README

   1 Userspace RCU Implementation
   2 by Mathieu Desnoyers and Paul E. McKenney
   3
   4 BUILDING
   5 --------
   6
   7         ./bootstrap (skip if using tarball)
   8         ./configure
   9         make
  10         make install
  11         ldconfig
  12
  13         Hints:  Forcing 32-bit build:
  14                 * CFLAGS="-m32 -g -O2" ./configure
  15
  16                 Forcing 64-bit build:
  17                 * CFLAGS="-m64 -g -O2" ./configure
  18
  19                 Forcing a 32-bit build with 386 backward compatibility:
  20                 * CFLAGS="-m32 -g -O2" ./configure --host=i386-pc-linux-gnu
  21
  22                 Forcing a 32-bit build for Sparcv9 (typical for Sparc v9)
  23                 * CFLAGS="-m32 -Wa,-Av9a -g -O2" ./configure
  24
  25
  26 ARCHITECTURES SUPPORTED
  27 -----------------------
  28
  29 Currently, Linux x86 (i386, i486, i586, i686), x86 64-bit, PowerPC 32/64,
  30 S390, S390x, ARM, MIPS, Alpha, ia64 and Sparcv9 32/64 are supported.
  31 Tested on Linux, FreeBSD 8.2/8.3/9.0/9.1/10.0 i386/amd64, and Cygwin.
  32 Should also work on: Android, NetBSD 5, OpenBSD, Darwin (more testing
  33 needed before claiming support for these OS).
  34
  35 Linux ARM depends on running a Linux kernel 2.6.15 or better, GCC 4.4 or
  36 better.
  37
  38 The gcc compiler versions 3.3, 3.4, 4.0, 4.1, 4.2, 4.3, 4.4 and 4.5 are
  39 supported, with the following exceptions:
  40
  41 - gcc 3.3 and 3.4 have a bug that prevents them from generating volatile
  42   accesses to offsets in a TLS structure on 32-bit x86. These versions are
  43   therefore not compatible with liburcu on x86 32-bit (i386, i486, i586, i686).
  44   The problem has been reported to the gcc community:
  45     http://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg281255.html
  46 - gcc 3.3 cannot match the "xchg" instruction on 32-bit x86 build.
  47   See: http://kerneltrap.org/node/7507
  48 - Alpha, ia64 and ARM architectures depend on gcc 4.x with atomic builtins
  49   support. For ARM this was introduced with gcc 4.4:
  50     http://gcc.gnu.org/gcc-4.4/changes.html
  51
  52 Clang version 3.0 (based on LLVM 3.0) is supported.
  53
  54 Building on MacOS X (Darwin) requires a work-around for processor
  55 detection:
  56   # 32-bit
  57   ./configure --build=i686-apple-darwin11
  58   # 64-bit
  59   ./configure --build=x86_64-apple-darwin11
  60
  61 For developers using the git tree:
  62
  63 This source tree is based on the autotools suite from GNU to simplify
  64 portability. Here are some things you should have on your system in order to
  65 compile the git repository tree :
  66
  67 - GNU autotools (automake >=1.10, autoconf >=2.50, autoheader >=2.50)
  68   (make sure your system wide "automake" points to a recent version!)
  69 - GNU Libtool >=2.2
  70   (for more information, go to http://www.gnu.org/software/autoconf/)
  71
  72 If you get the tree from the repository, you will need to use the "bootstrap"
  73 script in the root of the tree. It calls all the GNU tools needed to prepare the
  74 tree configuration.
  75
  76 Test scripts provided in the tests/ directory of the source tree depend
  77 on "bash" and the "seq" program.
  78
  79
  80 API
  81 ---
  82
  83 See the relevant API documentation files in doc/. The APIs provided by
  84 Userspace RCU are, by prefix:
  85
  86 - rcu_ : Read-Copy Update (see doc/rcu-api.txt)
  87 - cmm_ : Concurrent Memory Model
  88 - caa_ : Concurrent Architecture Abstraction
  89 - cds_ : Concurrent Data Structures (see doc/cds-api.txt)
  90 - uatomic_: Userspace Atomic (see doc/uatomic-api.txt)
  91
  92
  93 QUICK START GUIDE
  94 -----------------
  95
  96 Usage of all urcu libraries
  97
  98         * Define _LGPL_SOURCE (only) if your code is LGPL or GPL compatible
  99           before including the urcu.h or urcu-qsbr.h header. If your application
 100           is distributed under another license, function calls will be generated
 101           instead of inlines, so your application can link with the library.
