6857194: Add hotspot perf counters to aid class loading performance measurement

Summary: Add new jvmstat counters to measure detailed class loading time
Reviewed-by: acorn, kamg

src/share/vm/classfile/classFileParser.cpp

@@ -547,7 +547,6 @@ objArrayHandle ClassFileParser::parse_interfaces(constantPoolHandle cp,
                                                  int length,
                                                  Handle class_loader,
                                                  Handle protection_domain,
-                                                 PerfTraceTime* vmtimer,
                                                  symbolHandle class_name,
                                                  TRAPS) {
   ClassFileStream* cfs = stream();
@@ -575,13 +574,11 @@ objArrayHandle ClassFileParser::parse_interfaces(constantPoolHandle cp,
       guarantee_property(unresolved_klass->byte_at(0) != JVM_SIGNATURE_ARRAY,
                          "Bad interface name in class file %s", CHECK_(nullHandle));
 
-      vmtimer->suspend();  // do not count recursive loading twice
       // Call resolve_super so classcircularity is checked
       klassOop k = SystemDictionary::resolve_super_or_fail(class_name,
                     unresolved_klass, class_loader, protection_domain,
                     false, CHECK_(nullHandle));
       interf = KlassHandle(THREAD, k);
-      vmtimer->resume();
 
       if (LinkWellKnownClasses)  // my super type is well known to me
         cp->klass_at_put(interface_index, interf()); // eagerly resolve
@@ -2558,7 +2555,15 @@ instanceKlassHandle ClassFileParser::parseClassFile(symbolHandle name,
 
   ClassFileStream* cfs = stream();
   // Timing
-  PerfTraceTime vmtimer(ClassLoader::perf_accumulated_time());
+  assert(THREAD->is_Java_thread(), "must be a JavaThread");
+  JavaThread* jt = (JavaThread*) THREAD;
+
+  PerfClassTraceTime ctimer(ClassLoader::perf_class_parse_time(),
+                            ClassLoader::perf_class_parse_selftime(),
+                            NULL,
+                            jt->get_thread_stat()->perf_recursion_counts_addr(),
+                            jt->get_thread_stat()->perf_timers_addr(),
+                            PerfClassTraceTime::PARSE_CLASS);
 
   _has_finalizer = _has_empty_finalizer = _has_vanilla_constructor = false;
 
@@ -2738,7 +2743,7 @@ instanceKlassHandle ClassFileParser::parseClassFile(symbolHandle name,
     if (itfs_len == 0) {
       local_interfaces = objArrayHandle(THREAD, Universe::the_empty_system_obj_array());
     } else {
-      local_interfaces = parse_interfaces(cp, itfs_len, class_loader, protection_domain, &vmtimer, _class_name, CHECK_(nullHandle));
+      local_interfaces = parse_interfaces(cp, itfs_len, class_loader, protection_domain, _class_name, CHECK_(nullHandle));
     }
 
     // Fields (offsets are filled in later)
@@ -2782,6 +2787,7 @@ instanceKlassHandle ClassFileParser::parseClassFile(symbolHandle name,
                                                            protection_domain,
                                                            true,
                                                            CHECK_(nullHandle));
+
       KlassHandle kh (THREAD, k);
       super_klass = instanceKlassHandle(THREAD, kh());
       if (LinkWellKnownClasses)  // my super class is well known to me

src/share/vm/classfile/classFileParser.hpp

@@ -61,7 +61,6 @@ class ClassFileParser VALUE_OBJ_CLASS_SPEC {
                                   int length,
                                   Handle class_loader,
                                   Handle protection_domain,
-                                  PerfTraceTime* vmtimer,
                                   symbolHandle class_name,
                                   TRAPS);
 

