kernel/rcu/rcu.h

   1 /*
   2  * Read-Copy Update definitions shared among RCU implementations.
   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, you can access it online at
  16  * http://www.gnu.org/licenses/gpl-2.0.html.
  17  *
  18  * Copyright IBM Corporation, 2011
  19  *
  20  * Author: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
  21  */
  22
  23 #ifndef __LINUX_RCU_H
  24 #define __LINUX_RCU_H
  25
  26 #include <trace/events/rcu.h>
  27 #ifdef CONFIG_RCU_TRACE
  28 #define RCU_TRACE(stmt) stmt
  29 #else /* #ifdef CONFIG_RCU_TRACE */
  30 #define RCU_TRACE(stmt)
  31 #endif /* #else #ifdef CONFIG_RCU_TRACE */
  32
  33 /*
  34  * Process-level increment to ->dynticks_nesting field.  This allows for
  35  * architectures that use half-interrupts and half-exceptions from
  36  * process context.
  37  *
  38  * DYNTICK_TASK_NEST_MASK defines a field of width DYNTICK_TASK_NEST_WIDTH
  39  * that counts the number of process-based reasons why RCU cannot
  40  * consider the corresponding CPU to be idle, and DYNTICK_TASK_NEST_VALUE
  41  * is the value used to increment or decrement this field.
  42  *
  43  * The rest of the bits could in principle be used to count interrupts,
  44  * but this would mean that a negative-one value in the interrupt
  45  * field could incorrectly zero out the DYNTICK_TASK_NEST_MASK field.
  46  * We therefore provide a two-bit guard field defined by DYNTICK_TASK_MASK
  47  * that is set to DYNTICK_TASK_FLAG upon initial exit from idle.
  48  * The DYNTICK_TASK_EXIT_IDLE value is thus the combined value used upon
  49  * initial exit from idle.
  50  */
  51 #define DYNTICK_TASK_NEST_WIDTH 7
  52 #define DYNTICK_TASK_NEST_VALUE ((LLONG_MAX >> DYNTICK_TASK_NEST_WIDTH) + 1)
  53 #define DYNTICK_TASK_NEST_MASK  (LLONG_MAX - DYNTICK_TASK_NEST_VALUE + 1)
  54 #define DYNTICK_TASK_FLAG          ((DYNTICK_TASK_NEST_VALUE / 8) * 2)
  55 #define DYNTICK_TASK_MASK          ((DYNTICK_TASK_NEST_VALUE / 8) * 3)
  56 #define DYNTICK_TASK_EXIT_IDLE     (DYNTICK_TASK_NEST_VALUE + \
  57                                     DYNTICK_TASK_FLAG)
  58
  59
  60 /*
  61  * Grace-period counter management.
  62  */
  63
  64 #define RCU_SEQ_CTR_SHIFT       2
  65 #define RCU_SEQ_STATE_MASK      ((1 << RCU_SEQ_CTR_SHIFT) - 1)
  66
  67 /*
  68  * Return the counter portion of a sequence number previously returned
  69  * by rcu_seq_snap() or rcu_seq_current().
  70  */
  71 static inline unsigned long rcu_seq_ctr(unsigned long s)
  72 {
  73         return s >> RCU_SEQ_CTR_SHIFT;
  74 }
  75
  76 /*
  77  * Return the state portion of a sequence number previously returned
  78  * by rcu_seq_snap() or rcu_seq_current().
  79  */
  80 static inline int rcu_seq_state(unsigned long s)
  81 {
  82         return s & RCU_SEQ_STATE_MASK;
  83 }
  84
  85 /*
  86  * Set the state portion of the pointed-to sequence number.
  87  * The caller is responsible for preventing conflicting updates.
