Merge

hotspot/agent/src/share/classes/sun/jvm/hotspot/gc_implementation/g1/G1CollectedHeap.java

 /*
- * Copyright (c) 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011, 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
@@ -49,8 +49,12 @@ public class G1CollectedHeap extends SharedHeap {
     static private long g1CommittedFieldOffset;
     // size_t _summary_bytes_used;
     static private CIntegerField summaryBytesUsedField;
-    // G1MonitoringSupport* _g1mm
+    // G1MonitoringSupport* _g1mm;
     static private AddressField g1mmField;
+    // MasterOldRegionSet _old_set;
+    static private long oldSetFieldOffset;
+    // MasterHumongousRegionSet _humongous_set;
+    static private long humongousSetFieldOffset;
 
     static {
         VM.registerVMInitializedObserver(new Observer() {
@@ -67,12 +71,14 @@ public class G1CollectedHeap extends SharedHeap {
         g1CommittedFieldOffset = type.getField("_g1_committed").getOffset();
         summaryBytesUsedField = type.getCIntegerField("_summary_bytes_used");
         g1mmField = type.getAddressField("_g1mm");
+        oldSetFieldOffset = type.getField("_old_set").getOffset();
+        humongousSetFieldOffset = type.getField("_humongous_set").getOffset();
     }
 
     public long capacity() {
         Address g1CommittedAddr = addr.addOffsetTo(g1CommittedFieldOffset);
-        MemRegion g1_committed = new MemRegion(g1CommittedAddr);
-        return g1_committed.byteSize();
+        MemRegion g1Committed = new MemRegion(g1CommittedAddr);
+        return g1Committed.byteSize();
     }
 
     public long used() {
@@ -94,6 +100,18 @@ public class G1CollectedHeap extends SharedHeap {
         return (G1MonitoringSupport) VMObjectFactory.newObject(G1MonitoringSupport.class, g1mmAddr);
     }
 
+    public HeapRegionSetBase oldSet() {
+        Address oldSetAddr = addr.addOffsetTo(oldSetFieldOffset);
+        return (HeapRegionSetBase) VMObjectFactory.newObject(HeapRegionSetBase.class,
+                                                             oldSetAddr);
+    }
+
+    public HeapRegionSetBase humongousSet() {
+        Address humongousSetAddr = addr.addOffsetTo(humongousSetFieldOffset);
+        return (HeapRegionSetBase) VMObjectFactory.newObject(HeapRegionSetBase.class,
+                                                             humongousSetAddr);
+    }
+
     private Iterator<HeapRegion> heapRegionIterator() {
         return hrs().heapRegionIterator();
     }

hotspot/agent/src/share/classes/sun/jvm/hotspot/gc_implementation/g1/G1MonitoringSupport.java

 /*
- * Copyright (c) 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011, 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
@@ -77,6 +77,10 @@ public class G1MonitoringSupport extends VMObject {
         return edenUsedField.getValue(addr);
     }
 
+    public long edenRegionNum() {
+        return edenUsed() / HeapRegion.grainBytes();
+    }
+
     public long survivorCommitted() {
         return survivorCommittedField.getValue(addr);
     }
@@ -85,6 +89,10 @@ public class G1MonitoringSupport extends VMObject {
         return survivorUsedField.getValue(addr);
     }
 
+    public long survivorRegionNum() {
+        return survivorUsed() / HeapRegion.grainBytes();
+    }
+
     public long oldCommitted() {
         return oldCommittedField.getValue(addr);
     }

hotspot/agent/src/share/classes/sun/jvm/hotspot/gc_implementation/g1/HeapRegionSetBase.java

+/*
+ * 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.
+ *
+ */
+
+package sun.jvm.hotspot.gc_implementation.g1;
+
+import java.util.Iterator;
+import java.util.Observable;
+import java.util.Observer;
+
+import sun.jvm.hotspot.debugger.Address;
+import sun.jvm.hotspot.runtime.VM;
+import sun.jvm.hotspot.runtime.VMObject;
+import sun.jvm.hotspot.runtime.VMObjectFactory;
+import sun.jvm.hotspot.types.AddressField;
+import sun.jvm.hotspot.types.CIntegerField;
+import sun.jvm.hotspot.types.Type;
+import sun.jvm.hotspot.types.TypeDataBase;
+
+// Mirror class for HeapRegionSetBase. Represents a group of regions.
+
+public class HeapRegionSetBase extends VMObject {
+    // size_t _length;
+    static private CIntegerField lengthField;
+    // size_t _region_num;
+    static private CIntegerField regionNumField;
+    // size_t _total_used_bytes;
+    static private CIntegerField totalUsedBytesField;
+
+    static {
+        VM.registerVMInitializedObserver(new Observer() {
+                public void update(Observable o, Object data) {
+                    initialize(VM.getVM().getTypeDataBase());
+                }
+            });
+    }
+
+    static private synchronized void initialize(TypeDataBase db) {
+        Type type = db.lookupType("HeapRegionSetBase");
+
+        lengthField         = type.getCIntegerField("_length");
+        regionNumField      = type.getCIntegerField("_region_num");
+        totalUsedBytesField = type.getCIntegerField("_total_used_bytes");
+    }
+
+    public long length() {
+        return lengthField.getValue(addr);
+    }
+
+    public long regionNum() {
+        return regionNumField.getValue(addr);
+    }
+
+    public long totalUsedBytes() {
+        return totalUsedBytesField.getValue(addr);
+    }
+
+    public HeapRegionSetBase(Address addr) {
+        super(addr);
+    }
+}

hotspot/agent/src/share/classes/sun/jvm/hotspot/tools/HeapSummary.java

 /*
- * Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 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
@@ -67,6 +67,7 @@ public class HeapSummary extends Tool {
       printValue("SurvivorRatio    = ", getFlagValue("SurvivorRatio", flagMap));
       printValMB("PermSize         = ", getFlagValue("PermSize", flagMap));
       printValMB("MaxPermSize      = ", getFlagValue("MaxPermSize", flagMap));
+      printValMB("G1HeapRegionSize = ", HeapRegion.grainBytes());
 
       System.out.println();
       System.out.println("Heap Usage:");
@@ -100,11 +101,20 @@ public class HeapSummary extends Tool {
          } else if (sharedHeap instanceof G1CollectedHeap) {
              G1CollectedHeap g1h = (G1CollectedHeap) sharedHeap;
              G1MonitoringSupport g1mm = g1h.g1mm();
-             System.out.println("G1 Young Generation");
-             printG1Space("Eden Space:", g1mm.edenUsed(), g1mm.edenCommitted());
-             printG1Space("From Space:", g1mm.survivorUsed(), g1mm.survivorCommitted());
-             printG1Space("To Space:", 0, 0);
-             printG1Space("G1 Old Generation", g1mm.oldUsed(), g1mm.oldCommitted());
+             long edenRegionNum = g1mm.edenRegionNum();
+             long survivorRegionNum = g1mm.survivorRegionNum();
+             HeapRegionSetBase oldSet = g1h.oldSet();
+             HeapRegionSetBase humongousSet = g1h.humongousSet();
+             long oldRegionNum = oldSet.regionNum() + humongousSet.regionNum();
+             printG1Space("G1 Heap:", g1h.n_regions(),
+                          g1h.used(), g1h.capacity());
+             System.out.println("G1 Young Generation:");
+             printG1Space("Eden Space:", edenRegionNum,
+                          g1mm.edenUsed(), g1mm.edenCommitted());
+             printG1Space("Survivor Space:", survivorRegionNum,
+                          g1mm.survivorUsed(), g1mm.survivorCommitted());
+             printG1Space("G1 Old Generation:", oldRegionNum,
+                          g1mm.oldUsed(), g1mm.oldCommitted());
          } else {
              throw new RuntimeException("unknown SharedHeap type : " + heap.getClass());
          }
@@ -216,9 +226,11 @@ public class HeapSummary extends Tool {
       System.out.println(alignment +  (double)space.used() * 100.0 / space.capacity() + "% used");
    }
 
-   private void printG1Space(String spaceName, long used, long capacity) {
+   private void printG1Space(String spaceName, long regionNum,
+                             long used, long capacity) {
       long free = capacity - used;
       System.out.println(spaceName);
+      printValue("regions  = ", regionNum);
       printValMB("capacity = ", capacity);
       printValMB("used     = ", used);
       printValMB("free     = ", free);

hotspot/make/bsd/Makefile

@@ -208,7 +208,7 @@ TARGETS_ZERO      = $(addsuffix zero,$(TARGETS))
 TARGETS_SHARK     = $(addsuffix shark,$(TARGETS))
 
 BUILDTREE_MAKE    = $(GAMMADIR)/make/$(OSNAME)/makefiles/buildtree.make
-BUILDTREE_VARS    = GAMMADIR=$(GAMMADIR) OS_FAMILY=$(OSNAME) SRCARCH=$(SRCARCH) BUILDARCH=$(BUILDARCH) LIBARCH=$(LIBARCH)
+BUILDTREE_VARS    = GAMMADIR=$(GAMMADIR) OS_FAMILY=$(OSNAME) SRCARCH=$(SRCARCH) BUILDARCH=$(BUILDARCH) LIBARCH=$(LIBARCH) LIBRARY_SUFFIX=$(LIBRARY_SUFFIX)
 BUILDTREE_VARS   += HOTSPOT_RELEASE_VERSION=$(HOTSPOT_RELEASE_VERSION) HOTSPOT_BUILD_VERSION=$(HOTSPOT_BUILD_VERSION) JRE_RELEASE_VERSION=$(JRE_RELEASE_VERSION)
 
 BUILDTREE         = $(MAKE) -f $(BUILDTREE_MAKE) $(BUILDTREE_VARS)

hotspot/make/bsd/makefiles/buildtree.make

@@ -162,20 +162,6 @@ ifndef HOTSPOT_VM_DISTRO
   endif
 endif
 
-ifeq ($(OS_VENDOR), Darwin)
-  # MACOSX FIXME: we should be able to run test_gamma (see MACOSX_PORT-214)
-  ifeq ($(ALWAYS_PASS_TEST_GAMMA),)
-    # ALWAYS_PASS_TEST_GAMMA wasn't set so we default to true on MacOS X
-    # until MACOSX_PORT-214 is fixed
-    ALWAYS_PASS_TEST_GAMMA=true
-  endif
-endif
-ifeq ($(ALWAYS_PASS_TEST_GAMMA), true)
-  TEST_GAMMA_STATUS= echo 'exit 0';
-else
-  TEST_GAMMA_STATUS=
-endif
-
 BUILDTREE_VARS += HOTSPOT_RELEASE_VERSION=$(HS_BUILD_VER) HOTSPOT_BUILD_VERSION=  JRE_RELEASE_VERSION=$(JRE_RELEASE_VERSION)
 
 BUILDTREE	= \
@@ -353,12 +339,10 @@ env.sh: $(BUILDTREE_MAKE)
 	$(BUILDTREE_COMMENT); \
 	[ -n "$$JAVA_HOME" ] && { echo ": \$${JAVA_HOME:=$${JAVA_HOME}}"; }; \
 	{ \
-	echo "LD_LIBRARY_PATH=.:$${LD_LIBRARY_PATH:+$$LD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/${LIBARCH}/native_threads:\$${JAVA_HOME}/jre/lib/${LIBARCH}:${GCC_LIB}"; \
-	echo "DYLD_LIBRARY_PATH=.:$${DYLD_LIBRARY_PATH:+$$DYLD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/${LIBARCH}/native_threads:\$${JAVA_HOME}/jre/lib/${LIBARCH}:${GCC_LIB}"; \
 	echo "CLASSPATH=$${CLASSPATH:+$$CLASSPATH:}.:\$${JAVA_HOME}/jre/lib/rt.jar:\$${JAVA_HOME}/jre/lib/i18n.jar"; \
 	} | sed s:$${JAVA_HOME:--------}:\$${JAVA_HOME}:g; \
 	echo "HOTSPOT_BUILD_USER=\"$${LOGNAME:-$$USER} in `basename $(GAMMADIR)`\""; \
-	echo "export JAVA_HOME LD_LIBRARY_PATH DYLD_LIBRARY_PATH CLASSPATH HOTSPOT_BUILD_USER"; \
+	echo "export JAVA_HOME CLASSPATH HOTSPOT_BUILD_USER"; \
 	) > $@
 
 env.csh: env.sh
@@ -412,7 +396,7 @@ JAVA_FLAG/32 = -d32
 JAVA_FLAG/64 = -d64
 
 WRONG_DATA_MODE_MSG = \
-	echo "JAVA_HOME must point to $(DATA_MODE)bit JDK."
+	echo "JAVA_HOME must point to a $(DATA_MODE)-bit OpenJDK."
 
 CROSS_COMPILING_MSG = \
 	echo "Cross compiling for ARCH $(CROSS_COMPILE_ARCH), skipping gamma run."
@@ -420,20 +404,78 @@ CROSS_COMPILING_MSG = \
 test_gamma:  $(BUILDTREE_MAKE) $(GAMMADIR)/make/test/Queens.java
 	@echo Creating $@ ...
 	$(QUIETLY) ( \
-	echo '#!/bin/sh'; \
+	echo "#!/bin/sh"; \
+	echo ""; \
 	$(BUILDTREE_COMMENT); \
-	echo '. ./env.sh'; \
-	echo "if [ \"$(CROSS_COMPILE_ARCH)\" != \"\" ]; then { $(CROSS_COMPILING_MSG); exit 0; }; fi"; \
-	echo "if [ -z \$$JAVA_HOME ]; then { $(NO_JAVA_HOME_MSG); exit 0; }; fi"; \
-	echo "if ! \$${JAVA_HOME}/bin/java $(JAVA_FLAG) -fullversion 2>&1 > /dev/null"; \
-	echo "then"; \
-	echo "  $(WRONG_DATA_MODE_MSG); exit 0;"; \
+	echo ""; \
+	echo "# Include environment settings for gamma run"; \
+	echo ""; \
+	echo ". ./env.sh"; \
+	echo ""; \
+	echo "# Do not run gamma test for cross compiles"; \
+	echo ""; \
+	echo "if [ -n \"$(CROSS_COMPILE_ARCH)\" ]; then "; \
+	echo "  $(CROSS_COMPILING_MSG)"; \
+	echo "  exit 0"; \
 	echo "fi"; \
+	echo ""; \
+	echo "# Make sure JAVA_HOME is set as it is required for gamma"; \
+	echo ""; \
+	echo "if [ -z \"\$${JAVA_HOME}\" ]; then "; \
+	echo "  $(NO_JAVA_HOME_MSG)"; \
+	echo "  exit 0"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Check JAVA_HOME version to be used for the test"; \
+	echo ""; \
+	echo "\$${JAVA_HOME}/bin/java $(JAVA_FLAG) -fullversion > /dev/null 2>&1"; \
+	echo "if [ \$$? -ne 0 ]; then "; \
+	echo "  $(WRONG_DATA_MODE_MSG)"; \
+	echo "  exit 0"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Use gamma_g if it exists"; \
+	echo ""; \
+	echo "GAMMA_PROG=gamma"; \
+	echo "if [ -f gamma_g ]; then "; \
+	echo "  GAMMA_PROG=gamma_g"; \
+	echo "fi"; \
+	echo ""; \
+	echo "if [ \"$(OS_VENDOR)\" = \"Darwin\" ]; then "; \
+	echo "  # Ensure architecture for gamma and JAVA_HOME is the same."; \
+	echo "  # NOTE: gamma assumes the OpenJDK directory layout."; \
+	echo ""; \
+	echo "  GAMMA_ARCH=\"\`file \$${GAMMA_PROG} | awk '{print \$$NF}'\`\""; \
+	echo "  JVM_LIB=\"\$${JAVA_HOME}/jre/lib/libjava.$(LIBRARY_SUFFIX)\""; \
+	echo "  if [ ! -f \$${JVM_LIB} ]; then"; \
+	echo "    JVM_LIB=\"\$${JAVA_HOME}/jre/lib/$${LIBARCH}/libjava.$(LIBRARY_SUFFIX)\""; \
+	echo "  fi"; \
+	echo "  if [ ! -f \$${JVM_LIB} ] || [ -z \"\`file \$${JVM_LIB} | grep \$${GAMMA_ARCH}\`\" ]; then "; \
+	echo "    $(WRONG_DATA_MODE_MSG)"; \
+	echo "    exit 0"; \
+	echo "  fi"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Compile Queens program for test"; \
+	echo ""; \
 	echo "rm -f Queens.class"; \
 	echo "\$${JAVA_HOME}/bin/javac -d . $(GAMMADIR)/make/test/Queens.java"; \
-	echo '[ -f gamma_g ] && { gamma=gamma_g; }'; \
-	echo './$${gamma:-gamma} $(TESTFLAGS) Queens < /dev/null'; \
-	$(TEST_GAMMA_STATUS) \
+	echo ""; \
+	echo "# Set library path solely for gamma launcher test run"; \
+	echo ""; \
+	echo "LD_LIBRARY_PATH=.:$${LD_LIBRARY_PATH:+$$LD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/${LIBARCH}/native_threads:\$${JAVA_HOME}/jre/lib/${LIBARCH}:${GCC_LIB}"; \
+	echo "export LD_LIBRARY_PATH"; \
+	echo "unset LD_LIBRARY_PATH_32"; \
+	echo "unset LD_LIBRARY_PATH_64"; \
+	echo ""; \
+	echo "if [ \"$(OS_VENDOR)\" = \"Darwin\" ]; then "; \
+	echo "  DYLD_LIBRARY_PATH=.:$${DYLD_LIBRARY_PATH:+$$DYLD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/native_threads:\$${JAVA_HOME}/jre/lib:$${DYLD_LIBRARY_PATH:+$$DYLD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/${LIBARCH}/native_threads:\$${JAVA_HOME}/jre/lib/${LIBARCH}:${GCC_LIB}"; \
+	echo "  export DYLD_LIBRARY_PATH"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Use the gamma launcher and JAVA_HOME to run the test"; \
+	echo ""; \
+	echo "./\$${GAMMA_PROG} $(TESTFLAGS) Queens < /dev/null"; \
 	) > $@
 	$(QUIETLY) chmod +x $@
 

hotspot/make/bsd/makefiles/defs.make

@@ -142,6 +142,7 @@ EXPORT_LIST += $(EXPORT_DOCS_DIR)/platform/jvmti/jvmti.html
 # client and server subdirectories have symbolic links to ../libjsig.so
 EXPORT_LIST += $(EXPORT_JRE_LIB_ARCH_DIR)/libjsig.$(LIBRARY_SUFFIX)
 EXPORT_SERVER_DIR = $(EXPORT_JRE_LIB_ARCH_DIR)/server
+EXPORT_CLIENT_DIR = $(EXPORT_JRE_LIB_ARCH_DIR)/client
 
 ifndef BUILD_CLIENT_ONLY
 EXPORT_LIST += $(EXPORT_SERVER_DIR)/Xusage.txt
@@ -150,7 +151,6 @@ endif
 
 ifneq ($(ZERO_BUILD), true)
   ifeq ($(ARCH_DATA_MODEL), 32)
-    EXPORT_CLIENT_DIR = $(EXPORT_JRE_LIB_ARCH_DIR)/client
     EXPORT_LIST += $(EXPORT_CLIENT_DIR)/Xusage.txt
     EXPORT_LIST += $(EXPORT_CLIENT_DIR)/libjvm.$(LIBRARY_SUFFIX)
   endif

hotspot/make/bsd/makefiles/launcher.make

@@ -50,7 +50,24 @@ ifeq ($(LINK_INTO),AOUT)
   LIBS_LAUNCHER             += $(STATIC_STDCXX) $(LIBS)
 else
   LAUNCHER.o                 = launcher.o
-  LFLAGS_LAUNCHER           += -L`pwd`
+  LFLAGS_LAUNCHER           += -L`pwd` 
+
+  # The gamma launcher runs the JDK from $JAVA_HOME, overriding the JVM with a
+  # freshly built JVM at ./libjvm.{so|dylib}.  This is accomplished by setting 
+  # the library searchpath using ({DY}LD_LIBRARY_PATH) to find the local JVM 
+  # first.  Gamma dlopen()s libjava from $JAVA_HOME/jre/lib{/$arch}, which is
+  # statically linked with CoreFoundation framework libs. Unfortunately, gamma's
+  # unique searchpath results in some unresolved symbols in the framework 
+  # libraries, because JDK libraries are inadvertently discovered first on the
+  # searchpath, e.g. libjpeg.  On Mac OS X, filenames are case *insensitive*.
+  # So, the actual filename collision is libjpeg.dylib and libJPEG.dylib.
+  # To resolve this, gamma needs to also statically link with the CoreFoundation 
+  # framework libraries.
+
+  ifeq ($(OS_VENDOR),Darwin)
+    LFLAGS_LAUNCHER         += -framework CoreFoundation 
+  endif
+
   LIBS_LAUNCHER             += -l$(JVM) $(LIBS)
 endif
 

hotspot/make/bsd/makefiles/vm.make

@@ -337,8 +337,8 @@ ifeq ($(OS_VENDOR), Darwin)
 $(LIBJVM).dSYM: $(LIBJVM)
 	dsymutil $(LIBJVM)
 
