8046598: Scalable Native memory tracking development

Summary: Enhance scalability of native memory tracking
Reviewed-by: coleenp, ctornqvi, gtriantafill

make/excludeSrc.make

@@ -118,8 +118,8 @@ ifeq ($(INCLUDE_NMT), false)
       CFLAGS += -DINCLUDE_NMT=0
 
       Src_Files_EXCLUDE += \
-	 memBaseline.cpp memPtr.cpp memRecorder.cpp memReporter.cpp memSnapshot.cpp memTrackWorker.cpp \
-	 memTracker.cpp nmtDCmd.cpp
+	 memBaseline.cpp memReporter.cpp mallocTracker.cpp virtualMemoryTracker.cpp nmtCommon.cpp \
+	 memTracker.cpp nmtDCmd.cpp mallocSiteTable.cpp
 endif
 
 -include $(HS_ALT_MAKE)/excludeSrc.make

src/os/bsd/vm/os_bsd.cpp

@@ -2434,23 +2434,25 @@ char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr,
   }
 
   // The memory is committed
-  MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
+  MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
 
   return addr;
 }
 
 bool os::release_memory_special(char* base, size_t bytes) {
-  MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
-  // detaching the SHM segment will also delete it, see reserve_memory_special()
-  int rslt = shmdt(base);
-  if (rslt == 0) {
-    tkr.record((address)base, bytes);
-    return true;
+  if (MemTracker::tracking_level() > NMT_minimal) {
+    Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
+    // detaching the SHM segment will also delete it, see reserve_memory_special()
+    int rslt = shmdt(base);
+    if (rslt == 0) {
+      tkr.record((address)base, bytes);
+      return true;
+    } else {
+      return false;
+    }
   } else {
-    tkr.discard();
-    return false;
+    return shmdt(base) == 0;
   }
-
 }
 
 size_t os::large_page_size() {

src/os/bsd/vm/perfMemory_bsd.cpp

@@ -753,7 +753,7 @@ static char* mmap_create_shared(size_t size) {
   (void)::memset((void*) mapAddress, 0, size);
 
   // it does not go through os api, the operation has to record from here
-  MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC);
+  MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal);
 
   return mapAddress;
 }
@@ -918,7 +918,7 @@ static void mmap_attach_shared(const char* user, int vmid, PerfMemory::PerfMemor
   }
 
   // it does not go through os api, the operation has to record from here
-  MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC);
+  MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal);
 
   *addr = mapAddress;
   *sizep = size;

src/os/linux/vm/os_linux.cpp

@@ -3501,9 +3501,12 @@ char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t al
 
   assert(is_ptr_aligned(start, alignment), "Must be");
 
-  // os::reserve_memory_special will record this memory area.
-  // Need to release it here to prevent overlapping reservations.
-  MemTracker::record_virtual_memory_release((address)start, bytes);
+  if (MemTracker::tracking_level() > NMT_minimal) {
+    // os::reserve_memory_special will record this memory area.
+    // Need to release it here to prevent overlapping reservations.
+    Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
+    tkr.record((address)start, bytes);
+  }
 
   char* end = start + bytes;
 
@@ -3598,7 +3601,7 @@ char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr,
     }
 
     // The memory is committed
-    MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
+    MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
   }
 
   return addr;
@@ -3614,24 +3617,30 @@ bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
 }
 
 bool os::release_memory_special(char* base, size_t bytes) {
-  assert(UseLargePages, "only for large pages");
+  bool res;
+  if (MemTracker::tracking_level() > NMT_minimal) {
+    Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
+    res = os::Linux::release_memory_special_impl(base, bytes);
+    if (res) {
+      tkr.record((address)base, bytes);
+    }
 
-  MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
+  } else {
+    res = os::Linux::release_memory_special_impl(base, bytes);
+  }
+  return res;
+}
 
+bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
+  assert(UseLargePages, "only for large pages");
   bool res;
+
   if (UseSHM) {
     res = os::Linux::release_memory_special_shm(base, bytes);
   } else {
     assert(UseHugeTLBFS, "must be");
     res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
   }
-
-  if (res) {
-    tkr.record((address)base, bytes);
-  } else {
-    tkr.discard();
-  }
-
   return res;
 }
 

src/os/linux/vm/os_linux.hpp

@@ -108,6 +108,7 @@ class Linux {
   static char* reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec);
   static char* reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec);
 
+  static bool release_memory_special_impl(char* base, size_t bytes);
   static bool release_memory_special_shm(char* base, size_t bytes);
   static bool release_memory_special_huge_tlbfs(char* base, size_t bytes);
 

src/os/linux/vm/perfMemory_linux.cpp

@@ -753,7 +753,7 @@ static char* mmap_create_shared(size_t size) {
   (void)::memset((void*) mapAddress, 0, size);
 
   // it does not go through os api, the operation has to record from here
-  MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC);
+  MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal);
 
   return mapAddress;
 }
@@ -924,7 +924,7 @@ static void mmap_attach_shared(const char* user, int vmid, PerfMemory::PerfMemor
   }
 
   // it does not go through os api, the operation has to record from here
-  MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC);
+  MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size, CURRENT_PC, mtInternal);
 
   *addr = mapAddress;
   *sizep = size;

src/os/posix/vm/os_posix.cpp

@@ -74,21 +74,41 @@ void os::check_or_create_dump(void* exceptionRecord, void* contextRecord, char*
   VMError::report_coredump_status(buffer, success);
 }
 
-address os::get_caller_pc(int n) {
+int os::get_native_stack(address* stack, int frames, int toSkip) {
 #ifdef _NMT_NOINLINE_
-  n ++;
+  toSkip++;
 #endif
+
+  int frame_idx = 0;
+  int num_of_frames;  // number of frames captured
   frame fr = os::current_frame();
-  while (n > 0 && fr.pc() &&
-    !os::is_first_C_frame(&fr) && fr.sender_pc()) {
-    fr = os::get_sender_for_C_frame(&fr);
-    n --;
+  while (fr.pc() && frame_idx < frames) {
+    if (toSkip > 0) {
+      toSkip --;
+    } else {
+      stack[frame_idx ++] = fr.pc();
+    }
+    if (fr.fp() == NULL || os::is_first_C_frame(&fr)
+        ||fr.sender_pc() == NULL || fr.cb() != NULL) break;
+
+    if (fr.sender_pc() && !os::is_first_C_frame(&fr)) {
+      fr = os::get_sender_for_C_frame(&fr);
+    } else {
+      break;
+    }
   }
-  if (n == 0) {
-    return fr.pc();
-  } else {
-    return NULL;
+  num_of_frames = frame_idx;
+  for (; frame_idx < frames; frame_idx ++) {
+    stack[frame_idx] = NULL;
   }
+
+  return num_of_frames;
+}
+
+
+bool os::unsetenv(const char* name) {
+  assert(name != NULL, "Null pointer");
+  return (::unsetenv(name) == 0);
 }
 
 int os::get_last_error() {

src/os/solaris/vm/perfMemory_solaris.cpp

@@ -770,7 +770,8 @@ static char* mmap_create_shared(size_t size) {
   (void)::memset((void*) mapAddress, 0, size);
 
   // it does not go through os api, the operation has to record from here
-  MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC);
+  MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress,
+    size, CURRENT_PC, mtInternal);
 
   return mapAddress;
 }
@@ -941,7 +942,8 @@ static void mmap_attach_shared(const char* user, int vmid, PerfMemory::PerfMemor
   }
 
   // it does not go through os api, the operation has to record from here
-  MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC);
+  MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress,
+    size, CURRENT_PC, mtInternal);
 
   *addr = mapAddress;
   *sizep = size;

src/os/windows/vm/os_windows.cpp

@@ -131,6 +131,12 @@ BOOL WINAPI DllMain(HINSTANCE hinst, DWORD reason, LPVOID reserved) {
     case DLL_PROCESS_DETACH:
       if(ForceTimeHighResolution)
         timeEndPeriod(1L);
+
+       // Workaround for issue when a custom launcher doesn't call
+       // DestroyJavaVM and NMT is trying to track memory when free is
+       // called from a static destructor
+       MemTracker::shutdown();
+
       break;
     default:
       break;
@@ -153,6 +159,10 @@ bool os::getenv(const char* name, char* buffer, int len) {
  return result > 0 && result < len;
 }
 
+bool os::unsetenv(const char* name) {
+  assert(name != NULL, "Null pointer");
+  return (SetEnvironmentVariable(name, NULL) == TRUE);
+}
 
 // No setuid programs under Windows.
 bool os::have_special_privileges() {
@@ -311,15 +321,17 @@ extern "C" void breakpoint() {
  * So far, this method is only used by Native Memory Tracking, which is
  * only supported on Windows XP or later.
  */
-address os::get_caller_pc(int n) {
+
+int os::get_native_stack(address* stack, int frames, int toSkip) {
 #ifdef _NMT_NOINLINE_
-  n ++;
+  toSkip ++;
 #endif
-  address pc;
-  if (os::Kernel32Dll::RtlCaptureStackBackTrace(n + 1, 1, (PVOID*)&pc, NULL) == 1) {
-    return pc;
+  int captured = Kernel32Dll::RtlCaptureStackBackTrace(toSkip + 1, frames,
+    (PVOID*)stack, NULL);
+  for (int index = captured; index < frames; index ++) {
+    stack[index] = NULL;
   }
-  return NULL;
+  return captured;
 }
 
 
@@ -2904,7 +2916,7 @@ static char* allocate_pages_individually(size_t bytes, char* addr, DWORD flags,
                                 PAGE_READWRITE);
   // If reservation failed, return NULL
   if (p_buf == NULL) return NULL;
-  MemTracker::record_virtual_memory_reserve((address)p_buf, size_of_reserve, mtNone, CALLER_PC);
+  MemTracker::record_virtual_memory_reserve((address)p_buf, size_of_reserve, CALLER_PC);
   os::release_memory(p_buf, bytes + chunk_size);
 
   // we still need to round up to a page boundary (in case we are using large pages)
@@ -2970,7 +2982,7 @@ static char* allocate_pages_individually(size_t bytes, char* addr, DWORD flags,
         // need to create a dummy 'reserve' record to match
         // the release.
         MemTracker::record_virtual_memory_reserve((address)p_buf,
-          bytes_to_release, mtNone, CALLER_PC);
+          bytes_to_release, CALLER_PC);
         os::release_memory(p_buf, bytes_to_release);
       }
 #ifdef ASSERT
@@ -2989,11 +3001,10 @@ static char* allocate_pages_individually(size_t bytes, char* addr, DWORD flags,
   }
   // Although the memory is allocated individually, it is returned as one.
   // NMT records it as one block.
-  address pc = CALLER_PC;
   if ((flags & MEM_COMMIT) != 0) {
-    MemTracker::record_virtual_memory_reserve_and_commit((address)p_buf, bytes, mtNone, pc);
+    MemTracker::record_virtual_memory_reserve_and_commit((address)p_buf, bytes, CALLER_PC);
   } else {
-    MemTracker::record_virtual_memory_reserve((address)p_buf, bytes, mtNone, pc);
+    MemTracker::record_virtual_memory_reserve((address)p_buf, bytes, CALLER_PC);
   }
 
   // made it this far, success
@@ -3191,8 +3202,7 @@ char* os::reserve_memory_special(size_t bytes, size_t alignment, char* addr, boo
     DWORD flag = MEM_RESERVE | MEM_COMMIT | MEM_LARGE_PAGES;
     char * res = (char *)VirtualAlloc(addr, bytes, flag, prot);
     if (res != NULL) {
-      address pc = CALLER_PC;
-      MemTracker::record_virtual_memory_reserve_and_commit((address)res, bytes, mtNone, pc);
+      MemTracker::record_virtual_memory_reserve_and_commit((address)res, bytes, CALLER_PC);
     }
 
     return res;

src/os/windows/vm/perfMemory_windows.cpp

 /*
- * Copyright (c) 2001, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -1498,7 +1498,8 @@ static char* mapping_create_shared(size_t size) {
   (void)memset(mapAddress, '\0', size);
 
   // it does not go through os api, the operation has to record from here
-  MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC);
+  MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress,
+    size, CURRENT_PC, mtInternal);
 
   return (char*) mapAddress;
 }
@@ -1680,7 +1681,8 @@ static void open_file_mapping(const char* user, int vmid,
   }
 
   // it does not go through os api, the operation has to record from here
-  MemTracker::record_virtual_memory_reserve((address)mapAddress, size, mtInternal, CURRENT_PC);
+  MemTracker::record_virtual_memory_reserve_and_commit((address)mapAddress, size,
+    CURRENT_PC, mtInternal);
 
 
   *addrp = (char*)mapAddress;
@@ -1834,10 +1836,14 @@ void PerfMemory::detach(char* addr, size_t bytes, TRAPS) {
     return;
   }
 
-  MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
-  remove_file_mapping(addr);
-  // it does not go through os api, the operation has to record from here
-  tkr.record((address)addr, bytes);
+  if (MemTracker::tracking_level() > NMT_minimal) {
+    // it does not go through os api, the operation has to record from here
+    Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
+    remove_file_mapping(addr);
+    tkr.record((address)addr, bytes);
+  } else {
+    remove_file_mapping(addr);
+  }
 }
 
 char* PerfMemory::backing_store_filename() {

src/share/vm/asm/codeBuffer.cpp

@@ -268,7 +268,7 @@ address CodeBuffer::decode_begin() {
 
 GrowableArray<int>* CodeBuffer::create_patch_overflow() {
   if (_overflow_arena == NULL) {
-    _overflow_arena = new (mtCode) Arena();
+    _overflow_arena = new (mtCode) Arena(mtCode);
   }
   return new (_overflow_arena) GrowableArray<int>(_overflow_arena, 8, 0, 0);
 }

src/share/vm/c1/c1_Compiler.cpp

@@ -47,7 +47,7 @@ Compiler::Compiler () {}
 
 void Compiler::init_c1_runtime() {
   BufferBlob* buffer_blob = CompilerThread::current()->get_buffer_blob();
-  Arena* arena = new (mtCompiler) Arena();
+  Arena* arena = new (mtCompiler) Arena(mtCompiler);
   Runtime1::initialize(buffer_blob);
   FrameMap::initialize();
   // initialize data structures

src/share/vm/ci/ciEnv.cpp

@@ -86,7 +86,8 @@ static bool firstEnv = true;
 
 // ------------------------------------------------------------------
 // ciEnv::ciEnv
-ciEnv::ciEnv(CompileTask* task, int system_dictionary_modification_counter) {
+ciEnv::ciEnv(CompileTask* task, int system_dictionary_modification_counter)
+  : _ciEnv_arena(mtCompiler) {
   VM_ENTRY_MARK;
 
   // Set up ciEnv::current immediately, for the sake of ciObjectFactory, etc.
@@ -139,7 +140,7 @@ ciEnv::ciEnv(CompileTask* task, int system_dictionary_modification_counter) {
   _the_min_jint_string = NULL;
 }
 
-ciEnv::ciEnv(Arena* arena) {
+ciEnv::ciEnv(Arena* arena) : _ciEnv_arena(mtCompiler) {
   ASSERT_IN_VM;
 
   // Set up ciEnv::current immediately, for the sake of ciObjectFactory, etc.

src/share/vm/ci/ciObjectFactory.cpp

 /*
- * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -112,7 +112,7 @@ void ciObjectFactory::initialize() {
   // This Arena is long lived and exists in the resource mark of the
   // compiler thread that initializes the initial ciObjectFactory which
   // creates the shared ciObjects that all later ciObjectFactories use.
-  Arena* arena = new (mtCompiler) Arena();
+  Arena* arena = new (mtCompiler) Arena(mtCompiler);
   ciEnv initial(arena);
   ciEnv* env = ciEnv::current();
   env->_factory->init_shared_objects();

src/share/vm/classfile/symbolTable.cpp

@@ -74,9 +74,9 @@ Symbol* SymbolTable::allocate_symbol(const u1* name, int len, bool c_heap, TRAPS
 void SymbolTable::initialize_symbols(int arena_alloc_size) {
   // Initialize the arena for global symbols, size passed in depends on CDS.
   if (arena_alloc_size == 0) {
-    _arena = new (mtSymbol) Arena();
+    _arena = new (mtSymbol) Arena(mtSymbol);
   } else {
-    _arena = new (mtSymbol) Arena(arena_alloc_size);
+    _arena = new (mtSymbol) Arena(mtSymbol, arena_alloc_size);
   }
 }
 

src/share/vm/gc_implementation/concurrentMarkSweep/cmsCollectorPolicy.cpp

 /*
- * Copyright (c) 2007, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2007, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -53,7 +53,8 @@ void ConcurrentMarkSweepPolicy::initialize_alignments() {
 }
 
 void ConcurrentMarkSweepPolicy::initialize_generations() {
-  _generations = NEW_C_HEAP_ARRAY3(GenerationSpecPtr, number_of_generations(), mtGC, 0, AllocFailStrategy::RETURN_NULL);
+  _generations = NEW_C_HEAP_ARRAY3(GenerationSpecPtr, number_of_generations(), mtGC,
+    CURRENT_PC, AllocFailStrategy::RETURN_NULL);
   if (_generations == NULL)
     vm_exit_during_initialization("Unable to allocate gen spec");
 

src/share/vm/gc_implementation/g1/heapRegionRemSet.cpp

@@ -288,7 +288,7 @@ OtherRegionsTable::OtherRegionsTable(HeapRegion* hr, Mutex* m) :
   }
 
   _fine_grain_regions = NEW_C_HEAP_ARRAY3(PerRegionTablePtr, _max_fine_entries,
-                        mtGC, 0, AllocFailStrategy::RETURN_NULL);
+                        mtGC, CURRENT_PC, AllocFailStrategy::RETURN_NULL);
 
   if (_fine_grain_regions == NULL) {
     vm_exit_out_of_memory(sizeof(void*)*_max_fine_entries, OOM_MALLOC_ERROR,

src/share/vm/memory/allocation.cpp

@@ -438,24 +438,22 @@ void Chunk::start_chunk_pool_cleaner_task() {
 }
 
 //------------------------------Arena------------------------------------------
-NOT_PRODUCT(volatile jint Arena::_instance_count = 0;)
-
-Arena::Arena(size_t init_size) {
+Arena::Arena(MEMFLAGS flag, size_t init_size) : _flags(flag), _size_in_bytes(0)  {
   size_t round_size = (sizeof (char *)) - 1;
   init_size = (init_size+round_size) & ~round_size;
   _first = _chunk = new (AllocFailStrategy::EXIT_OOM, init_size) Chunk(init_size);
   _hwm = _chunk->bottom();      // Save the cached hwm, max
   _max = _chunk->top();
+  MemTracker::record_new_arena(flag);
   set_size_in_bytes(init_size);
-  NOT_PRODUCT(Atomic::inc(&_instance_count);)
 }
 
-Arena::Arena() {
+Arena::Arena(MEMFLAGS flag) : _flags(flag), _size_in_bytes(0) {
   _first = _chunk = new (AllocFailStrategy::EXIT_OOM, Chunk::init_size) Chunk(Chunk::init_size);
   _hwm = _chunk->bottom();      // Save the cached hwm, max
   _max = _chunk->top();
+  MemTracker::record_new_arena(flag);
   set_size_in_bytes(Chunk::init_size);
-  NOT_PRODUCT(Atomic::inc(&_instance_count);)
 }
 
 Arena *Arena::move_contents(Arena *copy) {
@@ -477,7 +475,7 @@ Arena *Arena::move_contents(Arena *copy) {
 
 Arena::~Arena() {
   destruct_contents();
-  NOT_PRODUCT(Atomic::dec(&_instance_count);)
+  MemTracker::record_arena_free(_flags);
 }
 
 void* Arena::operator new(size_t size) throw() {
@@ -493,21 +491,21 @@ void* Arena::operator new (size_t size, const std::nothrow_t&  nothrow_constant)
   // dynamic memory type binding
 void* Arena::operator new(size_t size, MEMFLAGS flags) throw() {
 #ifdef ASSERT
-  void* p = (void*)AllocateHeap(size, flags|otArena, CALLER_PC);
+  void* p = (void*)AllocateHeap(size, flags, CALLER_PC);
   if (PrintMallocFree) trace_heap_malloc(size, "Arena-new", p);
   return p;
 #else
-  return (void *) AllocateHeap(size, flags|otArena, CALLER_PC);
+  return (void *) AllocateHeap(size, flags, CALLER_PC);
 #endif
 }
 
 void* Arena::operator new(size_t size, const std::nothrow_t& nothrow_constant, MEMFLAGS flags) throw() {
 #ifdef ASSERT
-  void* p = os::malloc(size, flags|otArena, CALLER_PC);
+  void* p = os::malloc(size, flags, CALLER_PC);
   if (PrintMallocFree) trace_heap_malloc(size, "Arena-new", p);
   return p;
 #else
-  return os::malloc(size, flags|otArena, CALLER_PC);
+  return os::malloc(size, flags, CALLER_PC);
 #endif
 }
 
