blob: c2ec6c77874eacc0cd7a340b7acc7c386aea0ead [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 *
22 * Based on the original work by Paul McKenney <paul.mckenney@us.ibm.com>
23 * 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
69 spinlock_t lock ____cacheline_internodealigned_in_smp;
70 cpumask_t cpumask; /* CPUs that need to switch in order */
71 /* for current batch to proceed. */
Ravikiran G Thirumalai22fc6ec2006-01-08 01:01:27 -080072} ____cacheline_internodealigned_in_smp;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74/* Is batch a before batch b ? */
75static inline int rcu_batch_before(long a, long b)
76{
77 return (a - b) < 0;
78}
79
80/* Is batch a after batch b ? */
81static inline int rcu_batch_after(long a, long b)
82{
83 return (a - b) > 0;
84}
85
86/*
87 * Per-CPU data for Read-Copy UPdate.
88 * nxtlist - new callbacks are added here
89 * curlist - current batch for which quiescent cycle started if any
90 */
91struct rcu_data {
92 /* 1) quiescent state handling : */
93 long quiescbatch; /* Batch # for grace period */
94 int passed_quiesc; /* User-mode/idle loop etc. */
95 int qs_pending; /* core waits for quiesc state */
96
97 /* 2) batch handling */
98 long batch; /* Batch # for current RCU batch */
99 struct rcu_head *nxtlist;
100 struct rcu_head **nxttail;
Dipankar Sarma21a1ea92006-03-07 21:55:33 -0800101 long qlen; /* # of queued callbacks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 struct rcu_head *curlist;
103 struct rcu_head **curtail;
104 struct rcu_head *donelist;
105 struct rcu_head **donetail;
Dipankar Sarma21a1ea92006-03-07 21:55:33 -0800106 long blimit; /* Upper limit on a processed batch */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 int cpu;
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800108 struct rcu_head barrier;
Dipankar Sarma21a1ea92006-03-07 21:55:33 -0800109#ifdef CONFIG_SMP
110 long last_rs_qlen; /* qlen during the last resched */
111#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112};
113
114DECLARE_PER_CPU(struct rcu_data, rcu_data);
115DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
116extern struct rcu_ctrlblk rcu_ctrlblk;
117extern struct rcu_ctrlblk rcu_bh_ctrlblk;
118
119/*
120 * Increment the quiescent state counter.
121 * The counter is a bit degenerated: We do not need to know
122 * how many quiescent states passed, just if there was at least
123 * one since the start of the grace period. Thus just a flag.
124 */
125static inline void rcu_qsctr_inc(int cpu)
126{
127 struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
128 rdp->passed_quiesc = 1;
129}
130static inline void rcu_bh_qsctr_inc(int cpu)
131{
132 struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
133 rdp->passed_quiesc = 1;
134}
135
Oleg Nesterov67751772006-01-08 22:19:16 +0300136extern int rcu_pending(int cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
138/**
139 * rcu_read_lock - mark the beginning of an RCU read-side critical section.
140 *
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700141 * When synchronize_rcu() is invoked on one CPU while other CPUs
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 * are within RCU read-side critical sections, then the
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700143 * synchronize_rcu() is guaranteed to block until after all the other
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 * CPUs exit their critical sections. Similarly, if call_rcu() is invoked
145 * on one CPU while other CPUs are within RCU read-side critical
146 * sections, invocation of the corresponding RCU callback is deferred
147 * until after the all the other CPUs exit their critical sections.
148 *
149 * Note, however, that RCU callbacks are permitted to run concurrently
150 * with RCU read-side critical sections. One way that this can happen
151 * is via the following sequence of events: (1) CPU 0 enters an RCU
152 * read-side critical section, (2) CPU 1 invokes call_rcu() to register
153 * an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
154 * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
155 * callback is invoked. This is legal, because the RCU read-side critical
156 * section that was running concurrently with the call_rcu() (and which
157 * therefore might be referencing something that the corresponding RCU
158 * callback would free up) has completed before the corresponding
159 * RCU callback is invoked.
160 *
161 * RCU read-side critical sections may be nested. Any deferred actions
162 * will be deferred until the outermost RCU read-side critical section
163 * completes.
164 *
165 * It is illegal to block while in an RCU read-side critical section.
166 */
167#define rcu_read_lock() preempt_disable()
168
169/**
170 * rcu_read_unlock - marks the end of an RCU read-side critical section.
