1 Userspace RCU Implementation
2 ============================
4 by Mathieu Desnoyers and Paul E. McKenney
10 ./bootstrap # skip if using tarball
18 - Forcing 32-bit build:
20 CFLAGS="-m32 -g -O2" ./configure
22 - Forcing 64-bit build:
24 CFLAGS="-m64 -g -O2" ./configure
26 - Forcing a 32-bit build with 386 backward compatibility:
28 CFLAGS="-m32 -g -O2" ./configure --host=i386-pc-linux-gnu
30 - Forcing a 32-bit build for Sparcv9 (typical for Sparc v9)
32 CFLAGS="-m32 -Wa,-Av9a -g -O2" ./configure
35 Architectures supported
36 -----------------------
38 Currently, the following architectures are supported:
40 - x86 (i386, i486, i586, i686)
57 - Linux all architectures
58 - FreeBSD 8.2/8.3/9.0/9.1/10.0 i386/amd64
69 (more testing needed before claiming support for these OS).
71 Linux ARM depends on running a Linux kernel 2.6.15 or better, GCC 4.4 or
74 The C compiler used needs to support at least C99. The C++ compiler used
75 needs to support at least C++11.
77 The GCC compiler versions 3.3, 3.4, 4.0, 4.1, 4.2, 4.3, 4.4 and 4.5 are
78 supported, with the following exceptions:
80 - GCC 3.3 and 3.4 have a bug that prevents them from generating volatile
81 accesses to offsets in a TLS structure on 32-bit x86. These versions are
82 therefore not compatible with `liburcu` on x86 32-bit
83 (i386, i486, i586, i686).
84 The problem has been reported to the GCC community:
85 <http://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg281255.html>
86 - GCC 3.3 cannot match the "xchg" instruction on 32-bit x86 build.
87 See <http://kerneltrap.org/node/7507>
88 - Alpha, ia64 and ARM architectures depend on GCC 4.x with atomic builtins
89 support. For ARM this was introduced with GCC 4.4:
90 <http://gcc.gnu.org/gcc-4.4/changes.html>.
91 - Linux aarch64 depends on GCC 5.1 or better because prior versions
92 perform unsafe access to deallocated stack.
94 Clang version 3.0 (based on LLVM 3.0) is supported.
96 Glibc >= 2.4 should work but the older version we test against is
99 For developers using the Git tree:
101 This source tree is based on the autotools suite from GNU to simplify
102 portability. Here are some things you should have on your system in order to
103 compile the git repository tree :
105 - GNU autotools (automake >=1.12, autoconf >=2.69)
106 (make sure your system wide `automake` points to a recent version!)
108 (for more information, go to <http://www.gnu.org/software/autoconf/>)
110 If you get the tree from the repository, you will need to use the `bootstrap`
111 script in the root of the tree. It calls all the GNU tools needed to prepare
112 the tree configuration.
114 Test scripts provided in the `tests/` directory of the source tree depend
115 on `bash` and the `seq` program.
121 See the relevant API documentation files in `doc/`. The APIs provided by
122 Userspace RCU are, by prefix:
124 - `rcu_`: Read-Copy Update (see [`doc/rcu-api.md`](doc/rcu-api.md))
125 - `cmm_`: Concurrent Memory Model
126 - `caa_`: Concurrent Architecture Abstraction
127 - `cds_`: Concurrent Data Structures
128 (see [`doc/cds-api.md`](doc/cds-api.md))
129 - `uatomic_`: Userspace Atomic
130 (see [`doc/uatomic-api.md`](doc/uatomic-api.md))
136 ### Usage of all urcu libraries:
138 - Define `_LGPL_SOURCE` (only) if your code is LGPL or GPL compatible
139 before including the `urcu.h` or `urcu-qsbr.h` header. If your application
140 is distributed under another license, function calls will be generated
141 instead of inlines, so your application can link with the library.
142 - Linking with one of the libraries below is always necessary even for
143 LGPL and GPL applications.
144 - Define `URCU_INLINE_SMALL_FUNCTIONS` before including Userspace RCU
145 headers if you want Userspace RCU to inline small functions (10
146 lines or less) into the application. It can be used by applications
147 distributed under any kind of license, and does *not* make the
148 application a derived work of Userspace RCU.
