kernel/rcu/rcu.h

   1 /* SPDX-License-Identifier: GPL-2.0+ */
   2 /*
   3  * Read-Copy Update definitions shared among RCU implementations.
   4  *
   5  * Copyright IBM Corporation, 2011
   6  *
   7  * Author: Paul E. McKenney <paulmck@linux.ibm.com>
   8  */
   9
  10 #ifndef __LINUX_RCU_H
  11 #define __LINUX_RCU_H
  12
  13 #include <trace/events/rcu.h>
  14
  15 /* Offset to allow distinguishing irq vs. task-based idle entry/exit. */
  16 #define DYNTICK_IRQ_NONIDLE     ((LONG_MAX / 2) + 1)
  17
  18
  19 /*
  20  * Grace-period counter management.
  21  */
  22
  23 #define RCU_SEQ_CTR_SHIFT       2
  24 #define RCU_SEQ_STATE_MASK      ((1 << RCU_SEQ_CTR_SHIFT) - 1)
  25
  26 /*
  27  * Return the counter portion of a sequence number previously returned
  28  * by rcu_seq_snap() or rcu_seq_current().
  29  */
  30 static inline unsigned long rcu_seq_ctr(unsigned long s)
  31 {
  32         return s >> RCU_SEQ_CTR_SHIFT;
  33 }
  34
  35 /*
  36  * Return the state portion of a sequence number previously returned
  37  * by rcu_seq_snap() or rcu_seq_current().
  38  */
  39 static inline int rcu_seq_state(unsigned long s)
  40 {
  41         return s & RCU_SEQ_STATE_MASK;
  42 }
  43
  44 /*
  45  * Set the state portion of the pointed-to sequence number.
  46  * The caller is responsible for preventing conflicting updates.
  47  */
  48 static inline void rcu_seq_set_state(unsigned long *sp, int newstate)
  49 {
  50         WARN_ON_ONCE(newstate & ~RCU_SEQ_STATE_MASK);
  51         WRITE_ONCE(*sp, (*sp & ~RCU_SEQ_STATE_MASK) + newstate);
  52 }
  53
  54 /* Adjust sequence number for start of update-side operation. */
  55 static inline void rcu_seq_start(unsigned long *sp)
  56 {
  57         WRITE_ONCE(*sp, *sp + 1);
  58         smp_mb(); /* Ensure update-side operation after counter increment. */
  59         WARN_ON_ONCE(rcu_seq_state(*sp) != 1);
  60 }
  61
  62 /* Compute the end-of-grace-period value for the specified sequence number. */
  63 static inline unsigned long rcu_seq_endval(unsigned long *sp)
  64 {
  65         return (*sp | RCU_SEQ_STATE_MASK) + 1;
  66 }
  67
  68 /* Adjust sequence number for end of update-side operation. */
  69 static inline void rcu_seq_end(unsigned long *sp)
  70 {
  71         smp_mb(); /* Ensure update-side operation before counter increment. */
  72         WARN_ON_ONCE(!rcu_seq_state(*sp));
  73         WRITE_ONCE(*sp, rcu_seq_endval(sp));
  74 }
  75
  76 /*
  77  * rcu_seq_snap - Take a snapshot of the update side's sequence number.
  78  *
  79  * This function returns the earliest value of the grace-period sequence number
  80  * that will indicate that a full grace period has elapsed since the current
  81  * time.  Once the grace-period sequence number has reached this value, it will
  82  * be safe to invoke all callbacks that have been registered prior to the
  83  * current time. This value is the current grace-period number plus two to the
  84  * power of the number of low-order bits reserved for state, then rounded up to
  85  * the next value in which the state bits are all zero.
  86  */
  87 static inline unsigned long rcu_seq_snap(unsigned long *sp)
  88 {
  89         unsigned long s;
  90
  91         s = (READ_ONCE(*sp) + 2 * RCU_SEQ_STATE_MASK + 1) & ~RCU_SEQ_STATE_MASK;
  92         smp_mb(); /* Above access must not bleed into critical section. */
  93         return s;
  94 }
  95
  96 /* Return the current value the update side's sequence number, no ordering. */
  97 static inline unsigned long rcu_seq_current(unsigned long *sp)
  98 {
  99         return READ_ONCE(*sp);
 100 }
 101
 102 /*
 103  * Given a snapshot from rcu_seq_snap(), determine whether or not the
 104  * corresponding update-side operation has started.
