock() will result in infrequent
 514  * but very ugly failures.
 515  */
 516 #define rcu_dereference_protected(p, c) \
 517         __rcu_dereference_protected((p), (c), __rcu)
 518
 519 /**
 520  * rcu_dereference_bh_protected() - fetch RCU-bh pointer when updates prevented
 521  * @p: The pointer to read, prior to dereferencing
 522  * @c: The conditions under which the dereference will take place
 523  *
 524  * This is the RCU-bh counterpart to rcu_dereference_protected().
 525  */
 526 #define rcu_dereference_bh_protected(p, c) \
 527         __rcu_dereference_protected((p), (c), __rcu)
 528
 529 /**
 530  * rcu_dereference_sched_protected() - fetch RCU-sched pointer when updates prevented
 531  * @p: The pointer to read, prior to dereferencing
 532  * @c: The conditions under which the dereference will take place
 533  *
 534  * This is the RCU-sched counterpart to rcu_dereference_protected().
 535  */
 536 #define rcu_dereference_sched_protected(p, c) \
 537         __rcu_dereference_protected((p), (c), __rcu)
 538
 539
 540 /**
 541  * rcu_dereference() - fetch RCU-protected pointer for dereferencing
 542  * @p: The pointer to read, prior to dereferencing
 543  *
 544  * This is a simple wrapper around rcu_dereference_check().
 545  */
 546 #define rcu_dereference(p) rcu_dereference_check(p, 0)
 547
 548 /**
 549  * rcu_dereference_bh() - fetch an RCU-bh-protected pointer for dereferencing
 550  * @p: The pointer to read, prior to dereferencing
 551  *
 552  * Makes rcu_dereference_check() do the dirty work.
 553  */
 554 #define rcu_dereference_bh(p) rcu_dereference_bh_check(p, 0)
 555
 556 /**
 557  * rcu_dereference_sched() - fetch RCU-sched-protected pointer for dereferencing
 558  * @p: The pointer to read, prior to dereferencing
 559  *
 560  * Makes rcu_dereference_check() do the dirty work.
 561  */
 562 #define rcu_dereference_sched(p) rcu_dereference_sched_check(p, 0)
 563
 564 /**
 565  * rcu_read_lock() - mark the beginning of an RCU read-side critical section
 566  *
 567  * When synchronize_rcu() is invoked on one CPU while other CPUs
 568  * are within RCU read-side critical sections, then the
 569  * synchronize_rcu() is guaranteed to block until after all the other
 570  * CPUs exit their critical sections.  Similarly, if call_rcu() is invoked
 571  * on one CPU while other CPUs are within RCU read-side critical
 572  * sections, invocation of the corresponding RCU callback is deferred
 573  * until after the all the other CPUs exit their critical sections.
 574  *
 575  * Note, however, that RCU callbacks are permitted to run concurrently
 576  * with new RCU read-side critical sections.  One way that this can happen
 577  * is via the following sequence of events: (1) CPU 0 enters an RCU
 578  * read-side critical section, (2) CPU 1 invokes call_rcu() to register
 579  * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
 580  * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
 581  * callback is invoked.  This is legal, because the RCU read-side critical
 582  * section that was running concurrently with the call_rcu() (and which
 583  * therefore might be referencing something that the corresponding RCU
 584  * callback would free up) has completed before the corresponding
 585  * RCU callback is invoked.
 586  *
 587  * RCU read-side critical sections may be nested.  Any deferred actions
 588  * will be deferred until the outermost RCU read-side critical section
 589  * completes.
 590  *
 591  * You can avoid reading and understanding the next paragraph by
 592  * following this rule: don't put anything in an rcu_read_lock() RCU
 593  * read-side critical section that would block in a !PREEMPT kernel.
 594  * But if you want the full story, read on!
