blob: c6b7485eac7ce1e59c7f01cf71221e6637f524ce [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 *
18 * Copyright (C) IBM Corporation, 2001
19 *
20 * Author: Dipankar Sarma <dipankar@in.ibm.com>
21 *
Josh Triplett595182b2006-10-04 02:17:21 -070022 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
24 * Papers:
25 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
26 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
27 *
28 * For detailed explanation of Read-Copy Update mechanism see -
29 * http://lse.sourceforge.net/locking/rcupdate.html
30 *
31 */
32
33#ifndef __LINUX_RCUPDATE_H
34#define __LINUX_RCUPDATE_H
35
36#ifdef __KERNEL__
37
38#include <linux/cache.h>
39#include <linux/spinlock.h>
40#include <linux/threads.h>
41#include <linux/percpu.h>
42#include <linux/cpumask.h>
43#include <linux/seqlock.h>
44
45/**
46 * struct rcu_head - callback structure for use with RCU
47 * @next: next update requests in a list
48 * @func: actual update function to call after the grace period.
49 */
50struct rcu_head {
51 struct rcu_head *next;
52 void (*func)(struct rcu_head *head);
53};
54
Dipankar Sarma8b6490e2005-09-09 13:04:07 -070055#define RCU_HEAD_INIT { .next = NULL, .func = NULL }
56#define RCU_HEAD(head) struct rcu_head head = RCU_HEAD_INIT
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#define INIT_RCU_HEAD(ptr) do { \
58 (ptr)->next = NULL; (ptr)->func = NULL; \
59} while (0)
60
61
62
63/* Global control variables for rcupdate callback mechanism. */
64struct rcu_ctrlblk {
65 long cur; /* Current batch number. */
66 long completed; /* Number of the last completed batch */
67 int next_pending; /* Is the next batch already waiting? */
Oleg Nesterov69a0b312006-01-10 16:48:02 +030068
Oleg Nesterov20e97512006-10-04 02:17:17 -070069 int signaled;
70
Oleg Nesterov69a0b312006-01-10 16:48:02 +030071 spinlock_t lock ____cacheline_internodealigned_in_smp;
72 cpumask_t cpumask; /* CPUs that need to switch in order */
73 /* for current batch to proceed. */
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080074} ____cacheline_internodealigned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76/* Is batch a before batch b ? */
77static inline int rcu_batch_before(long a, long b)
78{
79 return (a - b) < 0;
80}
81
82/* Is batch a after batch b ? */
83static inline int rcu_batch_after(long a, long b)
84{
85 return (a - b) > 0;
86}
87
88/*
89 * Per-CPU data for Read-Copy UPdate.
90 * nxtlist - new callbacks are added here
91 * curlist - current batch for which quiescent cycle started if any
92 */
93struct rcu_data {
94 /* 1) quiescent state handling : */
95 long quiescbatch; /* Batch # for grace period */
96 int passed_quiesc; /* User-mode/idle loop etc. */
97 int qs_pending; /* core waits for quiesc state */
98
99 /* 2) batch handling */
100 long batch; /* Batch # for current RCU batch */
101 struct rcu_head *nxtlist;
102 struct rcu_head **nxttail;
Dipankar Sarma21a1ea92006-03-07 21:55:33 -0800103 long qlen; /* # of queued callbacks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 struct rcu_head *curlist;
105 struct rcu_head **curtail;
106 struct rcu_head *donelist;
107 struct rcu_head **donetail;
Dipankar Sarma21a1ea92006-03-07 21:55:33 -0800108 long blimit; /* Upper limit on a processed batch */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 int cpu;
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800110 struct rcu_head barrier;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111};
112
113DECLARE_PER_CPU(struct rcu_data, rcu_data);
114DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116/*
117 * Increment the quiescent state counter.
118 * The counter is a bit degenerated: We do not need to know
119 * how many quiescent states passed, just if there was at least
120 * one since the start of the grace period. Thus just a flag.
121 */
122static inline void rcu_qsctr_inc(int cpu)
123{
124 struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
125 rdp->passed_quiesc = 1;
126}
127static inline void rcu_bh_qsctr_inc(int cpu)
128{
129 struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
130 rdp->passed_quiesc = 1;
131}
132
Oleg Nesterov67751772006-01-08 22:19:16 +0300133extern int rcu_pending(int cpu);
Heiko Carstens986733e2006-05-15 09:43:58 -0700134extern int rcu_needs_cpu(int cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135
136/**
137 * rcu_read_lock - mark the beginning of an RCU read-side critical section.
138 *
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700139 * When synchronize_rcu() is invoked on one CPU while other CPUs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 * are within RCU read-side critical sections, then the
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700141 * synchronize_rcu() is guaranteed to block until after all the other
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
143 * on one CPU while other CPUs are within RCU read-side critical
144 * sections, invocation of the corresponding RCU callback is deferred
145 * until after the all the other CPUs exit their critical sections.
146 *
147 * Note, however, that RCU callbacks are permitted to run concurrently
148 * with RCU read-side critical sections. One way that this can happen
149 * is via the following sequence of events: (1) CPU 0 enters an RCU
150 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
151 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
152 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
153 * callback is invoked. This is legal, because the RCU read-side critical
154 * section that was running concurrently with the call_rcu() (and which
155 * therefore might be referencing something that the corresponding RCU
156 * callback would free up) has completed before the corresponding
157 * RCU callback is invoked.
