1 |
62 |
marcus.erl |
/*
|
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 <paulmck@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 |
|
|
#include <linux/lockdep.h>
|
45 |
|
|
|
46 |
|
|
/**
|
47 |
|
|
* struct rcu_head - callback structure for use with RCU
|
48 |
|
|
* @next: next update requests in a list
|
49 |
|
|
* @func: actual update function to call after the grace period.
|
50 |
|
|
*/
|
51 |
|
|
struct rcu_head {
|
52 |
|
|
struct rcu_head *next;
|
53 |
|
|
void (*func)(struct rcu_head *head);
|
54 |
|
|
};
|
55 |
|
|
|
56 |
|
|
#define RCU_HEAD_INIT { .next = NULL, .func = NULL }
|
57 |
|
|
#define RCU_HEAD(head) struct rcu_head head = RCU_HEAD_INIT
|
58 |
|
|
#define INIT_RCU_HEAD(ptr) do { \
|
59 |
|
|
(ptr)->next = NULL; (ptr)->func = NULL; \
|
60 |
|
|
} while (0)
|
61 |
|
|
|
62 |
|
|
|
63 |
|
|
|
64 |
|
|
/* Global control variables for rcupdate callback mechanism. */
|
65 |
|
|
struct rcu_ctrlblk {
|
66 |
|
|
long cur; /* Current batch number. */
|
67 |
|
|
long completed; /* Number of the last completed batch */
|
68 |
|
|
int next_pending; /* Is the next batch already waiting? */
|
69 |
|
|
|
70 |
|
|
int signaled;
|
71 |
|
|
|
72 |
|
|
spinlock_t lock ____cacheline_internodealigned_in_smp;
|
73 |
|
|
cpumask_t cpumask; /* CPUs that need to switch in order */
|
74 |
|
|
/* for current batch to proceed. */
|
75 |
|
|
} ____cacheline_internodealigned_in_smp;
|
76 |
|
|
|
77 |
|
|
/* Is batch a before batch b ? */
|
78 |
|
|
static inline int rcu_batch_before(long a, long b)
|
79 |
|
|
{
|
80 |
|
|
return (a - b) < 0;
|
81 |
|
|
}
|
82 |
|
|
|
83 |
|
|
/* Is batch a after batch b ? */
|
84 |
|
|
static inline int rcu_batch_after(long a, long b)
|
85 |
|
|
{
|
86 |
|
|
return (a - b) > 0;
|
87 |
|
|
}
|
88 |
|
|
|
89 |
|
|
/*
|
90 |
|
|
* Per-CPU data for Read-Copy UPdate.
|
91 |
|
|
* nxtlist - new callbacks are added here
|
92 |
|
|
* curlist - current batch for which quiescent cycle started if any
|
93 |
|
|
*/
|
94 |
|
|
struct rcu_data {
|
95 |
|
|
/* 1) quiescent state handling : */
|
96 |
|
|
long quiescbatch; /* Batch # for grace period */
|
97 |
|
|
int passed_quiesc; /* User-mode/idle loop etc. */
|
98 |
|
|
int qs_pending; /* core waits for quiesc state */
|
99 |
|
|
|
100 |
|
|
/* 2) batch handling */
|
101 |
|
|
long batch; /* Batch # for current RCU batch */
|
102 |
|
|
struct rcu_head *nxtlist;
|
103 |
|
|
struct rcu_head **nxttail;
|
104 |
|
|
long qlen; /* # of queued callbacks */
|
105 |
|
|
struct rcu_head *curlist;
|
106 |
|
|
struct rcu_head **curtail;
|
107 |
|
|
struct rcu_head *donelist;
|
108 |
|
|
struct rcu_head **donetail;
|
109 |
|
|
long blimit; /* Upper limit on a processed batch */
|
110 |
|
|
int cpu;
|
111 |
|
|
struct rcu_head barrier;
|
112 |
|
|
};
|
113 |
|
|
|
114 |
|
|
DECLARE_PER_CPU(struct rcu_data, rcu_data);
|
115 |
|
|
DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
|
116 |
|
|
|
117 |
|
|
/*
|
118 |
|
|
* Increment the quiescent state counter.
