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
70 (more testing needed before claiming support for these OS).
72 Linux ARM depends on running a Linux kernel 2.6.15 or better, GCC 4.4 or
75 The GCC compiler versions 3.3, 3.4, 4.0, 4.1, 4.2, 4.3, 4.4 and 4.5 are
76 supported, with the following exceptions:
78 - GCC 3.3 and 3.4 have a bug that prevents them from generating volatile
79 accesses to offsets in a TLS structure on 32-bit x86. These versions are
80 therefore not compatible with `liburcu` on x86 32-bit
81 (i386, i486, i586, i686).
82 The problem has been reported to the GCC community:
83 http://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg281255.html
84 - GCC 3.3 cannot match the "xchg" instruction on 32-bit x86 build.
85 See http://kerneltrap.org/node/7507
86 - Alpha, ia64 and ARM architectures depend on GCC 4.x with atomic builtins
87 support. For ARM this was introduced with GCC 4.4:
88 http://gcc.gnu.org/gcc-4.4/changes.html.
89 - Linux aarch64 depends on GCC 5.1 or better because prior versions
90 perform unsafe access to deallocated stack.
92 Clang version 3.0 (based on LLVM 3.0) is supported.
94 Building on MacOS X (Darwin) requires a work-around for processor
99 ./configure --build=i686-apple-darwin11
103 ./configure --build=x86_64-apple-darwin11
105 For developers using the Git tree:
107 This source tree is based on the autotools suite from GNU to simplify
108 portability. Here are some things you should have on your system in order to
109 compile the git repository tree :
111 - GNU autotools (automake >=1.10, autoconf >=2.50, autoheader >=2.50)
112 (make sure your system wide `automake` points to a recent version!)
114 (for more information, go to http://www.gnu.org/software/autoconf/)
116 If you get the tree from the repository, you will need to use the `bootstrap`
117 script in the root of the tree. It calls all the GNU tools needed to prepare
118 the tree configuration.
120 Test scripts provided in the `tests/` directory of the source tree depend
121 on `bash` and the `seq` program.
127 See the relevant API documentation files in `doc/`. The APIs provided by
128 Userspace RCU are, by prefix:
130 - `rcu_`: Read-Copy Update (see [`doc/rcu-api.md`](doc/rcu-api.md))
131 - `cmm_`: Concurrent Memory Model
132 - `caa_`: Concurrent Architecture Abstraction
133 - `cds_`: Concurrent Data Structures
134 (see [`doc/cds-api.md`](doc/cds-api.md))
135 - `uatomic_`: Userspace Atomic
136 (see [`doc/uatomic-api.md`](doc/uatomic-api.md))
142 ### Usage of all urcu libraries:
144 - Define `_LGPL_SOURCE` (only) if your code is LGPL or GPL compatible
145 before including the `urcu.h` or `urcu-qsbr.h` header. If your application
146 is distributed under another license, function calls will be generated
147 instead of inlines, so your application can link with the library.
148 - Linking with one of the libraries below is always necessary even for
149 LGPL and GPL applications.
150 - Define `URCU_INLINE_SMALL_FUNCTIONS` before including Userspace RCU
151 headers if you want Userspace RCU to inline small functions (10
152 lines or less) into the application. It can be used by applications
153 distributed under any kind of license, and does *not* make the
154 application a derived work of Userspace RCU.
156 Those small inlined functions are guaranteed to match the library
157 content as long as the library major version is unchanged.
158 Therefore, the application *must* be compiled with headers matching
159 the library major version number. Applications using
160 `URCU_INLINE_SMALL_FUNCTIONS` may be unable to use debugging
161 features of Userspace RCU without being recompiled.
163 There are multiple flavors of liburcu available:
171 The API members start with the prefix "urcu_<flavor>_", where
172 <flavor> is the chosen flavor name.
