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 ARCHITECTURES SUPPORTED
  26 -----------------------
  27
  28 Currently, x86 (i386, i486, i586, i686), x86 64-bit, PowerPC 32/64, S390, S390x,
  29 ARM, Alpha, ia64 and Sparcv9 32/64 are supported. Only tested on Linux so
  30 far, but should theoretically work on other operating systems.
  31
  32 ARM depends on running a Linux kernel 2.6.15 or better.
  33
  34 The gcc compiler versions 3.3, 3.4, 4.0, 4.1, 4.2, 4.3, 4.4 and 4.5 are
  35 supported, with the following exceptions:
  36
  37 - gcc 3.3 and 3.4 have a bug that prevents them from generating volatile
  38   accesses to offsets in a TLS structure on 32-bit x86. These versions are
  39   therefore not compatible with liburcu on x86 32-bit (i386, i486, i586, i686).
  40   The problem has been reported to the gcc community:
  41     http://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg281255.html
  42 - gcc 3.3 cannot match the "xchg" instruction on 32-bit x86 build.
  43   See: http://kerneltrap.org/node/7507
  44 - Alpha, ia64 and ARM architectures depend on 4.x gcc with atomic builtins
  45   support.
  46
  47 For developers using the git tree:
  48
  49 This source tree is based on the autotools suite from GNU to simplify
  50 portability. Here are some things you should have on your system in order to
  51 compile the git repository tree :
  52
  53 - GNU autotools (automake >=1.10, autoconf >=2.50, autoheader >=2.50)
  54   (make sure your system wide "automake" points to a recent version!)
  55 - GNU Libtool >=2.2
  56   (for more information, go to http://www.gnu.org/software/autoconf/)
  57
  58 If you get the tree from the repository, you will need to use the "bootstrap"
  59 script in the root of the tree. It calls all the GNU tools needed to prepare the
  60 tree configuration.
  61
  62
  63 QUICK START GUIDE
  64 -----------------
  65
  66 Usage of all urcu libraries
  67
  68         * Define _LGPL_SOURCE (only) if your code is LGPL or GPL compatible
  69           before including the urcu.h or urcu-qsbr.h header. If your application
  70           is distributed under another license, function calls will be generated
  71           instead of inlines, so your application can link with the library.
  72         * Linking with one of the libraries below is always necessary even for
  73           LGPL and GPL applications.
  74
  75 Usage of liburcu
  76
  77         * #include <urcu.h>
  78         * Link the application with "-lurcu".
  79         * This is the preferred version of the library, in terms of
  80           grace-period detection speed, read-side speed and flexibility.
  81           Dynamically detects kernel support for sys_membarrier(). Falls back
  82           on urcu-mb scheme if support is not present, which has slower
  83           read-side.
  84
  85 Usage of liburcu-qsbr
  86
  87         * #include <urcu-qsbr.h>
  88         * Link with "-lurcu-qsbr".
  89         * The QSBR flavor of RCU needs to have each reader thread executing
  90           rcu_quiescent_state() periodically to progress. rcu_thread_online()
  91           and rcu_thread_offline() can be used to mark long periods for which
  92           the threads are not active. It provides the fastest read-side at the
  93           expense of more intrusiveness in the application code.
  94
  95 Usage of liburcu-mb
  96
  97         * #include <urcu.h>
  98         * Compile any _LGPL_SOURCE code using this library with "-DRCU_MB".
  99         * Link with "-lurcu-mb".
 100         * This version of the urcu library uses memory barriers on the writer
 101           and reader sides. This results in faster grace-period detection, but
 102           results in slower reads.
 103
 104 Usage of liburcu-signal
 105
 106         * #include <urcu.h>
 107         * Compile any _LGPL_SOURCE code using this library with "-DRCU_SIGNAL".
 108         * Link the application with "-lurcu-signal".
 109         * Version of the library that requires a signal, typically SIGUSR1. Can
 110           be overridden with -DSIGRCU by modifying Makefile.build.inc.
 111
 112 Usage of liburcu-bp
 113
 114         * #include <urcu-bp.h>
 115         * Link with "-lurcu-bp".
 116         * The BP library flavor stands for "bulletproof". It is specifically
 117           designed to help tracing library to hook on applications without
 118           requiring to modify these applications. rcu_init(),
 119           rcu_register_thread() and rcu_unregister_thread() all become nops.
 120           The state is dealt with by the library internally at the expense of
 121           read-side and write-side performance.
 122
 123 Initialization
 124
 125         Each thread that has reader critical sections (that uses
 126         rcu_read_lock()/rcu_read_unlock() must first register to the URCU
 127         library. This is done by calling rcu_register_thread(). Unregistration
 128         must be performed before exiting the thread by using
 129         rcu_unregister_thread().
 130
 131 Reading
 132
 133         Reader critical sections must be protected by locating them between
 134         calls to rcu_read_lock() and rcu_read_unlock(). Inside that lock,
 135         rcu_dereference() may be called to read an RCU protected pointer.
 136
 137 Writing
 138
 139         rcu_assign_pointer() and rcu_xchg_pointer() may be called anywhere.
 140         After, synchronize_rcu() must be called. When it returns, the old
 141         values are not in usage anymore.
