acorn@2233: /*
coleenp@5614:  * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
acorn@2233:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
acorn@2233:  *
acorn@2233:  * This code is free software; you can redistribute it and/or modify it
acorn@2233:  * under the terms of the GNU General Public License version 2 only, as
acorn@2233:  * published by the Free Software Foundation.
acorn@2233:  *
acorn@2233:  * This code is distributed in the hope that it will be useful, but WITHOUT
acorn@2233:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
acorn@2233:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
acorn@2233:  * version 2 for more details (a copy is included in the LICENSE file that
acorn@2233:  * accompanied this code).
acorn@2233:  *
acorn@2233:  * You should have received a copy of the GNU General Public License version
acorn@2233:  * 2 along with this work; if not, write to the Free Software Foundation,
acorn@2233:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
acorn@2233:  *
acorn@2233:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
acorn@2233:  * or visit www.oracle.com if you need additional information or have any
acorn@2233:  * questions.
acorn@2233:  *
acorn@2233:  */
acorn@2233: 
stefank@2314: #include "precompiled.hpp"
stefank@2314: #include "runtime/thread.hpp"
acorn@2233: 
acorn@2233: 
acorn@2233: 
acorn@2233: // Lifecycle management for TSM ParkEvents.
acorn@2233: // ParkEvents are type-stable (TSM).
acorn@2233: // In our particular implementation they happen to be immortal.
acorn@2233: //
acorn@2233: // We manage concurrency on the FreeList with a CAS-based
acorn@2233: // detach-modify-reattach idiom that avoids the ABA problems
acorn@2233: // that would otherwise be present in a simple CAS-based
acorn@2233: // push-pop implementation.   (push-one and pop-all)
acorn@2233: //
acorn@2233: // Caveat: Allocate() and Release() may be called from threads
acorn@2233: // other than the thread associated with the Event!
acorn@2233: // If we need to call Allocate() when running as the thread in
acorn@2233: // question then look for the PD calls to initialize native TLS.
acorn@2233: // Native TLS (Win32/Linux/Solaris) can only be initialized or
acorn@2233: // accessed by the associated thread.
acorn@2233: // See also pd_initialize().
acorn@2233: //
acorn@2233: // Note that we could defer associating a ParkEvent with a thread
acorn@2233: // until the 1st time the thread calls park().  unpark() calls to
acorn@2233: // an unprovisioned thread would be ignored.  The first park() call
acorn@2233: // for a thread would allocate and associate a ParkEvent and return
acorn@2233: // immediately.
acorn@2233: 
acorn@2233: volatile int ParkEvent::ListLock = 0 ;
acorn@2233: ParkEvent * volatile ParkEvent::FreeList = NULL ;
acorn@2233: 
acorn@2233: ParkEvent * ParkEvent::Allocate (Thread * t) {
acorn@2233:   // In rare cases -- JVM_RawMonitor* operations -- we can find t == null.
acorn@2233:   ParkEvent * ev ;
acorn@2233: 
acorn@2233:   // Start by trying to recycle an existing but unassociated
acorn@2233:   // ParkEvent from the global free list.
acorn@2233:   for (;;) {
acorn@2233:     ev = FreeList ;
acorn@2233:     if (ev == NULL) break ;
acorn@2233:     // 1: Detach - sequester or privatize the list
acorn@2233:     // Tantamount to ev = Swap (&FreeList, NULL)
acorn@2233:     if (Atomic::cmpxchg_ptr (NULL, &FreeList, ev) != ev) {
acorn@2233:        continue ;
acorn@2233:     }
acorn@2233: 
acorn@2233:     // We've detached the list.  The list in-hand is now
acorn@2233:     // local to this thread.   This thread can operate on the
acorn@2233:     // list without risk of interference from other threads.
acorn@2233:     // 2: Extract -- pop the 1st element from the list.
acorn@2233:     ParkEvent * List = ev->FreeNext ;
acorn@2233:     if (List == NULL) break ;
acorn@2233:     for (;;) {
acorn@2233:         // 3: Try to reattach the residual list
acorn@2233:         guarantee (List != NULL, "invariant") ;
acorn@2233:         ParkEvent * Arv =  (ParkEvent *) Atomic::cmpxchg_ptr (List, &FreeList, NULL) ;
acorn@2233:         if (Arv == NULL) break ;
acorn@2233: 
acorn@2233:         // New nodes arrived.  Try to detach the recent arrivals.