 105  */
 106 static inline bool rcu_seq_started(unsigned long *sp, unsigned long s)
 107 {
 108         return ULONG_CMP_LT((s - 1) & ~RCU_SEQ_STATE_MASK, READ_ONCE(*sp));
 109 }
 110
 111 /*
 112  * Given a snapshot from rcu_seq_snap(), determine whether or not a
 113  * full update-side operation has occurred.
 114  */
 115 static inline bool rcu_seq_done(unsigned long *sp, unsigned long s)
 116 {
 117         return ULONG_CMP_GE(READ_ONCE(*sp), s);
 118 }
 119
 120 /*
 121  * Has a grace period completed since the time the old gp_seq was collected?
 122  */
 123 static inline bool rcu_seq_completed_gp(unsigned long old, unsigned long new)
 124 {
 125         return ULONG_CMP_LT(old, new & ~RCU_SEQ_STATE_MASK);
 126 }
 127
 128 /*
 129  * Has a grace period started since the time the old gp_seq was collected?
 130  */
 131 static inline bool rcu_seq_new_gp(unsigned long old, unsigned long new)
 132 {
 133         return ULONG_CMP_LT((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK,
 134                             new);
 135 }
 136
 137 /*
 138  * Roughly how many full grace periods have elapsed between the collection
 139  * of the two specified grace periods?
 140  */
 141 static inline unsigned long rcu_seq_diff(unsigned long new, unsigned long old)
 142 {
 143         unsigned long rnd_diff;
 144
 145         if (old == new)
 146                 return 0;
 147         /*
 148          * Compute the number of grace periods (still shifted up), plus
 149          * one if either of new and old is not an exact grace period.
 150          */
 151         rnd_diff = (new & ~RCU_SEQ_STATE_MASK) -
 152                    ((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK) +
 153                    ((new & RCU_SEQ_STATE_MASK) || (old & RCU_SEQ_STATE_MASK));
 154         if (ULONG_CMP_GE(RCU_SEQ_STATE_MASK, rnd_diff))
 155                 return 1; /* Definitely no grace period has elapsed. */
 156         return ((rnd_diff - RCU_SEQ_STATE_MASK - 1) >> RCU_SEQ_CTR_SHIFT) + 2;
 157 }
 158
 159 /*
 160  * debug_rcu_head_queue()/debug_rcu_head_unqueue() are used internally
 161  * by call_rcu() and rcu callback execution, and are therefore not part
 162  * of the RCU API. These are in rcupdate.h because they are used by all
 163  * RCU implementations.
 164  */
 165
 166 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
 167 # define STATE_RCU_HEAD_READY   0
 168 # define STATE_RCU_HEAD_QUEUED  1
 169
 170 extern struct debug_obj_descr rcuhead_debug_descr;
 171
 172 static inline int debug_rcu_head_queue(struct rcu_head *head)
 173 {
 174         int r1;
 175
 176         r1 = debug_object_activate(head, &rcuhead_debug_descr);
 177         debug_object_active_state(head, &rcuhead_debug_descr,
 178                                   STATE_RCU_HEAD_READY,
 179                                   STATE_RCU_HEAD_QUEUED);
 180         return r1;
 181 }
 182
 183 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
 184 {
 185         debug_object_active_state(head, &rcuhead_debug_descr,
 186                                   STATE_RCU_HEAD_QUEUED,
 187                                   STATE_RCU_HEAD_READY);
 188         debug_object_deactivate(head, &rcuhead_debug_descr);
 189 }
 190 #else   /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 191 static inline int debug_rcu_head_queue(struct rcu_head *head)
 192 {
 193         return 0;
 194 }
 195
 196 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
 197 {
 198 }
 199 #endif  /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
 200
 201 void kfree(const void *);
 202
 203 /*
 204  * Reclaim the specified callback, either by invoking it (non-lazy case)
 205  * or freeing it directly (lazy case).  Return true if lazy, false otherwise.
