2 * Read-Copy Update definitions shared among RCU implementations.
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.
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.
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.
18 * Copyright IBM Corporation, 2011
20 * Author: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
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 */
33 /* Offset to allow for unmatched rcu_irq_{enter,exit}(). */
34 #define DYNTICK_IRQ_NONIDLE ((LONG_MAX / 2) + 1)
38 * Grace-period counter management.
41 #define RCU_SEQ_CTR_SHIFT 2
42 #define RCU_SEQ_STATE_MASK ((1 << RCU_SEQ_CTR_SHIFT) - 1)
45 * Return the counter portion of a sequence number previously returned
46 * by rcu_seq_snap() or rcu_seq_current().
48 static inline unsigned long rcu_seq_ctr(unsigned long s)
50 return s >> RCU_SEQ_CTR_SHIFT;
54 * Return the state portion of a sequence number previously returned
55 * by rcu_seq_snap() or rcu_seq_current().
57 static inline int rcu_seq_state(unsigned long s)
59 return s & RCU_SEQ_STATE_MASK;
63 * Set the state portion of the pointed-to sequence number.
64 * The caller is responsible for preventing conflicting updates.
66 static inline void rcu_seq_set_state(unsigned long *sp, int newstate)
68 WARN_ON_ONCE(newstate & ~RCU_SEQ_STATE_MASK);
69 WRITE_ONCE(*sp, (*sp & ~RCU_SEQ_STATE_MASK) + newstate);
72 /* Adjust sequence number for start of update-side operation. */
73 static inline void rcu_seq_start(unsigned long *sp)
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);
80 /* Compute the end-of-grace-period value for the specified sequence number. */
81 static inline unsigned long rcu_seq_endval(unsigned long *sp)
83 return (*sp | RCU_SEQ_STATE_MASK) + 1;
86 /* Adjust sequence number for end of update-side operation. */
87 static inline void rcu_seq_end(unsigned long *sp)
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));
95 * rcu_seq_snap - Take a snapshot of the update side's sequence number.
97 * This function returns the earliest value of the grace-period sequence number
98 * that will indicate that a full grace period has elapsed since the current
99 * time. Once the grace-period sequence number has reached this value, it will
100 * be safe to invoke all callbacks that have been registered prior to the
101 * current time. This value is the current grace-period number plus two to the
102 * power of the number of low-order bits reserved for state, then rounded up to
103 * the next value in which the state bits are all zero.
105 static inline unsigned long rcu_seq_snap(unsigned long *sp)
109 s = (READ_ONCE(*sp) + 2 * RCU_SEQ_STATE_MASK + 1) & ~RCU_SEQ_STATE_MASK;
110 smp_mb(); /* Above access must not bleed into critical section. */
114 /* Return the current value the update side's sequence number, no ordering. */
115 static inline unsigned long rcu_seq_current(unsigned long *sp)
117 return READ_ONCE(*sp);
121 * Given a snapshot from rcu_seq_snap(), determine whether or not the
122 * corresponding update-side operation has started.
124 static inline bool rcu_seq_started(unsigned long *sp, unsigned long s)
126 return ULONG_CMP_LT((s - 1) & ~RCU_SEQ_STATE_MASK, READ_ONCE(*sp));
130 * Given a snapshot from rcu_seq_snap(), determine whether or not a
131 * full update-side operation has occurred.
133 static inline bool rcu_seq_done(unsigned long *sp, unsigned long s)
135 return ULONG_CMP_GE(READ_ONCE(*sp), s);
139 * Has a grace period completed since the time the old gp_seq was collected?
141 static inline bool rcu_seq_completed_gp(unsigned long old, unsigned long new)
143 return ULONG_CMP_LT(old, new & ~RCU_SEQ_STATE_MASK);
147 * Has a grace period started since the time the old gp_seq was collected?
149 static inline bool rcu_seq_new_gp(unsigned long old, unsigned long new)
151 return ULONG_CMP_LT((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK,
156 * Roughly how many full grace periods have elapsed between the collection
157 * of the two specified grace periods?
159 static inline unsigned long rcu_seq_diff(unsigned long new, unsigned long old)
161 unsigned long rnd_diff;
166 * Compute the number of grace periods (still shifted up), plus
167 * one if either of new and old is not an exact grace period.