|
119 |
|
|
* The counter is a bit degenerated: We do not need to know
|
120 |
|
|
* how many quiescent states passed, just if there was at least
|
121 |
|
|
* one since the start of the grace period. Thus just a flag.
|
122 |
|
|
*/
|
123 |
|
|
static inline void rcu_qsctr_inc(int cpu)
|
124 |
|
|
{
|
125 |
|
|
struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
|
126 |
|
|
rdp->passed_quiesc = 1;
|
127 |
|
|
}
|
128 |
|
|
static inline void rcu_bh_qsctr_inc(int cpu)
|
129 |
|
|
{
|
130 |
|
|
struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
|
131 |
|
|
rdp->passed_quiesc = 1;
|
132 |
|
|
}
|
133 |
|
|
|
134 |
|
|
extern int rcu_pending(int cpu);
|
135 |
|
|
extern int rcu_needs_cpu(int cpu);
|
136 |
|
|
|
137 |
|
|
#ifdef CONFIG_DEBUG_LOCK_ALLOC
|
138 |
|
|
extern struct lockdep_map rcu_lock_map;
|
139 |
|
|
# define rcu_read_acquire() lock_acquire(&rcu_lock_map, 0, 0, 2, 1, _THIS_IP_)
|
140 |
|
|
# define rcu_read_release() lock_release(&rcu_lock_map, 1, _THIS_IP_)
|
141 |
|
|
#else
|
142 |
|
|
# define rcu_read_acquire() do { } while (0)
|
143 |
|
|
# define rcu_read_release() do { } while (0)
|
144 |
|
|
#endif
|
145 |
|
|
|
146 |
|
|
/**
|
147 |
|
|
* rcu_read_lock - mark the beginning of an RCU read-side critical section.
|
148 |
|
|
*
|
149 |
|
|
* When synchronize_rcu() is invoked on one CPU while other CPUs
|
150 |
|
|
* are within RCU read-side critical sections, then the
|
151 |
|
|
* synchronize_rcu() is guaranteed to block until after all the other
|
152 |
|
|
* CPUs exit their critical sections. Similarly, if call_rcu() is invoked
|
153 |
|
|
* on one CPU while other CPUs are within RCU read-side critical
|
154 |
|
|
* sections, invocation of the corresponding RCU callback is deferred
|
155 |
|
|
* until after the all the other CPUs exit their critical sections.
|
156 |
|
|
*
|
157 |
|
|
* Note, however, that RCU callbacks are permitted to run concurrently
|
158 |
|
|
* with RCU read-side critical sections. One way that this can happen
|
159 |
|
|
* is via the following sequence of events: (1) CPU 0 enters an RCU
|
160 |
|
|
* read-side critical section, (2) CPU 1 invokes call_rcu() to register
|
161 |
|
|
* an RCU callback, (3) CPU 0 exits the RCU read-side critical section,
|
162 |
|
|
* (4) CPU 2 enters a RCU read-side critical section, (5) the RCU
|
163 |
|
|
* callback is invoked. This is legal, because the RCU read-side critical
|
164 |
|
|
* section that was running concurrently with the call_rcu() (and which
|
165 |
|
|
* therefore might be referencing something that the corresponding RCU
|
166 |
|
|
* callback would free up) has completed before the corresponding
|
167 |
|
|
* RCU callback is invoked.
|
168 |
|
|
*
|
169 |
|
|
* RCU read-side critical sections may be nested. Any deferred actions
|
170 |
|
|
* will be deferred until the outermost RCU read-side critical section
|
171 |
|
|
* completes.
|
172 |
|
|
*
|
173 |
|
|
* It is illegal to block while in an RCU read-side critical section.
