6797870: Add -XX:+{HeapDump,PrintClassHistogram}{Before,After}FullGC

Summary: Call newly created CollectedHeap::dump_{pre,post}_full_gc before and after every stop-world full collection cycle on GenCollectedHeap and ParallelScavengeHeap. (Support for G1CollectedHeap forthcoming under CR 6810861.) Small modifications to existing heap dumping and class histogram implementation, especially to allow multiple on-the-fly histos/dumps by the VM thread during a single safepoint.
Reviewed-by: jmasa, alanb, mchung

src/share/vm/gc_implementation/parallelScavenge/psMarkSweep.cpp

@@ -125,6 +125,8 @@ void PSMarkSweep::invoke_no_policy(bool clear_all_softrefs) {
     perm_gen->verify_object_start_array();
   }
 
+  heap->pre_full_gc_dump();
+
   // Filled in below to track the state of the young gen after the collection.
   bool eden_empty;
   bool survivors_empty;
@@ -363,6 +365,8 @@ void PSMarkSweep::invoke_no_policy(bool clear_all_softrefs) {
     Universe::print_heap_after_gc();
   }
 
+  heap->post_full_gc_dump();
+
 #ifdef TRACESPINNING
   ParallelTaskTerminator::print_termination_counts();
 #endif

src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp

@@ -1982,6 +1982,8 @@ void PSParallelCompact::invoke_no_policy(bool maximum_heap_compaction) {
     heap->record_gen_tops_before_GC();
   }
 
+  heap->pre_full_gc_dump();
+
   _print_phases = PrintGCDetails && PrintParallelOldGCPhaseTimes;
 
   // Make sure data structures are sane, make the heap parsable, and do other
@@ -2204,6 +2206,8 @@ void PSParallelCompact::invoke_no_policy(bool maximum_heap_compaction) {
     gc_task_manager()->print_task_time_stamps();
   }
 
+  heap->post_full_gc_dump();
+
 #ifdef TRACESPINNING
   ParallelTaskTerminator::print_termination_counts();
 #endif

src/share/vm/gc_implementation/shared/vmGCOperations.cpp

@@ -121,7 +121,7 @@ void VM_GC_HeapInspection::doit() {
     // make the heap parsable (no need to retire TLABs)
     ch->ensure_parsability(false);
   }
-  HeapInspection::heap_inspection(_out);
+  HeapInspection::heap_inspection(_out, _need_prologue /* need_prologue */);
 }
 
 

src/share/vm/gc_implementation/shared/vmGCOperations.hpp

@@ -112,13 +112,16 @@ class VM_GC_HeapInspection: public VM_GC_Operation {
  private:
   outputStream* _out;
   bool _full_gc;
+  bool _need_prologue;
  public:
-  VM_GC_HeapInspection(outputStream* out, bool request_full_gc) :
+  VM_GC_HeapInspection(outputStream* out, bool request_full_gc,
+                       bool need_prologue) :
     VM_GC_Operation(0 /* total collections,      dummy, ignored */,
                     0 /* total full collections, dummy, ignored */,
                     request_full_gc) {
     _out = out;
     _full_gc = request_full_gc;
+    _need_prologue = need_prologue;
   }
 