acorn@2233:         if (Atomic::cmpxchg_ptr (NULL, &FreeList, Arv) != Arv) {
acorn@2233:             continue ;
acorn@2233:         }
acorn@2233:         guarantee (Arv != NULL, "invariant") ;
acorn@2233:         // 4: Merge Arv into List
acorn@2233:         ParkEvent * Tail = List ;
acorn@2233:         while (Tail->FreeNext != NULL) Tail = Tail->FreeNext ;
acorn@2233:         Tail->FreeNext = Arv ;
acorn@2233:     }
acorn@2233:     break ;
acorn@2233:   }
acorn@2233: 
acorn@2233:   if (ev != NULL) {
acorn@2233:     guarantee (ev->AssociatedWith == NULL, "invariant") ;
acorn@2233:   } else {
acorn@2233:     // Do this the hard way -- materialize a new ParkEvent.
acorn@2233:     // In rare cases an allocating thread might detach a long list --
acorn@2233:     // installing null into FreeList -- and then stall or be obstructed.
acorn@2233:     // A 2nd thread calling Allocate() would see FreeList == null.
acorn@2233:     // The list held privately by the 1st thread is unavailable to the 2nd thread.
acorn@2233:     // In that case the 2nd thread would have to materialize a new ParkEvent,
acorn@2233:     // even though free ParkEvents existed in the system.  In this case we end up
acorn@2233:     // with more ParkEvents in circulation than we need, but the race is
acorn@2233:     // rare and the outcome is benign.  Ideally, the # of extant ParkEvents
acorn@2233:     // is equal to the maximum # of threads that existed at any one time.
acorn@2233:     // Because of the race mentioned above, segments of the freelist
acorn@2233:     // can be transiently inaccessible.  At worst we may end up with the
acorn@2233:     // # of ParkEvents in circulation slightly above the ideal.
acorn@2233:     // Note that if we didn't have the TSM/immortal constraint, then
acorn@2233:     // when reattaching, above, we could trim the list.
acorn@2233:     ev = new ParkEvent () ;
acorn@2233:     guarantee ((intptr_t(ev) & 0xFF) == 0, "invariant") ;
acorn@2233:   }
acorn@2233:   ev->reset() ;                     // courtesy to caller
acorn@2233:   ev->AssociatedWith = t ;          // Associate ev with t
acorn@2233:   ev->FreeNext       = NULL ;
acorn@2233:   return ev ;
acorn@2233: }
acorn@2233: 
acorn@2233: void ParkEvent::Release (ParkEvent * ev) {
acorn@2233:   if (ev == NULL) return ;
acorn@2233:   guarantee (ev->FreeNext == NULL      , "invariant") ;
acorn@2233:   ev->AssociatedWith = NULL ;
acorn@2233:   for (;;) {
acorn@2233:     // Push ev onto FreeList
acorn@2233:     // The mechanism is "half" lock-free.
acorn@2233:     ParkEvent * List = FreeList ;
acorn@2233:     ev->FreeNext = List ;
acorn@2233:     if (Atomic::cmpxchg_ptr (ev, &FreeList, List) == List) break ;
acorn@2233:   }
acorn@2233: }
acorn@2233: 
acorn@2233: // Override operator new and delete so we can ensure that the
acorn@2233: // least significant byte of ParkEvent addresses is 0.
acorn@2233: // Beware that excessive address alignment is undesirable
acorn@2233: // as it can result in D$ index usage imbalance as
acorn@2233: // well as bank access imbalance on Niagara-like platforms,
acorn@2233: // although Niagara's hash function should help.
acorn@2233: 
coleenp@5614: void * ParkEvent::operator new (size_t sz) throw() {
zgu@3900:   return (void *) ((intptr_t (AllocateHeap(sz + 256, mtInternal, CALLER_PC)) + 256) & -256) ;
acorn@2233: }
acorn@2233: 
acorn@2233: void ParkEvent::operator delete (void * a) {
acorn@2233:   // ParkEvents are type-stable and immortal ...
acorn@2233:   ShouldNotReachHere();
acorn@2233: }
acorn@2233: 
acorn@2233: 
acorn@2233: // 6399321 As a temporary measure we copied & modified the ParkEvent::
acorn@2233: // allocate() and release() code for use by Parkers.  The Parker:: forms
acorn@2233: // will eventually be removed as we consolide and shift over to ParkEvents
acorn@2233: // for both builtin synchronization and JSR166 operations.