150 Those small inlined functions are guaranteed to match the library
151 content as long as the library major version is unchanged.
152 Therefore, the application *must* be compiled with headers matching
153 the library major version number. Applications using
154 `URCU_INLINE_SMALL_FUNCTIONS` may be unable to use debugging
155 features of Userspace RCU without being recompiled.
157 There are multiple flavors of liburcu available:
165 The API members start with the prefix `urcu_<flavor>_`, where
166 `<flavor>` is the chosen flavor name.
169 ### Usage of `liburcu-memb`
171 1. `#include <urcu/urcu-memb.h>`
172 2. Link the application with `-lurcu-memb`
174 This is the preferred version of the library, in terms of
175 grace-period detection speed, read-side speed and flexibility.
176 Dynamically detects kernel support for `sys_membarrier()`. Falls back
177 on `urcu-mb` scheme if support is not present, which has slower
178 read-side. Use the `--disable-sys-membarrier-fallback` configure option
179 to disable the fall back, thus requiring `sys_membarrier()` to be
180 available. This gives a small speedup when `sys_membarrier()` is
181 supported by the kernel, and aborts in the library constructor if not
185 ### Usage of `liburcu-qsbr`
187 1. `#include <urcu/urcu-qsbr.h>`
188 2. Link with `-lurcu-qsbr`
190 The QSBR flavor of RCU needs to have each reader thread executing
191 `rcu_quiescent_state()` periodically to progress. `rcu_thread_online()`
192 and `rcu_thread_offline()` can be used to mark long periods for which
193 the threads are not active. It provides the fastest read-side at the
194 expense of more intrusiveness in the application code.
197 ### Usage of `liburcu-mb`
199 1. `#include <urcu/urcu-mb.h>`
200 2. Link with `-lurcu-mb`
202 This version of the urcu library uses memory barriers on the writer
203 and reader sides. This results in faster grace-period detection, but
204 results in slower reads.
207 ### Usage of `liburcu-signal`
209 1. `#include <urcu/urcu-signal.h>`
210 2. Link the application with `-lurcu-signal`
212 Version of the library that requires a signal, typically `SIGUSR1`. Can
213 be overridden with `-DSIGRCU` by modifying `Makefile.build.inc`.
216 ### Usage of `liburcu-bp`
218 1. `#include <urcu/urcu-bp.h>`
219 2. Link with `-lurcu-bp`
221 The BP library flavor stands for "bulletproof". It is specifically
222 designed to help tracing library to hook on applications without
223 requiring to modify these applications. `urcu_bp_init()`, and
224 `urcu_bp_unregister_thread()` all become nops, whereas calling
225 `urcu_bp_register_thread()` becomes optional. The state is dealt with by
226 the library internally at the expense of read-side and write-side
232 Each thread that has reader critical sections (that uses
233 `urcu_<flavor>_read_lock()`/`urcu_<flavor>_read_unlock()` must first
234 register to the URCU library. This is done by calling
235 `urcu_<flavor>_register_thread()`. Unregistration must be performed
236 before exiting the thread by using `urcu_<flavor>_unregister_thread()`.
241 Reader critical sections must be protected by locating them between
242 calls to `urcu_<flavor>_read_lock()` and `urcu_<flavor>_read_unlock()`.
243 Inside that lock, `rcu_dereference()` may be called to read an RCU
249 `rcu_assign_pointer()` and `rcu_xchg_pointer()` may be called anywhere.
250 After, `urcu_<flavor>_synchronize_rcu()` must be called. When it
251 returns, the old values are not in usage anymore.
254 ### Usage of `liburcu-defer`
256 - Follow instructions for either `liburcu-memb`, `liburcu-qsbr`,
257 `liburcu-mb`, `liburcu-signal`, or `liburcu-bp` above.
258 The `liburcu-defer` functionality is pulled into each of
259 those library modules.
260 - Provides `urcu_<flavor>_defer_rcu()` primitive to enqueue delayed
261 callbacks. Queued callbacks are executed in batch periodically after
262 a grace period. Do _not_ use `urcu_<flavor>_defer_rcu()` within a
263 read-side critical section, because it may call
264 `urcu_<flavor>_synchronize_rcu()` if the thread queue is full. This
265 can lead to deadlock or worse.