-# no launcher or libjvm_db for macosx
-build: $(LIBJVM) $(LIBJSIG) $(BUILDLIBSAPROC) dtraceCheck $(LIBJVM).dSYM
+# no libjvm_db for macosx
+build: $(LIBJVM) $(LAUNCHER) $(LIBJSIG) $(BUILDLIBSAPROC) dtraceCheck $(LIBJVM).dSYM
 	echo "Doing vm.make build:"
 else
 build: $(LIBJVM) $(LAUNCHER) $(LIBJSIG) $(LIBJVM_DB) $(BUILDLIBSAPROC)

hotspot/make/hotspot_version

@@ -35,7 +35,7 @@ HOTSPOT_VM_COPYRIGHT=Copyright 2011
 
 HS_MAJOR_VER=23
 HS_MINOR_VER=0
-HS_BUILD_NUMBER=11
+HS_BUILD_NUMBER=12
 
 JDK_MAJOR_VER=1
 JDK_MINOR_VER=8

hotspot/make/jprt.properties

@@ -174,6 +174,10 @@ jprt.my.linux.armsflt.ejdk6=linux_armsflt_2.6
 jprt.my.linux.armsflt.ejdk7=linux_armsflt_2.6
 jprt.my.linux.armsflt=${jprt.my.linux.armsflt.${jprt.tools.default.release}}
 
+jprt.my.macosx.x64.jdk8=macosx_x64_10.7
+jprt.my.macosx.x64.jdk7=macosx_x64_10.7
+jprt.my.macosx.x64=${jprt.my.macosx.x64.${jprt.tools.default.release}}
+
 jprt.my.windows.i586.jdk8=windows_i586_5.1
 jprt.my.windows.i586.jdk7=windows_i586_5.1
 jprt.my.windows.i586.jdk7b107=windows_i586_5.0
@@ -211,6 +215,7 @@ jprt.build.targets.standard= \
     ${jprt.my.solaris.x64}-{product|fastdebug|debug}, \
     ${jprt.my.linux.i586}-{product|fastdebug|debug}, \
     ${jprt.my.linux.x64}-{product|fastdebug}, \
+    ${jprt.my.macosx.x64}-{product|fastdebug|debug}, \
     ${jprt.my.windows.i586}-{product|fastdebug|debug}, \
     ${jprt.my.windows.x64}-{product|fastdebug|debug}
 
@@ -416,6 +421,30 @@ jprt.my.linux.x64.test.targets = \
     ${jprt.my.linux.x64}-{product|fastdebug}-c2-jbb_G1, \
     ${jprt.my.linux.x64}-{product|fastdebug}-c2-jbb_ParOldGC
 
+jprt.my.macosx.x64.test.targets = \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-jvm98, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-jvm98_nontiered, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-scimark, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCBasher_default, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCBasher_SerialGC, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCBasher_ParallelGC, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCBasher_ParNewGC, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCBasher_CMS, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCBasher_G1, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCBasher_ParOldGC, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCOld_default, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCOld_SerialGC, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCOld_ParallelGC, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCOld_ParNewGC, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCOld_CMS, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCOld_G1, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-GCOld_ParOldGC \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-jbb_default, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-jbb_default_tiered, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-jbb_ParallelGC, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-jbb_G1, \
+    ${jprt.my.macosx.x64}-{product|fastdebug}-c2-jbb_ParOldGC
+
 jprt.my.windows.i586.test.targets = \
     ${jprt.my.windows.i586}-{product|fastdebug}-{c1|c2}-jvm98, \
     ${jprt.my.windows.i586}-{product|fastdebug}-c2-jvm98_nontiered, \
@@ -492,6 +521,7 @@ jprt.test.targets.standard = \
   ${jprt.my.solaris.x64.test.targets}, \
   ${jprt.my.linux.i586.test.targets}, \
   ${jprt.my.linux.x64.test.targets}, \
+  ${jprt.my.macosx.x64.test.targets}, \
   ${jprt.my.windows.i586.test.targets}, \
   ${jprt.my.windows.x64.test.targets}, \
   ${jprt.test.targets.open}
@@ -538,6 +568,7 @@ jprt.make.rule.test.targets.standard.server = \
   ${jprt.my.solaris.x64}-*-c2-servertest, \
   ${jprt.my.linux.i586}-*-c2-servertest, \
   ${jprt.my.linux.x64}-*-c2-servertest, \
+  ${jprt.my.macosx.x64}-*-c2-servertest, \
   ${jprt.my.windows.i586}-*-c2-servertest, \
   ${jprt.my.windows.x64}-*-c2-servertest
 
@@ -548,6 +579,7 @@ jprt.make.rule.test.targets.standard.internalvmtests = \
   ${jprt.my.solaris.x64}-fastdebug-c2-internalvmtests, \
   ${jprt.my.linux.i586}-fastdebug-c2-internalvmtests, \
   ${jprt.my.linux.x64}-fastdebug-c2-internalvmtests, \
+  ${jprt.my.macosx.x64}-fastdebug-c2-internalvmtests, \
   ${jprt.my.windows.i586}-fastdebug-c2-internalvmtests, \
   ${jprt.my.windows.x64}-fastdebug-c2-internalvmtests
   

hotspot/make/linux/makefiles/buildtree.make

@@ -326,11 +326,10 @@ env.sh: $(BUILDTREE_MAKE)
 	$(BUILDTREE_COMMENT); \
 	[ -n "$$JAVA_HOME" ] && { echo ": \$${JAVA_HOME:=$${JAVA_HOME}}"; }; \
 	{ \
-	echo "LD_LIBRARY_PATH=.:$${LD_LIBRARY_PATH:+$$LD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/${LIBARCH}/native_threads:\$${JAVA_HOME}/jre/lib/${LIBARCH}:${GCC_LIB}"; \
 	echo "CLASSPATH=$${CLASSPATH:+$$CLASSPATH:}.:\$${JAVA_HOME}/jre/lib/rt.jar:\$${JAVA_HOME}/jre/lib/i18n.jar"; \
 	} | sed s:$${JAVA_HOME:--------}:\$${JAVA_HOME}:g; \
 	echo "HOTSPOT_BUILD_USER=\"$${LOGNAME:-$$USER} in `basename $(GAMMADIR)`\""; \
-	echo "export JAVA_HOME LD_LIBRARY_PATH CLASSPATH HOTSPOT_BUILD_USER"; \
+	echo "export JAVA_HOME CLASSPATH HOTSPOT_BUILD_USER"; \
 	) > $@
 
 env.csh: env.sh
@@ -384,7 +383,7 @@ JAVA_FLAG/32 = -d32
 JAVA_FLAG/64 = -d64
 
 WRONG_DATA_MODE_MSG = \
-	echo "JAVA_HOME must point to $(DATA_MODE)bit JDK."
+	echo "JAVA_HOME must point to a $(DATA_MODE)-bit OpenJDK."
 
 CROSS_COMPILING_MSG = \
 	echo "Cross compiling for ARCH $(CROSS_COMPILE_ARCH), skipping gamma run."
@@ -392,19 +391,78 @@ CROSS_COMPILING_MSG = \
 test_gamma:  $(BUILDTREE_MAKE) $(GAMMADIR)/make/test/Queens.java
 	@echo Creating $@ ...
 	$(QUIETLY) ( \
-	echo '#!/bin/sh'; \
+	echo "#!/bin/sh"; \
+	echo ""; \
 	$(BUILDTREE_COMMENT); \
-	echo '. ./env.sh'; \
-	echo "if [ \"$(CROSS_COMPILE_ARCH)\" != \"\" ]; then { $(CROSS_COMPILING_MSG); exit 0; }; fi"; \
-	echo "if [ -z \$$JAVA_HOME ]; then { $(NO_JAVA_HOME_MSG); exit 0; }; fi"; \
-	echo "if ! \$${JAVA_HOME}/bin/java $(JAVA_FLAG) -fullversion 2>&1 > /dev/null"; \
-	echo "then"; \
-	echo "  $(WRONG_DATA_MODE_MSG); exit 0;"; \
+	echo ""; \
+	echo "# Include environment settings for gamma run"; \
+	echo ""; \
+	echo ". ./env.sh"; \
+	echo ""; \
+	echo "# Do not run gamma test for cross compiles"; \
+	echo ""; \
+	echo "if [ -n \"$(CROSS_COMPILE_ARCH)\" ]; then "; \
+	echo "  $(CROSS_COMPILING_MSG)"; \
+	echo "  exit 0"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Make sure JAVA_HOME is set as it is required for gamma"; \
+	echo ""; \
+	echo "if [ -z \"\$${JAVA_HOME}\" ]; then "; \
+	echo "  $(NO_JAVA_HOME_MSG)"; \
+	echo "  exit 0"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Check JAVA_HOME version to be used for the test"; \
+	echo ""; \
+	echo "\$${JAVA_HOME}/bin/java $(JAVA_FLAG) -fullversion > /dev/null 2>&1"; \
+	echo "if [ \$$? -ne 0 ]; then "; \
+	echo "  $(WRONG_DATA_MODE_MSG)"; \
+	echo "  exit 0"; \
 	echo "fi"; \
+	echo ""; \
+	echo "# Use gamma_g if it exists"; \
+	echo ""; \
+	echo "GAMMA_PROG=gamma"; \
+	echo "if [ -f gamma_g ]; then "; \
+	echo "  GAMMA_PROG=gamma_g"; \
+	echo "fi"; \
+	echo ""; \
+	echo "if [ \"$(OS_VENDOR)\" = \"Darwin\" ]; then "; \
+	echo "  # Ensure architecture for gamma and JAVA_HOME is the same."; \
+	echo "  # NOTE: gamma assumes the OpenJDK directory layout."; \
+	echo ""; \
+	echo "  GAMMA_ARCH=\"\`file \$${GAMMA_PROG} | awk '{print \$$NF}'\`\""; \
+	echo "  JVM_LIB=\"\$${JAVA_HOME}/jre/lib/libjava.$(LIBRARY_SUFFIX)\""; \
+	echo "  if [ ! -f \$${JVM_LIB} ]; then"; \
+	echo "    JVM_LIB=\"\$${JAVA_HOME}/jre/lib/$${LIBARCH}/libjava.$(LIBRARY_SUFFIX)\""; \
+	echo "  fi"; \
+	echo "  if [ ! -f \$${JVM_LIB} ] || [ -z \"\`file \$${JVM_LIB} | grep \$${GAMMA_ARCH}\`\" ]; then "; \
+	echo "    $(WRONG_DATA_MODE_MSG)"; \
+	echo "    exit 0"; \
+	echo "  fi"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Compile Queens program for test"; \
+	echo ""; \
 	echo "rm -f Queens.class"; \
 	echo "\$${JAVA_HOME}/bin/javac -d . $(GAMMADIR)/make/test/Queens.java"; \
-	echo '[ -f gamma_g ] && { gamma=gamma_g; }'; \
-	echo './$${gamma:-gamma} $(TESTFLAGS) Queens < /dev/null'; \
+	echo ""; \
+	echo "# Set library path solely for gamma launcher test run"; \
+	echo ""; \
+	echo "LD_LIBRARY_PATH=.:$${LD_LIBRARY_PATH:+$$LD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/${LIBARCH}/native_threads:\$${JAVA_HOME}/jre/lib/${LIBARCH}:${GCC_LIB}"; \
+	echo "export LD_LIBRARY_PATH"; \
+	echo "unset LD_LIBRARY_PATH_32"; \
+	echo "unset LD_LIBRARY_PATH_64"; \
+	echo ""; \
+	echo "if [ \"$(OS_VENDOR)\" = \"Darwin\" ]; then "; \
+	echo "  DYLD_LIBRARY_PATH=.:$${DYLD_LIBRARY_PATH:+$$DYLD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/native_threads:\$${JAVA_HOME}/jre/lib:$${DYLD_LIBRARY_PATH:+$$DYLD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/${LIBARCH}/native_threads:\$${JAVA_HOME}/jre/lib/${LIBARCH}:${GCC_LIB}"; \
+	echo "  export DYLD_LIBRARY_PATH"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Use the gamma launcher and JAVA_HOME to run the test"; \
+	echo ""; \
+	echo "./\$${GAMMA_PROG} $(TESTFLAGS) Queens < /dev/null"; \
 	) > $@
 	$(QUIETLY) chmod +x $@
 

hotspot/make/solaris/makefiles/buildtree.make

@@ -118,7 +118,7 @@ SUBMAKE_DIRS = $(addprefix $(PLATFORM_DIR)/,$(TARGETS))
 BUILDTREE_MAKE	= $(GAMMADIR)/make/$(OS_FAMILY)/makefiles/buildtree.make
 
 BUILDTREE_TARGETS = Makefile flags.make flags_vm.make vm.make adlc.make jvmti.make sa.make \
-        env.ksh env.csh jdkpath.sh .dbxrc test_gamma
+        env.sh env.csh jdkpath.sh .dbxrc test_gamma
 
 BUILDTREE_VARS	= GAMMADIR=$(GAMMADIR) OS_FAMILY=$(OS_FAMILY) \
 	ARCH=$(ARCH) BUILDARCH=$(BUILDARCH) LIBARCH=$(LIBARCH) VARIANT=$(VARIANT)
@@ -313,22 +313,19 @@ sa.make: $(BUILDTREE_MAKE)
 	echo "include \$$(GAMMADIR)/make/$(OS_FAMILY)/makefiles/$(@F)"; \
 	) > $@
 
-env.ksh: $(BUILDTREE_MAKE)
+env.sh: $(BUILDTREE_MAKE)
 	@echo Creating $@ ...
 	$(QUIETLY) ( \
 	$(BUILDTREE_COMMENT); \
 	[ -n "$$JAVA_HOME" ] && { echo ": \$${JAVA_HOME:=$${JAVA_HOME}}"; }; \
 	{ \
-	echo "LD_LIBRARY_PATH=.:$${LD_LIBRARY_PATH:+$$LD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/${LIBARCH}/native_threads:\$${JAVA_HOME}/jre/lib/${LIBARCH}:${GCC_LIB}"; \
-	echo "unset LD_LIBRARY_PATH_32"; \
-	echo "unset LD_LIBRARY_PATH_64"; \
 	echo "CLASSPATH=$${CLASSPATH:+$$CLASSPATH:}.:\$${JAVA_HOME}/jre/lib/rt.jar:\$${JAVA_HOME}/jre/lib/i18n.jar"; \
 	} | sed s:$${JAVA_HOME:--------}:\$${JAVA_HOME}:g; \
 	echo "HOTSPOT_BUILD_USER=\"$${LOGNAME:-$$USER} in `basename $(GAMMADIR)`\""; \
 	echo "export JAVA_HOME LD_LIBRARY_PATH CLASSPATH HOTSPOT_BUILD_USER"; \
 	) > $@
 
-env.csh: env.ksh
+env.csh: env.sh
 	@echo Creating $@ ...
 	$(QUIETLY) ( \
 	$(BUILDTREE_COMMENT); \
@@ -384,23 +381,86 @@ JAVA_FLAG/32 = -d32
 JAVA_FLAG/64 = -d64
 
 WRONG_DATA_MODE_MSG = \
-	echo "JAVA_HOME must point to $(DATA_MODE)bit JDK."
+	echo "JAVA_HOME must point to a $(DATA_MODE)-bit OpenJDK."
+
+CROSS_COMPILING_MSG = \
+	echo "Cross compiling for ARCH $(CROSS_COMPILE_ARCH), skipping gamma run."
 
 test_gamma:  $(BUILDTREE_MAKE) $(GAMMADIR)/make/test/Queens.java
 	@echo Creating $@ ...
 	$(QUIETLY) ( \
-	echo '#!/bin/ksh'; \
+	echo "#!/bin/sh"; \
+	echo ""; \
 	$(BUILDTREE_COMMENT); \
-	echo '. ./env.ksh'; \
-	echo "if [ -z \$$JAVA_HOME ]; then { $(NO_JAVA_HOME_MSG); exit 0; }; fi"; \
-	echo "if ! \$${JAVA_HOME}/bin/java $(JAVA_FLAG) -fullversion 2>&1 > /dev/null"; \
-	echo "then"; \
-	echo "  $(WRONG_DATA_MODE_MSG); exit 0;"; \
+	echo ""; \
+	echo "# Include environment settings for gamma run"; \
+	echo ""; \
+	echo ". ./env.sh"; \
+	echo ""; \
+	echo "# Do not run gamma test for cross compiles"; \
+	echo ""; \
+	echo "if [ -n \"$(CROSS_COMPILE_ARCH)\" ]; then "; \
+	echo "  $(CROSS_COMPILING_MSG)"; \
+	echo "  exit 0"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Make sure JAVA_HOME is set as it is required for gamma"; \
+	echo ""; \
+	echo "if [ -z \"\$${JAVA_HOME}\" ]; then "; \
+	echo "  $(NO_JAVA_HOME_MSG)"; \
+	echo "  exit 0"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Check JAVA_HOME version to be used for the test"; \
+	echo ""; \
+	echo "\$${JAVA_HOME}/bin/java $(JAVA_FLAG) -fullversion > /dev/null 2>&1"; \
+	echo "if [ \$$? -ne 0 ]; then "; \
+	echo "  $(WRONG_DATA_MODE_MSG)"; \
+	echo "  exit 0"; \
 	echo "fi"; \
+	echo ""; \
+	echo "# Use gamma_g if it exists"; \
+	echo ""; \
+	echo "GAMMA_PROG=gamma"; \
+	echo "if [ -f gamma_g ]; then "; \
+	echo "  GAMMA_PROG=gamma_g"; \
+	echo "fi"; \
+	echo ""; \
+	echo "if [ \"$(OS_VENDOR)\" = \"Darwin\" ]; then "; \
+	echo "  # Ensure architecture for gamma and JAVA_HOME is the same."; \
+	echo "  # NOTE: gamma assumes the OpenJDK directory layout."; \
+	echo ""; \
+	echo "  GAMMA_ARCH=\"\`file \$${GAMMA_PROG} | awk '{print \$$NF}'\`\""; \
+	echo "  JVM_LIB=\"\$${JAVA_HOME}/jre/lib/libjava.$(LIBRARY_SUFFIX)\""; \
+	echo "  if [ ! -f \$${JVM_LIB} ]; then"; \
+	echo "    JVM_LIB=\"\$${JAVA_HOME}/jre/lib/$${LIBARCH}/libjava.$(LIBRARY_SUFFIX)\""; \
+	echo "  fi"; \
+	echo "  if [ ! -f \$${JVM_LIB} ] || [ -z \"\`file \$${JVM_LIB} | grep \$${GAMMA_ARCH}\`\" ]; then "; \
+	echo "    $(WRONG_DATA_MODE_MSG)"; \
+	echo "    exit 0"; \
+	echo "  fi"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Compile Queens program for test"; \
+	echo ""; \
 	echo "rm -f Queens.class"; \
 	echo "\$${JAVA_HOME}/bin/javac -d . $(GAMMADIR)/make/test/Queens.java"; \
-	echo '[ -f gamma_g ] && { gamma=gamma_g; }'; \
-	echo './$${gamma:-gamma} $(TESTFLAGS) Queens < /dev/null'; \
+	echo ""; \
+	echo "# Set library path solely for gamma launcher test run"; \
+	echo ""; \
+	echo "LD_LIBRARY_PATH=.:$${LD_LIBRARY_PATH:+$$LD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/${LIBARCH}/native_threads:\$${JAVA_HOME}/jre/lib/${LIBARCH}:${GCC_LIB}"; \
+	echo "export LD_LIBRARY_PATH"; \
+	echo "unset LD_LIBRARY_PATH_32"; \
+	echo "unset LD_LIBRARY_PATH_64"; \
+	echo ""; \
+	echo "if [ \"$(OS_VENDOR)\" = \"Darwin\" ]; then "; \
+	echo "  DYLD_LIBRARY_PATH=.:$${DYLD_LIBRARY_PATH:+$$DYLD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/native_threads:\$${JAVA_HOME}/jre/lib:$${DYLD_LIBRARY_PATH:+$$DYLD_LIBRARY_PATH:}\$${JAVA_HOME}/jre/lib/${LIBARCH}/native_threads:\$${JAVA_HOME}/jre/lib/${LIBARCH}:${GCC_LIB}"; \
+	echo "  export DYLD_LIBRARY_PATH"; \
+	echo "fi"; \
+	echo ""; \
+	echo "# Use the gamma launcher and JAVA_HOME to run the test"; \
+	echo ""; \
+	echo "./\$${GAMMA_PROG} $(TESTFLAGS) Queens < /dev/null"; \
 	) > $@
 	$(QUIETLY) chmod +x $@
 

hotspot/src/cpu/sparc/vm/frame_sparc.cpp

 /*
- * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -810,7 +810,7 @@ intptr_t* frame::interpreter_frame_tos_at(jint offset) const {
 }
 
 
-#ifdef ASSERT
+#ifndef PRODUCT
 
 #define DESCRIBE_FP_OFFSET(name) \
   values.describe(frame_no, fp() + frame::name##_offset, #name)
@@ -820,11 +820,19 @@ void frame::describe_pd(FrameValues& values, int frame_no) {
     values.describe(frame_no, sp() + w, err_msg("register save area word %d", w), 1);
   }
 
-  if (is_interpreted_frame()) {
+  if (is_ricochet_frame()) {
+    MethodHandles::RicochetFrame::describe(this, values, frame_no);
+  } else if (is_interpreted_frame()) {
     DESCRIBE_FP_OFFSET(interpreter_frame_d_scratch_fp);
     DESCRIBE_FP_OFFSET(interpreter_frame_l_scratch_fp);
     DESCRIBE_FP_OFFSET(interpreter_frame_padding);
     DESCRIBE_FP_OFFSET(interpreter_frame_oop_temp);
+
+    // esp, according to Lesp (e.g. not depending on bci), if seems valid
+    intptr_t* esp = *interpreter_frame_esp_addr();
+    if ((esp >= sp()) && (esp < fp())) {
+      values.describe(-1, esp, "*Lesp");
+    }
   }
 
   if (!is_compiled_frame()) {
@@ -844,4 +852,3 @@ intptr_t *frame::initial_deoptimization_info() {
   // unused... but returns fp() to minimize changes introduced by 7087445
   return fp();
 }
-

hotspot/src/cpu/sparc/vm/methodHandles_sparc.cpp

 /*
- * Copyright (c) 2008, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2008, 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
@@ -177,7 +177,7 @@ void MethodHandles::RicochetFrame::generate_ricochet_blob(MacroAssembler* _masm,
   BLOCK_COMMENT("ricochet_blob.bounce");
 
   if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
-  trace_method_handle(_masm, "ricochet_blob.bounce");
+  trace_method_handle(_masm, "return/ricochet_blob.bounce");
 
