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 /* Offset to allow for unmatched rcu_irq_{enter,exit}(). */
  34 #define DYNTICK_IRQ_NONIDLE     ((LONG_MAX / 2) + 1)
  35
  36
  37 /*
  38  * Grace-period counter management.
  39  */
  40
  41 #define RCU_SEQ_CTR_SHIFT       2
  42 #define RCU_SEQ_STATE_MASK      ((1 << RCU_SEQ_CTR_SHIFT) - 1)
  43
  44 /*
  45  * Return the counter portion of a sequence number previously returned
  46  * by rcu_seq_snap() or rcu_seq_current().
  47  */
  48 static inline unsigned long rcu_seq_ctr(unsigned long s)
  49 {
  50         return s >> RCU_SEQ_CTR_SHIFT;
  51 }
  52
  53 /*
  54  * Return the state portion of a sequence number previously returned
  55  * by rcu_seq_snap() or rcu_seq_current().
  56  */
  57 static inline int rcu_seq_state(unsigned long s)
  58 {
  59         return s & RCU_SEQ_STATE_MASK;
  60 }
  61
  62 /*
  63  * Set the state portion of the pointed-to sequence number.
  64  * The caller is responsible for preventing conflicting updates.
  65  */
  66 static inline void rcu_seq_set_state(unsigned long *sp, int newstate)
  67 {
  68         WARN_ON_ONCE(newstate & ~RCU_SEQ_STATE_MASK);
  69         WRITE_ONCE(*sp, (*sp & ~RCU_SEQ_STATE_MASK) + newstate);
  70 }
  71
  72 /* Adjust sequence number for start of update-side operation. */
  73 static inline void rcu_seq_start(unsigned long *sp)
  74 {
  75         WRITE_ONCE(*sp, *sp + 1);
  76         smp_mb(); /* Ensure update-side operation after counter increment. */
  77         WARN_ON_ONCE(rcu_seq_state(*sp) != 1);
  78 }
  79
  80 /* Compute the end-of-grace-period value for the specified sequence number. */
  81 static inline unsigned long rcu_seq_endval(unsigned long *sp)
  82 {
  83         return (*sp | RCU_SEQ_STATE_MASK) + 1;
  84 }
  85
  86 /* Adjust sequence number for end of update-side operation. */
  87 static inline void rcu_seq_end(unsigned long *sp)
  88 {
  89         smp_mb(); /* Ensure update-side operation before counter increment. */
  90         WARN_ON_ONCE(!rcu_seq_state(*sp));
  91         WRITE_ONCE(*sp, rcu_seq_endval(sp));
  92 }
  93
  94 /* Take a snapshot of the update side's sequence number. */
  95 static inline unsigned long rcu_seq_snap(unsigned long *sp)
  96 {
  97         unsigned long s;
  98
  99         s = (READ_ONCE(*sp) + 2 * RCU_SEQ_STATE_MASK + 1) & ~RCU_SEQ_STATE_MASK;
 100         smp_mb(); /* Above access must not bleed into critical section. */
 101         return s;
 102 }
 103
 104 /* Return the current value the update side's sequence number, no ordering. */
 105 static inline unsigned long rcu_seq_current(unsigned long *sp)
 106 {
 107         return READ_ONCE(*sp);
 108 }
 109
 110 /*
 111  * Given a snapshot from rcu_seq_snap(), determine whether or not a
 112  * full update-side operation has occurred.
 113  */
 114 static inline bool rcu_seq_done(unsigned long *sp, unsigned long s)
 115 {
 116         return ULONG_CMP_GE(READ_ONCE(*sp), s);
 117 }
 118
 119 /*
 120  * debug_rcu_head_queue()/debug_rcu_head_unqueue() are used internally
 121  * by call_rcu() and rcu callback execution, and are therefore not part of the
 122  * RCU API. Leaving in rcupdate.h because they are used by all RCU flavors.
 123  */
 124
 125 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
 126 # define STATE_RCU_HEAD_READY   0
 127 # define STATE_RCU_HEAD_QUEUED  1
 128
 129 extern struct debug_obj_descr rcuhead_debug_descr;
 130
 131 static inline int debug_rcu_head_queue(struct rcu_head *head)
 132 {
 133         int r1;
 134
 135         r1 = debug_object_activate(head, &rcuhead_debug_descr);
 136         debug_object_active_state(head, &rcuhead_debug_descr,
 137                                   STATE_RCU_HEAD_READY,
 138                                   STATE_RCU_HEAD_QUEUED);
 139         return r1;
 140 }
 141
 142 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
 143 {
 144         debug_object_active_state(head, &rcuhead_debug_descr,
 145                                   STATE_RCU_HEAD_QUEUED,
 146                                   STATE_RCU_HEAD_READY);
 147         debug_object_deactivate(head, &rcuhead_debug_descr);
 148 }
 149 #else   /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 150 static inline int debug_rcu_head_queue(struct rcu_head *head)
 151 {
 152         return 0;
 153 }
 154
 155 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
 156 {
 157 }
 158 #endif  /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 159
 160 void kfree(const void *);
 161
 162 /*
 163  * Reclaim the specified callback, either by invoking it (non-lazy case)
 164  * or freeing it directly (lazy case).  Return true if lazy, false otherwise.