175 ### Usage of `liburcu-memb`
177 1. `#include <urcu/urcu-memb.h>`
178 2. Link the application with `-lurcu-memb`
180 This is the preferred version of the library, in terms of
181 grace-period detection speed, read-side speed and flexibility.
182 Dynamically detects kernel support for `sys_membarrier()`. Falls back
183 on `urcu-mb` scheme if support is not present, which has slower
184 read-side. Use the --disable-sys-membarrier-fallback configure option
185 to disable the fall back, thus requiring `sys_membarrier()` to be
186 available. This gives a small speedup when `sys_membarrier()` is
187 supported by the kernel, and aborts in the library constructor if not
191 ### Usage of `liburcu-qsbr`
193 1. `#include <urcu/urcu-qsbr.h>`
194 2. Link with `-lurcu-qsbr`
196 The QSBR flavor of RCU needs to have each reader thread executing
197 `rcu_quiescent_state()` periodically to progress. `rcu_thread_online()`
198 and `rcu_thread_offline()` can be used to mark long periods for which
199 the threads are not active. It provides the fastest read-side at the
200 expense of more intrusiveness in the application code.
203 ### Usage of `liburcu-mb`
205 1. `#include <urcu/urcu-mb.h>`
206 2. Link with `-lurcu-mb`
208 This version of the urcu library uses memory barriers on the writer
209 and reader sides. This results in faster grace-period detection, but
210 results in slower reads.
213 ### Usage of `liburcu-signal`
215 1. `#include <urcu/urcu-signal.h>`
216 2. Link the application with `-lurcu-signal`
218 Version of the library that requires a signal, typically `SIGUSR1`. Can
219 be overridden with `-DSIGRCU` by modifying `Makefile.build.inc`.
222 ### Usage of `liburcu-bp`
224 1. `#include <urcu/urcu-bp.h>`
225 2. Link with `-lurcu-bp`
227 The BP library flavor stands for "bulletproof". It is specifically
228 designed to help tracing library to hook on applications without
229 requiring to modify these applications. `urcu_bp_init()`,
230 `urcu_bp_register_thread()` and `urcu_bp_unregister_thread()` all become
231 nops. The state is dealt with by the library internally at the expense
232 of read-side and write-side performance.
237 Each thread that has reader critical sections (that uses
238 `urcu_<flavor>_read_lock()`/`urcu_<flavor>_read_unlock()` must first
239 register to the URCU library. This is done by calling
240 `urcu_<flavor>_register_thread()`. Unregistration must be performed
241 before exiting the thread by using `urcu_<flavor>_unregister_thread()`.
246 Reader critical sections must be protected by locating them between
247 calls to `urcu_<flavor>_read_lock()` and `urcu_<flavor>_read_unlock()`.
248 Inside that lock, `rcu_dereference()` may be called to read an RCU
254 `rcu_assign_pointer()` and `rcu_xchg_pointer()` may be called anywhere.
255 After, `urcu_<flavor>_synchronize_rcu()` must be called. When it
256 returns, the old values are not in usage anymore.
259 ### Usage of `liburcu-defer`
261 - Follow instructions for either `liburcu-memb`, `liburcu-qsbr`,
262 `liburcu-mb`, `liburcu-signal`, or `liburcu-bp` above.
263 The `liburcu-defer` functionality is pulled into each of
264 those library modules.
265 - Provides `urcu_<flavor>_defer_rcu()` primitive to enqueue delayed
266 callbacks. Queued callbacks are executed in batch periodically after
267 a grace period. Do _not_ use `urcu_<flavor>_defer_rcu()` within a
268 read-side critical section, because it may call
269 `urcu_<flavor>_synchronize_rcu()` if the thread queue is full. This
270 can lead to deadlock or worse.
271 - Requires that `urcu_<flavor>_defer_barrier()` must be called in
272 library destructor if a library queues callbacks and is expected to
273 be unloaded with `dlclose()`.