   __ JMP(L1_continuation, 0);
   __ delayed()->nop();
@@ -268,14 +268,16 @@ void MethodHandles::RicochetFrame::leave_ricochet_frame(MacroAssembler* _masm,
 }
 
 // Emit code to verify that FP is pointing at a valid ricochet frame.
-#ifdef ASSERT
+#ifndef PRODUCT
 enum {
   ARG_LIMIT = 255, SLOP = 45,
   // use this parameter for checking for garbage stack movements:
   UNREASONABLE_STACK_MOVE = (ARG_LIMIT + SLOP)
   // the slop defends against false alarms due to fencepost errors
 };
+#endif
 
+#ifdef ASSERT
 void MethodHandles::RicochetFrame::verify_clean(MacroAssembler* _masm) {
   // The stack should look like this:
   //    ... keep1 | dest=42 | keep2 | magic | handler | magic | recursive args | [RF]
@@ -1000,32 +1002,143 @@ void MethodHandles::move_return_value(MacroAssembler* _masm, BasicType type,
   BLOCK_COMMENT("} move_return_value");
 }
 
+#ifndef PRODUCT
+void MethodHandles::RicochetFrame::describe(const frame* fr, FrameValues& values, int frame_no)  {
+    RicochetFrame* rf = new RicochetFrame(*fr);
+
+    // ricochet slots (kept in registers for sparc)
+    values.describe(frame_no, rf->register_addr(I5_savedSP), err_msg("exact_sender_sp reg for #%d", frame_no));
+    values.describe(frame_no, rf->register_addr(L5_conversion), err_msg("conversion reg for #%d", frame_no));
+    values.describe(frame_no, rf->register_addr(L4_saved_args_base), err_msg("saved_args_base reg for #%d", frame_no));
+    values.describe(frame_no, rf->register_addr(L3_saved_args_layout), err_msg("saved_args_layout reg for #%d", frame_no));
+    values.describe(frame_no, rf->register_addr(L2_saved_target), err_msg("saved_target reg for #%d", frame_no));
+    values.describe(frame_no, rf->register_addr(L1_continuation), err_msg("continuation reg for #%d", frame_no));
+
+    // relevant ricochet targets (in caller frame)
+    values.describe(-1, rf->saved_args_base(),  err_msg("*saved_args_base for #%d", frame_no));
+    values.describe(-1, (intptr_t *)(STACK_BIAS+(uintptr_t)rf->exact_sender_sp()),  err_msg("*exact_sender_sp+STACK_BIAS for #%d", frame_no));
+}
+#endif // ASSERT
+
 #ifndef PRODUCT
 extern "C" void print_method_handle(oop mh);
 void trace_method_handle_stub(const char* adaptername,
                               oopDesc* mh,
-                              intptr_t* saved_sp) {
+                              intptr_t* saved_sp,
+                              intptr_t* args,
+                              intptr_t* tracing_fp) {
   bool has_mh = (strstr(adaptername, "return/") == NULL);  // return adapters don't have mh
-  tty->print_cr("MH %s mh="INTPTR_FORMAT " saved_sp=" INTPTR_FORMAT, adaptername, (intptr_t) mh, saved_sp);
-  if (has_mh)
+
+  tty->print_cr("MH %s mh="INTPTR_FORMAT " saved_sp=" INTPTR_FORMAT " args=" INTPTR_FORMAT, adaptername, (intptr_t) mh, saved_sp, args);
+
+  if (Verbose) {
+    // dumping last frame with frame::describe
+
+    JavaThread* p = JavaThread::active();
+
+    ResourceMark rm;
+    PRESERVE_EXCEPTION_MARK; // may not be needed by safer and unexpensive here
+    FrameValues values;
+
+    // Note: We want to allow trace_method_handle from any call site.
+    // While trace_method_handle creates a frame, it may be entered
+    // without a valid return PC in O7 (e.g. not just after a call).
+    // Walking that frame could lead to failures due to that invalid PC.
+    // => carefully detect that frame when doing the stack walking
+
+    // walk up to the right frame using the "tracing_fp" argument
+    intptr_t* cur_sp = StubRoutines::Sparc::flush_callers_register_windows_func()();
+    frame cur_frame(cur_sp, frame::unpatchable, NULL);
+
+    while (cur_frame.fp() != (intptr_t *)(STACK_BIAS+(uintptr_t)tracing_fp)) {
+      cur_frame = os::get_sender_for_C_frame(&cur_frame);
+    }
+
+    // safely create a frame and call frame::describe
+    intptr_t *dump_sp = cur_frame.sender_sp();
+    intptr_t *dump_fp = cur_frame.link();
+
+    bool walkable = has_mh; // whether the traced frame shoud be walkable
+
+    // the sender for cur_frame is the caller of trace_method_handle
+    if (walkable) {
+      // The previous definition of walkable may have to be refined
+      // if new call sites cause the next frame constructor to start
+      // failing. Alternatively, frame constructors could be
+      // modified to support the current or future non walkable
+      // frames (but this is more intrusive and is not considered as
+      // part of this RFE, which will instead use a simpler output).
+      frame dump_frame = frame(dump_sp,
+                               cur_frame.sp(), // younger_sp
+                               false); // no adaptation
+      dump_frame.describe(values, 1);
+    } else {
+      // Robust dump for frames which cannot be constructed from sp/younger_sp
+      // Add descriptions without building a Java frame to avoid issues
+      values.describe(-1, dump_fp, "fp for #1 <not parsed, cannot trust pc>");
+      values.describe(-1, dump_sp, "sp");
+    }
+
+    bool has_args = has_mh; // whether Gargs is meaningful
+
+    // mark args, if seems valid (may not be valid for some adapters)
+    if (has_args) {
+      if ((args >= dump_sp) && (args < dump_fp)) {
+        values.describe(-1, args, "*G4_args");
+      }
+    }
+
+    // mark saved_sp, if seems valid (may not be valid for some adapters)
+    intptr_t *unbiased_sp = (intptr_t *)(STACK_BIAS+(uintptr_t)saved_sp);
+    if ((unbiased_sp >= dump_sp - UNREASONABLE_STACK_MOVE) && (unbiased_sp < dump_fp)) {
+      values.describe(-1, unbiased_sp, "*saved_sp+STACK_BIAS");
+    }
+
+    // Note: the unextended_sp may not be correct
+    tty->print_cr("  stack layout:");
+    values.print(p);
+  }
+
+  if (has_mh) {
     print_method_handle(mh);
+  }
 }
+
 void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) {
   if (!TraceMethodHandles)  return;
   BLOCK_COMMENT("trace_method_handle {");
   // save: Gargs, O5_savedSP
-  __ save_frame(16);
+  __ save_frame(16); // need space for saving required FPU state
+
   __ set((intptr_t) adaptername, O0);
   __ mov(G3_method_handle, O1);
   __ mov(I5_savedSP, O2);
+  __ mov(Gargs, O3);
+  __ mov(I6, O4); // frame identifier for safe stack walking
+
+  // Save scratched registers that might be needed. Robustness is more
+  // important than optimizing the saves for this debug only code.
+
+  // save FP result, valid at some call sites (adapter_opt_return_float, ...)
+  Address d_save(FP, -sizeof(jdouble) + STACK_BIAS);
+  __ stf(FloatRegisterImpl::D, Ftos_d, d_save);
+  // Safely save all globals but G2 (handled by call_VM_leaf) and G7
+  // (OS reserved).
   __ mov(G3_method_handle, L3);
   __ mov(Gargs, L4);
   __ mov(G5_method_type, L5);
-  __ call_VM_leaf(L7, CAST_FROM_FN_PTR(address, trace_method_handle_stub));
+  __ mov(G6, L6);
+  __ mov(G1, L1);
+
+  __ call_VM_leaf(L2 /* for G2 */, CAST_FROM_FN_PTR(address, trace_method_handle_stub));
 
   __ mov(L3, G3_method_handle);
   __ mov(L4, Gargs);
   __ mov(L5, G5_method_type);
+  __ mov(L6, G6);
+  __ mov(L1, G1);
+  __ ldf(FloatRegisterImpl::D, d_save, Ftos_d);
+
   __ restore();
   BLOCK_COMMENT("} trace_method_handle");
 }
@@ -1250,7 +1363,7 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
         move_typed_arg(_masm, arg_type, false,
                        prim_value_addr,
                        Address(O0_argslot, 0),
-                       O2_scratch);  // must be an even register for !_LP64 long moves (uses O2/O3)
+                      O2_scratch);  // must be an even register for !_LP64 long moves (uses O2/O3)
       }
 
       if (direct_to_method) {

hotspot/src/cpu/sparc/vm/methodHandles_sparc.hpp

 /*
- * Copyright (c) 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011, 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
@@ -145,6 +145,8 @@ class RicochetFrame : public ResourceObj {
   }
 
   static void verify_clean(MacroAssembler* _masm) NOT_DEBUG_RETURN;
+
+  static void describe(const frame* fr, FrameValues& values, int frame_no) PRODUCT_RETURN;
 };
 
 // Additional helper methods for MethodHandles code generation:

hotspot/src/cpu/x86/vm/frame_x86.cpp

@@ -651,13 +651,15 @@ intptr_t* frame::interpreter_frame_tos_at(jint offset) const {
   return &interpreter_frame_tos_address()[index];
 }
 
-#ifdef ASSERT
+#ifndef PRODUCT
 
 #define DESCRIBE_FP_OFFSET(name) \
   values.describe(frame_no, fp() + frame::name##_offset, #name)
 
 void frame::describe_pd(FrameValues& values, int frame_no) {
-  if (is_interpreted_frame()) {
+  if (is_ricochet_frame()) {
+    MethodHandles::RicochetFrame::describe(this, values, frame_no);
+  } else if (is_interpreted_frame()) {
     DESCRIBE_FP_OFFSET(interpreter_frame_sender_sp);
     DESCRIBE_FP_OFFSET(interpreter_frame_last_sp);
     DESCRIBE_FP_OFFSET(interpreter_frame_method);
@@ -667,7 +669,6 @@ void frame::describe_pd(FrameValues& values, int frame_no) {
     DESCRIBE_FP_OFFSET(interpreter_frame_bcx);
     DESCRIBE_FP_OFFSET(interpreter_frame_initial_sp);
   }
-
 }
 #endif
 

hotspot/src/cpu/x86/vm/methodHandles_x86.cpp

 /*
- * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -279,14 +279,16 @@ void MethodHandles::RicochetFrame::leave_ricochet_frame(MacroAssembler* _masm,
 }
 
 // Emit code to verify that RBP is pointing at a valid ricochet frame.
-#ifdef ASSERT
+#ifndef PRODUCT
 enum {
   ARG_LIMIT = 255, SLOP = 4,
   // use this parameter for checking for garbage stack movements:
   UNREASONABLE_STACK_MOVE = (ARG_LIMIT + SLOP)
   // the slop defends against false alarms due to fencepost errors
 };
+#endif
 
+#ifdef ASSERT
 void MethodHandles::RicochetFrame::verify_clean(MacroAssembler* _masm) {
   // The stack should look like this:
   //    ... keep1 | dest=42 | keep2 | RF | magic | handler | magic | recursive args |
@@ -990,6 +992,26 @@ void MethodHandles::move_return_value(MacroAssembler* _masm, BasicType type,
   BLOCK_COMMENT("} move_return_value");
 }
 
+#ifndef PRODUCT
+#define DESCRIBE_RICOCHET_OFFSET(rf, name) \
+  values.describe(frame_no, (intptr_t *) (((uintptr_t)rf) + MethodHandles::RicochetFrame::name##_offset_in_bytes()), #name)
+
+void MethodHandles::RicochetFrame::describe(const frame* fr, FrameValues& values, int frame_no)  {
+    address bp = (address) fr->fp();
+    RicochetFrame* rf = (RicochetFrame*)(bp - sender_link_offset_in_bytes());
+
+    // ricochet slots
+    DESCRIBE_RICOCHET_OFFSET(rf, exact_sender_sp);
+    DESCRIBE_RICOCHET_OFFSET(rf, conversion);
+    DESCRIBE_RICOCHET_OFFSET(rf, saved_args_base);
+    DESCRIBE_RICOCHET_OFFSET(rf, saved_args_layout);
+    DESCRIBE_RICOCHET_OFFSET(rf, saved_target);
+    DESCRIBE_RICOCHET_OFFSET(rf, continuation);
+
+    // relevant ricochet targets (in caller frame)
+    values.describe(-1, rf->saved_args_base(),  err_msg("*saved_args_base for #%d", frame_no));
+}
+#endif // ASSERT
 
 #ifndef PRODUCT
 extern "C" void print_method_handle(oop mh);
@@ -1001,6 +1023,7 @@ void trace_method_handle_stub(const char* adaptername,
                               intptr_t* saved_bp) {
   // called as a leaf from native code: do not block the JVM!
   bool has_mh = (strstr(adaptername, "return/") == NULL);  // return adapters don't have rcx_mh
+
   intptr_t* last_sp = (intptr_t*) saved_bp[frame::interpreter_frame_last_sp_offset];
   intptr_t* base_sp = last_sp;
   typedef MethodHandles::RicochetFrame RicochetFrame;
@@ -1030,13 +1053,64 @@ void trace_method_handle_stub(const char* adaptername,
     tty->cr();
     if (last_sp != saved_sp && last_sp != NULL)
       tty->print_cr("*** last_sp="PTR_FORMAT, (intptr_t)last_sp);
-    int stack_dump_count = 16;
-    if (stack_dump_count < (int)(saved_bp + 2 - saved_sp))
-      stack_dump_count = (int)(saved_bp + 2 - saved_sp);
-    if (stack_dump_count > 64)  stack_dump_count = 48;
-    for (i = 0; i < stack_dump_count; i += 4) {
-      tty->print_cr(" dump at SP[%d] "PTR_FORMAT": "PTR_FORMAT" "PTR_FORMAT" "PTR_FORMAT" "PTR_FORMAT,
-                    i, (intptr_t) &entry_sp[i+0], entry_sp[i+0], entry_sp[i+1], entry_sp[i+2], entry_sp[i+3]);
+
+    {
+     // dumping last frame with frame::describe
+
+      JavaThread* p = JavaThread::active();
+
+      ResourceMark rm;
+      PRESERVE_EXCEPTION_MARK; // may not be needed by safer and unexpensive here
+      FrameValues values;
+
+      // Note: We want to allow trace_method_handle from any call site.
+      // While trace_method_handle creates a frame, it may be entered
+      // without a PC on the stack top (e.g. not just after a call).
+      // Walking that frame could lead to failures due to that invalid PC.
+      // => carefully detect that frame when doing the stack walking
+
+      // Current C frame
+      frame cur_frame = os::current_frame();
+
+      // Robust search of trace_calling_frame (independant of inlining).
+      // Assumes saved_regs comes from a pusha in the trace_calling_frame.
+      assert(cur_frame.sp() < saved_regs, "registers not saved on stack ?");
+      frame trace_calling_frame = os::get_sender_for_C_frame(&cur_frame);
+      while (trace_calling_frame.fp() < saved_regs) {
+        trace_calling_frame = os::get_sender_for_C_frame(&trace_calling_frame);
+      }
+
+      // safely create a frame and call frame::describe
+      intptr_t *dump_sp = trace_calling_frame.sender_sp();
+      intptr_t *dump_fp = trace_calling_frame.link();
+
+      bool walkable = has_mh; // whether the traced frame shoud be walkable
+
+      if (walkable) {
+        // The previous definition of walkable may have to be refined
+        // if new call sites cause the next frame constructor to start
+        // failing. Alternatively, frame constructors could be
+        // modified to support the current or future non walkable
+        // frames (but this is more intrusive and is not considered as
+        // part of this RFE, which will instead use a simpler output).
+        frame dump_frame = frame(dump_sp, dump_fp);
+        dump_frame.describe(values, 1);
+      } else {
+        // Stack may not be walkable (invalid PC above FP):
+        // Add descriptions without building a Java frame to avoid issues
+        values.describe(-1, dump_fp, "fp for #1 <not parsed, cannot trust pc>");
+        values.describe(-1, dump_sp, "sp for #1");
+      }
+
+      // mark saved_sp if seems valid
+      if (has_mh) {
+        if ((saved_sp >= dump_sp - UNREASONABLE_STACK_MOVE) && (saved_sp < dump_fp)) {
+          values.describe(-1, saved_sp, "*saved_sp");
+        }
+      }
+
+      tty->print_cr("  stack layout:");
+      values.print(p);
     }
     if (has_mh)
       print_method_handle(mh);
@@ -1066,26 +1140,49 @@ void trace_method_handle_stub_wrapper(MethodHandleStubArguments* args) {
 void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) {
   if (!TraceMethodHandles)  return;
   BLOCK_COMMENT("trace_method_handle {");
-  __ push(rax);
-  __ lea(rax, Address(rsp, wordSize * NOT_LP64(6) LP64_ONLY(14))); // entry_sp  __ pusha();
-  __ pusha();
-  __ mov(rbx, rsp);
   __ enter();
+  __ andptr(rsp, -16); // align stack if needed for FPU state
+  __ pusha();
+  __ mov(rbx, rsp); // for retreiving saved_regs
+  // Note: saved_regs must be in the entered frame for the
+  // robust stack walking implemented in trace_method_handle_stub.
+
+  // save FP result, valid at some call sites (adapter_opt_return_float, ...)
+  __ increment(rsp, -2 * wordSize);
+  if  (UseSSE >= 2) {
+    __ movdbl(Address(rsp, 0), xmm0);
+  } else if (UseSSE == 1) {
+    __ movflt(Address(rsp, 0), xmm0);
+  } else {
+    __ fst_d(Address(rsp, 0));
+  }
+
   // incoming state:
   // rcx: method handle
   // r13 or rsi: saved sp
   // To avoid calling convention issues, build a record on the stack and pass the pointer to that instead.
+  // Note: fix the increment below if pushing more arguments
   __ push(rbp);               // saved_bp
-  __ push(rsi);               // saved_sp
-  __ push(rax);               // entry_sp
+  __ push(saved_last_sp_register()); // saved_sp
+  __ push(rbp);               // entry_sp (with extra align space)
   __ push(rbx);               // pusha saved_regs
   __ push(rcx);               // mh
-  __ push(rcx);               // adaptername
+  __ push(rcx);               // slot for adaptername
   __ movptr(Address(rsp, 0), (intptr_t) adaptername);
   __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub_wrapper), rsp);
-  __ leave();
+  __ increment(rsp, 6 * wordSize); // MethodHandleStubArguments
+
+  if  (UseSSE >= 2) {
+    __ movdbl(xmm0, Address(rsp, 0));
+  } else if (UseSSE == 1) {
+    __ movflt(xmm0, Address(rsp, 0));
+  } else {
+    __ fld_d(Address(rsp, 0));
+  }
+  __ increment(rsp, 2 * wordSize);
+
   __ popa();
-  __ pop(rax);
+  __ leave();
   BLOCK_COMMENT("} trace_method_handle");
 }
 #endif //PRODUCT

hotspot/src/cpu/x86/vm/methodHandles_x86.hpp

 /*
- * Copyright (c) 2010, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2010, 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
@@ -224,6 +224,8 @@ class RicochetFrame {
   }
 
   static void verify_clean(MacroAssembler* _masm) NOT_DEBUG_RETURN;
+
+  static void describe(const frame* fr, FrameValues& values, int frame_no) PRODUCT_RETURN;
 };
 
 // Additional helper methods for MethodHandles code generation:

hotspot/src/cpu/zero/vm/frame_zero.cpp

@@ -418,7 +418,7 @@ void ZeroFrame::identify_vp_word(int       frame_index,
   }
 }
 
-#ifdef ASSERT
+#ifndef PRODUCT
 
 void frame::describe_pd(FrameValues& values, int frame_no) {
 

hotspot/src/os/bsd/vm/os_bsd.cpp

 /*
- * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -301,6 +301,12 @@ static char cpu_arch[] = "sparc";
 #error Add appropriate cpu_arch setting
 #endif
 
+// Compiler variant
+#ifdef COMPILER2
+#define COMPILER_VARIANT "server"
+#else
+#define COMPILER_VARIANT "client"
+#endif
 
 #ifndef _ALLBSD_SOURCE
 // pid_t gettid()
@@ -2507,7 +2513,7 @@ void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
 
 static char saved_jvm_path[MAXPATHLEN] = {0};
 
