implement primitive heap dump facility for memory profiling, accessible via Runtime.dumpHeap

The proper way to do this is to implement a subset of JVMTI, but this
will do the job for now.

classpath/java/lang/Runtime.java

@@ -75,6 +75,8 @@ public class Runtime {
 
   public native long totalMemory();
 
+  public static native void dumpHeap(String outputFile);
+
   private static class MyProcess extends Process {
     private long pid;
     private final int in;

classpath/java/lang/reflect/Method.java

@@ -111,8 +111,8 @@ public class Method<T> extends AccessibleObject implements Member {
     }
   }
 
-  public static native Object invoke(Method method, Object instance,
-                                     Object ... arguments)
+  private static native Object invoke(Method method, Object instance,
+                                      Object ... arguments)
     throws InvocationTargetException, IllegalAccessException;
 
   public Class getReturnType() {

makefile

@@ -204,6 +204,11 @@ vm-sources = \
 	$(src)/process.cpp \
 	$(src)/$(asm).cpp
 
+ifeq ($(heapdump),true)
+	vm-sources += $(src)/heapdump.cpp
+	cflags += -DAVIAN_HEAPDUMP
+endif
+
 vm-asm-sources = $(src)/$(asm).S
 
 ifeq ($(process),compile)

src/builtin.cpp

@@ -585,6 +585,26 @@ Java_java_lang_Runtime_gc(Thread* t, jobject)
   collect(t, Heap::MajorCollection);
 }
 
+#ifdef AVIAN_HEAPDUMP
+
+extern "C" JNIEXPORT void JNICALL
+Java_java_lang_Runtime_dumpHeap(Thread* t, jclass, jstring outputFile)
+{
+  unsigned length = stringLength(t, *outputFile);
+  char n[length + 1];
+  stringChars(t, *outputFile, n);
+  FILE* out = fopen(n, "wb");
+  if (out) {
+    dumpHeap(t, out);
+    fclose(out);
+  } else {
+    object message = makeString(t, "file not found: %s", n);
+    t->exception = makeRuntimeException(t, message);
+  }
+}
+
+#endif//AVIAN_HEAPDUMP
+
 extern "C" JNIEXPORT void JNICALL
 Java_java_lang_Runtime_exit(Thread* t, jobject, jint code)
 {

