blob: fec804b7908032d4416ea9ff3531eae154bba8f9 [file] [log] [blame]
Paul E. McKenney00de9d72019-01-17 10:21:12 -08001// SPDX-License-Identifier: GPL-2.0+
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -07002/*
3 * Read-Copy Update mechanism for mutual exclusion, the Bloatwatch edition.
4 *
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -07005 * Copyright IBM Corporation, 2008
6 *
Paul E. McKenney00de9d72019-01-17 10:21:12 -08007 * Author: Paul E. McKenney <paulmck@linux.ibm.com>
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -07008 *
9 * For detailed explanation of Read-Copy Update mechanism see -
Ingo Molnar4ce5b902009-10-26 07:55:55 +010010 * Documentation/RCU
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070011 */
Ingo Molnar4ce5b902009-10-26 07:55:55 +010012#include <linux/completion.h>
13#include <linux/interrupt.h>
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070014#include <linux/notifier.h>
Ingo Molnarf9411eb2017-02-06 09:50:49 +010015#include <linux/rcupdate_wait.h>
Ingo Molnar4ce5b902009-10-26 07:55:55 +010016#include <linux/kernel.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040017#include <linux/export.h>
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070018#include <linux/mutex.h>
Ingo Molnar4ce5b902009-10-26 07:55:55 +010019#include <linux/sched.h>
20#include <linux/types.h>
21#include <linux/init.h>
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070022#include <linux/time.h>
Ingo Molnar4ce5b902009-10-26 07:55:55 +010023#include <linux/cpu.h>
Linus Torvalds268bb0c2011-05-20 12:50:29 -070024#include <linux/prefetch.h>
Joel Fernandes (Google)77a40f92019-08-30 12:36:32 -040025#include <linux/slab.h>
Uladzislau Rezki (Sony)64d1d062020-05-25 23:47:54 +020026#include <linux/mm.h>
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070027
Paul E. McKenney29c00b42011-06-17 15:53:19 -070028#include "rcu.h"
29
Paul E. McKenney6d481522017-05-17 10:54:29 -070030/* Global control variables for rcupdate callback mechanism. */
31struct rcu_ctrlblk {
32 struct rcu_head *rcucblist; /* List of pending callbacks (CBs). */
33 struct rcu_head **donetail; /* ->next pointer of last "done" CB. */
34 struct rcu_head **curtail; /* ->next pointer of last CB. */
Paul E. McKenney0909fc22021-02-25 17:36:06 -080035 unsigned long gp_seq; /* Grace-period counter. */
Paul E. McKenney6d481522017-05-17 10:54:29 -070036};
37
38/* Definition for rcupdate control block. */
Paul E. McKenney709fdce2018-07-03 10:44:44 -070039static struct rcu_ctrlblk rcu_ctrlblk = {
40 .donetail = &rcu_ctrlblk.rcucblist,
41 .curtail = &rcu_ctrlblk.rcucblist,
Paul E. McKenney0909fc22021-02-25 17:36:06 -080042 .gp_seq = 0 - 300UL,
Paul E. McKenney6d481522017-05-17 10:54:29 -070043};
44
Paul E. McKenney709fdce2018-07-03 10:44:44 -070045void rcu_barrier(void)
Ingo Molnarf9411eb2017-02-06 09:50:49 +010046{
Joel Fernandes (Google)3cb278e72022-10-16 16:22:54 +000047 wait_rcu_gp(call_rcu_hurry);
Ingo Molnarf9411eb2017-02-06 09:50:49 +010048}
Paul E. McKenney709fdce2018-07-03 10:44:44 -070049EXPORT_SYMBOL(rcu_barrier);
Ingo Molnarf9411eb2017-02-06 09:50:49 +010050
Paul E. McKenney65cfe352018-07-01 07:40:52 -070051/* Record an rcu quiescent state. */
Paul E. McKenney709fdce2018-07-03 10:44:44 -070052void rcu_qs(void)
Ingo Molnarf9411eb2017-02-06 09:50:49 +010053{
Eric Dumazetb554d7de2011-04-28 07:23:45 +020054 unsigned long flags;
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070055
Eric Dumazetb554d7de2011-04-28 07:23:45 +020056 local_irq_save(flags);
Paul E. McKenney709fdce2018-07-03 10:44:44 -070057 if (rcu_ctrlblk.donetail != rcu_ctrlblk.curtail) {
58 rcu_ctrlblk.donetail = rcu_ctrlblk.curtail;
Cyrill Gorcunov18d7e402019-01-24 21:14:37 +030059 raise_softirq_irqoff(RCU_SOFTIRQ);
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070060 }
Paul E. McKenney414c1232022-04-13 15:17:25 -070061 WRITE_ONCE(rcu_ctrlblk.gp_seq, rcu_ctrlblk.gp_seq + 2);
Eric Dumazetb554d7de2011-04-28 07:23:45 +020062 local_irq_restore(flags);
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070063}
64
65/*
66 * Check to see if the scheduling-clock interrupt came from an extended
Paul E. McKenney9b2e4f12011-09-30 12:10:22 -070067 * quiescent state, and, if so, tell RCU about it. This function must
68 * be called from hardirq context. It is normally called from the
69 * scheduling-clock interrupt.