   ~VM_GC_HeapInspection() {}

src/share/vm/gc_interface/collectedHeap.cpp

@@ -294,3 +294,29 @@ void CollectedHeap::resize_all_tlabs() {
     ThreadLocalAllocBuffer::resize_all_tlabs();
   }
 }
+
+void CollectedHeap::pre_full_gc_dump() {
+  if (HeapDumpBeforeFullGC) {
+    TraceTime tt("Heap Dump: ", PrintGCDetails, false, gclog_or_tty);
+    // We are doing a "major" collection and a heap dump before
+    // major collection has been requested.
+    HeapDumper::dump_heap();
+  }
+  if (PrintClassHistogramBeforeFullGC) {
+    TraceTime tt("Class Histogram: ", PrintGCDetails, true, gclog_or_tty);
+    VM_GC_HeapInspection inspector(gclog_or_tty, false /* ! full gc */, false /* ! prologue */);
+    inspector.doit();
+  }
+}
+
+void CollectedHeap::post_full_gc_dump() {
+  if (HeapDumpAfterFullGC) {
+    TraceTime tt("Heap Dump", PrintGCDetails, false, gclog_or_tty);
+    HeapDumper::dump_heap();
+  }
+  if (PrintClassHistogramAfterFullGC) {
+    TraceTime tt("Class Histogram", PrintGCDetails, true, gclog_or_tty);
+    VM_GC_HeapInspection inspector(gclog_or_tty, false /* ! full gc */, false /* ! prologue */);
+    inspector.doit();
+  }
+}

src/share/vm/gc_interface/collectedHeap.hpp

@@ -514,6 +514,10 @@ class CollectedHeap : public CHeapObj {
   // Perform any cleanup actions necessary before allowing a verification.
   virtual void prepare_for_verify() = 0;
 
+  // Generate any dumps preceding or following a full gc
+  void pre_full_gc_dump();
+  void post_full_gc_dump();
+
   virtual void print() const = 0;
   virtual void print_on(outputStream* st) const = 0;
 

src/share/vm/includeDB_gc

@@ -26,10 +26,12 @@
 
 collectedHeap.cpp                       collectedHeap.hpp
 collectedHeap.cpp                       collectedHeap.inline.hpp
+collectedHeap.cpp                       heapDumper.hpp
 collectedHeap.cpp                       init.hpp
 collectedHeap.cpp                       oop.inline.hpp
 collectedHeap.cpp                       systemDictionary.hpp
 collectedHeap.cpp                       thread_<os_family>.inline.hpp
+collectedHeap.cpp                       vmGCOperations.hpp
 
 collectedHeap.hpp                       allocation.hpp
 collectedHeap.hpp                       barrierSet.hpp

src/share/vm/memory/genCollectedHeap.cpp

@@ -456,6 +456,9 @@ void GenCollectedHeap::do_collection(bool  full,
     int max_level_collected = starting_level;
     for (int i = starting_level; i <= max_level; i++) {
       if (_gens[i]->should_collect(full, size, is_tlab)) {
+        if (i == n_gens() - 1) {  // a major collection is to happen
+          pre_full_gc_dump();    // do any pre full gc dumps
+        }
         // Timer for individual generations. Last argument is false: no CR
         TraceTime t1(_gens[i]->short_name(), PrintGCDetails, false, gclog_or_tty);
         TraceCollectorStats tcs(_gens[i]->counters());
@@ -573,6 +576,10 @@ void GenCollectedHeap::do_collection(bool  full,
     // a whole heap collection.
     complete = complete || (max_level_collected == n_gens() - 1);
 
+    if (complete) { // We did a "major" collection
+      post_full_gc_dump();   // do any post full gc dumps
+    }
+
     if (PrintGCDetails) {
       print_heap_change(gch_prev_used);
 

src/share/vm/memory/heapInspection.cpp

@@ -233,7 +233,7 @@ class RecordInstanceClosure : public ObjectClosure {
   size_t missed_count() { return _missed_count; }
 };
 
-void HeapInspection::heap_inspection(outputStream* st) {
+void HeapInspection::heap_inspection(outputStream* st, bool need_prologue) {
   ResourceMark rm;
   HeapWord* ref;
 
@@ -244,7 +244,9 @@ void HeapInspection::heap_inspection(outputStream* st) {
     case CollectedHeap::GenCollectedHeap: {
       is_shared_heap = true;
       SharedHeap* sh = (SharedHeap*)heap;
-      sh->gc_prologue(false /* !full */); // get any necessary locks, etc.
+      if (need_prologue) {
+        sh->gc_prologue(false /* !full */); // get any necessary locks, etc.
+      }
       ref = sh->perm_gen()->used_region().start();
       break;
     }
@@ -290,7 +292,7 @@ void HeapInspection::heap_inspection(outputStream* st) {
   }
   st->flush();
 