-// Find the full path to the current module, libjvm.so or libjvm_g.so
+// Find the full path to the current module, libjvm or libjvm_g
 void os::jvm_path(char *buf, jint buflen) {
   // Error checking.
   if (buflen < MAXPATHLEN) {
@@ -2532,11 +2538,11 @@ void os::jvm_path(char *buf, jint buflen) {
 
   if (Arguments::created_by_gamma_launcher()) {
     // Support for the gamma launcher.  Typical value for buf is
-    // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".  If "/jre/lib/" appears at
+    // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm".  If "/jre/lib/" appears at
     // the right place in the string, then assume we are installed in a JDK and
-    // we're done.  Otherwise, check for a JAVA_HOME environment variable and fix
-    // up the path so it looks like libjvm.so is installed there (append a
-    // fake suffix hotspot/libjvm.so).
+    // we're done.  Otherwise, check for a JAVA_HOME environment variable and
+    // construct a path to the JVM being overridden.
+
     const char *p = buf + strlen(buf) - 1;
     for (int count = 0; p > buf && count < 5; ++count) {
       for (--p; p > buf && *p != '/'; --p)
@@ -2550,7 +2556,7 @@ void os::jvm_path(char *buf, jint buflen) {
         char* jrelib_p;
         int len;
 
-        // Check the current module name "libjvm.so" or "libjvm_g.so".
+        // Check the current module name "libjvm" or "libjvm_g".
         p = strrchr(buf, '/');
         assert(strstr(p, "/libjvm") == p, "invalid library name");
         p = strstr(p, "_g") ? "_g" : "";
@@ -2563,19 +2569,32 @@ void os::jvm_path(char *buf, jint buflen) {
         // modules image doesn't have "jre" subdirectory
         len = strlen(buf);
         jrelib_p = buf + len;
-        snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
+
+        // Add the appropriate library subdir
+        snprintf(jrelib_p, buflen-len, "/jre/lib");
+        if (0 != access(buf, F_OK)) {
+          snprintf(jrelib_p, buflen-len, "/lib");
+        }
+
+        // Add the appropriate client or server subdir
+        len = strlen(buf);
+        jrelib_p = buf + len;
+        snprintf(jrelib_p, buflen-len, "/%s", COMPILER_VARIANT);
         if (0 != access(buf, F_OK)) {
-          snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
+          snprintf(jrelib_p, buflen-len, "");
         }
 
+        // If the path exists within JAVA_HOME, add the JVM library name
+        // to complete the path to JVM being overridden.  Otherwise fallback
+        // to the path to the current library.
         if (0 == access(buf, F_OK)) {
-          // Use current module name "libjvm[_g].so" instead of
-          // "libjvm"debug_only("_g")".so" since for fastdebug version
-          // we should have "libjvm.so" but debug_only("_g") adds "_g"!
+          // Use current module name "libjvm[_g]" instead of
+          // "libjvm"debug_only("_g")"" since for fastdebug version
+          // we should have "libjvm" but debug_only("_g") adds "_g"!
           len = strlen(buf);
-          snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
+          snprintf(buf + len, buflen-len, "/libjvm%s%s", p, JNI_LIB_SUFFIX);
         } else {
-          // Go back to path of .so
+          // Fall back to path of current library
           rp = realpath(dli_fname, buf);
           if (rp == NULL)
             return;
@@ -3570,26 +3589,28 @@ void os::loop_breaker(int attempts) {
 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
 
 #if defined(_ALLBSD_SOURCE) && !defined(__APPLE__)
-int os::java_to_os_priority[MaxPriority + 1] = {
+int os::java_to_os_priority[CriticalPriority + 1] = {
   19,              // 0 Entry should never be used
 
    0,              // 1 MinPriority
    3,              // 2
    6,              // 3
 
-   10,              // 4
-   15,              // 5 NormPriority
-   18,              // 6
+  10,              // 4
+  15,              // 5 NormPriority
+  18,              // 6
+
+  21,              // 7
+  25,              // 8
+  28,              // 9 NearMaxPriority
 
-   21,              // 7
-   25,              // 8
-   28,              // 9 NearMaxPriority
+  31,              // 10 MaxPriority
 
-   31              // 10 MaxPriority
+  31               // 11 CriticalPriority
 };
 #elif defined(__APPLE__)
 /* Using Mach high-level priority assignments */
-int os::java_to_os_priority[MaxPriority + 1] = {
+int os::java_to_os_priority[CriticalPriority + 1] = {
    0,              // 0 Entry should never be used (MINPRI_USER)
 
   27,              // 1 MinPriority
@@ -3604,10 +3625,12 @@ int os::java_to_os_priority[MaxPriority + 1] = {
   34,              // 8
   35,              // 9 NearMaxPriority
 
-  36               // 10 MaxPriority
+  36,              // 10 MaxPriority
+
+  36               // 11 CriticalPriority
 };
 #else
-int os::java_to_os_priority[MaxPriority + 1] = {
+int os::java_to_os_priority[CriticalPriority + 1] = {
   19,              // 0 Entry should never be used
 
    4,              // 1 MinPriority
@@ -3622,7 +3645,9 @@ int os::java_to_os_priority[MaxPriority + 1] = {
   -3,              // 8
   -4,              // 9 NearMaxPriority
 
-  -5               // 10 MaxPriority
+  -5,              // 10 MaxPriority
+
+  -5               // 11 CriticalPriority
 };
 #endif
 
@@ -3638,6 +3663,9 @@ static int prio_init() {
       ThreadPriorityPolicy = 0;
     }
   }
+  if (UseCriticalJavaThreadPriority) {
+    os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
+  }
   return 0;
 }
 

hotspot/src/os/linux/vm/os_linux.cpp

 /*
- * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -3383,7 +3383,7 @@ void os::loop_breaker(int attempts) {
 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
 
-int os::java_to_os_priority[MaxPriority + 1] = {
+int os::java_to_os_priority[CriticalPriority + 1] = {
   19,              // 0 Entry should never be used
 
    4,              // 1 MinPriority
@@ -3398,7 +3398,9 @@ int os::java_to_os_priority[MaxPriority + 1] = {
   -3,              // 8
   -4,              // 9 NearMaxPriority
 
-  -5               // 10 MaxPriority
+  -5,              // 10 MaxPriority
+
+  -5               // 11 CriticalPriority
 };
 
 static int prio_init() {
@@ -3413,6 +3415,9 @@ static int prio_init() {
       ThreadPriorityPolicy = 0;
     }
   }
+  if (UseCriticalJavaThreadPriority) {
+    os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
+  }
   return 0;
 }
 

hotspot/src/os/posix/launcher/java_md.c

@@ -701,6 +701,14 @@ GetJREPath(char *path, jint pathsize, char * arch, jboolean speculative)
     char libjava[MAXPATHLEN];
 
     if (GetApplicationHome(path, pathsize)) {
+
+        /* Is the JRE universal, i.e. no arch dir? */
+        sprintf(libjava, "%s/jre/lib/" JAVA_DLL, path);
+        if (access(libjava, F_OK) == 0) {
+            strcat(path, "/jre");
+            goto found;
+        }
+
         /* Is JRE co-located with the application? */
         sprintf(libjava, "%s/lib/%s/" JAVA_DLL, path, arch);
         if (access(libjava, F_OK) == 0) {
@@ -734,7 +742,7 @@ LoadJavaVM(const char *jvmpath, InvocationFunctions *ifn)
     ifn->GetDefaultJavaVMInitArgs = JNI_GetDefaultJavaVMInitArgs;
     return JNI_TRUE;
 #else
-   Dl_info dlinfo;
+    Dl_info dlinfo;
     void *libjvm;
 
     if (_launcher_debug) {

hotspot/src/os/solaris/vm/osThread_solaris.hpp

 /*
- * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -28,17 +28,17 @@
 // This is embedded via include into the class OSThread
 
  private:
-
-  thread_t _thread_id;      // Solaris thread id
-  unsigned int  _lwp_id;    // lwp ID, only used with bound threads
-  sigset_t _caller_sigmask; // Caller's signal mask
-  bool _vm_created_thread;  // true if the VM create this thread
-                            // false if primary thread or attached thread
+  thread_t _thread_id;         // Solaris thread id
+  uint     _lwp_id;            // lwp ID, only used with bound threads
+  int      _native_priority;   // Saved native priority when starting
+                               // a bound thread
+  sigset_t _caller_sigmask;    // Caller's signal mask
+  bool     _vm_created_thread; // true if the VM created this thread,
+                               // false if primary thread or attached thread
  public:
-
-  thread_t thread_id() const      { return _thread_id; }
-
-  unsigned int lwp_id() const     { return _lwp_id; }
+  thread_t thread_id() const       { return _thread_id; }
+  uint     lwp_id() const          { return _lwp_id; }
+  int      native_priority() const { return _native_priority; }
 
   // Set and get state of _vm_created_thread flag
   void set_vm_created()           { _vm_created_thread = true; }
@@ -62,8 +62,9 @@
     return true;
   }
 #endif
-  void set_thread_id(thread_t id) { _thread_id = id;   }
-  void set_lwp_id(unsigned int id){ _lwp_id = id;   }
+  void set_thread_id(thread_t id)    { _thread_id = id; }
+  void set_lwp_id(uint id)           { _lwp_id = id; }
+  void set_native_priority(int prio) { _native_priority = prio; }
 
  // ***************************************************************
  // interrupt support.  interrupts (using signals) are used to get

hotspot/src/os/solaris/vm/os_solaris.cpp

 /*
- * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -114,6 +114,7 @@
 # include <sys/rtpriocntl.h>
 # include <sys/tspriocntl.h>
 # include <sys/iapriocntl.h>
+# include <sys/fxpriocntl.h>
 # include <sys/loadavg.h>
 # include <string.h>
 # include <stdio.h>
@@ -129,8 +130,8 @@
 #ifdef _GNU_SOURCE
 // See bug #6514594
 extern "C" int madvise(caddr_t, size_t, int);
-extern "C"  int memcntl(caddr_t addr, size_t len, int cmd, caddr_t  arg,
-     int attr, int mask);
+extern "C" int memcntl(caddr_t addr, size_t len, int cmd, caddr_t arg,
+                       int attr, int mask);
 #endif //_GNU_SOURCE
 
 /*
@@ -215,8 +216,9 @@ struct memcntl_mha {
 #define MaximumPriority 127
 
 // Values for ThreadPriorityPolicy == 1
-int prio_policy1[MaxPriority+1] = { -99999, 0, 16, 32, 48, 64,
-                                        80, 96, 112, 124, 127 };
+int prio_policy1[CriticalPriority+1] = {
+  -99999,  0, 16,  32,  48,  64,
+          80, 96, 112, 124, 127, 127 };
 
 // System parameters used internally
 static clock_t clock_tics_per_sec = 100;
@@ -1048,15 +1050,22 @@ extern "C" void* java_start(void* thread_addr) {
   }
 
   // If the creator called set priority before we started,
-  // we need to call set priority now that we have an lwp.
-  // Get the priority from libthread and set the priority
-  // for the new Solaris lwp.
+  // we need to call set_native_priority now that we have an lwp.
+  // We used to get the priority from thr_getprio (we called
+  // thr_setprio way back in create_thread) and pass it to
+  // set_native_priority, but Solaris scales the priority
+  // in java_to_os_priority, so when we read it back here,
+  // we pass trash to set_native_priority instead of what's
+  // in java_to_os_priority. So we save the native priority
+  // in the osThread and recall it here.
+
   if ( osthr->thread_id() != -1 ) {
     if ( UseThreadPriorities ) {
-      thr_getprio(osthr->thread_id(), &prio);
+      int prio = osthr->native_priority();
       if (ThreadPriorityVerbose) {
-        tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is " INTPTR_FORMAT ", setting priority: %d\n",
-                      osthr->thread_id(), osthr->lwp_id(), prio );
+        tty->print_cr("Starting Thread " INTPTR_FORMAT ", LWP is "
+                      INTPTR_FORMAT ", setting priority: %d\n",
+                      osthr->thread_id(), osthr->lwp_id(), prio);
       }
       os::set_native_priority(thread, prio);
     }
@@ -1353,13 +1362,12 @@ bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
   // Remember that we created this thread so we can set priority on it
   osthread->set_vm_created();
 
-  // Set the default thread priority otherwise use NormalPriority
-
-  if ( UseThreadPriorities ) {
-     thr_setprio(tid, (DefaultThreadPriority == -1) ?
+  // Set the default thread priority.  If using bound threads, setting
+  // lwp priority will be delayed until thread start.
+  set_native_priority(thread,
+                      DefaultThreadPriority == -1 ?
                         java_to_os_priority[NormPriority] :
                         DefaultThreadPriority);
-  }
 
   // Initial thread state is INITIALIZED, not SUSPENDED
   osthread->set_state(INITIALIZED);
@@ -3728,7 +3736,7 @@ typedef struct {
 } SchedInfo;
 
 
-static SchedInfo tsLimits, iaLimits, rtLimits;
+static SchedInfo tsLimits, iaLimits, rtLimits, fxLimits;
 
 #ifdef ASSERT
 static int  ReadBackValidate = 1;
@@ -3739,6 +3747,8 @@ static int  myMax       = 0;
 static int  myCur       = 0;
 static bool priocntl_enable = false;
 
+static const int criticalPrio = 60; // FX/60 is critical thread class/priority on T4
+static int java_MaxPriority_to_os_priority = 0; // Saved mapping
 
 // Call the version of priocntl suitable for all supported versions
 // of Solaris. We need to call through this wrapper so that we can
@@ -3783,19 +3793,27 @@ int     lwp_priocntl_init ()
   if (os::Solaris::T2_libthread() || UseBoundThreads) {
     // If ThreadPriorityPolicy is 1, switch tables
     if (ThreadPriorityPolicy == 1) {
-      for (i = 0 ; i < MaxPriority+1; i++)
+      for (i = 0 ; i < CriticalPriority+1; i++)
         os::java_to_os_priority[i] = prio_policy1[i];
     }
+    if (UseCriticalJavaThreadPriority) {
+      // MaxPriority always maps to the FX scheduling class and criticalPrio.
+      // See set_native_priority() and set_lwp_class_and_priority().
+      // Save original MaxPriority mapping in case attempt to
+      // use critical priority fails.
+      java_MaxPriority_to_os_priority = os::java_to_os_priority[MaxPriority];
+      // Set negative to distinguish from other priorities
+      os::java_to_os_priority[MaxPriority] = -criticalPrio;
+    }
   }
   // Not using Bound Threads, set to ThreadPolicy 1
   else {
-    for ( i = 0 ; i < MaxPriority+1; i++ ) {
+    for ( i = 0 ; i < CriticalPriority+1; i++ ) {
       os::java_to_os_priority[i] = prio_policy1[i];
     }
     return 0;
   }
 
-
   // Get IDs for a set of well-known scheduling classes.
   // TODO-FIXME: GETCLINFO returns the current # of classes in the
   // the system.  We should have a loop that iterates over the
@@ -3828,24 +3846,33 @@ int     lwp_priocntl_init ()
   rtLimits.maxPrio = ((rtinfo_t*)ClassInfo.pc_clinfo)->rt_maxpri;
   rtLimits.minPrio = 0;
 
+  strcpy(ClassInfo.pc_clname, "FX");
+  ClassInfo.pc_cid = -1;
+  rslt = (*priocntl_ptr)(PC_VERSION, P_ALL, 0, PC_GETCID, (caddr_t)&ClassInfo);
+  if (rslt < 0) return errno;
+  assert(ClassInfo.pc_cid != -1, "cid for FX class is -1");
+  fxLimits.schedPolicy = ClassInfo.pc_cid;
+  fxLimits.maxPrio = ((fxinfo_t*)ClassInfo.pc_clinfo)->fx_maxupri;
+  fxLimits.minPrio = 0;
 
   // Query our "current" scheduling class.
-  // This will normally be IA,TS or, rarely, RT.
-  memset (&ParmInfo, 0, sizeof(ParmInfo));
+  // This will normally be IA, TS or, rarely, FX or RT.
+  memset(&ParmInfo, 0, sizeof(ParmInfo));
   ParmInfo.pc_cid = PC_CLNULL;
-  rslt = (*priocntl_ptr) (PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo );
-  if ( rslt < 0 ) return errno;
+  rslt = (*priocntl_ptr) (PC_VERSION, P_PID, P_MYID, PC_GETPARMS, (caddr_t)&ParmInfo);
+  if (rslt < 0) return errno;
   myClass = ParmInfo.pc_cid;
 
   // We now know our scheduling classId, get specific information
-  // the class.
+  // about the class.
   ClassInfo.pc_cid = myClass;
   ClassInfo.pc_clname[0] = 0;
-  rslt = (*priocntl_ptr) (PC_VERSION, (idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo );
-  if ( rslt < 0 ) return errno;
+  rslt = (*priocntl_ptr) (PC_VERSION, (idtype)0, 0, PC_GETCLINFO, (caddr_t)&ClassInfo);
+  if (rslt < 0) return errno;
 
-  if (ThreadPriorityVerbose)
-    tty->print_cr ("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
+  if (ThreadPriorityVerbose) {
+    tty->print_cr("lwp_priocntl_init: Class=%d(%s)...", myClass, ClassInfo.pc_clname);
+  }
 
   memset(&ParmInfo, 0, sizeof(pcparms_t));
   ParmInfo.pc_cid = PC_CLNULL;
@@ -3865,6 +3892,11 @@ int     lwp_priocntl_init ()
     myMin = tsLimits.minPrio;
     myMax = tsLimits.maxPrio;
     myMax = MIN2(myMax, (int)tsInfo->ts_uprilim);       // clamp - restrict
+  } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
+    fxparms_t *fxInfo = (fxparms_t*)ParmInfo.pc_clparms;
+    myMin = fxLimits.minPrio;
+    myMax = fxLimits.maxPrio;
+    myMax = MIN2(myMax, (int)fxInfo->fx_uprilim);       // clamp - restrict
   } else {
     // No clue - punt
     if (ThreadPriorityVerbose)
@@ -3872,8 +3904,9 @@ int     lwp_priocntl_init ()
     return EINVAL;      // no clue, punt
   }
 
-  if (ThreadPriorityVerbose)
-        tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
+  if (ThreadPriorityVerbose) {
+    tty->print_cr ("Thread priority Range: [%d..%d]\n", myMin, myMax);
+  }
 
   priocntl_enable = true;  // Enable changing priorities
   return 0;
@@ -3882,6 +3915,7 @@ int     lwp_priocntl_init ()
 #define IAPRI(x)        ((iaparms_t *)((x).pc_clparms))
 #define RTPRI(x)        ((rtparms_t *)((x).pc_clparms))
 #define TSPRI(x)        ((tsparms_t *)((x).pc_clparms))
+#define FXPRI(x)        ((fxparms_t *)((x).pc_clparms))
 
 
 // scale_to_lwp_priority
@@ -3900,13 +3934,13 @@ int     scale_to_lwp_priority (int rMin, int rMax, int x)
 }
 
 
-// set_lwp_priority
+// set_lwp_class_and_priority
 //
-// Set the priority of the lwp.  This call should only be made
-// when using bound threads (T2 threads are bound by default).
+// Set the class and priority of the lwp.  This call should only
+// be made when using bound threads (T2 threads are bound by default).
 //
-int     set_lwp_priority (int ThreadID, int lwpid, int newPrio )
-{
+int set_lwp_class_and_priority(int ThreadID, int lwpid,
+                               int newPrio, int new_class, bool scale) {
   int rslt;
   int Actual, Expected, prv;
   pcparms_t ParmInfo;                   // for GET-SET
@@ -3927,19 +3961,20 @@ int     set_lwp_priority (int ThreadID, int lwpid, int newPrio )
     return EINVAL;
   }
 
-
   // If lwp hasn't started yet, just return
   // the _start routine will call us again.
   if ( lwpid <= 0 ) {
     if (ThreadPriorityVerbose) {
-      tty->print_cr ("deferring the set_lwp_priority of thread " INTPTR_FORMAT " to %d, lwpid not set",
+      tty->print_cr ("deferring the set_lwp_class_and_priority of thread "
+                     INTPTR_FORMAT " to %d, lwpid not set",
                      ThreadID, newPrio);
     }
     return 0;
   }
 
   if (ThreadPriorityVerbose) {
-    tty->print_cr ("set_lwp_priority(" INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
+    tty->print_cr ("set_lwp_class_and_priority("
+                   INTPTR_FORMAT "@" INTPTR_FORMAT " %d) ",
                    ThreadID, lwpid, newPrio);
   }
 
@@ -3948,40 +3983,70 @@ int     set_lwp_priority (int ThreadID, int lwpid, int newPrio )
   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_GETPARMS, (caddr_t)&ParmInfo);
   if (rslt < 0) return errno;
 
