3 #ifndef CDSLIB_URCU_DETAILS_BASE_H
4 #define CDSLIB_URCU_DETAILS_BASE_H
6 #include <cds/algo/atomic.h>
7 #include <cds/gc/details/retired_ptr.h>
8 #include <cds/details/allocator.h>
9 #include <cds/os/thread.h>
10 #include <cds/details/marked_ptr.h>
13 /// User-space Read-Copy Update (URCU) namespace
14 /** @ingroup cds_garbage_collector
17 This namespace contains declarations for different types of Read-Copy Update (%RCU)
18 synchronization primitive and data structures developed for RCU.
19 In <b>libcds</b> %RCU is used as garbage collector.
22 - [2009] M.Desnoyers "Low-Impact Operating System Tracing" PhD Thesis,
23 Chapter 6 "User-Level Implementations of Read-Copy Update"
24 - [2011] M.Desnoyers, P.McKenney, A.Stern, M.Dagenias, J.Walpole "User-Level
25 Implementations of Read-Copy Update"
27 <b>Informal introduction to user-space %RCU</b>
29 [<i>From Desnoyer's papers</i>] %RCU is a synchronization mechanism that was added to the
30 Linux kernel in October of 2002. %RCU achieves scalability improvements by allowing
31 reads to occur concurrently with updates. In contrast to conventional locking
32 primitives that ensure mutual exclusion among concurrent threads regardless of whether
33 they be readers or updaters, or with reader-writer locks that allow concurrent reads
34 but not in the presence of updates, %RCU supports concurrency between multiple updaters
35 and multiple readers. %RCU ensures that data are not freed up until all pre-existing
36 critical sections complete. %RCU defines and uses efficient and scalable mechanisms
37 for deferring reclamation of old data. These mechanisms distribute the work among read and update
38 paths in such a way as to make read paths extremely fast.
40 %RCU readers execute within %RCU read-side critical sections. Each such critical section begins with
41 \p rcu_read_lock(), ends with \p rcu_read_unlock() (in \p libcds these primitives are the methods of
42 GC class and are usually called \p access_lock and \p access_unlock respectively). Read-side
43 critical sections can be nested.
44 The performance benefits of %RCU are due to the fact that \p rcu_read_lock()
45 and \p rcu_read_unlock() are exceedingly fast.
47 When a thread is not in an %RCU read-side critical section, it is in a quiescent state.
48 A quiescent state that persists for a significant time period is an extended quiescent state.
49 Any time period during which every thread has been in at least one quiescent state
50 is a grace period; this implies that every %RCU read-side critical section
51 that starts before a grace period must end before that grace period does.
52 Distinct grace periods may overlap, either partially or completely. Any time period
53 that includes a grace period is by definition itself a grace period.
54 Each grace period is guaranteed to complete as long as all read-side critical sections
55 are finite in duration; thus even a constant flow of such critical sections is unable to
56 extend an %RCU grace period indefinitely.
58 Suppose that readers enclose each of their data-structure traversals in
59 an %RCU read-side critical section. If an updater first removes an element
60 from such a data structure and then waits for a grace period, there can be
61 no more readers accessing that element. The updater can then carry out destructive
62 operations, for example freeing the element, without disturbing any readers.
64 The %RCU update is split into two phases, a removal phase and a reclamation phase.
65 These two phases must be separated by a grace period, for example via the \p synchronize_rcu()
66 primitive, which initiates a grace period and waits for it to finish.
67 During the removal phase, the %RCU update removes elements from a shared data structure.
68 The removed data elements will only be accessible to read-side critical sections
69 that ran concurrently with the removal phase, which are guaranteed to complete before the
70 grace period ends. Therefore the reclamation phase can safely free the data elements
71 removed by the removal phase.
73 Desnoyers describes several classes of user-space %RCU implementations:
74 - The Quiescent-State-Based Reclamation (QSBR) %RCU implementation offers
75 the best possible read-side performance, but requires that each thread periodically
76 calls a function to announce that it is in a quiescent state, thus strongly
77 constraining the application design. This type of %RCU is not implemented in \p libcds.
