1 /* SPDX-License-Identifier: GPL-2.0+ */
3 * Read-Copy Update definitions shared among RCU implementations.
5 * Copyright IBM Corporation, 2011
13 #include <trace/events/rcu.h>
14 #ifdef CONFIG_RCU_TRACE
15 #define RCU_TRACE(stmt) stmt
16 #else /* #ifdef CONFIG_RCU_TRACE */
17 #define RCU_TRACE(stmt)
18 #endif /* #else #ifdef CONFIG_RCU_TRACE */
20 /* Offset to allow distinguishing irq vs. task-based idle entry/exit. */
21 #define DYNTICK_IRQ_NONIDLE ((LONG_MAX / 2) + 1)
25 * Grace-period counter management.
28 #define RCU_SEQ_CTR_SHIFT 2
29 #define RCU_SEQ_STATE_MASK ((1 << RCU_SEQ_CTR_SHIFT) - 1)
32 * Return the counter portion of a sequence number previously returned
33 * by rcu_seq_snap() or rcu_seq_current().
35 static inline unsigned long rcu_seq_ctr(unsigned long s)
37 return s >> RCU_SEQ_CTR_SHIFT;
41 * Return the state portion of a sequence number previously returned
42 * by rcu_seq_snap() or rcu_seq_current().
44 static inline int rcu_seq_state(unsigned long s)
46 return s & RCU_SEQ_STATE_MASK;
50 * Set the state portion of the pointed-to sequence number.
51 * The caller is responsible for preventing conflicting updates.
53 static inline void rcu_seq_set_state(unsigned long *sp, int newstate)
55 WARN_ON_ONCE(newstate & ~RCU_SEQ_STATE_MASK);
56 WRITE_ONCE(*sp, (*sp & ~RCU_SEQ_STATE_MASK) + newstate);
59 /* Adjust sequence number for start of update-side operation. */
60 static inline void rcu_seq_start(unsigned long *sp)
62 WRITE_ONCE(*sp, *sp + 1);
63 smp_mb(); /* Ensure update-side operation after counter increment. */
64 WARN_ON_ONCE(rcu_seq_state(*sp) != 1);
67 /* Compute the end-of-grace-period value for the specified sequence number. */
68 static inline unsigned long rcu_seq_endval(unsigned long *sp)
70 return (*sp | RCU_SEQ_STATE_MASK) + 1;
73 /* Adjust sequence number for end of update-side operation. */
74 static inline void rcu_seq_end(unsigned long *sp)
76 smp_mb(); /* Ensure update-side operation before counter increment. */
77 WARN_ON_ONCE(!rcu_seq_state(*sp));
78 WRITE_ONCE(*sp, rcu_seq_endval(sp));
82 * rcu_seq_snap - Take a snapshot of the update side's sequence number.
84 * This function returns the earliest value of the grace-period sequence number
85 * that will indicate that a full grace period has elapsed since the current
86 * time. Once the grace-period sequence number has reached this value, it will
87 * be safe to invoke all callbacks that have been registered prior to the
88 * current time. This value is the current grace-period number plus two to the
89 * power of the number of low-order bits reserved for state, then rounded up to
90 * the next value in which the state bits are all zero.
92 static inline unsigned long rcu_seq_snap(unsigned long *sp)
96 s = (READ_ONCE(*sp) + 2 * RCU_SEQ_STATE_MASK + 1) & ~RCU_SEQ_STATE_MASK;
97 smp_mb(); /* Above access must not bleed into critical section. */
101 /* Return the current value the update side's sequence number, no ordering. */
102 static inline unsigned long rcu_seq_current(unsigned long *sp)
104 return READ_ONCE(*sp);
108 * Given a snapshot from rcu_seq_snap(), determine whether or not the
109 * corresponding update-side operation has started.
111 static inline bool rcu_seq_started(unsigned long *sp, unsigned long s)
113 return ULONG_CMP_LT((s - 1) & ~RCU_SEQ_STATE_MASK, READ_ONCE(*sp));
117 * Given a snapshot from rcu_seq_snap(), determine whether or not a
118 * full update-side operation has occurred.
120 static inline bool rcu_seq_done(unsigned long *sp, unsigned long s)
122 return ULONG_CMP_GE(READ_ONCE(*sp), s);
126 * Has a grace period completed since the time the old gp_seq was collected?
