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>
16 * Grace-period counter management.
19 #define RCU_SEQ_CTR_SHIFT 2
20 #define RCU_SEQ_STATE_MASK ((1 << RCU_SEQ_CTR_SHIFT) - 1)
22 /* Low-order bit definition for polled grace-period APIs. */
23 #define RCU_GET_STATE_COMPLETED 0x1
25 extern int sysctl_sched_rt_runtime;
28 * Return the counter portion of a sequence number previously returned
29 * by rcu_seq_snap() or rcu_seq_current().
31 static inline unsigned long rcu_seq_ctr(unsigned long s)
33 return s >> RCU_SEQ_CTR_SHIFT;
37 * Return the state portion of a sequence number previously returned
38 * by rcu_seq_snap() or rcu_seq_current().
40 static inline int rcu_seq_state(unsigned long s)
42 return s & RCU_SEQ_STATE_MASK;
46 * Set the state portion of the pointed-to sequence number.
47 * The caller is responsible for preventing conflicting updates.
49 static inline void rcu_seq_set_state(unsigned long *sp, int newstate)
51 WARN_ON_ONCE(newstate & ~RCU_SEQ_STATE_MASK);
52 WRITE_ONCE(*sp, (*sp & ~RCU_SEQ_STATE_MASK) + newstate);
55 /* Adjust sequence number for start of update-side operation. */
56 static inline void rcu_seq_start(unsigned long *sp)
58 WRITE_ONCE(*sp, *sp + 1);
59 smp_mb(); /* Ensure update-side operation after counter increment. */
60 WARN_ON_ONCE(rcu_seq_state(*sp) != 1);
63 /* Compute the end-of-grace-period value for the specified sequence number. */
64 static inline unsigned long rcu_seq_endval(unsigned long *sp)
66 return (*sp | RCU_SEQ_STATE_MASK) + 1;
69 /* Adjust sequence number for end of update-side operation. */
70 static inline void rcu_seq_end(unsigned long *sp)
72 smp_mb(); /* Ensure update-side operation before counter increment. */
73 WARN_ON_ONCE(!rcu_seq_state(*sp));
74 WRITE_ONCE(*sp, rcu_seq_endval(sp));
78 * rcu_seq_snap - Take a snapshot of the update side's sequence number.
80 * This function returns the earliest value of the grace-period sequence number
81 * that will indicate that a full grace period has elapsed since the current
82 * time. Once the grace-period sequence number has reached this value, it will
83 * be safe to invoke all callbacks that have been registered prior to the
84 * current time. This value is the current grace-period number plus two to the
85 * power of the number of low-order bits reserved for state, then rounded up to
86 * the next value in which the state bits are all zero.
88 static inline unsigned long rcu_seq_snap(unsigned long *sp)
92 s = (READ_ONCE(*sp) + 2 * RCU_SEQ_STATE_MASK + 1) & ~RCU_SEQ_STATE_MASK;
93 smp_mb(); /* Above access must not bleed into critical section. */
97 /* Return the current value the update side's sequence number, no ordering. */
98 static inline unsigned long rcu_seq_current(unsigned long *sp)
100 return READ_ONCE(*sp);
104 * Given a snapshot from rcu_seq_snap(), determine whether or not the
105 * corresponding update-side operation has started.
107 static inline bool rcu_seq_started(unsigned long *sp, unsigned long s)
109 return ULONG_CMP_LT((s - 1) & ~RCU_SEQ_STATE_MASK, READ_ONCE(*sp));
113 * Given a snapshot from rcu_seq_snap(), determine whether or not a
114 * full update-side operation has occurred.
116 static inline bool rcu_seq_done(unsigned long *sp, unsigned long s)
118 return ULONG_CMP_GE(READ_ONCE(*sp), s);
122 * Given a snapshot from rcu_seq_snap(), determine whether or not a
123 * full update-side operation has occurred, but do not allow the
124 * (ULONG_MAX / 2) safety-factor/guard-band.
126 static inline bool rcu_seq_done_exact(unsigned long *sp, unsigned long s)
128 unsigned long cur_s = READ_ONCE(*sp);
130 return ULONG_CMP_GE(cur_s, s) || ULONG_CMP_LT(cur_s, s - (2 * RCU_SEQ_STATE_MASK + 1));
134 * Has a grace period completed since the time the old gp_seq was collected?