78 - The general-purpose %RCU implementation places almost no constraints on the application
\92s
79 design, thus being appropriate for use within a general-purpose library, but it has
80 relatively higher read-side overhead. The \p libcds contains several implementations of general-purpose
81 %RCU: \ref general_instant, \ref general_buffered, \ref general_threaded.
82 - The signal-handling %RCU presents an implementation having low read-side overhead and
83 requiring only that the application give up one POSIX signal to %RCU update processing.
84 The \p libcds contains several implementations if signal-handling %RCU: \ref signal_buffered,
87 @note The signal-handled %RCU is defined only for UNIX-like systems, not for Windows.
90 <b>RCU implementation type</b>
92 There are several internal implementation of RCU (all declared in \p %cds::urcu namespace):
93 - \ref general_instant - general purpose RCU with immediate reclamation
94 - \ref general_buffered - general purpose RCU with deferred (buffered) reclamation
95 - \ref general_threaded - general purpose RCU with special reclamation thread
96 - \ref signal_buffered - signal-handling RCU with deferred (buffered) reclamation
97 - \ref signal_threaded - signal-handling RCU with special reclamation thread
99 You cannot create an object of any of those classes directly.
100 Instead, you should use wrapper classes.
101 The wrapper simplifies creation and usage of RCU singleton objects
102 and has the reacher interface that combines interfaces of wrapped class i.e. RCU global part like
103 \p synchronize, and corresponding RCU thread-specific interface like \p access_lock, \p access_unlock and \p retire_ptr.
106 There are several wrapper classes (all declared in \p %cds::urcu namespace)
107 - \ref cds_urcu_general_instant_gc "gc<general_instant>" - general purpose RCU with immediate reclamation,
108 include file <tt><cds/urcu/general_instant.h></tt>
109 - \ref cds_urcu_general_buffered_gc "gc<general_buffered>" - general purpose RCU with deferred (buffered) reclamation,
110 include file <tt><cds/urcu/general_buffered.h></tt>
111 - \ref cds_urcu_general_threaded_gc "gc<general_threaded>" - general purpose RCU with special reclamation thread
112 include file <tt><cds/urcu/general_threaded.h></tt>
113 - \ref cds_urcu_signal_buffered_gc "gc<signal_buffered>" - signal-handling RCU with deferred (buffered) reclamation
114 include file <tt><cds/urcu/signal_buffered.h></tt>
115 - \ref cds_urcu_signal_threaded_gc "gc<signal_threaded>" - signal-handling RCU with special reclamation thread
116 include file <tt><cds/urcu/signal_threaded.h></tt>
118 Any RCU-related container in \p libcds expects that its \p RCU template parameter is one of those wrapper.
120 @anchor cds_urcu_tags
121 For simplicity, in some algorithms instead of using RCU implementation type
122 you should specify corresponding RCU tags (all declared in \p %cds::urcu namespace):
123 - \ref general_instant_tag - for \ref general_instant
124 - \ref general_buffered_tag - for \ref general_buffered
125 - \ref general_threaded_tag - for \ref general_threaded
126 - \ref signal_buffered_tag - for \ref signal_buffered
127 - \ref signal_threaded_tag - for \ref signal_threaded
129 @anchor cds_urcu_performance
132 As a result of our experiments we can range above %RCU implementation in such order,
133 from high to low performance:
134 - <tt>gc<general_buffered></tt> - high
135 - <tt>gc<general_threaded></tt>
136 - <tt>gc<signal_buffered></tt>
137 - <tt>gc<signal_threaded></tt>
138 - <tt>gc<general_instant></tt> - low
140 This estimation is very rough and depends on many factors:
141 type of payload - mostly read-only (seeking) or read-write (inserting and deleting), -
142 a hardware, your application, and so on.
144 @anchor cds_urcu_howto
147 Usually, in your application you use only one \ref cds_urcu_gc "type of RCU" that is the best for your needs.
148 However, the library allows to apply several RCU singleton in one application.
149 The only limitation is that only one object of each RCU type can be instantiated.
150 Since each RCU type is a template class the creation of two object of one RCU type class
151 with different template arguments is an error and is not supported.
