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
26 ARCHITECTURES SUPPORTED
27 -----------------------
29 Currently, Linux x86 (i386, i486, i586, i686), x86 64-bit, PowerPC 32/64,
30 S390, S390x, ARM 32/64, MIPS, Alpha, ia64, Sparcv9 32/64, Tilera, and
31 hppa/PA-RISC are supported.
32 Tested on Linux, FreeBSD 8.2/8.3/9.0/9.1/10.0 i386/amd64, and Cygwin.
33 Should also work on: Android, NetBSD 5, OpenBSD, Darwin (more testing
34 needed before claiming support for these OS).
36 Linux ARM depends on running a Linux kernel 2.6.15 or better, GCC 4.4 or
39 The gcc compiler versions 3.3, 3.4, 4.0, 4.1, 4.2, 4.3, 4.4 and 4.5 are
40 supported, with the following exceptions:
42 - gcc 3.3 and 3.4 have a bug that prevents them from generating volatile
43 accesses to offsets in a TLS structure on 32-bit x86. These versions are
44 therefore not compatible with liburcu on x86 32-bit (i386, i486, i586, i686).
45 The problem has been reported to the gcc community:
46 http://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg281255.html
47 - gcc 3.3 cannot match the "xchg" instruction on 32-bit x86 build.
48 See: http://kerneltrap.org/node/7507
49 - Alpha, ia64 and ARM architectures depend on gcc 4.x with atomic builtins
50 support. For ARM this was introduced with gcc 4.4:
51 http://gcc.gnu.org/gcc-4.4/changes.html
53 Clang version 3.0 (based on LLVM 3.0) is supported.
55 Building on MacOS X (Darwin) requires a work-around for processor
58 ./configure --build=i686-apple-darwin11
60 ./configure --build=x86_64-apple-darwin11
62 For developers using the git tree:
64 This source tree is based on the autotools suite from GNU to simplify
65 portability. Here are some things you should have on your system in order to
66 compile the git repository tree :
68 - GNU autotools (automake >=1.10, autoconf >=2.50, autoheader >=2.50)
69 (make sure your system wide "automake" points to a recent version!)
71 (for more information, go to http://www.gnu.org/software/autoconf/)
73 If you get the tree from the repository, you will need to use the "bootstrap"
74 script in the root of the tree. It calls all the GNU tools needed to prepare the
77 Test scripts provided in the tests/ directory of the source tree depend
78 on "bash" and the "seq" program.
84 See the relevant API documentation files in doc/. The APIs provided by
85 Userspace RCU are, by prefix:
87 - rcu_ : Read-Copy Update (see doc/rcu-api.txt)
88 - cmm_ : Concurrent Memory Model
89 - caa_ : Concurrent Architecture Abstraction
90 - cds_ : Concurrent Data Structures (see doc/cds-api.txt)
91 - uatomic_: Userspace Atomic (see doc/uatomic-api.txt)
97 Usage of all urcu libraries
99 * Define _LGPL_SOURCE (only) if your code is LGPL or GPL compatible
100 before including the urcu.h or urcu-qsbr.h header. If your application
101 is distributed under another license, function calls will be generated
102 instead of inlines, so your application can link with the library.
103 * Linking with one of the libraries below is always necessary even for
104 LGPL and GPL applications.
105 * Define URCU_INLINE_SMALL_FUNCTIONS before including Userspace RCU
106 headers if you want Userspace RCU to inline small functions (10
107 lines or less) into the application. It can be used by applications
108 distributed under any kind of license, and does *not* make the
109 application a derived work of Userspace RCU.
111 Those small inlined functions are guaranteed to match the library
112 content as long as the library major version is unchanged.
113 Therefore, the application *must* be compiled with headers matching
114 the library major version number. Applications using
115 URCU_INLINE_SMALL_FUNCTIONS may be unable to use debugging
116 features of Userspace RCU without being recompiled.
122 * Link the application with "-lurcu".
123 * This is the preferred version of the library, in terms of
124 grace-period detection speed, read-side speed and flexibility.