169 rnd_diff = (new & ~RCU_SEQ_STATE_MASK) -
170 ((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK) +
171 ((new & RCU_SEQ_STATE_MASK) || (old & RCU_SEQ_STATE_MASK));
172 if (ULONG_CMP_GE(RCU_SEQ_STATE_MASK, rnd_diff))
173 return 1; /* Definitely no grace period has elapsed. */
174 return ((rnd_diff - RCU_SEQ_STATE_MASK - 1) >> RCU_SEQ_CTR_SHIFT) + 2;
178 * debug_rcu_head_queue()/debug_rcu_head_unqueue() are used internally
179 * by call_rcu() and rcu callback execution, and are therefore not part
180 * of the RCU API. These are in rcupdate.h because they are used by all
181 * RCU implementations.
184 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
185 # define STATE_RCU_HEAD_READY 0
186 # define STATE_RCU_HEAD_QUEUED 1
188 extern struct debug_obj_descr rcuhead_debug_descr;
190 static inline int debug_rcu_head_queue(struct rcu_head *head)
194 r1 = debug_object_activate(head, &rcuhead_debug_descr);
195 debug_object_active_state(head, &rcuhead_debug_descr,
196 STATE_RCU_HEAD_READY,
197 STATE_RCU_HEAD_QUEUED);
201 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
203 debug_object_active_state(head, &rcuhead_debug_descr,
204 STATE_RCU_HEAD_QUEUED,
205 STATE_RCU_HEAD_READY);
206 debug_object_deactivate(head, &rcuhead_debug_descr);
208 #else /* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
209 static inline int debug_rcu_head_queue(struct rcu_head *head)
214 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
217 #endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
219 void kfree(const void *);
222 * Reclaim the specified callback, either by invoking it (non-lazy case)
223 * or freeing it directly (lazy case). Return true if lazy, false otherwise.
225 static inline bool __rcu_reclaim(const char *rn, struct rcu_head *head)
228 unsigned long offset = (unsigned long)head->func;
230 rcu_lock_acquire(&rcu_callback_map);
231 if (__is_kfree_rcu_offset(offset)) {
232 RCU_TRACE(trace_rcu_invoke_kfree_callback(rn, head, offset);)
233 kfree((void *)head - offset);
234 rcu_lock_release(&rcu_callback_map);
237 RCU_TRACE(trace_rcu_invoke_callback(rn, head);)
239 WRITE_ONCE(head->func, (rcu_callback_t)0L);
241 rcu_lock_release(&rcu_callback_map);
246 #ifdef CONFIG_RCU_STALL_COMMON
248 extern int rcu_cpu_stall_suppress;
249 int rcu_jiffies_till_stall_check(void);
251 #define rcu_ftrace_dump_stall_suppress() \
253 if (!rcu_cpu_stall_suppress) \
254 rcu_cpu_stall_suppress = 3; \
257 #define rcu_ftrace_dump_stall_unsuppress() \
259 if (rcu_cpu_stall_suppress == 3) \
260 rcu_cpu_stall_suppress = 0; \
263 #else /* #endif #ifdef CONFIG_RCU_STALL_COMMON */
264 #define rcu_ftrace_dump_stall_suppress()
265 #define rcu_ftrace_dump_stall_unsuppress()
266 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */
269 * Strings used in tracepoints need to be exported via the
270 * tracing system such that tools like perf and trace-cmd can
271 * translate the string address pointers to actual text.
273 #define TPS(x) tracepoint_string(x)
276 * Dump the ftrace buffer, but only one time per callsite per boot.
278 #define rcu_ftrace_dump(oops_dump_mode) \
280 static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \
282 if (!atomic_read(&___rfd_beenhere) && \
283 !atomic_xchg(&___rfd_beenhere, 1)) { \
285 rcu_ftrace_dump_stall_suppress(); \
286 ftrace_dump(oops_dump_mode); \
287 rcu_ftrace_dump_stall_unsuppress(); \
291 void rcu_early_boot_tests(void);
292 void rcu_test_sync_prims(void);
295 * This function really isn't for public consumption, but RCU is special in
296 * that context switches can allow the state machine to make progress.