@@ -532,8 +530,9 @@ void Arena::destruct_contents() {
 // change the size
 void Arena::set_size_in_bytes(size_t size) {
   if (_size_in_bytes != size) {
+    long delta = (long)(size - size_in_bytes());
     _size_in_bytes = size;
-    MemTracker::record_arena_size((address)this, size);
+    MemTracker::record_arena_size_change(delta, _flags);
   }
 }
 

src/share/vm/memory/allocation.hpp

 /*
- * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -133,51 +133,34 @@ class AllocatedObj {
 
 
 /*
- * MemoryType bitmap layout:
- * | 16 15 14 13 12 11 10 09 | 08 07 06 05 | 04 03 02 01 |
- * |      memory type        |   object    | reserved    |
- * |                         |     type    |             |
+ * Memory types
  */
 enum MemoryType {
   // Memory type by sub systems. It occupies lower byte.
-  mtNone              = 0x0000,  // undefined
-  mtClass             = 0x0100,  // memory class for Java classes
-  mtThread            = 0x0200,  // memory for thread objects
-  mtThreadStack       = 0x0300,
-  mtCode              = 0x0400,  // memory for generated code
-  mtGC                = 0x0500,  // memory for GC
-  mtCompiler          = 0x0600,  // memory for compiler
-  mtInternal          = 0x0700,  // memory used by VM, but does not belong to
+  mtJavaHeap          = 0x00,  // Java heap
+  mtClass             = 0x01,  // memory class for Java classes
+  mtThread            = 0x02,  // memory for thread objects
+  mtThreadStack       = 0x03,
+  mtCode              = 0x04,  // memory for generated code
+  mtGC                = 0x05,  // memory for GC
+  mtCompiler          = 0x06,  // memory for compiler
+  mtInternal          = 0x07,  // memory used by VM, but does not belong to
                                  // any of above categories, and not used for
                                  // native memory tracking
-  mtOther             = 0x0800,  // memory not used by VM
-  mtSymbol            = 0x0900,  // symbol
-  mtNMT               = 0x0A00,  // memory used by native memory tracking
-  mtChunk             = 0x0B00,  // chunk that holds content of arenas
-  mtJavaHeap          = 0x0C00,  // Java heap
-  mtClassShared       = 0x0D00,  // class data sharing
-  mtTest              = 0x0E00,  // Test type for verifying NMT
-  mtTracing           = 0x0F00,  // memory used for Tracing
-  mt_number_of_types  = 0x000F,  // number of memory types (mtDontTrack
+  mtOther             = 0x08,  // memory not used by VM
+  mtSymbol            = 0x09,  // symbol
+  mtNMT               = 0x0A,  // memory used by native memory tracking
+  mtClassShared       = 0x0B,  // class data sharing
+  mtChunk             = 0x0C,  // chunk that holds content of arenas
+  mtTest              = 0x0D,  // Test type for verifying NMT
+  mtTracing           = 0x0E,  // memory used for Tracing
+  mtNone              = 0x0F,  // undefined
+  mt_number_of_types  = 0x10   // number of memory types (mtDontTrack
                                  // is not included as validate type)
-  mtDontTrack         = 0x0F00,  // memory we do not or cannot track
-  mt_masks            = 0x7F00,
-
-  // object type mask
-  otArena             = 0x0010, // an arena object
-  otNMTRecorder       = 0x0020, // memory recorder object
-  ot_masks            = 0x00F0
 };
 
-#define IS_MEMORY_TYPE(flags, type) ((flags & mt_masks) == type)
-#define HAS_VALID_MEMORY_TYPE(flags)((flags & mt_masks) != mtNone)
-#define FLAGS_TO_MEMORY_TYPE(flags) (flags & mt_masks)
-
-#define IS_ARENA_OBJ(flags)         ((flags & ot_masks) == otArena)
-#define IS_NMT_RECORDER(flags)      ((flags & ot_masks) == otNMTRecorder)
-#define NMT_CAN_TRACK(flags)        (!IS_NMT_RECORDER(flags) && !(IS_MEMORY_TYPE(flags, mtDontTrack)))
+typedef MemoryType MEMFLAGS;
 
-typedef unsigned short MEMFLAGS;
 
 #if INCLUDE_NMT
 
@@ -189,27 +172,23 @@ const bool NMT_track_callsite = false;
 
 #endif // INCLUDE_NMT
 
-// debug build does not inline
-#if defined(_NMT_NOINLINE_)
-  #define CURRENT_PC       (NMT_track_callsite ? os::get_caller_pc(1) : 0)
-  #define CALLER_PC        (NMT_track_callsite ? os::get_caller_pc(2) : 0)
-  #define CALLER_CALLER_PC (NMT_track_callsite ? os::get_caller_pc(3) : 0)
-#else
-  #define CURRENT_PC      (NMT_track_callsite? os::get_caller_pc(0) : 0)
-  #define CALLER_PC       (NMT_track_callsite ? os::get_caller_pc(1) : 0)
-  #define CALLER_CALLER_PC (NMT_track_callsite ? os::get_caller_pc(2) : 0)
-#endif
-
+class NativeCallStack;
 
 
 template <MEMFLAGS F> class CHeapObj ALLOCATION_SUPER_CLASS_SPEC {
  public:
-  _NOINLINE_ void* operator new(size_t size, address caller_pc = 0) throw();
+  _NOINLINE_ void* operator new(size_t size, const NativeCallStack& stack) throw();
+  _NOINLINE_ void* operator new(size_t size) throw();
   _NOINLINE_ void* operator new (size_t size, const std::nothrow_t&  nothrow_constant,
-                               address caller_pc = 0) throw();
-  _NOINLINE_ void* operator new [](size_t size, address caller_pc = 0) throw();
+                               const NativeCallStack& stack) throw();
+  _NOINLINE_ void* operator new (size_t size, const std::nothrow_t&  nothrow_constant)
+                               throw();
+  _NOINLINE_ void* operator new [](size_t size, const NativeCallStack& stack) throw();
+  _NOINLINE_ void* operator new [](size_t size) throw();
   _NOINLINE_ void* operator new [](size_t size, const std::nothrow_t&  nothrow_constant,
-                               address caller_pc = 0) throw();
+                               const NativeCallStack& stack) throw();
+  _NOINLINE_ void* operator new [](size_t size, const std::nothrow_t&  nothrow_constant)
+                               throw();
   void  operator delete(void* p);
   void  operator delete [] (void* p);
 };
@@ -384,13 +363,15 @@ class Chunk: CHeapObj<mtChunk> {
 
 //------------------------------Arena------------------------------------------
 // Fast allocation of memory
-class Arena : public CHeapObj<mtNone|otArena> {
+class Arena : public CHeapObj<mtNone> {
 protected:
   friend class ResourceMark;
   friend class HandleMark;
   friend class NoHandleMark;
   friend class VMStructs;
 
+  MEMFLAGS    _flags;           // Memory tracking flags
+
   Chunk *_first;                // First chunk
   Chunk *_chunk;                // current chunk
   char *_hwm, *_max;            // High water mark and max in current chunk
@@ -418,8 +399,8 @@ protected:
  }
 
  public:
-  Arena();
-  Arena(size_t init_size);
+  Arena(MEMFLAGS memflag);
+  Arena(MEMFLAGS memflag, size_t init_size);
   ~Arena();
   void  destruct_contents();
   char* hwm() const             { return _hwm; }
@@ -518,8 +499,6 @@ protected:
   static void free_malloced_objects(Chunk* chunk, char* hwm, char* max, char* hwm2)  PRODUCT_RETURN;
   static void free_all(char** start, char** end)                                     PRODUCT_RETURN;
 
-  // how many arena instances
-  NOT_PRODUCT(static volatile jint _instance_count;)
 private:
   // Reset this Arena to empty, access will trigger grow if necessary
   void   reset(void) {
@@ -681,7 +660,7 @@ class ResourceObj ALLOCATION_SUPER_CLASS_SPEC {
   NEW_C_HEAP_ARRAY3(type, (size), memflags, pc, AllocFailStrategy::RETURN_NULL)
 
 #define NEW_C_HEAP_ARRAY_RETURN_NULL(type, size, memflags)\
-  NEW_C_HEAP_ARRAY3(type, (size), memflags, (address)0, AllocFailStrategy::RETURN_NULL)
+  NEW_C_HEAP_ARRAY3(type, (size), memflags, CURRENT_PC, AllocFailStrategy::RETURN_NULL)
 
 #define REALLOC_C_HEAP_ARRAY(type, old, size, memflags)\
   (type*) (ReallocateHeap((char*)(old), (size) * sizeof(type), memflags))

src/share/vm/memory/allocation.inline.hpp

 /*
- * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -27,6 +27,7 @@
 
 #include "runtime/atomic.inline.hpp"
 #include "runtime/os.hpp"
+#include "services/memTracker.hpp"
 
 // Explicit C-heap memory management
 
@@ -49,12 +50,10 @@ inline void inc_stat_counter(volatile julong* dest, julong add_value) {
 #endif
 
 // allocate using malloc; will fail if no memory available
-inline char* AllocateHeap(size_t size, MEMFLAGS flags, address pc = 0,
+inline char* AllocateHeap(size_t size, MEMFLAGS flags,
+    const NativeCallStack& stack,
     AllocFailType alloc_failmode = AllocFailStrategy::EXIT_OOM) {
-  if (pc == 0) {
-    pc = CURRENT_PC;
-  }
-  char* p = (char*) os::malloc(size, flags, pc);
+  char* p = (char*) os::malloc(size, flags, stack);
   #ifdef ASSERT
   if (PrintMallocFree) trace_heap_malloc(size, "AllocateHeap", p);
   #endif
@@ -63,10 +62,14 @@ inline char* AllocateHeap(size_t size, MEMFLAGS flags, address pc = 0,
   }
   return p;
 }
+inline char* AllocateHeap(size_t size, MEMFLAGS flags,
+    AllocFailType alloc_failmode = AllocFailStrategy::EXIT_OOM) {
+  return AllocateHeap(size, flags, CURRENT_PC, alloc_failmode);
+}
 
-inline char* ReallocateHeap(char *old, size_t size, MEMFLAGS flags,
+inline char* ReallocateHeap(char *old, size_t size, MEMFLAGS flag,
     AllocFailType alloc_failmode = AllocFailStrategy::EXIT_OOM) {
-  char* p = (char*) os::realloc(old, size, flags, CURRENT_PC);
+  char* p = (char*) os::realloc(old, size, flag, CURRENT_PC);
   #ifdef ASSERT
   if (PrintMallocFree) trace_heap_malloc(size, "ReallocateHeap", p);
   #endif
@@ -85,32 +88,51 @@ inline void FreeHeap(void* p, MEMFLAGS memflags = mtInternal) {
 
 
 template <MEMFLAGS F> void* CHeapObj<F>::operator new(size_t size,
-      address caller_pc) throw() {
-    void* p = (void*)AllocateHeap(size, F, (caller_pc != 0 ? caller_pc : CALLER_PC));
+      const NativeCallStack& stack) throw() {
+  void* p = (void*)AllocateHeap(size, F, stack);
 #ifdef ASSERT
-    if (PrintMallocFree) trace_heap_malloc(size, "CHeapObj-new", p);
+  if (PrintMallocFree) trace_heap_malloc(size, "CHeapObj-new", p);
 #endif
-    return p;
-  }
+  return p;
+}
+
+template <MEMFLAGS F> void* CHeapObj<F>::operator new(size_t size) throw() {
+  return CHeapObj<F>::operator new(size, CALLER_PC);
+}
 
 template <MEMFLAGS F> void* CHeapObj<F>::operator new (size_t size,
-  const std::nothrow_t&  nothrow_constant, address caller_pc) throw() {
-  void* p = (void*)AllocateHeap(size, F, (caller_pc != 0 ? caller_pc : CALLER_PC),
+  const std::nothrow_t&  nothrow_constant, const NativeCallStack& stack) throw() {
+  void* p = (void*)AllocateHeap(size, F, stack,
       AllocFailStrategy::RETURN_NULL);
 #ifdef ASSERT
     if (PrintMallocFree) trace_heap_malloc(size, "CHeapObj-new", p);
 #endif
     return p;
+  }
+
+template <MEMFLAGS F> void* CHeapObj<F>::operator new (size_t size,
+  const std::nothrow_t& nothrow_constant) throw() {
+  return CHeapObj<F>::operator new(size, nothrow_constant, CALLER_PC);
+}
+
+template <MEMFLAGS F> void* CHeapObj<F>::operator new [](size_t size,
+      const NativeCallStack& stack) throw() {
+  return CHeapObj<F>::operator new(size, stack);
+}
+
+template <MEMFLAGS F> void* CHeapObj<F>::operator new [](size_t size)
+  throw() {
+  return CHeapObj<F>::operator new(size, CALLER_PC);
 }
 
 template <MEMFLAGS F> void* CHeapObj<F>::operator new [](size_t size,
-      address caller_pc) throw() {
-    return CHeapObj<F>::operator new(size, caller_pc);
+  const std::nothrow_t&  nothrow_constant, const NativeCallStack& stack) throw() {
+  return CHeapObj<F>::operator new(size, nothrow_constant, stack);
 }
 
 template <MEMFLAGS F> void* CHeapObj<F>::operator new [](size_t size,
-  const std::nothrow_t&  nothrow_constant, address caller_pc) throw() {
-    return CHeapObj<F>::operator new(size, nothrow_constant, caller_pc);
+  const std::nothrow_t& nothrow_constant) throw() {
+  return CHeapObj<F>::operator new(size, nothrow_constant, CALLER_PC);
 }
 
 template <MEMFLAGS F> void CHeapObj<F>::operator delete(void* p){

src/share/vm/memory/cardTableRS.cpp

@@ -56,7 +56,7 @@ CardTableRS::CardTableRS(MemRegion whole_heap,
   _ct_bs->initialize();
   set_bs(_ct_bs);
   _last_cur_val_in_gen = NEW_C_HEAP_ARRAY3(jbyte, GenCollectedHeap::max_gens + 1,
-                         mtGC, 0, AllocFailStrategy::RETURN_NULL);
+                         mtGC, CURRENT_PC, AllocFailStrategy::RETURN_NULL);
   if (_last_cur_val_in_gen == NULL) {
     vm_exit_during_initialization("Could not create last_cur_val_in_gen array.");
   }

src/share/vm/memory/collectorPolicy.cpp

@@ -969,7 +969,8 @@ void MarkSweepPolicy::initialize_alignments() {
 }
 
 void MarkSweepPolicy::initialize_generations() {
-  _generations = NEW_C_HEAP_ARRAY3(GenerationSpecPtr, number_of_generations(), mtGC, 0, AllocFailStrategy::RETURN_NULL);
+  _generations = NEW_C_HEAP_ARRAY3(GenerationSpecPtr, number_of_generations(), mtGC, CURRENT_PC,
+    AllocFailStrategy::RETURN_NULL);
   if (_generations == NULL) {
     vm_exit_during_initialization("Unable to allocate gen spec");
   }

src/share/vm/memory/heapInspection.cpp

@@ -135,7 +135,7 @@ KlassInfoTable::KlassInfoTable(bool need_class_stats) {
   _ref = (HeapWord*) Universe::boolArrayKlassObj();
   _buckets =
     (KlassInfoBucket*)  AllocateHeap(sizeof(KlassInfoBucket) * _num_buckets,
-                                            mtInternal, 0, AllocFailStrategy::RETURN_NULL);
+       mtInternal, CURRENT_PC, AllocFailStrategy::RETURN_NULL);
   if (_buckets != NULL) {
     _size = _num_buckets;
     for (int index = 0; index < _size; index++) {

src/share/vm/memory/memRegion.cpp

 /*
- * Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2000, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -103,11 +103,13 @@ MemRegion MemRegion::minus(const MemRegion mr2) const {
 }
 
 void* MemRegion::operator new(size_t size) throw() {
-  return (address)AllocateHeap(size, mtGC, 0, AllocFailStrategy::RETURN_NULL);
+  return (address)AllocateHeap(size, mtGC, CURRENT_PC,
+    AllocFailStrategy::RETURN_NULL);
 }
 
 void* MemRegion::operator new [](size_t size) throw() {
-  return (address)AllocateHeap(size, mtGC, 0, AllocFailStrategy::RETURN_NULL);
+  return (address)AllocateHeap(size, mtGC, CURRENT_PC,
+    AllocFailStrategy::RETURN_NULL);
 }
 void  MemRegion::operator delete(void* p) {
   FreeHeap(p, mtGC);

src/share/vm/memory/resourceArea.hpp

 /*
- * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -49,11 +49,11 @@ class ResourceArea: public Arena {
   debug_only(static int _warned;)       // to suppress multiple warnings
 
 public:
-  ResourceArea() {
+  ResourceArea() : Arena(mtThread) {
     debug_only(_nesting = 0;)
   }
 
-  ResourceArea(size_t init_size) : Arena(init_size) {
+  ResourceArea(size_t init_size) : Arena(mtThread, init_size) {
     debug_only(_nesting = 0;);
   }
 
@@ -64,7 +64,7 @@ public:
     if (UseMallocOnly) {
       // use malloc, but save pointer in res. area for later freeing
       char** save = (char**)internal_malloc_4(sizeof(char*));
-      return (*save = (char*)os::malloc(size, mtThread));
+      return (*save = (char*)os::malloc(size, mtThread, CURRENT_PC));
     }
 #endif
     return (char*)Amalloc(size, alloc_failmode);

src/share/vm/opto/compile.cpp

@@ -665,6 +665,10 @@ Compile::Compile( ciEnv* ci_env, C2Compiler* compiler, ciMethod* target, int osr
                   _printer(IdealGraphPrinter::printer()),
 #endif
                   _congraph(NULL),
+                  _comp_arena(mtCompiler),
+                  _node_arena(mtCompiler),
+                  _old_arena(mtCompiler),
+                  _Compile_types(mtCompiler),
                   _replay_inline_data(NULL),
                   _late_inlines(comp_arena(), 2, 0, NULL),
                   _string_late_inlines(comp_arena(), 2, 0, NULL),
@@ -972,6 +976,10 @@ Compile::Compile( ciEnv* ci_env,
     _in_dump_cnt(0),
     _printer(NULL),
 #endif
+    _comp_arena(mtCompiler),
+    _node_arena(mtCompiler),
+    _old_arena(mtCompiler),
+    _Compile_types(mtCompiler),
     _dead_node_list(comp_arena()),
     _dead_node_count(0),
     _congraph(NULL),

src/share/vm/opto/type.cpp

@@ -265,7 +265,7 @@ void Type::Initialize_shared(Compile* current) {
   // locking.
 