src/share/vm/classfile/classLoader.cpp

@@ -48,9 +48,26 @@ static canonicalize_fn_t CanonicalizeEntry  = NULL;
 PerfCounter*    ClassLoader::_perf_accumulated_time = NULL;
 PerfCounter*    ClassLoader::_perf_classes_inited = NULL;
 PerfCounter*    ClassLoader::_perf_class_init_time = NULL;
+PerfCounter*    ClassLoader::_perf_class_init_selftime = NULL;
+PerfCounter*    ClassLoader::_perf_classes_verified = NULL;
 PerfCounter*    ClassLoader::_perf_class_verify_time = NULL;
+PerfCounter*    ClassLoader::_perf_class_verify_selftime = NULL;
 PerfCounter*    ClassLoader::_perf_classes_linked = NULL;
 PerfCounter*    ClassLoader::_perf_class_link_time = NULL;
+PerfCounter*    ClassLoader::_perf_class_link_selftime = NULL;
+PerfCounter*    ClassLoader::_perf_class_parse_time = NULL;
+PerfCounter*    ClassLoader::_perf_class_parse_selftime = NULL;
+PerfCounter*    ClassLoader::_perf_sys_class_lookup_time = NULL;
+PerfCounter*    ClassLoader::_perf_shared_classload_time = NULL;
+PerfCounter*    ClassLoader::_perf_sys_classload_time = NULL;
+PerfCounter*    ClassLoader::_perf_app_classload_time = NULL;
+PerfCounter*    ClassLoader::_perf_app_classload_selftime = NULL;
+PerfCounter*    ClassLoader::_perf_app_classload_count = NULL;
+PerfCounter*    ClassLoader::_perf_define_appclasses = NULL;
+PerfCounter*    ClassLoader::_perf_define_appclass_time = NULL;
+PerfCounter*    ClassLoader::_perf_define_appclass_selftime = NULL;
+PerfCounter*    ClassLoader::_perf_app_classfile_bytes_read = NULL;
+PerfCounter*    ClassLoader::_perf_sys_classfile_bytes_read = NULL;
 PerfCounter*    ClassLoader::_sync_systemLoaderLockContentionRate = NULL;
 PerfCounter*    ClassLoader::_sync_nonSystemLoaderLockContentionRate = NULL;
 PerfCounter*    ClassLoader::_sync_JVMFindLoadedClassLockFreeCounter = NULL;
@@ -152,6 +169,9 @@ ClassFileStream* ClassPathDirEntry::open_stream(const char* name) {
       hpi::close(file_handle);
       // construct ClassFileStream
       if (num_read == (size_t)st.st_size) {
+        if (UsePerfData) {
+          ClassLoader::perf_sys_classfile_bytes_read()->inc(num_read);
+        }
         return new ClassFileStream(buffer, st.st_size, _dir);    // Resource allocated
       }
     }
@@ -198,6 +218,9 @@ ClassFileStream* ClassPathZipEntry::open_stream(const char* name) {
       buffer     = NEW_RESOURCE_ARRAY(u1, filesize);
       if (!(*ReadEntry)(_zip, entry, buffer, filename)) return NULL;
   }
+  if (UsePerfData) {
+    ClassLoader::perf_sys_classfile_bytes_read()->inc(filesize);
+  }
   // return result
   return new ClassFileStream(buffer, filesize, _zip_name);    // Resource allocated
 }
@@ -825,7 +848,9 @@ instanceKlassHandle ClassLoader::load_classfile(symbolHandle h_name, TRAPS) {
   ClassFileStream* stream = NULL;
   int classpath_index = 0;
   {
-    PerfTraceTime vmtimer(perf_accumulated_time());
+    PerfClassTraceTime vmtimer(perf_sys_class_lookup_time(),
+                               ((JavaThread*) THREAD)->get_thread_stat()->perf_timers_addr(),
+                               PerfClassTraceTime::CLASS_LOAD);
     ClassPathEntry* e = _first_entry;
     while (e != NULL) {
       stream = e->open_stream(name);
@@ -890,11 +915,29 @@ void ClassLoader::initialize() {
     // jvmstat performance counters
     NEWPERFTICKCOUNTER(_perf_accumulated_time, SUN_CLS, "time");
     NEWPERFTICKCOUNTER(_perf_class_init_time, SUN_CLS, "classInitTime");
+    NEWPERFTICKCOUNTER(_perf_class_init_selftime, SUN_CLS, "classInitTime.self");
     NEWPERFTICKCOUNTER(_perf_class_verify_time, SUN_CLS, "classVerifyTime");
+    NEWPERFTICKCOUNTER(_perf_class_verify_selftime, SUN_CLS, "classVerifyTime.self");
     NEWPERFTICKCOUNTER(_perf_class_link_time, SUN_CLS, "classLinkedTime");
-
+    NEWPERFTICKCOUNTER(_perf_class_link_selftime, SUN_CLS, "classLinkedTime.self");
     NEWPERFEVENTCOUNTER(_perf_classes_inited, SUN_CLS, "initializedClasses");
     NEWPERFEVENTCOUNTER(_perf_classes_linked, SUN_CLS, "linkedClasses");
+    NEWPERFEVENTCOUNTER(_perf_classes_verified, SUN_CLS, "verifiedClasses");
+
+    NEWPERFTICKCOUNTER(_perf_class_parse_time, SUN_CLS, "parseClassTime");
+    NEWPERFTICKCOUNTER(_perf_class_parse_selftime, SUN_CLS, "parseClassTime.self");
+    NEWPERFTICKCOUNTER(_perf_sys_class_lookup_time, SUN_CLS, "lookupSysClassTime");
+    NEWPERFTICKCOUNTER(_perf_shared_classload_time, SUN_CLS, "sharedClassLoadTime");
+    NEWPERFTICKCOUNTER(_perf_sys_classload_time, SUN_CLS, "sysClassLoadTime");
+    NEWPERFTICKCOUNTER(_perf_app_classload_time, SUN_CLS, "appClassLoadTime");
+    NEWPERFTICKCOUNTER(_perf_app_classload_selftime, SUN_CLS, "appClassLoadTime.self");
+    NEWPERFEVENTCOUNTER(_perf_app_classload_count, SUN_CLS, "appClassLoadCount");
+    NEWPERFTICKCOUNTER(_perf_define_appclasses, SUN_CLS, "defineAppClasses");
+    NEWPERFTICKCOUNTER(_perf_define_appclass_time, SUN_CLS, "defineAppClassTime");
+    NEWPERFTICKCOUNTER(_perf_define_appclass_selftime, SUN_CLS, "defineAppClassTime.self");
+    NEWPERFBYTECOUNTER(_perf_app_classfile_bytes_read, SUN_CLS, "appClassBytes");
+    NEWPERFBYTECOUNTER(_perf_sys_classfile_bytes_read, SUN_CLS, "sysClassBytes");
+
 