275 Its API is currently experimental. It may change in future library releases.
278 ### Usage of `urcu-call-rcu`
280 - Follow instructions for either `liburcu-memb`, `liburcu-qsbr`,
281 `liburcu-mb`, `liburcu-signal`, or `liburcu-bp` above.
282 The `urcu-call-rcu` functionality is pulled into each of
283 those library modules.
284 - Provides the `urcu_<flavor>_call_rcu()` primitive to enqueue delayed
285 callbacks in a manner similar to `urcu_<flavor>_defer_rcu()`, but
286 without ever delaying for a grace period. On the other hand,
287 `urcu_<flavor>_call_rcu()`'s best-case overhead is not quite as good
288 as that of `urcu_<flavor>_defer_rcu()`.
289 - Provides `urcu_<flavor>_call_rcu()` to allow asynchronous handling
290 of RCU grace periods. A number of additional functions are provided
291 to manage the helper threads used by `urcu_<flavor>_call_rcu()`, but
292 reasonable defaults are used if these additional functions are not
293 invoked. See [`doc/rcu-api.md`](doc/rcu-api.md) in userspace-rcu
294 documentation for more details.
297 ### Being careful with signals
299 The `liburcu-signal` library uses signals internally. The signal handler is
300 registered with the `SA_RESTART` flag. However, these signals may cause
301 some non-restartable system calls to fail with `errno = EINTR`. Care
302 should be taken to restart system calls manually if they fail with this
303 error. A list of non-restartable system calls may be found in
306 Read-side critical sections are allowed in a signal handler,
307 except those setup with `sigaltstack(2)`, with `liburcu-memb` and
308 `liburcu-mb`. Be careful, however, to disable these signals
309 between thread creation and calls to `urcu_<flavor>_register_thread()`,
310 because a signal handler nesting on an unregistered thread would not be
311 allowed to call `urcu_<flavor>_read_lock()`.
313 Read-side critical sections are _not_ allowed in a signal handler with
314 `liburcu-qsbr`, unless signals are disabled explicitly around each
315 `urcu_qsbr_quiescent_state()` calls, when threads are put offline and around
316 calls to `urcu_qsbr_synchronize_rcu()`. Even then, we do not recommend it.
319 ### Interaction with mutexes
321 One must be careful to do not cause deadlocks due to interaction of
322 `urcu_<flavor>_synchronize_rcu()` and RCU read-side with mutexes. If
323 `urcu_<flavor>_synchronize_rcu()` is called with a mutex held, this
324 mutex (or any mutex which has this mutex in its dependency chain) should
325 not be acquired from within a RCU read-side critical section.
327 This is especially important to understand in the context of the
328 QSBR flavor: a registered reader thread being "online" by
329 default should be considered as within a RCU read-side critical
330 section unless explicitly put "offline". Therefore, if
331 `urcu_qsbr_synchronize_rcu()` is called with a mutex held, this mutex,
332 as well as any mutex which has this mutex in its dependency chain should
333 only be taken when the RCU reader thread is "offline" (this can be
334 performed by calling `urcu_qsbr_thread_offline()`).
337 ### Interaction with `fork()`
339 Special care must be taken for applications performing `fork()` without
340 any following `exec()`. This is caused by the fact that Linux only clones
341 the thread calling `fork()`, and thus never replicates any of the other
342 parent thread into the child process. Most `liburcu` implementations
343 require that all registrations (as reader, `defer_rcu` and `call_rcu`
344 threads) should be released before a `fork()` is performed, except for the
345 rather common scenario where `fork()` is immediately followed by `exec()` in
346 the child process. The only implementation not subject to that rule is
347 `liburcu-bp`, which is designed to handle `fork()` by calling
348 `urcu_bp_before_fork`, `urcu_bp_after_fork_parent` and
349 `urcu_bp_after_fork_child`.
