]> git.saurik.com Git - apple/javascriptcore.git/blobdiff - runtime/Collector.cpp
JavaScriptCore-903.tar.gz
[apple/javascriptcore.git] / runtime / Collector.cpp
diff --git a/runtime/Collector.cpp b/runtime/Collector.cpp
deleted file mode 100644 (file)
index 2789e50..0000000
+++ /dev/null
@@ -1,1299 +0,0 @@
-/*
- *  Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009 Apple Inc. All rights reserved.
- *  Copyright (C) 2007 Eric Seidel <eric@webkit.org>
- *
- *  This library is free software; you can redistribute it and/or
- *  modify it under the terms of the GNU Lesser General Public
- *  License as published by the Free Software Foundation; either
- *  version 2 of the License, or (at your option) any later version.
- *
- *  This library is distributed in the hope that it will be useful,
- *  but WITHOUT ANY WARRANTY; without even the implied warranty of
- *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- *  Lesser General Public License for more details.
- *
- *  You should have received a copy of the GNU Lesser General Public
- *  License along with this library; if not, write to the Free Software
- *  Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
- *
- */
-
-#include "config.h"
-#include "Collector.h"
-
-#include "ArgList.h"
-#include "CallFrame.h"
-#include "CodeBlock.h"
-#include "CollectorHeapIterator.h"
-#include "Interpreter.h"
-#include "JSArray.h"
-#include "JSGlobalObject.h"
-#include "JSLock.h"
-#include "JSONObject.h"
-#include "JSString.h"
-#include "JSValue.h"
-#include "JSZombie.h"
-#include "MarkStack.h"
-#include "Nodes.h"
-#include "Tracing.h"
-#include <algorithm>
-#include <limits.h>
-#include <setjmp.h>
-#include <stdlib.h>
-#include <wtf/FastMalloc.h>
-#include <wtf/HashCountedSet.h>
-#include <wtf/UnusedParam.h>
-#include <wtf/VMTags.h>
-
-#if OS(DARWIN)
-
-#include <mach/mach_init.h>
-#include <mach/mach_port.h>
-#include <mach/task.h>
-#include <mach/thread_act.h>
-#include <mach/vm_map.h>
-
-#elif OS(WINDOWS)
-
-#include <windows.h>
-#include <malloc.h>
-
-#elif OS(HAIKU)
-
-#include <OS.h>
-
-#elif OS(UNIX)
-
-#include <stdlib.h>
-#if !OS(HAIKU)
-#include <sys/mman.h>
-#endif
-#include <unistd.h>
-
-#if OS(SOLARIS)
-#include <thread.h>
-#else
-#include <pthread.h>
-#endif
-
-#if HAVE(PTHREAD_NP_H)
-#include <pthread_np.h>
-#endif
-
-#if OS(QNX)
-#include <fcntl.h>
-#include <sys/procfs.h>
-#include <stdio.h>
-#include <errno.h>
-#endif
-
-#endif
-
-#define COLLECT_ON_EVERY_ALLOCATION 0
-
-using std::max;
-
-namespace JSC {
-
-// tunable parameters
-
-const size_t GROWTH_FACTOR = 2;
-const size_t LOW_WATER_FACTOR = 4;
-const size_t ALLOCATIONS_PER_COLLECTION = 3600;
-// This value has to be a macro to be used in max() without introducing
-// a PIC branch in Mach-O binaries, see <rdar://problem/5971391>.
-#define MIN_ARRAY_SIZE (static_cast<size_t>(14))
-
-#if ENABLE(JSC_MULTIPLE_THREADS)
-
-#if OS(DARWIN)
-typedef mach_port_t PlatformThread;
-#elif OS(WINDOWS)
-typedef HANDLE PlatformThread;
-#endif
-
-class Heap::Thread {
-public:
-    Thread(pthread_t pthread, const PlatformThread& platThread, void* base) 
-        : posixThread(pthread)
-        , platformThread(platThread)
-        , stackBase(base)
-    {
-    }
-
-    Thread* next;
-    pthread_t posixThread;
-    PlatformThread platformThread;
-    void* stackBase;
-};
-
-#endif
-
-Heap::Heap(JSGlobalData* globalData)
-    : m_markListSet(0)
-#if ENABLE(JSC_MULTIPLE_THREADS)
-    , m_registeredThreads(0)
-    , m_currentThreadRegistrar(0)
-#endif
-#if OS(SYMBIAN)
-    , m_blockallocator(JSCCOLLECTOR_VIRTUALMEM_RESERVATION, BLOCK_SIZE)
-#endif
-    , m_globalData(globalData)
-{
-    ASSERT(globalData);
-    memset(&m_heap, 0, sizeof(CollectorHeap));
-    allocateBlock();
-}
-
-Heap::~Heap()
-{
-    // The destroy function must already have been called, so assert this.
-    ASSERT(!m_globalData);
-}
-
-void Heap::destroy()
-{
-    JSLock lock(SilenceAssertionsOnly);
-
-    if (!m_globalData)
-        return;
-
-    ASSERT(!m_globalData->dynamicGlobalObject);
-    ASSERT(!isBusy());
-    
-    // The global object is not GC protected at this point, so sweeping may delete it
-    // (and thus the global data) before other objects that may use the global data.