-  if (is_shared_heap) {
+  if (need_prologue && is_shared_heap) {
     SharedHeap* sh = (SharedHeap*)heap;
     sh->gc_epilogue(false /* !full */); // release all acquired locks, etc.
   }

src/share/vm/memory/heapInspection.hpp

@@ -127,6 +127,6 @@ class KlassInfoHisto : public StackObj {
 
 class HeapInspection : public AllStatic {
  public:
-  static void heap_inspection(outputStream* st) KERNEL_RETURN;
+  static void heap_inspection(outputStream* st, bool need_prologue) KERNEL_RETURN;
   static void find_instances_at_safepoint(klassOop k, GrowableArray<oop>* result) KERNEL_RETURN;
 };

src/share/vm/runtime/globals.hpp

@@ -662,6 +662,12 @@ class CommandLineFlags {
   product(ccstrlist, OnOutOfMemoryError, "",                                \
           "Run user-defined commands on first java.lang.OutOfMemoryError")  \
                                                                             \
+  manageable(bool, HeapDumpBeforeFullGC, false,                             \
+          "Dump heap to file before any major stop-world GC")               \
+                                                                            \
+  manageable(bool, HeapDumpAfterFullGC, false,                              \
+          "Dump heap to file after any major stop-world GC")                \
+                                                                            \
   manageable(bool, HeapDumpOnOutOfMemoryError, false,                       \
           "Dump heap to file when java.lang.OutOfMemoryError is thrown")    \
                                                                             \
@@ -1971,6 +1977,12 @@ class CommandLineFlags {
   product(bool, PrintHeapAtSIGBREAK, true,                                  \
           "Print heap layout in response to SIGBREAK")                      \
                                                                             \
+  manageable(bool, PrintClassHistogramBeforeFullGC, false,                  \
+          "Print a class histogram before any major stop-world GC")         \
+                                                                            \
+  manageable(bool, PrintClassHistogramAfterFullGC, false,                   \
+          "Print a class histogram after any major stop-world GC")          \
+                                                                            \
   manageable(bool, PrintClassHistogram, false,                              \
           "Print a histogram of class instances")                           \
                                                                             \

src/share/vm/runtime/os.cpp

@@ -207,7 +207,8 @@ static void signal_thread_entry(JavaThread* thread, TRAPS) {
         VMThread::execute(&op1);
         Universe::print_heap_at_SIGBREAK();
         if (PrintClassHistogram) {
-          VM_GC_HeapInspection op1(gclog_or_tty, true /* force full GC before heap inspection */);
+          VM_GC_HeapInspection op1(gclog_or_tty, true /* force full GC before heap inspection */,
+                                   true /* need_prologue */);
           VMThread::execute(&op1);
         }
         if (JvmtiExport::should_post_data_dump()) {

src/share/vm/services/attachListener.cpp

@@ -194,7 +194,7 @@ static jint heap_inspection(AttachOperation* op, outputStream* out) {
     }
     live_objects_only = strcmp(arg0, "-live") == 0;
   }
-  VM_GC_HeapInspection heapop(out, live_objects_only /* request gc */);
+  VM_GC_HeapInspection heapop(out, live_objects_only /* request full gc */, true /* need_prologue */);
   VMThread::execute(&heapop);
   return JNI_OK;
 }

src/share/vm/services/heapDumper.cpp

@@ -347,7 +347,6 @@ enum {
   INITIAL_CLASS_COUNT = 200
 };
 