128 static inline bool rcu_seq_completed_gp(unsigned long old, unsigned long new)
130 return ULONG_CMP_LT(old, new & ~RCU_SEQ_STATE_MASK);
134 * Has a grace period started since the time the old gp_seq was collected?
136 static inline bool rcu_seq_new_gp(unsigned long old, unsigned long new)
138 return ULONG_CMP_LT((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK,
143 * Roughly how many full grace periods have elapsed between the collection
144 * of the two specified grace periods?
146 static inline unsigned long rcu_seq_diff(unsigned long new, unsigned long old)
148 unsigned long rnd_diff;
153 * Compute the number of grace periods (still shifted up), plus
154 * one if either of new and old is not an exact grace period.
156 rnd_diff = (new & ~RCU_SEQ_STATE_MASK) -
157 ((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK) +
158 ((new & RCU_SEQ_STATE_MASK) || (old & RCU_SEQ_STATE_MASK));
159 if (ULONG_CMP_GE(RCU_SEQ_STATE_MASK, rnd_diff))
160 return 1; /* Definitely no grace period has elapsed. */
161 return ((rnd_diff - RCU_SEQ_STATE_MASK - 1) >> RCU_SEQ_CTR_SHIFT) + 2;
165 * debug_rcu_head_queue()/debug_rcu_head_unqueue() are used internally
166 * by call_rcu() and rcu callback execution, and are therefore not part
167 * of the RCU API. These are in rcupdate.h because they are used by all
168 * RCU implementations.
171 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
172 # define STATE_RCU_HEAD_READY 0
173 # define STATE_RCU_HEAD_QUEUED 1
175 extern struct debug_obj_descr rcuhead_debug_descr;
177 static inline int debug_rcu_head_queue(struct rcu_head *head)
181 r1 = debug_object_activate(head, &rcuhead_debug_descr);
182 debug_object_active_state(head, &rcuhead_debug_descr,
183 STATE_RCU_HEAD_READY,
184 STATE_RCU_HEAD_QUEUED);
188 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
190 debug_object_active_state(head, &rcuhead_debug_descr,
191 STATE_RCU_HEAD_QUEUED,
192 STATE_RCU_HEAD_READY);
193 debug_object_deactivate(head, &rcuhead_debug_descr);
195 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
196 static inline int debug_rcu_head_queue(struct rcu_head *head)
201 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
204 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
206 void kfree(const void *);
209 * Reclaim the specified callback, either by invoking it (non-lazy case)
210 * or freeing it directly (lazy case). Return true if lazy, false otherwise.
212 static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head)
215 unsigned long offset = (unsigned long)head->func;
217 rcu_lock_acquire(&rcu_callback_map);
218 if (__is_kfree_rcu_offset(offset)) {
219 RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset);)
220 kfree((void *)head - offset);
221 rcu_lock_release(&rcu_callback_map);
224 RCU_TRACE(trace_rcu_invoke_callback(rn, head);)
226 WRITE_ONCE(head->func, (rcu_callback_t)0L);
228 rcu_lock_release(&rcu_callback_map);
233 #ifdef CONFIG_RCU_STALL_COMMON
235 extern int rcu_cpu_stall_suppress;
236 int rcu_jiffies_till_stall_check(void);
238 #define rcu_ftrace_dump_stall_suppress() \
240 if (!rcu_cpu_stall_suppress) \
241 rcu_cpu_stall_suppress = 3; \
244 #define rcu_ftrace_dump_stall_unsuppress() \
246 if (rcu_cpu_stall_suppress == 3) \
247 rcu_cpu_stall_suppress = 0; \
250 #else /* #endif #ifdef CONFIG_RCU_STALL_COMMON */
251 #define rcu_ftrace_dump_stall_suppress()
252 #define rcu_ftrace_dump_stall_unsuppress()
253 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
256 * Strings used in tracepoints need to be exported via the
257 * tracing system such that tools like perf and trace-cmd can
258 * translate the string address pointers to actual text.
260 #define TPS(x) tracepoint_string(x)
263 * Dump the ftrace buffer, but only one time per callsite per boot.
