Commit | Line | Data |
---|---|---|
d589a916 PP |
1 | Userspace RCU Implementation |
2 | ============================ | |
3 | ||
4 | by Mathieu Desnoyers and Paul E. McKenney | |
5 | ||
6 | ||
7 | Building | |
8 | -------- | |
9 | ||
10 | ./bootstrap # skip if using tarball | |
11 | ./configure | |
12 | make | |
13 | make install | |
14 | ldconfig | |
15 | ||
16 | Hints: | |
17 | ||
18 | - Forcing 32-bit build: | |
19 | ||
20 | CFLAGS="-m32 -g -O2" ./configure | |
21 | ||
22 | - Forcing 64-bit build: | |
23 | ||
24 | CFLAGS="-m64 -g -O2" ./configure | |
25 | ||
26 | - Forcing a 32-bit build with 386 backward compatibility: | |
27 | ||
28 | CFLAGS="-m32 -g -O2" ./configure --host=i386-pc-linux-gnu | |
29 | ||
30 | - Forcing a 32-bit build for Sparcv9 (typical for Sparc v9) | |
31 | ||
32 | CFLAGS="-m32 -Wa,-Av9a -g -O2" ./configure | |
33 | ||
34 | ||
35 | Architectures supported | |
36 | ----------------------- | |
37 | ||
38 | Currently, the following architectures are supported: | |
39 | ||
f328865f MD |
40 | - x86 (i386, i486, i586, i686) |
41 | - amd64 / x86_64 | |
d589a916 PP |
42 | - PowerPC 32/64 |
43 | - S390, S390x | |
44 | - ARM 32/64 | |
45 | - MIPS | |
859050b3 | 46 | - NIOS2 |
d589a916 PP |
47 | - Alpha |
48 | - ia64 | |
49 | - Sparcv9 32/64 | |
50 | - Tilera | |
51 | - hppa/PA-RISC | |
f328865f MD |
52 | - m68k |
53 | - RISC-V | |
54 | ||
55 | Tested on: | |
56 | ||
57 | - Linux all architectures | |
58 | - FreeBSD 8.2/8.3/9.0/9.1/10.0 i386/amd64 | |
59 | - Solaris 10/11 i386 | |
60 | - Cygwin i386/amd64 | |
0af4b40c | 61 | - MacOS amd64/arm64 |
d589a916 | 62 | |
d589a916 PP |
63 | Should also work on: |
64 | ||
65 | - Android | |
66 | - NetBSD 5 | |
67 | - OpenBSD | |
68 | - Darwin | |
69 | ||
70 | (more testing needed before claiming support for these OS). | |
71 | ||
72 | Linux ARM depends on running a Linux kernel 2.6.15 or better, GCC 4.4 or | |
73 | better. | |
74 | ||
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: | |
77 | ||
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. | |
ddec79fd MD |
89 | - Linux aarch64 depends on GCC 5.1 or better because prior versions |
90 | perform unsafe access to deallocated stack. | |
d589a916 PP |
91 | |
92 | Clang version 3.0 (based on LLVM 3.0) is supported. | |
93 | ||
ef6da886 MJ |
94 | Glibc >= 2.4 should work but the older version we test against is |
95 | currently 2.17. | |
96 | ||
d589a916 PP |
97 | For developers using the Git tree: |
98 | ||
99 | This source tree is based on the autotools suite from GNU to simplify | |
100 | portability. Here are some things you should have on your system in order to | |
101 | compile the git repository tree : | |
102 | ||
afb6113f | 103 | - GNU autotools (automake >=1.12, autoconf >=2.69) |
d589a916 PP |
104 | (make sure your system wide `automake` points to a recent version!) |
105 | - GNU Libtool >=2.2 | |
106 | (for more information, go to http://www.gnu.org/software/autoconf/) | |
107 | ||
108 | If you get the tree from the repository, you will need to use the `bootstrap` | |
109 | script in the root of the tree. It calls all the GNU tools needed to prepare | |
110 | the tree configuration. | |
111 | ||
112 | Test scripts provided in the `tests/` directory of the source tree depend | |
113 | on `bash` and the `seq` program. | |
114 | ||
115 | ||
116 | API | |
117 | --- | |
118 | ||
119 | See the relevant API documentation files in `doc/`. The APIs provided by | |
120 | Userspace RCU are, by prefix: | |
121 | ||
dcb9c05a | 122 | - `rcu_`: Read-Copy Update (see [`doc/rcu-api.md`](doc/rcu-api.md)) |
d589a916 PP |
123 | - `cmm_`: Concurrent Memory Model |
124 | - `caa_`: Concurrent Architecture Abstraction | |
125 | - `cds_`: Concurrent Data Structures | |
dcb9c05a | 126 | (see [`doc/cds-api.md`](doc/cds-api.md)) |
d589a916 | 127 | - `uatomic_`: Userspace Atomic |
