Merge branch 'rickard/timer-optimization/OTP-12650'

* rickard/timer-optimization/OTP-12650:
  Optimized timer implementation
  Reusable red-black tree implementation

Conflicts:
	erts/emulator/beam/erl_bif_timer.c

55 files changed, 6198 insertions, 2732 deletions

diff --git a/erts/emulator/Makefile.in b/erts/emulator/Makefile.in
index b4a17e76e7..659ea1b27f 100644
--- a/erts/emulator/Makefile.in
+++ b/erts/emulator/Makefile.in
@@ -779,7 +779,7 @@ RUN_OBJS = \
 	$(OBJDIR)/erl_fun.o             $(OBJDIR)/erl_bif_port.o \
 	$(OBJDIR)/erl_term.o 		$(OBJDIR)/erl_node_tables.o \
 	$(OBJDIR)/erl_monitors.o	$(OBJDIR)/erl_process_dump.o \
-	$(OBJDIR)/erl_bif_timer.o	$(OBJDIR)/erl_cpu_topology.o \
+	$(OBJDIR)/erl_hl_timer.o	$(OBJDIR)/erl_cpu_topology.o \
 	$(OBJDIR)/erl_drv_thread.o      $(OBJDIR)/erl_bif_chksum.o \
 	$(OBJDIR)/erl_bif_re.o		$(OBJDIR)/erl_unicode.o \
 	$(OBJDIR)/packet_parser.o	$(OBJDIR)/safe_hash.o \
diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names
index 8fdcbb4058..5ec1409adf 100644
--- a/erts/emulator/beam/atom.names
+++ b/erts/emulator/beam/atom.names
@@ -68,6 +68,7 @@ atom aborted
 atom abs_path
 atom absoluteURI
 atom ac
+atom accessor
 atom active
 atom all
 atom all_but_first
@@ -94,12 +95,14 @@ atom args
 atom arg0
 atom arity
 atom asn1
+atom async
 atom asynchronous
 atom atom
 atom atom_used
 atom attributes
 atom await_port_send_result
 atom await_proc_exit
+atom await_result
 atom await_sched_wall_time_modifications
 atom awaiting_load
 atom awaiting_unload
diff --git a/erts/emulator/beam/beam_bif_load.c b/erts/emulator/beam/beam_bif_load.c
index 8c32fc7854..500a98195b 100644
--- a/erts/emulator/beam/beam_bif_load.c
+++ b/erts/emulator/beam/beam_bif_load.c
@@ -368,7 +368,7 @@ staging_epilogue(Process* c_p, int commit, Eterm res, int is_blocking,
 	ASSERT(commiter_state.stager == NULL);
 	commiter_state.stager = c_p;
 	erts_schedule_thr_prgr_later_op(smp_code_ix_commiter, NULL, &commiter_state.lop);
-	erts_smp_proc_inc_refc(c_p);
+	erts_proc_inc_refc(c_p);
 	erts_suspend(c_p, ERTS_PROC_LOCK_MAIN, NULL);
 	/*
 	 * smp_code_ix_commiter() will do the rest "later"
@@ -395,7 +395,7 @@ static void smp_code_ix_commiter(void* null)
 	erts_resume(p, ERTS_PROC_LOCK_STATUS);
     }
     erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS);
-    erts_smp_proc_dec_refc(p);
+    erts_proc_dec_refc(p);
 }
 #endif /* ERTS_SMP */
 
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 21faf8fbf2..9fe02c3724 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -2119,44 +2119,32 @@ void process_main(void)
      }
      GetArg1(1, timeout_value);
      if (timeout_value != make_small(0)) {
-#if !defined(ARCH_64) || HALFWORD_HEAP
-	 Uint time_val;
-#endif
 
-	 if (is_small(timeout_value) && signed_val(timeout_value) > 0 &&
-#if defined(ARCH_64) && !HALFWORD_HEAP
-	     ((unsigned_val(timeout_value) >> 32) == 0)
-#else
-	     1
-#endif
-	     ) {
-	     /*
-	      * The timer routiner will set c_p->i to the value in
-	      * c_p->def_arg_reg[0].  Note that it is safe to use this
-	      * location because there are no living x registers in
-	      * a receive statement.
-	      * Note that for the halfword emulator, the two first elements
-	      * of the array are used.
-	      */
-	     BeamInstr** pi = (BeamInstr**) c_p->def_arg_reg;
-	     *pi = I+3;
-	     set_timer(c_p, unsigned_val(timeout_value));
-	 } else if (timeout_value == am_infinity) {
+	 if (timeout_value == am_infinity)
 	     c_p->flags |= F_TIMO;
-#if !defined(ARCH_64) || HALFWORD_HEAP
-	 } else if (term_to_Uint(timeout_value, &time_val)) {
-	     BeamInstr** pi = (BeamInstr**) c_p->def_arg_reg;
-	     *pi = I+3;
-	     set_timer(c_p, time_val);
-#endif
-	 } else {		/* Wrong time */
-	     OpCase(i_wait_error_locked): {
-		 erts_smp_proc_unlock(c_p, ERTS_PROC_LOCKS_MSG_RECEIVE);
-		 /* Fall through */
+	 else {
+	     int tres = erts_set_proc_timer_term(c_p, timeout_value);
+	     if (tres == 0) {
+		 /*
+		  * The timer routiner will set c_p->i to the value in
+		  * c_p->def_arg_reg[0].  Note that it is safe to use this
+		  * location because there are no living x registers in
+		  * a receive statement.
+		  * Note that for the halfword emulator, the two first elements
+		  * of the array are used.
+		  */
+		 BeamInstr** pi = (BeamInstr**) c_p->def_arg_reg;
+		 *pi = I+3;
 	     }
+	     else { /* Wrong time */
+	     OpCase(i_wait_error_locked): {
+		     erts_smp_proc_unlock(c_p, ERTS_PROC_LOCKS_MSG_RECEIVE);
+		     /* Fall through */
+		 }
 	     OpCase(i_wait_error): {
-		 c_p->freason = EXC_TIMEOUT_VALUE;
-		 goto find_func_info;
+		     c_p->freason = EXC_TIMEOUT_VALUE;
+		     goto find_func_info;
+		 }
 	     }
 	 }
 
@@ -2205,7 +2193,7 @@ void process_main(void)
      if ((c_p->flags & (F_INSLPQUEUE | F_TIMO)) == 0) {
 	 BeamInstr** p = (BeamInstr **) c_p->def_arg_reg;
 	 *p = I+3;
-	 set_timer(c_p, Arg(1));
+	 erts_set_proc_timer_uword(c_p, Arg(1));
      }
      goto wait2;
  }
@@ -5946,7 +5934,7 @@ build_stacktrace(Process* c_p, Eterm exc) {
      * (e.g. spawn_link(erlang, abs, [1])).
      */
     if (fi.current == NULL) {
-	erts_set_current_function(&fi, c_p->initial);
+	erts_set_current_function(&fi, c_p->u.initial);
 	args = am_true; /* Just in case */
     } else {
 	args = get_args_from_exc(exc);
diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c
index 4f2958c664..2b782f4484 100644
--- a/erts/emulator/beam/bif.c
+++ b/erts/emulator/beam/bif.c
@@ -44,6 +44,7 @@
 #include "erl_bits.h"
 #include "erl_bif_unique.h"
 
+Export *erts_await_result;
 static Export* flush_monitor_messages_trap = NULL;
 static Export* set_cpu_topology_trap = NULL;
 static Export* await_proc_exit_trap = NULL;
@@ -831,26 +832,6 @@ BIF_RETTYPE monitor_2(BIF_ALIST_2)
     return ret;
 }
 
-BIF_RETTYPE erts_internal_monitor_process_2(BIF_ALIST_2)
-{
-    if (is_not_internal_pid(BIF_ARG_1)) {
-	if (is_external_pid(BIF_ARG_1)
-	    && (external_pid_dist_entry(BIF_ARG_1)
-		== erts_this_dist_entry)) {
-	    BIF_RET(am_false);
-	}
-	goto badarg;
-    }
-
-    if (is_not_internal_ref(BIF_ARG_2))
-	goto badarg;
-
-    BIF_RET(local_pid_monitor(BIF_P, BIF_ARG_1, BIF_ARG_2, 1));
-
-badarg:
-    BIF_ERROR(BIF_P, BADARG);
-}
-
 /**********************************************************************/
 /* this is a combination of the spawn and link BIFs */
 
@@ -4904,6 +4885,10 @@ void erts_init_bif(void)
 #endif
 		     , &bif_return_trap);
 
+    erts_await_result = erts_export_put(am_erts_internal,
+					am_await_result,
+					1);
+
     erts_init_trap_export(&dsend_continue_trap_export,
 			  am_erts_internal, am_dsend_continue_trap, 1,
 			  dsend_continue_trap_1);
diff --git a/erts/emulator/beam/bif.h b/erts/emulator/beam/bif.h
index d461c3f479..b877711544 100644
--- a/erts/emulator/beam/bif.h
+++ b/erts/emulator/beam/bif.h
@@ -20,6 +20,7 @@
 #ifndef __BIF_H__
 #define __BIF_H__
 
+extern Export *erts_await_result;
 extern Export* erts_format_cpu_topology_trap;
 extern Export *erts_convert_time_unit_trap;
 
diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab
index 471f687101..eadba3eaff 100644
--- a/erts/emulator/beam/bif.tab
+++ b/erts/emulator/beam/bif.tab
@@ -171,11 +171,6 @@ bif erts_internal:map_hashmap_children/1
 
 bif erts_internal:time_unit/0
 
-bif erts_internal:get_bif_timer_servers/0
-bif erts_internal:create_bif_timer/0
-bif erts_internal:access_bif_timer/1
-
-bif erts_internal:monitor_process/2
 bif erts_internal:is_system_process/1
 
 # inet_db support
@@ -220,6 +215,15 @@ bif math:sqrt/1
 bif math:atan2/2
 bif math:pow/2
 
+bif erlang:start_timer/3
+bif erlang:start_timer/4
+bif erlang:send_after/3
+bif erlang:send_after/4
+bif erlang:cancel_timer/1
+bif erlang:cancel_timer/2
+bif erlang:read_timer/1
+bif erlang:read_timer/2
+
 bif erlang:make_tuple/2
 bif erlang:append_element/2
 bif erlang:make_tuple/3
diff --git a/erts/emulator/beam/break.c b/erts/emulator/beam/break.c
index e2fa572546..02e65cb9c6 100644
--- a/erts/emulator/beam/break.c
+++ b/erts/emulator/beam/break.c
@@ -36,7 +36,7 @@
 #include "atom.h"
 #include "beam_load.h"
 #include "erl_instrument.h"
-#include "erl_bif_timer.h"
+#include "erl_hl_timer.h"
 #include "erl_thr_progress.h"
 
 /* Forward declarations -- should really appear somewhere else */
@@ -108,7 +108,7 @@ process_killer(void)
 		case 'k': {
 		    ErtsProcLocks rp_locks = ERTS_PROC_LOCKS_XSIG_SEND;
 		    erts_aint32_t state;
-		    erts_smp_proc_inc_refc(rp);
+		    erts_proc_inc_refc(rp);
 		    erts_smp_proc_lock(rp, rp_locks);
 		    state = erts_smp_atomic32_read_acqb(&rp->state);
 		    if (state & (ERTS_PSFLG_FREE
@@ -131,7 +131,7 @@ process_killer(void)
 						     0);
 		    }
 		    erts_smp_proc_unlock(rp, rp_locks);
-		    erts_smp_proc_dec_refc(rp);
+		    erts_proc_dec_refc(rp);
 		}
 		case 'n': br = 1; break;
 		case 'r': return;
@@ -227,9 +227,9 @@ print_process_info(int to, void *to_arg, Process *p)
      * Display the initial function name
      */
     erts_print(to, to_arg, "Spawned as: %T:%T/%bpu\n",
-	       p->initial[INITIAL_MOD],
-	       p->initial[INITIAL_FUN],
-	       p->initial[INITIAL_ARI]);
+	       p->u.initial[INITIAL_MOD],
+	       p->u.initial[INITIAL_FUN],
+	       p->u.initial[INITIAL_ARI]);
     
     if (p->current != NULL) {
 	if (running) {
diff --git a/erts/emulator/beam/code_ix.c b/erts/emulator/beam/code_ix.c
index 4344558348..d925709bd0 100644
--- a/erts/emulator/beam/code_ix.c
+++ b/erts/emulator/beam/code_ix.c
@@ -130,7 +130,7 @@ int erts_try_seize_code_write_permission(Process* c_p)
 	ASSERT(code_writing_process != c_p);
 	qitem = erts_alloc(ERTS_ALC_T_CODE_IX_LOCK_Q, sizeof(*qitem));
 	qitem->p = c_p;
-	erts_smp_proc_inc_refc(c_p);
+	erts_proc_inc_refc(c_p);
 	qitem->next = code_write_queue;
 	code_write_queue = qitem;
 	erts_suspend(c_p, ERTS_PROC_LOCK_MAIN, NULL);
@@ -151,7 +151,7 @@ void erts_release_code_write_permission(void)
 	}
 	erts_smp_proc_unlock(qitem->p, ERTS_PROC_LOCK_STATUS);
 	code_write_queue = qitem->next;
-	erts_smp_proc_dec_refc(qitem->p);
+	erts_proc_dec_refc(qitem->p);
 	erts_free(ERTS_ALC_T_CODE_IX_LOCK_Q, qitem);
     }
     code_writing_process = NULL;
diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c
index e396395dde..b832626e7b 100644
--- a/erts/emulator/beam/erl_alloc.c
+++ b/erts/emulator/beam/erl_alloc.c
@@ -39,7 +39,7 @@
 #include "erl_instrument.h"
 #include "erl_mseg.h"
 #include "erl_monitors.h"
-#include "erl_bif_timer.h"
+#include "erl_hl_timer.h"
 #include "erl_cpu_topology.h"
 #include "erl_thr_queue.h"
 #if defined(ERTS_ALC_T_DRV_SEL_D_STATE) || defined(ERTS_ALC_T_DRV_EV_D_STATE)
@@ -575,6 +575,15 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
     fix_type_sizes[ERTS_ALC_FIX_TYPE_IX(ERTS_ALC_T_THR_Q_EL_SL)]
 	= sizeof(ErtsThrQElement_t);
 #endif
+    fix_type_sizes[ERTS_ALC_FIX_TYPE_IX(ERTS_ALC_T_LL_PTIMER)]
+	= erts_timer_type_size(ERTS_ALC_T_LL_PTIMER);
+    fix_type_sizes[ERTS_ALC_FIX_TYPE_IX(ERTS_ALC_T_HL_PTIMER)]
+	= erts_timer_type_size(ERTS_ALC_T_HL_PTIMER);
+    fix_type_sizes[ERTS_ALC_FIX_TYPE_IX(ERTS_ALC_T_BIF_TIMER)]
+	= erts_timer_type_size(ERTS_ALC_T_BIF_TIMER);
+    fix_type_sizes[ERTS_ALC_FIX_TYPE_IX(ERTS_ALC_T_ABIF_TIMER)]
+	= erts_timer_type_size(ERTS_ALC_T_ABIF_TIMER);
+
 #ifdef HARD_DEBUG
     hdbg_init();
 #endif
@@ -3186,7 +3195,7 @@ reply_alloc_info(void *vair)
 	rp_locks &= ~ERTS_PROC_LOCK_MAIN;
  
     erts_smp_proc_unlock(rp, rp_locks);
-    erts_smp_proc_dec_refc(rp);
+    erts_proc_dec_refc(rp);
 
     if (erts_smp_atomic32_dec_read_nob(&air->refc) == 0)
 	aireq_free(air);
@@ -3260,7 +3269,7 @@ erts_request_alloc_info(struct process *c_p,
     erts_smp_atomic32_init_nob(&air->refc,
 			       (erts_aint32_t) erts_no_schedulers);
 
-    erts_smp_proc_add_refc(c_p, (Sint32) erts_no_schedulers);
+    erts_proc_add_refc(c_p, (Sint) erts_no_schedulers);
 
 #ifdef ERTS_SMP
     if (erts_no_schedulers > 1)
diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types
index 074f864dee..57c506458c 100644
--- a/erts/emulator/beam/erl_alloc.types
+++ b/erts/emulator/beam/erl_alloc.types
@@ -163,9 +163,13 @@ type	MSG_ROOTS	TEMPORARY	PROCESSES	msg_roots
 type	ROOTSET		TEMPORARY	PROCESSES	root_set
 type	LOADER_TMP	TEMPORARY	CODE		loader_tmp
 type	PREPARED_CODE	SHORT_LIVED	CODE		prepared_code
-type	BIF_TIMER_TABLE	LONG_LIVED	SYSTEM		bif_timer_table
-type	SL_BIF_TIMER	SHORT_LIVED	PROCESSES	bif_timer_sl
-type	LL_BIF_TIMER	STANDARD	PROCESSES	bif_timer_ll
+type	TIMER_SERVICE	LONG_LIVED	SYSTEM		timer_service
+type    LL_PTIMER	FIXED_SIZE	PROCESSES	ll_ptimer
+type    HL_PTIMER	FIXED_SIZE	PROCESSES	hl_ptimer
+type    BIF_TIMER	FIXED_SIZE	PROCESSES	bif_timer
+type    ABIF_TIMER	FIXED_SIZE	PROCESSES	accessor_bif_timer
+type    TIMER_REQUEST	SHORT_LIVED	PROCESSES	timer_request
+type    BTM_YIELD_STATE	SHORT_LIVED	PROCESSES	btm_yield_state
 type	REG_TABLE	STANDARD	SYSTEM		reg_tab
 type	FUN_TABLE	STANDARD	CODE		fun_tab
 type	DIST_TABLE	STANDARD	SYSTEM		dist_tab
@@ -324,8 +328,6 @@ type	ACTIVE_PROCS	STANDARD	PROCESSES	active_procs
 +endif
 
 +if smp
-type	SL_PTIMER	SHORT_LIVED	SYSTEM		ptimer_sl
-type	LL_PTIMER	STANDARD	SYSTEM		ptimer_ll
 type	SYS_MSG_Q	SHORT_LIVED	PROCESSES	system_messages_queue
 type	FP_EXCEPTION	LONG_LIVED	SYSTEM		fp_exception
 type	LL_MPATHS	LONG_LIVED	SYSTEM		ll_migration_paths
@@ -365,7 +367,6 @@ type	AINFO_REQ	STANDARD_LOW	SYSTEM		alloc_info_request
 type	SCHED_WTIME_REQ	STANDARD_LOW	SYSTEM		sched_wall_time_request
 type	GC_INFO_REQ	STANDARD_LOW	SYSTEM		gc_info_request
 type	PORT_DATA_HEAP	STANDARD_LOW	SYSTEM		port_data_heap
-type	BIF_TIMER_DATA	LONG_LIVED_LOW	SYSTEM		bif_timer_data
 
 +else # "fullword"
 
@@ -386,7 +387,6 @@ type	AINFO_REQ	SHORT_LIVED	SYSTEM		alloc_info_request
 type	SCHED_WTIME_REQ	SHORT_LIVED	SYSTEM		sched_wall_time_request
 type	GC_INFO_REQ	SHORT_LIVED	SYSTEM		gc_info_request
 type	PORT_DATA_HEAP	STANDARD	SYSTEM		port_data_heap
-type	BIF_TIMER_DATA	LONG_LIVED	SYSTEM		bif_timer_data
 
 +endif
 
diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c
index fa7de23f00..f74aea80a7 100644
--- a/erts/emulator/beam/erl_bif_info.c
+++ b/erts/emulator/beam/erl_bif_info.c
@@ -1054,9 +1054,9 @@ process_info_aux(Process *BIF_P,
     case am_initial_call:
 	hp = HAlloc(BIF_P, 3+4);
 	res = TUPLE3(hp,
-		     rp->initial[INITIAL_MOD],
-		     rp->initial[INITIAL_FUN],
-		     make_small(rp->initial[INITIAL_ARI]));
+		     rp->u.initial[INITIAL_MOD],
+		     rp->u.initial[INITIAL_FUN],
+		     make_small(rp->u.initial[INITIAL_ARI]));
 	hp += 4;
 	break;
 