 102         * Linking with one of the libraries below is always necessary even for
 103           LGPL and GPL applications.
 104
 105 Usage of liburcu
 106
 107         * #include <urcu.h>
 108         * Link the application with "-lurcu".
 109         * This is the preferred version of the library, in terms of
 110           grace-period detection speed, read-side speed and flexibility.
 111           Dynamically detects kernel support for sys_membarrier(). Falls back
 112           on urcu-mb scheme if support is not present, which has slower
 113           read-side.
 114
 115 Usage of liburcu-qsbr
 116
 117         * #include <urcu-qsbr.h>
 118         * Link with "-lurcu-qsbr".
 119         * The QSBR flavor of RCU needs to have each reader thread executing
 120           rcu_quiescent_state() periodically to progress. rcu_thread_online()
 121           and rcu_thread_offline() can be used to mark long periods for which
 122           the threads are not active. It provides the fastest read-side at the
 123           expense of more intrusiveness in the application code.
 124
 125 Usage of liburcu-mb
 126
 127         * #include <urcu.h>
 128         * Compile any _LGPL_SOURCE code using this library with "-DRCU_MB".
 129         * Link with "-lurcu-mb".
 130         * This version of the urcu library uses memory barriers on the writer
 131           and reader sides. This results in faster grace-period detection, but
 132           results in slower reads.
 133
 134 Usage of liburcu-signal
 135
 136         * #include <urcu.h>
 137         * Compile any _LGPL_SOURCE code using this library with "-DRCU_SIGNAL".
 138         * Link the application with "-lurcu-signal".
 139         * Version of the library that requires a signal, typically SIGUSR1. Can
 140           be overridden with -DSIGRCU by modifying Makefile.build.inc.
 141
 142 Usage of liburcu-bp
 143
 144         * #include <urcu-bp.h>
 145         * Link with "-lurcu-bp".
 146         * The BP library flavor stands for "bulletproof". It is specifically
 147           designed to help tracing library to hook on applications without
 148           requiring to modify these applications. rcu_init(),
 149           rcu_register_thread() and rcu_unregister_thread() all become nops.
 150           The state is dealt with by the library internally at the expense of
 151           read-side and write-side performance.
 152
 153 Initialization
 154
 155         Each thread that has reader critical sections (that uses
 156         rcu_read_lock()/rcu_read_unlock() must first register to the URCU
 157         library. This is done by calling rcu_register_thread(). Unregistration
 158         must be performed before exiting the thread by using
 159         rcu_unregister_thread().
 160
 161 Reading
 162
 163         Reader critical sections must be protected by locating them between
 164         calls to rcu_read_lock() and rcu_read_unlock(). Inside that lock,
 165         rcu_dereference() may be called to read an RCU protected pointer.
 166
 167 Writing
 168
 169         rcu_assign_pointer() and rcu_xchg_pointer() may be called anywhere.
 170         After, synchronize_rcu() must be called. When it returns, the old
 171         values are not in usage anymore.
 172
 173 Usage of liburcu-defer
 174
 175         * Follow instructions for either liburcu, liburcu-qsbr,
 176           liburcu-mb, liburcu-signal, or liburcu-bp above.
 177           The liburcu-defer functionality is pulled into each of
 178           those library modules.
 179         * Provides defer_rcu() primitive to enqueue delayed callbacks. Queued
 180           callbacks are executed in batch periodically after a grace period.
 181           Do _not_ use defer_rcu() within a read-side critical section, because
 182           it may call synchronize_rcu() if the thread queue is full.
 183           This can lead to deadlock or worse.
 184         * Requires that rcu_defer_barrier() must be called in library destructor
 185           if a library queues callbacks and is expected to be unloaded with
 186           dlclose().
 187         * Its API is currently experimental. It may change in future library
 188           releases.
 189
 190 Usage of urcu-call-rcu
 191
 192         * Follow instructions for either liburcu, liburcu-qsbr,
 193           liburcu-mb, liburcu-signal, or liburcu-bp above.
 194           The urcu-call-rcu functionality is provided for each of
 195           these library modules.
 196         * Provides the call_rcu() primitive to enqueue delayed callbacks
 197           in a manner similar to defer_rcu(), but without ever delaying
 198           for a grace period.  On the other hand, call_rcu()'s best-case
 199           overhead is not quite as good as that of defer_rcu().
 200         * Provides call_rcu() to allow asynchronous handling of RCU
 201           grace periods.  A number of additional functions are provided
 202           to manage the helper threads used by call_rcu(), but reasonable
 203           defaults are used if these additional functions are not invoked.