125 Dynamically detects kernel support for sys_membarrier(). Falls back
126 on urcu-mb scheme if support is not present, which has slower
129 Usage of liburcu-qsbr
131 * #include <urcu-qsbr.h>
132 * Link with "-lurcu-qsbr".
133 * The QSBR flavor of RCU needs to have each reader thread executing
134 rcu_quiescent_state() periodically to progress. rcu_thread_online()
135 and rcu_thread_offline() can be used to mark long periods for which
136 the threads are not active. It provides the fastest read-side at the
137 expense of more intrusiveness in the application code.
142 * Compile any _LGPL_SOURCE code using this library with "-DRCU_MB".
143 * Link with "-lurcu-mb".
144 * This version of the urcu library uses memory barriers on the writer
145 and reader sides. This results in faster grace-period detection, but
146 results in slower reads.
148 Usage of liburcu-signal
151 * Compile any _LGPL_SOURCE code using this library with "-DRCU_SIGNAL".
152 * Link the application with "-lurcu-signal".
153 * Version of the library that requires a signal, typically SIGUSR1. Can
154 be overridden with -DSIGRCU by modifying Makefile.build.inc.
158 * #include <urcu-bp.h>
159 * Link with "-lurcu-bp".
160 * The BP library flavor stands for "bulletproof". It is specifically
161 designed to help tracing library to hook on applications without
162 requiring to modify these applications. rcu_init(),
163 rcu_register_thread() and rcu_unregister_thread() all become nops.
164 The state is dealt with by the library internally at the expense of
165 read-side and write-side performance.
169 Each thread that has reader critical sections (that uses
170 rcu_read_lock()/rcu_read_unlock() must first register to the URCU
171 library. This is done by calling rcu_register_thread(). Unregistration
172 must be performed before exiting the thread by using
173 rcu_unregister_thread().
177 Reader critical sections must be protected by locating them between
178 calls to rcu_read_lock() and rcu_read_unlock(). Inside that lock,
179 rcu_dereference() may be called to read an RCU protected pointer.
183 rcu_assign_pointer() and rcu_xchg_pointer() may be called anywhere.
184 After, synchronize_rcu() must be called. When it returns, the old
185 values are not in usage anymore.
187 Usage of liburcu-defer
189 * Follow instructions for either liburcu, liburcu-qsbr,
190 liburcu-mb, liburcu-signal, or liburcu-bp above.
191 The liburcu-defer functionality is pulled into each of
192 those library modules.
193 * Provides defer_rcu() primitive to enqueue delayed callbacks. Queued
194 callbacks are executed in batch periodically after a grace period.
195 Do _not_ use defer_rcu() within a read-side critical section, because
196 it may call synchronize_rcu() if the thread queue is full.
197 This can lead to deadlock or worse.
198 * Requires that rcu_defer_barrier() must be called in library destructor
199 if a library queues callbacks and is expected to be unloaded with
201 * Its API is currently experimental. It may change in future library
204 Usage of urcu-call-rcu
206 * Follow instructions for either liburcu, liburcu-qsbr,
207 liburcu-mb, liburcu-signal, or liburcu-bp above.
208 The urcu-call-rcu functionality is provided for each of
209 these library modules.
210 * Provides the call_rcu() primitive to enqueue delayed callbacks
211 in a manner similar to defer_rcu(), but without ever delaying
212 for a grace period. On the other hand, call_rcu()'s best-case
213 overhead is not quite as good as that of defer_rcu().
214 * Provides call_rcu() to allow asynchronous handling of RCU
215 grace periods. A number of additional functions are provided
216 to manage the helper threads used by call_rcu(), but reasonable
217 defaults are used if these additional functions are not invoked.
218 See rcu-api.txt in userspace-rcu documentation for more details.
220 Being careful with signals
222 The liburcu library uses signals internally. The signal handler is
223 registered with the SA_RESTART flag. However, these signals may cause
224 some non-restartable system calls to fail with errno = EINTR. Care
225 should be taken to restart system calls manually if they fail with this
226 error. A list of non-restartable system calls may be found in
227 signal(7). The liburcu-mb and liburcu-qsbr versions of the Userspace RCU
228 library do not require any signal.
