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