Loading...
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 */
30static 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 */
39static 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 */
48static 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. */
55static 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. */
63static 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. */
69static 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 */
87static 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. */
97static 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 */
106static 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 */
115static 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 */
123static 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 */
131static 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 */
141static 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
170extern struct debug_obj_descr rcuhead_debug_descr;
171
172static 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
183static 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 */
191static inline int debug_rcu_head_queue(struct rcu_head *head)
192{
193 return 0;
194}
195
196static inline void debug_rcu_head_unqueue(struct rcu_head *head)
197{
198}
199#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
200
201void 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 */
207static 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
230extern int rcu_cpu_stall_ftrace_dump;
231extern int rcu_cpu_stall_suppress;
232extern int rcu_cpu_stall_timeout;
233int rcu_jiffies_till_stall_check(void);
234
235#define rcu_ftrace_dump_stall_suppress() \
236do { \
237 if (!rcu_cpu_stall_suppress) \
238 rcu_cpu_stall_suppress = 3; \
239} while (0)
240
241#define rcu_ftrace_dump_stall_unsuppress() \
242do { \
243 if (rcu_cpu_stall_suppress == 3) \
244 rcu_cpu_stall_suppress = 0; \
245} while (0)
246
247#else /* #endif #ifdef CONFIG_RCU_STALL_COMMON */
248#define rcu_ftrace_dump_stall_suppress()
249#define rcu_ftrace_dump_stall_unsuppress()
250#endif /* #ifdef CONFIG_RCU_STALL_COMMON */
251
252/*
253 * Strings used in tracepoints need to be exported via the
254 * tracing system such that tools like perf and trace-cmd can
255 * translate the string address pointers to actual text.
256 */
257#define TPS(x) tracepoint_string(x)
258
259/*
260 * Dump the ftrace buffer, but only one time per callsite per boot.
261 */
262#define rcu_ftrace_dump(oops_dump_mode) \
263do { \
264 static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
265 \
266 if (!atomic_read(&___rfd_beenhere) && \
267 !atomic_xchg(&___rfd_beenhere, 1)) { \
268 tracing_off(); \
269 rcu_ftrace_dump_stall_suppress(); \
270 ftrace_dump(oops_dump_mode); \
271 rcu_ftrace_dump_stall_unsuppress(); \
272 } \
273} while (0)
274
275void rcu_early_boot_tests(void);
276void rcu_test_sync_prims(void);
277
278/*
279 * This function really isn't for public consumption, but RCU is special in
280 * that context switches can allow the state machine to make progress.
281 */
282extern void resched_cpu(int cpu);
283
284#if defined(SRCU) || !defined(TINY_RCU)
285
286#include <linux/rcu_node_tree.h>
287
288extern int rcu_num_lvls;
289extern int num_rcu_lvl[];
290extern int rcu_num_nodes;
291static bool rcu_fanout_exact;
292static int rcu_fanout_leaf;
293
294/*
295 * Compute the per-level fanout, either using the exact fanout specified
296 * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
297 */
298static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
299{
300 int i;
301
302 if (rcu_fanout_exact) {
303 levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
304 for (i = rcu_num_lvls - 2; i >= 0; i--)
305 levelspread[i] = RCU_FANOUT;
306 } else {
307 int ccur;
308 int cprv;
309
310 cprv = nr_cpu_ids;
311 for (i = rcu_num_lvls - 1; i >= 0; i--) {
312 ccur = levelcnt[i];
313 levelspread[i] = (cprv + ccur - 1) / ccur;
314 cprv = ccur;
315 }
316 }
317}
318
319/* Returns a pointer to the first leaf rcu_node structure. */
320#define rcu_first_leaf_node() (rcu_state.level[rcu_num_lvls - 1])
321
322/* Is this rcu_node a leaf? */
323#define rcu_is_leaf_node(rnp) ((rnp)->level == rcu_num_lvls - 1)
324
325/* Is this rcu_node the last leaf? */
326#define rcu_is_last_leaf_node(rnp) ((rnp) == &rcu_state.node[rcu_num_nodes - 1])
327
328/*
329 * Do a full breadth-first scan of the {s,}rcu_node structures for the
330 * specified state structure (for SRCU) or the only rcu_state structure
331 * (for RCU).