136 static inline bool rcu_seq_completed_gp(unsigned long old, unsigned long new)
138 return ULONG_CMP_LT(old, new & ~RCU_SEQ_STATE_MASK);
142 * Has a grace period started since the time the old gp_seq was collected?
144 static inline bool rcu_seq_new_gp(unsigned long old, unsigned long new)
146 return ULONG_CMP_LT((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK,
151 * Roughly how many full grace periods have elapsed between the collection
152 * of the two specified grace periods?
154 static inline unsigned long rcu_seq_diff(unsigned long new, unsigned long old)
156 unsigned long rnd_diff;
161 * Compute the number of grace periods (still shifted up), plus
162 * one if either of new and old is not an exact grace period.
164 rnd_diff = (new & ~RCU_SEQ_STATE_MASK) -
165 ((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK) +
166 ((new & RCU_SEQ_STATE_MASK) || (old & RCU_SEQ_STATE_MASK));
167 if (ULONG_CMP_GE(RCU_SEQ_STATE_MASK, rnd_diff))
168 return 1; /* Definitely no grace period has elapsed. */
169 return ((rnd_diff - RCU_SEQ_STATE_MASK - 1) >> RCU_SEQ_CTR_SHIFT) + 2;
173 * debug_rcu_head_queue()/debug_rcu_head_unqueue() are used internally
174 * by call_rcu() and rcu callback execution, and are therefore not part
175 * of the RCU API. These are in rcupdate.h because they are used by all
176 * RCU implementations.
179 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
180 # define STATE_RCU_HEAD_READY 0
181 # define STATE_RCU_HEAD_QUEUED 1
183 extern const struct debug_obj_descr rcuhead_debug_descr;
185 static inline int debug_rcu_head_queue(struct rcu_head *head)
189 r1 = debug_object_activate(head, &rcuhead_debug_descr);
190 debug_object_active_state(head, &rcuhead_debug_descr,
191 STATE_RCU_HEAD_READY,
192 STATE_RCU_HEAD_QUEUED);
196 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
198 debug_object_active_state(head, &rcuhead_debug_descr,
199 STATE_RCU_HEAD_QUEUED,
200 STATE_RCU_HEAD_READY);
201 debug_object_deactivate(head, &rcuhead_debug_descr);
203 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
204 static inline int debug_rcu_head_queue(struct rcu_head *head)
209 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
212 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
214 extern int rcu_cpu_stall_suppress_at_boot;
216 static inline bool rcu_stall_is_suppressed_at_boot(void)
218 return rcu_cpu_stall_suppress_at_boot && !rcu_inkernel_boot_has_ended();
221 #ifdef CONFIG_RCU_STALL_COMMON
223 extern int rcu_cpu_stall_ftrace_dump;
224 extern int rcu_cpu_stall_suppress;
225 extern int rcu_cpu_stall_timeout;
226 extern int rcu_exp_cpu_stall_timeout;
227 extern int rcu_cpu_stall_cputime;
228 extern bool rcu_exp_stall_task_details __read_mostly;
229 int rcu_jiffies_till_stall_check(void);
230 int rcu_exp_jiffies_till_stall_check(void);
232 static inline bool rcu_stall_is_suppressed(void)
234 return rcu_stall_is_suppressed_at_boot() || rcu_cpu_stall_suppress;
237 #define rcu_ftrace_dump_stall_suppress() \
239 if (!rcu_cpu_stall_suppress) \
240 rcu_cpu_stall_suppress = 3; \
243 #define rcu_ftrace_dump_stall_unsuppress() \
245 if (rcu_cpu_stall_suppress == 3) \
246 rcu_cpu_stall_suppress = 0; \
249 #else /* #endif #ifdef CONFIG_RCU_STALL_COMMON */
251 static inline bool rcu_stall_is_suppressed(void)
253 return rcu_stall_is_suppressed_at_boot();
255 #define rcu_ftrace_dump_stall_suppress()
256 #define rcu_ftrace_dump_stall_unsuppress()
257 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
260 * Strings used in tracepoints need to be exported via the
261 * tracing system such that tools like perf and trace-cmd can
262 * translate the string address pointers to actual text.
264 #define TPS(x) tracepoint_string(x)
267 * Dump the ftrace buffer, but only one time per callsite per boot.