src/heapdump.cpp

@@ -0,0 +1,331 @@
+/* Copyright (c) 2008, Avian Contributors
+
+   Permission to use, copy, modify, and/or distribute this software
+   for any purpose with or without fee is hereby granted, provided
+   that the above copyright notice and this permission notice appear
+   in all copies.
+
+   There is NO WARRANTY for this software.  See license.txt for
+   details. */
+
+#include "machine.h"
+
+using namespace vm;
+
+namespace {
+
+const uintptr_t PointerShift = log(BytesPerWord);
+
+class Set {
+ public:
+  class Entry {
+   public:
+    object value;
+    uint32_t number;
+    int next;
+  };
+
+  static unsigned footprint(unsigned capacity) {
+    return sizeof(Set)
+      + pad(sizeof(int) * capacity)
+      + pad(sizeof(Set::Entry) * capacity);
+  }
+
+  Set(unsigned capacity):
+    size(0),
+    capacity(capacity),
+    index(reinterpret_cast<int*>
+          (reinterpret_cast<uint8_t*>(this)
+           + sizeof(Set))),
+    entries(reinterpret_cast<Entry*>
+            (reinterpret_cast<uint8_t*>(index) 
+             + pad(sizeof(int) * capacity)))
+  { }
+
+  unsigned size;
+  unsigned capacity;
+  int* index;
+  Entry* entries;
+};
+
+class Stack {
+ public:
+  class Entry {
+   public:
+    object value;
+    int offset;
+  };
+
+  static const unsigned Capacity = 4096;
+
+  Stack(Stack* next): next(next), entryCount(0) { }
+
+  Stack* next;
+  unsigned entryCount;
+  Entry entries[Capacity];
+};
+
+class Context {
+ public:
+  Context(Thread* thread, FILE* out):
+    thread(thread), out(out), objects(0), stack(0), nextNumber(1)
+  { }
+
+  ~Context() {
+    if (objects) {
+      thread->m->heap->free(objects, Set::footprint(objects->capacity));
+    }
+    while (stack) {
+      Stack* dead = stack;
+      stack = dead->next;
+      thread->m->heap->free(stack, sizeof(Stack));
+    }
+  }
+
+  Thread* thread;
+  FILE* out;
+  Set* objects;
+  Stack* stack;
+  uint32_t nextNumber;
+};
+
+void
+push(Context* c, object p, int offset)
+{
+  if (c->stack == 0 or c->stack->entryCount == Stack::Capacity) {
+    c->stack = new (c->thread->m->heap->allocate(sizeof(Stack)))
+      Stack(c->stack);
+  }
+  Stack::Entry* e = c->stack->entries + (c->stack->entryCount++);
+  e->value = p;
+  e->offset = offset;
+}
+
+bool
+pop(Context* c, object* p, int* offset)
+{
+  if (c->stack) {
+    if (c->stack->entryCount == 0) {
+      if (c->stack->next) {
+        Stack* dead = c->stack;
+        c->stack = dead->next;
+        c->thread->m->heap->free(dead, sizeof(Stack));
+      } else {
+        return false;
+      }
+    }
+    Stack::Entry* e = c->stack->entries + (--c->stack->entryCount);
+    *p = e->value;
+    *offset = e->offset;
+    return true;
+  } else {
+    return false;
+  }
+}
+
+unsigned
+hash(object p, unsigned capacity)
+{
+  return (reinterpret_cast<uintptr_t>(p) >> PointerShift)
+    & (capacity - 1);
+}
+
+Set::Entry*
+find(Context* c, object p)
+{
+  if (c->objects == 0) return false;
+
+  for (int i = c->objects->index[hash(p, c->objects->capacity)]; i >= 0;) {
+    Set::Entry* e = c->objects->entries + i;
+    if (e->value == p) {
+      return e;
+    }
+    i = e->next;
+  }
+
+  return false;
+}
+
+Set::Entry*
+add(Context* c UNUSED, Set* set, object p)
+{
+  assert(c->thread, set->size < set->capacity);
+
+  unsigned index = hash(p, set->capacity);
+
+  int offset = set->size++;
+  Set::Entry* e = set->entries + offset;
+  e->value = p;
+  e->next = set->index[index];
+  set->index[index] = offset;
+  return e;
+}
+
+Set::Entry*
+add(Context* c, object p)
+{
+  if (c->objects == 0 or c->objects->size == c->objects->capacity) {
+    unsigned capacity;
+    if (c->objects) {
+      capacity = c->objects->capacity * 2;
+    } else {
+      capacity = 4096; // must be power of two
+    }
+
+    Set* set = new (c->thread->m->heap->allocate(Set::footprint(capacity)))
+      Set(capacity);
+
+    memset(set->index, 0xFF, sizeof(int) * capacity);
+
+    if (c->objects) {
+      for (unsigned i = 0; i < c->objects->capacity; ++i) {
+        for (int j = c->objects->index[i]; j >= 0;) {
+          Set::Entry* e = c->objects->entries + j;
+          add(c, set, e->value);
+          j = e->next;
+        }
+      }
+
+      c->thread->m->heap->free
+        (c->objects, Set::footprint(c->objects->capacity));
+    }
+
+    c->objects = set;
+  }
+
+  return add(c, c->objects, p);
+}
+
+enum {
+  Root,
+  ClassName,
+  Push,
+  LastChild,
+  Pop,
+  Size
+};
+
+inline object
+get(object o, unsigned offsetInWords)
+{
+  return static_cast<object>
+    (mask(cast<void*>(o, offsetInWords * BytesPerWord)));
+}
+
+void
+write1(Context* c, uint8_t v)
+{
+  fwrite(&v, 1, 1, c->out);
+}
+
+void
+write4(Context* c, uint32_t v)
+{
+  uint8_t b[] = { v >> 24, (v >> 16) & 0xFF, (v >> 8) & 0xFF, v & 0xFF };
+  fwrite(b, 4, 1, c->out);
+}
+
+void
+writeString(Context* c, int8_t* p, unsigned size)
+{
+  write4(c, size);
+  fwrite(p, size, 1, c->out);
+}
+
+unsigned
+objectSize(Thread* t, object o)
+{
+  unsigned n = baseSize(t, o, objectClass(t, o));
+  if (objectExtended(t, o)) {
+    ++ n;
+  }
+  return n;
+}
+
+
+
+void
+visit(Context* c, object p)
+{
+  Thread* t = c->thread;
+  int nextChildOffset;
+
+  write1(c, Root);
+
+ visit: {
+    Set::Entry* e = find(c, p);
+    if (e) {
+      write4(c, e->number);
+    } else {
+      e = add(c, p);
+      e->number = c->nextNumber++;
+
+      write4(c, e->number);
+
+      write1(c, Size);
+      write4(c, objectSize(t, p));
+
+      if (objectClass(t, p) == arrayBody(t, t->m->types, Machine::ClassType)) {
+        object name = className(t, p);
+        if (name) {
+          write1(c, ClassName);
+          writeString(c, &byteArrayBody(t, name, 0),
+                      byteArrayLength(t, name) - 1);
+        }
+      }
+
+      nextChildOffset = walkNext(t, p, -1);
+      if (nextChildOffset != -1) {
+        goto children;
+      }
+    }
+  }
+
+  goto pop;
+
+ children: {
+    int next = walkNext(t, p, nextChildOffset);
+    if (next >= 0) {
+      write1(c, Push);
+      push(c, p, next);
+    } else {
+      write1(c, LastChild);
+    }
+    p = get(p, nextChildOffset);
+    goto visit;
+  }
+
+ pop: {
+    if (pop(c, &p, &nextChildOffset)) {
+      write1(c, Pop);
+      goto children;
+    }
+  }
+}
+
+} // namespace
+
+namespace vm {
+
+void
+dumpHeap(Thread* t, FILE* out)
+{
+  Context context(t, out);
+
+  class Visitor : public Heap::Visitor {
+   public:
+    Visitor(Context* c): c(c) { }
+
+    virtual void visit(void* p) {
+      ::visit(c, static_cast<object>(mask(*static_cast<void**>(p))));
+    }
+
+    Context* c;
+  } v(&context);
+
+  add(&context, 0)->number = 0;
+
+  visitRoots(t, &v);
+}
+
+} // namespace vm

