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