332 */
333#define srcu_for_each_node_breadth_first(sp, rnp) \
334 for ((rnp) = &(sp)->node[0]; \
335 (rnp) < &(sp)->node[rcu_num_nodes]; (rnp)++)
336#define rcu_for_each_node_breadth_first(rnp) \
337 srcu_for_each_node_breadth_first(&rcu_state, rnp)
338
339/*
340 * Scan the leaves of the rcu_node hierarchy for the rcu_state structure.
341 * Note that if there is a singleton rcu_node tree with but one rcu_node
342 * structure, this loop -will- visit the rcu_node structure. It is still
343 * a leaf node, even if it is also the root node.
344 */
345#define rcu_for_each_leaf_node(rnp) \
346 for ((rnp) = rcu_first_leaf_node(); \
347 (rnp) < &rcu_state.node[rcu_num_nodes]; (rnp)++)
348
349/*
350 * Iterate over all possible CPUs in a leaf RCU node.
351 */
352#define for_each_leaf_node_possible_cpu(rnp, cpu) \
353 for ((cpu) = cpumask_next((rnp)->grplo - 1, cpu_possible_mask); \
354 (cpu) <= rnp->grphi; \
355 (cpu) = cpumask_next((cpu), cpu_possible_mask))
356
357/*
358 * Iterate over all CPUs in a leaf RCU node's specified mask.
359 */
360#define rcu_find_next_bit(rnp, cpu, mask) \
361 ((rnp)->grplo + find_next_bit(&(mask), BITS_PER_LONG, (cpu)))
362#define for_each_leaf_node_cpu_mask(rnp, cpu, mask) \
363 for ((cpu) = rcu_find_next_bit((rnp), 0, (mask)); \
364 (cpu) <= rnp->grphi; \
365 (cpu) = rcu_find_next_bit((rnp), (cpu) + 1 - (rnp->grplo), (mask)))
366
367/*
368 * Wrappers for the rcu_node::lock acquire and release.
369 *
370 * Because the rcu_nodes form a tree, the tree traversal locking will observe
371 * different lock values, this in turn means that an UNLOCK of one level
372 * followed by a LOCK of another level does not imply a full memory barrier;
373 * and most importantly transitivity is lost.
374 *
375 * In order to restore full ordering between tree levels, augment the regular
376 * lock acquire functions with smp_mb__after_unlock_lock().
377 *
378 * As ->lock of struct rcu_node is a __private field, therefore one should use
379 * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
380 */
381#define raw_spin_lock_rcu_node(p) \
382do { \
383 raw_spin_lock(&ACCESS_PRIVATE(p, lock)); \
384 smp_mb__after_unlock_lock(); \
385} while (0)
386
387#define raw_spin_unlock_rcu_node(p) raw_spin_unlock(&ACCESS_PRIVATE(p, lock))
388
389#define raw_spin_lock_irq_rcu_node(p) \
390do { \
391 raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock)); \
392 smp_mb__after_unlock_lock(); \
393} while (0)
394
395#define raw_spin_unlock_irq_rcu_node(p) \
396 raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock))
397
398#define raw_spin_lock_irqsave_rcu_node(p, flags) \
399do { \
400 raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags); \
401 smp_mb__after_unlock_lock(); \
402} while (0)
403
404#define raw_spin_unlock_irqrestore_rcu_node(p, flags) \
405 raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags)
406
407#define raw_spin_trylock_rcu_node(p) \
408({ \
409 bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock)); \
410 \
411 if (___locked) \
412 smp_mb__after_unlock_lock(); \
413 ___locked; \
414})
415
416#define raw_lockdep_assert_held_rcu_node(p) \
417 lockdep_assert_held(&ACCESS_PRIVATE(p, lock))
418
419#endif /* #if defined(SRCU) || !defined(TINY_RCU) */
420
421#ifdef CONFIG_SRCU
422void srcu_init(void);
423#else /* #ifdef CONFIG_SRCU */
424static inline void srcu_init(void) { }
425#endif /* #else #ifdef CONFIG_SRCU */
426
427#ifdef CONFIG_TINY_RCU
428/* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
429static inline bool rcu_gp_is_normal(void) { return true; }
430static inline bool rcu_gp_is_expedited(void) { return false; }