src/machine.cpp

@@ -177,34 +177,15 @@ footprint(Thread* t)
   return n;
 }
 
-void
-visitRoots(Thread* t, Heap::Visitor* v)
-{
-  if (t->state != Thread::ZombieState) {
-    v->visit(&(t->javaThread));
-    v->visit(&(t->exception));
-
-    t->m->processor->visitObjects(t, v);
-
-    for (Thread::Protector* p = t->protector; p; p = p->next) {
-      p->visit(v);
-    }
-  }
-
-  for (Thread* c = t->child; c; c = c->peer) {
-    visitRoots(c, v);
-  }
-}
-
 void
 walk(Thread*, Heap::Walker* w, uint32_t* mask, unsigned fixedSize,
-     unsigned arrayElementSize, unsigned arrayLength)
+     unsigned arrayElementSize, unsigned arrayLength, unsigned start)
 {
   unsigned fixedSizeInWords = ceiling(fixedSize, BytesPerWord);
   unsigned arrayElementSizeInWords
     = ceiling(arrayElementSize, BytesPerWord);
 
-  for (unsigned i = 0; i < fixedSizeInWords; ++i) {
+  for (unsigned i = start; i < fixedSizeInWords; ++i) {
     if (mask[i / 32] & (static_cast<uint32_t>(1) << (i % 32))) {
       if (not w->visit(i)) {
         return;
@@ -222,8 +203,19 @@ walk(Thread*, Heap::Walker* w, uint32_t* mask, unsigned fixedSize,
   }
 