171 *
172 * See rcu_read_lock() for more information.
173 */
174#define rcu_read_unlock() preempt_enable()
175
176/*
177 * So where is rcu_write_lock()? It does not exist, as there is no
178 * way for writers to lock out RCU readers. This is a feature, not
179 * a bug -- this property is what provides RCU's performance benefits.
180 * Of course, writers must coordinate with each other. The normal
181 * spinlock primitives work well for this, but any other technique may be
182 * used as well. RCU does not care how the writers keep out of each
183 * others' way, as long as they do so.
184 */
185
186/**
187 * rcu_read_lock_bh - mark the beginning of a softirq-only RCU critical section
188 *
189 * This is equivalent of rcu_read_lock(), but to be used when updates
190 * are being done using call_rcu_bh(). Since call_rcu_bh() callbacks
191 * consider completion of a softirq handler to be a quiescent state,
192 * a process in RCU read-side critical section must be protected by
193 * disabling softirqs. Read-side critical sections in interrupt context
194 * can use just rcu_read_lock().
195 *
196 */
197#define rcu_read_lock_bh() local_bh_disable()
198
199/*
200 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
201 *
202 * See rcu_read_lock_bh() for more information.
203 */
204#define rcu_read_unlock_bh() local_bh_enable()
205
206/**
207 * rcu_dereference - fetch an RCU-protected pointer in an
208 * RCU read-side critical section. This pointer may later
209 * be safely dereferenced.
210 *
211 * Inserts memory barriers on architectures that require them
212 * (currently only the Alpha), and, more importantly, documents
213 * exactly which pointers are protected by RCU.
214 */
215
216#define rcu_dereference(p) ({ \
217 typeof(p) _________p1 = p; \
218 smp_read_barrier_depends(); \
219 (_________p1); \
220 })
221
222/**
223 * rcu_assign_pointer - assign (publicize) a pointer to a newly
224 * initialized structure that will be dereferenced by RCU read-side
225 * critical sections. Returns the value assigned.
226 *
227 * Inserts memory barriers on architectures that require them
228 * (pretty much all of them other than x86), and also prevents
229 * the compiler from reordering the code that initializes the
230 * structure after the pointer assignment. More importantly, this
231 * call documents which pointers will be dereferenced by RCU read-side
232 * code.
233 */
234
235#define rcu_assign_pointer(p, v) ({ \
236 smp_wmb(); \
237 (p) = (v); \
238 })
239
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700240/**
241 * synchronize_sched - block until all CPUs have exited any non-preemptive
242 * kernel code sequences.
243 *
244 * This means that all preempt_disable code sequences, including NMI and
245 * hardware-interrupt handlers, in progress on entry will have completed
246 * before this primitive returns. However, this does not guarantee that
Paul E. McKenneybb3b9cf12006-02-03 03:04:38 -0800247 * softirq handlers will have completed, since in some kernels, these
248 * handlers can run in process context, and can block.
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700249 *
250 * This primitive provides the guarantees made by the (deprecated)
251 * synchronize_kernel() API. In contrast, synchronize_rcu() only
252 * guarantees that rcu_read_lock() sections will have completed.
Paul E. McKenneybb3b9cf12006-02-03 03:04:38 -0800253 * In "classic RCU", these two guarantees happen to be one and
254 * the same, but can differ in realtime RCU implementations.
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700255 */
256#define synchronize_sched() synchronize_rcu()
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258extern void rcu_init(void);
259extern void rcu_check_callbacks(int cpu, int user);
260extern void rcu_restart_cpu(int cpu);
Paul E. McKenneya241ec62005-10-30 15:03:12 -0800261extern long rcu_batches_completed(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
263/* Exported interfaces */
264extern void FASTCALL(call_rcu(struct rcu_head *head,
265 void (*func)(struct rcu_head *head)));
266extern void FASTCALL(call_rcu_bh(struct rcu_head *head,
267 void (*func)(struct rcu_head *head)));
Paul E. McKenney9b06e812005-05-01 08:59:04 -0700268extern __deprecated_for_modules void synchronize_kernel(void);
269extern void synchronize_rcu(void);
270void synchronize_idle(void);
Dipankar Sarmaab4720e2005-12-12 00:37:05 -0800271extern void rcu_barrier(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
273#endif /* __KERNEL__ */
274#endif /* __LINUX_RCUPDATE_H */