-
 // Supports I/O operations on a dump file
 
 class DumpWriter : public StackObj {
@@ -1303,7 +1302,9 @@ void HeapObjectDumper::do_object(oop o) {
 // The VM operation that performs the heap dump
 class VM_HeapDumper : public VM_GC_Operation {
  private:
-  DumpWriter* _writer;
+  static VM_HeapDumper* _global_dumper;
+  static DumpWriter*    _global_writer;
+  DumpWriter*           _local_writer;
   bool _gc_before_heap_dump;
   bool _is_segmented_dump;
   jlong _dump_start;
@@ -1311,8 +1312,20 @@ class VM_HeapDumper : public VM_GC_Operation {
   ThreadStackTrace** _stack_traces;
   int _num_threads;
 
-  // accessors
-  DumpWriter* writer() const                    { return _writer; }
+  // accessors and setters
+  static VM_HeapDumper* dumper()         {  assert(_global_dumper != NULL, "Error"); return _global_dumper; }
+  static DumpWriter* writer()            {  assert(_global_writer != NULL, "Error"); return _global_writer; }
+  void set_global_dumper() {
+    assert(_global_dumper == NULL, "Error");
+    _global_dumper = this;
+  }
+  void set_global_writer() {
+    assert(_global_writer == NULL, "Error");
+    _global_writer = _local_writer;
+  }
+  void clear_global_dumper() { _global_dumper = NULL; }
+  void clear_global_writer() { _global_writer = NULL; }
+
   bool is_segmented_dump() const                { return _is_segmented_dump; }
   void set_segmented_dump()                     { _is_segmented_dump = true; }
   jlong dump_start() const                      { return _dump_start; }
@@ -1357,7 +1370,7 @@ class VM_HeapDumper : public VM_GC_Operation {
     VM_GC_Operation(0 /* total collections,      dummy, ignored */,
                     0 /* total full collections, dummy, ignored */,
                     gc_before_heap_dump) {
-    _writer = writer;
+    _local_writer = writer;
     _gc_before_heap_dump = gc_before_heap_dump;
     _is_segmented_dump = false;
     _dump_start = (jlong)-1;
@@ -1381,6 +1394,9 @@ class VM_HeapDumper : public VM_GC_Operation {
   void doit();
 };
 
+VM_HeapDumper* VM_HeapDumper::_global_dumper = NULL;
+DumpWriter*    VM_HeapDumper::_global_writer = NULL;
+
 bool VM_HeapDumper::skip_operation() const {
   return false;
 }
@@ -1479,31 +1495,28 @@ void HeapObjectDumper::mark_end_of_record() {
 void VM_HeapDumper::do_load_class(klassOop k) {
   static u4 class_serial_num = 0;
 
-  VM_HeapDumper* dumper = ((VM_HeapDumper*)VMThread::vm_operation());
-  DumpWriter* writer = dumper->writer();
-
   // len of HPROF_LOAD_CLASS record
   u4 remaining = 2*oopSize + 2*sizeof(u4);
 
   // write a HPROF_LOAD_CLASS for the class and each array class
   do {
-    DumperSupport::write_header(writer, HPROF_LOAD_CLASS, remaining);
+    DumperSupport::write_header(writer(), HPROF_LOAD_CLASS, remaining);
 
     // class serial number is just a number
-    writer->write_u4(++class_serial_num);
+    writer()->write_u4(++class_serial_num);
 
     // class ID
     Klass* klass = Klass::cast(k);
-    writer->write_classID(klass);
+    writer()->write_classID(klass);
 
     // add the klassOop and class serial number pair
-    dumper->add_class_serial_number(klass, class_serial_num);
+    dumper()->add_class_serial_number(klass, class_serial_num);
 
-    writer->write_u4(STACK_TRACE_ID);
+    writer()->write_u4(STACK_TRACE_ID);
 
     // class name ID
     symbolOop name = klass->name();
-    writer->write_objectID(name);
+    writer()->write_objectID(name);
 
     // write a LOAD_CLASS record for the array type (if it exists)
     k = klass->array_klass_or_null();
@@ -1512,17 +1525,13 @@ void VM_HeapDumper::do_load_class(klassOop k) {
 
