1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/share/vm/runtime/park.cpp Fri Oct 22 15:59:34 2010 -0400
1.3 @@ -0,0 +1,237 @@
1.4 +/*
1.5 + * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
1.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
1.7 + *
1.8 + * This code is free software; you can redistribute it and/or modify it
1.9 + * under the terms of the GNU General Public License version 2 only, as
1.10 + * published by the Free Software Foundation.
1.11 + *
1.12 + * This code is distributed in the hope that it will be useful, but WITHOUT
1.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
1.14 + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
1.15 + * version 2 for more details (a copy is included in the LICENSE file that
1.16 + * accompanied this code).
1.17 + *
1.18 + * You should have received a copy of the GNU General Public License version
1.19 + * 2 along with this work; if not, write to the Free Software Foundation,
1.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
1.21 + *
1.22 + * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
1.23 + * or visit www.oracle.com if you need additional information or have any
1.24 + * questions.
1.25 + *
1.26 + */
1.27 +
1.28 +
1.29 +# include "incls/_precompiled.incl"
1.30 +# include "incls/_park.cpp.incl"
1.31 +
1.32 +
1.33 +// Lifecycle management for TSM ParkEvents.
1.34 +// ParkEvents are type-stable (TSM).
1.35 +// In our particular implementation they happen to be immortal.
1.36 +//
1.37 +// We manage concurrency on the FreeList with a CAS-based
1.38 +// detach-modify-reattach idiom that avoids the ABA problems
1.39 +// that would otherwise be present in a simple CAS-based
1.40 +// push-pop implementation. (push-one and pop-all)
1.41 +//
1.42 +// Caveat: Allocate() and Release() may be called from threads
1.43 +// other than the thread associated with the Event!
1.44 +// If we need to call Allocate() when running as the thread in
1.45 +// question then look for the PD calls to initialize native TLS.
1.46 +// Native TLS (Win32/Linux/Solaris) can only be initialized or
1.47 +// accessed by the associated thread.
1.48 +// See also pd_initialize().
1.49 +//
1.50 +// Note that we could defer associating a ParkEvent with a thread
1.51 +// until the 1st time the thread calls park(). unpark() calls to
1.52 +// an unprovisioned thread would be ignored. The first park() call
1.53 +// for a thread would allocate and associate a ParkEvent and return
1.54 +// immediately.
1.55 +
1.56 +volatile int ParkEvent::ListLock = 0 ;
1.57 +ParkEvent * volatile ParkEvent::FreeList = NULL ;
1.58 +
1.59 +ParkEvent * ParkEvent::Allocate (Thread * t) {
1.60 + // In rare cases -- JVM_RawMonitor* operations -- we can find t == null.
1.61 + ParkEvent * ev ;
1.62 +
1.63 + // Start by trying to recycle an existing but unassociated
1.64 + // ParkEvent from the global free list.
1.65 + for (;;) {
1.66 + ev = FreeList ;
1.67 + if (ev == NULL) break ;
1.68 + // 1: Detach - sequester or privatize the list
1.69 + // Tantamount to ev = Swap (&FreeList, NULL)
1.70 + if (Atomic::cmpxchg_ptr (NULL, &FreeList, ev) != ev) {
1.71 + continue ;
1.72 + }
1.73 +
1.74 + // We've detached the list. The list in-hand is now
1.75 + // local to this thread. This thread can operate on the
1.76 + // list without risk of interference from other threads.
1.77 + // 2: Extract -- pop the 1st element from the list.
1.78 + ParkEvent * List = ev->FreeNext ;
1.79 + if (List == NULL) break ;
1.80 + for (;;) {
1.81 + // 3: Try to reattach the residual list
1.82 + guarantee (List != NULL, "invariant") ;
1.83 + ParkEvent * Arv = (ParkEvent *) Atomic::cmpxchg_ptr (List, &FreeList, NULL) ;
1.84 + if (Arv == NULL) break ;
1.85 +
1.86 + // New nodes arrived. Try to detach the recent arrivals.