@@ -2129,6 +2129,8 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1)
 	BIF_RET(erts_has_time_correction() ? am_true : am_false);
     } else if (ERTS_IS_ATOM_STR("start_time", BIF_ARG_1)) {
 	BIF_RET(erts_get_monotonic_start_time(BIF_P));
+    } else if (ERTS_IS_ATOM_STR("end_time", BIF_ARG_1)) {
+	BIF_RET(erts_get_monotonic_end_time(BIF_P));
     } else if (ERTS_IS_ATOM_STR("time_warp_mode", BIF_ARG_1)) {
 	switch (erts_time_warp_mode()) {
 	case ERTS_NO_TIME_WARP_MODE: {
diff --git a/erts/emulator/beam/erl_bif_timer.c b/erts/emulator/beam/erl_bif_timer.c
deleted file mode 100644
index ac4a5644ac..0000000000
--- a/erts/emulator/beam/erl_bif_timer.c
+++ /dev/null
@@ -1,849 +0,0 @@
-/*
- * %CopyrightBegin%
- *
- * Copyright Ericsson AB 2005-2012. All Rights Reserved.
- *
- * The contents of this file are subject to the Erlang Public License,
- * Version 1.1, (the "License"); you may not use this file except in
- * compliance with the License. You should have received a copy of the
- * Erlang Public License along with this software. If not, it can be
- * retrieved online at http://www.erlang.org/.
- *
- * Software distributed under the License is distributed on an "AS IS"
- * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
- * the License for the specific language governing rights and limitations
- * under the License.
- *
- * %CopyrightEnd%
- */
-
-#ifdef HAVE_CONFIG_H
-#  include "config.h"
-#endif
-
-#include "erl_bif_timer.h"
-#include "global.h"
-#include "bif.h"
-#include "error.h"
-#include "big.h"
-#include "erl_thr_progress.h"
-#include "erl_bif_unique.h"
-
-/****************************************************************************
-** BIF Timer support
-****************************************************************************/
-
-#define BTM_FLG_SL_TIMER	(((Uint32) 1) << 0)
-#define BTM_FLG_CANCELED	(((Uint32) 1) << 1)
-#define BTM_FLG_HEAD		(((Uint32) 1) << 2)
-#define BTM_FLG_BYNAME		(((Uint32) 1) << 3)
-#define BTM_FLG_WRAP		(((Uint32) 1) << 4)
-
-struct ErtsBifTimer_ {
-    struct {
-	union {
-	    ErtsBifTimer **head;
-	    ErtsBifTimer *prev;
-	} u;
-	ErtsBifTimer *next;
-    } tab;
-    union {
-	Eterm name;
-	struct {
-	    ErtsBifTimer *prev;
-	    ErtsBifTimer *next;
-	    Process *ess;
-	} proc;
-    } receiver;
-    ErlTimer tm;
-    ErlHeapFragment* bp;
-    Uint32 flags;
-    Eterm message;
-    Uint32 ref_numbers[ERTS_REF_NUMBERS];
-};
-
-#ifdef SMALL_MEMORY
-#define TIMER_HASH_VEC_SZ	3331
-#define BTM_PREALC_SZ		10
-#else
-#define TIMER_HASH_VEC_SZ	10007
-#define BTM_PREALC_SZ		100
-#endif
-static ErtsBifTimer **bif_timer_tab;  
-static Uint no_bif_timers;
-
-
-static erts_smp_rwmtx_t bif_timer_lock;
-
-#define erts_smp_safe_btm_rwlock(P, L) \
-	safe_btm_lock((P), (L), 1)
-#define erts_smp_safe_btm_rlock(P, L) \
-	safe_btm_lock((P), (L), 0)
-#define erts_smp_btm_rwlock() \
-	erts_smp_rwmtx_rwlock(&bif_timer_lock)
-#define erts_smp_btm_tryrwlock() \
-	erts_smp_rwmtx_tryrwlock(&bif_timer_lock)
-#define erts_smp_btm_rwunlock() \
-	erts_smp_rwmtx_rwunlock(&bif_timer_lock)
-#define erts_smp_btm_rlock() \
-	erts_smp_rwmtx_rlock(&bif_timer_lock)
-#define erts_smp_btm_tryrlock() \
-	erts_smp_rwmtx_tryrlock(&bif_timer_lock)
-#define erts_smp_btm_runlock() \
-	erts_smp_rwmtx_runlock(&bif_timer_lock)
-#define erts_smp_btm_lock_init() \
-	erts_smp_rwmtx_init(&bif_timer_lock, "bif_timers")
-
-
-static ERTS_INLINE int
-safe_btm_lock(Process *c_p, ErtsProcLocks c_p_locks, int rw_lock)
-{
-    ASSERT(c_p && c_p_locks);
-#ifdef ERTS_SMP
-    if ((rw_lock ? erts_smp_btm_tryrwlock() : erts_smp_btm_tryrlock()) != EBUSY)
-	return 0;
-    erts_smp_proc_unlock(c_p, c_p_locks);
-    if (rw_lock)
-	erts_smp_btm_rwlock();
-    else
-	erts_smp_btm_rlock();
-    erts_smp_proc_lock(c_p, c_p_locks);
-    if (ERTS_PROC_IS_EXITING(c_p)) {
-	if (rw_lock)
-	    erts_smp_btm_rwunlock();
-	else
-	    erts_smp_btm_runlock();
-	return 1;
-    }
-#endif
-    return 0;
-}
-
-ERTS_SCHED_PREF_PALLOC_IMPL(btm_pre, ErtsBifTimer, BTM_PREALC_SZ)
-
-static ERTS_INLINE int
-get_index(Uint32 *ref_numbers, Uint32 len)
-{
-    Uint32 hash;
-    /* len can potentially be larger than ERTS_REF_NUMBERS
-       if it has visited another node... */
-    if (len > ERTS_REF_NUMBERS)
-	len = ERTS_REF_NUMBERS;
-
-#if ERTS_REF_NUMBERS != 3
-#error "ERTS_REF_NUMBERS changed. Update me..."
-#endif
-    switch (len) {
-    case 3: if (!ref_numbers[2]) len = 2;
-    case 2: if (!ref_numbers[1]) len = 1;
-    default:  break;
-    }
-
-    ASSERT(1 <= len && len <= ERTS_REF_NUMBERS);
-
-    hash = block_hash((byte *) ref_numbers, len * sizeof(Uint32), 0x08d12e65);
-    return (int) (hash % ((Uint32) TIMER_HASH_VEC_SZ));
-}
-
-static Eterm
-create_ref(Uint *hp, Uint32 *ref_numbers, Uint32 len)
-{
-    Uint32 *datap;
-    int i;
-
-
-    if (len > ERTS_MAX_REF_NUMBERS) {
-	/* Such large refs should no be able to appear in the emulator */
-	erl_exit(1, "%s:%d: Internal error\n", __FILE__, __LINE__);
-    }
-
-#if defined(ARCH_64) && !HALFWORD_HEAP
-    hp[0] = make_ref_thing_header(len/2 + 1);
-    datap = (Uint32 *) &hp[1];
-    *(datap++) = len;
-#else
-    hp[0] = make_ref_thing_header(len);
-    datap = (Uint32 *) &hp[1];
-#endif
-
-    for (i = 0; i < len; i++)
-	datap[i] = ref_numbers[i];
-
-    return make_internal_ref(hp);
-}
-
-static int
-eq_non_standard_ref_numbers(Uint32 *rn1, Uint32 len1, Uint32 *rn2, Uint32 len2)
-{
-#if defined(ARCH_64) && !HALFWORD_HEAP
-#define MAX_REF_HEAP_SZ (1+(ERTS_MAX_REF_NUMBERS/2+1))
-#else
-#define MAX_REF_HEAP_SZ (1+ERTS_MAX_REF_NUMBERS)
-#endif
-    DeclareTmpHeapNoproc(r1_hp,(MAX_REF_HEAP_SZ * 2));
-    Eterm *r2_hp = r1_hp +MAX_REF_HEAP_SZ;
-
-    return eq(create_ref(r1_hp, rn1, len1), create_ref(r2_hp, rn2, len2));
-#undef MAX_REF_HEAP_SZ
-}
-
-static ERTS_INLINE int
-eq_ref_numbers(Uint32 *rn1, Uint32 len1, Uint32 *rn2, Uint32 len2)
-{
-    int res;
-    if (len1 != ERTS_REF_NUMBERS || len2 != ERTS_REF_NUMBERS) {
-	/* Can potentially happen, but will never... */
-	return eq_non_standard_ref_numbers(rn1, len1, rn2, len2);
-    }
-
-#if ERTS_REF_NUMBERS != 3
-#error "ERTS_REF_NUMBERS changed. Update me..."
-#endif
-    res = rn1[0] == rn2[0] && rn1[1] == rn2[1] && rn1[2] == rn2[2];
-
-    ASSERT(res
-	   ? eq_non_standard_ref_numbers(rn1, len1, rn2, len2)
-	   : !eq_non_standard_ref_numbers(rn1, len1, rn2, len2));
-
-    return res;
-}
-
-static ERTS_INLINE ErtsBifTimer *
-tab_find(Eterm ref)
-{
-    Uint32 *ref_numbers = internal_ref_numbers(ref);
-    Uint32 ref_numbers_len = internal_ref_no_of_numbers(ref);
-    int ix = get_index(ref_numbers, ref_numbers_len);
-    ErtsBifTimer* btm;
-
-    for (btm = bif_timer_tab[ix]; btm; btm = btm->tab.next)
-	if (eq_ref_numbers(ref_numbers, ref_numbers_len,
-			   btm->ref_numbers, ERTS_REF_NUMBERS))
-	    return btm;
-    return NULL;
-}
-
-static ERTS_INLINE void
-tab_remove(ErtsBifTimer* btm)
-{
-    if (btm->flags & BTM_FLG_HEAD) {
-	*btm->tab.u.head = btm->tab.next;
-	if (btm->tab.next) {
-	    btm->tab.next->flags |= BTM_FLG_HEAD;
-	    btm->tab.next->tab.u.head = btm->tab.u.head;
-	}
-    }
-    else {
-	btm->tab.u.prev->tab.next = btm->tab.next;
-	if (btm->tab.next)
-	    btm->tab.next->tab.u.prev = btm->tab.u.prev;
-    }
-    btm->flags |= BTM_FLG_CANCELED;
-    ASSERT(no_bif_timers > 0);
-    no_bif_timers--;
-}
-
-static ERTS_INLINE void
-tab_insert(ErtsBifTimer* btm)
-{
-    int ix = get_index(btm->ref_numbers, ERTS_REF_NUMBERS);
-    ErtsBifTimer* btm_list = bif_timer_tab[ix];
-
-    if (btm_list) {
-	btm_list->flags &= ~BTM_FLG_HEAD;
-	btm_list->tab.u.prev = btm;
-    }
-
-    btm->flags |= BTM_FLG_HEAD;
-    btm->tab.u.head = &bif_timer_tab[ix];
-    btm->tab.next = btm_list;
-    bif_timer_tab[ix] = btm;
-    no_bif_timers++;
-}
-
-static ERTS_INLINE void
-link_proc(Process *p, ErtsBifTimer* btm)
-{
-    btm->receiver.proc.ess = p;
-    btm->receiver.proc.prev = NULL;
-    btm->receiver.proc.next = p->u.bif_timers;
-    if (p->u.bif_timers)	
-	p->u.bif_timers->receiver.proc.prev = btm;
-    p->u.bif_timers = btm;
-}
-
-static ERTS_INLINE void
-unlink_proc(ErtsBifTimer* btm)
-{
-    if (btm->receiver.proc.prev)
-	btm->receiver.proc.prev->receiver.proc.next = btm->receiver.proc.next;
-    else
-	btm->receiver.proc.ess->u.bif_timers = btm->receiver.proc.next;
-    if (btm->receiver.proc.next)
-	btm->receiver.proc.next->receiver.proc.prev = btm->receiver.proc.prev;
-}
-
-static void
-bif_timer_cleanup(ErtsBifTimer* btm)
-{
-    ASSERT(btm);
-
-    if (btm->bp)
-	free_message_buffer(btm->bp);
-
-    if (!btm_pre_free(btm)) {
-	if (btm->flags & BTM_FLG_SL_TIMER)
-	    erts_free(ERTS_ALC_T_SL_BIF_TIMER, (void *) btm);
-	else
-	    erts_free(ERTS_ALC_T_LL_BIF_TIMER, (void *) btm);
-    }
-}
-
-static void
-bif_timer_timeout(ErtsBifTimer* btm)
-{
-    ASSERT(btm);
-
-
-    erts_smp_btm_rwlock();
-
-    if (btm->flags & BTM_FLG_CANCELED) {
-    /*
-     * A concurrent cancel is ongoing. Do not send the timeout message,
-     * but cleanup here since the cancel call-back won't be called.
-     */
-#ifndef ERTS_SMP
-	ASSERT(0);
-#endif
-    }
-    else {
-	ErtsProcLocks rp_locks = 0;
-	Process* rp;
-
-	tab_remove(btm);
-
-	ASSERT(!erts_get_current_process());
-
-	if (btm->flags & BTM_FLG_BYNAME)
-	    rp = erts_whereis_process(NULL, 0, btm->receiver.name, 0, 0);
-	else {
-	    rp = btm->receiver.proc.ess;
-	    unlink_proc(btm);
-	}
-
-	if (rp) {
-	    Eterm message;
-	    ErlHeapFragment *bp;
-
-	    bp = btm->bp;
-	    btm->bp = NULL; /* Prevent cleanup of message buffer... */
-
-	    if (!(btm->flags & BTM_FLG_WRAP))
-		message = btm->message;
-	    else {
-#if ERTS_REF_NUMBERS != 3
-#error "ERTS_REF_NUMBERS changed. Update me..."
-#endif
-		Eterm ref;
-		Uint *hp;
-		Uint wrap_size = REF_THING_SIZE + 4;
-		message = btm->message;
-
-		if (!bp) {
-		    ErlOffHeap *ohp;
-		    ASSERT(is_immed(message));
-		    hp = erts_alloc_message_heap(wrap_size,
-						 &bp,
-						 &ohp,
-						 rp,
-						 &rp_locks);
-		} else {
-		    Eterm old_size = bp->used_size;
-		    bp = erts_resize_message_buffer(bp, old_size + wrap_size,
-						    &message, 1);
-		    hp = &bp->mem[0] + old_size;
-		}
-
-		write_ref_thing(hp,
-				btm->ref_numbers[0],
-				btm->ref_numbers[1],
-				btm->ref_numbers[2]);
-		ref = make_internal_ref(hp);
-		hp += REF_THING_SIZE;
-		message = TUPLE3(hp, am_timeout, ref, message);
-	    }
-
-	    erts_queue_message(rp, &rp_locks, bp, message, NIL);
-	    erts_smp_proc_unlock(rp, rp_locks);
-	}
-    }
-
-    erts_smp_btm_rwunlock();
-
-    bif_timer_cleanup(btm);
-}
-
-static Eterm
-setup_bif_timer(Uint32 xflags,
-		Process *c_p,
-		Eterm time,
-		Eterm receiver,
-		Eterm message)
-{
-    Process *rp;
-    ErtsBifTimer* btm;
-    Uint timeout;
-    Eterm ref;
-    Uint32 *ref_numbers;
-    
-    if (!term_to_Uint(time, &timeout))
-	return THE_NON_VALUE;
-#if defined(ARCH_64) && !HALFWORD_HEAP
-    if ((timeout >> 32) != 0)
-	return THE_NON_VALUE;
-#endif
-    if (is_not_internal_pid(receiver) && is_not_atom(receiver))
-	return THE_NON_VALUE;
-
-    ref = erts_make_ref(c_p);
-
-    if (is_atom(receiver))
-	rp = NULL;
-    else {
-	rp = erts_pid2proc(c_p, ERTS_PROC_LOCK_MAIN,
-			   receiver, ERTS_PROC_LOCK_MSGQ);
-	if (!rp)
-	    return ref;
-    }
-
-    if (timeout < ERTS_ALC_MIN_LONG_LIVED_TIME) {
-	if (timeout < 1000) {
-	    btm = btm_pre_alloc();
-	    if (!btm)
-		goto sl_timer_alloc;
-	    btm->flags = 0;
-	}
-	else {
-	sl_timer_alloc:
-	    btm = (ErtsBifTimer *) erts_alloc(ERTS_ALC_T_SL_BIF_TIMER,
-					      sizeof(ErtsBifTimer));
-	    btm->flags = BTM_FLG_SL_TIMER;
-	}
-    }
-    else {
-	btm = (ErtsBifTimer *) erts_alloc(ERTS_ALC_T_LL_BIF_TIMER,
-					 sizeof(ErtsBifTimer));
-	btm->flags = 0;
-    }
-
-    if (rp) {
-	link_proc(rp, btm);
-	erts_smp_proc_unlock(rp, ERTS_PROC_LOCK_MSGQ);
-    }
-    else {
-	ASSERT(is_atom(receiver));
-	btm->receiver.name = receiver;
-	btm->flags |= BTM_FLG_BYNAME;
-    }
-
-    btm->flags |= xflags;
-
-    ref_numbers = internal_ref_numbers(ref);
-    ASSERT(internal_ref_no_of_numbers(ref) == 3);
-#if ERTS_REF_NUMBERS != 3
-#error "ERTS_REF_NUMBERS changed. Update me..."
-#endif
-    btm->ref_numbers[0] = ref_numbers[0];
-    btm->ref_numbers[1] = ref_numbers[1];
-    btm->ref_numbers[2] = ref_numbers[2];
-
-    ASSERT(eq_ref_numbers(btm->ref_numbers, ERTS_REF_NUMBERS,
-			  ref_numbers, ERTS_REF_NUMBERS));
-
-    if (is_immed(message)) {
-	btm->bp = NULL;
-	btm->message = message;
-    }
-    else {
-	ErlHeapFragment* bp;
-	Eterm* hp;
-	Uint size;
-
-	size = size_object(message);
-	btm->bp = bp = new_message_buffer(size);
-	hp = bp->mem;
-	btm->message = copy_struct(message, size, &hp, &bp->off_heap);
-    }
-
-    tab_insert(btm);
-    ASSERT(btm == tab_find(ref));
-    erts_init_timer(&btm->tm);
-    erts_set_timer(&btm->tm,
-		  (ErlTimeoutProc) bif_timer_timeout,
-		  (ErlCancelProc) bif_timer_cleanup,
-		  (void *) btm,
-		  timeout);
-    return ref;
-}
-
-BIF_RETTYPE old_send_after_3(BIF_ALIST_3);
-/* send_after(Time, Pid, Message) -> Ref */
-BIF_RETTYPE old_send_after_3(BIF_ALIST_3)
-{
-    Eterm res;
-
-    if (erts_smp_safe_btm_rwlock(BIF_P, ERTS_PROC_LOCK_MAIN))
-	ERTS_BIF_EXITED(BIF_P);
-
-    res = setup_bif_timer(0, BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3);
-
-    erts_smp_btm_rwunlock();
-
-    if (is_non_value(res)) {
-	BIF_ERROR(BIF_P, BADARG);
-    }
-    else {
-	ASSERT(is_internal_ref(res));
-	BIF_RET(res);
-    }
-}
-
-BIF_RETTYPE old_start_timer_3(BIF_ALIST_3);
-/* start_timer(Time, Pid, Message) -> Ref */
-BIF_RETTYPE old_start_timer_3(BIF_ALIST_3)
-{
-    Eterm res;
-
-    if (erts_smp_safe_btm_rwlock(BIF_P, ERTS_PROC_LOCK_MAIN))
-	ERTS_BIF_EXITED(BIF_P);
-
-    res = setup_bif_timer(BTM_FLG_WRAP, BIF_P, BIF_ARG_1, BIF_ARG_2, BIF_ARG_3);
-
-    erts_smp_btm_rwunlock();
-
-    if (is_non_value(res)) {
-	BIF_ERROR(BIF_P, BADARG);
-    }
-    else {
-	ASSERT(is_internal_ref(res));
-	BIF_RET(res);
-    }
-}
-
-BIF_RETTYPE old_cancel_timer_1(BIF_ALIST_1);
-/* cancel_timer(Ref) -> false | RemainingTime */
-BIF_RETTYPE old_cancel_timer_1(BIF_ALIST_1)
-{
-    Eterm res;
-    ErtsBifTimer *btm;
-
-    if (is_not_internal_ref(BIF_ARG_1)) {
-	if (is_ref(BIF_ARG_1)) {
-	    BIF_RET(am_false);
-	}
-	BIF_ERROR(BIF_P, BADARG);
-    }
-
-    if (erts_smp_safe_btm_rwlock(BIF_P, ERTS_PROC_LOCK_MAIN))
-	ERTS_BIF_EXITED(BIF_P);
-
-    btm = tab_find(BIF_ARG_1);
-    if (!btm || btm->flags & BTM_FLG_CANCELED) {
-	erts_smp_btm_rwunlock();
-	res = am_false;
-    }
-    else {
-	Uint left = erts_time_left(&btm->tm);
-	if (!(btm->flags & BTM_FLG_BYNAME)) {
-	    erts_smp_proc_lock(btm->receiver.proc.ess, ERTS_PROC_LOCK_MSGQ);
-	    unlink_proc(btm);
-	    erts_smp_proc_unlock(btm->receiver.proc.ess, ERTS_PROC_LOCK_MSGQ);
-	}
-	tab_remove(btm);
-	ASSERT(!tab_find(BIF_ARG_1));
-	erts_cancel_timer(&btm->tm);
-	erts_smp_btm_rwunlock();
-	res = erts_make_integer(left, BIF_P);
-    }
-
-    BIF_RET(res);
-}
-
-BIF_RETTYPE old_read_timer_1(BIF_ALIST_1);
-/* read_timer(Ref) -> false | RemainingTime */
-BIF_RETTYPE old_read_timer_1(BIF_ALIST_1)
-{
-    Eterm res;
-    ErtsBifTimer *btm;
-
-    if (is_not_internal_ref(BIF_ARG_1)) {
-	if (is_ref(BIF_ARG_1)) {
-	    BIF_RET(am_false);
-	}
-	BIF_ERROR(BIF_P, BADARG);
-    }
-
-    if (erts_smp_safe_btm_rlock(BIF_P, ERTS_PROC_LOCK_MAIN))
-	ERTS_BIF_EXITED(BIF_P);
-
-    btm = tab_find(BIF_ARG_1);
-    if (!btm || btm->flags & BTM_FLG_CANCELED) {
-	res = am_false;
-    }
-    else {
-	Uint left = erts_time_left(&btm->tm);
-	res = erts_make_integer(left, BIF_P);
-    }
-
-    erts_smp_btm_runlock();
-
-    BIF_RET(res);
-}
-
-void
-erts_print_bif_timer_info(int to, void *to_arg)
-{
-    int i;
-    int lock = !ERTS_IS_CRASH_DUMPING;
-
-    if (lock)
-	erts_smp_btm_rlock();
-
-    for (i = 0; i < TIMER_HASH_VEC_SZ; i++) {
-	ErtsBifTimer *btm;
-	for (btm = bif_timer_tab[i]; btm; btm = btm->tab.next) {
-	    Eterm receiver = (btm->flags & BTM_FLG_BYNAME
-			      ? btm->receiver.name
-			      : btm->receiver.proc.ess->common.id);
-	    erts_print(to, to_arg, "=timer:%T\n", receiver);
-	    erts_print(to, to_arg, "Message: %T\n", btm->message);
-	    erts_print(to, to_arg, "Time left: %u\n",
-		       erts_time_left(&btm->tm));
-	}
-    }
-
-    if (lock)
-	erts_smp_btm_runlock();
-}
-
-
-void
-erts_cancel_bif_timers(Process *p, ErtsProcLocks plocks)
-{
-    ErtsBifTimer *btm;
-
-    if (erts_smp_btm_tryrwlock() == EBUSY) {
-	erts_smp_proc_unlock(p, plocks);
-	erts_smp_btm_rwlock();
-	erts_smp_proc_lock(p, plocks);
-    }
-
-    btm = p->u.bif_timers;
-    while (btm) {
-	ErtsBifTimer *tmp_btm;
-	ASSERT(!(btm->flags & BTM_FLG_CANCELED));
-	tab_remove(btm);
-	tmp_btm = btm;
-	btm = btm->receiver.proc.next;
-	erts_cancel_timer(&tmp_btm->tm);
-    }
-
-    p->u.bif_timers = NULL;
-
-    erts_smp_btm_rwunlock();
-}
-
-static void erts_old_bif_timer_init(void)
-{
-    int i;
-    no_bif_timers = 0;
-    init_btm_pre_alloc();
-    erts_smp_btm_lock_init();
-    bif_timer_tab = erts_alloc(ERTS_ALC_T_BIF_TIMER_TABLE,
-			       sizeof(ErtsBifTimer *)*TIMER_HASH_VEC_SZ);
-    for (i = 0; i < TIMER_HASH_VEC_SZ; ++i)
-	bif_timer_tab[i] = NULL;
-}
-
-Uint
-erts_bif_timer_memory_size(void)
-{
-    Uint res;
-    int lock = !ERTS_IS_CRASH_DUMPING;
-
-    if (lock)
-	erts_smp_btm_rlock();
-
-    res = (sizeof(ErtsBifTimer *)*TIMER_HASH_VEC_SZ
-	   + no_bif_timers*sizeof(ErtsBifTimer));
-
-    if (lock)
-	erts_smp_btm_runlock();
-
-    return res;
-}
-
-
-void
-erts_bif_timer_foreach(void (*func)(Eterm, Eterm, ErlHeapFragment *, void *),
-		       void *arg)
-{
-    int i;
-
-    ERTS_SMP_LC_ASSERT(erts_smp_thr_progress_is_blocking());
-
-    for (i = 0; i < TIMER_HASH_VEC_SZ; i++) {
-	ErtsBifTimer *btm;
-	for (btm = bif_timer_tab[i]; btm; btm = btm->tab.next) {
-	    (*func)((btm->flags & BTM_FLG_BYNAME
-		     ? btm->receiver.name
-		     : btm->receiver.proc.ess->common.id),
-		    btm->message,
-		    btm->bp,
-		    arg);
-	}
-    }
-}
-
-typedef struct {
-    Uint ref_heap[REF_THING_SIZE];
-    Eterm pid[1];
-} ErtsBifTimerServers;
-
-static ErtsBifTimerServers *bif_timer_servers;
-
-void erts_bif_timer_init(void)
-{
-    erts_old_bif_timer_init();
-}
-
-void
-erts_bif_timer_start_servers(Eterm parent)
-{
-    Process *parent_proc;
-    Eterm *hp, btr_ref, arg_list_end;
-    ErlSpawnOpts so;
-    int i;
-
-    bif_timer_servers = erts_alloc(ERTS_ALC_T_BIF_TIMER_DATA,
-				   (sizeof(ErtsBifTimerServers)
-				    + (sizeof(Eterm)*(erts_no_schedulers-1))));
-
-    so.flags = SPO_USE_ARGS|SPO_SYSTEM_PROC|SPO_PREFER_SCHED|SPO_OFF_HEAP_MSGS;
-    so.min_heap_size  = H_MIN_SIZE;
-    so.min_vheap_size = BIN_VH_MIN_SIZE;
-    so.priority       = PRIORITY_MAX;
-    so.max_gen_gcs    = (Uint16) erts_smp_atomic32_read_nob(&erts_max_gen_gcs);
-
-    /*
-     * Parent is "init" and schedulers have not yet been started, so it
-     * *should* be alive and well...
-     */
-    ASSERT(is_internal_pid(parent));
-    parent_proc = (Process *) erts_ptab_pix2intptr_ddrb(&erts_proc,
-							internal_pid_index(parent));
-    ASSERT(parent_proc);
-    ASSERT(parent_proc->common.id == parent);
-    ASSERT(!ERTS_PROC_IS_EXITING(parent_proc));
-
-    erts_smp_proc_lock(parent_proc, ERTS_PROC_LOCK_MAIN);
-
-    hp = HAlloc(parent_proc, 2*erts_no_schedulers + 2 + REF_THING_SIZE);
-
-    btr_ref = erts_make_ref_in_buffer(hp);
-    hp += REF_THING_SIZE;
-
-    arg_list_end = CONS(hp, btr_ref, NIL);
-    hp += 2;
-
-    for (i = 0; i < erts_no_schedulers; i++) {
-	int sched = i+1;
-	Eterm arg_list = CONS(hp, make_small(i+1), arg_list_end);
-	hp += 2;
-
-	so.scheduler = sched; /* Preferred scheduler */
-
-	bif_timer_servers->pid[i] = erl_create_process(parent_proc,
-						       am_erts_internal,
-						       am_bif_timer_server,
-						       arg_list,
-						       &so);
-    }
-
-    erts_smp_proc_unlock(parent_proc, ERTS_PROC_LOCK_MAIN);
-
-    hp = internal_ref_val(btr_ref);
-    for (i = 0; i < REF_THING_SIZE; i++)
-	bif_timer_servers->ref_heap[i] = hp[i];
-}
-
-BIF_RETTYPE
-erts_internal_get_bif_timer_servers_0(BIF_ALIST_0)
-{
-    int i;
-    Eterm *hp, res = NIL;
-
-    hp = HAlloc(BIF_P, erts_no_schedulers*2);
-    for (i = erts_no_schedulers-1; i >= 0; i--) {
-	res = CONS(hp, bif_timer_servers->pid[i], res);
-	hp += 2;
-    }
-    BIF_RET(res);
-}
-
-BIF_RETTYPE
-erts_internal_access_bif_timer_1(BIF_ALIST_1)
-{
-    int ix;
-    Uint32 *rdp;
-    Eterm ref, pid, *hp, res;
-
-    if (is_not_internal_ref(BIF_ARG_1)) {
-	if (is_not_ref(BIF_ARG_1))
-	    BIF_ERROR(BIF_P, BADARG);
-	BIF_RET(am_undefined);
-    }
-
-    rdp = internal_ref_numbers(BIF_ARG_1);
-    ix = (int) erts_get_ref_numbers_thr_id(rdp);
-    if (ix < 1 || erts_no_schedulers < ix)
-	BIF_RET(am_undefined);
-
-    pid = bif_timer_servers->pid[ix-1];
-    ASSERT(is_internal_pid(pid));
-
-    hp = HAlloc(BIF_P, 3 /* 2-tuple */ + REF_THING_SIZE);
-    for (ix = 0; ix < REF_THING_SIZE; ix++)
-	hp[ix] = bif_timer_servers->ref_heap[ix];
-    ref = make_internal_ref(&hp[0]);
-    hp += REF_THING_SIZE;
-
-    res = TUPLE2(hp, ref, pid);
-    BIF_RET(res);
-}
-
-BIF_RETTYPE
-erts_internal_create_bif_timer_0(BIF_ALIST_0)
-{
-    ErtsSchedulerData *esdp = ERTS_PROC_GET_SCHDATA(BIF_P);
-    Eterm *hp, btr_ref, t_ref, pid, res;
-    int ix;
-
-    hp = HAlloc(BIF_P, 4 /* 3-tuple */ + 2*REF_THING_SIZE);
-    for (ix = 0; ix < REF_THING_SIZE; ix++)
-	hp[ix] = bif_timer_servers->ref_heap[ix];
-    btr_ref = make_internal_ref(&hp[0]);
-    hp += REF_THING_SIZE;
-
-    t_ref = erts_sched_make_ref_in_buffer(esdp, hp);
-    hp += REF_THING_SIZE;
-
-    ASSERT(erts_get_ref_numbers_thr_id(internal_ref_numbers(t_ref))
-	   == (Uint32) esdp->no);
-
-    pid = bif_timer_servers->pid[((int) esdp->no) - 1];
-    
-    res = TUPLE3(hp, btr_ref, pid, t_ref);
-
-    BIF_RET(res);
-}
diff --git a/erts/emulator/beam/erl_bif_timer.h b/erts/emulator/beam/erl_bif_timer.h
deleted file mode 100644
index c2f5dfd3c3..0000000000
--- a/erts/emulator/beam/erl_bif_timer.h
+++ /dev/null
@@ -1,37 +0,0 @@
-/*
- * %CopyrightBegin%
- * 
- * Copyright Ericsson AB 2005-2009. All Rights Reserved.
- * 
- * The contents of this file are subject to the Erlang Public License,
- * Version 1.1, (the "License"); you may not use this file except in
- * compliance with the License. You should have received a copy of the
- * Erlang Public License along with this software. If not, it can be
- * retrieved online at http://www.erlang.org/.
- * 
- * Software distributed under the License is distributed on an "AS IS"
- * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
- * the License for the specific language governing rights and limitations
- * under the License.
- * 
- * %CopyrightEnd%
- */
-
-
-#ifndef ERL_BIF_TIMER_H__
-#define ERL_BIF_TIMER_H__
-
-typedef struct ErtsBifTimer_ ErtsBifTimer;
-
-#include "sys.h"
-#include "erl_process.h"
-#include "erl_message.h"
-
-Uint erts_bif_timer_memory_size(void);
-void erts_print_bif_timer_info(int to, void *to_arg);
-void erts_cancel_bif_timers(Process *p, ErtsProcLocks plocks);
-void erts_bif_timer_init(void);
-void erts_bif_timer_foreach(void (*func)(Eterm,Eterm,ErlHeapFragment *,void *),
-			    void *arg);
-void erts_bif_timer_start_servers(Eterm);
-#endif
diff --git a/erts/emulator/beam/erl_bif_trace.c b/erts/emulator/beam/erl_bif_trace.c
index ac57205c47..13e0160648 100644
--- a/erts/emulator/beam/erl_bif_trace.c
+++ b/erts/emulator/beam/erl_bif_trace.c
@@ -359,7 +359,7 @@ trace_pattern(Process* p, Eterm MFA, Eterm Pattern, Eterm flaglist)
 	ASSERT(finish_bp.stager == NULL);
 	finish_bp.stager = p;
 	erts_schedule_thr_prgr_later_op(smp_bp_finisher, NULL, &finish_bp.lop);
-	erts_smp_proc_inc_refc(p);
+	erts_proc_inc_refc(p);
 	erts_suspend(p, ERTS_PROC_LOCK_MAIN, NULL);
 	ERTS_BIF_YIELD_RETURN(p, make_small(matches));
     }
@@ -393,7 +393,7 @@ static void smp_bp_finisher(void* null)
 	    erts_resume(p, ERTS_PROC_LOCK_STATUS);
 	}
 	erts_smp_proc_unlock(p, ERTS_PROC_LOCK_STATUS);
-	erts_smp_proc_dec_refc(p);
+	erts_proc_dec_refc(p);
     }
 }
 #endif /* ERTS_SMP */
diff --git a/erts/emulator/beam/erl_db.c b/erts/emulator/beam/erl_db.c
index fff892ae54..2e2cb98354 100644
--- a/erts/emulator/beam/erl_db.c
+++ b/erts/emulator/beam/erl_db.c
@@ -620,7 +620,7 @@ BIF_RETTYPE ets_safe_fixtable_2(BIF_ALIST_2)
     erts_fprintf(stderr,
 		"ets:safe_fixtable(%T,%T); Process: %T, initial: %T:%T/%bpu\n",
 		BIF_ARG_1, BIF_ARG_2, BIF_P->common.id,
-		BIF_P->initial[0], BIF_P->initial[1], BIF_P->initial[2]);
+		BIF_P->u.initial[0], BIF_P->u.initial[1], BIF_P->u.initial[2]);
 #endif
     kind = (BIF_ARG_2 == am_true) ? LCK_READ : LCK_WRITE_REC; 
 
@@ -1247,7 +1247,7 @@ BIF_RETTYPE ets_rename_2(BIF_ALIST_2)
     erts_fprintf(stderr,
 		"ets:rename(%T,%T); Process: %T, initial: %T:%T/%bpu\n",
 		BIF_ARG_1, BIF_ARG_2, BIF_P->common.id,
-		BIF_P->initial[0], BIF_P->initial[1], BIF_P->initial[2]);
+		BIF_P->u.initial[0], BIF_P->u.initial[1], BIF_P->u.initial[2]);
 #endif
 
 
@@ -1563,7 +1563,7 @@ BIF_RETTYPE ets_new_2(BIF_ALIST_2)
     erts_fprintf(stderr,
 		"ets:new(%T,%T)=%T; Process: %T, initial: %T:%T/%bpu\n",
 		 BIF_ARG_1, BIF_ARG_2, ret, BIF_P->common.id,
-		 BIF_P->initial[0], BIF_P->initial[1], BIF_P->initial[2]);
+		 BIF_P->u.initial[0], BIF_P->u.initial[1], BIF_P->u.initial[2]);
 	erts_fprintf(stderr, "ets: new: meta_pid_to_tab common.memory_size = %ld\n",
 		     erts_smp_atomic_read_nob(&meta_pid_to_tab->common.memory_size));
 	erts_fprintf(stderr, "ets: new: meta_pid_to_fixed_tab common.memory_size = %ld\n",
@@ -1696,7 +1696,7 @@ BIF_RETTYPE ets_delete_1(BIF_ALIST_1)
     erts_fprintf(stderr,
 		"ets:delete(%T); Process: %T, initial: %T:%T/%bpu\n",
 		BIF_ARG_1, BIF_P->common.id,
-		BIF_P->initial[0], BIF_P->initial[1], BIF_P->initial[2]);
+		BIF_P->u.initial[0], BIF_P->u.initial[1], BIF_P->u.initial[2]);
 #endif
 
     CHECK_TABLES();
diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c
index eadd558dda..1785fc27be 100644
--- a/erts/emulator/beam/erl_gc.c
+++ b/erts/emulator/beam/erl_gc.c
@@ -97,10 +97,10 @@ typedef struct {
 
 static Uint setup_rootset(Process*, Eterm*, int, Rootset*);
 static void cleanup_rootset(Rootset *rootset);
-static Uint combined_message_size(Process* p, int off_heap_msgs);
+static Uint combined_message_size(Process* p);
 static void remove_message_buffers(Process* p);
-static int major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int off_heap_msgs);
-static int minor_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int off_heap_msgs);
+static int major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl);
+static int minor_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl);
 static void do_minor(Process *p, Uint new_sz, Eterm* objv, int nobj);
 static Eterm* sweep_rootset(Rootset *rootset, Eterm* htop, char* src, Uint src_size);
 static Eterm* sweep_one_area(Eterm* n_hp, Eterm* n_htop, char* src, Uint src_size);
@@ -403,9 +403,7 @@ erts_garbage_collect(Process* p, int need, Eterm* objv, int nobj)
 {
     Uint reclaimed_now = 0;
     int done = 0;
-    int off_heap_msgs;
     ErtsMonotonicTime start_time = 0; /* Shut up faulty warning... */
-    erts_aint32_t state;
     ErtsSchedulerData *esdp;
 #ifdef USE_VM_PROBES
     DTRACE_CHARBUF(pidbuf, DTRACE_TERM_BUF_SIZE);
@@ -422,10 +420,9 @@ erts_garbage_collect(Process* p, int need, Eterm* objv, int nobj)
         trace_gc(p, am_gc_start);
     }
 
-    state = erts_smp_atomic32_read_bor_nob(&p->state, ERTS_PSFLG_GC);
-    off_heap_msgs = state & ERTS_PSFLG_OFF_HEAP_MSGS;
+    (void) erts_smp_atomic32_read_bor_nob(&p->state, ERTS_PSFLG_GC);
     if (erts_system_monitor_long_gc != 0)
-	start_time = erts_get_monotonic_time();
+	start_time = erts_get_monotonic_time(esdp);
 
     ERTS_CHK_OFFHEAP(p);
 
@@ -448,11 +445,11 @@ erts_garbage_collect(Process* p, int need, Eterm* objv, int nobj)
     while (!done) {
 	if ((FLAGS(p) & F_NEED_FULLSWEEP) != 0) {
 	    DTRACE2(gc_major_start, pidbuf, need);
-	    done = major_collection(p, need, objv, nobj, &reclaimed_now, off_heap_msgs);
+	    done = major_collection(p, need, objv, nobj, &reclaimed_now);
 	    DTRACE2(gc_major_end, pidbuf, reclaimed_now);
 	} else {
 	    DTRACE2(gc_minor_start, pidbuf, need);
-	    done = minor_collection(p, need, objv, nobj, &reclaimed_now, off_heap_msgs);
+	    done = minor_collection(p, need, objv, nobj, &reclaimed_now);
 	    DTRACE2(gc_minor_end, pidbuf, reclaimed_now);
 	}
     }
@@ -477,7 +474,7 @@ erts_garbage_collect(Process* p, int need, Eterm* objv, int nobj)
 	Uint gc_time;
 	if (erts_test_long_gc_sleep)
 	    while (0 != erts_milli_sleep(erts_test_long_gc_sleep));
-	end_time = erts_get_monotonic_time();
+	end_time = erts_get_monotonic_time(esdp);
 	gc_time = (Uint) ERTS_MONOTONIC_TO_MSEC(end_time - start_time);
 	if (gc_time && gc_time > erts_system_monitor_long_gc) {
 	    monitor_long_gc(p, gc_time);
@@ -833,7 +830,7 @@ erts_garbage_collect_literals(Process* p, Eterm* literals,
 }
 
 static int
-minor_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int off_heap_msgs)
+minor_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl)
 {
     Uint mature = HIGH_WATER(p) - HEAP_START(p);
 
@@ -872,22 +869,20 @@ minor_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int of
 	Uint size_after;
 	Uint need_after;
 	Uint stack_size = STACK_SZ_ON_HEAP(p);
-	Uint fragments = MBUF_SIZE(p) + combined_message_size(p, off_heap_msgs);
+	Uint fragments = MBUF_SIZE(p) + combined_message_size(p);
 	Uint size_before = fragments + (HEAP_TOP(p) - HEAP_START(p));
 	Uint new_sz = next_heap_size(p, HEAP_SIZE(p) + fragments, 0);
 
         do_minor(p, new_sz, objv, nobj);
 
-	if (!off_heap_msgs) {
-	    /*
-	     * Copy newly received message onto the end of the new heap.
-	     */
-	    ErtsGcQuickSanityCheck(p);
-	    for (msgp = p->msg.first; msgp; msgp = msgp->next) {
-		if (msgp->data.attached) {
-		    erts_move_msg_attached_data_to_heap(&p->htop, &p->off_heap, msgp);
-		    ErtsGcQuickSanityCheck(p);
-		}
+	/*
+	 * Copy newly received message onto the end of the new heap.
+	 */
+	ErtsGcQuickSanityCheck(p);
+	for (msgp = p->msg.first; msgp; msgp = msgp->next) {
+	    if (msgp->data.attached) {
+		erts_move_msg_attached_data_to_heap(&p->htop, &p->off_heap, msgp);
+		ErtsGcQuickSanityCheck(p);
 	    }
 	}
 	ErtsGcQuickSanityCheck(p);
@@ -1213,7 +1208,7 @@ do_minor(Process *p, Uint new_sz, Eterm* objv, int nobj)
  */
 
 static int
-major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int off_heap_msgs)
+major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl)
 {
     Rootset rootset;
     Roots* roots;
@@ -1226,7 +1221,7 @@ major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int of
     Uint oh_size = (char *) OLD_HTOP(p) - oh;
     Uint n;
     Uint new_sz;
-    Uint fragments = MBUF_SIZE(p) + combined_message_size(p, off_heap_msgs);
+    Uint fragments = MBUF_SIZE(p) + combined_message_size(p);
 
     size_before = fragments + (HEAP_TOP(p) - HEAP_START(p));
 
@@ -1436,7 +1431,7 @@ major_collection(Process* p, int need, Eterm* objv, int nobj, Uint *recl, int of
 
     ErtsGcQuickSanityCheck(p);
 
-    if (!off_heap_msgs) {
+    {
 	ErlMessage *msgp;
 	/*
 	 * Copy newly received message onto the end of the new heap.
@@ -1506,14 +1501,11 @@ adjust_after_fullsweep(Process *p, Uint size_before, int need, Eterm *objv, int
  * mbuf list.
  */
 static Uint
-combined_message_size(Process* p, int off_heap_msgs)
+combined_message_size(Process* p)
 {
     Uint sz;
     ErlMessage *msgp;
 
-    if (off_heap_msgs)
-	return 0;
-
     for (sz = 0, msgp = p->msg.first; msgp; msgp = msgp->next) {
 	if (msgp->data.attached)
 	    sz += erts_msg_attached_data_size(msgp);
@@ -2662,7 +2654,7 @@ reply_gc_info(void *vgcirp)
     if (rp_locks)
 	erts_smp_proc_unlock(rp, rp_locks);
 
-    erts_smp_proc_dec_refc(rp);
+    erts_proc_dec_refc(rp);
 
     if (erts_smp_atomic32_dec_read_nob(&gcirp->refc) == 0)
 	gcireq_free(vgcirp);
@@ -2686,7 +2678,7 @@ erts_gc_info_request(Process *c_p)
     erts_smp_atomic32_init_nob(&gcirp->refc,
 			       (erts_aint32_t) erts_no_schedulers);
 
-    erts_smp_proc_add_refc(c_p, (Sint32) erts_no_schedulers);
+    erts_proc_add_refc(c_p, (Sint) erts_no_schedulers);
 