   Arena* save = current->type_arena();
-  Arena* shared_type_arena = new (mtCompiler)Arena();
+  Arena* shared_type_arena = new (mtCompiler)Arena(mtCompiler);
 
   current->set_type_arena(shared_type_arena);
   _shared_type_dict =

src/share/vm/precompiled/precompiled.hpp

 /*
- * Copyright (c) 2010, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2010, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -220,10 +220,17 @@
 # include "runtime/vmThread.hpp"
 # include "runtime/vm_operations.hpp"
 # include "runtime/vm_version.hpp"
+# include "services/allocationSite.hpp"
 # include "services/lowMemoryDetector.hpp"
+# include "services/mallocTracker.hpp"
+# include "services/memBaseline.hpp"
 # include "services/memoryPool.hpp"
 # include "services/memoryService.hpp"
 # include "services/memoryUsage.hpp"
+# include "services/memReporter.hpp"
+# include "services/memTracker.hpp"
+# include "services/nmtCommon.hpp"
+# include "services/virtualMemoryTracker.hpp"
 # include "utilities/accessFlags.hpp"
 # include "utilities/array.hpp"
 # include "utilities/bitMap.hpp"
@@ -237,6 +244,7 @@
 # include "utilities/hashtable.hpp"
 # include "utilities/histogram.hpp"
 # include "utilities/macros.hpp"
+# include "utilities/nativeCallStack.hpp"
 # include "utilities/numberSeq.hpp"
 # include "utilities/ostream.hpp"
 # include "utilities/preserveException.hpp"

src/share/vm/prims/jni.cpp

@@ -73,6 +73,7 @@
 #include "runtime/signature.hpp"
 #include "runtime/thread.inline.hpp"
 #include "runtime/vm_operations.hpp"
+#include "services/memTracker.hpp"
 #include "services/runtimeService.hpp"
 #include "trace/tracing.hpp"
 #include "utilities/defaultStream.hpp"
@@ -3582,6 +3583,7 @@ static char* get_bad_address() {
     if (bad_address != NULL) {
       os::protect_memory(bad_address, size, os::MEM_PROT_READ,
                          /*is_committed*/false);
+      MemTracker::record_virtual_memory_type((void*)bad_address, mtInternal);
     }
   }
   return bad_address;
@@ -5077,6 +5079,7 @@ void TestMetachunk_test();
 void TestVirtualSpaceNode_test();
 void TestNewSize_test();
 void TestKlass_test();
+void Test_linked_list();
 #if INCLUDE_ALL_GCS
 void TestOldFreeSpaceCalculation_test();
 void TestG1BiasedArray_test();
@@ -5104,6 +5107,7 @@ void execute_internal_vm_tests() {
     run_unit_test(test_loggc_filename());
     run_unit_test(TestNewSize_test());
     run_unit_test(TestKlass_test());
+    run_unit_test(Test_linked_list());
 #if INCLUDE_VM_STRUCTS
     run_unit_test(VMStructs::test());
 #endif

src/share/vm/prims/whitebox.cpp

@@ -49,8 +49,10 @@
 #include "gc_implementation/g1/heapRegionRemSet.hpp"
 #endif // INCLUDE_ALL_GCS
 
-#ifdef INCLUDE_NMT
+#if INCLUDE_NMT
+#include "services/mallocSiteTable.hpp"
 #include "services/memTracker.hpp"
+#include "utilities/nativeCallStack.hpp"
 #endif // INCLUDE_NMT
 
 #include "compiler/compileBroker.hpp"
@@ -276,14 +278,18 @@ WB_END
 // NMT picks it up correctly
 WB_ENTRY(jlong, WB_NMTMalloc(JNIEnv* env, jobject o, jlong size))
   jlong addr = 0;
-
-  if (MemTracker::is_on() && !MemTracker::shutdown_in_progress()) {
     addr = (jlong)(uintptr_t)os::malloc(size, mtTest);
-  }
-
   return addr;
 WB_END
 
+// Alloc memory with pseudo call stack. The test can create psudo malloc
+// allocation site to stress the malloc tracking.
+WB_ENTRY(jlong, WB_NMTMallocWithPseudoStack(JNIEnv* env, jobject o, jlong size, jint pseudo_stack))
+  address pc = (address)(size_t)pseudo_stack;
+  NativeCallStack stack(&pc, 1);
+  return (jlong)os::malloc(size, mtTest, stack);
+WB_END
+
 // Free the memory allocated by NMTAllocTest
 WB_ENTRY(void, WB_NMTFree(JNIEnv* env, jobject o, jlong mem))
   os::free((void*)(uintptr_t)mem, mtTest);
@@ -292,10 +298,8 @@ WB_END
 WB_ENTRY(jlong, WB_NMTReserveMemory(JNIEnv* env, jobject o, jlong size))
   jlong addr = 0;
 
-  if (MemTracker::is_on() && !MemTracker::shutdown_in_progress()) {
     addr = (jlong)(uintptr_t)os::reserve_memory(size);
     MemTracker::record_virtual_memory_type((address)addr, mtTest);
-  }
 
   return addr;
 WB_END
@@ -314,19 +318,19 @@ WB_ENTRY(void, WB_NMTReleaseMemory(JNIEnv* env, jobject o, jlong addr, jlong siz
   os::release_memory((char *)(uintptr_t)addr, size);
 WB_END
 
-// Block until the current generation of NMT data to be merged, used to reliably test the NMT feature
-WB_ENTRY(jboolean, WB_NMTWaitForDataMerge(JNIEnv* env))
+WB_ENTRY(jboolean, WB_NMTIsDetailSupported(JNIEnv* env))
+  return MemTracker::tracking_level() == NMT_detail;
+WB_END
 
-  if (!MemTracker::is_on() || MemTracker::shutdown_in_progress()) {
-    return false;
+WB_ENTRY(void, WB_NMTOverflowHashBucket(JNIEnv* env, jobject o, jlong num))
+  address pc = (address)1;
+  for (jlong index = 0; index < num; index ++) {
+    NativeCallStack stack(&pc, 1);
+    os::malloc(0, mtTest, stack);
+    pc += MallocSiteTable::hash_buckets();
   }
-
-  return MemTracker::wbtest_wait_for_data_merge();
 WB_END
 
-WB_ENTRY(jboolean, WB_NMTIsDetailSupported(JNIEnv* env))
-  return MemTracker::tracking_level() == MemTracker::NMT_detail;
-WB_END
 
 #endif // INCLUDE_NMT
 
@@ -861,12 +865,13 @@ static JNINativeMethod methods[] = {
 #endif // INCLUDE_ALL_GCS
 #if INCLUDE_NMT
   {CC"NMTMalloc",           CC"(J)J",                 (void*)&WB_NMTMalloc          },
+  {CC"NMTMallocWithPseudoStack", CC"(JI)J",           (void*)&WB_NMTMallocWithPseudoStack},
   {CC"NMTFree",             CC"(J)V",                 (void*)&WB_NMTFree            },
   {CC"NMTReserveMemory",    CC"(J)J",                 (void*)&WB_NMTReserveMemory   },
   {CC"NMTCommitMemory",     CC"(JJ)V",                (void*)&WB_NMTCommitMemory    },
   {CC"NMTUncommitMemory",   CC"(JJ)V",                (void*)&WB_NMTUncommitMemory  },
   {CC"NMTReleaseMemory",    CC"(JJ)V",                (void*)&WB_NMTReleaseMemory   },
-  {CC"NMTWaitForDataMerge", CC"()Z",                  (void*)&WB_NMTWaitForDataMerge},
+  {CC"NMTOverflowHashBucket", CC"(J)V",               (void*)&WB_NMTOverflowHashBucket},
   {CC"NMTIsDetailSupported",CC"()Z",                  (void*)&WB_NMTIsDetailSupported},
 #endif // INCLUDE_NMT
   {CC"deoptimizeAll",      CC"()V",                   (void*)&WB_DeoptimizeAll     },

src/share/vm/runtime/arguments.cpp

@@ -294,6 +294,7 @@ static ObsoleteFlag obsolete_jvm_flags[] = {
   { "UseMPSS",                       JDK_Version::jdk(8), JDK_Version::jdk(9) },
   { "UseStringCache",                JDK_Version::jdk(8), JDK_Version::jdk(9) },
   { "UseOldInlining",                JDK_Version::jdk(9), JDK_Version::jdk(10) },
+  { "AutoShutdownNMT",               JDK_Version::jdk(9), JDK_Version::jdk(10) },
 #ifdef PRODUCT
   { "DesiredMethodLimit",
                            JDK_Version::jdk_update(7, 2), JDK_Version::jdk(8) },
@@ -2343,7 +2344,7 @@ bool Arguments::check_vm_args_consistency() {
 
   if (PrintNMTStatistics) {
 #if INCLUDE_NMT
-    if (MemTracker::tracking_level() == MemTracker::NMT_off) {
+    if (MemTracker::tracking_level() == NMT_off) {
 #endif // INCLUDE_NMT
       warning("PrintNMTStatistics is disabled, because native memory tracking is not enabled");
       PrintNMTStatistics = false;
@@ -3533,15 +3534,24 @@ jint Arguments::parse(const JavaVMInitArgs* args) {
       CommandLineFlags::printFlags(tty, false);
       vm_exit(0);
     }
-    if (match_option(option, "-XX:NativeMemoryTracking", &tail)) {
 #if INCLUDE_NMT
-      MemTracker::init_tracking_options(tail);
-#else
-      jio_fprintf(defaultStream::error_stream(),
-        "Native Memory Tracking is not supported in this VM\n");
-      return JNI_ERR;
-#endif
+    if (match_option(option, "-XX:NativeMemoryTracking", &tail)) {
+      // The launcher did not setup nmt environment variable properly.
+//      if (!MemTracker::check_launcher_nmt_support(tail)) {
+//        warning("Native Memory Tracking did not setup properly, using wrong launcher?");
+//      }
+
+      // Verify if nmt option is valid.
+      if (MemTracker::verify_nmt_option()) {
+        // Late initialization, still in single-threaded mode.
+        if (MemTracker::tracking_level() >= NMT_summary) {
+          MemTracker::init();
+        }
+      } else {
+        vm_exit_during_initialization("Syntax error, expecting -XX:NativeMemoryTracking=[off|summary|detail]", NULL);
+      }
     }
+#endif
 
 
 #ifndef PRODUCT

src/share/vm/runtime/globals.hpp

@@ -943,11 +943,6 @@ class CommandLineFlags {
   diagnostic(bool, PrintNMTStatistics, false,                               \
           "Print native memory tracking summary data if it is on")          \
                                                                             \
-  diagnostic(bool, AutoShutdownNMT, true,                                   \
-          "Automatically shutdown native memory tracking under stress "     \
-          "situations. When set to false, native memory tracking tries to " \
-          "stay alive at the expense of JVM performance")                   \
-                                                                            \
   diagnostic(bool, LogCompilation, false,                                   \
           "Log compilation activity in detail to LogFile")                  \
                                                                             \

src/share/vm/runtime/handles.hpp

 /*
- * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -227,7 +227,7 @@ class HandleArea: public Arena {
   HandleArea* _prev;          // link to outer (older) area
  public:
   // Constructor
-  HandleArea(HandleArea* prev) : Arena(Chunk::tiny_size) {
+  HandleArea(HandleArea* prev) : Arena(mtThread, Chunk::tiny_size) {
     debug_only(_handle_mark_nesting    = 0);
     debug_only(_no_handle_mark_nesting = 0);
     _prev = prev;

src/share/vm/runtime/init.cpp

 /*
- * Copyright (c) 1997, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -34,8 +34,10 @@
 #include "runtime/init.hpp"
 #include "runtime/safepoint.hpp"
 #include "runtime/sharedRuntime.hpp"
+#include "services/memTracker.hpp"
 #include "utilities/macros.hpp"
 
+
 // Initialization done by VM thread in vm_init_globals()
 void check_ThreadShadow();
 void eventlog_init();
@@ -131,6 +133,12 @@ jint init_globals() {
   javaClasses_init();   // must happen after vtable initialization
   stubRoutines_init2(); // note: StubRoutines need 2-phase init
 
+#if INCLUDE_NMT
+  // Solaris stack is walkable only after stubRoutines are set up.
+  // On Other platforms, the stack is always walkable.
+  NMT_stack_walkable = true;
+#endif // INCLUDE_NMT
+
   // All the flags that get adjusted by VM_Version_init and os::init_2
   // have been set so dump the flags now.
   if (PrintFlagsFinal) {

src/share/vm/runtime/java.cpp

@@ -57,7 +57,6 @@
 #include "runtime/thread.inline.hpp"
 #include "runtime/timer.hpp"
 #include "runtime/vm_operations.hpp"
-#include "services/memReporter.hpp"
 #include "services/memTracker.hpp"
 #include "trace/tracing.hpp"
 #include "utilities/dtrace.hpp"
@@ -364,12 +363,7 @@ void print_statistics() {
 #endif // ENABLE_ZAP_DEAD_LOCALS
   // Native memory tracking data
   if (PrintNMTStatistics) {
-    if (MemTracker::is_on()) {
-      BaselineTTYOutputer outputer(tty);
-      MemTracker::print_memory_usage(outputer, K, false);
-    } else {
-      tty->print_cr("%s", MemTracker::reason());
-    }
+    MemTracker::final_report(tty);
   }
 }
 
@@ -401,12 +395,7 @@ void print_statistics() {
 
   // Native memory tracking data
   if (PrintNMTStatistics) {
-    if (MemTracker::is_on()) {
-      BaselineTTYOutputer outputer(tty);
-      MemTracker::print_memory_usage(outputer, K, false);
-    } else {
-      tty->print_cr("%s", MemTracker::reason());
-    }
+    MemTracker::final_report(tty);
   }
 }
 
@@ -555,10 +544,6 @@ void before_exit(JavaThread * thread) {
     BeforeExit_lock->notify_all();
   }
 
-  // Shutdown NMT before exit. Otherwise,
-  // it will run into trouble when system destroys static variables.
-  MemTracker::shutdown(MemTracker::NMT_normal);
-
   if (VerifyStringTableAtExit) {
     int fail_cnt = 0;
     {

src/share/vm/runtime/os.cpp

@@ -49,6 +49,7 @@
 #include "runtime/stubRoutines.hpp"
 #include "runtime/thread.inline.hpp"
 #include "services/attachListener.hpp"
+#include "services/nmtCommon.hpp"
 #include "services/memTracker.hpp"
 #include "services/threadService.hpp"
 #include "utilities/defaultStream.hpp"
@@ -561,7 +562,11 @@ static u_char* testMalloc(size_t alloc_size) {
   return ptr;
 }
 
-void* os::malloc(size_t size, MEMFLAGS memflags, address caller) {
+void* os::malloc(size_t size, MEMFLAGS flags) {
+  return os::malloc(size, flags, CALLER_PC);
+}
+
+void* os::malloc(size_t size, MEMFLAGS memflags, const NativeCallStack& stack) {
   NOT_PRODUCT(inc_stat_counter(&num_mallocs, 1));
   NOT_PRODUCT(inc_stat_counter(&alloc_bytes, size));
 
@@ -587,11 +592,15 @@ void* os::malloc(size_t size, MEMFLAGS memflags, address caller) {
     size = 1;
   }
 
+  // NMT support
+  NMT_TrackingLevel level = MemTracker::tracking_level();
+  size_t            nmt_header_size = MemTracker::malloc_header_size(level);
+
 #ifndef ASSERT
-  const size_t alloc_size = size;
+  const size_t alloc_size = size + nmt_header_size;
 #else
-  const size_t alloc_size = GuardedMemory::get_total_size(size);
-  if (size > alloc_size) { // Check for rollover.
+  const size_t alloc_size = GuardedMemory::get_total_size(size + nmt_header_size);
+  if (size + nmt_header_size > alloc_size) { // Check for rollover.
     return NULL;
   }
 #endif
@@ -610,7 +619,7 @@ void* os::malloc(size_t size, MEMFLAGS memflags, address caller) {
     return NULL;
   }
   // Wrap memory with guard
-  GuardedMemory guarded(ptr, size);
+  GuardedMemory guarded(ptr, size + nmt_header_size);
   ptr = guarded.get_user_ptr();
 #endif
   if ((intptr_t)ptr == (intptr_t)MallocCatchPtr) {
@@ -623,48 +632,50 @@ void* os::malloc(size_t size, MEMFLAGS memflags, address caller) {
   }
 
   // we do not track guard memory
-  MemTracker::record_malloc((address)ptr, size, memflags, caller == 0 ? CALLER_PC : caller);
-
-  return ptr;
+  return MemTracker::record_malloc((address)ptr, size, memflags, stack, level);
 }
 
+void* os::realloc(void *memblock, size_t size, MEMFLAGS flags) {
+  return os::realloc(memblock, size, flags, CALLER_PC);
+}
 
-void* os::realloc(void *memblock, size_t size, MEMFLAGS memflags, address caller) {
+void* os::realloc(void *memblock, size_t size, MEMFLAGS memflags, const NativeCallStack& stack) {
 #ifndef ASSERT
   NOT_PRODUCT(inc_stat_counter(&num_mallocs, 1));
   NOT_PRODUCT(inc_stat_counter(&alloc_bytes, size));
-  MemTracker::Tracker tkr = MemTracker::get_realloc_tracker();
-  void* ptr = ::realloc(memblock, size);
-  if (ptr != NULL) {
-    tkr.record((address)memblock, (address)ptr, size, memflags,
-     caller == 0 ? CALLER_PC : caller);
-  } else {
-    tkr.discard();
-  }
-  return ptr;
+   // NMT support
+  void* membase = MemTracker::record_free(memblock);
+  NMT_TrackingLevel level = MemTracker::tracking_level();
+  size_t  nmt_header_size = MemTracker::malloc_header_size(level);
+  void* ptr = ::realloc(membase, size + nmt_header_size);
+  return MemTracker::record_malloc(ptr, size, memflags, stack, level);
 #else
   if (memblock == NULL) {
-    return os::malloc(size, memflags, (caller == 0 ? CALLER_PC : caller));
+    return os::malloc(size, memflags, stack);
   }
   if ((intptr_t)memblock == (intptr_t)MallocCatchPtr) {
     tty->print_cr("os::realloc caught " PTR_FORMAT, memblock);
     breakpoint();
   }
-  verify_memory(memblock);
+  // NMT support
+  void* membase = MemTracker::malloc_base(memblock);
+  verify_memory(membase);
   NOT_PRODUCT(if (MallocVerifyInterval > 0) check_heap());
   if (size == 0) {
     return NULL;
   }
   // always move the block
-  void* ptr = os::malloc(size, memflags, caller == 0 ? CALLER_PC : caller);
+  void* ptr = os::malloc(size, memflags, stack);
   if (PrintMalloc) {
     tty->print_cr("os::remalloc " SIZE_FORMAT " bytes, " PTR_FORMAT " --> " PTR_FORMAT, size, memblock, ptr);
   }
   // Copy to new memory if malloc didn't fail
   if ( ptr != NULL ) {
-    GuardedMemory guarded(memblock);
-    memcpy(ptr, memblock, MIN2(size, guarded.get_user_size()));
-    if (paranoid) verify_memory(ptr);
+    GuardedMemory guarded(MemTracker::malloc_base(memblock));
+    // Guard's user data contains NMT header
+    size_t memblock_size = guarded.get_user_size() - MemTracker::malloc_header_size(memblock);
+    memcpy(ptr, memblock, MIN2(size, memblock_size));
+    if (paranoid) verify_memory(MemTracker::malloc_base(ptr));
     if ((intptr_t)ptr == (intptr_t)MallocCatchPtr) {
       tty->print_cr("os::realloc caught, " SIZE_FORMAT " bytes --> " PTR_FORMAT, size, ptr);
       breakpoint();
@@ -677,7 +688,6 @@ void* os::realloc(void *memblock, size_t size, MEMFLAGS memflags, address caller
 
 
 void  os::free(void *memblock, MEMFLAGS memflags) {
-  address trackp = (address) memblock;
   NOT_PRODUCT(inc_stat_counter(&num_frees, 1));
 #ifdef ASSERT
   if (memblock == NULL) return;
@@ -685,20 +695,22 @@ void  os::free(void *memblock, MEMFLAGS memflags) {
     if (tty != NULL) tty->print_cr("os::free caught " PTR_FORMAT, memblock);
     breakpoint();
   }
-  verify_memory(memblock);
+  void* membase = MemTracker::record_free(memblock);
+  verify_memory(membase);
   NOT_PRODUCT(if (MallocVerifyInterval > 0) check_heap());
 
-  GuardedMemory guarded(memblock);
+  GuardedMemory guarded(membase);
   size_t size = guarded.get_user_size();
   inc_stat_counter(&free_bytes, size);
-  memblock = guarded.release_for_freeing();
+  membase = guarded.release_for_freeing();
   if (PrintMalloc && tty != NULL) {
-      fprintf(stderr, "os::free " SIZE_FORMAT " bytes --> " PTR_FORMAT "\n", size, (uintptr_t)memblock);
+      fprintf(stderr, "os::free " SIZE_FORMAT " bytes --> " PTR_FORMAT "\n", size, (uintptr_t)membase);
   }
+  ::free(membase);
+#else
+  void* membase = MemTracker::record_free(memblock);
+  ::free(membase);
 #endif
-  MemTracker::record_free(trackp, memflags);
-
-  ::free(memblock);
 }
 
 void os::init_random(long initval) {
@@ -1412,7 +1424,7 @@ bool os::create_stack_guard_pages(char* addr, size_t bytes) {
 char* os::reserve_memory(size_t bytes, char* addr, size_t alignment_hint) {
   char* result = pd_reserve_memory(bytes, addr, alignment_hint);
   if (result != NULL) {
-    MemTracker::record_virtual_memory_reserve((address)result, bytes, mtNone, CALLER_PC);
+    MemTracker::record_virtual_memory_reserve((address)result, bytes, CALLER_PC);
   }
 
   return result;
@@ -1422,7 +1434,7 @@ char* os::reserve_memory(size_t bytes, char* addr, size_t alignment_hint,
    MEMFLAGS flags) {
   char* result = pd_reserve_memory(bytes, addr, alignment_hint);
   if (result != NULL) {
-    MemTracker::record_virtual_memory_reserve((address)result, bytes, mtNone, CALLER_PC);
+    MemTracker::record_virtual_memory_reserve((address)result, bytes, CALLER_PC);
     MemTracker::record_virtual_memory_type((address)result, flags);
   }
 