|
174 |
|
|
*/
|
175 |
|
|
#define rcu_read_lock() \
|
176 |
|
|
do { \
|
177 |
|
|
preempt_disable(); \
|
178 |
|
|
__acquire(RCU); \
|
179 |
|
|
rcu_read_acquire(); \
|
180 |
|
|
} while(0)
|
181 |
|
|
|
182 |
|
|
/**
|
183 |
|
|
* rcu_read_unlock - marks the end of an RCU read-side critical section.
|
184 |
|
|
*
|
185 |
|
|
* See rcu_read_lock() for more information.
|
186 |
|
|
*/
|
187 |
|
|
#define rcu_read_unlock() \
|
188 |
|
|
do { \
|
189 |
|
|
rcu_read_release(); \
|
190 |
|
|
__release(RCU); \
|
191 |
|
|
preempt_enable(); \
|
192 |
|
|
} while(0)
|
193 |
|
|
|
194 |
|
|
/*
|
195 |
|
|
* So where is rcu_write_lock()? It does not exist, as there is no
|
196 |
|
|
* way for writers to lock out RCU readers. This is a feature, not
|
197 |
|
|
* a bug -- this property is what provides RCU's performance benefits.
|
198 |
|
|
* Of course, writers must coordinate with each other. The normal
|
199 |
|
|
* spinlock primitives work well for this, but any other technique may be
|
200 |
|
|
* used as well. RCU does not care how the writers keep out of each
|
201 |
|
|
* others' way, as long as they do so.
|
202 |
|
|
*/
|
203 |
|
|
|
204 |
|
|
/**
|
205 |
|
|
* rcu_read_lock_bh - mark the beginning of a softirq-only RCU critical section
|
206 |
|
|
*
|
207 |
|
|
* This is equivalent of rcu_read_lock(), but to be used when updates
|
208 |
|
|
* are being done using call_rcu_bh(). Since call_rcu_bh() callbacks
|
209 |
|
|
* consider completion of a softirq handler to be a quiescent state,
|
210 |
|
|
* a process in RCU read-side critical section must be protected by
|
211 |
|
|
* disabling softirqs. Read-side critical sections in interrupt context
|
212 |
|
|
* can use just rcu_read_lock().
|
213 |
|
|
*
|
214 |
|
|
*/
|
215 |
|
|
#define rcu_read_lock_bh() \
|
216 |
|
|
do { \
|
217 |
|
|
local_bh_disable(); \
|
218 |
|
|
__acquire(RCU_BH); \
|
219 |
|
|
rcu_read_acquire(); \
|
220 |
|
|
} while(0)
|
221 |
|
|
|
222 |
|
|
/*
|
223 |
|
|
* rcu_read_unlock_bh - marks the end of a softirq-only RCU critical section
|
224 |
|
|
*
|
225 |
|
|
* See rcu_read_lock_bh() for more information.
|
226 |
|
|
*/
|
227 |
|
|
#define rcu_read_unlock_bh() \
|
228 |
|
|
do { \
|
229 |
|
|
rcu_read_release(); \
|
230 |
|
|
__release(RCU_BH); \
|
231 |
|
|
local_bh_enable(); \
|
232 |
|
|
} while(0)
|
233 |
|
|
|
234 |
|
|
/*
|
235 |
|
|
* Prevent the compiler from merging or refetching accesses. The compiler
|
236 |
|
|
* is also forbidden from reordering successive instances of ACCESS_ONCE(),
|
237 |
|
|
* but only when the compiler is aware of some particular ordering. One way
|
238 |
|
|
* to make the compiler aware of ordering is to put the two invocations of
|
239 |
|
|
* ACCESS_ONCE() in different C statements.
|
240 |
|
|
*
|
241 |
|
|
* This macro does absolutely -nothing- to prevent the CPU from reordering,
|
242 |
|
|
* merging, or refetching absolutely anything at any time.
|
243 |
|
|
*/
|
244 |
|
|
#define ACCESS_ONCE(x) (*(volatile typeof(x) *)&(x))
|
245 |
|
|
|
246 |
|
|
/**
|
247 |
|
|
* rcu_dereference - fetch an RCU-protected pointer in an
|
248 |
|
|
* RCU read-side critical section. This pointer may later
|
249 |
|
|
* be safely dereferenced.
|
250 |
|
|
*
|
251 |
|
|
* Inserts memory barriers on architectures that require them
|
252 |
|
|
* (currently only the Alpha), and, more importantly, documents
|
253 |
|
|
* exactly which pointers are protected by RCU.
|
254 |
|
|
*/
|
255 |
|
|
|
256 |
|
|
#define rcu_dereference(p) ({ \
|
257 |
|
|
typeof(p) _________p1 = ACCESS_ONCE(p); \
|
258 |
|
|
smp_read_barrier_depends(); \
|
259 |
|
|
(_________p1); \
|
260 |
|
|
})
|
261 |
|
|
|
262 |
|
|
/**
|
263 |
|
|
* rcu_assign_pointer - assign (publicize) a pointer to a newly
|
264 |
|
|
* initialized structure that will be dereferenced by RCU read-side
|
265 |
|
|
* critical sections. Returns the value assigned.
|
266 |
|
|
*
|
267 |
|
|
* Inserts memory barriers on architectures that require them
|
268 |
|
|
* (pretty much all of them other than x86), and also prevents
|
269 |
|
|
* the compiler from reordering the code that initializes the
|
270 |
|
|
* structure after the pointer assignment. More importantly, this
|
271 |
|
|
* call documents which pointers will be dereferenced by RCU read-side
|
272 |
|
|
* code.
|
273 |
|
|
*/
|
274 |
|
|
|
275 |
|
|
#define rcu_assign_pointer(p, v) ({ \
|
276 |
|
|
smp_wmb(); \
|
277 |
|
|
(p) = (v); \
|
278 |
|
|
})
|
279 |
|
|
|
280 |
|
|
/**
|
281 |
|
|
* synchronize_sched - block until all CPUs have exited any non-preemptive
|
282 |
|
|
* kernel code sequences.
|
283 |
|
|
*
|
284 |
|
|
* This means that all preempt_disable code sequences, including NMI and
|
285 |
|
|
* hardware-interrupt handlers, in progress on entry will have completed
|
286 |
|
|
* before this primitive returns. However, this does not guarantee that
|
287 |
|
|
* softirq handlers will have completed, since in some kernels, these
|
288 |
|
|
* handlers can run in process context, and can block.
|
289 |
|
|
*
|
290 |
|
|
* This primitive provides the guarantees made by the (now removed)
|
291 |
|
|
* synchronize_kernel() API. In contrast, synchronize_rcu() only
|
292 |
|
|
* guarantees that rcu_read_lock() sections will have completed.
|
293 |
|
|
* In "classic RCU", these two guarantees happen to be one and
|
294 |
|
|
* the same, but can differ in realtime RCU implementations.
|
295 |
|
|
*/
|
296 |
|
|
#define synchronize_sched() synchronize_rcu()
|
297 |
|
|
|
298 |
|
|
extern void rcu_init(void);
|
299 |
|
|
extern void rcu_check_callbacks(int cpu, int user);
|
300 |
|
|
extern void rcu_restart_cpu(int cpu);
|
301 |
|
|
extern long rcu_batches_completed(void);
|
302 |
|
|
extern long rcu_batches_completed_bh(void);
|
303 |
|
|
|
304 |
|
|
/* Exported interfaces */
|
305 |
|
|
extern void FASTCALL(call_rcu(struct rcu_head *head,
|
306 |
|
|
void (*func)(struct rcu_head *head)));
|
307 |
|
|
extern void FASTCALL(call_rcu_bh(struct rcu_head *head,
|
308 |
|
|
void (*func)(struct rcu_head *head)));
|
309 |
|
|
extern void synchronize_rcu(void);
|
310 |
|
|
extern void rcu_barrier(void);
|
311 |
|
|
|
312 |
|
|
#endif /* __KERNEL__ */
|
313 |
|
|
#endif /* __LINUX_RCUPDATE_H */
|