Merge branch 'maint'

Conflicts:
	erts/emulator/beam/beam_emu.c
	erts/emulator/beam/bif.tab
	erts/preloaded/ebin/prim_file.beam
	lib/hipe/cerl/erl_bif_types.erl

43 files changed, 3528 insertions, 199 deletions

diff --git a/erts/configure.in b/erts/configure.in
index b801994e14..cb1b00b8b1 100644
--- a/erts/configure.in
+++ b/erts/configure.in
@@ -276,6 +276,61 @@ else
               [Define to enable hrvtime() on Linux systems with perfctr extension])
 fi
 
+
+AC_ARG_WITH(dynamic-trace,
+AS_HELP_STRING([--with-dynamic-trace={dtrace|systemtap}],
+	       [specify use of dynamic trace framework, dtrace or systemtap])
+AS_HELP_STRING([--without-dynamic-trace], 
+               [don't enable any dynamic tracing (default)]))
+
+if test X"$with_dynamic_trace" = X""; then
+   with_dynamic_trace=no
+fi
+
+case "$with_dynamic_trace" in
+     no) DYNAMIC_TRACE_FRAMEWORK=;;
+     dtrace) 
+	    AC_DEFINE(USE_DTRACE,[1],
+		[Define if you want to use dtrace for dynamic tracing])
+     	     DYNAMIC_TRACE_FRAMEWORK=dtrace;;
+     systemtap) 
+	    AC_DEFINE(USE_SYSTEMTAP,[1],
+		[Define if you want to use systemtap for dynamic tracing])
+	    DYNAMIC_TRACE_FRAMEWORK=systemtap;;
+     *)	
+     	    AC_MSG_ERROR(Unknown dynamic tracing framework specified with --with-dynamic-trace!);;
+esac
+
+if test X"$DYNAMIC_TRACE_FRAMEWORK" != X""; then
+   AC_DEFINE(USE_DYNAMIC_TRACE,[1],
+		[Define if you want to use dynamic tracing])
+fi
+
+AC_ARG_ENABLE(vm-probes,
+AS_HELP_STRING([--enable-vm-probes],
+               [add dynamic trace probes to the Beam VM (only possible if --with-dynamic-trace is enabled, and then default)]),
+	       [ case "$enableval" in
+    	       	 no) use_vm_probes=no ;;
+    		 *)
+			if  test X"$DYNAMIC_TRACE_FRAMEWORK" != X""; then
+		 	    use_vm_probes=yes ;
+			else
+			    AC_MSG_ERROR(Can not enable VM probes without any dynamic tracing framework!);
+			fi;;  
+  		 esac ], if  test X"$DYNAMIC_TRACE_FRAMEWORK" != X""; then
+		 	    use_vm_probes=yes ;
+			else
+			    use_vm_probes=no
+			fi)
+
+AC_SUBST(USE_VM_PROBES)
+if test X"$use_vm_probes" = X"yes"; then
+   USE_VM_PROBES=yes
+   AC_DEFINE(USE_VM_PROBES,[1],
+	     [Define to enable VM dynamic trace probes])
+fi
+
+
 AC_ARG_ENABLE(clock-gettime,
 AS_HELP_STRING([--enable-clock-gettime],
                [use clock-gettime for time correction]),
@@ -3546,6 +3601,74 @@ dnl
 LM_FIND_EMU_CC
 
 dnl
+dnl DTrace
+dnl
+case $DYNAMIC_TRACE_FRAMEWORK in
+     dtrace|systemtap)
+        AC_CHECK_TOOL(DTRACE, dtrace, none)
+        test "$DTRACE" = "none" && AC_MSG_ERROR([No dtrace utility found.]);
+	enable_dtrace_test=yes;;
+     *) enable_dtrace_test=no;;
+esac        
+		     
+AC_SUBST(DTRACE)
+
+AC_SUBST(DTRACE_CPP)
+AC_SUBST(DTRACE_ENABLED)
+AC_SUBST(DTRACE_ENABLED_2STEP)
+DTRACE_CPP=-C
+DTRACE_ENABLED=
+DTRACE_ENABLED_2STEP=
+DTRACE_2STEP_TEST=./dtrace-test.o
+DTRACE_BITS_FLAG=
+case $OPSYS in
+    freebsd)
+	if test "$BITS64" = "yes" ; then
+		DTRACE_BITS_FLAG=-64
+	else
+		DTRACE_BITS_FLAG=-32
+	fi
+    ;;
+    *)
+	: # Nothing to do
+    ;;
+esac
+if test "$enable_dtrace_test" = "yes" ; then
+        if test "$DTRACE" = "dtrace" ; then
+                AC_CHECK_HEADERS(sys/sdt.h)
+                # The OS X version of dtrace prints a spurious line here.
+                if ! dtrace -h $DTRACE_CPP -Iemulator/beam -o ./foo-dtrace.h -s emulator/beam/erlang_dtrace.d; then
+                        AC_MSG_ERROR([Could not precompile erlang_dtrace.d: dtrace -h failed])
+                fi
+                rm -f foo-dtrace.h
+
+                $RM -f $DTRACE_2STEP_TEST
+                if dtrace -G $DTRACE_CPP $DTRACE_BITS_FLAG -Iemulator/beam -o $DTRACE_2STEP_TEST -s emulator/beam/erlang_dtrace.d 2> /dev/null && \
+                   test -f $DTRACE_2STEP_TEST ; then
+                        rm $DTRACE_2STEP_TEST
+                        DTRACE_ENABLED_2STEP=yes
+                        AC_MSG_NOTICE([dtrace precompilation for 2-stage DTrace successful])
+                else
+                        AC_MSG_NOTICE([dtrace precompilation for 1-stage DTrace successful])
+                fi
+                DTRACE_ENABLED=yes
+                case $OPSYS in
+                    linux)
+                        : # No extra libs to add to LIBS
+                    ;;
+                    freebsd)
+                        LIBS="$LIBS -lelf"
+                    ;;
+                    *)
+                        LIBS="$LIBS -ldtrace"
+                    ;;
+                esac
+        else
+                AC_MSG_ERROR([Dtrace preprocessing test failed.])
+        fi
+fi
+
+dnl
 dnl SSL, SSH and CRYPTO need the OpenSSL libraries
 dnl
 dnl Check flags --with-ssl, --without-ssl --with-ssl=PATH.
diff --git a/erts/doc/src/erlang.xml b/erts/doc/src/erlang.xml
index da0ef8fbe2..9c8c033e29 100644
--- a/erts/doc/src/erlang.xml
+++ b/erts/doc/src/erlang.xml
@@ -5213,6 +5213,8 @@ ok
       <name name="system_info" arity="1" clause_i="48"/>
       <name name="system_info" arity="1" clause_i="49"/>
       <name name="system_info" arity="1" clause_i="50"/>
+      <name name="system_info" arity="1" clause_i="51"/>
+      <name name="system_info" arity="1" clause_i="52"/>
       <fsummary>Information about the system</fsummary>
       <desc>
         <p>Returns various information about the current system
@@ -5306,6 +5308,29 @@ ok
               used by the runtime system. It will be on the form
               <seealso marker="erts:erl_driver#version_management">"&lt;major ver&gt;.&lt;minor ver&gt;"</seealso>.</p>
           </item>
+          <tag><c>dynamic_trace</c></tag>
+          <item>
+            <p>Returns an atom describing the dynamic trace framework
+            compiled into the virtual machine. It can currently be either
+            <c>dtrace</c>, <c>systemtap</c> or <c>none</c>. For a
+            commercial or standard build, this is always <c>none</c>,
+            the other return values indicate a custom configuration
+            (e.g. <c>./configure --with-dynamic-trace=dtrace</c>). See
+            the <seealso marker="runtime_tools:dyntrace">dyntrace
+            </seealso> manual page and the
+            <c>README.dtrace</c>/<c>README.systemtap</c> files in the
+            Erlang source code top directory for more information
+            about dynamic tracing.</p>
+          </item>
+          <tag><c>dynamic_trace_probes</c></tag>
+          <item>
+            <p>Returns a <c>boolean()</c> indicating if dynamic trace probes
+            (either dtrace or systemtap) are built into the
+            emulator. This can only be <c>true</c> if the virtual
+            machine was built for dynamic tracing
+            (i.e. <c>system_info(dynamic_trace)</c> returns
+            <c>dtrace</c> or <c>systemtap</c>).</p>
+          </item>
           <tag><c>elib_malloc</c></tag>
           <item>
             <p>This option will be removed in a future release.
@@ -5588,7 +5613,8 @@ ok
 	    <seealso marker="#system_info_schedulers">erlang:system_info(schedulers)</seealso>,
 	    and
 	    <seealso marker="#system_flag_schedulers_online">erlang:system_flag(schedulers_online, SchedulersOnline)</seealso>.
-	    </p>
+	    </p>      <name name="system_info" arity="1" clause_i="50"/>
+
           </item>
           <tag><c>smp_support</c></tag>
           <item>
diff --git a/erts/emulator/Makefile.in b/erts/emulator/Makefile.in
index 379943c7a6..4e66181a3f 100644
--- a/erts/emulator/Makefile.in
+++ b/erts/emulator/Makefile.in
@@ -23,6 +23,9 @@ include $(ERL_TOP)/make/$(TARGET)/otp.mk
 
 ENABLE_ALLOC_TYPE_VARS = @ENABLE_ALLOC_TYPE_VARS@
 HIPE_ENABLED=@HIPE_ENABLED@
+DTRACE_ENABLED=@DTRACE_ENABLED@
+DTRACE_ENABLED_2STEP=@DTRACE_ENABLED_2STEP@
+USE_VM_PROBES=@USE_VM_PROBES@
 LIBS = @LIBS@
 Z_LIB=@Z_LIB@
 NO_INLINE_FUNCTIONS=false
@@ -483,6 +486,14 @@ GENERATE +=	$(HIPE_ASM) \
 endif
 endif
 
+ifdef DTRACE_ENABLED
+# global.h causes problems by including dtrace-wrapper.h which includes
+# the autogenerated erlang_dtrace.h ... so make erlang_dtrace.h very early.
+generate: $(TARGET)/erlang_dtrace.h $(GENERATE)
+else
+generate: $(GENERATE)
+endif
+
 ifdef HIPE_ENABLED
 OPCODE_TABLES += hipe/hipe_ops.tab
 endif
@@ -498,6 +509,7 @@ $(TTF_DIR)/OPCODES-GENERATED: $(OPCODE_TABLES) utils/beam_makeops
 	LANG=C $(PERL) utils/beam_makeops \
 		-wordsize @EXTERNAL_WORD_SIZE@ \
 		-outdir $(TTF_DIR) \
+		-DUSE_VM_PROBES=$(if $(USE_VM_PROBES),1,0) \
 		-emulator $(OPCODE_TABLES) && echo $? >$(TTF_DIR)/OPCODES-GENERATED
 GENERATE += $(TTF_DIR)/OPCODES-GENERATED
 
@@ -590,6 +602,11 @@ $(TTF_DIR)/GENERATED: $(GENERATE)
 	echo $? >$(TTF_DIR)/GENERATED
 endif
 
+$(TARGET)/erlang_dtrace.h: beam/erlang_dtrace.d
+	dtrace -h -C -Ibeam -s $< -o ./erlang_dtrace.tmp
+	sed -e '/^#define[ 	]*ERLANG_[A-Z0-9_]*(.*)/y/ABCDEFGHIJKLMNOPQRSTUVWXYZ/abcdefghijklmnopqrstuvwxyz/' ./erlang_dtrace.tmp > $@
+	rm ./erlang_dtrace.tmp
+
 # ----------------------------------------------------------------------
 # Pattern rules
 #
@@ -633,7 +650,6 @@ $(OBJDIR)/beam_emu.o: beam/beam_emu.c
 	$(EMU_CC) $(subst -O2, $(GEN_OPT_FLGS), $(CFLAGS)) $(INCLUDES) -c $< -o $@
 endif
 
-
 $(OBJDIR)/%.o: beam/%.c
 	$(CC) $(subst -O2, $(GEN_OPT_FLGS), $(CFLAGS)) $(INCLUDES) -c $< -o $@
 
@@ -835,7 +851,18 @@ endif
 
 BASE_OBJS = $(RUN_OBJS) $(EMU_OBJS) $(OS_OBJS) $(EXTRA_BASE_OBJS)
 
-OBJS =	$(BASE_OBJS) $(DRV_OBJS)
+before_DTrace_OBJS =	$(BASE_OBJS) $(DRV_OBJS)
+
+DTRACE_OBJS =
+ifdef DTRACE_ENABLED_2STEP
+DTRACE_OBJS = $(OBJDIR)/erlang_dtrace.o
+$(OBJDIR)/erlang_dtrace.o: $(before_DTrace_OBJS) $(TARGET)/erlang_dtrace.h
+	dtrace -G -C -Ibeam \
+	  -s beam/erlang_dtrace.d \
+	  -o $@ $(before_DTrace_OBJS)
+endif
+
+OBJS = $(before_DTrace_OBJS) $(DTRACE_OBJS)
 
 $(INIT_OBJS): $(TTF_DIR)/GENERATED
 $(OBJS): $(TTF_DIR)/GENERATED
diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names
index 2c6e80a3c3..32b91ef02a 100644
--- a/erts/emulator/beam/atom.names
+++ b/erts/emulator/beam/atom.names
@@ -248,6 +248,7 @@ atom global_heaps_size
 atom Gt='>'
 atom grun
 atom group_leader
+atom have_dt_utag
 atom heap_block_size
 atom heap_size
 atom heap_sizes
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index d6189c6c65..8f8c469d9a 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -41,6 +41,7 @@
 #include "hipe_mode_switch.h"
 #include "hipe_bif1.h"
 #endif
+#include "dtrace-wrapper.h"
 
 /* #define HARDDEBUG 1 */
 
@@ -1050,6 +1051,101 @@ init_emulator(void)
 #  define REG_tmp_arg2
 #endif
 
+#ifdef USE_VM_PROBES
+#  define USE_VM_CALL_PROBES
+#endif
+
+#ifdef USE_VM_CALL_PROBES
+
+#define DTRACE_LOCAL_CALL(p, m, f, a)					\
+    if (DTRACE_ENABLED(local_function_entry)) {				\
+        DTRACE_CHARBUF(process_name, DTRACE_TERM_BUF_SIZE);		\
+        DTRACE_CHARBUF(mfa, DTRACE_TERM_BUF_SIZE);			\
+        int depth = STACK_START(p) - STACK_TOP(p);			\
+        dtrace_fun_decode(p, m, f, a,					\
+                          process_name, mfa);				\
+        DTRACE3(local_function_entry, process_name, mfa, depth);	\
+    }
+
+#define DTRACE_GLOBAL_CALL(p, m, f, a)					\
+    if (DTRACE_ENABLED(global_function_entry)) {			\
+        DTRACE_CHARBUF(process_name, DTRACE_TERM_BUF_SIZE);		\
+        DTRACE_CHARBUF(mfa, DTRACE_TERM_BUF_SIZE);			\
+        int depth = STACK_START(p) - STACK_TOP(p);			\
+        dtrace_fun_decode(p, m, f, a,					\
+                          process_name, mfa);				\
+        DTRACE3(global_function_entry, process_name, mfa, depth);	\
+    }
+
+#define DTRACE_RETURN(p, m, f, a)                               \
+    if (DTRACE_ENABLED(function_return)) {                      \
+        DTRACE_CHARBUF(process_name, DTRACE_TERM_BUF_SIZE);     \
+        DTRACE_CHARBUF(mfa, DTRACE_TERM_BUF_SIZE);              \
+        int depth = STACK_START(p) - STACK_TOP(p);              \
+        dtrace_fun_decode(p, m, f, a,                           \
+                          process_name, mfa);                   \
+        DTRACE3(function_return, process_name, mfa, depth);     \
+    }
+
+#define DTRACE_BIF_ENTRY(p, m, f, a)                            \
+    if (DTRACE_ENABLED(bif_entry)) {                            \
+        DTRACE_CHARBUF(process_name, DTRACE_TERM_BUF_SIZE);     \
+        DTRACE_CHARBUF(mfa, DTRACE_TERM_BUF_SIZE);              \
+        dtrace_fun_decode(p, m, f, a,                           \
+                          process_name, mfa);                   \
+        DTRACE2(bif_entry, process_name, mfa);                  \
+    }
+
+#define DTRACE_BIF_RETURN(p, m, f, a)                           \
+    if (DTRACE_ENABLED(bif_return)) {                           \
+        DTRACE_CHARBUF(process_name, DTRACE_TERM_BUF_SIZE);     \
+        DTRACE_CHARBUF(mfa, DTRACE_TERM_BUF_SIZE);              \
+        dtrace_fun_decode(p, m, f, a,                           \
+                          process_name, mfa);                   \
+        DTRACE2(bif_return, process_name, mfa);                 \
+    }
+
+#define DTRACE_NIF_ENTRY(p, m, f, a)                            \
+    if (DTRACE_ENABLED(nif_entry)) {                            \
+        DTRACE_CHARBUF(process_name, DTRACE_TERM_BUF_SIZE);     \
+        DTRACE_CHARBUF(mfa, DTRACE_TERM_BUF_SIZE);              \
+        dtrace_fun_decode(p, m, f, a,                           \
+                          process_name, mfa);                   \
+        DTRACE2(nif_entry, process_name, mfa);                  \
+    }
+
+#define DTRACE_NIF_RETURN(p, m, f, a)                           \
+    if (DTRACE_ENABLED(nif_return)) {                           \
+        DTRACE_CHARBUF(process_name, DTRACE_TERM_BUF_SIZE);     \
+        DTRACE_CHARBUF(mfa, DTRACE_TERM_BUF_SIZE);              \
+        dtrace_fun_decode(p, m, f, a,                           \
+                          process_name, mfa);                   \
+        DTRACE2(nif_return, process_name, mfa);                 \
+    }
+
+#else /* USE_VM_PROBES */
+
+#define DTRACE_LOCAL_CALL(p, m, f, a)  do {} while (0)
+#define DTRACE_GLOBAL_CALL(p, m, f, a) do {} while (0)
+#define DTRACE_RETURN(p, m, f, a)      do {} while (0)
+#define DTRACE_BIF_ENTRY(p, m, f, a)   do {} while (0)
+#define DTRACE_BIF_RETURN(p, m, f, a)  do {} while (0)
+#define DTRACE_NIF_ENTRY(p, m, f, a)   do {} while (0)
+#define DTRACE_NIF_RETURN(p, m, f, a)  do {} while (0)
+
+#endif /* USE_VM_PROBES */
+
+#ifdef USE_VM_PROBES
+void
+dtrace_drvport_str(ErlDrvPort drvport, char *port_buf)
+{
+    Port *port = erts_drvport2port(drvport);
+
+    erts_snprintf(port_buf, DTRACE_TERM_BUF_SIZE, "#Port<%lu.%lu>",
+                  port_channel_no(port->id),
+                  port_number(port->id));
+}
+#endif
 /*
  * process_main() is called twice:
  * The first call performs some initialisation, including exporting
@@ -1221,6 +1317,30 @@ void process_main(void)
 #endif
 	SWAPIN;
 	ASSERT(VALID_INSTR(next));
+
+#ifdef USE_VM_PROBES
+        if (DTRACE_ENABLED(process_scheduled)) {
+            DTRACE_CHARBUF(process_buf, DTRACE_TERM_BUF_SIZE);
+            DTRACE_CHARBUF(fun_buf, DTRACE_TERM_BUF_SIZE);
+            dtrace_proc_str(c_p, process_buf);
+
+            if (ERTS_PROC_IS_EXITING(c_p)) {
+                strcpy(fun_buf, "<exiting>");
+            } else {
+                BeamInstr *fptr = find_function_from_pc(c_p->i);
+                if (fptr) {
+                    dtrace_fun_decode(c_p, (Eterm)fptr[0],
+                                      (Eterm)fptr[1], (Uint)fptr[2],
+                                      NULL, fun_buf);
+                } else {
+                    erts_snprintf(fun_buf, sizeof(fun_buf),
+                                  "<unknown/%p>", next);
+                }
+            }
+
+            DTRACE2(process_scheduled, process_buf, fun_buf);
+        }
+#endif
 	Goto(next);
     }
 
@@ -1397,6 +1517,7 @@ void process_main(void)
  /* FALL THROUGH */
  OpCase(i_call_only_f): {
      SET_I((BeamInstr *) Arg(0));
+     DTRACE_LOCAL_CALL(c_p, (Eterm)I[-3], (Eterm)I[-2], I[-1]);
      Dispatch();
  }
 
@@ -1408,6 +1529,7 @@ void process_main(void)
      RESTORE_CP(E);
      E = ADD_BYTE_OFFSET(E, Arg(1));
      SET_I((BeamInstr *) Arg(0));
+     DTRACE_LOCAL_CALL(c_p, (Eterm)I[-3], (Eterm)I[-2], I[-1]);
      Dispatch();
  }
 
@@ -1419,6 +1541,7 @@ void process_main(void)
  OpCase(i_call_f): {
      SET_CP(c_p, I+2);
      SET_I((BeamInstr *) Arg(0));
+     DTRACE_LOCAL_CALL(c_p, (Eterm)I[-3], (Eterm)I[-2], I[-1]);
      Dispatch();
  }
 
@@ -1435,6 +1558,12 @@ void process_main(void)
      * is not loaded, it points to code which will invoke the error handler
      * (see lb_call_error_handler below).
      */
+#ifdef USE_VM_CALL_PROBES
+    if (DTRACE_ENABLED(global_function_entry)) {
+	BeamInstr* fp = (BeamInstr *) (((Export *) Arg(0))->address);
+	DTRACE_GLOBAL_CALL(c_p, (Eterm)fp[-3], (Eterm)fp[-2], fp[-1]);
+    }
+#endif
     Dispatchx();
 
  OpCase(i_move_call_ext_cre): {
@@ -1444,6 +1573,12 @@ void process_main(void)
  /* FALL THROUGH */
  OpCase(i_call_ext_e):
     SET_CP(c_p, I+2);
+#ifdef USE_VM_CALL_PROBES
+    if (DTRACE_ENABLED(global_function_entry)) {
+	BeamInstr* fp = (BeamInstr *) (((Export *) Arg(0))->address);
+	DTRACE_GLOBAL_CALL(c_p, (Eterm)fp[-3], (Eterm)fp[-2], fp[-1]);
+    }
+#endif
     Dispatchx();
 
  OpCase(i_move_call_ext_only_ecr): {
@@ -1451,6 +1586,12 @@ void process_main(void)
  }
  /* FALL THROUGH */
  OpCase(i_call_ext_only_e):
+#ifdef USE_VM_CALL_PROBES
+    if (DTRACE_ENABLED(global_function_entry)) {
+	BeamInstr* fp = (BeamInstr *) (((Export *) Arg(0))->address);
+	DTRACE_GLOBAL_CALL(c_p, (Eterm)fp[-3], (Eterm)fp[-2], fp[-1]);
+    }
+#endif
     Dispatchx();
 
  OpCase(init_y): {
@@ -1486,7 +1627,16 @@ void process_main(void)
 
 
  OpCase(return): {
+#ifdef USE_VM_CALL_PROBES
+    BeamInstr* fptr;
+#endif
     SET_I(c_p->cp);
+
+#ifdef USE_VM_CALL_PROBES
+    if (DTRACE_ENABLED(function_return) && (fptr = find_function_from_pc(c_p->cp))) {
+        DTRACE_RETURN(c_p, (Eterm)fptr[0], (Eterm)fptr[1], (Uint)fptr[2]);
+    }
+#endif
     /*
      * We must clear the CP to make sure that a stale value do not
      * create a false module dependcy preventing code upgrading.
@@ -1755,6 +1905,7 @@ void process_main(void)
 	  * remove it...
 	  */
 	 ASSERT(!msgp->data.attached);
+         /* TODO: Add DTrace probe for this bad message situation? */
 	 UNLINK_MESSAGE(c_p, msgp);
 	 free_message(msgp);
 	 goto loop_rec__;
@@ -1780,24 +1931,88 @@ void process_main(void)
 	 save_calls(c_p, &exp_receive);
      }
      if (ERL_MESSAGE_TOKEN(msgp) == NIL) {
-	 SEQ_TRACE_TOKEN(c_p) = NIL;
+#ifdef USE_VM_PROBES
+	 if (DT_UTAG(c_p) != NIL) {
+	     if (DT_UTAG_FLAGS(c_p) & DT_UTAG_PERMANENT) {
+		 SEQ_TRACE_TOKEN(c_p) = am_have_dt_utag;
+#ifdef DTRACE_TAG_HARDDEBUG
+		 if (DT_UTAG_FLAGS(c_p) & DT_UTAG_SPREADING) 
+		     erts_fprintf(stderr,
+				  "Dtrace -> (%T) stop spreading "
+				  "tag %T with message %T\r\n",
+				  c_p->id,DT_UTAG(c_p),ERL_MESSAGE_TERM(msgp));
+#endif
+	     } else {
+#ifdef DTRACE_TAG_HARDDEBUG
+		 erts_fprintf(stderr,
+			      "Dtrace -> (%T) kill tag %T with "
+			      "message %T\r\n",
+			      c_p->id,DT_UTAG(c_p),ERL_MESSAGE_TERM(msgp));
+#endif
+		 DT_UTAG(c_p) = NIL;
+		 SEQ_TRACE_TOKEN(c_p) = NIL;
+	     }
+	 } else {
+#endif
+	     SEQ_TRACE_TOKEN(c_p) = NIL;
+#ifdef USE_VM_PROBES
+	 }
+	 DT_UTAG_FLAGS(c_p) &= ~DT_UTAG_SPREADING;
+#endif
      } else if (ERL_MESSAGE_TOKEN(msgp) != am_undefined) {
 	 Eterm msg;
 	 SEQ_TRACE_TOKEN(c_p) = ERL_MESSAGE_TOKEN(msgp);
-	 ASSERT(is_tuple(SEQ_TRACE_TOKEN(c_p)));
-	 ASSERT(SEQ_TRACE_TOKEN_ARITY(c_p) == 5);
-	 ASSERT(is_small(SEQ_TRACE_TOKEN_SERIAL(c_p)));
-	 ASSERT(is_small(SEQ_TRACE_TOKEN_LASTCNT(c_p)));
-	 ASSERT(is_small(SEQ_TRACE_TOKEN_FLAGS(c_p)));
-	 ASSERT(is_pid(SEQ_TRACE_TOKEN_SENDER(c_p)));
-	 c_p->seq_trace_lastcnt = unsigned_val(SEQ_TRACE_TOKEN_SERIAL(c_p));
-	 if (c_p->seq_trace_clock < unsigned_val(SEQ_TRACE_TOKEN_SERIAL(c_p))) {
-	     c_p->seq_trace_clock = unsigned_val(SEQ_TRACE_TOKEN_SERIAL(c_p));
+#ifdef USE_VM_PROBES
+	 if (ERL_MESSAGE_TOKEN(msgp) == am_have_dt_utag) {
+	     if (DT_UTAG(c_p) == NIL) {
+		 DT_UTAG(c_p) = ERL_MESSAGE_DT_UTAG(msgp);
+	     }
+	     DT_UTAG_FLAGS(c_p) |= DT_UTAG_SPREADING;
+#ifdef DTRACE_TAG_HARDDEBUG
+	     erts_fprintf(stderr,
+			  "Dtrace -> (%T) receive tag (%T) "
+			  "with message %T\r\n",
+			  c_p->id, DT_UTAG(c_p), ERL_MESSAGE_TERM(msgp));
+#endif
+	 } else {
+#endif
+	     ASSERT(is_tuple(SEQ_TRACE_TOKEN(c_p)));
+	     ASSERT(SEQ_TRACE_TOKEN_ARITY(c_p) == 5);
+	     ASSERT(is_small(SEQ_TRACE_TOKEN_SERIAL(c_p)));
+	     ASSERT(is_small(SEQ_TRACE_TOKEN_LASTCNT(c_p)));
+	     ASSERT(is_small(SEQ_TRACE_TOKEN_FLAGS(c_p)));
+	     ASSERT(is_pid(SEQ_TRACE_TOKEN_SENDER(c_p)));
+	     c_p->seq_trace_lastcnt = unsigned_val(SEQ_TRACE_TOKEN_SERIAL(c_p));
+	     if (c_p->seq_trace_clock < unsigned_val(SEQ_TRACE_TOKEN_SERIAL(c_p))) {
+		 c_p->seq_trace_clock = unsigned_val(SEQ_TRACE_TOKEN_SERIAL(c_p));
+	     }
+	     msg = ERL_MESSAGE_TERM(msgp);
+	     seq_trace_output(SEQ_TRACE_TOKEN(c_p), msg, SEQ_TRACE_RECEIVE, 
+			      c_p->id, c_p);
+#ifdef USE_VM_PROBES
 	 }
-	 msg = ERL_MESSAGE_TERM(msgp);
-	 seq_trace_output(SEQ_TRACE_TOKEN(c_p), msg, SEQ_TRACE_RECEIVE, 
-			  c_p->id, c_p);
+#endif
+     }
+#ifdef USE_VM_PROBES
+     if (DTRACE_ENABLED(message_receive)) {
+         Eterm token2 = NIL;
+         DTRACE_CHARBUF(receiver_name, DTRACE_TERM_BUF_SIZE);
+         Sint tok_label = 0;
+         Sint tok_lastcnt = 0;
+         Sint tok_serial = 0;
+
+         dtrace_proc_str(c_p, receiver_name);
+         token2 = SEQ_TRACE_TOKEN(c_p);
+         if (token2 != NIL && token2 != am_have_dt_utag) {
+             tok_label = signed_val(SEQ_TRACE_T_LABEL(token2));
+             tok_lastcnt = signed_val(SEQ_TRACE_T_LASTCNT(token2));
+             tok_serial = signed_val(SEQ_TRACE_T_SERIAL(token2));
+         }
+         DTRACE6(message_receive,
+                 receiver_name, size_object(ERL_MESSAGE_TERM(msgp)),
+                 c_p->msg.len - 1, tok_label, tok_lastcnt, tok_serial);
      }
+#endif
      UNLINK_MESSAGE(c_p, msgp);
      JOIN_MESSAGE(c_p);
      CANCEL_TIMER(c_p);
@@ -3157,6 +3372,7 @@ void process_main(void)
 	     */
 	    BifFunction vbf;
 
+	    DTRACE_NIF_ENTRY(c_p, (Eterm)I[-3], (Eterm)I[-2], (Uint)I[-1]);
 	    c_p->current = I-3; /* current and vbf set to please handle_error */ 
 	    SWAPOUT;
 	    c_p->fcalls = FCALLS - 1;
@@ -3178,6 +3394,8 @@ void process_main(void)
 	    ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(nif_bif_result));
 	    PROCESS_MAIN_CHK_LOCKS(c_p);
 	    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
+
+	    DTRACE_NIF_RETURN(c_p, (Eterm)I[-3], (Eterm)I[-2], (Uint)I[-1]);
 	    goto apply_bif_or_nif_epilogue;
 	 
 	OpCase(apply_bif):
@@ -3197,6 +3415,8 @@ void process_main(void)
 	    c_p->arity = 0;		/* To allow garbage collection on ourselves
 					 * (check_process_code/2).
 					 */
+	    DTRACE_BIF_ENTRY(c_p, (Eterm)I[-3], (Eterm)I[-2], (Uint)I[-1]);
+
 	    SWAPOUT;
 	    c_p->fcalls = FCALLS - 1;
 	    vbf = (BifFunction) Arg(0);
@@ -3216,6 +3436,8 @@ void process_main(void)
 		PROCESS_MAIN_CHK_LOCKS(c_p);
 	    }
 
+	    DTRACE_BIF_RETURN(c_p, (Eterm)I[-3], (Eterm)I[-2], (Uint)I[-1]);
+
 	apply_bif_or_nif_epilogue:
 	    ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
 	    ERTS_HOLE_CHECK(c_p);
@@ -5900,6 +6122,12 @@ apply(Process* p, Eterm module, Eterm function, Eterm args, Eterm* reg)
 	save_calls(p, ep);
     }
 
+#ifdef USE_VM_CALL_PROBES
+    if (DTRACE_ENABLED(global_function_entry)) {
+        BeamInstr *fptr = (BeamInstr *) ep->addressv[erts_active_code_ix()];
+	DTRACE_GLOBAL_CALL(p, (Eterm)fptr[-3], (Eterm)fptr[-2], (Uint)fptr[-1]);
+    }
+#endif
     return ep->addressv[erts_active_code_ix()];
 }
 
@@ -5949,6 +6177,12 @@ fixed_apply(Process* p, Eterm* reg, Uint arity)
 	save_calls(p, ep);
     }
 
+#ifdef USE_VM_CALL_PROBES
+    if (DTRACE_ENABLED(global_function_entry)) {
+        BeamInstr *fptr = (BeamInstr *)  ep->addressv[erts_active_code_ix()];
+	DTRACE_GLOBAL_CALL(p, (Eterm)fptr[-3], (Eterm)fptr[-2], (Uint)fptr[-1]);
+    }
+#endif
     return ep->addressv[erts_active_code_ix()];
 }
 
