aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/srcu.c
diff options
context:
space:
mode:
authorPaul E. McKenney <paulmck@us.ibm.com>2006-10-04 05:17:02 -0400
committerLinus Torvalds <torvalds@g5.osdl.org>2006-10-04 10:55:30 -0400
commit621934ee7ed5b073c7fd638b347e632c53572761 (patch)
tree5722f9cda22c099ad60545f963410dcbc762ee65 /kernel/srcu.c
parent95d77884c77beed676036d2f74d10b470a483c63 (diff)
[PATCH] srcu-3: RCU variant permitting read-side blocking
Updated patch adding a variant of RCU that permits sleeping in read-side critical sections. SRCU is as follows: o Each use of SRCU creates its own srcu_struct, and each srcu_struct has its own set of grace periods. This is critical, as it prevents one subsystem with a blocking reader from holding up SRCU grace periods for other subsystems. o The SRCU primitives (srcu_read_lock(), srcu_read_unlock(), and synchronize_srcu()) all take a pointer to a srcu_struct. o The SRCU primitives must be called from process context. o srcu_read_lock() returns an int that must be passed to the matching srcu_read_unlock(). Realtime RCU avoids the need for this by storing the state in the task struct, but SRCU needs to allow a given code path to pass through multiple SRCU domains -- storing state in the task struct would therefore require either arbitrary space in the task struct or arbitrary limits on SRCU nesting. So I kicked the state-storage problem up to the caller. Of course, it is not permitted to call synchronize_srcu() while in an SRCU read-side critical section. o There is no call_srcu(). It would not be hard to implement one, but it seems like too easy a way to OOM the system. (Hey, we have enough trouble with call_rcu(), which does -not- permit readers to sleep!!!) So, if you want it, please tell me why... [josht@us.ibm.com: sparse notation] Signed-off-by: Paul E. McKenney <paulmck@us.ibm.com> Signed-off-by: Josh Triplett <josh@freedesktop.org> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'kernel/srcu.c')
-rw-r--r--kernel/srcu.c257
1 files changed, 257 insertions, 0 deletions
diff --git a/kernel/srcu.c b/kernel/srcu.c
new file mode 100644
index 000000000000..7e1979f624ba
--- /dev/null
+++ b/kernel/srcu.c
@@ -0,0 +1,257 @@
1/*
2 * Sleepable 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, 2006
19 *
20 * Author: Paul McKenney <paulmck@us.ibm.com>
21 *
22 * For detailed explanation of Read-Copy Update mechanism see -
23 * Documentation/RCU/ *.txt
24 *
25 */
26
27#include <linux/module.h>
28#include <linux/mutex.h>
29#include <linux/percpu.h>
30#include <linux/preempt.h>
31#include <linux/rcupdate.h>
32#include <linux/sched.h>
33#include <linux/slab.h>
34#include <linux/smp.h>
35#include <linux/srcu.h>
36
37/**
38 * init_srcu_struct - initialize a sleep-RCU structure
39 * @sp: structure to initialize.
40 *
41 * Must invoke this on a given srcu_struct before passing that srcu_struct
42 * to any other function. Each srcu_struct represents a separate domain
43 * of SRCU protection.
44 */
45void init_srcu_struct(struct srcu_struct *sp)
46{
47 sp->completed = 0;
48 sp->per_cpu_ref = alloc_percpu(struct srcu_struct_array);
49 mutex_init(&sp->mutex);
50}
51
52/*
53 * srcu_readers_active_idx -- returns approximate number of readers
54 * active on the specified rank of per-CPU counters.
55 */
56
57static int srcu_readers_active_idx(struct srcu_struct *sp, int idx)
58{
59 int cpu;
60 int sum;
61
62 sum = 0;
63 for_each_possible_cpu(cpu)
64 sum += per_cpu_ptr(sp->per_cpu_ref, cpu)->c[idx];
65 return sum;
66}
67
68/**
69 * srcu_readers_active - returns approximate number of readers.
70 * @sp: which srcu_struct to count active readers (holding srcu_read_lock).
71 *
72 * Note that this is not an atomic primitive, and can therefore suffer
73 * severe errors when invoked on an active srcu_struct. That said, it
74 * can be useful as an error check at cleanup time.
75 */
76int srcu_readers_active(struct srcu_struct *sp)
77{
78 return srcu_readers_active_idx(sp, 0) + srcu_readers_active_idx(sp, 1);
79}
80
81/**
82 * cleanup_srcu_struct - deconstruct a sleep-RCU structure
83 * @sp: structure to clean up.
84 *
85 * Must invoke this after you are finished using a given srcu_struct that
86 * was initialized via init_srcu_struct(), else you leak memory.
87 */
88void cleanup_srcu_struct(struct srcu_struct *sp)
89{
90 int sum;
91
92 sum = srcu_readers_active(sp);
93 WARN_ON(sum); /* Leakage unless caller handles error. */
94 if (sum != 0)
95 return;
96 free_percpu(sp->per_cpu_ref);
97 sp->per_cpu_ref = NULL;
98}
99
100/**
101 * srcu_read_lock - register a new reader for an SRCU-protected structure.
102 * @sp: srcu_struct in which to register the new reader.
103 *
104 * Counts the new reader in the appropriate per-CPU element of the
105 * srcu_struct. Must be called from process context.
106 * Returns an index that must be passed to the matching srcu_read_unlock().