   if (arrayObjectElements) {
-    for (unsigned i = 0; i < arrayLength; ++i) {
-      for (unsigned j = 0; j < arrayElementSizeInWords; ++j) {
+    unsigned arrayStart;
+    unsigned elementStart;
+    if (start > fixedSizeInWords) {
+      unsigned s = start - fixedSizeInWords;
+      arrayStart = s / arrayElementSizeInWords;
+      elementStart = s % arrayElementSizeInWords;
+    } else {
+      arrayStart = 0;
+      elementStart = 0;
+    }
+
+    for (unsigned i = arrayStart; i < arrayLength; ++i) {
+      for (unsigned j = elementStart; j < arrayElementSizeInWords; ++j) {
         unsigned k = fixedSizeInWords + j;
         if (mask[k / 32] & (static_cast<uint32_t>(1) << (k % 32))) {
           if (not w->visit
@@ -238,7 +230,7 @@ walk(Thread*, Heap::Walker* w, uint32_t* mask, unsigned fixedSize,
 }
 
 void
-walk(Thread* t, Heap::Walker* w, object o)
+walk(Thread* t, Heap::Walker* w, object o, unsigned start)
 {
   object class_ = static_cast<object>(t->m->heap->follow(objectClass(t, o)));
   object objectMask = static_cast<object>
@@ -255,16 +247,16 @@ walk(Thread* t, Heap::Walker* w, object o)
     memcpy(mask, &intArrayBody(t, objectMask, 0),
            intArrayLength(t, objectMask) * 4);
 
-    walk(t, w, mask, fixedSize, arrayElementSize, arrayLength);
+    walk(t, w, mask, fixedSize, arrayElementSize, arrayLength, start);
   } else if (classVmFlags(t, class_) & SingletonFlag) {
     unsigned length = singletonLength(t, o);
     if (length) {
       walk(t, w, singletonMask(t, o),
-           (singletonCount(t, o) + 2) * BytesPerWord, 0, 0);
-    } else {
+           (singletonCount(t, o) + 2) * BytesPerWord, 0, 0, start);
+    } else if (start == 0) {
       w->visit(0);
     }
-  } else {
+  } else if (start == 0) {
     w->visit(0);
   }
 }
@@ -1616,7 +1608,7 @@ class HeapClient: public Heap::Client {
 
   virtual void walk(void* p, Heap::Walker* w) {
     object o = static_cast<object>(m->heap->follow(mask(p)));
-    ::walk(m->rootThread, w, o);
+    ::walk(m->rootThread, w, o, 0);
   }
 
   void dispose() {
@@ -2772,6 +2764,44 @@ collect(Thread* t, Heap::CollectionType type)
 #endif
 }
 
+int
+walkNext(Thread* t, object o, int previous)
+{
+  class Walker: public Heap::Walker {
+   public:
+    Walker(): value(-1) { }
+
+    bool visit(unsigned offset) {
+      value = offset;
+      return false;
+    }
+
+    int value;
+  } walker;
+
+  walk(t, &walker, o, previous + 1);
+  return walker.value;
+}
+
+void
+visitRoots(Thread* t, Heap::Visitor* v)
+{
+  if (t->state != Thread::ZombieState) {
+    v->visit(&(t->javaThread));
+    v->visit(&(t->exception));
+
+    t->m->processor->visitObjects(t, v);
+
+    for (Thread::Protector* p = t->protector; p; p = p->next) {
+      p->visit(v);
+    }
+  }
+
+  for (Thread* c = t->child; c; c = c->peer) {
+    visitRoots(c, v);
+  }
+}
+
 void
 printTrace(Thread* t, object exception)
 {

src/machine.h

@@ -2204,6 +2204,12 @@ intern(Thread* t, object s);
 void
 exit(Thread* t);
 
+int
+walkNext(Thread* t, object o, int previous);
+
+void
+visitRoots(Thread* t, Heap::Visitor* v);
+
 inline jobject
 makeLocalReference(Thread* t, object o)
 {
@@ -2297,6 +2303,9 @@ makeSingleton(Thread* t, unsigned count)
   return o;
 }
 
+void
+dumpHeap(Thread* t, FILE* out);
+
 } // namespace vm
 
 void

test/Hello.java

@@ -1,5 +1,11 @@
 public class Hello {
   public static void main(String[] args) {
     System.out.println("hello, world!");
+    try {
+      Runtime.class.getMethod("dumpHeap", String.class)
+        .invoke(null, "/tmp/heap.bin");
+    } catch (Exception e) {
+      e.printStackTrace();
+    }
   }
 }