1 Userspace RCU Implementation
2 by Mathieu Desnoyers and Paul E. McKenney
7 ./bootstrap (skip if using tarball)
13 Hints: Forcing 32-bit build:
14 * CFLAGS="-m32 -g -O2" ./configure
17 * CFLAGS="-m64 -g -O2" ./configure
19 Forcing a 32-bit build with 386 backward compatibility:
20 * CFLAGS="-m32 -g -O2" ./configure --host=i386-pc-linux-gnu
22 Forcing a 32-bit build for Sparcv9 (typical for Sparc v9)
23 * CFLAGS="-m32 -Wa,-Av9a -g -O2" ./configure
25 ARCHITECTURES SUPPORTED
26 -----------------------
28 Currently, x86 (i386, i486, i586, i686), x86 64-bit, PowerPC 32/64, S390, S390x,
29 ARM, Alpha, ia64 and Sparcv9 32/64 are supported. Only tested on Linux so
30 far, but should theoretically work on other operating systems.
32 ARM depends on running a Linux kernel 2.6.15 or better.
34 The gcc compiler versions 3.3, 3.4, 4.0, 4.1, 4.2, 4.3, 4.4 and 4.5 are
35 supported, with the following exceptions:
37 - gcc 3.3 and 3.4 have a bug that prevents them from generating volatile
38 accesses to offsets in a TLS structure on 32-bit x86. These versions are
39 therefore not compatible with liburcu on x86 32-bit (i386, i486, i586, i686).
40 The problem has been reported to the gcc community:
41 http://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg281255.html
42 - gcc 3.3 cannot match the "xchg" instruction on 32-bit x86 build.
43 See: http://kerneltrap.org/node/7507
44 - Alpha, ia64 and ARM architectures depend on 4.x gcc with atomic builtins
51 Usage of all urcu libraries
53 * Define _LGPL_SOURCE (only) if your code is LGPL or GPL compatible
54 before including the urcu.h or urcu-qsbr.h header. If your application
55 is distributed under another license, function calls will be generated
56 instead of inlines, so your application can link with the library.
57 * Linking with one of the libraries below is always necessary even for
58 LGPL and GPL applications.
63 * Link the application with "-lurcu".
64 * This is the preferred version of the library, in terms of
65 grace-period detection speed, read-side speed and flexibility.
66 Dynamically detects kernel support for sys_membarrier(). Falls back
67 on urcu-mb scheme if support is not present, which has slower
72 * #include <urcu-qsbr.h>
73 * Link with "-lurcu-qsbr".
74 * The QSBR flavor of RCU needs to have each reader thread executing
75 rcu_quiescent_state() periodically to progress. rcu_thread_online()
76 and rcu_thread_offline() can be used to mark long periods for which
77 the threads are not active. It provides the fastest read-side at the
78 expense of more intrusiveness in the application code.
83 * Compile any _LGPL_SOURCE code using this library with "-DRCU_MB".
84 * Link with "-lurcu-mb".
85 * This version of the urcu library uses memory barriers on the writer
86 and reader sides. This results in faster grace-period detection, but
87 results in slower reads.
89 Usage of liburcu-signal
92 * Compile any _LGPL_SOURCE code using this library with "-DRCU_SIGNAL".
93 * Link the application with "-lurcu-signal".
94 * Version of the library that requires a signal, typically SIGUSR1. Can
95 be overridden with -DSIGRCU by modifying Makefile.build.inc.
99 * #include <urcu-bp.h>
100 * Link with "-lurcu-bp".
101 * The BP library flavor stands for "bulletproof". It is specifically
102 designed to help tracing library to hook on applications without
103 requiring to modify these applications. rcu_init(),
104 rcu_register_thread() and rcu_unregister_thread() all become nops.
105 The state is dealt with by the library internally at the expense of
106 read-side and write-side performance.
110 Each thread that has reader critical sections (that uses
111 rcu_read_lock()/rcu_read_unlock() must first register to the URCU
112 library. This is done by calling rcu_register_thread(). Unregistration
113 must be performed before exiting the thread by using
114 rcu_unregister_thread().
118 Reader critical sections must be protected by locating them between
119 calls to rcu_read_lock() and rcu_read_unlock(). Inside that lock,
120 rcu_dereference() may be called to read an RCU protected pointer.
124 rcu_assign_pointer() and rcu_xchg_pointer() may be called anywhere.