-    RefPtr<JSGlobalData> protect(m_globalData);
-
-    delete m_markListSet;
-    m_markListSet = 0;
-
-    freeBlocks();
-
-#if ENABLE(JSC_MULTIPLE_THREADS)
-    if (m_currentThreadRegistrar) {
-        int error = pthread_key_delete(m_currentThreadRegistrar);
-        ASSERT_UNUSED(error, !error);
-    }
-
-    MutexLocker registeredThreadsLock(m_registeredThreadsMutex);
-    for (Heap::Thread* t = m_registeredThreads; t;) {
-        Heap::Thread* next = t->next;
-        delete t;
-        t = next;
-    }
-#endif
-#if OS(SYMBIAN)
-    m_blockallocator.destroy();
-#endif
-    m_globalData = 0;
-}
-
-NEVER_INLINE CollectorBlock* Heap::allocateBlock()
-{
-#if OS(DARWIN)
-    vm_address_t address = 0;
-    vm_map(current_task(), &address, BLOCK_SIZE, BLOCK_OFFSET_MASK, VM_FLAGS_ANYWHERE | VM_TAG_FOR_COLLECTOR_MEMORY, MEMORY_OBJECT_NULL, 0, FALSE, VM_PROT_DEFAULT, VM_PROT_DEFAULT, VM_INHERIT_DEFAULT);
-#elif OS(SYMBIAN)
-    void* address = m_blockallocator.alloc();  
-    if (!address)
-        CRASH();
-#elif OS(WINCE)
-    void* address = VirtualAlloc(NULL, BLOCK_SIZE, MEM_COMMIT | MEM_RESERVE, PAGE_READWRITE);
-#elif OS(WINDOWS)
-#if COMPILER(MINGW) && !COMPILER(MINGW64)
-    void* address = __mingw_aligned_malloc(BLOCK_SIZE, BLOCK_SIZE);
-#else
-    void* address = _aligned_malloc(BLOCK_SIZE, BLOCK_SIZE);
-#endif
-    memset(address, 0, BLOCK_SIZE);
-#elif HAVE(POSIX_MEMALIGN)
-    void* address;
-    posix_memalign(&address, BLOCK_SIZE, BLOCK_SIZE);
-#else
-
-#if ENABLE(JSC_MULTIPLE_THREADS)
-#error Need to initialize pagesize safely.
-#endif
-    static size_t pagesize = getpagesize();
-
-    size_t extra = 0;
-    if (BLOCK_SIZE > pagesize)
-        extra = BLOCK_SIZE - pagesize;
-
-    void* mmapResult = mmap(NULL, BLOCK_SIZE + extra, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANON, -1, 0);
-    uintptr_t address = reinterpret_cast<uintptr_t>(mmapResult);
-
-    size_t adjust = 0;
-    if ((address & BLOCK_OFFSET_MASK) != 0)
-        adjust = BLOCK_SIZE - (address & BLOCK_OFFSET_MASK);
-
-    if (adjust > 0)
-        munmap(reinterpret_cast<char*>(address), adjust);
-
-    if (adjust < extra)
-        munmap(reinterpret_cast<char*>(address + adjust + BLOCK_SIZE), extra - adjust);
-
-    address += adjust;
-#endif
-
-    // Initialize block.
-
-    CollectorBlock* block = reinterpret_cast<CollectorBlock*>(address);
-    block->heap = this;
-    clearMarkBits(block);
-
-    Structure* dummyMarkableCellStructure = m_globalData->dummyMarkableCellStructure.get();
-    for (size_t i = 0; i < HeapConstants::cellsPerBlock; ++i)
-        new (block->cells + i) JSCell(dummyMarkableCellStructure);
-    
-    // Add block to blocks vector.
-
-    size_t numBlocks = m_heap.numBlocks;
-    if (m_heap.usedBlocks == numBlocks) {
-        static const size_t maxNumBlocks = ULONG_MAX / sizeof(CollectorBlock*) / GROWTH_FACTOR;
-        if (numBlocks > maxNumBlocks)
-            CRASH();
-        numBlocks = max(MIN_ARRAY_SIZE, numBlocks * GROWTH_FACTOR);
-        m_heap.numBlocks = numBlocks;
-        m_heap.blocks = static_cast<CollectorBlock**>(fastRealloc(m_heap.blocks, numBlocks * sizeof(CollectorBlock*)));
-    }
-    m_heap.blocks[m_heap.usedBlocks++] = block;
-
-    return block;
-}
-
-NEVER_INLINE void Heap::freeBlock(size_t block)
-{
-    m_heap.didShrink = true;
-
-    ObjectIterator it(m_heap, block);
-    ObjectIterator end(m_heap, block + 1);
-    for ( ; it != end; ++it)
-        (*it)->~JSCell();
-    freeBlockPtr(m_heap.blocks[block]);
-
-    // swap with the last block so we compact as we go
-    m_heap.blocks[block] = m_heap.blocks[m_heap.usedBlocks - 1];
-    m_heap.usedBlocks--;
-
-    if (m_heap.numBlocks > MIN_ARRAY_SIZE && m_heap.usedBlocks < m_heap.numBlocks / LOW_WATER_FACTOR) {
-        m_heap.numBlocks = m_heap.numBlocks / GROWTH_FACTOR; 
-        m_heap.blocks = static_cast<CollectorBlock**>(fastRealloc(m_heap.blocks, m_heap.numBlocks * sizeof(CollectorBlock*)));
-    }
-}
-
-NEVER_INLINE void Heap::freeBlockPtr(CollectorBlock* block)
-{
-#if OS(DARWIN)    
-    vm_deallocate(current_task(), reinterpret_cast<vm_address_t>(block), BLOCK_SIZE);
-#elif OS(SYMBIAN)
-    m_blockallocator.free(reinterpret_cast<void*>(block));
-#elif OS(WINCE)
-    VirtualFree(block, 0, MEM_RELEASE);
-#elif OS(WINDOWS)
-#if COMPILER(MINGW) && !COMPILER(MINGW64)
-    __mingw_aligned_free(block);
-#else
-    _aligned_free(block);
-#endif
-#elif HAVE(POSIX_MEMALIGN)
-    free(block);
-#else
-    munmap(reinterpret_cast<char*>(block), BLOCK_SIZE);
-#endif
-}
-
-void Heap::freeBlocks()
-{
-    ProtectCountSet protectedValuesCopy = m_protectedValues;
-
-    clearMarkBits();
-    ProtectCountSet::iterator protectedValuesEnd = protectedValuesCopy.end();
-    for (ProtectCountSet::iterator it = protectedValuesCopy.begin(); it != protectedValuesEnd; ++it)
-        markCell(it->first);
-
-    m_heap.nextCell = 0;
-    m_heap.nextBlock = 0;
-    DeadObjectIterator it(m_heap, m_heap.nextBlock, m_heap.nextCell);
-    DeadObjectIterator end(m_heap, m_heap.usedBlocks);
-    for ( ; it != end; ++it)
-        (*it)->~JSCell();
-
-    ASSERT(!protectedObjectCount());
-
-    protectedValuesEnd = protectedValuesCopy.end();
-    for (ProtectCountSet::iterator it = protectedValuesCopy.begin(); it != protectedValuesEnd; ++it)
-        it->first->~JSCell();
-
-    for (size_t block = 0; block < m_heap.usedBlocks; ++block)
-        freeBlockPtr(m_heap.blocks[block]);
-
-    fastFree(m_heap.blocks);
-
-    memset(&m_heap, 0, sizeof(CollectorHeap));
-}
-
-void Heap::recordExtraCost(size_t cost)
-{
-    // Our frequency of garbage collection tries to balance memory use against speed
-    // by collecting based on the number of newly created values. However, for values
-    // that hold on to a great deal of memory that's not in the form of other JS values,
-    // that is not good enough - in some cases a lot of those objects can pile up and
-    // use crazy amounts of memory without a GC happening. So we track these extra
-    // memory costs. Only unusually large objects are noted, and we only keep track
-    // of this extra cost until the next GC. In garbage collected languages, most values
-    // are either very short lived temporaries, or have extremely long lifetimes. So
-    // if a large value survives one garbage collection, there is not much point to
-    // collecting more frequently as long as it stays alive.