152 However, it is correct when your RCU objects relates to different RCU types.
154 @note If you want to use \p %signal_buffered and \p %signal_threaded RCU in your application simultaneously,
155 you should specify different signal number for each signal-handled RCU type on construction time,
156 for example, \p SIGUSR1 and \p SIGUSR2 respectively. By default, both signal-handled RCU implementation
157 share \p SIGUSR1 signal and cannot be applied together.
159 In \p libcds, many GC-based ordered list, set and map template classes have %RCU-related specializations
160 that hide the %RCU specific details.
162 RCU GC is initialized in usual way: you should declare an object of type \p cds::urcu::gc< RCU_type >,
165 #include <cds/urcu/general_buffered.h>
167 typedef cds::urcu::gc< cds::urcu::general_buffered<> > rcu_gpb;
173 // Initialize general_buffered RCU
176 // If main thread uses lock-free containers
177 // the main thread should be attached to libcds infrastructure
178 cds::threading::Manager::attachThread();
180 // Now you can use RCU-based containers in the main thread
188 Each thread that deals with RCU-based container should be initialized first:
190 #include <cds/urcu/general_buffered.h>
191 int myThreadEntryPoint(void *)
193 // Attach the thread to libcds infrastructure
194 cds::threading::Manager::attachThread();
196 // Now you can use RCU-based containers in the thread
199 // Detach thread when terminating
200 cds::threading::Manager::detachThread();
206 # if CDS_OS_INTERFACE == CDS_OSI_UNIX || defined(CDS_DOXYGEN_INVOKED)
207 # define CDS_URCU_SIGNAL_HANDLING_ENABLED 1
210 /// General-purpose URCU type
211 struct general_purpose_rcu {
213 static uint32_t const c_nControlBit = 0x80000000;
214 static uint32_t const c_nNestMask = c_nControlBit - 1;
218 # ifdef CDS_URCU_SIGNAL_HANDLING_ENABLED
219 /// Signal-handling URCU type
220 struct signal_handling_rcu {
222 static uint32_t const c_nControlBit = 0x80000000;
223 static uint32_t const c_nNestMask = c_nControlBit - 1;
228 /// Tag for general_instant URCU
229 struct general_instant_tag: public general_purpose_rcu {
230 typedef general_purpose_rcu rcu_class ; ///< The URCU type
233 /// Tag for general_buffered URCU
234 struct general_buffered_tag: public general_purpose_rcu
236 typedef general_purpose_rcu rcu_class ; ///< The URCU type
239 /// Tag for general_threaded URCU
240 struct general_threaded_tag: public general_purpose_rcu {
241 typedef general_purpose_rcu rcu_class ; ///< The URCU type
244 # ifdef CDS_URCU_SIGNAL_HANDLING_ENABLED
245 /// Tag for signal_buffered URCU
246 struct signal_buffered_tag: public signal_handling_rcu {
247 typedef signal_handling_rcu rcu_class ; ///< The URCU type
250 /// Tag for signal_threaded URCU
251 struct signal_threaded_tag: public signal_handling_rcu {
252 typedef signal_handling_rcu rcu_class ; ///< The URCU type
256 ///@anchor cds_urcu_retired_ptr Retired pointer, i.e. pointer that ready for reclamation
257 typedef cds::gc::details::retired_ptr retired_ptr;
259 /// Pointer to function to free (destruct and deallocate) retired pointer of specific type
260 typedef cds::gc::details::free_retired_ptr_func free_retired_ptr_func;
263 /// Implementation-specific URCU details
265 /// forward declarations
266 template <typename RCUtag>
269 template <typename RCUtag>
272 template <typename RCUtag >
276 class singleton_vtbl {
278 virtual ~singleton_vtbl()
281 virtual void retire_ptr( retired_ptr& p ) = 0;
287 template <typename MarkedPtr> using atomic_marked_ptr = atomics::atomic<MarkedPtr>;
