2 * Read-Copy Update mechanism for mutual exclusion
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.
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.
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.
18 * Copyright (C) IBM Corporation, 2001
20 * Author: Dipankar Sarma <dipankar@in.ibm.com>
22 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
23 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
25 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf
26 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001)
28 * For detailed explanation of Read-Copy Update mechanism see -
29 * http://lse.sourceforge.net/locking/rcupdate.html
33 #ifndef __LINUX_RCUPDATE_H
34 #define __LINUX_RCUPDATE_H
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>
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.
51 struct rcu_head *next;
52 void (*func)(struct rcu_head *head);
55 #define RCU_HEAD_INIT { .next = NULL, .func = NULL }
56 #define RCU_HEAD(head) struct rcu_head head = RCU_HEAD_INIT
57 #define INIT_RCU_HEAD(ptr) do { \
58 (ptr)->next = NULL; (ptr)->func = NULL; \
63 /* Global control variables for rcupdate callback mechanism. */
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? */
71 spinlock_t lock ____cacheline_internodealigned_in_smp;
72 cpumask_t cpumask; /* CPUs that need to switch in order */
73 /* for current batch to proceed. */
74 } ____cacheline_internodealigned_in_smp;
76 /* Is batch a before batch b ? */
77 static inline int rcu_batch_before(long a, long b)
82 /* Is batch a after batch b ? */
83 static inline int rcu_batch_after(long a, long b)
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
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 */
99 /* 2) batch handling */
100 long batch; /* Batch # for current RCU batch */
101 struct rcu_head *nxtlist;
102 struct rcu_head **nxttail;
103 long qlen; /* # of queued callbacks */
104 struct rcu_head *curlist;
105 struct rcu_head **curtail;
106 struct rcu_head *donelist;
107 struct rcu_head **donetail;
108 long blimit; /* Upper limit on a processed batch */
110 struct rcu_head barrier;
113 DECLARE_PER_CPU(struct rcu_data, rcu_data);
114 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
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.
122 static inline void rcu_qsctr_inc(int cpu)
124 struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
125 rdp->passed_quiesc = 1;
127 static inline void rcu_bh_qsctr_inc(int cpu)
129 struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
130 rdp->passed_quiesc = 1;
133 extern int rcu_pending(int cpu);
134 extern int rcu_needs_cpu(int cpu);
137 * rcu_read_lock - mark the beginning of an RCU read-side critical section.
139 * When synchronize_rcu() is invoked on one CPU while other CPUs
140 * are within RCU read-side critical sections, then the
141 * synchronize_rcu() is guaranteed to block until after all the other
142 * 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.
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.
159 * RCU read-side critical sections may be nested. Any deferred actions
160 * will be deferred until the outermost RCU read-side critical section
163 * It is illegal to block while in an RCU read-side critical section.
165 #define rcu_read_lock() \
172 * rcu_read_unlock - marks the end of an RCU read-side critical section.
174 * See rcu_read_lock() for more information.
176 #define rcu_read_unlock() \
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.
193 * rcu_read_lock_bh - mark the beginning of a softirq-only RCU critical section
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().
203 #define rcu_read_lock_bh() \
205 local_bh_disable(); \
210 * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
212 * See rcu_read_lock_bh() for more information.
214 #define rcu_read_unlock_bh() \
221 * rcu_dereference - fetch an RCU-protected pointer in an
222 * RCU read-side critical section. This pointer may later
223 * be safely dereferenced.
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.
230 #define rcu_dereference(p) ({ \
231 typeof(p) _________p1 = p; \
232 smp_read_barrier_depends(); \
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.
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
249 #define rcu_assign_pointer(p, v) ({ \
255 * synchronize_sched - block until all CPUs have exited any non-preemptive
256 * kernel code sequences.
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
261 * softirq handlers will have completed, since in some kernels, these
262 * handlers can run in process context, and can block.
264 * This primitive provides the guarantees made by the (now removed)
265 * synchronize_kernel() API. In contrast, synchronize_rcu() only
266 * guarantees that rcu_read_lock() sections will have completed.
267 * In "classic RCU", these two guarantees happen to be one and
268 * the same, but can differ in realtime RCU implementations.
270 #define synchronize_sched() synchronize_rcu()
272 extern void rcu_init(void);
273 extern void rcu_check_callbacks(int cpu, int user);
274 extern void rcu_restart_cpu(int cpu);
275 extern long rcu_batches_completed(void);
276 extern long rcu_batches_completed_bh(void);
278 /* Exported interfaces */
279 extern void FASTCALL(call_rcu(struct rcu_head *head,
280 void (*func)(struct rcu_head *head)));
281 extern void FASTCALL(call_rcu_bh(struct rcu_head *head,
282 void (*func)(struct rcu_head *head)));
283 extern void synchronize_rcu(void);
284 extern void rcu_barrier(void);
286 #endif /* __KERNEL__ */
287 #endif /* __LINUX_RCUPDATE_H */