xref: /linux/kernel/rcu/rcu.h (revision 856e7c4b619af622d56b3b454f7bec32a170ac99)
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  */
48 static 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  */
57 static 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  */
66 static 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. */
73 static 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. */
81 static 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. */
87 static 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. */
95 static 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. */
105 static 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  */
114 static 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 
129 extern struct debug_obj_descr rcuhead_debug_descr;
130 
131 static 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 
142 static 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 */
150 static inline int debug_rcu_head_queue(struct rcu_head *head)
151 {
152 	return 0;
153 }
154 
155 static inline void debug_rcu_head_unqueue(struct rcu_head *head)
156 {
157 }
158 #endif	/* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */
159 
160 void 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  */
166 static 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 
186 extern int rcu_cpu_stall_suppress;
187 int rcu_jiffies_till_stall_check(void);
188 
189 #define rcu_ftrace_dump_stall_suppress() \
190 do { \
191 	if (!rcu_cpu_stall_suppress) \
192 		rcu_cpu_stall_suppress = 3; \
193 } while (0)
194 
195 #define rcu_ftrace_dump_stall_unsuppress() \
196 do { \
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) \
217 do { \
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 
229 void rcu_early_boot_tests(void);
230 void 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  */
236 extern void resched_cpu(int cpu);
237 
238 #if defined(SRCU) || !defined(TINY_RCU)
239 
240 #include <linux/rcu_node_tree.h>
241 
242 extern int rcu_num_lvls;
243 extern int num_rcu_lvl[];
244 extern int rcu_num_nodes;
245 static bool rcu_fanout_exact;
246 static 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  */
252 static 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 /* Returns first leaf rcu_node of the specified RCU flavor. */
274 #define rcu_first_leaf_node(rsp) ((rsp)->level[rcu_num_lvls - 1])
275 
276 /* Is this rcu_node a leaf? */
277 #define rcu_is_leaf_node(rnp) ((rnp)->level == rcu_num_lvls - 1)
278 
279 /*
280  * Do a full breadth-first scan of the rcu_node structures for the
281  * specified rcu_state structure.
282  */
283 #define rcu_for_each_node_breadth_first(rsp, rnp) \
284 	for ((rnp) = &(rsp)->node[0]; \
285 	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
286 
287 /*
288  * Do a breadth-first scan of the non-leaf rcu_node structures for the
289  * specified rcu_state structure.  Note that if there is a singleton
290  * rcu_node tree with but one rcu_node structure, this loop is a no-op.
291  */
292 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
293 	for ((rnp) = &(rsp)->node[0]; !rcu_is_leaf_node(rsp, rnp); (rnp)++)
294 
295 /*
296  * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
297  * structure.  Note that if there is a singleton rcu_node tree with but
298  * one rcu_node structure, this loop -will- visit the rcu_node structure.
299  * It is still a leaf node, even if it is also the root node.
300  */
301 #define rcu_for_each_leaf_node(rsp, rnp) \
302 	for ((rnp) = rcu_first_leaf_node(rsp); \
303 	     (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
304 
305 /*
306  * Iterate over all possible CPUs in a leaf RCU node.
307  */
308 #define for_each_leaf_node_possible_cpu(rnp, cpu) \
309 	for ((cpu) = cpumask_next((rnp)->grplo - 1, cpu_possible_mask); \
310 	     (cpu) <= rnp->grphi; \
311 	     (cpu) = cpumask_next((cpu), cpu_possible_mask))
312 
313 /*
314  * Iterate over all CPUs in a leaf RCU node's specified mask.
315  */
316 #define rcu_find_next_bit(rnp, cpu, mask) \
317 	((rnp)->grplo + find_next_bit(&(mask), BITS_PER_LONG, (cpu)))
318 #define for_each_leaf_node_cpu_mask(rnp, cpu, mask) \
319 	for ((cpu) = rcu_find_next_bit((rnp), 0, (mask)); \
320 	     (cpu) <= rnp->grphi; \
321 	     (cpu) = rcu_find_next_bit((rnp), (cpu) + 1 - (rnp->grplo), (mask)))
322 
323 /*
324  * Wrappers for the rcu_node::lock acquire and release.
