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 - Alpha, ia64 and ARM architectures depend on 4.x gcc with atomic builtins
49 Usage of all urcu libraries
51 * Define _LGPL_SOURCE (only) if your code is LGPL or GPL compatible
52 before including the urcu.h or urcu-qsbr.h header. If your application
53 is distributed under another license, function calls will be generated
54 instead of inlines, so your application can link with the library.
55 * Linking with one of the libraries below is always necessary even for
56 LGPL and GPL applications.
61 * Link the application with "-lurcu".
62 * This is the preferred version of the library, in terms of
63 grace-period detection speed, read-side speed and flexibility.
64 Dynamically detects kernel support for sys_membarrier(). Falls back
65 on urcu-mb scheme if support is not present, which has slower
70 * #include <urcu-qsbr.h>
71 * Link with "-lurcu-qsbr".
72 * The QSBR flavor of RCU needs to have each reader thread executing
73 rcu_quiescent_state() periodically to progress. rcu_thread_online()
74 and rcu_thread_offline() can be used to mark long periods for which
75 the threads are not active. It provides the fastest read-side at the
76 expense of more intrusiveness in the application code.
81 * Compile any _LGPL_SOURCE code using this library with "-DRCU_MB".
82 * Link with "-lurcu-mb".
83 * This version of the urcu library uses memory barriers on the writer
84 and reader sides. This results in faster grace-period detection, but
85 results in slower reads.
87 Usage of liburcu-signal
90 * Compile any _LGPL_SOURCE code using this library with "-DRCU_SIGNAL".
91 * Link the application with "-lurcu-signal".
92 * Version of the library that requires a signal, typically SIGUSR1. Can
93 be overridden with -DSIGRCU by modifying Makefile.build.inc.
97 * #include <urcu-bp.h>
98 * Link with "-lurcu-bp".
99 * The BP library flavor stands for "bulletproof". It is specifically
100 designed to help tracing library to hook on applications without
101 requiring to modify these applications. rcu_init(),
102 rcu_register_thread() and rcu_unregister_thread() all become nops.
103 The state is dealt with by the library internally at the expense of
104 read-side and write-side performance.
108 Each thread that has reader critical sections (that uses
109 rcu_read_lock()/rcu_read_unlock() must first register to the URCU
110 library. This is done by calling rcu_register_thread(). Unregistration
111 must be performed before exiting the thread by using
112 rcu_unregister_thread().
116 Reader critical sections must be protected by locating them between
117 calls to rcu_read_lock() and rcu_read_unlock(). Inside that lock,
118 rcu_dereference() may be called to read an RCU protected pointer.
122 rcu_assign_pointer() and rcu_xchg_pointer() may be called anywhere.
123 After, synchronize_rcu() must be called. When it returns, the old
124 values are not in usage anymore.
126 Usage of liburcu-defer
128 * Follow instructions for either liburcu, liburcu-qsbr,
129 liburcu-mb, liburcu-signal, or liburcu-bp above.
130 The liburcu-defer functionality is pulled into each of
131 those library modules.
132 * Provides defer_rcu() primitive to enqueue delayed callbacks. Queued
133 callbacks are executed in batch periodically after a grace period.
134 Do _not_ use defer_rcu() within a read-side critical section, because
135 it may call synchronize_rcu() if the thread queue is full.
136 This can lead to deadlock or worse.
137 * Requires that rcu_defer_barrier() must be called in library destructor
138 if a library queues callbacks and is expected to be unloaded with
140 * Its API is currently experimental. It may change in future library
143 Usage of urcu-call-rcu
145 * Follow instructions for either liburcu, liburcu-qsbr,
146 liburcu-mb, liburcu-signal, or liburcu-bp above.
147 The urcu-call-rcu functionality is provided for each of
148 these library modules.
149 * Provides the call_rcu() primitive to enqueue delayed callbacks
150 in a manner similar to defer_rcu(), but without ever delaying
151 for a grace period. On the other hand, call_rcu()'s best-case
152 overhead is not quite as good as that of defer_rcu().
153 * Provides call_rcu() to allow asynchronous handling of RCU
154 grace periods. A number of additional functions are provided
155 to manage the helper threads used by call_rcu(), but reasonable
156 defaults are used if these additional functions are not invoked.
157 See API.txt for more details.
159 Being careful with signals
161 The liburcu library uses signals internally. The signal handler is
162 registered with the SA_RESTART flag. However, these signals may cause
163 some non-restartable system calls to fail with errno = EINTR. Care
164 should be taken to restart system calls manually if they fail with this
165 error. A list of non-restartable system calls may be found in
166 signal(7). The liburcu-mb and liburcu-qsbr versions of the Userspace RCU
167 library do not require any signal.
169 Read-side critical sections are allowed in a signal handler with
170 liburcu and liburcu-mb. Be careful, however, to disable these signals
171 between thread creation and calls to rcu_register_thread(), because a
172 signal handler nesting on an unregistered thread would not be allowed to
173 call rcu_read_lock().
175 Read-side critical sections are _not_ allowed in a signal handler with
176 liburcu-qsbr, unless signals are disabled explicitly around each
177 rcu_quiescent_state() calls, when threads are put offline and around
178 calls to synchronize_rcu(). Even then, we do not recommend it.
180 Interaction with mutexes
182 One must be careful to do not cause deadlocks due to interaction of
183 synchronize_rcu() and RCU read-side with mutexes. If synchronize_rcu()
184 is called with a mutex held, this mutex (or any mutex which has this
185 mutex in its dependency chain) should not be acquired from within a RCU
186 read-side critical section.
190 DEBUG_RCU is used to add internal debugging self-checks to the
191 RCU library. This define adds a performance penalty when enabled.
192 Can be enabled by uncommenting the corresponding line in
197 DEBUG_YIELD is used to add random delays in the code for testing
202 By default the library is configured to use synchronization primitives
203 adequate for SMP systems. On uniprocessor systems, support for SMP
204 systems can be disabled with:
206 ./configure --disable-smp-support
208 theoretically yielding slightly better performance.
210 Interaction with fork()
212 Special care must be taken for applications performing fork() without
213 any following exec(). This is caused by the fact that Linux only clones
214 the thread calling fork(), and thus never replicates any of the other
215 parent thread into the child process. Most liburcu implementations
216 require that all registrations (as reader, defer_rcu and call_rcu
217 threads) should be released before a fork() is performed, except for the
218 rather common scenario where fork() is immediately followed by exec() in
219 the child process. The only implementation not subject to that rule is
220 liburcu-bp, which is designed to handle fork() by calling
221 rcu_bp_before_fork, rcu_bp_after_fork_parent and
222 rcu_bp_after_fork_child.
224 Applications that use call_rcu() and that fork() without
225 doing an immediate exec() must take special action. The parent
226 must invoke call_rcu_before_fork() before the fork() and
227 call_rcu_after_fork_parent() after the fork(). The child
228 process must invoke call_rcu_after_fork_child().
229 These three APIs are suitable for passing to pthread_atfork().