diff options
Diffstat (limited to 'include/linux/rcupdate.h')
-rw-r--r-- | include/linux/rcupdate.h | 271 |
1 files changed, 271 insertions, 0 deletions
diff --git a/include/linux/rcupdate.h b/include/linux/rcupdate.h new file mode 100644 index 000000000000..4d747433916b --- /dev/null +++ b/include/linux/rcupdate.h | |||
@@ -0,0 +1,271 @@ | |||
1 | /* | ||
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 | */ | ||
50 | struct rcu_head { | ||
51 | struct rcu_head *next; | ||
52 | void (*func)(struct rcu_head *head); | ||
53 | }; | ||
54 | |||
55 | #define RCU_HEAD_INIT(head) { .next = NULL, .func = NULL } | ||
56 | #define RCU_HEAD(head) struct rcu_head head = RCU_HEAD_INIT(head) | ||
57 | #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. */ | ||
64 | struct 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? */ | ||
68 | } ____cacheline_maxaligned_in_smp; | ||
69 | |||
70 | /* Is batch a before batch b ? */ | ||
71 | static inline int rcu_batch_before(long a, long b) | ||
72 | { | ||
73 | return (a - b) < 0; | ||
74 | } | ||
75 | |||
76 | /* Is batch a after batch b ? */ | ||
77 | static inline int rcu_batch_after(long a, long b) | ||
78 | { | ||
79 | return (a - b) > 0; | ||
80 | } | ||
81 | |||
82 | /* | ||
83 | * Per-CPU data for Read-Copy UPdate. | ||
84 | * nxtlist - new callbacks are added here | ||
85 | * curlist - current batch for which quiescent cycle started if any | ||
86 | */ | ||
87 | struct rcu_data { | ||
88 | /* 1) quiescent state handling : */ | ||
89 | long quiescbatch; /* Batch # for grace period */ | ||
90 | int passed_quiesc; /* User-mode/idle loop etc. */ | ||
91 | int qs_pending; /* core waits for quiesc state */ | ||
92 | |||
93 | /* 2) batch handling */ | ||
94 | long batch; /* Batch # for current RCU batch */ | ||
95 | struct rcu_head *nxtlist; | ||
96 | struct rcu_head **nxttail; | ||
97 | struct rcu_head *curlist; | ||
98 | struct rcu_head **curtail; | ||
99 | struct rcu_head *donelist; | ||
100 | struct rcu_head **donetail; | ||
101 | int cpu; | ||
102 | }; | ||
103 | |||
104 | DECLARE_PER_CPU(struct rcu_data, rcu_data); | ||
105 | DECLARE_PER_CPU(struct rcu_data, rcu_bh_data); | ||
106 | extern struct rcu_ctrlblk rcu_ctrlblk; | ||
107 | extern struct rcu_ctrlblk rcu_bh_ctrlblk; | ||
108 | |||
109 | /* | ||
110 | * Increment the quiescent state counter. | ||
111 | * The counter is a bit degenerated: We do not need to know | ||
112 | * how many quiescent states passed, just if there was at least | ||
113 | * one since the start of the grace period. Thus just a flag. | ||
114 | */ | ||
115 | static inline void rcu_qsctr_inc(int cpu) | ||
116 | { | ||
117 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); | ||
118 | rdp->passed_quiesc = 1; | ||
119 | } | ||
120 | static inline void rcu_bh_qsctr_inc(int cpu) | ||
121 | { | ||
122 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); | ||
123 | rdp->passed_quiesc = 1; | ||
124 | } | ||
125 | |||
126 | static inline int __rcu_pending(struct rcu_ctrlblk *rcp, | ||
127 | struct rcu_data *rdp) | ||
128 | { | ||
129 | /* This cpu has pending rcu entries and the grace period | ||
130 | * for them has completed. | ||
131 | */ | ||
132 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) | ||
133 | return 1; | ||
134 | |||
135 | /* This cpu has no pending entries, but there are new entries */ | ||
136 | if (!rdp->curlist && rdp->nxtlist) | ||
137 | return 1; | ||
138 | |||
139 | /* This cpu has finished callbacks to invoke */ | ||
140 | if (rdp->donelist) | ||
141 | return 1; | ||
142 | |||
143 | /* The rcu core waits for a quiescent state from the cpu */ | ||
144 | if (rdp->quiescbatch != rcp->cur || rdp->qs_pending) | ||
145 | return 1; | ||
146 | |||
147 | /* nothing to do */ | ||
148 | return 0; | ||
149 | } | ||
150 | |||
151 | static inline int rcu_pending(int cpu) | ||
152 | { | ||
153 | return __rcu_pending(&rcu_ctrlblk, &per_cpu(rcu_data, cpu)) || | ||
154 | __rcu_pending(&rcu_bh_ctrlblk, &per_cpu(rcu_bh_data, cpu)); | ||
155 | } | ||
156 | |||
157 | /** | ||
158 | * rcu_read_lock - mark the beginning of an RCU read-side critical section. | ||
159 | * | ||
160 | * When synchronize_kernel() is invoked on one CPU while other CPUs | ||
161 | * are within RCU read-side critical sections, then the | ||
162 | * synchronize_kernel() is guaranteed to block until after all the other | ||
163 | * CPUs exit their critical sections. Similarly, if call_rcu() is invoked | ||
164 | * on one CPU while other CPUs are within RCU read-side critical | ||
165 | * sections, invocation of the corresponding RCU callback is deferred | ||
166 | * until after the all the other CPUs exit their critical sections. | ||
167 | * | ||
168 | * Note, however, that RCU callbacks are permitted to run concurrently | ||