dcb9c05a | 128 | (see [`doc/uatomic-api.md`](doc/uatomic-api.md)) |
d589a916 PP |
129 | |
130 | ||
131 | Quick start guide | |
132 | ----------------- | |
133 | ||
134 | ### Usage of all urcu libraries: | |
135 | ||
136 | - Define `_LGPL_SOURCE` (only) if your code is LGPL or GPL compatible | |
137 | before including the `urcu.h` or `urcu-qsbr.h` header. If your application | |
138 | is distributed under another license, function calls will be generated | |
139 | instead of inlines, so your application can link with the library. | |
140 | - Linking with one of the libraries below is always necessary even for | |
141 | LGPL and GPL applications. | |
142 | - Define `URCU_INLINE_SMALL_FUNCTIONS` before including Userspace RCU | |
143 | headers if you want Userspace RCU to inline small functions (10 | |
144 | lines or less) into the application. It can be used by applications | |
145 | distributed under any kind of license, and does *not* make the | |
146 | application a derived work of Userspace RCU. | |
147 | ||
148 | Those small inlined functions are guaranteed to match the library | |
149 | content as long as the library major version is unchanged. | |
150 | Therefore, the application *must* be compiled with headers matching | |
151 | the library major version number. Applications using | |
152 | `URCU_INLINE_SMALL_FUNCTIONS` may be unable to use debugging | |
153 | features of Userspace RCU without being recompiled. | |
154 | ||
f328865f MD |
155 | There are multiple flavors of liburcu available: |
156 | ||
157 | - `memb`, | |
158 | - `qsbr`, | |
159 | - `mb`, | |
160 | - `signal`, | |
161 | - `bp`. | |
162 | ||
163 | The API members start with the prefix "urcu_<flavor>_", where | |
164 | <flavor> is the chosen flavor name. | |
165 | ||
d589a916 | 166 | |
f328865f | 167 | ### Usage of `liburcu-memb` |
d589a916 | 168 | |
f328865f MD |
169 | 1. `#include <urcu/urcu-memb.h>` |
170 | 2. Link the application with `-lurcu-memb` | |
d589a916 PP |
171 | |
172 | This is the preferred version of the library, in terms of | |
173 | grace-period detection speed, read-side speed and flexibility. | |
174 | Dynamically detects kernel support for `sys_membarrier()`. Falls back | |
175 | on `urcu-mb` scheme if support is not present, which has slower | |
d8d9a340 MD |
176 | read-side. Use the --disable-sys-membarrier-fallback configure option |
177 | to disable the fall back, thus requiring `sys_membarrier()` to be | |
178 | available. This gives a small speedup when `sys_membarrier()` is | |
179 | supported by the kernel, and aborts in the library constructor if not | |
180 | supported. | |
d589a916 PP |
181 | |
182 | ||
183 | ### Usage of `liburcu-qsbr` | |
184 | ||
f328865f | 185 | 1. `#include <urcu/urcu-qsbr.h>` |
d589a916 PP |
186 | 2. Link with `-lurcu-qsbr` |
187 | ||
188 | The QSBR flavor of RCU needs to have each reader thread executing | |
189 | `rcu_quiescent_state()` periodically to progress. `rcu_thread_online()` | |
190 | and `rcu_thread_offline()` can be used to mark long periods for which | |
191 | the threads are not active. It provides the fastest read-side at the | |
192 | expense of more intrusiveness in the application code. | |
193 | ||
194 | ||
195 | ### Usage of `liburcu-mb` | |
196 | ||
f328865f MD |
197 | 1. `#include <urcu/urcu-mb.h>` |
198 | 2. Link with `-lurcu-mb` | |
d589a916 PP |
199 | |
200 | This version of the urcu library uses memory barriers on the writer | |
201 | and reader sides. This results in faster grace-period detection, but | |
202 | results in slower reads. | |
203 | ||
204 | ||
205 | ### Usage of `liburcu-signal` | |
206 | ||
f328865f MD |
207 | 1. `#include <urcu/urcu-signal.h>` |
208 | 2. Link the application with `-lurcu-signal` | |
d589a916 PP |
209 | |
210 | Version of the library that requires a signal, typically `SIGUSR1`. Can | |
211 | be overridden with `-DSIGRCU` by modifying `Makefile.build.inc`. | |