     // The following performance counters are added for measuring the impact
     // of the bug fix of 6365597. They are mainly focused on finding out

src/share/vm/classfile/classLoader.hpp

@@ -149,9 +149,26 @@ class ClassLoader: AllStatic {
   static PerfCounter* _perf_accumulated_time;
   static PerfCounter* _perf_classes_inited;
   static PerfCounter* _perf_class_init_time;
+  static PerfCounter* _perf_class_init_selftime;
+  static PerfCounter* _perf_classes_verified;
   static PerfCounter* _perf_class_verify_time;
+  static PerfCounter* _perf_class_verify_selftime;
   static PerfCounter* _perf_classes_linked;
   static PerfCounter* _perf_class_link_time;
+  static PerfCounter* _perf_class_link_selftime;
+  static PerfCounter* _perf_class_parse_time;
+  static PerfCounter* _perf_class_parse_selftime;
+  static PerfCounter* _perf_sys_class_lookup_time;
+  static PerfCounter* _perf_shared_classload_time;
+  static PerfCounter* _perf_sys_classload_time;
+  static PerfCounter* _perf_app_classload_time;
+  static PerfCounter* _perf_app_classload_selftime;
+  static PerfCounter* _perf_app_classload_count;
+  static PerfCounter* _perf_define_appclasses;
+  static PerfCounter* _perf_define_appclass_time;
+  static PerfCounter* _perf_define_appclass_selftime;
+  static PerfCounter* _perf_app_classfile_bytes_read;
+  static PerfCounter* _perf_sys_classfile_bytes_read;
 
   static PerfCounter* _sync_systemLoaderLockContentionRate;
   static PerfCounter* _sync_nonSystemLoaderLockContentionRate;
@@ -196,12 +213,29 @@ class ClassLoader: AllStatic {
   static void print_bootclasspath();
 