266 - Requires that `urcu_<flavor>_defer_barrier()` must be called in
267 library destructor if a library queues callbacks and is expected to
268 be unloaded with `dlclose()`.
270 Its API is currently experimental. It may change in future library releases.
273 ### Usage of `urcu-call-rcu`
275 - Follow instructions for either `liburcu-memb`, `liburcu-qsbr`,
276 `liburcu-mb`, `liburcu-signal`, or `liburcu-bp` above.
277 The `urcu-call-rcu` functionality is pulled into each of
278 those library modules.
279 - Provides the `urcu_<flavor>_call_rcu()` primitive to enqueue delayed
280 callbacks in a manner similar to `urcu_<flavor>_defer_rcu()`, but
281 without ever delaying for a grace period. On the other hand,
282 `urcu_<flavor>_call_rcu()`'s best-case overhead is not quite as good
283 as that of `urcu_<flavor>_defer_rcu()`.
284 - Provides `urcu_<flavor>_call_rcu()` to allow asynchronous handling
285 of RCU grace periods. A number of additional functions are provided
286 to manage the helper threads used by `urcu_<flavor>_call_rcu()`, but
287 reasonable defaults are used if these additional functions are not
288 invoked. See [`doc/rcu-api.md`](doc/rcu-api.md) in userspace-rcu
289 documentation for more details.
292 ### Being careful with signals
294 The `liburcu-signal` library uses signals internally. The signal handler is
295 registered with the `SA_RESTART` flag. However, these signals may cause
296 some non-restartable system calls to fail with `errno = EINTR`. Care
297 should be taken to restart system calls manually if they fail with this
298 error. A list of non-restartable system calls may be found in
301 Read-side critical sections are allowed in a signal handler,
302 except those setup with `sigaltstack(2)`, with `liburcu-memb` and
303 `liburcu-mb`. Be careful, however, to disable these signals
304 between thread creation and calls to `urcu_<flavor>_register_thread()`,
305 because a signal handler nesting on an unregistered thread would not be
306 allowed to call `urcu_<flavor>_read_lock()`.
308 Read-side critical sections are _not_ allowed in a signal handler with
309 `liburcu-qsbr`, unless signals are disabled explicitly around each
310 `urcu_qsbr_quiescent_state()` calls, when threads are put offline and around
311 calls to `urcu_qsbr_synchronize_rcu()`. Even then, we do not recommend it.
314 ### Interaction with mutexes
316 One must be careful to do not cause deadlocks due to interaction of
317 `urcu_<flavor>_synchronize_rcu()` and RCU read-side with mutexes. If
318 `urcu_<flavor>_synchronize_rcu()` is called with a mutex held, this
319 mutex (or any mutex which has this mutex in its dependency chain) should
320 not be acquired from within a RCU read-side critical section.
322 This is especially important to understand in the context of the
323 QSBR flavor: a registered reader thread being "online" by
324 default should be considered as within a RCU read-side critical
325 section unless explicitly put "offline". Therefore, if
326 `urcu_qsbr_synchronize_rcu()` is called with a mutex held, this mutex,
327 as well as any mutex which has this mutex in its dependency chain should
328 only be taken when the RCU reader thread is "offline" (this can be
329 performed by calling `urcu_qsbr_thread_offline()`).
332 ### Interaction with `fork()`
334 Special care must be taken for applications performing `fork()` without
335 any following `exec()`. This is caused by the fact that Linux only clones
336 the thread calling `fork()`, and thus never replicates any of the other
337 parent thread into the child process. Most `liburcu` implementations
338 require that all registrations (as reader, `defer_rcu` and `call_rcu`
339 threads) should be released before a `fork()` is performed, except for the
340 rather common scenario where `fork()` is immediately followed by `exec()` in
341 the child process. The only implementation not subject to that rule is
342 `liburcu-bp`, which is designed to handle `fork()` by calling
343 `urcu_bp_before_fork`, `urcu_bp_after_fork_parent` and
344 `urcu_bp_after_fork_child`.