212 | ||
213 | ||
214 | ### Usage of `liburcu-bp` | |
215 | ||
f328865f | 216 | 1. `#include <urcu/urcu-bp.h>` |
d589a916 PP |
217 | 2. Link with `-lurcu-bp` |
218 | ||
219 | The BP library flavor stands for "bulletproof". It is specifically | |
220 | designed to help tracing library to hook on applications without | |
5b46e39d MD |
221 | requiring to modify these applications. `urcu_bp_init()`, and |
222 | `urcu_bp_unregister_thread()` all become nops, whereas calling | |
223 | `urcu_bp_register_thread()` becomes optional. The state is dealt with by | |
224 | the library internally at the expense of read-side and write-side | |
225 | performance. | |
d589a916 PP |
226 | |
227 | ||
228 | ### Initialization | |
229 | ||
230 | Each thread that has reader critical sections (that uses | |
f328865f MD |
231 | `urcu_<flavor>_read_lock()`/`urcu_<flavor>_read_unlock()` must first |
232 | register to the URCU library. This is done by calling | |
233 | `urcu_<flavor>_register_thread()`. Unregistration must be performed | |
234 | before exiting the thread by using `urcu_<flavor>_unregister_thread()`. | |
d589a916 PP |
235 | |
236 | ||
237 | ### Reading | |
238 | ||
239 | Reader critical sections must be protected by locating them between | |
f328865f MD |
240 | calls to `urcu_<flavor>_read_lock()` and `urcu_<flavor>_read_unlock()`. |
241 | Inside that lock, `rcu_dereference()` may be called to read an RCU | |
242 | protected pointer. | |
d589a916 PP |
243 | |
244 | ||
245 | ### Writing | |
246 | ||
247 | `rcu_assign_pointer()` and `rcu_xchg_pointer()` may be called anywhere. | |
f328865f MD |
248 | After, `urcu_<flavor>_synchronize_rcu()` must be called. When it |
249 | returns, the old values are not in usage anymore. | |
d589a916 PP |
250 | |
251 | ||
252 | ### Usage of `liburcu-defer` | |
253 | ||
f328865f | 254 | - Follow instructions for either `liburcu-memb`, `liburcu-qsbr`, |
d589a916 PP |
255 | `liburcu-mb`, `liburcu-signal`, or `liburcu-bp` above. |
256 | The `liburcu-defer` functionality is pulled into each of | |
257 | those library modules. | |
f328865f MD |
258 | - Provides `urcu_<flavor>_defer_rcu()` primitive to enqueue delayed |
259 | callbacks. Queued callbacks are executed in batch periodically after | |
260 | a grace period. Do _not_ use `urcu_<flavor>_defer_rcu()` within a | |
261 | read-side critical section, because it may call | |
262 | `urcu_<flavor>_synchronize_rcu()` if the thread queue is full. This | |
263 | can lead to deadlock or worse. | |
264 | - Requires that `urcu_<flavor>_defer_barrier()` must be called in | |
265 | library destructor if a library queues callbacks and is expected to | |
266 | be unloaded with `dlclose()`. | |
d589a916 PP |
267 | |
268 | Its API is currently experimental. It may change in future library releases. | |
269 | ||
270 | ||
271 | ### Usage of `urcu-call-rcu` | |
272 | ||
f328865f | 273 | - Follow instructions for either `liburcu-memb`, `liburcu-qsbr`, |
d589a916 PP |
274 | `liburcu-mb`, `liburcu-signal`, or `liburcu-bp` above. |
275 | The `urcu-call-rcu` functionality is pulled into each of | |
276 | those library modules. | |
f328865f MD |
277 | - Provides the `urcu_<flavor>_call_rcu()` primitive to enqueue delayed |
278 | callbacks in a manner similar to `urcu_<flavor>_defer_rcu()`, but | |
279 | without ever delaying for a grace period. On the other hand, | |
280 | `urcu_<flavor>_call_rcu()`'s best-case overhead is not quite as good | |
281 | as that of `urcu_<flavor>_defer_rcu()`. | |
282 | - Provides `urcu_<flavor>_call_rcu()` to allow asynchronous handling | |
283 | of RCU grace periods. A number of additional functions are provided | |
284 | to manage the helper threads used by `urcu_<flavor>_call_rcu()`, but | |
285 | reasonable defaults are used if these additional functions are not | |
286 | invoked. See [`doc/rcu-api.md`](doc/rcu-api.md) in userspace-rcu | |