-
-    if (m_heap.extraCost > maxExtraCost && m_heap.extraCost > m_heap.usedBlocks * BLOCK_SIZE / 2) {
-        // If the last iteration through the heap deallocated blocks, we need
-        // to clean up remaining garbage before marking. Otherwise, the conservative
-        // marking mechanism might follow a pointer to unmapped memory.
-        if (m_heap.didShrink)
-            sweep();
-        reset();
-    }
-    m_heap.extraCost += cost;
-}
-
-void* Heap::allocate(size_t s)
-{
-    typedef HeapConstants::Block Block;
-    typedef HeapConstants::Cell Cell;
-    
-    ASSERT(JSLock::lockCount() > 0);
-    ASSERT(JSLock::currentThreadIsHoldingLock());
-    ASSERT_UNUSED(s, s <= HeapConstants::cellSize);
-
-    ASSERT(m_heap.operationInProgress == NoOperation);
-
-#if COLLECT_ON_EVERY_ALLOCATION
-    collectAllGarbage();
-    ASSERT(m_heap.operationInProgress == NoOperation);
-#endif
-
-allocate:
-
-    // Fast case: find the next garbage cell and recycle it.
-
-    do {
-        ASSERT(m_heap.nextBlock < m_heap.usedBlocks);
-        Block* block = reinterpret_cast<Block*>(m_heap.blocks[m_heap.nextBlock]);
-        do {
-            ASSERT(m_heap.nextCell < HeapConstants::cellsPerBlock);
-            if (!block->marked.get(m_heap.nextCell)) { // Always false for the last cell in the block
-                Cell* cell = block->cells + m_heap.nextCell;
-
-                m_heap.operationInProgress = Allocation;
-                JSCell* imp = reinterpret_cast<JSCell*>(cell);
-                imp->~JSCell();
-                m_heap.operationInProgress = NoOperation;
-
-                ++m_heap.nextCell;
-                return cell;
-            }
-        } while (++m_heap.nextCell != HeapConstants::cellsPerBlock);
-        m_heap.nextCell = 0;
-    } while (++m_heap.nextBlock != m_heap.usedBlocks);
-
-    // Slow case: reached the end of the heap. Mark live objects and start over.
-
-    reset();
-    goto allocate;
-}
-
-void Heap::resizeBlocks()
-{
-    m_heap.didShrink = false;
-
-    size_t usedCellCount = markedCells();
-    size_t minCellCount = usedCellCount + max(ALLOCATIONS_PER_COLLECTION, usedCellCount);
-    size_t minBlockCount = (minCellCount + HeapConstants::cellsPerBlock - 1) / HeapConstants::cellsPerBlock;
-
-    size_t maxCellCount = 1.25f * minCellCount;
-    size_t maxBlockCount = (maxCellCount + HeapConstants::cellsPerBlock - 1) / HeapConstants::cellsPerBlock;
-
-    if (m_heap.usedBlocks < minBlockCount)
-        growBlocks(minBlockCount);
-    else if (m_heap.usedBlocks > maxBlockCount)
-        shrinkBlocks(maxBlockCount);
-}
-
-void Heap::growBlocks(size_t neededBlocks)
-{
-    ASSERT(m_heap.usedBlocks < neededBlocks);
-    while (m_heap.usedBlocks < neededBlocks)
-        allocateBlock();
-}
-
-void Heap::shrinkBlocks(size_t neededBlocks)
-{
-    ASSERT(m_heap.usedBlocks > neededBlocks);
-    
-    // Clear the always-on last bit, so isEmpty() isn't fooled by it.
-    for (size_t i = 0; i < m_heap.usedBlocks; ++i)
-        m_heap.blocks[i]->marked.clear(HeapConstants::cellsPerBlock - 1);
-
-    for (size_t i = 0; i != m_heap.usedBlocks && m_heap.usedBlocks != neededBlocks; ) {
-        if (m_heap.blocks[i]->marked.isEmpty()) {
-            freeBlock(i);
-        } else
-            ++i;
-    }
-
-    // Reset the always-on last bit.