 // writes a HPROF_GC_CLASS_DUMP record for the given class
 void VM_HeapDumper::do_class_dump(klassOop k) {
-  VM_HeapDumper* dumper = ((VM_HeapDumper*)VMThread::vm_operation());
-  DumpWriter* writer = dumper->writer();
-  DumperSupport::dump_class_and_array_classes(writer, k);
+  DumperSupport::dump_class_and_array_classes(writer(), k);
 }
 
 // writes a HPROF_GC_CLASS_DUMP records for a given basic type
 // array (and each multi-dimensional array too)
 void VM_HeapDumper::do_basic_type_array_class_dump(klassOop k) {
-  VM_HeapDumper* dumper = ((VM_HeapDumper*)VMThread::vm_operation());
-  DumpWriter* writer = dumper->writer();
-  DumperSupport::dump_basic_type_array_class(writer, k);
+  DumperSupport::dump_basic_type_array_class(writer(), k);
 }
 
 // Walk the stack of the given thread.
@@ -1658,6 +1667,11 @@ void VM_HeapDumper::doit() {
     ch->ensure_parsability(false);
   }
 
+  // At this point we should be the only dumper active, so
+  // the following should be safe.
+  set_global_dumper();
+  set_global_writer();
+
   // Write the file header - use 1.0.2 for large heaps, otherwise 1.0.1
   size_t used = ch->used();
   const char* header;
@@ -1667,6 +1681,7 @@ void VM_HeapDumper::doit() {
   } else {
     header = "JAVA PROFILE 1.0.1";
   }
+
   // header is few bytes long - no chance to overflow int
   writer()->write_raw((void*)header, (int)strlen(header));
   writer()->write_u1(0); // terminator
@@ -1723,6 +1738,10 @@ void VM_HeapDumper::doit() {
   // fixes up the length of the dump record. In the case of a segmented
   // heap then the HPROF_HEAP_DUMP_END record is also written.
   end_of_dump();
+
+  // Now we clear the global variables, so that a future dumper might run.
+  clear_global_dumper();
+  clear_global_writer();
 }
 
 void VM_HeapDumper::dump_stack_traces() {
@@ -1790,7 +1809,12 @@ int HeapDumper::dump(const char* path) {
 
   // generate the dump
   VM_HeapDumper dumper(&writer, _gc_before_heap_dump);
-  VMThread::execute(&dumper);
+  if (Thread::current()->is_VM_thread()) {
+    assert(SafepointSynchronize::is_at_safepoint(), "Expected to be called at a safepoint");
+    dumper.doit();
+  } else {
+    VMThread::execute(&dumper);
+  }
 
   // close dump file and record any error that the writer may have encountered
   writer.close();
@@ -1845,49 +1869,68 @@ void HeapDumper::set_error(char* error) {
   }
 }
 
-
-// Called by error reporting
+// Called by error reporting by a single Java thread outside of a JVM safepoint,
+// or by heap dumping by the VM thread during a (GC) safepoint. Thus, these various
+// callers are strictly serialized and guaranteed not to interfere below. For more
+// general use, however, this method will need modification to prevent
+// inteference when updating the static variables base_path and dump_file_seq below.
 void HeapDumper::dump_heap() {
-  static char path[JVM_MAXPATHLEN];
+  static char base_path[JVM_MAXPATHLEN] = {'\0'};
+  static uint dump_file_seq = 0;
+  char   my_path[JVM_MAXPATHLEN] = {'\0'};
 