   // Timing
-  static PerfCounter* perf_accumulated_time()  { return _perf_accumulated_time; }
-  static PerfCounter* perf_classes_inited()    { return _perf_classes_inited; }
-  static PerfCounter* perf_class_init_time()   { return _perf_class_init_time; }
-  static PerfCounter* perf_class_verify_time() { return _perf_class_verify_time; }
-  static PerfCounter* perf_classes_linked()    { return _perf_classes_linked; }
-  static PerfCounter* perf_class_link_time() { return _perf_class_link_time; }
+  static PerfCounter* perf_accumulated_time()         { return _perf_accumulated_time; }
+  static PerfCounter* perf_classes_inited()           { return _perf_classes_inited; }
+  static PerfCounter* perf_class_init_time()          { return _perf_class_init_time; }
+  static PerfCounter* perf_class_init_selftime()      { return _perf_class_init_selftime; }
+  static PerfCounter* perf_classes_verified()         { return _perf_classes_verified; }
+  static PerfCounter* perf_class_verify_time()        { return _perf_class_verify_time; }
+  static PerfCounter* perf_class_verify_selftime()    { return _perf_class_verify_selftime; }
+  static PerfCounter* perf_classes_linked()           { return _perf_classes_linked; }
+  static PerfCounter* perf_class_link_time()          { return _perf_class_link_time; }
+  static PerfCounter* perf_class_link_selftime()      { return _perf_class_link_selftime; }
+  static PerfCounter* perf_class_parse_time()         { return _perf_class_parse_time; }
+  static PerfCounter* perf_class_parse_selftime()     { return _perf_class_parse_selftime; }
+  static PerfCounter* perf_sys_class_lookup_time()    { return _perf_sys_class_lookup_time; }
+  static PerfCounter* perf_shared_classload_time()    { return _perf_shared_classload_time; }
+  static PerfCounter* perf_sys_classload_time()       { return _perf_sys_classload_time; }
+  static PerfCounter* perf_app_classload_time()       { return _perf_app_classload_time; }
+  static PerfCounter* perf_app_classload_selftime()   { return _perf_app_classload_selftime; }
+  static PerfCounter* perf_app_classload_count()      { return _perf_app_classload_count; }
+  static PerfCounter* perf_define_appclasses()        { return _perf_define_appclasses; }
+  static PerfCounter* perf_define_appclass_time()     { return _perf_define_appclass_time; }
+  static PerfCounter* perf_define_appclass_selftime() { return _perf_define_appclass_selftime; }
+  static PerfCounter* perf_app_classfile_bytes_read() { return _perf_app_classfile_bytes_read; }
+  static PerfCounter* perf_sys_classfile_bytes_read() { return _perf_sys_classfile_bytes_read; }
 
