README.md

   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
  40   - x86 (i386, i486, i586, i686)
  41   - amd64 / x86_64
  42   - PowerPC 32/64
  43   - S390, S390x
  44   - ARM 32/64
  45   - MIPS
  46   - NIOS2
  47   - Alpha
  48   - ia64
  49   - Sparcv9 32/64
  50   - Tilera
  51   - hppa/PA-RISC
  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
  61   - MacOSX amd64
  62
  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.
  89
  90 Clang version 3.0 (based on LLVM 3.0) is supported.
  91
  92 Building on MacOS X (Darwin) requires a work-around for processor
  93 detection:
  94
  95   - 32-bit:
  96
  97         ./configure --build=i686-apple-darwin11
  98
  99   - 64-bit:
 100
 101         ./configure --build=x86_64-apple-darwin11
 102
 103 For developers using the Git tree:
 104
 105 This source tree is based on the autotools suite from GNU to simplify
 106 portability. Here are some things you should have on your system in order to
 107 compile the git repository tree :
 108
 109   - GNU autotools (automake >=1.10, autoconf >=2.50, autoheader >=2.50)
 110     (make sure your system wide `automake` points to a recent version!)
 111   - GNU Libtool >=2.2
 112     (for more information, go to http://www.gnu.org/software/autoconf/)
 113
 114 If you get the tree from the repository, you will need to use the `bootstrap`
 115 script in the root of the tree. It calls all the GNU tools needed to prepare
 116 the tree configuration.
 117
 118 Test scripts provided in the `tests/` directory of the source tree depend
 119 on `bash` and the `seq` program.
 120
 121
 122 API
 123 ---
 124
 125 See the relevant API documentation files in `doc/`. The APIs provided by
 126 Userspace RCU are, by prefix:
 127
 128   - `rcu_`: Read-Copy Update (see [`doc/rcu-api.md`](doc/rcu-api.md))
 129   - `cmm_`: Concurrent Memory Model
 130   - `caa_`: Concurrent Architecture Abstraction
 131   - `cds_`: Concurrent Data Structures
 132     (see [`doc/cds-api.md`](doc/cds-api.md))
 133   - `uatomic_`: Userspace Atomic
 134     (see [`doc/uatomic-api.md`](doc/uatomic-api.md))
 135
 136
 137 Quick start guide
 138 -----------------
 139
 140 ### Usage of all urcu libraries:
 141
 142   - Define `_LGPL_SOURCE` (only) if your code is LGPL or GPL compatible
 143     before including the `urcu.h` or `urcu-qsbr.h` header. If your application
 144     is distributed under another license, function calls will be generated
 145     instead of inlines, so your application can link with the library.
 146   - Linking with one of the libraries below is always necessary even for
 147     LGPL and GPL applications.
 148   - Define `URCU_INLINE_SMALL_FUNCTIONS` before including Userspace RCU
 149     headers if you want Userspace RCU to inline small functions (10
 150     lines or less) into the application. It can be used by applications
 151     distributed under any kind of license, and does *not* make the
 152     application a derived work of Userspace RCU.
 153
 154 Those small inlined functions are guaranteed to match the library
 155 content as long as the library major version is unchanged.
 156 Therefore, the application *must* be compiled with headers matching
 157 the library major version number. Applications using
 158 `URCU_INLINE_SMALL_FUNCTIONS` may be unable to use debugging
 159 features of Userspace RCU without being recompiled.
 160
 161 There are multiple flavors of liburcu available:
 162
 163   - `memb`,
 164   - `qsbr`,
 165   - `mb`,
 166   - `signal`,
 167   - `bp`.
 168
 169 The API members start with the prefix "urcu_<flavor>_", where
 170 <flavor> is the chosen flavor name.
 171
 172
 173 ### Usage of `liburcu-memb`
 174
 175   1. `#include <urcu/urcu-memb.h>`
 176   2. Link the application with `-lurcu-memb`
 177
 178 This is the preferred version of the library, in terms of
 179 grace-period detection speed, read-side speed and flexibility.