431static inline void rcu_expedite_gp(void) { }
432static inline void rcu_unexpedite_gp(void) { }
433static inline void rcu_request_urgent_qs_task(struct task_struct *t) { }
434#else /* #ifdef CONFIG_TINY_RCU */
435bool rcu_gp_is_normal(void); /* Internal RCU use. */
436bool rcu_gp_is_expedited(void); /* Internal RCU use. */
437void rcu_expedite_gp(void);
438void rcu_unexpedite_gp(void);
439void rcupdate_announce_bootup_oddness(void);
440void rcu_request_urgent_qs_task(struct task_struct *t);
441#endif /* #else #ifdef CONFIG_TINY_RCU */
442
443#define RCU_SCHEDULER_INACTIVE 0
444#define RCU_SCHEDULER_INIT 1
445#define RCU_SCHEDULER_RUNNING 2
446
447enum rcutorture_type {
448 RCU_FLAVOR,
449 RCU_TASKS_FLAVOR,
450 RCU_TRIVIAL_FLAVOR,
451 SRCU_FLAVOR,
452 INVALID_RCU_FLAVOR
453};
454
455#if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
456void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
457 unsigned long *gp_seq);
458void rcutorture_record_progress(unsigned long vernum);
459void do_trace_rcu_torture_read(const char *rcutorturename,
460 struct rcu_head *rhp,
461 unsigned long secs,
462 unsigned long c_old,
463 unsigned long c);
464#else
465static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
466 int *flags, unsigned long *gp_seq)
467{
468 *flags = 0;
469 *gp_seq = 0;
470}
471static inline void rcutorture_record_progress(unsigned long vernum) { }
472#ifdef CONFIG_RCU_TRACE
473void do_trace_rcu_torture_read(const char *rcutorturename,
474 struct rcu_head *rhp,
475 unsigned long secs,
476 unsigned long c_old,
477 unsigned long c);
478#else
479#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
480 do { } while (0)
481#endif
482#endif
483
484#if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST)
485long rcutorture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask);
486#endif
487
488#ifdef CONFIG_TINY_SRCU
489
490static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
491 struct srcu_struct *sp, int *flags,
492 unsigned long *gp_seq)
493{
494 if (test_type != SRCU_FLAVOR)
495 return;
496 *flags = 0;
497 *gp_seq = sp->srcu_idx;
498}
499
500#elif defined(CONFIG_TREE_SRCU)
501
502void srcutorture_get_gp_data(enum rcutorture_type test_type,
503 struct srcu_struct *sp, int *flags,
504 unsigned long *gp_seq);
505
506#endif
507
508#ifdef CONFIG_TINY_RCU
509static inline unsigned long rcu_get_gp_seq(void) { return 0; }
510static inline unsigned long rcu_exp_batches_completed(void) { return 0; }
511static inline unsigned long
512srcu_batches_completed(struct srcu_struct *sp) { return 0; }
513static inline void rcu_force_quiescent_state(void) { }
514static inline void show_rcu_gp_kthreads(void) { }
515static inline int rcu_get_gp_kthreads_prio(void) { return 0; }
516static inline void rcu_fwd_progress_check(unsigned long j) { }
517#else /* #ifdef CONFIG_TINY_RCU */
518unsigned long rcu_get_gp_seq(void);
519unsigned long rcu_exp_batches_completed(void);
520unsigned long srcu_batches_completed(struct srcu_struct *sp);
521void show_rcu_gp_kthreads(void);
522int rcu_get_gp_kthreads_prio(void);
523void rcu_fwd_progress_check(unsigned long j);
524void rcu_force_quiescent_state(void);
525extern struct workqueue_struct *rcu_gp_wq;
526extern struct workqueue_struct *rcu_par_gp_wq;
527#endif /* #else #ifdef CONFIG_TINY_RCU */
528
529#ifdef CONFIG_RCU_NOCB_CPU
530bool rcu_is_nocb_cpu(int cpu);
531void rcu_bind_current_to_nocb(void);
532#else
533static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
534static inline void rcu_bind_current_to_nocb(void) { }
535#endif
536
537#endif /* __LINUX_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/*
16 * Grace-period counter management.