269 #define rcu_ftrace_dump(oops_dump_mode) \
271 static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
273 if (!atomic_read(&___rfd_beenhere) && \
274 !atomic_xchg(&___rfd_beenhere, 1)) { \
276 rcu_ftrace_dump_stall_suppress(); \
277 ftrace_dump(oops_dump_mode); \
278 rcu_ftrace_dump_stall_unsuppress(); \
282 void rcu_early_boot_tests(void);
283 void rcu_test_sync_prims(void);
286 * This function really isn't for public consumption, but RCU is special in
287 * that context switches can allow the state machine to make progress.
289 extern void resched_cpu(int cpu);
291 #if !defined(CONFIG_TINY_RCU)
293 #include <linux/rcu_node_tree.h>
295 extern int rcu_num_lvls;
296 extern int num_rcu_lvl[];
297 extern int rcu_num_nodes;
298 static bool rcu_fanout_exact;
299 static int rcu_fanout_leaf;
302 * Compute the per-level fanout, either using the exact fanout specified
303 * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
305 static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
309 for (i = 0; i < RCU_NUM_LVLS; i++)
310 levelspread[i] = INT_MIN;
311 if (rcu_fanout_exact) {
312 levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
313 for (i = rcu_num_lvls - 2; i >= 0; i--)
314 levelspread[i] = RCU_FANOUT;
320 for (i = rcu_num_lvls - 1; i >= 0; i--) {
322 levelspread[i] = (cprv + ccur - 1) / ccur;
328 extern void rcu_init_geometry(void);
330 /* Returns a pointer to the first leaf rcu_node structure. */
331 #define rcu_first_leaf_node() (rcu_state.level[rcu_num_lvls - 1])
333 /* Is this rcu_node a leaf? */
334 #define rcu_is_leaf_node(rnp) ((rnp)->level == rcu_num_lvls - 1)
336 /* Is this rcu_node the last leaf? */
337 #define rcu_is_last_leaf_node(rnp) ((rnp) == &rcu_state.node[rcu_num_nodes - 1])
340 * Do a full breadth-first scan of the {s,}rcu_node structures for the
341 * specified state structure (for SRCU) or the only rcu_state structure
344 #define srcu_for_each_node_breadth_first(sp, rnp) \
345 for ((rnp) = &(sp)->node[0]; \
346 (rnp) < &(sp)->node[rcu_num_nodes]; (rnp)++)
347 #define rcu_for_each_node_breadth_first(rnp) \
348 srcu_for_each_node_breadth_first(&rcu_state, rnp)
351 * Scan the leaves of the rcu_node hierarchy for the rcu_state structure.
352 * Note that if there is a singleton rcu_node tree with but one rcu_node
353 * structure, this loop -will- visit the rcu_node structure. It is still
354 * a leaf node, even if it is also the root node.
356 #define rcu_for_each_leaf_node(rnp) \
357 for ((rnp) = rcu_first_leaf_node(); \
358 (rnp) < &rcu_state.node[rcu_num_nodes]; (rnp)++)
361 * Iterate over all possible CPUs in a leaf RCU node.
363 #define for_each_leaf_node_possible_cpu(rnp, cpu) \
364 for (WARN_ON_ONCE(!rcu_is_leaf_node(rnp)), \
365 (cpu) = cpumask_next((rnp)->grplo - 1, cpu_possible_mask); \
366 (cpu) <= rnp->grphi; \
367 (cpu) = cpumask_next((cpu), cpu_possible_mask))
370 * Iterate over all CPUs in a leaf RCU node's specified mask.
372 #define rcu_find_next_bit(rnp, cpu, mask) \
373 ((rnp)->grplo + find_next_bit(&(mask), BITS_PER_LONG, (cpu)))
374 #define for_each_leaf_node_cpu_mask(rnp, cpu, mask) \
375 for (WARN_ON_ONCE(!rcu_is_leaf_node(rnp)), \
376 (cpu) = rcu_find_next_bit((rnp), 0, (mask)); \
377 (cpu) <= rnp->grphi; \
378 (cpu) = rcu_find_next_bit((rnp), (cpu) + 1 - (rnp->grplo), (mask)))
380 #endif /* !defined(CONFIG_TINY_RCU) */
382 #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_TASKS_RCU_GENERIC)
385 * Wrappers for the rcu_node::lock acquire and release.
