blob: 0be1fa2ea521def28f0ccd22fe8a2eb53f6bffb9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Read-Copy Update mechanism for mutual exclusion
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, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
Paul E. McKenney01c1c662008-01-25 21:08:24 +010018 * Copyright IBM Corporation, 2001
Linus Torvalds1da177e2005-04-16 15:20:36 -070019 *
20 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
21 * Manfred Spraul <manfred@colorfullife.com>
Paul E. McKenneya71fca52009-09-18 10:28:19 -070022 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070023 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
24 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
25 * Papers:
26 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
27 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
28 *
29 * For detailed explanation of Read-Copy Update mechanism see -
Paul E. McKenneya71fca52009-09-18 10:28:19 -070030 * http://lse.sourceforge.net/locking/rcupdate.html
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 *
32 */
33#include <linux/types.h>
34#include <linux/kernel.h>
35#include <linux/init.h>
36#include <linux/spinlock.h>
37#include <linux/smp.h>
38#include <linux/interrupt.h>
39#include <linux/sched.h>
Arun Sharma600634972011-07-26 16:09:06 -070040#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/bitops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/percpu.h>
43#include <linux/notifier.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <linux/cpu.h>
Ingo Molnar9331b312006-03-23 03:00:19 -080045#include <linux/mutex.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040046#include <linux/export.h>
Paul E. McKenneye3818b82010-03-15 17:03:43 -070047#include <linux/hardirq.h>
Paul E. McKenneye3ebfb92012-07-02 14:42:01 -070048#include <linux/delay.h>
Antti P Miettinen3705b882012-10-05 09:59:15 +030049#include <linux/module.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Paul E. McKenney29c00b42011-06-17 15:53:19 -070051#define CREATE_TRACE_POINTS
52#include <trace/events/rcu.h>
53
54#include "rcu.h"
55
Antti P Miettinen3705b882012-10-05 09:59:15 +030056module_param(rcu_expedited, int, 0);
57
Paul E. McKenney9dd8fb12012-04-13 12:54:22 -070058#ifdef CONFIG_PREEMPT_RCU
59
60/*
Paul E. McKenney2a3fa842012-05-21 11:58:36 -070061 * Preemptible RCU implementation for rcu_read_lock().
62 * Just increment ->rcu_read_lock_nesting, shared state will be updated
63 * if we block.
64 */
65void __rcu_read_lock(void)
66{
67 current->rcu_read_lock_nesting++;
68 barrier(); /* critical section after entry code. */
69}
70EXPORT_SYMBOL_GPL(__rcu_read_lock);
71
72/*
73 * Preemptible RCU implementation for rcu_read_unlock().
74 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
75 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
76 * invoke rcu_read_unlock_special() to clean up after a context switch
77 * in an RCU read-side critical section and other special cases.
78 */
79void __rcu_read_unlock(void)
80{
81 struct task_struct *t = current;
82
83 if (t->rcu_read_lock_nesting != 1) {
84 --t->rcu_read_lock_nesting;
85 } else {
86 barrier(); /* critical section before exit code. */
87 t->rcu_read_lock_nesting = INT_MIN;
Paul E. McKenneye3ebfb92012-07-02 14:42:01 -070088#ifdef CONFIG_PROVE_RCU_DELAY
89 udelay(10); /* Make preemption more probable. */
90#endif /* #ifdef CONFIG_PROVE_RCU_DELAY */
Paul E. McKenney2a3fa842012-05-21 11:58:36 -070091 barrier(); /* assign before ->rcu_read_unlock_special load */
92 if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
93 rcu_read_unlock_special(t);
94 barrier(); /* ->rcu_read_unlock_special load before assign */
95 t->rcu_read_lock_nesting = 0;
96 }
97#ifdef CONFIG_PROVE_LOCKING
98 {
99 int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting);
100
101 WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2);
102 }
103#endif /* #ifdef CONFIG_PROVE_LOCKING */
104}
105EXPORT_SYMBOL_GPL(__rcu_read_unlock);
106
Paul E. McKenney2439b692013-04-11 10:15:52 -0700107#endif /* #ifdef CONFIG_PREEMPT_RCU */
Paul E. McKenney9dd8fb12012-04-13 12:54:22 -0700108
Paul E. McKenney162cc272009-09-23 16:18:13 -0700109#ifdef CONFIG_DEBUG_LOCK_ALLOC
110static struct lock_class_key rcu_lock_key;
111struct lockdep_map rcu_lock_map =
112 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
113EXPORT_SYMBOL_GPL(rcu_lock_map);
Paul E. McKenney632ee202010-02-22 17:04:45 -0800114
115static struct lock_class_key rcu_bh_lock_key;
116struct lockdep_map rcu_bh_lock_map =
117 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_bh", &rcu_bh_lock_key);
118EXPORT_SYMBOL_GPL(rcu_bh_lock_map);
119
120static struct lock_class_key rcu_sched_lock_key;
121struct lockdep_map rcu_sched_lock_map =
122 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key);
123EXPORT_SYMBOL_GPL(rcu_sched_lock_map);
Paul E. McKenney162cc272009-09-23 16:18:13 -0700124#endif
125
Paul E. McKenneye3818b82010-03-15 17:03:43 -0700126#ifdef CONFIG_DEBUG_LOCK_ALLOC
127
Paul E. McKenneybc293d62010-04-15 12:50:39 -0700128int debug_lockdep_rcu_enabled(void)
129{
130 return rcu_scheduler_active && debug_locks &&
131 current->lockdep_recursion == 0;
132}
133EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
134
Paul E. McKenneye3818b82010-03-15 17:03:43 -0700135/**
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700136 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
Paul E. McKenneye3818b82010-03-15 17:03:43 -0700137 *
138 * Check for bottom half being disabled, which covers both the
139 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
140 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
Paul E. McKenneyca5ecdd2010-04-28 14:39:09 -0700141 * will show the situation. This is useful for debug checks in functions
142 * that require that they be called within an RCU read-side critical
143 * section.