@@ -5998,6 +6232,15 @@ erts_hibernate(Process* c_p, Eterm module, Eterm function, Eterm args, Eterm* re
 	c_p->max_arg_reg = sizeof(c_p->def_arg_reg)/sizeof(c_p->def_arg_reg[0]);
     }
 
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(process_hibernate)) {
+        DTRACE_CHARBUF(process_name, DTRACE_TERM_BUF_SIZE);
+        DTRACE_CHARBUF(mfa, DTRACE_TERM_BUF_SIZE);
+        dtrace_fun_decode(c_p, module, function, arity,
+                          process_name, mfa);
+        DTRACE2(process_hibernate, process_name, mfa);
+    }
+#endif
     /*
      * Arrange for the process to be resumed at the given MFA with
      * the stack cleared.
@@ -6073,6 +6316,9 @@ call_fun(Process* p,		/* Current process. */
 	actual_arity = (int) code_ptr[-1];
 
 	if (actual_arity == arity+num_free) {
+	    DTRACE_LOCAL_CALL(p, (Eterm)code_ptr[-3],
+			(Eterm)code_ptr[-2],
+			code_ptr[-1]);
 	    if (num_free == 0) {
 		return code_ptr;
 	    } else {
@@ -6090,7 +6336,7 @@ call_fun(Process* p,		/* Current process. */
 	} else {
 	    /*
 	     * Something wrong here. First build a list of the arguments.
-	     */  
+	     */
 
 	    if (is_non_value(args)) {
 		Uint sz = 2 * arity;
@@ -6165,6 +6411,7 @@ call_fun(Process* p,		/* Current process. */
 	actual_arity = (int) ep->code[2];
 
 	if (arity == actual_arity) {
+	    DTRACE_GLOBAL_CALL(p, ep->code[0], ep->code[1], (Uint)ep->code[2]);
 	    return ep->addressv[erts_active_code_ix()];
 	} else {
 	    /*
@@ -6240,6 +6487,7 @@ call_fun(Process* p,		/* Current process. */
 	    reg[1] = function;
 	    reg[2] = args;
 	}
+	DTRACE_GLOBAL_CALL(p, module, function, arity);
 	return ep->addressv[erts_active_code_ix()];
     } else {
     badfun:
diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c
index d39beea1f3..ee93cde243 100644
--- a/erts/emulator/beam/bif.c
+++ b/erts/emulator/beam/bif.c
@@ -563,7 +563,11 @@ erts_queue_monitor_message(Process *p,
     ref_copy    = copy_struct(ref, ref_size, &hp, ohp);
 
     tup = TUPLE5(hp, am_DOWN, ref_copy, type, item_copy, reason_copy);
-    erts_queue_message(p, p_locksp, bp, tup, NIL);
+    erts_queue_message(p, p_locksp, bp, tup, NIL
+#ifdef USE_VM_PROBES
+		       , NIL
+#endif
+		       );
 }
 
 static BIF_RETTYPE
@@ -1944,7 +1948,11 @@ do_send(Process *p, Eterm to, Eterm msg, int suspend) {
 	if (ERTS_PROC_GET_SAVED_CALLS_BUF(p))
 	    save_calls(p, &exp_send);
 	
-	if (SEQ_TRACE_TOKEN(p) != NIL) {
+	if (SEQ_TRACE_TOKEN(p) != NIL
+#ifdef USE_VM_PROBES
+	    && SEQ_TRACE_TOKEN(p) != am_have_dt_utag
+#endif
+	    ) {
 	    seq_trace_update_send(p);
 	    seq_trace_output(SEQ_TRACE_TOKEN(p), msg, 
 			     SEQ_TRACE_SEND, portid, p);
@@ -4227,13 +4235,21 @@ BIF_RETTYPE system_flag_2(BIF_ALIST_2)
 	for (i = 0; i < erts_max_processes; i++) {
 	    if (process_tab[i] != (Process*) 0) {
 		Process* p = process_tab[i];
+#ifdef USE_VM_PROBES
+		p->seq_trace_token = (p->dt_utag != NIL) ? am_have_dt_utag : NIL;
+#else
 		p->seq_trace_token = NIL;
+#endif
 		p->seq_trace_clock = 0;
 		p->seq_trace_lastcnt = 0;
 		ERTS_SMP_MSGQ_MV_INQ2PRIVQ(p);
 		mp = p->msg.first;
 		while(mp != NULL) {
+#ifdef USE_VM_PROBES
+		    ERL_MESSAGE_TOKEN(mp) = (ERL_MESSAGE_DT_UTAG(mp) != NIL) ? am_have_dt_utag : NIL;
+#else
 		    ERL_MESSAGE_TOKEN(mp) = NIL;
+#endif
 		    mp = mp->next;
 		}
 	    }
@@ -4642,3 +4658,193 @@ BIF_RETTYPE get_module_info_2(BIF_ALIST_2)
     }
     BIF_RET(ret);
 }
+
+BIF_RETTYPE dt_put_tag_1(BIF_ALIST_1)
+{
+#ifdef USE_VM_PROBES
+    Eterm otag;
+    if (BIF_ARG_1 == am_undefined) {
+	otag = (DT_UTAG(BIF_P) == NIL) ? am_undefined : DT_UTAG(BIF_P);
+	DT_UTAG(BIF_P) = NIL;
+	DT_UTAG_FLAGS(BIF_P) = 0;
+	if (SEQ_TRACE_TOKEN(BIF_P) == am_have_dt_utag) {
+	    SEQ_TRACE_TOKEN(BIF_P) = NIL;
+	}
+	BIF_RET(otag);
+    }
+    if (!is_binary(BIF_ARG_1)) {
+	BIF_ERROR(BIF_P,BADARG);
+    }
+    otag = (DT_UTAG(BIF_P) == NIL) ? am_undefined : DT_UTAG(BIF_P);
+    DT_UTAG(BIF_P) = BIF_ARG_1;
+    DT_UTAG_FLAGS(BIF_P) |= DT_UTAG_PERMANENT;
+    if (SEQ_TRACE_TOKEN(BIF_P) == NIL) {
+	SEQ_TRACE_TOKEN(BIF_P) = am_have_dt_utag;
+    }
+    BIF_RET(otag);
+#else
+    BIF_RET(am_undefined);
+#endif
+}
+
+BIF_RETTYPE dt_get_tag_0(BIF_ALIST_0)
+{
+#ifdef USE_VM_PROBES
+    BIF_RET((DT_UTAG(BIF_P) == NIL || !(DT_UTAG_FLAGS(BIF_P) & DT_UTAG_PERMANENT)) ? am_undefined : DT_UTAG(BIF_P));
+#else
+    BIF_RET(am_undefined);
+#endif
+}
+BIF_RETTYPE dt_get_tag_data_0(BIF_ALIST_0)
+{
+#ifdef USE_VM_PROBES
+    BIF_RET((DT_UTAG(BIF_P) == NIL) ? am_undefined : DT_UTAG(BIF_P));
+#else
+    BIF_RET(am_undefined);
+#endif
+}
+BIF_RETTYPE dt_prepend_vm_tag_data_1(BIF_ALIST_1)
+{
+#ifdef USE_VM_PROBES
+    Eterm b; 
+    Eterm *hp;
+    hp = HAlloc(BIF_P,2);
+    if (is_binary((DT_UTAG(BIF_P)))) {
+	Uint sz = binary_size(DT_UTAG(BIF_P));
+	int i;
+	unsigned char *p,*q;
+	byte *temp_alloc = NULL;
+	b = new_binary(BIF_P,NULL,sz+1);
+	q = binary_bytes(b);
+	p = erts_get_aligned_binary_bytes(DT_UTAG(BIF_P),&temp_alloc);
+	for(i=0;i<sz;++i) {
+	    q[i] = p[i];
+	} 
+	erts_free_aligned_binary_bytes(temp_alloc);
+	q[sz] = '\0';
+    } else {
+	b = new_binary(BIF_P,(byte *)"\0",1);
+    }
+    BIF_RET(CONS(hp,b,BIF_ARG_1));
+#else
+    BIF_RET(BIF_ARG_1);
+#endif
+}
+BIF_RETTYPE dt_append_vm_tag_data_1(BIF_ALIST_1)
+{
+#ifdef USE_VM_PROBES
+    Eterm b; 
+    Eterm *hp;
+    hp = HAlloc(BIF_P,2);
+    if (is_binary((DT_UTAG(BIF_P)))) {
+	Uint sz = binary_size(DT_UTAG(BIF_P));
+	int i;
+	unsigned char *p,*q;
+	byte *temp_alloc = NULL;
+	b = new_binary(BIF_P,NULL,sz+1);
+	q = binary_bytes(b);
+	p = erts_get_aligned_binary_bytes(DT_UTAG(BIF_P),&temp_alloc);
+	for(i=0;i<sz;++i) {
+	    q[i] = p[i];
+	} 
+	erts_free_aligned_binary_bytes(temp_alloc);
+	q[sz] = '\0';
+    } else {
+	b = new_binary(BIF_P,(byte *)"\0",1);
+    }
+    BIF_RET(CONS(hp,BIF_ARG_1,b));
+#else
+    BIF_RET(BIF_ARG_1);
+#endif
+}
+BIF_RETTYPE dt_spread_tag_1(BIF_ALIST_1)
+{
+#ifdef USE_VM_PROBES
+    Eterm ret;
+    Eterm *hp;
+#endif
+    if (BIF_ARG_1 != am_true && BIF_ARG_1 != am_false) {
+	BIF_ERROR(BIF_P,BADARG);
+    }
+#ifdef USE_VM_PROBES
+    hp = HAlloc(BIF_P,3);
+    ret = TUPLE2(hp,make_small(DT_UTAG_FLAGS(BIF_P)),DT_UTAG(BIF_P));
+    if (DT_UTAG(BIF_P) != NIL) {
+	if (BIF_ARG_1 == am_true) {
+	    DT_UTAG_FLAGS(BIF_P) |= DT_UTAG_SPREADING;
+#ifdef DTRACE_TAG_HARDDEBUG
+	    erts_fprintf(stderr,
+			 "Dtrace -> (%T) start spreading tag %T\r\n",
+			 BIF_P->id,DT_UTAG(BIF_P));
+#endif
+	} else {
+	    DT_UTAG_FLAGS(BIF_P) &= ~DT_UTAG_SPREADING;
+#ifdef DTRACE_TAG_HARDDEBUG
+	    erts_fprintf(stderr,
+			 "Dtrace -> (%T) stop spreading tag %T\r\n",
+			 BIF_P->id,DT_UTAG(BIF_P));
+#endif
+	}
+    }
+    BIF_RET(ret);
+#else
+    BIF_RET(am_true);
+#endif
+}
+BIF_RETTYPE dt_restore_tag_1(BIF_ALIST_1)
+{
+#ifdef USE_VM_PROBES
+    Eterm *tpl;
+    Uint x;
+    if (is_not_tuple(BIF_ARG_1)) {
+	BIF_ERROR(BIF_P,BADARG);
+    }
+    tpl = tuple_val(BIF_ARG_1);
+    if(arityval(*tpl) != 2 || is_not_small(tpl[1]) || (is_not_binary(tpl[2]) && tpl[2] != NIL)) {
+	BIF_ERROR(BIF_P,BADARG);
+    }
+    if (tpl[2] == NIL) {
+	if (DT_UTAG(BIF_P) != NIL) {
+#ifdef DTRACE_TAG_HARDDEBUG
+	    erts_fprintf(stderr,
+			 "Dtrace -> (%T) restore Killing tag!\r\n",
+			 BIF_P->id);
+#endif
+	}
+	DT_UTAG(BIF_P) = NIL;
+	if (SEQ_TRACE_TOKEN(BIF_P) == am_have_dt_utag) {
+	    SEQ_TRACE_TOKEN(BIF_P) = NIL;
+	}
+	DT_UTAG_FLAGS(BIF_P) = 0;
+    } else {
+	x = unsigned_val(tpl[1]) & (DT_UTAG_SPREADING | DT_UTAG_PERMANENT);
+#ifdef DTRACE_TAG_HARDDEBUG
+
+	if (!(x & DT_UTAG_SPREADING) && (DT_UTAG_FLAGS(BIF_P) & 
+					 DT_UTAG_SPREADING)) {
+	    erts_fprintf(stderr,
+			 "Dtrace -> (%T) restore stop spreading "
+			 "tag %T\r\n",
+			 BIF_P->id, tpl[2]);
+	} else if ((x & DT_UTAG_SPREADING) && 
+		   !(DT_UTAG_FLAGS(BIF_P) & DT_UTAG_SPREADING)) {
+	    erts_fprintf(stderr,
+			 "Dtrace -> (%T) restore start spreading "
+			 "tag %T\r\n",BIF_P->id,tpl[2]);
+	}
+#endif
+	DT_UTAG_FLAGS(BIF_P) = x;
+	DT_UTAG(BIF_P) = tpl[2];
+	if (SEQ_TRACE_TOKEN(BIF_P) == NIL) {
+	    SEQ_TRACE_TOKEN(BIF_P) = am_have_dt_utag;
+	}
+    }
+#else    
+    if (BIF_ARG_1 != am_true) {
+	BIF_ERROR(BIF_P,BADARG);
+    }
+#endif
+    BIF_RET(am_true);
+}
+
+
diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab
index 1338b90374..698bbf0098 100644
--- a/erts/emulator/beam/bif.tab
+++ b/erts/emulator/beam/bif.tab
@@ -814,6 +814,23 @@ bif erlang:universaltime_to_posixtime/1
 bif erlang:posixtime_to_universaltime/1
 
 #
+# New in R15B01
+#
+
+# The dtrace BIF's are always present, but give dummy results if dynamic trace is not enabled in the build
+bif erlang:dt_put_tag/1
+bif erlang:dt_get_tag/0
+bif erlang:dt_get_tag_data/0
+bif erlang:dt_spread_tag/1
+bif erlang:dt_restore_tag/1
+
+# These are dummies even with enabled dynamic trace unless vm probes are enabled. 
+# They are also internal, for dtrace tags sent to the VM's own drivers (efile)
+bif erlang:dt_prepend_vm_tag_data/1
+bif erlang:dt_append_vm_tag_data/1
+
+
+#
 # New in R16B.
 #
 bif erlang:prepare_loading/2
diff --git a/erts/emulator/beam/copy.c b/erts/emulator/beam/copy.c
index 1d968fb147..2c355fadfa 100644
--- a/erts/emulator/beam/copy.c
+++ b/erts/emulator/beam/copy.c
@@ -30,6 +30,7 @@
 #include "big.h"
 #include "erl_binary.h"
 #include "erl_bits.h"
+#include "dtrace-wrapper.h"
 
 #ifdef HYBRID
 MA_STACK_DECLARE(src);
@@ -59,6 +60,14 @@ copy_object(Eterm obj, Process* to)
     Eterm* hp = HAlloc(to, size);
     Eterm res;
 
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(copy_object)) {
+        DTRACE_CHARBUF(proc_name, 64);
+
+        erts_snprintf(proc_name, sizeof(proc_name), "%T", to->id);
+        DTRACE2(copy_object, proc_name, size);
+    }
+#endif
     res = copy_struct(obj, size, &hp, &to->off_heap);
 #ifdef DEBUG
     if (eq(obj, res) == 0) {
@@ -213,6 +222,8 @@ Eterm copy_struct(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap)
     if (IS_CONST(obj))
 	return obj;
 
+    DTRACE1(copy_struct, (int32_t)sz);
+
     hp = htop = *hpp;
     hbot   = htop + sz;
     hstart = (char *)htop;
diff --git a/erts/emulator/beam/dist.c b/erts/emulator/beam/dist.c
index 00d57ceab5..fa68d041d1 100644
--- a/erts/emulator/beam/dist.c
+++ b/erts/emulator/beam/dist.c
@@ -42,6 +42,7 @@
 #include "external.h"
 #include "erl_binary.h"
 #include "erl_thr_progress.h"
+#include "dtrace-wrapper.h"
 
 /* Turn this on to get printouts of all distribution messages
  * which go on the line
@@ -381,7 +382,11 @@ static void doit_node_link_net_exits(ErtsLink *lnk, void *vnecp)
 	    Eterm tup;
 	    Eterm *hp = erts_alloc_message_heap(3,&bp,&ohp,rp,&rp_locks);
 	    tup = TUPLE2(hp, am_nodedown, name);
-	    erts_queue_message(rp, &rp_locks, bp, tup, NIL);
+	    erts_queue_message(rp, &rp_locks, bp, tup, NIL
+#ifdef USE_VM_PROBES
+			       , NIL
+#endif
+			       );
 	}
 	erts_smp_proc_unlock(rp, rp_locks);
     }
@@ -740,19 +745,50 @@ erts_dsig_send_msg(ErtsDSigData *dsdp, Eterm remote, Eterm message)
     Eterm token = NIL;
     Process *sender = dsdp->proc;
     int res;
+#ifdef USE_VM_PROBES
+    Sint tok_label = 0;
+    Sint tok_lastcnt = 0;
+    Sint tok_serial = 0;
+    Uint msize = 0;
+    DTRACE_CHARBUF(node_name, 64);
+    DTRACE_CHARBUF(sender_name, 64);
+    DTRACE_CHARBUF(receiver_name, 64);
+#endif
 
     UseTmpHeapNoproc(5);
-    if (SEQ_TRACE_TOKEN(sender) != NIL) {
+    if (SEQ_TRACE_TOKEN(sender) != NIL 
+#ifdef USE_VM_PROBES
+	&& SEQ_TRACE_TOKEN(sender) != am_have_dt_utag 
+#endif
+	) {
 	seq_trace_update_send(sender);
 	token = SEQ_TRACE_TOKEN(sender);
 	seq_trace_output(token, message, SEQ_TRACE_SEND, remote, sender);
     }
+#ifdef USE_VM_PROBES
+    *node_name = *sender_name = *receiver_name = '\0';
+    if (DTRACE_ENABLED(message_send) || DTRACE_ENABLED(message_send_remote)) {
+        erts_snprintf(node_name, sizeof(node_name), "%T", dsdp->dep->sysname);
+        erts_snprintf(sender_name, sizeof(sender_name), "%T", sender->id);
+        erts_snprintf(receiver_name, sizeof(receiver_name), "%T", remote);
+        msize = size_object(message);
+        if (token != NIL && token != am_have_dt_utag) {
+            tok_label = signed_val(SEQ_TRACE_T_LABEL(token));
+            tok_lastcnt = signed_val(SEQ_TRACE_T_LASTCNT(token));
+            tok_serial = signed_val(SEQ_TRACE_T_SERIAL(token));
+        }
+    }
+#endif
 
     if (token != NIL)
 	ctl = TUPLE4(&ctl_heap[0],
 		     make_small(DOP_SEND_TT), am_Cookie, remote, token);
     else
 	ctl = TUPLE3(&ctl_heap[0], make_small(DOP_SEND), am_Cookie, remote);
+    DTRACE6(message_send, sender_name, receiver_name,
+            msize, tok_label, tok_lastcnt, tok_serial);
+    DTRACE7(message_send_remote, sender_name, node_name, receiver_name,
+            msize, tok_label, tok_lastcnt, tok_serial);
     res = dsig_send(dsdp, ctl, message, 0);
     UnUseTmpHeapNoproc(5);
     return res;
@@ -766,13 +802,41 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message)
     Eterm token = NIL;
     Process *sender = dsdp->proc;
     int res;
+#ifdef USE_VM_PROBES
+    Sint tok_label = 0;
+    Sint tok_lastcnt = 0;
+    Sint tok_serial = 0;
+    Uint32 msize = 0;
+    DTRACE_CHARBUF(node_name, 64);
+    DTRACE_CHARBUF(sender_name, 64);
+    DTRACE_CHARBUF(receiver_name, 128);
+#endif
 
     UseTmpHeapNoproc(6);
-    if (SEQ_TRACE_TOKEN(sender) != NIL) {
+    if (SEQ_TRACE_TOKEN(sender) != NIL
+#ifdef USE_VM_PROBES
+	&& SEQ_TRACE_TOKEN(sender) != am_have_dt_utag 
+#endif
+	) {
 	seq_trace_update_send(sender);
 	token = SEQ_TRACE_TOKEN(sender);
 	seq_trace_output(token, message, SEQ_TRACE_SEND, remote_name, sender);
     }
+#ifdef USE_VM_PROBES
+    *node_name = *sender_name = *receiver_name = '\0';
+    if (DTRACE_ENABLED(message_send) || DTRACE_ENABLED(message_send_remote)) {
+        erts_snprintf(node_name, sizeof(node_name), "%T", dsdp->dep->sysname);
+        erts_snprintf(sender_name, sizeof(sender_name), "%T", sender->id);
+        erts_snprintf(receiver_name, sizeof(receiver_name),
+                      "{%T,%s}", remote_name, node_name);
+        msize = size_object(message);
+        if (token != NIL && token != am_have_dt_utag) {
+            tok_label = signed_val(SEQ_TRACE_T_LABEL(token));
+            tok_lastcnt = signed_val(SEQ_TRACE_T_LASTCNT(token));
+            tok_serial = signed_val(SEQ_TRACE_T_SERIAL(token));
+        }
+    }
+#endif
 
     if (token != NIL)
 	ctl = TUPLE5(&ctl_heap[0], make_small(DOP_REG_SEND_TT),
@@ -780,6 +844,10 @@ erts_dsig_send_reg_msg(ErtsDSigData *dsdp, Eterm remote_name, Eterm message)
     else
 	ctl = TUPLE4(&ctl_heap[0], make_small(DOP_REG_SEND),
 		     sender->id, am_Cookie, remote_name);
+    DTRACE6(message_send, sender_name, receiver_name,
+            msize, tok_label, tok_lastcnt, tok_serial);
+    DTRACE7(message_send_remote, sender_name, node_name, receiver_name,
+            msize, tok_label, tok_lastcnt, tok_serial);
     res = dsig_send(dsdp, ctl, message, 0);
     UnUseTmpHeapNoproc(6);
     return res;
@@ -793,9 +861,23 @@ erts_dsig_send_exit_tt(ErtsDSigData *dsdp, Eterm local, Eterm remote,
     Eterm ctl;
     DeclareTmpHeapNoproc(ctl_heap,6);
     int res;
+#ifdef USE_VM_PROBES
+    Process *sender = dsdp->proc;
+    Sint tok_label = 0;
+    Sint tok_lastcnt = 0;
+    Sint tok_serial = 0;
+    DTRACE_CHARBUF(node_name, 64);
+    DTRACE_CHARBUF(sender_name, 64);
+    DTRACE_CHARBUF(remote_name, 128);
+    DTRACE_CHARBUF(reason_str, 128);
+#endif
 
     UseTmpHeapNoproc(6);
-    if (token != NIL) {	
+    if (token != NIL 
+#ifdef USE_VM_PROBES
+	&& token != am_have_dt_utag
+#endif
+	) {	
 	seq_trace_update_send(dsdp->proc);
 	seq_trace_output_exit(token, reason, SEQ_TRACE_SEND, remote, local);
 	ctl = TUPLE5(&ctl_heap[0],
@@ -803,6 +885,23 @@ erts_dsig_send_exit_tt(ErtsDSigData *dsdp, Eterm local, Eterm remote,
     } else {
 	ctl = TUPLE4(&ctl_heap[0], make_small(DOP_EXIT), local, remote, reason);
     }
+#ifdef USE_VM_PROBES
+    *node_name = *sender_name = *remote_name = '\0';
+    if (DTRACE_ENABLED(process_exit_signal_remote)) {
+        erts_snprintf(node_name, sizeof(node_name), "%T", dsdp->dep->sysname);
+        erts_snprintf(sender_name, sizeof(sender_name), "%T", sender->id);
+        erts_snprintf(remote_name, sizeof(remote_name),
+                      "{%T,%s}", remote, node_name);
+        erts_snprintf(reason_str, sizeof(reason), "%T", reason);
+        if (token != NIL && token != am_have_dt_utag) {
+            tok_label = signed_val(SEQ_TRACE_T_LABEL(token));
+            tok_lastcnt = signed_val(SEQ_TRACE_T_LASTCNT(token));
+            tok_serial = signed_val(SEQ_TRACE_T_SERIAL(token));
+        }
+    }
+#endif
+    DTRACE7(process_exit_signal_remote, sender_name, node_name,
+            remote_name, reason_str, tok_label, tok_lastcnt, tok_serial);
     /* forced, i.e ignore busy */
     res = dsig_send(dsdp, ctl, THE_NON_VALUE, 1);
     UnUseTmpHeapNoproc(6);
@@ -1619,6 +1718,18 @@ dsig_send(ErtsDSigData *dsdp, Eterm ctl, Eterm msg, int force_busy)
 	    if (!(dep->qflgs & ERTS_DE_QFLG_BUSY)) {
 		if (suspended)
 		    resume = 1; /* was busy when we started, but isn't now */
+#ifdef USE_VM_PROBES
+                if (resume && DTRACE_ENABLED(dist_port_not_busy)) {
+                    DTRACE_CHARBUF(port_str, 64);
+                    DTRACE_CHARBUF(remote_str, 64);
+
+                    erts_snprintf(port_str, sizeof(port_str), "%T", cid);
+                    erts_snprintf(remote_str, sizeof(remote_str),
+                                  "%T", dep->sysname);
+                    DTRACE3(dist_port_not_busy, erts_this_node_sysname,
+                            port_str, remote_str);
+                }
+#endif
 	    }
 	    else {
 		/* Enqueue suspended process on dist entry */
@@ -1668,6 +1779,19 @@ dsig_send(ErtsDSigData *dsdp, Eterm ctl, Eterm msg, int force_busy)
     }
 
     if (suspended) {
+#ifdef USE_VM_PROBES
+        if (!resume && DTRACE_ENABLED(dist_port_busy)) {
+            DTRACE_CHARBUF(port_str, 64);
+            DTRACE_CHARBUF(remote_str, 64);
+            DTRACE_CHARBUF(pid_str, 16);
+
+            erts_snprintf(port_str, sizeof(port_str), "%T", cid);
+            erts_snprintf(remote_str, sizeof(remote_str), "%T", dep->sysname);
+            erts_snprintf(pid_str, sizeof(pid_str), "%T", c_p->id);
+            DTRACE4(dist_port_busy, erts_this_node_sysname,
+                    port_str, remote_str, pid_str);
+        }
+#endif
 	if (!resume && erts_system_monitor_flags.busy_dist_port)
 	    monitor_generic(c_p, am_busy_dist_port, cid);
 	return ERTS_DSIG_SEND_YIELD;
@@ -1691,6 +1815,18 @@ dist_port_command(Port *prt, ErtsDistOutputBuf *obuf)
 		 "(%beu bytes) passed.\n",
 		 size);
 
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(dist_output)) {
+        DTRACE_CHARBUF(port_str, 64);
+        DTRACE_CHARBUF(remote_str, 64);
+
+        erts_snprintf(port_str, sizeof(port_str), "%T", prt->id);
+        erts_snprintf(remote_str, sizeof(remote_str),
+                      "%T", prt->dist_entry->sysname);
+        DTRACE4(dist_output, erts_this_node_sysname, port_str,
+                remote_str, size);
+    }
+#endif
     prt->caller = NIL;
     fpe_was_unmasked = erts_block_fpe();
     (*prt->drv_ptr->output)((ErlDrvData) prt->drv_data,
@@ -1733,6 +1869,18 @@ dist_port_commandv(Port *prt, ErtsDistOutputBuf *obuf)
 
     ASSERT(prt->drv_ptr->outputv);
 
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(dist_outputv)) {
+        DTRACE_CHARBUF(port_str, 64);
+        DTRACE_CHARBUF(remote_str, 64);
+
+        erts_snprintf(port_str, sizeof(port_str), "%T", prt->id);
+        erts_snprintf(remote_str, sizeof(remote_str),
+                      "%T", prt->dist_entry->sysname);
+        DTRACE4(dist_outputv, erts_this_node_sysname, port_str,
+                remote_str, size);
+    }
+#endif
     prt->caller = NIL;
     fpe_was_unmasked = erts_block_fpe();
     (*prt->drv_ptr->outputv)((ErlDrvData) prt->drv_data, &eiov);
@@ -2052,6 +2200,18 @@ erts_dist_command(Port *prt, int reds_limit)
 void
 erts_dist_port_not_busy(Port *prt)
 {
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(dist_port_not_busy)) {
+        DTRACE_CHARBUF(port_str, 64);
+        DTRACE_CHARBUF(remote_str, 64);
+
+        erts_snprintf(port_str, sizeof(port_str), "%T", prt->id);
+        erts_snprintf(remote_str, sizeof(remote_str),
+                      "%T", prt->dist_entry->sysname);
+        DTRACE3(dist_port_not_busy, erts_this_node_sysname,
+                port_str, remote_str);
+    }
+#endif
     erts_schedule_dist_command(prt, NULL);
 }
 
@@ -2972,7 +3132,11 @@ send_nodes_mon_msg(Process *rp,
     }
 
     ASSERT(hend == hp);
-    erts_queue_message(rp, rp_locksp, bp, msg, NIL);
+    erts_queue_message(rp, rp_locksp, bp, msg, NIL
+#ifdef USE_VM_PROBES
+		       , NIL
+#endif
+		       );
 }
 
 static void
@@ -2985,6 +3149,21 @@ send_nodes_mon_msgs(Process *c_p, Eterm what, Eterm node, Eterm type, Eterm reas
     ASSERT(is_immed(what));
     ASSERT(is_immed(node));
     ASSERT(is_immed(type));
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(dist_monitor)) {
+        DTRACE_CHARBUF(what_str, 12);
+        DTRACE_CHARBUF(node_str, 64);
+        DTRACE_CHARBUF(type_str, 12);
+        DTRACE_CHARBUF(reason_str, 64);
+
+        erts_snprintf(what_str, sizeof(what_str), "%T", what);
+        erts_snprintf(node_str, sizeof(node_str), "%T", node);
+        erts_snprintf(type_str, sizeof(type_str), "%T", type);
+        erts_snprintf(reason_str, sizeof(reason_str), "%T", reason);
+        DTRACE5(dist_monitor, erts_this_node_sysname,
+                what_str, node_str, type_str, reason_str);
+    }
+#endif
 
