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