@@ -1432,7 +1444,7 @@ char* os::reserve_memory(size_t bytes, char* addr, size_t alignment_hint,
 char* os::attempt_reserve_memory_at(size_t bytes, char* addr) {
   char* result = pd_attempt_reserve_memory_at(bytes, addr);
   if (result != NULL) {
-    MemTracker::record_virtual_memory_reserve((address)result, bytes, mtNone, CALLER_PC);
+    MemTracker::record_virtual_memory_reserve((address)result, bytes, CALLER_PC);
   }
   return result;
 }
@@ -1472,23 +1484,29 @@ void os::commit_memory_or_exit(char* addr, size_t size, size_t alignment_hint,
 }
 
 bool os::uncommit_memory(char* addr, size_t bytes) {
-  MemTracker::Tracker tkr = MemTracker::get_virtual_memory_uncommit_tracker();
-  bool res = pd_uncommit_memory(addr, bytes);
-  if (res) {
-    tkr.record((address)addr, bytes);
+  bool res;
+  if (MemTracker::tracking_level() > NMT_minimal) {
+    Tracker tkr = MemTracker::get_virtual_memory_uncommit_tracker();
+    res = pd_uncommit_memory(addr, bytes);
+    if (res) {
+      tkr.record((address)addr, bytes);
+    }
   } else {
-    tkr.discard();
+    res = pd_uncommit_memory(addr, bytes);
   }
   return res;
 }
 
 bool os::release_memory(char* addr, size_t bytes) {
-  MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
-  bool res = pd_release_memory(addr, bytes);
-  if (res) {
-    tkr.record((address)addr, bytes);
+  bool res;
+  if (MemTracker::tracking_level() > NMT_minimal) {
+    Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
+    res = pd_release_memory(addr, bytes);
+    if (res) {
+      tkr.record((address)addr, bytes);
+    }
   } else {
-    tkr.discard();
+    res = pd_release_memory(addr, bytes);
   }
   return res;
 }
@@ -1499,7 +1517,7 @@ char* os::map_memory(int fd, const char* file_name, size_t file_offset,
                            bool allow_exec) {
   char* result = pd_map_memory(fd, file_name, file_offset, addr, bytes, read_only, allow_exec);
   if (result != NULL) {
-    MemTracker::record_virtual_memory_reserve_and_commit((address)result, bytes, mtNone, CALLER_PC);
+    MemTracker::record_virtual_memory_reserve_and_commit((address)result, bytes, CALLER_PC);
   }
   return result;
 }
@@ -1512,12 +1530,15 @@ char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
 }
 
 bool os::unmap_memory(char *addr, size_t bytes) {
-  MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
-  bool result = pd_unmap_memory(addr, bytes);
-  if (result) {
-    tkr.record((address)addr, bytes);
+  bool result;
+  if (MemTracker::tracking_level() > NMT_minimal) {
+    Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
+    result = pd_unmap_memory(addr, bytes);
+    if (result) {
+      tkr.record((address)addr, bytes);
+    }
   } else {
-    tkr.discard();
+    result = pd_unmap_memory(addr, bytes);
   }
   return result;
 }

src/share/vm/runtime/os.hpp

@@ -66,6 +66,8 @@ class JavaThread;
 class Event;
 class DLL;
 class FileHandle;
+class NativeCallStack;
+
 template<class E> class GrowableArray;
 
 // %%%%% Moved ThreadState, START_FN, OSThread to new osThread.hpp. -- Rose
@@ -97,9 +99,11 @@ const bool ExecMem = true;
 // Typedef for structured exception handling support
 typedef void (*java_call_t)(JavaValue* value, methodHandle* method, JavaCallArguments* args, Thread* thread);
 
+class MallocTracker;
+
 class os: AllStatic {
   friend class VMStructs;
-
+  friend class MallocTracker;
  public:
   enum { page_sizes_max = 9 }; // Size of _page_sizes array (8 plus a sentinel)
 
@@ -161,7 +165,10 @@ class os: AllStatic {
   // Override me as needed
   static int    file_name_strcmp(const char* s1, const char* s2);
 
+  // get/unset environment variable
   static bool getenv(const char* name, char* buffer, int len);
+  static bool unsetenv(const char* name);
+
   static bool have_special_privileges();
 
   static jlong  javaTimeMillis();
@@ -207,8 +214,13 @@ class os: AllStatic {
 
   // Interface for detecting multiprocessor system
   static inline bool is_MP() {
+#if !INCLUDE_NMT
     assert(_processor_count > 0, "invalid processor count");
     return _processor_count > 1 || AssumeMP;
+#else
+    // NMT needs atomic operations before this initialization.
+    return true;
+#endif
   }
   static julong available_memory();
   static julong physical_memory();
@@ -651,12 +663,20 @@ class os: AllStatic {
   static void* thread_local_storage_at(int index);
   static void  free_thread_local_storage(int index);
 
-  // Stack walk
-  static address get_caller_pc(int n = 0);
+  // Retrieve native stack frames.
+  // Parameter:
+  //   stack:  an array to storage stack pointers.
+  //   frames: size of above array.
+  //   toSkip: number of stack frames to skip at the beginning.
+  // Return: number of stack frames captured.
+  static int get_native_stack(address* stack, int size, int toSkip = 0);
 
   // General allocation (must be MT-safe)
-  static void* malloc  (size_t size, MEMFLAGS flags, address caller_pc = 0);
-  static void* realloc (void *memblock, size_t size, MEMFLAGS flags, address caller_pc = 0);
+  static void* malloc  (size_t size, MEMFLAGS flags, const NativeCallStack& stack);
+  static void* malloc  (size_t size, MEMFLAGS flags);
+  static void* realloc (void *memblock, size_t size, MEMFLAGS flag, const NativeCallStack& stack);
+  static void* realloc (void *memblock, size_t size, MEMFLAGS flag);
+
   static void  free    (void *memblock, MEMFLAGS flags = mtNone);
   static bool  check_heap(bool force = false);      // verify C heap integrity
   static char* strdup(const char *, MEMFLAGS flags = mtInternal);  // Like strdup

src/share/vm/runtime/safepoint.cpp

@@ -50,7 +50,6 @@
 #include "runtime/sweeper.hpp"
 #include "runtime/synchronizer.hpp"
 #include "runtime/thread.inline.hpp"
-#include "services/memTracker.hpp"
 #include "services/runtimeService.hpp"
 #include "utilities/events.hpp"
 #include "utilities/macros.hpp"
@@ -547,10 +546,6 @@ void SafepointSynchronize::do_cleanup_tasks() {
     TraceTime t7("purging class loader data graph", TraceSafepointCleanupTime);
     ClassLoaderDataGraph::purge_if_needed();
   }
-
-  if (MemTracker::is_on()) {
-    MemTracker::sync();
-  }
 }
 
 

src/share/vm/runtime/thread.cpp

@@ -331,8 +331,7 @@ void Thread::record_stack_base_and_size() {
 #if INCLUDE_NMT
   // record thread's native stack, stack grows downward
   address stack_low_addr = stack_base() - stack_size();
-  MemTracker::record_thread_stack(stack_low_addr, stack_size(), this,
-      CURRENT_PC);
+  MemTracker::record_thread_stack(stack_low_addr, stack_size());
 #endif // INCLUDE_NMT
 }
 
@@ -350,7 +349,7 @@ Thread::~Thread() {
 #if INCLUDE_NMT
   if (_stack_base != NULL) {
     address low_stack_addr = stack_base() - stack_size();
-    MemTracker::release_thread_stack(low_stack_addr, stack_size(), this);
+    MemTracker::release_thread_stack(low_stack_addr, stack_size());
 #ifdef ASSERT
     set_stack_base(NULL);
 #endif
@@ -1442,9 +1441,6 @@ void JavaThread::initialize() {
   set_monitor_chunks(NULL);
   set_next(NULL);
   set_thread_state(_thread_new);
-#if INCLUDE_NMT
-  set_recorder(NULL);
-#endif
   _terminated = _not_terminated;
   _privileged_stack_top = NULL;
   _array_for_gc = NULL;
@@ -1519,7 +1515,6 @@ JavaThread::JavaThread(bool is_attaching_via_jni) :
     _jni_attach_state = _not_attaching_via_jni;
   }
   assert(deferred_card_mark().is_empty(), "Default MemRegion ctor");
-  _safepoint_visible = false;
 }
 
 bool JavaThread::reguard_stack(address cur_sp) {
@@ -1582,7 +1577,6 @@ JavaThread::JavaThread(ThreadFunction entry_point, size_t stack_sz) :
   thr_type = entry_point == &compiler_thread_entry ? os::compiler_thread :
                                                      os::java_thread;
   os::create_thread(this, thr_type, stack_sz);
-  _safepoint_visible = false;
   // The _osthread may be NULL here because we ran out of memory (too many threads active).
   // We need to throw and OutOfMemoryError - however we cannot do this here because the caller
   // may hold a lock and all locks must be unlocked before throwing the exception (throwing
@@ -1600,13 +1594,6 @@ JavaThread::~JavaThread() {
       tty->print_cr("terminate thread %p", this);
   }
 
-  // By now, this thread should already be invisible to safepoint,
-  // and its per-thread recorder also collected.
-  assert(!is_safepoint_visible(), "wrong state");
-#if INCLUDE_NMT
-  assert(get_recorder() == NULL, "Already collected");
-#endif // INCLUDE_NMT
-
   // JSR166 -- return the parker to the free list
   Parker::Release(_parker);
   _parker = NULL ;
@@ -3370,11 +3357,6 @@ jint Threads::create_vm(JavaVMInitArgs* args, bool* canTryAgain) {
   // intialize TLS
   ThreadLocalStorage::init();
 
-  // Bootstrap native memory tracking, so it can start recording memory
-  // activities before worker thread is started. This is the first phase
-  // of bootstrapping, VM is currently running in single-thread mode.
-  MemTracker::bootstrap_single_thread();
-
   // Initialize output stream logging
   ostream_init_log();
 
@@ -3425,9 +3407,6 @@ jint Threads::create_vm(JavaVMInitArgs* args, bool* canTryAgain) {
   // Initialize Java-Level synchronization subsystem
   ObjectMonitor::Initialize() ;
 
-  // Second phase of bootstrapping, VM is about entering multi-thread mode
-  MemTracker::bootstrap_multi_thread();
-
   // Initialize global modules
   jint status = init_globals();
   if (status != JNI_OK) {
@@ -3449,9 +3428,6 @@ jint Threads::create_vm(JavaVMInitArgs* args, bool* canTryAgain) {
   // real raw monitor. VM is setup enough here for raw monitor enter.
   JvmtiExport::transition_pending_onload_raw_monitors();
 
-  // Fully start NMT
-  MemTracker::start();
-
   // Create the VMThread
   { TraceTime timer("Start VMThread", TraceStartupTime);
     VMThread::create();
@@ -4089,8 +4065,6 @@ void Threads::add(JavaThread* p, bool force_daemon) {
     daemon = false;
   }
 
-  p->set_safepoint_visible(true);
-
   ThreadService::add_thread(p, daemon);
 
   // Possible GC point.
@@ -4136,13 +4110,6 @@ void Threads::remove(JavaThread* p) {
     // to do callbacks into the safepoint code. However, the safepoint code is not aware
     // of this thread since it is removed from the queue.
     p->set_terminated_value();
-
-    // Now, this thread is not visible to safepoint
-    p->set_safepoint_visible(false);
-    // once the thread becomes safepoint invisible, we can not use its per-thread
-    // recorder. And Threads::do_threads() no longer walks this thread, so we have
-    // to release its per-thread recorder here.
-    MemTracker::thread_exiting(p);
   } // unlock Threads_lock
 
   // Since Events::log uses a lock, we grab it outside the Threads_lock

src/share/vm/runtime/thread.hpp

@@ -43,10 +43,6 @@
 #include "runtime/unhandledOops.hpp"
 #include "utilities/macros.hpp"
 
-#if INCLUDE_NMT
-#include "services/memRecorder.hpp"
-#endif // INCLUDE_NMT
-
 #include "trace/traceBackend.hpp"
 #include "trace/traceMacros.hpp"
 #include "utilities/exceptions.hpp"
@@ -1059,16 +1055,6 @@ class JavaThread: public Thread {
   bool do_not_unlock_if_synchronized()             { return _do_not_unlock_if_synchronized; }
   void set_do_not_unlock_if_synchronized(bool val) { _do_not_unlock_if_synchronized = val; }
 
-#if INCLUDE_NMT
-  // native memory tracking
-  inline MemRecorder* get_recorder() const          { return (MemRecorder*)_recorder; }
-  inline void         set_recorder(MemRecorder* rc) { _recorder = rc; }
-
- private:
-  // per-thread memory recorder
-  MemRecorder* volatile _recorder;
-#endif // INCLUDE_NMT
-
   // Suspend/resume support for JavaThread
  private:
   void set_ext_suspended()       { set_suspend_flag (_ext_suspended);  }
@@ -1511,19 +1497,6 @@ public:
      return result;
    }
 
- // NMT (Native memory tracking) support.
- // This flag helps NMT to determine if this JavaThread will be blocked
- // at safepoint. If not, ThreadCritical is needed for writing memory records.
- // JavaThread is only safepoint visible when it is in Threads' thread list,
- // it is not visible until it is added to the list and becomes invisible
- // once it is removed from the list.
- public:
-  bool is_safepoint_visible() const { return _safepoint_visible; }
-  void set_safepoint_visible(bool visible) { _safepoint_visible = visible; }
- private:
-  bool _safepoint_visible;
-
-  // Static operations
  public:
   // Returns the running thread as a JavaThread
   static inline JavaThread* current();

src/share/vm/services/allocationSite.hpp

+/*
+ * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code 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 General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_VM_SERVICES_ALLOCATION_SITE_HPP
+#define SHARE_VM_SERVICES_ALLOCATION_SITE_HPP
+
+#include "memory/allocation.hpp"
+#include "utilities/nativeCallStack.hpp"
+
+// Allocation site represents a code path that makes a memory
+// allocation
+template <class E> class AllocationSite VALUE_OBJ_CLASS_SPEC {
+ private:
+  NativeCallStack  _call_stack;
+  E                e;
+ public:
+  AllocationSite(const NativeCallStack& stack) : _call_stack(stack) { }
+  int hash() const { return _call_stack.hash(); }
+  bool equals(const NativeCallStack& stack) const {
+    return _call_stack.equals(stack);
+  }
+
+  bool equals(const AllocationSite<E>& other) const {
+    return other.equals(_call_stack);
+  }
+
+  const NativeCallStack* call_stack() const {
+    return &_call_stack;
+  }
+
+  // Information regarding this allocation
+  E* data()             { return &e; }
+  const E* peek() const { return &e; }
+};
+
+#endif  // SHARE_VM_SERVICES_ALLOCATION_SITE_HPP

src/share/vm/services/mallocSiteTable.cpp

+/*
+ * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code 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 General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+#include "precompiled.hpp"
+
+
+#include "memory/allocation.inline.hpp"
+#include "runtime/atomic.hpp"
+#include "services/mallocSiteTable.hpp"
+
+/*
+ * Early os::malloc() calls come from initializations of static variables, long before entering any
+ * VM code. Upon the arrival of the first os::malloc() call, malloc site hashtable has to be
+ * initialized, along with the allocation site for the hashtable entries.
+ * To ensure that malloc site hashtable can be initialized without triggering any additional os::malloc()
+ * call, the hashtable bucket array and hashtable entry allocation site have to be static.
+ * It is not a problem for hashtable bucket, since it is an array of pointer type, C runtime just
+ * allocates a block memory and zero the memory for it.
+ * But for hashtable entry allocation site object, things get tricky. C runtime not only allocates
+ * memory for it, but also calls its constructor at some later time. If we initialize the allocation site
+ * at the first os::malloc() call, the object will be reinitialized when its constructor is called
+ * by C runtime.
+ * To workaround above issue, we declare a static size_t array with the size of the CallsiteHashtableEntry,
+ * the memory is used to instantiate CallsiteHashtableEntry for the hashtable entry allocation site.
+ * Given it is a primitive type array, C runtime will do nothing other than assign the memory block for the variable,
+ * which is exactly what we want.
+ * The same trick is also applied to create NativeCallStack object for CallsiteHashtableEntry memory allocation.
+ *
+ * Note: C++ object usually aligns to particular alignment, depends on compiler implementation, we declare
+ * the memory as size_t arrays, to ensure the memory is aligned to native machine word alignment.
+ */
+
+// Reserve enough memory for NativeCallStack and MallocSiteHashtableEntry objects
+size_t MallocSiteTable::_hash_entry_allocation_stack[CALC_OBJ_SIZE_IN_TYPE(NativeCallStack, size_t)];
+size_t MallocSiteTable::_hash_entry_allocation_site[CALC_OBJ_SIZE_IN_TYPE(MallocSiteHashtableEntry, size_t)];
+
+// Malloc site hashtable buckets
+MallocSiteHashtableEntry*  MallocSiteTable::_table[MallocSiteTable::table_size];
+
+// concurrent access counter
+volatile int MallocSiteTable::_access_count = 0;
+
+// Tracking hashtable contention
+NOT_PRODUCT(int MallocSiteTable::_peak_count = 0;)
+
+
+/*
+ * Initialize malloc site table.
+ * Hashtable entry is malloc'd, so it can cause infinite recursion.
+ * To avoid above problem, we pre-initialize a hash entry for
+ * this allocation site.
+ * The method is called during C runtime static variable initialization
+ * time, it is in single-threaded mode from JVM perspective.
+ */
+bool MallocSiteTable::initialize() {
+  assert(sizeof(_hash_entry_allocation_stack) >= sizeof(NativeCallStack), "Sanity Check");
+  assert(sizeof(_hash_entry_allocation_site) >= sizeof(MallocSiteHashtableEntry),
+    "Sanity Check");
+  assert((size_t)table_size <= MAX_MALLOCSITE_TABLE_SIZE, "Hashtable overflow");
+
+  // Fake the call stack for hashtable entry allocation
+  assert(NMT_TrackingStackDepth > 1, "At least one tracking stack");
+
+  // Create pseudo call stack for hashtable entry allocation
+  address pc[3];
+  if (NMT_TrackingStackDepth >= 3) {
+    pc[2] = (address)MallocSiteTable::allocation_at;
+  }
+  if (NMT_TrackingStackDepth >= 2) {
+    pc[1] = (address)MallocSiteTable::lookup_or_add;
+  }
+  pc[0] = (address)MallocSiteTable::new_entry;
+
+  // Instantiate NativeCallStack object, have to use placement new operator. (see comments above)
+  NativeCallStack* stack = ::new ((void*)_hash_entry_allocation_stack)
+    NativeCallStack(pc, MIN2(((int)(sizeof(pc) / sizeof(address))), ((int)NMT_TrackingStackDepth)));
+
+  // Instantiate hash entry for hashtable entry allocation callsite
+  MallocSiteHashtableEntry* entry = ::new ((void*)_hash_entry_allocation_site)
+    MallocSiteHashtableEntry(*stack);
+
+  // Add the allocation site to hashtable.
+  int index = hash_to_index(stack->hash());
+  _table[index] = entry;
+
+  return true;
+}
+
+// Walks entries in the hashtable.
+// It stops walk if the walker returns false.
+bool MallocSiteTable::walk(MallocSiteWalker* walker) {
+  MallocSiteHashtableEntry* head;
+  for (int index = 0; index < table_size; index ++) {
+    head = _table[index];
+    while (head != NULL) {
+      if (!walker->do_malloc_site(head->peek())) {
+        return false;
+      }
+      head = (MallocSiteHashtableEntry*)head->next();
+    }
+  }
+  return true;
+}
+
+/*
+ *  The hashtable does not have deletion policy on individual entry,
+ *  and each linked list node is inserted via compare-and-swap,
+ *  so each linked list is stable, the contention only happens
+ *  at the end of linked list.
+ *  This method should not return NULL under normal circumstance.
+ *  If NULL is returned, it indicates:
+ *    1. Out of memory, it cannot allocate new hash entry.
+ *    2. Overflow hash bucket.
+ *  Under any of above circumstances, caller should handle the situation.
+ */
+MallocSite* MallocSiteTable::lookup_or_add(const NativeCallStack& key, size_t* bucket_idx,
+  size_t* pos_idx) {
+  int index = hash_to_index(key.hash());
+  assert(index >= 0, "Negative index");
+  *bucket_idx = (size_t)index;
+  *pos_idx = 0;
+
+  // First entry for this hash bucket
+  if (_table[index] == NULL) {
+    MallocSiteHashtableEntry* entry = new_entry(key);
+    // OOM check
+    if (entry == NULL) return NULL;
+
+    // swap in the head
+    if (Atomic::cmpxchg_ptr((void*)entry, (volatile void *)&_table[index], NULL) == NULL) {
+      return entry->data();
+    }
+
+    delete entry;
+  }
+
+  MallocSiteHashtableEntry* head = _table[index];
+  while (head != NULL && (*pos_idx) <= MAX_BUCKET_LENGTH) {
+    MallocSite* site = head->data();
+    if (site->equals(key)) {
+      // found matched entry
+      return head->data();
+    }
+
+    if (head->next() == NULL && (*pos_idx) < MAX_BUCKET_LENGTH) {
+      MallocSiteHashtableEntry* entry = new_entry(key);
+      // OOM check
+      if (entry == NULL) return NULL;
+      if (head->atomic_insert(entry)) {
+        (*pos_idx) ++;
+        return entry->data();
+      }
+      // contended, other thread won
+      delete entry;
+    }
+    head = (MallocSiteHashtableEntry*)head->next();
+    (*pos_idx) ++;
+  }
+  return NULL;
+}
+
+// Access malloc site
+MallocSite* MallocSiteTable::malloc_site(size_t bucket_idx, size_t pos_idx) {
+  assert(bucket_idx < table_size, "Invalid bucket index");
+  MallocSiteHashtableEntry* head = _table[bucket_idx];
+  for (size_t index = 0; index < pos_idx && head != NULL;
+    index ++, head = (MallocSiteHashtableEntry*)head->next());
+  assert(head != NULL, "Invalid position index");
+  return head->data();
+}
+
+// Allocates MallocSiteHashtableEntry object. Special call stack
+// (pre-installed allocation site) has to be used to avoid infinite
+// recursion.
+MallocSiteHashtableEntry* MallocSiteTable::new_entry(const NativeCallStack& key) {
+  void* p = AllocateHeap(sizeof(MallocSiteHashtableEntry), mtNMT,
+    *hash_entry_allocation_stack(), AllocFailStrategy::RETURN_NULL);
+  return ::new (p) MallocSiteHashtableEntry(key);
+}
+
+void MallocSiteTable::reset() {
+  for (int index = 0; index < table_size; index ++) {
+    MallocSiteHashtableEntry* head = _table[index];
+    _table[index] = NULL;
+    delete_linked_list(head);
+  }
+}
+
+void MallocSiteTable::delete_linked_list(MallocSiteHashtableEntry* head) {
+  MallocSiteHashtableEntry* p;
+  while (head != NULL) {
+    p = head;
+    head = (MallocSiteHashtableEntry*)head->next();
+    if (p != (MallocSiteHashtableEntry*)_hash_entry_allocation_site) {
+      delete p;
+    }
+  }
+}
+
+void MallocSiteTable::shutdown() {
+  AccessLock locker(&_access_count);
+  locker.exclusiveLock();
+  reset();
+}
+
+bool MallocSiteTable::walk_malloc_site(MallocSiteWalker* walker) {
+  assert(walker != NULL, "NuLL walker");
+  AccessLock locker(&_access_count);
+  if (locker.sharedLock()) {
+    NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);)
+    return walk(walker);
+  }
+  return false;
+}
+
+
+void MallocSiteTable::AccessLock::exclusiveLock() {
+  jint target;
+  jint val;
+
+  assert(_lock_state != ExclusiveLock, "Can only call once");
+  assert(*_lock >= 0, "Can not content exclusive lock");
+
+  // make counter negative to block out shared locks
+  do {
+    val = *_lock;
+    target = _MAGIC_ + *_lock;
+  } while (Atomic::cmpxchg(target, _lock, val) != val);
+
+  // wait for all readers to exit
+  while (*_lock != _MAGIC_) {
+#ifdef _WINDOWS
+    os::naked_short_sleep(1);
+#else
+    os::NakedYield();
+#endif
+  }
+  _lock_state = ExclusiveLock;
+}
+
+