 165  */
 166 static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head)
 167 {
 168         unsigned long offset = (unsigned long)head->func;
 169
 170         rcu_lock_acquire(&rcu_callback_map);
 171         if (__is_kfree_rcu_offset(offset)) {
 172                 RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset);)
 173                 kfree((void *)head - offset);
 174                 rcu_lock_release(&rcu_callback_map);
 175                 return true;
 176         } else {
 177                 RCU_TRACE(trace_rcu_invoke_callback(rn, head);)
 178                 head->func(head);
 179                 rcu_lock_release(&rcu_callback_map);
 180                 return false;
 181         }
 182 }
 183
 184 #ifdef CONFIG_RCU_STALL_COMMON
 185
 186 extern int rcu_cpu_stall_suppress;
 187 int rcu_jiffies_till_stall_check(void);
 188
 189 #define rcu_ftrace_dump_stall_suppress() \
 190 do { \
 191         if (!rcu_cpu_stall_suppress) \
 192                 rcu_cpu_stall_suppress = 3; \
 193 } while (0)
 194
 195 #define rcu_ftrace_dump_stall_unsuppress() \
 196 do { \
 197         if (rcu_cpu_stall_suppress == 3) \
 198                 rcu_cpu_stall_suppress = 0; \
 199 } while (0)
 200
 201 #else /* #endif #ifdef CONFIG_RCU_STALL_COMMON */
 202 #define rcu_ftrace_dump_stall_suppress()
 203 #define rcu_ftrace_dump_stall_unsuppress()
 204 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
 205
 206 /*
 207  * Strings used in tracepoints need to be exported via the
 208  * tracing system such that tools like perf and trace-cmd can
 209  * translate the string address pointers to actual text.
 210  */
 211 #define TPS(x)  tracepoint_string(x)
 212
 213 /*
 214  * Dump the ftrace buffer, but only one time per callsite per boot.
 215  */
 216 #define rcu_ftrace_dump(oops_dump_mode) \
 217 do { \
 218         static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
 219         \
 220         if (!atomic_read(&___rfd_beenhere) && \
 221             !atomic_xchg(&___rfd_beenhere, 1)) { \
 222                 tracing_off(); \
 223                 rcu_ftrace_dump_stall_suppress(); \
 224                 ftrace_dump(oops_dump_mode); \
 225                 rcu_ftrace_dump_stall_unsuppress(); \
 226         } \
 227 } while (0)
 228
 229 void rcu_early_boot_tests(void);
 230 void rcu_test_sync_prims(void);
 231
 232 /*
 233  * This function really isn't for public consumption, but RCU is special in
 234  * that context switches can allow the state machine to make progress.
 235  */
 236 extern void resched_cpu(int cpu);
 237
 238 #if defined(SRCU) || !defined(TINY_RCU)
 239
 240 #include <linux/rcu_node_tree.h>
 241
 242 extern int rcu_num_lvls;
 243 extern int num_rcu_lvl[];
 244 extern int rcu_num_nodes;
 245 static bool rcu_fanout_exact;
 246 static int rcu_fanout_leaf;
 247
 248 /*
 249  * Compute the per-level fanout, either using the exact fanout specified
 250  * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
 251  */
 252 static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
 253 {
 254         int i;
 255
 256         if (rcu_fanout_exact) {
 257                 levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
 258                 for (i = rcu_num_lvls - 2; i >= 0; i--)
 259                         levelspread[i] = RCU_FANOUT;
 260         } else {
 261                 int ccur;
 262                 int cprv;
 263
 264                 cprv = nr_cpu_ids;
 265                 for (i = rcu_num_lvls - 1; i >= 0; i--) {
 266                         ccur = levelcnt[i];
 267                         levelspread[i] = (cprv + ccur - 1) / ccur;
 268                         cprv = ccur;
 269                 }
 270         }
 271 }
 272
 273 /*
 274  * Do a full breadth-first scan of the rcu_node structures for the
 275  * specified rcu_state structure.
 276  */
 277 #define rcu_for_each_node_breadth_first(rsp, rnp) \
 278         for ((rnp) = &(rsp)->node[0]; \
 279              (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 280
 281 /*
 282  * Do a breadth-first scan of the non-leaf rcu_node structures for the
 283  * specified rcu_state structure.  Note that if there is a singleton
 284  * rcu_node tree with but one rcu_node structure, this loop is a no-op.