265 #define rcu_ftrace_dump(oops_dump_mode) \
267 static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
269 if (!atomic_read(&___rfd_beenhere) && \
270 !atomic_xchg(&___rfd_beenhere, 1)) { \
272 rcu_ftrace_dump_stall_suppress(); \
273 ftrace_dump(oops_dump_mode); \
274 rcu_ftrace_dump_stall_unsuppress(); \
278 void rcu_early_boot_tests(void);
279 void rcu_test_sync_prims(void);
282 * This function really isn't for public consumption, but RCU is special in
283 * that context switches can allow the state machine to make progress.
285 extern void resched_cpu(int cpu);
287 #if defined(SRCU) || !defined(TINY_RCU)
289 #include <linux/rcu_node_tree.h>
291 extern int rcu_num_lvls;
292 extern int num_rcu_lvl[];
293 extern int rcu_num_nodes;
294 static bool rcu_fanout_exact;
295 static int rcu_fanout_leaf;
298 * Compute the per-level fanout, either using the exact fanout specified
299 * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
301 static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
305 if (rcu_fanout_exact) {
306 levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
307 for (i = rcu_num_lvls - 2; i >= 0; i--)
308 levelspread[i] = RCU_FANOUT;
314 for (i = rcu_num_lvls - 1; i >= 0; i--) {
316 levelspread[i] = (cprv + ccur - 1) / ccur;
322 /* Returns a pointer to the first leaf rcu_node structure. */
323 #define rcu_first_leaf_node() (rcu_state.level[rcu_num_lvls - 1])
325 /* Is this rcu_node a leaf? */
326 #define rcu_is_leaf_node(rnp) ((rnp)->level == rcu_num_lvls - 1)
328 /* Is this rcu_node the last leaf? */
329 #define rcu_is_last_leaf_node(rnp) ((rnp) == &rcu_state.node[rcu_num_nodes - 1])
332 * Do a full breadth-first scan of the {s,}rcu_node structures for the
333 * specified state structure (for SRCU) or the only rcu_state structure
336 #define srcu_for_each_node_breadth_first(sp, rnp) \
337 for ((rnp) = &(sp)->node[0]; \
338 (rnp) < &(sp)->node[rcu_num_nodes]; (rnp)++)
339 #define rcu_for_each_node_breadth_first(rnp) \
340 srcu_for_each_node_breadth_first(&rcu_state, rnp)
343 * Scan the leaves of the rcu_node hierarchy for the rcu_state structure.
344 * Note that if there is a singleton rcu_node tree with but one rcu_node
345 * structure, this loop -will- visit the rcu_node structure. It is still
346 * a leaf node, even if it is also the root node.
348 #define rcu_for_each_leaf_node(rnp) \
349 for ((rnp) = rcu_first_leaf_node(); \
350 (rnp) < &rcu_state.node[rcu_num_nodes]; (rnp)++)
353 * Iterate over all possible CPUs in a leaf RCU node.
355 #define for_each_leaf_node_possible_cpu(rnp, cpu) \
356 for ((cpu) = cpumask_next((rnp)->grplo - 1, cpu_possible_mask); \
357 (cpu) <= rnp->grphi; \
358 (cpu) = cpumask_next((cpu), cpu_possible_mask))
361 * Iterate over all CPUs in a leaf RCU node's specified mask.
363 #define rcu_find_next_bit(rnp, cpu, mask) \
364 ((rnp)->grplo + find_next_bit(&(mask), BITS_PER_LONG, (cpu)))
365 #define for_each_leaf_node_cpu_mask(rnp, cpu, mask) \
366 for ((cpu) = rcu_find_next_bit((rnp), 0, (mask)); \
367 (cpu) <= rnp->grphi; \
368 (cpu) = rcu_find_next_bit((rnp), (cpu) + 1 - (rnp->grplo), (mask)))
371 * Wrappers for the rcu_node::lock acquire and release.
373 * Because the rcu_nodes form a tree, the tree traversal locking will observe
374 * different lock values, this in turn means that an UNLOCK of one level
375 * followed by a LOCK of another level does not imply a full memory barrier;
376 * and most importantly transitivity is lost.
378 * In order to restore full ordering between tree levels, augment the regular
379 * lock acquire functions with smp_mb__after_unlock_lock().