 #ifdef ERTS_SMP
     if (erts_no_schedulers > 1)
diff --git a/erts/emulator/beam/erl_hl_timer.c b/erts/emulator/beam/erl_hl_timer.c
new file mode 100644
index 0000000000..cf92ded415
--- /dev/null
+++ b/erts/emulator/beam/erl_hl_timer.c
@@ -0,0 +1,2884 @@
+/*
+ * %CopyrightBegin%
+ * 
+ * Copyright Ericsson AB 2015. All Rights Reserved.
+ * 
+ * The contents of this file are subject to the Erlang Public License,
+ * Version 1.1, (the "License"); you may not use this file except in
+ * compliance with the License. You should have received a copy of the
+ * Erlang Public License along with this software. If not, it can be
+ * retrieved online at http://www.erlang.org/.
+ * 
+ * Software distributed under the License is distributed on an "AS IS"
+ * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
+ * the License for the specific language governing rights and limitations
+ * under the License.
+ * 
+ * %CopyrightEnd%
+ */
+
+/*
+ * Description: High level timers implementing BIF timers
+ *              as well as process and port timers.
+ *
+ * Author: 	Rickard Green
+ */
+
+#ifdef HAVE_CONFIG_H
+#  include "config.h"
+#endif
+
+#include "sys.h"
+#include "global.h"
+#include "bif.h"
+#include "erl_bif_unique.h"
+#define ERTS_WANT_TIMER_WHEEL_API
+#include "erl_time.h"
+#include "erl_hl_timer.h"
+
+#define ERTS_TMR_CHECK_CANCEL_ON_CREATE 0
+
+#if 0
+#  define ERTS_HLT_HARD_DEBUG
+#endif
+#if 0
+#  define ERTS_HLT_DEBUG
+#endif
+
+#if defined(ERTS_HLT_HARD_DEBUG) || defined(DEBUG)
+#  if defined(ERTS_HLT_HARD_DEBUG)
+#    undef ERTS_RBT_HARD_DEBUG
+#    define ERTS_RBT_HARD_DEBUG 1
+#  endif
+#  ifndef ERTS_HLT_DEBUG
+#    define ERTS_HLT_DEBUG 1
+#  endif
+#endif
+
+#undef ERTS_HLT_ASSERT
+#if defined(ERTS_HLT_DEBUG)
+#  define ERTS_HLT_ASSERT(E) ERTS_ASSERT(E)
+#  undef ERTS_RBT_DEBUG
+#  define ERTS_RBT_DEBUG
+#else
+#  define ERTS_HLT_ASSERT(E) ((void) 1)
+#endif
+
+#if defined(ERTS_HLT_HARD_DEBUG) && defined(__GNUC__)
+#warning "* * * * * * * * * * * * * * * * * *"
+#warning "* ERTS_HLT_HARD_DEBUG IS ENABLED! *"
+#warning "* * * * * * * * * * * * * * * * * *"
+#endif
+
+#ifdef ERTS_HLT_HARD_DEBUG
+#  define ERTS_HLT_HDBG_CHK_SRV(SRV) hdbg_chk_srv((SRV))
+static void hdbg_chk_srv(ErtsHLTimerService *srv);
+#else
+#  define ERTS_HLT_HDBG_CHK_SRV(SRV) ((void) 1)
+#endif
+
+#if ERTS_REF_NUMBERS != 3
+#error "ERTS_REF_NUMBERS changed. Update me..."
+#endif
+
+#define ERTS_BIF_TIMER_SHORT_TIME 5000
+
+#ifdef ERTS_SMP
+#  define ERTS_HLT_SMP_MEMBAR_LoadLoad_LoadStore \
+    ETHR_MEMBAR(ETHR_LoadLoad|ETHR_LoadStore)
+#else
+#  define ERTS_HLT_SMP_MEMBAR_LoadLoad_LoadStore
+#endif
+
+/* Bit 0 to 9 contains scheduler id (see mask below) */
+#define ERTS_TMR_ROFLG_HLT		(((Uint32) 1) << 10)
+#define ERTS_TMR_ROFLG_BIF_TMR		(((Uint32) 1) << 11)
+#define ERTS_TMR_ROFLG_ABIF_TMR		(((Uint32) 1) << 12)
+#define ERTS_TMR_ROFLG_PRE_ALC		(((Uint32) 1) << 13)
+#define ERTS_TMR_ROFLG_REG_NAME		(((Uint32) 1) << 14)
+#define ERTS_TMR_ROFLG_PROC		(((Uint32) 1) << 15)
+#define ERTS_TMR_ROFLG_PORT		(((Uint32) 1) << 16)
+
+#define ERTS_TMR_ROFLG_SID_MASK	\
+    (ERTS_TMR_ROFLG_HLT - (Uint32) 1)
+
+#define ERTS_TMR_STATE_ACTIVE		((erts_aint32_t) 0)
+#define ERTS_TMR_STATE_CANCELED		((erts_aint32_t) 1)
+#define ERTS_TMR_STATE_TIMED_OUT	((erts_aint32_t) 2)
+
+typedef struct ErtsHLTimer_ ErtsHLTimer;
+
+#define ERTS_HLT_PFLG_RED		(((UWord) 1) << 0)
+#define ERTS_HLT_PFLG_SAME_TIME		(((UWord) 1) << 1)
+
+#define ERTS_HLT_PFLGS_MASK \
+    (ERTS_HLT_PFLG_RED|ERTS_HLT_PFLG_SAME_TIME)
+
+#define ERTS_HLT_PFIELD_NOT_IN_TABLE	(~((UWord) 0))
+
+typedef struct {
+    UWord parent; /* parent pointer and flags... */
+    union {
+	struct {
+	    ErtsHLTimer *right;
+	    ErtsHLTimer *left;
+	} t;
+	struct {
+	    ErtsHLTimer *prev;
+	    ErtsHLTimer *next;
+	} l;
+    } u;
+    ErtsHLTimer *same_time;
+} ErtsHLTimerTimeTree;
+
+typedef struct {
+    UWord parent; /* parent pointer and flags... */
+    ErtsHLTimer *right;
+    ErtsHLTimer *left;
+} ErtsHLTimerTree;
+
+typedef struct {
+    Uint32 roflgs;
+    erts_smp_atomic32_t refc;
+    union {
+	erts_atomic_t next;
+    } u;
+} ErtsTmrHead;
+
+struct ErtsHLTimer_ {
+    ErtsTmrHead head; /* NEED to be first! */
+    union {
+	ErtsThrPrgrLaterOp cleanup;
+	ErtsHLTimerTimeTree tree;
+    } time;
+    ErtsMonotonicTime timeout;
+    union {
+	Process *proc;
+	Port *port;
+	Eterm name;
+    } receiver;
+
+#ifdef ERTS_HLT_HARD_DEBUG
+    int pending_timeout;
+#endif
+
+    erts_smp_atomic32_t state;
+
+    /* BIF timer only fields follow... */
+    struct {
+	Uint32 refn[ERTS_REF_NUMBERS];
+	ErtsHLTimerTree proc_tree;
+	ErtsHLTimerTree tree;
+	Eterm message;
+	ErlHeapFragment *bp;
+    } btm;
+    struct {
+	Eterm accessor;
+	ErtsHLTimerTree tree;
+    } abtm;
+};
+
+#define ERTS_HL_PTIMER_SIZE offsetof(ErtsHLTimer, btm)
+#define ERTS_BIF_TIMER_SIZE offsetof(ErtsHLTimer, abtm)
+#define ERTS_ABIF_TIMER_SIZE sizeof(ErtsHLTimer)
+
+typedef struct {
+    ErtsTmrHead head; /* NEED to be first! */
+    void *p;
+    ErtsTWheelTimer tw_tmr;
+} ErtsTWTimer;
+
+typedef union {
+    ErtsTmrHead head;
+    ErtsHLTimer hlt;
+    ErtsTWTimer twt;
+} ErtsTimer;
+
+#ifdef SMALL_MEMORY
+#define BIF_TIMER_PREALC_SZ	10
+#define PTIMER_PREALC_SZ	10
+#else
+#define BIF_TIMER_PREALC_SZ	100
+#define PTIMER_PREALC_SZ	100
+#endif
+
+ERTS_SCHED_PREF_PALLOC_IMPL(bif_timer_pre,
+			    ErtsHLTimer,
+			    BIF_TIMER_PREALC_SZ)
+
+ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(tw_timer,
+				 ErtsTWTimer,
+				 PTIMER_PREALC_SZ,
+				 ERTS_ALC_T_LL_PTIMER)
+
+#ifdef ERTS_HLT_DEBUG
+#define ERTS_TMR_TIMEOUT_YIELD_LIMIT 5
+#else
+#define ERTS_TMR_TIMEOUT_YIELD_LIMIT 100
+#endif
+#define ERTS_TMR_CANCELED_TIMER_LIMIT 100
+#define ERTS_TMR_CANCELED_TIMER_SMALL_LIMIT 5
+
+#define ERTS_TMR_TIMEOUT_YIELD_STATE_T same_time_list_yield_state_t
+#define ERTS_TMR_YIELDING_TIMEOUT_STATE_INITER {NULL, {0}}
+typedef struct {
+    int dummy;
+} ERTS_TMR_TIMEOUT_YIELD_STATE_T;
+
+typedef struct {
+    ErtsTmrHead marker;
+    erts_atomic_t last;
+} ErtsHLTCncldTmrQTail;
+
+#ifdef ERTS_SMP
+
+typedef struct {
+    /*
+     * This structure needs to be cache line aligned for best
+     * performance.
+     */
+    union {
+	/*
+	 * Modified by threads returning canceled
+	 * timers to this timer service.
+	 */
+	ErtsHLTCncldTmrQTail data;
+	char align__[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(
+		sizeof(ErtsHLTCncldTmrQTail))];
+    } tail;
+    /*
+     * Everything below this point is *only* accessed by the
+     * thread managing this timer service.
+     */
+    struct {
+	ErtsTimer *first;
+	ErtsTimer *unref_end;
+	struct {
+	    ErtsThrPrgrVal thr_progress;
+	    int thr_progress_reached;
+	    ErtsTimer *unref_end;
+	} next;
+	int used_marker;
+    } head;
+} ErtsHLTCncldTmrQ;
+
+#endif /* ERTS_SMP */
+
+typedef struct {
+    ErtsHLTimer *root;
+    ERTS_TMR_TIMEOUT_YIELD_STATE_T state;
+} ErtsYieldingTimeoutState;
+
+struct ErtsHLTimerService_ {
+#ifdef ERTS_SMP
+    ErtsHLTCncldTmrQ canceled_queue;
+#endif
+    ErtsHLTimer *time_tree;
+    ErtsHLTimer *btm_tree;
+    ErtsHLTimer *next_timeout;
+    ErtsYieldingTimeoutState yield;
+    ErtsTWheelTimer service_timer;
+};
+
+static ERTS_INLINE int
+refn_is_lt(Uint32 *x, Uint32 *y)
+{
+    /* !0 if x < y */
+    if (x[2] < y[2])
+	return 1;
+    if (x[2] != y[2])
+	return 0;
+    if (x[1] < y[1])
+	return 1;
+    if (x[1] != y[1])
+	return 0;
+    return x[0] < y[0];
+}
+
+#define ERTS_RBT_PREFIX time
+#define ERTS_RBT_T ErtsHLTimer
+#define ERTS_RBT_KEY_T ErtsMonotonicTime
+#define ERTS_RBT_FLAGS_T UWord
+#define ERTS_RBT_INIT_EMPTY_TNODE(T)					\
+    do {								\
+	(T)->time.tree.parent = (UWord) NULL;				\
+	(T)->time.tree.u.t.right = NULL;				\
+	(T)->time.tree.u.t.left = NULL;					\
+    } while (0)
+#define ERTS_RBT_IS_RED(T)						\
+    ((int) ((T)->time.tree.parent & ERTS_HLT_PFLG_RED))
+#define ERTS_RBT_SET_RED(T)						\
+    ((T)->time.tree.parent |= ERTS_HLT_PFLG_RED)
+#define ERTS_RBT_IS_BLACK(T)						\
+    (!ERTS_RBT_IS_RED((T)))
+#define ERTS_RBT_SET_BLACK(T)						\
+    ((T)->time.tree.parent &= ~ERTS_HLT_PFLG_RED)
+#define ERTS_RBT_GET_FLAGS(T)						\
+    ((T)->time.tree.parent & ERTS_HLT_PFLGS_MASK)
+#define ERTS_RBT_SET_FLAGS(T, F)					\
+    do {								\
+	ERTS_HLT_ASSERT((((UWord) (F)) & ~ERTS_HLT_PFLGS_MASK) == 0);	\
+	(T)->time.tree.parent &= ~ERTS_HLT_PFLGS_MASK;			\
+	(T)->time.tree.parent |= (F);					\
+    } while (0)
+#define ERTS_RBT_GET_PARENT(T)						\
+    ((ErtsHLTimer *) ((T)->time.tree.parent & ~ERTS_HLT_PFLGS_MASK))
+#define ERTS_RBT_SET_PARENT(T, P)					\
+    do {								\
+	ERTS_HLT_ASSERT((((UWord) (P)) & ERTS_HLT_PFLGS_MASK) == 0);	\
+	(T)->time.tree.parent &= ERTS_HLT_PFLGS_MASK;			\
+	(T)->time.tree.parent |= (UWord) (P);				\
+    } while (0)
+#define ERTS_RBT_GET_RIGHT(T) ((T)->time.tree.u.t.right)
+#define ERTS_RBT_SET_RIGHT(T, R) ((T)->time.tree.u.t.right = (R))
+#define ERTS_RBT_GET_LEFT(T) ((T)->time.tree.u.t.left)
+#define ERTS_RBT_SET_LEFT(T, L) ((T)->time.tree.u.t.left = (L))
+#define ERTS_RBT_GET_KEY(T) ((T)->timeout)
+#define ERTS_RBT_IS_LT(KX, KY) ((KX) < (KY))
+#define ERTS_RBT_IS_EQ(KX, KY) ((KX) == (KY))
+#define ERTS_RBT_WANT_DELETE
+#define ERTS_RBT_WANT_SMALLEST
+#define ERTS_RBT_WANT_LOOKUP_INSERT
+#define ERTS_RBT_WANT_REPLACE
+#ifdef ERTS_HLT_HARD_DEBUG
+#  define ERTS_RBT_WANT_FOREACH
+#  define ERTS_RBT_WANT_LOOKUP
+#endif
+#define ERTS_RBT_UNDEF
+
+#include "erl_rbtree.h"
+
+/* Use circular list for timers at same time */
+
+static ERTS_INLINE void
+same_time_list_insert(ErtsHLTimer **root, ErtsHLTimer *tmr)
+{
+    ErtsHLTimer *first = *root;
+    if (!first) {
+	ERTS_HLT_ASSERT((((UWord) root) & ERTS_HLT_PFLG_SAME_TIME) == 0);
+	tmr->time.tree.parent = ((UWord) root) | ERTS_HLT_PFLG_SAME_TIME;
+	tmr->time.tree.u.l.next = tmr;
+	tmr->time.tree.u.l.prev = tmr;
+	*root = tmr;
+    }
+    else {
+	tmr->time.tree.parent = ERTS_HLT_PFLG_SAME_TIME;
+	tmr->time.tree.u.l.next = first;
+	tmr->time.tree.u.l.prev = first->time.tree.u.l.prev;
+	first->time.tree.u.l.prev = tmr;
+	tmr->time.tree.u.l.prev->time.tree.u.l.next = tmr;
+    }
+}
+
+static ERTS_INLINE void
+same_time_list_delete(ErtsHLTimer *tmr)
+{
+    ErtsHLTimer **root, *next;
+    
+    root = (ErtsHLTimer **) (tmr->time.tree.parent & ~ERTS_HLT_PFLG_SAME_TIME);
+    next = tmr->time.tree.u.l.next;
+
+    ERTS_HLT_ASSERT((tmr->time.tree.parent
+		     == (((UWord) root) | ERTS_HLT_PFLG_SAME_TIME))
+		    || (tmr->time.tree.parent
+			== ERTS_HLT_PFLG_SAME_TIME));
+
+    if (next == tmr) {
+	ERTS_HLT_ASSERT(root && *root == tmr);
+	ERTS_HLT_ASSERT(tmr->time.tree.u.l.prev == tmr);
+	*root = NULL;
+    }
+    else {
+	if (root) {
+	    ERTS_HLT_ASSERT(*root == tmr);
+	    *root = next;
+	    next->time.tree.parent = ((UWord) root) | ERTS_HLT_PFLG_SAME_TIME;
+	}
+	tmr->time.tree.u.l.next->time.tree.u.l.prev = tmr->time.tree.u.l.prev;
+	tmr->time.tree.u.l.prev->time.tree.u.l.next = next;
+    }
+}
+
+static ERTS_INLINE void
+same_time_list_new_root(ErtsHLTimer **root)
+{
+    ErtsHLTimer *tmr = *root;
+    if (tmr) {
+	ERTS_HLT_ASSERT(root);
+	tmr->time.tree.parent = ((UWord) root) | ERTS_HLT_PFLG_SAME_TIME;
+    }
+}
+
+static ERTS_INLINE int
+same_time_list_foreach_destroy_yielding(ErtsHLTimer **root,
+					void (*op)(ErtsHLTimer *, void *),
+					void *arg,
+					ERTS_TMR_TIMEOUT_YIELD_STATE_T *ys,
+					Sint ylimit)
+{
+    Sint ycnt = ylimit;
+    ErtsHLTimer *end, *tmr = *root;
+    if (!tmr)
+	return 0;
+
+    ERTS_HLT_ASSERT(tmr->time.tree.parent
+		    == (((UWord) root) | ERTS_HLT_PFLG_SAME_TIME));
+
+    end = tmr->time.tree.u.l.prev;
+    end->time.tree.u.l.next = NULL;
+
+    while (1) {
+	ErtsHLTimer *op_tmr = tmr;
+
+	ERTS_HLT_ASSERT((tmr->time.tree.parent
+			 == (((UWord) root) | ERTS_HLT_PFLG_SAME_TIME))
+			|| (tmr->time.tree.parent
+			    == ERTS_HLT_PFLG_SAME_TIME));
+
+	tmr = tmr->time.tree.u.l.next;
+	(*op)(op_tmr, arg);
+	if (!tmr) {
+	    *root = NULL;
+	    return 0;
+	}
+	if (--ycnt <= 0) {
+	    /* Make new circle of timers left to process... */
+	    *root = tmr;
+	    end->time.tree.u.l.next = tmr;
+	    tmr->time.tree.u.l.prev = end;
+	    tmr->time.tree.parent = ((UWord) root) | ERTS_HLT_PFLG_SAME_TIME;
+	    return 1;
+	}
+    }    
+}
+
+#ifdef ERTS_HLT_HARD_DEBUG
+
+static ERTS_INLINE void
+same_time_list_foreach(ErtsHLTimer *root,
+		       void (*op)(ErtsHLTimer *, void *),
+		       void *arg)
+{
+    if (root) {
+	ErtsHLTimer *tmr = root;
+	do {
+	    (*op)(tmr, arg);
+	    tmr = tmr->time.tree.u.l.next;
+	} while (root != tmr);
+    }
+}
+
+static ERTS_INLINE ErtsHLTimer *
+same_time_list_lookup(ErtsHLTimer *root, ErtsHLTimer *x)
+{
+    if (root) {
+	ErtsHLTimer *tmr = root;
+	do {
+	    if (tmr == x)
+		return tmr;
+	    tmr = tmr->time.tree.u.l.next;
+	} while (root != tmr);
+    }
+    return NULL;
+}
+
+#endif /* ERTS_HLT_HARD_DEBUG */
+
+#define ERTS_RBT_PREFIX btm
+#define ERTS_RBT_T ErtsHLTimer
+#define ERTS_RBT_KEY_T Uint32 *
+#define ERTS_RBT_FLAGS_T UWord
+#define ERTS_RBT_INIT_EMPTY_TNODE(T)					\
+    do {								\
+	(T)->btm.tree.parent = (UWord) NULL;				\
+	(T)->btm.tree.right = NULL;					\
+	(T)->btm.tree.left = NULL;					\
+    } while (0)
+#define ERTS_RBT_IS_RED(T)						\
+    ((int) ((T)->btm.tree.parent & ERTS_HLT_PFLG_RED))
+#define ERTS_RBT_SET_RED(T)						\
+    ((T)->btm.tree.parent |= ERTS_HLT_PFLG_RED)
+#define ERTS_RBT_IS_BLACK(T)						\
+    (!ERTS_RBT_IS_RED((T)))
+#define ERTS_RBT_SET_BLACK(T)						\
+    ((T)->btm.tree.parent &= ~ERTS_HLT_PFLG_RED)
+#define ERTS_RBT_GET_FLAGS(T)						\
+    ((T)->btm.tree.parent & ERTS_HLT_PFLGS_MASK)
+#define ERTS_RBT_SET_FLAGS(T, F)					\
+    do {								\
+	ERTS_HLT_ASSERT((((UWord) (F)) & ~ERTS_HLT_PFLGS_MASK) == 0);	\
+	(T)->btm.tree.parent &= ~ERTS_HLT_PFLGS_MASK;			\
+	(T)->btm.tree.parent |= (F);					\
+    } while (0)
+#define ERTS_RBT_GET_PARENT(T)						\
+    ((ErtsHLTimer *) ((T)->btm.tree.parent & ~ERTS_HLT_PFLGS_MASK))
+#define ERTS_RBT_SET_PARENT(T, P)					\
+    do {								\
+	ERTS_HLT_ASSERT((((UWord) (P)) & ERTS_HLT_PFLGS_MASK) == 0);	\
+	(T)->btm.tree.parent &= ERTS_HLT_PFLGS_MASK;			\
+	(T)->btm.tree.parent |= (UWord) (P);				\
+    } while (0)
+#define ERTS_RBT_GET_RIGHT(T) ((T)->btm.tree.right)
+#define ERTS_RBT_SET_RIGHT(T, R) ((T)->btm.tree.right = (R))
+#define ERTS_RBT_GET_LEFT(T) ((T)->btm.tree.left)
+#define ERTS_RBT_SET_LEFT(T, L) ((T)->btm.tree.left = (L))
+#define ERTS_RBT_GET_KEY(T) ((T)->btm.refn)
+#define ERTS_RBT_IS_LT(KX, KY) refn_is_lt((KX), (KY))
+#define ERTS_RBT_IS_EQ(KX, KY) \
+    (((KX)[0] == (KY)[0]) & ((KX)[1] == (KY)[1]) & ((KX)[2] == (KY)[2]))
+#define ERTS_RBT_WANT_DELETE
+#define ERTS_RBT_WANT_INSERT
+#define ERTS_RBT_WANT_LOOKUP
+#define ERTS_RBT_WANT_FOREACH
+#define ERTS_RBT_UNDEF
+
+#include "erl_rbtree.h"
+
+#define ERTS_RBT_PREFIX proc_btm
+#define ERTS_RBT_T ErtsHLTimer
+#define ERTS_RBT_KEY_T Uint32 *
+#define ERTS_RBT_FLAGS_T UWord
+#define ERTS_RBT_INIT_EMPTY_TNODE(T)					\
+    do {								\
+	(T)->btm.proc_tree.parent = (UWord) NULL;			\
+	(T)->btm.proc_tree.right = NULL;				\
+	(T)->btm.proc_tree.left = NULL;					\
+    } while (0)
+#define ERTS_RBT_IS_RED(T)						\
+    ((int) ((T)->btm.proc_tree.parent & ERTS_HLT_PFLG_RED))
+#define ERTS_RBT_SET_RED(T)						\
+    ((T)->btm.proc_tree.parent |= ERTS_HLT_PFLG_RED)
+#define ERTS_RBT_IS_BLACK(T)						\
+    (!ERTS_RBT_IS_RED((T)))
+#define ERTS_RBT_SET_BLACK(T)						\
+    ((T)->btm.proc_tree.parent &= ~ERTS_HLT_PFLG_RED)
+#define ERTS_RBT_GET_FLAGS(T)						\
+    ((T)->btm.proc_tree.parent & ERTS_HLT_PFLGS_MASK)
+#define ERTS_RBT_SET_FLAGS(T, F)					\
+    do {								\
+	ERTS_HLT_ASSERT((((UWord) (F)) & ~ERTS_HLT_PFLGS_MASK) == 0);	\
+	(T)->btm.proc_tree.parent &= ~ERTS_HLT_PFLGS_MASK;		\
+	(T)->btm.proc_tree.parent |= (F);				\
+    } while (0)
+#define ERTS_RBT_GET_PARENT(T)						\
+    ((ErtsHLTimer *) ((T)->btm.proc_tree.parent & ~ERTS_HLT_PFLGS_MASK))
+#define ERTS_RBT_SET_PARENT(T, P)					\
+    do {								\
+	ERTS_HLT_ASSERT((((UWord) (P)) & ERTS_HLT_PFLGS_MASK) == 0);	\
+	(T)->btm.proc_tree.parent &= ERTS_HLT_PFLGS_MASK;		\
+	(T)->btm.proc_tree.parent |= (UWord) (P);			\
+    } while (0)
+#define ERTS_RBT_GET_RIGHT(T) ((T)->btm.proc_tree.right)
+#define ERTS_RBT_SET_RIGHT(T, R) ((T)->btm.proc_tree.right = (R))
+#define ERTS_RBT_GET_LEFT(T) ((T)->btm.proc_tree.left)
+#define ERTS_RBT_SET_LEFT(T, L) ((T)->btm.proc_tree.left = (L))
+#define ERTS_RBT_GET_KEY(T) ((T)->btm.refn)
+#define ERTS_RBT_IS_LT(KX, KY) refn_is_lt((KX), (KY))
+#define ERTS_RBT_IS_EQ(KX, KY) \
+    (((KX)[0] == (KY)[0]) & ((KX)[1] == (KY)[1]) & ((KX)[2] == (KY)[2]))
+#define ERTS_RBT_WANT_DELETE
+#define ERTS_RBT_WANT_INSERT
+#define ERTS_RBT_WANT_LOOKUP
+#define ERTS_RBT_WANT_FOREACH_DESTROY_YIELDING
+#define ERTS_RBT_UNDEF
+
+#include "erl_rbtree.h"
+
+#define ERTS_RBT_PREFIX abtm
+#define ERTS_RBT_T ErtsHLTimer
+#define ERTS_RBT_KEY_T Uint32 *
+#define ERTS_RBT_FLAGS_T UWord
+#define ERTS_RBT_INIT_EMPTY_TNODE(T)					\
+    do {								\
+	(T)->abtm.tree.parent = (UWord) NULL;				\
+	(T)->abtm.tree.right = NULL;					\
+	(T)->abtm.tree.left = NULL;					\
+    } while (0)
+#define ERTS_RBT_IS_RED(T)						\
+    ((int) ((T)->abtm.tree.parent & ERTS_HLT_PFLG_RED))
+#define ERTS_RBT_SET_RED(T)						\
+    ((T)->abtm.tree.parent |= ERTS_HLT_PFLG_RED)
+#define ERTS_RBT_IS_BLACK(T)						\
+    (!ERTS_RBT_IS_RED((T)))
+#define ERTS_RBT_SET_BLACK(T)						\
+    ((T)->abtm.tree.parent &= ~ERTS_HLT_PFLG_RED)
+#define ERTS_RBT_GET_FLAGS(T)						\
+    ((T)->abtm.tree.parent & ERTS_HLT_PFLGS_MASK)
+#define ERTS_RBT_SET_FLAGS(T, F)					\
+    do {								\
+	ERTS_HLT_ASSERT((((UWord) (F)) & ~ERTS_HLT_PFLGS_MASK) == 0);	\
+	(T)->abtm.tree.parent &= ~ERTS_HLT_PFLGS_MASK;			\
+	(T)->abtm.tree.parent |= (F);					\
+    } while (0)
+#define ERTS_RBT_GET_PARENT(T)						\
+    ((ErtsHLTimer *) ((T)->abtm.tree.parent & ~ERTS_HLT_PFLGS_MASK))
+#define ERTS_RBT_SET_PARENT(T, P)					\
+    do {								\
+	ERTS_HLT_ASSERT((((UWord) (P)) & ERTS_HLT_PFLGS_MASK) == 0);	\
+	(T)->abtm.tree.parent &= ERTS_HLT_PFLGS_MASK;			\
+	(T)->abtm.tree.parent |= (UWord) (P);				\
+    } while (0)
+#define ERTS_RBT_GET_RIGHT(T) ((T)->abtm.tree.right)
+#define ERTS_RBT_SET_RIGHT(T, R) ((T)->abtm.tree.right = (R))
+#define ERTS_RBT_GET_LEFT(T) ((T)->abtm.tree.left)
+#define ERTS_RBT_SET_LEFT(T, L) ((T)->abtm.tree.left = (L))
+#define ERTS_RBT_GET_KEY(T) ((T)->btm.refn)
+#define ERTS_RBT_IS_LT(KX, KY) refn_is_lt((KX), (KY))
+#define ERTS_RBT_IS_EQ(KX, KY) \
+    (((KX)[0] == (KY)[0]) & ((KX)[1] == (KY)[1]) & ((KX)[2] == (KY)[2]))
+#define ERTS_RBT_WANT_DELETE
+#define ERTS_RBT_WANT_INSERT
+#define ERTS_RBT_WANT_LOOKUP
+#define ERTS_RBT_WANT_FOREACH_DESTROY_YIELDING
+#define ERTS_RBT_UNDEF
+
+#include "erl_rbtree.h"
+
+#ifdef ERTS_SMP
+static void init_canceled_queue(ErtsHLTCncldTmrQ *cq);
+#endif
+
+void
+erts_hl_timer_init(void)
+{
+    init_tw_timer_alloc();
+    init_bif_timer_pre_alloc();
+}
+
+ErtsHLTimerService *
+erts_create_timer_service(void)
+{
+    ErtsYieldingTimeoutState init_yield = ERTS_TMR_YIELDING_TIMEOUT_STATE_INITER;
+    ErtsHLTimerService *srv;
+
+    srv = erts_alloc_permanent_cache_aligned(ERTS_ALC_T_TIMER_SERVICE,
+					     sizeof(ErtsHLTimerService));
+    srv->time_tree = NULL;
+    srv->btm_tree = NULL;
+    srv->next_timeout = NULL;
+    srv->yield = init_yield;
+    erts_twheel_init_timer(&srv->service_timer);
+
+#ifdef ERTS_SMP
+    init_canceled_queue(&srv->canceled_queue);
+#endif
+
+    return srv;
+}
+
+size_t
+erts_timer_type_size(ErtsAlcType_t type)
+{
+    switch (type) {
+    case ERTS_ALC_T_LL_PTIMER: return sizeof(ErtsTWTimer);
+    case ERTS_ALC_T_HL_PTIMER: return ERTS_HL_PTIMER_SIZE;
+    case ERTS_ALC_T_BIF_TIMER: return ERTS_BIF_TIMER_SIZE;
+    case ERTS_ALC_T_ABIF_TIMER: return ERTS_ABIF_TIMER_SIZE;
+    default: ERTS_INTERNAL_ERROR("Unknown type");
+    }
+    return 0;
+}
+
+static ERTS_INLINE ErtsMonotonicTime
+get_timeout_pos(ErtsMonotonicTime now, ErtsMonotonicTime msec)
+{
+    ErtsMonotonicTime timeout_pos;
+    if (msec <= 0)
+	return ERTS_MONOTONIC_TO_CLKTCKS(now);
+    timeout_pos = ERTS_MONOTONIC_TO_CLKTCKS(now-1);
+    timeout_pos += ERTS_MSEC_TO_CLKTCKS(msec) + 1;
+    return timeout_pos;
+}
+
+static ERTS_INLINE Sint64
+get_time_left(ErtsSchedulerData *esdp, ErtsMonotonicTime timeout_pos)
+{
+    ErtsMonotonicTime now = erts_get_monotonic_time(esdp);
+
+    now = ERTS_MONOTONIC_TO_CLKTCKS(now-1)+1;
+    if (timeout_pos <= now)
+	return (Sint64) 0;
+    return (Sint64) ERTS_CLKTCKS_TO_MSEC(timeout_pos - now);
+}
+
+static ERTS_INLINE int
+proc_timeout_common(Process *proc, void *tmr)
+{
+    if (tmr == (void *) erts_smp_atomic_cmpxchg_mb(&proc->common.timer,
+						   ERTS_PTMR_TIMEDOUT,
+						   (erts_aint_t) tmr)) {
+	erts_aint32_t state = erts_smp_atomic32_read_acqb(&proc->state);
+	if (!(state & (ERTS_PSFLG_ACTIVE|ERTS_PSFLG_EXITING)))
+	    erts_schedule_process(proc, state, 0);
+	return 1;
+    }
+    return 0;
+}
+
+static ERTS_INLINE int
+port_timeout_common(Port *port, void *tmr)
+{
+    if (tmr == (void *) erts_smp_atomic_cmpxchg_mb(&port->common.timer,
+						   ERTS_PTMR_TIMEDOUT,
+						   (erts_aint_t) tmr)) {
+	erts_port_task_schedule(port->common.id,
+				&port->timeout_task,
+				ERTS_PORT_TASK_TIMEOUT);
+	return 1;
+    }
+    return 0;
+}
+
+/*
+ * Basic timer wheel timer stuff
+ */
+
+static void
+scheduled_tw_timer_destroy(void *vtmr)
+{
+    tw_timer_free((ErtsTWTimer *) vtmr);
+}
+
+static void
+schedule_tw_timer_destroy(ErtsTWTimer *tmr)
+{
+    /*
+     * Reference to process/port can be
+     * dropped at once...
+     */
+    if (tmr->head.roflgs & ERTS_TMR_ROFLG_PROC)
+	erts_proc_dec_refc((Process *) tmr->p);
+    else
+	erts_port_dec_refc((Port *) tmr->p);
+
+    erts_schedule_thr_prgr_later_cleanup_op(
+	scheduled_tw_timer_destroy,
+	(void *) tmr,
+	&tmr->tw_tmr.u.cleanup,
+	sizeof(ErtsTWTimer));
+}
+
+static ERTS_INLINE void
+tw_timer_dec_refc(ErtsTWTimer *tmr)
+{
+    if (erts_smp_atomic32_dec_read_relb(&tmr->head.refc) == 0) {
+	ERTS_HLT_SMP_MEMBAR_LoadLoad_LoadStore;
+	schedule_tw_timer_destroy(tmr);
+    }
+}
+
+static void
+tw_proc_timeout(void *vtwtp)
+{
+    ErtsTWTimer *twtp = (ErtsTWTimer *) vtwtp;
+    Process *proc = (Process *) twtp->p;
+    if (proc_timeout_common(proc, vtwtp))
+	tw_timer_dec_refc(twtp);
+    tw_timer_dec_refc(twtp);
+}
+
+static void
+tw_port_timeout(void *vtwtp)
+{
+    ErtsTWTimer *twtp = (ErtsTWTimer *) vtwtp;
+    Port *port = (Port *) twtp->p;
+    if (port_timeout_common(port, vtwtp))
+	tw_timer_dec_refc(twtp);
+    tw_timer_dec_refc(twtp);
+}
+
+static void
+tw_ptimer_cancel(void *vtwtp)
+{
+    tw_timer_dec_refc((ErtsTWTimer *) vtwtp);    
+}
+
+static void
+cancel_tw_timer(ErtsSchedulerData *esdp, ErtsTWTimer *tmr)
+{
+    ERTS_HLT_ASSERT((tmr->head.roflgs & ERTS_TMR_ROFLG_SID_MASK)
+		    == (Uint32) esdp->no);
+    erts_twheel_cancel_timer(esdp->timer_wheel, &tmr->tw_tmr);
+}
+
+static ErtsTWTimer *
+create_tw_timer(ErtsSchedulerData *esdp, 
+		void *p, int is_proc,
+		ErtsMonotonicTime timeout_pos)
+{
+    ErtsTWTimer *tmr;
+    void (*timeout_func)(void *);
+
+    tmr = tw_timer_alloc();
+    erts_twheel_init_timer(&tmr->tw_tmr);
+
+    tmr->head.roflgs = (Uint32) esdp->no;
+    ERTS_HLT_ASSERT((tmr->head.roflgs
+		     & ~ERTS_TMR_ROFLG_SID_MASK) == 0);
+    tmr->p = p;
+    if (is_proc) {
+	tmr->head.roflgs |= ERTS_TMR_ROFLG_PROC;
+	timeout_func = tw_proc_timeout;
+	erts_proc_inc_refc((Process *) p);
+    }
+    else {
+	tmr->head.roflgs |= ERTS_TMR_ROFLG_PORT;
+	timeout_func = tw_port_timeout;
+	erts_port_inc_refc((Port *) p);
+    }
+
+    erts_smp_atomic32_init_nob(&tmr->head.refc, 2);
+
+    erts_twheel_set_timer(esdp->timer_wheel,
+			  &tmr->tw_tmr,
+			  timeout_func,
+			  tw_ptimer_cancel,
+			  tmr,
+			  timeout_pos);
+
+    return tmr;
+}
+
+/*
+ * Basic high level timer stuff
+ */
+
+static ERTS_INLINE void
+hl_timer_destroy(ErtsHLTimer *tmr)
+{
+    Uint32 roflgs = tmr->head.roflgs;
+    if (!(roflgs & ERTS_TMR_ROFLG_BIF_TMR))
+	erts_free(ERTS_ALC_T_HL_PTIMER, tmr);
+    else {
+	if (tmr->btm.bp)
+	    free_message_buffer(tmr->btm.bp);
+	if (roflgs & ERTS_TMR_ROFLG_PRE_ALC)
+	    bif_timer_pre_free(tmr);
+	else if (roflgs & ERTS_TMR_ROFLG_ABIF_TMR)
+	    erts_free(ERTS_ALC_T_ABIF_TIMER, tmr);
+	else
+	    erts_free(ERTS_ALC_T_BIF_TIMER, tmr);
+    }
+}
+
+static void
+scheduled_hl_timer_destroy(void *vtmr)
+{
+    hl_timer_destroy((ErtsHLTimer *) vtmr);
+}
+
+static void
+schedule_hl_timer_destroy(ErtsHLTimer *tmr, Uint32 roflgs)
+{
+    UWord size;
+
+    /*
+     * Reference to process/port can be dropped
+     * at once...
+     */
+
+    ERTS_HLT_ASSERT(erts_smp_atomic32_read_nob(&tmr->head.refc) == 0);
+
+    if (roflgs & ERTS_TMR_ROFLG_REG_NAME) {
+	ERTS_HLT_ASSERT(is_atom(tmr->receiver.name));
+    }
+    else if (roflgs & ERTS_TMR_ROFLG_PROC) {
+	ERTS_HLT_ASSERT(tmr->receiver.proc);
+	erts_proc_dec_refc(tmr->receiver.proc);
+    }
+    else if (roflgs & ERTS_TMR_ROFLG_PORT) {
+	ERTS_HLT_ASSERT(tmr->receiver.port);
+	erts_port_dec_refc(tmr->receiver.port);
+    }
+
+    if (!(roflgs & ERTS_TMR_ROFLG_BIF_TMR))
+	size = ERTS_HL_PTIMER_SIZE;
+    else {
+	/*
+	 * Message buffer can be dropped at
+	 * once...
+	 */
+	if (tmr->btm.bp) {
+	    free_message_buffer(tmr->btm.bp);
+	    tmr->btm.bp = NULL;
+	}
+	size = sizeof(ErtsHLTimer);
+    }
+
+    erts_schedule_thr_prgr_later_cleanup_op(
+	scheduled_hl_timer_destroy, tmr,
+	&tmr->time.cleanup, size);
+}
+
+static ERTS_INLINE void
+hl_timer_pre_dec_refc(ErtsHLTimer *tmr)
+{
+#ifdef ERTS_HLT_DEBUG
+    erts_aint_t refc;
+    refc = erts_smp_atomic32_dec_read_nob(&tmr->head.refc);
+    ERTS_HLT_ASSERT(refc > 0);
+#else
+    erts_smp_atomic32_dec_nob(&tmr->head.refc);
+#endif
+}
+
+static ERTS_INLINE void
+hl_timer_dec_refc(ErtsHLTimer *tmr, Uint32 roflgs)
+{
+    if (erts_smp_atomic32_dec_read_relb(&tmr->head.refc) == 0) {
+	ERTS_HLT_SMP_MEMBAR_LoadLoad_LoadStore;
+	schedule_hl_timer_destroy(tmr, roflgs);
+    }
+}
+
+static void hlt_service_timeout(void *vesdp);
+#ifdef ERTS_SMP
+static void handle_canceled_queue(ErtsSchedulerData *esdp,
+				  ErtsHLTCncldTmrQ *cq,
+				  int use_limit,
+				  int ops_limit,
+				  int *need_thr_progress,
+				  ErtsThrPrgrVal *thr_prgr_p,
+				  int *need_more_work);
+#endif
+
+static ERTS_INLINE void
+check_canceled_queue(ErtsSchedulerData *esdp, ErtsHLTimerService *srv)
+{
+#if defined(ERTS_SMP) && ERTS_TMR_CHECK_CANCEL_ON_CREATE
+    ErtsHLTCncldTmrQ *cq = &srv->canceled_queue;
+    if (cq->head.first != cq->head.unref_end)
+	handle_canceled_queue(esdp, cq, 1,
+			      ERTS_TMR_CANCELED_TIMER_SMALL_LIMIT,
+			      NULL, NULL, NULL);
+#endif
+}
+
+static void