-  if (ParmInfo.pc_cid == rtLimits.schedPolicy) {
+  int cur_class = ParmInfo.pc_cid;
+  ParmInfo.pc_cid = (id_t)new_class;
+
+  if (new_class == rtLimits.schedPolicy) {
     rtparms_t *rtInfo  = (rtparms_t*)ParmInfo.pc_clparms;
-    rtInfo->rt_pri     = scale_to_lwp_priority (rtLimits.minPrio, rtLimits.maxPrio, newPrio);
+    rtInfo->rt_pri     = scale ? scale_to_lwp_priority(rtLimits.minPrio,
+                                                       rtLimits.maxPrio, newPrio)
+                               : newPrio;
     rtInfo->rt_tqsecs  = RT_NOCHANGE;
     rtInfo->rt_tqnsecs = RT_NOCHANGE;
     if (ThreadPriorityVerbose) {
       tty->print_cr("RT: %d->%d\n", newPrio, rtInfo->rt_pri);
     }
-  } else if (ParmInfo.pc_cid == iaLimits.schedPolicy) {
-    iaparms_t *iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
-    int maxClamped     = MIN2(iaLimits.maxPrio, (int)iaInfo->ia_uprilim);
-    iaInfo->ia_upri    = scale_to_lwp_priority(iaLimits.minPrio, maxClamped, newPrio);
-    iaInfo->ia_uprilim = IA_NOCHANGE;
+  } else if (new_class == iaLimits.schedPolicy) {
+    iaparms_t* iaInfo  = (iaparms_t*)ParmInfo.pc_clparms;
+    int maxClamped     = MIN2(iaLimits.maxPrio,
+                              cur_class == new_class
+                                ? (int)iaInfo->ia_uprilim : iaLimits.maxPrio);
+    iaInfo->ia_upri    = scale ? scale_to_lwp_priority(iaLimits.minPrio,
+                                                       maxClamped, newPrio)
+                               : newPrio;
+    iaInfo->ia_uprilim = cur_class == new_class
+                           ? IA_NOCHANGE : (pri_t)iaLimits.maxPrio;
     iaInfo->ia_mode    = IA_NOCHANGE;
+    iaInfo->ia_nice    = cur_class == new_class ? IA_NOCHANGE : NZERO;
     if (ThreadPriorityVerbose) {
-      tty->print_cr ("IA: [%d...%d] %d->%d\n",
-               iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
+      tty->print_cr("IA: [%d...%d] %d->%d\n",
+                    iaLimits.minPrio, maxClamped, newPrio, iaInfo->ia_upri);
     }
-  } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
-    tsparms_t *tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
-    int maxClamped     = MIN2(tsLimits.maxPrio, (int)tsInfo->ts_uprilim);
-    prv                = tsInfo->ts_upri;
-    tsInfo->ts_upri    = scale_to_lwp_priority(tsLimits.minPrio, maxClamped, newPrio);
-    tsInfo->ts_uprilim = IA_NOCHANGE;
+  } else if (new_class == tsLimits.schedPolicy) {
+    tsparms_t* tsInfo  = (tsparms_t*)ParmInfo.pc_clparms;
+    int maxClamped     = MIN2(tsLimits.maxPrio,
+                              cur_class == new_class
+                                ? (int)tsInfo->ts_uprilim : tsLimits.maxPrio);
+    tsInfo->ts_upri    = scale ? scale_to_lwp_priority(tsLimits.minPrio,
+                                                       maxClamped, newPrio)
+                               : newPrio;
+    tsInfo->ts_uprilim = cur_class == new_class
+                           ? TS_NOCHANGE : (pri_t)tsLimits.maxPrio;
     if (ThreadPriorityVerbose) {
-      tty->print_cr ("TS: %d [%d...%d] %d->%d\n",
-               prv, tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
+      tty->print_cr("TS: [%d...%d] %d->%d\n",
+                    tsLimits.minPrio, maxClamped, newPrio, tsInfo->ts_upri);
+    }
+  } else if (new_class == fxLimits.schedPolicy) {
+    fxparms_t* fxInfo  = (fxparms_t*)ParmInfo.pc_clparms;
+    int maxClamped     = MIN2(fxLimits.maxPrio,
+                              cur_class == new_class
+                                ? (int)fxInfo->fx_uprilim : fxLimits.maxPrio);
+    fxInfo->fx_upri    = scale ? scale_to_lwp_priority(fxLimits.minPrio,
+                                                       maxClamped, newPrio)
+                               : newPrio;
+    fxInfo->fx_uprilim = cur_class == new_class
+                           ? FX_NOCHANGE : (pri_t)fxLimits.maxPrio;
+    fxInfo->fx_tqsecs  = FX_NOCHANGE;
+    fxInfo->fx_tqnsecs = FX_NOCHANGE;
+    if (ThreadPriorityVerbose) {
+      tty->print_cr("FX: [%d...%d] %d->%d\n",
+                    fxLimits.minPrio, maxClamped, newPrio, fxInfo->fx_upri);
     }
-    if (prv == tsInfo->ts_upri) return 0;
   } else {
-    if ( ThreadPriorityVerbose ) {
-      tty->print_cr ("Unknown scheduling class\n");
+    if (ThreadPriorityVerbose) {
+      tty->print_cr("Unknown new scheduling class %d\n", new_class);
     }
-      return EINVAL;    // no clue, punt
+    return EINVAL;    // no clue, punt
   }
 
   rslt = (*priocntl_ptr)(PC_VERSION, P_LWPID, lwpid, PC_SETPARMS, (caddr_t)&ParmInfo);
@@ -4016,16 +4081,20 @@ int     set_lwp_priority (int ThreadID, int lwpid, int newPrio )
   } else if (ParmInfo.pc_cid == tsLimits.schedPolicy) {
     Actual   = TSPRI(ReadBack)->ts_upri;
     Expected = TSPRI(ParmInfo)->ts_upri;
+  } else if (ParmInfo.pc_cid == fxLimits.schedPolicy) {
+    Actual   = FXPRI(ReadBack)->fx_upri;
+    Expected = FXPRI(ParmInfo)->fx_upri;
   } else {
-    if ( ThreadPriorityVerbose ) {
-      tty->print_cr("set_lwp_priority: unexpected class in readback: %d\n", ParmInfo.pc_cid);
+    if (ThreadPriorityVerbose) {
+      tty->print_cr("set_lwp_class_and_priority: unexpected class in readback: %d\n",
+                    ParmInfo.pc_cid);
     }
   }
 
   if (Actual != Expected) {
-    if ( ThreadPriorityVerbose ) {
-      tty->print_cr ("set_lwp_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
-             lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
+    if (ThreadPriorityVerbose) {
+      tty->print_cr ("set_lwp_class_and_priority(%d %d) Class=%d: actual=%d vs expected=%d\n",
+                     lwpid, newPrio, ReadBack.pc_cid, Actual, Expected);
     }
   }
 #endif
@@ -4033,8 +4102,6 @@ int     set_lwp_priority (int ThreadID, int lwpid, int newPrio )
   return 0;
 }
 
-
-
 // Solaris only gives access to 128 real priorities at a time,
 // so we expand Java's ten to fill this range.  This would be better
 // if we dynamically adjusted relative priorities.
@@ -4055,8 +4122,7 @@ int     set_lwp_priority (int ThreadID, int lwpid, int newPrio )
 // which do not explicitly alter their thread priorities.
 //
 
-
-int os::java_to_os_priority[MaxPriority + 1] = {
+int os::java_to_os_priority[CriticalPriority + 1] = {
   -99999,         // 0 Entry should never be used
 
   0,              // 1 MinPriority
@@ -4071,17 +4137,51 @@ int os::java_to_os_priority[MaxPriority + 1] = {
   127,            // 8
   127,            // 9 NearMaxPriority
 
-  127             // 10 MaxPriority
-};
+  127,            // 10 MaxPriority
 
+  -criticalPrio   // 11 CriticalPriority
+};
 
 OSReturn os::set_native_priority(Thread* thread, int newpri) {
+  OSThread* osthread = thread->osthread();
+
+  // Save requested priority in case the thread hasn't been started
+  osthread->set_native_priority(newpri);
+
+  // Check for critical priority request
+  bool fxcritical = false;
+  if (newpri == -criticalPrio) {
+    fxcritical = true;
+    newpri = criticalPrio;
+  }
+
   assert(newpri >= MinimumPriority && newpri <= MaximumPriority, "bad priority mapping");
-  if ( !UseThreadPriorities ) return OS_OK;
-  int status = thr_setprio(thread->osthread()->thread_id(), newpri);
-  if ( os::Solaris::T2_libthread() || (UseBoundThreads && thread->osthread()->is_vm_created()) )
-    status |= (set_lwp_priority (thread->osthread()->thread_id(),
-                    thread->osthread()->lwp_id(), newpri ));
+  if (!UseThreadPriorities) return OS_OK;
+
+  int status = 0;
+
+  if (!fxcritical) {
+    // Use thr_setprio only if we have a priority that thr_setprio understands
+    status = thr_setprio(thread->osthread()->thread_id(), newpri);
+  }
+
+  if (os::Solaris::T2_libthread() ||
+      (UseBoundThreads && osthread->is_vm_created())) {
+    int lwp_status =
+      set_lwp_class_and_priority(osthread->thread_id(),
+                                 osthread->lwp_id(),
+                                 newpri,
+                                 fxcritical ? fxLimits.schedPolicy : myClass,
+                                 !fxcritical);
+    if (lwp_status != 0 && fxcritical) {
+      // Try again, this time without changing the scheduling class
+      newpri = java_MaxPriority_to_os_priority;
+      lwp_status = set_lwp_class_and_priority(osthread->thread_id(),
+                                              osthread->lwp_id(),
+                                              newpri, myClass, false);
+    }
+    status |= lwp_status;
+  }
   return (status == 0) ? OS_OK : OS_ERR;
 }
 

hotspot/src/os/windows/vm/os_windows.cpp

 /*
- * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 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
@@ -3296,7 +3296,7 @@ void os::yield_all(int attempts) {
 // so we compress Java's ten down to seven.  It would be better
 // if we dynamically adjusted relative priorities.
 
-int os::java_to_os_priority[MaxPriority + 1] = {
+int os::java_to_os_priority[CriticalPriority + 1] = {
   THREAD_PRIORITY_IDLE,                         // 0  Entry should never be used
   THREAD_PRIORITY_LOWEST,                       // 1  MinPriority
   THREAD_PRIORITY_LOWEST,                       // 2
@@ -3307,10 +3307,11 @@ int os::java_to_os_priority[MaxPriority + 1] = {
   THREAD_PRIORITY_ABOVE_NORMAL,                 // 7
   THREAD_PRIORITY_ABOVE_NORMAL,                 // 8
   THREAD_PRIORITY_HIGHEST,                      // 9  NearMaxPriority
-  THREAD_PRIORITY_HIGHEST                       // 10 MaxPriority
+  THREAD_PRIORITY_HIGHEST,                      // 10 MaxPriority
+  THREAD_PRIORITY_HIGHEST                       // 11 CriticalPriority
 };
 
-int prio_policy1[MaxPriority + 1] = {
+int prio_policy1[CriticalPriority + 1] = {
   THREAD_PRIORITY_IDLE,                         // 0  Entry should never be used
   THREAD_PRIORITY_LOWEST,                       // 1  MinPriority
   THREAD_PRIORITY_LOWEST,                       // 2
@@ -3321,17 +3322,21 @@ int prio_policy1[MaxPriority + 1] = {
   THREAD_PRIORITY_ABOVE_NORMAL,                 // 7
   THREAD_PRIORITY_HIGHEST,                      // 8
   THREAD_PRIORITY_HIGHEST,                      // 9  NearMaxPriority
-  THREAD_PRIORITY_TIME_CRITICAL                 // 10 MaxPriority
+  THREAD_PRIORITY_TIME_CRITICAL,                // 10 MaxPriority
+  THREAD_PRIORITY_TIME_CRITICAL                 // 11 CriticalPriority
 };
 
 static int prio_init() {
   // If ThreadPriorityPolicy is 1, switch tables
   if (ThreadPriorityPolicy == 1) {
     int i;
-    for (i = 0; i < MaxPriority + 1; i++) {
+    for (i = 0; i < CriticalPriority + 1; i++) {
       os::java_to_os_priority[i] = prio_policy1[i];
     }
   }
+  if (UseCriticalJavaThreadPriority) {
+    os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority] ;
+  }
   return 0;
 }
 

hotspot/src/share/vm/c1/c1_LIR.hpp

@@ -1354,9 +1354,10 @@ class LIR_OpBranch: public LIR_Op {
   CodeStub*     _stub;   // if this is a branch to a stub, this is the stub
 
  public:
-  LIR_OpBranch(LIR_Condition cond, Label* lbl)
+  LIR_OpBranch(LIR_Condition cond, BasicType type, Label* lbl)
     : LIR_Op(lir_branch, LIR_OprFact::illegalOpr, (CodeEmitInfo*) NULL)
     , _cond(cond)
+    , _type(type)
     , _label(lbl)
     , _block(NULL)
     , _ublock(NULL)
@@ -2053,7 +2054,7 @@ class LIR_List: public CompilationResourceObj {
   void jump(CodeStub* stub) {
     append(new LIR_OpBranch(lir_cond_always, T_ILLEGAL, stub));
   }
-  void branch(LIR_Condition cond, Label* lbl)        { append(new LIR_OpBranch(cond, lbl)); }
+  void branch(LIR_Condition cond, BasicType type, Label* lbl)        { append(new LIR_OpBranch(cond, type, lbl)); }
   void branch(LIR_Condition cond, BasicType type, BlockBegin* block) {
     assert(type != T_FLOAT && type != T_DOUBLE, "no fp comparisons");
     append(new LIR_OpBranch(cond, type, block));

hotspot/src/share/vm/c1/c1_LIRGenerator.cpp

@@ -2350,7 +2350,7 @@ void LIRGenerator::do_SwitchRanges(SwitchRangeArray* x, LIR_Opr value, BlockBegi
     } else {
       LabelObj* L = new LabelObj();
       __ cmp(lir_cond_less, value, low_key);
-      __ branch(lir_cond_less, L->label());
+      __ branch(lir_cond_less, T_INT, L->label());
       __ cmp(lir_cond_lessEqual, value, high_key);
       __ branch(lir_cond_lessEqual, T_INT, dest);
       __ branch_destination(L->label());

hotspot/src/share/vm/c1/c1_Runtime1.cpp

@@ -413,8 +413,9 @@ static nmethod* counter_overflow_helper(JavaThread* THREAD, int branch_bci, meth
     }
     bci = branch_bci + offset;
   }
-
+  assert(!HAS_PENDING_EXCEPTION, "Should not have any exceptions pending");
   osr_nm = CompilationPolicy::policy()->event(enclosing_method, method, branch_bci, bci, level, nm, THREAD);
+  assert(!HAS_PENDING_EXCEPTION, "Event handler should not throw any exceptions");
   return osr_nm;
 }
 

hotspot/src/share/vm/classfile/javaClasses.cpp

@@ -1347,7 +1347,13 @@ class BacktraceBuilder: public StackObj {
     return _backtrace();
   }
 
-  inline void push(methodOop method, short bci, TRAPS) {
+  inline void push(methodOop method, int bci, TRAPS) {
+    // Smear the -1 bci to 0 since the array only holds unsigned
+    // shorts.  The later line number lookup would just smear the -1
+    // to a 0 even if it could be recorded.
+    if (bci == SynchronizationEntryBCI) bci = 0;
+    assert(bci == (jushort)bci, "doesn't fit");
+
     if (_index >= trace_chunk_size) {
       methodHandle mhandle(THREAD, method);
       expand(CHECK);
@@ -1574,8 +1580,13 @@ void java_lang_Throwable::fill_in_stack_trace_of_preallocated_backtrace(Handle t
   int chunk_count = 0;
 
   for (;!st.at_end(); st.next()) {
-    // add element
-    bcis->ushort_at_put(chunk_count, st.bci());
+    // Add entry and smear the -1 bci to 0 since the array only holds
+    // unsigned shorts.  The later line number lookup would just smear
+    // the -1 to a 0 even if it could be recorded.
+    int bci = st.bci();
+    if (bci == SynchronizationEntryBCI) bci = 0;
+    assert(bci == (jushort)bci, "doesn't fit");
+    bcis->ushort_at_put(chunk_count, bci);
     methods->obj_at_put(chunk_count, st.method());
 
     chunk_count++;

hotspot/src/share/vm/compiler/compileBroker.cpp

 /*
- * Copyright (c) 1999, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 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
@@ -855,23 +855,23 @@ CompilerThread* CompileBroker::make_compiler_thread(const char* name, CompileQue
     // Note that this only sets the JavaThread _priority field, which by
     // definition is limited to Java priorities and not OS priorities.
     // The os-priority is set in the CompilerThread startup code itself
+
     java_lang_Thread::set_priority(thread_oop(), NearMaxPriority);
-    // CLEANUP PRIORITIES: This -if- statement hids a bug whereby the compiler
-    // threads never have their OS priority set.  The assumption here is to
-    // enable the Performance group to do flag tuning, figure out a suitable
-    // CompilerThreadPriority, and then remove this 'if' statement (and
-    // comment) and unconditionally set the priority.
-
-    // Compiler Threads should be at the highest Priority
-    if ( CompilerThreadPriority != -1 )
-      os::set_native_priority( compiler_thread, CompilerThreadPriority );
-    else
-      os::set_native_priority( compiler_thread, os::java_to_os_priority[NearMaxPriority]);
-
-      // Note that I cannot call os::set_priority because it expects Java
-      // priorities and I am *explicitly* using OS priorities so that it's
-      // possible to set the compiler thread priority higher than any Java
-      // thread.
+
+    // Note that we cannot call os::set_priority because it expects Java
+    // priorities and we are *explicitly* using OS priorities so that it's
+    // possible to set the compiler thread priority higher than any Java
+    // thread.
+
+    int native_prio = CompilerThreadPriority;
+    if (native_prio == -1) {
+      if (UseCriticalCompilerThreadPriority) {
+        native_prio = os::java_to_os_priority[CriticalPriority];
+      } else {
+        native_prio = os::java_to_os_priority[NearMaxPriority];
+      }
+    }
+    os::set_native_priority(compiler_thread, native_prio);
 
     java_lang_Thread::set_daemon(thread_oop());
 
@@ -879,6 +879,7 @@ CompilerThread* CompileBroker::make_compiler_thread(const char* name, CompileQue
     Threads::add(compiler_thread);
     Thread::start(compiler_thread);
   }
+
   // Let go of Threads_lock before yielding
   os::yield(); // make sure that the compiler thread is started early (especially helpful on SOLARIS)
 
@@ -961,7 +962,7 @@ void CompileBroker::compile_method_base(methodHandle method,
                                         methodHandle hot_method,
                                         int hot_count,
                                         const char* comment,
-                                        TRAPS) {
+                                        Thread* thread) {
   // do nothing if compiler thread(s) is not available
   if (!_initialized ) {
     return;
@@ -1037,7 +1038,7 @@ void CompileBroker::compile_method_base(methodHandle method,
 
   // Acquire our lock.
   {
-    MutexLocker locker(queue->lock(), THREAD);
+    MutexLocker locker(queue->lock(), thread);
 
     // Make sure the method has not slipped into the queues since
     // last we checked; note that those checks were "fast bail-outs".
@@ -1119,7 +1120,7 @@ void CompileBroker::compile_method_base(methodHandle method,
 nmethod* CompileBroker::compile_method(methodHandle method, int osr_bci,
                                        int comp_level,
                                        methodHandle hot_method, int hot_count,
-                                       const char* comment, TRAPS) {
+                                       const char* comment, Thread* THREAD) {
   // make sure arguments make sense
   assert(method->method_holder()->klass_part()->oop_is_instance(), "not an instance method");
   assert(osr_bci == InvocationEntryBci || (0 <= osr_bci && osr_bci < method->code_size()), "bci out of range");
@@ -1173,10 +1174,10 @@ nmethod* CompileBroker::compile_method(methodHandle method, int osr_bci,
   assert(!HAS_PENDING_EXCEPTION, "No exception should be present");
   // some prerequisites that are compiler specific
   if (compiler(comp_level)->is_c2() || compiler(comp_level)->is_shark()) {
-    method->constants()->resolve_string_constants(CHECK_0);
+    method->constants()->resolve_string_constants(CHECK_AND_CLEAR_NULL);
     // Resolve all classes seen in the signature of the method
     // we are compiling.
-    methodOopDesc::load_signature_classes(method, CHECK_0);
+    methodOopDesc::load_signature_classes(method, CHECK_AND_CLEAR_NULL);
   }
 
   // If the method is native, do the lookup in the thread requesting
@@ -1230,7 +1231,7 @@ nmethod* CompileBroker::compile_method(methodHandle method, int osr_bci,
       return NULL;
     }
   } else {
-    compile_method_base(method, osr_bci, comp_level, hot_method, hot_count, comment, CHECK_0);
+    compile_method_base(method, osr_bci, comp_level, hot_method, hot_count, comment, THREAD);
   }
 
   // return requested nmethod

hotspot/src/share/vm/compiler/compileBroker.hpp

@@ -333,7 +333,7 @@ class CompileBroker: AllStatic {
                                   methodHandle hot_method,
                                   int hot_count,
                                   const char* comment,
-                                  TRAPS);
+                                  Thread* thread);
   static CompileQueue* compile_queue(int comp_level) {
     if (is_c2_compile(comp_level)) return _c2_method_queue;
     if (is_c1_compile(comp_level)) return _c1_method_queue;
@@ -363,7 +363,7 @@ class CompileBroker: AllStatic {
                                  int comp_level,
                                  methodHandle hot_method,
                                  int hot_count,
-                                 const char* comment, TRAPS);
+                                 const char* comment, Thread* thread);
 
   static void compiler_thread_loop();
 

hotspot/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepThread.cpp

 /*
- * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 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
@@ -75,10 +75,25 @@ ConcurrentMarkSweepThread::ConcurrentMarkSweepThread(CMSCollector* collector)
   set_name("Concurrent Mark-Sweep GC Thread");
 
   if (os::create_thread(this, os::cgc_thread)) {
-    // XXX: need to set this to low priority
-    // unless "agressive mode" set; priority
-    // should be just less than that of VMThread.
-    os::set_priority(this, NearMaxPriority);
+    // An old comment here said: "Priority should be just less
+    // than that of VMThread".  Since the VMThread runs at
+    // NearMaxPriority, the old comment was inaccurate, but
+    // changing the default priority to NearMaxPriority-1
+    // could change current behavior, so the default of
+    // NearMaxPriority stays in place.
+    //
+    // Note that there's a possibility of the VMThread
+    // starving if UseCriticalCMSThreadPriority is on.
+    // That won't happen on Solaris for various reasons,
+    // but may well happen on non-Solaris platforms.
+    int native_prio;
+    if (UseCriticalCMSThreadPriority) {
+      native_prio = os::java_to_os_priority[CriticalPriority];
+    } else {
+      native_prio = os::java_to_os_priority[NearMaxPriority];
+    }
+    os::set_native_priority(this, native_prio);
+
     if (!DisableStartThread) {
       os::start_thread(this);
     }

hotspot/src/share/vm/gc_implementation/g1/concurrentMark.hpp

@@ -84,8 +84,8 @@ class CMBitMapRO VALUE_OBJ_CLASS_SPEC {
   }
 
   // iteration
-  bool iterate(BitMapClosure* cl) { return _bm.iterate(cl); }
-  bool iterate(BitMapClosure* cl, MemRegion mr);
+  inline bool iterate(BitMapClosure* cl, MemRegion mr);
+  inline bool iterate(BitMapClosure* cl);
 