Paul E. McKenneye3818b82010-03-15 17:03:43 -0700144 *
145 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
Paul E. McKenneyc0d6d012012-01-23 12:41:26 -0800146 *
147 * Note that rcu_read_lock() is disallowed if the CPU is either idle or
148 * offline from an RCU perspective, so check for those as well.
Paul E. McKenneye3818b82010-03-15 17:03:43 -0700149 */
150int rcu_read_lock_bh_held(void)
151{
152 if (!debug_lockdep_rcu_enabled())
153 return 1;
Frederic Weisbeckere6b80a32011-10-07 16:25:18 -0700154 if (rcu_is_cpu_idle())
155 return 0;
Paul E. McKenneyc0d6d012012-01-23 12:41:26 -0800156 if (!rcu_lockdep_current_cpu_online())
157 return 0;
Paul E. McKenney773e3f92010-10-05 14:03:02 -0700158 return in_softirq() || irqs_disabled();
Paul E. McKenneye3818b82010-03-15 17:03:43 -0700159}
160EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
161
162#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
163
Paul E. McKenney2c428182011-05-26 22:14:36 -0700164struct rcu_synchronize {
165 struct rcu_head head;
166 struct completion completion;
167};
168
Paul E. McKenneyd9f1bb62010-02-25 14:06:47 -0800169/*
Paul E. McKenneyfbf6bfc2008-02-13 15:03:15 -0800170 * Awaken the corresponding synchronize_rcu() instance now that a
171 * grace period has elapsed.
172 */
Paul E. McKenney2c428182011-05-26 22:14:36 -0700173static void wakeme_after_rcu(struct rcu_head *head)
Dipankar Sarma21a1ea92006-03-07 21:55:33 -0800174{
Paul E. McKenney01c1c662008-01-25 21:08:24 +0100175 struct rcu_synchronize *rcu;
176
177 rcu = container_of(head, struct rcu_synchronize, head);
178 complete(&rcu->completion);
Dipankar Sarma21a1ea92006-03-07 21:55:33 -0800179}
Paul E. McKenneyee84b822010-05-06 09:28:41 -0700180
Paul E. McKenney2c428182011-05-26 22:14:36 -0700181void wait_rcu_gp(call_rcu_func_t crf)
182{
183 struct rcu_synchronize rcu;
184
185 init_rcu_head_on_stack(&rcu.head);
186 init_completion(&rcu.completion);
187 /* Will wake me after RCU finished. */
188 crf(&rcu.head, wakeme_after_rcu);
189 /* Wait for it. */
190 wait_for_completion(&rcu.completion);
191 destroy_rcu_head_on_stack(&rcu.head);
192}
193EXPORT_SYMBOL_GPL(wait_rcu_gp);
194
Paul E. McKenneyee84b822010-05-06 09:28:41 -0700195#ifdef CONFIG_PROVE_RCU
196/*
197 * wrapper function to avoid #include problems.
198 */
199int rcu_my_thread_group_empty(void)
200{
201 return thread_group_empty(current);
202}
203EXPORT_SYMBOL_GPL(rcu_my_thread_group_empty);
204#endif /* #ifdef CONFIG_PROVE_RCU */
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400205
206#ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD
207static inline void debug_init_rcu_head(struct rcu_head *head)
208{
209 debug_object_init(head, &rcuhead_debug_descr);
210}
211
212static inline void debug_rcu_head_free(struct rcu_head *head)
213{
214 debug_object_free(head, &rcuhead_debug_descr);
215}
216
217/*
218 * fixup_init is called when:
219 * - an active object is initialized
220 */
221static int rcuhead_fixup_init(void *addr, enum debug_obj_state state)
222{
223 struct rcu_head *head = addr;
224
225 switch (state) {
226 case ODEBUG_STATE_ACTIVE:
227 /*
228 * Ensure that queued callbacks are all executed.