381 * As ->lock of struct rcu_node is a __private field, therefore one should use
382 * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
384 #define raw_spin_lock_rcu_node(p) \
386 raw_spin_lock(&ACCESS_PRIVATE(p, lock)); \
387 smp_mb__after_unlock_lock(); \
390 #define raw_spin_unlock_rcu_node(p) raw_spin_unlock(&ACCESS_PRIVATE(p, lock))
392 #define raw_spin_lock_irq_rcu_node(p) \
394 raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock)); \
395 smp_mb__after_unlock_lock(); \
398 #define raw_spin_unlock_irq_rcu_node(p) \
399 raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock))
401 #define raw_spin_lock_irqsave_rcu_node(p, flags) \
403 raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags); \
404 smp_mb__after_unlock_lock(); \
407 #define raw_spin_unlock_irqrestore_rcu_node(p, flags) \
408 raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags)
410 #define raw_spin_trylock_rcu_node(p) \
412 bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock)); \
415 smp_mb__after_unlock_lock(); \
419 #define raw_lockdep_assert_held_rcu_node(p) \
420 lockdep_assert_held(&ACCESS_PRIVATE(p, lock))
422 #endif /* #if defined(SRCU) || !defined(TINY_RCU) */
425 void srcu_init(void);
426 #else /* #ifdef CONFIG_SRCU */
427 static inline void srcu_init(void) { }
428 #endif /* #else #ifdef CONFIG_SRCU */
430 #ifdef CONFIG_TINY_RCU
431 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
432 static inline bool rcu_gp_is_normal(void) { return true; }
433 static inline bool rcu_gp_is_expedited(void) { return false; }
434 static inline void rcu_expedite_gp(void) { }
435 static inline void rcu_unexpedite_gp(void) { }
436 static inline void rcu_request_urgent_qs_task(struct task_struct *t) { }
437 #else /* #ifdef CONFIG_TINY_RCU */
438 bool rcu_gp_is_normal(void); /* Internal RCU use. */
439 bool rcu_gp_is_expedited(void); /* Internal RCU use. */
440 void rcu_expedite_gp(void);
441 void rcu_unexpedite_gp(void);
442 void rcupdate_announce_bootup_oddness(void);
443 void rcu_request_urgent_qs_task(struct task_struct *t);
444 #endif /* #else #ifdef CONFIG_TINY_RCU */
446 #define RCU_SCHEDULER_INACTIVE 0
447 #define RCU_SCHEDULER_INIT 1
448 #define RCU_SCHEDULER_RUNNING 2
450 enum rcutorture_type {
457 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
458 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
459 unsigned long *gp_seq);
460 void rcutorture_record_progress(unsigned long vernum);
461 void do_trace_rcu_torture_read(const char *rcutorturename,
462 struct rcu_head *rhp,
467 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
468 int *flags, unsigned long *gp_seq)
473 static inline void rcutorture_record_progress(unsigned long vernum) { }
474 #ifdef CONFIG_RCU_TRACE
475 void do_trace_rcu_torture_read(const char *rcutorturename,
476 struct rcu_head *rhp,
481 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
486 #ifdef CONFIG_TINY_SRCU
488 static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
489 struct srcu_struct *sp, int *flags,
490 unsigned long *gp_seq)
492 if (test_type != SRCU_FLAVOR)
495 *gp_seq = sp->srcu_idx;
498 #elif defined(CONFIG_TREE_SRCU)
500 void srcutorture_get_gp_data(enum rcutorture_type test_type,
501 struct srcu_struct *sp, int *flags,
502 unsigned long *gp_seq);
506 #ifdef CONFIG_TINY_RCU
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 unsigned long rcu_get_gp_seq(void);
517 unsigned long rcu_exp_batches_completed(void);
518 unsigned long srcu_batches_completed(struct srcu_struct *sp);
519 void show_rcu_gp_kthreads(void);
520 int rcu_get_gp_kthreads_prio(void);
521 void rcu_fwd_progress_check(unsigned long j);
522 void rcu_force_quiescent_state(void);
523 extern struct workqueue_struct *rcu_gp_wq;
524 extern struct workqueue_struct *rcu_par_gp_wq;
525 #endif /* #else #ifdef CONFIG_TINY_RCU */
527 #ifdef CONFIG_RCU_NOCB_CPU
528 bool rcu_is_nocb_cpu(int cpu);
529 void rcu_bind_current_to_nocb(void);
531 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
532 static inline void rcu_bind_current_to_nocb(void) { }
535 #endif /* __LINUX_RCU_H */