 180 Dynamically detects kernel support for `sys_membarrier()`. Falls back
 181 on `urcu-mb` scheme if support is not present, which has slower
 182 read-side. Use the --disable-sys-membarrier-fallback configure option
 183 to disable the fall back, thus requiring `sys_membarrier()` to be
 184 available. This gives a small speedup when `sys_membarrier()` is
 185 supported by the kernel, and aborts in the library constructor if not
 186 supported.
 187
 188
 189 ### Usage of `liburcu-qsbr`
 190
 191   1. `#include <urcu/urcu-qsbr.h>`
 192   2. Link with `-lurcu-qsbr`
 193
 194 The QSBR flavor of RCU needs to have each reader thread executing
 195 `rcu_quiescent_state()` periodically to progress. `rcu_thread_online()`
 196 and `rcu_thread_offline()` can be used to mark long periods for which
 197 the threads are not active. It provides the fastest read-side at the
 198 expense of more intrusiveness in the application code.
 199
 200
 201 ### Usage of `liburcu-mb`
 202
 203   1. `#include <urcu/urcu-mb.h>`
 204   2. Link with `-lurcu-mb`
 205
 206 This version of the urcu library uses memory barriers on the writer
 207 and reader sides. This results in faster grace-period detection, but
 208 results in slower reads.
 209
 210
 211 ### Usage of `liburcu-signal`
 212
 213   1. `#include <urcu/urcu-signal.h>`
 214   2. Link the application with `-lurcu-signal`
 215
 216 Version of the library that requires a signal, typically `SIGUSR1`. Can
 217 be overridden with `-DSIGRCU` by modifying `Makefile.build.inc`.
 218
 219
 220 ### Usage of `liburcu-bp`
 221
 222   1. `#include <urcu/urcu-bp.h>`
 223   2. Link with `-lurcu-bp`
 224
 225 The BP library flavor stands for "bulletproof". It is specifically
 226 designed to help tracing library to hook on applications without
 227 requiring to modify these applications. `urcu_bp_init()`, and
 228 `urcu_bp_unregister_thread()` all become nops, whereas calling
 229 `urcu_bp_register_thread()` becomes optional. The state is dealt with by
 230 the library internally at the expense of read-side and write-side
 231 performance.
 232
 233
 234 ### Initialization
 235
 236 Each thread that has reader critical sections (that uses
 237 `urcu_<flavor>_read_lock()`/`urcu_<flavor>_read_unlock()` must first
 238 register to the URCU library. This is done by calling
 239 `urcu_<flavor>_register_thread()`. Unregistration must be performed
 240 before exiting the thread by using `urcu_<flavor>_unregister_thread()`.
 241
 242
 243 ### Reading
 244
 245 Reader critical sections must be protected by locating them between
 246 calls to `urcu_<flavor>_read_lock()` and `urcu_<flavor>_read_unlock()`.
 247 Inside that lock, `rcu_dereference()` may be called to read an RCU
 248 protected pointer.
 249
 250
 251 ### Writing
 252
 253 `rcu_assign_pointer()` and `rcu_xchg_pointer()` may be called anywhere.
 254 After, `urcu_<flavor>_synchronize_rcu()` must be called. When it
 255 returns, the old values are not in usage anymore.
 256
 257
 258 ### Usage of `liburcu-defer`
 259
 260   - Follow instructions for either `liburcu-memb`, `liburcu-qsbr`,
 261     `liburcu-mb`, `liburcu-signal`, or `liburcu-bp` above.
 262     The `liburcu-defer` functionality is pulled into each of
 263     those library modules.
 264   - Provides `urcu_<flavor>_defer_rcu()` primitive to enqueue delayed
 265     callbacks. Queued callbacks are executed in batch periodically after
 266     a grace period.  Do _not_ use `urcu_<flavor>_defer_rcu()` within a
 267     read-side critical section, because it may call
 268     `urcu_<flavor>_synchronize_rcu()` if the thread queue is full.  This
 269     can lead to deadlock or worse.
