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/*
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 */
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 */