+hlt_delete_abtm(ErtsHLTimer *tmr)
+{
+    Process *proc;
+
+    ERTS_HLT_ASSERT(tmr->head.roflgs & ERTS_TMR_ROFLG_ABIF_TMR);
+
+    proc = erts_proc_lookup(tmr->abtm.accessor);
+
+    if (proc) {
+	int deref = 0;
+	erts_smp_proc_lock(proc, ERTS_PROC_LOCK_BTM);
+	if (tmr->abtm.tree.parent != ERTS_HLT_PFIELD_NOT_IN_TABLE) {
+	    abtm_rbt_delete(&proc->accessor_bif_timers, tmr);
+	    deref = 1;
+	    tmr->abtm.tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+	}
+	erts_smp_proc_unlock(proc, ERTS_PROC_LOCK_BTM);
+	if (deref)
+	    hl_timer_pre_dec_refc(tmr);
+    }
+}
+
+static ErtsHLTimer *
+create_hl_timer(ErtsSchedulerData *esdp,
+		ErtsMonotonicTime timeout_pos,
+		int short_time, int is_bif_tmr,
+		void *rcvrp, Eterm rcvr, Eterm acsr,
+		Eterm msg, Uint32 *refn)
+{
+    ErtsHLTimerService *srv = esdp->timer_service;
+    ErtsHLTimer *tmr, *st_tmr;
+    erts_aint32_t refc;
+    Uint32 roflgs;
+    int is_abif_tmr = is_bif_tmr && is_value(acsr) && acsr != rcvr;
+
+    check_canceled_queue(esdp, srv);
+
+    ERTS_HLT_ASSERT((esdp->no & ~ERTS_TMR_ROFLG_SID_MASK) == 0);
+
+    roflgs = ((Uint32) esdp->no) | ERTS_TMR_ROFLG_HLT;
+
+    if (!is_bif_tmr)
+	tmr = erts_alloc(ERTS_ALC_T_HL_PTIMER,
+			 ERTS_HL_PTIMER_SIZE);
+    else if (short_time) {
+	tmr = bif_timer_pre_alloc();
+	if (!tmr)
+	    goto alloc_bif_timer;
+	roflgs |= ERTS_TMR_ROFLG_PRE_ALC;
+    }
+    else {
+    alloc_bif_timer:
+	if (is_abif_tmr)
+	    tmr = erts_alloc(ERTS_ALC_T_ABIF_TIMER,
+			     ERTS_ABIF_TIMER_SIZE);
+	else
+	    tmr = erts_alloc(ERTS_ALC_T_BIF_TIMER,
+			     ERTS_BIF_TIMER_SIZE);
+    }
+
+    tmr->timeout = timeout_pos;
+
+    if (!is_bif_tmr) {
+	if (is_internal_pid(rcvr)) {
+	    erts_proc_inc_refc((Process *) rcvrp);
+	    tmr->receiver.proc = (Process *) rcvrp;
+	    roflgs |= ERTS_TMR_ROFLG_PROC;
+	}
+	else {
+	    erts_port_inc_refc((Port *) rcvrp);
+	    ERTS_HLT_ASSERT(is_internal_port(rcvr));
+	    tmr->receiver.port = (Port *) rcvrp;
+	    roflgs |= ERTS_TMR_ROFLG_PORT;
+	}
+	refc = 2;
+    }
+    else {
+	Uint hsz;
+
+	roflgs |= ERTS_TMR_ROFLG_BIF_TMR;
+	if (is_internal_pid(rcvr)) {
+	    roflgs |= ERTS_TMR_ROFLG_PROC;
+	    tmr->receiver.proc = (Process *) rcvrp;
+	    refc = 2;
+	}
+	else {
+	    ERTS_HLT_ASSERT(is_atom(rcvr));
+	    roflgs |= ERTS_TMR_ROFLG_REG_NAME;
+	    tmr->receiver.name = rcvr;
+	    refc = 1;
+	}
+
+	hsz = is_immed(msg) ? ((Uint) 0) : size_object(msg);
+	if (!hsz) {
+	    tmr->btm.message = msg;
+	    tmr->btm.bp = NULL;
+	}
+	else {
+	    ErlHeapFragment *bp = new_message_buffer(hsz);
+	    Eterm *hp = bp->mem;
+	    tmr->btm.message = copy_struct(msg, hsz, &hp, &bp->off_heap);
+	    tmr->btm.bp = bp;
+	}
+	tmr->btm.refn[0] = refn[0];
+	tmr->btm.refn[1] = refn[1];
+	tmr->btm.refn[2] = refn[2];
+
+	tmr->btm.proc_tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+	if (is_abif_tmr) {
+	    Process *aproc;
+	    roflgs |= ERTS_TMR_ROFLG_ABIF_TMR;
+	    tmr->abtm.accessor = acsr;
+	    aproc = erts_proc_lookup(acsr);
+	    if (!aproc)
+		tmr->abtm.tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+	    else {
+		refc++;
+		erts_smp_proc_lock(aproc, ERTS_PROC_LOCK_BTM);
+		abtm_rbt_insert(&aproc->accessor_bif_timers, tmr);
+		erts_smp_proc_unlock(aproc, ERTS_PROC_LOCK_BTM);
+	    }
+	}
+    }
+
+    tmr->head.roflgs = roflgs;
+    erts_smp_atomic32_init_nob(&tmr->head.refc, refc);
+    erts_smp_atomic32_init_nob(&tmr->state, ERTS_TMR_STATE_ACTIVE);
+
+    ERTS_HLT_HDBG_CHK_SRV(srv);
+
+    if (!srv->next_timeout
+	|| tmr->timeout < srv->next_timeout->timeout) {
+	if (srv->next_timeout)
+	    erts_twheel_cancel_timer(esdp->timer_wheel,
+				     &srv->service_timer);
+	erts_twheel_set_timer(esdp->timer_wheel,
+			      &srv->service_timer,
+			      hlt_service_timeout,
+			      NULL,
+			      (void *) esdp,
+			      tmr->timeout);
+	srv->next_timeout = tmr;
+    }
+
+    st_tmr = time_rbt_lookup_insert(&srv->time_tree, tmr);
+    tmr->time.tree.same_time = st_tmr;
+    if (st_tmr)
+	same_time_list_insert(&st_tmr->time.tree.same_time, tmr);
+
+    if (is_bif_tmr)
+	btm_rbt_insert(&srv->btm_tree, tmr);
+
+#ifdef ERTS_HLT_HARD_DEBUG
+    tmr->pending_timeout = 0;
+#endif
+
+    ERTS_HLT_HDBG_CHK_SRV(srv);
+
+    return tmr;
+}
+
+static ERTS_INLINE void
+hlt_bif_timer_timeout(ErtsHLTimer *tmr, Uint32 roflgs)
+{
+    ErtsProcLocks proc_locks = ERTS_PROC_LOCKS_MSG_SEND;
+    Process *proc;
+    int dec_refc = 0;
+    Uint32 is_reg_name = (roflgs & ERTS_TMR_ROFLG_REG_NAME);
+    ERTS_HLT_ASSERT(roflgs & ERTS_TMR_ROFLG_BIF_TMR);
+
+    if (tmr->head.roflgs & ERTS_TMR_ROFLG_ABIF_TMR)
+	hlt_delete_abtm(tmr);
+
+    if (is_reg_name) {
+	Eterm pid;
+	ERTS_HLT_ASSERT(is_atom(tmr->receiver.name));
+	pid = erts_whereis_name_to_id(NULL, tmr->receiver.name);
+	proc = erts_proc_lookup(pid);
+    }
+    else {
+	ERTS_HLT_ASSERT(roflgs & ERTS_TMR_ROFLG_PROC);
+	ERTS_HLT_ASSERT(tmr->receiver.proc);
+
+	proc = tmr->receiver.proc;
+	proc_locks |= ERTS_PROC_LOCK_BTM;
+    }
+    if (proc) {
+	erts_smp_proc_lock(proc, proc_locks);
+	/*
+	 * If process is exiting, let it clean up
+	 * the btm tree by itself (it may be in
+	 * the middle of tree destruction).
+	 */
+	if (!ERTS_PROC_IS_EXITING(proc)) {
+	    erts_queue_message(proc, &proc_locks, tmr->btm.bp,
+			       tmr->btm.message, NIL);
+	    erts_smp_proc_unlock(proc, ERTS_PROC_LOCKS_MSG_SEND);
+	    proc_locks &= ~ERTS_PROC_LOCKS_MSG_SEND;
+	    tmr->btm.bp = NULL;
+	    if (tmr->btm.proc_tree.parent != ERTS_HLT_PFIELD_NOT_IN_TABLE) {
+		proc_btm_rbt_delete(&proc->bif_timers, tmr);
+		tmr->btm.proc_tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+		dec_refc = 1;
+	    }
+	}
+	if (proc_locks)
+	    erts_smp_proc_unlock(proc, proc_locks);
+	if (dec_refc)
+	    hl_timer_pre_dec_refc(tmr);
+    }
+}
+
+static ERTS_INLINE void
+hlt_proc_timeout(ErtsHLTimer *tmr)
+{
+    if (proc_timeout_common(tmr->receiver.proc, (void *) tmr))
+	hl_timer_dec_refc(tmr, tmr->head.roflgs);
+}
+
+static ERTS_INLINE void
+hlt_port_timeout(ErtsHLTimer *tmr)
+{
+    if (port_timeout_common(tmr->receiver.port, (void *) tmr))
+	hl_timer_dec_refc(tmr, tmr->head.roflgs);
+}
+
+static void hlt_timeout(ErtsHLTimer *tmr, void *vsrv)
+{
+    ErtsHLTimerService *srv = (ErtsHLTimerService *) vsrv;
+    Uint32 roflgs;
+    erts_aint32_t state;
+
+    ERTS_HLT_HDBG_CHK_SRV(srv);
+
+    roflgs = tmr->head.roflgs;
+    ERTS_HLT_ASSERT(roflgs & ERTS_TMR_ROFLG_HLT);
+
+    state = erts_smp_atomic32_cmpxchg_acqb(&tmr->state,
+					   ERTS_TMR_STATE_TIMED_OUT,
+					   ERTS_TMR_STATE_ACTIVE);
+
+    ERTS_HLT_ASSERT(state == ERTS_TMR_STATE_CANCELED
+		    || state == ERTS_TMR_STATE_ACTIVE);
+
+    if (state == ERTS_TMR_STATE_ACTIVE) {
+
+	if (roflgs & ERTS_TMR_ROFLG_BIF_TMR)
+	    hlt_bif_timer_timeout(tmr, roflgs);
+	else if (roflgs & ERTS_TMR_ROFLG_PROC)
+	    hlt_proc_timeout(tmr);
+	else {
+	    ERTS_HLT_ASSERT(roflgs & ERTS_TMR_ROFLG_PORT);
+	    hlt_port_timeout(tmr);
+	}
+    }
+
+    tmr->time.tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+    if ((roflgs & ERTS_TMR_ROFLG_BIF_TMR)
+	&& tmr->btm.tree.parent != ERTS_HLT_PFIELD_NOT_IN_TABLE) {
+	btm_rbt_delete(&srv->btm_tree, tmr);
+	tmr->btm.tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+    }
+
+    ERTS_HLT_HDBG_CHK_SRV(srv);
+
+    hl_timer_dec_refc(tmr, roflgs);
+}
+
+#ifdef ERTS_HLT_HARD_DEBUG
+static void
+set_pending_timeout(ErtsHLTimer *tmr, void *unused)
+{
+    tmr->pending_timeout = -1;
+}
+#endif
+
+static void
+hlt_service_timeout(void *vesdp)
+{
+    ErtsSchedulerData *esdp = (ErtsSchedulerData *) vesdp;
+    ErtsHLTimerService *srv = esdp->timer_service;
+    ErtsHLTimer *tmr = srv->next_timeout;
+    int yield;
+
+    ERTS_HLT_HDBG_CHK_SRV(srv);
+
+    ERTS_HLT_ASSERT(esdp == erts_get_scheduler_data());
+
+    ERTS_HLT_ASSERT(!srv->yield.root || srv->yield.root == tmr);
+    ERTS_HLT_ASSERT(tmr);
+    ERTS_HLT_ASSERT(tmr->timeout <= erts_get_monotonic_time(esdp));
+
+    if (!srv->yield.root) {
+	ERTS_HLT_ASSERT(tmr->time.tree.parent
+			!= ERTS_HLT_PFIELD_NOT_IN_TABLE);
+	time_rbt_delete(&srv->time_tree, tmr);
+	tmr->time.tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+#ifdef ERTS_HLT_HARD_DEBUG
+	tmr->pending_timeout = 1;
+	if (tmr->time.tree.same_time)
+	    same_time_list_foreach(tmr->time.tree.same_time, set_pending_timeout, NULL);
+#endif
+    }
+
+    if (!tmr->time.tree.same_time && !srv->yield.root)
+	yield = 0;
+    else {
+	yield = same_time_list_foreach_destroy_yielding(
+	    &tmr->time.tree.same_time, hlt_timeout, (void *) srv,
+	    &srv->yield.state, ERTS_TMR_TIMEOUT_YIELD_LIMIT);
+    }
+
+    if (yield)
+	srv->yield.root = tmr;
+    else {
+	srv->yield.root = NULL;
+	hlt_timeout(tmr, (void *) srv);
+
+	tmr = time_rbt_smallest(srv->time_tree);
+	srv->next_timeout = tmr;
+    }
+
+    ERTS_HLT_HDBG_CHK_SRV(srv);
+
+    if (tmr)
+	erts_twheel_set_timer(esdp->timer_wheel,
+			      &srv->service_timer,
+			      hlt_service_timeout,
+			      NULL,
+			      vesdp,
+			      tmr->timeout);
+}
+
+static void
+hlt_delete_timer(ErtsSchedulerData *esdp, ErtsHLTimer *tmr)
+{
+    ErtsHLTimerService *srv = esdp->timer_service;
+
+    ERTS_HLT_HDBG_CHK_SRV(srv);
+
+    if (tmr->head.roflgs & ERTS_TMR_ROFLG_BIF_TMR) {
+
+	if (tmr->btm.tree.parent != ERTS_HLT_PFIELD_NOT_IN_TABLE) {
+	    btm_rbt_delete(&srv->btm_tree, tmr);
+	    tmr->btm.tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+	}
+
+	if (tmr->head.roflgs & ERTS_TMR_ROFLG_ABIF_TMR)
+	    hlt_delete_abtm(tmr);
+    }
+
+    if (tmr->time.tree.parent == ERTS_HLT_PFIELD_NOT_IN_TABLE) {
+	/* Already removed... */
+	ERTS_HLT_HDBG_CHK_SRV(srv);
+	return;
+    }
+
+    if (tmr->time.tree.parent & ERTS_HLT_PFLG_SAME_TIME) {
+	same_time_list_delete(tmr);
+    }
+    else if (tmr->time.tree.same_time) {
+	ErtsHLTimer *st_container;
+
+	ERTS_HLT_ASSERT((tmr->time.tree.parent & ERTS_HLT_PFLG_SAME_TIME) == 0);
+	st_container = tmr->time.tree.same_time->time.tree.u.l.prev;
+
+	ERTS_HLT_ASSERT(st_container);
+	ERTS_HLT_ASSERT(st_container->time.tree.parent
+			 & ERTS_HLT_PFLG_SAME_TIME);
+	ERTS_HLT_ASSERT(tmr->timeout == st_container->timeout);
+
+	same_time_list_delete(st_container);
+	st_container->time.tree.same_time = tmr->time.tree.same_time;
+	same_time_list_new_root(&st_container->time.tree.same_time);
+
+	time_rbt_replace(&srv->time_tree, tmr, st_container);
+	ERTS_HLT_ASSERT((st_container->time.tree.parent
+			 & ERTS_HLT_PFLG_SAME_TIME) == 0);
+
+	if (srv->next_timeout == tmr)
+	    srv->next_timeout = st_container;
+    }
+    else {
+	ERTS_HLT_ASSERT((tmr->time.tree.parent & ERTS_HLT_PFLG_SAME_TIME) == 0);
+	time_rbt_delete(&srv->time_tree, tmr);
+	if (tmr == srv->next_timeout) {
+	    ErtsHLTimer *smlst;
+	    erts_twheel_cancel_timer(esdp->timer_wheel,
+				     &srv->service_timer);
+	    smlst = time_rbt_smallest(srv->time_tree);
+	    srv->next_timeout = smlst;
+	    if (smlst) {
+		ERTS_HLT_ASSERT(smlst->timeout > tmr->timeout);
+		erts_twheel_set_timer(esdp->timer_wheel,
+				      &srv->service_timer,
+				      hlt_service_timeout,
+				      NULL,
+				      (void *) esdp,
+				      smlst->timeout);
+	    }
+	}
+    }
+    tmr->time.tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+
+    hl_timer_dec_refc(tmr, tmr->head.roflgs);
+
+    ERTS_HLT_HDBG_CHK_SRV(srv);
+}
+
+/*
+ * Pass canceled timers back to originating scheduler
+ */
+
+static ERTS_INLINE void
+cleanup_sched_local_canceled_timer(ErtsSchedulerData *esdp,
+				   ErtsTimer *tmr)
+{
+    Uint32 roflgs = tmr->head.roflgs;
+    ERTS_HLT_ASSERT(esdp == erts_get_scheduler_data());
+    ERTS_HLT_ASSERT((tmr->head.roflgs & ERTS_TMR_ROFLG_SID_MASK)
+		    == (Uint32) esdp->no);
+    if (roflgs & ERTS_TMR_ROFLG_HLT) {
+	hlt_delete_timer(esdp, &tmr->hlt);
+	hl_timer_dec_refc(&tmr->hlt, roflgs);
+    }
+    else {
+	cancel_tw_timer(esdp, &tmr->twt);
+	tw_timer_dec_refc(&tmr->twt);
+    }
+}
+
+#ifdef ERTS_SMP
+
+static void
+init_canceled_queue(ErtsHLTCncldTmrQ *cq)
+{
+    erts_atomic_init_nob(&cq->tail.data.marker.u.next, ERTS_AINT_NULL);
+    erts_atomic_init_nob(&cq->tail.data.last,
+			 (erts_aint_t) &cq->tail.data.marker);
+    cq->head.first = (ErtsTimer *) &cq->tail.data.marker;
+    cq->head.unref_end = (ErtsTimer *) &cq->tail.data.marker;
+    cq->head.next.thr_progress = erts_thr_progress_current();
+    cq->head.next.thr_progress_reached = 1;
+    cq->head.next.unref_end = (ErtsTimer *) &cq->tail.data.marker;
+    cq->head.used_marker = 1;
+}
+
+static ERTS_INLINE int
+cq_enqueue(ErtsHLTCncldTmrQ *cq, ErtsTimer *tmr, int cinit)
+{
+    erts_aint_t itmp;
+    ErtsTimer *enq, *this = tmr;
+
+    erts_atomic_init_nob(&this->head.u.next, ERTS_AINT_NULL);
+    /* Enqueue at end of list... */
+
+    enq = (ErtsTimer *) erts_atomic_read_nob(&cq->tail.data.last);
+    itmp = erts_atomic_cmpxchg_relb(&enq->head.u.next,
+				    (erts_aint_t) this,
+				    ERTS_AINT_NULL);
+    if (itmp == ERTS_AINT_NULL) {
+	/* We are required to move last pointer */
+#ifdef DEBUG
+	ASSERT(ERTS_AINT_NULL == erts_atomic_read_nob(&this->head.u.next));
+	ASSERT(((erts_aint_t) enq)
+	       == erts_atomic_xchg_relb(&cq->tail.data.last,
+					(erts_aint_t) this));
+#else
+	erts_atomic_set_relb(&cq->tail.data.last, (erts_aint_t) this);
+#endif
+	return 1;
+    }
+    else {
+	/*
+	 * We *need* to insert element somewhere in between the
+	 * last element we read earlier and the actual last element.
+	 */
+	int i = cinit;
+
+	while (1) {
+	    erts_aint_t itmp2;
+	    erts_atomic_set_nob(&this->head.u.next, itmp);
+	    itmp2 = erts_atomic_cmpxchg_relb(&enq->head.u.next,
+					     (erts_aint_t) this,
+					     itmp);
+	    if (itmp == itmp2)
+		return 0; /* inserted this */
+	    if ((i & 1) == 0)
+		itmp = itmp2;
+	    else {
+		enq = (ErtsTimer *) itmp2;
+		itmp = erts_atomic_read_acqb(&enq->head.u.next);
+		ASSERT(itmp != ERTS_AINT_NULL);
+	    }
+	    i++;
+	}
+    }
+}
+
+static ERTS_INLINE erts_aint_t
+check_insert_marker(ErtsHLTCncldTmrQ *cq, erts_aint_t ilast)
+{
+    if (!cq->head.used_marker
+	&& cq->head.unref_end == (ErtsTimer *) ilast) {
+	erts_aint_t itmp;
+	ErtsTimer *last = (ErtsTimer *) ilast;
+
+	erts_atomic_init_nob(&cq->tail.data.marker.u.next, ERTS_AINT_NULL);
+	itmp = erts_atomic_cmpxchg_relb(&last->head.u.next,
+					(erts_aint_t) &cq->tail.data.marker,
+					ERTS_AINT_NULL);
+	if (itmp == ERTS_AINT_NULL) {
+	    ilast = (erts_aint_t) &cq->tail.data.marker;
+	    cq->head.used_marker = !0;
+	    erts_atomic_set_relb(&cq->tail.data.last, ilast);
+	}
+    }
+    return ilast;
+}
+
+static ERTS_INLINE ErtsTimer *
+cq_dequeue(ErtsHLTCncldTmrQ *cq)
+{
+    ErtsTimer *tmr;
+
+    if (cq->head.first == cq->head.unref_end)
+	return NULL;
+
+    tmr = cq->head.first;
+    if (tmr == (ErtsTimer *) &cq->tail.data.marker) {
+	ASSERT(cq->head.used_marker);
+	cq->head.used_marker = 0;
+	tmr = (ErtsTimer *) erts_atomic_read_nob(&tmr->head.u.next);
+	if (tmr == cq->head.unref_end) {
+	    cq->head.first = tmr;
+	    return NULL;
+	}
+    }
+
+    cq->head.first = (ErtsTimer *) erts_atomic_read_nob(&tmr->head.u.next);
+
+    ASSERT(cq->head.first);
+
+    return tmr;
+}
+
+static int
+cq_check_incoming(ErtsSchedulerData *esdp, ErtsHLTCncldTmrQ *cq)
+{
+    erts_aint_t ilast = erts_atomic_read_nob(&cq->tail.data.last);
+    if (((ErtsTimer *) ilast) == (ErtsTimer *) &cq->tail.data.marker
+	&& cq->head.first == (ErtsTimer *) &cq->tail.data.marker) {
+	/* Nothing more to do... */
+	return 0;
+    }
+
+    if (cq->head.next.thr_progress_reached
+	|| erts_thr_progress_has_reached(cq->head.next.thr_progress)) {
+	cq->head.next.thr_progress_reached = 1;
+	/* Move unreferenced end pointer forward... */
+
+	ERTS_HLT_SMP_MEMBAR_LoadLoad_LoadStore;
+
+	cq->head.unref_end = cq->head.next.unref_end;
+
+	ilast = check_insert_marker(cq, ilast);
+
+	if (cq->head.unref_end != (ErtsTimer *) ilast) {
+	    cq->head.next.unref_end = (ErtsTimer *) ilast;
+	    cq->head.next.thr_progress = erts_thr_progress_later(esdp);
+	    cq->head.next.thr_progress_reached = 0;
+	}
+    }
+    return 1;
+}
+
+static ERTS_INLINE void
+store_earliest_thr_prgr(ErtsThrPrgrVal *prev_val, ErtsHLTCncldTmrQ *cq)
+{
+    if (!cq->head.next.thr_progress_reached
+	&& (*prev_val == ERTS_THR_PRGR_INVALID
+	    || erts_thr_progress_cmp(cq->head.next.thr_progress,
+				     *prev_val) < 0)) {
+	*prev_val = cq->head.next.thr_progress;
+    }
+}
+
+static void
+handle_canceled_queue(ErtsSchedulerData *esdp,
+		      ErtsHLTCncldTmrQ *cq,
+		      int use_limit,
+		      int ops_limit,
+		      int *need_thr_progress,
+		      ErtsThrPrgrVal *thr_prgr_p,
+		      int *need_more_work)
+{
+    int need_thr_prgr = 0;
+    int need_mr_wrk = 0;
+    int have_checked_incoming = 0;
+    int ops = 0;
+
+    ERTS_HLT_ASSERT(cq == &esdp->timer_service->canceled_queue);
+
+    while (1) {
+	ErtsTimer *tmr = cq_dequeue(cq);
+
+	if (tmr)
+	    cleanup_sched_local_canceled_timer(esdp, tmr);
+	else {
+	    if (have_checked_incoming)
+		break;
+	    need_thr_prgr = cq_check_incoming(esdp, cq);
+	    if (need_thr_progress) {
+		*need_thr_progress |= need_thr_prgr;
+		if (need_thr_prgr)
+		    store_earliest_thr_prgr(thr_prgr_p, cq);
+	    }
+	    have_checked_incoming = 1;
+	    continue;
+	}
+
+	if (use_limit && ++ops >= ops_limit) {
+	    if (cq->head.first != cq->head.unref_end) {
+		need_mr_wrk = 1;
+		if (need_more_work)
+		    *need_more_work |= 1;
+	    }
+	    break;
+	}
+    }
+
+    if (need_thr_progress && !(need_thr_prgr | need_mr_wrk)) {
+	need_thr_prgr = cq_check_incoming(esdp, cq);
+	*need_thr_progress |= need_thr_prgr;
+	if (need_thr_prgr)
+	    store_earliest_thr_prgr(thr_prgr_p, cq);
+    }
+}
+
+void
+erts_handle_canceled_timers(void *vesdp,
+			    int *need_thr_progress,
+			    ErtsThrPrgrVal *thr_prgr_p,
+			    int *need_more_work)
+{
+    ErtsSchedulerData *esdp = (ErtsSchedulerData *) vesdp;
+    ERTS_HLT_ASSERT(esdp == erts_get_scheduler_data());
+
+    handle_canceled_queue(esdp, &esdp->timer_service->canceled_queue,
+			  1, ERTS_TMR_CANCELED_TIMER_LIMIT,
+			  need_thr_progress, thr_prgr_p,
+			  need_more_work);
+}
+
+#endif /* ERTS_SMP */
+
+static void
+queue_canceled_timer(ErtsSchedulerData *esdp, int rsched_id, ErtsTimer *tmr)
+{
+#ifdef ERTS_SMP
+    ErtsHLTCncldTmrQ *cq;
+    cq = &ERTS_SCHEDULER_IX(rsched_id-1)->timer_service->canceled_queue;
+    if (cq_enqueue(cq, tmr, rsched_id - (int) esdp->no))
+	erts_notify_canceled_timer(esdp, rsched_id);
+#else
+    ERTS_INTERNAL_ERROR("Unexpected enqueue of canceled timer");
+#endif
+}
+
+static void
+continue_cancel_ptimer(ErtsSchedulerData *esdp, ErtsTimer *tmr)
+{
+#ifdef ERTS_SMP
+    Uint32 sid = (tmr->head.roflgs & ERTS_TMR_ROFLG_SID_MASK);
+
+    if (esdp->no != sid)
+	queue_canceled_timer(esdp, sid, tmr);
+    else
+#endif
+	cleanup_sched_local_canceled_timer(esdp, tmr);
+}
+
+/*
+ * BIF timer specific
+ */
+
+Uint erts_bif_timer_memory_size(void)
+{
+    return (Uint) 0;
+}
+
+static BIF_RETTYPE
+setup_bif_timer(Process *c_p, ErtsMonotonicTime timeout_pos,
+		int short_time, Eterm rcvr, Eterm acsr,
+		Eterm msg, int wrap)
+{
+    BIF_RETTYPE ret;
+    Eterm ref, tmo_msg, *hp;
+    ErtsHLTimer *tmr;
+    ErtsSchedulerData *esdp;
+    DeclareTmpHeap(tmp_hp, 4, c_p);
+
+    if (is_not_internal_pid(rcvr) && is_not_atom(rcvr))
+	goto badarg;
+
+    esdp = ERTS_PROC_GET_SCHDATA(c_p);
+
+    hp = HAlloc(c_p, REF_THING_SIZE);
+    ref = erts_sched_make_ref_in_buffer(esdp, hp);
+
+    ASSERT(erts_get_ref_numbers_thr_id(
+	       internal_ref_numbers(ref)) == (Uint32) esdp->no);
+
+    UseTmpHeap(4, c_p);
+
+    tmo_msg = wrap ? TUPLE3(tmp_hp, am_timeout, ref, msg) : msg;
+
+    tmr = create_hl_timer(esdp, timeout_pos, short_time, 1, NULL, 
+			  rcvr, acsr, tmo_msg, internal_ref_numbers(ref));
+
+    UnUseTmpHeap(4, c_p);
+
+    if (is_internal_pid(rcvr)) {
+	Process *proc = erts_pid2proc_opt(c_p, ERTS_PROC_LOCK_MAIN,
+					  rcvr, ERTS_PROC_LOCK_BTM,
+					  ERTS_P2P_FLG_INC_REFC);
+	if (!proc) {
+	    hlt_delete_timer(esdp, tmr);
+	    hl_timer_destroy(tmr);
+	}
+	else {
+	    proc_btm_rbt_insert(&proc->bif_timers, tmr);
+	    erts_smp_proc_unlock(proc, ERTS_PROC_LOCK_BTM);
+	    tmr->receiver.proc = proc;
+	}
+    }
+
+    ERTS_BIF_PREP_RET(ret, ref);
+    return ret;
+
+badarg:
+
+    ERTS_BIF_PREP_ERROR(ret, c_p, BADARG);
+    return ret;
+}
+
+static int
+cancel_bif_timer(ErtsHLTimer *tmr)
+{
+    erts_aint_t state;
+    Uint32 roflgs;
+    int res;
+
+    state = erts_smp_atomic32_cmpxchg_acqb(&tmr->state,
+					   ERTS_TMR_STATE_CANCELED,
+					   ERTS_TMR_STATE_ACTIVE);
+    if (state != ERTS_TMR_STATE_ACTIVE)
+	return 0;
+
+    res = -1;
+
+    roflgs = tmr->head.roflgs;
+    if (roflgs & ERTS_TMR_ROFLG_PROC) {
+	Process *proc = tmr->receiver.proc;
+	ERTS_HLT_ASSERT(!(tmr->head.roflgs & ERTS_TMR_ROFLG_REG_NAME));
+
+	erts_smp_proc_lock(proc, ERTS_PROC_LOCK_BTM);
+	/*
+	 * If process is exiting, let it clean up
+	 * the btm tree by itself (it may be in
+	 * the middle of tree destruction).
+	 */
+	if (!ERTS_PROC_IS_EXITING(proc)
+	    && tmr->btm.proc_tree.parent != ERTS_HLT_PFIELD_NOT_IN_TABLE) {
+	    proc_btm_rbt_delete(&proc->bif_timers, tmr);
+	    tmr->btm.proc_tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+	    res = 1;
+	}
+	erts_smp_proc_unlock(proc, ERTS_PROC_LOCK_BTM);
+    }
+
+    return res;
+}
+
+static ERTS_INLINE Eterm
+access_sched_local_btm(Process *c_p, Eterm pid,
+		       Eterm tref, Uint32 *trefn,
+		       Uint32 *rrefn,
+		       int async, int cancel,
+		       int return_res,
+		       int info)
+{
+    ErtsSchedulerData *esdp;
+    ErtsHLTimerService *srv;
+    ErtsHLTimer *tmr;
+    Sint64 time_left;
+    Process *proc;
+    ErtsProcLocks proc_locks;
+
+    time_left = -1;
+
+    if (!c_p)
+	esdp = erts_get_scheduler_data();
+    else {
+	esdp = ERTS_PROC_GET_SCHDATA(c_p);
+	ERTS_HLT_ASSERT(esdp == erts_get_scheduler_data());
+    }
+
+    ERTS_HLT_ASSERT(erts_get_ref_numbers_thr_id(trefn)
+		    == (Uint32) esdp->no);
+
+    srv = esdp->timer_service;
+
+    tmr = btm_rbt_lookup(srv->btm_tree, trefn);
+    if (tmr) {
+	if (!cancel) {
+	    erts_aint32_t state = erts_smp_atomic32_read_acqb(&tmr->state);
+	    if (state == ERTS_TMR_STATE_ACTIVE)
+		time_left = get_time_left(esdp, tmr->timeout);
+	}
+	else {
+	    int cncl_res = cancel_bif_timer(tmr);
+	    if (cncl_res) {
+
+		time_left = get_time_left(esdp, tmr->timeout);
+
+		if (cncl_res > 0)
+		    hl_timer_dec_refc(tmr, tmr->head.roflgs);
+
+		hlt_delete_timer(esdp, tmr);
+	    }
+	}
+    }
+
+    if (!info)
+	return am_ok;
+
+    if (return_res) {
+	ERTS_HLT_ASSERT(c_p);
+	if (time_left < 0)
+	    return am_false;
+	else if (time_left <= (Sint64) MAX_SMALL)
+	    return make_small((Sint) time_left);
+	else {
+	    Uint hsz = ERTS_SINT64_HEAP_SIZE(time_left);
+	    Eterm *hp = HAlloc(c_p, hsz);
+	    return erts_sint64_to_big(time_left, &hp);
+	}
+    }
+
+    if (c_p) {
+	proc = c_p;
+	proc_locks = ERTS_PROC_LOCK_MAIN;
+    }
+    else {
+	proc = erts_proc_lookup(pid);
+	proc_locks = 0;
+    }
+
+    if (proc) {
+	Uint hsz;
+	ErlOffHeap *ohp;
+	ErlHeapFragment* bp;
+	Eterm *hp, msg, ref, result;
+#ifdef ERTS_HLT_DEBUG
+	Eterm *hp_end;
+#endif
+
+	hsz = 3; /* 2-tuple */
+	if (!async)
+	    hsz += REF_THING_SIZE;
+	else {
+	    if (is_non_value(tref))
+		hsz += REF_THING_SIZE;
+	    hsz += 1; /* upgrade to 3-tuple */
+	}
+	if (time_left > (Sint64) MAX_SMALL)
+	    hsz += ERTS_SINT64_HEAP_SIZE(time_left);
+
+	hp = erts_alloc_message_heap(hsz,
+				     &bp,
+				     &ohp,
+				     proc,
+				     &proc_locks);
+
+#ifdef ERTS_HLT_DEBUG
+	hp_end = hp + hsz;
+#endif
+
+	if (time_left < 0)
+	    result = am_false;
+	else if (time_left <= (Sint64) MAX_SMALL)
+	    result = make_small((Sint) time_left);
+	else
+	    result = erts_sint64_to_big(time_left, &hp);
+
+	if (!async) {
+	    write_ref_thing(hp,
+			    rrefn[0],
+			    rrefn[1],
+			    rrefn[2]);
+	    ref = make_internal_ref(hp);
+	    hp += REF_THING_SIZE;
+	    msg = TUPLE2(hp, ref, result);
+
+	    ERTS_HLT_ASSERT(hp + 3 == hp_end);
+	}
+	else {
+	    Eterm tag = cancel ? am_cancel_timer : am_read_timer;
+	    if (is_value(tref))
+		ref = tref;
+	    else {
+		write_ref_thing(hp,
+				trefn[0],
+				trefn[1],
+				trefn[2]);
+		ref = make_internal_ref(hp);
+		hp += REF_THING_SIZE;
+	    }
+	    msg = TUPLE3(hp, tag, ref, result);
+
+	    ERTS_HLT_ASSERT(hp + 4 == hp_end);
+
+	}
+	erts_queue_message(proc, &proc_locks, bp, msg, NIL);
+
+	if (c_p)
+	    proc_locks &= ~ERTS_PROC_LOCK_MAIN;
+	if (proc_locks)
+	    erts_smp_proc_unlock(proc, proc_locks);
+    }
+
+    return am_ok;
+}
+
+#define ERTS_BTM_REQ_FLG_ASYNC		(((Uint32) 1) << 0)
+#define ERTS_BTM_REQ_FLG_CANCEL		(((Uint32) 1) << 1)
+#define ERTS_BTM_REQ_FLG_INFO		(((Uint32) 1) << 2)
+
+typedef struct {
+    Eterm pid;
+    Uint32 trefn[ERTS_REF_NUMBERS];
+    Uint32 rrefn[ERTS_REF_NUMBERS];
+    Uint32 flags;
+} ErtsBifTimerRequest;
+
+static void
+bif_timer_access_request(void *vreq)
+{
+    ErtsBifTimerRequest *req = (ErtsBifTimerRequest *) vreq;
+    int async = (int) (req->flags & ERTS_BTM_REQ_FLG_ASYNC);
+    int cancel = (int) (req->flags & ERTS_BTM_REQ_FLG_CANCEL);
+    int info = (int) (req->flags & ERTS_BTM_REQ_FLG_INFO);
+    (void) access_sched_local_btm(NULL, req->pid, THE_NON_VALUE,
+				  req->trefn, req->rrefn, async,
+				  cancel, 0, info);
+    erts_free(ERTS_ALC_T_TIMER_REQUEST, vreq);
+}
+
+static int
+try_access_sched_remote_btm(ErtsSchedulerData *esdp,
+			    Process *c_p, Uint32 sid,
+			    Eterm tref, Uint32 *trefn,
+			    int async, int cancel,
+			    int info, Eterm *resp)
+{
+    ErtsHLTimer *tmr;
+    Sint64 time_left;
+
+    ERTS_HLT_ASSERT(c_p);
+
+    /*
+     * Check if the timer is aimed at current
+     * process of if this process is an accessor
+     * of the timer...
+     */
+    erts_smp_proc_lock(c_p, ERTS_PROC_LOCK_BTM);
+    tmr = proc_btm_rbt_lookup(c_p->bif_timers, trefn);
+    if (!tmr)
+	tmr = abtm_rbt_lookup(c_p->accessor_bif_timers, trefn);
+    erts_smp_proc_unlock(c_p, ERTS_PROC_LOCK_BTM);
+    if (!tmr)
+	return 0;
+
+    if (!cancel) {
+	erts_aint32_t state = erts_smp_atomic32_read_acqb(&tmr->state);
+	if (state == ERTS_TMR_STATE_ACTIVE)
+	    time_left = get_time_left(esdp, tmr->timeout);
+	else
+	    time_left = -1;
+    }
+    else {
+	int cncl_res = cancel_bif_timer(tmr);
+	if (!cncl_res)
+	    time_left = -1;
+	else {
+	    time_left = get_time_left(esdp, tmr->timeout);
+	    if (cncl_res > 0)
+		queue_canceled_timer(esdp, sid, (ErtsTimer *) tmr);
+	}
+    }
+
+    if (!info) {
+	*resp = am_ok;
+	return 1;
+    }
+
+    if (!async) {
+	if (time_left < 0)
+	    *resp = am_false;
+	else if (time_left <= (Sint64) MAX_SMALL)
+	    *resp = make_small((Sint) time_left);
+	else {