229 * If we detect that we are nested in a RCU read-side critical
230 * section, we should simply fail, otherwise we would deadlock.
Mathieu Desnoyersfc2ecf72011-02-23 09:42:14 -0800231 * In !PREEMPT configurations, there is no way to tell if we are
232 * in a RCU read-side critical section or not, so we never
233 * attempt any fixup and just print a warning.
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400234 */
Mathieu Desnoyersfc2ecf72011-02-23 09:42:14 -0800235#ifndef CONFIG_PREEMPT
Paul E. McKenney108aae22011-02-23 09:56:00 -0800236 WARN_ON_ONCE(1);
Mathieu Desnoyersfc2ecf72011-02-23 09:42:14 -0800237 return 0;
238#endif
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400239 if (rcu_preempt_depth() != 0 || preempt_count() != 0 ||
240 irqs_disabled()) {
Paul E. McKenney108aae22011-02-23 09:56:00 -0800241 WARN_ON_ONCE(1);
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400242 return 0;
243 }
244 rcu_barrier();
245 rcu_barrier_sched();
246 rcu_barrier_bh();
247 debug_object_init(head, &rcuhead_debug_descr);
248 return 1;
249 default:
250 return 0;
251 }
252}
253
254/*
255 * fixup_activate is called when:
256 * - an active object is activated
257 * - an unknown object is activated (might be a statically initialized object)
258 * Activation is performed internally by call_rcu().
259 */
260static int rcuhead_fixup_activate(void *addr, enum debug_obj_state state)
261{
262 struct rcu_head *head = addr;
263
264 switch (state) {
265
266 case ODEBUG_STATE_NOTAVAILABLE:
267 /*
268 * This is not really a fixup. We just make sure that it is
269 * tracked in the object tracker.
270 */
271 debug_object_init(head, &rcuhead_debug_descr);
272 debug_object_activate(head, &rcuhead_debug_descr);
273 return 0;
274
275 case ODEBUG_STATE_ACTIVE:
276 /*
277 * Ensure that queued callbacks are all executed.
278 * If we detect that we are nested in a RCU read-side critical
279 * section, we should simply fail, otherwise we would deadlock.
Mathieu Desnoyersfc2ecf72011-02-23 09:42:14 -0800280 * In !PREEMPT configurations, there is no way to tell if we are
281 * in a RCU read-side critical section or not, so we never
282 * attempt any fixup and just print a warning.
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400283 */
Mathieu Desnoyersfc2ecf72011-02-23 09:42:14 -0800284#ifndef CONFIG_PREEMPT
Paul E. McKenney108aae22011-02-23 09:56:00 -0800285 WARN_ON_ONCE(1);
Mathieu Desnoyersfc2ecf72011-02-23 09:42:14 -0800286 return 0;
287#endif
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400288 if (rcu_preempt_depth() != 0 || preempt_count() != 0 ||
289 irqs_disabled()) {
Paul E. McKenney108aae22011-02-23 09:56:00 -0800290 WARN_ON_ONCE(1);
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400291 return 0;
292 }
293 rcu_barrier();
294 rcu_barrier_sched();
295 rcu_barrier_bh();
296 debug_object_activate(head, &rcuhead_debug_descr);
297 return 1;
298 default:
299 return 0;
300 }
301}
302
303/*
304 * fixup_free is called when:
305 * - an active object is freed
306 */
307static int rcuhead_fixup_free(void *addr, enum debug_obj_state state)
308{
309 struct rcu_head *head = addr;
310
311 switch (state) {
312 case ODEBUG_STATE_ACTIVE:
313 /*
314 * Ensure that queued callbacks are all executed.
315 * If we detect that we are nested in a RCU read-side critical
316 * section, we should simply fail, otherwise we would deadlock.
Mathieu Desnoyersfc2ecf72011-02-23 09:42:14 -0800317 * In !PREEMPT configurations, there is no way to tell if we are
318 * in a RCU read-side critical section or not, so we never
319 * attempt any fixup and just print a warning.