 595  *
 596  * In non-preemptible RCU implementations (TREE_RCU and TINY_RCU), it
 597  * is illegal to block while in an RCU read-side critical section.  In
 598  * preemptible RCU implementations (TREE_PREEMPT_RCU and TINY_PREEMPT_RCU)
 599  * in CONFIG_PREEMPT kernel builds, RCU read-side critical sections may
 600  * be preempted, but explicit blocking is illegal.  Finally, in preemptible
 601  * RCU implementations in real-time (CONFIG_PREEMPT_RT) kernel builds,
 602  * RCU read-side critical sections may be preempted and they may also
 603  * block, but only when acquiring spinlocks that are subject to priority
 604  * inheritance.
 605  */
 606 static inline void rcu_read_lock(void)
 607 {
 608         __rcu_read_lock();
 609         __acquire(RCU);
 610         rcu_read_acquire();
 611 }
 612
 613 /*
 614  * So where is rcu_write_lock()?  It does not exist, as there is no
 615  * way for writers to lock out RCU readers.  This is a feature, not
 616  * a bug -- this property is what provides RCU's performance benefits.
 617  * Of course, writers must coordinate with each other.  The normal
 618  * spinlock primitives work well for this, but any other technique may be
 619  * used as well.  RCU does not care how the writers keep out of each
 620  * others' way, as long as they do so.
 621  */
 622
 623 /**
 624  * rcu_read_unlock() - marks the end of an RCU read-side critical section.
 625  *
 626  * See rcu_read_lock() for more information.
 627  */
 628 static inline void rcu_read_unlock(void)
 629 {
 630         rcu_read_release();
 631         __release(RCU);
 632         __rcu_read_unlock();
 633 }
 634
 635 /**
 636  * rcu_read_lock_bh() - mark the beginning of an RCU-bh critical section
 637  *
 638  * This is equivalent of rcu_read_lock(), but to be used when updates
 639  * are being done using call_rcu_bh() or synchronize_rcu_bh(). Since
 640  * both call_rcu_bh() and synchronize_rcu_bh() consider completion of a
 641  * softirq handler to be a quiescent state, a process in RCU read-side
 642  * critical section must be protected by disabling softirqs. Read-side
 643  * critical sections in interrupt context can use just rcu_read_lock(),
 644  * though this should at least be commented to avoid confusing people
 645  * reading the code.
 646  */
 647 static inline void rcu_read_lock_bh(void)
 648 {
 649         __rcu_read_lock_bh();
 650         __acquire(RCU_BH);
 651         rcu_read_acquire_bh();
 652 }
 653
 654 /*
 655  * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
 656  *
 657  * See rcu_read_lock_bh() for more information.
 658  */
 659 static inline void rcu_read_unlock_bh(void)
 660 {
 661         rcu_read_release_bh();
 662         __release(RCU_BH);
 663         __rcu_read_unlock_bh();
 664 }
 665
 666 /**
 667  * rcu_read_lock_sched() - mark the beginning of a RCU-sched critical section
 668  *
 669  * This is equivalent of rcu_read_lock(), but to be used when updates
 670  * are being done using call_rcu_sched() or synchronize_rcu_sched().
 671  * Read-side critical sections can also be introduced by anything that
 672  * disables preemption, including local_irq_disable() and friends.
 673  */
 674 static inline void rcu_read_lock_sched(void)
 675 {
 676         preempt_disable();
 677         __acquire(RCU_SCHED);
 678         rcu_read_acquire_sched();
 679 }
 680
 681 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
 682 static inline notrace void rcu_read_lock_sched_notrace(void)
 683 {
 684         preempt_disable_notrace();
 685         __acquire(RCU_SCHED);
 686 }
 687
 688 /*
 689  * rcu_read_unlock_sched - marks the end of a RCU-classic critical section
 690  *
 691  * See rcu_read_lock_sched for more information.