+	    Uint hsz = ERTS_SINT64_HEAP_SIZE(time_left);
+	    Eterm *hp = HAlloc(c_p, hsz);
+	    *resp = erts_sint64_to_big(time_left, &hp);
+	}
+    }
+    else {
+	Eterm tag, res, msg;
+	ErlOffHeap *ohp;
+	ErlHeapFragment* bp;
+	Uint hsz;
+	Eterm *hp;
+	ErtsProcLocks proc_locks = ERTS_PROC_LOCK_MAIN;
+
+	hsz = 4;
+	if (time_left > (Sint64) MAX_SMALL)
+	    hsz += ERTS_SINT64_HEAP_SIZE(time_left);
+
+	hp = erts_alloc_message_heap(hsz,
+				     &bp,
+				     &ohp,
+				     c_p,
+				     &proc_locks);
+	if (cancel)
+	    tag = am_cancel_timer;
+	else
+	    tag = am_read_timer;
+
+	if (time_left < 0)
+	    res = am_false;
+	else if (time_left <= (Sint64) MAX_SMALL)
+	    res = make_small((Sint) time_left);
+	else
+	    res = erts_sint64_to_big(time_left, &hp);
+
+	msg = TUPLE3(hp, tag, tref, res);
+
+	erts_queue_message(c_p, &proc_locks, bp, msg, NIL);
+
+	proc_locks &= ~ERTS_PROC_LOCK_MAIN;
+	if (proc_locks)
+	    erts_smp_proc_unlock(c_p, proc_locks);
+
+	*resp = am_ok;
+    }
+    return 1;
+}
+
+static BIF_RETTYPE
+access_bif_timer(Process *c_p, Eterm tref, int cancel, int async, int info)
+{
+    BIF_RETTYPE ret;
+    ErtsSchedulerData *esdp;
+    Uint32 sid;
+    Uint32 *trefn;
+    Eterm res;
+
+    if (is_not_internal_ref(tref)) {
+	if (is_not_ref(tref))
+	    goto badarg;
+	else
+	    goto no_timer;
+    }
+
+    esdp = ERTS_PROC_GET_SCHDATA(c_p);
+
+    trefn = internal_ref_numbers(tref);
+    sid = erts_get_ref_numbers_thr_id(trefn);
+    if (sid < 1 || erts_no_schedulers < sid)
+	goto no_timer;
+
+    if (sid == (Uint32) esdp->no) {
+	res = access_sched_local_btm(c_p, c_p->common.id,
+				     tref, trefn, NULL,
+				     async, cancel, !async,
+				     info);
+	ERTS_BIF_PREP_RET(ret, res);
+    }
+    else if (try_access_sched_remote_btm(esdp, c_p, sid,
+					 tref, trefn,
+					 async, cancel,
+					 info, &res)) {
+	ERTS_BIF_PREP_RET(ret, res);
+    }
+    else {
+	/*
+	 * Schedule access for execution on
+	 * remote scheduler...
+	 */
+	ErtsBifTimerRequest *req = erts_alloc(ERTS_ALC_T_TIMER_REQUEST,
+					      sizeof(ErtsBifTimerRequest));
+
+	req->flags = 0;
+	if (cancel)
+	    req->flags |= ERTS_BTM_REQ_FLG_CANCEL;
+	if (async)
+	    req->flags |= ERTS_BTM_REQ_FLG_ASYNC;
+	if (info)
+	    req->flags |= ERTS_BTM_REQ_FLG_INFO;
+
+	req->pid = c_p->common.id;
+
+	req->trefn[0] = trefn[0];
+	req->trefn[1] = trefn[1];
+	req->trefn[2] = trefn[2];
+
+	if (async)
+	    ERTS_BIF_PREP_RET(ret, am_ok);
+	else {
+	    Eterm *hp, rref;
+	    Uint32 *rrefn;
+
+	    hp = HAlloc(c_p, REF_THING_SIZE);
+	    rref = erts_sched_make_ref_in_buffer(esdp, hp);
+	    rrefn = internal_ref_numbers(rref);
+
+	    req->rrefn[0] = rrefn[0];
+	    req->rrefn[1] = rrefn[1];
+	    req->rrefn[2] = rrefn[2];
+
+	    erts_smp_proc_lock(c_p, ERTS_PROC_LOCKS_MSG_RECEIVE);
+
+	    if (ERTS_PROC_PENDING_EXIT(c_p))
+		ERTS_VBUMP_ALL_REDS(c_p);
+	    else {
+		/*
+		 * Caller needs to wait for a message containing
+		 * the ref that we just created. No such message
+		 * can exist in callers message queue at this time.
+		 * We therefore move the save pointer of the
+		 * callers message queue to the end of the queue.
+		 *
+		 * NOTE: It is of vital importance that the caller
+		 *       immediately do a receive unconditionaly
+		 *       waiting for the message with the reference;
+		 *       otherwise, next receive will *not* work
+		 *       as expected!
+		 */
+		ERTS_SMP_MSGQ_MV_INQ2PRIVQ(c_p);
+		c_p->msg.save = c_p->msg.last;
+	    }
+	    erts_smp_proc_unlock(c_p, ERTS_PROC_LOCKS_MSG_RECEIVE);
+
+	    ERTS_BIF_PREP_TRAP1(ret, erts_await_result, c_p, rref);
+	}
+
+	erts_schedule_misc_aux_work(sid,
+				    bif_timer_access_request,
+				    (void *) req);
+    }
+
+    return ret;
+
+badarg:
+    ERTS_BIF_PREP_ERROR(ret, c_p, BADARG);
+    return ret;
+
+no_timer:
+    ERTS_BIF_PREP_RET(ret, am_false);
+    return ret;
+    
+}
+
+static ERTS_INLINE int
+bool_arg(Eterm val, int *argp)
+{
+    switch (val) {
+    case am_true: *argp = 1; return 1;
+    case am_false: *argp = 0; return 1;
+    default: return 0;
+    }
+}
+
+static ERTS_INLINE int
+parse_bif_timer_options(Eterm option_list, int *async, int *info,
+			int *abs, Eterm *accessor)
+{
+    Eterm list = option_list;
+
+    if (async)
+	*async = 0;
+    if (info)
+	*info = 0;
+    if (abs)
+	*abs = 0;
+    if (accessor)
+	*accessor = THE_NON_VALUE;
+
+    while (is_list(list)) {
+	Eterm *consp, *tp, opt;
+
+	consp = list_val(list);
+	opt = CAR(consp);
+	if (is_not_tuple(opt))
+	    return 0;
+
+	tp = tuple_val(opt);
+	if (arityval(tp[0]) != 2)
+	    return 0;
+
+	switch (tp[1]) {
+	case am_async:
+	    if (!async || !bool_arg(tp[2], async))
+		return 0;
+	    break;
+	case am_info:
+	    if (!info || !bool_arg(tp[2], info))
+		return 0;
+	    break;
+	case am_abs:
+	    if (!abs || !bool_arg(tp[2], abs))
+		return 0;
+	    break;
+	case am_accessor:
+	    if (!accessor || is_not_internal_pid(tp[2]))
+		return 0;
+	    *accessor = tp[2];
+	    break;
+	default:
+	    return 0;
+	}
+
+	list = CDR(consp);	
+    }
+
+    if (is_not_nil(list))
+	return 0;
+    return 1;
+}
+
+static void
+exit_cancel_bif_timer(ErtsHLTimer *tmr, void *vesdp)
+{
+    ErtsSchedulerData *esdp = (ErtsSchedulerData *) vesdp;
+    Uint32 sid, roflgs;
+    erts_aint_t state;
+
+    state = erts_smp_atomic32_cmpxchg_acqb(&tmr->state,
+					   ERTS_TMR_STATE_CANCELED,
+					   ERTS_TMR_STATE_ACTIVE);
+
+    roflgs = tmr->head.roflgs;
+    sid = roflgs & ERTS_TMR_ROFLG_SID_MASK;
+
+    ERTS_HLT_ASSERT(sid == erts_get_ref_numbers_thr_id(tmr->btm.refn));
+    ERTS_HLT_ASSERT(tmr->btm.proc_tree.parent
+		    != ERTS_HLT_PFIELD_NOT_IN_TABLE);
+
+    tmr->btm.proc_tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+
+    if (sid == (Uint32) esdp->no) {
+	if (state == ERTS_TMR_STATE_ACTIVE)
+	    hlt_delete_timer(esdp, tmr);
+	hl_timer_dec_refc(tmr, roflgs);
+    }
+    else {
+	if (state == ERTS_TMR_STATE_ACTIVE)
+	    queue_canceled_timer(esdp, sid, (ErtsTimer *) tmr);
+	else
+	    hl_timer_dec_refc(tmr, roflgs);
+    }
+}
+
+#ifdef ERTS_HLT_DEBUG
+#  define ERTS_BTM_MAX_DESTROY_LIMIT 2
+#else
+#  define ERTS_BTM_MAX_DESTROY_LIMIT 50
+#endif
+
+typedef struct {
+    ErtsBifTimers *bif_timers;
+    union {
+	proc_btm_rbt_yield_state_t proc_btm_yield_state;
+	abtm_rbt_yield_state_t abtm_yield_state;
+    } u;
+} ErtsBifTimerYieldState;
+
+int erts_cancel_bif_timers(Process *p, ErtsBifTimers *btm, void **vyspp)
+{
+    ErtsSchedulerData *esdp = ERTS_PROC_GET_SCHDATA(p);
+    ErtsBifTimerYieldState ys = {btm, {ERTS_RBT_YIELD_STAT_INITER}};
+    ErtsBifTimerYieldState *ysp;
+    int res;
+
+    ysp = (ErtsBifTimerYieldState *) *vyspp;
+    if (!ysp)
+	ysp = &ys;
+
+    res = proc_btm_rbt_foreach_destroy_yielding(&ysp->bif_timers,
+						exit_cancel_bif_timer,
+						(void *) esdp,
+						&ysp->u.proc_btm_yield_state,
+						ERTS_BTM_MAX_DESTROY_LIMIT);
+
+    if (res == 0) {
+	if (ysp != &ys)
+	    erts_free(ERTS_ALC_T_BTM_YIELD_STATE, ysp);
+	*vyspp = NULL;
+    }
+    else {
+
+	if (ysp == &ys) {
+	    ysp = erts_alloc(ERTS_ALC_T_BTM_YIELD_STATE,
+			     sizeof(ErtsBifTimerYieldState));
+	    sys_memcpy((void *) ysp, (void *) &ys,
+		       sizeof(ErtsBifTimerYieldState));
+	}
+
+	*vyspp = (void *) ysp;
+    }
+
+    return res;
+}
+
+static void
+detach_bif_timer(ErtsHLTimer *tmr, void *vesdp)
+{
+    tmr->abtm.tree.parent = ERTS_HLT_PFIELD_NOT_IN_TABLE;
+    hl_timer_dec_refc(tmr, tmr->head.roflgs);
+}
+
+int erts_detach_accessor_bif_timers(Process *p, ErtsBifTimers *btm, void **vyspp)
+{
+    ErtsSchedulerData *esdp = ERTS_PROC_GET_SCHDATA(p);
+    ErtsBifTimerYieldState ys = {btm, {ERTS_RBT_YIELD_STAT_INITER}};
+    ErtsBifTimerYieldState *ysp;
+    int res;
+
+    ysp = (ErtsBifTimerYieldState *) *vyspp;
+    if (!ysp)
+	ysp = &ys;
+
+    res = abtm_rbt_foreach_destroy_yielding(&ysp->bif_timers,
+					    detach_bif_timer,
+					    (void *) esdp,
+					    &ysp->u.abtm_yield_state,
+					    ERTS_BTM_MAX_DESTROY_LIMIT);
+
+    if (res == 0) {
+	if (ysp != &ys)
+	    erts_free(ERTS_ALC_T_BTM_YIELD_STATE, ysp);
+	*vyspp = NULL;
+    }
+    else {
+
+	if (ysp == &ys) {
+	    ysp = erts_alloc(ERTS_ALC_T_BTM_YIELD_STATE,
+			     sizeof(ErtsBifTimerYieldState));
+	    sys_memcpy((void *) ysp, (void *) &ys,
+		       sizeof(ErtsBifTimerYieldState));
+	}
+
+	*vyspp = (void *) ysp;
+    }
+
+    return res;
+}
+
+static ERTS_INLINE int
+parse_timeout_pos(ErtsSchedulerData *esdp, Eterm arg,
+		  ErtsMonotonicTime *conv_arg, int abs,
+		  ErtsMonotonicTime *tposp, int *stimep)
+{
+    ErtsMonotonicTime t;
+
+    if (!term_to_Sint64(arg, &t)) {
+	ERTS_HLT_ASSERT(!is_small(arg));
+	if (!is_big(arg))
+	    return -1;
+
+	if (abs || !big_sign(arg))
+	    return 1;
+
+	return -1;
+    }
+
+    if (conv_arg)
+	*conv_arg = t;
+
+    if (abs) {
+	t += -1*ERTS_MONOTONIC_OFFSET_MSEC; /* external to internal */
+	if (t < ERTS_MONOTONIC_TO_MSEC(ERTS_MONOTONIC_BEGIN))
+	    return 1;
+	if (t > ERTS_MONOTONIC_TO_MSEC(ERTS_MONOTONIC_END))
+	    return 1;
+	*stimep = (t - ERTS_MONOTONIC_TO_MSEC(esdp->last_monotonic_time)
+		   < ERTS_BIF_TIMER_SHORT_TIME);
+	*tposp = ERTS_MSEC_TO_CLKTCKS(t);
+    }
+    else {
+	ErtsMonotonicTime now, ticks;
+	
+	if (t < 0)
+	    return -1;
+
+	ticks = ERTS_MSEC_TO_CLKTCKS(t);
+
+	if (ERTS_CLKTCK_RESOLUTION > 1000 && ticks < 0)
+	    return 1;
+
+	ERTS_HLT_ASSERT(ticks >= 0);
+
+	now = erts_get_monotonic_time(esdp);
+	ticks += ERTS_MONOTONIC_TO_CLKTCKS(now-1);
+	ticks += 1;
+
+	if (ticks < ERTS_MONOTONIC_TO_CLKTCKS(ERTS_MONOTONIC_BEGIN))
+	    return 1;
+	if (ticks > ERTS_MONOTONIC_TO_CLKTCKS(ERTS_MONOTONIC_END))
+	    return 1;
+
+	*stimep = (t < ERTS_BIF_TIMER_SHORT_TIME);
+	*tposp = ticks;
+    }
+
+    return 0;
+}
+
+/*
+ *
+ * The BIF timer BIFs...
+ */
+
+BIF_RETTYPE send_after_3(BIF_ALIST_3)
+{
+    ErtsMonotonicTime timeout_pos;
+    int short_time, tres;
+
+    tres = parse_timeout_pos(ERTS_PROC_GET_SCHDATA(BIF_P), BIF_ARG_1, NULL,
+			     0, &timeout_pos, &short_time);
+    if (tres != 0)
+	BIF_ERROR(BIF_P, tres < 0 ? BADARG : SYSTEM_LIMIT);
+
+    return setup_bif_timer(BIF_P, timeout_pos, short_time,
+			   BIF_ARG_2, BIF_ARG_2, BIF_ARG_3, 0);
+}
+
+BIF_RETTYPE send_after_4(BIF_ALIST_4)
+{
+    ErtsMonotonicTime timeout_pos;
+    Eterm accessor;
+    int short_time, abs, tres;
+
+    if (!parse_bif_timer_options(BIF_ARG_4, NULL, NULL, &abs, &accessor))
+	BIF_ERROR(BIF_P, BADARG);
+    
+    tres = parse_timeout_pos(ERTS_PROC_GET_SCHDATA(BIF_P), BIF_ARG_1, NULL,
+			     abs, &timeout_pos, &short_time);
+    if (tres != 0)
+	BIF_ERROR(BIF_P, tres < 0 ? BADARG : SYSTEM_LIMIT);
+
+    return setup_bif_timer(BIF_P, timeout_pos, short_time,
+			   BIF_ARG_2, accessor, BIF_ARG_3, 0);
+}
+
+BIF_RETTYPE start_timer_3(BIF_ALIST_3)
+{
+    ErtsMonotonicTime timeout_pos;
+    int short_time, tres;
+
+    tres = parse_timeout_pos(ERTS_PROC_GET_SCHDATA(BIF_P), BIF_ARG_1, NULL,
+			     0, &timeout_pos, &short_time);
+    if (tres != 0)
+	BIF_ERROR(BIF_P, tres < 0 ? BADARG : SYSTEM_LIMIT);
+
+    return setup_bif_timer(BIF_P, timeout_pos, short_time,
+			   BIF_ARG_2, BIF_ARG_2, BIF_ARG_3, !0);
+}
+
+BIF_RETTYPE start_timer_4(BIF_ALIST_4)
+{
+    ErtsMonotonicTime timeout_pos;
+    Eterm accessor;
+    int short_time, abs, tres;
+
+    if (!parse_bif_timer_options(BIF_ARG_4, NULL, NULL, &abs, &accessor))
+	BIF_ERROR(BIF_P, BADARG);
+
+    tres = parse_timeout_pos(ERTS_PROC_GET_SCHDATA(BIF_P), BIF_ARG_1, NULL,
+			     abs, &timeout_pos, &short_time);
+    if (tres != 0)
+	BIF_ERROR(BIF_P, tres < 0 ? BADARG : SYSTEM_LIMIT);
+
+    return setup_bif_timer(BIF_P, timeout_pos, short_time,
+			   BIF_ARG_2, accessor, BIF_ARG_3, !0);
+}
+
+BIF_RETTYPE cancel_timer_1(BIF_ALIST_1)
+{
+    return access_bif_timer(BIF_P, BIF_ARG_1, 1, 0, 1);
+}
+
+BIF_RETTYPE cancel_timer_2(BIF_ALIST_2)
+{
+    BIF_RETTYPE ret;
+    int async, info;
+
+    if (parse_bif_timer_options(BIF_ARG_2, &async, &info, NULL, NULL))
+	return access_bif_timer(BIF_P, BIF_ARG_1, 1, async, info);
+
+    ERTS_BIF_PREP_ERROR(ret, BIF_P, BADARG);
+    return ret;
+}
+
+BIF_RETTYPE read_timer_1(BIF_ALIST_1)
+{
+    return access_bif_timer(BIF_P, BIF_ARG_1, 0, 0, 1);
+}
+
+BIF_RETTYPE read_timer_2(BIF_ALIST_2)
+{
+    BIF_RETTYPE ret;
+    int async;
+
+    if (parse_bif_timer_options(BIF_ARG_2, &async, NULL, NULL, NULL))
+	return access_bif_timer(BIF_P, BIF_ARG_1, 0, async, 1);
+
+    ERTS_BIF_PREP_ERROR(ret, BIF_P, BADARG);
+    return ret;
+}
+
+/*
+ * Process and Port timer functionality.
+ *
+ * NOTE! These are only allowed to be called by a
+ *       scheduler thread that currently is
+ *       executing the process or port.
+ */
+
+static ERTS_INLINE void
+set_proc_timer_common(Process *c_p, ErtsSchedulerData *esdp, Sint64 tmo,
+		      ErtsMonotonicTime timeout_pos, int short_time)
+{
+    void *tmr;
+    check_canceled_queue(esdp, esdp->timer_service);
+
+    if (tmo == 0)
+	c_p->flags |= F_TIMO;
+    else {
+
+	c_p->flags |= F_INSLPQUEUE;
+	c_p->flags &= ~F_TIMO;
+
+	if (tmo < ERTS_TIMER_WHEEL_MSEC)
+	    tmr = (void *) create_tw_timer(esdp, (void *) c_p, 1, timeout_pos);
+	else
+	    tmr = (void *) create_hl_timer(esdp, timeout_pos,
+					   short_time, 0, (void *) c_p,
+					   c_p->common.id, NIL, NIL, NULL);
+	erts_smp_atomic_set_relb(&c_p->common.timer, (erts_aint_t) tmr);
+    }
+}
+
+int
+erts_set_proc_timer_term(Process *c_p, Eterm etmo)
+{
+    ErtsSchedulerData *esdp = ERTS_PROC_GET_SCHDATA(c_p);
+    ErtsMonotonicTime tmo, timeout_pos;
+    int short_time, tres;
+
+    ERTS_HLT_ASSERT(erts_smp_atomic_read_nob(&c_p->common.timer)
+		    == ERTS_PTMR_NONE);
+
+    tres = parse_timeout_pos(esdp, etmo, &tmo, 0,
+			     &timeout_pos, &short_time);
+    if (tres != 0)
+	return tres;
+
+    if ((tmo >> 32) != 0)
+	return 1;
+
+    set_proc_timer_common(c_p, esdp, tmo, timeout_pos, short_time);
+    return 0;
+}
+
+void
+erts_set_proc_timer_uword(Process *c_p, UWord tmo)
+{
+    ErtsSchedulerData *esdp = ERTS_PROC_GET_SCHDATA(c_p);
+
+    ERTS_HLT_ASSERT(erts_smp_atomic_read_nob(&c_p->common.timer)
+		    == ERTS_PTMR_NONE);
+
+#ifndef ARCH_32
+    ERTS_HLT_ASSERT((tmo >> 32) == (UWord) 0);
+#endif
+
+    if (tmo == 0)
+	c_p->flags |= F_TIMO;
+    else {
+	ErtsMonotonicTime timeout_pos;
+	timeout_pos = get_timeout_pos(erts_get_monotonic_time(esdp),
+				      (ErtsMonotonicTime) tmo);
+	set_proc_timer_common(c_p, esdp, (ErtsMonotonicTime) tmo,
+			      timeout_pos,
+			      tmo < ERTS_BIF_TIMER_SHORT_TIME);
+    }
+}
+
+void
+erts_cancel_proc_timer(Process *c_p)
+{
+    erts_aint_t tval;
+    tval = erts_smp_atomic_xchg_acqb(&c_p->common.timer,
+				     ERTS_PTMR_NONE);
+    c_p->flags &= ~(F_INSLPQUEUE|F_TIMO);
+    if (tval == ERTS_PTMR_NONE)
+	return;
+    if (tval == ERTS_PTMR_TIMEDOUT) {
+	erts_smp_atomic_set_nob(&c_p->common.timer, ERTS_PTMR_NONE);
+	return;
+    }
+    continue_cancel_ptimer(ERTS_PROC_GET_SCHDATA(c_p),
+			   (ErtsTimer *) tval);
+}
+
+void
+erts_set_port_timer(Port *c_prt, Sint64 tmo)
+{
+    void *tmr;
+    ErtsSchedulerData *esdp = erts_get_scheduler_data();
+    ErtsMonotonicTime timeout_pos;
+
+    if (erts_smp_atomic_read_nob(&c_prt->common.timer) != ERTS_PTMR_NONE)
+	erts_cancel_port_timer(c_prt);
+
+    check_canceled_queue(esdp, esdp->timer_service);
+
+    timeout_pos = get_timeout_pos(erts_get_monotonic_time(esdp), tmo);
+
+    if (tmo < ERTS_TIMER_WHEEL_MSEC)
+	tmr = (void *) create_tw_timer(esdp, (void *) c_prt, 0,
+				       timeout_pos);
+    else
+	tmr = (void *) create_hl_timer(esdp, timeout_pos, 0, 0,
+				       (void *) c_prt,
+				       c_prt->common.id, NIL, NIL,
+				       NULL);
+    erts_smp_atomic_set_relb(&c_prt->common.timer, (erts_aint_t) tmr);
+}
+
+void
+erts_cancel_port_timer(Port *c_prt)
+{
+    erts_aint_t tval;
+    tval = erts_smp_atomic_xchg_acqb(&c_prt->common.timer,
+				     ERTS_PTMR_NONE);
+    if (tval == ERTS_PTMR_NONE)
+	return;
+    if (tval == ERTS_PTMR_TIMEDOUT) {
+	while (!erts_port_task_is_scheduled(&c_prt->timeout_task))
+	    erts_thr_yield();
+	erts_port_task_abort(&c_prt->timeout_task);
+	erts_smp_atomic_set_nob(&c_prt->common.timer, ERTS_PTMR_NONE);
+	return;
+    }
+    continue_cancel_ptimer(erts_get_scheduler_data(),
+			   (ErtsTimer *) tval);
+}
+
+Sint64
+erts_read_port_timer(Port *c_prt)
+{
+    ErtsTimer *tmr;
+    erts_aint_t itmr;
+    ErtsMonotonicTime timeout_pos;
+
+    itmr = erts_smp_atomic_read_acqb(&c_prt->common.timer);
+    if (itmr == ERTS_PTMR_NONE)
+	return (Sint64) -1;
+    if (itmr == ERTS_PTMR_TIMEDOUT)
+	return (Sint64) 0;
+    tmr = (ErtsTimer *) itmr;
+    if (tmr->head.roflgs & ERTS_TMR_ROFLG_HLT)
+	timeout_pos = tmr->hlt.timeout;
+    else
+	timeout_pos = tmr->twt.tw_tmr.timeout_pos;
+    return get_time_left(NULL, timeout_pos);
+}
+
+/*
+ * Debug stuff...
+ */
+
+typedef struct {
+    int to;
+    void *to_arg;
+    ErtsMonotonicTime now;
+} ErtsBTMPrint;
+
+static void
+btm_print(ErtsHLTimer *tmr, void *vbtmp)
+{
+    ErtsBTMPrint *btmp = (ErtsBTMPrint *) vbtmp;
+    ErtsMonotonicTime left;
+    Eterm receiver;
+
+    if (tmr->timeout <= btmp->now)
+	left = 0;
+    left = ERTS_CLKTCKS_TO_MSEC(tmr->timeout - btmp->now);
+
+    receiver = ((tmr->head.roflgs & ERTS_TMR_ROFLG_REG_NAME)
+		? tmr->receiver.name
+		: tmr->receiver.proc->common.id);
+
+    erts_print(btmp->to, btmp->to_arg,
+	       "=timer:%T\n"
+	       "Message: %T\n"
+	       "Time left: %b64d\n",
+	       receiver,
+	       tmr->btm.message,
+	       (Sint64) left);
+}
+
+void
+erts_print_bif_timer_info(int to, void *to_arg)
+{
+    ErtsBTMPrint btmp;
+    int six;
+
+    if (!ERTS_IS_CRASH_DUMPING)
+	ERTS_INTERNAL_ERROR("Not crash dumping");
+
+    btmp.to = to;
+    btmp.to_arg = to_arg;
+    btmp.now = erts_get_monotonic_time(NULL);
+    btmp.now = ERTS_MONOTONIC_TO_CLKTCKS(btmp.now);
+
+    for (six = 0; six < erts_no_schedulers; six++) {
+	ErtsHLTimerService *srv =
+	    erts_aligned_scheduler_data[six].esd.timer_service;
+	btm_rbt_foreach(srv->btm_tree, btm_print, (void *) &btmp);
+    }
+}
+
+typedef struct {
+    void (*func)(Eterm,
+		 Eterm,
+		 ErlHeapFragment *,
+		 void *);
+    void *arg;
+} ErtsBTMForeachDebug;
+
+static void
+debug_btm_foreach(ErtsHLTimer *tmr, void *vbtmfd)
+{
+    if (erts_smp_atomic32_read_nob(&tmr->state) == ERTS_TMR_STATE_ACTIVE) {
+	ErtsBTMForeachDebug *btmfd = (ErtsBTMForeachDebug *) vbtmfd;
+	(*btmfd->func)(((tmr->head.roflgs & ERTS_TMR_ROFLG_REG_NAME)
+			? tmr->receiver.name
+			: tmr->receiver.proc->common.id),
+		       tmr->btm.message,
+		       tmr->btm.bp,
+		       btmfd->arg);
+    }
+}
+
+void
+erts_debug_bif_timer_foreach(void (*func)(Eterm,
+					  Eterm,
+					  ErlHeapFragment *,
+					  void *),
+			     void *arg)
+{
+    ErtsBTMForeachDebug btmfd;
+    int six;
+
+    btmfd.func = func;
+    btmfd.arg = arg;
+
+    if (!erts_smp_thr_progress_is_blocking())
+	ERTS_INTERNAL_ERROR("Not blocking thread progress");
+
+    for (six = 0; six < erts_no_schedulers; six++) {
+	ErtsHLTimerService *srv =
+	    erts_aligned_scheduler_data[six].esd.timer_service;
+	btm_rbt_foreach(srv->btm_tree,
+			debug_btm_foreach,
+			(void *) &btmfd);
+    }
+}
+
+#ifdef ERTS_HLT_HARD_DEBUG
+
+typedef struct {
+    ErtsHLTimerService *srv;
+    int found_root;
+    ErtsHLTimer **rootpp;
+} ErtsHdbgHLT;
+
+static void
+st_hdbg_func(ErtsHLTimer *tmr, void *vhdbg)
+{
+    ErtsHdbgHLT *hdbg = (ErtsHdbgHLT *) vhdbg;
+    ErtsHLTimer **rootpp;
+    ERTS_HLT_ASSERT(tmr->time.tree.parent & ERTS_HLT_PFLG_SAME_TIME);
+    if (tmr->time.tree.parent == ERTS_HLT_PFLG_SAME_TIME) {
+	ERTS_HLT_ASSERT(tmr != *hdbg->rootpp);
+    }
+    else {
+	rootpp = (ErtsHLTimer **) (tmr->time.tree.parent
+				   & ~ERTS_HLT_PFLG_SAME_TIME);
+	ERTS_HLT_ASSERT(rootpp == hdbg->rootpp);
+	ERTS_HLT_ASSERT(tmr == *rootpp);
+	ERTS_HLT_ASSERT(!hdbg->found_root);
+	hdbg->found_root = 1;
+    }
+    ERTS_HLT_ASSERT(tmr->time.tree.u.l.next->time.tree.u.l.prev == tmr);
+    ERTS_HLT_ASSERT(tmr->time.tree.u.l.prev->time.tree.u.l.next == tmr);
+    ERTS_HLT_ASSERT(btm_rbt_lookup(hdbg->srv->btm_tree, tmr->btm.refn) == tmr);
+}
+
+static void
+tt_hdbg_func(ErtsHLTimer *tmr, void *vhdbg)
+{
+    ErtsHdbgHLT *hdbg = (ErtsHdbgHLT *) vhdbg;
+    ErtsHLTimer *prnt;
+    ERTS_HLT_ASSERT((tmr->time.tree.parent & ERTS_HLT_PFLG_SAME_TIME) == 0);
+    prnt = (ErtsHLTimer *) (tmr->time.tree.parent & ~ERTS_HLT_PFLGS_MASK);
+    if (prnt) {
+	ERTS_HLT_ASSERT(prnt->time.tree.u.t.left == tmr
+			|| prnt->time.tree.u.t.right == tmr);
+    }
+    else {
+	ERTS_HLT_ASSERT(!hdbg->found_root);
+	hdbg->found_root = 1;
+	ERTS_HLT_ASSERT(tmr == *hdbg->rootpp);
+    }
+    if (tmr->time.tree.u.t.left) {
+	prnt = (ErtsHLTimer *) (tmr->time.tree.u.t.left->time.tree.parent
+				& ~ERTS_HLT_PFLGS_MASK);
+	ERTS_HLT_ASSERT(tmr == prnt);
+    }
+    if (tmr->time.tree.u.t.right) {
+	prnt = (ErtsHLTimer *) (tmr->time.tree.u.t.right->time.tree.parent
+				& ~ERTS_HLT_PFLGS_MASK);
+	ERTS_HLT_ASSERT(tmr == prnt);
+    }
+    ERTS_HLT_ASSERT(btm_rbt_lookup(hdbg->srv->btm_tree, tmr->btm.refn) == tmr);
+    if (tmr->time.tree.same_time) {
+	ErtsHdbgHLT st_hdbg;
+	st_hdbg.srv = hdbg->srv;
+	st_hdbg.found_root = 0;
+	st_hdbg.rootpp = &tmr->time.tree.same_time;
+	same_time_list_foreach(tmr->time.tree.same_time, st_hdbg_func, (void *) &st_hdbg);
+	ERTS_HLT_ASSERT(st_hdbg.found_root);
+    }
+}
+
+static void
+bt_hdbg_func(ErtsHLTimer *tmr, void *vhdbg)
+{
+    ErtsHdbgHLT *hdbg = (ErtsHdbgHLT *) vhdbg;
+    ErtsHLTimer *prnt;
+    ERTS_HLT_ASSERT((tmr->btm.tree.parent & ERTS_HLT_PFLG_SAME_TIME) == 0);
+    prnt = (ErtsHLTimer *) (tmr->btm.tree.parent & ~ERTS_HLT_PFLGS_MASK);
+    if (prnt) {
+	ERTS_HLT_ASSERT(prnt->btm.tree.left == tmr
+			|| prnt->btm.tree.right == tmr);
+    }
+    else {
+	ERTS_HLT_ASSERT(!hdbg->found_root);
+	hdbg->found_root = 1;
+	ERTS_HLT_ASSERT(tmr == *hdbg->rootpp);
+    }
+    if (tmr->btm.tree.left) {
+	prnt = (ErtsHLTimer *) (tmr->btm.tree.left->btm.tree.parent
+				& ~ERTS_HLT_PFLGS_MASK);
+	ERTS_HLT_ASSERT(tmr == prnt);
+    }
+    if (tmr->btm.tree.right) {
+	prnt = (ErtsHLTimer *) (tmr->btm.tree.right->btm.tree.parent
+				& ~ERTS_HLT_PFLGS_MASK);
+	ERTS_HLT_ASSERT(tmr == prnt);
+    }
+    if (tmr->pending_timeout) {
+	if (tmr->pending_timeout > 0) /* container > 0 */
+	    ERTS_HLT_ASSERT(tmr->time.tree.parent == ERTS_HLT_PFIELD_NOT_IN_TABLE);
+	else {
+	    ERTS_HLT_ASSERT(tmr->time.tree.parent != ERTS_HLT_PFIELD_NOT_IN_TABLE);
+	    ERTS_HLT_ASSERT(tmr->time.tree.parent & ERTS_HLT_PFLG_SAME_TIME);
+	}
+    }
+    else {
+	ErtsHLTimer *ttmr = time_rbt_lookup(hdbg->srv->time_tree, tmr->timeout);
+	ERTS_HLT_ASSERT(ttmr);
+	if (ttmr != tmr) {
+	    ERTS_HLT_ASSERT(ttmr->time.tree.same_time);
+	    ERTS_HLT_ASSERT(tmr == same_time_list_lookup(ttmr->time.tree.same_time, tmr));
+	}
+    }
+}
+
+static void
+hdbg_chk_srv(ErtsHLTimerService *srv)
+{
+    if (srv->time_tree) {
+	ErtsHdbgHLT hdbg;
+	hdbg.srv = srv;
+	hdbg.found_root = 0;
+	hdbg.rootpp = &srv->time_tree;
+	time_rbt_foreach(srv->time_tree, tt_hdbg_func, (void *) &hdbg);
+	ERTS_HLT_ASSERT(hdbg.found_root);
+    }
+    if (srv->btm_tree) {
+	ErtsHdbgHLT hdbg;
+	hdbg.srv = srv;
+	hdbg.found_root = 0;
+	hdbg.rootpp = &srv->btm_tree;
+	btm_rbt_foreach(srv->btm_tree, bt_hdbg_func, (void *) &hdbg);
+	ERTS_HLT_ASSERT(hdbg.found_root);
+    }
+}
+
+#endif /* ERTS_HLT_HARD_DEBUG */
diff --git a/erts/emulator/beam/erl_hl_timer.h b/erts/emulator/beam/erl_hl_timer.h
new file mode 100644
index 0000000000..30889a71da
--- /dev/null
+++ b/erts/emulator/beam/erl_hl_timer.h
@@ -0,0 +1,80 @@
+/*
+ * %CopyrightBegin%
+ * 
+ * Copyright Ericsson AB 2015. All Rights Reserved.
+ * 
+ * The contents of this file are subject to the Erlang Public License,
+ * Version 1.1, (the "License"); you may not use this file except in
+ * compliance with the License. You should have received a copy of the
+ * Erlang Public License along with this software. If not, it can be
+ * retrieved online at http://www.erlang.org/.
+ * 
+ * Software distributed under the License is distributed on an "AS IS"
+ * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
+ * the License for the specific language governing rights and limitations
+ * under the License.
+ * 
+ * %CopyrightEnd%
+ */
+
+#ifndef ERL_HL_TIMER_H__
+#define ERL_HL_TIMER_H__
+
+typedef struct ErtsHLTimer_ ErtsBifTimers;
+typedef struct ErtsHLTimerService_ ErtsHLTimerService;
+
+#include "sys.h"
+#include "erl_process.h"
+#define ERL_PORT_GET_PORT_TYPE_ONLY__
+#include "erl_port.h"
+#undef ERL_PORT_GET_PORT_TYPE_ONLY__
+#include "erl_message.h"
+#include "erl_alloc_types.h"
+
+#define ERTS_PTMR_NONE ((erts_aint_t) NULL)
+#define ERTS_PTMR_TIMEDOUT (ERTS_PTMR_NONE + ((erts_aint_t) 1))
+
+#define ERTS_PTMR_INIT(P) \
+    erts_smp_atomic_init_nob(&(P)->common.timer, ERTS_PTMR_NONE)
+#define ERTS_PTMR_IS_SET(P) \
+    (ERTS_PTMR_NONE != erts_smp_atomic_read_nob(&(P)->common.timer))
+#define ERTS_PTMR_IS_TIMED_OUT(P) \
+    (ERTS_PTMR_TIMEDOUT == erts_smp_atomic_read_nob(&(P)->common.timer))
+
+#define ERTS_PTMR_CLEAR(P)					\
+    do {							\
+	ASSERT(ERTS_PTMR_IS_TIMED_OUT((P)));			\
+	erts_smp_atomic_set_nob(&(P)->common.timer,		\
+				ERTS_PTMR_NONE);		\
+    } while (0)
+
+size_t erts_timer_type_size(ErtsAlcType_t type);
+int erts_set_proc_timer_term(Process *, Eterm);
+void erts_set_proc_timer_uword(Process *, UWord);
+void erts_cancel_proc_timer(Process *);
+void erts_set_port_timer(Port *, Sint64);
+void erts_cancel_port_timer(Port *);
+Sint64 erts_read_port_timer(Port *);
+int erts_cancel_bif_timers(Process *, ErtsBifTimers *, void **);
+int erts_detach_accessor_bif_timers(Process *, ErtsBifTimers *, void **);
+ErtsHLTimerService *erts_create_timer_service(void);
+void erts_hl_timer_init(void);
+
+#ifdef ERTS_SMP
+void
+erts_handle_canceled_timers(void *vesdp,
+			    int *need_thr_progress,
+			    ErtsThrPrgrVal *thr_prgr_p,
+			    int *need_more_work);
+#endif
+
+Uint erts_bif_timer_memory_size(void);
+void erts_print_bif_timer_info(int to, void *to_arg);
+
+void erts_debug_bif_timer_foreach(void (*func)(Eterm,
+					       Eterm,
+					       ErlHeapFragment *,
+					       void *),
+				  void *arg);
+
+#endif /* ERL_HL_TIMER_H__ */
diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c
index 4f12727044..988ff0e2b5 100644
--- a/erts/emulator/beam/erl_init.c
+++ b/erts/emulator/beam/erl_init.c
@@ -35,7 +35,7 @@
 #include "dist.h"
 #include "erl_mseg.h"
 #include "erl_threads.h"
-#include "erl_bif_timer.h"
+#include "erl_hl_timer.h"
 #include "erl_instrument.h"
 #include "erl_printf_term.h"
 #include "erl_misc_utils.h"
@@ -46,6 +46,8 @@
 #include "erl_async.h"
 #include "erl_ptab.h"
 #include "erl_bif_unique.h"
+#define ERTS_WANT_TIMER_WHEEL_API
+#include "erl_time.h"
 