1.87 + if (Atomic::cmpxchg_ptr (NULL, &FreeList, Arv) != Arv) {
1.88 + continue ;
1.89 + }
1.90 + guarantee (Arv != NULL, "invariant") ;
1.91 + // 4: Merge Arv into List
1.92 + ParkEvent * Tail = List ;
1.93 + while (Tail->FreeNext != NULL) Tail = Tail->FreeNext ;
1.94 + Tail->FreeNext = Arv ;
1.95 + }
1.96 + break ;
1.97 + }
1.98 +
1.99 + if (ev != NULL) {
1.100 + guarantee (ev->AssociatedWith == NULL, "invariant") ;
1.101 + } else {
1.102 + // Do this the hard way -- materialize a new ParkEvent.
1.103 + // In rare cases an allocating thread might detach a long list --
1.104 + // installing null into FreeList -- and then stall or be obstructed.
1.105 + // A 2nd thread calling Allocate() would see FreeList == null.
1.106 + // The list held privately by the 1st thread is unavailable to the 2nd thread.
1.107 + // In that case the 2nd thread would have to materialize a new ParkEvent,
1.108 + // even though free ParkEvents existed in the system. In this case we end up
1.109 + // with more ParkEvents in circulation than we need, but the race is
1.110 + // rare and the outcome is benign. Ideally, the # of extant ParkEvents
1.111 + // is equal to the maximum # of threads that existed at any one time.
1.112 + // Because of the race mentioned above, segments of the freelist
1.113 + // can be transiently inaccessible. At worst we may end up with the
1.114 + // # of ParkEvents in circulation slightly above the ideal.
1.115 + // Note that if we didn't have the TSM/immortal constraint, then
1.116 + // when reattaching, above, we could trim the list.
1.117 + ev = new ParkEvent () ;
1.118 + guarantee ((intptr_t(ev) & 0xFF) == 0, "invariant") ;
1.119 + }
1.120 + ev->reset() ; // courtesy to caller
1.121 + ev->AssociatedWith = t ; // Associate ev with t
1.122 + ev->FreeNext = NULL ;
1.123 + return ev ;
1.124 +}
1.125 +
1.126 +void ParkEvent::Release (ParkEvent * ev) {
1.127 + if (ev == NULL) return ;
1.128 + guarantee (ev->FreeNext == NULL , "invariant") ;
1.129 + ev->AssociatedWith = NULL ;
1.130 + for (;;) {
1.131 + // Push ev onto FreeList
1.132 + // The mechanism is "half" lock-free.
1.133 + ParkEvent * List = FreeList ;
1.134 + ev->FreeNext = List ;
1.135 + if (Atomic::cmpxchg_ptr (ev, &FreeList, List) == List) break ;
1.136 + }
1.137 +}
1.138 +
1.139 +// Override operator new and delete so we can ensure that the
1.140 +// least significant byte of ParkEvent addresses is 0.
1.141 +// Beware that excessive address alignment is undesirable
1.142 +// as it can result in D$ index usage imbalance as
1.143 +// well as bank access imbalance on Niagara-like platforms,
1.144 +// although Niagara's hash function should help.
1.145 +
1.146 +void * ParkEvent::operator new (size_t sz) {
1.147 + return (void *) ((intptr_t (CHeapObj::operator new (sz + 256)) + 256) & -256) ;
1.148 +}
1.149 +
1.150 +void ParkEvent::operator delete (void * a) {
1.151 + // ParkEvents are type-stable and immortal ...
1.152 + ShouldNotReachHere();
1.153 +}
1.154 +
1.155 +
1.156 +// 6399321 As a temporary measure we copied & modified the ParkEvent::
1.157 +// allocate() and release() code for use by Parkers. The Parker:: forms
1.158 +// will eventually be removed as we consolide and shift over to ParkEvents
1.159 +// for both builtin synchronization and JSR166 operations.