   // The dump file defaults to java_pid<pid>.hprof in the current working
   // directory. HeapDumpPath=<file> can be used to specify an alternative
   // dump file name or a directory where dump file is created.
-  bool use_default_filename = true;
-  if (HeapDumpPath == NULL || HeapDumpPath[0] == '\0') {
-    path[0] = '\0'; // HeapDumpPath=<file> not specified
-  } else {
-    assert(strlen(HeapDumpPath) < sizeof(path), "HeapDumpPath too long");
-    strcpy(path, HeapDumpPath);
-    // check if the path is a directory (must exist)
-    DIR* dir = os::opendir(path);
-    if (dir == NULL) {
-      use_default_filename = false;
+  if (dump_file_seq == 0) { // first time in, we initialize base_path
+    bool use_default_filename = true;
+    if (HeapDumpPath == NULL || HeapDumpPath[0] == '\0') {
+      // HeapDumpPath=<file> not specified
     } else {
-      // HeapDumpPath specified a directory. We append a file separator
-      // (if needed).
-      os::closedir(dir);
-      size_t fs_len = strlen(os::file_separator());
-      if (strlen(path) >= fs_len) {
-        char* end = path;
-        end += (strlen(path) - fs_len);
-        if (strcmp(end, os::file_separator()) != 0) {
-          assert(strlen(path) + strlen(os::file_separator()) < sizeof(path),
-            "HeapDumpPath too long");
-          strcat(path, os::file_separator());
+      assert(strlen(HeapDumpPath) < sizeof(base_path), "HeapDumpPath too long");
+      strcpy(base_path, HeapDumpPath);
+      // check if the path is a directory (must exist)
+      DIR* dir = os::opendir(base_path);
+      if (dir == NULL) {
+        use_default_filename = false;
+      } else {
+        // HeapDumpPath specified a directory. We append a file separator
+        // (if needed).
+        os::closedir(dir);
+        size_t fs_len = strlen(os::file_separator());
+        if (strlen(base_path) >= fs_len) {
+          char* end = base_path;
+          end += (strlen(base_path) - fs_len);
+          if (strcmp(end, os::file_separator()) != 0) {
+            assert(strlen(base_path) + strlen(os::file_separator()) < sizeof(base_path),
+              "HeapDumpPath too long");
+            strcat(base_path, os::file_separator());
+          }
         }
       }
     }
+    // If HeapDumpPath wasn't a file name then we append the default name
+    if (use_default_filename) {
+      char fn[32];
+      sprintf(fn, "java_pid%d", os::current_process_id());
+      assert(strlen(base_path) + strlen(fn) < sizeof(base_path), "HeapDumpPath too long");
+      strcat(base_path, fn);
+    }
+    assert(strlen(base_path) < sizeof(my_path), "Buffer too small");
+    strcpy(my_path, base_path);
+  } else {
+    // Append a sequence number id for dumps following the first
+    char fn[33];
+    sprintf(fn, ".%d", dump_file_seq);
+    assert(strlen(base_path) + strlen(fn) < sizeof(my_path), "HeapDumpPath too long");
+    strcpy(my_path, base_path);
+    strcat(my_path, fn);
   }
-  // If HeapDumpPath wasn't a file name then we append the default name
-  if (use_default_filename) {
-    char fn[32];
-    sprintf(fn, "java_pid%d.hprof", os::current_process_id());
-    assert(strlen(path) + strlen(fn) < sizeof(path), "HeapDumpPath too long");
-    strcat(path, fn);
-  }
+  dump_file_seq++;   // increment seq number for next time we dump
+  assert(strlen(".hprof") + strlen(my_path) < sizeof(my_path), "HeapDumpPath too long");
+  strcat(my_path, ".hprof");
 
   HeapDumper dumper(false /* no GC before heap dump */,
                     true  /* send to tty */);
-  dumper.dump(path);
+  dumper.dump(my_path);
 }

src/share/vm/services/heapDumper.hpp

@@ -53,7 +53,7 @@ class HeapDumper : public StackObj {
 
  public:
   HeapDumper(bool gc_before_heap_dump) :
-    _gc_before_heap_dump(gc_before_heap_dump), _error(NULL), _print_to_tty(false)  { }
+    _gc_before_heap_dump(gc_before_heap_dump), _error(NULL), _print_to_tty(false) { }
   HeapDumper(bool gc_before_heap_dump, bool print_to_tty) :
     _gc_before_heap_dump(gc_before_heap_dump), _error(NULL), _print_to_tty(print_to_tty) { }