 #ifdef HIPE
 #include "hipe_mode_switch.h"	/* for hipe_mode_switch_init() */
@@ -365,7 +367,6 @@ erl_init(int ncpu,
     erts_init_binary(); /* Must be after init_emulator() */
     erts_bp_init();
     init_db(); /* Must be after init_emulator */
-    erts_bif_timer_init();
     erts_init_node_tables();
     init_dist();
     erl_drv_thr_init();
@@ -2095,11 +2096,8 @@ erl_start(int argc, char **argv)
 
     erts_initialized = 1;
 
-    {
-	Eterm init = erl_first_process_otp("otp_ring0", NULL, 0,
-					   boot_argc, boot_argv);
-	erts_bif_timer_start_servers(init);
-    }
+    (void) erl_first_process_otp("otp_ring0", NULL, 0,
+				 boot_argc, boot_argv);
 
 #ifdef ERTS_SMP
     erts_start_schedulers();
@@ -2107,13 +2105,17 @@ erl_start(int argc, char **argv)
 
     erts_sys_main_thread(); /* May or may not return! */
 #else
-    erts_thr_set_main_status(1, 1);
+    {
+	ErtsSchedulerData *esdp = erts_get_scheduler_data();
+	erts_thr_set_main_status(1, 1);
 #if ERTS_USE_ASYNC_READY_Q
-    erts_get_scheduler_data()->aux_work_data.async_ready.queue
-	= erts_get_async_ready_queue(1);
+	esdp->aux_work_data.async_ready.queue
+	    = erts_get_async_ready_queue(1);
 #endif
-    set_main_stack_size();
-    process_main();
+	set_main_stack_size();
+	erts_sched_init_time_sup(esdp);
+	process_main();
+    }
 #endif
 }
 
diff --git a/erts/emulator/beam/erl_lock_check.c b/erts/emulator/beam/erl_lock_check.c
index 261460d054..617ce84895 100644
--- a/erts/emulator/beam/erl_lock_check.c
+++ b/erts/emulator/beam/erl_lock_check.c
@@ -91,6 +91,7 @@ static erts_lc_lock_order_t erts_lock_order[] = {
     {   "driver_list",                          NULL                    },
     {	"proc_link",				"pid"			},
     {	"proc_msgq",				"pid"			},
+    {	"proc_btm",				"pid"			},
     {	"dist_entry",				"address"		},
     {	"dist_entry_links",			"address"		},
     {   "code_write_permission",                NULL                    },
diff --git a/erts/emulator/beam/erl_message.c b/erts/emulator/beam/erl_message.c
index 247ea10764..ccfc2e6458 100644
--- a/erts/emulator/beam/erl_message.c
+++ b/erts/emulator/beam/erl_message.c
@@ -990,7 +990,7 @@ erts_send_message(Process* sender,
 #endif
 	    );
         BM_SWAP_TIMER(send,system);
-    } else if (sender == receiver && !(sender->flags & F_OFF_HEAP_MSGS)) {
+    } else if (sender == receiver) {
 	/* Drop message if receiver has a pending exit ... */
 #ifdef ERTS_SMP
 	ErtsProcLocks need_locks = (~(*receiver_locks)
diff --git a/erts/emulator/beam/erl_message.h b/erts/emulator/beam/erl_message.h
index 8f9ea939e8..1e1dafee90 100644
--- a/erts/emulator/beam/erl_message.h
+++ b/erts/emulator/beam/erl_message.h
@@ -213,25 +213,15 @@ do {									\
     if ((M)->data.attached) {						\
 	Uint need__ = erts_msg_attached_data_size((M));			\
  	if ((ST) - (HT) >= need__) {					\
-	    Uint *htop__;						\
-	move__attached__msg__data____:					\
-	    htop__ = (HT);						\
+	    Uint *htop__ = (HT);					\
 	    erts_move_msg_attached_data_to_heap(&htop__, &MSO((P)), (M));\
 	    ASSERT(htop__ - (HT) <= need__);				\
 	    (HT) = htop__;						\
 	}								\
 	else {								\
-	    int off_heap_msgs__ = (int) (P)->flags & F_OFF_HEAP_MSGS;	\
-	    if (!off_heap_msgs__)					\
-		need__ = 0;						\
 	    { SWPO ; }							\
-	    (FC) -= erts_garbage_collect((P), need__, NULL, 0);		\
+	    (FC) -= erts_garbage_collect((P), 0, NULL, 0);		\
 	    { SWPI ; }							\
-	    if (off_heap_msgs__) {					\
-		ASSERT((M)->data.attached);				\
-		ASSERT((ST) - (HT) >= need__);				\
-		goto move__attached__msg__data____;			\
-	    }								\
 	}								\
 	ASSERT(!(M)->data.attached);					\
     }									\
diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c
index 25caaa4e44..426a00304e 100644
--- a/erts/emulator/beam/erl_nif.c
+++ b/erts/emulator/beam/erl_nif.c
@@ -336,7 +336,7 @@ int enif_send(ErlNifEnv* env, const ErlNifPid* to_pid,
     rp = (scheduler
 	  ? erts_proc_lookup(receiver)
 	  : erts_pid2proc_opt(c_p, ERTS_PROC_LOCK_MAIN,
-			      receiver, rp_locks, ERTS_P2P_FLG_SMP_INC_REFC));
+			      receiver, rp_locks, ERTS_P2P_FLG_INC_REFC));
     if (rp == NULL) {
 	ASSERT(env == NULL || receiver != c_p->common.id);
 	return 0;
@@ -364,7 +364,7 @@ int enif_send(ErlNifEnv* env, const ErlNifPid* to_pid,
     if (rp_locks)
 	erts_smp_proc_unlock(rp, rp_locks);
     if (!scheduler)
-	erts_smp_proc_dec_refc(rp);
+	erts_proc_dec_refc(rp);
     if (flush_me) {
 	cache_env(env);
     }
diff --git a/erts/emulator/beam/erl_node_tables.c b/erts/emulator/beam/erl_node_tables.c
index c6d136f951..bcf6311079 100644
--- a/erts/emulator/beam/erl_node_tables.c
+++ b/erts/emulator/beam/erl_node_tables.c
@@ -1469,7 +1469,7 @@ setup_reference_table(void)
     erts_db_foreach_table(insert_ets_table, NULL);
 
     /* Insert all bif timers */
-    erts_bif_timer_foreach(insert_bif_timer, NULL);
+    erts_debug_bif_timer_foreach(insert_bif_timer, NULL);
 
     /* Insert node table (references to dist) */
     hash_foreach(&erts_node_table, insert_erl_node, NULL);
diff --git a/erts/emulator/beam/erl_port.h b/erts/emulator/beam/erl_port.h
index ad3f104a68..3920fae2d9 100644
--- a/erts/emulator/beam/erl_port.h
+++ b/erts/emulator/beam/erl_port.h
@@ -350,6 +350,7 @@ int erts_lc_is_port_locked(Port *);
 ERTS_GLB_INLINE void erts_port_inc_refc(Port *prt);
 ERTS_GLB_INLINE void erts_port_dec_refc(Port *prt);
 ERTS_GLB_INLINE void erts_port_add_refc(Port *prt, Sint32 add_refc);
+ERTS_GLB_INLINE Sint erts_port_read_refc(Port *prt);
 
 ERTS_GLB_INLINE int erts_smp_port_trylock(Port *prt);
 ERTS_GLB_INLINE void erts_smp_port_lock(Port *prt);
@@ -359,37 +360,26 @@ ERTS_GLB_INLINE void erts_smp_port_unlock(Port *prt);
 
 ERTS_GLB_INLINE void erts_port_inc_refc(Port *prt)
 {
-#ifdef ERTS_SMP
-    erts_ptab_inc_refc(&prt->common);
-#else
-    erts_atomic32_inc_nob(&prt->refc);
-#endif
+    erts_ptab_atmc_inc_refc(&prt->common);
 }
 
 ERTS_GLB_INLINE void erts_port_dec_refc(Port *prt)
 {
-#ifdef ERTS_SMP
-    int referred = erts_ptab_dec_test_refc(&prt->common);
+    int referred = erts_ptab_atmc_dec_test_refc(&prt->common);
     if (!referred)
 	erts_port_free(prt);
-#else
-    int refc = erts_atomic32_dec_read_nob(&prt->refc);
-    if (refc == 0)
-	erts_port_free(prt);	
-#endif
 }
 
 ERTS_GLB_INLINE void erts_port_add_refc(Port *prt, Sint32 add_refc)
 {
-#ifdef ERTS_SMP
-    int referred = erts_ptab_add_test_refc(&prt->common, add_refc);
+    int referred = erts_ptab_atmc_add_test_refc(&prt->common, add_refc);
     if (!referred)
 	erts_port_free(prt);
-#else
-    int refc = erts_atomic32_add_read_nob(&prt->refc, add_refc);
-    if (refc == 0)
-	erts_port_free(prt);	
-#endif
+}
+
+ERTS_GLB_INLINE Sint erts_port_read_refc(Port *prt)
+{
+    return erts_ptab_atmc_read_refc(&prt->common);
 }
 
 ERTS_GLB_INLINE int
diff --git a/erts/emulator/beam/erl_port_task.c b/erts/emulator/beam/erl_port_task.c
index 2aa0a27197..c701737e26 100644
--- a/erts/emulator/beam/erl_port_task.c
+++ b/erts/emulator/beam/erl_port_task.c
@@ -1646,6 +1646,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp)
     erts_aint32_t state;
     int active;
     Uint64 start_time = 0;
+    ErtsSchedulerData *esdp = runq->scheduler;
 
     ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(runq));
 
@@ -1662,7 +1663,6 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp)
     *curr_port_pp = pp;
     
     if (erts_sched_stat.enabled) {
-	ErtsSchedulerData *esdp = erts_get_scheduler_data();
 	Uint old = ERTS_PORT_SCHED_ID(pp, esdp->no);
 	int migrated = old && old != esdp->no;
 
@@ -1718,11 +1718,16 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp)
 	switch (ptp->type) {
 	case ERTS_PORT_TASK_TIMEOUT:
 	    reset_handle(ptp);
-	    reds = ERTS_PORT_REDS_TIMEOUT;
-	    if (!(state & ERTS_PORT_SFLGS_DEAD)) {
-                DTRACE_DRIVER(driver_timeout, pp);
-		(*pp->drv_ptr->timeout)((ErlDrvData) pp->drv_data);
-            }
+	    if (!ERTS_PTMR_IS_TIMED_OUT(pp))
+		reds = 0;
+	    else {
+		ERTS_PTMR_CLEAR(pp);
+		reds = ERTS_PORT_REDS_TIMEOUT;
+		if (!(state & ERTS_PORT_SFLGS_DEAD)) {
+		    DTRACE_DRIVER(driver_timeout, pp);
+		    (*pp->drv_ptr->timeout)((ErlDrvData) pp->drv_data);
+		}
+	    }
 	    break;
 	case ERTS_PORT_TASK_INPUT:
 	    reds = ERTS_PORT_REDS_INPUT;
@@ -1879,7 +1884,7 @@ erts_port_task_execute(ErtsRunQueue *runq, Port **curr_port_pp)
     runq->scheduler->reductions += reds;
 
     ERTS_SMP_LC_ASSERT(erts_smp_lc_runq_is_locked(runq));
-    ERTS_PORT_REDUCTIONS_EXECUTED(runq, reds);
+    ERTS_PORT_REDUCTIONS_EXECUTED(esdp, runq, reds);
 
     return res;
 }
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 00c7b163c2..af8db519d4 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -44,8 +44,10 @@
 #include "dtrace-wrapper.h"
 #include "erl_ptab.h"
 #include "erl_bif_unique.h"
+#define ERTS_WANT_TIMER_WHEEL_API
+#include "erl_time.h"
 
-
+#define ERTS_CHECK_TIME_REDS CONTEXT_REDS
 #define ERTS_DELAYED_WAKEUP_INFINITY (~(Uint64) 0)
 #define ERTS_DELAYED_WAKEUP_REDUCTIONS ((Uint64) CONTEXT_REDS/2)
 
@@ -463,7 +465,7 @@ static int stack_element_dump(int to, void *to_arg, Eterm* sp, int yreg);
 static void aux_work_timeout(void *unused);
 static void aux_work_timeout_early_init(int no_schedulers);
 static void aux_work_timeout_late_init(void);
-static void setup_aux_work_timer(void);
+static void setup_aux_work_timer(ErtsSchedulerData *esdp);
 
 static int execute_sys_tasks(Process *c_p,
 			     erts_aint32_t *statep,
@@ -493,6 +495,8 @@ dbg_chk_aux_work_val(erts_aint32_t value)
     valid |= ERTS_SSI_AUX_WORK_MISC_THR_PRGR;
     valid |= ERTS_SSI_AUX_WORK_DD;
     valid |= ERTS_SSI_AUX_WORK_DD_THR_PRGR;
+    valid |= ERTS_SSI_AUX_WORK_CNCLD_TMRS;
+    valid |= ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR;
     valid |= ERTS_SSI_AUX_WORK_THR_PRGR_LATER_OP;
 #endif
 #if HAVE_ERTS_MSEG
@@ -610,13 +614,11 @@ erts_pre_init_process(void)
 #endif
 }
 
-#ifdef ERTS_SMP
 static void
 release_process(void *vproc)
 {
-    erts_smp_proc_dec_refc((Process *) vproc);
+    erts_proc_dec_refc((Process *) vproc);
 }
-#endif
 
 /* initialize the scheduler */
 void
@@ -632,16 +634,18 @@ erts_init_process(int ncpu, int proc_tab_size, int legacy_proc_tab)
 
     erts_ptab_init_table(&erts_proc,
 			 ERTS_ALC_T_PROC_TABLE,
-#ifdef ERTS_SMP
 			 release_process,
-#else
-			 NULL,
-#endif
 			 (ErtsPTabElementCommon *) &erts_invalid_process.common,
 			 proc_tab_size,
 			 sizeof(Process),
 			 "process_table",
-			 legacy_proc_tab);
+			 legacy_proc_tab,
+#ifdef ERTS_SMP
+			 1
+#else
+			 0
+#endif
+	);
 
     last_reductions = 0;
     last_exact_reductions = 0;
@@ -1031,7 +1035,7 @@ reply_sched_wall_time(void *vswtrp)
     if (rp_locks)
 	erts_smp_proc_unlock(rp, rp_locks);
 
-    erts_smp_proc_dec_refc(rp);
+    erts_proc_dec_refc(rp);
 
     if (erts_smp_atomic32_dec_read_nob(&swtrp->refc) == 0)
 	swtreq_free(vswtrp);
@@ -1063,7 +1067,7 @@ erts_sched_wall_time_request(Process *c_p, int set, int enable)
     erts_smp_atomic32_init_nob(&swtrp->refc,
 			       (erts_aint32_t) erts_no_schedulers);
 
-    erts_smp_proc_add_refc(c_p, (Sint32) erts_no_schedulers);
+    erts_proc_add_refc(c_p, (Sint32) erts_no_schedulers);
 
 #ifdef ERTS_SMP
     if (erts_no_schedulers > 1)
@@ -1124,7 +1128,7 @@ erts_psd_set_init(Process *p, ErtsProcLocks plocks, int ix, void *data)
     xplocks &= ~plocks;
     if (xplocks && erts_smp_proc_trylock(p, xplocks) == EBUSY) {
 	if (xplocks & ERTS_PROC_LOCK_MAIN) {
-	    erts_smp_proc_inc_refc(p);
+	    erts_proc_inc_refc(p);
 	    erts_smp_proc_unlock(p, plocks);
 	    erts_smp_proc_lock(p, ERTS_PROC_LOCKS_ALL);
 	    refc = 1;
@@ -1140,7 +1144,7 @@ erts_psd_set_init(Process *p, ErtsProcLocks plocks, int ix, void *data)
     if (xplocks)
 	erts_smp_proc_unlock(p, xplocks);
     if (refc)
-	erts_smp_proc_dec_refc(p);
+	erts_proc_dec_refc(p);
     ASSERT(p->psd);
     if (p->psd != psd)
 	erts_free(ERTS_ALC_T_PSD, psd);
@@ -1768,6 +1772,101 @@ handle_delayed_dealloc_thr_prgr(ErtsAuxWorkData *awdp, erts_aint32_t aux_work, i
 }
 
 /*
+ * Canceled timers
+ */
+
+void
+erts_notify_canceled_timer(ErtsSchedulerData *esdp, int rsid)
+{
+    ASSERT(esdp && esdp == erts_get_scheduler_data());
+    if (esdp && !ERTS_SCHEDULER_IS_DIRTY(esdp))
+	schedule_aux_work_wakeup(&esdp->aux_work_data,
+				 rsid,
+				 ERTS_SSI_AUX_WORK_CNCLD_TMRS);
+    else
+	set_aux_work_flags_wakeup_relb(ERTS_SCHED_SLEEP_INFO_IX(rsid-1),
+				       ERTS_SSI_AUX_WORK_CNCLD_TMRS);
+}
+
+static ERTS_INLINE erts_aint32_t
+handle_canceled_timers(ErtsAuxWorkData *awdp, erts_aint32_t aux_work, int waiting)
+{
+    ErtsSchedulerSleepInfo *ssi = awdp->ssi;
+    int need_thr_progress = 0;
+    ErtsThrPrgrVal wakeup = ERTS_THR_PRGR_INVALID;
+    int more_work = 0;
+
+#ifdef ERTS_DIRTY_SCHEDULERS
+    ASSERT(!awdp->esdp || !ERTS_SCHEDULER_IS_DIRTY(awdp->esdp));
+#endif
+    unset_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_CNCLD_TMRS);
+    erts_handle_canceled_timers((void *) awdp->esdp,
+				&need_thr_progress,
+				&wakeup,
+				&more_work);
+    if (more_work) {
+	if (set_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_CNCLD_TMRS)
+	    & ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR) {
+	    unset_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR);
+	    aux_work &= ~ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR;
+	}
+	return aux_work;
+    }
+
+    if (need_thr_progress) {
+	if (wakeup == ERTS_THR_PRGR_INVALID)
+	    wakeup = erts_thr_progress_later(awdp->esdp);
+	awdp->cncld_tmrs.thr_prgr = wakeup;
+	set_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR);
+	haw_thr_prgr_soft_wakeup(awdp, wakeup);
+    }
+    return aux_work & ~ERTS_SSI_AUX_WORK_CNCLD_TMRS;
+}
+
+static ERTS_INLINE erts_aint32_t
+handle_canceled_timers_thr_prgr(ErtsAuxWorkData *awdp, erts_aint32_t aux_work, int waiting)
+{
+    ErtsSchedulerSleepInfo *ssi;
+    int need_thr_progress;
+    int more_work;
+    ErtsThrPrgrVal wakeup = ERTS_THR_PRGR_INVALID;
+    ErtsThrPrgrVal current = haw_thr_prgr_current(awdp);
+
+#ifdef ERTS_DIRTY_SCHEDULERS
+    ASSERT(!awdp->esdp || !ERTS_SCHEDULER_IS_DIRTY(awdp->esdp));
+#endif
+    if (!erts_thr_progress_has_reached_this(current, awdp->cncld_tmrs.thr_prgr))
+	return aux_work & ~ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR;
+
+    ssi = awdp->ssi;
+    need_thr_progress = 0;
+    more_work = 0;
+
+    erts_handle_canceled_timers((void *) awdp->esdp,
+				&need_thr_progress,
+				&wakeup,
+				&more_work);
+    if (more_work) {
+	set_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_CNCLD_TMRS);
+	unset_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR);
+	return ((aux_work & ~ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR)
+		| ERTS_SSI_AUX_WORK_CNCLD_TMRS);
+    }
+
+    if (need_thr_progress) {
+	if (wakeup == ERTS_THR_PRGR_INVALID)
+	    wakeup = erts_thr_progress_later(awdp->esdp);
+	awdp->cncld_tmrs.thr_prgr = wakeup;
+	haw_thr_prgr_soft_wakeup(awdp, wakeup);
+    }
+    else {
+	unset_aux_work_flags(ssi, ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR);
+    }
+
+    return aux_work & ~ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR;
+}
+
+/*
  * Handle scheduled thread progress later operations.
  */
 #define ERTS_MAX_THR_PRGR_LATER_OPS 50
@@ -1865,7 +1964,7 @@ completed_dealloc(void *vproc)
 {
     if (erts_atomic32_dec_read_mb(&completed_dealloc_count) == 0) {
 	erts_resume((Process *) vproc, (ErtsProcLocks) 0);
-	erts_smp_proc_dec_refc((Process *) vproc);
+	erts_proc_dec_refc((Process *) vproc);
     }
 }
 
@@ -1914,7 +2013,7 @@ erts_debug_wait_deallocations(Process *c_p)
 				      count,
 				      0)) {
 	erts_suspend(c_p, ERTS_PROC_LOCK_MAIN, NULL);
-	erts_smp_proc_inc_refc(c_p);
+	erts_proc_inc_refc(c_p);
 	/* scheduler threads */
 	erts_schedule_multi_misc_aux_work(0,
 					  erts_no_schedulers,
@@ -2029,7 +2128,7 @@ static ERTS_INLINE erts_aint32_t
 handle_setup_aux_work_timer(ErtsAuxWorkData *awdp, erts_aint32_t aux_work, int waiting)
 {
     unset_aux_work_flags(awdp->ssi, ERTS_SSI_AUX_WORK_SET_TMO);
-    setup_aux_work_timer();
+    setup_aux_work_timer(awdp->esdp);
     return aux_work & ~ERTS_SSI_AUX_WORK_SET_TMO;
 }
 
@@ -2089,6 +2188,11 @@ handle_aux_work(ErtsAuxWorkData *awdp, erts_aint32_t orig_aux_work, int waiting)
 #ifdef ERTS_SMP
     HANDLE_AUX_WORK(ERTS_SSI_AUX_WORK_THR_PRGR_LATER_OP,
 		    handle_thr_prgr_later_op);
+    HANDLE_AUX_WORK(ERTS_SSI_AUX_WORK_CNCLD_TMRS,
+		    handle_canceled_timers);
+    /* CNCLD_TMRS must be before CNCLD_TMRS_THR_PRGR */
+    HANDLE_AUX_WORK(ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR,
+		    handle_canceled_timers_thr_prgr);
 #endif
 
 #if ERTS_USE_ASYNC_READY_Q
@@ -2138,8 +2242,8 @@ handle_aux_work(ErtsAuxWorkData *awdp, erts_aint32_t orig_aux_work, int waiting)
 
 typedef struct {
     union {
-	ErlTimer data;
-	char align__[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(ErlTimer))];
+	ErtsTWheelTimer data;
+	char align__[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(ErtsTWheelTimer))];
     } timer;
 
     int initialized;
@@ -2149,6 +2253,22 @@ typedef struct {
 
 static ErtsAuxWorkTmo *aux_work_tmo;
 
+static ERTS_INLINE void
+start_aux_work_timer(ErtsSchedulerData *esdp)
+{
+    ErtsMonotonicTime tmo = erts_get_monotonic_time(esdp);
+    tmo = ERTS_MONOTONIC_TO_CLKTCKS(tmo-1);
+    tmo += ERTS_MSEC_TO_CLKTCKS(1000) + 1;
+    erts_twheel_init_timer(&aux_work_tmo->timer.data);
+    ASSERT(esdp);
+    erts_twheel_set_timer(esdp->timer_wheel,
+			  &aux_work_tmo->timer.data,
+			  aux_work_timeout,
+			  NULL,
+			  (void *) esdp,
+			  tmo);
+}
+
 static void
 aux_work_timeout_early_init(int no_schedulers)
 {
@@ -2181,18 +2301,12 @@ void
 aux_work_timeout_late_init(void)
 {
     aux_work_tmo->initialized = 1;
-    if (erts_atomic32_read_nob(&aux_work_tmo->refc)) {
-	erts_init_timer(&aux_work_tmo->timer.data);
-	erts_set_timer(&aux_work_tmo->timer.data,
-		       aux_work_timeout,
-		       NULL,
-		       NULL,
-		       1000);
-    }
+    if (erts_atomic32_read_nob(&aux_work_tmo->refc))
+	start_aux_work_timer(erts_get_scheduler_data());
 }
 
 static void
-aux_work_timeout(void *unused)
+aux_work_timeout(void *vesdp)
 {
     erts_aint32_t refc;
     int i;
@@ -2215,31 +2329,18 @@ aux_work_timeout(void *unused)
     if (refc != 1
 	|| 1 != erts_atomic32_cmpxchg_relb(&aux_work_tmo->refc, 0, 1)) {
 	/* Setup next timeout... */
-	erts_set_timer(&aux_work_tmo->timer.data,
-		       aux_work_timeout,
-		       NULL,
-		       NULL,
-		       1000);
+	start_aux_work_timer((ErtsSchedulerData *) vesdp);
     }
 }
 
 static void
-setup_aux_work_timer(void)
+setup_aux_work_timer(ErtsSchedulerData *esdp)
 {
-#ifndef ERTS_SMP
-    if (!erts_get_scheduler_data())
+    if (!esdp || !esdp->timer_wheel)
 	set_aux_work_flags_wakeup_nob(ERTS_SCHED_SLEEP_INFO_IX(0),
 				      ERTS_SSI_AUX_WORK_SET_TMO);
     else
-#endif
-    {
-	erts_init_timer(&aux_work_tmo->timer.data);
-	erts_set_timer(&aux_work_tmo->timer.data,
-		       aux_work_timeout,
-		       NULL,
-		       NULL,
-		       1000);
-    }
+	start_aux_work_timer(esdp);
 }
 
 erts_aint32_t
@@ -2270,7 +2371,7 @@ erts_set_aux_work_timeout(int ix, erts_aint32_t type, int enable)
 	if (refc == 1) {
 	    erts_atomic32_inc_acqb(&aux_work_tmo->refc);
 	    if (aux_work_tmo->initialized) 
-		setup_aux_work_timer();
+		setup_aux_work_timer(erts_get_scheduler_data());
 	}
     }
     return old;
@@ -2649,11 +2750,6 @@ aux_thread(void *unused)
     erts_aint32_t aux_work;
     ErtsThrPrgrCallbacks callbacks;
     int thr_prgr_active = 1;
-    ErtsTimerWheel *timer_wheel = erts_default_timer_wheel;
-    ErtsNextTimeoutRef nxt_tmo_ref = erts_get_next_timeout_reference(timer_wheel);
-
-    if (!timer_wheel)
-	ERTS_INTERNAL_ERROR("Missing aux timer wheel");
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
     {
@@ -2677,7 +2773,6 @@ aux_thread(void *unused)
     sched_prep_spin_wait(ssi);
 
     while (1) {
-	ErtsMonotonicTime current_time;
 	erts_aint32_t flgs;
 
 	aux_work = erts_atomic32_read_acqb(&ssi->aux_work);
@@ -2689,56 +2784,28 @@ aux_thread(void *unused)
 		erts_thr_progress_leader_update(NULL);
 	}
 
-	if (aux_work) {
-	    current_time = erts_get_monotonic_time();
-	    if (current_time >= erts_next_timeout_time(nxt_tmo_ref)) {
-		if (!thr_prgr_active)
-		    erts_thr_progress_active(NULL, thr_prgr_active = 1);
-		erts_bump_timers(timer_wheel, current_time);
-	    }
-	}
-	else {
-	    ErtsMonotonicTime timeout_time;
-	    timeout_time = erts_check_next_timeout_time(timer_wheel,
-							ERTS_SEC_TO_MONOTONIC(10*60));
-	    current_time = erts_get_monotonic_time();
-	    if (current_time >= timeout_time) {
-		if (!thr_prgr_active)
-		    erts_thr_progress_active(NULL, thr_prgr_active = 1);
-	    }
-	    else {
-		if (thr_prgr_active)
-		    erts_thr_progress_active(NULL, thr_prgr_active = 0);
-		erts_thr_progress_prepare_wait(NULL);
+	if (!aux_work) {
+	    if (thr_prgr_active)
+		erts_thr_progress_active(NULL, thr_prgr_active = 0);
+	    erts_thr_progress_prepare_wait(NULL);
 
-		ERTS_SCHED_FAIR_YIELD();
+	    ERTS_SCHED_FAIR_YIELD();
 
-		flgs = sched_spin_wait(ssi, 0);
+	    flgs = sched_spin_wait(ssi, 0);
 
+	    if (flgs & ERTS_SSI_FLG_SLEEPING) {
+		ASSERT(flgs & ERTS_SSI_FLG_WAITING);
+		flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING);
 		if (flgs & ERTS_SSI_FLG_SLEEPING) {
+		    int res;
+		    ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING);
 		    ASSERT(flgs & ERTS_SSI_FLG_WAITING);
-		    flgs = sched_set_sleeptype(ssi, ERTS_SSI_FLG_TSE_SLEEPING);
-		    if (flgs & ERTS_SSI_FLG_SLEEPING) {
-			int res;
-			ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING);
-			ASSERT(flgs & ERTS_SSI_FLG_WAITING);
-			current_time = erts_get_monotonic_time();
-			do {
-			    Sint64 timeout;
-			    if (current_time >= timeout_time)
-				break;
-			    timeout = ERTS_MONOTONIC_TO_NSEC(timeout_time
-							     - current_time
-							     - 1) + 1;
-			    res = erts_tse_twait(ssi->event, timeout);
-			    current_time = erts_get_monotonic_time();
-			} while (res == EINTR);
-		    }
+		    do {
+			res = erts_tse_wait(ssi->event);
+		    } while (res == EINTR);
 		}
-		erts_thr_progress_finalize_wait(NULL);
 	    }
-	    if (current_time >= timeout_time)
-		erts_bump_timers(timer_wheel, current_time);
+	    erts_thr_progress_finalize_wait(NULL);
 	}
 
 	flgs = sched_prep_spin_wait(ssi);
@@ -2821,7 +2888,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 
 	    if (aux_work) {
 		flgs = erts_smp_atomic32_read_acqb(&ssi->flags);
-		current_time = erts_get_monotonic_time();
+		current_time = erts_get_monotonic_time(esdp);
 		if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) {
 		    if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && !thr_prgr_active) {
 			erts_thr_progress_active(esdp, thr_prgr_active = 1);
@@ -2832,9 +2899,8 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 	    }
 	    else {
 		ErtsMonotonicTime timeout_time;
-		timeout_time = erts_check_next_timeout_time(esdp->timer_wheel,
-							    ERTS_SEC_TO_MONOTONIC(10*60));
-		current_time = erts_get_monotonic_time();
+		timeout_time = erts_check_next_timeout_time(esdp);
+		current_time = erts_get_monotonic_time(esdp);
 		if (current_time >= timeout_time) {
 		    if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && !thr_prgr_active) {
 			erts_thr_progress_active(esdp, thr_prgr_active = 1);
@@ -2860,7 +2926,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 			    int res;
 			    ASSERT(flgs & ERTS_SSI_FLG_TSE_SLEEPING);
 			    ASSERT(flgs & ERTS_SSI_FLG_WAITING);
-			    current_time = erts_get_monotonic_time();
+			    current_time = erts_get_monotonic_time(esdp);
 			    do {
 				Sint64 timeout;
 				if (current_time >= timeout_time)
@@ -2869,7 +2935,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 								 - current_time
 								 - 1) + 1;
 				res = erts_tse_twait(ssi->event, timeout);
-				current_time = erts_get_monotonic_time();
+				current_time = erts_get_monotonic_time(esdp);
 			    } while (res == EINTR);
 			}
 		    }
@@ -2944,7 +3010,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 	    ASSERT(!erts_port_task_have_outstanding_io_tasks());
 	    erl_sys_schedule(1); /* Might give us something to do */
 