1.160 +
1.161 +volatile int Parker::ListLock = 0 ;
1.162 +Parker * volatile Parker::FreeList = NULL ;
1.163 +
1.164 +Parker * Parker::Allocate (JavaThread * t) {
1.165 + guarantee (t != NULL, "invariant") ;
1.166 + Parker * p ;
1.167 +
1.168 + // Start by trying to recycle an existing but unassociated
1.169 + // Parker from the global free list.
1.170 + for (;;) {
1.171 + p = FreeList ;
1.172 + if (p == NULL) break ;
1.173 + // 1: Detach
1.174 + // Tantamount to p = Swap (&FreeList, NULL)
1.175 + if (Atomic::cmpxchg_ptr (NULL, &FreeList, p) != p) {
1.176 + continue ;
1.177 + }
1.178 +
1.179 + // We've detached the list. The list in-hand is now
1.180 + // local to this thread. This thread can operate on the
1.181 + // list without risk of interference from other threads.
1.182 + // 2: Extract -- pop the 1st element from the list.
1.183 + Parker * List = p->FreeNext ;
1.184 + if (List == NULL) break ;
1.185 + for (;;) {
1.186 + // 3: Try to reattach the residual list
1.187 + guarantee (List != NULL, "invariant") ;
1.188 + Parker * Arv = (Parker *) Atomic::cmpxchg_ptr (List, &FreeList, NULL) ;
1.189 + if (Arv == NULL) break ;
1.190 +
1.191 + // New nodes arrived. Try to detach the recent arrivals.
1.192 + if (Atomic::cmpxchg_ptr (NULL, &FreeList, Arv) != Arv) {
1.193 + continue ;
1.194 + }
1.195 + guarantee (Arv != NULL, "invariant") ;
1.196 + // 4: Merge Arv into List
1.197 + Parker * Tail = List ;
1.198 + while (Tail->FreeNext != NULL) Tail = Tail->FreeNext ;
1.199 + Tail->FreeNext = Arv ;
1.200 + }
1.201 + break ;
1.202 + }
1.203 +
1.204 + if (p != NULL) {
1.205 + guarantee (p->AssociatedWith == NULL, "invariant") ;
1.206 + } else {
1.207 + // Do this the hard way -- materialize a new Parker..
1.208 + // In rare cases an allocating thread might detach
1.209 + // a long list -- installing null into FreeList --and
1.210 + // then stall. Another thread calling Allocate() would see
1.211 + // FreeList == null and then invoke the ctor. In this case we
1.212 + // end up with more Parkers in circulation than we need, but
1.213 + // the race is rare and the outcome is benign.
1.214 + // Ideally, the # of extant Parkers is equal to the
1.215 + // maximum # of threads that existed at any one time.
1.216 + // Because of the race mentioned above, segments of the
1.217 + // freelist can be transiently inaccessible. At worst
1.218 + // we may end up with the # of Parkers in circulation
1.219 + // slightly above the ideal.
1.220 + p = new Parker() ;
1.221 + }
1.222 + p->AssociatedWith = t ; // Associate p with t
1.223 + p->FreeNext = NULL ;
1.224 + return p ;
1.225 +}
1.226 +
1.227 +
1.228 +void Parker::Release (Parker * p) {
1.229 + if (p == NULL) return ;
1.230 + guarantee (p->AssociatedWith != NULL, "invariant") ;
1.231 + guarantee (p->FreeNext == NULL , "invariant") ;
1.232 + p->AssociatedWith = NULL ;
1.233 + for (;;) {
1.234 + // Push p onto FreeList
1.235 + Parker * List = FreeList ;
1.236 + p->FreeNext = List ;
1.237 + if (Atomic::cmpxchg_ptr (p, &FreeList, List) == List) break ;
1.238 + }
1.239 +}
1.240 +