325  *
326  * Because the rcu_nodes form a tree, the tree traversal locking will observe
327  * different lock values, this in turn means that an UNLOCK of one level
328  * followed by a LOCK of another level does not imply a full memory barrier;
329  * and most importantly transitivity is lost.
330  *
331  * In order to restore full ordering between tree levels, augment the regular
332  * lock acquire functions with smp_mb__after_unlock_lock().
333  *
334  * As ->lock of struct rcu_node is a __private field, therefore one should use
335  * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock.
336  */
337 #define raw_spin_lock_rcu_node(p)					\
338 do {									\
339 	raw_spin_lock(&ACCESS_PRIVATE(p, lock));			\
340 	smp_mb__after_unlock_lock();					\
341 } while (0)
342 
343 #define raw_spin_unlock_rcu_node(p) raw_spin_unlock(&ACCESS_PRIVATE(p, lock))
344 
345 #define raw_spin_lock_irq_rcu_node(p)					\
346 do {									\
347 	raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock));			\
348 	smp_mb__after_unlock_lock();					\
349 } while (0)
350 
351 #define raw_spin_unlock_irq_rcu_node(p)					\
352 	raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock))
353 
354 #define raw_spin_lock_irqsave_rcu_node(p, flags)			\
355 do {									\
356 	raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags);	\
357 	smp_mb__after_unlock_lock();					\
358 } while (0)
359 
360 #define raw_spin_unlock_irqrestore_rcu_node(p, flags)			\
361 	raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags)
362 
363 #define raw_spin_trylock_rcu_node(p)					\
364 ({									\
365 	bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock));	\
366 									\
367 	if (___locked)							\
368 		smp_mb__after_unlock_lock();				\
369 	___locked;							\
370 })
371 
372 #define raw_lockdep_assert_held_rcu_node(p)				\
373 	lockdep_assert_held(&ACCESS_PRIVATE(p, lock))
374 
375 #endif /* #if defined(SRCU) || !defined(TINY_RCU) */
376 
377 #ifdef CONFIG_TINY_RCU
378 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
379 static inline bool rcu_gp_is_normal(void) { return true; }
380 static inline bool rcu_gp_is_expedited(void) { return false; }
381 static inline void rcu_expedite_gp(void) { }
382 static inline void rcu_unexpedite_gp(void) { }
383 static inline void rcu_request_urgent_qs_task(struct task_struct *t) { }
384 #else /* #ifdef CONFIG_TINY_RCU */
385 bool rcu_gp_is_normal(void);     /* Internal RCU use. */
386 bool rcu_gp_is_expedited(void);  /* Internal RCU use. */
387 void rcu_expedite_gp(void);
388 void rcu_unexpedite_gp(void);
389 void rcupdate_announce_bootup_oddness(void);
390 void rcu_request_urgent_qs_task(struct task_struct *t);
391 #endif /* #else #ifdef CONFIG_TINY_RCU */
392 
393 #define RCU_SCHEDULER_INACTIVE	0
394 #define RCU_SCHEDULER_INIT	1
395 #define RCU_SCHEDULER_RUNNING	2
396 
397 enum rcutorture_type {
398 	RCU_FLAVOR,
399 	RCU_BH_FLAVOR,
400 	RCU_SCHED_FLAVOR,
401 	RCU_TASKS_FLAVOR,
402 	SRCU_FLAVOR,
403 	INVALID_RCU_FLAVOR
404 };
405 
406 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_PREEMPT_RCU)
407 void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
408 			    unsigned long *gpnum, unsigned long *completed);
409 void rcutorture_record_test_transition(void);
410 void rcutorture_record_progress(unsigned long vernum);
411 void do_trace_rcu_torture_read(const char *rcutorturename,
412 			       struct rcu_head *rhp,
413 			       unsigned long secs,
414 			       unsigned long c_old,