17 */
18
19#define RCU_SEQ_CTR_SHIFT 2
20#define RCU_SEQ_STATE_MASK ((1 << RCU_SEQ_CTR_SHIFT) - 1)
21
22/* Low-order bit definition for polled grace-period APIs. */
23#define RCU_GET_STATE_COMPLETED 0x1
24
25extern int sysctl_sched_rt_runtime;
26
27/*
28 * Return the counter portion of a sequence number previously returned
29 * by rcu_seq_snap() or rcu_seq_current().
30 */
31static inline unsigned long rcu_seq_ctr(unsigned long s)
32{
33 return s >> RCU_SEQ_CTR_SHIFT;
34}
35
36/*
37 * Return the state portion of a sequence number previously returned
38 * by rcu_seq_snap() or rcu_seq_current().
39 */
40static inline int rcu_seq_state(unsigned long s)
41{
42 return s & RCU_SEQ_STATE_MASK;
43}
44
45/*
46 * Set the state portion of the pointed-to sequence number.
47 * The caller is responsible for preventing conflicting updates.
48 */
49static inline void rcu_seq_set_state(unsigned long *sp, int newstate)
50{
51 WARN_ON_ONCE(newstate & ~RCU_SEQ_STATE_MASK);
52 WRITE_ONCE(*sp, (*sp & ~RCU_SEQ_STATE_MASK) + newstate);
53}
54
55/* Adjust sequence number for start of update-side operation. */
56static inline void rcu_seq_start(unsigned long *sp)
57{
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);
61}
62
63/* Compute the end-of-grace-period value for the specified sequence number. */
64static inline unsigned long rcu_seq_endval(unsigned long *sp)
65{
66 return (*sp | RCU_SEQ_STATE_MASK) + 1;
67}
68
69/* Adjust sequence number for end of update-side operation. */
70static inline void rcu_seq_end(unsigned long *sp)
71{
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));
75}
76
77/*
78 * rcu_seq_snap - Take a snapshot of the update side's sequence number.
79 *
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.
87 */
88static inline unsigned long rcu_seq_snap(unsigned long *sp)
89{
90 unsigned long s;
91
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. */
94 return s;
95}
96
97/* Return the current value the update side's sequence number, no ordering. */
98static inline unsigned long rcu_seq_current(unsigned long *sp)
99{
100 return READ_ONCE(*sp);
101}
102
103/*
104 * Given a snapshot from rcu_seq_snap(), determine whether or not the
105 * corresponding update-side operation has started.
106 */
107static inline bool rcu_seq_started(unsigned long *sp, unsigned long s)
108{
109 return ULONG_CMP_LT((s - 1) & ~RCU_SEQ_STATE_MASK, READ_ONCE(*sp));
110}
111
112/*
113 * Given a snapshot from rcu_seq_snap(), determine whether or not a
114 * full update-side operation has occurred.
115 */
116static inline bool rcu_seq_done(unsigned long *sp, unsigned long s)
117{
118 return ULONG_CMP_GE(READ_ONCE(*sp), s);
119}
120
121/*
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.
125 */
126static inline bool rcu_seq_done_exact(unsigned long *sp, unsigned long s)
127{
128 unsigned long cur_s = READ_ONCE(*sp);
129
130 return ULONG_CMP_GE(cur_s, s) || ULONG_CMP_LT(cur_s, s - (2 * RCU_SEQ_STATE_MASK + 1));
131}
132
133/*
134 * Has a grace period completed since the time the old gp_seq was collected?
135 */
136static inline bool rcu_seq_completed_gp(unsigned long old, unsigned long new)
137{
138 return ULONG_CMP_LT(old, new & ~RCU_SEQ_STATE_MASK);
139}
140
141/*
142 * Has a grace period started since the time the old gp_seq was collected?
143 */
144static inline bool rcu_seq_new_gp(unsigned long old, unsigned long new)
145{
146 return ULONG_CMP_LT((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK,
147 new);
148}
149
150/*
151 * Roughly how many full grace periods have elapsed between the collection
152 * of the two specified grace periods?