   // Return the address corresponding to the next marked bit at or after
   // "addr", and before "limit", if "limit" is non-NULL.  If there is no
@@ -349,10 +349,62 @@ typedef enum {
   high_verbose       // per object verbose
 } CMVerboseLevel;
 
+class YoungList;
+
+// Root Regions are regions that are not empty at the beginning of a
+// marking cycle and which we might collect during an evacuation pause
+// while the cycle is active. Given that, during evacuation pauses, we
+// do not copy objects that are explicitly marked, what we have to do
+// for the root regions is to scan them and mark all objects reachable
+// from them. According to the SATB assumptions, we only need to visit
+// each object once during marking. So, as long as we finish this scan
+// before the next evacuation pause, we can copy the objects from the
+// root regions without having to mark them or do anything else to them.
+//
+// Currently, we only support root region scanning once (at the start
+// of the marking cycle) and the root regions are all the survivor
+// regions populated during the initial-mark pause.
+class CMRootRegions VALUE_OBJ_CLASS_SPEC {
+private:
+  YoungList*           _young_list;
+  ConcurrentMark*      _cm;
+
+  volatile bool        _scan_in_progress;
+  volatile bool        _should_abort;
+  HeapRegion* volatile _next_survivor;
+
+public:
+  CMRootRegions();
+  // We actually do most of the initialization in this method.
+  void init(G1CollectedHeap* g1h, ConcurrentMark* cm);
+
+  // Reset the claiming / scanning of the root regions.
+  void prepare_for_scan();
+
+  // Forces get_next() to return NULL so that the iteration aborts early.
+  void abort() { _should_abort = true; }
+
+  // Return true if the CM thread are actively scanning root regions,
+  // false otherwise.
+  bool scan_in_progress() { return _scan_in_progress; }
+
+  // Claim the next root region to scan atomically, or return NULL if
+  // all have been claimed.
+  HeapRegion* claim_next();
+
+  // Flag that we're done with root region scanning and notify anyone
+  // who's waiting on it. If aborted is false, assume that all regions
+  // have been claimed.
+  void scan_finished();
+
+  // If CM threads are still scanning root regions, wait until they
+  // are done. Return true if we had to wait, false otherwise.
+  bool wait_until_scan_finished();
+};
 
 class ConcurrentMarkThread;
 
-class ConcurrentMark: public CHeapObj {
+class ConcurrentMark : public CHeapObj {
   friend class ConcurrentMarkThread;
   friend class CMTask;
   friend class CMBitMapClosure;
@@ -386,7 +438,7 @@ protected:
 
   FreeRegionList        _cleanup_list;
 
-  // CMS marking support structures
+  // Concurrent marking support structures
   CMBitMap                _markBitMap1;
   CMBitMap                _markBitMap2;
   CMBitMapRO*             _prevMarkBitMap; // completed mark bitmap
@@ -400,6 +452,9 @@ protected:
   HeapWord*               _heap_start;
   HeapWord*               _heap_end;
 
+  // Root region tracking and claiming.
+  CMRootRegions           _root_regions;
+
   // For gray objects
   CMMarkStack             _markStack; // Grey objects behind global finger.
   CMRegionStack           _regionStack; // Grey regions behind global finger.
@@ -426,7 +481,6 @@ protected:
   WorkGangBarrierSync     _first_overflow_barrier_sync;
   WorkGangBarrierSync     _second_overflow_barrier_sync;
 
-
   // this is set by any task, when an overflow on the global data
   // structures is detected.
   volatile bool           _has_overflown;
@@ -554,9 +608,9 @@ protected:
   bool has_overflown()           { return _has_overflown; }
   void set_has_overflown()       { _has_overflown = true; }
   void clear_has_overflown()     { _has_overflown = false; }
+  bool restart_for_overflow()    { return _restart_for_overflow; }
 
   bool has_aborted()             { return _has_aborted; }
-  bool restart_for_overflow()    { return _restart_for_overflow; }
 
   // Methods to enter the two overflow sync barriers
   void enter_first_sync_barrier(int task_num);
@@ -578,6 +632,27 @@ protected:
     }
   }
 
+  // Live Data Counting data structures...
+  // These data structures are initialized at the start of
+  // marking. They are written to while marking is active.
+  // They are aggregated during remark; the aggregated values
+  // are then used to populate the _region_bm, _card_bm, and
+  // the total live bytes, which are then subsequently updated
+  // during cleanup.
+
+  // An array of bitmaps (one bit map per task). Each bitmap
+  // is used to record the cards spanned by the live objects
+  // marked by that task/worker.
+  BitMap*  _count_card_bitmaps;
+
+  // Used to record the number of marked live bytes
+  // (for each region, by worker thread).
+  size_t** _count_marked_bytes;
+
+  // Card index of the bottom of the G1 heap. Used for biasing indices into
+  // the card bitmaps.
+  intptr_t _heap_bottom_card_num;
+
 public:
   // Manipulation of the global mark stack.
   // Notice that the first mark_stack_push is CAS-based, whereas the
@@ -671,6 +746,8 @@ public:
   // Returns true if there are any aborted memory regions.
   bool has_aborted_regions();
 
+  CMRootRegions* root_regions() { return &_root_regions; }
+
   bool concurrent_marking_in_progress() {
     return _concurrent_marking_in_progress;
   }
@@ -703,6 +780,7 @@ public:
 
   ConcurrentMark(ReservedSpace rs, int max_regions);
   ~ConcurrentMark();
+
   ConcurrentMarkThread* cmThread() { return _cmThread; }
 
   CMBitMapRO* prevMarkBitMap() const { return _prevMarkBitMap; }
@@ -720,8 +798,17 @@ public:
   // G1CollectedHeap
 
   // This notifies CM that a root during initial-mark needs to be
-  // grayed. It is MT-safe.
-  inline void grayRoot(oop obj, size_t word_size);
+  // grayed. It is MT-safe. word_size is the size of the object in
+  // words. It is passed explicitly as sometimes we cannot calculate
+  // it from the given object because it might be in an inconsistent
+  // state (e.g., in to-space and being copied). So the caller is
+  // responsible for dealing with this issue (e.g., get the size from
+  // the from-space image when the to-space image might be
+  // inconsistent) and always passing the size. hr is the region that
+  // contains the object and it's passed optionally from callers who
+  // might already have it (no point in recalculating it).
+  inline void grayRoot(oop obj, size_t word_size,
+                       uint worker_id, HeapRegion* hr = NULL);
 
   // It's used during evacuation pauses to gray a region, if
   // necessary, and it's MT-safe. It assumes that the caller has
@@ -772,6 +859,13 @@ public:
   void checkpointRootsInitialPre();
   void checkpointRootsInitialPost();
 
+  // Scan all the root regions and mark everything reachable from
+  // them.
+  void scanRootRegions();
+
+  // Scan a single root region and mark everything reachable from it.
+  void scanRootRegion(HeapRegion* hr, uint worker_id);
+
   // Do concurrent phase of marking, to a tentative transitive closure.
   void markFromRoots();
 
@@ -781,15 +875,13 @@ public:
 
   void checkpointRootsFinal(bool clear_all_soft_refs);
   void checkpointRootsFinalWork();
-  void calcDesiredRegions();
   void cleanup();
   void completeCleanup();
 
   // Mark in the previous bitmap.  NB: this is usually read-only, so use
   // this carefully!
   inline void markPrev(oop p);
-  inline void markNext(oop p);
-  void clear(oop p);
+
   // Clears marks for all objects in the given range, for the prev,
   // next, or both bitmaps.  NB: the previous bitmap is usually
   // read-only, so use this carefully!
@@ -913,6 +1005,114 @@ public:
   bool verbose_high() {
     return _MARKING_VERBOSE_ && _verbose_level >= high_verbose;
   }
+
+  // Counting data structure accessors
+
+  // Returns the card number of the bottom of the G1 heap.
+  // Used in biasing indices into accounting card bitmaps.
+  intptr_t heap_bottom_card_num() const {
+    return _heap_bottom_card_num;
+  }
+
+  // Returns the card bitmap for a given task or worker id.
+  BitMap* count_card_bitmap_for(uint worker_id) {
+    assert(0 <= worker_id && worker_id < _max_task_num, "oob");
+    assert(_count_card_bitmaps != NULL, "uninitialized");
+    BitMap* task_card_bm = &_count_card_bitmaps[worker_id];
+    assert(task_card_bm->size() == _card_bm.size(), "size mismatch");
+    return task_card_bm;
+  }
+
+  // Returns the array containing the marked bytes for each region,
+  // for the given worker or task id.
+  size_t* count_marked_bytes_array_for(uint worker_id) {
+    assert(0 <= worker_id && worker_id < _max_task_num, "oob");
+    assert(_count_marked_bytes != NULL, "uninitialized");
+    size_t* marked_bytes_array = _count_marked_bytes[worker_id];
+    assert(marked_bytes_array != NULL, "uninitialized");
+    return marked_bytes_array;
+  }
+
+  // Returns the index in the liveness accounting card table bitmap
+  // for the given address
+  inline BitMap::idx_t card_bitmap_index_for(HeapWord* addr);
+
+  // Counts the size of the given memory region in the the given
+  // marked_bytes array slot for the given HeapRegion.
+  // Sets the bits in the given card bitmap that are associated with the
+  // cards that are spanned by the memory region.
+  inline void count_region(MemRegion mr, HeapRegion* hr,
+                           size_t* marked_bytes_array,
+                           BitMap* task_card_bm);
+
+  // Counts the given memory region in the task/worker counting
+  // data structures for the given worker id.
+  inline void count_region(MemRegion mr, HeapRegion* hr, uint worker_id);
+
+  // Counts the given memory region in the task/worker counting
+  // data structures for the given worker id.
+  inline void count_region(MemRegion mr, uint worker_id);
+
+  // Counts the given object in the given task/worker counting
+  // data structures.
+  inline void count_object(oop obj, HeapRegion* hr,
+                           size_t* marked_bytes_array,
+                           BitMap* task_card_bm);
+
+  // Counts the given object in the task/worker counting data
+  // structures for the given worker id.
+  inline void count_object(oop obj, HeapRegion* hr, uint worker_id);
+
+  // Attempts to mark the given object and, if successful, counts
+  // the object in the given task/worker counting structures.
+  inline bool par_mark_and_count(oop obj, HeapRegion* hr,
+                                 size_t* marked_bytes_array,
+                                 BitMap* task_card_bm);
+
+  // Attempts to mark the given object and, if successful, counts
+  // the object in the task/worker counting structures for the
+  // given worker id.
+  inline bool par_mark_and_count(oop obj, size_t word_size,
+                                 HeapRegion* hr, uint worker_id);
+
+  // Attempts to mark the given object and, if successful, counts
+  // the object in the task/worker counting structures for the
+  // given worker id.
+  inline bool par_mark_and_count(oop obj, HeapRegion* hr, uint worker_id);
+
+  // Similar to the above routine but we don't know the heap region that
+  // contains the object to be marked/counted, which this routine looks up.
+  inline bool par_mark_and_count(oop obj, uint worker_id);
+
+  // Similar to the above routine but there are times when we cannot
+  // safely calculate the size of obj due to races and we, therefore,
+  // pass the size in as a parameter. It is the caller's reponsibility
+  // to ensure that the size passed in for obj is valid.
+  inline bool par_mark_and_count(oop obj, size_t word_size, uint worker_id);
+
+  // Unconditionally mark the given object, and unconditinally count
+  // the object in the counting structures for worker id 0.
+  // Should *not* be called from parallel code.
+  inline bool mark_and_count(oop obj, HeapRegion* hr);
+
+  // Similar to the above routine but we don't know the heap region that
+  // contains the object to be marked/counted, which this routine looks up.
+  // Should *not* be called from parallel code.
+  inline bool mark_and_count(oop obj);
+
+protected:
+  // Clear all the per-task bitmaps and arrays used to store the
+  // counting data.
+  void clear_all_count_data();
+
+  // Aggregates the counting data for each worker/task
+  // that was constructed while marking. Also sets
+  // the amount of marked bytes for each region and
+  // the top at concurrent mark count.
+  void aggregate_count_data();
+
+  // Verification routine
+  void verify_count_data();
 };
 
 // A class representing a marking task.
@@ -1031,6 +1231,12 @@ private:
 
   TruncatedSeq                _marking_step_diffs_ms;
 
+  // Counting data structures. Embedding the task's marked_bytes_array
+  // and card bitmap into the actual task saves having to go through
+  // the ConcurrentMark object.
+  size_t*                     _marked_bytes_array;
+  BitMap*                     _card_bm;
+
   // LOTS of statistics related with this task
 #if _MARKING_STATS_
   NumberSeq                   _all_clock_intervals_ms;
@@ -1196,6 +1402,7 @@ public:
   }
 
   CMTask(int task_num, ConcurrentMark *cm,
+         size_t* marked_bytes, BitMap* card_bm,
          CMTaskQueue* task_queue, CMTaskQueueSet* task_queues);
 
   // it prints statistics associated with this task

hotspot/src/share/vm/gc_implementation/g1/concurrentMark.inline.hpp

@@ -28,6 +28,214 @@
 #include "gc_implementation/g1/concurrentMark.hpp"
 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
 
+// Returns the index in the liveness accounting card bitmap
+// for the given address
+inline BitMap::idx_t ConcurrentMark::card_bitmap_index_for(HeapWord* addr) {
+  // Below, the term "card num" means the result of shifting an address
+  // by the card shift -- address 0 corresponds to card number 0.  One
+  // must subtract the card num of the bottom of the heap to obtain a
+  // card table index.
+
+  intptr_t card_num = intptr_t(uintptr_t(addr) >> CardTableModRefBS::card_shift);
+  return card_num - heap_bottom_card_num();
+}
+
+// Counts the given memory region in the given task/worker
+// counting data structures.
+inline void ConcurrentMark::count_region(MemRegion mr, HeapRegion* hr,
+                                         size_t* marked_bytes_array,
+                                         BitMap* task_card_bm) {
+  G1CollectedHeap* g1h = _g1h;
+  HeapWord* start = mr.start();
+  HeapWord* last = mr.last();
+  size_t region_size_bytes = mr.byte_size();
+  size_t index = hr->hrs_index();
+
+  assert(!hr->continuesHumongous(), "should not be HC region");
+  assert(hr == g1h->heap_region_containing(start), "sanity");
+  assert(hr == g1h->heap_region_containing(mr.last()), "sanity");
+  assert(marked_bytes_array != NULL, "pre-condition");
+  assert(task_card_bm != NULL, "pre-condition");
+
+  // Add to the task local marked bytes for this region.
+  marked_bytes_array[index] += region_size_bytes;
+
+  BitMap::idx_t start_idx = card_bitmap_index_for(start);
+  BitMap::idx_t last_idx = card_bitmap_index_for(last);
+
+  // The card bitmap is task/worker specific => no need to use 'par' routines.
+  // Set bits in the inclusive bit range [start_idx, last_idx].
+  //
+  // For small ranges use a simple loop; otherwise use set_range
+  // The range are the cards that are spanned by the object/region
+  // so 8 cards will allow objects/regions up to 4K to be handled
+  // using the loop.
+  if ((last_idx - start_idx) <= 8) {
+    for (BitMap::idx_t i = start_idx; i <= last_idx; i += 1) {
+     task_card_bm->set_bit(i);
+    }
+  } else {
+    assert(last_idx < task_card_bm->size(), "sanity");
+    // Note: BitMap::set_range() is exclusive.
+    task_card_bm->set_range(start_idx, last_idx+1);
+  }
+}
+
+// Counts the given memory region in the task/worker counting
+// data structures for the given worker id.
+inline void ConcurrentMark::count_region(MemRegion mr,
+                                         HeapRegion* hr,
+                                         uint worker_id) {
+  size_t* marked_bytes_array = count_marked_bytes_array_for(worker_id);
+  BitMap* task_card_bm = count_card_bitmap_for(worker_id);
+  count_region(mr, hr, marked_bytes_array, task_card_bm);
+}
+
+// Counts the given memory region, which may be a single object, in the
+// task/worker counting data structures for the given worker id.
+inline void ConcurrentMark::count_region(MemRegion mr, uint worker_id) {
+  HeapWord* addr = mr.start();
+  HeapRegion* hr = _g1h->heap_region_containing_raw(addr);
+  count_region(mr, hr, worker_id);
+}
+
+// Counts the given object in the given task/worker counting data structures.
+inline void ConcurrentMark::count_object(oop obj,
+                                         HeapRegion* hr,
+                                         size_t* marked_bytes_array,
+                                         BitMap* task_card_bm) {
+  MemRegion mr((HeapWord*)obj, obj->size());
+  count_region(mr, hr, marked_bytes_array, task_card_bm);
+}
+
+// Counts the given object in the task/worker counting data
+// structures for the given worker id.
+inline void ConcurrentMark::count_object(oop obj,
+                                         HeapRegion* hr,
+                                         uint worker_id) {
+  size_t* marked_bytes_array = count_marked_bytes_array_for(worker_id);
+  BitMap* task_card_bm = count_card_bitmap_for(worker_id);
+  HeapWord* addr = (HeapWord*) obj;
+  count_object(obj, hr, marked_bytes_array, task_card_bm);
+}
+
+// Attempts to mark the given object and, if successful, counts
+// the object in the given task/worker counting structures.
+inline bool ConcurrentMark::par_mark_and_count(oop obj,
+                                               HeapRegion* hr,
+                                               size_t* marked_bytes_array,
+                                               BitMap* task_card_bm) {
+  HeapWord* addr = (HeapWord*)obj;
+  if (_nextMarkBitMap->parMark(addr)) {
+    // Update the task specific count data for the object.
+    count_object(obj, hr, marked_bytes_array, task_card_bm);
+    return true;
+  }
+  return false;
+}
+
+// Attempts to mark the given object and, if successful, counts
+// the object in the task/worker counting structures for the
+// given worker id.
+inline bool ConcurrentMark::par_mark_and_count(oop obj,
+                                               size_t word_size,
+                                               HeapRegion* hr,
+                                               uint worker_id) {
+  HeapWord* addr = (HeapWord*)obj;
+  if (_nextMarkBitMap->parMark(addr)) {
+    MemRegion mr(addr, word_size);
+    count_region(mr, hr, worker_id);
+    return true;
+  }
+  return false;
+}
+
+// Attempts to mark the given object and, if successful, counts
+// the object in the task/worker counting structures for the
+// given worker id.
+inline bool ConcurrentMark::par_mark_and_count(oop obj,
+                                               HeapRegion* hr,
+                                               uint worker_id) {
+  HeapWord* addr = (HeapWord*)obj;
+  if (_nextMarkBitMap->parMark(addr)) {
+    // Update the task specific count data for the object.
+    count_object(obj, hr, worker_id);
+    return true;
+  }
+  return false;
+}
+
+// As above - but we don't know the heap region containing the
+// object and so have to supply it.
+inline bool ConcurrentMark::par_mark_and_count(oop obj, uint worker_id) {
+  HeapWord* addr = (HeapWord*)obj;
+  HeapRegion* hr = _g1h->heap_region_containing_raw(addr);
+  return par_mark_and_count(obj, hr, worker_id);
+}
+
+// Similar to the above routine but we already know the size, in words, of
+// the object that we wish to mark/count
+inline bool ConcurrentMark::par_mark_and_count(oop obj,
+                                               size_t word_size,
+                                               uint worker_id) {
+  HeapWord* addr = (HeapWord*)obj;
+  if (_nextMarkBitMap->parMark(addr)) {
+    // Update the task specific count data for the object.
+    MemRegion mr(addr, word_size);
+    count_region(mr, worker_id);
+    return true;
+  }
+  return false;
+}
+
+// Unconditionally mark the given object, and unconditinally count
+// the object in the counting structures for worker id 0.
+// Should *not* be called from parallel code.
+inline bool ConcurrentMark::mark_and_count(oop obj, HeapRegion* hr) {
+  HeapWord* addr = (HeapWord*)obj;
+  _nextMarkBitMap->mark(addr);
+  // Update the task specific count data for the object.
+  count_object(obj, hr, 0 /* worker_id */);
+  return true;
+}
+
+// As above - but we don't have the heap region containing the
+// object, so we have to supply it.
+inline bool ConcurrentMark::mark_and_count(oop obj) {
+  HeapWord* addr = (HeapWord*)obj;
+  HeapRegion* hr = _g1h->heap_region_containing_raw(addr);
+  return mark_and_count(obj, hr);
+}
+
+inline bool CMBitMapRO::iterate(BitMapClosure* cl, MemRegion mr) {
+  HeapWord* start_addr = MAX2(startWord(), mr.start());
+  HeapWord* end_addr = MIN2(endWord(), mr.end());
+
+  if (end_addr > start_addr) {
+    // Right-open interval [start-offset, end-offset).
+    BitMap::idx_t start_offset = heapWordToOffset(start_addr);
+    BitMap::idx_t end_offset = heapWordToOffset(end_addr);
+
+    start_offset = _bm.get_next_one_offset(start_offset, end_offset);
+    while (start_offset < end_offset) {
+      HeapWord* obj_addr = offsetToHeapWord(start_offset);
+      oop obj = (oop) obj_addr;
+      if (!cl->do_bit(start_offset)) {
+        return false;
+      }
+      HeapWord* next_addr = MIN2(obj_addr + obj->size(), end_addr);
+      BitMap::idx_t next_offset = heapWordToOffset(next_addr);
+      start_offset = _bm.get_next_one_offset(next_offset, end_offset);
+    }
+  }
+  return true;
+}
+
+inline bool CMBitMapRO::iterate(BitMapClosure* cl) {
+  MemRegion mr(startWord(), sizeInWords());
+  return iterate(cl, mr);
+}
+
 inline void CMTask::push(oop obj) {
   HeapWord* objAddr = (HeapWord*) obj;
   assert(_g1h->is_in_g1_reserved(objAddr), "invariant");
@@ -84,7 +292,7 @@ inline void CMTask::deal_with_reference(oop obj) {
 