-    for (size_t i = 0; i < m_heap.usedBlocks; ++i)
-        m_heap.blocks[i]->marked.set(HeapConstants::cellsPerBlock - 1);
-}
-
-#if OS(WINCE)
-JS_EXPORTDATA void* g_stackBase = 0;
-
-inline bool isPageWritable(void* page)
-{
-    MEMORY_BASIC_INFORMATION memoryInformation;
-    DWORD result = VirtualQuery(page, &memoryInformation, sizeof(memoryInformation));
-
-    // return false on error, including ptr outside memory
-    if (result != sizeof(memoryInformation))
-        return false;
-
-    DWORD protect = memoryInformation.Protect & ~(PAGE_GUARD | PAGE_NOCACHE);
-    return protect == PAGE_READWRITE
-        || protect == PAGE_WRITECOPY
-        || protect == PAGE_EXECUTE_READWRITE
-        || protect == PAGE_EXECUTE_WRITECOPY;
-}
-
-static void* getStackBase(void* previousFrame)
-{
-    // find the address of this stack frame by taking the address of a local variable
-    bool isGrowingDownward;
-    void* thisFrame = (void*)(&isGrowingDownward);
-
-    isGrowingDownward = previousFrame < &thisFrame;
-    static DWORD pageSize = 0;
-    if (!pageSize) {
-        SYSTEM_INFO systemInfo;
-        GetSystemInfo(&systemInfo);
-        pageSize = systemInfo.dwPageSize;
-    }
-
-    // scan all of memory starting from this frame, and return the last writeable page found
-    register char* currentPage = (char*)((DWORD)thisFrame & ~(pageSize - 1));
-    if (isGrowingDownward) {
-        while (currentPage > 0) {
-            // check for underflow
-            if (currentPage >= (char*)pageSize)
-                currentPage -= pageSize;
-            else
-                currentPage = 0;
-            if (!isPageWritable(currentPage))
-                return currentPage + pageSize;
-        }
-        return 0;
-    } else {
-        while (true) {
-            // guaranteed to complete because isPageWritable returns false at end of memory
-            currentPage += pageSize;
-            if (!isPageWritable(currentPage))
-                return currentPage;
-        }
-    }
-}
-#endif
-
-#if OS(QNX)
-static inline void *currentThreadStackBaseQNX()
-{
-    static void* stackBase = 0;
-    static size_t stackSize = 0;
-    static pthread_t stackThread;
-    pthread_t thread = pthread_self();
-    if (stackBase == 0 || thread != stackThread) {
-        struct _debug_thread_info threadInfo;
-        memset(&threadInfo, 0, sizeof(threadInfo));
-        threadInfo.tid = pthread_self();
-        int fd = open("/proc/self", O_RDONLY);
-        if (fd == -1) {
-            LOG_ERROR("Unable to open /proc/self (errno: %d)", errno);
-            return 0;
-        }
-        devctl(fd, DCMD_PROC_TIDSTATUS, &threadInfo, sizeof(threadInfo), 0);
-        close(fd);
-        stackBase = reinterpret_cast<void*>(threadInfo.stkbase);
-        stackSize = threadInfo.stksize;
-        ASSERT(stackBase);
-        stackThread = thread;
-    }
-    return static_cast<char*>(stackBase) + stackSize;
-}
-#endif
-
-static inline void* currentThreadStackBase()
-{
-#if OS(DARWIN)
-    pthread_t thread = pthread_self();
-    return pthread_get_stackaddr_np(thread);
-#elif OS(WINDOWS) && CPU(X86) && COMPILER(MSVC)
-    // offset 0x18 from the FS segment register gives a pointer to
-    // the thread information block for the current thread
-    NT_TIB* pTib;
-    __asm {
-        MOV EAX, FS:[18h]
-        MOV pTib, EAX
-    }
-    return static_cast<void*>(pTib->StackBase);
-#elif OS(WINDOWS) && CPU(X86) && COMPILER(GCC)
-    // offset 0x18 from the FS segment register gives a pointer to
-    // the thread information block for the current thread
-    NT_TIB* pTib;
-    asm ( "movl %%fs:0x18, %0\n"
-          : "=r" (pTib)
-        );
-    return static_cast<void*>(pTib->StackBase);
-#elif OS(WINDOWS) && CPU(X86_64)
-    PNT_TIB64 pTib = reinterpret_cast<PNT_TIB64>(NtCurrentTeb());
-    return reinterpret_cast<void*>(pTib->StackBase);
-#elif OS(QNX)
-    AtomicallyInitializedStatic(Mutex&, mutex = *new Mutex);
-    MutexLocker locker(mutex);
-    return currentThreadStackBaseQNX();
-#elif OS(SOLARIS)
-    stack_t s;
-    thr_stksegment(&s);
-    return s.ss_sp;
-#elif OS(OPENBSD)
-    pthread_t thread = pthread_self();
-    stack_t stack;
-    pthread_stackseg_np(thread, &stack);
-    return stack.ss_sp;
-#elif OS(SYMBIAN)
-    TThreadStackInfo info;
-    RThread thread;
-    thread.StackInfo(info);
-    return (void*)info.iBase;
-#elif OS(HAIKU)
-    thread_info threadInfo;
-    get_thread_info(find_thread(NULL), &threadInfo);
-    return threadInfo.stack_end;
-#elif OS(UNIX)
-    AtomicallyInitializedStatic(Mutex&, mutex = *new Mutex);
-    MutexLocker locker(mutex);
-    static void* stackBase = 0;
-    static size_t stackSize = 0;
-    static pthread_t stackThread;
-    pthread_t thread = pthread_self();
-    if (stackBase == 0 || thread != stackThread) {
-        pthread_attr_t sattr;
-        pthread_attr_init(&sattr);
-#if HAVE(PTHREAD_NP_H) || OS(NETBSD)
-        // e.g. on FreeBSD 5.4, neundorf@kde.org
-        pthread_attr_get_np(thread, &sattr);
-#else
-        // FIXME: this function is non-portable; other POSIX systems may have different np alternatives
-        pthread_getattr_np(thread, &sattr);
-#endif
-        int rc = pthread_attr_getstack(&sattr, &stackBase, &stackSize);
-        (void)rc; // FIXME: Deal with error code somehow? Seems fatal.