107 */
108int srcu_read_lock(struct srcu_struct *sp)
109{
110 int idx;
111
112 preempt_disable();
113 idx = sp->completed & 0x1;
114 barrier(); /* ensure compiler looks -once- at sp->completed. */
115 per_cpu_ptr(sp->per_cpu_ref, smp_processor_id())->c[idx]++;
116 srcu_barrier(); /* ensure compiler won't misorder critical section. */
117 preempt_enable();
118 return idx;
119}
120
121/**
122 * srcu_read_unlock - unregister a old reader from an SRCU-protected structure.
123 * @sp: srcu_struct in which to unregister the old reader.
124 * @idx: return value from corresponding srcu_read_lock().
125 *
126 * Removes the count for the old reader from the appropriate per-CPU
127 * element of the srcu_struct. Note that this may well be a different
128 * CPU than that which was incremented by the corresponding srcu_read_lock().
129 * Must be called from process context.
130 */
131void srcu_read_unlock(struct srcu_struct *sp, int idx)
132{
133 preempt_disable();
134 srcu_barrier(); /* ensure compiler won't misorder critical section. */
135 per_cpu_ptr(sp->per_cpu_ref, smp_processor_id())->c[idx]--;
136 preempt_enable();
137}
138
139/**
140 * synchronize_srcu - wait for prior SRCU read-side critical-section completion
141 * @sp: srcu_struct with which to synchronize.
142 *
143 * Flip the completed counter, and wait for the old count to drain to zero.
144 * As with classic RCU, the updater must use some separate means of
145 * synchronizing concurrent updates. Can block; must be called from
146 * process context.
147 *
148 * Note that it is illegal to call synchornize_srcu() from the corresponding
149 * SRCU read-side critical section; doing so will result in deadlock.
150 * However, it is perfectly legal to call synchronize_srcu() on one
151 * srcu_struct from some other srcu_struct's read-side critical section.
152 */
153void synchronize_srcu(struct srcu_struct *sp)
154{
155 int idx;
156
157 idx = sp->completed;
158 mutex_lock(&sp->mutex);
159
160 /*
161 * Check to see if someone else did the work for us while we were
162 * waiting to acquire the lock. We need -two- advances of
163 * the counter, not just one. If there was but one, we might have
164 * shown up -after- our helper's first synchronize_sched(), thus
165 * having failed to prevent CPU-reordering races with concurrent
166 * srcu_read_unlock()s on other CPUs (see comment below). So we
167 * either (1) wait for two or (2) supply the second ourselves.
168 */
169
170 if ((sp->completed - idx) >= 2) {
171 mutex_unlock(&sp->mutex);
172 return;
173 }
174
175 synchronize_sched(); /* Force memory barrier on all CPUs. */
176
177 /*
178 * The preceding synchronize_sched() ensures that any CPU that
179 * sees the new value of sp->completed will also see any preceding
180 * changes to data structures made by this CPU. This prevents
181 * some other CPU from reordering the accesses in its SRCU
182 * read-side critical section to precede the corresponding
183 * srcu_read_lock() -- ensuring that such references will in
184 * fact be protected.
185 *
186 * So it is now safe to do the flip.
187 */
188
189 idx = sp->completed & 0x1;
190 sp->completed++;
191
192 synchronize_sched(); /* Force memory barrier on all CPUs. */
193
194 /*
195 * At this point, because of the preceding synchronize_sched(),
196 * all srcu_read_lock() calls using the old counters have completed.
197 * Their corresponding critical sections might well be still
198 * executing, but the srcu_read_lock() primitives themselves
199 * will have finished executing.
200 */
201
202 while (srcu_readers_active_idx(sp, idx))
203 schedule_timeout_interruptible(1);
204
205 synchronize_sched(); /* Force memory barrier on all CPUs. */
206
207 /*
208 * The preceding synchronize_sched() forces all srcu_read_unlock()
209 * primitives that were executing concurrently with the preceding
210 * for_each_possible_cpu() loop to have completed by this point.
211 * More importantly, it also forces the corresponding SRCU read-side
212 * critical sections to have also completed, and the corresponding
213 * references to SRCU-protected data items to be dropped.
214 *
215 * Note:
216 *
217 * Despite what you might think at first glance, the
218 * preceding synchronize_sched() -must- be within the
219 * critical section ended by the following mutex_unlock().
220 * Otherwise, a task taking the early exit can race
221 * with a srcu_read_unlock(), which might have executed
222 * just before the preceding srcu_readers_active() check,
223 * and whose CPU might have reordered the srcu_read_unlock()
224 * with the preceding critical section. In this case, there
225 * is nothing preventing the synchronize_sched() task that is
226 * taking the early exit from freeing a data structure that
227 * is still being referenced (out of order) by the task
228 * doing the srcu_read_unlock().
229 *
230 * Alternatively, the comparison with "2" on the early exit
231 * could be changed to "3", but this increases synchronize_srcu()
232 * latency for bulk loads. So the current code is preferred.
233 */
234
235 mutex_unlock(&sp->mutex);
236}
237
238/**
239 * srcu_batches_completed - return batches completed.
240 * @sp: srcu_struct on which to report batch completion.
241 *
242 * Report the number of batches, correlated with, but not necessarily
243 * precisely the same as, the number of grace periods that have elapsed.
244 */
245
246long srcu_batches_completed(struct srcu_struct *sp)
247{
248 return sp->completed;
249}
250
251EXPORT_SYMBOL_GPL(init_srcu_struct);
252EXPORT_SYMBOL_GPL(cleanup_srcu_struct);
253EXPORT_SYMBOL_GPL(srcu_read_lock);
254EXPORT_SYMBOL_GPL(srcu_read_unlock);
255EXPORT_SYMBOL_GPL(synchronize_srcu);
256EXPORT_SYMBOL_GPL(srcu_batches_completed);
257EXPORT_SYMBOL_GPL(srcu_readers_active);