298 extern void resched_cpu(int cpu);
300 #if defined(SRCU) || !defined(TINY_RCU)
302 #include <linux/rcu_node_tree.h>
304 extern int rcu_num_lvls;
305 extern int num_rcu_lvl[];
306 extern int rcu_num_nodes;
307 static bool rcu_fanout_exact;
308 static int rcu_fanout_leaf;
311 * Compute the per-level fanout, either using the exact fanout specified
312 * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
314 static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
318 if (rcu_fanout_exact) {
319 levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
320 for (i = rcu_num_lvls - 2; i >= 0; i--)
321 levelspread[i] = RCU_FANOUT;
327 for (i = rcu_num_lvls - 1; i >= 0; i--) {
329 levelspread[i] = (cprv + ccur - 1) / ccur;
335 /* Returns a pointer to the first leaf rcu_node structure. */
336 #define rcu_first_leaf_node() (rcu_state.level[rcu_num_lvls - 1])
338 /* Is this rcu_node a leaf? */
339 #define rcu_is_leaf_node(rnp) ((rnp)->level == rcu_num_lvls - 1)
341 /* Is this rcu_node the last leaf? */
342 #define rcu_is_last_leaf_node(rnp) ((rnp) == &rcu_state.node[rcu_num_nodes - 1])
345 * Do a full breadth-first scan of the {s,}rcu_node structures for the
346 * specified state structure (for SRCU) or the only rcu_state structure
349 #define srcu_for_each_node_breadth_first(sp, rnp) \
350 for ((rnp) = &(sp)->node[0]; \
351 (rnp) < &(sp)->node[rcu_num_nodes]; (rnp)++)
352 #define rcu_for_each_node_breadth_first(rnp) \
353 srcu_for_each_node_breadth_first(&rcu_state, rnp)
356 * Scan the leaves of the rcu_node hierarchy for the rcu_state structure.
357 * Note that if there is a singleton rcu_node tree with but one rcu_node
358 * structure, this loop -will- visit the rcu_node structure. It is still
359 * a leaf node, even if it is also the root node.
361 #define rcu_for_each_leaf_node(rnp) \
362 for ((rnp) = rcu_first_leaf_node(); \
363 (rnp) < &rcu_state.node[rcu_num_nodes]; (rnp)++)
366 * Iterate over all possible CPUs in a leaf RCU node.
368 #define for_each_leaf_node_possible_cpu(rnp, cpu) \
369 for ((cpu) = cpumask_next((rnp)->grplo - 1, cpu_possible_mask); \
370 (cpu) <= rnp->grphi; \
371 (cpu) = cpumask_next((cpu), cpu_possible_mask))
374 * Iterate over all CPUs in a leaf RCU node's specified mask.
376 #define rcu_find_next_bit(rnp, cpu, mask) \
377 ((rnp)->grplo + find_next_bit(&(mask), BITS_PER_LONG, (cpu)))
378 #define for_each_leaf_node_cpu_mask(rnp, cpu, mask) \
379 for ((cpu) = rcu_find_next_bit((rnp), 0, (mask)); \
380 (cpu) <= rnp->grphi; \
381 (cpu) = rcu_find_next_bit((rnp), (cpu) + 1 - (rnp->grplo), (mask)))
384 * Wrappers for the rcu_node::lock acquire and release.
386 * Because the rcu_nodes form a tree, the tree traversal locking will observe
387 * different lock values, this in turn means that an UNLOCK of one level
388 * followed by a LOCK of another level does not imply a full memory barrier;
389 * and most importantly transitivity is lost.
391 * In order to restore full ordering between tree levels, augment the regular
392 * lock acquire functions with smp_mb__after_unlock_lock().