   HeapWord* objAddr = (HeapWord*) obj;
   assert(obj->is_oop_or_null(true /* ignore mark word */), "Error");
- if (_g1h->is_in_g1_reserved(objAddr)) {
+  if (_g1h->is_in_g1_reserved(objAddr)) {
     assert(obj != NULL, "null check is implicit");
     if (!_nextMarkBitMap->isMarked(objAddr)) {
       // Only get the containing region if the object is not marked on the
@@ -98,9 +306,9 @@ inline void CMTask::deal_with_reference(oop obj) {
         }
 
         // we need to mark it first
-        if (_nextMarkBitMap->parMark(objAddr)) {
+        if (_cm->par_mark_and_count(obj, hr, _marked_bytes_array, _card_bm)) {
           // No OrderAccess:store_load() is needed. It is implicit in the
-          // CAS done in parMark(objAddr) above
+          // CAS done in CMBitMap::parMark() call in the routine above.
           HeapWord* global_finger = _cm->finger();
 
 #if _CHECK_BOTH_FINGERS_
@@ -160,25 +368,20 @@ inline void ConcurrentMark::markPrev(oop p) {
   ((CMBitMap*)_prevMarkBitMap)->mark((HeapWord*) p);
 }
 
-inline void ConcurrentMark::markNext(oop p) {
-  assert(!_nextMarkBitMap->isMarked((HeapWord*) p), "sanity");
-  _nextMarkBitMap->mark((HeapWord*) p);
-}
-
-inline void ConcurrentMark::grayRoot(oop obj, size_t word_size) {
+inline void ConcurrentMark::grayRoot(oop obj, size_t word_size,
+                                     uint worker_id, HeapRegion* hr) {
+  assert(obj != NULL, "pre-condition");
   HeapWord* addr = (HeapWord*) obj;
-
-  // Currently we don't do anything with word_size but we will use it
-  // in the very near future in the liveness calculation piggy-backing
-  // changes.
-
-#ifdef ASSERT
-  HeapRegion* hr = _g1h->heap_region_containing(addr);
+  if (hr == NULL) {
+    hr = _g1h->heap_region_containing_raw(addr);
+  } else {
+    assert(hr->is_in(addr), "pre-condition");
+  }
   assert(hr != NULL, "sanity");
-  assert(!hr->is_survivor(), "should not allocate survivors during IM");
-  assert(addr < hr->next_top_at_mark_start(),
-         err_msg("addr: "PTR_FORMAT" hr: "HR_FORMAT" NTAMS: "PTR_FORMAT,
-                 addr, HR_FORMAT_PARAMS(hr), hr->next_top_at_mark_start()));
+  // Given that we're looking for a region that contains an object
+  // header it's impossible to get back a HC region.
+  assert(!hr->continuesHumongous(), "sanity");
+
   // We cannot assert that word_size == obj->size() given that obj
   // might not be in a consistent state (another thread might be in
   // the process of copying it). So the best thing we can do is to
@@ -188,10 +391,11 @@ inline void ConcurrentMark::grayRoot(oop obj, size_t word_size) {
          err_msg("size: "SIZE_FORMAT" capacity: "SIZE_FORMAT" "HR_FORMAT,
                  word_size * HeapWordSize, hr->capacity(),
                  HR_FORMAT_PARAMS(hr)));
-#endif // ASSERT
 
-  if (!_nextMarkBitMap->isMarked(addr)) {
-    _nextMarkBitMap->parMark(addr);
+  if (addr < hr->next_top_at_mark_start()) {
+    if (!_nextMarkBitMap->isMarked(addr)) {
+      par_mark_and_count(obj, word_size, hr, worker_id);
+    }
   }
 }
 

hotspot/src/share/vm/gc_implementation/g1/concurrentMarkThread.cpp

 /*
- * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 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
@@ -44,9 +44,7 @@ ConcurrentMarkThread::ConcurrentMarkThread(ConcurrentMark* cm) :
   _started(false),
   _in_progress(false),
   _vtime_accum(0.0),
-  _vtime_mark_accum(0.0),
-  _vtime_count_accum(0.0)
-{
+  _vtime_mark_accum(0.0) {
   create_and_start();
 }
 
@@ -94,9 +92,36 @@ void ConcurrentMarkThread::run() {
       ResourceMark rm;
       HandleMark   hm;
       double cycle_start = os::elapsedVTime();
-      double mark_start_sec = os::elapsedTime();
       char verbose_str[128];
 
+      // We have to ensure that we finish scanning the root regions
+      // before the next GC takes place. To ensure this we have to
+      // make sure that we do not join the STS until the root regions
+      // have been scanned. If we did then it's possible that a
+      // subsequent GC could block us from joining the STS and proceed
+      // without the root regions have been scanned which would be a
+      // correctness issue.
+
+      double scan_start = os::elapsedTime();
+      if (!cm()->has_aborted()) {
+        if (PrintGC) {
+          gclog_or_tty->date_stamp(PrintGCDateStamps);
+          gclog_or_tty->stamp(PrintGCTimeStamps);
+          gclog_or_tty->print_cr("[GC concurrent-root-region-scan-start]");
+        }
+
+        _cm->scanRootRegions();
+
+        double scan_end = os::elapsedTime();
+        if (PrintGC) {
+          gclog_or_tty->date_stamp(PrintGCDateStamps);
+          gclog_or_tty->stamp(PrintGCTimeStamps);
+          gclog_or_tty->print_cr("[GC concurrent-root-region-scan-end, %1.7lf]",
+                                 scan_end - scan_start);
+        }
+      }
+
+      double mark_start_sec = os::elapsedTime();
       if (PrintGC) {
         gclog_or_tty->date_stamp(PrintGCDateStamps);
         gclog_or_tty->stamp(PrintGCTimeStamps);
@@ -148,36 +173,12 @@ void ConcurrentMarkThread::run() {
         }
       } while (cm()->restart_for_overflow());
 
-      double counting_start_time = os::elapsedVTime();
-      if (!cm()->has_aborted()) {
-        double count_start_sec = os::elapsedTime();
-        if (PrintGC) {
-          gclog_or_tty->date_stamp(PrintGCDateStamps);
-          gclog_or_tty->stamp(PrintGCTimeStamps);
-          gclog_or_tty->print_cr("[GC concurrent-count-start]");
-        }
-
-        _sts.join();
-        _cm->calcDesiredRegions();
-        _sts.leave();
-
-        if (!cm()->has_aborted()) {
-          double count_end_sec = os::elapsedTime();
-          if (PrintGC) {
-            gclog_or_tty->date_stamp(PrintGCDateStamps);
-            gclog_or_tty->stamp(PrintGCTimeStamps);
-            gclog_or_tty->print_cr("[GC concurrent-count-end, %1.7lf]",
-                                   count_end_sec - count_start_sec);
-          }
-        }
-      }
-
       double end_time = os::elapsedVTime();
-      _vtime_count_accum += (end_time - counting_start_time);
       // Update the total virtual time before doing this, since it will try
       // to measure it to get the vtime for this marking.  We purposely
       // neglect the presumably-short "completeCleanup" phase here.
       _vtime_accum = (end_time - _vtime_start);
+
       if (!cm()->has_aborted()) {
         if (g1_policy->adaptive_young_list_length()) {
           double now = os::elapsedTime();

hotspot/src/share/vm/gc_implementation/g1/concurrentMarkThread.hpp

 /*
- * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 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
@@ -40,7 +40,6 @@ class ConcurrentMarkThread: public ConcurrentGCThread {
   double _vtime_accum;  // Accumulated virtual time.
 
   double _vtime_mark_accum;
-  double _vtime_count_accum;
 
  public:
   virtual void run();
@@ -69,8 +68,6 @@ class ConcurrentMarkThread: public ConcurrentGCThread {
   double vtime_accum();
   // Marking virtual time so far
   double vtime_mark_accum();
-  // Counting virtual time so far.
-  double vtime_count_accum() { return _vtime_count_accum; }
 
   ConcurrentMark* cm()     { return _cm; }
 

hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp

@@ -174,13 +174,10 @@ public:
   }
 };
 
-YoungList::YoungList(G1CollectedHeap* g1h)
-  : _g1h(g1h), _head(NULL),
-    _length(0),
-    _last_sampled_rs_lengths(0),
-    _survivor_head(NULL), _survivor_tail(NULL), _survivor_length(0)
-{
-  guarantee( check_list_empty(false), "just making sure..." );
+YoungList::YoungList(G1CollectedHeap* g1h) :
+    _g1h(g1h), _head(NULL), _length(0), _last_sampled_rs_lengths(0),
+    _survivor_head(NULL), _survivor_tail(NULL), _survivor_length(0) {
+  guarantee(check_list_empty(false), "just making sure...");
 }
 
 void YoungList::push_region(HeapRegion *hr) {
@@ -1029,6 +1026,15 @@ HeapWord* G1CollectedHeap::attempt_allocation_humongous(size_t word_size,
   assert(isHumongous(word_size), "attempt_allocation_humongous() "
          "should only be called for humongous allocations");
 
+  // Humongous objects can exhaust the heap quickly, so we should check if we
+  // need to start a marking cycle at each humongous object allocation. We do
+  // the check before we do the actual allocation. The reason for doing it
+  // before the allocation is that we avoid having to keep track of the newly
+  // allocated memory while we do a GC.
+  if (g1_policy()->need_to_start_conc_mark("concurrent humongous allocation", word_size)) {
+    collect(GCCause::_g1_humongous_allocation);
+  }
+
   // We will loop until a) we manage to successfully perform the
   // allocation or b) we successfully schedule a collection which
   // fails to perform the allocation. b) is the only case when we'll
@@ -1111,7 +1117,11 @@ HeapWord* G1CollectedHeap::attempt_allocation_at_safepoint(size_t word_size,
     return _mutator_alloc_region.attempt_allocation_locked(word_size,
                                                       false /* bot_updates */);
   } else {
-    return humongous_obj_allocate(word_size);
+    HeapWord* result = humongous_obj_allocate(word_size);
+    if (result != NULL && g1_policy()->need_to_start_conc_mark("STW humongous allocation")) {
+      g1_policy()->set_initiate_conc_mark_if_possible();
+    }
+    return result;
   }
 
   ShouldNotReachHere();
@@ -1257,7 +1267,18 @@ bool G1CollectedHeap::do_collection(bool explicit_gc,
     double start = os::elapsedTime();
     g1_policy()->record_full_collection_start();
 
+    // Note: When we have a more flexible GC logging framework that
+    // allows us to add optional attributes to a GC log record we
+    // could consider timing and reporting how long we wait in the
+    // following two methods.
     wait_while_free_regions_coming();
+    // If we start the compaction before the CM threads finish
+    // scanning the root regions we might trip them over as we'll
+    // be moving objects / updating references. So let's wait until
+    // they are done. By telling them to abort, they should complete
+    // early.
+    _cm->root_regions()->abort();
+    _cm->root_regions()->wait_until_scan_finished();
     append_secondary_free_list_if_not_empty_with_lock();
 
     gc_prologue(true);
@@ -1286,7 +1307,8 @@ bool G1CollectedHeap::do_collection(bool explicit_gc,
     ref_processor_cm()->verify_no_references_recorded();
 
     // Abandon current iterations of concurrent marking and concurrent
-    // refinement, if any are in progress.
+    // refinement, if any are in progress. We have to do this before
+    // wait_until_scan_finished() below.
     concurrent_mark()->abort();
 
     // Make sure we'll choose a new allocation region afterwards.
@@ -2295,7 +2317,8 @@ size_t G1CollectedHeap::unsafe_max_alloc() {
 bool G1CollectedHeap::should_do_concurrent_full_gc(GCCause::Cause cause) {
   return
     ((cause == GCCause::_gc_locker           && GCLockerInvokesConcurrent) ||
-     (cause == GCCause::_java_lang_system_gc && ExplicitGCInvokesConcurrent));
+     (cause == GCCause::_java_lang_system_gc && ExplicitGCInvokesConcurrent) ||
+      cause == GCCause::_g1_humongous_allocation);
 }
 
 #ifndef PRODUCT
@@ -3545,19 +3568,25 @@ G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) {
   verify_region_sets_optional();
   verify_dirty_young_regions();
 
-  {
-    // This call will decide whether this pause is an initial-mark
-    // pause. If it is, during_initial_mark_pause() will return true
-    // for the duration of this pause.
-    g1_policy()->decide_on_conc_mark_initiation();
+  // This call will decide whether this pause is an initial-mark
+  // pause. If it is, during_initial_mark_pause() will return true
+  // for the duration of this pause.
+  g1_policy()->decide_on_conc_mark_initiation();
 
-    // We do not allow initial-mark to be piggy-backed on a mixed GC.
-    assert(!g1_policy()->during_initial_mark_pause() ||
-            g1_policy()->gcs_are_young(), "sanity");
+  // We do not allow initial-mark to be piggy-backed on a mixed GC.
+  assert(!g1_policy()->during_initial_mark_pause() ||
+          g1_policy()->gcs_are_young(), "sanity");
 
-    // We also do not allow mixed GCs during marking.
-    assert(!mark_in_progress() || g1_policy()->gcs_are_young(), "sanity");
+  // We also do not allow mixed GCs during marking.
+  assert(!mark_in_progress() || g1_policy()->gcs_are_young(), "sanity");
 
+  // Record whether this pause is an initial mark. When the current
+  // thread has completed its logging output and it's safe to signal
+  // the CM thread, the flag's value in the policy has been reset.
+  bool should_start_conc_mark = g1_policy()->during_initial_mark_pause();
+
+  // Inner scope for scope based logging, timers, and stats collection
+  {
     char verbose_str[128];
     sprintf(verbose_str, "GC pause ");
     if (g1_policy()->gcs_are_young()) {
@@ -3613,7 +3642,6 @@ G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) {
         Universe::verify(/* allow dirty */ false,
                          /* silent      */ false,
                          /* option      */ VerifyOption_G1UsePrevMarking);
-
       }
 
       COMPILER2_PRESENT(DerivedPointerTable::clear());
@@ -3656,6 +3684,18 @@ G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) {
         g1_policy()->record_collection_pause_start(start_time_sec,
                                                    start_used_bytes);
 
+        double scan_wait_start = os::elapsedTime();
+        // We have to wait until the CM threads finish scanning the
+        // root regions as it's the only way to ensure that all the
+        // objects on them have been correctly scanned before we start
+        // moving them during the GC.
+        bool waited = _cm->root_regions()->wait_until_scan_finished();
+        if (waited) {
+          double scan_wait_end = os::elapsedTime();
+          double wait_time_ms = (scan_wait_end - scan_wait_start) * 1000.0;
+          g1_policy()->record_root_region_scan_wait_time(wait_time_ms);
+        }
+
 #if YOUNG_LIST_VERBOSE
         gclog_or_tty->print_cr("\nAfter recording pause start.\nYoung_list:");
         _young_list->print();
@@ -3765,16 +3805,14 @@ G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) {
         }
 
         if (g1_policy()->during_initial_mark_pause()) {
+          // We have to do this before we notify the CM threads that
+          // they can start working to make sure that all the
+          // appropriate initialization is done on the CM object.
           concurrent_mark()->checkpointRootsInitialPost();
           set_marking_started();
-          // CAUTION: after the doConcurrentMark() call below,
-          // the concurrent marking thread(s) could be running
-          // concurrently with us. Make sure that anything after
-          // this point does not assume that we are the only GC thread
-          // running. Note: of course, the actual marking work will
-          // not start until the safepoint itself is released in
-          // ConcurrentGCThread::safepoint_desynchronize().
-          doConcurrentMark();
+          // Note that we don't actually trigger the CM thread at
+          // this point. We do that later when we're sure that
+          // the current thread has completed its logging output.
         }
 
         allocate_dummy_regions();
@@ -3884,6 +3922,15 @@ G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) {
     }
   }
 
+  // The closing of the inner scope, immediately above, will complete
+  // the PrintGC logging output. The record_collection_pause_end() call
+  // above will complete the logging output of PrintGCDetails.
+  //
+  // It is not yet to safe, however, to tell the concurrent mark to
+  // start as we have some optional output below. We don't want the
+  // output from the concurrent mark thread interfering with this
+  // logging output either.
+
   _hrs.verify_optional();
   verify_region_sets_optional();
 
@@ -3901,6 +3948,21 @@ G1CollectedHeap::do_collection_pause_at_safepoint(double target_pause_time_ms) {
     g1_rem_set()->print_summary_info();
   }
 
+  // It should now be safe to tell the concurrent mark thread to start
+  // without its logging output interfering with the logging output
+  // that came from the pause.
+
+  if (should_start_conc_mark) {
+    // CAUTION: after the doConcurrentMark() call below,
+    // the concurrent marking thread(s) could be running
+    // concurrently with us. Make sure that anything after
+    // this point does not assume that we are the only GC thread
+    // running. Note: of course, the actual marking work will
+    // not start until the safepoint itself is released in
+    // ConcurrentGCThread::safepoint_desynchronize().
+    doConcurrentMark();
+  }
+
   return true;
 }
 
@@ -4162,7 +4224,7 @@ HeapWord* G1CollectedHeap::par_allocate_during_gc(GCAllocPurpose purpose,
 G1ParGCAllocBuffer::G1ParGCAllocBuffer(size_t gclab_word_size) :
   ParGCAllocBuffer(gclab_word_size), _retired(false) { }
 
-G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, int queue_num)
+G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num)
   : _g1h(g1h),
     _refs(g1h->task_queue(queue_num)),
     _dcq(&g1h->dirty_card_queue_set()),
@@ -4283,6 +4345,7 @@ G1ParClosureSuper::G1ParClosureSuper(G1CollectedHeap* g1,
                                      G1ParScanThreadState* par_scan_state) :
   _g1(g1), _g1_rem(_g1->g1_rem_set()), _cm(_g1->concurrent_mark()),
   _par_scan_state(par_scan_state),
+  _worker_id(par_scan_state->queue_num()),
   _during_initial_mark(_g1->g1_policy()->during_initial_mark_pause()),
   _mark_in_progress(_g1->mark_in_progress()) { }
 
@@ -4294,7 +4357,7 @@ void G1ParCopyHelper::mark_object(oop obj) {
 #endif // ASSERT
 
   // We know that the object is not moving so it's safe to read its size.
-  _cm->grayRoot(obj, (size_t) obj->size());
+  _cm->grayRoot(obj, (size_t) obj->size(), _worker_id);
 }
 
 void G1ParCopyHelper::mark_forwarded_object(oop from_obj, oop to_obj) {
@@ -4316,7 +4379,7 @@ void G1ParCopyHelper::mark_forwarded_object(oop from_obj, oop to_obj) {
   // worker so we cannot trust that its to-space image is
   // well-formed. So we have to read its size from its from-space
   // image which we know should not be changing.
-  _cm->grayRoot(to_obj, (size_t) from_obj->size());
+  _cm->grayRoot(to_obj, (size_t) from_obj->size(), _worker_id);
 }
 
 oop G1ParCopyHelper::copy_to_survivor_space(oop old) {
@@ -4406,6 +4469,8 @@ void G1ParCopyClosure<do_gen_barrier, barrier, do_mark_object>
   assert(barrier != G1BarrierRS || obj != NULL,
          "Precondition: G1BarrierRS implies obj is non-NULL");
 
+  assert(_worker_id == _par_scan_state->queue_num(), "sanity");
+
   // here the null check is implicit in the cset_fast_test() test
   if (_g1->in_cset_fast_test(obj)) {
     oop forwardee;
@@ -4424,7 +4489,7 @@ void G1ParCopyClosure<do_gen_barrier, barrier, do_mark_object>
 
     // When scanning the RS, we only care about objs in CS.
     if (barrier == G1BarrierRS) {
-      _par_scan_state->update_rs(_from, p, _par_scan_state->queue_num());
+      _par_scan_state->update_rs(_from, p, _worker_id);
     }
   } else {
     // The object is not in collection set. If we're a root scanning
@@ -4436,7 +4501,7 @@ void G1ParCopyClosure<do_gen_barrier, barrier, do_mark_object>
   }
 
   if (barrier == G1BarrierEvac && obj != NULL) {
-    _par_scan_state->update_rs(_from, p, _par_scan_state->queue_num());
+    _par_scan_state->update_rs(_from, p, _worker_id);
   }
 
   if (do_gen_barrier && obj != NULL) {
@@ -5666,16 +5731,6 @@ void G1CollectedHeap::free_collection_set(HeapRegion* cs_head) {
 