287 | documentation for more details. | |
d589a916 PP |
288 | |
289 | ||
290 | ### Being careful with signals | |
291 | ||
f328865f | 292 | The `liburcu-signal` library uses signals internally. The signal handler is |
d589a916 PP |
293 | registered with the `SA_RESTART` flag. However, these signals may cause |
294 | some non-restartable system calls to fail with `errno = EINTR`. Care | |
295 | should be taken to restart system calls manually if they fail with this | |
296 | error. A list of non-restartable system calls may be found in | |
f328865f | 297 | `signal(7)`. |
d589a916 PP |
298 | |
299 | Read-side critical sections are allowed in a signal handler, | |
f328865f | 300 | except those setup with `sigaltstack(2)`, with `liburcu-memb` and |
d589a916 | 301 | `liburcu-mb`. Be careful, however, to disable these signals |
f328865f MD |
302 | between thread creation and calls to `urcu_<flavor>_register_thread()`, |
303 | because a signal handler nesting on an unregistered thread would not be | |
304 | allowed to call `urcu_<flavor>_read_lock()`. | |
d589a916 PP |
305 | |
306 | Read-side critical sections are _not_ allowed in a signal handler with | |
307 | `liburcu-qsbr`, unless signals are disabled explicitly around each | |
f328865f MD |
308 | `urcu_qsbr_quiescent_state()` calls, when threads are put offline and around |
309 | calls to `urcu_qsbr_synchronize_rcu()`. Even then, we do not recommend it. | |
d589a916 PP |
310 | |
311 | ||
312 | ### Interaction with mutexes | |
313 | ||
314 | One must be careful to do not cause deadlocks due to interaction of | |
f328865f MD |
315 | `urcu_<flavor>_synchronize_rcu()` and RCU read-side with mutexes. If |
316 | `urcu_<flavor>_synchronize_rcu()` is called with a mutex held, this | |
317 | mutex (or any mutex which has this mutex in its dependency chain) should | |
318 | not be acquired from within a RCU read-side critical section. | |
d589a916 PP |
319 | |
320 | This is especially important to understand in the context of the | |
321 | QSBR flavor: a registered reader thread being "online" by | |
322 | default should be considered as within a RCU read-side critical | |
323 | section unless explicitly put "offline". Therefore, if | |
f328865f MD |
324 | `urcu_qsbr_synchronize_rcu()` is called with a mutex held, this mutex, |
325 | as well as any mutex which has this mutex in its dependency chain should | |
326 | only be taken when the RCU reader thread is "offline" (this can be | |
327 | performed by calling `urcu_qsbr_thread_offline()`). | |
d589a916 PP |
328 | |
329 | ||
330 | ### Interaction with `fork()` | |
331 | ||
332 | Special care must be taken for applications performing `fork()` without | |
333 | any following `exec()`. This is caused by the fact that Linux only clones | |
334 | the thread calling `fork()`, and thus never replicates any of the other | |
335 | parent thread into the child process. Most `liburcu` implementations | |
336 | require that all registrations (as reader, `defer_rcu` and `call_rcu` | |
337 | threads) should be released before a `fork()` is performed, except for the | |
338 | rather common scenario where `fork()` is immediately followed by `exec()` in | |
339 | the child process. The only implementation not subject to that rule is | |
340 | `liburcu-bp`, which is designed to handle `fork()` by calling | |
f328865f MD |
341 | `urcu_bp_before_fork`, `urcu_bp_after_fork_parent` and |
342 | `urcu_bp_after_fork_child`. | |
343 | ||
344 | Applications that use `urcu_<flavor>_call_rcu()` and that `fork()` | |
345 | without doing an immediate `exec()` must take special action. The | |
346 | parent must invoke `urcu_<flavor>_call_rcu_before_fork()` before the | |
347 | `fork()` and `urcu_<flavor>_call_rcu_after_fork_parent()` after the | |
348 | `fork()`. The child process must invoke | |
349 | `urcu_<flavor>_call_rcu_after_fork_child()`. Even though these three | |