351 Applications that use `urcu_<flavor>_call_rcu()` and that `fork()`
352 without doing an immediate `exec()` must take special action. The
353 parent must invoke `urcu_<flavor>_call_rcu_before_fork()` before the
354 `fork()` and `urcu_<flavor>_call_rcu_after_fork_parent()` after the
355 `fork()`. The child process must invoke
356 `urcu_<flavor>_call_rcu_after_fork_child()`. Even though these three
357 APIs are suitable for passing to `pthread_atfork()`, use of
358 `pthread_atfork()` is **STRONGLY DISCOURAGED** for programs calling the
359 glibc memory allocator (`malloc()`, `calloc()`, `free()`, ...) within
360 `urcu_<flavor>_call_rcu` callbacks. This is due to limitations in the
361 way glibc memory allocator handles calls to the memory allocator from
362 concurrent threads while the `pthread_atfork()` handlers are executing.
366 - call to `free()` from callbacks executed within
367 `urcu_<flavor>_call_rcu` worker threads,
368 - executing `urcu_<flavor>_call_rcu` atfork handlers within the glibc
369 pthread atfork mechanism,
371 will sometimes trigger interesting process hangs. This usually
372 hangs on a memory allocator lock within glibc.
375 ### Thread Local Storage (TLS)
377 Userspace RCU can fall back on `pthread_getspecific()` to emulate
378 TLS variables on systems where it is not available. This behavior
379 can be forced by specifying `--disable-compiler-tls` as configure
383 ### Usage of `DEBUG_RCU` & `--enable-rcu-debug`
385 By default the library is configured with internal debugging
386 self-checks disabled.
388 For always-on debugging self-checks:
389 ./configure --enable-rcu-debug
391 For fine grained enabling of debugging self-checks, build
392 userspace-rcu with DEBUG_RCU defined and compile dependent
393 applications with DEBUG_RCU defined when necessary.
395 Warning: Enabling this feature result in a performance penalty.
398 ### Usage of `DEBUG_YIELD`
400 `DEBUG_YIELD` is used to add random delays in the code for testing
406 By default the library is configured to use synchronization primitives
407 adequate for SMP systems. On uniprocessor systems, support for SMP
408 systems can be disabled with:
410 ./configure --disable-smp-support
412 theoretically yielding slightly better performance.
415 ### Usage of `--enable-cds-lfht-iter-debug`
417 By default the library is configured with extra debugging checks for
418 lock-free hash table iterator traversal disabled.
420 Building liburcu with --enable-cds-lfht-iter-debug and rebuilding
421 application to match the ABI change allows finding cases where the hash
422 table iterator is re-purposed to be used on a different hash table while
423 still being used to iterate on a hash table.
425 This option alters the rculfhash ABI. Make sure to compile both library
426 and application with matching configuration.
432 In addition to the usual `make check` target, Userspace RCU features
433 `make regtest` and `make bench` targets:
435 - `make check`: short tests, meant to be run when rebuilding or
436 porting Userspace RCU.
437 - `make regtest`: long (many hours) test, meant to be run when
438 modifying Userspace RCU or porting it to a new architecture or
440 - `make bench`: long (many hours) benchmarks.
446 There is an application vs library compatibility issue between
447 applications built using Userspace RCU 0.10 headers linked against
448 Userspace RCU 0.11 or 0.12 shared objects. The problem occurs as
451 - An application executable is built with _LGPL_SOURCE defined, includes
452 any of the Userspace RCU 0.10 urcu flavor headers, and is built
453 without the -fpic compiler option.
455 - The Userspace RCU 0.10 library shared objects are updated to 0.11
456 or 0.12 without rebuilding the application.
458 - The application will hang, typically when RCU grace period
459 (synchronize_rcu) is invoked.
461 Some possible work-arounds for this are:
463 - Rebuild the application against Userspace RCU 0.11+.
465 - Rebuild the application with -fpic.
467 - Upgrade Userspace RCU to 0.13+ without installing 0.11 nor 0.12.
473 You can contact the maintainers on the following mailing list:
474 `lttng-dev@lists.lttng.org`.