   // Record how often system loader lock object is contended
   static PerfCounter* sync_systemLoaderLockContentionRate() {
@@ -307,3 +341,118 @@ class ClassLoader: AllStatic {
   static int  compile_the_world_counter() { return _compile_the_world_counter; }
 #endif //PRODUCT
 };
+
+// PerfClassTraceTime is used to measure time for class loading related events.
+// This class tracks cumulative time and exclusive time for specific event types.
+// During the execution of one event, other event types (e.g. class loading and
+// resolution) as well as recursive calls of the same event type could happen.
+// Only one elapsed timer (cumulative) and one thread-local self timer (exclusive)
+// (i.e. only one event type) are active at a time even multiple PerfClassTraceTime
+// instances have been created as multiple events are happening.
+class PerfClassTraceTime {
+  public:
+    enum {
+       CLASS_LOAD   = 0,
+       PARSE_CLASS  = 1,
+       CLASS_LINK   = 2,
+       CLASS_VERIFY = 3,
+       CLASS_CLINIT = 4,
+       DEFINE_CLASS = 5,
+       EVENT_TYPE_COUNT = 6
+    };
+  protected:
+    // _t tracks time from initialization to destruction of this timer instance
+    // including time for all other event types, and recursive calls of this type.
+    // When a timer is called recursively, the elapsedTimer _t would not be used.
+    elapsedTimer     _t;
+    PerfLongCounter* _timep;
+    PerfLongCounter* _selftimep;
+    PerfLongCounter* _eventp;
+    // pointer to thread-local recursion counter and timer array
+    // The thread_local timers track cumulative time for specific event types
+    // exclusive of time for other event types, but including recursive calls
+    // of the same type.
+    int*             _recursion_counters;
+    elapsedTimer*    _timers;
+    int              _event_type;
+    int              _prev_active_event;
+
+  public:
+
+    inline PerfClassTraceTime(PerfLongCounter* timep,     /* counter incremented with inclusive time */
+                              PerfLongCounter* selftimep, /* counter incremented with exclusive time */
+                              PerfLongCounter* eventp,    /* event counter */
+                              int* recursion_counters,    /* thread-local recursion counter array */
+                              elapsedTimer* timers,       /* thread-local timer array */
+                              int type                    /* event type */ ) :
+        _timep(timep), _selftimep(selftimep), _eventp(eventp), _recursion_counters(recursion_counters), _timers(timers), _event_type(type) {
+      initialize();
+    }
+
+    inline PerfClassTraceTime(PerfLongCounter* timep,     /* counter incremented with inclusive time */
+                              elapsedTimer* timers,       /* thread-local timer array */
+                              int type                    /* event type */ ) :
+        _timep(timep), _selftimep(NULL), _eventp(NULL), _recursion_counters(NULL), _timers(timers), _event_type(type) {
+      initialize();
+    }
+
+    void initialize() {
+      if (!UsePerfData) return;
+
+      if (_eventp != NULL) {
+        // increment the event counter
+        _eventp->inc();
+      }
+
+      // stop the current active thread-local timer to measure inclusive time
+      _prev_active_event = -1;
+      for (int i=0; i < EVENT_TYPE_COUNT; i++) {
+         if (_timers[i].is_active()) {
+           assert(_prev_active_event == -1, "should have only one active timer");
+           _prev_active_event = i;
+           _timers[i].stop();
+         }
+      }
+
+      if (_recursion_counters == NULL || (_recursion_counters[_event_type])++ == 0) {
+        // start the inclusive timer if not recursively called
+        _t.start();
+      }
+
+      // start thread-local timer of the given event type
+      if (!_timers[_event_type].is_active()) {
+        _timers[_event_type].start();
+      }
+    }
+
+    inline void suspend() { _t.stop(); _timers[_event_type].stop(); }
+    inline void resume()  { _t.start(); _timers[_event_type].start(); }
+
+    ~PerfClassTraceTime() {
+      if (!UsePerfData) return;
+
+      // stop the thread-local timer as the event completes
+      // and resume the thread-local timer of the event next on the stack
+      _timers[_event_type].stop();
+      jlong selftime = _timers[_event_type].ticks();
+
+      if (_prev_active_event >= 0) {
+        _timers[_prev_active_event].start();
+      }
+
+      if (_recursion_counters != NULL && --(_recursion_counters[_event_type]) > 0) return;
+
+      // increment the counters only on the leaf call
+      _t.stop();
+      _timep->inc(_t.ticks());
+      if (_selftimep != NULL) {
+        _selftimep->inc(selftime);
+      }
+      // add all class loading related event selftime to the accumulated time counter
+      ClassLoader::perf_accumulated_time()->inc(selftime);
+
+      // reset the timer
+      _timers[_event_type].reset();
+    }
+};
+

src/share/vm/classfile/systemDictionary.cpp

@@ -1306,13 +1306,18 @@ static instanceKlassHandle download_and_retry_class_load(
 instanceKlassHandle SystemDictionary::load_instance_class(symbolHandle class_name, Handle class_loader, TRAPS) {
   instanceKlassHandle nh = instanceKlassHandle(); // null Handle
   if (class_loader.is_null()) {
+
     // Search the shared system dictionary for classes preloaded into the
     // shared spaces.
     instanceKlassHandle k;
-    k = load_shared_class(class_name, class_loader, THREAD);
+    {
+      PerfTraceTime vmtimer(ClassLoader::perf_shared_classload_time());
+      k = load_shared_class(class_name, class_loader, THREAD);
+    }
 
     if (k.is_null()) {
       // Use VM class loader
+      PerfTraceTime vmtimer(ClassLoader::perf_sys_classload_time());
       k = ClassLoader::load_classfile(class_name, CHECK_(nh));
     }
 
@@ -1334,6 +1339,16 @@ instanceKlassHandle SystemDictionary::load_instance_class(symbolHandle class_nam
     // Use user specified class loader to load class. Call loadClass operation on class_loader.
     ResourceMark rm(THREAD);
 