 285  */
 286 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
 287         for ((rnp) = &(rsp)->node[0]; \
 288              (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
 289
 290 /*
 291  * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
 292  * structure.  Note that if there is a singleton rcu_node tree with but
 293  * one rcu_node structure, this loop -will- visit the rcu_node structure.
 294  * It is still a leaf node, even if it is also the root node.
 295  */
 296 #define rcu_for_each_leaf_node(rsp, rnp) \
 297         for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
 298              (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
 299
 300 /*
 301  * Iterate over all possible CPUs in a leaf RCU node.
 302  */
 303 #define for_each_leaf_node_possible_cpu(rnp, cpu) \
 304         for ((cpu) = cpumask_next((rnp)->grplo - 1, cpu_possible_mask); \
 305              (cpu) <= rnp->grphi; \
 306              (cpu) = cpumask_next((cpu), cpu_possible_mask))
 307
 308 /*
 309  * Iterate over all CPUs in a leaf RCU node's specified mask.
 310  */
 311 #define rcu_find_next_bit(rnp, cpu, mask) \
 312         ((rnp)->grplo + find_next_bit(&(mask), BITS_PER_LONG, (cpu)))
 313 #define for_each_leaf_node_cpu_mask(rnp, cpu, mask) \
 314         for ((cpu) = rcu_find_next_bit((rnp), 0, (mask)); \
 315              (cpu) <= rnp->grphi; \
 316              (cpu) = rcu_find_next_bit((rnp), (cpu) + 1 - (rnp->grplo), (mask)))
 317
 318 /*
 319  * Wrappers for the rcu_node::lock acquire and release.
 320  *
 321  * Because the rcu_nodes form a tree, the tree traversal locking will observe
 322  * different lock values, this in turn means that an UNLOCK of one level
 323  * followed by a LOCK of another level does not imply a full memory barrier;
 324  * and most importantly transitivity is lost.
 325  *
 326  * In order to restore full ordering between tree levels, augment the regular
 327  * lock acquire functions with smp_mb__after_unlock_lock().
 328  *
 329  * As ->lock of struct rcu_node is a __private field, therefore one should use
 330  * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
 331  */
 332 #define raw_spin_lock_rcu_node(p)                                       \
 333 do {                                                                    \
 334         raw_spin_lock(&ACCESS_PRIVATE(p, lock));                        \
 335         smp_mb__after_unlock_lock();                                    \
 336 } while (0)
 337
 338 #define raw_spin_unlock_rcu_node(p) raw_spin_unlock(&ACCESS_PRIVATE(p, lock))
 339
 340 #define raw_spin_lock_irq_rcu_node(p)                                   \
 341 do {                                                                    \
 342         raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock));                    \
 343         smp_mb__after_unlock_lock();                                    \
 344 } while (0)
 345
 346 #define raw_spin_unlock_irq_rcu_node(p)                                 \
 347         raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock))
 348
 349 #define raw_spin_lock_irqsave_rcu_node(p, flags)                        \
 350 do {                                                                    \
 351         raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags); \
 352         smp_mb__after_unlock_lock();                                    \
 353 } while (0)
 354
 355 #define raw_spin_unlock_irqrestore_rcu_node(p, flags)                   \
 356         raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags)
 357
 358 #define raw_spin_trylock_rcu_node(p)                                    \
 359 ({                                                                      \
 360         bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock));    \
 361                                                                         \
 362         if (___locked)                                                  \
 363                 smp_mb__after_unlock_lock();                            \
 364         ___locked;                                                      \
 365 })
 366
 367 #define raw_lockdep_assert_held_rcu_node(p)                             \
 368         lockdep_assert_held(&ACCESS_PRIVATE(p, lock))
 369
 370 #endif /* #if defined(SRCU) || !defined(TINY_RCU) */
 371
 372 #ifdef CONFIG_TINY_RCU
 373 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
 374 static inline bool rcu_gp_is_normal(void) { return true; }
 375 static inline bool rcu_gp_is_expedited(void) { return false; }
 376 static inline void rcu_expedite_gp(void) { }
 377 static inline void rcu_unexpedite_gp(void) { }
 378 static inline void rcu_request_urgent_qs_task(struct task_struct *t) { }
 379 #else /* #ifdef CONFIG_TINY_RCU */
 380 bool rcu_gp_is_normal(void);     /* Internal RCU use. */
 381 bool rcu_gp_is_expedited(void);  /* Internal RCU use. */
 382 void rcu_expedite_gp(void);