158 *
159 * RCU read-side critical sections may be nested. Any deferred actions
160 * will be deferred until the outermost RCU read-side critical section
161 * completes.
162 *
163 * It is illegal to block while in an RCU read-side critical section.
164 */
Josh Triplett7f04ac02006-06-30 01:56:05 -0700165#define rcu_read_lock() \
166 do { \
167 preempt_disable(); \
168 __acquire(RCU); \
169 } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
171/**
172 * rcu_read_unlock - marks the end of an RCU read-side critical section.
173 *
174 * See rcu_read_lock() for more information.
175 */
Josh Triplett7f04ac02006-06-30 01:56:05 -0700176#define rcu_read_unlock() \
177 do { \
178 __release(RCU); \
179 preempt_enable(); \
180 } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
182/*
183 * So where is rcu_write_lock()? It does not exist, as there is no
184 * way for writers to lock out RCU readers. This is a feature, not
185 * a bug -- this property is what provides RCU's performance benefits.
186 * Of course, writers must coordinate with each other. The normal
187 * spinlock primitives work well for this, but any other technique may be
188 * used as well. RCU does not care how the writers keep out of each
189 * others' way, as long as they do so.
190 */
191
192/**
193 * rcu_read_lock_bh - mark the beginning of a softirq-only RCU critical section
194 *
195 * This is equivalent of rcu_read_lock(), but to be used when updates
196 * are being done using call_rcu_bh(). Since call_rcu_bh() callbacks
197 * consider completion of a softirq handler to be a quiescent state,
198 * a process in RCU read-side critical section must be protected by
199 * disabling softirqs. Read-side critical sections in interrupt context
200 * can use just rcu_read_lock().
201 *
202 */
Josh Triplett7f04ac02006-06-30 01:56:05 -0700203#define rcu_read_lock_bh() \
204 do { \
205 local_bh_disable(); \
206 __acquire(RCU_BH); \
207 } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
209/*
210 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
211 *
212 * See rcu_read_lock_bh() for more information.
213 */
Josh Triplett7f04ac02006-06-30 01:56:05 -0700214#define rcu_read_unlock_bh() \
215 do { \
216 __release(RCU_BH); \
217 local_bh_enable(); \
218 } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
220/**
221 * rcu_dereference - fetch an RCU-protected pointer in an
222 * RCU read-side critical section. This pointer may later
223 * be safely dereferenced.
224 *
225 * Inserts memory barriers on architectures that require them
226 * (currently only the Alpha), and, more importantly, documents
227 * exactly which pointers are protected by RCU.
228 */
229
230#define rcu_dereference(p) ({ \
231 typeof(p) _________p1 = p; \
232 smp_read_barrier_depends(); \
233 (_________p1); \
234 })
235
236/**
237 * rcu_assign_pointer - assign (publicize) a pointer to a newly
238 * initialized structure that will be dereferenced by RCU read-side
239 * critical sections. Returns the value assigned.
240 *
241 * Inserts memory barriers on architectures that require them
242 * (pretty much all of them other than x86), and also prevents
243 * the compiler from reordering the code that initializes the
244 * structure after the pointer assignment. More importantly, this
245 * call documents which pointers will be dereferenced by RCU read-side
246 * code.
247 */
248
249#define rcu_assign_pointer(p, v) ({ \
250 smp_wmb(); \
251 (p) = (v); \
252 })
253
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700254/**
255 * synchronize_sched - block until all CPUs have exited any non-preemptive
256 * kernel code sequences.
257 *
258 * This means that all preempt_disable code sequences, including NMI and
259 * hardware-interrupt handlers, in progress on entry will have completed
260 * before this primitive returns. However, this does not guarantee that
Paul E. McKenneybb3b9cf2006-02-03 03:04:38 -0800261 * softirq handlers will have completed, since in some kernels, these
262 * handlers can run in process context, and can block.
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700263 *
Paul E. McKenneyd83015b2006-06-23 02:05:51 -0700264 * This primitive provides the guarantees made by the (now removed)
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700265 * synchronize_kernel() API. In contrast, synchronize_rcu() only
266 * guarantees that rcu_read_lock() sections will have completed.
Paul E. McKenneybb3b9cf2006-02-03 03:04:38 -0800267 * In "classic RCU", these two guarantees happen to be one and
268 * the same, but can differ in realtime RCU implementations.
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700269 */
270#define synchronize_sched() synchronize_rcu()
271
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272extern void rcu_init(void);
273extern void rcu_check_callbacks(int cpu, int user);
274extern void rcu_restart_cpu(int cpu);
Paul E. McKenneya241ec62005-10-30 15:03:12 -0800275extern long rcu_batches_completed(void);
Paul E. McKenneyc32e0662006-06-27 02:54:04 -0700276extern long rcu_batches_completed_bh(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277
278/* Exported interfaces */
279extern void FASTCALL(call_rcu(struct rcu_head *head,
280 void (*func)(struct rcu_head *head)));
281extern void FASTCALL(call_rcu_bh(struct rcu_head *head,
282 void (*func)(struct rcu_head *head)));
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700283extern void synchronize_rcu(void);
284void synchronize_idle(void);
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800285extern void rcu_barrier(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287#endif /* __KERNEL__ */
288#endif /* __LINUX_RCUPDATE_H */