346 Applications that use `urcu_<flavor>_call_rcu()` and that `fork()`
347 without doing an immediate `exec()` must take special action. The
348 parent must invoke `urcu_<flavor>_call_rcu_before_fork()` before the
349 `fork()` and `urcu_<flavor>_call_rcu_after_fork_parent()` after the
350 `fork()`. The child process must invoke
351 `urcu_<flavor>_call_rcu_after_fork_child()`. Even though these three
352 APIs are suitable for passing to `pthread_atfork()`, use of
353 `pthread_atfork()` is **STRONGLY DISCOURAGED** for programs calling the
354 glibc memory allocator (`malloc()`, `calloc()`, `free()`, ...) within
355 `urcu_<flavor>_call_rcu` callbacks. This is due to limitations in the
356 way glibc memory allocator handles calls to the memory allocator from
357 concurrent threads while the `pthread_atfork()` handlers are executing.
361 - call to `free()` from callbacks executed within
362 `urcu_<flavor>_call_rcu` worker threads,
363 - executing `urcu_<flavor>_call_rcu` atfork handlers within the glibc
364 pthread atfork mechanism,
366 will sometimes trigger interesting process hangs. This usually
367 hangs on a memory allocator lock within glibc.
370 ### Thread Local Storage (TLS)
372 Userspace RCU can fall back on `pthread_getspecific()` to emulate
373 TLS variables on systems where it is not available. This behavior
374 can be forced by specifying `--disable-compiler-tls` as configure
378 ### Usage of `DEBUG_RCU` & `--enable-rcu-debug`
380 By default the library is configured with internal debugging
381 self-checks disabled.
383 For always-on debugging self-checks:
385 ./configure --enable-rcu-debug
387 For fine grained enabling of debugging self-checks, build
388 userspace-rcu with `DEBUG_RCU` defined and compile dependent
389 applications with `DEBUG_RCU` defined when necessary.
391 Warning: Enabling this feature result in a performance penalty.
394 ### Usage of `DEBUG_YIELD`
396 `DEBUG_YIELD` is used to add random delays in the code for testing
402 By default the library is configured to use synchronization primitives
403 adequate for SMP systems. On uniprocessor systems, support for SMP
404 systems can be disabled with:
406 ./configure --disable-smp-support
408 theoretically yielding slightly better performance.
411 ### Usage of `--enable-cds-lfht-iter-debug`
413 By default the library is configured with extra debugging checks for
414 lock-free hash table iterator traversal disabled.
416 Building liburcu with `--enable-cds-lfht-iter-debug` and rebuilding
417 application to match the ABI change allows finding cases where the hash
418 table iterator is re-purposed to be used on a different hash table while
419 still being used to iterate on a hash table.
421 This option alters the rculfhash ABI. Make sure to compile both library
422 and application with matching configuration.
428 In addition to the usual `make check` target, Userspace RCU features
429 `make regtest`, `make short_bench` and `make long_bench` targets:
431 - `make check`: short tests, meant to be run when rebuilding or
432 porting Userspace RCU.
433 - `make regtest`: long (many hours) test, meant to be run when
434 modifying Userspace RCU or porting it to a new architecture or
436 - `make short_bench`: short benchmarks, 3 seconds per test.
437 - `make long_bench`: long (many hours) benchmarks, 30 seconds per test.
443 There is an application vs library compatibility issue between
444 applications built using Userspace RCU 0.10 headers linked against
445 Userspace RCU 0.11 or 0.12 shared objects. The problem occurs as
448 - An application executable is built with `_LGPL_SOURCE` defined, includes
449 any of the Userspace RCU 0.10 urcu flavor headers, and is built
450 without the `-fpic` compiler option.
452 - The Userspace RCU 0.10 library shared objects are updated to 0.11
453 or 0.12 without rebuilding the application.
455 - The application will hang, typically when RCU grace period
456 (synchronize_rcu) is invoked.
458 Some possible work-arounds for this are:
460 - Rebuild the application against Userspace RCU 0.11+.
462 - Rebuild the application with `-fpic`.
464 - Upgrade Userspace RCU to 0.13+ without installing 0.11 nor 0.12.
470 You can contact the maintainers on the following mailing list:
471 `lttng-dev@lists.lttng.org`.