 692  */
 693 static inline void rcu_read_unlock_sched(void)
 694 {
 695         rcu_read_release_sched();
 696         __release(RCU_SCHED);
 697         preempt_enable();
 698 }
 699
 700 /* Used by lockdep and tracing: cannot be traced, cannot call lockdep. */
 701 static inline notrace void rcu_read_unlock_sched_notrace(void)
 702 {
 703         __release(RCU_SCHED);
 704         preempt_enable_notrace();
 705 }
 706
 707 /**
 708  * rcu_assign_pointer() - assign to RCU-protected pointer
 709  * @p: pointer to assign to
 710  * @v: value to assign (publish)
 711  *
 712  * Assigns the specified value to the specified RCU-protected
 713  * pointer, ensuring that any concurrent RCU readers will see
 714  * any prior initialization.  Returns the value assigned.
 715  *
 716  * Inserts memory barriers on architectures that require them
 717  * (pretty much all of them other than x86), and also prevents
 718  * the compiler from reordering the code that initializes the
 719  * structure after the pointer assignment.  More importantly, this
 720  * call documents which pointers will be dereferenced by RCU read-side
 721  * code.
 722  */
 723 #define rcu_assign_pointer(p, v) \
 724         __rcu_assign_pointer((p), (v), __rcu)
 725
 726 /**
 727  * RCU_INIT_POINTER() - initialize an RCU protected pointer
 728  *
 729  * Initialize an RCU-protected pointer in such a way to avoid RCU-lockdep
 730  * splats.
 731  */
 732 #define RCU_INIT_POINTER(p, v) \
 733                 p = (typeof(*v) __force __rcu *)(v)
 734
 735 /* Infrastructure to implement the synchronize_() primitives. */
 736
 737 struct rcu_synchronize {
 738         struct rcu_head head;
 739         struct completion completion;
 740 };
 741
 742 extern void wakeme_after_rcu(struct rcu_head  *head);
 743
 744 #ifdef CONFIG_PREEMPT_RCU
 745
 746 /**
 747  * call_rcu() - Queue an RCU callback for invocation after a grace period.
 748  * @head: structure to be used for queueing the RCU updates.
 749  * @func: actual callback function to be invoked after the grace period
 750  *
 751  * The callback function will be invoked some time after a full grace
 752  * period elapses, in other words after all pre-existing RCU read-side
 753  * critical sections have completed.  However, the callback function
 754  * might well execute concurrently with RCU read-side critical sections
 755  * that started after call_rcu() was invoked.  RCU read-side critical
 756  * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
 757  * and may be nested.
 758  */
 759 extern void call_rcu(struct rcu_head *head,
 760                               void (*func)(struct rcu_head *head));
 761
 762 #else /* #ifdef CONFIG_PREEMPT_RCU */
 763
 764 /* In classic RCU, call_rcu() is just call_rcu_sched(). */
 765 #define call_rcu        call_rcu_sched
 766
 767 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
 768
 769 /**
 770  * call_rcu_bh() - Queue an RCU for invocation after a quicker grace period.
 771  * @head: structure to be used for queueing the RCU updates.
 772  * @func: actual callback function to be invoked after the grace period
 773  *
 774  * The callback function will be invoked some time after a full grace
 775  * period elapses, in other words after all currently executing RCU
 776  * read-side critical sections have completed. call_rcu_bh() assumes
 777  * that the read-side critical sections end on completion of a softirq
 778  * handler. This means that read-side critical sections in process
 779  * context must not be interrupted by softirqs. This interface is to be
 780  * used when most of the read-side critical sections are in softirq context.
 781  * RCU read-side critical sections are delimited by :
 782  *  - rcu_read_lock() and  rcu_read_unlock(), if in interrupt context.
 783  *  OR
 784  *  - rcu_read_lock_bh() and rcu_read_unlock_bh(), if in process context.
 785  *  These may be nested.
 786  */
 787 extern void call_rcu_bh(struct rcu_head *head,
 788                         void (*func)(struct rcu_head *head));
 789
 790 /*
 791  * debug_rcu_head_queue()/debug_rcu_head_unqueue() are used internally
 792  * by call_rcu() and rcu callback execution, and are therefore not part of the
 793  * RCU API. Leaving in rcupdate.h because they are used by all RCU flavors.