 270   - Requires that `urcu_<flavor>_defer_barrier()` must be called in
 271     library destructor if a library queues callbacks and is expected to
 272     be unloaded with `dlclose()`.
 273
 274 Its API is currently experimental. It may change in future library releases.
 275
 276
 277 ### Usage of `urcu-call-rcu`
 278
 279   - Follow instructions for either `liburcu-memb`, `liburcu-qsbr`,
 280     `liburcu-mb`, `liburcu-signal`, or `liburcu-bp` above.
 281     The `urcu-call-rcu` functionality is pulled into each of
 282     those library modules.
 283   - Provides the `urcu_<flavor>_call_rcu()` primitive to enqueue delayed
 284     callbacks in a manner similar to `urcu_<flavor>_defer_rcu()`, but
 285     without ever delaying for a grace period.  On the other hand,
 286     `urcu_<flavor>_call_rcu()`'s best-case overhead is not quite as good
 287     as that of `urcu_<flavor>_defer_rcu()`.
 288   - Provides `urcu_<flavor>_call_rcu()` to allow asynchronous handling
 289     of RCU grace periods.  A number of additional functions are provided
 290     to manage the helper threads used by `urcu_<flavor>_call_rcu()`, but
 291     reasonable defaults are used if these additional functions are not
 292     invoked. See [`doc/rcu-api.md`](doc/rcu-api.md) in userspace-rcu
 293     documentation for more details.
 294
 295
 296 ### Being careful with signals
 297
 298 The `liburcu-signal` library uses signals internally. The signal handler is
 299 registered with the `SA_RESTART` flag. However, these signals may cause
 300 some non-restartable system calls to fail with `errno = EINTR`. Care
 301 should be taken to restart system calls manually if they fail with this
 302 error. A list of non-restartable system calls may be found in
 303 `signal(7)`.
 304
 305 Read-side critical sections are allowed in a signal handler,
 306 except those setup with `sigaltstack(2)`, with `liburcu-memb` and
 307 `liburcu-mb`. Be careful, however, to disable these signals
 308 between thread creation and calls to `urcu_<flavor>_register_thread()`,
 309 because a signal handler nesting on an unregistered thread would not be
 310 allowed to call `urcu_<flavor>_read_lock()`.
 311
 312 Read-side critical sections are _not_ allowed in a signal handler with
 313 `liburcu-qsbr`, unless signals are disabled explicitly around each
 314 `urcu_qsbr_quiescent_state()` calls, when threads are put offline and around
 315 calls to `urcu_qsbr_synchronize_rcu()`. Even then, we do not recommend it.
 316
 317
 318 ### Interaction with mutexes
 319
 320 One must be careful to do not cause deadlocks due to interaction of
 321 `urcu_<flavor>_synchronize_rcu()` and RCU read-side with mutexes. If
 322 `urcu_<flavor>_synchronize_rcu()` is called with a mutex held, this
 323 mutex (or any mutex which has this mutex in its dependency chain) should
 324 not be acquired from within a RCU read-side critical section.
 325
 326 This is especially important to understand in the context of the
 327 QSBR flavor: a registered reader thread being "online" by
 328 default should be considered as within a RCU read-side critical
 329 section unless explicitly put "offline". Therefore, if
 330 `urcu_qsbr_synchronize_rcu()` is called with a mutex held, this mutex,
 331 as well as any mutex which has this mutex in its dependency chain should
 332 only be taken when the RCU reader thread is "offline" (this can be
 333 performed by calling `urcu_qsbr_thread_offline()`).
 334
 335
 336 ### Interaction with `fork()`
 337
 338 Special care must be taken for applications performing `fork()` without
 339 any following `exec()`. This is caused by the fact that Linux only clones
 340 the thread calling `fork()`, and thus never replicates any of the other
 341 parent thread into the child process. Most `liburcu` implementations
 342 require that all registrations (as reader, `defer_rcu` and `call_rcu`
 343 threads) should be released before a `fork()` is performed, except for the
 344 rather common scenario where `fork()` is immediately followed by `exec()` in
 345 the child process. The only implementation not subject to that rule is
 346 `liburcu-bp`, which is designed to handle `fork()` by calling
 347 `urcu_bp_before_fork`, `urcu_bp_after_fork_parent` and
 348 `urcu_bp_after_fork_child`.