src/share/vm/services/mallocSiteTable.hpp

+/*
+ * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code 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 General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_VM_SERVICES_MALLOC_SITE_TABLE_HPP
+#define SHARE_VM_SERVICES_MALLOC_SITE_TABLE_HPP
+
+#if INCLUDE_NMT
+
+#include "memory/allocation.hpp"
+#include "runtime/atomic.hpp"
+#include "services/allocationSite.hpp"
+#include "services/mallocTracker.hpp"
+#include "services/nmtCommon.hpp"
+
+// MallocSite represents a code path that eventually calls
+// os::malloc() to allocate memory
+class MallocSite : public AllocationSite<MemoryCounter> {
+ public:
+  MallocSite() :
+    AllocationSite<MemoryCounter>(emptyStack) { }
+
+  MallocSite(const NativeCallStack& stack) :
+    AllocationSite<MemoryCounter>(stack) { }
+
+  void allocate(size_t size)      { data()->allocate(size);   }
+  void deallocate(size_t size)    { data()->deallocate(size); }
+
+  // Memory allocated from this code path
+  size_t size()  const { return peek()->size(); }
+  // The number of calls were made
+  size_t count() const { return peek()->count(); }
+};
+
+// Malloc site hashtable entry
+class MallocSiteHashtableEntry : public CHeapObj<mtNMT> {
+ private:
+  MallocSite                _malloc_site;
+  MallocSiteHashtableEntry* _next;
+
+ public:
+  MallocSiteHashtableEntry() : _next(NULL) { }
+
+  MallocSiteHashtableEntry(NativeCallStack stack):
+    _malloc_site(stack), _next(NULL) { }
+
+  inline const MallocSiteHashtableEntry* next() const {
+    return _next;
+  }
+
+  // Insert an entry atomically.
+  // Return true if the entry is inserted successfully.
+  // The operation can be failed due to contention from other thread.
+  bool atomic_insert(const MallocSiteHashtableEntry* entry) {
+    return (Atomic::cmpxchg_ptr((void*)entry, (volatile void*)&_next,
+      NULL) == NULL);
+  }
+
+  void set_callsite(const MallocSite& site) {
+    _malloc_site = site;
+  }
+
+  inline const MallocSite* peek() const { return &_malloc_site; }
+  inline MallocSite* data()             { return &_malloc_site; }
+
+  inline long hash() const { return _malloc_site.hash(); }
+  inline bool equals(const NativeCallStack& stack) const {
+    return _malloc_site.equals(stack);
+  }
+  // Allocation/deallocation on this allocation site
+  inline void allocate(size_t size)   { _malloc_site.allocate(size);   }
+  inline void deallocate(size_t size) { _malloc_site.deallocate(size); }
+  // Memory counters
+  inline size_t size() const  { return _malloc_site.size();  }
+  inline size_t count() const { return _malloc_site.count(); }
+};
+
+// The walker walks every entry on MallocSiteTable
+class MallocSiteWalker : public StackObj {
+ public:
+   virtual bool do_malloc_site(const MallocSite* e) { return false; }
+};
+
+/*
+ * Native memory tracking call site table.
+ * The table is only needed when detail tracking is enabled.
+ */
+class MallocSiteTable : AllStatic {
+ private:
+  // The number of hash bucket in this hashtable. The number should
+  // be tuned if malloc activities changed significantly.
+  // The statistics data can be obtained via Jcmd
+  // jcmd <pid> VM.native_memory statistics.
+
+  // Currently, (number of buckets / number of entires) ratio is
+  // about 1 / 6
+  enum {
+    table_base_size = 128,   // The base size is calculated from statistics to give
+                             // table ratio around 1:6
+    table_size = (table_base_size * NMT_TrackingStackDepth - 1)
+  };
+
+
+  // This is a very special lock, that allows multiple shared accesses (sharedLock), but
+  // once exclusive access (exclusiveLock) is requested, all shared accesses are
+  // rejected forever.
+  class AccessLock : public StackObj {
+    enum LockState {
+      NoLock,
+      SharedLock,
+      ExclusiveLock
+    };
+
+   private:
+    // A very large negative number. The only possibility to "overflow"
+    // this number is when there are more than -min_jint threads in
+    // this process, which is not going to happen in foreseeable future.
+    const static int _MAGIC_ = min_jint;
+
+    LockState      _lock_state;
+    volatile int*  _lock;
+   public:
+    AccessLock(volatile int* lock) :
+      _lock(lock), _lock_state(NoLock) {
+    }
+
+    ~AccessLock() {
+      if (_lock_state == SharedLock) {
+        Atomic::dec((volatile jint*)_lock);
+      }
+    }
+    // Acquire shared lock.
+    // Return true if shared access is granted.
+    inline bool sharedLock() {
+      jint res = Atomic::add(1, _lock);
+      if (res < 0) {
+        Atomic::add(-1, _lock);
+        return false;
+      }
+      _lock_state = SharedLock;
+      return true;
+    }
+    // Acquire exclusive lock
+    void exclusiveLock();
+ };
+
+ public:
+  static bool initialize();
+  static void shutdown();
+
+  NOT_PRODUCT(static int access_peak_count() { return _peak_count; })
+
+  // Number of hash buckets
+  static inline int hash_buckets()      { return (int)table_size; }
+
+  // Access and copy a call stack from this table. Shared lock should be
+  // acquired before access the entry.
+  static inline bool access_stack(NativeCallStack& stack, size_t bucket_idx,
+    size_t pos_idx) {
+    AccessLock locker(&_access_count);
+    if (locker.sharedLock()) {
+      NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);)
+      MallocSite* site = malloc_site(bucket_idx, pos_idx);
+      if (site != NULL) {
+        stack = *site->call_stack();
+        return true;
+      }
+    }
+    return false;
+  }
+
+  // Record a new allocation from specified call path.
+  // Return true if the allocation is recorded successfully, bucket_idx
+  // and pos_idx are also updated to indicate the entry where the allocation
+  // information was recorded.
+  // Return false only occurs under rare scenarios:
+  //  1. out of memory
+  //  2. overflow hash bucket
+  static inline bool allocation_at(const NativeCallStack& stack, size_t size,
+    size_t* bucket_idx, size_t* pos_idx) {
+    AccessLock locker(&_access_count);
+    if (locker.sharedLock()) {
+      NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);)
+      MallocSite* site = lookup_or_add(stack, bucket_idx, pos_idx);
+      if (site != NULL) site->allocate(size);
+      return site != NULL;
+    }
+    return false;
+  }
+
+  // Record memory deallocation. bucket_idx and pos_idx indicate where the allocation
+  // information was recorded.
+  static inline bool deallocation_at(size_t size, size_t bucket_idx, size_t pos_idx) {
+    AccessLock locker(&_access_count);
+    if (locker.sharedLock()) {
+      NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);)
+      MallocSite* site = malloc_site(bucket_idx, pos_idx);
+      if (site != NULL) {
+        site->deallocate(size);
+        return true;
+      }
+    }
+    return false;
+  }
+
+  // Walk this table.
+  static bool walk_malloc_site(MallocSiteWalker* walker);
+
+ private:
+  static MallocSiteHashtableEntry* new_entry(const NativeCallStack& key);
+  static void reset();
+
+  // Delete a bucket linked list
+  static void delete_linked_list(MallocSiteHashtableEntry* head);
+
+  static MallocSite* lookup_or_add(const NativeCallStack& key, size_t* bucket_idx, size_t* pos_idx);
+  static MallocSite* malloc_site(size_t bucket_idx, size_t pos_idx);
+  static bool walk(MallocSiteWalker* walker);
+
+  static inline int hash_to_index(int  hash) {
+    hash = (hash > 0) ? hash : (-hash);
+    return (hash % table_size);
+  }
+
+  static inline const NativeCallStack* hash_entry_allocation_stack() {
+    return (NativeCallStack*)_hash_entry_allocation_stack;
+  }
+
+ private:
+  // Counter for counting concurrent access
+  static volatile int                _access_count;
+
+  // The callsite hashtable. It has to be a static table,
+  // since malloc call can come from C runtime linker.
+  static MallocSiteHashtableEntry*   _table[table_size];
+
+
+  // Reserve enough memory for placing the objects
+
+  // The memory for hashtable entry allocation stack object
+  static size_t _hash_entry_allocation_stack[CALC_OBJ_SIZE_IN_TYPE(NativeCallStack, size_t)];
+  // The memory for hashtable entry allocation callsite object
+  static size_t _hash_entry_allocation_site[CALC_OBJ_SIZE_IN_TYPE(MallocSiteHashtableEntry, size_t)];
+  NOT_PRODUCT(static int     _peak_count;)
+};
+
+#endif // INCLUDE_NMT
+#endif // SHARE_VM_SERVICES_MALLOC_SITE_TABLE_HPP

src/share/vm/services/mallocTracker.cpp

+/*
+ * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code 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 General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+#include "precompiled.hpp"
+
+#include "runtime/atomic.hpp"
+#include "runtime/atomic.inline.hpp"
+#include "services/mallocSiteTable.hpp"
+#include "services/mallocTracker.hpp"
+#include "services/mallocTracker.inline.hpp"
+#include "services/memTracker.hpp"
+
+size_t MallocMemorySummary::_snapshot[CALC_OBJ_SIZE_IN_TYPE(MallocMemorySnapshot, size_t)];
+
+// Total malloc'd memory amount
+size_t MallocMemorySnapshot::total() const {
+  size_t amount = 0;
+  for (int index = 0; index < mt_number_of_types; index ++) {
+    amount += _malloc[index].malloc_size();
+  }
+  amount += _tracking_header.size() + total_arena();
+  return amount;
+}
+
+// Total malloc'd memory used by arenas
+size_t MallocMemorySnapshot::total_arena() const {
+  size_t amount = 0;
+  for (int index = 0; index < mt_number_of_types; index ++) {
+    amount += _malloc[index].arena_size();
+  }
+  return amount;
+}
+
+
+void MallocMemorySnapshot::reset() {
+  _tracking_header.reset();
+  for (int index = 0; index < mt_number_of_types; index ++) {
+    _malloc[index].reset();
+  }
+}
+
+// Make adjustment by subtracting chunks used by arenas
+// from total chunks to get total free chunck size
+void MallocMemorySnapshot::make_adjustment() {
+  size_t arena_size = total_arena();
+  int chunk_idx = NMTUtil::flag_to_index(mtChunk);
+  _malloc[chunk_idx].record_free(arena_size);
+}
+
+
+void MallocMemorySummary::initialize() {
+  assert(sizeof(_snapshot) >= sizeof(MallocMemorySnapshot), "Sanity Check");
+  // Uses placement new operator to initialize static area.
+  ::new ((void*)_snapshot)MallocMemorySnapshot();
+}
+
+void MallocHeader::release() const {
+  // Tracking already shutdown, no housekeeping is needed anymore
+  if (MemTracker::tracking_level() <= NMT_minimal) return;
+
+  MallocMemorySummary::record_free(size(), flags());
+  MallocMemorySummary::record_free_malloc_header(sizeof(MallocHeader));
+  if (tracking_level() == NMT_detail) {
+    MallocSiteTable::deallocation_at(size(), _bucket_idx, _pos_idx);
+  }
+}
+
+bool MallocHeader::record_malloc_site(const NativeCallStack& stack, size_t size,
+  size_t* bucket_idx, size_t* pos_idx) const {
+  bool ret =  MallocSiteTable::allocation_at(stack, size, bucket_idx, pos_idx);
+
+  // Something went wrong, could be OOM or overflow malloc site table.
+  // We want to keep tracking data under OOM circumstance, so transition to
+  // summary tracking.
+  if (!ret) {
+    MemTracker::transition_to(NMT_summary);
+  }
+  return ret;
+}
+
+bool MallocHeader::get_stack(NativeCallStack& stack) const {
+  return MallocSiteTable::access_stack(stack, _bucket_idx, _pos_idx);
+}
+
+bool MallocTracker::initialize(NMT_TrackingLevel level) {
+  if (level >= NMT_summary) {
+    MallocMemorySummary::initialize();
+  }
+
+  if (level == NMT_detail) {
+    return MallocSiteTable::initialize();
+  }
+  return true;
+}
+
+bool MallocTracker::transition(NMT_TrackingLevel from, NMT_TrackingLevel to) {
+  assert(from != NMT_off, "Can not transition from off state");
+  assert(to != NMT_off, "Can not transition to off state");
+  if (from == NMT_minimal) {
+    MallocMemorySummary::reset();
+  }
+
+  if (to == NMT_detail) {
+    assert(from == NMT_minimal || from == NMT_summary, "Just check");
+    return MallocSiteTable::initialize();
+  } else if (from == NMT_detail) {
+    assert(to == NMT_minimal || to == NMT_summary, "Just check");
+    MallocSiteTable::shutdown();
+  }
+  return true;
+}
+
+// Record a malloc memory allocation
+void* MallocTracker::record_malloc(void* malloc_base, size_t size, MEMFLAGS flags,
+  const NativeCallStack& stack, NMT_TrackingLevel level) {
+  void*         memblock;      // the address for user data
+  MallocHeader* header = NULL;
+
+  if (malloc_base == NULL) {
+    return NULL;
+  }
+
+  // Check malloc size, size has to <= MAX_MALLOC_SIZE. This is only possible on 32-bit
+  // systems, when malloc size >= 1GB, but is is safe to assume it won't happen.
+  if (size > MAX_MALLOC_SIZE) {
+    fatal("Should not use malloc for big memory block, use virtual memory instead");
+  }
+  // Uses placement global new operator to initialize malloc header
+  switch(level) {
+    case NMT_off:
+      return malloc_base;
+    case NMT_minimal: {
+      MallocHeader* hdr = ::new (malloc_base) MallocHeader();
+      break;
+    }
+    case NMT_summary: {
+      header = ::new (malloc_base) MallocHeader(size, flags);
+      break;
+    }
+    case NMT_detail: {
+      header = ::new (malloc_base) MallocHeader(size, flags, stack);
+      break;
+    }
+    default:
+      ShouldNotReachHere();
+  }
+  memblock = (void*)((char*)malloc_base + sizeof(MallocHeader));
+
+  // The alignment check: 8 bytes alignment for 32 bit systems.
+  //                      16 bytes alignment for 64-bit systems.
+  assert(((size_t)memblock & (sizeof(size_t) * 2 - 1)) == 0, "Alignment check");
+
+  // Sanity check
+  assert(get_memory_tracking_level(memblock) == level,
+    "Wrong tracking level");
+
+#ifdef ASSERT
+  if (level > NMT_minimal) {
+    // Read back
+    assert(get_size(memblock) == size,   "Wrong size");
+    assert(get_flags(memblock) == flags, "Wrong flags");
+  }
+#endif
+
+  return memblock;
+}
+
+void* MallocTracker::record_free(void* memblock) {
+  // Never turned on
+  if (MemTracker::tracking_level() == NMT_off ||
+      memblock == NULL) {
+    return memblock;
+  }
+  MallocHeader* header = malloc_header(memblock);
+  header->release();
+
+  return (void*)header;
+}
+
+

