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