-        ASSERT(stackBase);
-        pthread_attr_destroy(&sattr);
-        stackThread = thread;
-    }
-    return static_cast<char*>(stackBase) + stackSize;
-#elif OS(WINCE)
-    AtomicallyInitializedStatic(Mutex&, mutex = *new Mutex);
-    MutexLocker locker(mutex);
-    if (g_stackBase)
-        return g_stackBase;
-    else {
-        int dummy;
-        return getStackBase(&dummy);
-    }
-#else
-#error Need a way to get the stack base on this platform
-#endif
-}
-
-#if ENABLE(JSC_MULTIPLE_THREADS)
-
-static inline PlatformThread getCurrentPlatformThread()
-{
-#if OS(DARWIN)
-    return pthread_mach_thread_np(pthread_self());
-#elif OS(WINDOWS)
-    return pthread_getw32threadhandle_np(pthread_self());
-#endif
-}
-
-void Heap::makeUsableFromMultipleThreads()
-{
-    if (m_currentThreadRegistrar)
-        return;
-
-    int error = pthread_key_create(&m_currentThreadRegistrar, unregisterThread);
-    if (error)
-        CRASH();
-}
-
-void Heap::registerThread()
-{
-    ASSERT(!m_globalData->exclusiveThread || m_globalData->exclusiveThread == currentThread());
-
-    if (!m_currentThreadRegistrar || pthread_getspecific(m_currentThreadRegistrar))
-        return;
-
-    pthread_setspecific(m_currentThreadRegistrar, this);
-    Heap::Thread* thread = new Heap::Thread(pthread_self(), getCurrentPlatformThread(), currentThreadStackBase());
-
-    MutexLocker lock(m_registeredThreadsMutex);
-
-    thread->next = m_registeredThreads;
-    m_registeredThreads = thread;
-}
-
-void Heap::unregisterThread(void* p)
-{
-    if (p)
-        static_cast<Heap*>(p)->unregisterThread();
-}
-
-void Heap::unregisterThread()
-{
-    pthread_t currentPosixThread = pthread_self();
-
-    MutexLocker lock(m_registeredThreadsMutex);
-
-    if (pthread_equal(currentPosixThread, m_registeredThreads->posixThread)) {
-        Thread* t = m_registeredThreads;
-        m_registeredThreads = m_registeredThreads->next;
-        delete t;
-    } else {
-        Heap::Thread* last = m_registeredThreads;
-        Heap::Thread* t;
-        for (t = m_registeredThreads->next; t; t = t->next) {
-            if (pthread_equal(t->posixThread, currentPosixThread)) {
-                last->next = t->next;
-                break;
-            }
-            last = t;
-        }
-        ASSERT(t); // If t is NULL, we never found ourselves in the list.
-        delete t;
-    }
-}
-
-#else // ENABLE(JSC_MULTIPLE_THREADS)
-
-void Heap::registerThread()
-{
-}
-
-#endif
-
-inline bool isPointerAligned(void* p)
-{
-    return (((intptr_t)(p) & (sizeof(char*) - 1)) == 0);
-}
-
-// Cell size needs to be a power of two for isPossibleCell to be valid.
-COMPILE_ASSERT(sizeof(CollectorCell) % 2 == 0, Collector_cell_size_is_power_of_two);
-
-#if USE(JSVALUE32)
-static bool isHalfCellAligned(void *p)
-{
-    return (((intptr_t)(p) & (CELL_MASK >> 1)) == 0);
-}
-
-static inline bool isPossibleCell(void* p)
-{
-    return isHalfCellAligned(p) && p;
-}
-
-#else
-
-static inline bool isCellAligned(void *p)
-{
-    return (((intptr_t)(p) & CELL_MASK) == 0);
-}
-
-static inline bool isPossibleCell(void* p)
-{
-    return isCellAligned(p) && p;
-}
-#endif // USE(JSVALUE32)
-
-void Heap::markConservatively(MarkStack& markStack, void* start, void* end)
-{
-    if (start > end) {
-        void* tmp = start;
-        start = end;
-        end = tmp;
-    }
-
-    ASSERT((static_cast<char*>(end) - static_cast<char*>(start)) < 0x1000000);
-    ASSERT(isPointerAligned(start));
-    ASSERT(isPointerAligned(end));
-
-    char** p = static_cast<char**>(start);
-    char** e = static_cast<char**>(end);
-
-    CollectorBlock** blocks = m_heap.blocks;
-    while (p != e) {
-        char* x = *p++;
-        if (isPossibleCell(x)) {
-            size_t usedBlocks;
-            uintptr_t xAsBits = reinterpret_cast<uintptr_t>(x);
-            xAsBits &= CELL_ALIGN_MASK;
-
-            uintptr_t offset = xAsBits & BLOCK_OFFSET_MASK;
-            const size_t lastCellOffset = sizeof(CollectorCell) * (CELLS_PER_BLOCK - 1);
-            if (offset > lastCellOffset)
-                continue;
-
-            CollectorBlock* blockAddr = reinterpret_cast<CollectorBlock*>(xAsBits - offset);
-            usedBlocks = m_heap.usedBlocks;
-            for (size_t block = 0; block < usedBlocks; block++) {
-                if (blocks[block] != blockAddr)
-                    continue;
-                markStack.append(reinterpret_cast<JSCell*>(xAsBits));
-                markStack.drain();
-            }
-        }
-    }
-}
-
-void NEVER_INLINE Heap::markCurrentThreadConservativelyInternal(MarkStack& markStack)
-{
-    void* dummy;
-    void* stackPointer = &dummy;
-    void* stackBase = currentThreadStackBase();
-    markConservatively(markStack, stackPointer, stackBase);
-}
-
-#if COMPILER(GCC)
-#define REGISTER_BUFFER_ALIGNMENT __attribute__ ((aligned (sizeof(void*))))
-#else
-#define REGISTER_BUFFER_ALIGNMENT
-#endif
-
-void Heap::markCurrentThreadConservatively(MarkStack& markStack)
-{
-    // setjmp forces volatile registers onto the stack