src/share/vm/services/mallocTracker.hpp

+/*
+ * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This code is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code 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 General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_VM_SERVICES_MALLOC_TRACKER_HPP
+#define SHARE_VM_SERVICES_MALLOC_TRACKER_HPP
+
+#if INCLUDE_NMT
+
+#include "memory/allocation.hpp"
+#include "runtime/atomic.hpp"
+#include "services/nmtCommon.hpp"
+#include "utilities/nativeCallStack.hpp"
+
+/*
+ * This counter class counts memory allocation and deallocation,
+ * records total memory allocation size and number of allocations.
+ * The counters are updated atomically.
+ */
+class MemoryCounter VALUE_OBJ_CLASS_SPEC {
+ private:
+  size_t   _count;
+  size_t   _size;
+
+  DEBUG_ONLY(size_t   _peak_count;)
+  DEBUG_ONLY(size_t   _peak_size; )
+
+ public:
+  MemoryCounter() : _count(0), _size(0) {
+    DEBUG_ONLY(_peak_count = 0;)
+    DEBUG_ONLY(_peak_size  = 0;)
+  }
+
+  // Reset counters
+  void reset() {
+    _size  = 0;
+    _count = 0;
+    DEBUG_ONLY(_peak_size = 0;)
+    DEBUG_ONLY(_peak_count = 0;)
+  }
+
+  inline void allocate(size_t sz) {
+    Atomic::add(1, (volatile MemoryCounterType*)&_count);
+    if (sz > 0) {
+      Atomic::add((MemoryCounterType)sz, (volatile MemoryCounterType*)&_size);
+      DEBUG_ONLY(_peak_size = MAX2(_peak_size, _size));
+    }
+    DEBUG_ONLY(_peak_count = MAX2(_peak_count, _count);)
+  }
+
+  inline void deallocate(size_t sz) {
+    assert(_count > 0, "Negative counter");
+    assert(_size >= sz, "Negative size");
+    Atomic::add(-1, (volatile MemoryCounterType*)&_count);
+    if (sz > 0) {
+      Atomic::add(-(MemoryCounterType)sz, (volatile MemoryCounterType*)&_size);
+    }
+  }
+
+  inline void resize(long sz) {
+    if (sz != 0) {
+      Atomic::add((MemoryCounterType)sz, (volatile MemoryCounterType*)&_size);
+      DEBUG_ONLY(_peak_size = MAX2(_size, _peak_size);)
+    }
+  }
+
+  inline size_t count() const { return _count; }
+  inline size_t size()  const { return _size;  }
+  DEBUG_ONLY(inline size_t peak_count() const { return _peak_count; })
+  DEBUG_ONLY(inline size_t peak_size()  const { return _peak_size; })
+
+};
+
+/*
+ * Malloc memory used by a particular subsystem.
+ * It includes the memory acquired through os::malloc()
+ * call and arena's backing memory.
+ */
+class MallocMemory VALUE_OBJ_CLASS_SPEC {
+ private:
+  MemoryCounter _malloc;
+  MemoryCounter _arena;
+
+ public:
+  MallocMemory() { }
+
+  inline void record_malloc(size_t sz) {
+    _malloc.allocate(sz);
+  }
+
+  inline void record_free(size_t sz) {
+    _malloc.deallocate(sz);
+  }
+
+  inline void record_new_arena() {
+    _arena.allocate(0);
+  }
+
+  inline void record_arena_free() {
+    _arena.deallocate(0);
+  }
+
+  inline void record_arena_size_change(long sz) {
+    _arena.resize(sz);
+  }
+
+  void reset() {
+    _malloc.reset();
+    _arena.reset();
+  }
+
+  inline size_t malloc_size()  const { return _malloc.size(); }
+  inline size_t malloc_count() const { return _malloc.count();}
+  inline size_t arena_size()   const { return _arena.size();  }
+  inline size_t arena_count()  const { return _arena.count(); }
+
+  DEBUG_ONLY(inline const MemoryCounter& malloc_counter() const { return _malloc; })
+  DEBUG_ONLY(inline const MemoryCounter& arena_counter()  const { return _arena;  })
+};
+
+class MallocMemorySummary;
+
+// A snapshot of malloc'd memory, includes malloc memory
+// usage by types and memory used by tracking itself.
+class MallocMemorySnapshot : public ResourceObj {
+  friend class MallocMemorySummary;
+
+ private:
+  MallocMemory      _malloc[mt_number_of_types];
+  MemoryCounter     _tracking_header;
+
+
+ public:
+  inline MallocMemory*  by_type(MEMFLAGS flags) {
+    int index = NMTUtil::flag_to_index(flags);
+    return &_malloc[index];
+  }
+
+  inline MallocMemory* by_index(int index) {
+    assert(index >= 0, "Index out of bound");
+    assert(index < mt_number_of_types, "Index out of bound");
+    return &_malloc[index];
+  }
+
+  inline MemoryCounter* malloc_overhead() {
+    return &_tracking_header;
+  }
+
+  // Total malloc'd memory amount
+  size_t total() const;
+  // Total malloc'd memory used by arenas
+  size_t total_arena() const;
+
+  inline size_t thread_count() {
+    return by_type(mtThreadStack)->malloc_count();
+  }
+
+  void reset();
+
+  void copy_to(MallocMemorySnapshot* s) {
+    s->_tracking_header = _tracking_header;
+    for (int index = 0; index < mt_number_of_types; index ++) {
+      s->_malloc[index] = _malloc[index];
+    }
+  }
+
+  // Make adjustment by subtracting chunks used by arenas
+  // from total chunks to get total free chunk size
+  void make_adjustment();
+};
+
+/*
+ * This class is for collecting malloc statistics at summary level
+ */
+class MallocMemorySummary : AllStatic {
+ private:
+  // Reserve memory for placement of MallocMemorySnapshot object
+  static size_t _snapshot[CALC_OBJ_SIZE_IN_TYPE(MallocMemorySnapshot, size_t)];
+
+ public:
+   static void initialize();
+
+   static inline void record_malloc(size_t size, MEMFLAGS flag) {
+     as_snapshot()->by_type(flag)->record_malloc(size);
+   }
+
+   static inline void record_free(size_t size, MEMFLAGS flag) {
+     as_snapshot()->by_type(flag)->record_free(size);
+   }
+
+   static inline void record_new_arena(MEMFLAGS flag) {
+     as_snapshot()->by_type(flag)->record_new_arena();
+   }
+
+   static inline void record_arena_free(MEMFLAGS flag) {
+     as_snapshot()->by_type(flag)->record_arena_free();
+   }
+
+   static inline void record_arena_size_change(long size, MEMFLAGS flag) {
+     as_snapshot()->by_type(flag)->record_arena_size_change(size);
+   }
+
+   static void snapshot(MallocMemorySnapshot* s) {
+     as_snapshot()->copy_to(s);
+     s->make_adjustment();
+   }
+
+   // Record memory used by malloc tracking header
+   static inline void record_new_malloc_header(size_t sz) {
+     as_snapshot()->malloc_overhead()->allocate(sz);
+   }
+
+   static inline void record_free_malloc_header(size_t sz) {
+     as_snapshot()->malloc_overhead()->deallocate(sz);
+   }
+
+   // The memory used by malloc tracking headers
+   static inline size_t tracking_overhead() {
+     return as_snapshot()->malloc_overhead()->size();
+   }
+
+   // Reset all counters to zero
+   static void reset() {
+     as_snapshot()->reset();
+   }
+
+  static MallocMemorySnapshot* as_snapshot() {
+    return (MallocMemorySnapshot*)_snapshot;
+  }
+};
+
+
+/*
+ * Malloc tracking header.
+ * To satisfy malloc alignment requirement, NMT uses 2 machine words for tracking purpose,
+ * which ensures 8-bytes alignment on 32-bit systems and 16-bytes on 64-bit systems (Product build).
+ */
+
+class MallocHeader VALUE_OBJ_CLASS_SPEC {
+#ifdef _LP64
+  size_t           _size      : 62;
+  size_t           _level     : 2;
+  size_t           _flags     : 8;
+  size_t           _pos_idx   : 16;
+  size_t           _bucket_idx: 40;
+#define MAX_MALLOCSITE_TABLE_SIZE ((size_t)1 << 40)
+#define MAX_BUCKET_LENGTH         ((size_t)(1 << 16))
+#define MAX_MALLOC_SIZE           (((size_t)1 << 62) - 1)
+#else
+  size_t           _size      : 30;
+  size_t           _level     : 2;
+  size_t           _flags     : 8;
+  size_t           _pos_idx   : 8;
+  size_t           _bucket_idx: 16;
+#define MAX_MALLOCSITE_TABLE_SIZE  ((size_t)(1 << 16))
+#define MAX_BUCKET_LENGTH          ((size_t)(1 << 8))
+// Max malloc size = 1GB - 1 on 32 bit system, such has total 4GB memory
+#define MAX_MALLOC_SIZE            ((size_t)(1 << 30) - 1)
+#endif  // _LP64
+
+ public:
+  // Summary tracking header
+  MallocHeader(size_t size, MEMFLAGS flags) {
+    assert(sizeof(MallocHeader) == sizeof(void*) * 2,
+      "Wrong header size");
+
+    _level = NMT_summary;
+    _flags = flags;
+    set_size(size);
+    MallocMemorySummary::record_malloc(size, flags);
+    MallocMemorySummary::record_new_malloc_header(sizeof(MallocHeader));
+  }
+  // Detail tracking header
+  MallocHeader(size_t size, MEMFLAGS flags, const NativeCallStack& stack) {
+    assert(sizeof(MallocHeader) == sizeof(void*) * 2,
+      "Wrong header size");
+
+    _level = NMT_detail;
+    _flags = flags;
+    set_size(size);
+    size_t bucket_idx;
+    size_t pos_idx;
+    if (record_malloc_site(stack, size, &bucket_idx, &pos_idx)) {
+      assert(bucket_idx <= MAX_MALLOCSITE_TABLE_SIZE, "Overflow bucket index");
+      assert(pos_idx <= MAX_BUCKET_LENGTH, "Overflow bucket position index");
+      _bucket_idx = bucket_idx;
+      _pos_idx = pos_idx;
+    }
+    MallocMemorySummary::record_malloc(size, flags);
+    MallocMemorySummary::record_new_malloc_header(sizeof(MallocHeader));
+  }
+  // Minimal tracking header
+  MallocHeader() {
+    assert(sizeof(MallocHeader) == sizeof(void*) * 2,
+      "Wrong header size");
+
+    _level = (unsigned short)NMT_minimal;
+  }
+
+  inline NMT_TrackingLevel tracking_level() const {
+    return (NMT_TrackingLevel)_level;
+  }
+
+  inline size_t   size()  const { return _size; }
+  inline MEMFLAGS flags() const { return (MEMFLAGS)_flags; }
+  bool get_stack(NativeCallStack& stack) const;
+
+  // Cleanup tracking information before the memory is released.
+  void release() const;
+
+ private:
+  inline void set_size(size_t size) {
+    assert(size <= MAX_MALLOC_SIZE, "Malloc size too large, should use virtual memory?");
+    _size = size;
+  }
+  bool record_malloc_site(const NativeCallStack& stack, size_t size,
+    size_t* bucket_idx, size_t* pos_idx) const;
+};
+
+
+// Main class called from MemTracker to track malloc activities
+class MallocTracker : AllStatic {
+ public:
+  // Initialize malloc tracker for specific tracking level
+  static bool initialize(NMT_TrackingLevel level);
+
+  static bool transition(NMT_TrackingLevel from, NMT_TrackingLevel to);
+
+  // malloc tracking header size for specific tracking level
+  static inline size_t malloc_header_size(NMT_TrackingLevel level) {
+    return (level == NMT_off) ? 0 : sizeof(MallocHeader);
+  }
+
+  // Parameter name convention:
+  // memblock :   the beginning address for user data
+  // malloc_base: the beginning address that includes malloc tracking header
+  //
+  // The relationship:
+  // memblock = (char*)malloc_base + sizeof(nmt header)
+  //
+
+  // Record  malloc on specified memory block
+  static void* record_malloc(void* malloc_base, size_t size, MEMFLAGS flags,
+    const NativeCallStack& stack, NMT_TrackingLevel level);
+
+  // Record free on specified memory block
+  static void* record_free(void* memblock);
+
+  // Get tracking level of specified memory block
+  static inline NMT_TrackingLevel get_memory_tracking_level(void* memblock);
+
+
+  // Offset memory address to header address
+  static inline void* get_base(void* memblock);
+  static inline void* get_base(void* memblock, NMT_TrackingLevel level) {
+    if (memblock == NULL || level == NMT_off) return memblock;
+    return (char*)memblock - malloc_header_size(level);
+  }
+
+  // Get memory size
+  static inline size_t get_size(void* memblock) {
+    MallocHeader* header = malloc_header(memblock);
+    assert(header->tracking_level() >= NMT_summary,
+      "Wrong tracking level");
+    return header->size();
+  }
+
+  // Get memory type
+  static inline MEMFLAGS get_flags(void* memblock) {
+    MallocHeader* header = malloc_header(memblock);
+    assert(header->tracking_level() >= NMT_summary,
+      "Wrong tracking level");
+    return header->flags();
+  }
+
+  // Get header size
+  static inline size_t get_header_size(void* memblock) {
+    return (memblock == NULL) ? 0 : sizeof(MallocHeader);
+  }
+
+  static inline void record_new_arena(MEMFLAGS flags) {
+    MallocMemorySummary::record_new_arena(flags);
+  }
+
+  static inline void record_arena_free(MEMFLAGS flags) {
+    MallocMemorySummary::record_arena_free(flags);
+  }
+
+  static inline void record_arena_size_change(int size, MEMFLAGS flags) {
+    MallocMemorySummary::record_arena_size_change(size, flags);
+  }
+ private:
+  static inline MallocHeader* malloc_header(void *memblock) {
+    assert(memblock != NULL, "NULL pointer");
+    MallocHeader* header = (MallocHeader*)((char*)memblock - sizeof(MallocHeader));
+    assert(header->tracking_level() >= NMT_minimal, "Bad header");
+    return header;
+  }
+};
+
+#endif // INCLUDE_NMT
+
+
+#endif //SHARE_VM_SERVICES_MALLOC_TRACKER_HPP

src/share/vm/services/memPtr.cpp → src/share/vm/services/mallocTracker.inline.hpp

 /*
- * Copyright (c) 2012, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -22,21 +22,22 @@
  *
  */
 
-#include "precompiled.hpp"
-#include "services/memPtr.hpp"
+#ifndef SHARE_VM_SERVICES_MALLOC_TRACKER_INLINE_HPP
+#define SHARE_VM_SERVICES_MALLOC_TRACKER_INLINE_HPP
+
+#include "services/mallocTracker.hpp"
 #include "services/memTracker.hpp"
 
-volatile jint SequenceGenerator::_seq_number = 1;
-volatile unsigned long SequenceGenerator::_generation = 1;
-NOT_PRODUCT(jint SequenceGenerator::_max_seq_number = 1;)
+inline NMT_TrackingLevel MallocTracker::get_memory_tracking_level(void* memblock) {
+  assert(memblock != NULL, "Sanity check");
+  if (MemTracker::tracking_level() == NMT_off) return NMT_off;
+  MallocHeader* header = malloc_header(memblock);
+  return header->tracking_level();
+}
 
-jint SequenceGenerator::next() {
-  jint seq = Atomic::add(1, &_seq_number);
-  if (seq < 0) {
-    MemTracker::shutdown(MemTracker::NMT_sequence_overflow);
-  } else {
-    NOT_PRODUCT(_max_seq_number = (seq > _max_seq_number) ? seq : _max_seq_number;)
-  }
-  return seq;
+inline void* MallocTracker::get_base(void* memblock){
+  return get_base(memblock, MemTracker::tracking_level());
 }
 
+#endif // SHARE_VM_SERVICES_MALLOC_TRACKER_INLINE_HPP
+

src/share/vm/services/memBaseline.hpp

 /*
- * Copyright (c) 2012, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -25,425 +25,205 @@
 #ifndef SHARE_VM_SERVICES_MEM_BASELINE_HPP
 #define SHARE_VM_SERVICES_MEM_BASELINE_HPP
 
+#if INCLUDE_NMT
+
 #include "memory/allocation.hpp"
 #include "runtime/mutex.hpp"
-#include "services/memPtr.hpp"
-#include "services/memSnapshot.hpp"
+#include "services/mallocSiteTable.hpp"
+#include "services/mallocTracker.hpp"
+#include "services/nmtCommon.hpp"
+#include "services/virtualMemoryTracker.hpp"
+#include "utilities/linkedlist.hpp"
 
-// compare unsigned number
-#define UNSIGNED_COMPARE(a, b)  ((a > b) ? 1 : ((a == b) ? 0 : -1))
+typedef LinkedListIterator<MallocSite>                   MallocSiteIterator;
+typedef LinkedListIterator<VirtualMemoryAllocationSite>  VirtualMemorySiteIterator;
+typedef LinkedListIterator<ReservedMemoryRegion>         VirtualMemoryAllocationIterator;
 
 /*
- * MallocCallsitePointer and VMCallsitePointer are used
- * to baseline memory blocks with their callsite information.
- * They are only available when detail tracking is turned
- * on.
+ * Baseline a memory snapshot
  */
-
-/* baselined malloc record aggregated by callsite */
-class MallocCallsitePointer : public MemPointer {
- private:
-  size_t    _count;   // number of malloc invocation from this callsite
-  size_t    _amount;  // total amount of memory malloc-ed from this callsite
-
+class MemBaseline VALUE_OBJ_CLASS_SPEC {
  public:
-  MallocCallsitePointer() {
-    _count = 0;
-    _amount = 0;
-  }
-
-  MallocCallsitePointer(address pc) : MemPointer(pc) {
-    _count = 0;
-    _amount = 0;
-  }
-
-  MallocCallsitePointer& operator=(const MallocCallsitePointer& p) {
-    MemPointer::operator=(p);
-    _count = p.count();
-    _amount = p.amount();
-    return *this;
-  }
-
-  inline void inc(size_t size) {
-    _count ++;
-    _amount += size;
+  enum BaselineThreshold {
+    SIZE_THRESHOLD = K        // Only allocation size over this threshold will be baselined.
   };
 
-  inline size_t count() const {
-    return _count;
-  }
+  enum BaselineType {
+    Not_baselined,
+    Summary_baselined,
+    Detail_baselined
+  };
 
-  inline size_t amount() const {
-    return _amount;
-  }
-};
+  enum SortingOrder {
+    by_address,   // by memory address
+    by_size,      // by memory size
+    by_site       // by call site where the memory is allocated from
+  };
 
-// baselined virtual memory record aggregated by callsite
-class VMCallsitePointer : public MemPointer {
  private:
-  size_t     _count;              // number of invocation from this callsite
-  size_t     _reserved_amount;    // total reserved amount
-  size_t     _committed_amount;   // total committed amount
-
- public:
-  VMCallsitePointer() {
-    _count = 0;
-    _reserved_amount = 0;
-    _committed_amount = 0;
-  }
-
-  VMCallsitePointer(address pc) : MemPointer(pc) {
-    _count = 0;
-    _reserved_amount = 0;
-    _committed_amount = 0;
-  }
-
-  VMCallsitePointer& operator=(const VMCallsitePointer& p) {
-    MemPointer::operator=(p);
-    _count = p.count();
-    _reserved_amount = p.reserved_amount();
-    _committed_amount = p.committed_amount();
-    return *this;
-  }
-
-  inline void inc(size_t reserved, size_t committed) {
-    _count ++;
-    _reserved_amount += reserved;
-    _committed_amount += committed;
-  }
+  // All baseline data is stored in this arena
+  Arena*                  _arena;
 
-  inline size_t count() const {
-    return _count;
-  }
+  // Summary information
+  MallocMemorySnapshot*   _malloc_memory_snapshot;
+  VirtualMemorySnapshot*  _virtual_memory_snapshot;
 
-  inline size_t reserved_amount() const {
-    return _reserved_amount;
-  }
+  size_t               _class_count;
 
-  inline size_t committed_amount() const {
-    return _committed_amount;
-  }
-};
+  // Allocation sites information
+  // Malloc allocation sites
+  LinkedListImpl<MallocSite, ResourceObj::ARENA>
+                       _malloc_sites;
 
-// maps a memory type flag to readable name
-typedef struct _memType2Name {
-  MEMFLAGS     _flag;
-  const char*  _name;
-} MemType2Name;
+  // All virtual memory allocations
+  LinkedListImpl<ReservedMemoryRegion, ResourceObj::ARENA>
+                       _virtual_memory_allocations;
 
+  // Virtual memory allocations by allocation sites, always in by_address
+  // order
+  LinkedListImpl<VirtualMemoryAllocationSite, ResourceObj::ARENA>
+                       _virtual_memory_sites;
 
-// This class aggregates malloc'd records by memory type
-class MallocMem VALUE_OBJ_CLASS_SPEC {
- private:
-  MEMFLAGS       _type;
+  SortingOrder         _malloc_sites_order;
+  SortingOrder         _virtual_memory_sites_order;
 