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070070 */
Paul E. McKenneyc98cac62018-11-21 11:35:03 -080071void rcu_sched_clock_irq(int user)
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070072{
Paul E. McKenneyc5bacd92018-07-20 14:18:23 -070073 if (user) {
Paul E. McKenney709fdce2018-07-03 10:44:44 -070074 rcu_qs();
Paul E. McKenneyc5bacd92018-07-20 14:18:23 -070075 } else if (rcu_ctrlblk.donetail != rcu_ctrlblk.curtail) {
76 set_tsk_need_resched(current);
77 set_preempt_need_resched();
78 }
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -070079}
80
Joel Fernandes (Google)77a40f92019-08-30 12:36:32 -040081/*
82 * Reclaim the specified callback, either by invoking it for non-kfree cases or
83 * freeing it directly (for kfree). Return true if kfreeing, false otherwise.
84 */
85static inline bool rcu_reclaim_tiny(struct rcu_head *head)
86{
87 rcu_callback_t f;
88 unsigned long offset = (unsigned long)head->func;
89
90 rcu_lock_acquire(&rcu_callback_map);
Uladzislau Rezki (Sony)c408b212020-05-25 23:47:55 +020091 if (__is_kvfree_rcu_offset(offset)) {
92 trace_rcu_invoke_kvfree_callback("", head, offset);
Uladzislau Rezki (Sony)64d1d062020-05-25 23:47:54 +020093 kvfree((void *)head - offset);
Joel Fernandes (Google)77a40f92019-08-30 12:36:32 -040094 rcu_lock_release(&rcu_callback_map);
95 return true;
96 }
97
98 trace_rcu_invoke_callback("", head);
99 f = head->func;
Zhen Lei2cbc4822023-08-05 11:17:26 +0800100 debug_rcu_head_callback(head);
Joel Fernandes (Google)77a40f92019-08-30 12:36:32 -0400101 WRITE_ONCE(head->func, (rcu_callback_t)0L);
102 f(head);
103 rcu_lock_release(&rcu_callback_map);
104 return false;
105}
106
Paul E. McKenney65cfe352018-07-01 07:40:52 -0700107/* Invoke the RCU callbacks whose grace period has elapsed. */
108static __latent_entropy void rcu_process_callbacks(struct softirq_action *unused)
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700109{
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700110 struct rcu_head *next, *list;
Ingo Molnar4ce5b902009-10-26 07:55:55 +0100111 unsigned long flags;
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700112
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700113 /* Move the ready-to-invoke callbacks to a local list. */
114 local_irq_save(flags);
Paul E. McKenney709fdce2018-07-03 10:44:44 -0700115 if (rcu_ctrlblk.donetail == &rcu_ctrlblk.rcucblist) {
Paul E. McKenney6e91f8c2015-05-11 11:13:05 -0700116 /* No callbacks ready, so just leave. */
117 local_irq_restore(flags);
118 return;
119 }
Paul E. McKenney709fdce2018-07-03 10:44:44 -0700120 list = rcu_ctrlblk.rcucblist;
121 rcu_ctrlblk.rcucblist = *rcu_ctrlblk.donetail;
122 *rcu_ctrlblk.donetail = NULL;
123 if (rcu_ctrlblk.curtail == rcu_ctrlblk.donetail)
124 rcu_ctrlblk.curtail = &rcu_ctrlblk.rcucblist;
125 rcu_ctrlblk.donetail = &rcu_ctrlblk.rcucblist;
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700126 local_irq_restore(flags);
127
128 /* Invoke the callbacks on the local list. */
129 while (list) {
130 next = list->next;
131 prefetch(next);
Mathieu Desnoyers551d55a2010-04-17 08:48:42 -0400132 debug_rcu_head_unqueue(list);
Paul E. McKenneyb2c07102010-09-09 13:40:39 -0700133 local_bh_disable();
Joel Fernandes (Google)77a40f92019-08-30 12:36:32 -0400134 rcu_reclaim_tiny(list);
Paul E. McKenneyb2c07102010-09-09 13:40:39 -0700135 local_bh_enable();
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700136 list = next;
137 }
138}
139
Paul E. McKenneyb2c07102010-09-09 13:40:39 -0700140/*
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700141 * Wait for a grace period to elapse. But it is illegal to invoke
Paul E. McKenney679d3f32018-07-07 18:12:26 -0700142 * synchronize_rcu() from within an RCU read-side critical section.