+    assert(THREAD->is_Java_thread(), "must be a JavaThread");
+    JavaThread* jt = (JavaThread*) THREAD;
+
+    PerfClassTraceTime vmtimer(ClassLoader::perf_app_classload_time(),
+                               ClassLoader::perf_app_classload_selftime(),
+                               ClassLoader::perf_app_classload_count(),
+                               jt->get_thread_stat()->perf_recursion_counts_addr(),
+                               jt->get_thread_stat()->perf_timers_addr(),
+                               PerfClassTraceTime::CLASS_LOAD);
+
     Handle s = java_lang_String::create_from_symbol(class_name, CHECK_(nh));
     // Translate to external class name format, i.e., convert '/' chars to '.'
     Handle string = java_lang_String::externalize_classname(s, CHECK_(nh));

src/share/vm/includeDB_core

@@ -874,6 +874,7 @@ classFileParser.cpp                     signature.hpp
 classFileParser.cpp                     symbolOop.hpp
 classFileParser.cpp                     symbolTable.hpp
 classFileParser.cpp                     systemDictionary.hpp
+classFileParser.cpp                     threadService.hpp
 classFileParser.cpp                     timer.hpp
 classFileParser.cpp                     universe.inline.hpp
 classFileParser.cpp                     verificationType.hpp
@@ -926,6 +927,7 @@ classLoader.cpp                         os_<os_family>.inline.hpp
 classLoader.cpp                         symbolOop.hpp
 classLoader.cpp                         systemDictionary.hpp
 classLoader.cpp                         threadCritical.hpp
+classLoader.cpp                         threadService.hpp
 classLoader.cpp                         timer.hpp
 classLoader.cpp                         universe.inline.hpp
 classLoader.cpp                         vmSymbols.hpp
@@ -4026,6 +4028,7 @@ systemDictionary.cpp                    placeholders.hpp
 systemDictionary.cpp                    resolutionErrors.hpp
 systemDictionary.cpp                    signature.hpp
 systemDictionary.cpp                    systemDictionary.hpp
+systemDictionary.cpp                    threadService.hpp
 systemDictionary.cpp                    typeArrayKlass.hpp
 systemDictionary.cpp                    vmSymbols.hpp
 

src/share/vm/oops/instanceKlass.cpp

@@ -158,9 +158,6 @@ bool instanceKlass::link_class_impl(
   // timer handles recursion
   assert(THREAD->is_Java_thread(), "non-JavaThread in link_class_impl");
   JavaThread* jt = (JavaThread*)THREAD;
-  PerfTraceTimedEvent vmtimer(ClassLoader::perf_class_link_time(),
-                        ClassLoader::perf_classes_linked(),
-                        jt->get_thread_stat()->class_link_recursion_count_addr());
 
   // link super class before linking this class
   instanceKlassHandle super(THREAD, this_oop->super());
@@ -194,6 +191,15 @@ bool instanceKlass::link_class_impl(
     return true;
   }
 
+  // trace only the link time for this klass that includes
+  // the verification time
+  PerfClassTraceTime vmtimer(ClassLoader::perf_class_link_time(),
+                             ClassLoader::perf_class_link_selftime(),
+                             ClassLoader::perf_classes_linked(),
+                             jt->get_thread_stat()->perf_recursion_counts_addr(),
+                             jt->get_thread_stat()->perf_timers_addr(),
+                             PerfClassTraceTime::CLASS_LINK);
+
   // verification & rewriting
   {
     ObjectLocker ol(this_oop, THREAD);
@@ -203,12 +209,14 @@ bool instanceKlass::link_class_impl(
     if (!this_oop->is_linked()) {
       if (!this_oop->is_rewritten()) {
         {
-          assert(THREAD->is_Java_thread(), "non-JavaThread in link_class_impl");
-          JavaThread* jt = (JavaThread*)THREAD;
           // Timer includes any side effects of class verification (resolution,
           // etc), but not recursive entry into verify_code().
-          PerfTraceTime timer(ClassLoader::perf_class_verify_time(),
-                            jt->get_thread_stat()->class_verify_recursion_count_addr());
+          PerfClassTraceTime timer(ClassLoader::perf_class_verify_time(),
+                                   ClassLoader::perf_class_verify_selftime(),
+                                   ClassLoader::perf_classes_verified(),
+                                   jt->get_thread_stat()->perf_recursion_counts_addr(),
+                                   jt->get_thread_stat()->perf_timers_addr(),
+                                   PerfClassTraceTime::CLASS_VERIFY);
           bool verify_ok = verify_code(this_oop, throw_verifyerror, THREAD);
           if (!verify_ok) {
             return false;
@@ -350,9 +358,12 @@ void instanceKlass::initialize_impl(instanceKlassHandle this_oop, TRAPS) {
     JavaThread* jt = (JavaThread*)THREAD;
     // Timer includes any side effects of class initialization (resolution,
     // etc), but not recursive entry into call_class_initializer().
-    PerfTraceTimedEvent timer(ClassLoader::perf_class_init_time(),
-                              ClassLoader::perf_classes_inited(),
-                              jt->get_thread_stat()->class_init_recursion_count_addr());
+    PerfClassTraceTime timer(ClassLoader::perf_class_init_time(),
+                             ClassLoader::perf_class_init_selftime(),
+                             ClassLoader::perf_classes_inited(),
+                             jt->get_thread_stat()->perf_recursion_counts_addr(),
+                             jt->get_thread_stat()->perf_timers_addr(),
+                             PerfClassTraceTime::CLASS_CLINIT);
     this_oop->call_class_initializer(THREAD);
   }
 