-  size_t         _count;
-  size_t         _amount;
+  BaselineType         _baseline_type;
 
  public:
-  MallocMem() {
-    _type = mtNone;
-    _count = 0;
-    _amount = 0;
-  }
-
-  MallocMem(MEMFLAGS flags) {
-    assert(HAS_VALID_MEMORY_TYPE(flags), "no type");
-    _type = FLAGS_TO_MEMORY_TYPE(flags);
-    _count = 0;
-    _amount = 0;
-  }
-
-  inline void set_type(MEMFLAGS flag) {
-    _type = flag;
-  }
-
-  inline void clear() {
-    _count = 0;
-    _amount = 0;
-    _type = mtNone;
-  }
-
-  MallocMem& operator=(const MallocMem& m) {
-    assert(_type == m.type(), "different type");
-    _count = m.count();
-    _amount = m.amount();
-    return *this;
-  }
-
-  inline void inc(size_t amt) {
-    _amount += amt;
-    _count ++;
-  }
-
-  inline void reduce(size_t amt) {
-    assert(_amount >= amt, "Just check");
-    _amount -= amt;
-  }
-
-  inline void overwrite_counter(size_t count) {
-    _count = count;
-  }
-
-  inline MEMFLAGS type() const {
-    return _type;
+  // create a memory baseline
+  MemBaseline():
+    _baseline_type(Not_baselined),
+    _class_count(0),
+    _arena(NULL),
+    _malloc_memory_snapshot(NULL),
+    _virtual_memory_snapshot(NULL),
+    _malloc_sites(NULL) {
   }
 
-  inline bool is_type(MEMFLAGS flags) const {
-    return FLAGS_TO_MEMORY_TYPE(flags) == _type;
+  ~MemBaseline() {
+    reset();
+    if (_arena != NULL) {
+      delete _arena;
+    }
   }
 
-  inline size_t count() const {
-    return _count;
-  }
+  bool baseline(bool summaryOnly = true);
 
-  inline size_t amount() const {
-    return _amount;
-  }
-};
+  BaselineType baseline_type() const { return _baseline_type; }
 
-// This class records live arena's memory usage
-class ArenaMem : public MallocMem {
- public:
-  ArenaMem(MEMFLAGS typeflag): MallocMem(typeflag) {
+  MallocMemorySnapshot* malloc_memory_snapshot() const {
+    return _malloc_memory_snapshot;
   }
-  ArenaMem() { }
-};
-
-// This class aggregates virtual memory by its memory type
-class VMMem VALUE_OBJ_CLASS_SPEC {
- private:
-  MEMFLAGS       _type;
 
-  size_t         _count;
-  size_t         _reserved_amount;
-  size_t         _committed_amount;
-
- public:
-  VMMem() {
-    _type = mtNone;
-    _count = 0;
-    _reserved_amount = 0;
-    _committed_amount = 0;
+  VirtualMemorySnapshot* virtual_memory_snapshot() const {
+    return _virtual_memory_snapshot;
   }
 
-  VMMem(MEMFLAGS flags) {
-    assert(HAS_VALID_MEMORY_TYPE(flags), "no type");
-    _type = FLAGS_TO_MEMORY_TYPE(flags);
-    _count = 0;
-    _reserved_amount = 0;
-    _committed_amount = 0;
-  }
+  MallocSiteIterator malloc_sites(SortingOrder order);
+  VirtualMemorySiteIterator virtual_memory_sites(SortingOrder order);
 
-  inline void clear() {
-    _type = mtNone;
-    _count = 0;
-    _reserved_amount = 0;
-    _committed_amount = 0;
+  // Virtual memory allocation iterator always returns in virtual memory
+  // base address order.
+  VirtualMemoryAllocationIterator virtual_memory_allocations() {
+    assert(!_virtual_memory_allocations.is_empty(), "Not detail baseline");
+    return VirtualMemoryAllocationIterator(_virtual_memory_allocations.head());
   }
 
-  inline void set_type(MEMFLAGS flags) {
-    _type = FLAGS_TO_MEMORY_TYPE(flags);
+  // Total reserved memory = total malloc'd memory + total reserved virtual
+  // memory
+  size_t total_reserved_memory() const {
+    assert(baseline_type() != Not_baselined, "Not yet baselined");
+    assert(_virtual_memory_snapshot != NULL, "No virtual memory snapshot");
+    assert(_malloc_memory_snapshot != NULL,  "No malloc memory snapshot");
+    size_t amount = _malloc_memory_snapshot->total() +
+           _virtual_memory_snapshot->total_reserved();
+    return amount;
   }
 
-  VMMem& operator=(const VMMem& m) {
-    assert(_type == m.type(), "different type");
-
-    _count = m.count();
-    _reserved_amount = m.reserved_amount();
-    _committed_amount = m.committed_amount();
-    return *this;
+  // Total committed memory = total malloc'd memory + total committed
+  // virtual memory
+  size_t total_committed_memory() const {
+    assert(baseline_type() != Not_baselined, "Not yet baselined");
+    assert(_virtual_memory_snapshot != NULL,
+      "Not a snapshot");
+    size_t amount = _malloc_memory_snapshot->total() +
+           _virtual_memory_snapshot->total_committed();
+    return amount;
   }
 
-
-  inline MEMFLAGS type() const {
-    return _type;
+  size_t total_arena_memory() const {
+    assert(baseline_type() != Not_baselined, "Not yet baselined");
+    assert(_malloc_memory_snapshot != NULL, "Not yet baselined");
+    return _malloc_memory_snapshot->total_arena();
   }
 
-  inline bool is_type(MEMFLAGS flags) const {
-    return FLAGS_TO_MEMORY_TYPE(flags) == _type;
+  size_t malloc_tracking_overhead() const {
+    assert(baseline_type() != Not_baselined, "Not yet baselined");
+    return _malloc_memory_snapshot->malloc_overhead()->size();
   }
 
-  inline void inc(size_t reserved_amt, size_t committed_amt) {
-    _reserved_amount += reserved_amt;
-    _committed_amount += committed_amt;
-    _count ++;
+  const MallocMemory* malloc_memory(MEMFLAGS flag) const {
+    assert(_malloc_memory_snapshot != NULL, "Not a snapshot");
+    return _malloc_memory_snapshot->by_type(flag);
   }
 
-  inline size_t count() const {
-    return _count;
+  const VirtualMemory* virtual_memory(MEMFLAGS flag) const {
+    assert(_virtual_memory_snapshot != NULL, "Not a snapshot");
+    return _virtual_memory_snapshot->by_type(flag);
   }
 
-  inline size_t reserved_amount() const {
-    return _reserved_amount;
-  }
 
-  inline size_t committed_amount() const {
-    return _committed_amount;
+  size_t class_count() const {
+    assert(baseline_type() != Not_baselined, "Not yet baselined");
+    return _class_count;
   }
-};
-
-
-
-#define NUMBER_OF_MEMORY_TYPE    (mt_number_of_types + 1)
-
-class BaselineReporter;
-class BaselineComparisonReporter;
-
-/*
- * This class baselines current memory snapshot.
- * A memory baseline summarizes memory usage by memory type,
- * aggregates memory usage by callsites when detail tracking
- * is on.
- */
-class MemBaseline VALUE_OBJ_CLASS_SPEC {
-  friend class BaselineReporter;
-  friend class BaselineComparisonReporter;
-
- private:
-  // overall summaries
-  size_t        _total_malloced;
-  size_t        _total_vm_reserved;
-  size_t        _total_vm_committed;
-  size_t        _number_of_classes;
-  size_t        _number_of_threads;
-
-  // if it has properly baselined
-  bool          _baselined;
-
-  // we categorize memory into three categories within the memory type
-  MallocMem     _malloc_data[NUMBER_OF_MEMORY_TYPE];
-  VMMem         _vm_data[NUMBER_OF_MEMORY_TYPE];
-  ArenaMem      _arena_data[NUMBER_OF_MEMORY_TYPE];
-
-  // memory records that aggregate memory usage by callsites.
-  // only available when detail tracking is on.
-  MemPointerArray*  _malloc_cs;
-  MemPointerArray*  _vm_cs;
-  // virtual memory map
-  MemPointerArray*  _vm_map;
-
- private:
-  static MemType2Name  MemType2NameMap[NUMBER_OF_MEMORY_TYPE];
-
- private:
-  // should not use copy constructor
-  MemBaseline(MemBaseline& copy) { ShouldNotReachHere(); }
-
-  // check and block at a safepoint
-  static inline void check_safepoint(JavaThread* thr);
-
- public:
-  // create a memory baseline
-  MemBaseline();
-
-  ~MemBaseline();
 
-  inline bool baselined() const {
-    return _baselined;
+  size_t thread_count() const {
+    assert(baseline_type() != Not_baselined, "Not yet baselined");
+    assert(_malloc_memory_snapshot != NULL, "Baselined?");
+    return _malloc_memory_snapshot->thread_count();
   }
 
-  MemBaseline& operator=(const MemBaseline& other);
-
   // reset the baseline for reuse
-  void clear();
-
-  // baseline the snapshot
-  bool baseline(MemSnapshot& snapshot, bool summary_only = true);
-
-  bool baseline(const MemPointerArray* malloc_records,
-                const MemPointerArray* vm_records,
-                bool summary_only = true);
-
-  // total malloc'd memory of specified memory type
-  inline size_t malloc_amount(MEMFLAGS flag) const {
-    return _malloc_data[flag2index(flag)].amount();
-  }
-  // number of malloc'd memory blocks of specified memory type
-  inline size_t malloc_count(MEMFLAGS flag) const {
-    return _malloc_data[flag2index(flag)].count();
-  }
-  // total memory used by arenas of specified memory type
-  inline size_t arena_amount(MEMFLAGS flag) const {
-    return _arena_data[flag2index(flag)].amount();
-  }
-  // number of arenas of specified memory type
-  inline size_t arena_count(MEMFLAGS flag) const {
-    return _arena_data[flag2index(flag)].count();
-  }
-  // total reserved memory of specified memory type
-  inline size_t reserved_amount(MEMFLAGS flag) const {
-    return _vm_data[flag2index(flag)].reserved_amount();
-  }
-  // total committed memory of specified memory type
-  inline size_t committed_amount(MEMFLAGS flag) const {
-    return _vm_data[flag2index(flag)].committed_amount();
-  }
-  // total memory (malloc'd + mmap'd + arena) of specified
-  // memory type
-  inline size_t total_amount(MEMFLAGS flag) const {
-    int index = flag2index(flag);
-    return _malloc_data[index].amount() +
-           _vm_data[index].reserved_amount() +
-           _arena_data[index].amount();
-  }
+  void reset() {
+    _baseline_type = Not_baselined;
+    _malloc_memory_snapshot = NULL;
+    _virtual_memory_snapshot = NULL;
+    _class_count  = 0;
 
-  /* overall summaries */
+    _malloc_sites = NULL;
+    _virtual_memory_sites = NULL;
+    _virtual_memory_allocations = NULL;
 
-  // total malloc'd memory in snapshot
-  inline size_t total_malloc_amount() const {
-    return _total_malloced;
-  }
-  // total mmap'd memory in snapshot
-  inline size_t total_reserved_amount() const {
-    return _total_vm_reserved;
-  }
-  // total committed memory in snapshot
-  inline size_t total_committed_amount() const {
-    return _total_vm_committed;
+    if (_arena != NULL) {
+      _arena->destruct_contents();
+    }
   }
-  // number of loaded classes
-  inline size_t number_of_classes() const {
-    return _number_of_classes;
-  }
-  // number of running threads
-  inline size_t number_of_threads() const {
-    return _number_of_threads;
-  }
-  // lookup human readable name of a memory type
-  static const char* type2name(MEMFLAGS type);
 
  private:
-  // convert memory flag to the index to mapping table
-  int         flag2index(MEMFLAGS flag) const;
+  // Baseline summary information
+  bool baseline_summary();
 
-  // reset baseline values
-  void reset();
+  // Baseline allocation sites (detail tracking only)
+  bool baseline_allocation_sites();
 
-  // summarize the records in global snapshot
-  bool baseline_malloc_summary(const MemPointerArray* malloc_records);
-  bool baseline_vm_summary(const MemPointerArray* vm_records);
-  bool baseline_malloc_details(const MemPointerArray* malloc_records);
-  bool baseline_vm_details(const MemPointerArray* vm_records);
+  // Aggregate virtual memory allocation by allocation sites
+  bool aggregate_virtual_memory_allocation_sites();
 
-  // print a line of malloc'd memory aggregated by callsite
-  void print_malloc_callsite(outputStream* st, address pc, size_t size,
-    size_t count, int diff_amt, int diff_count) const;
-  // print a line of mmap'd memory aggregated by callsite
-  void print_vm_callsite(outputStream* st, address pc, size_t rsz,
-    size_t csz, int diff_rsz, int diff_csz) const;
+  Arena* arena() { return _arena; }
 
-  // sorting functions for raw records
-  static int malloc_sort_by_pc(const void* p1, const void* p2);
-  static int malloc_sort_by_addr(const void* p1, const void* p2);
+  // Sorting allocation sites in different orders
+  // Sort allocation sites in size order
+  void malloc_sites_to_size_order();
+  // Sort allocation sites in call site address order
+  void malloc_sites_to_allocation_site_order();
 
- private:
-  // sorting functions for baselined records
-  static int bl_malloc_sort_by_size(const void* p1, const void* p2);
-  static int bl_vm_sort_by_size(const void* p1, const void* p2);
-  static int bl_malloc_sort_by_pc(const void* p1, const void* p2);
-  static int bl_vm_sort_by_pc(const void* p1, const void* p2);
+  // Sort allocation sites in reserved size order
+  void virtual_memory_sites_to_size_order();
+  // Sort allocation sites in call site address order
+  void virtual_memory_sites_to_reservation_site_order();
 };
 