Paul E. McKenney7f453532022-04-14 11:49:58 -0700143 * Therefore, any legal call to synchronize_rcu() is a quiescent state,
144 * and so on a UP system, synchronize_rcu() need do nothing, other than
145 * let the polled APIs know that another grace period elapsed.
146 *
Paul E. McKenney65cfe352018-07-01 07:40:52 -0700147 * (But Lai Jiangshan points out the benefits of doing might_sleep()
148 * to reduce latency.)
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700149 *
150 * Cool, huh? (Due to Josh Triplett.)
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700151 */
Paul E. McKenney709fdce2018-07-03 10:44:44 -0700152void synchronize_rcu(void)
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700153{
Paul E. McKenneyf78f5b92015-06-18 15:50:02 -0700154 RCU_LOCKDEP_WARN(lock_is_held(&rcu_bh_lock_map) ||
155 lock_is_held(&rcu_lock_map) ||
156 lock_is_held(&rcu_sched_lock_map),
Paul E. McKenney679d3f32018-07-07 18:12:26 -0700157 "Illegal synchronize_rcu() in RCU read-side critical section");
Paul E. McKenney7f453532022-04-14 11:49:58 -0700158 WRITE_ONCE(rcu_ctrlblk.gp_seq, rcu_ctrlblk.gp_seq + 2);
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700159}
Paul E. McKenney709fdce2018-07-03 10:44:44 -0700160EXPORT_SYMBOL_GPL(synchronize_rcu);
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700161
Zqiang6ca02922022-07-01 10:44:04 +0800162static void tiny_rcu_leak_callback(struct rcu_head *rhp)
163{
164}
165
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700166/*
Paul E. McKenney679d3f32018-07-07 18:12:26 -0700167 * Post an RCU callback to be invoked after the end of an RCU grace
Paul E. McKenney65cfe352018-07-01 07:40:52 -0700168 * period. But since we have but one CPU, that would be after any
169 * quiescent state.
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700170 */
Paul E. McKenney709fdce2018-07-03 10:44:44 -0700171void call_rcu(struct rcu_head *head, rcu_callback_t func)
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700172{
Zqiang6ca02922022-07-01 10:44:04 +0800173 static atomic_t doublefrees;
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700174 unsigned long flags;
175
Zqiang6ca02922022-07-01 10:44:04 +0800176 if (debug_rcu_head_queue(head)) {
177 if (atomic_inc_return(&doublefrees) < 4) {
178 pr_err("%s(): Double-freed CB %p->%pS()!!! ", __func__, head, head->func);
179 mem_dump_obj(head);
180 }
181
182 if (!__is_kvfree_rcu_offset((unsigned long)head->func))
183 WRITE_ONCE(head->func, tiny_rcu_leak_callback);
184 return;
185 }
186
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700187 head->func = func;
188 head->next = NULL;
Ingo Molnar4ce5b902009-10-26 07:55:55 +0100189
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700190 local_irq_save(flags);
Paul E. McKenney709fdce2018-07-03 10:44:44 -0700191 *rcu_ctrlblk.curtail = head;
192 rcu_ctrlblk.curtail = &head->next;
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700193 local_irq_restore(flags);
Lai Jiangshan5f6130f2014-12-09 17:53:34 +0800194
195 if (unlikely(is_idle_task(current))) {
Paul E. McKenney709fdce2018-07-03 10:44:44 -0700196 /* force scheduling for rcu_qs() */
Lai Jiangshan5f6130f2014-12-09 17:53:34 +0800197 resched_cpu(0);
198 }
Paul E. McKenney9b1d82f2009-10-25 19:03:50 -0700199}
Paul E. McKenney709fdce2018-07-03 10:44:44 -0700200EXPORT_SYMBOL_GPL(call_rcu);
Paul E. McKenney9dc5ad32013-03-27 10:11:15 -0700201
Paul E. McKenney0909fc22021-02-25 17:36:06 -0800202/*
Paul E. McKenney91a967f2022-07-28 15:37:05 -0700203 * Store a grace-period-counter "cookie". For more information,
204 * see the Tree RCU header comment.