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400320 */
Mathieu Desnoyersfc2ecf72011-02-23 09:42:14 -0800321#ifndef CONFIG_PREEMPT
Paul E. McKenney108aae22011-02-23 09:56:00 -0800322 WARN_ON_ONCE(1);
Mathieu Desnoyersfc2ecf72011-02-23 09:42:14 -0800323 return 0;
324#endif
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400325 if (rcu_preempt_depth() != 0 || preempt_count() != 0 ||
326 irqs_disabled()) {
Paul E. McKenney108aae22011-02-23 09:56:00 -0800327 WARN_ON_ONCE(1);
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400328 return 0;
329 }
330 rcu_barrier();
331 rcu_barrier_sched();
332 rcu_barrier_bh();
333 debug_object_free(head, &rcuhead_debug_descr);
334 return 1;
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400335 default:
336 return 0;
337 }
338}
339
340/**
341 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects
342 * @head: pointer to rcu_head structure to be initialized
343 *
344 * This function informs debugobjects of a new rcu_head structure that
345 * has been allocated as an auto variable on the stack. This function
346 * is not required for rcu_head structures that are statically defined or
347 * that are dynamically allocated on the heap. This function has no
348 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
349 */
350void init_rcu_head_on_stack(struct rcu_head *head)
351{
352 debug_object_init_on_stack(head, &rcuhead_debug_descr);
353}
354EXPORT_SYMBOL_GPL(init_rcu_head_on_stack);
355
356/**
357 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects
358 * @head: pointer to rcu_head structure to be initialized
359 *
360 * This function informs debugobjects that an on-stack rcu_head structure
361 * is about to go out of scope. As with init_rcu_head_on_stack(), this
362 * function is not required for rcu_head structures that are statically
363 * defined or that are dynamically allocated on the heap. Also as with
364 * init_rcu_head_on_stack(), this function has no effect for
365 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds.
366 */
367void destroy_rcu_head_on_stack(struct rcu_head *head)
368{
369 debug_object_free(head, &rcuhead_debug_descr);
370}
371EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack);
372
373struct debug_obj_descr rcuhead_debug_descr = {
374 .name = "rcu_head",
375 .fixup_init = rcuhead_fixup_init,
376 .fixup_activate = rcuhead_fixup_activate,
377 .fixup_free = rcuhead_fixup_free,
378};
379EXPORT_SYMBOL_GPL(rcuhead_debug_descr);
380#endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */
Paul E. McKenney91afaf32011-10-02 07:44:32 -0700381
382#if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) || defined(CONFIG_RCU_TRACE)
Paul E. McKenney52494532012-11-14 16:26:40 -0800383void do_trace_rcu_torture_read(char *rcutorturename, struct rcu_head *rhp,
384 unsigned long secs,
385 unsigned long c_old, unsigned long c)
Paul E. McKenney91afaf32011-10-02 07:44:32 -0700386{
Paul E. McKenney52494532012-11-14 16:26:40 -0800387 trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c);
Paul E. McKenney91afaf32011-10-02 07:44:32 -0700388}
389EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read);
390#else
Paul E. McKenney52494532012-11-14 16:26:40 -0800391#define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \
392 do { } while (0)
Paul E. McKenney91afaf32011-10-02 07:44:32 -0700393#endif
Paul E. McKenney6bfc09e2012-10-19 12:49:17 -0700394
395#ifdef CONFIG_RCU_STALL_COMMON
396
397#ifdef CONFIG_PROVE_RCU
398#define RCU_STALL_DELAY_DELTA (5 * HZ)
399#else
400#define RCU_STALL_DELAY_DELTA 0
401#endif
402
403int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */
404int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
405
406module_param(rcu_cpu_stall_suppress, int, 0644);
407module_param(rcu_cpu_stall_timeout, int, 0644);
408
409int rcu_jiffies_till_stall_check(void)
410{
411 int till_stall_check = ACCESS_ONCE(rcu_cpu_stall_timeout);
412
413 /*
414 * Limit check must be consistent with the Kconfig limits
415 * for CONFIG_RCU_CPU_STALL_TIMEOUT.
416 */
417 if (till_stall_check < 3) {
418 ACCESS_ONCE(rcu_cpu_stall_timeout) = 3;
419 till_stall_check = 3;
420 } else if (till_stall_check > 300) {
421 ACCESS_ONCE(rcu_cpu_stall_timeout) = 300;
422 till_stall_check = 300;
423 }
424 return till_stall_check * HZ + RCU_STALL_DELAY_DELTA;
425}
426
427static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
428{
429 rcu_cpu_stall_suppress = 1;
430 return NOTIFY_DONE;
431}
432
433static struct notifier_block rcu_panic_block = {
434 .notifier_call = rcu_panic,
435};
436
437static int __init check_cpu_stall_init(void)
438{
439 atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block);
440 return 0;
441}
442early_initcall(check_cpu_stall_init);
443
444#endif /* #ifdef CONFIG_RCU_STALL_COMMON */