+#endif // INCLUDE_NMT
 
 #endif // SHARE_VM_SERVICES_MEM_BASELINE_HPP

src/share/vm/services/memPtr.hpp

-/*
- * Copyright (c) 2012, 2013, Oracle and/or its affiliates. All rights reserved.
- * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
- *
- * This code is free software; you can redistribute it and/or modify it
- * under the terms of the GNU General Public License version 2 only, as
- * published by the Free Software Foundation.
- *
- * This code 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 General Public License
- * version 2 for more details (a copy is included in the LICENSE file that
- * accompanied this code).
- *
- * You should have received a copy of the GNU General Public License version
- * 2 along with this work; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
- *
- * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
- * or visit www.oracle.com if you need additional information or have any
- * questions.
- *
- */
-
-#ifndef SHARE_VM_SERVICES_MEM_PTR_HPP
-#define SHARE_VM_SERVICES_MEM_PTR_HPP
-
-#include "memory/allocation.hpp"
-#include "runtime/atomic.hpp"
-#include "runtime/os.hpp"
-#include "runtime/safepoint.hpp"
-
-/*
- * global sequence generator that generates sequence numbers to serialize
- * memory records.
- */
-class SequenceGenerator : AllStatic {
- public:
-  static jint next();
-
-  // peek last sequence number
-  static jint peek() {
-    return _seq_number;
-  }
-
-  // reset sequence number
-  static void reset() {
-    assert(SafepointSynchronize::is_at_safepoint(), "Safepoint required");
-    _seq_number = 1;
-    _generation ++;
-  };
-
-  static unsigned long current_generation() { return _generation; }
-  NOT_PRODUCT(static jint max_seq_num() { return _max_seq_number; })
-
- private:
-  static volatile jint             _seq_number;
-  static volatile unsigned long    _generation;
-  NOT_PRODUCT(static jint          _max_seq_number; )
-};
-
-/*
- * followings are the classes that are used to hold memory activity records in different stages.
- *   MemPointer
- *     |--------MemPointerRecord
- *                     |
- *                     |----MemPointerRecordEx
- *                     |           |
- *                     |           |-------SeqMemPointerRecordEx
- *                     |
- *                     |----SeqMemPointerRecord
- *                     |
- *                     |----VMMemRegion
- *                               |
- *                               |-----VMMemRegionEx
- *
- *
- *  prefix 'Seq' - sequenced, the record contains a sequence number
- *  surfix 'Ex'  - extension, the record contains a caller's pc
- *
- *  per-thread recorder : SeqMemPointerRecord(Ex)
- *  snapshot staging    : SeqMemPointerRecord(Ex)
- *  snapshot            : MemPointerRecord(Ex) and VMMemRegion(Ex)
- *
- */
-
-/*
- * class that wraps an address to a memory block,
- * the memory pointer either points to a malloc'd
- * memory block, or a mmap'd memory block
- */
-class MemPointer VALUE_OBJ_CLASS_SPEC {
- public:
-  MemPointer(): _addr(0) { }
-  MemPointer(address addr): _addr(addr) { }
-
-  MemPointer(const MemPointer& copy_from) {
-    _addr = copy_from.addr();
-  }
-
-  inline address addr() const {
-    return _addr;
-  }
-
-  inline operator address() const {
-    return addr();
-  }
-
-  inline bool operator == (const MemPointer& other) const {
-    return addr() == other.addr();
-  }
-
-  inline MemPointer& operator = (const MemPointer& other) {
-    _addr = other.addr();
-    return *this;
-  }
-
- protected:
-  inline void set_addr(address addr) { _addr = addr; }
-
- protected:
-  // memory address
-  address    _addr;
-};
-
-/* MemPointerRecord records an activityand associated
- * attributes on a memory block.
- */
-class MemPointerRecord : public MemPointer {
- private:
-  MEMFLAGS       _flags;
-  size_t         _size;
-
-public:
-  /* extension of MemoryType enum
-   * see share/vm/memory/allocation.hpp for details.
-   *
-   * The tag values are associated to sorting orders, so be
-   * careful if changes are needed.
-   * The allocation records should be sorted ahead of tagging
-   * records, which in turn ahead of deallocation records
-   */
-  enum MemPointerTags {
-    tag_alloc            = 0x0001, // malloc or reserve record
-    tag_commit           = 0x0002, // commit record
-    tag_type             = 0x0003, // tag virtual memory to a memory type
-    tag_uncommit         = 0x0004, // uncommit record
-    tag_release          = 0x0005, // free or release record
-    tag_size             = 0x0006, // arena size
-    tag_masks            = 0x0007, // all tag bits
-    vmBit                = 0x0008
-  };
-
-  /* helper functions to interpret the tagging flags */
-
-  inline static bool is_allocation_record(MEMFLAGS flags) {
-    return (flags & tag_masks) == tag_alloc;
-  }
-
-  inline static bool is_deallocation_record(MEMFLAGS flags) {
-    return (flags & tag_masks) == tag_release;
-  }
-
-  inline static bool is_arena_record(MEMFLAGS flags) {
-    return (flags & (otArena | tag_size)) == otArena;
-  }
-
-  inline static bool is_arena_memory_record(MEMFLAGS flags) {
-    return (flags & (otArena | tag_size)) == (otArena | tag_size);
-  }
-
-  inline static bool is_virtual_memory_record(MEMFLAGS flags) {
-    return (flags & vmBit) != 0;
-  }
-
-  inline static bool is_virtual_memory_reserve_record(MEMFLAGS flags) {
-    return (flags & 0x0F) == (tag_alloc | vmBit);
-  }
-
-  inline static bool is_virtual_memory_commit_record(MEMFLAGS flags) {
-    return (flags & 0x0F) == (tag_commit | vmBit);
-  }
-
-  inline static bool is_virtual_memory_uncommit_record(MEMFLAGS flags) {
-    return (flags & 0x0F) == (tag_uncommit | vmBit);
-  }
-
-  inline static bool is_virtual_memory_release_record(MEMFLAGS flags) {
-    return (flags & 0x0F) == (tag_release | vmBit);
-  }
-
-  inline static bool is_virtual_memory_type_record(MEMFLAGS flags) {
-    return (flags & 0x0F) == (tag_type | vmBit);
-  }
-
-  /* tagging flags */
-  inline static MEMFLAGS malloc_tag()                 { return tag_alloc;   }
-  inline static MEMFLAGS free_tag()                   { return tag_release; }
-  inline static MEMFLAGS arena_size_tag()             { return tag_size | otArena; }
-  inline static MEMFLAGS virtual_memory_tag()         { return vmBit; }
-  inline static MEMFLAGS virtual_memory_reserve_tag() { return (tag_alloc | vmBit); }
-  inline static MEMFLAGS virtual_memory_commit_tag()  { return (tag_commit | vmBit); }
-  inline static MEMFLAGS virtual_memory_uncommit_tag(){ return (tag_uncommit | vmBit); }
-  inline static MEMFLAGS virtual_memory_release_tag() { return (tag_release | vmBit); }
-  inline static MEMFLAGS virtual_memory_type_tag()    { return (tag_type | vmBit); }
-
- public:
-  MemPointerRecord(): _size(0), _flags(mtNone) { }
-
-  MemPointerRecord(address addr, MEMFLAGS memflags, size_t size = 0):
-      MemPointer(addr), _flags(memflags), _size(size) { }
-
-  MemPointerRecord(const MemPointerRecord& copy_from):
-    MemPointer(copy_from), _flags(copy_from.flags()),
-    _size(copy_from.size()) {
-  }
-
-  /* MemPointerRecord is not sequenced, it always return
-   * 0 to indicate non-sequenced
-   */
-  virtual jint seq() const               { return 0; }
-
-  inline size_t   size()  const          { return _size; }
-  inline void set_size(size_t size)      { _size = size; }
-
-  inline MEMFLAGS flags() const          { return _flags; }
-  inline void set_flags(MEMFLAGS flags)  { _flags = flags; }
-
-  MemPointerRecord& operator= (const MemPointerRecord& ptr) {
-    MemPointer::operator=(ptr);
-    _flags = ptr.flags();
-#ifdef ASSERT
-    if (IS_ARENA_OBJ(_flags)) {
-      assert(!is_vm_pointer(), "wrong flags");
-      assert((_flags & ot_masks) == otArena, "wrong flags");
-    }
-#endif
-    _size = ptr.size();
-    return *this;
-  }
-
-  // if the pointer represents a malloc-ed memory address
-  inline bool is_malloced_pointer() const {
-    return !is_vm_pointer();
-  }
-
-  // if the pointer represents a virtual memory address
-  inline bool is_vm_pointer() const {
-    return is_virtual_memory_record(_flags);
-  }
-
-  // if this record records a 'malloc' or virtual memory
-  // 'reserve' call
-  inline bool is_allocation_record() const {
-    return is_allocation_record(_flags);
-  }
-
-  // if this record records a size information of an arena
-  inline bool is_arena_memory_record() const {
-    return is_arena_memory_record(_flags);
-  }
-
-  // if this pointer represents an address to an arena object
-  inline bool is_arena_record() const {
-    return is_arena_record(_flags);
-  }
-
-  // if this record represents a size information of specific arena
-  inline bool is_memory_record_of_arena(const MemPointerRecord* arena_rc) {
-    assert(is_arena_memory_record(), "not size record");
-    assert(arena_rc->is_arena_record(), "not arena record");
-    return (arena_rc->addr() + sizeof(void*)) == addr();
-  }
-
-  // if this record records a 'free' or virtual memory 'free' call
-  inline bool is_deallocation_record() const {
-    return is_deallocation_record(_flags);
-  }
-
-  // if this record records a virtual memory 'commit' call
-  inline bool is_commit_record() const {
-    return is_virtual_memory_commit_record(_flags);
-  }
-
-  // if this record records a virtual memory 'uncommit' call
-  inline bool is_uncommit_record() const {
-    return is_virtual_memory_uncommit_record(_flags);
-  }
-
-  // if this record is a tagging record of a virtual memory block
-  inline bool is_type_tagging_record() const {
-    return is_virtual_memory_type_record(_flags);
-  }
-
-  // if the two memory pointer records actually represent the same
-  // memory block
-  inline bool is_same_region(const MemPointerRecord* other) const {
-    return (addr() == other->addr() && size() == other->size());
-  }
-
-  // if this memory region fully contains another one
-  inline bool contains_region(const MemPointerRecord* other) const {
-    return contains_region(other->addr(), other->size());
-  }
-
-  // if this memory region fully contains specified memory range
-  inline bool contains_region(address add, size_t sz) const {
-    return (addr() <= add && addr() + size() >= add + sz);
-  }
-
-  inline bool contains_address(address add) const {
-    return (addr() <= add && addr() + size() > add);
-  }
-
-  // if this memory region overlaps another region
-  inline bool overlaps_region(const MemPointerRecord* other) const {
-    assert(other != NULL, "Just check");
-    assert(size() > 0 && other->size() > 0, "empty range");
-    return contains_address(other->addr()) ||
-           contains_address(other->addr() + other->size() - 1) || // exclude end address
-           other->contains_address(addr()) ||
-           other->contains_address(addr() + size() - 1); // exclude end address
-  }
-
-};
-
-// MemPointerRecordEx also records callsite pc, from where
-// the memory block is allocated
-class MemPointerRecordEx : public MemPointerRecord {
- private:
-  address      _pc;  // callsite pc
-
- public:
-  MemPointerRecordEx(): _pc(0) { }
-
-  MemPointerRecordEx(address addr, MEMFLAGS memflags, size_t size = 0, address pc = 0):
-    MemPointerRecord(addr, memflags, size), _pc(pc) {}
-
-  MemPointerRecordEx(const MemPointerRecordEx& copy_from):
-    MemPointerRecord(copy_from), _pc(copy_from.pc()) {}
-
-  inline address pc() const { return _pc; }
-
-  void init(const MemPointerRecordEx* mpe) {
-    MemPointerRecord::operator=(*mpe);
-    _pc = mpe->pc();
-  }
-
-  void init(const MemPointerRecord* mp) {
-    MemPointerRecord::operator=(*mp);
-    _pc = 0;
-  }
-};
-
-// a virtual memory region. The region can represent a reserved
-// virtual memory region or a committed memory region
-class VMMemRegion : public MemPointerRecord {
-public:
-  VMMemRegion() { }
-
-  void init(const MemPointerRecord* mp) {
-    assert(mp->is_vm_pointer(), "Sanity check");
-    _addr = mp->addr();
-      set_size(mp->size());
-    set_flags(mp->flags());
-  }
-
-  VMMemRegion& operator=(const VMMemRegion& other) {
-    MemPointerRecord::operator=(other);
-    return *this;
-  }
-
-  inline bool is_reserved_region() const {
-    return is_allocation_record();
-  }
-
-  inline bool is_committed_region() const {
-    return is_commit_record();
-  }
-
-  /* base address of this virtual memory range */
-  inline address base() const {
-    return addr();
-  }
-
-  /* tag this virtual memory range to the specified memory type */
-  inline void tag(MEMFLAGS f) {
-    set_flags(flags() | (f & mt_masks));
-  }
-
-  // expand this region to also cover specified range.
-  // The range has to be on either end of the memory region.
-  void expand_region(address addr, size_t sz) {
-    if (addr < base()) {
-      assert(addr + sz == base(), "Sanity check");
-      _addr = addr;
-      set_size(size() + sz);
-    } else {
-      assert(base() + size() == addr, "Sanity check");
-      set_size(size() + sz);
-    }
-  }
-
-  // exclude the specified address range from this region.
-  // The excluded memory range has to be on either end of this memory
-  // region.
-  inline void exclude_region(address add, size_t sz) {
-    assert(is_reserved_region() || is_committed_region(), "Sanity check");
-    assert(addr() != NULL && size() != 0, "Sanity check");
-    assert(add >= addr() && add < addr() + size(), "Sanity check");
-    assert(add == addr() || (add + sz) == (addr() + size()),
-      "exclude in the middle");
-    if (add == addr()) {
-      set_addr(add + sz);
-      set_size(size() - sz);
-    } else {
-      set_size(size() - sz);
-    }
-  }
-};
-
-class VMMemRegionEx : public VMMemRegion {
- private:
-  jint   _seq;  // sequence number
-
- public:
-  VMMemRegionEx(): _pc(0) { }
-
-  void init(const MemPointerRecordEx* mpe) {
-    VMMemRegion::init(mpe);
-    _pc = mpe->pc();
-  }
-
-  void init(const MemPointerRecord* mpe) {
-    VMMemRegion::init(mpe);
-    _pc = 0;
-  }
-
-  VMMemRegionEx& operator=(const VMMemRegionEx& other) {
-    VMMemRegion::operator=(other);
-    _pc = other.pc();
-    return *this;
-  }
-
-  inline address pc() const { return _pc; }
- private:
-  address   _pc;
-};
-
-/*
- * Sequenced memory record
- */
-class SeqMemPointerRecord : public MemPointerRecord {
- private:
-   jint _seq;  // sequence number
-
- public:
-  SeqMemPointerRecord(): _seq(0){ }
-
-  SeqMemPointerRecord(address addr, MEMFLAGS flags, size_t size, jint seq)
-    : MemPointerRecord(addr, flags, size), _seq(seq)  {
-  }
-
-  SeqMemPointerRecord(const SeqMemPointerRecord& copy_from)
-    : MemPointerRecord(copy_from) {
-    _seq = copy_from.seq();
-  }
-
-  SeqMemPointerRecord& operator= (const SeqMemPointerRecord& ptr) {
-    MemPointerRecord::operator=(ptr);
-    _seq = ptr.seq();
-    return *this;
-  }
-
-  inline jint seq() const {
-    return _seq;
-  }
-};
-
-
-
-class SeqMemPointerRecordEx : public MemPointerRecordEx {
- private:
-  jint    _seq;  // sequence number
-
- public:
-  SeqMemPointerRecordEx(): _seq(0) { }
-
-  SeqMemPointerRecordEx(address addr, MEMFLAGS flags, size_t size,
-    jint seq, address pc):
-    MemPointerRecordEx(addr, flags, size, pc), _seq(seq)  {
-  }
-
-  SeqMemPointerRecordEx(const SeqMemPointerRecordEx& copy_from)
-    : MemPointerRecordEx(copy_from) {
-    _seq = copy_from.seq();
-  }
-
-  SeqMemPointerRecordEx& operator= (const SeqMemPointerRecordEx& ptr) {
-    MemPointerRecordEx::operator=(ptr);
-    _seq = ptr.seq();
-    return *this;
-  }
-
-  inline jint seq() const {
-    return _seq;
-  }
-};
-
-#endif // SHARE_VM_SERVICES_MEM_PTR_HPP

src/share/vm/services/memPtrArray.hpp

-/*
- * Copyright (c) 2012, Oracle and/or its affiliates. All rights reserved.
- * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
- *
- * This code is free software; you can redistribute it and/or modify it
- * under the terms of the GNU General Public License version 2 only, as
- * published by the Free Software Foundation.
- *
- * This code 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 General Public License
- * version 2 for more details (a copy is included in the LICENSE file that
- * accompanied this code).
- *
- * You should have received a copy of the GNU General Public License version
- * 2 along with this work; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
- *
- * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
- * or visit www.oracle.com if you need additional information or have any
- * questions.
- *
- */
-#ifndef SHARE_VM_UTILITIES_MEM_PTR_ARRAY_HPP
-#define SHARE_VM_UTILITIES_MEM_PTR_ARRAY_HPP
-
-#include "memory/allocation.hpp"
-#include "services/memPtr.hpp"
-
-class MemPtr;
-class MemRecorder;
-class ArenaInfo;
-class MemSnapshot;
-
-extern "C" {
-  typedef int (*FN_SORT)(const void *, const void *);
-}
-
-
-// Memory pointer array interface. This array is used by NMT to hold
-// various memory block information.
-// The memory pointer arrays are usually walked with their iterators.
-
-class MemPointerArray : public CHeapObj<mtNMT> {
- public:
-  virtual ~MemPointerArray() { }
-
-  // return true if it can not allocate storage for the data
-  virtual bool out_of_memory() const = 0;
-  virtual bool is_empty() const = 0;
-  virtual bool is_full() = 0;
-  virtual int  length() const = 0;
-  virtual void clear() = 0;
-  virtual bool append(MemPointer* ptr) = 0;
-  virtual bool insert_at(MemPointer* ptr, int pos) = 0;
-  virtual bool remove_at(int pos) = 0;
-  virtual MemPointer* at(int index) const = 0;
-  virtual void sort(FN_SORT fn) = 0;
-  virtual size_t instance_size() const = 0;
-  virtual bool shrink() = 0;
-
-  NOT_PRODUCT(virtual int capacity() const = 0;)
-};
-
-// Iterator interface
-class MemPointerArrayIterator VALUE_OBJ_CLASS_SPEC {
- public:
-  // return the pointer at current position
-  virtual MemPointer* current() const = 0;
-  // return the next pointer and advance current position
-  virtual MemPointer* next() = 0;
-  // return next pointer without advancing current position
-  virtual MemPointer* peek_next() const = 0;
-  // return previous pointer without changing current position
-  virtual MemPointer* peek_prev() const = 0;
-  // remove the pointer at current position
-  virtual void        remove() = 0;
-  // insert the pointer at current position
-  virtual bool        insert(MemPointer* ptr) = 0;
-  // insert specified element after current position and
-  // move current position to newly inserted position
-  virtual bool        insert_after(MemPointer* ptr) = 0;
-};
-
-// implementation class
-class MemPointerArrayIteratorImpl : public MemPointerArrayIterator {
- protected:
-  MemPointerArray*  _array;
-  int               _pos;
-
- public:
-  MemPointerArrayIteratorImpl(MemPointerArray* arr) {
-    assert(arr != NULL, "Parameter check");
-    _array = arr;
-    _pos = 0;
-  }
-
-  virtual MemPointer* current() const {
-    if (_pos < _array->length()) {
-      return _array->at(_pos);
-    }
-    return NULL;
-  }
-
-  virtual MemPointer* next() {
-    if (_pos + 1 < _array->length()) {
-      return _array->at(++_pos);
-    }
-    _pos = _array->length();
-    return NULL;
-  }
-
-  virtual MemPointer* peek_next() const {
-    if (_pos + 1 < _array->length()) {
-      return _array->at(_pos + 1);
-    }
-    return NULL;
-  }
-
-  virtual MemPointer* peek_prev() const {
-    if (_pos > 0) {
-      return _array->at(_pos - 1);
-    }
-    return NULL;
-  }
-
-  virtual void remove() {
-    if (_pos < _array->length()) {
-      _array->remove_at(_pos);
-    }
-  }
-
-  virtual bool insert(MemPointer* ptr) {
-    return _array->insert_at(ptr, _pos);
-  }
-
-  virtual bool insert_after(MemPointer* ptr) {
-    if (_array->insert_at(ptr, _pos + 1)) {
-      _pos ++;
-      return true;
-    }
-    return false;
-  }
-};
-
-
-
-// Memory pointer array implementation.
-// This implementation implements expandable array
-#define DEFAULT_PTR_ARRAY_SIZE 1024
-
-template <class E> class MemPointerArrayImpl : public MemPointerArray {
- private:
-  int                   _max_size;
-  int                   _size;
-  bool                  _init_elements;
-  E*                    _data;
-
- public:
-  MemPointerArrayImpl(int initial_size = DEFAULT_PTR_ARRAY_SIZE, bool init_elements = true):
-   _max_size(initial_size), _size(0), _init_elements(init_elements) {
-    _data = (E*)raw_allocate(sizeof(E), initial_size);
-    if (_init_elements) {
-      for (int index = 0; index < _max_size; index ++) {
-        ::new ((void*)&_data[index]) E();
-      }
-    }
-  }
-
-  virtual ~MemPointerArrayImpl() {
-    if (_data != NULL) {
-      raw_free(_data);
-    }
-  }
-
- public:
-  bool out_of_memory() const {
-    return (_data == NULL);
-  }
-
-  size_t instance_size() const {
-    return sizeof(MemPointerArrayImpl<E>) + _max_size * sizeof(E);
-  }
-
-  bool is_empty() const {
-    assert(_data != NULL, "Just check");
-    return _size == 0;
-  }
-
-  bool is_full() {
-    assert(_data != NULL, "Just check");
-    if (_size < _max_size) {
-      return false;
-    } else {
-      return !expand_array();
-    }
-  }
-
-  int length() const {
-    assert(_data != NULL, "Just check");
-    return _size;
-  }
-
-  NOT_PRODUCT(int capacity() const { return _max_size; })
-
-  void clear() {
-    assert(_data != NULL, "Just check");
-    _size = 0;
-  }
-
-  bool append(MemPointer* ptr) {
-    assert(_data != NULL, "Just check");
-    if (is_full()) {
-      return false;
-    }
-    _data[_size ++] = *(E*)ptr;
-    return true;
-  }
-
-  bool insert_at(MemPointer* ptr, int pos) {
-    assert(_data != NULL, "Just check");
-    if (is_full()) {
-      return false;
-    }
-    for (int index = _size; index > pos; index --) {
-      _data[index] = _data[index - 1];
-    }
-    _data[pos] = *(E*)ptr;
-    _size ++;
-    return true;
-  }
-
-  bool remove_at(int pos) {
-    assert(_data != NULL, "Just check");
-    if (_size <= pos && pos >= 0) {
-      return false;
-    }
-    -- _size;
-
-    for (int index = pos; index < _size; index ++) {
-      _data[index] = _data[index + 1];
-    }
-    return true;
-  }
-
-  MemPointer* at(int index) const {
-    assert(_data != NULL, "Just check");
-    assert(index >= 0 && index < _size, "illegal index");
-    return &_data[index];
-  }
-
-  bool shrink() {
-    float used = ((float)_size) / ((float)_max_size);
-    if (used < 0.40) {
-      E* old_ptr = _data;
-      int new_size = ((_max_size) / (2 * DEFAULT_PTR_ARRAY_SIZE) + 1) * DEFAULT_PTR_ARRAY_SIZE;
-      _data = (E*)raw_reallocate(_data, sizeof(E), new_size);
-      if (_data == NULL) {
-        _data = old_ptr;
-        return false;
-      } else {
-        _max_size = new_size;
-        return true;
-      }
-    }
-    return false;
-  }
-
-  void sort(FN_SORT fn) {
-    assert(_data != NULL, "Just check");
-    qsort((void*)_data, _size, sizeof(E), fn);
-  }
-
- private:
-  bool  expand_array() {
-    assert(_data != NULL, "Not yet allocated");
-    E* old_ptr = _data;
-    if ((_data = (E*)raw_reallocate((void*)_data, sizeof(E),
-      _max_size + DEFAULT_PTR_ARRAY_SIZE)) == NULL) {
-      _data = old_ptr;
-      return false;
-    } else {
-      _max_size += DEFAULT_PTR_ARRAY_SIZE;
-      if (_init_elements) {
-        for (int index = _size; index < _max_size; index ++) {
-          ::new ((void*)&_data[index]) E();
-        }
-      }
-      return true;
-    }
-  }
-
-  void* raw_allocate(size_t elementSize, int items) {
-    return os::malloc(elementSize * items, mtNMT);
-  }
-
-  void* raw_reallocate(void* ptr, size_t elementSize, int items) {
-    return os::realloc(ptr, elementSize * items, mtNMT);
-  }
-
-  void  raw_free(void* ptr) {
-    os::free(ptr, mtNMT);
-  }
-};
-
-#endif // SHARE_VM_UTILITIES_MEM_PTR_ARRAY_HPP

src/share/vm/services/nmtDCmd.hpp

 /*
- * Copyright (c) 2012, 2013, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2012, 2014, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -25,8 +25,12 @@
 #ifndef SHARE_VM_SERVICES_NMT_DCMD_HPP
 #define SHARE_VM_SERVICES_NMT_DCMD_HPP
 
+#if INCLUDE_NMT
+
 #include "services/diagnosticArgument.hpp"
 #include "services/diagnosticFramework.hpp"
+#include "services/memBaseline.hpp"
+#include "services/mallocTracker.hpp"
 
 /**
  * Native memory tracking DCmd implementation
@@ -39,10 +43,7 @@ class NMTDCmd: public DCmdWithParser {
   DCmdArgument<bool>  _summary_diff;
   DCmdArgument<bool>  _detail_diff;
   DCmdArgument<bool>  _shutdown;
-  DCmdArgument<bool>  _auto_shutdown;
-#ifndef PRODUCT
-  DCmdArgument<bool>  _debug;
-#endif
+  DCmdArgument<bool>  _statistics;
   DCmdArgument<char*> _scale;
 
  public:
@@ -61,6 +62,17 @@ class NMTDCmd: public DCmdWithParser {
   }
   static int num_arguments();
   virtual void execute(DCmdSource source, TRAPS);
+
+ private:
+  void report(bool summaryOnly, size_t scale);
+  void report_diff(bool summaryOnly, size_t scale);
+
+  size_t get_scale(const char* scale) const;
+
+  // check if NMT running at detail tracking level
+  bool check_detail_tracking_level(outputStream* out);
 };
 
+#endif // INCLUDE_NMT
+
 #endif // SHARE_VM_SERVICES_NMT_DCMD_HPP

test/testlibrary/whitebox/sun/hotspot/WhiteBox.java

@@ -93,7 +93,8 @@ public class WhiteBox {
   public native void NMTCommitMemory(long addr, long size);
   public native void NMTUncommitMemory(long addr, long size);
   public native void NMTReleaseMemory(long addr, long size);
-  public native boolean NMTWaitForDataMerge();
+  public native void NMTOverflowHashBucket(long num);
+  public native long NMTMallocWithPseudoStack(long size, int index);
   public native boolean NMTIsDetailSupported();
 
   // Compiler