205 */
206void get_completed_synchronize_rcu_full(struct rcu_gp_oldstate *rgosp)
207{
208 rgosp->rgos_norm = RCU_GET_STATE_COMPLETED;
209}
210EXPORT_SYMBOL_GPL(get_completed_synchronize_rcu_full);
211
212/*
Paul E. McKenney0909fc22021-02-25 17:36:06 -0800213 * Return a grace-period-counter "cookie". For more information,
214 * see the Tree RCU header comment.
215 */
216unsigned long get_state_synchronize_rcu(void)
217{
218 return READ_ONCE(rcu_ctrlblk.gp_seq);
219}
220EXPORT_SYMBOL_GPL(get_state_synchronize_rcu);
221
222/*
223 * Return a grace-period-counter "cookie" and ensure that a future grace
224 * period completes. For more information, see the Tree RCU header comment.
225 */
226unsigned long start_poll_synchronize_rcu(void)
227{
228 unsigned long gp_seq = get_state_synchronize_rcu();
229
230 if (unlikely(is_idle_task(current))) {
231 /* force scheduling for rcu_qs() */
232 resched_cpu(0);
233 }
234 return gp_seq;
235}
236EXPORT_SYMBOL_GPL(start_poll_synchronize_rcu);
237
238/*
239 * Return true if the grace period corresponding to oldstate has completed
240 * and false otherwise. For more information, see the Tree RCU header
241 * comment.
242 */
243bool poll_state_synchronize_rcu(unsigned long oldstate)
244{
Paul E. McKenney414c1232022-04-13 15:17:25 -0700245 return oldstate == RCU_GET_STATE_COMPLETED || READ_ONCE(rcu_ctrlblk.gp_seq) != oldstate;
Paul E. McKenney0909fc22021-02-25 17:36:06 -0800246}
247EXPORT_SYMBOL_GPL(poll_state_synchronize_rcu);
248
Johannes Berg800d6ac2022-05-27 17:07:45 +0200249#ifdef CONFIG_KASAN_GENERIC
Uladzislau Rezki (Sony)04a522b2022-10-25 16:46:12 +0200250void kvfree_call_rcu(struct rcu_head *head, void *ptr)
Johannes Berg800d6ac2022-05-27 17:07:45 +0200251{
Uladzislau Rezki (Sony)04a522b2022-10-25 16:46:12 +0200252 if (head)
Johannes Berg800d6ac2022-05-27 17:07:45 +0200253 kasan_record_aux_stack_noalloc(ptr);
Johannes Berg800d6ac2022-05-27 17:07:45 +0200254
Uladzislau Rezki (Sony)04a522b2022-10-25 16:46:12 +0200255 __kvfree_call_rcu(head, ptr);
Johannes Berg800d6ac2022-05-27 17:07:45 +0200256}
257EXPORT_SYMBOL_GPL(kvfree_call_rcu);
258#endif
259
Pranith Kumaraa23c6fbc2014-09-19 11:32:29 -0400260void __init rcu_init(void)
Paul E. McKenney9dc5ad32013-03-27 10:11:15 -0700261{
262 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
Pranith Kumaraa23c6fbc2014-09-19 11:32:29 -0400263 rcu_early_boot_tests();
Paul E. McKenney9dc5ad32013-03-27 10:11:15 -0700264}