  88  */
  89 static inline void rcu_seq_set_state(unsigned long *sp, int newstate)
  90 {
  91         WARN_ON_ONCE(newstate & ~RCU_SEQ_STATE_MASK);
  92         WRITE_ONCE(*sp, (*sp & ~RCU_SEQ_STATE_MASK) + newstate);
  93 }
  94
  95 /* Adjust sequence number for start of update-side operation. */
  96 static inline void rcu_seq_start(unsigned long *sp)
  97 {
  98         WRITE_ONCE(*sp, *sp + 1);
  99         smp_mb(); /* Ensure update-side operation after counter increment. */
 100         WARN_ON_ONCE(rcu_seq_state(*sp) != 1);
 101 }
 102
 103 /* Adjust sequence number for end of update-side operation. */
 104 static inline void rcu_seq_end(unsigned long *sp)
 105 {
 106         smp_mb(); /* Ensure update-side operation before counter increment. */
 107         WARN_ON_ONCE(!rcu_seq_state(*sp));
 108         WRITE_ONCE(*sp, (*sp | RCU_SEQ_STATE_MASK) + 1);
 109 }
 110
 111 /* Take a snapshot of the update side's sequence number. */
 112 static inline unsigned long rcu_seq_snap(unsigned long *sp)
 113 {
 114         unsigned long s;
 115
 116         s = (READ_ONCE(*sp) + 2 * RCU_SEQ_STATE_MASK + 1) & ~RCU_SEQ_STATE_MASK;
 117         smp_mb(); /* Above access must not bleed into critical section. */
 118         return s;
 119 }
 120
 121 /* Return the current value the update side's sequence number, no ordering. */
 122 static inline unsigned long rcu_seq_current(unsigned long *sp)
 123 {
 124         return READ_ONCE(*sp);
 125 }
 126
 127 /*
 128  * Given a snapshot from rcu_seq_snap(), determine whether or not a
 129  * full update-side operation has occurred.
 130  */
 131 static inline bool rcu_seq_done(unsigned long *sp, unsigned long s)
 132 {
 133         return ULONG_CMP_GE(READ_ONCE(*sp), s);
 134 }
 135
 136 /*
 137  * debug_rcu_head_queue()/debug_rcu_head_unqueue() are used internally
 138  * by call_rcu() and rcu callback execution, and are therefore not part of the
 139  * RCU API. Leaving in rcupdate.h because they are used by all RCU flavors.
 140  */
 141
 142 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
 143 # define STATE_RCU_HEAD_READY   0
 144 # define STATE_RCU_HEAD_QUEUED  1
 145
 146 extern struct debug_obj_descr rcuhead_debug_descr;
 147
 148 static inline int debug_rcu_head_queue(struct rcu_head *head)
 149 {
 150         int r1;
 151
 152         r1 = debug_object_activate(head, &rcuhead_debug_descr);
 153         debug_object_active_state(head, &rcuhead_debug_descr,
 154                                   STATE_RCU_HEAD_READY,
 155                                   STATE_RCU_HEAD_QUEUED);
 156         return r1;
 157 }
 158
 159 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
 160 {
 161         debug_object_active_state(head, &rcuhead_debug_descr,
 162                                   STATE_RCU_HEAD_QUEUED,
 163                                   STATE_RCU_HEAD_READY);
 164         debug_object_deactivate(head, &rcuhead_debug_descr);
 165 }
 166 #else   /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 167 static inline int debug_rcu_head_queue(struct rcu_head *head)
 168 {
 169         return 0;
 170 }
 171
 172 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
 173 {
 174 }
 175 #endif  /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 176
 177 void kfree(const void *);
 178
 179 /*
 180  * Reclaim the specified callback, either by invoking it (non-lazy case)
 181  * or freeing it directly (lazy case).  Return true if lazy, false otherwise.