 206  */
 207 static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head)
 208 {
 209         rcu_callback_t f;
 210         unsigned long offset = (unsigned long)head->func;
 211
 212         rcu_lock_acquire(&rcu_callback_map);
 213         if (__is_kfree_rcu_offset(offset)) {
 214                 trace_rcu_invoke_kfree_callback(rn, head, offset);
 215                 kfree((void *)head - offset);
 216                 rcu_lock_release(&rcu_callback_map);
 217                 return true;
 218         } else {
 219                 trace_rcu_invoke_callback(rn, head);
 220                 f = head->func;
 221                 WRITE_ONCE(head->func, (rcu_callback_t)0L);
 222                 f(head);
 223                 rcu_lock_release(&rcu_callback_map);
 224                 return false;
 225         }
 226 }
 227
 228 #ifdef CONFIG_RCU_STALL_COMMON
 229
 230 extern int rcu_cpu_stall_suppress;
 231 extern int rcu_cpu_stall_timeout;
 232 int rcu_jiffies_till_stall_check(void);
 233
 234 #define rcu_ftrace_dump_stall_suppress() \
 235 do { \
 236         if (!rcu_cpu_stall_suppress) \
 237                 rcu_cpu_stall_suppress = 3; \
 238 } while (0)
 239
 240 #define rcu_ftrace_dump_stall_unsuppress() \
 241 do { \
 242         if (rcu_cpu_stall_suppress == 3) \
 243                 rcu_cpu_stall_suppress = 0; \
 244 } while (0)
 245
 246 #else /* #endif #ifdef CONFIG_RCU_STALL_COMMON */
 247 #define rcu_ftrace_dump_stall_suppress()
 248 #define rcu_ftrace_dump_stall_unsuppress()
 249 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
 250
 251 /*
 252  * Strings used in tracepoints need to be exported via the
 253  * tracing system such that tools like perf and trace-cmd can
 254  * translate the string address pointers to actual text.
 255  */
 256 #define TPS(x)  tracepoint_string(x)
 257
 258 /*
 259  * Dump the ftrace buffer, but only one time per callsite per boot.
 260  */
 261 #define rcu_ftrace_dump(oops_dump_mode) \
 262 do { \
 263         static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
 264         \
 265         if (!atomic_read(&___rfd_beenhere) && \
 266             !atomic_xchg(&___rfd_beenhere, 1)) { \
 267                 tracing_off(); \
 268                 rcu_ftrace_dump_stall_suppress(); \
 269                 ftrace_dump(oops_dump_mode); \
 270                 rcu_ftrace_dump_stall_unsuppress(); \
 271         } \
 272 } while (0)
 273
 274 void rcu_early_boot_tests(void);
 275 void rcu_test_sync_prims(void);
 276
 277 /*
 278  * This function really isn't for public consumption, but RCU is special in
 279  * that context switches can allow the state machine to make progress.
 280  */
 281 extern void resched_cpu(int cpu);
 282
 283 #if defined(SRCU) || !defined(TINY_RCU)
 284
 285 #include <linux/rcu_node_tree.h>
 286
 287 extern int rcu_num_lvls;
 288 extern int num_rcu_lvl[];
 289 extern int rcu_num_nodes;
 290 static bool rcu_fanout_exact;
 291 static int rcu_fanout_leaf;
 292
 293 /*
 294  * Compute the per-level fanout, either using the exact fanout specified
 295  * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
 296  */
 297 static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
 298 {
 299         int i;
 300
 301         if (rcu_fanout_exact) {
 302                 levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
 303                 for (i = rcu_num_lvls - 2; i >= 0; i--)
 304                         levelspread[i] = RCU_FANOUT;
 305         } else {
 306                 int ccur;
 307                 int cprv;
 308
 309                 cprv = nr_cpu_ids;
 310                 for (i = rcu_num_lvls - 1; i >= 0; i--) {
 311                         ccur = levelcnt[i];
 312                         levelspread[i] = (cprv + ccur - 1) / ccur;
 313                         cprv = ccur;
 314                 }
 315         }
 316 }
 317
 318 /* Returns a pointer to the first leaf rcu_node structure. */
 319 #define rcu_first_leaf_node() (rcu_state.level[rcu_num_lvls - 1])
 320
 321 /* Is this rcu_node a leaf? */
 322 #define rcu_is_leaf_node(rnp) ((rnp)->level == rcu_num_lvls - 1)
 323
 324 /* Is this rcu_node the last leaf? */
 325 #define rcu_is_last_leaf_node(rnp) ((rnp) == &rcu_state.node[rcu_num_nodes - 1])
 326
 327 /*
 328  * Do a full breadth-first scan of the {s,}rcu_node structures for the
 329  * specified state structure (for SRCU) or the only rcu_state structure
 330  * (for RCU).
 331  */
 332 #define srcu_for_each_node_breadth_first(sp, rnp) \
 333         for ((rnp) = &(sp)->node[0]; \
 334              (rnp) < &(sp)->node[rcu_num_nodes]; (rnp)++)
 335 #define rcu_for_each_node_breadth_first(rnp) \
 336         srcu_for_each_node_breadth_first(&rcu_state, rnp)
 337
 338 /*
 339  * Scan the leaves of the rcu_node hierarchy for the rcu_state structure.