394 * As ->lock of struct rcu_node is a __private field, therefore one should use
395 * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
397 #define raw_spin_lock_rcu_node(p) \
399 raw_spin_lock(&ACCESS_PRIVATE(p, lock)); \
400 smp_mb__after_unlock_lock(); \
403 #define raw_spin_unlock_rcu_node(p) raw_spin_unlock(&ACCESS_PRIVATE(p, lock))
405 #define raw_spin_lock_irq_rcu_node(p) \
407 raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock)); \
408 smp_mb__after_unlock_lock(); \
411 #define raw_spin_unlock_irq_rcu_node(p) \
412 raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock))
414 #define raw_spin_lock_irqsave_rcu_node(p, flags) \
416 raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags); \
417 smp_mb__after_unlock_lock(); \
420 #define raw_spin_unlock_irqrestore_rcu_node(p, flags) \
421 raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags)
423 #define raw_spin_trylock_rcu_node(p) \
425 bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock)); \
428 smp_mb__after_unlock_lock(); \
432 #define raw_lockdep_assert_held_rcu_node(p) \
433 lockdep_assert_held(&ACCESS_PRIVATE(p, lock))
435 #endif /* #if defined(SRCU) || !defined(TINY_RCU) */
438 void srcu_init(void);
439 #else /* #ifdef CONFIG_SRCU */
440 static inline void srcu_init(void) { }
441 #endif /* #else #ifdef CONFIG_SRCU */
443 #ifdef CONFIG_TINY_RCU
444 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
445 static inline bool rcu_gp_is_normal(void) { return true; }
446 static inline bool rcu_gp_is_expedited(void) { return false; }
447 static inline void rcu_expedite_gp(void) { }
448 static inline void rcu_unexpedite_gp(void) { }
449 static inline void rcu_request_urgent_qs_task(struct task_struct *t) { }
450 #else /* #ifdef CONFIG_TINY_RCU */
451 bool rcu_gp_is_normal(void); /* Internal RCU use. */
452 bool rcu_gp_is_expedited(void); /* Internal RCU use. */
453 void rcu_expedite_gp(void);
454 void rcu_unexpedite_gp(void);
455 void rcupdate_announce_bootup_oddness(void);
456 void rcu_request_urgent_qs_task(struct task_struct *t);
457 #endif /* #else #ifdef CONFIG_TINY_RCU */
459 #define RCU_SCHEDULER_INACTIVE 0
460 #define RCU_SCHEDULER_INIT 1
461 #define RCU_SCHEDULER_RUNNING 2
463 enum rcutorture_type {
472 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
473 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
474 unsigned long *gp_seq);
475 void rcutorture_record_progress(unsigned long vernum);
476 void do_trace_rcu_torture_read(const char *rcutorturename,
477 struct rcu_head *rhp,
482 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
483 int *flags, unsigned long *gp_seq)
488 static inline void rcutorture_record_progress(unsigned long vernum) { }
489 #ifdef CONFIG_RCU_TRACE
490 void do_trace_rcu_torture_read(const char *rcutorturename,
491 struct rcu_head *rhp,
496 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
501 #ifdef CONFIG_TINY_SRCU
503 static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
504 struct srcu_struct *sp, int *flags,
505 unsigned long *gp_seq)
507 if (test_type != SRCU_FLAVOR)
510 *gp_seq = sp->srcu_idx;
513 #elif defined(CONFIG_TREE_SRCU)
515 void srcutorture_get_gp_data(enum rcutorture_type test_type,
516 struct srcu_struct *sp, int *flags,
517 unsigned long *gp_seq);
521 #ifdef CONFIG_TINY_RCU
522 static inline unsigned long rcu_get_gp_seq(void) { return 0; }
523 static inline unsigned long rcu_exp_batches_completed(void) { return 0; }
524 static inline unsigned long
525 srcu_batches_completed(struct srcu_struct *sp) { return 0; }
526 static inline void rcu_force_quiescent_state(void) { }
527 static inline void show_rcu_gp_kthreads(void) { }
528 static inline int rcu_get_gp_kthreads_prio(void) { return 0; }
529 static inline void rcu_fwd_progress_check(unsigned long j) { }
530 #else /* #ifdef CONFIG_TINY_RCU */
531 unsigned long rcu_get_gp_seq(void);
532 unsigned long rcu_exp_batches_completed(void);
533 unsigned long srcu_batches_completed(struct srcu_struct *sp);
534 void show_rcu_gp_kthreads(void);
535 int rcu_get_gp_kthreads_prio(void);
536 void rcu_fwd_progress_check(unsigned long j);
537 void rcu_force_quiescent_state(void);
538 extern struct workqueue_struct *rcu_gp_wq;
539 extern struct workqueue_struct *rcu_par_gp_wq;
540 #endif /* #else #ifdef CONFIG_TINY_RCU */
542 #ifdef CONFIG_RCU_NOCB_CPU
543 bool rcu_is_nocb_cpu(int cpu);
544 void rcu_bind_current_to_nocb(void);
546 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
547 static inline void rcu_bind_current_to_nocb(void) { }
550 #endif /* __LINUX_RCU_H */