     ERTS_SMP_LC_ASSERT(!c_p
 		       || (erts_proc_lc_my_proc_locks(c_p)
diff --git a/erts/emulator/beam/dtrace-wrapper.h b/erts/emulator/beam/dtrace-wrapper.h
new file mode 100644
index 0000000000..9d1e55fc43
--- /dev/null
+++ b/erts/emulator/beam/dtrace-wrapper.h
@@ -0,0 +1,109 @@
+/*
+ * %CopyrightBegin%
+ *
+ * Copyright Dustin Sallings, Michal Ptaszek, Scott Lystig Fritchie 2011.
+ * All Rights Reserved.
+ *
+ * The contents of this file are subject to the Erlang Public License,
+ * Version 1.1, (the "License"); you may not use this file except in
+ * compliance with the License. You should have received a copy of the
+ * Erlang Public License along with this software. If not, it can be
+ * retrieved online at http://www.erlang.org/.
+ *
+ * Software distributed under the License is distributed on an "AS IS"
+ * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
+ * the License for the specific language governing rights and limitations
+ * under the License.
+ *
+ * %CopyrightEnd%
+ */
+
+#ifndef __DTRACE_WRAPPER_H
+#define __DTRACE_WRAPPER_H
+
+#define DTRACE_TERM_BUF_SIZE 256
+
+/*
+ * Some varieties of SystemTap macros do not like statically-sized
+ * char[N] buffers.  (For example, CentOS 6's macros.)
+ * So, we'll play a game to humor them.
+ *
+ * The code necessary to play nice with CentOS 6's SystemTap looks
+ * stupid to a C programmer's eyes, so we hide the ugliness with this
+ * macro, which expands:
+ *
+ *    DTRACE_CHARBUF(proc_name, 64);
+ *
+ * to become:
+ *
+ *    char proc_name_BUFFER[64], *proc_name = proc_name_BUFFER;
+ */
+
+#define DTRACE_CHARBUF(name, size) \
+    char name##_BUFFER[size], *name = name##_BUFFER
+
+#if defined(USE_DYNAMIC_TRACE) && defined(USE_VM_PROBES) 
+
+#include "erlang_dtrace.h"
+
+#define DTRACE_ENABLED(name)                         \
+    erlang_##name##_enabled()
+#define DTRACE0(name)                                \
+    erlang_##name()
+#define DTRACE1(name, a0)                            \
+    erlang_##name(a0)
+#define DTRACE2(name, a0, a1)                        \
+    erlang_##name((a0), (a1))
+#define DTRACE3(name, a0, a1, a2)                    \
+    erlang_##name((a0), (a1), (a2))
+#define DTRACE4(name, a0, a1, a2, a3)                \
+    erlang_##name((a0), (a1), (a2), (a3))
+#define DTRACE5(name, a0, a1, a2, a3, a4)            \
+    erlang_##name((a0), (a1), (a2), (a3), (a4))
+#define DTRACE6(name, a0, a1, a2, a3, a4, a5)        \
+    erlang_##name((a0), (a1), (a2), (a3), (a4), (a5))
+#define DTRACE7(name, a0, a1, a2, a3, a4, a5, a6)    \
+    erlang_##name((a0), (a1), (a2), (a3), (a4), (a5), (a6))
+#define DTRACE10(name, a0, a1, a2, a3, a4, a5, a6, a7, a8, a9) \
+    erlang_##name((a0), (a1), (a2), (a3), (a4), (a5), (a6), (a7), (a8), (a9))
+#define DTRACE11(name, a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10) \
+    erlang_##name((a0), (a1), (a2), (a3), (a4), (a5), (a6), (a7), (a8), (a9), (a10))
+
+#if defined(_SDT_PROBE) && !defined(STAP_PROBE11)
+/* SLF: This is Ubuntu 11-style SystemTap hackery */
+/* work arround for missing STAP macro */
+#define STAP_PROBE11(provider,name,arg1,arg2,arg3,arg4,arg5,arg6,arg7,arg8,arg9,arg10,arg11) \
+  _SDT_PROBE(provider, name, 11, \
+             (arg1,arg2,arg3,arg4,arg5,arg6,arg7,arg8,arg9,arg10,arg11))
+#define _SDT_ASM_OPERANDS_11(arg1,arg2,arg3,arg4,arg5,arg6,arg7,arg8,arg9,arg10,arg11) \
+  _SDT_ASM_OPERANDS_10(arg1, arg2, arg3, arg4, arg5, arg6, arg7, arg8, arg9,arg10), \
+    _SDT_ARG(11, arg11)
+#endif
+
+#ifdef STAP_PROBE_ADDR
+/* SLF: This is CentOS 5-style SystemTap hackery */
+/* SystemTap compat mode cannot support 11 args. We'll ignore the 11th */
+#define STAP_PROBE11(provider,probe,parm1,parm2,parm3,parm4,parm5,parm6,parm7,parm8,parm9,parm10,parm11) \
+    STAP_PROBE10(provider,probe,(parm1),(parm2),(parm3),(parm4),(parm5),(parm6),(parm7),(parm8),(parm9),(parm10))
+#endif /* STAP_PROBE_ADDR */
+
+#else   /* USE_DYNAMIC_TRACE && USE_VM_PROBES */
+
+/* Render all macros to do nothing */
+#define DTRACE_ENABLED(name)                         0
+#define DTRACE0(name)                                do {} while (0)
+#define DTRACE1(name, a0)                            do {} while (0)
+#define DTRACE2(name, a0, a1)                        do {} while (0)
+#define DTRACE3(name, a0, a1, a2)                    do {} while (0)
+#define DTRACE4(name, a0, a1, a2, a3)                do {} while (0)
+#define DTRACE5(name, a0, a1, a2, a3, a4)            do {} while (0)
+#define DTRACE6(name, a0, a1, a2, a3, a4, a5)        do {} while (0)
+#define DTRACE7(name, a0, a1, a2, a3, a4, a5, a6)    do {} while (0)
+#define DTRACE10(name, a0, a1, a2, a3, a4, a5, a6, a7, a8, a9) \
+                                                     do {} while (0)
+#define DTRACE11(name, a0, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10) \
+                                                     do {} while (0)
+
+#endif  /* USE_DYNAMIC_TRACE && USE_VM_PROBES */
+
+#endif  /* __DTRACE_WRAPPER_H */
diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c
index ac92822aca..c501b79b1f 100644
--- a/erts/emulator/beam/erl_alloc.c
+++ b/erts/emulator/beam/erl_alloc.c
@@ -3001,7 +3001,11 @@ reply_alloc_info(void *vair)
 	HRelease(rp, hp_end, hp);	    
     }
 
-    erts_queue_message(rp, &rp_locks, bp, msg, NIL);
+    erts_queue_message(rp, &rp_locks, bp, msg, NIL
+#ifdef USE_VM_PROBES
+		       , NIL
+#endif
+		       );
 
     if (air->req_sched == sched_id)
 	rp_locks &= ~ERTS_PROC_LOCK_MAIN;
diff --git a/erts/emulator/beam/erl_async.c b/erts/emulator/beam/erl_async.c
index 8bca9ae582..f0e98b33a5 100644
--- a/erts/emulator/beam/erl_async.c
+++ b/erts/emulator/beam/erl_async.c
@@ -26,6 +26,7 @@
 #include "erl_threads.h"
 #include "erl_thr_queue.h"
 #include "erl_async.h"
+#include "dtrace-wrapper.h"
 
 #define ERTS_MAX_ASYNC_READY_CALLS_IN_SEQ 20
 
@@ -121,6 +122,14 @@ typedef struct {
 #endif
 } ErtsAsyncData;
 
+/*
+ * Some compilers, e.g. GCC 4.2.1 and -O3, will optimize away DTrace
+ * calls if they're the last thing in the function.  :-(
+ * Many thanks to Trond Norbye, via:
+ * https://github.com/memcached/memcached/commit/6298b3978687530bc9d219b6ac707a1b681b2a46
+ */
+static unsigned gcc_optimizer_hack = 0;
+
 int erts_async_max_threads; /* Initialized by erl_init.c */
 int erts_async_thread_suggested_stack_size; /* Initialized by erl_init.c */
 
@@ -244,6 +253,8 @@ erts_get_async_ready_queue(Uint sched_id)
 
 static ERTS_INLINE void async_add(ErtsAsync *a, ErtsAsyncQ* q)
 {
+    int len;
+
     if (is_internal_port(a->port)) {
 #if ERTS_USE_ASYNC_READY_Q
 	ErtsAsyncReadyQ *arq = async_ready_q(a->sched_id);
@@ -259,6 +270,17 @@ static ERTS_INLINE void async_add(ErtsAsync *a, ErtsAsyncQ* q)
 #endif
 
     erts_thr_q_enqueue(&q->thr_q, a);
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(aio_pool_add)) {
+        DTRACE_CHARBUF(port_str, 16);
+
+        erts_snprintf(port_str, sizeof(port_str), "%T", a->port);
+        /* DTRACE TODO: Get the queue length from erts_thr_q_enqueue() ? */
+        len = -1;
+        DTRACE2(aio_pool_add, port_str, len);
+    }
+#endif
+    gcc_optimizer_hack++;
 }
 
 static ERTS_INLINE ErtsAsync *async_get(ErtsThrQ_t *q,
@@ -269,6 +291,7 @@ static ERTS_INLINE ErtsAsync *async_get(ErtsThrQ_t *q,
     int saved_fin_deq = 0;
     ErtsThrQFinDeQ_t fin_deq;
 #endif
+    int len;
 
     while (1) {
 	ErtsAsync *a = (ErtsAsync *) erts_thr_q_dequeue(q);
@@ -280,7 +303,16 @@ static ERTS_INLINE ErtsAsync *async_get(ErtsThrQ_t *q,
 	    if (saved_fin_deq)
 		erts_thr_q_append_finalize_dequeue_data(&a->q.fin_deq, &fin_deq);
 #endif
+#ifdef USE_VM_PROBES
+            if (DTRACE_ENABLED(aio_pool_get)) {
+                DTRACE_CHARBUF(port_str, 16);
 
+                erts_snprintf(port_str, sizeof(port_str), "%T", a->port);
+                /* DTRACE TODO: Get the length from erts_thr_q_dequeue() ? */
+                len = -1;
+                DTRACE2(aio_pool_get, port_str, len);
+            }
+#endif
 	    return a;
 	}
 
diff --git a/erts/emulator/beam/erl_bif_ddll.c b/erts/emulator/beam/erl_bif_ddll.c
index 37d540b41b..bcfdacb91c 100644
--- a/erts/emulator/beam/erl_bif_ddll.c
+++ b/erts/emulator/beam/erl_bif_ddll.c
@@ -45,6 +45,7 @@
 #include "big.h"
 #include "dist.h"
 #include "erl_version.h"
+#include "dtrace-wrapper.h"
 
 #ifdef ERTS_SMP
 #define DDLL_SMP 1
@@ -1647,6 +1648,7 @@ static int do_unload_driver_entry(DE_Handle *dh, Eterm *save_name)
 	       diver_list lock here!*/
 	    if (q->finish) {
 		int fpe_was_unmasked = erts_block_fpe();
+		DTRACE1(driver_finish, q->name);
 		(*(q->finish))();
 		erts_unblock_fpe(fpe_was_unmasked);
 	    }
@@ -1760,7 +1762,11 @@ static void notify_proc(Process *proc, Eterm ref, Eterm driver_name, Eterm type,
 	hp += REF_THING_SIZE;
 	mess = TUPLE5(hp,type,r,am_driver,driver_name,tag);
     }
-    erts_queue_message(proc, &rp_locks, bp, mess, am_undefined);
+    erts_queue_message(proc, &rp_locks, bp, mess, am_undefined
+#ifdef USE_VM_PROBES
+		       , NIL
+#endif
+		       );
     erts_smp_proc_unlock(proc, rp_locks);
     ERTS_SMP_CHK_NO_PROC_LOCKS;
 }
diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c
index e198be67c7..a469dfc929 100644
--- a/erts/emulator/beam/erl_bif_info.c
+++ b/erts/emulator/beam/erl_bif_info.c
@@ -115,6 +115,12 @@ static char erts_system_version[] = ("Erlang " ERLANG_OTP_RELEASE
 #ifdef VALGRIND
 				     " [valgrind-compiled]"
 #endif
+#ifdef USE_DTRACE
+				     " [dtrace]"
+#endif
+#ifdef USE_SYSTEMTAP
+				     " [systemtap]"
+#endif
 				     "\n");
 
 #define ASIZE(a) (sizeof(a)/sizeof(a[0]))
@@ -2720,6 +2726,24 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1)
 #endif
 	BIF_RET(am_true);
     }
+    else if (ERTS_IS_ATOM_STR("dynamic_trace", BIF_ARG_1)) {
+#if defined(USE_DTRACE)
+	DECL_AM(dtrace);
+	BIF_RET(AM_dtrace);
+#elif defined(USE_SYSTEMTAP)
+	DECL_AM(systemtap);
+	BIF_RET(AM_systemtap);
+#else
+	BIF_RET(am_none);
+#endif
+    }	    
+    else if (ERTS_IS_ATOM_STR("dynamic_trace_probes", BIF_ARG_1)) {
+#if defined(USE_VM_PROBES)
+	BIF_RET(am_true);
+#else
+	BIF_RET(am_false);
+#endif	
+    }
 #ifdef ERTS_SMP
     else if (ERTS_IS_ATOM_STR("thread_progress", BIF_ARG_1)) {
 	erts_thr_progress_dbg_print_state();
diff --git a/erts/emulator/beam/erl_bif_port.c b/erts/emulator/beam/erl_bif_port.c
index cd423eb200..3056319809 100644
--- a/erts/emulator/beam/erl_bif_port.c
+++ b/erts/emulator/beam/erl_bif_port.c
@@ -40,6 +40,7 @@
 #include "external.h"
 #include "packet_parser.h"
 #include "erl_bits.h"
+#include "dtrace-wrapper.h"
 
 static int open_port(Process* p, Eterm name, Eterm settings, int *err_nump);
 static byte* convert_environment(Process* p, Eterm env);
@@ -343,6 +344,16 @@ port_call(Process* c_p, Eterm arg1, Eterm arg2, Eterm arg3)
 		 __FILE__, __LINE__, endp - (bytes + size));
     }
     erts_smp_proc_unlock(c_p, ERTS_PROC_LOCK_MAIN);
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(driver_call)) {
+        DTRACE_CHARBUF(process_str, DTRACE_TERM_BUF_SIZE);
+        DTRACE_CHARBUF(port_str, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_pid_str(p->connected, process_str);
+        dtrace_port_str(p, port_str);
+        DTRACE5(driver_call, process_str, port_str, p->name, op, real_size);
+    }
+#endif
     prc  = (char *) port_resp;
     fpe_was_unmasked = erts_block_fpe();
     ret = drv->call((ErlDrvData)p->drv_data, 
@@ -539,6 +550,18 @@ BIF_RETTYPE port_connect_2(BIF_ALIST_2)
 
     prt->connected = pid; /* internal pid */
     erts_smp_port_unlock(prt);
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(port_connect)) {
+        DTRACE_CHARBUF(process_str, DTRACE_TERM_BUF_SIZE);
+        DTRACE_CHARBUF(port_str, DTRACE_TERM_BUF_SIZE);
+        DTRACE_CHARBUF(newprocess_str, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_pid_str(prt->connected, process_str);
+        erts_snprintf(port_str, sizeof(port_str), "%T", prt->id);
+        dtrace_proc_str(rp, newprocess_str);
+        DTRACE4(port_connect, process_str, port_str, prt->name, newprocess_str);
+    }
+#endif
     BIF_RET(am_true);
 }
 
@@ -904,7 +927,16 @@ open_port(Process* p, Eterm name, Eterm settings, int *err_nump)
     erts_smp_proc_unlock(p, ERTS_PROC_LOCK_MAIN);
 
     port_num = erts_open_driver(driver, p->id, name_buf, &opts, err_nump);
+#ifdef USE_VM_PROBES
+    if (port_num >= 0 && DTRACE_ENABLED(port_open)) {
+        DTRACE_CHARBUF(process_str, DTRACE_TERM_BUF_SIZE);
+        DTRACE_CHARBUF(port_str, DTRACE_TERM_BUF_SIZE);
 
+        dtrace_proc_str(p, process_str);
+        erts_snprintf(port_str, sizeof(port_str), "%T", erts_port[port_num].id);
+        DTRACE3(port_open, process_str, name_buf, port_str);
+    }
+#endif
     erts_smp_proc_lock(p, ERTS_PROC_LOCK_MAIN);
 
     if (port_num < 0) {
diff --git a/erts/emulator/beam/erl_bif_timer.c b/erts/emulator/beam/erl_bif_timer.c
index a922a33da3..0002f8374f 100644
--- a/erts/emulator/beam/erl_bif_timer.c
+++ b/erts/emulator/beam/erl_bif_timer.c
@@ -373,7 +373,11 @@ bif_timer_timeout(ErtsBifTimer* btm)
 		message = TUPLE3(hp, am_timeout, ref, message);
 	    }
 
-	    erts_queue_message(rp, &rp_locks, bp, message, NIL);
+	    erts_queue_message(rp, &rp_locks, bp, message, NIL
+#ifdef USE_VM_PROBES
+			       , NIL
+#endif
+			       );
 	    erts_smp_proc_unlock(rp, rp_locks);
 	    erts_smp_proc_dec_refc(rp);
 	}
diff --git a/erts/emulator/beam/erl_bif_trace.c b/erts/emulator/beam/erl_bif_trace.c
index 1c99fcdaa6..2af037d146 100644
--- a/erts/emulator/beam/erl_bif_trace.c
+++ b/erts/emulator/beam/erl_bif_trace.c
@@ -1759,9 +1759,17 @@ Eterm erts_seq_trace(Process *p, Eterm arg1, Eterm arg2,
 	    return THE_NON_VALUE;
         }
 	if (build_result) {
+#ifdef USE_VM_PROBES
+	    old_value = (SEQ_TRACE_TOKEN(p) == am_have_dt_utag) ? NIL : SEQ_TRACE_TOKEN(p);
+#else
 	    old_value = SEQ_TRACE_TOKEN(p);
+#endif
 	}
+#ifdef USE_VM_PROBES
+        SEQ_TRACE_TOKEN(p) = (DT_UTAG(p) != NIL) ? am_have_dt_utag : NIL;
+#else
         SEQ_TRACE_TOKEN(p) = NIL;
+#endif
         return old_value;
     }
     else {
@@ -1774,7 +1782,11 @@ new_seq_trace_token(Process* p)
 {
     Eterm* hp;
 
-    if (SEQ_TRACE_TOKEN(p) == NIL) {
+    if (SEQ_TRACE_TOKEN(p) == NIL
+#ifdef USE_VM_PROBES
+	|| SEQ_TRACE_TOKEN(p) == am_have_dt_utag
+#endif
+	) {
 	hp = HAlloc(p, 6);
 	SEQ_TRACE_TOKEN(p) = TUPLE5(hp, make_small(0),		/* Flags  */ 
 				    make_small(0),		/* Label  */
@@ -1794,7 +1806,11 @@ BIF_RETTYPE erl_seq_trace_info(Process *p, Eterm item)
 	BIF_ERROR(p, BADARG);
     }
 
-    if (SEQ_TRACE_TOKEN(p) == NIL) {
+    if (SEQ_TRACE_TOKEN(p) == NIL
+#ifdef USE_VM_PROBES
+	|| SEQ_TRACE_TOKEN(p) == am_have_dt_utag
+#endif
+	) {
 	if ((item == am_send) || (item == am_receive) || 
 	    (item == am_print) || (item == am_timestamp)) {
 	    hp = HAlloc(p,3);
@@ -1851,8 +1867,13 @@ BIF_RETTYPE seq_trace_info_1(BIF_ALIST_1)
  */
 BIF_RETTYPE seq_trace_print_1(BIF_ALIST_1)    
 {
-    if (SEQ_TRACE_TOKEN(BIF_P) == NIL) 
+    if (SEQ_TRACE_TOKEN(BIF_P) == NIL 
+#ifdef USE_VM_PROBES
+	|| SEQ_TRACE_TOKEN(BIF_P) == am_have_dt_utag
+#endif
+	) {
 	BIF_RET(am_false);
+    }
     seq_trace_update_send(BIF_P);
     seq_trace_output(SEQ_TRACE_TOKEN(BIF_P), BIF_ARG_1, 
 		     SEQ_TRACE_PRINT, NIL, BIF_P);
@@ -1869,8 +1890,13 @@ BIF_RETTYPE seq_trace_print_1(BIF_ALIST_1)
  */
 BIF_RETTYPE seq_trace_print_2(BIF_ALIST_2)    
 {
-    if (SEQ_TRACE_TOKEN(BIF_P) == NIL) 
+    if (SEQ_TRACE_TOKEN(BIF_P) == NIL
+#ifdef USE_VM_PROBES
+	|| SEQ_TRACE_TOKEN(BIF_P) == am_have_dt_utag
+#endif
+	) {
 	BIF_RET(am_false);
+    }
     if (!(is_atom(BIF_ARG_1) || is_small(BIF_ARG_1))) {
 	BIF_ERROR(BIF_P, BADARG);
     }
diff --git a/erts/emulator/beam/erl_db_util.c b/erts/emulator/beam/erl_db_util.c
index 4821a7d9fb..be345e7c9b 100644
--- a/erts/emulator/beam/erl_db_util.c
+++ b/erts/emulator/beam/erl_db_util.c
@@ -2203,7 +2203,11 @@ restart:
 	    *esp++ = am_true;
 	    break;
 	case matchIsSeqTrace:
-	    if (SEQ_TRACE_TOKEN(c_p) != NIL)
+	    if (SEQ_TRACE_TOKEN(c_p) != NIL
+#ifdef USE_VM_PROBES
+		&& SEQ_TRACE_TOKEN(c_p) != am_have_dt_utag
+#endif
+		)
 		*esp++ = am_true;
 	    else
 		*esp++ = am_false;
@@ -2227,7 +2231,11 @@ restart:
 	    --esp;
 	    break;
 	case matchGetSeqToken:
-	    if (SEQ_TRACE_TOKEN(c_p) == NIL) 
+	    if (SEQ_TRACE_TOKEN(c_p) == NIL
+#ifdef USE_VM_PROBES
+		|| SEQ_TRACE_TOKEN(c_p) == am_have_dt_utag
+#endif
+		) 
 		*esp++ = NIL;
 	    else {
 		Eterm sender = SEQ_TRACE_TOKEN_SENDER(c_p);
diff --git a/erts/emulator/beam/erl_driver.h b/erts/emulator/beam/erl_driver.h
index 7510f6b724..1ae9a211d7 100644
--- a/erts/emulator/beam/erl_driver.h
+++ b/erts/emulator/beam/erl_driver.h
@@ -649,6 +649,8 @@ EXTERN int erl_drv_getenv(char *key, char *value, size_t *value_size);
 
 #endif
 
+/* also in global.h, but driver's can't include global.h */
+void dtrace_drvport_str(ErlDrvPort port, char *port_buf);
 
 
 
diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c
index bde87b8346..82f2dc6091 100644
--- a/erts/emulator/beam/erl_gc.c
+++ b/erts/emulator/beam/erl_gc.c
@@ -35,6 +35,7 @@
 #include "hipe_stack.h"
 #include "hipe_mode_switch.h"
 #endif
+#include "dtrace-wrapper.h"
 
 #define ERTS_INACT_WR_PB_LEAVE_MUCH_LIMIT 1
 #define ERTS_INACT_WR_PB_LEAVE_MUCH_PERCENTAGE 20
@@ -349,7 +350,9 @@ erts_garbage_collect(Process* p, int need, Eterm* objv, int nobj)
     Uint reclaimed_now = 0;
     int done = 0;
     Uint ms1, s1, us1;
-
+#ifdef USE_VM_PROBES
+    DTRACE_CHARBUF(pidbuf, DTRACE_TERM_BUF_SIZE);
+#endif
     if (IS_TRACED_FL(p, F_TRACE_GC)) {
         trace_gc(p, am_gc_start);
     }
@@ -369,15 +372,27 @@ erts_garbage_collect(Process* p, int need, Eterm* objv, int nobj)
     if (GEN_GCS(p) >= MAX_GEN_GCS(p)) {
         FLAGS(p) |= F_NEED_FULLSWEEP;
     }
-
+#ifdef USE_VM_PROBES
+    *pidbuf = '\0';
+    if (DTRACE_ENABLED(gc_major_start)
+        || DTRACE_ENABLED(gc_major_end)
+        || DTRACE_ENABLED(gc_minor_start)
+        || DTRACE_ENABLED(gc_minor_end)) {
+        dtrace_proc_str(p, pidbuf);
+    }
+#endif
     /*
      * Test which type of GC to do.
      */
     while (!done) {
 	if ((FLAGS(p) & F_NEED_FULLSWEEP) != 0) {
+	    DTRACE2(gc_major_start, pidbuf, need);
 	    done = major_collection(p, need, objv, nobj, &reclaimed_now);
+	    DTRACE2(gc_major_end, pidbuf, reclaimed_now);
 	} else {
+	    DTRACE2(gc_minor_start, pidbuf, need);
 	    done = minor_collection(p, need, objv, nobj, &reclaimed_now);
+	    DTRACE2(gc_minor_end, pidbuf, reclaimed_now);
 	}
     }
 
@@ -1118,6 +1133,15 @@ do_minor(Process *p, Uint new_sz, Eterm* objv, int nobj)
     sys_memcpy(n_heap + new_sz - n, p->stop, n * sizeof(Eterm));
     p->stop = n_heap + new_sz - n;
 
+#ifdef USE_VM_PROBES
+    if (HEAP_SIZE(p) != new_sz && DTRACE_ENABLED(process_heap_grow)) {
+        DTRACE_CHARBUF(pidbuf, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_proc_str(p, pidbuf);
+        DTRACE3(process_heap_grow, pidbuf, HEAP_SIZE(p), new_sz);
+    }
+#endif
+
     ERTS_HEAP_FREE(ERTS_ALC_T_HEAP,
 		   (void*)HEAP_START(p),
 		   HEAP_SIZE(p) * sizeof(Eterm));
@@ -1339,6 +1363,15 @@ major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl)
     sys_memcpy(n_heap + new_sz - n, p->stop, n * sizeof(Eterm));
     p->stop = n_heap + new_sz - n;
 
+#ifdef USE_VM_PROBES
+    if (HEAP_SIZE(p) != new_sz && DTRACE_ENABLED(process_heap_grow)) {
+        DTRACE_CHARBUF(pidbuf, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_proc_str(p, pidbuf);
+        DTRACE3(process_heap_grow, pidbuf, HEAP_SIZE(p), new_sz);
+    }
+#endif
+
     ERTS_HEAP_FREE(ERTS_ALC_T_HEAP,
 		   (void *) HEAP_START(p),
 		   (HEAP_END(p) - HEAP_START(p)) * sizeof(Eterm));
@@ -1907,7 +1940,13 @@ setup_rootset(Process *p, Eterm *objv, int nobj, Rootset *rootset)
 	roots[n].sz = 1;
 	n++;
     }
-
+#ifdef USE_VM_PROBES
+    if (is_not_immed(p->dt_utag)) {
+	roots[n].v = &p->dt_utag;
+	roots[n].sz = 1;
+	n++;
+    }
+#endif
     ASSERT(is_nil(p->tracer_proc) ||
 	   is_internal_pid(p->tracer_proc) ||
 	   is_internal_port(p->tracer_proc));
@@ -2009,6 +2048,16 @@ grow_new_heap(Process *p, Uint new_sz, Eterm* objv, int nobj)
         HEAP_TOP(p) = new_heap + heap_size;
         HEAP_START(p) = new_heap;
     }
+
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(process_heap_grow)) {
+	DTRACE_CHARBUF(pidbuf, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_proc_str(p, pidbuf);
+	DTRACE3(process_heap_grow, pidbuf, HEAP_SIZE(p), new_sz);
+    }
+#endif
+
     HEAP_SIZE(p) = new_sz;
 }
 
@@ -2018,7 +2067,6 @@ shrink_new_heap(Process *p, Uint new_sz, Eterm *objv, int nobj)
     Eterm* new_heap;
     Uint heap_size = HEAP_TOP(p) - HEAP_START(p);
     Sint offs;
-
     Uint stack_size = p->hend - p->stop;
 
     ASSERT(new_sz < p->heap_sz);
@@ -2047,6 +2095,16 @@ shrink_new_heap(Process *p, Uint new_sz, Eterm *objv, int nobj)
         HEAP_TOP(p) = new_heap + heap_size;
         HEAP_START(p) = new_heap;
     }
+
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(process_heap_shrink)) {
+	DTRACE_CHARBUF(pidbuf, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_proc_str(p, pidbuf);
+	DTRACE3(process_heap_shrink, pidbuf, HEAP_SIZE(p), new_sz);
+    }
+#endif
+
     HEAP_SIZE(p) = new_sz;
 }
 
@@ -2429,6 +2487,13 @@ offset_mqueue(Process *p, Sint offs, char* area, Uint area_size)
 	if (is_boxed(mesg) && in_area(ptr_val(mesg), area, area_size)) {
 	    ERL_MESSAGE_TOKEN(mp) = offset_ptr(mesg, offs);
         }
+#ifdef USE_VM_PROBES
+	mesg = ERL_MESSAGE_DT_UTAG(mp);
+	if (is_boxed(mesg) && in_area(ptr_val(mesg), area, area_size)) {
+	    ERL_MESSAGE_DT_UTAG(mp) = offset_ptr(mesg, offs);
+        }
+#endif	
+	
         ASSERT((is_nil(ERL_MESSAGE_TOKEN(mp)) ||
 		is_tuple(ERL_MESSAGE_TOKEN(mp)) ||
 		is_atom(ERL_MESSAGE_TOKEN(mp))));
@@ -2448,6 +2513,9 @@ offset_one_rootset(Process *p, Sint offs, char* area, Uint area_size,
     offset_heap_ptr(&p->fvalue, 1, offs, area, area_size);
     offset_heap_ptr(&p->ftrace, 1, offs, area, area_size);
     offset_heap_ptr(&p->seq_trace_token, 1, offs, area, area_size);
+#ifdef USE_VM_PROBES
+    offset_heap_ptr(&p->dt_utag, 1, offs, area, area_size);
+#endif
     offset_heap_ptr(&p->group_leader, 1, offs, area, area_size);
     offset_mqueue(p, offs, area, area_size);
     offset_heap_ptr(p->stop, (STACK_START(p) - p->stop), offs, area, area_size);
diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c
index 8c7e1f36ee..e9f2e94dc6 100644
--- a/erts/emulator/beam/erl_lock_check.c
+++ b/erts/emulator/beam/erl_lock_check.c
@@ -185,6 +185,9 @@ static erts_lc_lock_order_t erts_lock_order[] = {
     {   "save_ops_lock",                        NULL                    },
 #endif
 #endif
+#ifdef	USE_VM_PROBES
+    {   "efile_drv dtrace mutex",               NULL                    },
+#endif
     {	"mtrace_buf",				NULL			},
     {	"erts_alloc_hard_debug",		NULL			}
 };
diff --git a/erts/emulator/beam/erl_message.c b/erts/emulator/beam/erl_message.c
index ab1ab7b1ea..4cdf2d7d09 100644
--- a/erts/emulator/beam/erl_message.c
+++ b/erts/emulator/beam/erl_message.c
@@ -31,6 +31,7 @@
 #include "erl_process.h"
 #include "erl_nmgc.h"
 #include "erl_binary.h"
+#include "dtrace-wrapper.h"
 
 ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(message,
 				 ErlMessage,
@@ -335,6 +336,11 @@ erts_queue_dist_message(Process *rcvr,
 			Eterm token)
 {
     ErlMessage* mp;
+#ifdef USE_VM_PROBES
+    Sint tok_label = 0;
+    Sint tok_lastcnt = 0;
+    Sint tok_serial = 0;
+#endif
 #ifdef ERTS_SMP
     ErtsProcLocks need_locks;
 #endif
@@ -376,15 +382,61 @@ erts_queue_dist_message(Process *rcvr,
 	message_free(mp);
 	msg = erts_msg_distext2heap(rcvr, rcvr_locks, &mbuf, &token, dist_ext);
 	if (is_value(msg))
-	    erts_queue_message(rcvr, rcvr_locks, mbuf, msg, token);
+#ifdef USE_VM_PROBES
+            if (DTRACE_ENABLED(message_queued)) {
+                DTRACE_CHARBUF(receiver_name, DTRACE_TERM_BUF_SIZE);
+
+                dtrace_proc_str(rcvr, receiver_name);
+                if (token != NIL && token != am_have_dt_utag) {
+                    tok_label = signed_val(SEQ_TRACE_T_LABEL(token));
+                    tok_lastcnt = signed_val(SEQ_TRACE_T_LASTCNT(token));
+                    tok_serial = signed_val(SEQ_TRACE_T_SERIAL(token));
+                }
+                DTRACE6(message_queued,
+                        receiver_name, size_object(msg), rcvr->msg.len,
+                        tok_label, tok_lastcnt, tok_serial);
+            }
+#endif
+	    erts_queue_message(rcvr, rcvr_locks, mbuf, msg, token
+#ifdef USE_VM_PROBES
+			       , NIL
+#endif
+			       );
     }
     else {
 	/* Enqueue message on external format */
 
 	ERL_MESSAGE_TERM(mp) = THE_NON_VALUE;
-	ERL_MESSAGE_TOKEN(mp) = token;
+#ifdef USE_VM_PROBES
+	ERL_MESSAGE_DT_UTAG(mp) = NIL;
+	if (token == am_have_dt_utag) {
+	    ERL_MESSAGE_TOKEN(mp) = NIL;
+	} else {
+#endif
+	    ERL_MESSAGE_TOKEN(mp) = token;
+#ifdef USE_VM_PROBES
+	}
+#endif
 	mp->next = NULL;
 
+#ifdef USE_VM_PROBES
+        if (DTRACE_ENABLED(message_queued)) {
+            DTRACE_CHARBUF(receiver_name, DTRACE_TERM_BUF_SIZE);
+
+            dtrace_proc_str(rcvr, receiver_name);
+            if (token != NIL && token != am_have_dt_utag) {
+                tok_label = signed_val(SEQ_TRACE_T_LABEL(token));
+                tok_lastcnt = signed_val(SEQ_TRACE_T_LASTCNT(token));
+                tok_serial = signed_val(SEQ_TRACE_T_SERIAL(token));
+            }
+            /*
+             * TODO: We don't know the real size of the external message here.
+             *       -1 will appear to a D script as 4294967295.
+             */
+            DTRACE6(message_queued, receiver_name, -1, rcvr->msg.len + 1,
+                    tok_label, tok_lastcnt, tok_serial);
+        }
+#endif
 	mp->data.dist_ext = dist_ext;
 	LINK_MESSAGE(rcvr, mp);
 
@@ -398,7 +450,11 @@ erts_queue_message(Process* receiver,
 		   ErtsProcLocks *receiver_locks,
 		   ErlHeapFragment* bp,
 		   Eterm message,
-		   Eterm seq_trace_token)
+		   Eterm seq_trace_token
+#ifdef USE_VM_PROBES
+		   , Eterm dt_utag
+#endif
+)
 {
     ErlMessage* mp;
 #ifdef ERTS_SMP
@@ -439,6 +495,9 @@ erts_queue_message(Process* receiver,
 
     ERL_MESSAGE_TERM(mp) = message;
     ERL_MESSAGE_TOKEN(mp) = seq_trace_token;
+#ifdef USE_VM_PROBES
+    ERL_MESSAGE_DT_UTAG(mp) = dt_utag;
+#endif
     mp->next = NULL;
     mp->data.heap_frag = bp;
 
@@ -462,12 +521,30 @@ erts_queue_message(Process* receiver,
     LINK_MESSAGE(receiver, mp);
 #endif
 
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(message_queued)) {
+        DTRACE_CHARBUF(receiver_name, DTRACE_TERM_BUF_SIZE);
+        Sint tok_label = 0;
+        Sint tok_lastcnt = 0;
+        Sint tok_serial = 0;
+
+        dtrace_proc_str(receiver, receiver_name);
+        if (seq_trace_token != NIL && is_tuple(seq_trace_token)) {
+            tok_label = signed_val(SEQ_TRACE_T_LABEL(seq_trace_token));
+            tok_lastcnt = signed_val(SEQ_TRACE_T_LASTCNT(seq_trace_token));
+            tok_serial = signed_val(SEQ_TRACE_T_SERIAL(seq_trace_token));
+        }
+        DTRACE6(message_queued,
+                receiver_name, size_object(message), receiver->msg.len,
+                tok_label, tok_lastcnt, tok_serial);
+    }
+#endif
     notify_new_message(receiver);
 
     if (IS_TRACED_FL(receiver, F_TRACE_RECEIVE)) {
 	trace_receive(receiver, message);
     }
-    
+
 #ifndef ERTS_SMP
     ERTS_HOLE_CHECK(receiver);
 #endif
@@ -497,6 +574,9 @@ erts_move_msg_mbuf_to_heap(Eterm** hpp, ErlOffHeap* off_heap, ErlMessage *msg)
     Sint offs;
     Uint sz;
     ErlHeapFragment *bp;
+#ifdef USE_VM_PROBES
+    Eterm utag;
+#endif
 
 #ifdef HARD_DEBUG
     ProcBin *dbg_mso_start = off_heap->mso;
@@ -506,32 +586,56 @@ erts_move_msg_mbuf_to_heap(Eterm** hpp, ErlOffHeap* off_heap, ErlMessage *msg)
     ErlHeapFragment *dbg_bp;
     Uint *dbg_hp, *dbg_thp_start;
     Uint dbg_term_sz, dbg_token_sz;
+#ifdef USE_VM_PROBES
+    Eterm dbg_utag;
+    Uint dbg_utag_sz;
+#endif
 #endif
 
     bp = msg->data.heap_frag;
     term = ERL_MESSAGE_TERM(msg);
     token = ERL_MESSAGE_TOKEN(msg);
+#ifdef USE_VM_PROBES
+    utag = ERL_MESSAGE_DT_UTAG(msg);
+#endif
     if (!bp) {
+#ifdef USE_VM_PROBES
+	ASSERT(is_immed(term) && is_immed(token) && is_immed(utag));
+#else
 	ASSERT(is_immed(term) && is_immed(token));
+#endif
 	return;
     }
 
 #ifdef HARD_DEBUG
     dbg_term_sz = size_object(term);
     dbg_token_sz = size_object(token);
+    dbg_bp = new_message_buffer(dbg_term_sz + dbg_token_sz);
+#ifdef USE_VM_PROBES
+    dbg_utag_sz = size_object(utag);
+    dbg_bp = new_message_buffer(dbg_term_sz + dbg_token_sz + dbg_utag_sz );
+#endif
     /*ASSERT(dbg_term_sz + dbg_token_sz == erts_msg_used_frag_sz(msg));
       Copied size may be smaller due to removed SubBins's or garbage.
       Copied size may be larger due to duplicated shared terms.
     */
-    dbg_bp = new_message_buffer(dbg_term_sz + dbg_token_sz);
     dbg_hp = dbg_bp->mem;
     dbg_term = copy_struct(term, dbg_term_sz, &dbg_hp, &dbg_bp->off_heap);
     dbg_token = copy_struct(token, dbg_token_sz, &dbg_hp, &dbg_bp->off_heap);
-    dbg_thp_start = *hpp;
+#ifdef USE_VM_PROBES
+    dbg_utag = copy_struct(utag, dbg_utag_sz, &dbg_hp, &dbg_bp->off_heap);
+#endif
+   dbg_thp_start = *hpp;
 #endif
 
     if (bp->next != NULL) {
-	move_multi_frags(hpp, off_heap, bp, msg->m, 2);
+	move_multi_frags(hpp, off_heap, bp, msg->m, 
+#ifdef USE_VM_PROBES
+			 3
+#else
+			 2
+#endif
+			 );
 	goto copy_done;
     }
 
@@ -633,6 +737,16 @@ erts_move_msg_mbuf_to_heap(Eterm** hpp, ErlOffHeap* off_heap, ErlMessage *msg)
 	ASSERT(hp > ptr_val(ERL_MESSAGE_TERM(msg)));
 #endif
     }
+#ifdef USE_VM_PROBES
+    if (is_not_immed(utag)) {
+	ASSERT(in_heapfrag(ptr_val(utag), bp));
+	ERL_MESSAGE_DT_UTAG(msg) = offset_ptr(utag, offs);
+#ifdef HARD_DEBUG
+	ASSERT(dbg_thp_start <= ptr_val(ERL_MESSAGE_DT_UTAG(msg)));
+	ASSERT(hp > ptr_val(ERL_MESSAGE_DT_UTAG(msg)));
+#endif
+    }
+#endif
 
 copy_done:
 
@@ -699,6 +813,9 @@ copy_done:
 #ifdef HARD_DEBUG
     ASSERT(eq(ERL_MESSAGE_TERM(msg), dbg_term));
     ASSERT(eq(ERL_MESSAGE_TOKEN(msg), dbg_token));
+#ifdef USE_VM_PROBES
+    ASSERT(eq(ERL_MESSAGE_DT_UTAG(msg), dbg_utag));
+#endif
     free_message_buffer(dbg_bp);
 #endif
 
@@ -774,39 +891,101 @@ erts_send_message(Process* sender,
     Uint msize;
     ErlHeapFragment* bp = NULL;
     Eterm token = NIL;
-
+#ifdef USE_VM_PROBES
+    DTRACE_CHARBUF(sender_name, 64);
+    DTRACE_CHARBUF(receiver_name, 64);
+    Sint tok_label = 0;
+    Sint tok_lastcnt = 0;
+    Sint tok_serial = 0;
+#endif
     BM_STOP_TIMER(system);
     BM_MESSAGE(message,sender,receiver);
     BM_START_TIMER(send);
 
+ #ifdef USE_VM_PROBES
+    *sender_name = *receiver_name = '\0';
+   if (DTRACE_ENABLED(message_send)) {
+        erts_snprintf(sender_name, sizeof(sender_name), "%T", sender->id);
+        erts_snprintf(receiver_name, sizeof(receiver_name), "%T", receiver->id);
+    }
+#endif
     if (SEQ_TRACE_TOKEN(sender) != NIL && !(flags & ERTS_SND_FLG_NO_SEQ_TRACE)) {
         Eterm* hp;
+	Eterm stoken = SEQ_TRACE_TOKEN(sender);
+	Uint seq_trace_size = 0;
+#ifdef USE_VM_PROBES
+	Uint dt_utag_size = 0;
+	Eterm utag = NIL;
+#endif
 
-        BM_SWAP_TIMER(send,size);
+	BM_SWAP_TIMER(send,size);
 	msize = size_object(message);
-        BM_SWAP_TIMER(size,send);
+	BM_SWAP_TIMER(size,send);
+
+#ifdef USE_VM_PROBES
+	if (stoken != am_have_dt_utag) {
+#endif
+
+	    seq_trace_update_send(sender);
+	    seq_trace_output(stoken, message, SEQ_TRACE_SEND, 
+			     receiver->id, sender);
+	    seq_trace_size = 6; /* TUPLE5 */
+#ifdef USE_VM_PROBES
+	}
+	if (DT_UTAG_FLAGS(sender) & DT_UTAG_SPREADING) {
+	    dt_utag_size = size_object(DT_UTAG(sender));
+	} else if (stoken == am_have_dt_utag ) {
+	    stoken = NIL;
+	}
+#endif
 
-	seq_trace_update_send(sender);
-	seq_trace_output(SEQ_TRACE_TOKEN(sender), message, SEQ_TRACE_SEND, 
-			 receiver->id, sender);
-	bp = new_message_buffer(msize + 6 /* TUPLE5 */);
+	bp = new_message_buffer(msize + seq_trace_size 
+#ifdef USE_VM_PROBES
+				+ dt_utag_size
+#endif
+				);
 	hp = bp->mem;
 
         BM_SWAP_TIMER(send,copy);
-	token = copy_struct(SEQ_TRACE_TOKEN(sender),
-			    6 /* TUPLE5 */,
+	token = copy_struct(stoken,
+			    seq_trace_size,
 			    &hp,
 			    &bp->off_heap);
 
 	message = copy_struct(message, msize, &hp, &bp->off_heap);
+#ifdef USE_VM_PROBES
+	if (DT_UTAG_FLAGS(sender) & DT_UTAG_SPREADING) {
+	    utag = copy_struct(DT_UTAG(sender), dt_utag_size, &hp, &bp->off_heap);
+#ifdef DTRACE_TAG_HARDDEBUG
+	    erts_fprintf(stderr,
+			 "Dtrace -> (%T) Spreading tag (%T) with "
+			 "message %T!\r\n",sender->id, utag, message);
+#endif
+	}
+#endif
         BM_MESSAGE_COPIED(msize);
         BM_SWAP_TIMER(copy,send);
 
+#ifdef USE_VM_PROBES
+        if (DTRACE_ENABLED(message_send)) {
+	    if (stoken != NIL && stoken != am_have_dt_utag) {
+		tok_label = signed_val(SEQ_TRACE_T_LABEL(stoken));
+		tok_lastcnt = signed_val(SEQ_TRACE_T_LASTCNT(stoken));
+		tok_serial = signed_val(SEQ_TRACE_T_SERIAL(stoken));
+	    }
+	    DTRACE6(message_send, sender_name, receiver_name,
+		    msize, tok_label, tok_lastcnt, tok_serial);
+        }
+#endif
         erts_queue_message(receiver,
 			   receiver_locks,
 			   bp,
 			   message,
-			   token);
+			   token
+#ifdef USE_VM_PROBES
+			   , utag
+#endif
+			   );
         BM_SWAP_TIMER(send,system);
 #ifdef HYBRID
     } else {
@@ -835,8 +1014,13 @@ erts_send_message(Process* sender,
 #endif
         LAZY_COPY(sender,message);
         BM_SWAP_TIMER(copy,send);
+        DTRACE6(message_send, sender_name, receiver_name,
+                size_object(message)msize, tok_label, tok_lastcnt, tok_serial);
         ERL_MESSAGE_TERM(mp) = message;
         ERL_MESSAGE_TOKEN(mp) = NIL;
+#ifdef USE_VM_PROBES
+	ERL_MESSAGE_DT_UTAG(mp) = NIL;
+#endif
         mp->next = NULL;
 	LINK_MESSAGE(receiver, mp);
         ACTIVATE(receiver);
@@ -874,9 +1058,14 @@ erts_send_message(Process* sender,
 	{
 	    ErlMessage* mp = message_alloc();
 
+            DTRACE6(message_send, sender_name, receiver_name,
+                    size_object(message), tok_label, tok_lastcnt, tok_serial);
 	    mp->data.attached = NULL;
 	    ERL_MESSAGE_TERM(mp) = message;
 	    ERL_MESSAGE_TOKEN(mp) = NIL;
+#ifdef USE_VM_PROBES
+	    ERL_MESSAGE_DT_UTAG(mp) = NIL;
+#endif
 	    mp->next = NULL;
 	    /*
 	     * We move 'in queue' to 'private queue' and place
@@ -908,7 +1097,13 @@ erts_send_message(Process* sender,
 	message = copy_struct(message, msize, &hp, ohp);
 	BM_MESSAGE_COPIED(msz);
 	BM_SWAP_TIMER(copy,send);
-	erts_queue_message(receiver, receiver_locks, bp, message, token);
+        DTRACE6(message_send, sender_name, receiver_name,
+                msize, tok_label, tok_lastcnt, tok_serial);
+	erts_queue_message(receiver, receiver_locks, bp, message, token
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+			   );
         BM_SWAP_TIMER(send,system);
 #else
 	ErlMessage* mp = message_alloc();
@@ -928,8 +1123,13 @@ erts_send_message(Process* sender,
 	message = copy_struct(message, msize, &hp, &receiver->off_heap);
 	BM_MESSAGE_COPIED(msize);
         BM_SWAP_TIMER(copy,send);
+        DTRACE6(message_send, sender_name, receiver_name,
+                (uint32_t)msize, tok_label, tok_lastcnt, tok_serial);
 	ERL_MESSAGE_TERM(mp) = message;
 	ERL_MESSAGE_TOKEN(mp) = NIL;
+#ifdef USE_VM_PROBES
+	ERL_MESSAGE_DT_UTAG(mp) = NIL;
+#endif
 	mp->next = NULL;
 	mp->data.attached = NULL;
 	LINK_MESSAGE(receiver, mp);
@@ -968,7 +1168,11 @@ erts_deliver_exit_message(Eterm from, Process *to, ErtsProcLocks *to_locksp,
     Eterm temptoken;
     ErlHeapFragment* bp = NULL;
 
-    if (token != NIL) {
+    if (token != NIL
+#ifdef USE_VM_PROBES
+	&& token != am_have_dt_utag
+#endif
+	) {
 
 	ASSERT(is_tuple(token));
 	sz_reason = size_object(reason);
@@ -983,7 +1187,11 @@ erts_deliver_exit_message(Eterm from, Process *to, ErtsProcLocks *to_locksp,
 	/* the trace token must in this case be updated by the caller */
 	seq_trace_output(token, save, SEQ_TRACE_SEND, to->id, NULL);
 	temptoken = copy_struct(token, sz_token, &hp, &bp->off_heap);
-	erts_queue_message(to, to_locksp, bp, save, temptoken);
+	erts_queue_message(to, to_locksp, bp, save, temptoken
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+			   );
     } else {
 	ErlOffHeap *ohp;
 	sz_reason = size_object(reason);
@@ -1000,7 +1208,11 @@ erts_deliver_exit_message(Eterm from, Process *to, ErtsProcLocks *to_locksp,
 		     ? from
 		     : copy_struct(from, sz_from, &hp, ohp));
 	save = TUPLE3(hp, am_EXIT, from_copy, mess);
-	erts_queue_message(to, to_locksp, bp, save, NIL);
+	erts_queue_message(to, to_locksp, bp, save, NIL
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+			   );
     }
 }
 
diff --git a/erts/emulator/beam/erl_message.h b/erts/emulator/beam/erl_message.h
index 5aca0db6fe..7678c7c753 100644
--- a/erts/emulator/beam/erl_message.h
+++ b/erts/emulator/beam/erl_message.h
@@ -70,11 +70,18 @@ typedef struct erl_mesg {
 	ErlHeapFragment *heap_frag;
 	void *attached;
     } data;
+#ifdef USE_VM_PROBES
+    Eterm m[3];                 /* m[0] = message, m[1] = seq trace token, m[3] = dynamic trace user tag */
+#else
     Eterm m[2];			/* m[0] = message, m[1] = seq trace token */
+#endif
 } ErlMessage;
 
 #define ERL_MESSAGE_TERM(mp) ((mp)->m[0])
 #define ERL_MESSAGE_TOKEN(mp) ((mp)->m[1])
+#ifdef USE_VM_PROBES
+#define ERL_MESSAGE_DT_UTAG(mp) ((mp)->m[2])
+#endif
 
 /* Size of default message buffer (erl_message.c) */
 #define ERL_MESSAGE_BUF_SZ 500
@@ -221,7 +228,11 @@ ErlHeapFragment* erts_resize_message_buffer(ErlHeapFragment *, Uint,
 					    Eterm *, Uint);
 void free_message_buffer(ErlHeapFragment *);
 void erts_queue_dist_message(Process*, ErtsProcLocks*, ErtsDistExternal *, Eterm);
-void erts_queue_message(Process*, ErtsProcLocks*, ErlHeapFragment*, Eterm, Eterm);
+void erts_queue_message(Process*, ErtsProcLocks*, ErlHeapFragment*, Eterm, Eterm
+#ifdef USE_VM_PROBES
+		   , Eterm dt_utag
+#endif
+);
 void erts_deliver_exit_message(Eterm, Process*, ErtsProcLocks *, Eterm, Eterm);
 void erts_send_message(Process*, Process*, ErtsProcLocks*, Eterm, unsigned);
 void erts_link_mbuf_to_proc(Process *proc, ErlHeapFragment *bp);
diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c
index bdbeb3bcf6..0b1e87df54 100644
--- a/erts/emulator/beam/erl_nif.c
+++ b/erts/emulator/beam/erl_nif.c
@@ -66,6 +66,9 @@ static void add_readonly_check(ErlNifEnv*, unsigned char* ptr, unsigned sz);
 static int is_offheap(const ErlOffHeap* off_heap);
 #endif
 
+#ifdef USE_VM_PROBES
+void dtrace_nifenv_str(ErlNifEnv *, char *);
+#endif
 
 #define MIN_HEAP_FRAG_SZ 200
 static Eterm* alloc_heap_heavy(ErlNifEnv* env, unsigned need, Eterm* hp);
@@ -350,7 +353,11 @@ int enif_send(ErlNifEnv* env, const ErlNifPid* to_pid,
     if (flush_me) {	
 	flush_env(env); /* Needed for ERTS_HOLE_CHECK */ 
     }
-    erts_queue_message(rp, &rp_locks, frags, msg, am_undefined);
+    erts_queue_message(rp, &rp_locks, frags, msg, am_undefined
+#ifdef USE_VM_PROBES
+		       , NIL
+#endif
+		       );
     if (rp_locks) {	
 	ERTS_SMP_LC_ASSERT(rp_locks == (rp_had_locks | (ERTS_PROC_LOCK_MSGQ | 
 							ERTS_PROC_LOCK_STATUS)));
@@ -1787,6 +1794,13 @@ void erl_nif_init()
     resource_type_list.name = THE_NON_VALUE;
 }
 
+#ifdef USE_VM_PROBES
+void dtrace_nifenv_str(ErlNifEnv *env, char *process_buf)
+{
+    dtrace_pid_str(env->proc->id, process_buf);
+}
+#endif
+
 #ifdef READONLY_CHECK
 /* Use checksums to assert that NIFs do not write into inspected binaries
 */
diff --git a/erts/emulator/beam/erl_node_tables.c b/erts/emulator/beam/erl_node_tables.c
index 908ba755ed..1481f66b55 100644
--- a/erts/emulator/beam/erl_node_tables.c
+++ b/erts/emulator/beam/erl_node_tables.c
@@ -27,6 +27,7 @@
 #include "big.h"
 #include "error.h"
 #include "erl_thr_progress.h"
+#include "dtrace-wrapper.h"
 
 Hash erts_dist_table;
 Hash erts_node_table;
@@ -42,6 +43,8 @@ Sint erts_no_of_not_connected_dist_entries;
 
 DistEntry *erts_this_dist_entry;
 ErlNode *erts_this_node;
+char erts_this_node_sysname_BUFFER[256],
+    *erts_this_node_sysname = "uninitialized yet";
 
 static Uint node_entries;
 static Uint dist_entries;
@@ -702,6 +705,9 @@ erts_set_this_node(Eterm sysname, Uint creation)
     (void) hash_erase(&erts_node_table, (void *) erts_this_node);
     erts_this_node->sysname = sysname;
     erts_this_node->creation = creation;
+    erts_this_node_sysname = erts_this_node_sysname_BUFFER;
+    erts_snprintf(erts_this_node_sysname, sizeof(erts_this_node_sysname),
+                  "%T", sysname);
     (void) hash_put(&erts_node_table, (void *) erts_this_node);
 
     erts_smp_rwmtx_rwunlock(&erts_dist_table_rwmtx);
@@ -789,6 +795,9 @@ void erts_init_node_tables(void)
     erts_this_node->sysname			= am_Noname;
     erts_this_node->creation			= 0;
     erts_this_node->dist_entry			= erts_this_dist_entry;
+    erts_this_node_sysname = erts_this_node_sysname_BUFFER;
+    erts_snprintf(erts_this_node_sysname, sizeof(erts_this_node_sysname),
+                  "%T", erts_this_node->sysname);
 
     (void) hash_put(&erts_node_table, (void *) erts_this_node);
 
diff --git a/erts/emulator/beam/erl_node_tables.h b/erts/emulator/beam/erl_node_tables.h
index b0a63ae035..5cfd0ac641 100644
--- a/erts/emulator/beam/erl_node_tables.h
+++ b/erts/emulator/beam/erl_node_tables.h
@@ -169,6 +169,7 @@ extern Sint erts_no_of_not_connected_dist_entries;
 
 extern DistEntry *erts_this_dist_entry;
 extern ErlNode *erts_this_node;
+extern char *erts_this_node_sysname; /* must match erl_node_tables.c */
 
 DistEntry *erts_channel_no_to_dist_entry(Uint);
 DistEntry *erts_sysname_to_connected_dist_entry(Eterm);
diff --git a/erts/emulator/beam/erl_port_task.c b/erts/emulator/beam/erl_port_task.c
index a2b08fcf56..a8cb4563d6 100644
--- a/erts/emulator/beam/erl_port_task.c
+++ b/erts/emulator/beam/erl_port_task.c
@@ -32,6 +32,7 @@
 #include "global.h"
 #include "erl_port_task.h"
 #include "dist.h"
+#include "dtrace-wrapper.h"
 
 #if defined(DEBUG) && 0
 #define HARD_DEBUG
@@ -61,6 +62,20 @@ do {					\
     (P)->sched.next = NULL;		\
 } while (0)
 
+#ifdef USE_VM_PROBES
+#define DTRACE_DRIVER(PROBE_NAME, PP)                              \
+    if (DTRACE_ENABLED(driver_ready_input)) {                      \
+        DTRACE_CHARBUF(process_str, DTRACE_TERM_BUF_SIZE);         \
+        DTRACE_CHARBUF(port_str, DTRACE_TERM_BUF_SIZE);            \
+                                                                   \
+        dtrace_pid_str(PP->connected, process_str);                \
+        dtrace_port_str(PP, port_str);                             \
+        DTRACE3(PROBE_NAME, process_str, port_str, PP->name);      \
+    }
+#else
+#define  DTRACE_DRIVER(PROBE_NAME, PP) do {} while(0)
+#endif
+
 erts_smp_atomic_t erts_port_task_outstanding_io_tasks;
 
 struct ErtsPortTaskQueue_ {
@@ -823,12 +838,15 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp)
 	    goto tasks_done;
 	case ERTS_PORT_TASK_TIMEOUT:
 	    reds += ERTS_PORT_REDS_TIMEOUT;
-	    if (!(pp->status & ERTS_PORT_SFLGS_DEAD))
+	    if (!(pp->status & ERTS_PORT_SFLGS_DEAD)) {
+                DTRACE_DRIVER(driver_timeout, pp);
 		(*pp->drv_ptr->timeout)((ErlDrvData) pp->drv_data);
+            }
 	    break;
 	case ERTS_PORT_TASK_INPUT:
 	    reds += ERTS_PORT_REDS_INPUT;
 	    ASSERT((pp->status & ERTS_PORT_SFLGS_DEAD) == 0);
+            DTRACE_DRIVER(driver_ready_input, pp);
 	    /* NOTE some windows drivers use ->ready_input for input and output */
 	    (*pp->drv_ptr->ready_input)((ErlDrvData) pp->drv_data, ptp->event);
 	    io_tasks_executed++;
@@ -836,12 +854,14 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp)
 	case ERTS_PORT_TASK_OUTPUT:
 	    reds += ERTS_PORT_REDS_OUTPUT;
 	    ASSERT((pp->status & ERTS_PORT_SFLGS_DEAD) == 0);
+            DTRACE_DRIVER(driver_ready_output, pp);
 	    (*pp->drv_ptr->ready_output)((ErlDrvData) pp->drv_data, ptp->event);
 	    io_tasks_executed++;
 	    break;
 	case ERTS_PORT_TASK_EVENT:
 	    reds += ERTS_PORT_REDS_EVENT;
 	    ASSERT((pp->status & ERTS_PORT_SFLGS_DEAD) == 0);
+            DTRACE_DRIVER(driver_event, pp);
 	    (*pp->drv_ptr->event)((ErlDrvData) pp->drv_data, ptp->event, ptp->event_data);
 	    io_tasks_executed++;
 	    break;
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index ec85dd487a..7942a05153 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -42,6 +42,7 @@
 #include "erl_thr_progress.h"
 #include "erl_thr_queue.h"
 #include "erl_async.h"
+#include "dtrace-wrapper.h"
 
 #define ERTS_RUNQ_CHECK_BALANCE_REDS_PER_SCHED (2000*CONTEXT_REDS)
 #define ERTS_RUNQ_CALL_CHECK_BALANCE_REDS \
@@ -681,7 +682,11 @@ reply_sched_wall_time(void *vswtrp)
 	hpp = &hp;
     }
 
-    erts_queue_message(rp, &rp_locks, bp, msg, NIL);
+    erts_queue_message(rp, &rp_locks, bp, msg, NIL
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+		       );
 
     if (swtrp->req_sched == esdp->no)
 	rp_locks &= ~ERTS_PROC_LOCK_MAIN;
@@ -6260,6 +6265,15 @@ Process *schedule(Process *p, int calls)
     int actual_reds;
     int reds;
 
+#ifdef USE_VM_PROBES
+    if (p != NULL && DTRACE_ENABLED(process_unscheduled)) {
+        DTRACE_CHARBUF(process_buf, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_proc_str(p, process_buf);
+        DTRACE1(process_unscheduled, process_buf);
+    }
+#endif
+
     if (ERTS_USE_MODIFIED_TIMING()) {
 	context_reds = ERTS_MODIFIED_TIMING_CONTEXT_REDS;
 	input_reductions = ERTS_MODIFIED_TIMING_INPUT_REDS;
@@ -7298,6 +7312,10 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
     p->seq_trace_lastcnt = 0;
     p->seq_trace_clock = 0;
     SEQ_TRACE_TOKEN(p) = NIL;
+#ifdef USE_VM_PROBES
+    DT_UTAG(p) = NIL;
+    DT_UTAG_FLAGS(p) = 0;
+#endif
     p->parent = parent->id == ERTS_INVALID_PID ? NIL : parent->id;
 
 #ifdef HYBRID
@@ -7430,6 +7448,16 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
 
     VERBOSE(DEBUG_PROCESSES, ("Created a new process: %T\n",p->id));
 
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(process_spawn)) {
+        DTRACE_CHARBUF(process_name, DTRACE_TERM_BUF_SIZE);
+        DTRACE_CHARBUF(mfa, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_fun_decode(p, mod, func, arity, process_name, mfa);
+        DTRACE2(process_spawn, process_name, mfa);
+    }
+#endif
+
  error:
 
     erts_smp_proc_unlock(parent, ERTS_PROC_LOCKS_ALL_MINOR);
@@ -7881,7 +7909,11 @@ static ERTS_INLINE void
 send_exit_message(Process *to, ErtsProcLocks *to_locksp,
 		  Eterm exit_term, Uint term_size, Eterm token)
 {
-    if (token == NIL) {
+    if (token == NIL 
+#ifdef USE_VM_PROBES
+	|| token == am_have_dt_utag
+#endif
+	) {
 	Eterm* hp;
 	Eterm mess;
 	ErlHeapFragment* bp;
@@ -7889,7 +7921,11 @@ send_exit_message(Process *to, ErtsProcLocks *to_locksp,
 
 	hp = erts_alloc_message_heap(term_size, &bp, &ohp, to, to_locksp);
 	mess = copy_struct(exit_term, term_size, &hp, ohp);
-	erts_queue_message(to, to_locksp, bp, mess, NIL);
+	erts_queue_message(to, to_locksp, bp, mess, NIL
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+			   );
     } else {
 	ErlHeapFragment* bp;
 	Eterm* hp;
@@ -7905,7 +7941,11 @@ send_exit_message(Process *to, ErtsProcLocks *to_locksp,
 	/* the trace token must in this case be updated by the caller */
 	seq_trace_output(token, mess, SEQ_TRACE_SEND, to->id, NULL);
 	temp_token = copy_struct(token, sz_token, &hp, &bp->off_heap);
-	erts_queue_message(to, to_locksp, bp, mess, temp_token);
+	erts_queue_message(to, to_locksp, bp, mess, temp_token
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+			   );
     }
 }
 
@@ -7998,9 +8038,26 @@ send_exit_signal(Process *c_p,		/* current process if and only
 
     ASSERT(reason != THE_NON_VALUE);
 
+#ifdef USE_VM_PROBES
+    if(DTRACE_ENABLED(process_exit_signal) && is_pid(from)) {
+        DTRACE_CHARBUF(sender_str, DTRACE_TERM_BUF_SIZE);
+        DTRACE_CHARBUF(receiver_str, DTRACE_TERM_BUF_SIZE);
+        DTRACE_CHARBUF(reason_buf, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_pid_str(from, sender_str);
+        dtrace_proc_str(rp, receiver_str);
+        erts_snprintf(reason_buf, sizeof(reason_buf) - 1, "%T", reason);
+        DTRACE3(process_exit_signal, sender_str, receiver_str, reason_buf);
+    }
+#endif
+
     if (ERTS_PROC_IS_TRAPPING_EXITS(rp)
 	&& (reason != am_kill || (flags & ERTS_XSIG_FLG_IGN_KILL))) {
-	if (is_not_nil(token) && token_update)
+	if (is_not_nil(token) 
+#ifdef USE_VM_PROBES
+	    && token != am_have_dt_utag
+#endif
+	    && token_update)
 	    seq_trace_update_send(token_update);
 	if (is_value(exit_tuple))
 	    send_exit_message(rp, rp_locks, exit_tuple, exit_tuple_sz, token);
@@ -8424,7 +8481,18 @@ erts_do_exit_process(Process* p, Eterm reason)
 
     p->arity = 0;		/* No live registers */
     p->fvalue = reason;
-    
+
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(process_exit)) {
+        DTRACE_CHARBUF(process_buf, DTRACE_TERM_BUF_SIZE);
+        DTRACE_CHARBUF(reason_buf, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_proc_str(p, process_buf);
+        erts_snprintf(reason_buf, DTRACE_TERM_BUF_SIZE - 1, "%T", reason);
+        DTRACE2(process_exit, process_buf, reason_buf);
+    }
+#endif
+
 #ifdef ERTS_SMP
     ERTS_SMP_CHK_HAVE_ONLY_MAIN_PROC_LOCK(p);
     /* By locking all locks (main lock is already locked) when going
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index e20cb493f0..45a96f1e96 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -690,6 +690,10 @@ struct process {
     Uint seq_trace_lastcnt;
     Eterm seq_trace_token;	/* Sequential trace token (tuple size 5 see below) */
 
+#ifdef USE_VM_PROBES
+    Eterm dt_utag;              /* Place to store the dynamc trace user tag */
+    Uint dt_utag_flags;         /* flag field for the dt_utag */
+#endif       
     BeamInstr initial[3];	/* Initial module(0), function(1), arity(2), often used instead
 				   of pointer to funcinfo instruction, hence the BeamInstr datatype */
     BeamInstr* current;		/* Current Erlang function, part of the funcinfo:
@@ -1005,6 +1009,14 @@ extern struct erts_system_profile_flags_t erts_system_profile_flags;
 #define SEQ_TRACE_PRINT    (1 << 2)
 #define SEQ_TRACE_TIMESTAMP (1 << 3)
 
+#ifdef USE_VM_PROBES
+#define DT_UTAG_PERMANENT (1 << 0)
+#define DT_UTAG_SPREADING (1 << 1)
+#define DT_UTAG(P) ((P)->dt_utag)
+#define DT_UTAG_FLAGS(P)  ((P)->dt_utag_flags) 
+#endif
+
+
 #ifdef ERTS_SMP
 /* Status flags ... */
 #define ERTS_PROC_SFLG_PENDADD2SCHEDQ	(((Uint32) 1) << 0)	/* Pending
diff --git a/erts/emulator/beam/erl_trace.c b/erts/emulator/beam/erl_trace.c
index b1d1e1d9b0..4261cd03be 100644
--- a/erts/emulator/beam/erl_trace.c
+++ b/erts/emulator/beam/erl_trace.c
@@ -125,8 +125,13 @@ do { \
     enqueue_sys_msg_unlocked(SYS_MSG_TYPE_TRACE, (FPID), (TPID), (MSG), (BP)); \
 } while(0)
 #else
+#ifdef USE_VM_PROBES
 #define ERTS_ENQ_TRACE_MSG(FPID, TPROC, MSG, BP) \
-  erts_queue_message((TPROC), NULL, (BP), (MSG), NIL)
+    erts_queue_message((TPROC), NULL, (BP), (MSG), NIL, NIL)
+#else
+#define ERTS_ENQ_TRACE_MSG(FPID, TPROC, MSG, BP) \
+    erts_queue_message((TPROC), NULL, (BP), (MSG), NIL)
+#endif
 #endif
 
 /*
@@ -583,7 +588,11 @@ profile_send(Eterm from, Eterm message) {
 	hp = erts_alloc_message_heap(sz, &bp, &off_heap, profile_p, 0);
 	msg = copy_struct(message, sz, &hp, &bp->off_heap);
 	
-    	erts_queue_message(profile_p, NULL, bp, msg, NIL);
+    	erts_queue_message(profile_p, NULL, bp, msg, NIL
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+			   );
     }
 }
 
@@ -994,9 +1003,13 @@ seq_trace_update_send(Process *p)
 {
     Eterm seq_tracer = erts_get_system_seq_tracer();
     ASSERT((is_tuple(SEQ_TRACE_TOKEN(p)) || is_nil(SEQ_TRACE_TOKEN(p))));
-    if ( (p->id == seq_tracer) || (SEQ_TRACE_TOKEN(p) == NIL))
+    if ( (p->id == seq_tracer) || (SEQ_TRACE_TOKEN(p) == NIL)
+#ifdef USE_VM_PROBES
+	 || (SEQ_TRACE_TOKEN(p) == am_have_dt_utag)
+#endif
+	 ) {
 	return 0;
-
+    }
     SEQ_TRACE_TOKEN_SENDER(p) = p->id; /* Internal pid */
     SEQ_TRACE_TOKEN_SERIAL(p) = 
 	make_small(++(p -> seq_trace_clock));
@@ -1178,7 +1191,11 @@ seq_trace_output_generic(Eterm token, Eterm msg, Uint type,
 	enqueue_sys_msg_unlocked(SYS_MSG_TYPE_SEQTRACE, NIL, NIL, mess, bp);
 	erts_smp_mtx_unlock(&smq_mtx);
 #else
-	erts_queue_message(tracer, NULL, bp, mess, NIL); /* trace_token must be NIL here */
+	erts_queue_message(tracer, NULL, bp, mess, NIL
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+			   ); /* trace_token must be NIL here */
 #endif
     }
 }
@@ -2469,7 +2486,11 @@ monitor_long_gc(Process *p, Uint time) {
 #ifdef ERTS_SMP
     enqueue_sys_msg(SYS_MSG_TYPE_SYSMON, p->id, NIL, msg, bp);
 #else
-    erts_queue_message(monitor_p, NULL, bp, msg, NIL);
+    erts_queue_message(monitor_p, NULL, bp, msg, NIL
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+		       );
 #endif
 }
 
@@ -2541,7 +2562,11 @@ monitor_large_heap(Process *p) {
 #ifdef ERTS_SMP
     enqueue_sys_msg(SYS_MSG_TYPE_SYSMON, p->id, NIL, msg, bp);
 #else
-    erts_queue_message(monitor_p, NULL, bp, msg, NIL);
+    erts_queue_message(monitor_p, NULL, bp, msg, NIL
+#ifdef USE_VM_PROBES
+		       , NIL
+#endif
+		       );
 #endif
 }
 
@@ -2571,7 +2596,11 @@ monitor_generic(Process *p, Eterm type, Eterm spec) {
 #ifdef ERTS_SMP
     enqueue_sys_msg(SYS_MSG_TYPE_SYSMON, p->id, NIL, msg, bp);
 #else
-    erts_queue_message(monitor_p, NULL, bp, msg, NIL);
+    erts_queue_message(monitor_p, NULL, bp, msg, NIL
+#ifdef USE_VM_PROBES
+		       , NIL
+#endif
+		       );
 #endif
 
 }
@@ -3357,7 +3386,11 @@ sys_msg_dispatcher_func(void *unused)
 		}
 		else {
 		queue_proc_msg:
-		    erts_queue_message(proc,&proc_locks,smqp->bp,smqp->msg,NIL);
+		    erts_queue_message(proc,&proc_locks,smqp->bp,smqp->msg,NIL
+#ifdef USE_VM_PROBES
+				       , NIL
+#endif
+				       );
 #ifdef DEBUG_PRINTOUTS
 		    erts_fprintf(stderr, "delivered\n");
 #endif
diff --git a/erts/emulator/beam/erlang_dtrace.d b/erts/emulator/beam/erlang_dtrace.d
new file mode 100644
index 0000000000..587e51cb67
--- /dev/null
+++ b/erts/emulator/beam/erlang_dtrace.d
@@ -0,0 +1,726 @@
+/*
+ * %CopyrightBegin%
+ *
+ * Copyright Dustin Sallings, Michal Ptaszek, Scott Lystig Fritchie 2011.
+ * All Rights Reserved.
+ *
+ * The contents of this file are subject to the Erlang Public License,
+ * Version 1.1, (the "License"); you may not use this file except in
+ * compliance with the License. You should have received a copy of the
+ * Erlang Public License along with this software. If not, it can be
+ * retrieved online at http://www.erlang.org/.
+ *
+ * Software distributed under the License is distributed on an "AS IS"
+ * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
+ * the License for the specific language governing rights and limitations
+ * under the License.
+ *
+ * %CopyrightEnd%
+ */
+
+/*
+ * A note on probe naming: if "__" appears in a provider probe
+ * definition, then two things happen during compilation:
+ *
+ *    1. The "__" will turn into a hypen, "-", for the probe name.
+ *    2. The "__" will turn into a single underscore, "_", for the
+ *       macro names and function definitions that the compiler and
+ *       C developers will see.
+ *
+ * We'll try to use the following naming convention.  We're a bit
+ * limited because, as a USDT probe, we can only specify the 4th part
+ * of the probe name, e.g. erlang*:::mumble.  The 2nd part of the
+ * probe name is always going to be "beam" or "beam.smp", and the 3rd
+ * part of the probe name will always be the name of the function
+ * that's calling the probe.
+ *
+ * So, all probes will be have names defined in this file using the
+ * convention category__name or category__sub_category__name.  This
+ * will translate to probe names of category-name or
+ * category-sub_category-name.
+ *
+ * Each of "category", "sub_category", and "name" may have underscores
+ * but may not have hyphens.
+ */
+
+provider erlang {
+    /**
+     * Fired when a message is sent from one local process to another.
+     *
+     * NOTE: The 'size' parameter is in machine-dependent words and
+     *       that the actual size of any binary terms in the message
+     *       are not included.
+     *
+     * @param sender the PID (string form) of the sender
+     * @param receiver the PID (string form) of the receiver
+     * @param size the size of the message being delivered (words)
+     * @param token_label for the sender's sequential trace token
+     * @param token_previous count for the sender's sequential trace token
+     * @param token_current count for the sender's sequential trace token
+     */
+    probe message__send(char *sender, char *receiver, uint32_t size,
+                        int token_label, int token_previous, int token_current);
+
+    /**
+     * Fired when a message is sent from a local process to a remote process.
+     *
+     * NOTE: The 'size' parameter is in machine-dependent words and
+     *       that the actual size of any binary terms in the message
+     *       are not included.
+     *
+     * @param sender the PID (string form) of the sender
+     * @param node_name the Erlang node name (string form) of the receiver
+     * @param receiver the PID/name (string form) of the receiver
+     * @param size the size of the message being delivered (words)
+     * @param token_label for the sender's sequential trace token
+     * @param token_previous count for the sender's sequential trace token
+     * @param token_current count for the sender's sequential trace token
+     */
+    probe message__send__remote(char *sender, char *node_name, char *receiver,
+                                uint32_t size,
+                        int token_label, int token_previous, int token_current);
+
+    /**
+     * Fired when a message is queued to a local process.  This probe
+     * will not fire if the sender's pid == receiver's pid.
+     *
+     * NOTE: The 'size' parameter is in machine-dependent words and
+     *       that the actual size of any binary terms in the message
+     *       are not included.
+     *
+     * NOTE: In cases of messages in external format (i.e. from another
+     *       Erlang node), we probably don't know the message size
+     *       without performing substantial extra computation.  To
+     *       avoid the extra CPU overhead, the message size may be
+     *       reported as -1, which can appear to a D script as 4294967295.
+     *
+     * @param receiver the PID (string form) of the receiver
+     * @param size the size of the message being delivered (words)
+     * @param queue_len length of the queue of the receiving process
+     * @param token_label for the sender's sequential trace token
+     * @param token_previous count for the sender's sequential trace token
+     * @param token_current count for the sender's sequential trace token
+     */
+    probe message__queued(char *receiver, uint32_t size, uint32_t queue_len,
+                        int token_label, int token_previous, int token_current);
+
+    /**
+     * Fired when a message is 'receive'd by a local process and removed
+     * from its mailbox.
+     *
+     * NOTE: The 'size' parameter is in machine-dependent words and
+     *       that the actual size of any binary terms in the message
+     *       are not included.
+     *
+     * NOTE: In cases of messages in external format (i.e. from another
+     *       Erlang node), we probably don't know the message size
+     *       without performing substantial extra computation.  To
+     *       avoid the extra CPU overhead, the message size may be
+     *       reported as -1, which can appear to a D script as 4294967295.
+     *
+     * @param receiver the PID (string form) of the receiver
+     * @param size the size of the message being delivered (words)
+     * @param queue_len length of the queue of the receiving process
+     * @param token_label for the sender's sequential trace token
+     * @param token_previous count for the sender's sequential trace token
+     * @param token_current count for the sender's sequential trace token
+     */
+    probe message__receive(char *receiver, uint32_t size, uint32_t queue_len,
+                        int token_label, int token_previous, int token_current);
+
+    /**
+     * Fired when an Eterm structure is being copied.
+     *
+     * NOTE: Due to the placement of this probe, the process ID of
+     *       owner of the Eterm is not available.
+     *
+     * @param size the size of the structure
+     */
+    probe copy__struct(uint32_t size);
+
+    /**
+     * Fired when an Eterm is being copied onto a process.
+     *
+     * @param proc the PID (string form) of the recipient process
+     * @param size the size of the structure
+     */
+    probe copy__object(char *proc, uint32_t size);
+
+    /* PID, Module, Function, Arity */
+
+    /**
+     * Fired whenever a user function is being called locally.
+     *
+     * @param p the PID (string form) of the process
+     * @param mfa the m:f/a of the function
+     * @param depth the stack depth
+     */
+    probe local__function__entry(char *p, char *mfa, int depth);
+
+    /**
+     * Fired whenever a user function is called externally
+     * (through an export entry).
+     *
+     * @param p the PID (string form) of the process
+     * @param mfa the m:f/a of the function
+     * @param depth the stack depth
+     */
+    probe global__function__entry(char *p, char *mfa, int depth);
+
+    /**
+     * Fired whenever a user function returns.
+     *
+     * @param p the PID (string form) of the process
+     * @param mfa the m:f/a of the function
+     * @param depth the stack depth
+     */
+    probe function__return(char *p, char *mfa, int depth);
+
+    /**
+     * Fired whenever a Built In Function is called.
+     *
+     * @param p the PID (string form) of the process
+     * @param mfa the m:f/a of the function
+     */
+    probe bif__entry(char *p, char *mfa);
+
+    /**
+     * Fired whenever a Built In Function returns.
+     *
+     * @param p the PID (string form) of the process
+     * @param mfa the m:f/a of the function
+     */
+    probe bif__return(char *p, char *mfa);
+
+    /**
+     * Fired whenever a Native Function is called.
+     *
+     * @param p the PID (string form) of the process
+     * @param mfa the m:f/a of the function
+     */
+    probe nif__entry(char *p, char *mfa);
+
+    /**
+     * Fired whenever a Native Function returns.
+     *
+     * @param p the PID (string form) of the process
+     * @param mfa the m:f/a of the function
+     */
+    probe nif__return(char *p, char *mfa);
+
+    /**
+     * Fired when a major GC is starting.
+     *
+     * @param p the PID (string form) of the exiting process
+     * @param need the number of words needed on the heap
+     */
+    probe gc_major__start(char *p, int need);
+
+    /**
+     * Fired when a minor GC is starting.
+     *
+     * @param p the PID (string form) of the exiting process
+     * @param need the number of words needed on the heap
+     */
+    probe gc_minor__start(char *p, int need);
+
+    /**
+     * Fired when a major GC is starting.
+     *
+     * @param p the PID (string form) of the exiting process
+     * @param reclaimed the amount of space reclaimed
+     */
+    probe gc_major__end(char *p, int reclaimed);
+
+    /**
+     * Fired when a minor GC is starting.
+     *
+     * @param p the PID (string form) of the exiting process
+     * @param reclaimed the amount of space reclaimed
+     */
+    probe gc_minor__end(char *p, int reclaimed);
+
+    /**
+     * Fired when a process is spawned.
+     *
+     * @param p the PID (string form) of the new process.
+     * @param mfa the m:f/a of the function
+     */
+    probe process__spawn(char *p, char *mfa);
+
+    /**
+     * Fired when a process is exiting.
+     *
+     * @param p the PID (string form) of the exiting process
+     * @param reason the reason for the exit (may be truncated)
+     */
+    probe process__exit(char *p, char *reason);
+
+    /**
+     * Fired when exit signal is delivered to a local process.
+     *
+     * @param sender the PID (string form) of the exiting process
+     * @param receiver the PID (string form) of the process receiving EXIT signal
+     * @param reason the reason for the exit (may be truncated)
+     */
+    probe process__exit_signal(char *sender, char *receiver, char *reason);
+
+    /**
+     * Fired when exit signal is delivered to a remote process.
+     *
+     * @param sender the PID (string form) of the exiting process
+     * @param node_name the Erlang node name (string form) of the receiver
+     * @param receiver the PID (string form) of the process receiving EXIT signal
+     * @param reason the reason for the exit (may be truncated)
+     * @param token_label for the sender's sequential trace token
+     * @param token_previous count for the sender's sequential trace token
+     * @param token_current count for the sender's sequential trace token
+     */
+    probe process__exit_signal__remote(char *sender, char *node_name,
+                                       char *receiver, char *reason,
+                        int token_label, int token_previous, int token_current);
+
+    /**
+     * Fired when a process is scheduled.
+     *
+     * @param p the PID (string form) of the newly scheduled process
+     * @param mfa the m:f/a of the function it should run next
+     */
+    probe process__scheduled(char *p, char *mfa);
+
+    /**
+     * Fired when a process is unscheduled.
+     *
+     * @param p the PID (string form) of the process that has been
+     * unscheduled.
+     */
+    probe process__unscheduled(char *p);
+
+    /**
+     * Fired when a process goes into hibernation.
+     *
+     * @param p the PID (string form) of the process entering hibernation
+     * @param mfa the m:f/a of the location to resume
+     */
+    probe process__hibernate(char *p, char *mfa);
+
+    /**
+     * Fired when a process is unblocked after a port has been unblocked.
+     *
+     * @param p the PID (string form) of the process that has been
+     * unscheduled.
+     * @param port the port that is no longer busy (i.e., is now unblocked)
+     */
+    probe process__port_unblocked(char *p, char *port);
+
+    /**
+     * Fired when process' heap is growing.
+     *
+     * @param p the PID (string form)
+     * @param old_size the size of the old heap
+     * @param new_size the size of the new heap
+     */
+    probe process__heap_grow(char *p, int old_size, int new_size);
+
+    /**
+     * Fired when process' heap is shrinking.
+     *
+     * @param p the PID (string form)
+     * @param old_size the size of the old heap
+     * @param new_size the size of the new heap
+     */
+    probe process__heap_shrink(char *p, int old_size, int new_size);
+
+    /* network distribution */
+
+    /**
+     * Fired when network distribution event monitor events are triggered.
+     *
+     * @param node the name of the reporting node
+     * @param what the type of event, e.g., nodeup, nodedown
+     * @param monitored_node the name of the monitored node
+     * @param type the type of node, e.g., visible, hidden
+     * @param reason the reason term, e.g., normal, connection_closed, term()
+     */
+    probe dist__monitor(char *node, char *what, char *monitored_node,
+                        char *type, char *reason);
+
+    /**
+     * Fired when network distribution port is busy (i.e. blocked),
+     * usually due to the remote node not consuming distribution
+     * data quickly enough.
+     *
+     * @param node the name of the reporting node
+     * @param port the port ID of the busy port
+     * @param remote_node the name of the remote node.
+     * @param pid the PID (string form) of the local process that has
+     *        become unschedulable until the port becomes unblocked.
+     */
+    probe dist__port_busy(char *node, char *port, char *remote_node,
+                          char *pid);
+
+    /**
+     * Fired when network distribution's driver's "output" callback is called
+     *
+     * @param node the name of the reporting node
+     * @param port the port ID of the busy port
+     * @param remote_node the name of the remote node.
+     * @param bytes the number of bytes written
+     */
+    probe dist__output(char *node, char *port, char *remote_node, int bytes);
+
+    /**
+     * Fired when network distribution's driver's "outputv" callback is called
+     *
+     * @param node the name of the reporting node
+     * @param port the port ID of the busy port
+     * @param remote_node the name of the remote node.
+     * @param bytes the number of bytes written
+     */
+    probe dist__outputv(char *node, char *port, char *remote_node, int bytes);
+
+    /**
+     * Fired when network distribution port is no longer busy (i.e. blocked).
+     *
+     * NOTE: This probe may fire multiple times after the same single
+     *       dist-port_busy probe firing.
+     *
+     * @param node the name of the reporting node
+     * @param port the port ID of the busy port
+     * @param remote_node the name of the remote node.
+     */
+    probe dist__port_not_busy(char *node, char *port, char *remote_node);
+
+    /* ports */
+
+    /**
+     * Fired when new port is opened.
+     *
+     * @param process the PID (string form)
+     * @param port_name the string used when the port was opened
+     * @param port the Port (string form) of the new port
+     */
+    probe port__open(char *process, char *port_name, char *port);
+
+    /**
+     * Fired when port_command is issued.
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     * @param command_type type of the issued command, one of: "close", "command" or "connect"
+     */
+    probe port__command(char *process, char *port, char *port_name, char *command_type);
+
+    /**
+     * Fired when port_control is issued.
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     * @param command_no command number that has been issued to the port
+     */
+    probe port__control(char *process, char *port, char *port_name, int command_no);
+
+    /**
+     * Fired when port is closed via port_close/1 (reason = 'normal')
+     * or is sent an exit signal.
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     * @param reason Erlang term representing the exit signal, e.g. 'normal'
+     */
+    probe port__exit(char *process, char *port, char *port_name,
+                     char *new_process);
+
+    /**
+     * Fired when port_connect is issued.
+     *
+     * @param process the PID (string form) of the current port owner
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     * @param new_process the PID (string form) of the new port owner
+     */
+    probe port__connect(char *process, char *port, char *port_name,
+                        char *new_process);
+
+    /**
+     * Fired when a port is busy (i.e. blocked)
+     *
+     * @param port the port ID of the busy port
+     */
+    probe port__busy(char *port);
+
+    /**
+     * Fired when a port is no longer busy (i.e. no longer blocked)
+     *
+     * @param port the port ID of the not busy port
+     */
+    probe port__not_busy(char *port);
+
+    /* drivers */
+
+    /**
+     * Fired when drivers's "init" callback is called.
+     *
+     * @param name the name of the driver
+     * @param major the major version number
+     * @param minor the minor version number
+     * @param flags the flags argument
+     */
+    probe driver__init(char *name, int major, int minor, int flags);
+
+    /**
+     * Fired when drivers's "start" callback is called.
+     *
+     * @param process the PID (string form) of the calling process
+     * @param name the name of the driver
+     * @param port the Port (string form) of the driver's port
+     */
+     probe driver__start(char *process, char *name, char *port);
+
+    /**
+     * Fired when drivers's "stop" callback is called.
+     *
+     * @param process the PID (string form) of the calling process
+     * @param name the name of the driver
+     * @param port the Port (string form) of the driver's port
+     */
+     probe driver__stop(char *process, char *name, char *port);
+
+    /**
+     * Fired when drivers's "finish" callback is called.
+     *
+     * @param name the name of the driver
+     */
+     probe driver__finish(char *name);
+
+    /**
+     * Fired when drivers's "flush" callback is called.
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     */
+    probe driver__flush(char *process, char *port, char *port_name);
+
+    /**
+     * Fired when driver's "output" callback is called
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     * @param bytes the number of bytes written
+     */
+    probe driver__output(char *node, char *port, char *port_name, int bytes);
+
+    /**
+     * Fired when driver's "outputv" callback is called
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     * @param bytes the number of bytes written
+     */
+    probe driver__outputv(char *node, char *port, char *port_name, int bytes);
+
+    /**
+     * Fired when driver's "control" callback is called
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     * @param command the command #
+     * @param bytes the number of bytes written
+     */
+    probe driver__control(char *node, char *port, char *port_name,
+                          int command, int bytes);
+
+    /**
+     * Fired when driver's "call" callback is called
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     * @param command the command #
+     * @param bytes the number of bytes written
+     */
+    probe driver__call(char *node, char *port, char *port_name,
+                       int command, int bytes);
+
+    /**
+     * Fired when driver's "event" callback is called
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     */
+    probe driver__event(char *node, char *port, char *port_name);
+
+    /**
+     * Fired when driver's "ready_input" callback is called
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     */
+    probe driver__ready_input(char *node, char *port, char *port_name);
+
+    /**
+     * Fired when driver's "read_output" callback is called
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     */
+    probe driver__ready_output(char *node, char *port, char *port_name);
+
+    /**
+     * Fired when driver's "timeout" callback is called
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     */
+    probe driver__timeout(char *node, char *port, char *port_name);
+
+    /**
+     * Fired when drivers's "ready_async" callback is called.
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     */
+    probe driver__ready_async(char *process, char *port, char *port_name);
+
+    /**
+     * Fired when driver's "process_exit" callback is called
+     *
+     * @param process the PID (string form)
+     * @param port the Port (string form)
+     * @param port_name the string used when the port was opened
+     */
+    probe driver__process_exit(char *node, char *port, char *port_name);
+
+    /**
+     * Fired when driver's "stop_select" callback is called
+     *
+     * @param name the name of the driver
+     */
+    probe driver__stop_select(char *name);
+
+
+    /* Async driver pool */
+
+    /**
+     * Show the post-add length of the async driver thread pool member's queue.
+     *
+     * NOTE: The port name is not available: additional lock(s) must
+     *       be acquired in order to get the port name safely in an SMP
+     *       environment.  The same is true for the aio__pool_get probe.
+     *
+     * @param port the Port (string form)
+     * @param new queue length
+     */
+    probe aio_pool__add(char *, int);
+
+    /**
+     * Show the post-get length of the async driver thread pool member's queue.
+     *
+     * @param port the Port (string form)
+     * @param new queue length
+     */
+    probe aio_pool__get(char *, int);
+
+    /* Probes for efile_drv.c */
+
+    /**
+     * Entry into the efile_drv.c file I/O driver
+     *
+     * For a list of command numbers used by this driver, see the section
+     * "Guide to probe arguments" in ../../../README.md.  That section
+     * also contains explanation of the various integer and string
+     * arguments that may be present when any particular probe fires.
+     *
+     * NOTE: Not all Linux platforms (using SystemTap) can support
+     *       arguments beyond arg9.
+     *
+     *
+     * TODO: Adding the port string, args[10], is a pain.  Making that
+     *       port string available to all the other efile_drv.c probes
+     *       will be more pain.  Is the pain worth it?  If yes, then
+     *       add them everywhere else and grit our teeth.  If no, then
+     *       rip it out.
+     *
+     * @param thread-id number of the scheduler Pthread                   arg0
+     * @param tag number: {thread-id, tag} uniquely names a driver operation
+     * @param user-tag string                                             arg2
+     * @param command number                                              arg3
+     * @param string argument 1                                           arg4
+     * @param string argument 2                                           arg5
+     * @param integer argument 1                                          arg6
+     * @param integer argument 2                                          arg7
+     * @param integer argument 3                                          arg8
+     * @param integer argument 4                                          arg9
+     * @param port the port ID of the busy port                       args[10]
+     */
+    probe efile_drv__entry(int, int, char *, int, char *, char *,
+                           int64_t, int64_t, int64_t, int64_t, char *);
+
+    /**
+     * Entry into the driver's internal work function.  Computation here
+     * is performed by a async worker pool Pthread.
+     *
+     * @param thread-id number
+     * @param tag number
+     * @param command number
+     */
+    probe efile_drv__int_entry(int, int, int);
+
+    /**
+     * Return from the driver's internal work function.
+     *
+     * @param thread-id number
+     * @param tag number
+     * @param command number
+     */
+    probe efile_drv__int_return(int, int, int);
+
+    /**
+     * Return from the efile_drv.c file I/O driver
+     *
+     * @param thread-id number                                            arg0
+     * @param tag number                                                  arg1
+     * @param user-tag string                                             arg2
+     * @param command number                                              arg3
+     * @param Success? 1 is success, 0 is failure                         arg4
+     * @param If failure, the errno of the error.                         arg5
+     */
+    probe efile_drv__return(int, int, char *, int, int, int);
+
+/*
+ * NOTE:
+ * For formatting int64_t arguments within a D script, see:
+ *
+ *   http://mail.opensolaris.org/pipermail/dtrace-discuss/2006-November/002830.html
+ *   Summary:
+ *       "1) you don't need the 'l' printf() modifiers with DTrace ever"
+ */
+
+/*
+ * NOTE: For file_drv_return + SMP + R14B03 (and perhaps other
+ *       releases), the sched-thread-id will be the same as the
+ *       work-thread-id: erl_async.c's async_main() function
+ *       will call the asynchronous invoke function and then
+ *       immediately call the drivers ready_async function while
+ *       inside the same I/O worker pool thread.
+ *       For R14B03's source, see erl_async.c lines 302-317.
+ */
+};
+
+#pragma D attributes Evolving/Evolving/Common provider erlang provider
+#pragma D attributes Private/Private/Common provider erlang module
+#pragma D attributes Private/Private/Common provider erlang function
+#pragma D attributes Evolving/Evolving/Common provider erlang name
+#pragma D attributes Evolving/Evolving/Common provider erlang args
diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h
index b259ef3694..12e40288bd 100644
--- a/erts/emulator/beam/global.h
+++ b/erts/emulator/beam/global.h
@@ -1988,4 +1988,46 @@ erts_alloc_message_heap(Uint size,
 #  define UseTmpHeapNoproc(Size) /* Nothing */
 #  define UnUseTmpHeapNoproc(Size) /* Nothing */
 #endif /* HEAP_ON_C_STACK */
+
+#if ERTS_GLB_INLINE_INCL_FUNC_DEF
+
+#include "dtrace-wrapper.h"
+
+ERTS_GLB_INLINE void
+dtrace_pid_str(Eterm pid, char *process_buf)
+{
+    erts_snprintf(process_buf, DTRACE_TERM_BUF_SIZE, "<%lu.%lu.%lu>",
+                  pid_channel_no(pid),
+                  pid_number(pid),
+                  pid_serial(pid));
+}
+
+ERTS_GLB_INLINE void
+dtrace_proc_str(Process *process, char *process_buf)
+{
+    dtrace_pid_str(process->id, process_buf);
+}
+
+ERTS_GLB_INLINE void
+dtrace_port_str(Port *port, char *port_buf)
+{
+    erts_snprintf(port_buf, DTRACE_TERM_BUF_SIZE, "#Port<%lu.%lu>",
+                  port_channel_no(port->id),
+                  port_number(port->id));
+}
+
+ERTS_GLB_INLINE void
+dtrace_fun_decode(Process *process,
+                  Eterm module, Eterm function, int arity,
+                  char *process_buf, char *mfa_buf)
+{
+    if (process_buf) {
+        dtrace_proc_str(process, process_buf);
+    }
+
+    erts_snprintf(mfa_buf, DTRACE_TERM_BUF_SIZE, "%T:%T/%d",
+                  module, function, arity);
+}
+#endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */
+
 #endif /* !__GLOBAL_H__ */
diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c
index 7604365ab6..001d8d0362 100644
--- a/erts/emulator/beam/io.c
+++ b/erts/emulator/beam/io.c
@@ -43,6 +43,7 @@
 #include "erl_version.h"
 #include "error.h"
 #include "erl_async.h"
+#include "dtrace-wrapper.h"
 
 extern ErlDrvEntry fd_driver_entry;
 extern ErlDrvEntry vanilla_driver_entry;
@@ -180,6 +181,20 @@ typedef struct line_buf_context {
 
 #define LINEBUF_INITIAL 100
 
+#ifdef USE_VM_PROBES
+#define DTRACE_FORMAT_COMMON_PID_AND_PORT(PID, PORT)         \
+    DTRACE_CHARBUF(process_str, DTRACE_TERM_BUF_SIZE);       \
+    DTRACE_CHARBUF(port_str, DTRACE_TERM_BUF_SIZE);          \
+                                                             \
+    dtrace_pid_str((PID), process_str);                      \
+    dtrace_port_str((PORT), port_str);
+#define DTRACE_FORMAT_COMMON_PROC_AND_PORT(PID, PORT)        \
+    DTRACE_CHARBUF(process_str, DTRACE_TERM_BUF_SIZE);       \
+    DTRACE_CHARBUF(port_str, DTRACE_TERM_BUF_SIZE);          \
+                                                             \
+    dtrace_proc_str((PID), process_str);                     \
+    dtrace_port_str((PORT), port_str);
+#endif
 
 /* The 'number' field in a port now has two parts: the lowest bits
    contain the index in the port table, and the higher bits are a counter
@@ -640,6 +655,12 @@ erts_open_driver(erts_driver_t* driver,	/* Pointer to driver. */
 	    trace_sched_ports_where(port, am_in, am_start);
 	}
 	port->caller = pid;
+#ifdef USE_VM_PROBES
+        if (DTRACE_ENABLED(driver_start)) {
+            DTRACE_FORMAT_COMMON_PID_AND_PORT(pid, port)
+            DTRACE3(driver_start, process_str, driver->name, port_str);
+        }
+#endif
 	fpe_was_unmasked = erts_block_fpe();
 	drv_data = (*driver->start)((ErlDrvPort)(port_ix),
 				    name, opts);
@@ -1171,6 +1192,12 @@ int erts_write_to_port(Eterm caller_id, Port *p, Eterm list)
 	ev.size = size;  /* total size */
 	ev.iov = ivp;
 	ev.binv = bvp;
+#ifdef USE_VM_PROBES
+        if (DTRACE_ENABLED(driver_outputv)) {
+            DTRACE_FORMAT_COMMON_PID_AND_PORT(caller_id, p)
+            DTRACE4(driver_outputv, process_str, port_str, p->name, size);
+        }
+#endif
 	fpe_was_unmasked = erts_block_fpe();
 	(*drv->outputv)((ErlDrvData)p->drv_data, &ev);
 	erts_unblock_fpe(fpe_was_unmasked);
@@ -1190,8 +1217,21 @@ int erts_write_to_port(Eterm caller_id, Port *p, Eterm list)
 	buf = erts_alloc(ERTS_ALC_T_TMP, size+1);
 	r = io_list_to_buf(list, buf, size);
 
+#ifdef USE_VM_PROBES
+	if(DTRACE_ENABLED(port_command)) {
+            DTRACE_FORMAT_COMMON_PID_AND_PORT(caller_id, p)
+	    DTRACE4(port_command, process_str, port_str, p->name, "command");
+	}
+#endif
+
 	if (r >= 0) {
 	    size -= r;
+#ifdef USE_VM_PROBES
+            if (DTRACE_ENABLED(driver_output)) {
+                DTRACE_FORMAT_COMMON_PID_AND_PORT(caller_id, p)
+                DTRACE4(driver_output, process_str, port_str, p->name, size);
+            }
+#endif
 	    fpe_was_unmasked = erts_block_fpe();
 	    (*drv->output)((ErlDrvData)p->drv_data, buf, size);
 	    erts_unblock_fpe(fpe_was_unmasked);
@@ -1215,6 +1255,12 @@ int erts_write_to_port(Eterm caller_id, Port *p, Eterm list)
 	     */
 	    buf = erts_alloc(ERTS_ALC_T_TMP, size+1); 
 	    r = io_list_to_buf(list, buf, size);
+#ifdef USE_VM_PROBES
+            if (DTRACE_ENABLED(driver_output)) {
+                DTRACE_FORMAT_COMMON_PID_AND_PORT(caller_id, p)
+                DTRACE4(driver_output, process_str, port_str, p->name, size);
+            }
+#endif
 	    fpe_was_unmasked = erts_block_fpe();
 	    (*drv->output)((ErlDrvData)p->drv_data, buf, size);
 	    erts_unblock_fpe(fpe_was_unmasked);
@@ -1530,7 +1576,11 @@ deliver_result(Eterm sender, Eterm pid, Eterm res)
 	    hp = erts_alloc_message_heap(sz_res + 3, &bp, &ohp, rp, &rp_locks);
 	    res = copy_struct(res, sz_res, &hp, ohp);
 	    tuple = TUPLE2(hp, sender, res);
-	erts_queue_message(rp, &rp_locks, bp, tuple, NIL);
+	erts_queue_message(rp, &rp_locks, bp, tuple, NIL
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+			   );
 	erts_smp_proc_unlock(rp, rp_locks);
 	erts_smp_proc_dec_refc(rp);
     }
@@ -1619,7 +1669,11 @@ static void deliver_read_message(Port* prt, Eterm to,
     tuple = TUPLE2(hp, prt->id, tuple);
     hp += 3;
 
-    erts_queue_message(rp, &rp_locks, bp, tuple, am_undefined);
+    erts_queue_message(rp, &rp_locks, bp, tuple, am_undefined
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+		       );
     erts_smp_proc_unlock(rp, rp_locks);
     erts_smp_proc_dec_refc(rp);
 }
@@ -1772,7 +1826,11 @@ deliver_vec_message(Port* prt,			/* Port */
     tuple = TUPLE2(hp, prt->id, tuple);
     hp += 3;
 
-    erts_queue_message(rp, &rp_locks, bp, tuple, am_undefined);
+    erts_queue_message(rp, &rp_locks, bp, tuple, am_undefined
+#ifdef USE_VM_PROBES
+		       , NIL
+#endif
+		       );
     erts_smp_proc_unlock(rp, rp_locks);
     erts_smp_proc_dec_refc(rp);
 }
@@ -1811,6 +1869,12 @@ static void flush_port(Port *p)
     ERTS_SMP_LC_ASSERT(erts_lc_is_port_locked(p));
 
     if (p->drv_ptr->flush != NULL) {
+#ifdef USE_VM_PROBES
+        if (DTRACE_ENABLED(driver_flush)) {
+            DTRACE_FORMAT_COMMON_PID_AND_PORT(p->connected, p)
+            DTRACE3(driver_flush, process_str, port_str, p->name);
+        }
+#endif
         if (IS_TRACED_FL(p, F_TRACE_SCHED_PORTS)) {
 	    trace_sched_ports_where(p, am_in, am_flush);
 	}
@@ -1866,6 +1930,12 @@ terminate_port(Port *prt)
     drv = prt->drv_ptr;
     if ((drv != NULL) && (drv->stop != NULL)) {
 	int fpe_was_unmasked = erts_block_fpe();
+#ifdef USE_VM_PROBES
+        if (DTRACE_ENABLED(driver_stop)) {
+            DTRACE_FORMAT_COMMON_PID_AND_PORT(prt->connected, prt)
+            DTRACE3(driver_stop, process_str, drv->name, port_str);
+        }
+#endif
 	(*drv->stop)((ErlDrvData)prt->drv_data);
 	erts_unblock_fpe(fpe_was_unmasked);
 #ifdef ERTS_SMP
@@ -2027,6 +2097,19 @@ erts_do_exit_port(Port *p, Eterm from, Eterm reason)
 
    rreason = (reason == am_kill) ? am_killed : reason;
 
+#ifdef USE_VM_PROBES
+   if (DTRACE_ENABLED(port_exit)) {
+       DTRACE_CHARBUF(from_str, DTRACE_TERM_BUF_SIZE);
+       DTRACE_CHARBUF(port_str, DTRACE_TERM_BUF_SIZE);
+       DTRACE_CHARBUF(rreason_str, 64);
+
+       erts_snprintf(from_str, sizeof(from_str), "%T", from);
+       dtrace_port_str(p, port_str);
+       erts_snprintf(rreason_str, sizeof(rreason_str), "%T", rreason);
+       DTRACE4(port_exit, from_str, port_str, p->name, rreason_str);
+   }
+#endif
+
    if ((p->status & (ERTS_PORT_SFLGS_DEAD
 		     | ERTS_PORT_SFLG_EXITING
 		     | ERTS_PORT_SFLG_IMMORTAL))
@@ -2127,6 +2210,13 @@ void erts_port_command(Process *proc,
 	    if (tp[2] == am_close) {
 		erts_port_status_bor_set(port, ERTS_PORT_SFLG_SEND_CLOSED);
 		erts_do_exit_port(port, pid, am_normal);
+
+#ifdef USE_VM_PROBES
+		if(DTRACE_ENABLED(port_command)) {
+                    DTRACE_FORMAT_COMMON_PROC_AND_PORT(proc, port)
+		    DTRACE4(port_command, process_str, port_str, port->name, "close");
+		}
+#endif
 		goto done;
 	    } else if (is_tuple_arity(tp[2], 2)) {
 		tp = tuple_val(tp[2]);
@@ -2134,6 +2224,12 @@ void erts_port_command(Process *proc,
 		    if (erts_write_to_port(caller_id, port, tp[2]) == 0)
 			goto done;
 		} else if ((tp[1] == am_connect) && is_internal_pid(tp[2])) {
+#ifdef USE_VM_PROBES
+		    if(DTRACE_ENABLED(port_command)) {
+                        DTRACE_FORMAT_COMMON_PROC_AND_PORT(proc, port)
+			DTRACE4(port_command, process_str, port_str, port->name, "connect");
+		    }
+#endif
 		    port->connected = tp[2];
 		    deliver_result(port->id, pid, am_connected);
 		    goto done;
@@ -2236,6 +2332,15 @@ erts_port_control(Process* p, Port* prt, Uint command, Eterm iolist)
     erts_smp_proc_unlock(p, ERTS_PROC_LOCK_MAIN);
     ERTS_SMP_CHK_NO_PROC_LOCKS;
 
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(port_control) || DTRACE_ENABLED(driver_control)) {
+        DTRACE_FORMAT_COMMON_PROC_AND_PORT(p, prt);
+        DTRACE4(port_control, process_str, port_str, prt->name, command);
+        DTRACE5(driver_control, process_str, port_str, prt->name,
+                command, to_len);
+    }
+#endif
+
     /*
      * Call the port's control routine.
      */
@@ -2376,6 +2481,10 @@ print_port_info(int to, void *arg, int i)
 void
 set_busy_port(ErlDrvPort port_num, int on)
 {
+#ifdef USE_VM_PROBES
+    DTRACE_CHARBUF(port_str, 16);
+#endif
+
     ERTS_SMP_CHK_NO_PROC_LOCKS;
 
     ERTS_SMP_LC_ASSERT(erts_lc_is_port_locked(&erts_port[port_num]));
@@ -2383,12 +2492,26 @@ set_busy_port(ErlDrvPort port_num, int on)
     if (on) {
         erts_port_status_bor_set(&erts_port[port_num],
 				 ERTS_PORT_SFLG_PORT_BUSY);
+#ifdef USE_VM_PROBES
+        if (DTRACE_ENABLED(port_busy)) {
+            erts_snprintf(port_str, sizeof(port_str),
+                          "%T", erts_port[port_num].id);
+            DTRACE1(port_busy, port_str);
+        }
+#endif
     } else {
         ErtsProcList* plp = erts_port[port_num].suspended;
         erts_port_status_band_set(&erts_port[port_num],
 				  ~ERTS_PORT_SFLG_PORT_BUSY);
         erts_port[port_num].suspended = NULL;
 
+#ifdef USE_VM_PROBES
+        if (DTRACE_ENABLED(port_not_busy)) {
+            erts_snprintf(port_str, sizeof(port_str),
+                          "%T", erts_port[port_num].id);
+            DTRACE1(port_not_busy, port_str);
+        }
+#endif
 	if (erts_port[port_num].dist_entry) {
 	    /*
 	     * Processes suspended on distribution ports are
@@ -2406,6 +2529,28 @@ set_busy_port(ErlDrvPort port_num, int on)
 	 */
 
         if (plp) {
+#ifdef USE_VM_PROBES
+            /*
+             * Hrm, for blocked dist ports, plp always seems to be NULL.
+             * That's not so fun.
+             * Well, another way to get the same info is using a D
+             * script to correlate an earlier process-port_blocked+pid
+             * event with a later process-scheduled event.  That's
+             * subject to the multi-CPU races with how events are
+             * handled, but hey, that way works most of the time.
+             */
+            if (DTRACE_ENABLED(process_port_unblocked)) {
+                DTRACE_CHARBUF(pid_str, 16);
+                ErtsProcList* plp2 = plp;
+
+                erts_snprintf(port_str, sizeof(port_str),
+                             "%T", erts_port[port_num]);
+                while (plp2 != NULL) {
+                    erts_snprintf(pid_str, sizeof(pid_str), "%T", plp2->pid);
+                    DTRACE2(process_port_unblocked, pid_str, port_str);
+                }
+            }
+#endif
             /* First proc should be resumed last */
 	    if (plp->next) {
 		erts_resume_processes(plp->next);
@@ -2452,6 +2597,14 @@ void erts_raw_port_command(Port* p, byte* buf, Uint len)
 		 p->drv_ptr->name ? p->drv_ptr->name : "unknown");
 
     p->caller = NIL;
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(driver_output)) {
+        DTRACE_CHARBUF(port_str, DTRACE_TERM_BUF_SIZE);
+
+        dtrace_port_str(p, port_str);
+        DTRACE4(driver_output, "-raw-", port_str, p->name, len);
+    }
+#endif
     fpe_was_unmasked = erts_block_fpe();
     (*p->drv_ptr->output)((ErlDrvData)p->drv_data, (char*) buf, (int) len);
     erts_unblock_fpe(fpe_was_unmasked);
@@ -2467,6 +2620,12 @@ int async_ready(Port *p, void* data)
 	ERTS_SMP_LC_ASSERT(erts_lc_is_port_locked(p));
 	ASSERT(!(p->status & ERTS_PORT_SFLGS_DEAD));
 	if (p->drv_ptr->ready_async != NULL) {
+#ifdef USE_VM_PROBES
+            if (DTRACE_ENABLED(driver_ready_async)) {
+                DTRACE_FORMAT_COMMON_PID_AND_PORT(p->connected, p)
+                DTRACE3(driver_ready_async, process_str, port_str, p->name);
+            }
+#endif
 	    (*p->drv_ptr->ready_async)((ErlDrvData)p->drv_data, data);
 	    need_free = 0;
 #ifdef ERTS_SMP
@@ -2661,7 +2820,11 @@ void driver_report_exit(int ix, int status)
    hp += 3;
    tuple = TUPLE2(hp, prt->id, tuple);
 
-   erts_queue_message(rp, &rp_locks, bp, tuple, am_undefined);
+   erts_queue_message(rp, &rp_locks, bp, tuple, am_undefined
+#ifdef USE_VM_PROBES
+		      , NIL
+#endif
+		      );
 
    erts_smp_proc_unlock(rp, rp_locks);
    erts_smp_proc_dec_refc(rp);
@@ -3211,7 +3374,11 @@ driver_deliver_term(ErlDrvPort port,
 	    HRelease(rp, hp_end, hp);	    
 	}
 	/* send message */
-	erts_queue_message(rp, &rp_locks, bp, mess, am_undefined);
+	erts_queue_message(rp, &rp_locks, bp, mess, am_undefined
+#ifdef USE_VM_PROBES
+			   , NIL
+#endif
+			   );
     }
     else {
 	if (b2t.ix > b2t.used)
@@ -4442,6 +4609,12 @@ void erts_fire_port_monitor(Port *prt, Eterm ref)
     ASSERT(callback != NULL);
     ref_to_driver_monitor(ref,&drv_monitor);
     DRV_MONITOR_UNLOCK_PDL(prt);
+#ifdef USE_VM_PROBES
+    if (DTRACE_ENABLED(driver_process_exit)) {
+        DTRACE_FORMAT_COMMON_PID_AND_PORT(prt->connected, prt)
+        DTRACE3(driver_process_exit, process_str, port_str, prt->name);
+    }
+#endif
     fpe_was_unmasked = erts_block_fpe();
     (*callback)((ErlDrvData) (prt->drv_data), &drv_monitor);
     erts_unblock_fpe(fpe_was_unmasked);
@@ -4885,6 +5058,8 @@ init_driver(erts_driver_t *drv, ErlDrvEntry *de, DE_Handle *handle)
     else {
 	int res;
 	int fpe_was_unmasked = erts_block_fpe();
+        DTRACE4(driver_init, drv->name, drv->version.major, drv->version.minor,
+                drv->flags);
 	res = (*de->init)();
 	erts_unblock_fpe(fpe_was_unmasked);
 	return res;
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index f051f6aa5b..a0165107d2 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -887,6 +887,95 @@ call_ext_last u==3 u$func:erlang:hibernate/3 D => i_hibernate
 call_ext_only u==3 u$func:erlang:hibernate/3 => i_hibernate
 
 #
+# If VM probes are not enabled, we want to short-circult calls to
+# the dt tag BIFs to make them as cheap as possible.
+#
+
+%unless USE_VM_PROBES
+
+call_ext Arity u$func:erlang:dt_get_tag/0 => \
+    move a=am_undefined r
+call_ext_last Arity u$func:erlang:dt_get_tag/0 D => \
+    move a=am_undefined r | deallocate D | return
+call_ext_only Arity u$func:erlang:dt_get_tag/0 => \
+    move a=am_undefined r | return
+
+move Any r | call_ext Arity u$func:erlang:dt_put_tag/1 => \
+    move a=am_undefined r
+move Any r | call_ext_last Arity u$func:erlang:dt_put_tag/1 D => \
+    move a=am_undefined r | deallocate D | return
+move Any r | call_ext_only Arity u$func:erlang:dt_put_tag/1 => \
+    move a=am_undefined r | return
+call_ext Arity u$func:erlang:dt_put_tag/1 => \
+    move a=am_undefined r
+call_ext_last Arity u$func:erlang:dt_put_tag/1 D => \
+    move a=am_undefined r | deallocate D | return
+call_ext_only Arity u$func:erlang:dt_put_tag/1 => \
+    move a=am_undefined r | return
+
+call_ext Arity u$func:erlang:dt_get_tag_data/0 => \
+    move a=am_undefined r
+call_ext_last Arity u$func:erlang:dt_get_tag_data/0 D => \
+    move a=am_undefined r | deallocate D | return
+call_ext_only Arity u$func:erlang:dt_get_tag_data/0 => \
+    move a=am_undefined r | return
+
+move Any r | call_ext Arity u$func:erlang:dt_spread_tag/1 => \
+    move a=am_true r
+move Any r | call_ext_last Arity u$func:erlang:dt_spread_tag/1 D => \
+    move a=am_true r | deallocate D | return
+move Any r | call_ext_only Arity u$func:erlang:dt_spread_tag/1 => \
+    move a=am_true r | return
+call_ext Arity u$func:erlang:dt_spread_tag/1 => \
+    move a=am_true r
+call_ext_last Arity u$func:erlang:dt_spread_tag/1 D => \
+    move a=am_true r | deallocate D | return
+call_ext_only Arity u$func:erlang:dt_spread_tag/1 => \
+    move a=am_true r | return
+
+move Any r | call_ext Arity u$func:erlang:dt_restore_tag/1 => \
+    move a=am_true r
+move Any r | call_ext_last Arity u$func:erlang:dt_restore_tag/1 D => \
+    move a=am_true r | deallocate D | return
+move Any r | call_ext_only Arity u$func:erlang:dt_restore_tag/1 => \
+    move a=am_true r | return
+call_ext Arity u$func:erlang:dt_restore_tag/1 => \
+    move a=am_true r
+call_ext_last Arity u$func:erlang:dt_restore_tag/1 D => \
+    move a=am_true r | deallocate D | return
+call_ext_only Arity u$func:erlang:dt_restore_tag/1 => \
+    move a=am_true r | return
+
+move Any r | call_ext Arity u$func:erlang:dt_prepend_vm_tag_data/1 => \
+    move Any r
+move Any r | call_ext_last Arity u$func:erlang:dt_prepend_vm_tag_data/1 D => \
+    move Any r | deallocate D | return
+move Any r | call_ext_only Arity u$func:erlang:dt_prepend_vm_tag_data/1 => \
+    move Any r | return
+call_ext Arity u$func:erlang:dt_prepend_vm_tag_data/1 =>
+call_ext_last Arity u$func:erlang:dt_prepend_vm_tag_data/1 D => \
+    deallocate D | return
+call_ext_only Arity u$func:erlang:dt_prepend_vm_tag_data/1 => \
+    return
+
+move Any r | call_ext Arity u$func:erlang:dt_append_vm_tag_data/1 => \
+    move Any r
+move Any r | call_ext_last Arity u$func:erlang:dt_append_vm_tag_data/1 D => \
+    move Any r | deallocate D | return
+move Any r | call_ext_only Arity u$func:erlang:dt_append_vm_tag_data/1 => \
+    move Any r | return
+call_ext Arity u$func:erlang:dt_append_vm_tag_data/1 =>
+call_ext_last Arity u$func:erlang:dt_append_vm_tag_data/1 D => \
+    deallocate D | return
+call_ext_only Arity u$func:erlang:dt_append_vm_tag_data/1 => \
+    return
+
+# Can happen after one of the transformations above.
+move Discarded r | move Something r => move Something r
+
+%endif
+
+#
 # The general case for BIFs that have no special instructions.
 # A BIF used in the tail must be followed by a return instruction.
 #
diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c
index 49b6618f73..5ab51fab50 100644
--- a/erts/emulator/beam/utils.c
+++ b/erts/emulator/beam/utils.c
@@ -1697,7 +1697,11 @@ static int do_send_to_logger(Eterm tag, Eterm gleader, char *buf, int len)
 	erts_queue_error_logger_message(from, tuple3, bp);
     }
 #else
-    erts_queue_message(p, NULL /* only used for smp build */, bp, tuple3, NIL);
+    erts_queue_message(p, NULL /* only used for smp build */, bp, tuple3, NIL
+#ifdef USE_VM_PROBES
+		       , NIL
+#endif
+		       );
 #endif
     return 0;
 }
diff --git a/erts/emulator/drivers/common/efile_drv.c b/erts/emulator/drivers/common/efile_drv.c
index a251b064da..b33a8d210b 100644
--- a/erts/emulator/drivers/common/efile_drv.c
+++ b/erts/emulator/drivers/common/efile_drv.c
@@ -100,7 +100,7 @@
 #endif
 #include <stdlib.h>
 
-// Need (NON)BLOCKING macros for sendfile
+/* Need (NON)BLOCKING macros for sendfile */
 #ifndef WANT_NONBLOCKING
 #define WANT_NONBLOCKING
 #endif
@@ -112,6 +112,7 @@
 #include "erl_threads.h"
 #include "zlib.h"
 #include "gzio.h"
+#include "dtrace-wrapper.h" 
 #include <ctype.h>
 #include <sys/types.h>
 
@@ -119,6 +120,39 @@ void erl_exit(int n, char *fmt, ...);
 
 static ErlDrvSysInfo sys_info;
 
+/* For explanation of this var, see comment for same var in erl_async.c */
+static unsigned gcc_optimizer_hack = 0;
+
+#ifdef  USE_VM_PROBES
+
+#define DTRACE_EFILE_BUFSIZ 128
+
+#define DTRACE_INVOKE_SETUP(op) \
+    do { DTRACE3(efile_drv_int_entry, d->sched_i1, d->sched_i2, op); } while (0)
+#define DTRACE_INVOKE_SETUP_BY_NAME(op) \
+    struct t_data *d = (struct t_data *) data ; \
+    DTRACE_INVOKE_SETUP(op)
+#define DTRACE_INVOKE_RETURN(op) \
+    do { DTRACE3(efile_drv_int_return, d->sched_i1, d->sched_i2, \
+                 op); } while (0) ; gcc_optimizer_hack++ ;
+
+/* Assign human-friendlier id numbers to scheduler & I/O worker threads */
+int             dt_driver_idnum = 0;
+int             dt_driver_io_worker_base = 5000;
+erts_mtx_t      dt_driver_mutex;
+pthread_key_t   dt_driver_key;
+
+typedef struct {
+    int         thread_num;
+    Uint64      tag;
+} dt_private;
+
+dt_private *get_dt_private(int);
+#else  /* USE_VM_PROBES */
+#define DTRACE_INVOKE_SETUP(op)            do {} while (0)
+#define DTRACE_INVOKE_SETUP_BY_NAME(op)    do {} while (0)
+#define DTRACE_INVOKE_RETURN(op)           do {} while (0)
+#endif  /* USE_VM_PROBES */
 
 /* #define TRACE 1 */
 #ifdef TRACE
@@ -174,6 +208,9 @@ static ErlDrvSysInfo sys_info;
 
 
 #ifdef FILENAMES_16BIT
+#ifdef USE_VM_PROBES
+#error 16bit characters in filenames and dtrace in combination is not supported.
+#endif
 #  define FILENAME_BYTELEN(Str) filename_len_16bit(Str)
 #  define FILENAME_COPY(To,From) filename_cpy_16bit((To),(From)) 
 #  define FILENAME_CHARSIZE 2
@@ -289,6 +326,10 @@ typedef struct {
     ErlDrvPDL       q_mtx;    /* Mutex for the driver queue, known by the emulator. Also used for
 				 mutual exclusion when accessing field(s) below. */
     size_t          write_buffered;
+#ifdef USE_VM_PROBES
+    int             idnum;      /* Unique ID # for this driver thread/desc */
+    char            port_str[DTRACE_TERM_BUF_SIZE];
+#endif
 } file_descriptor;
 
 
@@ -386,6 +427,11 @@ struct t_data
     void         (*free)(void *);
     int            again;
     int            reply;
+#ifdef  USE_VM_PROBES
+    int               sched_i1;
+    Uint64            sched_i2;
+    char              sched_utag[DTRACE_EFILE_BUFSIZ+1];
+#endif
     int            result_ok;
     Efile_error    errInfo;
     int            flags;
@@ -458,8 +504,6 @@ struct t_data
     char b[1];
 };
 
-
-
 #define EF_ALLOC(S)		driver_alloc((S))
 #define EF_REALLOC(P, S)	driver_realloc((P), (S))
 #define EF_SAFE_ALLOC(S)	ef_safe_alloc((S))
@@ -488,7 +532,7 @@ static void *ef_safe_realloc(void *op, Uint s)
  * ErlIOVec manipulation functions.
  */
 
-/* char EV_CHAR(ErlIOVec *ev, int p, int q) */
+/* char EV_CHAR_P(ErlIOVec *ev, int p, int q) */
 #define EV_CHAR_P(ev, p, q)                   \
     (((char *)(ev)->iov[(q)].iov_base) + (p))
 
@@ -683,6 +727,11 @@ file_init(void)
 			    : 0);
     driver_system_info(&sys_info, sizeof(ErlDrvSysInfo));
 
+#ifdef  USE_VM_PROBES
+    erts_mtx_init(&dt_driver_mutex, "efile_drv dtrace mutex");
+    pthread_key_create(&dt_driver_key, NULL);
+#endif  /* USE_VM_PROBES */
+
     return 0;
 }
 
@@ -722,6 +771,10 @@ file_start(ErlDrvPort port, char* command)
     desc->write_error = 0;
     MUTEX_INIT(desc->q_mtx, port); /* Refc is one, referenced by emulator now */
     desc->write_buffered = 0;
+#ifdef  USE_VM_PROBES
+    dtrace_drvport_str(port, desc->port_str);
+    get_dt_private(0);           /* throw away return value */
+#endif  /* USE_VM_PROBES */
     return (ErlDrvData) desc;
 }
 
@@ -741,8 +794,10 @@ static void do_close(int flags, SWord fd) {
 static void invoke_close(void *data)
 {
     struct t_data *d = (struct t_data *) data;
+    DTRACE_INVOKE_SETUP(FILE_CLOSE);
     d->again = 0;
     do_close(d->flags, d->fd);
+    DTRACE_INVOKE_RETURN(FILE_CLOSE);
 }
 
 /*********************************************************************
@@ -972,49 +1027,63 @@ static void invoke_name(void *data, int (*f)(Efile_error *, char *))
 
 static void invoke_mkdir(void *data)
 {
+    DTRACE_INVOKE_SETUP_BY_NAME(FILE_MKDIR);
     invoke_name(data, efile_mkdir);
+    DTRACE_INVOKE_RETURN(FILE_MKDIR);
 }
 
 static void invoke_rmdir(void *data)
 {
+    DTRACE_INVOKE_SETUP_BY_NAME(FILE_RMDIR);
     invoke_name(data, efile_rmdir);
+    DTRACE_INVOKE_RETURN(FILE_RMDIR);
 }
 
 static void invoke_delete_file(void *data)
 {
+    DTRACE_INVOKE_SETUP_BY_NAME(FILE_DELETE);
     invoke_name(data, efile_delete_file);
+    DTRACE_INVOKE_RETURN(FILE_DELETE);
 }
 
 static void invoke_chdir(void *data)
 {
+    DTRACE_INVOKE_SETUP_BY_NAME(FILE_CHDIR);
     invoke_name(data, efile_chdir);
+    DTRACE_INVOKE_RETURN(FILE_CHDIR);
 }
 
 static void invoke_fdatasync(void *data)
 {
     struct t_data *d = (struct t_data *) data;
     int fd = (int) d->fd;
+    DTRACE_INVOKE_SETUP(FILE_FDATASYNC);
 
     d->again = 0;
     d->result_ok = efile_fdatasync(&d->errInfo, fd);
+    DTRACE_INVOKE_RETURN(FILE_FDATASYNC);
 }
 
 static void invoke_fsync(void *data)
 {
     struct t_data *d = (struct t_data *) data;
     int fd = (int) d->fd;
+    DTRACE_INVOKE_SETUP(FILE_FSYNC);
 
     d->again = 0;
     d->result_ok = efile_fsync(&d->errInfo, fd);
+    DTRACE_INVOKE_RETURN(FILE_FSYNC);
 }
 
 static void invoke_truncate(void *data)
 {
     struct t_data *d = (struct t_data *) data;
     int fd = (int) d->fd;
+    DTRACE_INVOKE_SETUP(FILE_TRUNCATE);
 
     d->again = 0;
     d->result_ok = efile_truncate_file(&d->errInfo, &fd, d->flags);
+    DTRACE_INVOKE_RETURN(FILE_TRUNCATE);
 }
 
 static void invoke_read(void *data)
@@ -1022,6 +1091,7 @@ static void invoke_read(void *data)
     struct t_data *d = (struct t_data *) data;
     int status, segment;
     size_t size, read_size;
+    DTRACE_INVOKE_SETUP(FILE_READ);
 
     segment = d->again && d->c.read.bin_size >= 2*FILE_SEGMENT_READ;
     if (segment) {
@@ -1056,6 +1126,7 @@ static void invoke_read(void *data)
     } else {
 	d->again = 0;
     }
+    DTRACE_INVOKE_RETURN(FILE_READ);
 }
 
 static void free_read(void *data)
@@ -1072,6 +1143,7 @@ static void invoke_read_line(void *data)
     int status;
     size_t read_size;
     int local_loop = (d->again == 0);
+    DTRACE_INVOKE_SETUP(FILE_READ_LINE);
 
     do {
 	size_t size = (d->c.read_line.binp)->orig_size - 
@@ -1163,6 +1235,7 @@ static void invoke_read_line(void *data)
 	    break;
 	}
     } while (local_loop);
+    DTRACE_INVOKE_RETURN(FILE_READ_LINE);
 }
 
 static void free_read_line(void *data)
@@ -1178,6 +1251,7 @@ static void invoke_read_file(void *data)
     struct t_data *d = (struct t_data *) data;
     size_t read_size;
     int chop;
+    DTRACE_INVOKE_SETUP(FILE_READ_FILE);
     
     if (! d->c.read_file.binp) { /* First invocation only */
 	int fd;
@@ -1214,12 +1288,14 @@ static void invoke_read_file(void *data)
 		   &read_size);
     if (d->result_ok) {
 	d->c.read_file.offset += read_size;
-	if (chop) return; /* again */
+	if (chop) goto chop_done; /* again */
     }
  close:
     efile_closefile((int) d->fd);
  done:
     d->again = 0;
+ chop_done:
+    DTRACE_INVOKE_RETURN(FILE_READ_FILE);
 }
 
 static void free_read_file(void *data)
@@ -1239,6 +1315,7 @@ static void invoke_preadv(void *data)
     ErlIOVec        *ev = &c->eiov;
     size_t           bytes_read_so_far = 0;
     unsigned char   *p = (unsigned char *)ev->iov[0].iov_base + 4+4+8*c->cnt;
+    DTRACE_INVOKE_SETUP(FILE_PREADV);
 
     while (c->cnt < c->n) {
 	size_t read_size = ev->iov[1 + c->cnt].iov_len - c->size;
@@ -1260,7 +1337,7 @@ static void invoke_preadv(void *data)
 	    bytes_read_so_far += bytes_read;
 	    if (chop && bytes_read == read_size) {
 		c->size += bytes_read;
-		return;
+		goto done;
 	    }
 	    ASSERT(bytes_read <= read_size);
 	    ev->iov[1 + c->cnt].iov_len = bytes_read + c->size;
@@ -1271,7 +1348,7 @@ static void invoke_preadv(void *data)
 	    if (d->again 
 		&& bytes_read_so_far >= FILE_SEGMENT_READ
 		&& c->cnt < c->n) {
-		return;
+		goto done;
 	    }
 	} else {
 	    /* In case of a read error, ev->size will not be correct,
@@ -1282,6 +1359,8 @@ static void invoke_preadv(void *data)
 	}
     }					
     d->again = 0;
+ done:
+    DTRACE_INVOKE_RETURN(FILE_PREADV);
 }
 
 static void free_preadv(void *data) {
@@ -1303,6 +1382,7 @@ static void invoke_ipread(void *data)
     size_t bytes_read = 0;
     char buf[2*sizeof(Uint32)];
     Uint32 offset, size;
+    DTRACE_INVOKE_SETUP(FILE_IPREAD);
     
     /* Read indirection header */
     if (! efile_pread(&d->errInfo, (int) d->fd, c->offsets[0], 
@@ -1341,14 +1421,17 @@ static void invoke_ipread(void *data)
     /* Read data block */
     d->invoke = invoke_preadv;
     invoke_preadv(data);
+    DTRACE_INVOKE_RETURN(FILE_IPREAD);
     return;
  error:
     d->result_ok = 0;
     d->again = 0;
+    DTRACE_INVOKE_RETURN(FILE_IPREAD);
     return;
  done:
     d->result_ok = !0;
     d->again = 0;
+    DTRACE_INVOKE_RETURN(FILE_IPREAD);
 }
 
 /* invoke_writev and invoke_pwritev are the only thread functions that
@@ -1371,6 +1454,7 @@ static void invoke_writev(void *data) {
     size_t         size;
     size_t         p;
     int            segment;
+    DTRACE_INVOKE_SETUP(FILE_WRITE);
 
     segment = d->again && d->c.writev.size >= 2*FILE_SEGMENT_WRITE;
     if (segment) {
@@ -1444,6 +1528,7 @@ static void invoke_writev(void *data) {
 	TRACE_F(("w%lu", (unsigned long)size));
 
     }
+    DTRACE_INVOKE_RETURN(FILE_WRITE);
 }
 
 static void free_writev(void *data) {
@@ -1457,34 +1542,40 @@ static void free_writev(void *data) {
 static void invoke_pwd(void *data)
 {
     struct t_data *d = (struct t_data *) data;
+    DTRACE_INVOKE_SETUP(FILE_PWD);
 
     d->again = 0;
     d->result_ok = efile_getdcwd(&d->errInfo,d->drive, d->b+1,
 				 RESBUFSIZE-1);
+    DTRACE_INVOKE_RETURN(FILE_PWD);
 }
 
 static void invoke_readlink(void *data)
 {
     struct t_data *d = (struct t_data *) data;
     char resbuf[RESBUFSIZE];	/* Result buffer. */
+    DTRACE_INVOKE_SETUP(FILE_READLINK);
 
     d->again = 0;
     d->result_ok = efile_readlink(&d->errInfo, d->b, resbuf+1,
 				  RESBUFSIZE-1);
     if (d->result_ok != 0)
 	FILENAME_COPY((char *) d->b + 1, resbuf+1);
+    DTRACE_INVOKE_RETURN(FILE_READLINK);
 }
 
 static void invoke_altname(void *data)
 {
     struct t_data *d = (struct t_data *) data;
     char resbuf[RESBUFSIZE];	/* Result buffer. */
+    DTRACE_INVOKE_SETUP(FILE_ALTNAME);
 
     d->again = 0;
     d->result_ok = efile_altname(&d->errInfo, d->b, resbuf+1,
 				  RESBUFSIZE-1);
     if (d->result_ok != 0)
 	FILENAME_COPY((char *) d->b + 1, resbuf+1);
+    DTRACE_INVOKE_RETURN(FILE_ALTNAME);
 }
 
 static void invoke_pwritev(void *data) {
@@ -1497,6 +1588,7 @@ static void invoke_pwritev(void *data) {
     size_t            p;
     int               segment;
     size_t            size, write_size;
+    DTRACE_INVOKE_SETUP(FILE_PWRITEV);
 
     segment = d->again && c->size >= 2*FILE_SEGMENT_WRITE;
     if (segment) {
@@ -1576,6 +1668,7 @@ static void invoke_pwritev(void *data) {
     }
  done:
     EF_FREE(iov); /* Free our copy of the vector, nothing to restore */
+    DTRACE_INVOKE_RETURN(FILE_PWRITEV);
 }
 
 static void free_pwritev(void *data) {
@@ -1591,9 +1684,14 @@ static void invoke_flstat(void *data)
 {
     struct t_data *d = (struct t_data *) data;
 
+    DTRACE3(efile_drv_int_entry, d->sched_i1, d->sched_i2,
+            d->command == FILE_LSTAT ? FILE_LSTAT : FILE_FSTAT);
     d->again = 0;
     d->result_ok = efile_fileinfo(&d->errInfo, &d->info,
 				  d->b, d->command == FILE_LSTAT);
+    DTRACE3(efile_drv_int_entry, d->sched_i1, d->sched_i2,
+            d->command == FILE_LSTAT ? FILE_LSTAT : FILE_FSTAT);
+    gcc_optimizer_hack++;
 }
 
 static void invoke_link(void *data)
@@ -1601,10 +1699,12 @@ static void invoke_link(void *data)
     struct t_data *d = (struct t_data *) data;
     char *name = d->b;
     char *new_name;
+    DTRACE_INVOKE_SETUP(FILE_LINK);
 
     d->again = 0;
     new_name = name+FILENAME_BYTELEN(name)+FILENAME_CHARSIZE;
     d->result_ok = efile_link(&d->errInfo, name, new_name);
+    DTRACE_INVOKE_RETURN(FILE_LINK);
 }
 
 static void invoke_symlink(void *data)
@@ -1612,10 +1712,12 @@ static void invoke_symlink(void *data)
     struct t_data *d = (struct t_data *) data;
     char *name = d->b;
     char *new_name;
+    DTRACE_INVOKE_SETUP(FILE_SYMLINK);
 
     d->again = 0;
     new_name = name+FILENAME_BYTELEN(name)+FILENAME_CHARSIZE;
     d->result_ok = efile_symlink(&d->errInfo, name, new_name);
+    DTRACE_INVOKE_RETURN(FILE_SYMLINK);
 }
 
 static void invoke_rename(void *data)
@@ -1623,24 +1725,29 @@ static void invoke_rename(void *data)
     struct t_data *d = (struct t_data *) data;
     char *name = d->b;
     char *new_name;
+    DTRACE_INVOKE_SETUP(FILE_RENAME);
 
     d->again = 0;
     new_name = name+FILENAME_BYTELEN(name)+FILENAME_CHARSIZE;
     d->result_ok = efile_rename(&d->errInfo, name, new_name);
+    DTRACE_INVOKE_RETURN(FILE_RENAME);
 }
 
 static void invoke_write_info(void *data)
 {
     struct t_data *d = (struct t_data *) data;
+    DTRACE_INVOKE_SETUP(FILE_WRITE_INFO);
 
     d->again = 0;
     d->result_ok = efile_write_info(&d->errInfo, &d->info, d->b);
+    DTRACE_INVOKE_RETURN(FILE_WRITE_INFO);
 }
 
 static void invoke_lseek(void *data)
 {
     struct t_data *d = (struct t_data *) data;
     int status;
+    DTRACE_INVOKE_SETUP(FILE_LSEEK);
 
     d->again = 0;
     if (d->flags & EFILE_COMPRESSED) {
@@ -1665,6 +1772,7 @@ static void invoke_lseek(void *data)
 			    &d->c.lseek.location);
     }
     d->result_ok = status;
+    DTRACE_INVOKE_RETURN(FILE_LSEEK);
 }
 
 static void invoke_readdir(void *data)
@@ -1675,6 +1783,7 @@ static void invoke_readdir(void *data)
     size_t n = 0, total = 0;
     struct t_readdir_buf *b = NULL;
     int res = 0;
+    DTRACE_INVOKE_SETUP(FILE_READDIR);
 
     d->again = 0;
     d->errInfo.posix_errno = 0;
@@ -1710,13 +1819,14 @@ static void invoke_readdir(void *data)
     } while(res);
 
     d->result_ok = (d->errInfo.posix_errno == 0);
+    DTRACE_INVOKE_RETURN(FILE_READDIR);
 }
 
 static void invoke_open(void *data)
 {
     struct t_data *d = (struct t_data *) data;
-    
     int status = 1;		/* Status of open call. */
+    DTRACE_INVOKE_SETUP(FILE_OPEN);
 
     d->again = 0;
     if ((d->flags & EFILE_COMPRESSED) == 0) {
@@ -1749,6 +1859,7 @@ static void invoke_open(void *data)
     }
 
     d->result_ok = status;
+    DTRACE_INVOKE_RETURN(FILE_OPEN);
 }
 
 static void invoke_fadvise(void *data)
@@ -1758,9 +1869,11 @@ static void invoke_fadvise(void *data)
     off_t offset = (off_t) d->c.fadvise.offset;
     off_t length = (off_t) d->c.fadvise.length;
     int advise = (int) d->c.fadvise.advise;
+    DTRACE_INVOKE_SETUP(FILE_FADVISE);
 
     d->again = 0;
     d->result_ok = efile_fadvise(&d->errInfo, fd, offset, length, advise);
+    DTRACE_INVOKE_RETURN(FILE_FADVISE);
 }
 
 #ifdef HAVE_SENDFILE
@@ -1841,6 +1954,7 @@ static void free_readdir(void *data)
 {
     struct t_data *d = (struct t_data *) data;
     struct t_readdir_buf *b1 = d->c.read_dir.first_buf;
+
     while (b1) {
 	struct t_readdir_buf *b2 = b1;
 	b1 = b1->next;
@@ -1909,12 +2023,16 @@ static void cq_execute(file_descriptor *desc) {
     DRIVER_ASYNC(d->level, desc, d->invoke, void_ptr=d, d->free);
 }
 
-static int async_write(file_descriptor *desc, int *errp,
-		       int reply, Uint32 reply_size) {
+static struct t_data *async_write(file_descriptor *desc, int *errp,
+		       int reply, Uint32 reply_size
+#ifdef USE_VM_PROBES
+		       ,Sint64 *dt_i1, Sint64 *dt_i2, Sint64 *dt_i3
+#endif
+) {
     struct t_data *d;
     if (! (d = EF_ALLOC(sizeof(struct t_data) - 1))) {
 	if (errp) *errp = ENOMEM;
-	return -1;
+	return NULL;
     }
     TRACE_F(("w%lu", (unsigned long)desc->write_buffered));
     d->command = FILE_WRITE;
@@ -1923,6 +2041,13 @@ static int async_write(file_descriptor *desc, int *errp,
     d->c.writev.port = desc->port;
     d->c.writev.q_mtx = desc->q_mtx;
     d->c.writev.size = desc->write_buffered;
+#ifdef USE_VM_PROBES
+    if (dt_i1 != NULL) {
+        *dt_i1 = d->fd;
+        *dt_i2 = d->flags;
+        *dt_i3 = d->c.writev.size;
+    }
+#endif
     d->reply = reply;
     d->c.writev.free_size = 0;
     d->c.writev.reply_size = reply_size;
@@ -1931,18 +2056,49 @@ static int async_write(file_descriptor *desc, int *errp,
     d->level = 1;
     cq_enq(desc, d);
     desc->write_buffered = 0;
-    return 0;
+    return d;
 }
 
-static int flush_write(file_descriptor *desc, int *errp) {
-    int    result;
+static int flush_write(file_descriptor *desc, int *errp
+#ifdef USE_VM_PROBES
+                       , dt_private *dt_priv, char *dt_utag
+#endif
+) {
+    int    result = 0;
+#ifdef USE_VM_PROBES
+    Sint64 dt_i1 = 0, dt_i2 = 0, dt_i3 = 0;
+#endif
+    struct t_data *d = NULL;
+
     MUTEX_LOCK(desc->q_mtx);
     if (desc->write_buffered > 0) {
-	result = async_write(desc, errp, 0, 0);
-    } else {
-	result = 0;
+	if ((d = async_write(desc, errp, 0, 0
+#ifdef USE_VM_PROBES
+			     ,&dt_i1, &dt_i2, &dt_i3
+#endif
+			     )) == NULL) {
+            result = -1;
+        }
     }
     MUTEX_UNLOCK(desc->q_mtx);
+#ifdef USE_VM_PROBES
+    if (d != NULL) {
+        d->sched_i1 = dt_priv->thread_num;
+        d->sched_i2 = dt_priv->tag;
+        d->sched_utag[0] = '\0';
+        if (dt_utag != NULL) {
+            if (dt_utag[0] == '\0') {
+                dt_utag = NULL;
+            } else {
+                strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1);
+                d->sched_utag[sizeof(d->sched_utag) - 1] = '\0';
+            }
+        }
+        DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++,
+                 dt_utag, FILE_WRITE,
+                 NULL, NULL, dt_i1, dt_i2, dt_i3, 0, desc->port_str);
+    }
+#endif /* USE_VM_PROBES */
     return result;
 }
 
@@ -1955,9 +2111,17 @@ static int check_write_error(file_descriptor *desc, int *errp) {
     return 0;
 }
 
-static int flush_write_check_error(file_descriptor *desc, int *errp) {
+static int flush_write_check_error(file_descriptor *desc, int *errp
+#ifdef USE_VM_PROBES
+                                   , dt_private *dt_priv, char *dt_utag
+#endif
+				   ) {
     int r;
-    if ( (r = flush_write(desc, errp)) != 0) {
+    if ( (r = flush_write(desc, errp
+#ifdef USE_VM_PROBES
+			  , dt_priv, dt_utag
+#endif
+			  )) != 0) {
 	check_write_error(desc, NULL);
 	return r;
     } else {
@@ -1965,12 +2129,16 @@ static int flush_write_check_error(file_descriptor *desc, int *errp) {
     }
 }
 
-static int async_lseek(file_descriptor *desc, int *errp, int reply, 
-		       Sint64 offset, int origin) {
+static struct t_data *async_lseek(file_descriptor *desc, int *errp, int reply,
+				  Sint64 offset, int origin
+#ifdef USE_VM_PROBES
+				  , Sint64 *dt_i1, Sint64 *dt_i2, Sint64 *dt_i3
+#endif
+				  ) {
     struct t_data *d;
     if (! (d = EF_ALLOC(sizeof(struct t_data)))) {
 	*errp = ENOMEM;
-	return -1;
+	return NULL;
     }
     d->flags = desc->flags;
     d->fd = desc->fd;
@@ -1978,11 +2146,18 @@ static int async_lseek(file_descriptor *desc, int *errp, int reply,
     d->reply = reply;
     d->c.lseek.offset = offset;
     d->c.lseek.origin = origin;
+#ifdef USE_VM_PROBES
+    if (dt_i1 != NULL) {
+        *dt_i1 = d->fd;
+        *dt_i2 = d->c.lseek.offset;
+        *dt_i3 = d->c.lseek.origin;
+    }
+#endif
     d->invoke = invoke_lseek;
     d->free = free_data;
     d->level = 1;
     cq_enq(desc, d);
-    return 0;
+    return d;
 }
 
 static void flush_read(file_descriptor *desc) {
@@ -1994,18 +2169,45 @@ static void flush_read(file_descriptor *desc) {
     }
 }
 
-static int lseek_flush_read(file_descriptor *desc, int *errp) {
+static int lseek_flush_read(file_descriptor *desc, int *errp
+#ifdef USE_VM_PROBES
+			    ,dt_private *dt_priv, char *dt_utag
+#endif
+			    ) {
     int r = 0;
     size_t read_size = desc->read_size;
+#ifdef USE_VM_PROBES
+    Sint64 dt_i1 = 0, dt_i2 = 0, dt_i3 = 0;
+#endif
+    struct t_data *d;
+
+    flush_read(desc);
     if (read_size != 0) {
-	flush_read(desc);
-	if ((r = async_lseek(desc, errp, 0, 
-			     -((ssize_t)read_size), EFILE_SEEK_CUR)) 
-	    < 0) {
-	    return r;
-	}
-    } else {
-	flush_read(desc);
+	if ((d = async_lseek(desc, errp, 0,
+                             -((ssize_t)read_size), EFILE_SEEK_CUR
+#ifdef USE_VM_PROBES
+			     , &dt_i1, &dt_i2, &dt_i3
+#endif
+			     )) == NULL) {
+            r = -1;
+        } else {
+#ifdef USE_VM_PROBES
+            d->sched_i1 = dt_priv->thread_num;
+            d->sched_i2 = dt_priv->tag;
+            d->sched_utag[0] = '\0';
+            if (dt_utag != NULL) {
+                if (dt_utag[0] == '\0') {
+                    dt_utag = NULL;
+                } else {
+                    strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1);
+                    d->sched_utag[sizeof(d->sched_utag) - 1] = '\0';
+                }
+            }
+            DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++,
+                     dt_utag, FILE_LSEEK,
+                     NULL, NULL, dt_i1, dt_i2, dt_i3, 0, desc->port_str);
+#endif /* USE_VM_PROBES */
+        }
     }
     return r;
 }
@@ -2022,11 +2224,23 @@ file_async_ready(ErlDrvData e, ErlDrvThreadData data)
     struct t_data *d = (struct t_data *) data;
     char header[5];		/* result code + count */
     char resbuf[RESBUFSIZE];	/* Result buffer. */
-    
+#ifdef  USE_VM_PROBES
+    int sched_i1 = d->sched_i1, sched_i2 = d->sched_i2, command = d->command,
+        result_ok = d->result_ok,
+        posix_errno = d->result_ok ? 0 : d->errInfo.posix_errno;
+    DTRACE_CHARBUF(sched_utag, DTRACE_EFILE_BUFSIZ+1);
+
+    sched_utag[0] = '\0';
+    if (DTRACE_ENABLED(efile_drv_return)) {
+        strncpy(sched_utag, d->sched_utag, DTRACE_EFILE_BUFSIZ);
+        sched_utag[DTRACE_EFILE_BUFSIZ] = '\0';
+    }
+#endif  /* USE_VM_PROBES */
 
     TRACE_C('r');
 
     if (try_again(desc, d)) {
+        /* DTRACE TODO: what kind of probe makes sense here? */
 	return;
     }
 
@@ -2224,6 +2438,9 @@ file_async_ready(ErlDrvData e, ErlDrvThreadData data)
 	  if (d->reply) {
 	      TRACE_C('K');
 	      reply_ok(desc);
+#ifdef USE_VM_PROBES
+              result_ok = 1;
+#endif
 	  }
 	  free_data(data);
 	  break;
@@ -2279,6 +2496,8 @@ file_async_ready(ErlDrvData e, ErlDrvThreadData data)
       default:
 	abort();
     }
+    DTRACE6(efile_drv_return, sched_i1, sched_i2, sched_utag,
+            command, result_ok, posix_errno);
     if (desc->write_buffered != 0 && desc->timer_state == timer_idle) {
 	desc->timer_state = timer_write;
 	driver_set_timer(desc->port, desc->write_delay);
@@ -2301,7 +2520,15 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
     char* name;			/* Points to the filename in buf. */
     int command;
     struct t_data *d = NULL;
-
+#ifdef  USE_VM_PROBES
+    char *dt_utag = NULL;
+    char *dt_s1 = NULL, *dt_s2 = NULL;
+    Sint64 dt_i1 = 0;
+    Sint64 dt_i2 = 0;
+    Sint64 dt_i3 = 0;
+    Sint64 dt_i4 = 0;
+    dt_private *dt_priv = get_dt_private(0);
+#endif  /* USE_VM_PROBES */
 
     TRACE_C('o');
 
@@ -2316,6 +2543,10 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE);
 	
 	FILENAME_COPY(d->b, name);
+#ifdef USE_VM_PROBES
+	dt_s1 = d->b;
+	dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
+#endif
 	d->command = command;
 	d->invoke = invoke_mkdir;
 	d->free = free_data;
@@ -2327,6 +2558,10 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE);
 	
 	FILENAME_COPY(d->b, name);
+#ifdef USE_VM_PROBES
+	dt_s1 = d->b;
+	dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
+#endif
 	d->command = command;
 	d->invoke = invoke_rmdir;
 	d->free = free_data;
@@ -2338,6 +2573,10 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE);
 	
 	FILENAME_COPY(d->b, name);
+#ifdef USE_VM_PROBES
+	dt_s1 = d->b;
+	dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
+#endif
 	d->command = command;
 	d->invoke = invoke_delete_file;
 	d->free = free_data;
@@ -2355,6 +2594,11 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	
 	    FILENAME_COPY(d->b, name);
 	    FILENAME_COPY(d->b + namelen, new_name);
+#ifdef USE_VM_PROBES
+	    dt_s1 = d->b;
+	    dt_s2 = d->b + namelen;
+	    dt_utag = buf + namelen + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
+#endif
 	    d->flags = desc->flags;
 	    d->fd = fd;
 	    d->command = command;
@@ -2368,6 +2612,10 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE);
 	
 	FILENAME_COPY(d->b, name);
+#ifdef USE_VM_PROBES
+	dt_s1 = d->b;
+	dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
+#endif
 	d->command = command;
 	d->invoke = invoke_chdir;
 	d->free = free_data;
@@ -2379,6 +2627,9 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	    d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + RESBUFSIZE + 1);
 	
 	    d->drive = *(uchar*)buf;
+#ifdef USE_VM_PROBES
+	    dt_utag = buf + 1;
+#endif
 	    d->command = command;
 	    d->invoke = invoke_pwd;
 	    d->free = free_data;
@@ -2394,6 +2645,10 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 			      FILENAME_CHARSIZE);
 	
 	    FILENAME_COPY(d->b, name);
+#ifdef USE_VM_PROBES
+	    dt_s1 = d->b;
+	    dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
+#endif
 	    d->dir_handle = NULL;
 	    d->command = command;
 	    d->invoke = invoke_readdir;
@@ -2418,6 +2673,10 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	    dir_handle          = NULL;
 	    resbuf[0]           = FILE_RESP_LFNAME;
 
+#ifdef USE_VM_PROBES
+	    dt_s1 = name;
+	    dt_utag = name + FILENAME_BYTELEN(name) + FILENAME_CHARSIZE;
+#endif
 	    /* Fill the buffer with multiple directory listings before sending it to the
 	     * receiving process. READDIR_CHUNKS is minimum number of files sent to the
 	     * receiver.
@@ -2451,6 +2710,17 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 		reply_error(desc, &errInfo);
 		return;
 	    }
+#ifdef USE_VM_PROBES
+	    if (dt_utag != NULL && dt_utag[0] == '\0') {
+                dt_utag = NULL;
+            } 
+
+	    DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag,
+		     dt_utag, command, name, dt_s2,
+		     dt_i1, dt_i2, dt_i3, dt_i4, desc->port_str);
+	    DTRACE6(efile_drv_return, dt_priv->thread_num, dt_priv->tag++, 
+		    dt_utag, command, 1, 0);
+#endif
 	    TRACE_C('R');
 	    driver_output2(desc->port, resbuf, 1, NULL, 0);
 	    return;
@@ -2463,6 +2733,11 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	    d->flags = get_int32((uchar*)buf);
 	    name = buf+4;
 	    FILENAME_COPY(d->b, name);
+#ifdef USE_VM_PROBES
+	    dt_i1 = d->flags;
+	    dt_s1 = d->b;
+	    dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE;
+#endif
 	    d->command = command;
 	    d->invoke = invoke_open;
 	    d->free = free_data;
@@ -2475,6 +2750,10 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	    d = EF_SAFE_ALLOC(sizeof(struct t_data));
 	    
 	    d->fd = fd;
+#ifdef USE_VM_PROBES
+	    dt_utag = name;
+	    dt_i1 = fd;
+#endif
 	    d->command = command;
 	    d->invoke = invoke_fdatasync;
 	    d->free = free_data;
@@ -2487,6 +2766,10 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	    d = EF_SAFE_ALLOC(sizeof(struct t_data));
 	    
 	    d->fd = fd;
+#ifdef USE_VM_PROBES
+	    dt_utag = name;
+	    dt_i1 = fd;
+#endif
 	    d->command = command;
 	    d->invoke = invoke_fsync;
 	    d->free = free_data;
@@ -2503,6 +2786,14 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	    
 	    FILENAME_COPY(d->b, name);
 	    d->fd = fd;
+#ifdef USE_VM_PROBES
+	    dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE;
+	    if (command == FILE_LSTAT) {
+		dt_s1 = d->b;
+	    } else {
+		dt_i1 = fd;
+	    }
+#endif
 	    d->command = command;
 	    d->invoke = invoke_flstat;
 	    d->free = free_data;
@@ -2516,6 +2807,11 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	    
 	    d->flags = desc->flags;
 	    d->fd = fd;
+#ifdef USE_VM_PROBES
+	    dt_utag = name;
+	    dt_i1 = fd;
+	    dt_i2 = d->flags;
+#endif
 	    d->command = command;
 	    d->invoke = invoke_truncate;
 	    d->free = free_data;
@@ -2536,6 +2832,13 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	    d->info.cTime      = (time_t)((Sint64)get_int64(buf +  7 * 4));
 
 	    FILENAME_COPY(d->b, buf + 9*4);
+#ifdef USE_VM_PROBES
+	    dt_i1              = d->info.mode;
+	    dt_i2              = d->info.uid;
+	    dt_i3              = d->info.gid;
+	    dt_s1 = d->b;
+	    dt_utag = buf + 9 * 4 + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE;
+#endif
 	    d->command = command;
 	    d->invoke = invoke_write_info;
 	    d->free = free_data;
@@ -2548,6 +2851,10 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	    d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + RESBUFSIZE + 1);
 	
 	    FILENAME_COPY(d->b, name);
+#ifdef USE_VM_PROBES
+	    dt_s1 = d->b;
+	    dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE;
+#endif
 	    d->command = command;
 	    d->invoke = invoke_readlink;
 	    d->free = free_data;
@@ -2559,6 +2866,10 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	{
 	    d = EF_SAFE_ALLOC(sizeof(struct t_data) - 1 + RESBUFSIZE + 1);
 	    FILENAME_COPY(d->b, name);
+#ifdef USE_VM_PROBES
+	    dt_s1 = d->b;
+	    dt_utag = name + FILENAME_BYTELEN(d->b) + FILENAME_CHARSIZE;
+#endif
 	    d->command = command;
 	    d->invoke = invoke_altname;
 	    d->free = free_data;
@@ -2579,6 +2890,11 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	
 	    FILENAME_COPY(d->b, name);
 	    FILENAME_COPY(d->b + namelen, new_name);
+#ifdef USE_VM_PROBES
+	    dt_s1 = d->b;
+	    dt_s2 = d->b + namelen;
+	    dt_utag = buf + namelen + FILENAME_BYTELEN(dt_s2) + FILENAME_CHARSIZE;
+#endif
 	    d->flags = desc->flags;
 	    d->fd = fd;
 	    d->command = command;
@@ -2600,6 +2916,11 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 	
 	    FILENAME_COPY(d->b, name);
 	    FILENAME_COPY(d->b + namelen, new_name);
+#ifdef USE_VM_PROBES
+	    dt_s1 = d->b;
+	    dt_s2 = d->b + namelen;
+	    dt_utag = buf + namelen + FILENAME_BYTELEN(dt_s2) + FILENAME_CHARSIZE;
+#endif
 	    d->flags = desc->flags;
 	    d->fd = fd;
 	    d->command = command;
@@ -2621,6 +2942,13 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
         d->c.fadvise.offset = get_int64((uchar*) buf);
         d->c.fadvise.length = get_int64(((uchar*) buf) + sizeof(Sint64));
         d->c.fadvise.advise = get_int32(((uchar*) buf) + 2 * sizeof(Sint64));
+#ifdef USE_VM_PROBES
+        dt_i1 = d->fd;
+        dt_i2 = d->c.fadvise.offset;
+        dt_i3 = d->c.fadvise.length;
+        dt_i4 = d->c.fadvise.advise;
+        dt_utag = buf + 3 * sizeof(Sint64);
+#endif
         goto done;
     }
 
@@ -2634,6 +2962,22 @@ file_output(ErlDrvData e, char* buf, ErlDrvSizeT count)
 
  done:
     if (d) {
+#ifdef USE_VM_PROBES
+	d->sched_i1 = dt_priv->thread_num;
+	d->sched_i2 = dt_priv->tag;
+	d->sched_utag[0] = '\0';
+	if (dt_utag != NULL) {
+	    if (dt_utag[0] == '\0') {
+		dt_utag = NULL;
+	    } else {
+		strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1);
+		d->sched_utag[sizeof(d->sched_utag) - 1] = '\0';
+	    }
+	}
+	DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++,
+		 dt_utag, command, dt_s1, dt_s2,
+		 dt_i1, dt_i2, dt_i3, dt_i4, desc->port_str);
+#endif
 	cq_enq(desc, d);
     }
 }
@@ -2647,6 +2991,9 @@ file_flush(ErlDrvData e) {
 #ifdef DEBUG
     int r;
 #endif
+#ifdef  USE_VM_PROBES
+    dt_private *dt_priv = get_dt_private(dt_driver_io_worker_base);
+#endif
 
     TRACE_C('f');
 
@@ -2657,7 +3004,11 @@ file_flush(ErlDrvData e) {
 #ifdef DEBUG
     r = 
 #endif
-	flush_write(desc, NULL);
+         flush_write(desc, NULL
+#ifdef USE_VM_PROBES
+		     , dt_priv, (desc->d == NULL) ? NULL : desc->d->sched_utag
+#endif
+		     );
     /* Only possible reason for bad return value is ENOMEM, and 
      * there is nobody to tell...
      */
@@ -2699,6 +3050,9 @@ static void
 file_timeout(ErlDrvData e) {
     file_descriptor *desc = (file_descriptor *)e;
     enum e_timer timer_state = desc->timer_state;
+#ifdef  USE_VM_PROBES
+    dt_private *dt_priv = get_dt_private(dt_driver_io_worker_base);
+#endif
 
     TRACE_C('t');
 
@@ -2716,7 +3070,11 @@ file_timeout(ErlDrvData e) {
 #ifdef DEBUG
 	int r = 
 #endif
-	    flush_write(desc, NULL);
+	         flush_write(desc, NULL
+#ifdef USE_VM_PROBES
+			     , dt_priv, (desc->d == NULL) ? NULL : desc->d->sched_utag
+#endif
+			     );
 	/* Only possible reason for bad return value is ENOMEM, and 
 	 * there is nobody to tell...
 	 */
@@ -2737,6 +3095,14 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
     char command;
     int p, q;
     int err;
+    struct t_data *d = NULL;
+#ifdef USE_VM_PROBES
+    Sint64 dt_i1 = 0, dt_i2 = 0, dt_i3 = 0;
+    Sint64 dt_i4 = 0;
+    char *dt_utag = NULL;
+    char *dt_s1 = NULL;
+    dt_private *dt_priv = get_dt_private(dt_driver_io_worker_base);
+#endif
 
     TRACE_C('v');
 
@@ -2756,18 +3122,19 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
     switch (command) {
 
     case FILE_CLOSE: {
+#ifdef USE_VM_PROBES
+	dt_utag = EV_CHAR_P(ev, p, q);
+#endif
 	flush_read(desc);
-	if (flush_write_check_error(desc, &err) < 0) {
+	if (flush_write_check_error(desc, &err
+#ifdef USE_VM_PROBES
+				    , dt_priv, dt_utag
+#endif
+				    ) < 0) {
 	    reply_posix_error(desc, err);
 	    goto done;
 	}
-	if (ev->size != 1) {
-	    /* Wrong command length */
-	    reply_posix_error(desc, EINVAL);
-	    goto done;
-	}
 	if (desc->fd != FILE_FD_INVALID) {
-	    struct t_data *d;
 	    if (! (d = EF_ALLOC(sizeof(struct t_data)))) {
 		reply_posix_error(desc, ENOMEM);
 	    } else {
@@ -2775,6 +3142,10 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 		d->reply = !0;
 		d->fd = desc->fd;
 		d->flags = desc->flags;
+#ifdef USE_VM_PROBES
+		dt_i1 = d->fd;
+		dt_i2 = d->flags;
+#endif
 		d->invoke = invoke_close;
 		d->free = free_data;
 		d->level = 2;
@@ -2790,8 +3161,21 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
     case FILE_READ: {
 	Uint32 sizeH, sizeL;
 	size_t size, alloc_size;
-	struct t_data *d;
-	if (flush_write_check_error(desc, &err) < 0) {
+
+	if (!EV_GET_UINT32(ev, &sizeH, &p, &q)
+	    || !EV_GET_UINT32(ev, &sizeL, &p, &q)) {
+	    /* Wrong buffer length to contain the read count */
+	    reply_posix_error(desc, EINVAL);
+	    goto done;
+	}
+#ifdef USE_VM_PROBES
+	dt_utag = EV_CHAR_P(ev, p, q);
+#endif
+	if (flush_write_check_error(desc, &err
+#ifdef USE_VM_PROBES
+				    , dt_priv, dt_utag
+#endif
+				    ) < 0) {
 	    reply_posix_error(desc, err);
 	    goto done;
 	}
@@ -2799,19 +3183,16 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	if (desc->read_bufsize == 0 && desc->read_binp != NULL && desc->read_size > 0) {
 	    /* We have allocated a buffer for line mode but should not really have a 
 	       read-ahead buffer... */
-	    if (lseek_flush_read(desc, &err) < 0) {
+	    if (lseek_flush_read(desc, &err
+#ifdef USE_VM_PROBES
+				 , dt_priv, dt_utag
+#endif
+				 ) < 0) {
 		reply_posix_error(desc, err);
 		goto done;
 	    }
 	}
 #endif
-	if (ev->size != 1+8
-	    || !EV_GET_UINT32(ev, &sizeH, &p, &q)
-	    || !EV_GET_UINT32(ev, &sizeL, &p, &q)) {
-	    /* Wrong buffer length to contain the read count */
-	    reply_posix_error(desc, EINVAL);
-	    goto done;
-	}
 #if SIZEOF_SIZE_T == 4
 	if (sizeH != 0) {
 	    reply_posix_error(desc, EINVAL);
@@ -2887,6 +3268,11 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	d->c.read.bin_offset = desc->read_offset + desc->read_size;
 	d->c.read.bin_size = desc->read_binp->orig_size - d->c.read.bin_offset;
 	d->c.read.size = size;
+#ifdef USE_VM_PROBES
+	dt_i1 = d->fd;
+	dt_i2 = d->flags;
+	dt_i3 = d->c.read.size;
+#endif
 	driver_binary_inc_refc(d->c.read.binp);
 	d->invoke = invoke_read;
 	d->free = free_read;
@@ -2904,12 +3290,22 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	 *    allocated binary + dealing with offsets and lengts are done in file_async ready
 	 *    for this OP.
 	 */
-	struct t_data *d;
-	if (flush_write_check_error(desc, &err) < 0) {
+#ifdef USE_VM_PROBES
+	dt_utag = EV_CHAR_P(ev, p, q);
+#endif
+	if (flush_write_check_error(desc, &err
+#ifdef USE_VM_PROBES
+				    , dt_priv, dt_utag
+#endif
+				    ) < 0) {
 	    reply_posix_error(desc, err);
 	    goto done;
 	}
-	if (ev->size != 1) {
+	if (ev->size != 1
+#ifdef USE_VM_PROBES
+	    + FILENAME_BYTELEN(dt_utag) + FILENAME_CHARSIZE
+#endif
+	    ) {
 	    /* Wrong command length */
 	    reply_posix_error(desc, EINVAL);
 	    goto done;
@@ -2965,8 +3361,16 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	d->c.read_line.binp = desc->read_binp;
 	d->c.read_line.read_offset = desc->read_offset;
 	d->c.read_line.read_size = desc->read_size;
+#ifdef USE_VM_PROBES
+	dt_i1 = d->fd;
+	dt_i2 = d->flags;
+	dt_i3 = d->c.read_line.read_offset;
+#endif
 #if !ALWAYS_READ_LINE_AHEAD
 	d->c.read_line.read_ahead = (desc->read_bufsize > 0);
+#ifdef USE_VM_PROBES
+	dt_i4 = d->c.read_line.read_ahead;
+#endif
 #endif 
 	driver_binary_inc_refc(d->c.read.binp);
 	d->invoke = invoke_read_line;
@@ -2974,10 +3378,22 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	d->level = 1;
 	cq_enq(desc, d);
     } goto done;
-    case FILE_WRITE: {
+    case FILE_WRITE: { /* Dtrace: The dtrace user tag is not last in message, 
+			  but follows the message tag directly. 
+			  This is handled specially in prim_file.erl */
 	ErlDrvSizeT skip = 1;
 	ErlDrvSizeT size = ev->size - skip;
-	if (lseek_flush_read(desc, &err) < 0) {
+
+#ifdef USE_VM_PROBES
+	dt_utag = EV_CHAR_P(ev, p, q);
+	skip += FILENAME_BYTELEN(dt_utag) + FILENAME_CHARSIZE;
+	size = ev->size - skip;
+#endif
+	if (lseek_flush_read(desc, &err
+#ifdef USE_VM_PROBES
+			     , dt_priv, dt_utag
+#endif
+			     ) < 0) {
 	    reply_posix_error(desc, err);
 	    goto done;
 	}
@@ -3004,7 +3420,11 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 		driver_set_timer(desc->port, desc->write_delay);
 	    }
 	} else {
-	    if (async_write(desc, &err, !0, size) != 0) {
+	    if ((d = async_write(desc, &err, !0, size
+#ifdef USE_VM_PROBES
+				 , &dt_i1, &dt_i2, &dt_i3
+#endif
+				 )) == NULL) {
 		MUTEX_UNLOCK(desc->q_mtx);
 		reply_posix_error(desc, err);
 		goto done;
@@ -3014,24 +3434,49 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	}
     } goto done; /* case FILE_WRITE */
 