292 template <typename ThreadData>
293 struct thread_list_record {
294 ThreadData * m_pNext ; ///< Next item in thread list
295 atomics::atomic<OS::ThreadId> m_idOwner ; ///< Owner thread id; 0 - the record is free (not owned)
299 , m_idOwner( cds::OS::c_NullThreadId )
302 ~thread_list_record()
308 template <typename RCUtag, class Alloc = CDS_DEFAULT_ALLOCATOR >
311 typedef thread_data<RCUtag> thread_record;
312 typedef cds::details::Allocator< thread_record, Alloc > allocator_type;
315 atomics::atomic<thread_record *> m_pHead;
327 thread_record * alloc()
329 thread_record * pRec;
330 cds::OS::ThreadId const nullThreadId = cds::OS::c_NullThreadId;
331 cds::OS::ThreadId const curThreadId = cds::OS::get_current_thread_id();
333 // First try to reuse a retired (non-active) HP record
334 for ( pRec = m_pHead.load( atomics::memory_order_acquire ); pRec; pRec = pRec->m_list.m_pNext ) {
335 cds::OS::ThreadId thId = nullThreadId;
336 if ( !pRec->m_list.m_idOwner.compare_exchange_strong( thId, curThreadId, atomics::memory_order_seq_cst, atomics::memory_order_relaxed ) )
341 // No records available for reuse
342 // Allocate and push a new record
343 pRec = allocator_type().New();
344 pRec->m_list.m_idOwner.store( curThreadId, atomics::memory_order_relaxed );
346 atomics::atomic_thread_fence( atomics::memory_order_release );
348 thread_record * pOldHead = m_pHead.load( atomics::memory_order_acquire );
350 pRec->m_list.m_pNext = pOldHead;
351 } while ( !m_pHead.compare_exchange_weak( pOldHead, pRec, atomics::memory_order_release, atomics::memory_order_relaxed ));
356 void retire( thread_record * pRec )
358 assert( pRec != nullptr );
359 pRec->m_list.m_idOwner.store( cds::OS::c_NullThreadId, atomics::memory_order_release );
364 thread_record * pNext = nullptr;
365 cds::OS::ThreadId const nullThreadId = cds::OS::c_NullThreadId;
367 for ( thread_record * pRec = m_pHead.load(atomics::memory_order_acquire); pRec; pRec = pNext ) {
368 pNext = pRec->m_list.m_pNext;
369 if ( pRec->m_list.m_idOwner.load(atomics::memory_order_relaxed) != nullThreadId ) {
375 thread_record * head( atomics::memory_order mo ) const
377 return m_pHead.load( mo );
384 CDS_DEBUG_ONLY( cds::OS::ThreadId const nullThreadId = cds::OS::c_NullThreadId; )
385 CDS_DEBUG_ONLY( cds::OS::ThreadId const mainThreadId = cds::OS::get_current_thread_id() ;)
387 thread_record * p = m_pHead.exchange( nullptr, atomics::memory_order_seq_cst );
389 thread_record * pNext = p->m_list.m_pNext;
391 assert( p->m_list.m_idOwner.load( atomics::memory_order_relaxed ) == nullThreadId
392 || p->m_list.m_idOwner.load( atomics::memory_order_relaxed ) == mainThreadId
393 || !cds::OS::is_thread_alive( p->m_list.m_idOwner.load( atomics::memory_order_relaxed ) )
404 template <class ThreadGC>
407 typedef ThreadGC thread_gc;
408 typedef typename thread_gc::rcu_tag rcu_tag;
412 scoped_lock(bool bLock = true)
416 thread_gc::access_lock();
422 thread_gc::access_unlock();
426 } // namespace details
431 template <typename RCUimpl> class gc;
434 /// Epoch-based retired ptr
436 Retired pointer with additional epoch field that prevents early reclamation.
437 This type of retired pointer is used in buffered RCU implementations.
439 struct epoch_retired_ptr: public retired_ptr
441 uint64_t m_nEpoch ; ///< The epoch when the object has been retired
448 /// Constructor creates a copy of \p rp retired pointer and saves \p nEpoch reclamation epoch.
449 epoch_retired_ptr( retired_ptr const& rp, uint64_t nEpoch )
458 #endif // #ifndef CDSLIB_URCU_DETAILS_BASE_H