 142
 143 Usage of liburcu-defer
 144
 145         * Follow instructions for either liburcu, liburcu-qsbr,
 146           liburcu-mb, liburcu-signal, or liburcu-bp above.
 147           The liburcu-defer functionality is pulled into each of
 148           those library modules.
 149         * Provides defer_rcu() primitive to enqueue delayed callbacks. Queued
 150           callbacks are executed in batch periodically after a grace period.
 151           Do _not_ use defer_rcu() within a read-side critical section, because
 152           it may call synchronize_rcu() if the thread queue is full.
 153           This can lead to deadlock or worse.
 154         * Requires that rcu_defer_barrier() must be called in library destructor
 155           if a library queues callbacks and is expected to be unloaded with
 156           dlclose().
 157         * Its API is currently experimental. It may change in future library
 158           releases.
 159
 160 Usage of urcu-call-rcu
 161
 162         * Follow instructions for either liburcu, liburcu-qsbr,
 163           liburcu-mb, liburcu-signal, or liburcu-bp above.
 164           The urcu-call-rcu functionality is provided for each of
 165           these library modules.
 166         * Provides the call_rcu() primitive to enqueue delayed callbacks
 167           in a manner similar to defer_rcu(), but without ever delaying
 168           for a grace period.  On the other hand, call_rcu()'s best-case
 169           overhead is not quite as good as that of defer_rcu().
 170         * Provides call_rcu() to allow asynchronous handling of RCU
 171           grace periods.  A number of additional functions are provided
 172           to manage the helper threads used by call_rcu(), but reasonable
 173           defaults are used if these additional functions are not invoked.
 174           See API.txt for more details.
 175
 176 Being careful with signals
 177
 178         The liburcu library uses signals internally. The signal handler is
 179         registered with the SA_RESTART flag. However, these signals may cause
 180         some non-restartable system calls to fail with errno = EINTR. Care
 181         should be taken to restart system calls manually if they fail with this
 182         error. A list of non-restartable system calls may be found in
 183         signal(7). The liburcu-mb and liburcu-qsbr versions of the Userspace RCU
 184         library do not require any signal.
 185
 186         Read-side critical sections are allowed in a signal handler with
 187         liburcu and liburcu-mb. Be careful, however, to disable these signals
 188         between thread creation and calls to rcu_register_thread(), because a
 189         signal handler nesting on an unregistered thread would not be allowed to
 190         call rcu_read_lock().
 191
 192         Read-side critical sections are _not_ allowed in a signal handler with
 193         liburcu-qsbr, unless signals are disabled explicitly around each
 194         rcu_quiescent_state() calls, when threads are put offline and around
 195         calls to synchronize_rcu(). Even then, we do not recommend it.
 196
 197 Interaction with mutexes
 198
 199         One must be careful to do not cause deadlocks due to interaction of
 200         synchronize_rcu() and RCU read-side with mutexes. If synchronize_rcu()
 201         is called with a mutex held, this mutex (or any mutex which has this
 202         mutex in its dependency chain) should not be acquired from within a RCU
 203         read-side critical section.
 204
 205         This is especially important to understand in the context of the
 206         QSBR flavor: a registered reader thread being "online" by
 207         default should be considered as within a RCU read-side critical
 208         section unless explicitly put "offline". Therefore, if
 209         synchronize_rcu() is called with a mutex held, this mutex, as
 210         well as any mutex which has this mutex in its dependency chain
 211         should only be taken when the RCU reader thread is "offline"
 212         (this can be performed by calling rcu_thread_offline()).
 213
 214 Usage of DEBUG_RCU
 215
 216         DEBUG_RCU is used to add internal debugging self-checks to the
 217         RCU library. This define adds a performance penalty when enabled.
 218         Can be enabled by uncommenting the corresponding line in
 219         Makefile.build.inc.
 220
 221 Usage of DEBUG_YIELD
 222
 223         DEBUG_YIELD is used to add random delays in the code for testing
 224         purposes.
 225
 226 SMP support
 227
 228         By default the library is configured to use synchronization primitives
 229         adequate for SMP systems. On uniprocessor systems, support for SMP
 230         systems can be disabled with:
 231
 232                 ./configure --disable-smp-support
 233
 234         theoretically yielding slightly better performance.
 235
 236 Interaction with fork()
 237
 238         Special care must be taken for applications performing fork() without
 239         any following exec(). This is caused by the fact that Linux only clones
 240         the thread calling fork(), and thus never replicates any of the other
 241         parent thread into the child process. Most liburcu implementations
 242         require that all registrations (as reader, defer_rcu and call_rcu
 243         threads) should be released before a fork() is performed, except for the
 244         rather common scenario where fork() is immediately followed by exec() in
 245         the child process. The only implementation not subject to that rule is
 246         liburcu-bp, which is designed to handle fork() by calling
 247         rcu_bp_before_fork, rcu_bp_after_fork_parent and
 248         rcu_bp_after_fork_child.
 249
 250         Applications that use call_rcu() and that fork() without
 251         doing an immediate exec() must take special action.  The parent
 252         must invoke call_rcu_before_fork() before the fork() and
 253         call_rcu_after_fork_parent() after the fork().  The child
 254         process must invoke call_rcu_after_fork_child().
 255         These three APIs are suitable for passing to pthread_atfork().