 182  */
 183 static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head)
 184 {
 185         unsigned long offset = (unsigned long)head->func;
 186
 187         rcu_lock_acquire(&rcu_callback_map);
 188         if (__is_kfree_rcu_offset(offset)) {
 189                 RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset);)
 190                 kfree((void *)head - offset);
 191                 rcu_lock_release(&rcu_callback_map);
 192                 return true;
 193         } else {
 194                 RCU_TRACE(trace_rcu_invoke_callback(rn, head);)
 195                 head->func(head);
 196                 rcu_lock_release(&rcu_callback_map);
 197                 return false;
 198         }
 199 }
 200
 201 #ifdef CONFIG_RCU_STALL_COMMON
 202
 203 extern int rcu_cpu_stall_suppress;
 204 int rcu_jiffies_till_stall_check(void);
 205
 206 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
 207
 208 /*
 209  * Strings used in tracepoints need to be exported via the
 210  * tracing system such that tools like perf and trace-cmd can
 211  * translate the string address pointers to actual text.
 212  */
 213 #define TPS(x)  tracepoint_string(x)
 214
 215 /*
 216  * Dump the ftrace buffer, but only one time per callsite per boot.
 217  */
 218 #define rcu_ftrace_dump(oops_dump_mode) \
 219 do { \
 220         static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
 221         \
 222         if (!atomic_read(&___rfd_beenhere) && \
 223             !atomic_xchg(&___rfd_beenhere, 1)) \
 224                 ftrace_dump(oops_dump_mode); \
 225 } while (0)
 226
 227 void rcu_early_boot_tests(void);
 228 void rcu_test_sync_prims(void);
 229
 230 /*
 231  * This function really isn't for public consumption, but RCU is special in
 232  * that context switches can allow the state machine to make progress.
 233  */
 234 extern void resched_cpu(int cpu);
 235
 236 #if defined(SRCU) || !defined(TINY_RCU)
 237
 238 #include <linux/rcu_node_tree.h>
 239
 240 extern int rcu_num_lvls;
 241 extern int num_rcu_lvl[];
 242 extern int rcu_num_nodes;
 243 static bool rcu_fanout_exact;
 244 static int rcu_fanout_leaf;
 245
 246 /*
 247  * Compute the per-level fanout, either using the exact fanout specified
 248  * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
 249  */
 250 static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
 251 {
 252         int i;
 253
 254         if (rcu_fanout_exact) {
 255                 levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
 256                 for (i = rcu_num_lvls - 2; i >= 0; i--)
 257                         levelspread[i] = RCU_FANOUT;
 258         } else {
 259                 int ccur;
 260                 int cprv;
 261
 262                 cprv = nr_cpu_ids;
 263                 for (i = rcu_num_lvls - 1; i >= 0; i--) {
 264                         ccur = levelcnt[i];
 265                         levelspread[i] = (cprv + ccur - 1) / ccur;
 266                         cprv = ccur;
 267                 }
 268         }
 269 }
 270
 271 /*
 272  * Do a full breadth-first scan of the rcu_node structures for the
 273  * specified rcu_state structure.
 274  */
 275 #define rcu_for_each_node_breadth_first(rsp, rnp) \
 276         for ((rnp) = &(rsp)->node[0]; \
 277              (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 278
 279 /*
 280  * Do a breadth-first scan of the non-leaf rcu_node structures for the
 281  * specified rcu_state structure.  Note that if there is a singleton
 282  * rcu_node tree with but one rcu_node structure, this loop is a no-op.
 283  */
 284 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
 285         for ((rnp) = &(rsp)->node[0]; \
 286              (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
 287
 288 /*
 289  * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
 290  * structure.  Note that if there is a singleton rcu_node tree with but
 291  * one rcu_node structure, this loop -will- visit the rcu_node structure.
 292  * It is still a leaf node, even if it is also the root node.
 293  */
 294 #define rcu_for_each_leaf_node(rsp, rnp) \
 295         for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
 296              (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 297
 298 /*
 299  * Iterate over all possible CPUs in a leaf RCU node.
 300  */
 301 #define for_each_leaf_node_possible_cpu(rnp, cpu) \
 302         for ((cpu) = cpumask_next(rnp->grplo - 1, cpu_possible_mask); \
 303              cpu <= rnp->grphi; \
 304              cpu = cpumask_next((cpu), cpu_possible_mask))
 305
 306 /*
 307  * Wrappers for the rcu_node::lock acquire and release.