src/share/vm/prims/jvm.cpp

@@ -756,6 +756,20 @@ static void is_lock_held_by_thread(Handle loader, PerfCounter* counter, TRAPS) {
 static jclass jvm_define_class_common(JNIEnv *env, const char *name, jobject loader, const jbyte *buf, jsize len, jobject pd, const char *source, TRAPS) {
   if (source == NULL)  source = "__JVM_DefineClass__";
 
+  assert(THREAD->is_Java_thread(), "must be a JavaThread");
+  JavaThread* jt = (JavaThread*) THREAD;
+
+  PerfClassTraceTime vmtimer(ClassLoader::perf_define_appclass_time(),
+                             ClassLoader::perf_define_appclass_selftime(),
+                             ClassLoader::perf_define_appclasses(),
+                             jt->get_thread_stat()->perf_recursion_counts_addr(),
+                             jt->get_thread_stat()->perf_timers_addr(),
+                             PerfClassTraceTime::DEFINE_CLASS);
+
+  if (UsePerfData) {
+    ClassLoader::perf_app_classfile_bytes_read()->inc(len);
+  }
+
   // Since exceptions can be thrown, class initialization can take place
   // if name is NULL no check for class name in .class stream has to be made.
   symbolHandle class_name;

src/share/vm/runtime/perfData.hpp

@@ -868,6 +868,10 @@ class PerfDataManager : AllStatic {
   {counter = PerfDataManager::create_counter(counter_ns, counter_name, \
                                              PerfData::U_Events,CHECK);}
 
+#define NEWPERFBYTECOUNTER(counter, counter_ns, counter_name)  \
+  {counter = PerfDataManager::create_counter(counter_ns, counter_name, \
+                                             PerfData::U_Bytes,CHECK);}
+
 // Utility Classes
 
 /*

src/share/vm/services/threadService.cpp

@@ -688,10 +688,9 @@ ThreadStatistics::ThreadStatistics() {
   _contended_enter_count = 0;
   _monitor_wait_count = 0;
   _sleep_count = 0;
-  _class_init_recursion_count = 0;
-  _class_verify_recursion_count = 0;
   _count_pending_reset = false;
   _timer_pending_reset = false;
+  memset((void*) _perf_recursion_counts, 0, sizeof(_perf_recursion_counts));
 }
 
 ThreadSnapshot::ThreadSnapshot(JavaThread* thread) {

src/share/vm/services/threadService.hpp

@@ -120,9 +120,8 @@ private:
   bool         _timer_pending_reset;
 
   // Keep accurate times for potentially recursive class operations
-  int          _class_init_recursion_count;
-  int          _class_verify_recursion_count;
-  int          _class_link_recursion_count;
+  int           _perf_recursion_counts[6];
+  elapsedTimer  _perf_timers[6];
 
   // utility functions
   void  check_and_reset_count()            {
@@ -165,9 +164,8 @@ public:
   void reset_count_stat()                  { _count_pending_reset = true; }
   void reset_time_stat()                   { _timer_pending_reset = true; }
 
-  int* class_init_recursion_count_addr()   { return &_class_init_recursion_count; }
-  int* class_verify_recursion_count_addr() { return &_class_verify_recursion_count; }
-  int* class_link_recursion_count_addr()   { return &_class_link_recursion_count; }
+  int* perf_recursion_counts_addr()        { return _perf_recursion_counts; }
+  elapsedTimer* perf_timers_addr()         { return _perf_timers; }
 };
 
 // Thread snapshot to represent the thread state and statistics