-	    current_time = erts_get_monotonic_time();
+	    current_time = erts_get_monotonic_time(esdp);
 	    if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref))
 		erts_bump_timers(esdp->timer_wheel, current_time);
 
@@ -3062,7 +3128,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 	erl_sys_schedule(0);
 
 	{
-	    ErtsMonotonicTime current_time = erts_get_monotonic_time();
+	    ErtsMonotonicTime current_time = erts_get_monotonic_time(esdp);
 	    if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref))
 		erts_bump_timers(esdp->timer_wheel, current_time);
 	}
@@ -5273,6 +5339,7 @@ init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp, char *dawwp)
     awdp->dd.thr_prgr = ERTS_THR_PRGR_VAL_WAITING;
     awdp->dd.completed_callback = NULL;
     awdp->dd.completed_arg = NULL;
+    awdp->cncld_tmrs.thr_prgr = ERTS_THR_PRGR_VAL_WAITING;
     awdp->later_op.thr_prgr = ERTS_THR_PRGR_VAL_FIRST;
     awdp->later_op.size = 0;
     awdp->later_op.first = NULL;
@@ -5338,9 +5405,6 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num,
     esdp->no = (Uint) num;
 #endif
 
-    esdp->timer_wheel = erts_default_timer_wheel;
-    esdp->next_tmo_ref = erts_get_next_timeout_reference(esdp->timer_wheel);
-
     esdp->ssi = ssi;
     esdp->current_process = NULL;
     esdp->current_port = NULL;
@@ -5353,6 +5417,9 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num,
     esdp->run_queue = runq;
     esdp->run_queue->scheduler = esdp;
 
+    esdp->last_monotonic_time = 0;
+    esdp->check_time_reds = 0;
+
     esdp->thr_id = (Uint32) num;
     erts_sched_bif_unique_init(esdp);
 
@@ -5916,13 +5983,6 @@ schedule_out_process(ErtsRunQueue *c_rq, erts_aint32_t state, Process *p, Proces
     int check_emigration_need;
 #endif
 
-#ifdef ERTS_SMP
-    if ((p->static_flags & ERTS_STC_FLG_PREFER_SCHED)
-	&& p->preferred_run_queue != RUNQ_READ_RQ(&p->run_queue)) {
-	RUNQ_SET_RQ(&p->run_queue, p->preferred_run_queue);
-    }
-#endif
-
     a = state;
 
     while (1) {
@@ -6730,6 +6790,7 @@ suspend_scheduler(ErtsSchedulerData *esdp)
 	}
     }
 
+    (void) erts_get_monotonic_time(esdp);
     erts_smp_runq_lock(esdp->run_queue);
     non_empty_runq(esdp->run_queue);
 
@@ -6863,7 +6924,7 @@ suspend_scheduler(ErtsSchedulerData *esdp)
 		}
 
 		if (aux_work) {
-		    current_time = erts_get_monotonic_time();
+		    current_time = erts_get_monotonic_time(esdp);
 		    if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) {
 			if (!thr_prgr_active) {
 			    erts_thr_progress_active(esdp, thr_prgr_active = 1);
@@ -6874,9 +6935,8 @@ suspend_scheduler(ErtsSchedulerData *esdp)
 		}
 		else {
 		    ErtsMonotonicTime timeout_time;
-		    timeout_time = erts_check_next_timeout_time(esdp->timer_wheel,
-								ERTS_SEC_TO_MONOTONIC(60*60));
-		    current_time = erts_get_monotonic_time();
+		    timeout_time = erts_check_next_timeout_time(esdp);
+		    current_time = erts_get_monotonic_time(esdp);
 
 		    if (current_time >= timeout_time) {
 			if (!thr_prgr_active) {
@@ -6902,7 +6962,7 @@ suspend_scheduler(ErtsSchedulerData *esdp)
 					 | ERTS_SSI_FLG_SUSPENDED)) {
 				int res;
 
-				current_time = erts_get_monotonic_time();
+				current_time = erts_get_monotonic_time(esdp);
 				do {
 				    Sint64 timeout;
 				    if (current_time >= timeout_time)
@@ -6911,7 +6971,7 @@ suspend_scheduler(ErtsSchedulerData *esdp)
 								     - current_time
 								     - 1) + 1;
 				    res = erts_tse_twait(ssi->event, timeout);
-				    current_time = erts_get_monotonic_time();
+				    current_time = erts_get_monotonic_time(esdp);
 				} while (res == EINTR);
 			    }
 			}
@@ -7739,8 +7799,8 @@ sched_thread_func(void *vesdp)
     ErtsSchedulerData *esdp = vesdp;
     Uint no = esdp->no;
 
-    esdp->timer_wheel = erts_create_timer_wheel((int) no);
-    esdp->next_tmo_ref = erts_get_next_timeout_reference(esdp->timer_wheel);
+    erts_sched_init_time_sup(esdp);
+
 #ifdef ERTS_SMP
     ERTS_SCHED_SLEEP_INFO_IX(no - 1)->event = erts_tse_fetch();
     callbacks.arg = (void *) esdp->ssi;
@@ -9109,7 +9169,7 @@ Process *schedule(Process *p, int calls)
 	schedule_out_process(rq, state, p, proxy_p); /* Returns with rq locked! */
 	proxy_p = NULL;
 
-	ERTS_PROC_REDUCTIONS_EXECUTED(rq,
+	ERTS_PROC_REDUCTIONS_EXECUTED(esdp, rq,
 				      (int) ERTS_PSFLGS_GET_USR_PRIO(state),
 				      reds,
 				      actual_reds);
@@ -9126,12 +9186,9 @@ Process *schedule(Process *p, int calls)
 	    ASSERT(esdp->free_process == p);
 	    esdp->free_process = NULL;
 #else
-	    state = erts_smp_atomic32_read_nob(&p->state);
-	    if (!(state & ERTS_PSFLG_IN_RUNQ))
-		erts_free_proc(p);
+	    erts_proc_dec_refc(p);
 #endif
 	}
-
 #ifdef ERTS_SMP
 	ASSERT(!esdp->free_process);
 #endif
@@ -9139,13 +9196,13 @@ Process *schedule(Process *p, int calls)
 
 	ERTS_SMP_CHK_NO_PROC_LOCKS;
 
-	{
-	    ErtsMonotonicTime current_time = erts_get_monotonic_time();
-	    if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref)) {
-		erts_smp_runq_unlock(rq);
-		erts_bump_timers(esdp->timer_wheel, current_time);
-		erts_smp_runq_lock(rq);
-	    }
+	if (esdp->check_time_reds >= ERTS_CHECK_TIME_REDS)
+	    (void) erts_get_monotonic_time(esdp);
+
+	if (esdp->last_monotonic_time >= erts_next_timeout_time(esdp->next_tmo_ref)) {
+	    erts_smp_runq_unlock(rq);
+	    erts_bump_timers(esdp->timer_wheel, esdp->last_monotonic_time);
+	    erts_smp_runq_lock(rq);
 	}
 	BM_STOP_TIMER(system);
 
@@ -9305,7 +9362,7 @@ Process *schedule(Process *p, int calls)
 	    erts_smp_runq_unlock(rq);
 	    erl_sys_schedule(1);
 
-	    current_time = erts_get_monotonic_time();
+	    current_time = erts_get_monotonic_time(esdp);
 	    if (current_time >= erts_next_timeout_time(esdp->next_tmo_ref))
 		erts_bump_timers(esdp->timer_wheel, current_time);
 
@@ -9446,13 +9503,8 @@ Process *schedule(Process *p, int calls)
 				      | ERTS_PSFLG_PENDING_EXIT
 				      | ERTS_PSFLG_ACTIVE_SYS))
 			    == ERTS_PSFLG_SUSPENDED)) {
-			if (state & ERTS_PSFLG_FREE) {
-#ifdef ERTS_SMP
-			    erts_smp_proc_dec_refc(p);
-#else
-			    erts_free_proc(p);
-#endif
-			}
+			if (state & ERTS_PSFLG_FREE)
+			    erts_proc_dec_refc(p);
 			if (proxy_p) {
 			    free_proxy_proc(proxy_p);
 			    proxy_p = NULL;
@@ -9595,6 +9647,20 @@ Process *schedule(Process *p, int calls)
 	/* Never run a suspended process */
 	ASSERT(!(ERTS_PSFLG_SUSPENDED & erts_smp_atomic32_read_nob(&p->state)));
 
+	ASSERT(erts_proc_read_refc(p) > 0);
+
+	if (ERTS_PTMR_IS_TIMED_OUT(p)) {
+	    BeamInstr** pi;
+#ifdef ERTS_SMP
+	    ETHR_MEMBAR(ETHR_LoadLoad|ETHR_LoadStore);
+#endif
+	    pi = (BeamInstr **) p->def_arg_reg;
+	    p->i = *pi;
+	    p->flags &= ~F_INSLPQUEUE;
+	    p->flags |= F_TIMO;
+	    ERTS_PTMR_CLEAR(p);
+	}
+
 	return p;
     }
 }
@@ -10503,6 +10569,8 @@ erts_free_proc(Process *p)
 #ifdef ERTS_SMP
     erts_proc_lock_fin(p);
 #endif
+    ASSERT(erts_smp_atomic32_read_nob(&p->state) & ERTS_PSFLG_FREE);
+    ASSERT(0 == erts_proc_read_refc(p));
     erts_free(ERTS_ALC_T_PROC, (void *) p);
 }
 
@@ -10555,6 +10623,8 @@ alloc_process(ErtsRunQueue *rq, erts_aint32_t state)
 	return NULL;
     }
 
+    ASSERT(erts_proc_read_refc(p) > 0);
+
     ASSERT(internal_pid_serial(p->common.id) <= ERTS_MAX_PID_SERIAL);
     
     p->approx_started = erts_get_approx_time();
@@ -10604,10 +10674,8 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
 	    int ix = so->scheduler-1;
 	    ASSERT(0 <= ix && ix < erts_no_run_queues);
 	    rq = ERTS_RUNQ_IX(ix);
-	    if (!(so->flags & SPO_PREFER_SCHED)) {
-		/* Unsupported feature... */
-		state |= ERTS_PSFLG_BOUND;
-	    }
+	    /* Unsupported feature... */
+	    state |= ERTS_PSFLG_BOUND;
 	}
 	prio = (erts_aint32_t) so->priority;
     }
@@ -10615,9 +10683,6 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
     state |= (((prio & ERTS_PSFLGS_PRIO_MASK) << ERTS_PSFLGS_ACT_PRIO_OFFSET)
 	      | ((prio & ERTS_PSFLGS_PRIO_MASK) << ERTS_PSFLGS_USR_PRIO_OFFSET));
 
-    if (so->flags & SPO_OFF_HEAP_MSGS)
-	state |= ERTS_PSFLG_OFF_HEAP_MSGS;
-
     if (!rq)
 	rq = erts_get_runq_proc(parent);
 
@@ -10641,12 +10706,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
     heap_need = arg_size;
 
     p->flags = erts_default_process_flags;
-    if (so->flags & SPO_OFF_HEAP_MSGS)
-	p->flags |= F_OFF_HEAP_MSGS;
 
-#ifdef ERTS_SMP
-    p->preferred_run_queue = NULL;
-#endif
     p->static_flags = 0;
     if (so->flags & SPO_SYSTEM_PROC)
 	p->static_flags |= ERTS_STC_FLG_SYSTEM_PROC;
@@ -10654,12 +10714,6 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
 	p->min_heap_size  = so->min_heap_size;
 	p->min_vheap_size = so->min_vheap_size;
 	p->max_gen_gcs    = so->max_gen_gcs;
-	if (so->flags & SPO_PREFER_SCHED) {
-#ifdef ERTS_SMP
-	    p->preferred_run_queue = rq;
-#endif
-	    p->static_flags |= ERTS_STC_FLG_PREFER_SCHED;
-	}
     } else {
 	p->min_heap_size  = H_MIN_SIZE;
 	p->min_vheap_size = BIN_VH_MIN_SIZE;
@@ -10668,9 +10722,9 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
     p->schedule_count = 0;
     ASSERT(p->min_heap_size == erts_next_heap_size(p->min_heap_size, 0));
     
-    p->initial[INITIAL_MOD] = mod;
-    p->initial[INITIAL_FUN] = func;
-    p->initial[INITIAL_ARI] = (Uint) arity;
+    p->u.initial[INITIAL_MOD] = mod;
+    p->u.initial[INITIAL_FUN] = func;
+    p->u.initial[INITIAL_ARI] = (Uint) arity;
 
     /*
      * Must initialize binary lists here before copying binaries to process.
@@ -10711,7 +10765,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
 
     /* No need to initialize p->fcalls. */
 
-    p->current = p->initial+INITIAL_MOD;
+    p->current = p->u.initial+INITIAL_MOD;
 
     p->i = (BeamInstr *) beam_apply;
     p->cp = (BeamInstr *) beam_apply+1;
@@ -10734,11 +10788,7 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
     p->ftrace = NIL;
     p->reds = 0;
 
-#ifdef ERTS_SMP
-    p->common.u.alive.ptimer = NULL;
-#else
-    erts_init_timer(&p->common.u.alive.tm);
-#endif
+    ERTS_PTMR_INIT(p);
 
     p->common.u.alive.reg = NULL;
     ERTS_P_LINKS(p) = NULL;
@@ -10769,7 +10819,8 @@ erl_create_process(Process* parent, /* Parent of process (default group leader).
     p->msg_inq.last = &p->msg_inq.first;
     p->msg_inq.len = 0;
 #endif
-    p->u.bif_timers = NULL;
+    p->bif_timers = NULL;
+    p->accessor_bif_timers = NULL;
     p->mbuf = NULL;
     p->mbuf_sz = 0;
     p->psd = NULL;
@@ -10927,11 +10978,7 @@ void erts_init_empty_process(Process *p)
     p->bin_old_vheap = 0;
     p->sys_task_qs = NULL;
     p->bin_vheap_mature = 0;
-#ifdef ERTS_SMP
-    p->common.u.alive.ptimer = NULL;
-#else
-    erts_init_timer(&p->common.u.alive.tm);
-#endif
+    ERTS_PTMR_INIT(p);
     p->next = NULL;
     p->off_heap.first = NULL;
     p->off_heap.overhead = 0;
@@ -10952,14 +10999,15 @@ void erts_init_empty_process(Process *p)
     p->msg.last = &p->msg.first;
     p->msg.save = &p->msg.first;
     p->msg.len = 0;
-    p->u.bif_timers = NULL;
+    p->bif_timers = NULL;
+    p->accessor_bif_timers = NULL;
     p->dictionary = NULL;
     p->seq_trace_clock = 0;
     p->seq_trace_lastcnt = 0;
     p->seq_trace_token = NIL;
-    p->initial[0] = 0;
-    p->initial[1] = 0;
-    p->initial[2] = 0;
+    p->u.initial[0] = 0;
+    p->u.initial[1] = 0;
+    p->u.initial[2] = 0;
     p->catches = 0;
     p->cp = NULL;
     p->i = NULL;
@@ -11007,7 +11055,6 @@ void erts_init_empty_process(Process *p)
     p->pending_suspenders = NULL;
     p->pending_exit.reason = THE_NON_VALUE;
     p->pending_exit.bp = NULL;
-    p->preferred_run_queue = NULL;
     erts_proc_lock_init(p);
     erts_smp_proc_unlock(p, ERTS_PROC_LOCKS_ALL);
     RUNQ_SET_RQ(&p->run_queue, ERTS_RUNQ_IX(0));
@@ -11047,7 +11094,8 @@ erts_debug_verify_clean_empty_process(Process* p)
     ASSERT(p->suspend_monitors == NULL);
     ASSERT(p->msg.first == NULL);
     ASSERT(p->msg.len == 0);
-    ASSERT(p->u.bif_timers == NULL);
+    ASSERT(p->bif_timers == NULL);
+    ASSERT(p->accessor_bif_timers == NULL);
     ASSERT(p->dictionary == NULL);
     ASSERT(p->catches == 0);
     ASSERT(p->cp == NULL);
@@ -11211,7 +11259,6 @@ set_proc_exiting(Process *p,
      */
     p->freason = EXTAG_EXIT;
     KILL_CATCHES(p);
-    cancel_timer(p);
     p->i = (BeamInstr *) beam_exit;
 
     if (enqueue)
@@ -11901,6 +11948,7 @@ erts_do_exit_process(Process* p, Eterm reason)
 {
     p->arity = 0;		/* No live registers */
     p->fvalue = reason;
+    
 
 #ifdef USE_VM_PROBES
     if (DTRACE_ENABLED(process_exit)) {
@@ -11955,20 +12003,20 @@ erts_do_exit_process(Process* p, Eterm reason)
     ASSERT((ERTS_TRACE_FLAGS(p) & F_INITIAL_TRACE_FLAGS)
 	   == F_INITIAL_TRACE_FLAGS);
 
-    cancel_timer(p);		/* Always cancel timer just in case */
-
-    if (p->u.bif_timers)
-	erts_cancel_bif_timers(p, ERTS_PROC_LOCKS_ALL);
+    ASSERT(erts_proc_read_refc(p) > 0);
+    if (ERTS_PTMR_IS_SET(p)) {
+	erts_cancel_proc_timer(p);
+	ASSERT(erts_proc_read_refc(p) > 0);
+    }
 
     erts_smp_proc_unlock(p, ERTS_PROC_LOCKS_ALL_MINOR);
 
     /*
-     * The p->u.bif_timers of this process can *not* be used anymore;
+     * p->u.initial of this process can *not* be used anymore;
      * will be overwritten by misc termination data.
      */
     p->u.terminate = NULL;
 
-
     erts_continue_exit_process(p);
 }
 
@@ -11993,6 +12041,27 @@ erts_continue_exit_process(Process *p)
 
     ASSERT(ERTS_PROC_IS_EXITING(p));
 
+    ASSERT(erts_proc_read_refc(p) > 0);
+    if (p->bif_timers) {
+	if (erts_cancel_bif_timers(p, p->bif_timers, &p->u.terminate)) {
+	    ASSERT(erts_proc_read_refc(p) > 0);
+	    goto yield;
+	}
+	ASSERT(erts_proc_read_refc(p) > 0);
+	p->bif_timers = NULL;
+    }
+
+    if (p->accessor_bif_timers) {
+	if (erts_detach_accessor_bif_timers(p,
+					    p->accessor_bif_timers,
+					    &p->u.terminate)) {
+	    ASSERT(erts_proc_read_refc(p) > 0);
+	    goto yield;
+	}
+	ASSERT(erts_proc_read_refc(p) > 0);
+	p->accessor_bif_timers = NULL;
+    }
+
 #ifdef ERTS_SMP
     if (p->flags & F_HAVE_BLCKD_MSCHED) {
 	ErtsSchedSuspendResult ssr;
@@ -12086,6 +12155,8 @@ erts_continue_exit_process(Process *p)
 
 	p->scheduler_data->current_process = NULL;
 	p->scheduler_data->free_process = p;
+#else
+	erts_proc_inc_refc(p); /* Decremented in schedule() */
 #endif
 
 	/* Time of death! */
@@ -12104,29 +12175,23 @@ erts_continue_exit_process(Process *p)
     {
 	/* Inactivate and notify free */
 	erts_aint32_t n, e, a = erts_smp_atomic32_read_nob(&p->state);
-#ifdef ERTS_SMP
 	int refc_inced = 0;
-#endif
 	while (1) {
 	    n = e = a;
 	    ASSERT(a & ERTS_PSFLG_EXITING);
 	    n |= ERTS_PSFLG_FREE;
 	    n &= ~ERTS_PSFLG_ACTIVE;
-#ifdef ERTS_SMP
 	    if ((n & ERTS_PSFLG_IN_RUNQ) && !refc_inced) {
-		erts_smp_proc_inc_refc(p);
+		erts_proc_inc_refc(p);
 		refc_inced = 1;
 	    }
-#endif
 	    a = erts_smp_atomic32_cmpxchg_mb(&p->state, n, e);
 	    if (a == e)
 		break;
 	}
 
-#ifdef ERTS_SMP
 	if (refc_inced && !(n & ERTS_PSFLG_IN_RUNQ))
-	    erts_smp_proc_dec_refc(p);
-#endif
+	    erts_proc_dec_refc(p);
     }
     
     dep = ((p->flags & F_DISTRIBUTION)
@@ -12217,64 +12282,6 @@ erts_continue_exit_process(Process *p)
 
 }
 
-/* Callback for process timeout */
-static void
-timeout_proc(Process* p)
-{
-    erts_aint32_t state;
-    BeamInstr** pi = (BeamInstr **) p->def_arg_reg;
-    p->i = *pi;
-    p->flags |= F_TIMO;
-    p->flags &= ~F_INSLPQUEUE;
-
-    state = erts_smp_atomic32_read_acqb(&p->state);
-    if (!(state & ERTS_PSFLG_ACTIVE))
-	schedule_process(p, state, ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS);
-}
-
-
-void
-cancel_timer(Process* p)
-{
-    ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_MAIN & erts_proc_lc_my_proc_locks(p));
-    p->flags &= ~(F_INSLPQUEUE|F_TIMO);
-#ifdef ERTS_SMP
-    erts_cancel_smp_ptimer(p->common.u.alive.ptimer);
-#else
-    erts_cancel_timer(&p->common.u.alive.tm);
-#endif
-}
-
-/*
- * Insert a process into the time queue, with a timeout 'timeout' in ms.
- */
-void
-set_timer(Process* p, Uint timeout)
-{
-    ERTS_SMP_LC_ASSERT(ERTS_PROC_LOCK_MAIN & erts_proc_lc_my_proc_locks(p));
-
-    /* check for special case timeout=0 DONT ADD TO time queue */
-    if (timeout == 0) {
-	p->flags |= F_TIMO;
-	return;
-    }
-    p->flags |= F_INSLPQUEUE;
-    p->flags &= ~F_TIMO;
-
-#ifdef ERTS_SMP
-    erts_create_smp_ptimer(&p->common.u.alive.ptimer,
-			   p->common.id,
-			   (ErlTimeoutProc) timeout_proc,
-			   timeout);
-#else
-    erts_set_timer(&p->common.u.alive.tm,
-		  (ErlTimeoutProc) timeout_proc,
-		  NULL,
-		  (void*) p,
-		  timeout);
-#endif
-}
-
 /*
  * Stack dump functions follow.
  */
@@ -12432,6 +12439,10 @@ erts_print_scheduler_info(int to, void *to_arg, ErtsSchedulerData *esdp) {
                 erts_print(to, to_arg, "FIX_ALLOC_LOWER_LIM"); break;
             case ERTS_SSI_AUX_WORK_THR_PRGR_LATER_OP:
                 erts_print(to, to_arg, "THR_PRGR_LATER_OP"); break;
+            case ERTS_SSI_AUX_WORK_CNCLD_TMRS:
+                erts_print(to, to_arg, "CANCELED_TIMERS"); break;
+            case ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR:
+                erts_print(to, to_arg, "CANCELED_TIMERS_THR_PRGR"); break;
             case ERTS_SSI_AUX_WORK_ASYNC_READY:
                 erts_print(to, to_arg, "ASYNC_READY"); break;
             case ERTS_SSI_AUX_WORK_ASYNC_READY_CLEAN:
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index 743711cc3b..b1c30e7652 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -52,7 +52,7 @@ typedef struct process Process;
 #include "erl_node_container_utils.h"
 #include "erl_node_tables.h"
 #include "erl_monitors.h"
-#include "erl_bif_timer.h"
+#include "erl_hl_timer.h"
 #include "erl_time.h"
 #include "erl_atom_table.h"
 #include "external.h"
@@ -278,16 +278,18 @@ typedef enum {
 #define ERTS_SSI_AUX_WORK_FIX_ALLOC_DEALLOC	(((erts_aint32_t) 1) << 3)
 #define ERTS_SSI_AUX_WORK_FIX_ALLOC_LOWER_LIM	(((erts_aint32_t) 1) << 4)
 #define ERTS_SSI_AUX_WORK_THR_PRGR_LATER_OP	(((erts_aint32_t) 1) << 5)
-#define ERTS_SSI_AUX_WORK_ASYNC_READY		(((erts_aint32_t) 1) << 6)
-#define ERTS_SSI_AUX_WORK_ASYNC_READY_CLEAN	(((erts_aint32_t) 1) << 7)
-#define ERTS_SSI_AUX_WORK_MISC_THR_PRGR		(((erts_aint32_t) 1) << 8)
-#define ERTS_SSI_AUX_WORK_MISC			(((erts_aint32_t) 1) << 9)
-#define ERTS_SSI_AUX_WORK_CHECK_CHILDREN	(((erts_aint32_t) 1) << 10)
-#define ERTS_SSI_AUX_WORK_SET_TMO		(((erts_aint32_t) 1) << 11)
-#define ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK	(((erts_aint32_t) 1) << 12)
-#define ERTS_SSI_AUX_WORK_REAP_PORTS		(((erts_aint32_t) 1) << 13)
-
-#define ERTS_SSI_AUX_WORK_MAX                                           14
+#define ERTS_SSI_AUX_WORK_CNCLD_TMRS		(((erts_aint32_t) 1) << 6)
+#define ERTS_SSI_AUX_WORK_CNCLD_TMRS_THR_PRGR	(((erts_aint32_t) 1) << 7)
+#define ERTS_SSI_AUX_WORK_ASYNC_READY		(((erts_aint32_t) 1) << 8)
+#define ERTS_SSI_AUX_WORK_ASYNC_READY_CLEAN	(((erts_aint32_t) 1) << 9)
+#define ERTS_SSI_AUX_WORK_MISC_THR_PRGR		(((erts_aint32_t) 1) << 10)
+#define ERTS_SSI_AUX_WORK_MISC			(((erts_aint32_t) 1) << 11)
+#define ERTS_SSI_AUX_WORK_CHECK_CHILDREN	(((erts_aint32_t) 1) << 12)
+#define ERTS_SSI_AUX_WORK_SET_TMO		(((erts_aint32_t) 1) << 13)
+#define ERTS_SSI_AUX_WORK_MSEG_CACHE_CHECK	(((erts_aint32_t) 1) << 14)
+#define ERTS_SSI_AUX_WORK_REAP_PORTS		(((erts_aint32_t) 1) << 15)
+
+#define ERTS_SSI_AUX_WORK_MAX                                           16
 
 typedef struct ErtsSchedulerSleepInfo_ ErtsSchedulerSleepInfo;
 
@@ -463,19 +465,21 @@ typedef union {
 
 extern ErtsAlignedRunQueue *erts_aligned_run_queues;
 
-#define ERTS_PROC_REDUCTIONS_EXECUTED(RQ, PRIO, REDS, AREDS)	\
+#define ERTS_PROC_REDUCTIONS_EXECUTED(SD, RQ, PRIO, REDS, AREDS)\
 do {								\
     (RQ)->procs.reductions += (AREDS);				\
     (RQ)->procs.prio_info[(PRIO)].reds += (REDS);		\
     (RQ)->check_balance_reds -= (REDS);				\
     (RQ)->wakeup_other_reds += (AREDS);				\
+    (SD)->check_time_reds += (AREDS);				\
 } while (0)
 
-#define ERTS_PORT_REDUCTIONS_EXECUTED(RQ, REDS)			\
+#define ERTS_PORT_REDUCTIONS_EXECUTED(SD, RQ, REDS)		\
 do {								\
     (RQ)->ports.info.reds += (REDS);				\
     (RQ)->check_balance_reds -= (REDS);				\
     (RQ)->wakeup_other_reds += (REDS);				\
+    (SD)->check_time_reds += (REDS);				\
 } while (0)
 
 typedef struct {
@@ -516,6 +520,9 @@ typedef struct {
     } dd;
     struct {
 	ErtsThrPrgrVal thr_prgr;
+    } cncld_tmrs;
+    struct {
+	ErtsThrPrgrVal thr_prgr;
 	UWord size;
 	ErtsThrPrgrLaterOp *first;
 	ErtsThrPrgrLaterOp *last;
@@ -566,6 +573,7 @@ struct ErtsSchedulerData_ {
 
     ErtsTimerWheel *timer_wheel;
     ErtsNextTimeoutRef next_tmo_ref;
+    ErtsHLTimerService *timer_service;
 #ifdef ERTS_SMP
     ethr_tid tid;		/* Thread id */
     struct erl_bits_state erl_bits_state; /* erl_bits.c state */
@@ -592,6 +600,9 @@ struct ErtsSchedulerData_ {
     ErtsAuxWorkData aux_work_data;
     ErtsAtomCacheMap atom_cache_map;
 
+    ErtsMonotonicTime last_monotonic_time;
+    int check_time_reds;
+
     Uint32 thr_id;
     Uint64 unique;
     Uint64 ref;
@@ -913,10 +924,8 @@ struct process {
 
     ErlMessageQueue msg;	/* Message queue */
 
-    union {
-	ErtsBifTimer *bif_timers;	/* Bif timers aiming at this process */
-	void *terminate;
-    } u;
+    ErtsBifTimers *bif_timers;	/* Bif timers aiming at this process */
+    ErtsBifTimers *accessor_bif_timers;	/* Accessor bif timers */
 
     ProcDict  *dictionary;       /* Process dictionary, may be NULL */
 
@@ -927,9 +936,12 @@ struct process {
 #ifdef USE_VM_PROBES
     Eterm dt_utag;              /* Place to store the dynamc trace user tag */
     Uint dt_utag_flags;         /* flag field for the dt_utag */
-#endif       
-    BeamInstr initial[3];	/* Initial module(0), function(1), arity(2), often used instead
+#endif
+    union {
+	void *terminate;
+	BeamInstr initial[3];	/* Initial module(0), function(1), arity(2), often used instead
 				   of pointer to funcinfo instruction, hence the BeamInstr datatype */
+    } u;
     BeamInstr* current;		/* Current Erlang function, part of the funcinfo:
 				 * module(0), function(1), arity(2)
 				 * (module and functions are tagged atoms;
@@ -975,7 +987,6 @@ struct process {
     ErtsSchedulerData *scheduler_data;
     Eterm suspendee;
     ErtsPendingSuspend *pending_suspenders;
-    ErtsRunQueue *preferred_run_queue;
     erts_smp_atomic_t run_queue;
 #ifdef HIPE
     struct hipe_process_state_smp hipe_smp;
@@ -1085,15 +1096,14 @@ void erts_check_for_holes(Process* p);
 #define ERTS_PSFLG_RUNNING_SYS		ERTS_PSFLG_BIT(15)
 #define ERTS_PSFLG_PROXY		ERTS_PSFLG_BIT(16)
 #define ERTS_PSFLG_DELAYED_SYS		ERTS_PSFLG_BIT(17)
-#define ERTS_PSFLG_OFF_HEAP_MSGS	ERTS_PSFLG_BIT(18)
 #ifdef ERTS_DIRTY_SCHEDULERS
-#define ERTS_PSFLG_DIRTY_CPU_PROC	ERTS_PSFLG_BIT(19)
-#define ERTS_PSFLG_DIRTY_IO_PROC	ERTS_PSFLG_BIT(20)
-#define ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q	ERTS_PSFLG_BIT(21)
-#define ERTS_PSFLG_DIRTY_IO_PROC_IN_Q	ERTS_PSFLG_BIT(22)
-#define ERTS_PSFLG_MAX  (ERTS_PSFLGS_ZERO_BIT_OFFSET + 23)
+#define ERTS_PSFLG_DIRTY_CPU_PROC	ERTS_PSFLG_BIT(18)
+#define ERTS_PSFLG_DIRTY_IO_PROC	ERTS_PSFLG_BIT(19)
+#define ERTS_PSFLG_DIRTY_CPU_PROC_IN_Q	ERTS_PSFLG_BIT(20)
+#define ERTS_PSFLG_DIRTY_IO_PROC_IN_Q	ERTS_PSFLG_BIT(21)
+#define ERTS_PSFLG_MAX  (ERTS_PSFLGS_ZERO_BIT_OFFSET + 22)
 #else
-#define ERTS_PSFLG_MAX  (ERTS_PSFLGS_ZERO_BIT_OFFSET + 19)
+#define ERTS_PSFLG_MAX  (ERTS_PSFLGS_ZERO_BIT_OFFSET + 18)
 #endif
 
 #define ERTS_PSFLGS_IN_PRQ_MASK 	(ERTS_PSFLG_IN_PRQ_MAX		\
@@ -1112,7 +1122,6 @@ void erts_check_for_holes(Process* p);
  * Static flags that do not change after process creation.
  */
 #define ERTS_STC_FLG_SYSTEM_PROC	(((Uint32) 1) << 0)
-#define ERTS_STC_FLG_PREFER_SCHED	(((Uint32) 1) << 1)
 
 /* The sequential tracing token is a tuple of size 5:
  *
@@ -1141,9 +1150,7 @@ void erts_check_for_holes(Process* p);
 #define SPO_LINK 1
 #define SPO_USE_ARGS 2
 #define SPO_MONITOR 4
-#define SPO_OFF_HEAP_MSGS 8
-#define SPO_SYSTEM_PROC 16
-#define SPO_PREFER_SCHED 32
+#define SPO_SYSTEM_PROC 8
 
 /*
  * The following struct contains options for a process to be spawned.
@@ -1231,7 +1238,6 @@ extern struct erts_system_profile_flags_t erts_system_profile_flags;
 #define F_P2PNR_RESCHED      (1 <<  9) /* Process has been rescheduled via erts_pid2proc_not_running() */
 #define F_FORCE_GC           (1 << 10) /* Force gc at process in-scheduling */
 #define F_DISABLE_GC         (1 << 11) /* Disable GC */
-#define F_OFF_HEAP_MSGS      (1 << 12)
 
 /* process trace_flags */
 #define F_SENSITIVE          (1 << 0)
@@ -1267,8 +1273,6 @@ extern struct erts_system_profile_flags_t erts_system_profile_flags;
 #  define F_INITIAL_TRACE_FLAGS 0
 #endif
 
-
-
 #define TRACEE_FLAGS ( F_TRACE_PROCS | F_TRACE_CALLS \
 		     | F_TRACE_SOS |  F_TRACE_SOS1| F_TRACE_RECEIVE  \
 		     | F_TRACE_SOL | F_TRACE_SOL1 | F_TRACE_SEND \
@@ -1302,12 +1306,14 @@ extern struct erts_system_profile_flags_t erts_system_profile_flags;
 #define ERTS_XSIG_FLG_IGN_KILL		(((Uint32) 1) << 0)
 #define ERTS_XSIG_FLG_NO_IGN_NORMAL	(((Uint32) 1) << 1)
 
-#define CANCEL_TIMER(p) \
-    do { \
-	if ((p)->flags & (F_INSLPQUEUE)) \
-	    cancel_timer(p); \
-	else \
-	    (p)->flags &= ~F_TIMO; \
+#define CANCEL_TIMER(P)					\
+    do {						\
+	if ((P)->flags & (F_INSLPQUEUE|F_TIMO)) {	\
+	    if ((P)->flags & F_INSLPQUEUE)		\
+		erts_cancel_proc_timer((P));		\
+	    else					\
+		(P)->flags &= ~F_TIMO;			\
+	}						\
     } while (0)
 