 308  *
 309  * Because the rcu_nodes form a tree, the tree traversal locking will observe
 310  * different lock values, this in turn means that an UNLOCK of one level
 311  * followed by a LOCK of another level does not imply a full memory barrier;
 312  * and most importantly transitivity is lost.
 313  *
 314  * In order to restore full ordering between tree levels, augment the regular
 315  * lock acquire functions with smp_mb__after_unlock_lock().
 316  *
 317  * As ->lock of struct rcu_node is a __private field, therefore one should use
 318  * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
 319  */
 320 #define raw_spin_lock_rcu_node(p)                                       \
 321 do {                                                                    \
 322         raw_spin_lock(&ACCESS_PRIVATE(p, lock));                        \
 323         smp_mb__after_unlock_lock();                                    \
 324 } while (0)
 325
 326 #define raw_spin_unlock_rcu_node(p) raw_spin_unlock(&ACCESS_PRIVATE(p, lock))
 327
 328 #define raw_spin_lock_irq_rcu_node(p)                                   \
 329 do {                                                                    \
 330         raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock));                    \
 331         smp_mb__after_unlock_lock();                                    \
 332 } while (0)
 333
 334 #define raw_spin_unlock_irq_rcu_node(p)                                 \
 335         raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock))
 336
 337 #define raw_spin_lock_irqsave_rcu_node(p, flags)                        \
 338 do {                                                                    \
 339         raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags); \
 340         smp_mb__after_unlock_lock();                                    \
 341 } while (0)
 342
 343 #define raw_spin_unlock_irqrestore_rcu_node(p, flags)                   \
 344         raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags)     \
 345
 346 #define raw_spin_trylock_rcu_node(p)                                    \
 347 ({                                                                      \
 348         bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock));    \
 349                                                                         \
 350         if (___locked)                                                  \
 351                 smp_mb__after_unlock_lock();                            \
 352         ___locked;                                                      \
 353 })
 354
 355 #endif /* #if defined(SRCU) || !defined(TINY_RCU) */
 356
 357 #ifdef CONFIG_TINY_RCU
 358 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
 359 static inline bool rcu_gp_is_normal(void) { return true; }
 360 static inline bool rcu_gp_is_expedited(void) { return false; }
 361 static inline void rcu_expedite_gp(void) { }
 362 static inline void rcu_unexpedite_gp(void) { }
 363 #else /* #ifdef CONFIG_TINY_RCU */
 364 bool rcu_gp_is_normal(void);     /* Internal RCU use. */
 365 bool rcu_gp_is_expedited(void);  /* Internal RCU use. */
 366 void rcu_expedite_gp(void);
 367 void rcu_unexpedite_gp(void);
 368 void rcupdate_announce_bootup_oddness(void);
 369 #endif /* #else #ifdef CONFIG_TINY_RCU */
 370
 371 #define RCU_SCHEDULER_INACTIVE  0
 372 #define RCU_SCHEDULER_INIT      1
 373 #define RCU_SCHEDULER_RUNNING   2
 374
 375 #ifdef CONFIG_TINY_RCU
 376 static inline void rcu_request_urgent_qs_task(struct task_struct *t) { }
 377 #else /* #ifdef CONFIG_TINY_RCU */
 378 void rcu_request_urgent_qs_task(struct task_struct *t);
 379 #endif /* #else #ifdef CONFIG_TINY_RCU */
 380
 381 enum rcutorture_type {
 382         RCU_FLAVOR,
 383         RCU_BH_FLAVOR,
 384         RCU_SCHED_FLAVOR,
 385         RCU_TASKS_FLAVOR,
 386         SRCU_FLAVOR,
 387         INVALID_RCU_FLAVOR
 388 };
 389
 390 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
 391 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
 392                             unsigned long *gpnum, unsigned long *completed);