153 */
154static inline unsigned long rcu_seq_diff(unsigned long new, unsigned long old)
155{
156 unsigned long rnd_diff;
157
158 if (old == new)
159 return 0;
160 /*
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.
163 */
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;
170}
171
172/*
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.
177 */
178
179#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
180# define STATE_RCU_HEAD_READY 0
181# define STATE_RCU_HEAD_QUEUED 1
182
183extern const struct debug_obj_descr rcuhead_debug_descr;
184
185static inline int debug_rcu_head_queue(struct rcu_head *head)
186{
187 int r1;
188
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);
193 return r1;
194}
195
196static inline void debug_rcu_head_unqueue(struct rcu_head *head)
197{
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);
202}
203#else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
204static inline int debug_rcu_head_queue(struct rcu_head *head)
205{
206 return 0;
207}
208
209static inline void debug_rcu_head_unqueue(struct rcu_head *head)
210{
211}
212#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
213
214extern int rcu_cpu_stall_suppress_at_boot;
215
216static inline bool rcu_stall_is_suppressed_at_boot(void)
217{
218 return rcu_cpu_stall_suppress_at_boot && !rcu_inkernel_boot_has_ended();
219}
220
221#ifdef CONFIG_RCU_STALL_COMMON
222
223extern int rcu_cpu_stall_ftrace_dump;
224extern int rcu_cpu_stall_suppress;
225extern int rcu_cpu_stall_timeout;
226extern int rcu_exp_cpu_stall_timeout;
227int rcu_jiffies_till_stall_check(void);
228int rcu_exp_jiffies_till_stall_check(void);
229
230static inline bool rcu_stall_is_suppressed(void)
231{
232 return rcu_stall_is_suppressed_at_boot() || rcu_cpu_stall_suppress;
233}
234
235#define rcu_ftrace_dump_stall_suppress() \
236do { \
237 if (!rcu_cpu_stall_suppress) \
238 rcu_cpu_stall_suppress = 3; \
239} while (0)
240
241#define rcu_ftrace_dump_stall_unsuppress() \
242do { \
243 if (rcu_cpu_stall_suppress == 3) \
244 rcu_cpu_stall_suppress = 0; \
245} while (0)
246
247#else /* #endif #ifdef CONFIG_RCU_STALL_COMMON */
248
249static inline bool rcu_stall_is_suppressed(void)
250{
251 return rcu_stall_is_suppressed_at_boot();
252}
253#define rcu_ftrace_dump_stall_suppress()
254#define rcu_ftrace_dump_stall_unsuppress()
255#endif /* #ifdef CONFIG_RCU_STALL_COMMON */
256
257/*
258 * Strings used in tracepoints need to be exported via the
259 * tracing system such that tools like perf and trace-cmd can
260 * translate the string address pointers to actual text.
261 */
262#define TPS(x) tracepoint_string(x)
263
264/*
265 * Dump the ftrace buffer, but only one time per callsite per boot.
266 */
267#define rcu_ftrace_dump(oops_dump_mode) \
268do { \
269 static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
270 \
271 if (!atomic_read(&___rfd_beenhere) && \
272 !atomic_xchg(&___rfd_beenhere, 1)) { \
273 tracing_off(); \
274 rcu_ftrace_dump_stall_suppress(); \
275 ftrace_dump(oops_dump_mode); \
276 rcu_ftrace_dump_stall_unsuppress(); \
277 } \
278} while (0)
279
280void rcu_early_boot_tests(void);
281void rcu_test_sync_prims(void);
282
283/*
284 * This function really isn't for public consumption, but RCU is special in
285 * that context switches can allow the state machine to make progress.