169 | * with RCU read-side critical sections. One way that this can happen | ||
170 | * is via the following sequence of events: (1) CPU 0 enters an RCU | ||
171 | * read-side critical section, (2) CPU 1 invokes call_rcu() to register | ||
172 | * an RCU callback, (3) CPU 0 exits the RCU read-side critical section, | ||
173 | * (4) CPU 2 enters a RCU read-side critical section, (5) the RCU | ||
174 | * callback is invoked. This is legal, because the RCU read-side critical | ||
175 | * section that was running concurrently with the call_rcu() (and which | ||
176 | * therefore might be referencing something that the corresponding RCU | ||
177 | * callback would free up) has completed before the corresponding | ||
178 | * RCU callback is invoked. | ||
179 | * | ||
180 | * RCU read-side critical sections may be nested. Any deferred actions | ||
181 | * will be deferred until the outermost RCU read-side critical section | ||
182 | * completes. | ||
183 | * | ||
184 | * It is illegal to block while in an RCU read-side critical section. | ||
185 | */ | ||
186 | #define rcu_read_lock() preempt_disable() | ||
187 | |||
188 | /** | ||
189 | * rcu_read_unlock - marks the end of an RCU read-side critical section. | ||
190 | * | ||
191 | * See rcu_read_lock() for more information. | ||
192 | */ | ||
193 | #define rcu_read_unlock() preempt_enable() | ||
194 | |||
195 | /* | ||
196 | * So where is rcu_write_lock()? It does not exist, as there is no | ||
197 | * way for writers to lock out RCU readers. This is a feature, not | ||
198 | * a bug -- this property is what provides RCU's performance benefits. | ||
199 | * Of course, writers must coordinate with each other. The normal | ||
200 | * spinlock primitives work well for this, but any other technique may be | ||
201 | * used as well. RCU does not care how the writers keep out of each | ||
202 | * others' way, as long as they do so. | ||
203 | */ | ||
204 | |||
205 | /** | ||
206 | * rcu_read_lock_bh - mark the beginning of a softirq-only RCU critical section | ||
207 | * | ||
208 | * This is equivalent of rcu_read_lock(), but to be used when updates | ||
209 | * are being done using call_rcu_bh(). Since call_rcu_bh() callbacks | ||
210 | * consider completion of a softirq handler to be a quiescent state, | ||
211 | * a process in RCU read-side critical section must be protected by | ||
212 | * disabling softirqs. Read-side critical sections in interrupt context | ||
213 | * can use just rcu_read_lock(). | ||
214 | * | ||
215 | */ | ||
216 | #define rcu_read_lock_bh() local_bh_disable() | ||
217 | |||
218 | /* | ||
219 | * rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section | ||
220 | * | ||
221 | * See rcu_read_lock_bh() for more information. | ||
222 | */ | ||
223 | #define rcu_read_unlock_bh() local_bh_enable() | ||
224 | |||
225 | /** | ||
226 | * rcu_dereference - fetch an RCU-protected pointer in an | ||
227 | * RCU read-side critical section. This pointer may later | ||
228 | * be safely dereferenced. | ||
229 | * | ||
230 | * Inserts memory barriers on architectures that require them | ||
231 | * (currently only the Alpha), and, more importantly, documents | ||
232 | * exactly which pointers are protected by RCU. | ||
233 | */ | ||
234 | |||
235 | #define rcu_dereference(p) ({ \ | ||
236 | typeof(p) _________p1 = p; \ | ||
237 | smp_read_barrier_depends(); \ | ||
238 | (_________p1); \ | ||
239 | }) | ||
240 | |||
241 | /** | ||
242 | * rcu_assign_pointer - assign (publicize) a pointer to a newly | ||
243 | * initialized structure that will be dereferenced by RCU read-side | ||
244 | * critical sections. Returns the value assigned. | ||
245 | * | ||
246 | * Inserts memory barriers on architectures that require them | ||
247 | * (pretty much all of them other than x86), and also prevents | ||
248 | * the compiler from reordering the code that initializes the | ||
249 | * structure after the pointer assignment. More importantly, this | ||
250 | * call documents which pointers will be dereferenced by RCU read-side | ||
251 | * code. | ||
252 | */ | ||
253 | |||
254 | #define rcu_assign_pointer(p, v) ({ \ | ||
255 | smp_wmb(); \ | ||
256 | (p) = (v); \ | ||
257 | }) | ||
258 | |||
259 | extern void rcu_init(void); | ||
260 | extern void rcu_check_callbacks(int cpu, int user); | ||
261 | extern void rcu_restart_cpu(int cpu); | ||
262 | |||
263 | /* Exported interfaces */ | ||
264 | extern void FASTCALL(call_rcu(struct rcu_head *head, | ||
265 | void (*func)(struct rcu_head *head))); | ||
266 | extern void FASTCALL(call_rcu_bh(struct rcu_head *head, | ||
267 | void (*func)(struct rcu_head *head))); | ||
268 | extern void synchronize_kernel(void); | ||
269 | |||
270 | #endif /* __KERNEL__ */ | ||
271 | #endif /* __LINUX_RCUPDATE_H */ | ||