350 | APIs are suitable for passing to `pthread_atfork()`, use of | |
351 | `pthread_atfork()` is **STRONGLY DISCOURAGED** for programs calling the | |
352 | glibc memory allocator (`malloc()`, `calloc()`, `free()`, ...) within | |
353 | `urcu_<flavor>_call_rcu` callbacks. This is due to limitations in the | |
354 | way glibc memory allocator handles calls to the memory allocator from | |
355 | concurrent threads while the `pthread_atfork()` handlers are executing. | |
d589a916 PP |
356 | |
357 | Combining e.g.: | |
358 | ||
f328865f MD |
359 | - call to `free()` from callbacks executed within |
360 | `urcu_<flavor>_call_rcu` worker threads, | |
361 | - executing `urcu_<flavor>_call_rcu` atfork handlers within the glibc | |
362 | pthread atfork mechanism, | |
d589a916 PP |
363 | |
364 | will sometimes trigger interesting process hangs. This usually | |
365 | hangs on a memory allocator lock within glibc. | |
366 | ||
367 | ||
368 | ### Thread Local Storage (TLS) | |
369 | ||
370 | Userspace RCU can fall back on `pthread_getspecific()` to emulate | |
371 | TLS variables on systems where it is not available. This behavior | |
372 | can be forced by specifying `--disable-compiler-tls` as configure | |
373 | argument. | |
374 | ||
375 | ||
d4e640c0 | 376 | ### Usage of `DEBUG_RCU` & `--enable-rcu-debug` |
d589a916 | 377 | |
d4e640c0 JR |
378 | By default the library is configured with internal debugging |
379 | self-checks disabled. | |
380 | ||
381 | For always-on debugging self-checks: | |
382 | ./configure --enable-rcu-debug | |
383 | ||
384 | For fine grained enabling of debugging self-checks, build | |
e400ca4a | 385 | userspace-rcu with DEBUG_RCU defined and compile dependent |
d4e640c0 JR |
386 | applications with DEBUG_RCU defined when necessary. |
387 | ||
388 | Warning: Enabling this feature result in a performance penalty. | |
d589a916 PP |
389 | |
390 | ||
391 | ### Usage of `DEBUG_YIELD` | |
392 | ||
393 | `DEBUG_YIELD` is used to add random delays in the code for testing | |
394 | purposes. | |
395 | ||
396 | ||
397 | ### SMP support | |
398 | ||
399 | By default the library is configured to use synchronization primitives | |
400 | adequate for SMP systems. On uniprocessor systems, support for SMP | |
401 | systems can be disabled with: | |
402 | ||
403 | ./configure --disable-smp-support | |
404 | ||
405 | theoretically yielding slightly better performance. | |
406 | ||
407 | ||
d7c76f85 MD |
408 | ### Usage of `--enable-cds-lfht-iter-debug` |
409 | ||
410 | By default the library is configured with extra debugging checks for | |
411 | lock-free hash table iterator traversal disabled. | |
412 | ||
413 | Building liburcu with --enable-cds-lfht-iter-debug and rebuilding | |
414 | application to match the ABI change allows finding cases where the hash | |
415 | table iterator is re-purposed to be used on a different hash table while | |
416 | still being used to iterate on a hash table. | |
417 | ||
418 | This option alters the rculfhash ABI. Make sure to compile both library | |
419 | and application with matching configuration. | |
420 | ||
421 | ||
d589a916 PP |
422 | Make targets |
423 | ------------ | |
424 | ||
425 | In addition to the usual `make check` target, Userspace RCU features | |
426 | `make regtest` and `make bench` targets: | |
427 | ||
428 | - `make check`: short tests, meant to be run when rebuilding or | |
429 | porting Userspace RCU. | |
430 | - `make regtest`: long (many hours) test, meant to be run when | |
431 | modifying Userspace RCU or porting it to a new architecture or | |
432 | operating system. | |
433 | - `make bench`: long (many hours) benchmarks. | |
434 | ||
435 | ||
436 | Contacts | |
437 | -------- | |
438 | ||
439 | You can contact the maintainers on the following mailing list: | |
dcb9c05a | 440 | `lttng-dev@lists.lttng.org`. |