-    jmp_buf registers REGISTER_BUFFER_ALIGNMENT;
-#if COMPILER(MSVC)
-#pragma warning(push)
-#pragma warning(disable: 4611)
-#endif
-    setjmp(registers);
-#if COMPILER(MSVC)
-#pragma warning(pop)
-#endif
-
-    markCurrentThreadConservativelyInternal(markStack);
-}
-
-#if ENABLE(JSC_MULTIPLE_THREADS)
-
-static inline void suspendThread(const PlatformThread& platformThread)
-{
-#if OS(DARWIN)
-    thread_suspend(platformThread);
-#elif OS(WINDOWS)
-    SuspendThread(platformThread);
-#else
-#error Need a way to suspend threads on this platform
-#endif
-}
-
-static inline void resumeThread(const PlatformThread& platformThread)
-{
-#if OS(DARWIN)
-    thread_resume(platformThread);
-#elif OS(WINDOWS)
-    ResumeThread(platformThread);
-#else
-#error Need a way to resume threads on this platform
-#endif
-}
-
-typedef unsigned long usword_t; // word size, assumed to be either 32 or 64 bit
-
-#if OS(DARWIN)
-
-#if CPU(X86)
-typedef i386_thread_state_t PlatformThreadRegisters;
-#elif CPU(X86_64)
-typedef x86_thread_state64_t PlatformThreadRegisters;
-#elif CPU(PPC)
-typedef ppc_thread_state_t PlatformThreadRegisters;
-#elif CPU(PPC64)
-typedef ppc_thread_state64_t PlatformThreadRegisters;
-#elif CPU(ARM)
-typedef arm_thread_state_t PlatformThreadRegisters;
-#else
-#error Unknown Architecture
-#endif
-
-#elif OS(WINDOWS) && CPU(X86)
-typedef CONTEXT PlatformThreadRegisters;
-#else
-#error Need a thread register struct for this platform
-#endif
-
-static size_t getPlatformThreadRegisters(const PlatformThread& platformThread, PlatformThreadRegisters& regs)
-{
-#if OS(DARWIN)
-
-#if CPU(X86)
-    unsigned user_count = sizeof(regs)/sizeof(int);
-    thread_state_flavor_t flavor = i386_THREAD_STATE;
-#elif CPU(X86_64)
-    unsigned user_count = x86_THREAD_STATE64_COUNT;
-    thread_state_flavor_t flavor = x86_THREAD_STATE64;
-#elif CPU(PPC) 
-    unsigned user_count = PPC_THREAD_STATE_COUNT;
-    thread_state_flavor_t flavor = PPC_THREAD_STATE;
-#elif CPU(PPC64)
-    unsigned user_count = PPC_THREAD_STATE64_COUNT;
-    thread_state_flavor_t flavor = PPC_THREAD_STATE64;
-#elif CPU(ARM)
-    unsigned user_count = ARM_THREAD_STATE_COUNT;
-    thread_state_flavor_t flavor = ARM_THREAD_STATE;
-#else
-#error Unknown Architecture
-#endif
-
-    kern_return_t result = thread_get_state(platformThread, flavor, (thread_state_t)&regs, &user_count);
-    if (result != KERN_SUCCESS) {
-        WTFReportFatalError(__FILE__, __LINE__, WTF_PRETTY_FUNCTION, 
-                            "JavaScript garbage collection failed because thread_get_state returned an error (%d). This is probably the result of running inside Rosetta, which is not supported.", result);
-        CRASH();
-    }
-    return user_count * sizeof(usword_t);
-// end OS(DARWIN)
-
-#elif OS(WINDOWS) && CPU(X86)
-    regs.ContextFlags = CONTEXT_INTEGER | CONTEXT_CONTROL | CONTEXT_SEGMENTS;
-    GetThreadContext(platformThread, &regs);
-    return sizeof(CONTEXT);
-#else
-#error Need a way to get thread registers on this platform
-#endif
-}
-
-static inline void* otherThreadStackPointer(const PlatformThreadRegisters& regs)
-{
-#if OS(DARWIN)
-
-#if __DARWIN_UNIX03
-
-#if CPU(X86)
-    return reinterpret_cast<void*>(regs.__esp);
-#elif CPU(X86_64)
-    return reinterpret_cast<void*>(regs.__rsp);
-#elif CPU(PPC) || CPU(PPC64)
-    return reinterpret_cast<void*>(regs.__r1);
-#elif CPU(ARM)
-    return reinterpret_cast<void*>(regs.__sp);
-#else
-#error Unknown Architecture
-#endif
-
-#else // !__DARWIN_UNIX03
-
-#if CPU(X86)
-    return reinterpret_cast<void*>(regs.esp);
-#elif CPU(X86_64)
-    return reinterpret_cast<void*>(regs.rsp);
-#elif CPU(PPC) || CPU(PPC64)
-    return reinterpret_cast<void*>(regs.r1);
-#else
-#error Unknown Architecture
-#endif
-
-#endif // __DARWIN_UNIX03
-
-// end OS(DARWIN)
-#elif CPU(X86) && OS(WINDOWS)
-    return reinterpret_cast<void*>((uintptr_t) regs.Esp);
-#else
-#error Need a way to get the stack pointer for another thread on this platform
-#endif
-}
-
-void Heap::markOtherThreadConservatively(MarkStack& markStack, Thread* thread)
-{
-    suspendThread(thread->platformThread);
-
-    PlatformThreadRegisters regs;
-    size_t regSize = getPlatformThreadRegisters(thread->platformThread, regs);
-
-    // mark the thread's registers
-    markConservatively(markStack, static_cast<void*>(&regs), static_cast<void*>(reinterpret_cast<char*>(&regs) + regSize));
-
-    void* stackPointer = otherThreadStackPointer(regs);
-    markConservatively(markStack, stackPointer, thread->stackBase);
-
-    resumeThread(thread->platformThread);
-}
-
-#endif
-
-void Heap::markStackObjectsConservatively(MarkStack& markStack)
-{
-    markCurrentThreadConservatively(markStack);
-
-#if ENABLE(JSC_MULTIPLE_THREADS)
-
-    if (m_currentThreadRegistrar) {
-
-        MutexLocker lock(m_registeredThreadsMutex);
-
-#ifndef NDEBUG
-        // Forbid malloc during the mark phase. Marking a thread suspends it, so 
-        // a malloc inside markChildren() would risk a deadlock with a thread that had been 
-        // suspended while holding the malloc lock.
