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 extern int rcu_cpu_stall_suppress_at_boot;
 202
 203 static inline bool rcu_stall_is_suppressed_at_boot(void)
 204 {
 205         return rcu_cpu_stall_suppress_at_boot && !rcu_inkernel_boot_has_ended();
 206 }
 207
 208 #ifdef CONFIG_RCU_STALL_COMMON
 209
 210 extern int rcu_cpu_stall_ftrace_dump;
 211 extern int rcu_cpu_stall_suppress;
 212 extern int rcu_cpu_stall_timeout;
 213 int rcu_jiffies_till_stall_check(void);
 214
 215 static inline bool rcu_stall_is_suppressed(void)
 216 {
 217         return rcu_stall_is_suppressed_at_boot() || rcu_cpu_stall_suppress;
 218 }
 219
 220 #define rcu_ftrace_dump_stall_suppress() \
 221 do { \
 222         if (!rcu_cpu_stall_suppress) \
 223                 rcu_cpu_stall_suppress = 3; \
 224 } while (0)
 225
 226 #define rcu_ftrace_dump_stall_unsuppress() \
 227 do { \
 228         if (rcu_cpu_stall_suppress == 3) \
 229                 rcu_cpu_stall_suppress = 0; \
 230 } while (0)
 231
 232 #else /* #endif #ifdef CONFIG_RCU_STALL_COMMON */
 233
 234 static inline bool rcu_stall_is_suppressed(void)
 235 {
 236         return rcu_stall_is_suppressed_at_boot();
 237 }
 238 #define rcu_ftrace_dump_stall_suppress()
 239 #define rcu_ftrace_dump_stall_unsuppress()
 240 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
 241
 242 /*
 243  * Strings used in tracepoints need to be exported via the
 244  * tracing system such that tools like perf and trace-cmd can
 245  * translate the string address pointers to actual text.
 246  */
 247 #define TPS(x)  tracepoint_string(x)
 248
 249 /*
 250  * Dump the ftrace buffer, but only one time per callsite per boot.
 251  */
 252 #define rcu_ftrace_dump(oops_dump_mode) \
 253 do { \
 254         static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
 255         \
 256         if (!atomic_read(&___rfd_beenhere) && \
 257             !atomic_xchg(&___rfd_beenhere, 1)) { \
 258                 tracing_off(); \
 259                 rcu_ftrace_dump_stall_suppress(); \
 260                 ftrace_dump(oops_dump_mode); \
 261                 rcu_ftrace_dump_stall_unsuppress(); \
 262         } \
 263 } while (0)
 264
 265 void rcu_early_boot_tests(void);
 266 void rcu_test_sync_prims(void);
 267
 268 /*
 269  * This function really isn't for public consumption, but RCU is special in
 270  * that context switches can allow the state machine to make progress.
 271  */
 272 extern void resched_cpu(int cpu);
 273
 274 #if defined(CONFIG_SRCU) || !defined(CONFIG_TINY_RCU)
 275
 276 #include <linux/rcu_node_tree.h>
 277
 278 extern int rcu_num_lvls;
 279 extern int num_rcu_lvl[];
 280 extern int rcu_num_nodes;
 281 static bool rcu_fanout_exact;
 282 static int rcu_fanout_leaf;
 283
 284 /*
 285  * Compute the per-level fanout, either using the exact fanout specified
 286  * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
 287  */
 288 static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
 289 {
 290         int i;
 291
 292         for (i = 0; i < RCU_NUM_LVLS; i++)
 293                 levelspread[i] = INT_MIN;
 294         if (rcu_fanout_exact) {
 295                 levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
 296                 for (i = rcu_num_lvls - 2; i >= 0; i--)
 297                         levelspread[i] = RCU_FANOUT;
 298         } else {
 299                 int ccur;
 300                 int cprv;
 301
 302                 cprv = nr_cpu_ids;
 303                 for (i = rcu_num_lvls - 1; i >= 0; i--) {
 304                         ccur = levelcnt[i];
 305                         levelspread[i] = (cprv + ccur - 1) / ccur;
 306                         cprv = ccur;
 307                 }
 308         }
 309 }
 310
 311 /* Returns a pointer to the first leaf rcu_node structure. */
 312 #define rcu_first_leaf_node() (rcu_state.level[rcu_num_lvls - 1])
 313
 314 /* Is this rcu_node a leaf? */
 315 #define rcu_is_leaf_node(rnp) ((rnp)->level == rcu_num_lvls - 1)
 316
 317 /* Is this rcu_node the last leaf? */
 318 #define rcu_is_last_leaf_node(rnp) ((rnp) == &rcu_state.node[rcu_num_nodes - 1])
 319
 320 /*
 321  * Do a full breadth-first scan of the {s,}rcu_node structures for the
 322  * specified state structure (for SRCU) or the only rcu_state structure
 323  * (for RCU).
 324  */
 325 #define srcu_for_each_node_breadth_first(sp, rnp) \
 326         for ((rnp) = &(sp)->node[0]; \
 327              (rnp) < &(sp)->node[rcu_num_nodes]; (rnp)++)
 328 #define rcu_for_each_node_breadth_first(rnp) \
 329         srcu_for_each_node_breadth_first(&rcu_state, rnp)
 330
 331 /*
 332  * Scan the leaves of the rcu_node hierarchy for the rcu_state structure.
 333  * Note that if there is a singleton rcu_node tree with but one rcu_node
 334  * structure, this loop -will- visit the rcu_node structure.  It is still
 335  * a leaf node, even if it is also the root node.
 336  */
 337 #define rcu_for_each_leaf_node(rnp) \
 338         for ((rnp) = rcu_first_leaf_node(); \
 339              (rnp) < &rcu_state.node[rcu_num_nodes]; (rnp)++)
 340
 341 /*
 342  * Iterate over all possible CPUs in a leaf RCU node.
 343  */
 344 #define for_each_leaf_node_possible_cpu(rnp, cpu) \
 345         for (WARN_ON_ONCE(!rcu_is_leaf_node(rnp)), \
 346              (cpu) = cpumask_next((rnp)->grplo - 1, cpu_possible_mask); \
 347              (cpu) <= rnp->grphi; \
 348              (cpu) = cpumask_next((cpu), cpu_possible_mask))
 349
 350 /*
 351  * Iterate over all CPUs in a leaf RCU node's specified mask.
 352  */
 353 #define rcu_find_next_bit(rnp, cpu, mask) \
 354         ((rnp)->grplo + find_next_bit(&(mask), BITS_PER_LONG, (cpu)))
 355 #define for_each_leaf_node_cpu_mask(rnp, cpu, mask) \
 356         for (WARN_ON_ONCE(!rcu_is_leaf_node(rnp)), \
 357              (cpu) = rcu_find_next_bit((rnp), 0, (mask)); \
 358              (cpu) <= rnp->grphi; \
 359              (cpu) = rcu_find_next_bit((rnp), (cpu) + 1 - (rnp->grplo), (mask)))
 360
 361 /*
 362  * Wrappers for the rcu_node::lock acquire and release.
 363  *
 364  * Because the rcu_nodes form a tree, the tree traversal locking will observe
 365  * different lock values, this in turn means that an UNLOCK of one level
 366  * followed by a LOCK of another level does not imply a full memory barrier;
 367  * and most importantly transitivity is lost.
 368  *
 369  * In order to restore full ordering between tree levels, augment the regular
 370  * lock acquire functions with smp_mb__after_unlock_lock().
 371  *
 372  * As ->lock of struct rcu_node is a __private field, therefore one should use
 373  * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
 374  */
 375 #define raw_spin_lock_rcu_node(p)                                       \
 376 do {                                                                    \
 377         raw_spin_lock(&ACCESS_PRIVATE(p, lock));                        \
 378         smp_mb__after_unlock_lock();                                    \
 379 } while (0)
 380
 381 #define raw_spin_unlock_rcu_node(p) raw_spin_unlock(&ACCESS_PRIVATE(p, lock))
 382
 383 #define raw_spin_lock_irq_rcu_node(p)                                   \
 384 do {                                                                    \
 385         raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock));                    \
 386         smp_mb__after_unlock_lock();                                    \
 387 } while (0)
 388
 389 #define raw_spin_unlock_irq_rcu_node(p)                                 \
 390         raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock))
 391
 392 #define raw_spin_lock_irqsave_rcu_node(p, flags)                        \
 393 do {                                                                    \
 394         raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags); \
 395         smp_mb__after_unlock_lock();                                    \
 396 } while (0)
 397
 398 #define raw_spin_unlock_irqrestore_rcu_node(p, flags)                   \
 399         raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags)
 400
 401 #define raw_spin_trylock_rcu_node(p)                                    \
 402 ({                                                                      \
 403         bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock));    \
 404                                                                         \
 405         if (___locked)                                                  \
 406                 smp_mb__after_unlock_lock();                            \
 407         ___locked;                                                      \
 408 })
 409
 410 #define raw_lockdep_assert_held_rcu_node(p)                             \
 411         lockdep_assert_held(&ACCESS_PRIVATE(p, lock))
 412
 413 #endif /* #if defined(CONFIG_SRCU) || !defined(CONFIG_TINY_RCU) */
 414
 415 #ifdef CONFIG_SRCU
 416 void srcu_init(void);
 417 #else /* #ifdef CONFIG_SRCU */
 418 static inline void srcu_init(void) { }
 419 #endif /* #else #ifdef CONFIG_SRCU */