 340  * Note that if there is a singleton rcu_node tree with but one rcu_node
 341  * structure, this loop -will- visit the rcu_node structure.  It is still
 342  * a leaf node, even if it is also the root node.
 343  */
 344 #define rcu_for_each_leaf_node(rnp) \
 345         for ((rnp) = rcu_first_leaf_node(); \
 346              (rnp) < &rcu_state.node[rcu_num_nodes]; (rnp)++)
 347
 348 /*
 349  * Iterate over all possible CPUs in a leaf RCU node.
 350  */
 351 #define for_each_leaf_node_possible_cpu(rnp, cpu) \
 352         for ((cpu) = cpumask_next((rnp)->grplo - 1, cpu_possible_mask); \
 353              (cpu) <= rnp->grphi; \
 354              (cpu) = cpumask_next((cpu), cpu_possible_mask))
 355
 356 /*
 357  * Iterate over all CPUs in a leaf RCU node's specified mask.
 358  */
 359 #define rcu_find_next_bit(rnp, cpu, mask) \
 360         ((rnp)->grplo + find_next_bit(&(mask), BITS_PER_LONG, (cpu)))
 361 #define for_each_leaf_node_cpu_mask(rnp, cpu, mask) \
 362         for ((cpu) = rcu_find_next_bit((rnp), 0, (mask)); \
 363              (cpu) <= rnp->grphi; \
 364              (cpu) = rcu_find_next_bit((rnp), (cpu) + 1 - (rnp->grplo), (mask)))
 365
 366 /*
 367  * Wrappers for the rcu_node::lock acquire and release.
 368  *
 369  * Because the rcu_nodes form a tree, the tree traversal locking will observe
 370  * different lock values, this in turn means that an UNLOCK of one level
 371  * followed by a LOCK of another level does not imply a full memory barrier;
 372  * and most importantly transitivity is lost.
 373  *
 374  * In order to restore full ordering between tree levels, augment the regular
 375  * lock acquire functions with smp_mb__after_unlock_lock().
 376  *
 377  * As ->lock of struct rcu_node is a __private field, therefore one should use
 378  * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
 379  */
 380 #define raw_spin_lock_rcu_node(p)                                       \
 381 do {                                                                    \
 382         raw_spin_lock(&ACCESS_PRIVATE(p, lock));                        \
 383         smp_mb__after_unlock_lock();                                    \
 384 } while (0)
 385
 386 #define raw_spin_unlock_rcu_node(p) raw_spin_unlock(&ACCESS_PRIVATE(p, lock))
 387
 388 #define raw_spin_lock_irq_rcu_node(p)                                   \
 389 do {                                                                    \
 390         raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock));                    \
 391         smp_mb__after_unlock_lock();                                    \
 392 } while (0)
 393
 394 #define raw_spin_unlock_irq_rcu_node(p)                                 \
 395         raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock))
 396
 397 #define raw_spin_lock_irqsave_rcu_node(p, flags)                        \
 398 do {                                                                    \
 399         raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags); \
 400         smp_mb__after_unlock_lock();                                    \
 401 } while (0)
 402
 403 #define raw_spin_unlock_irqrestore_rcu_node(p, flags)                   \
 404         raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags)
 405
 406 #define raw_spin_trylock_rcu_node(p)                                    \
 407 ({                                                                      \
 408         bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock));    \
 409                                                                         \
 410         if (___locked)                                                  \
 411                 smp_mb__after_unlock_lock();                            \
 412         ___locked;                                                      \
 413 })
 414
 415 #define raw_lockdep_assert_held_rcu_node(p)                             \
 416         lockdep_assert_held(&ACCESS_PRIVATE(p, lock))
 417
 418 #endif /* #if defined(SRCU) || !defined(TINY_RCU) */
 419
 420 #ifdef CONFIG_SRCU
 421 void srcu_init(void);
 422 #else /* #ifdef CONFIG_SRCU */
 423 static inline void srcu_init(void) { }
 424 #endif /* #else #ifdef CONFIG_SRCU */
 425
 426 #ifdef CONFIG_TINY_RCU