 383 void rcu_unexpedite_gp(void);
 384 void rcupdate_announce_bootup_oddness(void);
 385 void rcu_request_urgent_qs_task(struct task_struct *t);
 386 #endif /* #else #ifdef CONFIG_TINY_RCU */
 387
 388 #define RCU_SCHEDULER_INACTIVE  0
 389 #define RCU_SCHEDULER_INIT      1
 390 #define RCU_SCHEDULER_RUNNING   2
 391
 392 enum rcutorture_type {
 393         RCU_FLAVOR,
 394         RCU_BH_FLAVOR,
 395         RCU_SCHED_FLAVOR,
 396         RCU_TASKS_FLAVOR,
 397         SRCU_FLAVOR,
 398         INVALID_RCU_FLAVOR
 399 };
 400
 401 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
 402 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
 403                             unsigned long *gpnum, unsigned long *completed);
 404 void rcutorture_record_test_transition(void);
 405 void rcutorture_record_progress(unsigned long vernum);
 406 void do_trace_rcu_torture_read(const char *rcutorturename,
 407                                struct rcu_head *rhp,
 408                                unsigned long secs,
 409                                unsigned long c_old,
 410                                unsigned long c);
 411 #else
 412 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
 413                                           int *flags,
 414                                           unsigned long *gpnum,
 415                                           unsigned long *completed)
 416 {
 417         *flags = 0;
 418         *gpnum = 0;
 419         *completed = 0;
 420 }
 421 static inline void rcutorture_record_test_transition(void) { }
 422 static inline void rcutorture_record_progress(unsigned long vernum) { }
 423 #ifdef CONFIG_RCU_TRACE
 424 void do_trace_rcu_torture_read(const char *rcutorturename,
 425                                struct rcu_head *rhp,
 426                                unsigned long secs,
 427                                unsigned long c_old,
 428                                unsigned long c);
 429 #else
 430 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
 431         do { } while (0)
 432 #endif
 433 #endif
 434
 435 #ifdef CONFIG_TINY_SRCU
 436
 437 static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
 438                                            struct srcu_struct *sp, int *flags,
 439                                            unsigned long *gpnum,
 440                                            unsigned long *completed)
 441 {
 442         if (test_type != SRCU_FLAVOR)
 443                 return;
 444         *flags = 0;
 445         *completed = sp->srcu_idx;
 446         *gpnum = *completed;
 447 }
 448
 449 #elif defined(CONFIG_TREE_SRCU)
 450
 451 void srcutorture_get_gp_data(enum rcutorture_type test_type,
 452                              struct srcu_struct *sp, int *flags,
 453                              unsigned long *gpnum, unsigned long *completed);
 454
 455 #endif
 456
 457 #ifdef CONFIG_TINY_RCU
 458 static inline unsigned long rcu_batches_started(void) { return 0; }
 459 static inline unsigned long rcu_batches_started_bh(void) { return 0; }
 460 static inline unsigned long rcu_batches_started_sched(void) { return 0; }
 461 static inline unsigned long rcu_batches_completed(void) { return 0; }
 462 static inline unsigned long rcu_batches_completed_bh(void) { return 0; }
 463 static inline unsigned long rcu_batches_completed_sched(void) { return 0; }
 464 static inline unsigned long rcu_exp_batches_completed(void) { return 0; }
 465 static inline unsigned long rcu_exp_batches_completed_sched(void) { return 0; }
 466 static inline unsigned long
 467 srcu_batches_completed(struct srcu_struct *sp) { return 0; }
 468 static inline void rcu_force_quiescent_state(void) { }
 469 static inline void rcu_bh_force_quiescent_state(void) { }
 470 static inline void rcu_sched_force_quiescent_state(void) { }
 471 static inline void show_rcu_gp_kthreads(void) { }
 472 #else /* #ifdef CONFIG_TINY_RCU */
 473 extern unsigned long rcutorture_testseq;
 474 extern unsigned long rcutorture_vernum;
 475 unsigned long rcu_batches_started(void);
 476 unsigned long rcu_batches_started_bh(void);
 477 unsigned long rcu_batches_started_sched(void);
 478 unsigned long rcu_batches_completed(void);
 479 unsigned long rcu_batches_completed_bh(void);
 480 unsigned long rcu_batches_completed_sched(void);
 481 unsigned long rcu_exp_batches_completed(void);
 482 unsigned long rcu_exp_batches_completed_sched(void);
 483 unsigned long srcu_batches_completed(struct srcu_struct *sp);
 484 void show_rcu_gp_kthreads(void);
 485 void rcu_force_quiescent_state(void);
 486 void rcu_bh_force_quiescent_state(void);
 487 void rcu_sched_force_quiescent_state(void);
 488 extern struct workqueue_struct *rcu_gp_wq;
 489 #endif /* #else #ifdef CONFIG_TINY_RCU */
 490
 491 #ifdef CONFIG_RCU_NOCB_CPU
 492 bool rcu_is_nocb_cpu(int cpu);
 493 #else
 494 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
 495 #endif
 496
 497 #endif /* __LINUX_RCU_H */