 794  */
 795
 796 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
 797 # define STATE_RCU_HEAD_READY   0
 798 # define STATE_RCU_HEAD_QUEUED  1
 799
 800 extern struct debug_obj_descr rcuhead_debug_descr;
 801
 802 static inline void debug_rcu_head_queue(struct rcu_head *head)
 803 {
 804         WARN_ON_ONCE((unsigned long)head & 0x3);
 805         debug_object_activate(head, &rcuhead_debug_descr);
 806         debug_object_active_state(head, &rcuhead_debug_descr,
 807                                   STATE_RCU_HEAD_READY,
 808                                   STATE_RCU_HEAD_QUEUED);
 809 }
 810
 811 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
 812 {
 813         debug_object_active_state(head, &rcuhead_debug_descr,
 814                                   STATE_RCU_HEAD_QUEUED,
 815                                   STATE_RCU_HEAD_READY);
 816         debug_object_deactivate(head, &rcuhead_debug_descr);
 817 }
 818 #else   /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 819 static inline void debug_rcu_head_queue(struct rcu_head *head)
 820 {
 821 }
 822
 823 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
 824 {
 825 }
 826 #endif  /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 827
 828 static __always_inline bool __is_kfree_rcu_offset(unsigned long offset)
 829 {
 830         return offset < 4096;
 831 }
 832
 833 static __always_inline
 834 void __kfree_rcu(struct rcu_head *head, unsigned long offset)
 835 {
 836         typedef void (*rcu_callback)(struct rcu_head *);
 837
 838         BUILD_BUG_ON(!__builtin_constant_p(offset));
 839
 840         /* See the kfree_rcu() header comment. */
 841         BUILD_BUG_ON(!__is_kfree_rcu_offset(offset));
 842
 843         call_rcu(head, (rcu_callback)offset);
 844 }
 845
 846 extern void kfree(const void *);
 847
 848 static inline void __rcu_reclaim(struct rcu_head *head)
 849 {
 850         unsigned long offset = (unsigned long)head->func;
 851
 852         if (__is_kfree_rcu_offset(offset))
 853                 kfree((void *)head - offset);
 854         else
 855                 head->func(head);
 856 }
 857
 858 /**
 859  * kfree_rcu() - kfree an object after a grace period.
 860  * @ptr:        pointer to kfree
 861  * @rcu_head:   the name of the struct rcu_head within the type of @ptr.
 862  *
 863  * Many rcu callbacks functions just call kfree() on the base structure.
 864  * These functions are trivial, but their size adds up, and furthermore
 865  * when they are used in a kernel module, that module must invoke the
 866  * high-latency rcu_barrier() function at module-unload time.
 867  *
 868  * The kfree_rcu() function handles this issue.  Rather than encoding a
 869  * function address in the embedded rcu_head structure, kfree_rcu() instead
 870  * encodes the offset of the rcu_head structure within the base structure.
 871  * Because the functions are not allowed in the low-order 4096 bytes of
 872  * kernel virtual memory, offsets up to 4095 bytes can be accommodated.
 873  * If the offset is larger than 4095 bytes, a compile-time error will
 874  * be generated in __kfree_rcu().  If this error is triggered, you can
 875  * either fall back to use of call_rcu() or rearrange the structure to
 876  * position the rcu_head structure into the first 4096 bytes.
 877  *
 878  * Note that the allowable offset might decrease in the future, for example,
 879  * to allow something like kmem_cache_free_rcu().
 880  */
 881 #define kfree_rcu(ptr, rcu_head)                                        \
 882         __kfree_rcu(&((ptr)->rcu_head), offsetof(typeof(*(ptr)), rcu_head))
 883
 884 #endif /* __LINUX_RCUPDATE_H */