286 */
287extern void resched_cpu(int cpu);
288
289#if !defined(CONFIG_TINY_RCU)
290
291#include <linux/rcu_node_tree.h>
292
293extern int rcu_num_lvls;
294extern int num_rcu_lvl[];
295extern int rcu_num_nodes;
296static bool rcu_fanout_exact;
297static int rcu_fanout_leaf;
298
299/*
300 * Compute the per-level fanout, either using the exact fanout specified
301 * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
302 */
303static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
304{
305 int i;
306
307 for (i = 0; i < RCU_NUM_LVLS; i++)
308 levelspread[i] = INT_MIN;
309 if (rcu_fanout_exact) {
310 levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
311 for (i = rcu_num_lvls - 2; i >= 0; i--)
312 levelspread[i] = RCU_FANOUT;
313 } else {
314 int ccur;
315 int cprv;
316
317 cprv = nr_cpu_ids;
318 for (i = rcu_num_lvls - 1; i >= 0; i--) {
319 ccur = levelcnt[i];
320 levelspread[i] = (cprv + ccur - 1) / ccur;
321 cprv = ccur;
322 }
323 }
324}
325
326extern void rcu_init_geometry(void);
327
328/* Returns a pointer to the first leaf rcu_node structure. */
329#define rcu_first_leaf_node() (rcu_state.level[rcu_num_lvls - 1])
330
331/* Is this rcu_node a leaf? */
332#define rcu_is_leaf_node(rnp) ((rnp)->level == rcu_num_lvls - 1)
333
334/* Is this rcu_node the last leaf? */
335#define rcu_is_last_leaf_node(rnp) ((rnp) == &rcu_state.node[rcu_num_nodes - 1])
336
337/*
338 * Do a full breadth-first scan of the {s,}rcu_node structures for the
339 * specified state structure (for SRCU) or the only rcu_state structure
340 * (for RCU).
341 */
342#define srcu_for_each_node_breadth_first(sp, rnp) \
343 for ((rnp) = &(sp)->node[0]; \
344 (rnp) < &(sp)->node[rcu_num_nodes]; (rnp)++)
345#define rcu_for_each_node_breadth_first(rnp) \
346 srcu_for_each_node_breadth_first(&rcu_state, rnp)
347
348/*
349 * Scan the leaves of the rcu_node hierarchy for the rcu_state structure.
350 * Note that if there is a singleton rcu_node tree with but one rcu_node
351 * structure, this loop -will- visit the rcu_node structure. It is still
352 * a leaf node, even if it is also the root node.
353 */
354#define rcu_for_each_leaf_node(rnp) \
355 for ((rnp) = rcu_first_leaf_node(); \
356 (rnp) < &rcu_state.node[rcu_num_nodes]; (rnp)++)
357
358/*
359 * Iterate over all possible CPUs in a leaf RCU node.
360 */
361#define for_each_leaf_node_possible_cpu(rnp, cpu) \
362 for (WARN_ON_ONCE(!rcu_is_leaf_node(rnp)), \
363 (cpu) = cpumask_next((rnp)->grplo - 1, cpu_possible_mask); \
364 (cpu) <= rnp->grphi; \
365 (cpu) = cpumask_next((cpu), cpu_possible_mask))
366
367/*
368 * Iterate over all CPUs in a leaf RCU node's specified mask.
369 */
370#define rcu_find_next_bit(rnp, cpu, mask) \
371 ((rnp)->grplo + find_next_bit(&(mask), BITS_PER_LONG, (cpu)))
372#define for_each_leaf_node_cpu_mask(rnp, cpu, mask) \
373 for (WARN_ON_ONCE(!rcu_is_leaf_node(rnp)), \
374 (cpu) = rcu_find_next_bit((rnp), 0, (mask)); \
375 (cpu) <= rnp->grphi; \
376 (cpu) = rcu_find_next_bit((rnp), (cpu) + 1 - (rnp->grplo), (mask)))
377
378#endif /* !defined(CONFIG_TINY_RCU) */
379
380#if !defined(CONFIG_TINY_RCU) || defined(CONFIG_TASKS_RCU_GENERIC)
381
382/*
383 * Wrappers for the rcu_node::lock acquire and release.
384 *
385 * Because the rcu_nodes form a tree, the tree traversal locking will observe
386 * different lock values, this in turn means that an UNLOCK of one level
387 * followed by a LOCK of another level does not imply a full memory barrier;
388 * and most importantly transitivity is lost.
389 *
390 * In order to restore full ordering between tree levels, augment the regular
391 * lock acquire functions with smp_mb__after_unlock_lock().