-        fastMallocForbid();
-#endif
-        // It is safe to access the registeredThreads list, because we earlier asserted that locks are being held,
-        // and since this is a shared heap, they are real locks.
-        for (Thread* thread = m_registeredThreads; thread; thread = thread->next) {
-            if (!pthread_equal(thread->posixThread, pthread_self()))
-                markOtherThreadConservatively(markStack, thread);
-        }
-#ifndef NDEBUG
-        fastMallocAllow();
-#endif
-    }
-#endif
-}
-
-void Heap::protect(JSValue k)
-{
-    ASSERT(k);
-    ASSERT(JSLock::currentThreadIsHoldingLock() || !m_globalData->isSharedInstance());
-
-    if (!k.isCell())
-        return;
-
-    m_protectedValues.add(k.asCell());
-}
-
-bool Heap::unprotect(JSValue k)
-{
-    ASSERT(k);
-    ASSERT(JSLock::currentThreadIsHoldingLock() || !m_globalData->isSharedInstance());
-
-    if (!k.isCell())
-        return false;
-
-    return m_protectedValues.remove(k.asCell());
-}
-
-void Heap::markProtectedObjects(MarkStack& markStack)
-{
-    ProtectCountSet::iterator end = m_protectedValues.end();
-    for (ProtectCountSet::iterator it = m_protectedValues.begin(); it != end; ++it) {
-        markStack.append(it->first);
-        markStack.drain();
-    }
-}
-
-void Heap::pushTempSortVector(Vector<ValueStringPair>* tempVector)
-{
-    m_tempSortingVectors.append(tempVector);
-}
-
-void Heap::popTempSortVector(Vector<ValueStringPair>* tempVector)
-{
-    ASSERT_UNUSED(tempVector, tempVector == m_tempSortingVectors.last());
-    m_tempSortingVectors.removeLast();
-}
-    
-void Heap::markTempSortVectors(MarkStack& markStack)
-{
-    typedef Vector<Vector<ValueStringPair>* > VectorOfValueStringVectors;
-
-    VectorOfValueStringVectors::iterator end = m_tempSortingVectors.end();
-    for (VectorOfValueStringVectors::iterator it = m_tempSortingVectors.begin(); it != end; ++it) {
-        Vector<ValueStringPair>* tempSortingVector = *it;
-
-        Vector<ValueStringPair>::iterator vectorEnd = tempSortingVector->end();
-        for (Vector<ValueStringPair>::iterator vectorIt = tempSortingVector->begin(); vectorIt != vectorEnd; ++vectorIt)
-            if (vectorIt->first)
-                markStack.append(vectorIt->first);
-        markStack.drain();
-    }
-}
-    
-void Heap::clearMarkBits()
-{
-    for (size_t i = 0; i < m_heap.usedBlocks; ++i)
-        clearMarkBits(m_heap.blocks[i]);
-}
-
-void Heap::clearMarkBits(CollectorBlock* block)
-{
-    // allocate assumes that the last cell in every block is marked.
-    block->marked.clearAll();
-    block->marked.set(HeapConstants::cellsPerBlock - 1);
-}
-
-size_t Heap::markedCells(size_t startBlock, size_t startCell) const
-{
-    ASSERT(startBlock <= m_heap.usedBlocks);
-    ASSERT(startCell < HeapConstants::cellsPerBlock);
-
-    if (startBlock >= m_heap.usedBlocks)
-        return 0;
-
-    size_t result = 0;
-    result += m_heap.blocks[startBlock]->marked.count(startCell);
-    for (size_t i = startBlock + 1; i < m_heap.usedBlocks; ++i)
-        result += m_heap.blocks[i]->marked.count();
-
-    return result;
-}
-
-void Heap::sweep()
-{
-    ASSERT(m_heap.operationInProgress == NoOperation);
-    if (m_heap.operationInProgress != NoOperation)
-        CRASH();
-    m_heap.operationInProgress = Collection;
-    
-#if !ENABLE(JSC_ZOMBIES)
-    Structure* dummyMarkableCellStructure = m_globalData->dummyMarkableCellStructure.get();
-#endif
-
-    DeadObjectIterator it(m_heap, m_heap.nextBlock, m_heap.nextCell);
-    DeadObjectIterator end(m_heap, m_heap.usedBlocks);
-    for ( ; it != end; ++it) {
-        JSCell* cell = *it;
-#if ENABLE(JSC_ZOMBIES)
-        if (!cell->isZombie()) {
-            const ClassInfo* info = cell->classInfo();
-            cell->~JSCell();
-            new (cell) JSZombie(info, JSZombie::leakedZombieStructure());
-            Heap::markCell(cell);
-        }
-#else
-        cell->~JSCell();
-        // Callers of sweep assume it's safe to mark any cell in the heap.
-        new (cell) JSCell(dummyMarkableCellStructure);
-#endif
-    }
-
-    m_heap.operationInProgress = NoOperation;
-}
-
-void Heap::markRoots()
-{
-#ifndef NDEBUG
-    if (m_globalData->isSharedInstance()) {
-        ASSERT(JSLock::lockCount() > 0);
-        ASSERT(JSLock::currentThreadIsHoldingLock());
-    }
-#endif
-
-    ASSERT(m_heap.operationInProgress == NoOperation);
-    if (m_heap.operationInProgress != NoOperation)
-        CRASH();
-
-    m_heap.operationInProgress = Collection;
-
-    MarkStack& markStack = m_globalData->markStack;
-
-    // Reset mark bits.