       // And the region is empty.
       assert(!used_mr.is_empty(), "Should not have empty regions in a CS.");
-
-      // If marking is in progress then clear any objects marked in
-      // the current region. Note mark_in_progress() returns false,
-      // even during an initial mark pause, until the set_marking_started()
-      // call which takes place later in the pause.
-      if (mark_in_progress()) {
-        assert(!g1_policy()->during_initial_mark_pause(), "sanity");
-        _cm->nextMarkBitMap()->clearRange(used_mr);
-      }
-
       free_region(cur, &pre_used, &local_free_list, false /* par */);
     } else {
       cur->uninstall_surv_rate_group();
@@ -5742,8 +5797,9 @@ void G1CollectedHeap::set_free_regions_coming() {
 }
 
 void G1CollectedHeap::reset_free_regions_coming() {
+  assert(free_regions_coming(), "pre-condition");
+
   {
-    assert(free_regions_coming(), "pre-condition");
     MutexLockerEx x(SecondaryFreeList_lock, Mutex::_no_safepoint_check_flag);
     _free_regions_coming = false;
     SecondaryFreeList_lock->notify_all();

hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp

@@ -355,6 +355,7 @@ private:
   // explicitly started if:
   // (a) cause == _gc_locker and +GCLockerInvokesConcurrent, or
   // (b) cause == _java_lang_system_gc and +ExplicitGCInvokesConcurrent.
+  // (c) cause == _g1_humongous_allocation
   bool should_do_concurrent_full_gc(GCCause::Cause cause);
 
   // Keeps track of how many "full collections" (i.e., Full GCs or
@@ -1172,6 +1173,10 @@ public:
     _old_set.remove(hr);
   }
 
+  size_t non_young_capacity_bytes() {
+    return _old_set.total_capacity_bytes() + _humongous_set.total_capacity_bytes();
+  }
+
   void set_free_regions_coming();
   void reset_free_regions_coming();
   bool free_regions_coming() { return _free_regions_coming; }
@@ -1904,7 +1909,7 @@ protected:
   G1ParScanPartialArrayClosure* _partial_scan_cl;
 
   int _hash_seed;
-  int _queue_num;
+  uint _queue_num;
 
   size_t _term_attempts;
 
@@ -1948,7 +1953,7 @@ protected:
   }
 
 public:
-  G1ParScanThreadState(G1CollectedHeap* g1h, int queue_num);
+  G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num);
 
   ~G1ParScanThreadState() {
     FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base);
@@ -2040,7 +2045,7 @@ public:
   }
 
   int* hash_seed() { return &_hash_seed; }
-  int  queue_num() { return _queue_num; }
+  uint queue_num() { return _queue_num; }
 
   size_t term_attempts() const  { return _term_attempts; }
   void note_term_attempt() { _term_attempts++; }

hotspot/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp

@@ -141,6 +141,7 @@ G1CollectorPolicy::G1CollectorPolicy() :
 
   _cur_clear_ct_time_ms(0.0),
   _mark_closure_time_ms(0.0),
+  _root_region_scan_wait_time_ms(0.0),
 
   _cur_ref_proc_time_ms(0.0),
   _cur_ref_enq_time_ms(0.0),
@@ -213,8 +214,6 @@ G1CollectorPolicy::G1CollectorPolicy() :
   _survivor_bytes_before_gc(0),
   _capacity_before_gc(0),
 
-  _prev_collection_pause_used_at_end_bytes(0),
-
   _eden_cset_region_length(0),
   _survivor_cset_region_length(0),
   _old_cset_region_length(0),
@@ -905,19 +904,10 @@ void G1CollectorPolicy::record_collection_pause_start(double start_time_sec,
     gclog_or_tty->print(" (%s)", gcs_are_young() ? "young" : "mixed");
   }
 
-  if (!during_initial_mark_pause()) {
-    // We only need to do this here as the policy will only be applied
-    // to the GC we're about to start. so, no point is calculating this
-    // every time we calculate / recalculate the target young length.
-    update_survivors_policy();
-  } else {
-    // The marking phase has a "we only copy implicitly live
-    // objects during marking" invariant. The easiest way to ensure it
-    // holds is not to allocate any survivor regions and tenure all
-    // objects. In the future we might change this and handle survivor
-    // regions specially during marking.
-    tenure_all_objects();
-  }
+  // We only need to do this here as the policy will only be applied
+  // to the GC we're about to start. so, no point is calculating this
+  // every time we calculate / recalculate the target young length.
+  update_survivors_policy();
 
   assert(_g1->used() == _g1->recalculate_used(),
          err_msg("sanity, used: "SIZE_FORMAT" recalculate_used: "SIZE_FORMAT,
@@ -969,6 +959,9 @@ void G1CollectorPolicy::record_collection_pause_start(double start_time_sec,
   // This is initialized to zero here and is set during
   // the evacuation pause if marking is in progress.
   _cur_satb_drain_time_ms = 0.0;
+  // This is initialized to zero here and is set during the evacuation
+  // pause if we actually waited for the root region scanning to finish.
+  _root_region_scan_wait_time_ms = 0.0;
 
   _last_gc_was_young = false;
 
@@ -1140,6 +1133,50 @@ double G1CollectorPolicy::max_sum(double* data1, double* data2) {
   return ret;
 }
 
+bool G1CollectorPolicy::need_to_start_conc_mark(const char* source, size_t alloc_word_size) {
+  if (_g1->concurrent_mark()->cmThread()->during_cycle()) {
+    return false;
+  }
+
+  size_t marking_initiating_used_threshold =
+    (_g1->capacity() / 100) * InitiatingHeapOccupancyPercent;
+  size_t cur_used_bytes = _g1->non_young_capacity_bytes();
+  size_t alloc_byte_size = alloc_word_size * HeapWordSize;
+
+  if ((cur_used_bytes + alloc_byte_size) > marking_initiating_used_threshold) {
+    if (gcs_are_young()) {
+      ergo_verbose5(ErgoConcCycles,
+        "request concurrent cycle initiation",
+        ergo_format_reason("occupancy higher than threshold")
+        ergo_format_byte("occupancy")
+        ergo_format_byte("allocation request")
+        ergo_format_byte_perc("threshold")
+        ergo_format_str("source"),
+        cur_used_bytes,
+        alloc_byte_size,
+        marking_initiating_used_threshold,
+        (double) InitiatingHeapOccupancyPercent,
+        source);
+      return true;
+    } else {
+      ergo_verbose5(ErgoConcCycles,
+        "do not request concurrent cycle initiation",
+        ergo_format_reason("still doing mixed collections")
+        ergo_format_byte("occupancy")
+        ergo_format_byte("allocation request")
+        ergo_format_byte_perc("threshold")
+        ergo_format_str("source"),
+        cur_used_bytes,
+        alloc_byte_size,
+        marking_initiating_used_threshold,
+        (double) InitiatingHeapOccupancyPercent,
+        source);
+    }
+  }
+
+  return false;
+}
+
 // Anything below that is considered to be zero
 #define MIN_TIMER_GRANULARITY 0.0000001
 
@@ -1166,44 +1203,16 @@ void G1CollectorPolicy::record_collection_pause_end(int no_of_gc_threads) {
 #endif // PRODUCT
 
   last_pause_included_initial_mark = during_initial_mark_pause();
-  if (last_pause_included_initial_mark)
+  if (last_pause_included_initial_mark) {
     record_concurrent_mark_init_end(0.0);
-
-  size_t marking_initiating_used_threshold =
-    (_g1->capacity() / 100) * InitiatingHeapOccupancyPercent;
-
-  if (!_g1->mark_in_progress() && !_last_young_gc) {
-    assert(!last_pause_included_initial_mark, "invariant");
-    if (cur_used_bytes > marking_initiating_used_threshold) {
-      if (cur_used_bytes > _prev_collection_pause_used_at_end_bytes) {
-        assert(!during_initial_mark_pause(), "we should not see this here");
-
-        ergo_verbose3(ErgoConcCycles,
-                      "request concurrent cycle initiation",
-                      ergo_format_reason("occupancy higher than threshold")
-                      ergo_format_byte("occupancy")
-                      ergo_format_byte_perc("threshold"),
-                      cur_used_bytes,
-                      marking_initiating_used_threshold,
-                      (double) InitiatingHeapOccupancyPercent);
-
-        // Note: this might have already been set, if during the last
-        // pause we decided to start a cycle but at the beginning of
-        // this pause we decided to postpone it. That's OK.
-        set_initiate_conc_mark_if_possible();
-      } else {
-        ergo_verbose2(ErgoConcCycles,
-                  "do not request concurrent cycle initiation",
-                  ergo_format_reason("occupancy lower than previous occupancy")
-                  ergo_format_byte("occupancy")
-                  ergo_format_byte("previous occupancy"),
-                  cur_used_bytes,
-                  _prev_collection_pause_used_at_end_bytes);
-      }
-    }
   }
 
-  _prev_collection_pause_used_at_end_bytes = cur_used_bytes;
+  if (!_last_young_gc && need_to_start_conc_mark("end of GC")) {
+    // Note: this might have already been set, if during the last
+    // pause we decided to start a cycle but at the beginning of
+    // this pause we decided to postpone it. That's OK.
+    set_initiate_conc_mark_if_possible();
+  }
 
   _mmu_tracker->add_pause(end_time_sec - elapsed_ms/1000.0,
                           end_time_sec, false);
@@ -1257,6 +1266,10 @@ void G1CollectorPolicy::record_collection_pause_end(int no_of_gc_threads) {
   // is in progress.
   other_time_ms -= _cur_satb_drain_time_ms;
 
+  // Subtract the root region scanning wait time. It's initialized to
+  // zero at the start of the pause.
+  other_time_ms -= _root_region_scan_wait_time_ms;
+
   if (parallel) {
     other_time_ms -= _cur_collection_par_time_ms;
   } else {
@@ -1289,6 +1302,8 @@ void G1CollectorPolicy::record_collection_pause_end(int no_of_gc_threads) {
     // each other. Therefore we unconditionally record the SATB drain
     // time - even if it's zero.
     body_summary->record_satb_drain_time_ms(_cur_satb_drain_time_ms);
+    body_summary->record_root_region_scan_wait_time_ms(
+                                               _root_region_scan_wait_time_ms);
 
     body_summary->record_ext_root_scan_time_ms(ext_root_scan_time);
     body_summary->record_satb_filtering_time_ms(satb_filtering_time);
@@ -1385,6 +1400,9 @@ void G1CollectorPolicy::record_collection_pause_end(int no_of_gc_threads) {
                            (last_pause_included_initial_mark) ? " (initial-mark)" : "",
                            elapsed_ms / 1000.0);
 
+    if (_root_region_scan_wait_time_ms > 0.0) {
+      print_stats(1, "Root Region Scan Waiting", _root_region_scan_wait_time_ms);
+    }
     if (parallel) {
       print_stats(1, "Parallel Time", _cur_collection_par_time_ms);
       print_par_stats(2, "GC Worker Start", _par_last_gc_worker_start_times_ms);
@@ -1988,6 +2006,7 @@ void G1CollectorPolicy::print_summary(PauseSummary* summary) const {
   if (summary->get_total_seq()->num() > 0) {
     print_summary_sd(0, "Evacuation Pauses", summary->get_total_seq());
     if (body_summary != NULL) {
+      print_summary(1, "Root Region Scan Wait", body_summary->get_root_region_scan_wait_seq());
       if (parallel) {
         print_summary(1, "Parallel Time", body_summary->get_parallel_seq());
         print_summary(2, "Ext Root Scanning", body_summary->get_ext_root_scan_seq());
@@ -2029,15 +2048,17 @@ void G1CollectorPolicy::print_summary(PauseSummary* summary) const {
           // parallel
           NumberSeq* other_parts[] = {
             body_summary->get_satb_drain_seq(),
+            body_summary->get_root_region_scan_wait_seq(),
             body_summary->get_parallel_seq(),
             body_summary->get_clear_ct_seq()
           };
           calc_other_times_ms = NumberSeq(summary->get_total_seq(),
-                                                3, other_parts);
+                                          4, other_parts);
         } else {
           // serial
           NumberSeq* other_parts[] = {
             body_summary->get_satb_drain_seq(),
+            body_summary->get_root_region_scan_wait_seq(),
             body_summary->get_update_rs_seq(),
             body_summary->get_ext_root_scan_seq(),
             body_summary->get_satb_filtering_seq(),
@@ -2045,7 +2066,7 @@ void G1CollectorPolicy::print_summary(PauseSummary* summary) const {
             body_summary->get_obj_copy_seq()
           };
           calc_other_times_ms = NumberSeq(summary->get_total_seq(),
-                                                6, other_parts);
+                                          7, other_parts);
         }
         check_other_times(1,  summary->get_other_seq(), &calc_other_times_ms);
       }

hotspot/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp

@@ -65,6 +65,7 @@ public:
 
 class MainBodySummary: public CHeapObj {
   define_num_seq(satb_drain) // optional
+  define_num_seq(root_region_scan_wait)
   define_num_seq(parallel) // parallel only
     define_num_seq(ext_root_scan)
     define_num_seq(satb_filtering)
@@ -177,7 +178,6 @@ private:
   double _cur_collection_start_sec;
   size_t _cur_collection_pause_used_at_start_bytes;
   size_t _cur_collection_pause_used_regions_at_start;
-  size_t _prev_collection_pause_used_at_end_bytes;
   double _cur_collection_par_time_ms;
   double _cur_satb_drain_time_ms;
   double _cur_clear_ct_time_ms;
@@ -716,6 +716,7 @@ private:
   double _mark_remark_start_sec;
   double _mark_cleanup_start_sec;
   double _mark_closure_time_ms;
+  double _root_region_scan_wait_time_ms;
 
   // Update the young list target length either by setting it to the
   // desired fixed value or by calculating it using G1's pause
@@ -800,6 +801,8 @@ public:
 
   GenRemSet::Name  rem_set_name()     { return GenRemSet::CardTable; }
 
+  bool need_to_start_conc_mark(const char* source, size_t alloc_word_size = 0);
+
   // Update the heuristic info to record a collection pause of the given
   // start time, where the given number of bytes were used at the start.
   // This may involve changing the desired size of a collection set.
@@ -816,6 +819,10 @@ public:
     _mark_closure_time_ms = mark_closure_time_ms;
   }
 
+  void record_root_region_scan_wait_time(double time_ms) {
+    _root_region_scan_wait_time_ms = time_ms;
+  }
+
   void record_concurrent_mark_remark_start();
   void record_concurrent_mark_remark_end();
 
@@ -1146,11 +1153,6 @@ public:
     _survivor_surv_rate_group->stop_adding_regions();
   }
 
-  void tenure_all_objects() {
-    _max_survivor_regions = 0;
-    _tenuring_threshold = 0;
-  }
-
   void record_survivor_regions(size_t      regions,
                                HeapRegion* head,
                                HeapRegion* tail) {

hotspot/src/share/vm/gc_implementation/g1/g1EvacFailure.hpp

@@ -70,16 +70,20 @@ private:
   OopsInHeapRegionClosure *_update_rset_cl;
   bool _during_initial_mark;
   bool _during_conc_mark;
+  uint _worker_id;
+
 public:
   RemoveSelfForwardPtrObjClosure(G1CollectedHeap* g1, ConcurrentMark* cm,
                                  HeapRegion* hr,
                                  OopsInHeapRegionClosure* update_rset_cl,
                                  bool during_initial_mark,
-                                 bool during_conc_mark) :
+                                 bool during_conc_mark,
+                                 uint worker_id) :
     _g1(g1), _cm(cm), _hr(hr), _marked_bytes(0),
     _update_rset_cl(update_rset_cl),
     _during_initial_mark(during_initial_mark),
-    _during_conc_mark(during_conc_mark) { }
+    _during_conc_mark(during_conc_mark),
+    _worker_id(worker_id) { }
 
   size_t marked_bytes() { return _marked_bytes; }
 
@@ -123,7 +127,7 @@ public:
         // explicitly and all objects in the CSet are considered
         // (implicitly) live. So, we won't mark them explicitly and
         // we'll leave them over NTAMS.
-        _cm->markNext(obj);
+        _cm->grayRoot(obj, obj_size, _worker_id, _hr);
       }
       _marked_bytes += (obj_size * HeapWordSize);
       obj->set_mark(markOopDesc::prototype());
@@ -155,12 +159,14 @@ class RemoveSelfForwardPtrHRClosure: public HeapRegionClosure {
   G1CollectedHeap* _g1h;
   ConcurrentMark* _cm;
   OopsInHeapRegionClosure *_update_rset_cl;
+  uint _worker_id;
 
 public:
   RemoveSelfForwardPtrHRClosure(G1CollectedHeap* g1h,
-                                OopsInHeapRegionClosure* update_rset_cl) :
+                                OopsInHeapRegionClosure* update_rset_cl,
+                                uint worker_id) :
     _g1h(g1h), _update_rset_cl(update_rset_cl),
-    _cm(_g1h->concurrent_mark()) { }
+    _worker_id(worker_id), _cm(_g1h->concurrent_mark()) { }
 
   bool doHeapRegion(HeapRegion *hr) {
     bool during_initial_mark = _g1h->g1_policy()->during_initial_mark_pause();
@@ -173,7 +179,8 @@ public:
       if (hr->evacuation_failed()) {
         RemoveSelfForwardPtrObjClosure rspc(_g1h, _cm, hr, _update_rset_cl,
                                             during_initial_mark,
-                                            during_conc_mark);
+                                            during_conc_mark,
+                                            _worker_id);
 
         MemRegion mr(hr->bottom(), hr->end());
         // We'll recreate the prev marking info so we'll first clear
@@ -226,7 +233,7 @@ public:
       update_rset_cl = &immediate_update;
     }
 
-    RemoveSelfForwardPtrHRClosure rsfp_cl(_g1h, update_rset_cl);
+    RemoveSelfForwardPtrHRClosure rsfp_cl(_g1h, update_rset_cl, worker_id);
 
     HeapRegion* hr = _g1h->start_cset_region_for_worker(worker_id);
     _g1h->collection_set_iterate_from(hr, &rsfp_cl);

hotspot/src/share/vm/gc_implementation/g1/g1MonitoringSupport.hpp

@@ -89,16 +89,15 @@ class G1CollectedHeap;
 //
 // * Min Capacity
 //
-//    We set this to 0 for all spaces. We could consider setting the old
-//    min capacity to the min capacity of the heap (see 7078465).
+//    We set this to 0 for all spaces.
 //
 // * Max Capacity
 //
 //    For jstat, we set the max capacity of all spaces to heap_capacity,
-//    given that we don't always have a reasonably upper bound on how big
-//    each space can grow. For the memory pools, we actually make the max
-//    capacity undefined. We could consider setting the old max capacity
-//    to the max capacity of the heap (see 7078465).
+//    given that we don't always have a reasonable upper bound on how big
+//    each space can grow. For the memory pools, we make the max
+//    capacity undefined with the exception of the old memory pool for
+//    which we make the max capacity same as the max heap capacity.
 //
 // If we had more accurate occupancy / capacity information per
 // region set the above calculations would be greatly simplified and

hotspot/src/share/vm/gc_implementation/g1/g1RemSet.hpp

 /*
- * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 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
@@ -191,44 +191,5 @@ public:
   virtual void do_oop(      oop* p) { do_oop_work(p); }
 };
 
-class UpdateRSOrPushRefOopClosure: public OopClosure {
-  G1CollectedHeap* _g1;
-  G1RemSet* _g1_rem_set;
-  HeapRegion* _from;
-  OopsInHeapRegionClosure* _push_ref_cl;
-  bool _record_refs_into_cset;
-  int _worker_i;
-
-  template <class T> void do_oop_work(T* p);
-
-public:
-  UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h,
-                              G1RemSet* rs,
-                              OopsInHeapRegionClosure* push_ref_cl,
-                              bool record_refs_into_cset,
-                              int worker_i = 0) :
-    _g1(g1h),
-    _g1_rem_set(rs),
-    _from(NULL),
-    _record_refs_into_cset(record_refs_into_cset),
-    _push_ref_cl(push_ref_cl),
-    _worker_i(worker_i) { }
-
-  void set_from(HeapRegion* from) {
-    assert(from != NULL, "from region must be non-NULL");
-    _from = from;
-  }
-
-  bool self_forwarded(oop obj) {
-    bool result = (obj->is_forwarded() && (obj->forwardee()== obj));
-    return result;
-  }
-
-  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
-  virtual void do_oop(oop* p)       { do_oop_work(p); }
-
-  bool apply_to_weak_ref_discovered_field() { return true; }
-};
-
 
 #endif // SHARE_VM_GC_IMPLEMENTATION_G1_G1REMSET_HPP

hotspot/src/share/vm/gc_implementation/g1/heapRegion.hpp

@@ -374,7 +374,9 @@ class HeapRegion: public G1OffsetTableContigSpace {
     ParVerifyClaimValue        = 4,
     RebuildRSClaimValue        = 5,
     CompleteMarkCSetClaimValue = 6,
-    ParEvacFailureClaimValue   = 7
+    ParEvacFailureClaimValue   = 7,
+    AggregateCountClaimValue   = 8,
+    VerifyCountClaimValue      = 9
   };
 
   inline HeapWord* par_allocate_no_bot_updates(size_t word_size) {

hotspot/src/share/vm/gc_implementation/g1/heapRegionSet.hpp

 /*
- * Copyright (c) 2011, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2011, 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
@@ -59,6 +59,7 @@ class HRSPhaseSetter;
 class HeapRegionSetBase VALUE_OBJ_CLASS_SPEC {
   friend class hrs_ext_msg;
   friend class HRSPhaseSetter;
+  friend class VMStructs;
 
 protected:
   static size_t calculate_region_num(HeapRegion* hr);