 393 void rcutorture_record_test_transition(void);
 394 void rcutorture_record_progress(unsigned long vernum);
 395 void do_trace_rcu_torture_read(const char *rcutorturename,
 396                                struct rcu_head *rhp,
 397                                unsigned long secs,
 398                                unsigned long c_old,
 399                                unsigned long c);
 400 #else
 401 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
 402                                           int *flags,
 403                                           unsigned long *gpnum,
 404                                           unsigned long *completed)
 405 {
 406         *flags = 0;
 407         *gpnum = 0;
 408         *completed = 0;
 409 }
 410 static inline void rcutorture_record_test_transition(void) { }
 411 static inline void rcutorture_record_progress(unsigned long vernum) { }
 412 #ifdef CONFIG_RCU_TRACE
 413 void do_trace_rcu_torture_read(const char *rcutorturename,
 414                                struct rcu_head *rhp,
 415                                unsigned long secs,
 416                                unsigned long c_old,
 417                                unsigned long c);
 418 #else
 419 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
 420         do { } while (0)
 421 #endif
 422 #endif
 423
 424 #ifdef CONFIG_TINY_SRCU
 425
 426 static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
 427                                            struct srcu_struct *sp, int *flags,
 428                                            unsigned long *gpnum,
 429                                            unsigned long *completed)
 430 {
 431         if (test_type != SRCU_FLAVOR)
 432                 return;
 433         *flags = 0;
 434         *completed = sp->srcu_idx;
 435         *gpnum = *completed;
 436 }
 437
 438 #elif defined(CONFIG_TREE_SRCU)
 439
 440 void srcutorture_get_gp_data(enum rcutorture_type test_type,
 441                              struct srcu_struct *sp, int *flags,
 442                              unsigned long *gpnum, unsigned long *completed);
 443
 444 #endif
 445
 446 #ifdef CONFIG_TINY_RCU
 447 static inline unsigned long rcu_batches_started(void) { return 0; }
 448 static inline unsigned long rcu_batches_started_bh(void) { return 0; }
 449 static inline unsigned long rcu_batches_started_sched(void) { return 0; }
 450 static inline unsigned long rcu_batches_completed(void) { return 0; }
 451 static inline unsigned long rcu_batches_completed_bh(void) { return 0; }
 452 static inline unsigned long rcu_batches_completed_sched(void) { return 0; }
 453 static inline unsigned long rcu_exp_batches_completed(void) { return 0; }
 454 static inline unsigned long rcu_exp_batches_completed_sched(void) { return 0; }
 455 static inline unsigned long
 456 srcu_batches_completed(struct srcu_struct *sp) { return 0; }
 457 static inline void rcu_force_quiescent_state(void) { }
 458 static inline void rcu_bh_force_quiescent_state(void) { }
 459 static inline void rcu_sched_force_quiescent_state(void) { }
 460 static inline void show_rcu_gp_kthreads(void) { }
 461 #else /* #ifdef CONFIG_TINY_RCU */
 462 extern unsigned long rcutorture_testseq;
 463 extern unsigned long rcutorture_vernum;
 464 unsigned long rcu_batches_started(void);
 465 unsigned long rcu_batches_started_bh(void);
 466 unsigned long rcu_batches_started_sched(void);
 467 unsigned long rcu_batches_completed(void);
 468 unsigned long rcu_batches_completed_bh(void);
 469 unsigned long rcu_batches_completed_sched(void);
 470 unsigned long rcu_exp_batches_completed(void);
 471 unsigned long rcu_exp_batches_completed_sched(void);
 472 unsigned long srcu_batches_completed(struct srcu_struct *sp);
 473 void show_rcu_gp_kthreads(void);
 474 void rcu_force_quiescent_state(void);
 475 void rcu_bh_force_quiescent_state(void);
 476 void rcu_sched_force_quiescent_state(void);
 477 #endif /* #else #ifdef CONFIG_TINY_RCU */
 478
 479 #ifdef CONFIG_RCU_NOCB_CPU
 480 bool rcu_is_nocb_cpu(int cpu);
 481 #else
 482 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
 483 #endif
 484
 485 #endif /* __LINUX_RCU_H */