230 Read-side critical sections are allowed in a signal handler,
231 except those setup with sigaltstack(2), with liburcu and
232 liburcu-mb. Be careful, however, to disable these signals
233 between thread creation and calls to rcu_register_thread(), because a
234 signal handler nesting on an unregistered thread would not be
235 allowed to call rcu_read_lock().
237 Read-side critical sections are _not_ allowed in a signal handler with
238 liburcu-qsbr, unless signals are disabled explicitly around each
239 rcu_quiescent_state() calls, when threads are put offline and around
240 calls to synchronize_rcu(). Even then, we do not recommend it.
242 Interaction with mutexes
244 One must be careful to do not cause deadlocks due to interaction of
245 synchronize_rcu() and RCU read-side with mutexes. If synchronize_rcu()
246 is called with a mutex held, this mutex (or any mutex which has this
247 mutex in its dependency chain) should not be acquired from within a RCU
248 read-side critical section.
250 This is especially important to understand in the context of the
251 QSBR flavor: a registered reader thread being "online" by
252 default should be considered as within a RCU read-side critical
253 section unless explicitly put "offline". Therefore, if
254 synchronize_rcu() is called with a mutex held, this mutex, as
255 well as any mutex which has this mutex in its dependency chain
256 should only be taken when the RCU reader thread is "offline"
257 (this can be performed by calling rcu_thread_offline()).
259 Interaction with fork()
261 Special care must be taken for applications performing fork() without
262 any following exec(). This is caused by the fact that Linux only clones
263 the thread calling fork(), and thus never replicates any of the other
264 parent thread into the child process. Most liburcu implementations
265 require that all registrations (as reader, defer_rcu and call_rcu
266 threads) should be released before a fork() is performed, except for the
267 rather common scenario where fork() is immediately followed by exec() in
268 the child process. The only implementation not subject to that rule is
269 liburcu-bp, which is designed to handle fork() by calling
270 rcu_bp_before_fork, rcu_bp_after_fork_parent and
271 rcu_bp_after_fork_child.
273 Applications that use call_rcu() and that fork() without
274 doing an immediate exec() must take special action. The parent
275 must invoke call_rcu_before_fork() before the fork() and
276 call_rcu_after_fork_parent() after the fork(). The child
277 process must invoke call_rcu_after_fork_child().
278 Even though these three APIs are suitable for passing to
279 pthread_atfork(), use of pthread_atfork() is *STRONGLY
280 DISCOURAGED* for programs calling the glibc memory allocator
281 (malloc(), calloc(), free(), ...) within call_rcu callbacks.
282 This is due to limitations in the way glibc memory allocator
283 handles calls to the memory allocator from concurrent threads
284 while the pthread_atfork() handlers are executing.
286 * call to free() from callbacks executed within call_rcu worker
288 * executing call_rcu atfork handlers within the glibc pthread
290 will sometimes trigger interesting process hangs. This usually
291 hangs on a memory allocator lock within glibc.
293 Thread Local Storage (TLS)
295 Userspace RCU can fall back on pthread_getspecific() to emulate
296 TLS variables on systems where it is not available. This behavior
297 can be forced by specifying --disable-compiler-tls as configure
302 DEBUG_RCU is used to add internal debugging self-checks to the
303 RCU library. This define adds a performance penalty when enabled.
304 Can be enabled by uncommenting the corresponding line in
309 DEBUG_YIELD is used to add random delays in the code for testing
314 By default the library is configured to use synchronization primitives
315 adequate for SMP systems. On uniprocessor systems, support for SMP
316 systems can be disabled with:
318 ./configure --disable-smp-support
320 theoretically yielding slightly better performance.
325 In addition to the usual "make check" target, Userspace RCU features
326 "make regtest" and "make bench" targets.
328 make check: Short tests, meant to be run when rebuilding or porting
331 make regtest: Long (many hours) test, meant to be run when modifying
332 Userspace RCU or porting it to a new architecture or
335 make bench: Long (many hours) benchmarks.
340 You can contact the maintainers on the following mailing list:
341 lttng-dev@lists.lttng.org