387 * Because the rcu_nodes form a tree, the tree traversal locking will observe
388 * different lock values, this in turn means that an UNLOCK of one level
389 * followed by a LOCK of another level does not imply a full memory barrier;
390 * and most importantly transitivity is lost.
392 * In order to restore full ordering between tree levels, augment the regular
393 * lock acquire functions with smp_mb__after_unlock_lock().
395 * As ->lock of struct rcu_node is a __private field, therefore one should use
396 * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
398 #define raw_spin_lock_rcu_node(p) \
400 raw_spin_lock(&ACCESS_PRIVATE(p, lock)); \
401 smp_mb__after_unlock_lock(); \
404 #define raw_spin_unlock_rcu_node(p) \
406 lockdep_assert_irqs_disabled(); \
407 raw_spin_unlock(&ACCESS_PRIVATE(p, lock)); \
410 #define raw_spin_lock_irq_rcu_node(p) \
412 raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock)); \
413 smp_mb__after_unlock_lock(); \
416 #define raw_spin_unlock_irq_rcu_node(p) \
418 lockdep_assert_irqs_disabled(); \
419 raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock)); \
422 #define raw_spin_lock_irqsave_rcu_node(p, flags) \
424 raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags); \
425 smp_mb__after_unlock_lock(); \
428 #define raw_spin_unlock_irqrestore_rcu_node(p, flags) \
430 lockdep_assert_irqs_disabled(); \
431 raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags); \
434 #define raw_spin_trylock_rcu_node(p) \
436 bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock)); \
439 smp_mb__after_unlock_lock(); \
443 #define raw_lockdep_assert_held_rcu_node(p) \
444 lockdep_assert_held(&ACCESS_PRIVATE(p, lock))
446 #endif // #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_TASKS_RCU_GENERIC)
448 #ifdef CONFIG_TINY_RCU
449 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
450 static inline bool rcu_gp_is_normal(void) { return true; }
451 static inline bool rcu_gp_is_expedited(void) { return false; }
452 static inline bool rcu_async_should_hurry(void) { return false; }
453 static inline void rcu_expedite_gp(void) { }
454 static inline void rcu_unexpedite_gp(void) { }
455 static inline void rcu_async_hurry(void) { }
456 static inline void rcu_async_relax(void) { }
457 static inline void rcu_request_urgent_qs_task(struct task_struct *t) { }
458 #else /* #ifdef CONFIG_TINY_RCU */
459 bool rcu_gp_is_normal(void); /* Internal RCU use. */
460 bool rcu_gp_is_expedited(void); /* Internal RCU use. */
461 bool rcu_async_should_hurry(void); /* Internal RCU use. */
462 void rcu_expedite_gp(void);
463 void rcu_unexpedite_gp(void);
464 void rcu_async_hurry(void);
465 void rcu_async_relax(void);
466 void rcupdate_announce_bootup_oddness(void);
467 #ifdef CONFIG_TASKS_RCU_GENERIC
468 void show_rcu_tasks_gp_kthreads(void);
469 #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */
470 static inline void show_rcu_tasks_gp_kthreads(void) {}
471 #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */
472 void rcu_request_urgent_qs_task(struct task_struct *t);
473 #endif /* #else #ifdef CONFIG_TINY_RCU */
475 #define RCU_SCHEDULER_INACTIVE 0
476 #define RCU_SCHEDULER_INIT 1
477 #define RCU_SCHEDULER_RUNNING 2
479 enum rcutorture_type {
482 RCU_TASKS_RUDE_FLAVOR,
483 RCU_TASKS_TRACING_FLAVOR,
489 #if defined(CONFIG_RCU_LAZY)
490 unsigned long rcu_lazy_get_jiffies_till_flush(void);
491 void rcu_lazy_set_jiffies_till_flush(unsigned long j);
493 static inline unsigned long rcu_lazy_get_jiffies_till_flush(void) { return 0; }
494 static inline void rcu_lazy_set_jiffies_till_flush(unsigned long j) { }
