Restrict supported arch ot P6+ on Intel x86 32.
[urcu.git] / urcu-static.h
CommitLineData
adcfce54
MD
1#ifndef _URCU_STATIC_H
2#define _URCU_STATIC_H
3
4/*
5 * urcu-static.h
6 *
d2d23035 7 * Userspace RCU header.
adcfce54 8 *
d2d23035
MD
9 * TO BE INCLUDED ONLY IN LGPL-COMPATIBLE CODE. See urcu.h for linking
10 * dynamically with the userspace rcu library.
adcfce54 11 *
d2d23035
MD
12 * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
13 * Copyright (c) 2009 Paul E. McKenney, IBM Corporation.
adcfce54 14 *
d2d23035
MD
15 * This library is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU Lesser General Public
17 * License as published by the Free Software Foundation; either
18 * version 2.1 of the License, or (at your option) any later version.
19 *
20 * This library is distributed in the hope that it will be useful,
21 * but WITHOUT ANY WARRANTY; without even the implied warranty of
22 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
23 * Lesser General Public License for more details.
24 *
25 * You should have received a copy of the GNU Lesser General Public
26 * License along with this library; if not, write to the Free Software
27 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
adcfce54
MD
28 *
29 * IBM's contributions to this file may be relicensed under LGPLv2 or later.
30 */
31
32#include <stdlib.h>
33#include <pthread.h>
bc6c15bb
MD
34#include <syscall.h>
35#include <unistd.h>
adcfce54 36
ec4e58a3
MD
37#include <urcu/compiler.h>
38#include <urcu/arch.h>
7e30abe3 39#include <urcu/system.h>
48d848c7 40#include <urcu/uatomic_arch.h>
e3b0cef0 41#include <urcu/list.h>
adcfce54 42
bc6c15bb
MD
43#define futex(...) syscall(__NR_futex, __VA_ARGS__)
44#define FUTEX_WAIT 0
45#define FUTEX_WAKE 1
46
adcfce54
MD
47/*
48 * This code section can only be included in LGPL 2.1 compatible source code.
49 * See below for the function call wrappers which can be used in code meant to
50 * be only linked with the Userspace RCU library. This comes with a small
51 * performance degradation on the read-side due to the added function calls.
52 * This is required to permit relinking with newer versions of the library.
53 */
54
55/*
56 * The signal number used by the RCU library can be overridden with
57 * -DSIGURCU= when compiling the library.
58 */
59#ifndef SIGURCU
60#define SIGURCU SIGUSR1
61#endif
62
63/*
64 * If a reader is really non-cooperative and refuses to commit its
65 * urcu_active_readers count to memory (there is no barrier in the reader
66 * per-se), kick it after a few loops waiting for it.
67 */
68#define KICK_READER_LOOPS 10000
69
bc6c15bb
MD
70/*
71 * Active attempts to check for reader Q.S. before calling futex().
72 */
73#define RCU_QS_ACTIVE_ATTEMPTS 100
74
7ac06cef
MD
75#ifdef DEBUG_RCU
76#define rcu_assert(args...) assert(args)
77#else
78#define rcu_assert(args...)
79#endif
80
adcfce54
MD
81#ifdef DEBUG_YIELD
82#include <sched.h>
83#include <time.h>
84#include <pthread.h>
85#include <unistd.h>
86
87#define YIELD_READ (1 << 0)
88#define YIELD_WRITE (1 << 1)
89
b4ce1526 90/*
0a1d290b 91 * Updates without URCU_MB are much slower. Account this in
b4ce1526
MD
92 * the delay.
93 */
0a1d290b 94#ifdef URCU_MB
adcfce54
MD
95/* maximum sleep delay, in us */
96#define MAX_SLEEP 50
97#else
98#define MAX_SLEEP 30000
99#endif
100
101extern unsigned int yield_active;
102extern unsigned int __thread rand_yield;
103
104static inline void debug_yield_read(void)
105{
106 if (yield_active & YIELD_READ)
107 if (rand_r(&rand_yield) & 0x1)
108 usleep(rand_r(&rand_yield) % MAX_SLEEP);
109}
110
111static inline void debug_yield_write(void)
112{
113 if (yield_active & YIELD_WRITE)
114 if (rand_r(&rand_yield) & 0x1)
115 usleep(rand_r(&rand_yield) % MAX_SLEEP);
116}
117
118static inline void debug_yield_init(void)
119{
120 rand_yield = time(NULL) ^ pthread_self();
121}
122#else
123static inline void debug_yield_read(void)
124{
125}
126
127static inline void debug_yield_write(void)
128{
129}
130
131static inline void debug_yield_init(void)
132{
133
134}
135#endif
136
0a1d290b 137#ifdef URCU_MB
adcfce54
MD
138static inline void reader_barrier()
139{
140 smp_mb();
141}
142#else
143static inline void reader_barrier()
144{
145 barrier();
146}
147#endif
148
149/*
150 * The trick here is that RCU_GP_CTR_BIT must be a multiple of 8 so we can use a
151 * full 8-bits, 16-bits or 32-bits bitmask for the lower order bits.