125 After, synchronize_rcu() must be called. When it returns, the old
126 values are not in usage anymore.
128 Usage of liburcu-defer
130 * Follow instructions for either liburcu, liburcu-qsbr,
131 liburcu-mb, liburcu-signal, or liburcu-bp above.
132 The liburcu-defer functionality is pulled into each of
133 those library modules.
134 * Provides defer_rcu() primitive to enqueue delayed callbacks. Queued
135 callbacks are executed in batch periodically after a grace period.
136 Do _not_ use defer_rcu() within a read-side critical section, because
137 it may call synchronize_rcu() if the thread queue is full.
138 This can lead to deadlock or worse.
139 * Requires that rcu_defer_barrier() must be called in library destructor
140 if a library queues callbacks and is expected to be unloaded with
142 * Its API is currently experimental. It may change in future library
145 Usage of urcu-call-rcu
147 * Follow instructions for either liburcu, liburcu-qsbr,
148 liburcu-mb, liburcu-signal, or liburcu-bp above.
149 The urcu-call-rcu functionality is provided for each of
150 these library modules.
151 * Provides the call_rcu() primitive to enqueue delayed callbacks
152 in a manner similar to defer_rcu(), but without ever delaying
153 for a grace period. On the other hand, call_rcu()'s best-case
154 overhead is not quite as good as that of defer_rcu().
155 * Provides call_rcu() to allow asynchronous handling of RCU
156 grace periods. A number of additional functions are provided
157 to manage the helper threads used by call_rcu(), but reasonable
158 defaults are used if these additional functions are not invoked.
159 See API.txt for more details.
161 Being careful with signals
163 The liburcu library uses signals internally. The signal handler is
164 registered with the SA_RESTART flag. However, these signals may cause
165 some non-restartable system calls to fail with errno = EINTR. Care
166 should be taken to restart system calls manually if they fail with this
167 error. A list of non-restartable system calls may be found in
168 signal(7). The liburcu-mb and liburcu-qsbr versions of the Userspace RCU
169 library do not require any signal.
171 Read-side critical sections are allowed in a signal handler with
172 liburcu and liburcu-mb. Be careful, however, to disable these signals
173 between thread creation and calls to rcu_register_thread(), because a
174 signal handler nesting on an unregistered thread would not be allowed to
175 call rcu_read_lock().
177 Read-side critical sections are _not_ allowed in a signal handler with
178 liburcu-qsbr, unless signals are disabled explicitly around each
179 rcu_quiescent_state() calls, when threads are put offline and around
180 calls to synchronize_rcu(). Even then, we do not recommend it.
182 Interaction with mutexes
184 One must be careful to do not cause deadlocks due to interaction of
185 synchronize_rcu() and RCU read-side with mutexes. If synchronize_rcu()
186 is called with a mutex held, this mutex (or any mutex which has this
187 mutex in its dependency chain) should not be acquired from within a RCU
188 read-side critical section.
192 DEBUG_RCU is used to add internal debugging self-checks to the
193 RCU library. This define adds a performance penalty when enabled.
194 Can be enabled by uncommenting the corresponding line in
199 DEBUG_YIELD is used to add random delays in the code for testing
204 By default the library is configured to use synchronization primitives
205 adequate for SMP systems. On uniprocessor systems, support for SMP
206 systems can be disabled with:
208 ./configure --disable-smp-support
210 theoretically yielding slightly better performance.
212 Interaction with fork()
214 Special care must be taken for applications performing fork() without
215 any following exec(). This is caused by the fact that Linux only clones
216 the thread calling fork(), and thus never replicates any of the other
217 parent thread into the child process. Most liburcu implementations
218 require that all registrations (as reader, defer_rcu and call_rcu
219 threads) should be released before a fork() is performed, except for the
220 rather common scenario where fork() is immediately followed by exec() in
221 the child process. The only implementation not subject to that rule is
222 liburcu-bp, which is designed to handle fork() by calling
223 rcu_bp_before_fork, rcu_bp_after_fork_parent and
224 rcu_bp_after_fork_child.
226 Applications that use call_rcu() and that fork() without
227 doing an immediate exec() must take special action. The parent
228 must invoke call_rcu_before_fork() before the fork() and
229 call_rcu_after_fork_parent() after the fork(). The child
230 process must invoke call_rcu_after_fork_child().
231 These three APIs are suitable for passing to pthread_atfork().