497 #if defined(CONFIG_TREE_RCU)
498 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
499 unsigned long *gp_seq);
500 void do_trace_rcu_torture_read(const char *rcutorturename,
501 struct rcu_head *rhp,
505 void rcu_gp_set_torture_wait(int duration);
507 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
508 int *flags, unsigned long *gp_seq)
513 #ifdef CONFIG_RCU_TRACE
514 void do_trace_rcu_torture_read(const char *rcutorturename,
515 struct rcu_head *rhp,
520 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
523 static inline void rcu_gp_set_torture_wait(int duration) { }
526 #if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST)
527 long rcutorture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask);
530 #ifdef CONFIG_TINY_SRCU
532 static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
533 struct srcu_struct *sp, int *flags,
534 unsigned long *gp_seq)
536 if (test_type != SRCU_FLAVOR)
539 *gp_seq = sp->srcu_idx;
542 #elif defined(CONFIG_TREE_SRCU)
544 void srcutorture_get_gp_data(enum rcutorture_type test_type,
545 struct srcu_struct *sp, int *flags,
546 unsigned long *gp_seq);
550 #ifdef CONFIG_TINY_RCU
551 static inline bool rcu_dynticks_zero_in_eqs(int cpu, int *vp) { return false; }
552 static inline unsigned long rcu_get_gp_seq(void) { return 0; }
553 static inline unsigned long rcu_exp_batches_completed(void) { return 0; }
554 static inline unsigned long
555 srcu_batches_completed(struct srcu_struct *sp) { return 0; }
556 static inline void rcu_force_quiescent_state(void) { }
557 static inline bool rcu_check_boost_fail(unsigned long gp_state, int *cpup) { return true; }
558 static inline void show_rcu_gp_kthreads(void) { }
559 static inline int rcu_get_gp_kthreads_prio(void) { return 0; }
560 static inline void rcu_fwd_progress_check(unsigned long j) { }
561 static inline void rcu_gp_slow_register(atomic_t *rgssp) { }
562 static inline void rcu_gp_slow_unregister(atomic_t *rgssp) { }
563 #else /* #ifdef CONFIG_TINY_RCU */
564 bool rcu_dynticks_zero_in_eqs(int cpu, int *vp);
565 unsigned long rcu_get_gp_seq(void);
566 unsigned long rcu_exp_batches_completed(void);
567 unsigned long srcu_batches_completed(struct srcu_struct *sp);
568 bool rcu_check_boost_fail(unsigned long gp_state, int *cpup);
569 void show_rcu_gp_kthreads(void);
570 int rcu_get_gp_kthreads_prio(void);
571 void rcu_fwd_progress_check(unsigned long j);
572 void rcu_force_quiescent_state(void);
573 extern struct workqueue_struct *rcu_gp_wq;
574 #ifdef CONFIG_RCU_EXP_KTHREAD
575 extern struct kthread_worker *rcu_exp_gp_kworker;
576 extern struct kthread_worker *rcu_exp_par_gp_kworker;
577 #else /* !CONFIG_RCU_EXP_KTHREAD */
578 extern struct workqueue_struct *rcu_par_gp_wq;
579 #endif /* CONFIG_RCU_EXP_KTHREAD */
580 void rcu_gp_slow_register(atomic_t *rgssp);
581 void rcu_gp_slow_unregister(atomic_t *rgssp);
582 #endif /* #else #ifdef CONFIG_TINY_RCU */
584 #ifdef CONFIG_RCU_NOCB_CPU
585 void rcu_bind_current_to_nocb(void);
587 static inline void rcu_bind_current_to_nocb(void) { }
590 #if !defined(CONFIG_TINY_RCU) && defined(CONFIG_TASKS_RCU)
591 void show_rcu_tasks_classic_gp_kthread(void);
593 static inline void show_rcu_tasks_classic_gp_kthread(void) {}
595 #if !defined(CONFIG_TINY_RCU) && defined(CONFIG_TASKS_RUDE_RCU)
596 void show_rcu_tasks_rude_gp_kthread(void);
598 static inline void show_rcu_tasks_rude_gp_kthread(void) {}
600 #if !defined(CONFIG_TINY_RCU) && defined(CONFIG_TASKS_TRACE_RCU)
601 void show_rcu_tasks_trace_gp_kthread(void);
603 static inline void show_rcu_tasks_trace_gp_kthread(void) {}
606 #endif /* __LINUX_RCU_H */