415 			       unsigned long c);
416 #else
417 static inline void rcutorture_get_gp_data(enum rcutorture_type test_type,
418 					  int *flags,
419 					  unsigned long *gpnum,
420 					  unsigned long *completed)
421 {
422 	*flags = 0;
423 	*gpnum = 0;
424 	*completed = 0;
425 }
426 static inline void rcutorture_record_test_transition(void) { }
427 static inline void rcutorture_record_progress(unsigned long vernum) { }
428 #ifdef CONFIG_RCU_TRACE
429 void do_trace_rcu_torture_read(const char *rcutorturename,
430 			       struct rcu_head *rhp,
431 			       unsigned long secs,
432 			       unsigned long c_old,
433 			       unsigned long c);
434 #else
435 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
436 	do { } while (0)
437 #endif
438 #endif
439 
440 #ifdef CONFIG_TINY_SRCU
441 
442 static inline void srcutorture_get_gp_data(enum rcutorture_type test_type,
443 					   struct srcu_struct *sp, int *flags,
444 					   unsigned long *gpnum,
445 					   unsigned long *completed)
446 {
447 	if (test_type != SRCU_FLAVOR)
448 		return;
449 	*flags = 0;
450 	*completed = sp->srcu_idx;
451 	*gpnum = *completed;
452 }
453 
454 #elif defined(CONFIG_TREE_SRCU)
455 
456 void srcutorture_get_gp_data(enum rcutorture_type test_type,
457 			     struct srcu_struct *sp, int *flags,
458 			     unsigned long *gpnum, unsigned long *completed);
459 
460 #endif
461 
462 #ifdef CONFIG_TINY_RCU
463 static inline unsigned long rcu_batches_started(void) { return 0; }
464 static inline unsigned long rcu_batches_started_bh(void) { return 0; }
465 static inline unsigned long rcu_batches_started_sched(void) { return 0; }
466 static inline unsigned long rcu_batches_completed(void) { return 0; }
467 static inline unsigned long rcu_batches_completed_bh(void) { return 0; }
468 static inline unsigned long rcu_batches_completed_sched(void) { return 0; }
469 static inline unsigned long rcu_exp_batches_completed(void) { return 0; }
470 static inline unsigned long rcu_exp_batches_completed_sched(void) { return 0; }
471 static inline unsigned long
472 srcu_batches_completed(struct srcu_struct *sp) { return 0; }
473 static inline void rcu_force_quiescent_state(void) { }
474 static inline void rcu_bh_force_quiescent_state(void) { }
475 static inline void rcu_sched_force_quiescent_state(void) { }
476 static inline void show_rcu_gp_kthreads(void) { }
477 #else /* #ifdef CONFIG_TINY_RCU */
478 extern unsigned long rcutorture_testseq;
479 extern unsigned long rcutorture_vernum;
480 unsigned long rcu_batches_started(void);
481 unsigned long rcu_batches_started_bh(void);
482 unsigned long rcu_batches_started_sched(void);
483 unsigned long rcu_batches_completed(void);
484 unsigned long rcu_batches_completed_bh(void);
485 unsigned long rcu_batches_completed_sched(void);
486 unsigned long rcu_exp_batches_completed(void);
487 unsigned long rcu_exp_batches_completed_sched(void);
488 unsigned long srcu_batches_completed(struct srcu_struct *sp);
489 void show_rcu_gp_kthreads(void);
490 void rcu_force_quiescent_state(void);
491 void rcu_bh_force_quiescent_state(void);
492 void rcu_sched_force_quiescent_state(void);
493 extern struct workqueue_struct *rcu_gp_wq;
494 extern struct workqueue_struct *rcu_par_gp_wq;
495 #endif /* #else #ifdef CONFIG_TINY_RCU */
496 
497 #ifdef CONFIG_RCU_NOCB_CPU
498 bool rcu_is_nocb_cpu(int cpu);
499 #else
500 static inline bool rcu_is_nocb_cpu(int cpu) { return false; }
501 #endif
502 
503 #endif /* __LINUX_RCU_H */
504