 420
 421 #ifdef CONFIG_TINY_RCU
 422 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
 423 static inline bool rcu_gp_is_normal(void) { return true; }
 424 static inline bool rcu_gp_is_expedited(void) { return false; }
 425 static inline void rcu_expedite_gp(void) { }
 426 static inline void rcu_unexpedite_gp(void) { }
 427 static inline void rcu_request_urgent_qs_task(struct task_struct *t) { }
 428 #else /* #ifdef CONFIG_TINY_RCU */
 429 bool rcu_gp_is_normal(void);     /* Internal RCU use. */
 430 bool rcu_gp_is_expedited(void);  /* Internal RCU use. */
 431 void rcu_expedite_gp(void);
 432 void rcu_unexpedite_gp(void);
 433 void rcupdate_announce_bootup_oddness(void);
 434 void show_rcu_tasks_gp_kthreads(void);
 435 void rcu_request_urgent_qs_task(struct task_struct *t);
 436 #endif /* #else #ifdef CONFIG_TINY_RCU */
 437
 438 #define RCU_SCHEDULER_INACTIVE  0
 439 #define RCU_SCHEDULER_INIT      1
 440 #define RCU_SCHEDULER_RUNNING   2
 441
 442 enum rcutorture_type {
 443         RCU_FLAVOR,
 444         RCU_TASKS_FLAVOR,
 445         RCU_TASKS_RUDE_FLAVOR,
 446         RCU_TASKS_TRACING_FLAVOR,
 447         RCU_TRIVIAL_FLAVOR,
 448         SRCU_FLAVOR,
 449         INVALID_RCU_FLAVOR
 450 };
 451
 452 #if defined(CONFIG_TREE_RCU)
 453 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
 454                             unsigned long *gp_seq);
 455 void do_trace_rcu_torture_read(const char *rcutorturename,
 456                                struct rcu_head *rhp,
 457                                unsigned long secs,
 458                                unsigned long c_old,
 459                                unsigned long c);
 460 void rcu_gp_set_torture_wait(int duration);
 461 #else
 462 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
 463                                           int *flags, unsigned long *gp_seq)
 464 {
 465         *flags = 0;
 466         *gp_seq = 0;
 467 }
 468 #ifdef CONFIG_RCU_TRACE
 469 void do_trace_rcu_torture_read(const char *rcutorturename,
 470                                struct rcu_head *rhp,
 471                                unsigned long secs,
 472                                unsigned long c_old,
 473                                unsigned long c);
 474 #else
 475 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
 476         do { } while (0)
 477 #endif
 478 static inline void rcu_gp_set_torture_wait(int duration) { }
 479 #endif
 480
 481 #if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST)
 482 long rcutorture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask);
 483 #endif
 484
 485 #ifdef CONFIG_TINY_SRCU
 486
 487 static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
 488                                            struct srcu_struct *sp, int *flags,
 489                                            unsigned long *gp_seq)
 490 {
 491         if (test_type != SRCU_FLAVOR)
 492                 return;
 493         *flags = 0;
 494         *gp_seq = sp->srcu_idx;
 495 }
 496
 497 #elif defined(CONFIG_TREE_SRCU)
 498
 499 void srcutorture_get_gp_data(enum rcutorture_type test_type,
 500                              struct srcu_struct *sp, int *flags,
 501                              unsigned long *gp_seq);
 502
 503 #endif
 504
 505 #ifdef CONFIG_TINY_RCU
 506 static inline bool rcu_dynticks_zero_in_eqs(int cpu, int *vp) { return false; }
 507 static inline unsigned long rcu_get_gp_seq(void) { return 0; }
 508 static inline unsigned long rcu_exp_batches_completed(void) { return 0; }
 509 static inline unsigned long
 510 srcu_batches_completed(struct srcu_struct *sp) { return 0; }
 511 static inline void rcu_force_quiescent_state(void) { }
 512 static inline void show_rcu_gp_kthreads(void) { }
 513 static inline int rcu_get_gp_kthreads_prio(void) { return 0; }
 514 static inline void rcu_fwd_progress_check(unsigned long j) { }
 515 #else /* #ifdef CONFIG_TINY_RCU */
 516 bool rcu_dynticks_zero_in_eqs(int cpu, int *vp);
 517 unsigned long rcu_get_gp_seq(void);
 518 unsigned long rcu_exp_batches_completed(void);
 519 unsigned long srcu_batches_completed(struct srcu_struct *sp);
 520 void show_rcu_gp_kthreads(void);
 521 int rcu_get_gp_kthreads_prio(void);
 522 void rcu_fwd_progress_check(unsigned long j);
 523 void rcu_force_quiescent_state(void);
 524 extern struct workqueue_struct *rcu_gp_wq;
 525 extern struct workqueue_struct *rcu_par_gp_wq;
 526 #endif /* #else #ifdef CONFIG_TINY_RCU */
 527
 528 #ifdef CONFIG_RCU_NOCB_CPU
 529 bool rcu_is_nocb_cpu(int cpu);
 530 void rcu_bind_current_to_nocb(void);
 531 #else
 532 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
 533 static inline void rcu_bind_current_to_nocb(void) { }
 534 #endif
 535
 536 #endif /* __LINUX_RCU_H */