8046598: Scalable Native memory tracking development

Enhance scalability of native memory tracking

Reviewed-by: coleenp, ctornqvi, gtriantafill

hotspot/src/share/vm/services/mallocSiteTable.hpp

@@ -0,0 +1,268 @@
+/*
+ * 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