 349
 350 Applications that use `urcu_<flavor>_call_rcu()` and that `fork()`
 351 without doing an immediate `exec()` must take special action.  The
 352 parent must invoke `urcu_<flavor>_call_rcu_before_fork()` before the
 353 `fork()` and `urcu_<flavor>_call_rcu_after_fork_parent()` after the
 354 `fork()`. The child process must invoke
 355 `urcu_<flavor>_call_rcu_after_fork_child()`.  Even though these three
 356 APIs are suitable for passing to `pthread_atfork()`, use of
 357 `pthread_atfork()` is **STRONGLY DISCOURAGED** for programs calling the
 358 glibc memory allocator (`malloc()`, `calloc()`, `free()`, ...) within
 359 `urcu_<flavor>_call_rcu` callbacks.  This is due to limitations in the
 360 way glibc memory allocator handles calls to the memory allocator from
 361 concurrent threads while the `pthread_atfork()` handlers are executing.
 362
 363 Combining e.g.:
 364
 365   - call to `free()` from callbacks executed within
 366     `urcu_<flavor>_call_rcu` worker threads,
 367   - executing `urcu_<flavor>_call_rcu` atfork handlers within the glibc
 368     pthread atfork mechanism,
 369
 370 will sometimes trigger interesting process hangs. This usually
 371 hangs on a memory allocator lock within glibc.
 372
 373
 374 ### Thread Local Storage (TLS)
 375
 376 Userspace RCU can fall back on `pthread_getspecific()` to emulate
 377 TLS variables on systems where it is not available. This behavior
 378 can be forced by specifying `--disable-compiler-tls` as configure
 379 argument.
 380
 381
 382 ### Usage of `DEBUG_RCU` & `--enable-rcu-debug`
 383
 384 By default the library is configured with internal debugging
 385 self-checks disabled.
 386
 387 For always-on debugging self-checks:
 388         ./configure --enable-rcu-debug
 389
 390 For fine grained enabling of debugging self-checks, build
 391 userspace-rcu with DEBUG_RCU defined and compile dependent
 392 applications with DEBUG_RCU defined when necessary.
 393
 394 Warning: Enabling this feature result in a performance penalty.
 395
 396
 397 ### Usage of `DEBUG_YIELD`
 398
 399 `DEBUG_YIELD` is used to add random delays in the code for testing
 400 purposes.
 401
 402
 403 ### SMP support
 404
 405 By default the library is configured to use synchronization primitives
 406 adequate for SMP systems. On uniprocessor systems, support for SMP
 407 systems can be disabled with:
 408
 409     ./configure --disable-smp-support
 410
 411 theoretically yielding slightly better performance.
 412
 413
 414 ### Usage of `--enable-cds-lfht-iter-debug`
 415
 416 By default the library is configured with extra debugging checks for
 417 lock-free hash table iterator traversal disabled.
 418
 419 Building liburcu with --enable-cds-lfht-iter-debug and rebuilding
 420 application to match the ABI change allows finding cases where the hash
 421 table iterator is re-purposed to be used on a different hash table while
 422 still being used to iterate on a hash table.
 423
 424 This option alters the rculfhash ABI. Make sure to compile both library
 425 and application with matching configuration.
 426
 427
 428 Make targets
 429 ------------
 430
 431 In addition to the usual `make check` target, Userspace RCU features
 432 `make regtest` and `make bench` targets:
 433
 434   - `make check`: short tests, meant to be run when rebuilding or
 435     porting Userspace RCU.
 436   - `make regtest`: long (many hours) test, meant to be run when
 437     modifying Userspace RCU or porting it to a new architecture or
 438     operating system.
 439   - `make bench`: long (many hours) benchmarks.
 440
 441
 442 Contacts
 443 --------
 444
 445 You can contact the maintainers on the following mailing list:
 446 `lttng-dev@lists.lttng.org`.