 204           See rcu-api.txt in userspace-rcu documentation for more details.
 205
 206 Being careful with signals
 207
 208         The liburcu library uses signals internally. The signal handler is
 209         registered with the SA_RESTART flag. However, these signals may cause
 210         some non-restartable system calls to fail with errno = EINTR. Care
 211         should be taken to restart system calls manually if they fail with this
 212         error. A list of non-restartable system calls may be found in
 213         signal(7). The liburcu-mb and liburcu-qsbr versions of the Userspace RCU
 214         library do not require any signal.
 215
 216         Read-side critical sections are allowed in a signal handler,
 217         except those setup with sigaltstack(2), with liburcu and
 218         liburcu-mb. Be careful, however, to disable these signals
 219         between thread creation and calls to rcu_register_thread(), because a
 220         signal handler nesting on an unregistered thread would not be
 221         allowed to call rcu_read_lock().
 222
 223         Read-side critical sections are _not_ allowed in a signal handler with
 224         liburcu-qsbr, unless signals are disabled explicitly around each
 225         rcu_quiescent_state() calls, when threads are put offline and around
 226         calls to synchronize_rcu(). Even then, we do not recommend it.
 227
 228 Interaction with mutexes
 229
 230         One must be careful to do not cause deadlocks due to interaction of
 231         synchronize_rcu() and RCU read-side with mutexes. If synchronize_rcu()
 232         is called with a mutex held, this mutex (or any mutex which has this
 233         mutex in its dependency chain) should not be acquired from within a RCU
 234         read-side critical section.
 235
 236         This is especially important to understand in the context of the
 237         QSBR flavor: a registered reader thread being "online" by
 238         default should be considered as within a RCU read-side critical
 239         section unless explicitly put "offline". Therefore, if
 240         synchronize_rcu() is called with a mutex held, this mutex, as
 241         well as any mutex which has this mutex in its dependency chain
 242         should only be taken when the RCU reader thread is "offline"
 243         (this can be performed by calling rcu_thread_offline()).
 244
 245 Interaction with fork()
 246
 247         Special care must be taken for applications performing fork() without
 248         any following exec(). This is caused by the fact that Linux only clones
 249         the thread calling fork(), and thus never replicates any of the other
 250         parent thread into the child process. Most liburcu implementations
 251         require that all registrations (as reader, defer_rcu and call_rcu
 252         threads) should be released before a fork() is performed, except for the
 253         rather common scenario where fork() is immediately followed by exec() in
 254         the child process. The only implementation not subject to that rule is
 255         liburcu-bp, which is designed to handle fork() by calling
 256         rcu_bp_before_fork, rcu_bp_after_fork_parent and
 257         rcu_bp_after_fork_child.
 258
 259         Applications that use call_rcu() and that fork() without
 260         doing an immediate exec() must take special action.  The parent
 261         must invoke call_rcu_before_fork() before the fork() and
 262         call_rcu_after_fork_parent() after the fork().  The child
 263         process must invoke call_rcu_after_fork_child().
 264         Even though these three APIs are suitable for passing to
 265         pthread_atfork(), use of pthread_atfork() is *STRONGLY
 266         DISCOURAGED* for programs calling the glibc memory allocator
 267         (malloc(), calloc(), free(), ...) within call_rcu callbacks.
 268         This is due to limitations in the way glibc memory allocator
 269         handles calls to the memory allocator from concurrent threads
 270         while the pthread_atfork() handlers are executing.
 271         Combining e.g.:
 272         * call to free() from callbacks executed within call_rcu worker
 273           threads,
 274         * executing call_rcu atfork handlers within the glibc pthread
 275           atfork mechanism,
 276         will sometimes trigger interesting process hangs. This usually
 277         hangs on a memory allocator lock within glibc.
 278
 279 Thread Local Storage (TLS)
 280
 281         Userspace RCU can fall back on pthread_getspecific() to emulate
 282         TLS variables on systems where it is not available. This behavior
 283         can be forced by specifying --disable-compiler-tls as configure
 284         argument.
 285
 286 Usage of DEBUG_RCU
 287
 288         DEBUG_RCU is used to add internal debugging self-checks to the
 289         RCU library. This define adds a performance penalty when enabled.
 290         Can be enabled by uncommenting the corresponding line in
 291         Makefile.build.inc.
 292
 293 Usage of DEBUG_YIELD
 294
 295         DEBUG_YIELD is used to add random delays in the code for testing
 296         purposes.
 297
 298 SMP support
 299
 300         By default the library is configured to use synchronization primitives
 301         adequate for SMP systems. On uniprocessor systems, support for SMP
 302         systems can be disabled with:
 303
 304                 ./configure --disable-smp-support
 305
 306         theoretically yielding slightly better performance.