392 *
393 * As ->lock of struct rcu_node is a __private field, therefore one should use
394 * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
395 */
396#define raw_spin_lock_rcu_node(p) \
397do { \
398 raw_spin_lock(&ACCESS_PRIVATE(p, lock)); \
399 smp_mb__after_unlock_lock(); \
400} while (0)
401
402#define raw_spin_unlock_rcu_node(p) \
403do { \
404 lockdep_assert_irqs_disabled(); \
405 raw_spin_unlock(&ACCESS_PRIVATE(p, lock)); \
406} while (0)
407
408#define raw_spin_lock_irq_rcu_node(p) \
409do { \
410 raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock)); \
411 smp_mb__after_unlock_lock(); \
412} while (0)
413
414#define raw_spin_unlock_irq_rcu_node(p) \
415do { \
416 lockdep_assert_irqs_disabled(); \
417 raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock)); \
418} while (0)
419
420#define raw_spin_lock_irqsave_rcu_node(p, flags) \
421do { \
422 raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags); \
423 smp_mb__after_unlock_lock(); \
424} while (0)
425
426#define raw_spin_unlock_irqrestore_rcu_node(p, flags) \
427do { \
428 lockdep_assert_irqs_disabled(); \
429 raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags); \
430} while (0)
431
432#define raw_spin_trylock_rcu_node(p) \
433({ \
434 bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock)); \
435 \
436 if (___locked) \
437 smp_mb__after_unlock_lock(); \
438 ___locked; \
439})
440
441#define raw_lockdep_assert_held_rcu_node(p) \
442 lockdep_assert_held(&ACCESS_PRIVATE(p, lock))
443
444#endif // #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_TASKS_RCU_GENERIC)
445
446#ifdef CONFIG_TINY_RCU
447/* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
448static inline bool rcu_gp_is_normal(void) { return true; }
449static inline bool rcu_gp_is_expedited(void) { return false; }
450static inline void rcu_expedite_gp(void) { }
451static inline void rcu_unexpedite_gp(void) { }
452static inline void rcu_request_urgent_qs_task(struct task_struct *t) { }
453#else /* #ifdef CONFIG_TINY_RCU */
454bool rcu_gp_is_normal(void); /* Internal RCU use. */
455bool rcu_gp_is_expedited(void); /* Internal RCU use. */
456void rcu_expedite_gp(void);
457void rcu_unexpedite_gp(void);
458void rcupdate_announce_bootup_oddness(void);
459#ifdef CONFIG_TASKS_RCU_GENERIC
460void show_rcu_tasks_gp_kthreads(void);
461#else /* #ifdef CONFIG_TASKS_RCU_GENERIC */
462static inline void show_rcu_tasks_gp_kthreads(void) {}
463#endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */
464void rcu_request_urgent_qs_task(struct task_struct *t);
465#endif /* #else #ifdef CONFIG_TINY_RCU */
466
467#define RCU_SCHEDULER_INACTIVE 0
468#define RCU_SCHEDULER_INIT 1
469#define RCU_SCHEDULER_RUNNING 2
470
471enum rcutorture_type {
472 RCU_FLAVOR,
473 RCU_TASKS_FLAVOR,
474 RCU_TASKS_RUDE_FLAVOR,
475 RCU_TASKS_TRACING_FLAVOR,
476 RCU_TRIVIAL_FLAVOR,
477 SRCU_FLAVOR,
478 INVALID_RCU_FLAVOR
479};
480
481#if defined(CONFIG_RCU_LAZY)
482unsigned long rcu_lazy_get_jiffies_till_flush(void);
483void rcu_lazy_set_jiffies_till_flush(unsigned long j);
484#else
485static inline unsigned long rcu_lazy_get_jiffies_till_flush(void) { return 0; }
486static inline void rcu_lazy_set_jiffies_till_flush(unsigned long j) { }
487#endif
488
489#if defined(CONFIG_TREE_RCU)
490void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
491 unsigned long *gp_seq);
492void do_trace_rcu_torture_read(const char *rcutorturename,
493 struct rcu_head *rhp,
494 unsigned long secs,
495 unsigned long c_old,
496 unsigned long c);
497void rcu_gp_set_torture_wait(int duration);
498#else
499static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