-    clearMarkBits();
-
-    // Mark stack roots.
-    markStackObjectsConservatively(markStack);
-    m_globalData->interpreter->registerFile().markCallFrames(markStack, this);
-
-    // Mark explicitly registered roots.
-    markProtectedObjects(markStack);
-    
-    // Mark temporary vector for Array sorting
-    markTempSortVectors(markStack);
-
-    // Mark misc. other roots.
-    if (m_markListSet && m_markListSet->size())
-        MarkedArgumentBuffer::markLists(markStack, *m_markListSet);
-    if (m_globalData->exception)
-        markStack.append(m_globalData->exception);
-    if (m_globalData->functionCodeBlockBeingReparsed)
-        m_globalData->functionCodeBlockBeingReparsed->markAggregate(markStack);
-    if (m_globalData->firstStringifierToMark)
-        JSONObject::markStringifiers(markStack, m_globalData->firstStringifierToMark);
-
-    // Mark the small strings cache last, since it will clear itself if nothing
-    // else has marked it.
-    m_globalData->smallStrings.markChildren(markStack);
-
-    markStack.drain();
-    markStack.compact();
-
-    m_heap.operationInProgress = NoOperation;
-}
-
-size_t Heap::objectCount() const
-{
-    return m_heap.nextBlock * HeapConstants::cellsPerBlock // allocated full blocks
-           + m_heap.nextCell // allocated cells in current block
-           + markedCells(m_heap.nextBlock, m_heap.nextCell) // marked cells in remainder of m_heap
-           - m_heap.usedBlocks; // 1 cell per block is a dummy sentinel
-}
-
-void Heap::addToStatistics(Heap::Statistics& statistics) const
-{
-    statistics.size += m_heap.usedBlocks * BLOCK_SIZE;
-    statistics.free += m_heap.usedBlocks * BLOCK_SIZE - (objectCount() * HeapConstants::cellSize);
-}
-
-Heap::Statistics Heap::statistics() const
-{
-    Statistics statistics = { 0, 0 };
-    addToStatistics(statistics);
-    return statistics;
-}
-
-size_t Heap::globalObjectCount()
-{
-    size_t count = 0;
-    if (JSGlobalObject* head = m_globalData->head) {
-        JSGlobalObject* o = head;
-        do {
-            ++count;
-            o = o->next();
-        } while (o != head);
-    }
-    return count;
-}
-
-size_t Heap::protectedGlobalObjectCount()
-{
-    size_t count = 0;
-    if (JSGlobalObject* head = m_globalData->head) {
-        JSGlobalObject* o = head;
-        do {
-            if (m_protectedValues.contains(o))
-                ++count;
-            o = o->next();
-        } while (o != head);
-    }
-
-    return count;
-}
-
-size_t Heap::protectedObjectCount()
-{
-    return m_protectedValues.size();
-}
-
-static const char* typeName(JSCell* cell)
-{
-    if (cell->isString())
-        return "string";
-#if USE(JSVALUE32)
-    if (cell->isNumber())
-        return "number";
-#endif
-    if (cell->isGetterSetter())
-        return "Getter-Setter";
-    if (cell->isAPIValueWrapper())
-        return "API wrapper";
-    if (cell->isPropertyNameIterator())
-        return "For-in iterator";
-    if (!cell->isObject())
-        return "[empty cell]";
-    const ClassInfo* info = cell->classInfo();
-    return info ? info->className : "Object";
-}
-
-HashCountedSet<const char*>* Heap::protectedObjectTypeCounts()
-{
-    HashCountedSet<const char*>* counts = new HashCountedSet<const char*>;
-
-    ProtectCountSet::iterator end = m_protectedValues.end();
-    for (ProtectCountSet::iterator it = m_protectedValues.begin(); it != end; ++it)
-        counts->add(typeName(it->first));
-
-    return counts;
-}
-
-HashCountedSet<const char*>* Heap::objectTypeCounts()
-{
-    HashCountedSet<const char*>* counts = new HashCountedSet<const char*>;
-
-    LiveObjectIterator it = primaryHeapBegin();
-    LiveObjectIterator heapEnd = primaryHeapEnd();
-    for ( ; it != heapEnd; ++it)
-        counts->add(typeName(*it));
-
-    return counts;
-}
-
-bool Heap::isBusy()
-{
-    return m_heap.operationInProgress != NoOperation;
-}
-
-void Heap::reset()
-{
-    JAVASCRIPTCORE_GC_BEGIN();
-
-    markRoots();
-
-    JAVASCRIPTCORE_GC_MARKED();
-
-    m_heap.nextCell = 0;
-    m_heap.nextBlock = 0;
-    m_heap.nextNumber = 0;
-    m_heap.extraCost = 0;
-#if ENABLE(JSC_ZOMBIES)
-    sweep();
-#endif
-    resizeBlocks();
-
-    JAVASCRIPTCORE_GC_END();
-}
-
-void Heap::collectAllGarbage()
-{
-    JAVASCRIPTCORE_GC_BEGIN();
-
-    // If the last iteration through the heap deallocated blocks, we need
-    // to clean up remaining garbage before marking. Otherwise, the conservative
-    // marking mechanism might follow a pointer to unmapped memory.
-    if (m_heap.didShrink)
-        sweep();
-
-    markRoots();
-
-    JAVASCRIPTCORE_GC_MARKED();
-
-    m_heap.nextCell = 0;
-    m_heap.nextBlock = 0;
-    m_heap.nextNumber = 0;
-    m_heap.extraCost = 0;
-    sweep();
-    resizeBlocks();
-
-    JAVASCRIPTCORE_GC_END();
-}
-
-LiveObjectIterator Heap::primaryHeapBegin()
-{
-    return LiveObjectIterator(m_heap, 0);
-}
-
-LiveObjectIterator Heap::primaryHeapEnd()
-{
-    return LiveObjectIterator(m_heap, m_heap.usedBlocks);
-}
-
-} // namespace JSC