152 */
153#define RCU_GP_COUNT (1UL << 0)
154/* Use the amount of bits equal to half of the architecture long size */
155#define RCU_GP_CTR_BIT (1UL << (sizeof(long) << 2))
156#define RCU_GP_CTR_NEST_MASK (RCU_GP_CTR_BIT - 1)
157
158/*
159 * Global quiescent period counter with low-order bits unused.
160 * Using a int rather than a char to eliminate false register dependencies
161 * causing stalls on some architectures.
162 */
163extern long urcu_gp_ctr;
164
e3b0cef0 165struct urcu_reader {
bd1a5e15 166 /* Data used by both reader and synchronize_rcu() */
e3b0cef0 167 long ctr;
e3b0cef0 168 char need_mb;
bd1a5e15
MD
169 /* Data used for registry */
170 struct list_head head __attribute__((aligned(CACHE_LINE_SIZE)));
171 pthread_t tid;
e3b0cef0
MD
172};
173
174extern struct urcu_reader __thread urcu_reader;
adcfce54 175
bc6c15bb
MD
176extern int gp_futex;
177
178/*
179 * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
180 */
181static inline void wake_up_gp(void)
182{
ec4e58a3
MD
183 if (unlikely(uatomic_read(&gp_futex) == -1)) {
184 uatomic_set(&gp_futex, 0);
bc6c15bb
MD
185 futex(&gp_futex, FUTEX_WAKE, 1,
186 NULL, NULL, 0);
187 }
188}
189
adcfce54
MD
190static inline int rcu_old_gp_ongoing(long *value)
191{
192 long v;
193
194 if (value == NULL)
195 return 0;
196 /*
197 * Make sure both tests below are done on the same version of *value
198 * to insure consistency.
199 */
200 v = LOAD_SHARED(*value);
201 return (v & RCU_GP_CTR_NEST_MASK) &&
202 ((v ^ urcu_gp_ctr) & RCU_GP_CTR_BIT);
203}
204
205static inline void _rcu_read_lock(void)
206{
207 long tmp;
208
e3b0cef0 209 tmp = urcu_reader.ctr;
adcfce54 210 /* urcu_gp_ctr = RCU_GP_COUNT | (~RCU_GP_CTR_BIT or RCU_GP_CTR_BIT) */
67ef1a2c 211 if (likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
e3b0cef0 212 _STORE_SHARED(urcu_reader.ctr, _LOAD_SHARED(urcu_gp_ctr));
67ef1a2c
MD
213 /*
214 * Set active readers count for outermost nesting level before
215 * accessing the pointer. See force_mb_all_threads().
216 */
217 reader_barrier();
218 } else {
e3b0cef0 219 _STORE_SHARED(urcu_reader.ctr, tmp + RCU_GP_COUNT);
67ef1a2c 220 }
adcfce54
MD
221}
222
223static inline void _rcu_read_unlock(void)
224{
bc6c15bb
MD
225 long tmp;
226
e3b0cef0 227 tmp = urcu_reader.ctr;
adcfce54
MD
228 /*
229 * Finish using rcu before decrementing the pointer.
230 * See force_mb_all_threads().
231 */
bc6c15bb
MD
232 if (likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) {
233 reader_barrier();
e3b0cef0
MD
234 _STORE_SHARED(urcu_reader.ctr, urcu_reader.ctr - RCU_GP_COUNT);
235 /* write urcu_reader.ctr before read futex */
bc6c15bb
MD
236 reader_barrier();
237 wake_up_gp();
238 } else {
e3b0cef0 239 _STORE_SHARED(urcu_reader.ctr, urcu_reader.ctr - RCU_GP_COUNT);
bc6c15bb 240 }
adcfce54
MD
241}
242
adcfce54 243#endif /* _URCU_STATIC_H */
This page took 0.050555 seconds and 4 git commands to generate.