500 int *flags, unsigned long *gp_seq)
501{
502 *flags = 0;
503 *gp_seq = 0;
504}
505#ifdef CONFIG_RCU_TRACE
506void do_trace_rcu_torture_read(const char *rcutorturename,
507 struct rcu_head *rhp,
508 unsigned long secs,
509 unsigned long c_old,
510 unsigned long c);
511#else
512#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
513 do { } while (0)
514#endif
515static inline void rcu_gp_set_torture_wait(int duration) { }
516#endif
517
518#if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST)
519long rcutorture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask);
520#endif
521
522#ifdef CONFIG_TINY_SRCU
523
524static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
525 struct srcu_struct *sp, int *flags,
526 unsigned long *gp_seq)
527{
528 if (test_type != SRCU_FLAVOR)
529 return;
530 *flags = 0;
531 *gp_seq = sp->srcu_idx;
532}
533
534#elif defined(CONFIG_TREE_SRCU)
535
536void srcutorture_get_gp_data(enum rcutorture_type test_type,
537 struct srcu_struct *sp, int *flags,
538 unsigned long *gp_seq);
539
540#endif
541
542#ifdef CONFIG_TINY_RCU
543static inline bool rcu_dynticks_zero_in_eqs(int cpu, int *vp) { return false; }
544static inline unsigned long rcu_get_gp_seq(void) { return 0; }
545static inline unsigned long rcu_exp_batches_completed(void) { return 0; }
546static inline unsigned long
547srcu_batches_completed(struct srcu_struct *sp) { return 0; }
548static inline void rcu_force_quiescent_state(void) { }
549static inline bool rcu_check_boost_fail(unsigned long gp_state, int *cpup) { return true; }
550static inline void show_rcu_gp_kthreads(void) { }
551static inline int rcu_get_gp_kthreads_prio(void) { return 0; }
552static inline void rcu_fwd_progress_check(unsigned long j) { }
553static inline void rcu_gp_slow_register(atomic_t *rgssp) { }
554static inline void rcu_gp_slow_unregister(atomic_t *rgssp) { }
555#else /* #ifdef CONFIG_TINY_RCU */
556bool rcu_dynticks_zero_in_eqs(int cpu, int *vp);
557unsigned long rcu_get_gp_seq(void);
558unsigned long rcu_exp_batches_completed(void);
559unsigned long srcu_batches_completed(struct srcu_struct *sp);
560bool rcu_check_boost_fail(unsigned long gp_state, int *cpup);
561void show_rcu_gp_kthreads(void);
562int rcu_get_gp_kthreads_prio(void);
563void rcu_fwd_progress_check(unsigned long j);
564void rcu_force_quiescent_state(void);
565extern struct workqueue_struct *rcu_gp_wq;
566#ifdef CONFIG_RCU_EXP_KTHREAD
567extern struct kthread_worker *rcu_exp_gp_kworker;
568extern struct kthread_worker *rcu_exp_par_gp_kworker;
569#else /* !CONFIG_RCU_EXP_KTHREAD */
570extern struct workqueue_struct *rcu_par_gp_wq;
571#endif /* CONFIG_RCU_EXP_KTHREAD */
572void rcu_gp_slow_register(atomic_t *rgssp);
573void rcu_gp_slow_unregister(atomic_t *rgssp);
574#endif /* #else #ifdef CONFIG_TINY_RCU */
575
576#ifdef CONFIG_RCU_NOCB_CPU
577void rcu_bind_current_to_nocb(void);
578#else
579static inline void rcu_bind_current_to_nocb(void) { }
580#endif
581
582#if !defined(CONFIG_TINY_RCU) && defined(CONFIG_TASKS_RCU)
583void show_rcu_tasks_classic_gp_kthread(void);
584#else
585static inline void show_rcu_tasks_classic_gp_kthread(void) {}
586#endif
587#if !defined(CONFIG_TINY_RCU) && defined(CONFIG_TASKS_RUDE_RCU)
588void show_rcu_tasks_rude_gp_kthread(void);
589#else
590static inline void show_rcu_tasks_rude_gp_kthread(void) {}
591#endif
592#if !defined(CONFIG_TINY_RCU) && defined(CONFIG_TASKS_TRACE_RCU)
593void show_rcu_tasks_trace_gp_kthread(void);
594#else
595static inline void show_rcu_tasks_trace_gp_kthread(void) {}
596#endif
597
598#endif /* __LINUX_RCU_H */