acorn@2233: 
acorn@2233: volatile int Parker::ListLock = 0 ;
acorn@2233: Parker * volatile Parker::FreeList = NULL ;
acorn@2233: 
acorn@2233: Parker * Parker::Allocate (JavaThread * t) {
acorn@2233:   guarantee (t != NULL, "invariant") ;
acorn@2233:   Parker * p ;
acorn@2233: 
acorn@2233:   // Start by trying to recycle an existing but unassociated
acorn@2233:   // Parker from the global free list.
acorn@2233:   for (;;) {
acorn@2233:     p = FreeList ;
acorn@2233:     if (p  == NULL) break ;
acorn@2233:     // 1: Detach
acorn@2233:     // Tantamount to p = Swap (&FreeList, NULL)
acorn@2233:     if (Atomic::cmpxchg_ptr (NULL, &FreeList, p) != p) {
acorn@2233:        continue ;
acorn@2233:     }
acorn@2233: 
acorn@2233:     // We've detached the list.  The list in-hand is now
acorn@2233:     // local to this thread.   This thread can operate on the
acorn@2233:     // list without risk of interference from other threads.
acorn@2233:     // 2: Extract -- pop the 1st element from the list.
acorn@2233:     Parker * List = p->FreeNext ;
acorn@2233:     if (List == NULL) break ;
acorn@2233:     for (;;) {
acorn@2233:         // 3: Try to reattach the residual list
acorn@2233:         guarantee (List != NULL, "invariant") ;
acorn@2233:         Parker * Arv =  (Parker *) Atomic::cmpxchg_ptr (List, &FreeList, NULL) ;
acorn@2233:         if (Arv == NULL) break ;
acorn@2233: 
acorn@2233:         // New nodes arrived.  Try to detach the recent arrivals.
acorn@2233:         if (Atomic::cmpxchg_ptr (NULL, &FreeList, Arv) != Arv) {
acorn@2233:             continue ;
acorn@2233:         }
acorn@2233:         guarantee (Arv != NULL, "invariant") ;
acorn@2233:         // 4: Merge Arv into List
acorn@2233:         Parker * Tail = List ;
acorn@2233:         while (Tail->FreeNext != NULL) Tail = Tail->FreeNext ;
acorn@2233:         Tail->FreeNext = Arv ;
acorn@2233:     }
acorn@2233:     break ;
acorn@2233:   }
acorn@2233: 
acorn@2233:   if (p != NULL) {
acorn@2233:     guarantee (p->AssociatedWith == NULL, "invariant") ;
acorn@2233:   } else {
acorn@2233:     // Do this the hard way -- materialize a new Parker..
acorn@2233:     // In rare cases an allocating thread might detach
acorn@2233:     // a long list -- installing null into FreeList --and
acorn@2233:     // then stall.  Another thread calling Allocate() would see
acorn@2233:     // FreeList == null and then invoke the ctor.  In this case we
acorn@2233:     // end up with more Parkers in circulation than we need, but
acorn@2233:     // the race is rare and the outcome is benign.
acorn@2233:     // Ideally, the # of extant Parkers is equal to the
acorn@2233:     // maximum # of threads that existed at any one time.
acorn@2233:     // Because of the race mentioned above, segments of the
acorn@2233:     // freelist can be transiently inaccessible.  At worst
acorn@2233:     // we may end up with the # of Parkers in circulation
acorn@2233:     // slightly above the ideal.
acorn@2233:     p = new Parker() ;
acorn@2233:   }
acorn@2233:   p->AssociatedWith = t ;          // Associate p with t
acorn@2233:   p->FreeNext       = NULL ;
acorn@2233:   return p ;
acorn@2233: }
acorn@2233: 
acorn@2233: 
acorn@2233: void Parker::Release (Parker * p) {
acorn@2233:   if (p == NULL) return ;
acorn@2233:   guarantee (p->AssociatedWith != NULL, "invariant") ;
acorn@2233:   guarantee (p->FreeNext == NULL      , "invariant") ;
acorn@2233:   p->AssociatedWith = NULL ;
acorn@2233:   for (;;) {
acorn@2233:     // Push p onto FreeList
acorn@2233:     Parker * List = FreeList ;
acorn@2233:     p->FreeNext = List ;
acorn@2233:     if (Atomic::cmpxchg_ptr (p, &FreeList, List) == List) break ;
acorn@2233:   }
acorn@2233: }
acorn@2233: