Implement magic references

Magic references are *intentionally* indistinguishable from ordinary
references for the Erlang software. Magic references do not change
the language, and are intended as a pure runtime internal optimization.

An ordinary reference is typically used as a key in some table. A
magic reference has a direct pointer to a reference counted magic
binary. This makes it possible to implement various things without
having to do lookups in a table, but instead access the data directly.
Besides very fast lookups this can also improve scalability by
removing a potentially contended table. A couple of examples of
planned future usage of magic references are ETS table identifiers,
and BIF timer identifiers.

Besides future optimizations using magic references it should also
be possible to replace the exposed magic binary cludge with magic
references. That is, magic binaries that are exposed as empty
binaries to the Erlang software.

36 files changed, 1475 insertions, 445 deletions

diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names
index 1efe5fe300..df2866b40e 100644
--- a/erts/emulator/beam/atom.names
+++ b/erts/emulator/beam/atom.names
@@ -379,6 +379,7 @@ atom long_schedule
 atom low
 atom Lt='<'
 atom machine
+atom magic_ref
 atom major
 atom match
 atom match_limit
diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c
index a2d2433ed8..bda7b9224c 100644
--- a/erts/emulator/beam/bif.c
+++ b/erts/emulator/beam/bif.c
@@ -199,7 +199,7 @@ BIF_RETTYPE link_1(BIF_ALIST_1)
 		goto res_no_proc;
 	    case ERTS_PORT_OP_SCHEDULED:
 		if (refp) {
-		    ASSERT(is_internal_ref(ref));
+		    ASSERT(is_internal_ordinary_ref(ref));
 		    BIF_TRAP3(await_port_send_result_trap, BIF_P, ref, am_true, am_true);
 		}
 	    default:
@@ -782,7 +782,7 @@ local_name_monitor(Process *self, Eterm type, Eterm target_name)
         case ERTS_PORT_OP_DONE:
             return ret;
         case ERTS_PORT_OP_SCHEDULED: { /* Scheduled a signal */
-            ASSERT(is_internal_ref(ret));
+            ASSERT(is_internal_ordinary_ref(ret));
             BIF_TRAP3(await_port_send_result_trap, self,
                       ret, am_true, ret);
             /* bif_trap returns */
@@ -1180,7 +1180,7 @@ BIF_RETTYPE unlink_1(BIF_ALIST_1)
 		res = erts_port_unlink(BIF_P, prt, BIF_P->common.id, refp);
 
 		if (refp && res == ERTS_PORT_OP_SCHEDULED) {
-		    ASSERT(is_internal_ref(ref));
+		    ASSERT(is_internal_ordinary_ref(ref));
 		    BIF_TRAP3(await_port_send_result_trap, BIF_P, ref, am_true, am_true);
 		}
 	    }
@@ -1586,7 +1586,7 @@ BIF_RETTYPE exit_2(BIF_ALIST_2)
 	     ERTS_BIF_CHK_EXITED(BIF_P);
 
 	     if (refp && res == ERTS_PORT_OP_SCHEDULED) {
-		 ASSERT(is_internal_ref(ref));
+		 ASSERT(is_internal_ordinary_ref(ref));
 		 BIF_TRAP3(await_port_send_result_trap, BIF_P, ref, am_true, am_true);
 	     }
 
@@ -2221,7 +2221,7 @@ do_send(Process *p, Eterm to, Eterm msg, Eterm *refp, ErtsSendContext* ctx)
 		/* Fall through */
 	    case ERTS_PORT_OP_SCHEDULED:
 		if (is_not_nil(*refp)) {
-		    ASSERT(is_internal_ref(*refp));
+		    ASSERT(is_internal_ordinary_ref(*refp));
 		    ret_val = SEND_AWAIT_RESULT;
 		}
 		break;
@@ -2409,7 +2409,7 @@ BIF_RETTYPE send_3(BIF_ALIST_3)
 	ERTS_BIF_PREP_YIELD_RETURN(retval, p, am_ok);
         break;
     case SEND_AWAIT_RESULT:
-	ASSERT(is_internal_ref(ref));
+	ASSERT(is_internal_ordinary_ref(ref));
 	ERTS_BIF_PREP_TRAP3(retval, await_port_send_result_trap, p, ref, am_nosuspend, am_ok);
 	break;
     case SEND_BADARG:
@@ -2526,7 +2526,7 @@ Eterm erl_send(Process *p, Eterm to, Eterm msg)
 	ERTS_BIF_PREP_YIELD_RETURN(retval, p, msg);
         break;
     case SEND_AWAIT_RESULT:
-	ASSERT(is_internal_ref(ref));
+	ASSERT(is_internal_ordinary_ref(ref));
 	ERTS_BIF_PREP_TRAP3(retval,
 			    await_port_send_result_trap, p, ref, msg, msg);
 	break;
@@ -4109,6 +4109,7 @@ BIF_RETTYPE ref_to_list_1(BIF_ALIST_1)
 {
     if (is_not_ref(BIF_ARG_1))
 	BIF_ERROR(BIF_P, BADARG);
+    erts_magic_ref_save_bin(BIF_ARG_1);
     BIF_RET(term2list_dsprintf(BIF_P, BIF_ARG_1));
 }
 
diff --git a/erts/emulator/beam/binary.c b/erts/emulator/beam/binary.c
index 071a356260..a09950ec40 100644
--- a/erts/emulator/beam/binary.c
+++ b/erts/emulator/beam/binary.c
@@ -47,13 +47,8 @@ void
 erts_init_binary(void)
 {
     /* Verify Binary alignment... */
-    if ((((UWord) &((Binary *) 0)->orig_bytes[0]) % ((UWord) 8)) != 0) {
-	/* I assume that any compiler should be able to optimize this
-	   away. If not, this test is not very expensive... */
-	erts_exit(ERTS_ABORT_EXIT,
-		 "Internal error: Address of orig_bytes[0] of a Binary"
-		 " is *not* 8-byte aligned\n");
-    }
+    ERTS_CT_ASSERT((offsetof(Binary,orig_bytes) % 8) == 0);
+    ERTS_CT_ASSERT((offsetof(ErtsMagicBinary,u.aligned.data) % 8) == 0);
 
     erts_init_trap_export(&binary_to_list_continue_export,
 			  am_erts_internal, am_binary_to_list_continue, 1,
diff --git a/erts/emulator/beam/copy.c b/erts/emulator/beam/copy.c
index d20102cb2c..85db23393c 100644
--- a/erts/emulator/beam/copy.c
+++ b/erts/emulator/beam/copy.c
@@ -858,20 +858,24 @@ Eterm copy_struct_x(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap, Uint
 	    case EXTERNAL_PORT_SUBTAG:
 	    case EXTERNAL_REF_SUBTAG:
 		{
-		  ExternalThing *etp = (ExternalThing *) htop;
-
+		  ExternalThing *etp = (ExternalThing *) objp;
+		  erts_smp_refc_inc(&etp->node->refc, 2);
+		}
+	    L_off_heap_node_container_common:
+		{
+		  struct erl_off_heap_header *ohhp;
+		  ohhp = (struct erl_off_heap_header *) htop;
 		  i  = thing_arityval(hdr) + 1;
+		  *argp = make_boxed(htop);
 		  tp = htop;
 
 		  while (i--)  {
 		    *htop++ = *objp++;
 		  }
 
-		  etp->next = off_heap->first;
-		  off_heap->first = (struct erl_off_heap_header*)etp;
-		  erts_smp_refc_inc(&etp->node->refc, 2);
+		  ohhp->next = off_heap->first;
+		  off_heap->first = ohhp;
 
-		  *argp = make_external(tp);
 		}
 		break;
 	    case MAP_SUBTAG:
@@ -899,6 +903,13 @@ Eterm copy_struct_x(Eterm obj, Uint sz, Eterm** hpp, ErlOffHeap* off_heap, Uint
 	    case BIN_MATCHSTATE_SUBTAG:
 		erts_exit(ERTS_ABORT_EXIT,
 			 "copy_struct: matchstate term not allowed");
+	    case REF_SUBTAG:
+		if (is_magic_ref_thing(objp)) {
+		    ErtsMRefThing *mreft = (ErtsMRefThing *) objp;
+		    erts_refc_inc(&mreft->mb->refc, 2);
+		    goto L_off_heap_node_container_common;
+		}
+		/* Fall through... */
 	    default:
 		i = thing_arityval(hdr)+1;
 		hbot -= i;
@@ -1649,20 +1660,33 @@ Uint copy_shared_perform(Eterm obj, Uint size, erts_shcopy_t *info,
 	    }
 	    case EXTERNAL_PID_SUBTAG:
 	    case EXTERNAL_PORT_SUBTAG:
-	    case EXTERNAL_REF_SUBTAG: {
-		ExternalThing *etp = (ExternalThing *) hp;
+	    case EXTERNAL_REF_SUBTAG:
+	    {
+		ExternalThing *etp = (ExternalThing *) ptr;
+		erts_smp_refc_inc(&etp->node->refc, 2);
+	    }
+	  off_heap_node_container_common:
+	    {
+		struct erl_off_heap_header *ohhp;
+		ohhp = (struct erl_off_heap_header *) hp;
 		sz = thing_arityval(hdr);
-		*resp = make_external(hp);
+		*resp = make_boxed(hp);
 		*hp++ = hdr;
 		ptr++;
 		while (sz-- > 0) {
 		    *hp++ = *ptr++;
 		}
-		etp->next = off_heap->first;
-		off_heap->first = (struct erl_off_heap_header*) etp;
-		erts_smp_refc_inc(&etp->node->refc, 2);
+		ohhp->next = off_heap->first;
+		off_heap->first = ohhp;
 		goto cleanup_next;
 	    }
+	    case REF_SUBTAG:
+		if (is_magic_ref_thing(ptr)) {
+		    ErtsMRefThing *mreft = (ErtsMRefThing *) ptr;
+		    erts_refc_inc(&mreft->mb->refc, 2);
+		    goto off_heap_node_container_common;
+		}
+		/* Fall through... */
 	    default:
 		sz = thing_arityval(hdr);
 		*resp = make_boxed(hp);
@@ -1859,6 +1883,15 @@ Eterm copy_shallow(Eterm* ptr, Uint sz, Eterm** hpp, ErlOffHeap* off_heap)
 		    off_heap->first = ohh;
 		}
 		break;
+	    case REF_SUBTAG: {
+		ErtsRefThing *rtp = (ErtsRefThing *) (tp - 1);
+		if (is_magic_ref_thing(rtp)) {
+		    ErtsMRefThing *mreft = (ErtsMRefThing *) rtp;
+		    erts_refc_inc(&mreft->mb->refc, 2);
+		    goto off_heap_common;
+		}
+		/* Fall through... */
+	    }
 	    default:
 		{
 		    int tari = header_arity(val);
@@ -1959,6 +1992,9 @@ move_one_frag(Eterm** hpp, ErlHeapFragment* frag, ErlOffHeap* off_heap, int lite
 	    ASSERT(ptr + header_arity(val) < end);
 	    MOVE_BOXED(ptr, val, hp, &dummy_ref);
 	    switch (val & _HEADER_SUBTAG_MASK) {
+	    case REF_SUBTAG:
+		if (is_ordinary_ref_thing(hdr))
+		    break;
 	    case REFC_BINARY_SUBTAG:
 	    case FUN_SUBTAG:
 	    case EXTERNAL_PID_SUBTAG:
diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c
index 4d990a9c56..e433de0f35 100644
--- a/erts/emulator/beam/erl_alloc.c
+++ b/erts/emulator/beam/erl_alloc.c
@@ -48,6 +48,7 @@
 #if defined(ERTS_ALC_T_DRV_SEL_D_STATE) || defined(ERTS_ALC_T_DRV_EV_D_STATE)
 #include "erl_check_io.h"
 #endif
+#include "erl_bif_unique.h"
 
 #define GET_ERL_GF_ALLOC_IMPL
 #include "erl_goodfit_alloc.h"
@@ -152,8 +153,7 @@ typedef struct {
     int internal;
     Uint req_sched;
     Process *proc;
-    Eterm ref;
-    Eterm ref_heap[REF_THING_SIZE];
+    ErtsIRefStorage iref;
     int allocs[ERTS_ALC_INFO_A_END - ERTS_ALC_A_MIN + 1];
 } ErtsAllocInfoReq;
 
@@ -682,6 +682,8 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
 #endif
     fix_type_sizes[ERTS_ALC_FIX_TYPE_IX(ERTS_ALC_T_NIF_EXP_TRACE)]
 	= sizeof(NifExportTrace);
+    fix_type_sizes[ERTS_ALC_FIX_TYPE_IX(ERTS_ALC_T_MREF_NSCHED_ENT)]
+	= sizeof(ErtsNSchedMagicRefTableEntry);
 
 #ifdef HARD_DEBUG
     hdbg_init();
@@ -3139,9 +3141,10 @@ reply_alloc_info(void *vair)
     while (1) {
 
 	if (hpp)
-	    ref_copy = STORE_NC(hpp, ohp, air->ref);
+	    ref_copy = erts_iref_storage_make_ref(&air->iref,
+                                                  hpp, ohp, 0);
 	else
-	    *szp += REF_THING_SIZE;
+	    *szp += erts_iref_storage_heap_size(&air->iref);
 
 	ai_list = NIL;
 	for (i = 0; air->allocs[i] != ERTS_ALC_A_INVALID; i++);
@@ -3370,8 +3373,10 @@ reply_alloc_info(void *vair)
     erts_smp_proc_unlock(rp, rp_locks);
     erts_proc_dec_refc(rp);
 
-    if (erts_smp_atomic32_dec_read_nob(&air->refc) == 0)
+    if (erts_smp_atomic32_dec_read_nob(&air->refc) == 0) {
+        erts_iref_storage_clean(&air->iref);
 	aireq_free(air);
+    }
 }
 
 int
@@ -3384,7 +3389,6 @@ erts_request_alloc_info(struct process *c_p,
     ErtsAllocInfoReq *air = aireq_alloc();
     Eterm req_ai[ERTS_ALC_INFO_A_END] = {0};
     Eterm alist;
-    Eterm *hp;
     int airix = 0, ai;
 
     air->req_sched = erts_get_scheduler_id();
@@ -3398,8 +3402,7 @@ erts_request_alloc_info(struct process *c_p,
     if (is_not_internal_ref(ref))
 	return 0;
 
-    hp = &air->ref_heap[0];
-    air->ref = STORE_NC(&hp, NULL, ref);
+    erts_iref_storage_save(&air->iref, ref);
 
     if (is_not_list(allocs))
 	return 0;
diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types
index 70eca5b49c..c253ea82c8 100644
--- a/erts/emulator/beam/erl_alloc.types
+++ b/erts/emulator/beam/erl_alloc.types
@@ -280,6 +280,10 @@ type    TRACE_MSG_QUEUE SHORT_LIVED     SYSTEM          trace_message_queue
 type    SCHED_ASYNC_JOB SHORT_LIVED     SYSTEM          async_calls
 type    DIRTY_START     STANDARD	PROCESSES       dirty_start
 type    DIRTY_SL	SHORT_LIVED	SYSTEM		dirty_short_lived
+type    MREF_NSCHED_ENT	FIXED_SIZE	SYSTEM		nsched_magic_ref_entry
+type    MREF_ENT	STANDARD	SYSTEM		magic_ref_entry
+type	MREF_TAB_BKTS	STANDARD	SYSTEM		magic_ref_table_buckets
+type	MREF_TAB	LONG_LIVED	SYSTEM		magic_ref_table
 
 +if threads_no_smp
 # Need thread safe allocs, but std_alloc and fix_alloc are not;
diff --git a/erts/emulator/beam/erl_bif_ddll.c b/erts/emulator/beam/erl_bif_ddll.c
index af90cfe40e..2ce872d2bc 100644
--- a/erts/emulator/beam/erl_bif_ddll.c
+++ b/erts/emulator/beam/erl_bif_ddll.c
@@ -1476,8 +1476,10 @@ static void add_proc_loaded_deref(DE_Handle *dh, Process *proc)
 
 static Eterm copy_ref(Eterm ref, Eterm *hp)
 {
-    RefThing *ptr = ref_thing_ptr(ref);
-    memcpy(hp, ptr, sizeof(RefThing));
+    ErtsORefThing *ptr;
+    ASSERT(is_internal_ordinary_ref(ref));
+    ptr = ordinary_ref_thing_ptr(ref);
+    memcpy(hp, ptr, sizeof(ErtsORefThing));
     return (make_internal_ref(hp));
 }
 
@@ -1720,10 +1722,10 @@ static void notify_proc(Process *proc, Eterm ref, Eterm driver_name, Eterm type,
 	Eterm e;
 	mp = erts_alloc_message_heap(proc, &rp_locks,
 				     (6 /* tuple */ + 3 /* Error tuple */ + 
-				      REF_THING_SIZE + need),
+				      ERTS_REF_THING_SIZE + need),
 				     &hp, &ohp);
 	r = copy_ref(ref,hp);
-	hp += REF_THING_SIZE;
+	hp += ERTS_REF_THING_SIZE;
 	e = build_load_error_hp(hp, errcode);
 	hp += need;
 	mess = TUPLE2(hp,tag,e);
@@ -1731,10 +1733,10 @@ static void notify_proc(Process *proc, Eterm ref, Eterm driver_name, Eterm type,
 	mess = TUPLE5(hp,type,r,am_driver,driver_name,mess);
     } else {	
 	mp = erts_alloc_message_heap(proc, &rp_locks,
-				     6 /* tuple */ + REF_THING_SIZE,
+				     6 /* tuple */ + ERTS_REF_THING_SIZE,
 				     &hp, &ohp);
 	r = copy_ref(ref,hp);
-	hp += REF_THING_SIZE;
+	hp += ERTS_REF_THING_SIZE;
 	mess = TUPLE5(hp,type,r,am_driver,driver_name,tag);
     }
     erts_queue_message(proc, rp_locks, mp, mess, am_system);
diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c
index 37dcb5e7a6..be113b7c88 100644
--- a/erts/emulator/beam/erl_bif_info.c
+++ b/erts/emulator/beam/erl_bif_info.c
@@ -187,6 +187,32 @@ bld_bin_list(Uint **hpp, Uint *szp, ErlOffHeap* oh)
     return res;
 }
 
+static Eterm
+bld_magic_ref_bin_list(Uint **hpp, Uint *szp, ErlOffHeap* oh)
+{
+    struct erl_off_heap_header* ohh;
+    Eterm res = NIL;
+    Eterm tuple;
+
+    for (ohh = oh->first; ohh; ohh = ohh->next) {
+	if (is_ref_thing_header((*((Eterm *) ohh)))) {
+            ErtsMRefThing *mrtp = (ErtsMRefThing *) ohh;
+	    Eterm val = erts_bld_uword(hpp, szp, (UWord) mrtp->mb);
+	    Eterm orig_size = erts_bld_uint(hpp, szp, mrtp->mb->orig_size);
+    
+	    if (szp)
+		*szp += 4+2;
+	    if (hpp) {
+		Uint refc = (Uint) erts_refc_read(&mrtp->mb->refc, 1);
+		tuple = TUPLE3(*hpp, val, orig_size, make_small(refc));
+		res = CONS(*hpp + 4, tuple, res);
+		*hpp += 4+2;
+	    }
+	}
+    }
+    return res;
+}
+
 
 /*
   make_monitor_list:
@@ -613,7 +639,8 @@ static Eterm pi_args[] = {
     am_current_location,
     am_current_stacktrace,
     am_message_queue_data,
-    am_garbage_collection_info
+    am_garbage_collection_info,
+    am_magic_ref
 };
 
 #define ERTS_PI_ARGS ((int) (sizeof(pi_args)/sizeof(Eterm)))
@@ -664,6 +691,7 @@ pi_arg2ix(Eterm arg)
     case am_current_stacktrace:			return 31;
     case am_message_queue_data:			return 32;
     case am_garbage_collection_info:		return 33;
+    case am_magic_ref:                          return 34;
     default:					return -1;
     }
 }
@@ -1599,6 +1627,14 @@ process_info_aux(Process *BIF_P,
 	hp = HAlloc(BIF_P, 3);
 	break;
 
+    case am_magic_ref: {
+	Uint sz = 3;
+	(void) bld_magic_ref_bin_list(NULL, &sz, &MSO(rp));
+	hp = HAlloc(BIF_P, sz);
+	res = bld_magic_ref_bin_list(&hp, NULL, &MSO(rp));
+	break;
+    }
+
     default:
 	return THE_NON_VALUE; /* will produce badarg */
 
@@ -3531,6 +3567,11 @@ BIF_RETTYPE error_logger_warning_map_0(BIF_ALIST_0)
 
 static erts_smp_atomic_t available_internal_state;
 
+static void empty_magic_ref_destructor(Binary *bin)
+{
+
+}
+
 BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1)
 {
     /*
@@ -3896,6 +3937,34 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1)
 		}
 		return make_atom(ix);
 	    }
+	    else if (ERTS_IS_ATOM_STR("magic_ref", tp[1])) {
+                Binary *bin;
+                UWord bin_addr, refc;
+                Eterm bin_addr_term, refc_term, test_type;
+                Uint sz;
+                Eterm *hp;
+                if (!is_internal_magic_ref(tp[2])) {
+                    if (is_internal_ordinary_ref(tp[2])) {
+                        ErtsORefThing *rtp;
+                        rtp = (ErtsORefThing *) internal_ref_val(tp[2]);
+                        if (erts_is_ref_numbers_magic(rtp->num))
+                            BIF_RET(am_true);
+                    }
+                    BIF_RET(am_false);
+                }
+                bin = erts_magic_ref2bin(tp[2]);
+                refc = erts_refc_read(&bin->refc, 1);
+                bin_addr = (UWord) bin;
+                sz = 4;
+                erts_bld_uword(NULL, &sz, bin_addr);
+                erts_bld_uword(NULL, &sz, refc);
+                hp = HAlloc(BIF_P, sz);
+                bin_addr_term = erts_bld_uword(&hp, NULL, bin_addr);
+                refc_term = erts_bld_uword(&hp, NULL, refc);
+                test_type = (ERTS_MAGIC_BIN_DESTRUCTOR(bin) == empty_magic_ref_destructor
+                             ? am_true : am_false);
+                BIF_RET(TUPLE3(hp, bin_addr_term, refc_term, test_type));
+	    }
 
 	    break;
 	}
@@ -3984,7 +4053,6 @@ static void broken_halt_test(Eterm bif_arg_2)
     erts_exit(ERTS_DUMP_EXIT, "%T", bif_arg_2);
 }
 
-
 BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2)
 {
     /*
@@ -4314,6 +4382,21 @@ BIF_RETTYPE erts_debug_set_internal_state_2(BIF_ALIST_2)
             /* Used by ets_SUITE (stdlib) */
             BIF_RET(erts_ets_restore_meta_state(BIF_P, BIF_ARG_2));
         }
+        else if (ERTS_IS_ATOM_STR("make", BIF_ARG_1)) {
+            if (ERTS_IS_ATOM_STR("magic_ref", BIF_ARG_2)) {
+                Binary *bin = erts_create_magic_binary(0, empty_magic_ref_destructor);
+                UWord bin_addr = (UWord) bin;
+                Eterm bin_addr_term, magic_ref, res;
+                Eterm *hp;
+                Uint sz = ERTS_MAGIC_REF_THING_SIZE + 3;
+                erts_bld_uword(NULL, &sz, bin_addr);
+                hp = HAlloc(BIF_P, sz);
+                bin_addr_term = erts_bld_uword(&hp, NULL, bin_addr);
+                magic_ref = erts_mk_magic_ref(&hp, &BIF_P->off_heap, bin);
+                res = TUPLE2(hp, magic_ref, bin_addr_term);
+                BIF_RET(res);
+            }
+        }
     }
 
     BIF_ERROR(BIF_P, BADARG);
diff --git a/erts/emulator/beam/erl_bif_port.c b/erts/emulator/beam/erl_bif_port.c
index 90e78a9b0b..09d2c5376b 100644
--- a/erts/emulator/beam/erl_bif_port.c
+++ b/erts/emulator/beam/erl_bif_port.c
@@ -214,7 +214,7 @@ BIF_RETTYPE erts_internal_port_command_3(BIF_ALIST_3)
 	ASSERT(!(flags & ERTS_PORT_SIG_FLG_FORCE));
 	/* Fall through... */
     case ERTS_PORT_OP_SCHEDULED:
-	ASSERT(is_internal_ref(ref));
+	ASSERT(is_internal_ordinary_ref(ref));
 	ERTS_BIF_PREP_RET(res, ref);
 	break;
     case ERTS_PORT_OP_DONE:
@@ -260,7 +260,7 @@ BIF_RETTYPE erts_internal_port_call_3(BIF_ALIST_3)
 	retval = am_badarg;
 	break;
     case ERTS_PORT_OP_SCHEDULED:
-	ASSERT(is_internal_ref(retval));
+	ASSERT(is_internal_ordinary_ref(retval));
 	break;
     case ERTS_PORT_OP_DONE:
 	ASSERT(is_not_internal_ref(retval));
@@ -310,7 +310,7 @@ BIF_RETTYPE erts_internal_port_control_3(BIF_ALIST_3)
 	retval = am_badarg;
 	break;
     case ERTS_PORT_OP_SCHEDULED:
-	ASSERT(is_internal_ref(retval));
+	ASSERT(is_internal_ordinary_ref(retval));
 	break;
     case ERTS_PORT_OP_DONE:
 	ASSERT(is_not_internal_ref(retval));
@@ -356,7 +356,7 @@ BIF_RETTYPE erts_internal_port_close_1(BIF_ALIST_1)
     case ERTS_PORT_OP_DROPPED:
 	BIF_RET(am_badarg);
     case ERTS_PORT_OP_SCHEDULED:
-	ASSERT(is_internal_ref(ref));
+	ASSERT(is_internal_ordinary_ref(ref));
 	BIF_RET(ref);
     case ERTS_PORT_OP_DONE:
 	BIF_RET(am_true);
@@ -389,7 +389,7 @@ BIF_RETTYPE erts_internal_port_connect_2(BIF_ALIST_2)
     case ERTS_PORT_OP_DROPPED:
 	BIF_RET(am_badarg);
     case ERTS_PORT_OP_SCHEDULED:
-	ASSERT(is_internal_ref(ref));
+	ASSERT(is_internal_ordinary_ref(ref));
 	BIF_RET(ref);
 	break;
     case ERTS_PORT_OP_DONE:
@@ -428,7 +428,7 @@ BIF_RETTYPE erts_internal_port_info_1(BIF_ALIST_1)
     case ERTS_PORT_OP_DROPPED:
 	BIF_RET(am_undefined);
     case ERTS_PORT_OP_SCHEDULED:
-	ASSERT(is_internal_ref(retval));
+	ASSERT(is_internal_ordinary_ref(retval));
 	BIF_RET(retval);
     case ERTS_PORT_OP_DONE:
 	ASSERT(is_not_internal_ref(retval));
@@ -467,7 +467,7 @@ BIF_RETTYPE erts_internal_port_info_2(BIF_ALIST_2)
     case ERTS_PORT_OP_DROPPED:
 	BIF_RET(am_undefined);
     case ERTS_PORT_OP_SCHEDULED:
-	ASSERT(is_internal_ref(retval));
+	ASSERT(is_internal_ordinary_ref(retval));
 	BIF_RET(retval);
     case ERTS_PORT_OP_DONE:
 	ASSERT(is_not_internal_ref(retval));
diff --git a/erts/emulator/beam/erl_bif_trace.c b/erts/emulator/beam/erl_bif_trace.c
index a480754bdb..f471390501 100644
--- a/erts/emulator/beam/erl_bif_trace.c
+++ b/erts/emulator/beam/erl_bif_trace.c
@@ -2363,7 +2363,7 @@ BIF_RETTYPE system_profile_2(BIF_ALIST_2)
 typedef struct {
     Process *proc;
     Eterm ref;
-    Eterm ref_heap[REF_THING_SIZE];
+    Eterm ref_heap[ERTS_REF_THING_SIZE];
     Eterm target;
     erts_smp_atomic32_t refc;
 } ErtsTraceDeliveredAll;
diff --git a/erts/emulator/beam/erl_bif_unique.c b/erts/emulator/beam/erl_bif_unique.c
index 7c70217d8d..3fd4c87094 100644
--- a/erts/emulator/beam/erl_bif_unique.c
+++ b/erts/emulator/beam/erl_bif_unique.c
@@ -22,12 +22,15 @@
 #  include "config.h"
 #endif
 
+#define ERL_BIF_UNIQUE_C__
 #include "sys.h"
 #include "erl_vm.h"
 #include "erl_alloc.h"
 #include "export.h"
 #include "bif.h"
 #include "erl_bif_unique.h"
+#include "hash.h"
+#include "erl_binary.h"
 
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\
  * Reference                                                         *
@@ -58,9 +61,20 @@ static union {
 static Uint32 max_thr_id;
 #endif
 
+static void init_magic_ref_tables(void);
+
+static Uint64 ref_init_value;
+
 static void
 init_reference(void)
 {
+    SysTimeval tv;
+    sys_gettimeofday(&tv);
+    ref_init_value = 0;
+    ref_init_value |= (Uint64) tv.tv_sec;
+    ref_init_value |= ((Uint64) tv.tv_usec) << 32;
+    ref_init_value *= (Uint64) 268438039;
+    ref_init_value += (Uint64) tv.tv_usec;
 #ifdef DEBUG
     max_thr_id = (Uint32) erts_no_schedulers;
 #ifdef ERTS_DIRTY_SCHEDULERS
@@ -68,11 +82,13 @@ init_reference(void)
     max_thr_id += (Uint32) erts_no_dirty_io_schedulers;
 #endif
 #endif
-    erts_atomic64_init_nob(&global_reference.count, 0);
+    erts_atomic64_init_nob(&global_reference.count,
+			   (erts_aint64_t) ref_init_value);
+    init_magic_ref_tables();
 }
 
 static ERTS_INLINE void
-global_make_ref_in_array(Uint32 thr_id, Uint32 ref[ERTS_MAX_REF_NUMBERS])
+global_make_ref_in_array(Uint32 thr_id, Uint32 ref[ERTS_REF_NUMBERS])
 {
     Uint64 value;
 
@@ -82,7 +98,7 @@ global_make_ref_in_array(Uint32 thr_id, Uint32 ref[ERTS_MAX_REF_NUMBERS])
 }
 
 static ERTS_INLINE void
-make_ref_in_array(Uint32 ref[ERTS_MAX_REF_NUMBERS])
+make_ref_in_array(Uint32 ref[ERTS_REF_NUMBERS])
 {
     ErtsSchedulerData *esdp = erts_get_scheduler_data();
     if (esdp)
@@ -92,15 +108,23 @@ make_ref_in_array(Uint32 ref[ERTS_MAX_REF_NUMBERS])
 }
 
 void
-erts_make_ref_in_array(Uint32 ref[ERTS_MAX_REF_NUMBERS])
+erts_make_ref_in_array(Uint32 ref[ERTS_REF_NUMBERS])
+{
+    make_ref_in_array(ref);
+}
+
+void
+erts_make_magic_ref_in_array(Uint32 ref[ERTS_REF_NUMBERS])
 {
     make_ref_in_array(ref);
+    ASSERT(!(ref[1] & ERTS_REF1_MAGIC_MARKER_BIT__));
+    ref[1] |= ERTS_REF1_MAGIC_MARKER_BIT__;
 }
 
-Eterm erts_make_ref_in_buffer(Eterm buffer[REF_THING_SIZE])
+Eterm erts_make_ref_in_buffer(Eterm buffer[ERTS_REF_THING_SIZE])
 {
     Eterm* hp = buffer;
-    Uint32 ref[ERTS_MAX_REF_NUMBERS];
+    Uint32 ref[ERTS_REF_NUMBERS];
 
     make_ref_in_array(ref);
     write_ref_thing(hp, ref[0], ref[1], ref[2]);
@@ -110,11 +134,11 @@ Eterm erts_make_ref_in_buffer(Eterm buffer[REF_THING_SIZE])
 Eterm erts_make_ref(Process *c_p)
 {
     Eterm* hp;
-    Uint32 ref[ERTS_MAX_REF_NUMBERS];
+    Uint32 ref[ERTS_REF_NUMBERS];
 
     ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_MAIN & erts_proc_lc_my_proc_locks(c_p));
 
-    hp = HAlloc(c_p, REF_THING_SIZE);
+    hp = HAlloc(c_p, ERTS_REF_THING_SIZE);
 
     make_ref_in_array(ref);
     write_ref_thing(hp, ref[0], ref[1], ref[2]);
@@ -122,6 +146,272 @@ Eterm erts_make_ref(Process *c_p)
     return make_internal_ref(hp);
 }
 
+/*
+ * Magic reference tables
+ */
+
+typedef struct {
+    HashBucket hash;
+    ErtsMagicBinary *mb;
+    Uint64 value;
+    Uint32 thr_id;
+} ErtsMagicRefTableEntry;
+
+typedef struct {
+    erts_rwmtx_t rwmtx;
+    Hash hash;
+    char name[32];
+} ErtsMagicRefTable;
+
+typedef struct {
+    union {
+	ErtsMagicRefTable table;
+	char align__[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(ErtsMagicRefTable))];
+    } u;
+} ErtsAlignedMagicRefTable;
+
+ErtsAlignedMagicRefTable *magic_ref_table;
+
+ErtsMagicBinary *
+erts_magic_ref_lookup_bin__(Uint32 refn[ERTS_REF_NUMBERS])
+{
+    ErtsMagicRefTableEntry tmpl;
+    ErtsMagicRefTableEntry *tep;
+    ErtsMagicBinary *mb;
+    ErtsMagicRefTable *tblp;
+
+    ASSERT(erts_is_ref_numbers_magic(refn));
+
+    tmpl.value = erts_get_ref_numbers_value(refn);
+    tmpl.thr_id = erts_get_ref_numbers_thr_id(refn);
+    if (tmpl.thr_id > erts_no_schedulers)
+	tblp = &magic_ref_table[0].u.table;
+    else
+	tblp = &magic_ref_table[tmpl.thr_id].u.table;
+
+    erts_rwmtx_rlock(&tblp->rwmtx);
+
+    tep = (ErtsMagicRefTableEntry *) hash_get(&tblp->hash, &tmpl);
+    if (!tep)
+	mb = NULL;
+    else {
+	erts_aint_t refc;
+	mb = tep->mb;
+        refc = erts_refc_inc_unless(&mb->refc, 0, 0);
+        if (refc == 0)
+            mb = NULL;
+    }
+
+    erts_rwmtx_runlock(&tblp->rwmtx);
+
+    return mb;
+}
+
+void
+erts_magic_ref_save_bin__(Eterm ref)
+{
+    ErtsMagicRefTableEntry tmpl;
+    ErtsMagicRefTableEntry *tep;
+    ErtsMRefThing *mrtp;
+    ErtsMagicRefTable *tblp;
+    Uint32 *refn;
+
+    ASSERT(is_internal_magic_ref(ref));
+
+    mrtp = (ErtsMRefThing *) internal_ref_val(ref);
+    refn = mrtp->mb->refn;
+
+    tmpl.value = erts_get_ref_numbers_value(refn);
+    tmpl.thr_id = erts_get_ref_numbers_thr_id(refn);
+
+    if (tmpl.thr_id > erts_no_schedulers)
+	tblp = &magic_ref_table[0].u.table;
+    else
+	tblp = &magic_ref_table[tmpl.thr_id].u.table;
+
+    erts_rwmtx_rlock(&tblp->rwmtx);
+
+    tep = (ErtsMagicRefTableEntry *) hash_get(&tblp->hash, &tmpl);
+
+    erts_rwmtx_runlock(&tblp->rwmtx);
+
+    if (!tep) {
+	ErtsMagicRefTableEntry *used_tep;
+	
+	ASSERT(tmpl.value == erts_get_ref_numbers_value(refn));
+	ASSERT(tmpl.thr_id == erts_get_ref_numbers_thr_id(refn));
+
+	if (tblp != &magic_ref_table[0].u.table) {
+	    tep = erts_alloc(ERTS_ALC_T_MREF_NSCHED_ENT,
+			     sizeof(ErtsNSchedMagicRefTableEntry));
+	}
+	else {
+	    tep = erts_alloc(ERTS_ALC_T_MREF_ENT,
+			     sizeof(ErtsMagicRefTableEntry));
+	    tep->thr_id = tmpl.thr_id;
+	}
+
+	tep->value = tmpl.value;
+	tep->mb = mrtp->mb;
+
+	erts_rwmtx_rwlock(&tblp->rwmtx);
+
+	used_tep = hash_put(&tblp->hash, tep);
+
+	erts_rwmtx_rwunlock(&tblp->rwmtx);
+
+	if (used_tep != tep) {
+	    if (tblp != &magic_ref_table[0].u.table)
+		erts_free(ERTS_ALC_T_MREF_NSCHED_ENT, (void *) tep);
+	    else
+		erts_free(ERTS_ALC_T_MREF_ENT, (void *) tep);
+	}
+    }
+}
+
+void
+erts_magic_ref_remove_bin(Uint32 refn[ERTS_REF_NUMBERS])
+{
+    ErtsMagicRefTableEntry tmpl;
+    ErtsMagicRefTableEntry *tep;
+    ErtsMagicRefTable *tblp;
+
+    tmpl.value = erts_get_ref_numbers_value(refn);
+    tmpl.thr_id = erts_get_ref_numbers_thr_id(refn);
+
+    if (tmpl.thr_id > erts_no_schedulers)
+	tblp = &magic_ref_table[0].u.table;
+    else
+	tblp = &magic_ref_table[tmpl.thr_id].u.table;
+
+    erts_rwmtx_rlock(&tblp->rwmtx);
+
+    tep = (ErtsMagicRefTableEntry *) hash_get(&tblp->hash, &tmpl);
+
+    erts_rwmtx_runlock(&tblp->rwmtx);
+
+    if (tep) {
+
+	ASSERT(tmpl.value == erts_get_ref_numbers_value(refn));
+	ASSERT(tmpl.thr_id == erts_get_ref_numbers_thr_id(refn));
+
+	erts_rwmtx_rwlock(&tblp->rwmtx);
+
+	tep = hash_remove(&tblp->hash, &tmpl);
+	ASSERT(tep);
+
+	erts_rwmtx_rwunlock(&tblp->rwmtx);
+
+	if (tblp != &magic_ref_table[0].u.table)
+	    erts_free(ERTS_ALC_T_MREF_NSCHED_ENT, (void *) tep);
+	else
+	    erts_free(ERTS_ALC_T_MREF_ENT, (void *) tep);
+    }
+}
+
+static int nsched_mreft_cmp(void *ve1, void *ve2)
+{
+    ErtsNSchedMagicRefTableEntry *e1 = ve1;
+    ErtsNSchedMagicRefTableEntry *e2 = ve2;
+    return e1->value != e2->value;
+}
+
+static int non_nsched_mreft_cmp(void *ve1, void *ve2)
+{
+    ErtsMagicRefTableEntry *e1 = ve1;
+    ErtsMagicRefTableEntry *e2 = ve2;
+    return e1->value != e2->value || e1->thr_id != e2->thr_id;
+}
+
+static HashValue nsched_mreft_hash(void *ve)
+{
+    ErtsNSchedMagicRefTableEntry *e = ve;
+    return (HashValue) e->value;
+}
+
+static HashValue non_nsched_mreft_hash(void *ve)
+{
+    ErtsMagicRefTableEntry *e = ve;
+    HashValue h;
+    h = (HashValue) e->thr_id;
+    h *= 268440163;
+    h += (HashValue) e->value;
+    return h;
+}
+
+static void *mreft_alloc(void *ve)
+{
+    /*
+     * We allocate the element before
+     * hash_put() and pass it as
+     * template which we get as
+     * input...
+     */
+    return ve;
+}
+
+static void mreft_free(void *ve)
+{
+    /*
+     * We free the element ourselves
+     * after hash_remove()...
+     */
+}
+
+static void *mreft_meta_alloc(int i, size_t size)
+{
+    return erts_alloc(ERTS_ALC_T_MREF_TAB_BKTS, size);
+}
+
+static void mreft_meta_free(int i, void *ptr)
+{
+    erts_free(ERTS_ALC_T_MREF_TAB_BKTS, ptr);
+}
+
+static void
+init_magic_ref_tables(void)
+{
+    HashFunctions hash_funcs;
+    int i;
+    ErtsMagicRefTable *tblp;
+
+    magic_ref_table = erts_alloc_permanent_cache_aligned(ERTS_ALC_T_MREF_TAB,
+							 (sizeof(ErtsAlignedMagicRefTable)
+							  * (erts_no_schedulers + 1)));
+
+    hash_funcs.hash = non_nsched_mreft_hash;
+    hash_funcs.cmp = non_nsched_mreft_cmp;
+
+    hash_funcs.alloc = mreft_alloc;
+    hash_funcs.free = mreft_free;
+    hash_funcs.meta_alloc = mreft_meta_alloc;
+    hash_funcs.meta_free = mreft_meta_free;
+    hash_funcs.meta_print = erts_print;
+
+    tblp = &magic_ref_table[0].u.table;
+    erts_snprintf(&tblp->name[0], sizeof(tblp->name),
+		  "magic_ref_table_0");
+    hash_init(0, &tblp->hash, &tblp->name[0], 1, hash_funcs);
+    erts_rwmtx_init(&tblp->rwmtx, "magic_ref_table");
+
+    hash_funcs.hash = nsched_mreft_hash;
+    hash_funcs.cmp = nsched_mreft_cmp;
+
+    for (i = 1; i <= erts_no_schedulers; i++) {
+	ErtsMagicRefTable *tblp = &magic_ref_table[i].u.table;
+	erts_snprintf(&tblp->name[0], sizeof(tblp->name),
+		      "magic_ref_table_%d", i);
+	hash_init(0, &tblp->hash, &tblp->name[0], 1, hash_funcs);
+	erts_rwmtx_init(&tblp->rwmtx, "magic_ref_table");
+    }
+}
+
+void erts_ref_bin_free(ErtsMagicBinary *mb)
+{
+    erts_bin_free((Binary *) mb);
+}
+
+
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\
  * Unique Integer                                                    *
 \*                                                                   */
@@ -498,7 +788,7 @@ void
 erts_sched_bif_unique_init(ErtsSchedulerData *esdp)
 {
     esdp->unique = (Uint64) 0;
-    esdp->ref = (Uint64) 0;
+    esdp->ref = (Uint64) ref_init_value;
 }
 
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *\
@@ -513,7 +803,7 @@ BIF_RETTYPE make_ref_0(BIF_ALIST_0)
 
     ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_MAIN & erts_proc_lc_my_proc_locks(BIF_P));
 
-    hp = HAlloc(BIF_P, REF_THING_SIZE);
+    hp = HAlloc(BIF_P, ERTS_REF_THING_SIZE);
 
     res = erts_sched_make_ref_in_buffer(erts_proc_sched_data(BIF_P), hp);
 
diff --git a/erts/emulator/beam/erl_bif_unique.h b/erts/emulator/beam/erl_bif_unique.h
index c6481864d0..6cd6cc3e09 100644
--- a/erts/emulator/beam/erl_bif_unique.h
+++ b/erts/emulator/beam/erl_bif_unique.h
@@ -21,16 +21,25 @@
 #ifndef ERTS_BIF_UNIQUE_H__
 #define ERTS_BIF_UNIQUE_H__
 
+#include "erl_term.h"
 #include "erl_process.h"
 #include "big.h"
+#define ERTS_BINARY_TYPES_ONLY__
+#include "erl_binary.h"
+#undef ERTS_BINARY_TYPES_ONLY__
 
 void erts_bif_unique_init(void);
 void erts_sched_bif_unique_init(ErtsSchedulerData *esdp);
 
 /* reference */
 Eterm erts_make_ref(Process *);
-Eterm erts_make_ref_in_buffer(Eterm buffer[REF_THING_SIZE]);
-void erts_make_ref_in_array(Uint32 ref[ERTS_MAX_REF_NUMBERS]);
+Eterm erts_make_ref_in_buffer(Eterm buffer[ERTS_REF_THING_SIZE]);
+void erts_make_ref_in_array(Uint32 ref[ERTS_REF_NUMBERS]);
+void erts_make_magic_ref_in_array(Uint32 ref[ERTS_REF_NUMBERS]);
+void erts_magic_ref_remove_bin(Uint32 refn[ERTS_REF_NUMBERS]);
+void erts_magic_ref_save_bin__(Eterm ref);
+ErtsMagicBinary *erts_magic_ref_lookup_bin__(Uint32 refn[ERTS_REF_NUMBERS]);
+
 
 /* strict monotonic counter */
 
@@ -67,19 +76,35 @@ Eterm erts_debug_make_unique_integer(Process *c_p,
 				     Eterm etval1);
 
 
-ERTS_GLB_INLINE void erts_set_ref_numbers(Uint32 ref[ERTS_MAX_REF_NUMBERS],
+ERTS_GLB_INLINE void erts_set_ref_numbers(Uint32 ref[ERTS_REF_NUMBERS],
 					  Uint32 thr_id, Uint64 value);
-ERTS_GLB_INLINE Uint32 erts_get_ref_numbers_thr_id(Uint32 ref[ERTS_MAX_REF_NUMBERS]);
-ERTS_GLB_INLINE Uint64 erts_get_ref_numbers_value(Uint32 ref[ERTS_MAX_REF_NUMBERS]);
+ERTS_GLB_INLINE Uint32 erts_get_ref_numbers_thr_id(Uint32 ref[ERTS_REF_NUMBERS]);
+ERTS_GLB_INLINE int erts_is_ref_numbers_magic(Uint32 ref[ERTS_REF_NUMBERS]);
+ERTS_GLB_INLINE Uint64 erts_get_ref_numbers_value(Uint32 ref[ERTS_REF_NUMBERS]);
 ERTS_GLB_INLINE void erts_sched_make_ref_in_array(ErtsSchedulerData *esdp,
-						  Uint32 ref[ERTS_MAX_REF_NUMBERS]);
+						  Uint32 ref[ERTS_REF_NUMBERS]);
+ERTS_GLB_INLINE void erts_sched_make_magic_ref_in_array(ErtsSchedulerData *esdp,
+							Uint32 ref[ERTS_REF_NUMBERS]);
 ERTS_GLB_INLINE Eterm erts_sched_make_ref_in_buffer(ErtsSchedulerData *esdp,
-						    Eterm buffer[REF_THING_SIZE]);
+						    Eterm buffer[ERTS_REF_THING_SIZE]);
+ERTS_GLB_INLINE Eterm erts_mk_magic_ref(Eterm **hpp, ErlOffHeap *ohp, Binary *mbp);
+ERTS_GLB_INLINE Binary *erts_magic_ref2bin(Eterm mref);
+ERTS_GLB_INLINE void erts_magic_ref_save_bin(Eterm ref);
+ERTS_GLB_INLINE ErtsMagicBinary *erts_magic_ref_lookup_bin(Uint32 ref[ERTS_REF_NUMBERS]);
+
+#define ERTS_REF1_MAGIC_MARKER_BIT_NO__ \
+    (_REF_NUM_SIZE-1)
+#define ERTS_REF1_MAGIC_MARKER_BIT__ \
+    (((Uint32) 1) << ERTS_REF1_MAGIC_MARKER_BIT_NO__)
+#define ERTS_REF1_THR_ID_MASK__ \
+    (ERTS_REF1_MAGIC_MARKER_BIT__-1)
+#define ERTS_REF1_NUM_MASK__ \
+    (~(ERTS_REF1_THR_ID_MASK__|ERTS_REF1_MAGIC_MARKER_BIT__))
 
 #if ERTS_GLB_INLINE_INCL_FUNC_DEF
 
 ERTS_GLB_INLINE void
-erts_set_ref_numbers(Uint32 ref[ERTS_MAX_REF_NUMBERS], Uint32 thr_id, Uint64 value)
+erts_set_ref_numbers(Uint32 ref[ERTS_REF_NUMBERS], Uint32 thr_id, Uint64 value)
 {
     /*
      * We cannot use thread id in the first 18-bit word since
@@ -87,30 +112,39 @@ erts_set_ref_numbers(Uint32 ref[ERTS_MAX_REF_NUMBERS], Uint32 thr_id, Uint64 val
      * we did, we would get really poor hash values. Instead
      * we have to shuffle the bits a bit.
      */
-    ASSERT(thr_id == (thr_id & ((Uint32) 0x3ffff)));
-    ref[0] = (Uint32) (value & ((Uint64) 0x3ffff));
-    ref[1] = (((Uint32) (value & ((Uint64) 0xfffc0000)))
-	      | (thr_id & ((Uint32) 0x3ffff)));
+    ASSERT(thr_id == (thr_id & ((Uint32) ERTS_REF1_THR_ID_MASK__)));
+    ref[0] = (Uint32) (value & ((Uint64) REF_MASK));
+    ref[1] = (((Uint32) (value & ((Uint64) ERTS_REF1_NUM_MASK__)))
+	      | (thr_id & ((Uint32) ERTS_REF1_THR_ID_MASK__)));
     ref[2] = (Uint32) ((value >> 32) & ((Uint64) 0xffffffff));
 }
 
 ERTS_GLB_INLINE Uint32
-erts_get_ref_numbers_thr_id(Uint32 ref[ERTS_MAX_REF_NUMBERS])
+erts_get_ref_numbers_thr_id(Uint32 ref[ERTS_REF_NUMBERS])
 {
-    return ref[1] & ((Uint32) 0x3ffff);
+    return ref[1] & ((Uint32) ERTS_REF1_THR_ID_MASK__);
+}
+
+ERTS_GLB_INLINE int
+erts_is_ref_numbers_magic(Uint32 ref[ERTS_REF_NUMBERS])
+{
+    return !!(ref[1] & ERTS_REF1_MAGIC_MARKER_BIT__);
 }
 
 ERTS_GLB_INLINE Uint64
-erts_get_ref_numbers_value(Uint32 ref[ERTS_MAX_REF_NUMBERS])
+erts_get_ref_numbers_value(Uint32 ref[ERTS_REF_NUMBERS])
 {
+    ERTS_CT_ASSERT((ERTS_REF1_NUM_MASK__ | REF_MASK) == 0xffffffff);
+    ERTS_CT_ASSERT((ERTS_REF1_NUM_MASK__ & REF_MASK) == 0);
+
     return (((((Uint64) ref[2]) & ((Uint64) 0xffffffff)) << 32)
-	    | (((Uint64) ref[1]) & ((Uint64) 0xfffc0000))
-	    | (((Uint64) ref[0]) & ((Uint64) 0x3ffff)));
+	    | (((Uint64) ref[1]) & ((Uint64) ERTS_REF1_NUM_MASK__))
+	    | (((Uint64) ref[0]) & ((Uint64) REF_MASK)));
 }
 
 ERTS_GLB_INLINE void
 erts_sched_make_ref_in_array(ErtsSchedulerData *esdp,
-			     Uint32 ref[ERTS_MAX_REF_NUMBERS])
+			     Uint32 ref[ERTS_REF_NUMBERS])
 {
     Uint64 value;
 
@@ -119,18 +153,286 @@ erts_sched_make_ref_in_array(ErtsSchedulerData *esdp,
     erts_set_ref_numbers(ref, (Uint32) esdp->thr_id, value);
 }
 
+ERTS_GLB_INLINE void
+erts_sched_make_magic_ref_in_array(ErtsSchedulerData *esdp,
+				   Uint32 ref[ERTS_REF_NUMBERS])
+{
+    erts_sched_make_ref_in_array(esdp, ref);
+    ASSERT(!(ref[1] & ERTS_REF1_MAGIC_MARKER_BIT__));
+    ref[1] |= ERTS_REF1_MAGIC_MARKER_BIT__;
+}
+
 ERTS_GLB_INLINE Eterm
 erts_sched_make_ref_in_buffer(ErtsSchedulerData *esdp,
-			      Eterm buffer[REF_THING_SIZE])
+			      Eterm buffer[ERTS_REF_THING_SIZE])
 {
     Eterm* hp = buffer;
-    Uint32 ref[ERTS_MAX_REF_NUMBERS];
+    Uint32 ref[ERTS_REF_NUMBERS];
 
     erts_sched_make_ref_in_array(esdp, ref);
     write_ref_thing(hp, ref[0], ref[1], ref[2]);
     return make_internal_ref(hp);
 }
 
+ERTS_GLB_INLINE Eterm
+erts_mk_magic_ref(Eterm **hpp, ErlOffHeap *ohp, Binary *bp)
+{
+    Eterm *hp = *hpp;
+    ASSERT(bp->flags & BIN_FLAG_MAGIC);
+    write_magic_ref_thing(hp, ohp, (ErtsMagicBinary *) bp);
+    *hpp += ERTS_MAGIC_REF_THING_SIZE;
+    erts_refc_inc(&bp->refc, 1);
+    return make_internal_ref(hp);
+}
+
+ERTS_GLB_INLINE Binary *
+erts_magic_ref2bin(Eterm mref)
+{
+    ErtsMRefThing *mrtp;
+    ASSERT(is_internal_magic_ref(mref));
+    mrtp = (ErtsMRefThing *) internal_ref_val(mref);
+    return (Binary *) mrtp->mb;
+}
+
+/*
+ * Save the magic binary of a ref when the
+ * ref is exposed to the outside world...
+ */
+ERTS_GLB_INLINE void
+erts_magic_ref_save_bin(Eterm ref)
+{
+    if (is_internal_magic_ref(ref))
+	erts_magic_ref_save_bin__(ref);
+}
+
+/*
+ * Look up the magic binary of a magic ref
+ * when the ref comes from the outside world...
+ */
+ERTS_GLB_INLINE ErtsMagicBinary *
+erts_magic_ref_lookup_bin(Uint32 ref[ERTS_REF_NUMBERS])
+{
+    if (!erts_is_ref_numbers_magic(ref))
+	return NULL;
+    return erts_magic_ref_lookup_bin__(ref);
+}
+
+
 #endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */
 
+
+/*
+ * Storage of internal refs in misc structures...
+ */
+
+#include "erl_message.h"
+
+#if ERTS_REF_NUMBERS != 3
+#  error fix this...
+#endif
+
+ERTS_GLB_INLINE int erts_internal_ref_number_cmp(Uint32 num1[ERTS_REF_NUMBERS],
+						 Uint32 num2[ERTS_REF_NUMBERS]);
+
+#if ERTS_GLB_INLINE_INCL_FUNC_DEF
+
+ERTS_GLB_INLINE int
+erts_internal_ref_number_cmp(Uint32 num1[ERTS_REF_NUMBERS],
+			     Uint32 num2[ERTS_REF_NUMBERS])
+{
+    if (num1[2] != num2[2])
+	return (int) ((Sint64) num1[2] - (Sint64) num2[2]);
+    if (num1[1] != num2[1])
+	return (int) ((Sint64) num1[1] - (Sint64) num2[1]);
+    if (num1[0] != num2[0])
+	return (int) ((Sint64) num1[0] - (Sint64) num2[0]);
+    return 0;
+}
+
+#endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */
+
+/* Iref storage for all internal references... */ 
+typedef struct {
+    Uint32 is_magic;
+    union {
+	ErtsMagicBinary *mb;
+	Uint32 num[ERTS_REF_NUMBERS];
+    } u;
+} ErtsIRefStorage;
+
+void erts_ref_bin_free(ErtsMagicBinary *mb);
+
+ERTS_GLB_INLINE void erts_iref_storage_save(ErtsIRefStorage *iref, Eterm ref);
+ERTS_GLB_INLINE void erts_iref_storage_clean(ErtsIRefStorage *iref);
+ERTS_GLB_INLINE Uint erts_iref_storage_heap_size(ErtsIRefStorage *iref);
+ERTS_GLB_INLINE Eterm erts_iref_storage_make_ref(ErtsIRefStorage *iref,
+						 Eterm **hpp, ErlOffHeap *ohp,
+						 int clean_storage);
+ERTS_GLB_INLINE int erts_iref_storage_cmp(ErtsIRefStorage *iref1,
+					  ErtsIRefStorage *iref2);
+
+
+#if ERTS_GLB_INLINE_INCL_FUNC_DEF
+
+ERTS_GLB_INLINE void
+erts_iref_storage_save(ErtsIRefStorage *iref, Eterm ref)
+{
+    Eterm *hp;
+
+    ERTS_CT_ASSERT(ERTS_REF_NUMBERS == 3);
+    ASSERT(is_internal_ref(ref));
+
+    hp = boxed_val(ref);
+
+    if (is_ordinary_ref_thing(hp)) {
+	ErtsORefThing *rtp = (ErtsORefThing *) hp;
+	iref->is_magic = 0;
+	iref->u.num[0] = rtp->num[0];
+	iref->u.num[1] = rtp->num[1];
+	iref->u.num[2] = rtp->num[2];
+    }
+    else {
+	ErtsMRefThing *mrtp = (ErtsMRefThing *) hp;
+	ASSERT(is_magic_ref_thing(hp));
+	iref->is_magic = 1;
+	iref->u.mb = mrtp->mb;
+	erts_refc_inc(&mrtp->mb->refc, 1);
+    }
+}
+
+ERTS_GLB_INLINE void
+erts_iref_storage_clean(ErtsIRefStorage *iref)
+{
+    if (iref->is_magic && erts_refc_dectest(&iref->u.mb->refc, 0) == 0)
+	erts_ref_bin_free(iref->u.mb);
+#ifdef DEBUG
+    memset((void *) iref, 0xf, sizeof(ErtsIRefStorage));
+#endif
+}
+
+ERTS_GLB_INLINE Uint
+erts_iref_storage_heap_size(ErtsIRefStorage *iref)
+{
+    return iref->is_magic ? ERTS_MAGIC_REF_THING_SIZE : ERTS_REF_THING_SIZE;
+}
+
+ERTS_GLB_INLINE Eterm
+erts_iref_storage_make_ref(ErtsIRefStorage *iref,
+			   Eterm **hpp, ErlOffHeap *ohp,
+			   int clean_storage)
+{
+    Eterm *hp = *hpp;
+    if (!iref->is_magic) {
+	write_ref_thing(hp, iref->u.num[0], iref->u.num[1],
+			iref->u.num[2]);
+	*hpp += ERTS_REF_THING_SIZE;
+    }
+    else {
+	write_magic_ref_thing(hp, ohp, iref->u.mb);
+	*hpp += ERTS_MAGIC_REF_THING_SIZE;
+	/*
+	 * If we clean storage, the term inherits the
+	 * refc increment of the cleaned storage...
+	 */
+	if (!clean_storage)
+	    erts_refc_inc(&iref->u.mb->refc, 1);
+    }
+
+#ifdef DEBUG
+    if (clean_storage)
+	memset((void *) iref, 0xf, sizeof(ErtsIRefStorage));
+#endif
+
+    return make_internal_ref(hp);
+}
+
+ERTS_GLB_INLINE int
+erts_iref_storage_cmp(ErtsIRefStorage *iref1,
+		      ErtsIRefStorage *iref2)
+{
+    Uint32 *num1 = iref1->is_magic ? iref1->u.mb->refn : iref1->u.num;
+    Uint32 *num2 = iref2->is_magic ? iref2->u.mb->refn : iref2->u.num;
+    return erts_internal_ref_number_cmp(num1, num2);
+}
+
+#endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */
+
+
+/* OIref storage for ordinary internal references only... */ 
+typedef struct {
+    Uint32 num[ERTS_REF_NUMBERS];
+} ErtsOIRefStorage;
+
+ERTS_GLB_INLINE void erts_oiref_storage_save(ErtsOIRefStorage *oiref,
+					     Eterm ref);
+ERTS_GLB_INLINE Eterm erts_oiref_storage_make_ref(ErtsOIRefStorage *oiref,
+						  Eterm **hpp);
+ERTS_GLB_INLINE int erts_oiref_storage_cmp(ErtsOIRefStorage *oiref1,
+					   ErtsOIRefStorage *oiref2);
+
+#if ERTS_GLB_INLINE_INCL_FUNC_DEF
+
+ERTS_GLB_INLINE void
+erts_oiref_storage_save(ErtsOIRefStorage *oiref, Eterm ref)
+{
+    ErtsORefThing *rtp;
+    ERTS_CT_ASSERT(ERTS_REF_NUMBERS == 3);
+    ASSERT(is_internal_ordinary_ref(ref));
+
+    rtp = (ErtsORefThing *) internal_ref_val(ref);
+
+    oiref->num[0] = rtp->num[0];
+    oiref->num[1] = rtp->num[1];
+    oiref->num[2] = rtp->num[2];
+}
+
+ERTS_GLB_INLINE Eterm
+erts_oiref_storage_make_ref(ErtsOIRefStorage *oiref, Eterm **hpp)
+{
+    Eterm *hp = *hpp;
+    ERTS_CT_ASSERT(ERTS_REF_NUMBERS == 3);
+    write_ref_thing(hp, oiref->num[0], oiref->num[1], oiref->num[2]);
+    *hpp += ERTS_REF_THING_SIZE;
+    return make_internal_ref(hp);
+}
+
+ERTS_GLB_INLINE int
+erts_oiref_storage_cmp(ErtsOIRefStorage *oiref1,
+		       ErtsOIRefStorage *oiref2)
+{
+    return erts_internal_ref_number_cmp(oiref1->num, oiref2->num);
+}
+
+#endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */
+
+
+ERTS_GLB_INLINE Eterm
+erts_proc_store_ref(Process *c_p, Uint32 ref[ERTS_MAX_REF_NUMBERS]);
+
+#if ERTS_GLB_INLINE_INCL_FUNC_DEF
+
+ERTS_GLB_INLINE Eterm
+erts_proc_store_ref(Process *c_p, Uint32 ref[ERTS_MAX_REF_NUMBERS])
+{
+    Eterm *hp = HAlloc(c_p, ERTS_REF_THING_SIZE);
+    write_ref_thing(hp, ref[0], ref[1], ref[2]);
+    return make_internal_ref(hp);
+}
+
+#endif
+
 #endif /*  ERTS_BIF_UNIQUE_H__ */
+
+#if (defined(ERTS_ALLOC_C__) || defined(ERL_BIF_UNIQUE_C__)) \
+    && !defined(ERTS_BIF_UNIQUE_H__FIX_ALLOC_TYPES__)
+#define ERTS_BIF_UNIQUE_H__FIX_ALLOC_TYPES__
+
+#include "hash.h"
+
+typedef struct {
+    HashBucket hash;
+    ErtsMagicBinary *mb;
+    Uint64 value;
+} ErtsNSchedMagicRefTableEntry;
+
+#endif
diff --git a/erts/emulator/beam/erl_binary.h b/erts/emulator/beam/erl_binary.h
index fdc5efea98..c811a002ce 100644
--- a/erts/emulator/beam/erl_binary.h
+++ b/erts/emulator/beam/erl_binary.h
@@ -18,11 +18,146 @@
  * %CopyrightEnd%
  */
 
-#ifndef __ERL_BINARY_H
-#define __ERL_BINARY_H
+#ifndef ERL_BINARY_H__TYPES__
+#define ERL_BINARY_H__TYPES__
+
+/*
+** Just like the driver binary but with initial flags
+** Note that the two structures Binary and ErlDrvBinary HAVE to
+** be equal except for extra fields in the beginning of the struct.
+** ErlDrvBinary is defined in erl_driver.h.
+** When driver_alloc_binary is called, a Binary is allocated, but 
+** the pointer returned is to the address of the first element that
+** also occurs in the ErlDrvBinary struct (driver.*binary takes care if this).
+** The driver need never know about additions to the internal Binary of the
+** emulator. One should however NEVER be sloppy when mixing ErlDrvBinary
+** and Binary, the macros below can convert one type to the other, as they both
+** in reality are equal.
+*/
+
+#ifdef ARCH_32
+ /* *DO NOT USE* only for alignment. */
+#define ERTS_BINARY_STRUCT_ALIGNMENT Uint32 align__;
+#else
+#define ERTS_BINARY_STRUCT_ALIGNMENT
+#endif
+
+/* Add fields in ERTS_INTERNAL_BINARY_FIELDS, otherwise the drivers crash */
+#define ERTS_INTERNAL_BINARY_FIELDS				\
+    UWord flags;						\
+    erts_refc_t refc;						\
+    ERTS_BINARY_STRUCT_ALIGNMENT
+
+typedef struct binary {
+    ERTS_INTERNAL_BINARY_FIELDS
+    SWord orig_size;
+    char orig_bytes[1]; /* to be continued */
+} Binary;
+
+#define ERTS_SIZEOF_Binary(Sz) \
+    (offsetof(Binary,orig_bytes) + (Sz))
+
+#if ERTS_REF_NUMBERS != 3
+#error "Update ErtsMagicBinary"
+#endif
+
+typedef struct magic_binary ErtsMagicBinary;
+struct magic_binary {
+    ERTS_INTERNAL_BINARY_FIELDS
+    SWord orig_size;
+    void (*destructor)(Binary *);
+    Uint32 refn[ERTS_REF_NUMBERS];
+    ErtsAlcType_t alloc_type;
+    union {
+        struct {
+            ERTS_BINARY_STRUCT_ALIGNMENT
+            char data[1];
+        } aligned;
+        struct {
+            char data[1];
+        } unaligned;
+    } u;
+};
+
+#ifdef ARCH_32
+#define ERTS_MAGIC_BIN_BYTES_TO_ALIGN 4
+#else
+#define ERTS_MAGIC_BIN_BYTES_TO_ALIGN 0
+#endif
+
+typedef union {
+    Binary binary;
+    ErtsMagicBinary magic_binary;
+    struct {
+	ERTS_INTERNAL_BINARY_FIELDS
+	ErlDrvBinary binary;
+    } driver;
+} ErtsBinary;
+
+/*
+ * 'Binary' alignment:
+ *   Address of orig_bytes[0] of a Binary should always be 8-byte aligned.
+ * It is assumed that the flags, refc, and orig_size fields are 4 bytes on
+ * 32-bits architectures and 8 bytes on 64-bits architectures.
+ */
+
+#define ERTS_MAGIC_BIN_REFN(BP) \
+  ((ErtsBinary *) (BP))->magic_binary.refn
+#define ERTS_MAGIC_BIN_ATYPE(BP) \
+  ((ErtsBinary *) (BP))->magic_binary.alloc_type
+#define ERTS_MAGIC_DATA_OFFSET \
+  (offsetof(ErtsMagicBinary,u.aligned.data) - offsetof(Binary,orig_bytes))
+#define ERTS_MAGIC_BIN_DESTRUCTOR(BP) \
+  ((ErtsBinary *) (BP))->magic_binary.destructor
+#define ERTS_MAGIC_BIN_DATA(BP) \
+  ((void *) ((ErtsBinary *) (BP))->magic_binary.u.aligned.data)
+#define ERTS_MAGIC_BIN_DATA_SIZE(BP) \
+  ((BP)->orig_size - ERTS_MAGIC_DATA_OFFSET)
+#define ERTS_MAGIC_DATA_OFFSET \
+  (offsetof(ErtsMagicBinary,u.aligned.data) - offsetof(Binary,orig_bytes))
+#define ERTS_MAGIC_BIN_ORIG_SIZE(Sz) \
+  (ERTS_MAGIC_DATA_OFFSET + (Sz))
+#define ERTS_MAGIC_BIN_SIZE(Sz) \
+  (offsetof(ErtsMagicBinary,u.aligned.data) + (Sz))
+#define ERTS_MAGIC_BIN_FROM_DATA(DATA) \
+  ((ErtsBinary*)((char*)(DATA) - offsetof(ErtsMagicBinary,u.aligned.data)))
+
+/* On 32-bit arch these macro variants will save memory
+   by not forcing 8-byte alignment for the magic payload.
+*/
+#define ERTS_MAGIC_BIN_UNALIGNED_DATA(BP) \
+  ((void *) ((ErtsBinary *) (BP))->magic_binary.u.unaligned.data)
+#define ERTS_MAGIC_UNALIGNED_DATA_OFFSET \
+  (offsetof(ErtsMagicBinary,u.unaligned.data) - offsetof(Binary,orig_bytes))
+#define ERTS_MAGIC_BIN_UNALIGNED_DATA_SIZE(BP) \
+  ((BP)->orig_size - ERTS_MAGIC_UNALIGNED_DATA_OFFSET)
+#define ERTS_MAGIC_BIN_UNALIGNED_ORIG_SIZE(Sz) \
+  (ERTS_MAGIC_UNALIGNED_DATA_OFFSET + (Sz))
+#define ERTS_MAGIC_BIN_UNALIGNED_SIZE(Sz) \
+  (offsetof(ErtsMagicBinary,u.unaligned.data) + (Sz))
+#define ERTS_MAGIC_BIN_FROM_UNALIGNED_DATA(DATA) \
+  ((ErtsBinary*)((char*)(DATA) - offsetof(ErtsMagicBinary,u.unaligned.data)))
+
+
+#define Binary2ErlDrvBinary(B) (&((ErtsBinary *) (B))->driver.binary)
+#define ErlDrvBinary2Binary(D) ((Binary *) \
+				(((char *) (D)) \
+				 - offsetof(ErtsBinary, driver.binary)))
+
+/* A "magic" binary flag */
+#define BIN_FLAG_MAGIC      1
+#define BIN_FLAG_USR1       2 /* Reserved for use by different modules too mark */
+#define BIN_FLAG_USR2       4 /*  certain binaries as special (used by ets) */
+#define BIN_FLAG_DRV        8
+
+#endif /* ERL_BINARY_H__TYPES__ */
+
+#if !defined(ERL_BINARY_H__) && !defined(ERTS_BINARY_TYPES_ONLY__)
+#define ERL_BINARY_H__
 
 #include "erl_threads.h"
 #include "bif.h"
+#include "erl_bif_unique.h"
 
 /*
  * Maximum number of bytes to place in a heap binary.
@@ -190,6 +325,7 @@ ERTS_GLB_INLINE Binary *erts_bin_realloc(Binary *bp, Uint size);
 ERTS_GLB_INLINE void erts_bin_free(Binary *bp);
 ERTS_GLB_INLINE Binary *erts_create_magic_binary_x(Uint size,
                                                   void (*destructor)(Binary *),
+                                                   ErtsAlcType_t alloc_type,
                                                   int unaligned);
 ERTS_GLB_INLINE Binary *erts_create_magic_binary(Uint size,
 						 void (*destructor)(Binary *));
@@ -320,9 +456,12 @@ erts_bin_realloc(Binary *bp, Uint size)
 ERTS_GLB_INLINE void
 erts_bin_free(Binary *bp)
 {
-    if (bp->flags & BIN_FLAG_MAGIC)
+    if (bp->flags & BIN_FLAG_MAGIC) {
 	ERTS_MAGIC_BIN_DESTRUCTOR(bp)(bp);
-    if (bp->flags & BIN_FLAG_DRV)
+	erts_magic_ref_remove_bin(ERTS_MAGIC_BIN_REFN(bp));
+        erts_free(ERTS_MAGIC_BIN_ATYPE(bp), (void *) bp);
+    }
+    else if (bp->flags & BIN_FLAG_DRV)
 	erts_free(ERTS_ALC_T_DRV_BINARY, (void *) bp);
     else
 	erts_free(ERTS_ALC_T_BINARY, (void *) bp);
@@ -330,29 +469,33 @@ erts_bin_free(Binary *bp)
 
 ERTS_GLB_INLINE Binary *
 erts_create_magic_binary_x(Uint size, void (*destructor)(Binary *),
+                           ErtsAlcType_t alloc_type,
                            int unaligned)
 {
     Uint bsize = unaligned ? ERTS_MAGIC_BIN_UNALIGNED_SIZE(size)
                            : ERTS_MAGIC_BIN_SIZE(size);
-    Binary* bptr = erts_alloc_fnf(ERTS_ALC_T_BINARY, bsize);
+    Binary* bptr = erts_alloc_fnf(alloc_type, bsize);
     ASSERT(bsize > size);
     if (!bptr)
-	erts_alloc_n_enomem(ERTS_ALC_T2N(ERTS_ALC_T_BINARY), bsize);
+	erts_alloc_n_enomem(ERTS_ALC_T2N(alloc_type), bsize);
     ERTS_CHK_BIN_ALIGNMENT(bptr);
     bptr->flags = BIN_FLAG_MAGIC;
     bptr->orig_size = unaligned ? ERTS_MAGIC_BIN_UNALIGNED_ORIG_SIZE(size)
                                 : ERTS_MAGIC_BIN_ORIG_SIZE(size);
     erts_refc_init(&bptr->refc, 0);
     ERTS_MAGIC_BIN_DESTRUCTOR(bptr) = destructor;
+    ERTS_MAGIC_BIN_ATYPE(bptr) = alloc_type;
+    erts_make_magic_ref_in_array(ERTS_MAGIC_BIN_REFN(bptr));
     return bptr;
 }
 
 ERTS_GLB_INLINE Binary *
 erts_create_magic_binary(Uint size, void (*destructor)(Binary *))
 {
-    return erts_create_magic_binary_x(size, destructor, 0);
+    return erts_create_magic_binary_x(size, destructor,
+                                      ERTS_ALC_T_BINARY, 0);
 }
 
 #endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */
 
-#endif /* !__ERL_BINARY_H */
+#endif /* !ERL_BINARY_H__ */
diff --git a/erts/emulator/beam/erl_db_util.c b/erts/emulator/beam/erl_db_util.c
index 5537935802..0ab42394ce 100644
--- a/erts/emulator/beam/erl_db_util.c
+++ b/erts/emulator/beam/erl_db_util.c
@@ -5289,24 +5289,35 @@ void db_match_dis(Binary *bp)
 	case matchEqRef:
 	    ++t;
 	    {
-		RefThing *rt = (RefThing *) t;
+		Uint32 *num;
 		int ri;
-		n = thing_arityval(rt->header);
-		erts_printf("EqRef\t(%d) {", (int) n);
+
+		if (is_ordinary_ref_thing(t)) {
+		    ErtsORefThing *rt = (ErtsORefThing *) t;
+		    num = rt->num;
+		    t += TermWords(ERTS_REF_THING_SIZE);
+		}
+		else {
+		    ErtsMRefThing *mrt = (ErtsMRefThing *) t;
+		    ASSERT(is_magic_ref_thing(t));
+		    num = mrt->mb->refn;
+		    t += TermWords(ERTS_MAGIC_REF_THING_SIZE);
+		}
+
+		erts_printf("EqRef\t(%d) {", (int) ERTS_REF_NUMBERS);
 		first = 1;
-		for (ri = 0; ri < n; ++ri) {
+		for (ri = 0; ri < ERTS_REF_NUMBERS; ++ri) {
 		    if (first)
 			first = 0;
 		    else
 			erts_printf(", ");
 #if defined(ARCH_64)
-		    erts_printf("0x%016bex", rt->data.ui[ri]);
+		    erts_printf("0x%016bex", num[ri]);
 #else
-		    erts_printf("0x%08bex", rt->data.ui[ri]);
+		    erts_printf("0x%08bex", num[ri]);
 #endif
 		}
 	    }
-	    t += TermWords(REF_THING_SIZE);
 	    erts_printf("}\n");
 	    break;
 	case matchEqBig:
diff --git a/erts/emulator/beam/erl_debug.c b/erts/emulator/beam/erl_debug.c
index 3526bb684d..517dad9cbb 100644
--- a/erts/emulator/beam/erl_debug.c
+++ b/erts/emulator/beam/erl_debug.c
@@ -130,7 +130,7 @@ pdisplay1(fmtfn_t to, void *to_arg, Process* p, Eterm obj)
 	Uint32 *ref_num;
 	erts_print(to, to_arg, "#Ref<%lu", ref_channel_no(obj));
 	ref_num = ref_numbers(obj);
-	for (i = ref_no_of_numbers(obj)-1; i >= 0; i--)
+	for (i = ref_no_numbers(obj)-1; i >= 0; i--)
 	    erts_print(to, to_arg, ",%lu", ref_num[i]);
 	erts_print(to, to_arg, ">");
 	break;
diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c
index 897dbbe82b..807d9d15e4 100644
--- a/erts/emulator/beam/erl_gc.c
+++ b/erts/emulator/beam/erl_gc.c
@@ -178,7 +178,7 @@ Uint erts_test_long_gc_sleep; /* Only used for testing... */
 typedef struct {
     Process *proc;
     Eterm ref;
-    Eterm ref_heap[REF_THING_SIZE];
+    Eterm ref_heap[ERTS_REF_THING_SIZE];
     Uint req_sched;
     erts_smp_atomic32_t refc;
 } ErtsGCInfoReq;
@@ -2354,6 +2354,9 @@ copy_one_frag(Eterm** hpp, ErlOffHeap* off_heap,
 	    switch (val & _HEADER_SUBTAG_MASK) {
 	    case ARITYVAL_SUBTAG:
 		break;
+	    case REF_SUBTAG:
+		if (is_ordinary_ref_thing(fhp - 1))
+		    goto the_default;
 	    case REFC_BINARY_SUBTAG:
 	    case FUN_SUBTAG:
 	    case EXTERNAL_PID_SUBTAG:
@@ -2363,6 +2366,7 @@ copy_one_frag(Eterm** hpp, ErlOffHeap* off_heap,
 		cpy_sz = thing_arityval(val);
 		goto cpy_words;
 	    default:
+	    the_default:
 		cpy_sz = header_arity(val);
 
 	    cpy_words:
@@ -2862,6 +2866,15 @@ sweep_off_heap(Process *p, int fullsweep)
 		    }
 		    break;
 		}
+	    case REF_SUBTAG:
+		{
+		    ErtsMagicBinary *bptr;
+		    ASSERT(is_magic_ref_thing(ptr));
+		    bptr = ((ErtsMRefThing *) ptr)->mb;
+		    if (erts_refc_dectest(&bptr->refc, 0) == 0)
+			erts_bin_free((Binary *) bptr);		    
+		    break;
+		}
 	    default:
 		ASSERT(is_external_header(ptr->thing_word));
 		erts_deref_node_entry(((ExternalThing*)ptr)->node);
@@ -2883,7 +2896,8 @@ sweep_off_heap(Process *p, int fullsweep)
 	}
 	else {
 	    ASSERT(is_fun_header(ptr->thing_word) ||
-		   is_external_header(ptr->thing_word));
+		   is_external_header(ptr->thing_word)
+		   || is_magic_ref_thing(ptr));
 	    prev = &ptr->next;
 	    ptr = ptr->next;
 	}
@@ -2978,6 +2992,9 @@ offset_heap(Eterm* hp, Uint sz, Sint offs, char* area, Uint area_size)
 	      }
 	      tari = thing_arityval(val);
 	      switch (thing_subtag(val)) {
+	      case REF_SUBTAG:
+		  if (is_ordinary_ref_thing(hp))
+		      break;
 	      case REFC_BINARY_SUBTAG:
 	      case FUN_SUBTAG:
 	      case EXTERNAL_PID_SUBTAG:
@@ -3175,7 +3192,7 @@ reply_gc_info(void *vgcirp)
 	if (hpp)
 	    ref_copy = STORE_NC(hpp, ohp, gcirp->ref);
 	else
-	    *szp += REF_THING_SIZE;
+	    *szp += ERTS_REF_THING_SIZE;
 
 	msg = erts_bld_tuple(hpp, szp, 3,
 			     make_small(esdp->no),
@@ -3526,6 +3543,10 @@ erts_check_off_heap2(Process *p, Eterm *htop)
 	case EXTERNAL_REF_SUBTAG:
 	    refc = erts_smp_refc_read(&u.ext->node->refc, 1);
 	    break;
+	case REF_SUBTAG:
+	    ASSERT(is_magic_ref_thing(u.hdr));
+	    refc = erts_refc_read(&u.mref->mb->refc, 1);
+	    break;
 	default:
 	    ASSERT(!"erts_check_off_heap2: Invalid thing_word");
 	}
diff --git a/erts/emulator/beam/erl_hl_timer.c b/erts/emulator/beam/erl_hl_timer.c
index 647fa26811..26be8c7edf 100644
--- a/erts/emulator/beam/erl_hl_timer.c
+++ b/erts/emulator/beam/erl_hl_timer.c
@@ -1771,7 +1771,7 @@ setup_bif_timer(Process *c_p, ErtsMonotonicTime timeout_pos,
 
     esdp = erts_proc_sched_data(c_p);
 
-    hp = HAlloc(c_p, REF_THING_SIZE);
+    hp = HAlloc(c_p, ERTS_REF_THING_SIZE);
     ref = erts_sched_make_ref_in_buffer(esdp, hp);
 
     ASSERT(erts_get_ref_numbers_thr_id(
@@ -1939,7 +1939,7 @@ access_sched_local_btm(Process *c_p, Eterm pid,
 	Eterm *hp_end;
 #endif
 
-	hsz = REF_THING_SIZE;
+	hsz = ERTS_REF_THING_SIZE;
 	if (async) {
 	    refn = trefn; /* timer ref */
 	    hsz += 4; /* 3-tuple */
@@ -1973,7 +1973,7 @@ access_sched_local_btm(Process *c_p, Eterm pid,
 			refn[1],
 			refn[2]);
 	ref = make_internal_ref(hp);
-	hp += REF_THING_SIZE;
+	hp += ERTS_REF_THING_SIZE;
 
 	msg = (async
 	       ? TUPLE3(hp, (cancel
@@ -2087,7 +2087,7 @@ try_access_sched_remote_btm(ErtsSchedulerData *esdp,
 	ErtsProcLocks proc_locks = ERTS_PROC_LOCK_MAIN;
 	ErlOffHeap *ohp;
 
-	hsz = 4 + REF_THING_SIZE;
+	hsz = 4 + ERTS_REF_THING_SIZE;
 	if (time_left > (Sint64) MAX_SMALL)
 	    hsz += ERTS_SINT64_HEAP_SIZE(time_left);
 
@@ -2103,7 +2103,7 @@ try_access_sched_remote_btm(ErtsSchedulerData *esdp,
 			trefn[1],
 			trefn[2]);
 	tref = make_internal_ref(hp);
-	hp += REF_THING_SIZE;
+	hp += ERTS_REF_THING_SIZE;
 
 	if (time_left < 0)
 	    res = am_false;
@@ -2189,7 +2189,7 @@ access_bif_timer(Process *c_p, Eterm tref, int cancel, int async, int info)
 	    Eterm *hp, rref;
 	    Uint32 *rrefn;
 
-	    hp = HAlloc(c_p, REF_THING_SIZE);
+	    hp = HAlloc(c_p, ERTS_REF_THING_SIZE);
 	    rref = erts_sched_make_ref_in_buffer(esdp, hp);
 	    rrefn = internal_ref_numbers(rref);
 
diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c
index 0646665b07..b2c307e826 100644
--- a/erts/emulator/beam/erl_init.c
+++ b/erts/emulator/beam/erl_init.c
@@ -111,10 +111,12 @@ const int etp_lock_check = 1;
 #else
 const int etp_lock_check = 0;
 #endif
-#ifdef WORDS_BIGENDIAN
-const int etp_big_endian = 1;
+const int etp_endianness = ERTS_ENDIANNESS;
+const Eterm etp_ref_header = ERTS_REF_THING_HEADER;
+#ifdef ERTS_MAGIC_REF_THING_HEADER
+const Eterm etp_magic_ref_header = ERTS_MAGIC_REF_THING_HEADER;
 #else
-const int etp_big_endian = 0;
+const Eterm etp_magic_ref_header = ERTS_REF_THING_HEADER;
 #endif
 const Eterm etp_the_non_value = THE_NON_VALUE;
 #ifdef ERTS_HOLE_MARKER
@@ -772,6 +774,8 @@ early_init(int *argc, char **argv) /*
     H_MAX_SIZE = H_DEFAULT_MAX_SIZE;
     H_MAX_FLAGS = MAX_HEAP_SIZE_KILL|MAX_HEAP_SIZE_LOG;
 
+    erts_term_init();
+
     erts_initialized = 0;
 
     erts_use_sender_punish = 1;
@@ -1143,6 +1147,8 @@ early_init(int *argc, char **argv) /*
     no_schedulers_online = schdlrs_onln;
 
     erts_no_schedulers = (Uint) no_schedulers;
+#else
+    erts_no_schedulers = 1;
 #endif
 #ifdef ERTS_DIRTY_SCHEDULERS
     erts_no_dirty_cpu_schedulers = no_dirty_cpu_schedulers = dirty_cpu_scheds;
diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c
index 08dcbed91c..c98581a45e 100644
--- a/erts/emulator/beam/erl_lock_check.c
+++ b/erts/emulator/beam/erl_lock_check.c
@@ -156,6 +156,7 @@ static erts_lc_lock_order_t erts_lock_order[] = {
     {	"tracer_mtx", 				NULL			},
     {   "port_table",                           NULL                    },
 #endif
+    {	"magic_ref_table",			"address"		},
     {	"mtrace_op",				NULL			},
     {	"instr_x",				NULL			},
     {	"instr",				NULL			},
diff --git a/erts/emulator/beam/erl_message.c b/erts/emulator/beam/erl_message.c
index 5e160de080..f181c1e3cb 100644
--- a/erts/emulator/beam/erl_message.c
+++ b/erts/emulator/beam/erl_message.c
@@ -176,6 +176,11 @@ erts_cleanup_offheap(ErlOffHeap *offheap)
 		erts_erase_fun_entry(u.fun->fe);		    
 	    }
 	    break;
+	case REF_SUBTAG:
+	    ASSERT(is_magic_ref_thing(u.hdr));
+	    if (erts_refc_dectest(&u.mref->mb->refc, 0) == 0)
+		erts_bin_free((Binary *)u.mref->mb);		    
+	    break;
 	default:
 	    ASSERT(is_external_header(u.hdr->thing_word));
 	    erts_deref_node_entry(u.ext->node);
diff --git a/erts/emulator/beam/erl_monitors.c b/erts/emulator/beam/erl_monitors.c
index c207dea10f..bdfc7f30d9 100644
--- a/erts/emulator/beam/erl_monitors.c
+++ b/erts/emulator/beam/erl_monitors.c
@@ -45,6 +45,7 @@
 #include "bif.h"
 #include "big.h"
 #include "erl_monitors.h"
+#include "erl_bif_unique.h"
 
 #define STACK_NEED 50
 #define MAX_MONITORS 0xFFFFFFFFUL
@@ -79,7 +80,24 @@ static ERTS_INLINE int cmp_mon_ref(Eterm ref1, Eterm ref2)
     b2 = boxed_val(ref2);
     if (is_ref_thing_header(*b1)) {
 	if (is_ref_thing_header(*b2)) {
-	    return memcmp(b1+1,b2+1,ERTS_REF_WORDS*sizeof(Uint));
+	    Uint32 *num1, *num2;
+	    if (is_ordinary_ref_thing(b1)) {
+		ErtsORefThing *rtp = (ErtsORefThing *) b1;
+		num1 = rtp->num;
+	    }
+	    else {
+		ErtsMRefThing *mrtp = (ErtsMRefThing *) b1;
+		num1 = mrtp->mb->refn;
+	    }
+	    if (is_ordinary_ref_thing(b2)) {
+		ErtsORefThing *rtp = (ErtsORefThing *) b2;
+		num2 = rtp->num;
+	    }
+	    else {
+		ErtsMRefThing *mrtp = (ErtsMRefThing *) b2;
+		num2 = mrtp->mb->refn;
+	    }
+	    return erts_internal_ref_number_cmp(num1, num2);
 	}
 	return -1;
     }
@@ -97,7 +115,8 @@ do {								\
 	Uint i__;						\
 	Uint len__;						\
 	ASSERT((Hp));						\
-	ASSERT(is_internal_ref((From)) || is_external((From)));	\
+	ASSERT(is_internal_ordinary_ref((From))                 \
+               || is_external((From)));                         \
 	(To) = make_boxed((Hp));				\
 	len__ = thing_arityval(*boxed_val((From))) + 1;		\
 	for(i__ = 0; i__ < len__; i__++)			\
@@ -339,6 +358,8 @@ void erts_add_monitor(ErtsMonitor **root, Uint type, Eterm ref, Eterm pid,
     int state = 0;
     ErtsMonitor **this = root;
     Sint c;
+
+    ASSERT(is_internal_ordinary_ref(ref) || is_external_ref(ref));
   
     dstack[0] = DIR_END;
     for (;;) {
diff --git a/erts/emulator/beam/erl_monitors.h b/erts/emulator/beam/erl_monitors.h
index 9e2beedea3..bb493bf336 100644
--- a/erts/emulator/beam/erl_monitors.h
+++ b/erts/emulator/beam/erl_monitors.h
@@ -91,7 +91,7 @@
 
 /* Size of a monitor without heap, in words (fixalloc) */
 #define ERTS_MONITOR_SIZE ((sizeof(ErtsMonitor) - sizeof(Uint))/sizeof(Uint))
-#define ERTS_MONITOR_SH_SIZE (ERTS_MONITOR_SIZE + REF_THING_SIZE)
+#define ERTS_MONITOR_SH_SIZE (ERTS_MONITOR_SIZE + ERTS_REF_THING_SIZE)
 #define ERTS_LINK_SIZE ((sizeof(ErtsLink) - sizeof(Uint))/sizeof(Uint))
 #define ERTS_LINK_SH_SIZE ERTS_LINK_SIZE /* Size of fix-alloced links */
 
diff --git a/erts/emulator/beam/erl_msacc.c b/erts/emulator/beam/erl_msacc.c
index 66bb55e6c8..1c3160efaf 100644
--- a/erts/emulator/beam/erl_msacc.c
+++ b/erts/emulator/beam/erl_msacc.c
@@ -66,9 +66,9 @@ static Uint msacc_unmanaged_count = 0;
 #endif
 
 #if ERTS_MSACC_STATE_COUNT < MAP_SMALL_MAP_LIMIT
-#define DEFAULT_MSACC_MSG_SIZE (3 + 1 + ERTS_MSACC_STATE_COUNT * 2 + 3 + REF_THING_SIZE)
+#define DEFAULT_MSACC_MSG_SIZE (3 + 1 + ERTS_MSACC_STATE_COUNT * 2 + 3 + ERTS_REF_THING_SIZE)
 #else
-#define DEFAULT_MSACC_MSG_SIZE (3 + ERTS_MSACC_STATE_COUNT * 3 + 3 + REF_THING_SIZE)
+#define DEFAULT_MSACC_MSG_SIZE (3 + ERTS_MSACC_STATE_COUNT * 3 + 3 + ERTS_REF_THING_SIZE)
 #endif
 
 /* we have to split initiation as atoms are not inited in early init */
@@ -212,7 +212,7 @@ typedef struct {
     int action;
     Process *proc;
     Eterm ref;
-    Eterm ref_heap[REF_THING_SIZE];
+    Eterm ref_heap[ERTS_REF_THING_SIZE];
     Uint req_sched;
     erts_smp_atomic32_t refc;
 } ErtsMSAccReq;
@@ -245,7 +245,7 @@ static void send_reply(ErtsMsAcc *msacc, ErtsMSAccReq *msaccrp) {
 
         if (msacc->unmanaged) erts_mtx_lock(&msacc->mtx);
 
-        hp = erts_produce_heap(&factory, REF_THING_SIZE + 3 /* tuple */, 0);
+        hp = erts_produce_heap(&factory, ERTS_REF_THING_SIZE + 3 /* tuple */, 0);
         ref_copy = STORE_NC(&hp, &msgp->hfrag.off_heap, msaccrp->ref);
         msg = erts_msacc_gather_stats(msacc, &factory);
         msg = TUPLE2(hp, ref_copy, msg);
@@ -255,7 +255,7 @@ static void send_reply(ErtsMsAcc *msacc, ErtsMSAccReq *msaccrp) {
         erts_factory_close(&factory);
     } else {
         ErlOffHeap *ohp = NULL;
-        msgp = erts_alloc_message_heap(rp, &rp_locks, REF_THING_SIZE, &hp, &ohp);
+        msgp = erts_alloc_message_heap(rp, &rp_locks, ERTS_REF_THING_SIZE, &hp, &ohp);
         msg = STORE_NC(&hp, &msgp->hfrag.off_heap, msaccrp->ref);
     }
 
diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c
index 7c7a32f234..20b00bd4ba 100644
--- a/erts/emulator/beam/erl_nif.c
+++ b/erts/emulator/beam/erl_nif.c
@@ -1721,7 +1721,7 @@ ERL_NIF_TERM enif_make_string_len(ErlNifEnv* env, const char* string,
 
 ERL_NIF_TERM enif_make_ref(ErlNifEnv* env)
 {
-    Eterm* hp = alloc_heap(env, REF_THING_SIZE);
+    Eterm* hp = alloc_heap(env, ERTS_REF_THING_SIZE);
     return erts_make_ref_in_buffer(hp);
 }
 
@@ -1967,7 +1967,6 @@ typedef struct enif_resource_t
     byte align__[4];
 # endif
 #endif
-
     char data[1];
 }ErlNifResource;
 
@@ -2166,6 +2165,7 @@ void* enif_alloc_resource(ErlNifResourceType* type, size_t size)
 {
     Binary* bin = erts_create_magic_binary_x(SIZEOF_ErlNifResource(size),
                                              &nif_resource_dtor,
+                                             ERTS_ALC_T_BINARY,
                                              1); /* unaligned */
     ErlNifResource* resource = ERTS_MAGIC_BIN_UNALIGNED_DATA(bin);
 
diff --git a/erts/emulator/beam/erl_node_container_utils.h b/erts/emulator/beam/erl_node_container_utils.h
index 0c76c6fe7d..e27f1d121a 100644
--- a/erts/emulator/beam/erl_node_container_utils.h
+++ b/erts/emulator/beam/erl_node_container_utils.h
@@ -255,19 +255,16 @@ extern ErtsPTab erts_port;
  * Refs                                                                    *
 \*                                                                         */
 
+#define internal_ref_no_numbers(x)	ERTS_REF_NUMBERS
+#define internal_ref_numbers(x)		(is_internal_ordinary_ref((x)) \
+					 ? internal_ordinary_ref_numbers((x)) \
+					 : (ASSERT(is_internal_magic_ref((x))), \
+					    internal_magic_ref_numbers((x))))
 #if defined(ARCH_64)
 
-#define internal_ref_no_of_numbers(x)					\
-  (internal_ref_data((x))[0])
-#define internal_thing_ref_no_of_numbers(thing)			        \
-  (internal_thing_ref_data(thing)[0])
-#define internal_ref_numbers(x)						\
-  (&internal_ref_data((x))[1])
-#define internal_thing_ref_numbers(thing)				\
-  (&internal_thing_ref_data(thing)[1])
-#define external_ref_no_of_numbers(x)					\
+#define external_ref_no_numbers(x)					\
   (external_ref_data((x))[0])
-#define external_thing_ref_no_of_numbers(thing)			        \
+#define external_thing_ref_no_numbers(thing)			        \
   (external_thing_ref_data(thing)[0])
 #define external_ref_numbers(x)						\
   (&external_ref_data((x))[1])
@@ -277,12 +274,8 @@ extern ErtsPTab erts_port;
 
 #else
 
-#define internal_ref_no_of_numbers(x)	(internal_ref_data_words((x)))
-#define internal_thing_ref_no_of_numbers(t) (internal_thing_ref_data_words(t))
-#define internal_ref_numbers(x)		(internal_ref_data((x)))
-#define internal_thing_ref_numbers(t)   (internal_thing_ref_data(t))
-#define external_ref_no_of_numbers(x)	(external_ref_data_words((x)))
-#define external_thing_ref_no_of_numbers(t) (external_thing_ref_data_words((t)))
+#define external_ref_no_numbers(x)	(external_ref_data_words((x)))
+#define external_thing_ref_no_numbers(t) (external_thing_ref_data_words((t)))
 #define external_ref_numbers(x)		(external_ref_data((x)))
 #define external_thing_ref_numbers(t)   (external_thing_ref_data((t)))
 
@@ -299,15 +292,9 @@ extern ErtsPTab erts_port;
 #define internal_ref_channel_no(x)	(internal_channel_no((x)))
 #define external_ref_channel_no(x)	(external_channel_no((x)))
 
-#define ref_data_words(x)		(is_internal_ref((x))		\
-					 ? internal_ref_data_words((x))	\
-					 : external_ref_data_words((x)))
-#define ref_data(x)			(is_internal_ref((x))		\
-					 ? internal_ref_data((x))	\
-					 : external_ref_data((x)))
-#define ref_no_of_numbers(x)		(is_internal_ref((x))		\
-					 ? internal_ref_no_of_numbers((x))\
-					 : external_ref_no_of_numbers((x)))
+#define ref_no_numbers(x)		(is_internal_ref((x))		\
+					 ? internal_ref_no_numbers((x))\
+					 : external_ref_no_numbers((x)))
 #define ref_numbers(x)			(is_internal_ref((x))		\
 					 ? internal_ref_numbers((x))	\
 					 : external_ref_numbers((x)))
diff --git a/erts/emulator/beam/erl_node_tables.c b/erts/emulator/beam/erl_node_tables.c
index 467a05f950..060fbd5883 100644
--- a/erts/emulator/beam/erl_node_tables.c
+++ b/erts/emulator/beam/erl_node_tables.c
@@ -1126,8 +1126,8 @@ insert_offheap(ErlOffHeap *oh, int type, Eterm id)
 
     for (u.hdr = oh->first; u.hdr; u.hdr = u.hdr->next) {
 	switch (thing_subtag(u.hdr->thing_word)) {
-	case REFC_BINARY_SUBTAG:
-	    if(IsMatchProgBinary(u.pb->val)) {
+	case REF_SUBTAG:
+	    if(IsMatchProgBinary(u.mref->mb)) {
 		InsertedBin *ib;
 		int insert_bin = 1;
 		for (ib = inserted_bins; ib; ib = ib->next)
@@ -1152,6 +1152,7 @@ insert_offheap(ErlOffHeap *oh, int type, Eterm id)
 		}
 	    }		
 	    break;
+	case REFC_BINARY_SUBTAG:
 	case FUN_SUBTAG:
 	    break; /* No need to */
 	default:
diff --git a/erts/emulator/beam/erl_printf_term.c b/erts/emulator/beam/erl_printf_term.c
index b43a1b0190..6b64bbd2f1 100644
--- a/erts/emulator/beam/erl_printf_term.c
+++ b/erts/emulator/beam/erl_printf_term.c
@@ -382,12 +382,15 @@ print_term(fmtfn_t fn, void* arg, Eterm obj, long *dcount) {
 	    break;
 	}
 	case REF_DEF:
+            if (!ERTS_IS_CRASH_DUMPING)
+                erts_magic_ref_save_bin(obj);
+            /* fall through... */
 	case EXTERNAL_REF_DEF:
 	    PRINT_STRING(res, fn, arg, "#Ref<");
 	    PRINT_UWORD(res, fn, arg, 'u', 0, 1,
 			(ErlPfUWord) ref_channel_no(wobj));
 	    ref_num = ref_numbers(wobj);
-	    for (i = ref_no_of_numbers(wobj)-1; i >= 0; i--) {
+	    for (i = ref_no_numbers(wobj)-1; i >= 0; i--) {
 		PRINT_CHAR(res, fn, arg, '.');
 		PRINT_UWORD(res, fn, arg, 'u', 0, 1, (ErlPfUWord) ref_num[i]);
 	    }
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 3691e31922..87d71fdd22 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -1173,7 +1173,7 @@ typedef struct {
     int enable;
     Process *proc;
     Eterm ref;
-    Eterm ref_heap[REF_THING_SIZE];
+    Eterm ref_heap[ERTS_REF_THING_SIZE];
     Uint req_sched;
     erts_smp_atomic32_t refc;
 #ifdef ERTS_DIRTY_SCHEDULERS
@@ -1185,7 +1185,7 @@ typedef struct {
 typedef struct {
     Process *proc;
     Eterm ref;
-    Eterm ref_heap[REF_THING_SIZE];
+    Eterm ref_heap[ERTS_REF_THING_SIZE];
     Uint req_sched;
     erts_smp_atomic32_t refc;
 } ErtsSystemCheckReq;
@@ -1297,7 +1297,7 @@ reply_sched_wall_time(void *vswtrp)
             if (hpp)
                 ref_copy = STORE_NC(hpp, ohp, swtrp->ref);
             else
-                *szp += REF_THING_SIZE;
+                *szp += ERTS_REF_THING_SIZE;
 
             ASSERT(!swtrp->set);
 
@@ -1342,7 +1342,7 @@ reply_sched_wall_time(void *vswtrp)
             if (hpp)
                 ref_copy = STORE_NC(hpp, ohp, swtrp->ref);
             else
-                *szp += REF_THING_SIZE;
+                *szp += ERTS_REF_THING_SIZE;
 
             if (swtrp->set)
                 msg = ref_copy;
@@ -1444,7 +1444,7 @@ reply_system_check(void *vscrp)
     ASSERT(!ERTS_SCHEDULER_IS_DIRTY(esdp));
 #endif
 
-    sz = REF_THING_SIZE;
+    sz = ERTS_REF_THING_SIZE;
     mp = erts_alloc_message_heap(rp, &rp_locks, sz, &hp, &ohp);
     hpp = &hp;
     msg = STORE_NC(hpp, ohp, scrp->ref);
@@ -6381,7 +6381,6 @@ erts_init_scheduling(int no_schedulers, int no_schedulers_online
 	erts_tsd_set(sched_data_key, (void *) esdp);
 #endif
     }
-    erts_no_schedulers = 1;
     erts_no_dirty_cpu_schedulers = 0;
     erts_no_dirty_io_schedulers = 0;
 #endif
diff --git a/erts/emulator/beam/erl_term.c b/erts/emulator/beam/erl_term.c
index 7d857ad326..6b25728af7 100644
--- a/erts/emulator/beam/erl_term.c
+++ b/erts/emulator/beam/erl_term.c
@@ -59,6 +59,42 @@ erts_set_literal_tag(Eterm *term, Eterm *hp_start, Eterm hsz)
 #endif
 }
 
+void
+erts_term_init(void)
+{
+#ifdef ERTS_ORDINARY_REF_MARKER
+    /* Ordinary and magic references of same size... */
+
+    ErtsRefThing ref_thing;
+
+    ERTS_CT_ASSERT(ERTS_ORDINARY_REF_MARKER == ~((Uint32)0));
+    ref_thing.m.header = ERTS_REF_THING_HEADER;
+    ref_thing.m.mb = (ErtsMagicBinary *) ~((UWord) 3);
+    ref_thing.m.next = (struct erl_off_heap_header *) ~((UWord) 3);
+    if (ref_thing.o.marker == ERTS_ORDINARY_REF_MARKER)
+        ERTS_INTERNAL_ERROR("Cannot differentiate between magic and ordinary references");
+
+    ERTS_CT_ASSERT(offsetof(ErtsORefThing,marker) != 0);
+    ERTS_CT_ASSERT(sizeof(ErtsORefThing) == sizeof(ErtsMRefThing));
+#  ifdef ERTS_MAGIC_REF_THING_HEADER
+#    error Magic ref thing header should not have been defined...
+#  endif
+
+#else
+    /* Ordinary and magic references of different sizes... */
+
+#  ifndef ERTS_MAGIC_REF_THING_HEADER
+#    error Magic ref thing header should have been defined...
+#  endif
+    ERTS_CT_ASSERT(sizeof(ErtsORefThing) != sizeof(ErtsMRefThing));
+
+#endif
+
+    ERTS_CT_ASSERT(ERTS_REF_THING_SIZE*sizeof(Eterm) == sizeof(ErtsORefThing));
+    ERTS_CT_ASSERT(ERTS_MAGIC_REF_THING_SIZE*sizeof(Eterm) == sizeof(ErtsMRefThing));
+
+}
+
 /*
  * XXX: define NUMBER_CODE() here when new representation is used
  */
@@ -95,8 +131,8 @@ ET_DEFINE_CHECKED(Eterm*,tuple_val,Wterm,is_tuple);
 ET_DEFINE_CHECKED(struct erl_node_*,internal_pid_node,Eterm,is_internal_pid);
 ET_DEFINE_CHECKED(struct erl_node_*,internal_port_node,Eterm,is_internal_port);
 ET_DEFINE_CHECKED(Eterm*,internal_ref_val,Wterm,is_internal_ref);
-ET_DEFINE_CHECKED(Uint,internal_ref_data_words,Wterm,is_internal_ref);
-ET_DEFINE_CHECKED(Uint32*,internal_ref_data,Wterm,is_internal_ref);
+ET_DEFINE_CHECKED(Uint32*,internal_magic_ref_numbers,Wterm,is_internal_magic_ref);
+ET_DEFINE_CHECKED(Uint32*,internal_ordinary_ref_numbers,Wterm,is_internal_ordinary_ref);
 ET_DEFINE_CHECKED(struct erl_node_*,internal_ref_node,Eterm,is_internal_ref);
 ET_DEFINE_CHECKED(Eterm*,external_val,Wterm,is_external);
 ET_DEFINE_CHECKED(Uint,external_data_words,Wterm,is_external);
diff --git a/erts/emulator/beam/erl_term.h b/erts/emulator/beam/erl_term.h
index c3234ee349..a602a8f7c6 100644
--- a/erts/emulator/beam/erl_term.h
+++ b/erts/emulator/beam/erl_term.h
@@ -23,6 +23,8 @@
 
 #include "erl_mmap.h"
 
+void erts_term_init(void);
+
 typedef UWord Wterm;  /* Full word terms */
 
 struct erl_node_; /* Declared in erl_node_tables.h */
@@ -718,73 +720,224 @@ _ET_DECLARE_CHECKED(struct erl_node_*,internal_port_node,Eterm)
 #define ERTS_MAX_REF_NUMBERS	3
 #define ERTS_REF_NUMBERS	ERTS_MAX_REF_NUMBERS
 
-#if defined(ARCH_64)
-#  define ERTS_REF_WORDS	(ERTS_REF_NUMBERS/2 + 1)
-#  define ERTS_REF_32BIT_WORDS  (ERTS_REF_NUMBERS+1)
-#else
-#  define ERTS_REF_WORDS	ERTS_REF_NUMBERS
-#  define ERTS_REF_32BIT_WORDS  ERTS_REF_NUMBERS
+#ifndef ERTS_ENDIANNESS
+# error ERTS_ENDIANNESS not defined...
 #endif
 
-typedef struct {
-    Eterm      header;
-    union {
-	Uint32 ui32[ERTS_REF_32BIT_WORDS];
-	Uint   ui[ERTS_REF_WORDS];
-    } data;
-} RefThing;
-
-#define REF_THING_SIZE (sizeof(RefThing)/sizeof(Uint))
-#define REF_THING_HEAD_SIZE (sizeof(Eterm)/sizeof(Uint))
+#if ERTS_REF_NUMBERS != 3
+#  error "A new reference layout for 64-bit needs to be implemented..."
+#endif
 
-#define make_ref_thing_header(DW) \
-  _make_header((DW)+REF_THING_HEAD_SIZE-1,_TAG_HEADER_REF)
+struct magic_binary;
 
 #if defined(ARCH_64)
 
+# define ERTS_ORDINARY_REF_MARKER (~((Uint32) 0))
+
+typedef struct {
+    Eterm header;
+#if ERTS_ENDIANNESS <= 0
+    Uint32 marker;
+#endif
+    Uint32 num[ERTS_REF_NUMBERS];
+#if ERTS_ENDIANNESS > 0
+    Uint32 marker;
+#endif
+} ErtsORefThing;
+
+typedef struct {
+    Eterm header;
+    struct magic_binary *mb;
+    struct erl_off_heap_header* next;
+#if !ERTS_ENDIANNESS
+    Uint32 num[ERTS_REF_NUMBERS];
+    Uint32 marker;
+#endif
+} ErtsMRefThing;
+
 /*
- * Ref layout on a 64-bit little endian machine:
+ * Ordinary ref layout on a 64-bit little endian machine:
  *
  * 63             31             0
  * +--------------+--------------+
  * |         Thing word          |
  * +--------------+--------------+
- * |  Data 0      | 32-bit arity |
+ * |  Data 0      | 0xffffffff   |
  * +--------------+--------------+
  * |  Data 2      | Data 1       |
  * +--------------+--------------+
  *
- * Data is stored as an Uint32 array with 32-bit arity as first number.
+ * Ordinary ref layout on a 64-bit big endian machine:
+ *
+ * 63             31             0
+ * +--------------+--------------+
+ * |         Thing word          |
+ * +--------------+--------------+
+ * |  Data 0      | Data 1       |
+ * +--------------+--------------+
+ * |  Data 2      | 0xffffffff   |
+ * +--------------+--------------+
+ *
+ * Magic Ref layout on a 64-bit machine:
+ *
+ * 63             31             0
+ * +--------------+--------------+
+ * |         Thing word          |
+ * +--------------+--------------+
+ * |  Magic Binary Pointer       |
+ * +--------------+--------------+
+ * |  Next Off Heap Pointer      |
+ * +--------------+--------------+
+ *
+ * Both pointers in the magic ref are 64-bit aligned. That is,
+ * least significant bits are zero. The marker 32-bit word is
+ * placed over the least significant bits of one of the pointers.
+ * That is, we can distinguish between magic and ordinary ref
+ * by looking at the marker field.
+ * 
  */
 
 #define write_ref_thing(Hp, R0, R1, R2)					\
 do {									\
-  ((RefThing *) (Hp))->header  = make_ref_thing_header(ERTS_REF_WORDS);	\
-  ((RefThing *) (Hp))->data.ui32[0] = ERTS_REF_NUMBERS;			\
-  ((RefThing *) (Hp))->data.ui32[1] = (R0);				\
-  ((RefThing *) (Hp))->data.ui32[2] = (R1);				\
-  ((RefThing *) (Hp))->data.ui32[3] = (R2);				\
+  ((ErtsORefThing *) (Hp))->header = ERTS_REF_THING_HEADER;		\
+  ((ErtsORefThing *) (Hp))->marker = ERTS_ORDINARY_REF_MARKER;		\
+  ((ErtsORefThing *) (Hp))->num[0] = (R0);				\
+  ((ErtsORefThing *) (Hp))->num[1] = (R1);				\
+  ((ErtsORefThing *) (Hp))->num[2] = (R2);				\
 } while (0)
 
-#else
+#if ERTS_ENDIANNESS
+/* Known big or little endian */
+
+#define write_magic_ref_thing(Hp, Ohp, Binp)				\
+do {									\
+  ((ErtsMRefThing *) (Hp))->header = ERTS_REF_THING_HEADER;		\
+  ((ErtsMRefThing *) (Hp))->mb = (Binp);				\
+  ((ErtsMRefThing *) (Hp))->next = (Ohp)->first;			\
+  (Ohp)->first = (struct erl_off_heap_header*) (Hp);			\
+  ASSERT(erts_is_ref_numbers_magic((Binp)->refn));			\
+} while (0)
+
+#else /* !ERTS_ENDIANNESS */
+
+#define write_magic_ref_thing(Hp, Ohp, Binp)				\
+do {									\
+  ((ErtsMRefThing *) (Hp))->header = ERTS_MAGIC_REF_THING_HEADER;	\
+  ((ErtsMRefThing *) (Hp))->mb = (Binp);				\
+  ((ErtsMRefThing *) (Hp))->next = (Ohp)->first;			\
+  (Ohp)->first = (struct erl_off_heap_header*) (Hp);			\
+  ((ErtsMRefThing *) (Hp))->marker = 0;					\
+  ((ErtsMRefThing *) (Hp))->num[0] = (Binp)->refn[0];			\
+  ((ErtsMRefThing *) (Hp))->num[1] = (Binp)->refn[1];			\
+  ((ErtsMRefThing *) (Hp))->num[2] = (Binp)->refn[2];			\
+  ASSERT(erts_is_ref_numbers_magic((Binp)->refn));			\
+} while (0)
+
+#endif /* !ERTS_ENDIANNESS */
+
+#else /* ARCH_32 */
+
+typedef struct {
+    Eterm header;
+    Uint32 num[ERTS_REF_NUMBERS];
+} ErtsORefThing;
+
+typedef struct {
+    Eterm header;
+    struct magic_binary *mb;
+    struct erl_off_heap_header* next;
+} ErtsMRefThing;
+
 
 #define write_ref_thing(Hp, R0, R1, R2)					\
 do {									\
-  ((RefThing *) (Hp))->header  = make_ref_thing_header(ERTS_REF_WORDS);	\
-  ((RefThing *) (Hp))->data.ui32[0] = (R0);				\
-  ((RefThing *) (Hp))->data.ui32[1] = (R1);				\
-  ((RefThing *) (Hp))->data.ui32[2] = (R2);				\
+  ((ErtsORefThing *) (Hp))->header = ERTS_REF_THING_HEADER;		\
+  ((ErtsORefThing *) (Hp))->num[0] = (R0);				\
+  ((ErtsORefThing *) (Hp))->num[1] = (R1);				\
+  ((ErtsORefThing *) (Hp))->num[2] = (R2);				\
 } while (0)
 
+#define write_magic_ref_thing(Hp, Ohp, Binp)				\
+do {									\
+  ((ErtsMRefThing *) (Hp))->header = ERTS_MAGIC_REF_THING_HEADER;	\
+  ((ErtsMRefThing *) (Hp))->mb = (Binp);				\
+  ((ErtsMRefThing *) (Hp))->next = (Ohp)->first;			\
+  (Ohp)->first = (struct erl_off_heap_header*) (Hp);			\
+  ASSERT(erts_is_ref_numbers_magic(&(Binp)->refn));			\
+} while (0)
+
+#endif /* ARCH_32 */
+
+typedef union {
+    ErtsMRefThing m;
+    ErtsORefThing o;
+} ErtsRefThing;
+
+#define ERTS_REF_THING_SIZE (sizeof(ErtsORefThing)/sizeof(Uint))
+#define ERTS_MAGIC_REF_THING_SIZE (sizeof(ErtsMRefThing)/sizeof(Uint))
+#define ERTS_MAX_INTERNAL_REF_SIZE (sizeof(ErtsRefThing)/sizeof(Uint))
+
+#define make_ref_thing_header(Words) \
+    _make_header((Words)-1,_TAG_HEADER_REF)
+
+#define ERTS_REF_THING_HEADER _make_header(ERTS_REF_THING_SIZE-1,_TAG_HEADER_REF)
+
+#if defined(ARCH_64) && ERTS_ENDIANNESS /* All internal refs of same size... */
+
+#  undef ERTS_MAGIC_REF_THING_HEADER
+
+#  define is_ref_thing_header(x) ((x) == ERTS_REF_THING_HEADER)
+
+#define is_ordinary_ref_thing(x)					\
+    (ASSERT(is_ref_thing_header(*((Eterm *)(x)))),			\
+     ((ErtsRefThing *) (x))->o.marker == ERTS_ORDINARY_REF_MARKER)
+
+#define is_magic_ref_thing(x)						\
+    (!is_ordinary_ref_thing((x)))
+
+#define is_internal_magic_ref(x)					\
+    ((_unchecked_is_boxed((x)) && *boxed_val((x)) == ERTS_REF_THING_HEADER) \
+     && is_magic_ref_thing(boxed_val((x))))
+
+#define is_internal_ordinary_ref(x)					\
+    ((_unchecked_is_boxed((x)) && *boxed_val((x)) == ERTS_REF_THING_HEADER) \
+     && is_ordinary_ref_thing(boxed_val((x))))
+
+#else /* Ordinary and magic references of different sizes... */
+
+#  define ERTS_MAGIC_REF_THING_HEADER					\
+    _make_header(ERTS_MAGIC_REF_THING_SIZE-1,_TAG_HEADER_REF)
+
+#  define is_ref_thing_header(x)					\
+    (((x) & _TAG_HEADER_MASK) == _TAG_HEADER_REF)
+
+#define is_ordinary_ref_thing(x)					\
+    (ASSERT(is_ref_thing_header(*((Eterm *)(x)))),			\
+     *((Eterm *)(x)) == ERTS_REF_THING_HEADER)
+
+#define is_magic_ref_thing(x)						\
+    (ASSERT(is_ref_thing_header(*((Eterm *)(x)))),			\
+     *((Eterm *)(x)) == ERTS_MAGIC_REF_THING_HEADER)
+
+#define is_internal_magic_ref(x)					\
+    (_unchecked_is_boxed((x)) && *boxed_val((x)) == ERTS_MAGIC_REF_THING_HEADER)
+
+#define is_internal_ordinary_ref(x)					\
+    (_unchecked_is_boxed((x)) && *boxed_val((x)) == ERTS_REF_THING_HEADER)
+
 #endif
 
-#define is_ref_thing_header(x)	(((x) & _TAG_HEADER_MASK) == _TAG_HEADER_REF)
 #define make_internal_ref(x)	make_boxed((Eterm*)(x))
 
-#define _unchecked_ref_thing_ptr(x) \
-  ((RefThing*) _unchecked_internal_ref_val(x))
-#define ref_thing_ptr(x) \
-  ((RefThing*) internal_ref_val(x))
+#define _unchecked_ordinary_ref_thing_ptr(x) \
+  ((ErtsORefThing*) _unchecked_internal_ref_val(x))
+#define ordinary_ref_thing_ptr(x) \
+  ((ErtsORefThing*) internal_ref_val(x))
+
+#define _unchecked_magic_ref_thing_ptr(x) \
+  ((ErtsMRefThing*) _unchecked_internal_ref_val(x))
+#define magic_ref_thing_ptr(x) \
+  ((ErtsMRefThing*) internal_ref_val(x))
 
 #define is_internal_ref(x) \
     (_unchecked_is_boxed((x)) && is_ref_thing_header(*boxed_val((x))))
@@ -796,16 +949,21 @@ do {									\
 _ET_DECLARE_CHECKED(Eterm*,internal_ref_val,Wterm)
 #define internal_ref_val(x) _ET_APPLY(internal_ref_val,(x))
 
-#define internal_thing_ref_data_words(t) (thing_arityval(*(Eterm*)(t)))
-#define _unchecked_internal_ref_data_words(x) \
- (_unchecked_thing_arityval(*_unchecked_internal_ref_val(x)))
-_ET_DECLARE_CHECKED(Uint,internal_ref_data_words,Wterm)
-#define internal_ref_data_words(x) _ET_APPLY(internal_ref_data_words,(x))
+#define internal_ordinary_thing_ref_numbers(ort) (((ErtsORefThing *)(ort))->num)
+#define _unchecked_internal_ordinary_ref_numbers(x) (internal_ordinary_thing_ref_numbers(_unchecked_ordinary_ref_thing_ptr(x)))
+_ET_DECLARE_CHECKED(Uint32*,internal_ordinary_ref_numbers,Wterm)
+#define internal_ordinary_ref_numbers(x) _ET_APPLY(internal_ordinary_ref_numbers,(x))
+
+#if defined(ARCH_64) && !ERTS_ENDIANNESS
+#define internal_magic_thing_ref_numbers(mrt) (((ErtsMRefThing *)(mrt))->num)
+#else
+#define internal_magic_thing_ref_numbers(mrt) (((ErtsMRefThing *)(mrt))->mb->refn)
+#endif
+
+#define _unchecked_internal_magic_ref_numbers(x) (internal_magic_thing_ref_numbers(_unchecked_magic_ref_thing_ptr(x)))
+_ET_DECLARE_CHECKED(Uint32*,internal_magic_ref_numbers,Wterm)
+#define internal_magic_ref_numbers(x) _ET_APPLY(internal_magic_ref_numbers,(x))
 
-#define internal_thing_ref_data(thing) ((thing)->data.ui32)
-#define _unchecked_internal_ref_data(x) (internal_thing_ref_data(_unchecked_ref_thing_ptr(x)))
-_ET_DECLARE_CHECKED(Uint32*,internal_ref_data,Wterm)
-#define internal_ref_data(x) _ET_APPLY(internal_ref_data,(x))
 
 #define _unchecked_internal_ref_node(x) erts_this_node
 _ET_DECLARE_CHECKED(struct erl_node_*,internal_ref_node,Eterm)
diff --git a/erts/emulator/beam/erl_time_sup.c b/erts/emulator/beam/erl_time_sup.c
index 6aa2a7500f..1d747e5fab 100644
--- a/erts/emulator/beam/erl_time_sup.c
+++ b/erts/emulator/beam/erl_time_sup.c
@@ -1831,7 +1831,10 @@ erts_demonitor_time_offset(Eterm ref)
     ErtsMonitor *mon;
     ASSERT(is_internal_ref(ref));
     erts_smp_mtx_lock(&erts_get_time_mtx);
-    mon = erts_remove_monitor(&time_offset_monitors, ref);
+    if (is_internal_ordinary_ref(ref))
+        mon = erts_remove_monitor(&time_offset_monitors, ref);
+    else
+        mon = NULL;
     if (!mon)
 	res = 0;
     else {
@@ -1848,7 +1851,7 @@ erts_demonitor_time_offset(Eterm ref)
 typedef struct {
     Eterm pid;
     Eterm ref;
-    Eterm heap[REF_THING_SIZE];
+    Eterm heap[ERTS_REF_THING_SIZE];
 } ErtsTimeOffsetMonitorInfo;
 
 typedef struct {
@@ -1869,11 +1872,11 @@ save_time_offset_monitor(ErtsMonitor *mon, void *vcntxt)
     cntxt->to_mon_info[mix].pid = mon->pid;
     to_hp = &cntxt->to_mon_info[mix].heap[0];
 
-    ASSERT(is_internal_ref(mon->ref));
+    ASSERT(is_internal_ordinary_ref(mon->ref));
     from_hp = internal_ref_val(mon->ref);
-    ASSERT(thing_arityval(*from_hp) + 1 == REF_THING_SIZE);
+    ASSERT(thing_arityval(*from_hp) + 1 == ERTS_REF_THING_SIZE);
 
-    for (hix = 0; hix < REF_THING_SIZE; hix++)
+    for (hix = 0; hix < ERTS_REF_THING_SIZE; hix++)
 	to_hp[hix] = from_hp[hix];
 
     cntxt->to_mon_info[mix].ref
@@ -1933,7 +1936,7 @@ send_time_offset_changed_notifications(void *new_offsetp)
 	hp = (Eterm *) (tmp + no_monitors*sizeof(ErtsTimeOffsetMonitorInfo));
 
 	hsz = 6; /* 5-tuple */
-	hsz += REF_THING_SIZE;
+	hsz += ERTS_REF_THING_SIZE;
 	hsz += ERTS_SINT64_HEAP_SIZE(new_offset);
 
 	if (IS_SSMALL(new_offset))
diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c
index 6bcddb9e69..64b17a3f57 100644
--- a/erts/emulator/beam/external.c
+++ b/erts/emulator/beam/external.c
@@ -2579,7 +2579,9 @@ enc_term_int(TTBEncodeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj, byte* ep,
 
 	    ASSERT(dflags & DFLAG_EXTENDED_REFERENCES);
 
-	    i = ref_no_of_numbers(obj);
+	    erts_magic_ref_save_bin(obj);
+
+	    i = ref_no_numbers(obj);
 	    put_int16(i, ep);
 	    ep += 2;
 	    ep = enc_atom(acmp, sysname, ep, dflags);
@@ -3435,26 +3437,35 @@ dec_term_atom_common:
 		r0 = get_int32(ep); /* allow full word */
 		ep += 4;
 
-	    ref_ext_common:
+	    ref_ext_common: {
+		ErtsORefThing *rtp;
+
 		if (ref_words > ERTS_MAX_REF_NUMBERS)
 		    goto error;
 
 		node = dec_get_node(sysname, cre);
 		if(node == erts_this_node) {
-		    RefThing *rtp = (RefThing *) hp;
-		    ref_num = (Uint32 *) (hp + REF_THING_HEAD_SIZE);
+	
+		    rtp = (ErtsORefThing *) hp;
+		    ref_num = &rtp->num[0];
+		    if (ref_words != ERTS_REF_NUMBERS) {
+                        int i;
+                        if (ref_words > ERTS_REF_NUMBERS)
+                            goto error; /* Not a ref that we created... */
+                        for (i = ref_words; i < ERTS_REF_NUMBERS; i++)
+                            ref_num[i] = 0;
+		    }
 
-#if defined(ARCH_64)
-		    hp += REF_THING_HEAD_SIZE + ref_words/2 + 1;
-		    rtp->header = make_ref_thing_header(ref_words/2 + 1);
-#else
-		    hp += REF_THING_HEAD_SIZE + ref_words;
-		    rtp->header = make_ref_thing_header(ref_words);
+#ifdef ERTS_ORDINARY_REF_MARKER
+		    rtp->marker = ERTS_ORDINARY_REF_MARKER;
 #endif
+		    hp += ERTS_REF_THING_SIZE;
+		    rtp->header = ERTS_REF_THING_HEADER;
 		    *objp = make_internal_ref(rtp);
 		}
 		else {
 		    ExternalThing *etp = (ExternalThing *) hp;
+		    rtp = NULL;
 #if defined(ARCH_64)
 		    hp += EXTERNAL_THING_HEAD_SIZE + ref_words/2 + 1;
 #else
@@ -3472,12 +3483,13 @@ dec_term_atom_common:
 		    factory->off_heap->first = (struct erl_off_heap_header*)etp;
 		    *objp = make_external_ref(etp);
 		    ref_num = &(etp->data.ui32[0]);
-		}
-
 #if defined(ARCH_64)
-		*(ref_num++) = ref_words /* 32-bit arity */;
+		    *(ref_num++) = ref_words /* 32-bit arity */;
 #endif
+		}
+
 		ref_num[0] = r0;
+
 		for(i = 1; i < ref_words; i++) {
 		    ref_num[i] = get_int32(ep);
 		    ep += 4;
@@ -3486,8 +3498,24 @@ dec_term_atom_common:
 		if ((1 + ref_words) % 2)
 		    ref_num[ref_words] = 0;
 #endif
+		if (node == erts_this_node) {
+		    /* Check if it was a magic reference... */
+		    ErtsMagicBinary *mb = erts_magic_ref_lookup_bin(ref_num);
+		    if (mb) {
+			/*
+			 * Was a magic ref; adjust it...
+			 *
+			 * Refc on binary was increased by lookup above...
+			 */
+			ASSERT(rtp);
+			hp = (Eterm *) rtp;
+			write_magic_ref_thing(hp, factory->off_heap, mb);
+			hp += ERTS_MAGIC_REF_THING_SIZE;
+		    }
+		}
 		break;
 	    }
+	    }
 	case BINARY_EXT:
 	    {
 		n = get_int32(ep);
@@ -4109,7 +4137,7 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj,
             /*fall through*/
 	case REF_DEF:
 	    ASSERT(dflags & DFLAG_EXTENDED_REFERENCES);
-	    i = ref_no_of_numbers(obj);
+	    i = ref_no_numbers(obj);
 	    result += (1 + 2 + encode_size_struct2(acmp, ref_node_name(obj), dflags) +
 		       1 + 4*i);
 	    break;
diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h
index 713fb81c77..187a948a7b 100644
--- a/erts/emulator/beam/global.h
+++ b/erts/emulator/beam/global.h
@@ -42,6 +42,9 @@
 #include "erl_utils.h"
 #include "erl_port.h"
 #include "erl_gc.h"
+#define ERTS_BINARY_TYPES_ONLY__
+#include "erl_binary.h"
+#undef ERTS_BINARY_TYPES_ONLY__
 
 struct enif_func_t;
 
@@ -126,7 +129,7 @@ typedef struct de_proc_entry {
 			                PROC_AWAIT_LOAD == Wants to be notified when we
 			                reloaded the driver (old was locked) */
     Uint    flags;                   /* ERL_FL_DE_DEREFERENCED when reload in progress */
-    Eterm   heap[REF_THING_SIZE];    /* "ref heap" */
+    Eterm   heap[ERTS_REF_THING_SIZE];    /* "ref heap" */
     struct  de_proc_entry *next;
 } DE_ProcEntry;
 
@@ -214,118 +217,6 @@ extern Eterm erts_ddll_monitor_driver(Process *p,
 				      ErtsProcLocks plocks);
 
 /*
-** Just like the driver binary but with initial flags
-** Note that the two structures Binary and ErlDrvBinary HAVE to
-** be equal except for extra fields in the beginning of the struct.
-** ErlDrvBinary is defined in erl_driver.h.
-** When driver_alloc_binary is called, a Binary is allocated, but 
-** the pointer returned is to the address of the first element that
-** also occurs in the ErlDrvBinary struct (driver.*binary takes care if this).
-** The driver need never know about additions to the internal Binary of the
-** emulator. One should however NEVER be sloppy when mixing ErlDrvBinary
-** and Binary, the macros below can convert one type to the other, as they both
-** in reality are equal.
-*/
-
-#ifdef ARCH_32
- /* *DO NOT USE* only for alignment. */
-#define ERTS_BINARY_STRUCT_ALIGNMENT Uint32 align__;
-#else
-#define ERTS_BINARY_STRUCT_ALIGNMENT
-#endif
-
-/* Add fields in ERTS_INTERNAL_BINARY_FIELDS, otherwise the drivers crash */
-#define ERTS_INTERNAL_BINARY_FIELDS				\
-    UWord flags;							\
-    erts_refc_t refc;						\
-    ERTS_BINARY_STRUCT_ALIGNMENT
-
-typedef struct binary {
-    ERTS_INTERNAL_BINARY_FIELDS
-    SWord orig_size;
-    char orig_bytes[1]; /* to be continued */
-} Binary;
-
-#define ERTS_SIZEOF_Binary(Sz) \
-    (offsetof(Binary,orig_bytes) + (Sz))
-
-typedef struct {
-    ERTS_INTERNAL_BINARY_FIELDS
-    SWord orig_size;
-    void (*destructor)(Binary *);
-    union {
-        struct {
-            ERTS_BINARY_STRUCT_ALIGNMENT
-            char data[1];
-        } aligned;
-        struct {
-            char data[1];
-        } unaligned;
-    } u;
-} ErtsMagicBinary;
-
-#ifdef ARCH_32
-#define ERTS_MAGIC_BIN_BYTES_TO_ALIGN 4
-#else
-#define ERTS_MAGIC_BIN_BYTES_TO_ALIGN 0
-#endif
-
-typedef union {
-    Binary binary;
-    ErtsMagicBinary magic_binary;
-    struct {
-	ERTS_INTERNAL_BINARY_FIELDS
-	ErlDrvBinary binary;
-    } driver;
-} ErtsBinary;
-
-/*
- * 'Binary' alignment:
- *   Address of orig_bytes[0] of a Binary should always be 8-byte aligned.
- * It is assumed that the flags, refc, and orig_size fields are 4 bytes on
- * 32-bits architectures and 8 bytes on 64-bits architectures.
- */
-
-#define ERTS_MAGIC_BIN_DESTRUCTOR(BP) \
-  ((ErtsBinary *) (BP))->magic_binary.destructor
-#define ERTS_MAGIC_BIN_DATA(BP) \
-  ((void *) ((ErtsBinary *) (BP))->magic_binary.u.aligned.data)
-#define ERTS_MAGIC_DATA_OFFSET \
-  (offsetof(ErtsMagicBinary,u.aligned.data) - offsetof(Binary,orig_bytes))
-#define ERTS_MAGIC_BIN_ORIG_SIZE(Sz) \
-  (ERTS_MAGIC_DATA_OFFSET + (Sz))
-#define ERTS_MAGIC_BIN_SIZE(Sz) \
-  (offsetof(ErtsMagicBinary,u.aligned.data) + (Sz))
-
-/* On 32-bit arch these macro variants will save memory
-   by not forcing 8-byte alignment for the magic payload.
-*/
-#define ERTS_MAGIC_BIN_UNALIGNED_DATA(BP) \
-  ((void *) ((ErtsBinary *) (BP))->magic_binary.u.unaligned.data)
-#define ERTS_MAGIC_UNALIGNED_DATA_OFFSET \
-  (offsetof(ErtsMagicBinary,u.unaligned.data) - offsetof(Binary,orig_bytes))
-#define ERTS_MAGIC_BIN_UNALIGNED_DATA_SIZE(BP) \
-  ((BP)->orig_size - ERTS_MAGIC_UNALIGNED_DATA_OFFSET)
-#define ERTS_MAGIC_BIN_UNALIGNED_ORIG_SIZE(Sz) \
-  (ERTS_MAGIC_UNALIGNED_DATA_OFFSET + (Sz))
-#define ERTS_MAGIC_BIN_UNALIGNED_SIZE(Sz) \
-  (offsetof(ErtsMagicBinary,u.unaligned.data) + (Sz))
-#define ERTS_MAGIC_BIN_FROM_UNALIGNED_DATA(DATA) \
-  ((ErtsBinary*)((char*)(DATA) - offsetof(ErtsMagicBinary,u.unaligned.data)))
-
-
-#define Binary2ErlDrvBinary(B) (&((ErtsBinary *) (B))->driver.binary)
-#define ErlDrvBinary2Binary(D) ((Binary *) \
-				(((char *) (D)) \
-				 - offsetof(ErtsBinary, driver.binary)))
-
-/* A "magic" binary flag */
-#define BIN_FLAG_MAGIC      1
-#define BIN_FLAG_USR1       2 /* Reserved for use by different modules too mark */
-#define BIN_FLAG_USR2       4 /*  certain binaries as special (used by ets) */
-#define BIN_FLAG_DRV        8
-
-/*
  * This structure represents one type of a binary in a process.
  */
 
@@ -386,6 +277,7 @@ union erl_off_heap_ptr {
     ProcBin *pb;
     struct erl_fun_thing* fun;
     struct external_thing_* ext;
+    ErtsMRefThing *mref;
     Eterm* ep;
     void* voidp;
 };
@@ -978,21 +870,6 @@ void erts_bif_info_init(void);
 
 /* bif.c */
 
-ERTS_GLB_INLINE Eterm
-erts_proc_store_ref(Process *c_p, Uint32 ref[ERTS_MAX_REF_NUMBERS]);
-
-#if ERTS_GLB_INLINE_INCL_FUNC_DEF
-
-ERTS_GLB_INLINE Eterm
-erts_proc_store_ref(Process *c_p, Uint32 ref[ERTS_MAX_REF_NUMBERS])
-{
-    Eterm *hp = HAlloc(c_p, REF_THING_SIZE);
-    write_ref_thing(hp, ref[0], ref[1], ref[2]);
-    return make_internal_ref(hp);
-}
-
-#endif
-
 void erts_queue_monitor_message(Process *,
 				ErtsProcLocks*,
 				Eterm,
diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c
index 5446ebaab0..dad24d7ef6 100644
--- a/erts/emulator/beam/io.c
+++ b/erts/emulator/beam/io.c
@@ -1445,7 +1445,7 @@ finalize_force_imm_drv_call(ErtsTryImmDrvCallState *sp)
     erts_unblock_fpe(sp->fpe_was_unmasked);
 }
 
-#define ERTS_QUEUE_PORT_SCHED_OP_REPLY_SIZE (REF_THING_SIZE + 3)
+#define ERTS_QUEUE_PORT_SCHED_OP_REPLY_SIZE (ERTS_REF_THING_SIZE + 3)
 
 static ERTS_INLINE void
 queue_port_sched_op_reply(Process *rp,
@@ -1460,7 +1460,7 @@ queue_port_sched_op_reply(Process *rp,
 
     ref= make_internal_ref(hp);
     write_ref_thing(hp, ref_num[0], ref_num[1], ref_num[2]);
-    hp += REF_THING_SIZE;
+    hp += ERTS_REF_THING_SIZE;
 
     msg = TUPLE2(hp, ref, msg);
 
@@ -3099,7 +3099,7 @@ port_monitor(Port *prt, erts_aint32_t state, Eterm origin,
 
     ASSERT(is_pid(origin));
     ASSERT(is_atom(name) || is_port(name) || name == NIL);
-    ASSERT(is_internal_ref(ref));
+    ASSERT(is_internal_ordinary_ref(ref));
 
     if (!(state & ERTS_PORT_SFLGS_INVALID_LOOKUP)) {
         ErtsProcLocks p_locks = ERTS_PROC_LOCK_LINK;
@@ -3124,7 +3124,7 @@ static int
 port_sig_monitor(Port *prt, erts_aint32_t state, int op,
                  ErtsProc2PortSigData *sigdp)
 {
-    Eterm hp[REF_THING_SIZE];
+    Eterm hp[ERTS_REF_THING_SIZE];
     Eterm ref = make_internal_ref(&hp);
     write_ref_thing(hp, sigdp->ref[0], sigdp->ref[1], sigdp->ref[2]);
 
@@ -3245,7 +3245,7 @@ static int
 port_sig_demonitor(Port *prt, erts_aint32_t state, int op,
                    ErtsProc2PortSigData *sigdp)
 {
-    Eterm hp[REF_THING_SIZE];
+    Eterm hp[ERTS_REF_THING_SIZE];
     Eterm ref = make_internal_ref(&hp);
     write_ref_thing(hp, sigdp->u.demonitor.ref[0],
                     sigdp->u.demonitor.ref[1],
@@ -3302,10 +3302,10 @@ ErtsPortOpResult erts_port_demonitor(Process *origin, ErtsDemonitorMode mode,
     sigdp->u.demonitor.origin = origin->common.id;
     sigdp->u.demonitor.name = target->common.id;
     {
-        RefThing *reft = ref_thing_ptr(ref);
+	Uint32 *nums = internal_ref_numbers(ref);
         /* Start from 1 skip ref arity */
         sys_memcpy(sigdp->u.demonitor.ref,
-                   internal_thing_ref_numbers(reft),
+		   nums,
                    sizeof(sigdp->u.demonitor.ref));
     }
 
@@ -5413,7 +5413,7 @@ reply_io_bytes(void *vreq)
 	    rp_locks = ERTS_PROC_LOCK_MAIN;
 	}
 
-	hsz = 5 /* 4-tuple */ + REF_THING_SIZE;
+	hsz = 5 /* 4-tuple */ + ERTS_REF_THING_SIZE;
 
 	erts_bld_uint64(NULL, &hsz, in);
 	erts_bld_uint64(NULL, &hsz, out);
@@ -5422,7 +5422,7 @@ reply_io_bytes(void *vreq)
 
 	ref = make_internal_ref(hp);
 	write_ref_thing(hp, req->refn[0], req->refn[1], req->refn[2]);
-	hp += REF_THING_SIZE;
+	hp += ERTS_REF_THING_SIZE;
 
 	ein = erts_bld_uint64(&hp, NULL, in);
 	eout = erts_bld_uint64(&hp, NULL, out);
@@ -5451,7 +5451,7 @@ erts_request_io_bytes(Process *c_p)
     ErtsIOBytesReq *req = erts_alloc(ERTS_ALC_T_IOB_REQ,
 				     sizeof(ErtsIOBytesReq));
 
-    hp = HAlloc(c_p, REF_THING_SIZE);
+    hp = HAlloc(c_p, ERTS_REF_THING_SIZE);
     ref = erts_sched_make_ref_in_buffer(esdp, hp);
     refn = internal_ref_numbers(ref);
 
@@ -7609,20 +7609,16 @@ erl_drv_convert_time_unit(ErlDrvTime val,
 
 static void ref_to_driver_monitor(Eterm ref, ErlDrvMonitor *mon)
 {
-    RefThing *refp;
-    ASSERT(is_internal_ref(ref));
-    ERTS_CT_ASSERT(sizeof(RefThing) <= sizeof(ErlDrvMonitor));
-    refp = ref_thing_ptr(ref);
-    memset(mon,0,sizeof(ErlDrvMonitor));
-    memcpy(mon,refp,sizeof(RefThing));
+    ERTS_CT_ASSERT(sizeof(ErtsOIRefStorage) <= sizeof(ErlDrvMonitor));
+    erts_oiref_storage_save((ErtsOIRefStorage *) mon, ref);
 }
 
 
 static int do_driver_monitor_process(Port *prt,
-				     Eterm *buf,
 				     ErlDrvTermData process,
 				     ErlDrvMonitor *monitor)
 {
+    Eterm buf[ERTS_REF_THING_SIZE];
     Process *rp;
     Eterm ref;
 
@@ -7665,26 +7661,22 @@ int driver_monitor_process(ErlDrvPort drvport,
     /* Now (in SMP) we should have either the port lock (if we have a scheduler) or the port data lock
        (if we're a driver thread) */
     ERTS_SMP_LC_ASSERT((sched != NULL || prt->port_data_lock));
-    {
-	DeclareTmpHeapNoproc(buf,REF_THING_SIZE);
-	UseTmpHeapNoproc(REF_THING_SIZE);
-	ret = do_driver_monitor_process(prt,buf,process,monitor);
-	UnUseTmpHeapNoproc(REF_THING_SIZE);
-    }
+    ret = do_driver_monitor_process(prt,process,monitor);
     DRV_MONITOR_UNLOCK_PDL(prt);
     return ret;
 }
 
-static int do_driver_demonitor_process(Port *prt, Eterm *buf,
-				       const ErlDrvMonitor *monitor)
+static int do_driver_demonitor_process(Port *prt, const ErlDrvMonitor *monitor)
 {
+    Eterm heap[ERTS_REF_THING_SIZE];
+    Eterm *hp = &heap[0];
     Process *rp;
     Eterm ref;
     ErtsMonitor *mon;
     Eterm to;
 
-    memcpy(buf,monitor,sizeof(Eterm)*REF_THING_SIZE);
-    ref = make_internal_ref(buf);
+    ref = erts_oiref_storage_make_ref((ErtsOIRefStorage *) monitor, &hp),
+
     mon = erts_lookup_monitor(ERTS_P_MONITORS(prt), ref);
     if (mon == NULL) {
 	return 1;
@@ -7728,25 +7720,21 @@ int driver_demonitor_process(ErlDrvPort drvport,
     /* Now we should have either the port lock (if we have a scheduler) or the port data lock
        (if we're a driver thread) */
     ERTS_SMP_LC_ASSERT((sched != NULL || prt->port_data_lock));
-    {
-	DeclareTmpHeapNoproc(buf,REF_THING_SIZE);
-	UseTmpHeapNoproc(REF_THING_SIZE);
-	ret = do_driver_demonitor_process(prt,buf,monitor);
-	UnUseTmpHeapNoproc(REF_THING_SIZE);
-    }
+    ret = do_driver_demonitor_process(prt,monitor);
     DRV_MONITOR_UNLOCK_PDL(prt);
     return ret;
 }
 
-static ErlDrvTermData do_driver_get_monitored_process(Port *prt, Eterm *buf,
-					    const ErlDrvMonitor *monitor)
+static ErlDrvTermData do_driver_get_monitored_process(Port *prt,const ErlDrvMonitor *monitor)
 {
     Eterm ref;
     ErtsMonitor *mon;
     Eterm to;
+    Eterm heap[ERTS_REF_THING_SIZE];
+    Eterm *hp = &heap[0];
+
+    ref = erts_oiref_storage_make_ref((ErtsOIRefStorage *) monitor, &hp),
 
-    memcpy(buf,monitor,sizeof(Eterm)*REF_THING_SIZE);
-    ref = make_internal_ref(buf);
     mon = erts_lookup_monitor(ERTS_P_MONITORS(prt), ref);
     if (mon == NULL) {
 	return driver_term_nil;
@@ -7774,12 +7762,7 @@ ErlDrvTermData driver_get_monitored_process(ErlDrvPort drvport,
     /* Now we should have either the port lock (if we have a scheduler) or the port data lock
        (if we're a driver thread) */
     ERTS_SMP_LC_ASSERT((sched != NULL || prt->port_data_lock));
-    {
-	DeclareTmpHeapNoproc(buf,REF_THING_SIZE);
-	UseTmpHeapNoproc(REF_THING_SIZE);
-	ret = do_driver_get_monitored_process(prt,buf,monitor);
-	UnUseTmpHeapNoproc(REF_THING_SIZE);
-    }
+    ret = do_driver_get_monitored_process(prt,monitor);
     DRV_MONITOR_UNLOCK_PDL(prt);
     return ret;
 }
@@ -7788,7 +7771,8 @@ ErlDrvTermData driver_get_monitored_process(ErlDrvPort drvport,
 int driver_compare_monitors(const ErlDrvMonitor *monitor1,
 			    const ErlDrvMonitor *monitor2)
 {
-    return memcmp(monitor1,monitor2,sizeof(ErlDrvMonitor));
+    return erts_oiref_storage_cmp((ErtsOIRefStorage *) monitor1,
+				  (ErtsOIRefStorage *) monitor2);
 }
 
 void erts_fire_port_monitor(Port *prt, Eterm ref)
diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c
index 25940edab7..fc459e17ad 100644
--- a/erts/emulator/beam/utils.c
+++ b/erts/emulator/beam/utils.c
@@ -1838,7 +1838,7 @@ make_internal_hash(Eterm term)
 	    break;
 	    case REF_SUBTAG:
 		UINT32_HASH(internal_ref_numbers(term)[0], HCONST_7);
-                ASSERT(internal_ref_no_of_numbers(term) == 3);
+                ASSERT(internal_ref_no_numbers(term) == 3);
                 UINT32_HASH_2(internal_ref_numbers(term)[1],
                               internal_ref_numbers(term)[2], HCONST_8);
                 goto pop_next;
@@ -1847,7 +1847,7 @@ make_internal_hash(Eterm term)
             {
                 ExternalThing* thing = external_thing_ptr(term);
 
-                ASSERT(external_thing_ref_no_of_numbers(thing) == 3);
+                ASSERT(external_thing_ref_no_numbers(thing) == 3);
                 /* See limitation #2 */
             #ifdef ARCH_64
                 POINTER_HASH(thing->node, HCONST_7);
@@ -2486,22 +2486,20 @@ tailrecur_ne:
 
 		anum = external_thing_ref_numbers(athing);
 		bnum = external_thing_ref_numbers(bthing);
-		alen = external_thing_ref_no_of_numbers(athing);
-		blen = external_thing_ref_no_of_numbers(bthing);
+		alen = external_thing_ref_no_numbers(athing);
+		blen = external_thing_ref_no_numbers(bthing);
 
 		goto ref_common;
+
 	    case REF_SUBTAG:
-		    if (!is_internal_ref(b))
-			goto not_equal;
 
-		    {
-			RefThing* athing = ref_thing_ptr(a);
-			RefThing* bthing = ref_thing_ptr(b);
-			alen = internal_thing_ref_no_of_numbers(athing);
-			blen = internal_thing_ref_no_of_numbers(bthing);
-			anum = internal_thing_ref_numbers(athing);
-			bnum = internal_thing_ref_numbers(bthing);
-		    }
+		if (!is_internal_ref(b))
+		    goto not_equal;
+
+		alen = internal_ref_no_numbers(a);
+		anum = internal_ref_numbers(a);
+		blen = internal_ref_no_numbers(b);
+		bnum = internal_ref_numbers(b);
 
 	    ref_common:
 		    ASSERT(alen > 0 && blen > 0);
@@ -3133,25 +3131,21 @@ tailrecur_ne:
 		 */
 
 		if (is_internal_ref(b)) {
-		    RefThing* bthing = ref_thing_ptr(b);
 		    bnode = erts_this_node;
-		    bnum = internal_thing_ref_numbers(bthing);
-		    blen = internal_thing_ref_no_of_numbers(bthing);
+		    blen = internal_ref_no_numbers(b);
+		    bnum = internal_ref_numbers(b);
 		} else if(is_external_ref(b)) {
 		    ExternalThing* bthing = external_thing_ptr(b);
 		    bnode = bthing->node;
 		    bnum = external_thing_ref_numbers(bthing);
-		    blen = external_thing_ref_no_of_numbers(bthing);
+		    blen = external_thing_ref_no_numbers(bthing);
 		} else {
 		    a_tag = REF_DEF;
 		    goto mixed_types;
 		}
-		{
-		    RefThing* athing = ref_thing_ptr(a);
-		    anode = erts_this_node;
-		    anum = internal_thing_ref_numbers(athing);
-		    alen = internal_thing_ref_no_of_numbers(athing);
-		}
+		anode = erts_this_node;
+		alen = internal_ref_no_numbers(a);
+		anum = internal_ref_numbers(a);
 
 	    ref_common:
 		CMP_NODES(anode, bnode);
@@ -3181,15 +3175,14 @@ tailrecur_ne:
 		goto pop_next;
 	    case (_TAG_HEADER_EXTERNAL_REF >> _TAG_PRIMARY_SIZE):
 		if (is_internal_ref(b)) {
-		    RefThing* bthing = ref_thing_ptr(b);
 		    bnode = erts_this_node;
-		    bnum = internal_thing_ref_numbers(bthing);
-		    blen = internal_thing_ref_no_of_numbers(bthing);
+		    blen = internal_ref_no_numbers(b);
+		    bnum = internal_ref_numbers(b);
 		} else if (is_external_ref(b)) {
 		    ExternalThing* bthing = external_thing_ptr(b);
 		    bnode = bthing->node;
 		    bnum = external_thing_ref_numbers(bthing);
-		    blen = external_thing_ref_no_of_numbers(bthing);
+		    blen = external_thing_ref_no_numbers(bthing);
 		} else {
 		    a_tag = EXTERNAL_REF_DEF;
 		    goto mixed_types;
@@ -3198,7 +3191,7 @@ tailrecur_ne:
 		    ExternalThing* athing = external_thing_ptr(a);
 		    anode = athing->node;
 		    anum = external_thing_ref_numbers(athing);
-		    alen = external_thing_ref_no_of_numbers(athing);
+		    alen = external_thing_ref_no_numbers(athing);
 		}
 		goto ref_common;
 	    default:
@@ -3575,40 +3568,39 @@ not_equal:
 Eterm
 store_external_or_ref_(Uint **hpp, ErlOffHeap* oh, Eterm ns)
 {
+    struct erl_off_heap_header *ohhp;
     Uint i;
     Uint size;
-    Uint *from_hp;
-    Uint *to_hp = *hpp;
+    Eterm *from_hp;
+    Eterm *to_hp = *hpp;
 
     ASSERT(is_external(ns) || is_internal_ref(ns));
 
-    if(is_external(ns)) {
-	from_hp = external_val(ns);
-	size = thing_arityval(*from_hp) + 1;
-	*hpp += size;
-
-	for(i = 0; i < size; i++)
-	    to_hp[i] = from_hp[i];
-
-	erts_smp_refc_inc(&((ExternalThing *) to_hp)->node->refc, 2);
-
-	((struct erl_off_heap_header*) to_hp)->next = oh->first;
-	oh->first = (struct erl_off_heap_header*) to_hp;
-
-	return make_external(to_hp);
-    }
-
-    /* Internal ref */
-    from_hp = internal_ref_val(ns);
-
+    from_hp = boxed_val(ns);
     size = thing_arityval(*from_hp) + 1;
-
     *hpp += size;
 
     for(i = 0; i < size; i++)
 	to_hp[i] = from_hp[i];
 
-    return make_internal_ref(to_hp);
+    if (is_external_header(*from_hp)) {
+	ExternalThing *etp = (ExternalThing *) from_hp;
+	ASSERT(is_external(ns));
+	erts_smp_refc_inc(&etp->node->refc, 2);
+    }
+    else if (is_ordinary_ref_thing(from_hp))
+	return make_internal_ref(to_hp);
+    else {
+	ErtsMRefThing *mreft = (ErtsMRefThing *) from_hp;
+	ASSERT(is_magic_ref_thing(from_hp));
+	erts_refc_inc(&mreft->mb->refc, 2);
+    }
+
+    ohhp = (struct erl_off_heap_header*) to_hp;
+    ohhp->next = oh->first;
+    oh->first = ohhp;
+
+    return make_boxed(to_hp);
 }
 
 Eterm
diff --git a/erts/emulator/test/node_container_SUITE.erl b/erts/emulator/test/node_container_SUITE.erl
index af18545bff..7356f31b75 100644
--- a/erts/emulator/test/node_container_SUITE.erl
+++ b/erts/emulator/test/node_container_SUITE.erl
@@ -50,7 +50,8 @@
          port_wrap/1,
          bad_nc/1,
          unique_pid/1,
-         iter_max_procs/1]).
+         iter_max_procs/1,
+         magic_ref/1]).
 
 suite() ->
     [{ct_hooks,[ts_install_cth]},
@@ -62,7 +63,7 @@ all() ->
      node_table_gc, dist_link_refc, dist_monitor_refc,
      node_controller_refc, ets_refc, match_spec_refc,
      timer_refc, otp_4715, pid_wrap, port_wrap, bad_nc,
-     unique_pid, iter_max_procs].
+     unique_pid, iter_max_procs, magic_ref].
 
 init_per_suite(Config) ->
     Config.
@@ -889,10 +890,46 @@ chk_max_proc_line_until(NoMoreTests, Res) ->
               chk_max_proc_line_until(NoMoreTests, Res)
     end.
 
+magic_ref(Config) when is_list(Config) ->
+    {MRef0, Addr0} = erts_debug:set_internal_state(make, magic_ref),
+    true = is_reference(MRef0),
+    {Addr0, 1, true} = erts_debug:get_internal_state({magic_ref,MRef0}),
+    MRef1 = binary_to_term(term_to_binary(MRef0)),
+    {Addr0, 2, true} = erts_debug:get_internal_state({magic_ref,MRef1}),
+    MRef0 = MRef1,
+    Me = self(),
+    {Pid, Mon} = spawn_opt(fun () ->
+				   receive
+				       {Me, MRef} ->
+					   Me ! {self(), erts_debug:get_internal_state({magic_ref,MRef})}
+				   end
+			   end,
+			   [link, monitor]),
+    Pid ! {self(), MRef0},
+    receive
+	{Pid, Info} ->
+	    {Addr0, 3, true} = Info
+    end,
+    receive
+	{'DOWN', Mon, process, Pid, _} ->
+	    ok
+    end,
+    {Addr0, 2, true} = erts_debug:get_internal_state({magic_ref,MRef0}),
+    id(MRef0),
+    id(MRef1),
+    MRefExt = term_to_binary(erts_debug:set_internal_state(make, magic_ref)),
+    garbage_collect(),
+    {MRef2, _Addr2} = binary_to_term(MRefExt),
+    true = is_reference(MRef2),
+    true = erts_debug:get_internal_state({magic_ref,MRef2}),
+    ok.
 %%
 %% -- Internal utils ---------------------------------------------------------
 %%
 
+id(X) ->
+    X.
+
 -define(ND_REFS, erts_debug:get_internal_state(node_and_dist_references)).
 
 node_container_refc_check(Node) when is_atom(Node) ->