-    case FILE_PWRITEV: {
+    case FILE_PWRITEV: { /* Dtrace: The dtrace user tag is not last in message, 
+			   but follows the message tag directly. 
+			   This is handled specially in prim_file.erl */
 	Uint32 i, j, n; 
 	size_t total;
-	struct t_data *d;
-	if (lseek_flush_read(desc, &err) < 0) {
-	    reply_Uint_posix_error(desc, 0, err);
-	    goto done;
-	}
-	if (flush_write_check_error(desc, &err) < 0) {
-	    reply_Uint_posix_error(desc, 0, err);
-	    goto done;
+#ifdef USE_VM_PROBES
+	char dt_tmp;
+	int dt_utag_bytes = 1;
+
+	dt_utag = EV_CHAR_P(ev, p, q);
+	/* This will work for UTF-8, but not for UTF-16 - extra reminder here */
+#ifdef FILENAMES_16BIT 
+#error 16bit characters in filenames and dtrace in combination is not supported.
+#endif
+	while (EV_GET_CHAR(ev, &dt_tmp, &p, &q) && dt_tmp != '\0') {
+	    dt_utag_bytes++;
 	}
+#endif
 	if (ev->size < 1+4
+#ifdef USE_VM_PROBES
+	    + dt_utag_bytes
+#endif
 	    || !EV_GET_UINT32(ev, &n, &p, &q)) {
 	    /* Buffer too short to contain even the number of pos/size specs */
 	    reply_Uint_posix_error(desc, 0, EINVAL);
 	    goto done;
 	}
+	if (lseek_flush_read(desc, &err
+#ifdef USE_VM_PROBES
+			     , dt_priv, dt_utag
+#endif
+			     ) < 0) {
+	    reply_Uint_posix_error(desc, 0, err);
+	    goto done;
+	}
+	if (flush_write_check_error(desc, &err
+#ifdef USE_VM_PROBES
+				    , dt_priv, dt_utag
+#endif
+				    ) < 0) {
+	    reply_Uint_posix_error(desc, 0, err);
+	    goto done;
+	}
 	if (n == 0) {
 	    /* Trivial case - nothing to write */
 	    if (ev->size != 1+4) {
@@ -3041,7 +3486,11 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	    }
 	    goto done;
 	}
-	if (ev->size < 1+4+8*(2*n)) {
+	if (ev->size < 1+4+8*(2*n)
+#ifdef USE_VM_PROBES
+	    + dt_utag_bytes
+#endif
+	    ) {
 	    /* Buffer too short to contain even the pos/size specs */
 	    reply_Uint_posix_error(desc, 0, EINVAL);
 	    goto done;
@@ -3056,6 +3505,10 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	d->reply = !0;
 	d->fd = desc->fd;
 	d->flags = desc->flags;
+#ifdef USE_VM_PROBES
+	dt_i1 = d->fd;
+	dt_i2 = d->flags;
+#endif
 	d->c.pwritev.port = desc->port;
 	d->c.pwritev.q_mtx = desc->q_mtx;
 	d->c.pwritev.n = n;
@@ -3093,13 +3546,20 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	    }
 	}
 	d->c.pwritev.size = total;
+#ifdef USE_VM_PROBES
+	dt_i3 = d->c.pwritev.size;
+#endif
 	d->c.pwritev.free_size = 0;
 	if (j == 0) {
 	    /* Trivial case - nothing to write */
 	    EF_FREE(d);
 	    reply_Uint(desc, 0);
 	} else {
-	    ErlDrvSizeT skip = 1 + 4 + 8*(2*n);
+	    ErlDrvSizeT skip = 1 + 4 + 8 * (2*n) 
+#ifdef USE_VM_PROBES
+		+ dt_utag_bytes
+#endif
+		;
 	    if (skip + total != ev->size) {
 		/* Actual amount of data does not match 
 		 * total of all pos/size specs
@@ -3120,27 +3580,55 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	}
     } goto done; /* case FILE_PWRITEV: */
 
-    case FILE_PREADV: {
+    case FILE_PREADV: { /* Dtrace: The dtrace user tag is not last in message, 
+			   but follows the message tag directly. 
+			   This is handled specially in prim_file.erl */
 	register void * void_ptr;
 	Uint32 i, n;
-	struct t_data *d;
 	ErlIOVec *res_ev;
-	if (lseek_flush_read(desc, &err) < 0) {
+#ifdef USE_VM_PROBES
+	char dt_tmp;
+	int dt_utag_bytes = 1;
+	/* This will work for UTF-8, but not for UTF-16 - extra reminder here */
+#ifdef FILENAMES_16BIT 
+#error 16bit characters in filenames and dtrace in combination is not supported.
+#endif
+	dt_utag = EV_CHAR_P(ev, p, q);
+	while (EV_GET_CHAR(ev, &dt_tmp, &p, &q) && dt_tmp != '\0') {
+	    dt_utag_bytes++;
+	}
+#endif
+	if (lseek_flush_read(desc, &err
+#ifdef USE_VM_PROBES
+			     , dt_priv, dt_utag
+#endif
+			     ) < 0) {
 	    reply_posix_error(desc, err);
 	    goto done;
 	}
-	if (flush_write_check_error(desc, &err) < 0) {
+	if (flush_write_check_error(desc, &err
+#ifdef USE_VM_PROBES
+				    , dt_priv, dt_utag
+#endif
+				    ) < 0) {
 	    reply_posix_error(desc, err);
 	    goto done;
 	}
 	if (ev->size < 1+8
+#ifdef USE_VM_PROBES
+	    + dt_utag_bytes
+#endif
 	    || !EV_GET_UINT32(ev, &n, &p, &q)
 	    || !EV_GET_UINT32(ev, &n, &p, &q)) {
 	    /* Buffer too short to contain even the number of pos/size specs */
 	    reply_posix_error(desc, EINVAL);
 	    goto done;
 	}
-	if (ev->size != 1+8+8*(2*n)) {
+	if (ev->size < 1+8+8*(2*n)
+#ifdef USE_VM_PROBES
+	    + dt_utag_bytes
+#endif
+	    ) {
 	    /* Buffer wrong length to contain the pos/size specs */
 	    reply_posix_error(desc, EINVAL);
 	    goto done;
@@ -3160,6 +3648,10 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	d->reply = !0;
 	d->fd = desc->fd;
 	d->flags = desc->flags;
+#ifdef USE_VM_PROBES
+	dt_i1 = d->fd;
+	dt_i2 = d->flags;
+#endif
 	d->c.preadv.n = n;
 	d->c.preadv.cnt = 0;
 	d->c.preadv.size = 0;
@@ -3187,6 +3679,9 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 #else
 	    size = ((size_t)sizeH<<32) | sizeL;
 #endif
+#ifdef USE_VM_PROBES
+	    dt_i3 += size;
+#endif
 	    if (! (res_ev->binv[i] = driver_alloc_binary(size))) {
 		reply_posix_error(desc, ENOMEM);
 		break;
@@ -3233,42 +3728,68 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
     } goto done; /* case FILE_PREADV: */
 
     case FILE_LSEEK: {
-	Sint64 offset;          /* Offset for seek */
+	Sint64 offset;		/* Offset for seek */
 	Uint32 origin;		/* Origin of seek. */
-	if (lseek_flush_read(desc, &err) < 0) {
-	    reply_posix_error(desc, err);
+
+	if (ev->size < 1+8+4
+	    || !EV_GET_UINT64(ev, &offset, &p, &q)
+	    || !EV_GET_UINT32(ev, &origin, &p, &q)) {
+	    /* Wrong length of buffer to contain offset and origin */
+	    reply_posix_error(desc, EINVAL);
 	    goto done;
 	}
-	if (flush_write_check_error(desc, &err) < 0) {
+#ifdef USE_VM_PROBES
+	dt_utag = EV_CHAR_P(ev, p, q);
+#endif
+	if (lseek_flush_read(desc, &err
+#ifdef USE_VM_PROBES
+			     , dt_priv, dt_utag
+#endif
+			     ) < 0) {
 	    reply_posix_error(desc, err);
 	    goto done;
 	}
-	if (ev->size != 1+8+4
-	    || !EV_GET_UINT64(ev, &offset, &p, &q)
-	    || !EV_GET_UINT32(ev, &origin, &p, &q)) {
-	    /* Wrong length of buffer to contain offset and origin */
-	    reply_posix_error(desc, EINVAL);
+	if (flush_write_check_error(desc, &err
+#ifdef USE_VM_PROBES
+				    , dt_priv, dt_utag
+#endif
+				    ) < 0) {
+	    reply_posix_error(desc, err);
 	    goto done;
 	}
-	if (async_lseek(desc, &err, !0, offset, origin) < 0) {
+	if ((d = async_lseek(desc, &err, !0, offset, origin
+#ifdef USE_VM_PROBES
+			     , &dt_i1, &dt_i2, &dt_i3
+#endif
+			     )) == NULL) {
 	    reply_posix_error(desc, err);
 	    goto done;
 	}
     } goto done;
 
     case FILE_READ_FILE: {
-	struct t_data *d;
 	char *filename;
 	if (ev->size < 1+1) {
 	    /* Buffer contains empty name */
 	    reply_posix_error(desc, ENOENT);
 	    goto done;
 	}
+#ifndef USE_VM_PROBES
+	/* In the dtrace case, the iov has an extra element, the dtrace utag - we will need 
+	   another test to see that
+	   the filename is in a single buffer: */
 	if (ev->size-1 != ev->iov[q].iov_len-p) {
 	    /* Name not in one single buffer */
 	    reply_posix_error(desc, EINVAL);
 	    goto done;
 	}
+#else
+	if (((byte *)ev->iov[q].iov_base)[ev->iov[q].iov_len-1] != '\0') {
+	    /* Name not in one single buffer */
+	    reply_posix_error(desc, EINVAL);
+	    goto done;
+	}	
+#endif
 	filename = EV_CHAR_P(ev, p, q);
 	d = EF_ALLOC(sizeof(struct t_data) -1 + FILENAME_BYTELEN(filename) + FILENAME_CHARSIZE);
 	if (! d) {
@@ -3279,6 +3800,20 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	d->reply = !0;
 	/* Copy name */
 	FILENAME_COPY(d->b, filename);
+#ifdef USE_VM_PROBES
+	{
+	    char dt_tmp;
+
+	    /* This will work for UTF-8, but not for UTF-16 - extra reminder here */
+#ifdef FILENAMES_16BIT 
+#error 16bit characters in filenames and dtrace in combination is not supported.
+#endif
+	    while (EV_GET_CHAR(ev, &dt_tmp, &p, &q) && dt_tmp != '\0') 
+		;
+	    dt_s1 = d->b;
+	    dt_utag = EV_CHAR_P(ev, p, q);
+	}
+#endif
 	d->c.read_file.binp = NULL;
 	d->invoke = invoke_read_file;
 	d->free = free_read_file;
@@ -3298,7 +3833,6 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	char mode;
 	Sint64 hdr_offset;
 	Uint32 max_size;
-	struct t_data *d;
 	ErlIOVec *res_ev;
 	int vsize;
 	if (! EV_GET_CHAR(ev, &mode, &p, &q)) {
@@ -3310,14 +3844,6 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	    reply_posix_error(desc, EINVAL);
 	    goto done;
 	}
-	if (lseek_flush_read(desc, &err) < 0) {
-	    reply_posix_error(desc, err);
-	    goto done;
-	}
-	if (flush_write_check_error(desc, &err) < 0) {
-	    reply_posix_error(desc, err);
-	    goto done;
-	}
 	if (ev->size < 1+1+8+4
 	    || !EV_GET_UINT64(ev, &hdr_offset, &p, &q)
 	    || !EV_GET_UINT32(ev, &max_size, &p, &q)) {
@@ -3326,6 +3852,25 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	    reply_posix_error(desc, EINVAL);
 	    goto done;
 	}
+#ifdef USE_VM_PROBES
+	dt_utag = EV_CHAR_P(ev, p, q);
+#endif
+	if (lseek_flush_read(desc, &err
+#ifdef USE_VM_PROBES
+			     , dt_priv, dt_utag
+#endif
+			     ) < 0) {
+	    reply_posix_error(desc, err);
+	    goto done;
+	}
+	if (flush_write_check_error(desc, &err
+#ifdef USE_VM_PROBES
+				    , dt_priv, dt_utag
+#endif
+				    ) < 0) {
+	    reply_posix_error(desc, err);
+	    goto done;
+	}
 	/* Create the thread data structure with the contained ErlIOVec 
 	 * and corresponding binaries for the response 
 	 */
@@ -3342,6 +3887,12 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 	d->flags = desc->flags;
 	d->c.preadv.offsets[0] = hdr_offset;
 	d->c.preadv.size = max_size;
+#ifdef USE_VM_PROBES
+	dt_i1 = d->fd;
+	dt_i2 = d->flags;
+	dt_i3 = d->c.preadv.offsets[0];
+	dt_i4 = d->c.preadv.size;
+#endif
 	res_ev = &d->c.preadv.eiov;
 	/* XXX possible alignment problems here for weird machines */
 	res_ev->iov = void_ptr = d + 1;
@@ -3356,16 +3907,24 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 
     case FILE_SETOPT: {
 	char opt;
+
 	if (ev->size < 1+1
 	    || !EV_GET_CHAR(ev, &opt, &p, &q)) {
 	    /* Buffer too short to contain even the option type */
 	    reply_posix_error(desc, EINVAL);
 	    goto done;
 	}
+#ifdef USE_VM_PROBES
+	dt_i1 = opt;
+	dt_utag = EV_CHAR_P(ev, p, q);
+#endif
 	switch (opt) {
 	case FILE_OPT_DELAYED_WRITE: {
 	    Uint32 sizeH, sizeL, delayH, delayL;
 	    if (ev->size != 1+1+4*sizeof(Uint32)
+#ifdef USE_VM_PROBES
+		+ FILENAME_BYTELEN(dt_utag) + FILENAME_CHARSIZE
+#endif
 		|| !EV_GET_UINT32(ev, &sizeH, &p, &q)
 		|| !EV_GET_UINT32(ev, &sizeL, &p, &q)
 		|| !EV_GET_UINT32(ev, &delayH, &p, &q)
@@ -3392,12 +3951,18 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 #else
 	    desc->write_delay = ((unsigned long)delayH << 32) | delayL;
 #endif
+#ifdef USE_VM_PROBES
+	    dt_i2 = desc->write_delay;
+#endif
 	    TRACE_C('K');
 	    reply_ok(desc);
 	} goto done;
 	case FILE_OPT_READ_AHEAD: {
 	    Uint32 sizeH, sizeL;
 	    if (ev->size != 1+1+2*sizeof(Uint32)
+#ifdef USE_VM_PROBES
+		+ FILENAME_BYTELEN(dt_utag)+FILENAME_CHARSIZE
+#endif
 		|| !EV_GET_UINT32(ev, &sizeH, &p, &q)
 		|| !EV_GET_UINT32(ev, &sizeL, &p, &q)) {
 		/* Buffer has wrong length to contain the option values */
@@ -3413,6 +3978,9 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 #else
 	    desc->read_bufsize = ((size_t)sizeH << 32) | sizeL;
 #endif
+#ifdef USE_VM_PROBES
+	    dt_i2 = desc->read_bufsize;
+#endif
 	    TRACE_C('K');
 	    reply_ok(desc);
 	} goto done;
@@ -3499,11 +4067,19 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
 
     } /* switch(command) */
 
-    if (lseek_flush_read(desc, &err) < 0) {
+    if (lseek_flush_read(desc, &err
+#ifdef USE_VM_PROBES
+			 , dt_priv, dt_utag
+#endif
+			 ) < 0) {
 	reply_posix_error(desc, err);
 	goto done;
     }
-    if (flush_write_check_error(desc, &err) < 0) {
+    if (flush_write_check_error(desc, &err
+#ifdef USE_VM_PROBES
+				, dt_priv, dt_utag
+#endif
+				) < 0) {
 	reply_posix_error(desc, err);
 	goto done;
     } else {
@@ -3521,5 +4097,50 @@ file_outputv(ErlDrvData e, ErlIOVec *ev) {
     }
 
  done:
+    if (d != NULL) {
+#ifdef USE_VM_PROBES
+	/*
+	 * If d == NULL, then either:
+	 *    1). There was an error of some sort, or
+	 *    2). The command given to us is actually implemented
+	 *	  by file_output() instead.
+	 *
+	 * Case #1 is probably a TODO item, perhaps?
+	 * Case #2 we definitely don't want to activate a probe.
+	 */
+	d->sched_i1 = dt_priv->thread_num;
+	d->sched_i2 = dt_priv->tag;
+	d->sched_utag[0] = '\0';
+	if (dt_utag != NULL) {
+	    if (dt_utag[0] == '\0') {
+                dt_utag = NULL;
+            } else {
+		strncpy(d->sched_utag, dt_utag, sizeof(d->sched_utag) - 1);
+		d->sched_utag[sizeof(d->sched_utag) - 1] = '\0';
+	    }
+	}
+	DTRACE11(efile_drv_entry, dt_priv->thread_num, dt_priv->tag++,
+		 dt_utag, command, dt_s1, NULL, dt_i1, dt_i2, dt_i3, dt_i4,
+		 desc->port_str);
+#endif
+    }
     cq_execute(desc);
 }
+
+#ifdef  USE_VM_PROBES
+dt_private *
+get_dt_private(int base)
+{
+    dt_private *dt_priv = (dt_private *) pthread_getspecific(dt_driver_key);
+
+    if (dt_priv == NULL) {
+	dt_priv = EF_SAFE_ALLOC(sizeof(dt_private));
+	erts_mtx_lock(&dt_driver_mutex);
+	dt_priv->thread_num = (base + dt_driver_idnum++);
+	erts_mtx_unlock(&dt_driver_mutex);
+	dt_priv->tag = 0;
+	pthread_setspecific(dt_driver_key, dt_priv);
+    }
+    return dt_priv;
+}
+#endif  /* USE_VM_PROBES */
diff --git a/erts/emulator/sys/common/erl_check_io.c b/erts/emulator/sys/common/erl_check_io.c
index 23a4bf1b04..1d173a758a 100644
--- a/erts/emulator/sys/common/erl_check_io.c
+++ b/erts/emulator/sys/common/erl_check_io.c
@@ -36,6 +36,7 @@
 #include "global.h"
 #include "erl_check_io.h"
 #include "erl_thr_progress.h"
+#include "dtrace-wrapper.h"
 
 #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS
 #  define ERTS_DRV_EV_STATE_EXTRA_SIZE 128
@@ -314,6 +315,7 @@ forget_removed(struct pollset_info* psi)
 	erts_smp_mtx_unlock(mtx);
 	if (drv_ptr) {
 	    int was_unmasked = erts_block_fpe();
+	    DTRACE1(driver_stop_select, drv_ptr->name);
 	    (*drv_ptr->stop_select) ((ErlDrvEvent) fd, NULL);
 	    erts_unblock_fpe(was_unmasked);
 	    if (drv_ptr->handle) {
@@ -496,6 +498,9 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix,
     ErtsDrvEventState *state;
     int wake_poller;
     int ret;
+#ifdef USE_VM_PROBES
+    DTRACE_CHARBUF(name, 64);
+#endif
     
     ERTS_SMP_LC_ASSERT(erts_drvport2port(ix)
 		       && erts_lc_is_port_locked(erts_drvport2port(ix)));
@@ -525,6 +530,10 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix,
 	if (IS_FD_UNKNOWN(state)) {
 	    /* fast track to stop_select callback */
 	    stop_select_fn = erts_drvport2port(ix)->drv_ptr->stop_select;
+#ifdef USE_VM_PROBES
+	    strncpy(name, erts_drvport2port(ix)->drv_ptr->name, sizeof(name)-1);
+	    name[sizeof(name)-1] = '\0';
+#endif
 	    ret = 0;
 	    goto done_unknown;
 	}
@@ -661,6 +670,10 @@ ERTS_CIO_EXPORT(driver_select)(ErlDrvPort ix,
 		    /* Safe to close fd now as it is not in pollset
 		       or there was no need to eject fd (kernel poll) */
 		    stop_select_fn = drv_ptr->stop_select;
+#ifdef USE_VM_PROBES
+		    strncpy(name, erts_drvport2port(ix)->drv_ptr->name, sizeof(name)-1);
+		    name[sizeof(name)-1] = '\0';
+#endif
 		}
 		else {
 		    /* Not safe to close fd, postpone stop_select callback. */
@@ -686,6 +699,7 @@ done_unknown:
     erts_smp_mtx_unlock(fd_mtx(fd));
     if (stop_select_fn) {
 	int was_unmasked = erts_block_fpe();
+	DTRACE1(driver_stop_select, name);
 	(*stop_select_fn)(e, NULL);
 	erts_unblock_fpe(was_unmasked);
     }
diff --git a/erts/emulator/utils/beam_makeops b/erts/emulator/utils/beam_makeops
index 58c36c3bdc..ea57000c82 100755
--- a/erts/emulator/utils/beam_makeops
+++ b/erts/emulator/utils/beam_makeops
@@ -28,6 +28,7 @@ my $verbose = 0;
 my $hot = 1;
 my $num_file_opcodes = 0;
 my $wordsize = 32;
+my %defs;			# Defines (from command line).
 
 # This is shift counts and mask for the packer.
 my $WHOLE_WORD = '';
@@ -96,6 +97,12 @@ my %unnumbered;
 my %is_transformed;
 
 #
+# Pre-processor.
+#
+my @if_val;
+my @if_line;
+
+#
 # Code transformations.
 #
 my $te_max_vars = 0;		# Max number of variables ever needed.
@@ -223,6 +230,7 @@ while (@ARGV && $ARGV[0] =~ /^-(.*)/) {
     ($outdir = shift), next if /^outdir/;
     ($wordsize = shift), next if /^wordsize/;
     ($verbose = 1), next if /^v/;
+    ($defs{$1} = $2), next if /^D(\w+)=(\w+)/;
     die "$0: Bad option: -$_\n";
 }
 
@@ -239,7 +247,43 @@ while (<>) {
     }
     next if /^\s*$/;
     next if /^\#/;
-    
+
+    #
+    # Handle %if.
+    #
+    if (/^\%if (\w+)/) {
+	my $name = $1;
+	my $val = $defs{$name};
+	defined $val or error("'$name' is undefined");
+	push @if_val, $val;
+	push @if_line, $.;
+	next;
+    } elsif (/^\%unless (\w+)/) {
+	my $name = $1;
+	my $val = $defs{$name};
+	defined $val or error("'$name' is undefined");
+	push @if_val, !$val;
+	push @if_line, $.;
+	next;
+    } elsif (/^\%else$/) {
+	unless (@if_line) {
+	    error("%else without a preceding %if/%unless");
+	}
+	$if_line[$#if_line] = $.;
+	$if_val[$#if_val] = !$if_val[$#if_val];
+	next;
+    } elsif (/^\%endif$/) {
+	unless (@if_line) {
+	    error("%endif without a preceding %if/%unless/%else");
+	}
+	pop @if_val;
+	pop @if_line;
+	next;
+    }
+    if (@if_val and not $if_val[$#if_val]) {
+	next;
+    }
+
     #
     # Handle assignments.
     #
@@ -349,7 +393,13 @@ while (<>) {
 	$unnumbered{$name,$arity} = 1;
     }
 } continue {
-    close(ARGV) if eof(ARGV);
+    if (eof(ARGV)) {
+	close(ARGV);
+	if (@if_line) {
+	    error("Unterminated %if/%unless/%else at " .
+		  "line $if_line[$#if_line]\n");
+	}
+    }
 }
 
 $num_file_opcodes = @gen_opname;
diff --git a/erts/lib_src/common/erl_printf.c b/erts/lib_src/common/erl_printf.c
index 108a8bb531..399c83384e 100644
--- a/erts/lib_src/common/erl_printf.c
+++ b/erts/lib_src/common/erl_printf.c
@@ -173,6 +173,7 @@ typedef struct {
 static int
 write_sn(void *vwsnap, char* buf, size_t len)
 {
+    int rv = 0;
     write_sn_arg_t *wsnap = (write_sn_arg_t *) vwsnap;
     ASSERT(wsnap);
     ASSERT(len > 0);
@@ -180,12 +181,13 @@ write_sn(void *vwsnap, char* buf, size_t len)
 	size_t sz = len;
 	if (sz >= wsnap->len)
 	    sz = wsnap->len;
+	rv = (int)sz;
 	memcpy((void *) wsnap->buf, (void *) buf, sz);
 	wsnap->buf += sz;
 	wsnap->len -= sz;
 	return sz;
     }
-    return 0;
+    return rv;
 }
 
 static int
diff --git a/erts/preloaded/ebin/erlang.beam b/erts/preloaded/ebin/erlang.beam
index 9b65302b6e..c01c121779 100644
--- a/erts/preloaded/ebin/erlang.beam
+++ b/erts/preloaded/ebin/erlang.beam
diff --git a/erts/preloaded/ebin/prim_file.beam b/erts/preloaded/ebin/prim_file.beam
index 5cd0722a24..538aa264d3 100644
--- a/erts/preloaded/ebin/prim_file.beam
+++ b/erts/preloaded/ebin/prim_file.beam
diff --git a/erts/preloaded/src/erlang.erl b/erts/preloaded/src/erlang.erl
index 274764ac74..f0264ba658 100644
--- a/erts/preloaded/src/erlang.erl
+++ b/erts/preloaded/src/erlang.erl
@@ -128,6 +128,8 @@
          term_to_binary/1, term_to_binary/2, tl/1, trace_pattern/2,
          trace_pattern/3, tuple_to_list/1, system_info/1,
          universaltime_to_localtime/1]).
+-export([dt_get_tag/0, dt_get_tag_data/0, dt_prepend_vm_tag_data/1, dt_append_vm_tag_data/1,
+	 dt_put_tag/1, dt_restore_tag/1, dt_spread_tag/1]). 
 
 
 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
@@ -572,6 +574,48 @@ display_string(_P1) ->
 dist_exit(_P1, _P2, _P3) ->
     erlang:nif_error(undefined).
 
+%% dt_append_vm_tag_data/1
+-spec dt_append_vm_tag_data(IoData) -> IoDataRet when
+      IoData :: iodata(),
+      IoDataRet :: iodata().
+dt_append_vm_tag_data(_IoData) ->
+    erlang:nif_error(undefined).
+
+%% dt_get_tag/0
+-spec erlang:dt_get_tag() -> binary() | undefined.
+dt_get_tag() ->
+    erlang:nif_error(undefined).
+
+%%  dt_get_tag_data/0
+-spec erlang:dt_get_tag_data() -> binary() | undefined.
+dt_get_tag_data() ->
+    erlang:nif_error(undefined).
+
+%% dt_prepend_vm_tag_data/1
+-spec dt_prepend_vm_tag_data(IoData) -> IoDataRet when
+      IoData :: iodata(),
+      IoDataRet :: iodata().
+dt_prepend_vm_tag_data(_IoData) ->
+    erlang:nif_error(undefined).
+
+%% dt_put_tag/1
+-spec erlang:dt_put_tag(IoData) -> binary() | undefined when
+      IoData :: iodata().
+dt_put_tag(_IoData) ->
+    erlang:nif_error(undefined).
+
+%% dt_restore_tag/1
+-spec erlang:dt_restore_tag(TagData) -> true when
+      TagData :: term().
+dt_restore_tag(_TagData) ->
+    erlang:nif_error(undefined).
+    
+%% dt_spread_tag/1
+-spec erlang:dt_spread_tag(boolean()) -> TagData when
+      TagData :: term().
+dt_spread_tag(_Bool) ->   
+    erlang:nif_error(undefined).
+
 %% erase/0
 -spec erase() -> [{Key, Val}] when
       Key :: term(),
@@ -2071,6 +2115,8 @@ tuple_to_list(_Tuple) ->
          (dist_ctrl) -> {Node :: node(),
                          ControllingEntity :: port() | pid()};
          (driver_version) -> string();
+	 (dynamic_trace) -> none | dtrace | systemtap;
+         (dynamic_trace_probes) -> boolean();
          (elib_malloc) -> false;
          (dist_buf_busy_limit) -> non_neg_integer();
          (fullsweep_after) -> {fullsweep_after, non_neg_integer()};
diff --git a/erts/preloaded/src/prim_file.erl b/erts/preloaded/src/prim_file.erl
index fcf94c5fa2..d66f0e09cc 100644
--- a/erts/preloaded/src/prim_file.erl
+++ b/erts/preloaded/src/prim_file.erl
@@ -294,7 +294,7 @@ advise(#file_descriptor{module = ?MODULE, data = {Port, _}},
 
 %% Returns {error, Reason} | ok.
 write(#file_descriptor{module = ?MODULE, data = {Port, _}}, Bytes) ->
-    case drv_command(Port, [?FILE_WRITE,Bytes]) of
+    case drv_command_nt(Port, [?FILE_WRITE,erlang:dt_prepend_vm_tag_data(Bytes)],undefined) of
 	{ok, _Size} ->
 	    ok;
 	Error ->
@@ -309,8 +309,8 @@ pwrite(#file_descriptor{module = ?MODULE, data = {Port, _}}, L)
 pwrite_int(_, [], 0, [], []) ->
     ok;
 pwrite_int(Port, [], N, Spec, Data) ->
-    Header = list_to_binary([<<?FILE_PWRITEV, N:32>> | reverse(Spec)]),
-    case drv_command_raw(Port, [Header | reverse(Data)]) of
+    Header = list_to_binary([?FILE_PWRITEV, erlang:dt_prepend_vm_tag_data(<<N:32>>) | reverse(Spec)]),
+    case drv_command_nt(Port, [Header | reverse(Data)], undefined) of
 	{ok, _Size} ->
 	    ok;
 	Error ->
@@ -428,7 +428,7 @@ pread(#file_descriptor{module = ?MODULE, data = {Port, _}}, L)
 pread_int(_, [], 0, []) ->
     {ok, []};
 pread_int(Port, [], N, Spec) ->
-    drv_command(Port, [<<?FILE_PREADV, 0:32, N:32>> | reverse(Spec)]);
+    drv_command_nt(Port, [?FILE_PREADV, erlang:dt_prepend_vm_tag_data(<<0:32, N:32>>) | reverse(Spec)],undefined);
 pread_int(Port, [{Offs, Size} | T], N, Spec)
   when is_integer(Offs), is_integer(Size), 0 =< Size ->
     if
@@ -449,9 +449,9 @@ pread(#file_descriptor{module = ?MODULE, data = {Port, _}}, Offs, Size)
     if
 	-(?LARGEFILESIZE) =< Offs, Offs < ?LARGEFILESIZE,
 	Size < ?LARGEFILESIZE ->
-	    case drv_command(Port, 
-			     <<?FILE_PREADV, 0:32, 1:32,
-			      Offs:64/signed, Size:64>>) of
+	    case drv_command_nt(Port, 
+				[?FILE_PREADV, erlang:dt_prepend_vm_tag_data(<<0:32, 1:32,
+									     Offs:64/signed, Size:64>>)], undefined) of
 		{ok, [eof]} ->
 		    eof;
 		{ok, [Data]} ->
@@ -949,12 +949,17 @@ drv_open(Driver, Portopts) ->
 %% Closes a port in a safe way. Returns ok.
 
 drv_close(Port) ->
-    try erlang:port_close(Port) catch error:_ -> ok end,
-    receive %% Ugly workaround in case the caller==owner traps exits
-	{'EXIT', Port, _Reason} -> 
-	    ok
-    after 0 -> 
-	    ok
+    Save = erlang:dt_spread_tag(false),
+    try
+	try erlang:port_close(Port) catch error:_ -> ok end,
+	receive %% Ugly workaround in case the caller==owner traps exits
+	    {'EXIT', Port, _Reason} -> 
+		ok
+	after 0 -> 
+		ok
+	end
+    after
+	erlang:dt_restore_tag(Save)
     end.
 
 
@@ -964,9 +969,6 @@ drv_close(Port) ->
 %% then closed after the result has been received.
 %% Returns {ok, Result} or {error, Reason}.
 
-drv_command_raw(Port, Command) ->
-    drv_command(Port, Command, false, undefined).
-
 drv_command(Port, Command) ->
     drv_command(Port, Command, undefined).
 
@@ -982,7 +984,8 @@ drv_command(Port, Command, R) ->
     end.
 
 drv_command(Port, Command, Validated, R) when is_port(Port) ->
-    try erlang:port_command(Port, Command) of
+    Save = erlang:dt_spread_tag(false),
+    try erlang:port_command(Port, erlang:dt_append_vm_tag_data(Command)) of
 	true ->
 	    drv_get_response(Port, R)
     catch
@@ -1001,6 +1004,8 @@ drv_command(Port, Command, Validated, R) when is_port(Port) ->
 	    end;
 	error:Reason ->
 	    {error, Reason}
+    after
+	erlang:dt_restore_tag(Save)
     end;
 drv_command({Driver, Portopts}, Command, Validated, R) ->
     case drv_open(Driver, Portopts) of
@@ -1011,6 +1016,25 @@ drv_command({Driver, Portopts}, Command, Validated, R) ->
 	Error ->
 	    Error
     end.
+drv_command_nt(Port, Command, R) when is_port(Port) ->
+    Save = erlang:dt_spread_tag(false),
+    try erlang:port_command(Port, Command) of
+	true ->
+	    drv_get_response(Port, R)
+    catch
+	error:badarg ->
+	    try erlang:iolist_size(Command) of
+		_ -> % Valid
+		    {error, einval}
+	    catch
+		error:_ ->
+		    {error, badarg}
+	    end;
+	error:Reason ->
+	    {error, Reason}
+    after
+	erlang:dt_restore_tag(Save)
+    end.