 427 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
 428 static inline bool rcu_gp_is_normal(void) { return true; }
 429 static inline bool rcu_gp_is_expedited(void) { return false; }
 430 static inline void rcu_expedite_gp(void) { }
 431 static inline void rcu_unexpedite_gp(void) { }
 432 static inline void rcu_request_urgent_qs_task(struct task_struct *t) { }
 433 #else /* #ifdef CONFIG_TINY_RCU */
 434 bool rcu_gp_is_normal(void);     /* Internal RCU use. */
 435 bool rcu_gp_is_expedited(void);  /* Internal RCU use. */
 436 void rcu_expedite_gp(void);
 437 void rcu_unexpedite_gp(void);
 438 void rcupdate_announce_bootup_oddness(void);
 439 void rcu_request_urgent_qs_task(struct task_struct *t);
 440 #endif /* #else #ifdef CONFIG_TINY_RCU */
 441
 442 #define RCU_SCHEDULER_INACTIVE  0
 443 #define RCU_SCHEDULER_INIT      1
 444 #define RCU_SCHEDULER_RUNNING   2
 445
 446 enum rcutorture_type {
 447         RCU_FLAVOR,
 448         RCU_TASKS_FLAVOR,
 449         SRCU_FLAVOR,
 450         INVALID_RCU_FLAVOR
 451 };
 452
 453 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
 454 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
 455                             unsigned long *gp_seq);
 456 void rcutorture_record_progress(unsigned long vernum);
 457 void do_trace_rcu_torture_read(const char *rcutorturename,
 458                                struct rcu_head *rhp,
 459                                unsigned long secs,
 460                                unsigned long c_old,
 461                                unsigned long c);
 462 #else
 463 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
 464                                           int *flags, unsigned long *gp_seq)
 465 {
 466         *flags = 0;
 467         *gp_seq = 0;
 468 }
 469 static inline void rcutorture_record_progress(unsigned long vernum) { }
 470 #ifdef CONFIG_RCU_TRACE
 471 void do_trace_rcu_torture_read(const char *rcutorturename,
 472                                struct rcu_head *rhp,
 473                                unsigned long secs,
 474                                unsigned long c_old,
 475                                unsigned long c);
 476 #else
 477 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
 478         do { } while (0)
 479 #endif
 480 #endif
 481
 482 #ifdef CONFIG_TINY_SRCU
 483
 484 static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
 485                                            struct srcu_struct *sp, int *flags,
 486                                            unsigned long *gp_seq)
 487 {
 488         if (test_type != SRCU_FLAVOR)
 489                 return;
 490         *flags = 0;
 491         *gp_seq = sp->srcu_idx;
 492 }
 493
 494 #elif defined(CONFIG_TREE_SRCU)
 495
 496 void srcutorture_get_gp_data(enum rcutorture_type test_type,
 497                              struct srcu_struct *sp, int *flags,
 498                              unsigned long *gp_seq);
 499
 500 #endif
 501
 502 #ifdef CONFIG_TINY_RCU
 503 static inline unsigned long rcu_get_gp_seq(void) { return 0; }
 504 static inline unsigned long rcu_exp_batches_completed(void) { return 0; }
 505 static inline unsigned long
 506 srcu_batches_completed(struct srcu_struct *sp) { return 0; }
 507 static inline void rcu_force_quiescent_state(void) { }
 508 static inline void show_rcu_gp_kthreads(void) { }
 509 static inline int rcu_get_gp_kthreads_prio(void) { return 0; }
 510 static inline void rcu_fwd_progress_check(unsigned long j) { }
 511 #else /* #ifdef CONFIG_TINY_RCU */
 512 unsigned long rcu_get_gp_seq(void);
 513 unsigned long rcu_exp_batches_completed(void);
 514 unsigned long srcu_batches_completed(struct srcu_struct *sp);
 515 void show_rcu_gp_kthreads(void);
 516 int rcu_get_gp_kthreads_prio(void);
 517 void rcu_fwd_progress_check(unsigned long j);
 518 void rcu_force_quiescent_state(void);
 519 extern struct workqueue_struct *rcu_gp_wq;
 520 extern struct workqueue_struct *rcu_par_gp_wq;
 521 #endif /* #else #ifdef CONFIG_TINY_RCU */
 522
 523 #ifdef CONFIG_RCU_NOCB_CPU
 524 bool rcu_is_nocb_cpu(int cpu);
 525 void rcu_bind_current_to_nocb(void);
 526 #else
 527 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
 528 static inline void rcu_bind_current_to_nocb(void) { }
 529 #endif
 530
 531 #endif /* __LINUX_RCU_H */