 #if defined(ERTS_DIRTY_SCHEDULERS) && defined(ERTS_SMP)
@@ -1594,6 +1600,9 @@ Eterm erts_multi_scheduling_blockers(Process *);
 void erts_start_schedulers(void);
 void erts_alloc_notify_delayed_dealloc(int);
 void erts_alloc_ensure_handle_delayed_dealloc_call(int);
+#ifdef ERTS_SMP
+void erts_notify_canceled_timer(ErtsSchedulerData *, int);
+#endif
 void erts_smp_notify_check_children_needed(void);
 #endif
 #if ERTS_USE_ASYNC_READY_Q
@@ -1628,8 +1637,6 @@ void erts_schedule_misc_op(void (*)(void *), void *);
 Eterm erl_create_process(Process*, Eterm, Eterm, Eterm, ErlSpawnOpts*);
 void erts_do_exit_process(Process*, Eterm);
 void erts_continue_exit_process(Process *);
-void set_timer(Process*, Uint);
-void cancel_timer(Process*);
 /* Begin System profile */
 Uint erts_runnable_process_count(void);
 /* End System profile */
diff --git a/erts/emulator/beam/erl_process_lock.c b/erts/emulator/beam/erl_process_lock.c
index 82cc68222d..fff267ff2a 100644
--- a/erts/emulator/beam/erl_process_lock.c
+++ b/erts/emulator/beam/erl_process_lock.c
@@ -103,6 +103,7 @@ static struct {
     Sint16 proc_lock_main;
     Sint16 proc_lock_link;
     Sint16 proc_lock_msgq;
+    Sint16 proc_lock_btm;
     Sint16 proc_lock_status;
 } lc_id;
 #endif
@@ -145,6 +146,7 @@ erts_init_proc_lock(int cpus)
     lc_id.proc_lock_main	= erts_lc_get_lock_order_id("proc_main");
     lc_id.proc_lock_link	= erts_lc_get_lock_order_id("proc_link");
     lc_id.proc_lock_msgq	= erts_lc_get_lock_order_id("proc_msgq");
+    lc_id.proc_lock_btm		= erts_lc_get_lock_order_id("proc_btm");
     lc_id.proc_lock_status	= erts_lc_get_lock_order_id("proc_status");
 #endif
 }
@@ -707,7 +709,7 @@ proc_safelock(int is_managed,
 	    need_locks1 |= unlock_locks;
 	    if (!is_managed && !have_locks1) {
 		refc1 = 1;
-		erts_smp_proc_inc_refc(p1);
+		erts_proc_inc_refc(p1);
 	    }
 	    erts_smp_proc_unlock(p1, unlock_locks);
 	}
@@ -717,7 +719,7 @@ proc_safelock(int is_managed,
 	    need_locks2 |= unlock_locks;
 	    if (!is_managed && !have_locks2) {
 		refc2 = 1;
-		erts_smp_proc_inc_refc(p2);
+		erts_proc_inc_refc(p2);
 	    }
 	    erts_smp_proc_unlock(p2, unlock_locks);
 	}
@@ -798,9 +800,9 @@ proc_safelock(int is_managed,
 
     if (!is_managed) {
 	if (refc1)
-	    erts_smp_proc_dec_refc(p1);
+	    erts_proc_dec_refc(p1);
 	if (refc2)
-	    erts_smp_proc_dec_refc(p2);
+	    erts_proc_dec_refc(p2);
     }
 }
 
@@ -861,8 +863,8 @@ erts_pid2proc_opt(Process *c_p,
 	    return NULL;
 	need_locks &= ~c_p_have_locks;
 	if (!need_locks) {
-	    if (flags & ERTS_P2P_FLG_SMP_INC_REFC)
-		erts_smp_proc_inc_refc(c_p);
+	    if (flags & ERTS_P2P_FLG_INC_REFC)
+		erts_proc_inc_refc(c_p);
 	    return c_p;
 	}
     }
@@ -875,8 +877,8 @@ erts_pid2proc_opt(Process *c_p,
 	if (proc->common.id != pid)
 	    proc = NULL;
 	else if (!need_locks) {
-	    if (flags & ERTS_P2P_FLG_SMP_INC_REFC)
-		erts_smp_proc_inc_refc(proc);
+	    if (flags & ERTS_P2P_FLG_INC_REFC)
+		erts_proc_inc_refc(proc);
 	}
 	else {
 	    int busy;
@@ -916,8 +918,8 @@ erts_pid2proc_opt(Process *c_p,
 #endif
 
 	    if (!busy) {
-		if (flags & ERTS_P2P_FLG_SMP_INC_REFC)
-		    erts_smp_proc_inc_refc(proc);
+		if (flags & ERTS_P2P_FLG_INC_REFC)
+		    erts_proc_inc_refc(proc);
 
 #if ERTS_PROC_LOCK_OWN_IMPL && defined(ERTS_ENABLE_LOCK_COUNT)
 	    	/* all is great */
@@ -932,8 +934,8 @@ erts_pid2proc_opt(Process *c_p,
 		    proc = ERTS_PROC_LOCK_BUSY;
 		else {
 		    int managed;
-		    if (flags & ERTS_P2P_FLG_SMP_INC_REFC)
-			erts_smp_proc_inc_refc(proc);
+		    if (flags & ERTS_P2P_FLG_INC_REFC)
+			erts_proc_inc_refc(proc);
 
 #if ERTS_PROC_LOCK_OWN_IMPL && defined(ERTS_ENABLE_LOCK_COUNT)
 		    erts_lcnt_proc_lock_unaquire(&proc->lock, lcnt_locks);
@@ -941,7 +943,7 @@ erts_pid2proc_opt(Process *c_p,
 
 		    managed = dhndl == ERTS_THR_PRGR_DHANDLE_MANAGED;
 		    if (!managed) {
-			erts_smp_proc_inc_refc(proc);
+			erts_proc_inc_refc(proc);
 			erts_thr_progress_unmanaged_continue(dhndl);
 			dec_refc_proc = proc;
 
@@ -978,14 +980,14 @@ erts_pid2proc_opt(Process *c_p,
 
 	erts_smp_proc_unlock(proc, need_locks);
 
-	if (flags & ERTS_P2P_FLG_SMP_INC_REFC)
+	if (flags & ERTS_P2P_FLG_INC_REFC)
 	    dec_refc_proc = proc;
 	proc = NULL;
 
     }
 
     if (dec_refc_proc)
-	erts_smp_proc_dec_refc(dec_refc_proc);
+	erts_proc_dec_refc(dec_refc_proc);
 
 #if ERTS_PROC_LOCK_OWN_IMPL && defined(ERTS_PROC_LOCK_DEBUG)
     ERTS_LC_ASSERT(!proc
@@ -1038,6 +1040,11 @@ erts_proc_lock_init(Process *p)
 #ifdef ERTS_ENABLE_LOCK_CHECK
     erts_lc_trylock(1, &p->lock.msgq.lc);
 #endif
+    erts_mtx_init_x(&p->lock.btm, "proc_btm", p->common.id, do_lock_count);
+    ethr_mutex_lock(&p->lock.btm.mtx);
+#ifdef ERTS_ENABLE_LOCK_CHECK
+    erts_lc_trylock(1, &p->lock.btm.lc);
+#endif
     erts_mtx_init_x(&p->lock.status, "proc_status", p->common.id,
 		    do_lock_count);
     ethr_mutex_lock(&p->lock.status.mtx);
@@ -1045,7 +1052,6 @@ erts_proc_lock_init(Process *p)
     erts_lc_trylock(1, &p->lock.status.lc);
 #endif
 #endif
-    erts_atomic32_init_nob(&p->lock.refc, 1);
 #ifdef ERTS_PROC_LOCK_DEBUG
     for (i = 0; i <= ERTS_PROC_LOCK_MAX_BIT; i++)
 	erts_smp_atomic32_init_nob(&p->lock.locked[i], (erts_aint32_t) 1);
@@ -1064,6 +1070,7 @@ erts_proc_lock_fin(Process *p)
     erts_mtx_destroy(&p->lock.main);
     erts_mtx_destroy(&p->lock.link);
     erts_mtx_destroy(&p->lock.msgq);
+    erts_mtx_destroy(&p->lock.btm);
     erts_mtx_destroy(&p->lock.status);
 #endif
 #if defined(ERTS_ENABLE_LOCK_COUNT) && defined(ERTS_SMP)
@@ -1079,17 +1086,20 @@ void erts_lcnt_proc_lock_init(Process *p) {
     if (p->common.id != ERTS_INVALID_PID) {
 	erts_lcnt_init_lock_x(&(p->lock.lcnt_main),   "proc_main",   ERTS_LCNT_LT_PROCLOCK, p->common.id);
 	erts_lcnt_init_lock_x(&(p->lock.lcnt_msgq),   "proc_msgq",   ERTS_LCNT_LT_PROCLOCK, p->common.id);
+	erts_lcnt_init_lock_x(&(p->lock.lcnt_btm),    "proc_btm",   ERTS_LCNT_LT_PROCLOCK, p->common.id);
 	erts_lcnt_init_lock_x(&(p->lock.lcnt_link),   "proc_link",   ERTS_LCNT_LT_PROCLOCK, p->common.id);
 	erts_lcnt_init_lock_x(&(p->lock.lcnt_status), "proc_status", ERTS_LCNT_LT_PROCLOCK, p->common.id);
     } else {
 	erts_lcnt_init_lock(&(p->lock.lcnt_main),   "proc_main",   ERTS_LCNT_LT_PROCLOCK);
 	erts_lcnt_init_lock(&(p->lock.lcnt_msgq),   "proc_msgq",   ERTS_LCNT_LT_PROCLOCK);
+	erts_lcnt_init_lock(&(p->lock.lcnt_btm),    "proc_btm",   ERTS_LCNT_LT_PROCLOCK);
 	erts_lcnt_init_lock(&(p->lock.lcnt_link),   "proc_link",   ERTS_LCNT_LT_PROCLOCK);
 	erts_lcnt_init_lock(&(p->lock.lcnt_status), "proc_status", ERTS_LCNT_LT_PROCLOCK);
     }
     } else {
 	sys_memzero(&(p->lock.lcnt_main), sizeof(p->lock.lcnt_main));
 	sys_memzero(&(p->lock.lcnt_msgq), sizeof(p->lock.lcnt_msgq));
+	sys_memzero(&(p->lock.lcnt_btm), sizeof(p->lock.lcnt_btm));
 	sys_memzero(&(p->lock.lcnt_link), sizeof(p->lock.lcnt_link));
 	sys_memzero(&(p->lock.lcnt_status), sizeof(p->lock.lcnt_status));
     }
@@ -1099,6 +1109,7 @@ void erts_lcnt_proc_lock_init(Process *p) {
 void erts_lcnt_proc_lock_destroy(Process *p) {
     erts_lcnt_destroy_lock(&(p->lock.lcnt_main));
     erts_lcnt_destroy_lock(&(p->lock.lcnt_msgq));
+    erts_lcnt_destroy_lock(&(p->lock.lcnt_btm));
     erts_lcnt_destroy_lock(&(p->lock.lcnt_link));
     erts_lcnt_destroy_lock(&(p->lock.lcnt_status));
 }
@@ -1111,6 +1122,9 @@ void erts_lcnt_proc_lock(erts_proc_lock_t *lock, ErtsProcLocks locks) {
     if (locks & ERTS_PROC_LOCK_MSGQ) {
         erts_lcnt_lock(&(lock->lcnt_msgq));
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+        erts_lcnt_lock(&(lock->lcnt_btm));
+    }
     if (locks & ERTS_PROC_LOCK_LINK) {
 	erts_lcnt_lock(&(lock->lcnt_link));
     }
@@ -1128,6 +1142,9 @@ void erts_lcnt_proc_lock_post_x(erts_proc_lock_t *lock, ErtsProcLocks locks, cha
     if (locks & ERTS_PROC_LOCK_MSGQ) {
         erts_lcnt_lock_post_x(&(lock->lcnt_msgq), file, line);
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+        erts_lcnt_lock_post_x(&(lock->lcnt_btm), file, line);
+    }
     if (locks & ERTS_PROC_LOCK_LINK) {
 	erts_lcnt_lock_post_x(&(lock->lcnt_link), file, line);
     }
@@ -1145,6 +1162,9 @@ void erts_lcnt_proc_lock_unaquire(erts_proc_lock_t *lock, ErtsProcLocks locks) {
     if (locks & ERTS_PROC_LOCK_MSGQ) {
         erts_lcnt_lock_unaquire(&(lock->lcnt_msgq));
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+        erts_lcnt_lock_unaquire(&(lock->lcnt_btm));
+    }
     if (locks & ERTS_PROC_LOCK_LINK) {
 	erts_lcnt_lock_unaquire(&(lock->lcnt_link));
     }
@@ -1162,6 +1182,9 @@ void erts_lcnt_proc_unlock(erts_proc_lock_t *lock, ErtsProcLocks locks) {
     if (locks & ERTS_PROC_LOCK_MSGQ) {
         erts_lcnt_unlock(&(lock->lcnt_msgq));
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+        erts_lcnt_unlock(&(lock->lcnt_btm));
+    }
     if (locks & ERTS_PROC_LOCK_LINK) {
 	erts_lcnt_unlock(&(lock->lcnt_link));
     }
@@ -1178,6 +1201,9 @@ void erts_lcnt_proc_trylock(erts_proc_lock_t *lock, ErtsProcLocks locks, int res
     if (locks & ERTS_PROC_LOCK_MSGQ) {
         erts_lcnt_trylock(&(lock->lcnt_msgq), res);
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+        erts_lcnt_trylock(&(lock->lcnt_btm), res);
+    }
     if (locks & ERTS_PROC_LOCK_LINK) {
 	erts_lcnt_trylock(&(lock->lcnt_link), res);
     }
@@ -1235,6 +1261,10 @@ erts_proc_lc_lock(Process *p, ErtsProcLocks locks, char *file, unsigned int line
 	lck.id = lc_id.proc_lock_msgq;
 	erts_lc_lock_x(&lck,file,line);
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+	lck.id = lc_id.proc_lock_btm;
+	erts_lc_lock_x(&lck,file,line);
+    }
     if (locks & ERTS_PROC_LOCK_STATUS) {
 	lck.id = lc_id.proc_lock_status;
 	erts_lc_lock_x(&lck,file,line);
@@ -1260,6 +1290,10 @@ erts_proc_lc_trylock(Process *p, ErtsProcLocks locks, int locked,
 	lck.id = lc_id.proc_lock_msgq;
 	erts_lc_trylock_x(locked, &lck, file, line);
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+	lck.id = lc_id.proc_lock_btm;
+	erts_lc_trylock_x(locked, &lck, file, line);
+    }
     if (locks & ERTS_PROC_LOCK_STATUS) {
 	lck.id = lc_id.proc_lock_status;
 	erts_lc_trylock_x(locked, &lck, file, line);
@@ -1276,6 +1310,10 @@ erts_proc_lc_unlock(Process *p, ErtsProcLocks locks)
 	lck.id = lc_id.proc_lock_status;
 	erts_lc_unlock(&lck);
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+	lck.id = lc_id.proc_lock_btm;
+	erts_lc_unlock(&lck);
+    }
     if (locks & ERTS_PROC_LOCK_MSGQ) {
 	lck.id = lc_id.proc_lock_msgq;
 	erts_lc_unlock(&lck);
@@ -1303,6 +1341,10 @@ erts_proc_lc_might_unlock(Process *p, ErtsProcLocks locks)
 	lck.id = lc_id.proc_lock_status;
 	erts_lc_might_unlock(&lck);
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+	lck.id = lc_id.proc_lock_btm;
+	erts_lc_might_unlock(&lck);
+    }
     if (locks & ERTS_PROC_LOCK_MSGQ) {
 	lck.id = lc_id.proc_lock_msgq;
 	erts_lc_might_unlock(&lck);
@@ -1322,6 +1364,8 @@ erts_proc_lc_might_unlock(Process *p, ErtsProcLocks locks)
 	erts_lc_might_unlock(&p->lock.link.lc);
     if (locks & ERTS_PROC_LOCK_MSGQ)
 	erts_lc_might_unlock(&p->lock.msgq.lc);
+    if (locks & ERTS_PROC_LOCK_BTM)
+	erts_lc_might_unlock(&p->lock.btm.lc);
     if (locks & ERTS_PROC_LOCK_STATUS)
 	erts_lc_might_unlock(&p->lock.status.lc);
 #endif
@@ -1347,6 +1391,10 @@ erts_proc_lc_require_lock(Process *p, ErtsProcLocks locks, char *file,
 	lck.id = lc_id.proc_lock_msgq;
 	erts_lc_require_lock(&lck, file, line);
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+	lck.id = lc_id.proc_lock_btm;
+	erts_lc_require_lock(&lck, file, line);
+    }
     if (locks & ERTS_PROC_LOCK_STATUS) {
 	lck.id = lc_id.proc_lock_status;
 	erts_lc_require_lock(&lck, file, line);
@@ -1358,6 +1406,8 @@ erts_proc_lc_require_lock(Process *p, ErtsProcLocks locks, char *file,
 	erts_lc_require_lock(&p->lock.link.lc, file, line);
     if (locks & ERTS_PROC_LOCK_MSGQ)
 	erts_lc_require_lock(&p->lock.msgq.lc, file, line);
+    if (locks & ERTS_PROC_LOCK_BTM)
+	erts_lc_require_lock(&p->lock.btm.lc, file, line);
     if (locks & ERTS_PROC_LOCK_STATUS)
 	erts_lc_require_lock(&p->lock.status.lc, file, line);
 #endif
@@ -1374,6 +1424,10 @@ erts_proc_lc_unrequire_lock(Process *p, ErtsProcLocks locks)
 	lck.id = lc_id.proc_lock_status;
 	erts_lc_unrequire_lock(&lck);
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+	lck.id = lc_id.proc_lock_btm;
+	erts_lc_unrequire_lock(&lck);
+    }
     if (locks & ERTS_PROC_LOCK_MSGQ) {
 	lck.id = lc_id.proc_lock_msgq;
 	erts_lc_unrequire_lock(&lck);
@@ -1393,6 +1447,8 @@ erts_proc_lc_unrequire_lock(Process *p, ErtsProcLocks locks)
 	erts_lc_unrequire_lock(&p->lock.link.lc);
     if (locks & ERTS_PROC_LOCK_MSGQ)
 	erts_lc_unrequire_lock(&p->lock.msgq.lc);
+    if (locks & ERTS_PROC_LOCK_BTM)
+	erts_lc_unrequire_lock(&p->lock.btm.lc);
     if (locks & ERTS_PROC_LOCK_STATUS)
 	erts_lc_unrequire_lock(&p->lock.status.lc);
 #endif
@@ -1414,6 +1470,8 @@ erts_proc_lc_trylock_force_busy(Process *p, ErtsProcLocks locks)
 	    lck.id = lc_id.proc_lock_link;
 	else if (locks & ERTS_PROC_LOCK_MSGQ)
 	    lck.id = lc_id.proc_lock_msgq;
+	else if (locks & ERTS_PROC_LOCK_BTM)
+	    lck.id = lc_id.proc_lock_btm;
 	else if (locks & ERTS_PROC_LOCK_STATUS)
 	    lck.id = lc_id.proc_lock_status;
 	else
@@ -1448,7 +1506,8 @@ erts_proc_lc_chk_have_proc_locks(Process *p, ErtsProcLocks locks)
 {
     int have_locks_len = 0;
 #if ERTS_PROC_LOCK_OWN_IMPL
-    erts_lc_lock_t have_locks[4] = {ERTS_PROC_LC_EMPTY_LOCK_INIT,
+    erts_lc_lock_t have_locks[5] = {ERTS_PROC_LC_EMPTY_LOCK_INIT,
+				    ERTS_PROC_LC_EMPTY_LOCK_INIT,
 				    ERTS_PROC_LC_EMPTY_LOCK_INIT,
 				    ERTS_PROC_LC_EMPTY_LOCK_INIT,
 				    ERTS_PROC_LC_EMPTY_LOCK_INIT};
@@ -1464,18 +1523,24 @@ erts_proc_lc_chk_have_proc_locks(Process *p, ErtsProcLocks locks)
 	have_locks[have_locks_len].id = lc_id.proc_lock_msgq;
 	have_locks[have_locks_len++].extra = p->common.id;
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+	have_locks[have_locks_len].id = lc_id.proc_lock_btm;
+	have_locks[have_locks_len++].extra = p->common.id;
+    }
     if (locks & ERTS_PROC_LOCK_STATUS) {
 	have_locks[have_locks_len].id = lc_id.proc_lock_status;
 	have_locks[have_locks_len++].extra = p->common.id;
     }
 #elif ERTS_PROC_LOCK_RAW_MUTEX_IMPL
-    erts_lc_lock_t have_locks[4];
+    erts_lc_lock_t have_locks[5];
     if (locks & ERTS_PROC_LOCK_MAIN)
 	have_locks[have_locks_len++] = p->lock.main.lc;
     if (locks & ERTS_PROC_LOCK_LINK)
 	have_locks[have_locks_len++] = p->lock.link.lc;
     if (locks & ERTS_PROC_LOCK_MSGQ)
 	have_locks[have_locks_len++] = p->lock.msgq.lc;
+    if (locks & ERTS_PROC_LOCK_BTM)
+	have_locks[have_locks_len++] = p->lock.btm.lc;
     if (locks & ERTS_PROC_LOCK_STATUS)
 	have_locks[have_locks_len++] = p->lock.status.lc;
 #endif
@@ -1488,11 +1553,11 @@ erts_proc_lc_chk_proc_locks(Process *p, ErtsProcLocks locks)
     int have_locks_len = 0;
     int have_not_locks_len = 0;
 #if ERTS_PROC_LOCK_OWN_IMPL
-    erts_lc_lock_t have_locks[4] = {ERTS_PROC_LC_EMPTY_LOCK_INIT,
+    erts_lc_lock_t have_locks[5] = {ERTS_PROC_LC_EMPTY_LOCK_INIT,
 				    ERTS_PROC_LC_EMPTY_LOCK_INIT,
 				    ERTS_PROC_LC_EMPTY_LOCK_INIT,
 				    ERTS_PROC_LC_EMPTY_LOCK_INIT};
-    erts_lc_lock_t have_not_locks[4] = {ERTS_PROC_LC_EMPTY_LOCK_INIT,
+    erts_lc_lock_t have_not_locks[5] = {ERTS_PROC_LC_EMPTY_LOCK_INIT,
 					ERTS_PROC_LC_EMPTY_LOCK_INIT,
 					ERTS_PROC_LC_EMPTY_LOCK_INIT,
 					ERTS_PROC_LC_EMPTY_LOCK_INIT};
@@ -1521,6 +1586,14 @@ erts_proc_lc_chk_proc_locks(Process *p, ErtsProcLocks locks)
 	have_not_locks[have_not_locks_len].id = lc_id.proc_lock_msgq;
 	have_not_locks[have_not_locks_len++].extra = p->common.id;
     }
+    if (locks & ERTS_PROC_LOCK_BTM) {
+	have_locks[have_locks_len].id = lc_id.proc_lock_btm;
+	have_locks[have_locks_len++].extra = p->common.id;
+    }
+    else {
+	have_not_locks[have_not_locks_len].id = lc_id.proc_lock_btm;
+	have_not_locks[have_not_locks_len++].extra = p->common.id;
+    }
     if (locks & ERTS_PROC_LOCK_STATUS) {
 	have_locks[have_locks_len].id = lc_id.proc_lock_status;
 	have_locks[have_locks_len++].extra = p->common.id;
@@ -1530,8 +1603,8 @@ erts_proc_lc_chk_proc_locks(Process *p, ErtsProcLocks locks)
 	have_not_locks[have_not_locks_len++].extra = p->common.id;
     }
 #elif ERTS_PROC_LOCK_RAW_MUTEX_IMPL
-    erts_lc_lock_t have_locks[4];
-    erts_lc_lock_t have_not_locks[4];
+    erts_lc_lock_t have_locks[5];
+    erts_lc_lock_t have_not_locks[5];
 
     if (locks & ERTS_PROC_LOCK_MAIN)
 	have_locks[have_locks_len++] = p->lock.main.lc;
@@ -1545,6 +1618,10 @@ erts_proc_lc_chk_proc_locks(Process *p, ErtsProcLocks locks)
 	have_locks[have_locks_len++] = p->lock.msgq.lc;
     else
 	have_not_locks[have_not_locks_len++] = p->lock.msgq.lc;
+    if (locks & ERTS_PROC_LOCK_BTM)
+	have_locks[have_locks_len++] = p->lock.btm.lc;
+    else
+	have_not_locks[have_not_locks_len++] = p->lock.btm.lc;
     if (locks & ERTS_PROC_LOCK_STATUS)
 	have_locks[have_locks_len++] = p->lock.status.lc;
     else
@@ -1558,10 +1635,10 @@ erts_proc_lc_chk_proc_locks(Process *p, ErtsProcLocks locks)
 ErtsProcLocks
 erts_proc_lc_my_proc_locks(Process *p)
 {
-    int resv[4];
+    int resv[5];
     ErtsProcLocks res = 0;
 #if ERTS_PROC_LOCK_OWN_IMPL
-    erts_lc_lock_t locks[4] = {ERTS_LC_LOCK_INIT(lc_id.proc_lock_main,
+    erts_lc_lock_t locks[5] = {ERTS_LC_LOCK_INIT(lc_id.proc_lock_main,
 						 p->common.id,
 						 ERTS_LC_FLG_LT_PROCLOCK),
 			       ERTS_LC_LOCK_INIT(lc_id.proc_lock_link,
@@ -1570,17 +1647,21 @@ erts_proc_lc_my_proc_locks(Process *p)
 			       ERTS_LC_LOCK_INIT(lc_id.proc_lock_msgq,
 						 p->common.id,
 						 ERTS_LC_FLG_LT_PROCLOCK),
+			       ERTS_LC_LOCK_INIT(lc_id.proc_lock_btm,
+						 p->common.id,
+						 ERTS_LC_FLG_LT_PROCLOCK),
 			       ERTS_LC_LOCK_INIT(lc_id.proc_lock_status,
 						 p->common.id,
 						 ERTS_LC_FLG_LT_PROCLOCK)};
 #elif ERTS_PROC_LOCK_RAW_MUTEX_IMPL
-    erts_lc_lock_t locks[4] = {p->lock.main.lc,
+    erts_lc_lock_t locks[5] = {p->lock.main.lc,
 			       p->lock.link.lc,
 			       p->lock.msgq.lc,
+			       p->lock.btm.lc,
 			       p->lock.status.lc};
 #endif
 
-    erts_lc_have_locks(resv, locks, 4);
+    erts_lc_have_locks(resv, locks, 5);
     if (resv[0])
 	res |= ERTS_PROC_LOCK_MAIN;
     if (resv[1])
@@ -1588,6 +1669,8 @@ erts_proc_lc_my_proc_locks(Process *p)
     if (resv[2])
 	res |= ERTS_PROC_LOCK_MSGQ;
     if (resv[3])
+	res |= ERTS_PROC_LOCK_BTM;
+    if (resv[4])
 	res |= ERTS_PROC_LOCK_STATUS;
 
     return res;
@@ -1596,13 +1679,14 @@ erts_proc_lc_my_proc_locks(Process *p)
 void
 erts_proc_lc_chk_no_proc_locks(char *file, int line)
 {
-    int resv[4];
-    int ids[4] = {lc_id.proc_lock_main,
+    int resv[5];
+    int ids[5] = {lc_id.proc_lock_main,
 		  lc_id.proc_lock_link,
 		  lc_id.proc_lock_msgq,
+		  lc_id.proc_lock_btm,
 		  lc_id.proc_lock_status};
-    erts_lc_have_lock_ids(resv, ids, 4);
-    if (!ERTS_IS_CRASH_DUMPING && (resv[0] || resv[1] || resv[2] || resv[3])) {
+    erts_lc_have_lock_ids(resv, ids, 5);
+    if (!ERTS_IS_CRASH_DUMPING && (resv[0] || resv[1] || resv[2] || resv[3] || resv[4])) {
 	erts_lc_fail("%s:%d: Thread has process locks locked when expected "
 		     "not to have any process locks locked",
 		     file, line);
diff --git a/erts/emulator/beam/erl_process_lock.h b/erts/emulator/beam/erl_process_lock.h
index 052d992d3f..8957e7773b 100644
--- a/erts/emulator/beam/erl_process_lock.h
+++ b/erts/emulator/beam/erl_process_lock.h
@@ -65,7 +65,7 @@
 
 #endif
 
-#define ERTS_PROC_LOCK_MAX_BIT 3
+#define ERTS_PROC_LOCK_MAX_BIT 4
 
 typedef erts_aint32_t ErtsProcLocks;
 
@@ -81,17 +81,18 @@ typedef struct erts_proc_lock_t_ {
     erts_lcnt_lock_t lcnt_main;
     erts_lcnt_lock_t lcnt_link;
     erts_lcnt_lock_t lcnt_msgq;
+    erts_lcnt_lock_t lcnt_btm;
     erts_lcnt_lock_t lcnt_status;
 #endif
 #elif ERTS_PROC_LOCK_RAW_MUTEX_IMPL
     erts_mtx_t main;
     erts_mtx_t link;
     erts_mtx_t msgq;
+    erts_mtx_t btm;
     erts_mtx_t status;
 #else
 #  error "no implementation"
 #endif
-    erts_atomic32_t refc;
 #ifdef ERTS_PROC_LOCK_DEBUG
     erts_smp_atomic32_t locked[ERTS_PROC_LOCK_MAX_BIT+1];
 #endif
@@ -120,11 +121,17 @@ typedef struct erts_proc_lock_t_ {
  * Message queue lock:
  *   Protects the following fields in the process structure:
  *   * msg_inq
- *   * bif_timers
  */
 #define ERTS_PROC_LOCK_MSGQ		(((ErtsProcLocks) 1) << 2)
 
 /*
+ * Bif timer lock:
+ *   Protects the following fields in the process structure:
+ *   * bif_timers
+ */
+#define ERTS_PROC_LOCK_BTM		(((ErtsProcLocks) 1) << 3)
+
+/*
  * Status lock:
  *   Protects the following fields in the process structure:
  *   * pending_suspenders
@@ -463,6 +470,9 @@ erts_smp_proc_raw_trylock__(Process *p, ErtsProcLocks locks)
     if (locks & ERTS_PROC_LOCK_MSGQ)
 	if (erts_mtx_trylock(&p->lock.msgq) == EBUSY)
 	    goto busy_msgq;
+    if (locks & ERTS_PROC_LOCK_BTM)
+	if (erts_mtx_trylock(&p->lock.btm) == EBUSY)
+	    goto busy_btm;
     if (locks & ERTS_PROC_LOCK_STATUS)
 	if (erts_mtx_trylock(&p->lock.status) == EBUSY)
 	    goto busy_status;
@@ -470,6 +480,9 @@ erts_smp_proc_raw_trylock__(Process *p, ErtsProcLocks locks)
     return 0;
 
 busy_status:
+    if (locks & ERTS_PROC_LOCK_BTM)
+	erts_mtx_unlock(&p->lock.btm);
+busy_btm:
     if (locks & ERTS_PROC_LOCK_MSGQ)
 	erts_mtx_unlock(&p->lock.msgq);
 busy_msgq:
@@ -549,6 +562,8 @@ erts_smp_proc_lock__(Process *p,
 	erts_mtx_lock(&p->lock.link);
     if (locks & ERTS_PROC_LOCK_MSGQ)
 	erts_mtx_lock(&p->lock.msgq);
+    if (locks & ERTS_PROC_LOCK_BTM)
+	erts_mtx_lock(&p->lock.btm);
     if (locks & ERTS_PROC_LOCK_STATUS)
 	erts_mtx_lock(&p->lock.status);
 
@@ -638,6 +653,8 @@ erts_smp_proc_unlock__(Process *p,
 
     if (locks & ERTS_PROC_LOCK_STATUS)
 	erts_mtx_unlock(&p->lock.status);
+    if (locks & ERTS_PROC_LOCK_BTM)
+	erts_mtx_unlock(&p->lock.btm);
     if (locks & ERTS_PROC_LOCK_MSGQ)
 	erts_mtx_unlock(&p->lock.msgq);
     if (locks & ERTS_PROC_LOCK_LINK)
@@ -752,9 +769,10 @@ ERTS_GLB_INLINE void erts_smp_proc_lock(Process *, ErtsProcLocks);
 ERTS_GLB_INLINE void erts_smp_proc_unlock(Process *, ErtsProcLocks);
 ERTS_GLB_INLINE int erts_smp_proc_trylock(Process *, ErtsProcLocks);
 
-ERTS_GLB_INLINE void erts_smp_proc_inc_refc(Process *);
-ERTS_GLB_INLINE void erts_smp_proc_dec_refc(Process *);
-ERTS_GLB_INLINE void erts_smp_proc_add_refc(Process *, Sint32);
+ERTS_GLB_INLINE void erts_proc_inc_refc(Process *);
+ERTS_GLB_INLINE void erts_proc_dec_refc(Process *);
+ERTS_GLB_INLINE void erts_proc_add_refc(Process *, Sint);
+ERTS_GLB_INLINE Sint erts_proc_read_refc(Process *);
 
 #if ERTS_GLB_INLINE_INCL_FUNC_DEF
 
@@ -814,28 +832,59 @@ erts_smp_proc_trylock(Process *p, ErtsProcLocks locks)
 #endif
 }
 
-ERTS_GLB_INLINE void erts_smp_proc_inc_refc(Process *p)
+ERTS_GLB_INLINE void erts_proc_inc_refc(Process *p)
 {
+    ASSERT(!(erts_smp_atomic32_read_nob(&p->state) & ERTS_PSFLG_PROXY));
 #ifdef ERTS_SMP
+    erts_ptab_atmc_inc_refc(&p->common);
+#else
     erts_ptab_inc_refc(&p->common);
 #endif
 }
 
-ERTS_GLB_INLINE void erts_smp_proc_dec_refc(Process *p)
+ERTS_GLB_INLINE void erts_proc_dec_refc(Process *p)
 {
+    Sint referred;
+    ASSERT(!(erts_smp_atomic32_read_nob(&p->state) & ERTS_PSFLG_PROXY));
 #ifdef ERTS_SMP
-    int referred = erts_ptab_dec_test_refc(&p->common);
-    if (!referred)
-	erts_free_proc(p);
+    referred = erts_ptab_atmc_dec_test_refc(&p->common);
+#else
+    referred = erts_ptab_dec_test_refc(&p->common);
 #endif
+    if (!referred) {
+	ASSERT(ERTS_PROC_IS_EXITING(p));
+	ASSERT(ERTS_AINT_NULL
+	       == erts_ptab_pix2intptr_ddrb(&erts_proc,
+					    internal_pid_index(p->common.id)));
+	erts_free_proc(p);
+    }
 }
 
-ERTS_GLB_INLINE void erts_smp_proc_add_refc(Process *p, Sint32 add_refc)
+ERTS_GLB_INLINE void erts_proc_add_refc(Process *p, Sint add_refc)
 {
+    Sint referred;
+    ASSERT(!(erts_smp_atomic32_read_nob(&p->state) & ERTS_PSFLG_PROXY));
 #ifdef ERTS_SMP
-    int referred = erts_ptab_add_test_refc(&p->common, add_refc);
-    if (!referred)
+    referred = erts_ptab_atmc_add_test_refc(&p->common, add_refc);
+#else
+    referred = erts_ptab_add_test_refc(&p->common, add_refc);
+#endif
+    if (!referred) {
+	ASSERT(ERTS_PROC_IS_EXITING(p));
+	ASSERT(ERTS_AINT_NULL
+	       == erts_ptab_pix2intptr_ddrb(&erts_proc,
+					    internal_pid_index(p->common.id)));
 	erts_free_proc(p);
+    }
+}
+
+ERTS_GLB_INLINE Sint erts_proc_read_refc(Process *p)
+{
+    ASSERT(!(erts_smp_atomic32_read_nob(&p->state) & ERTS_PSFLG_PROXY));
+#ifdef ERTS_SMP
+    return erts_ptab_atmc_read_refc(&p->common);
+#else
+    return erts_ptab_read_refc(&p->common);
 #endif
 }
 
@@ -868,7 +917,7 @@ void erts_proc_safelock(Process *a_proc,
 
 #define ERTS_P2P_FLG_ALLOW_OTHER_X	(1 <<  0)
 #define ERTS_P2P_FLG_TRY_LOCK		(1 <<  1)
-#define ERTS_P2P_FLG_SMP_INC_REFC	(1 <<  2)
+#define ERTS_P2P_FLG_INC_REFC		(1 <<  2)
 
 #define ERTS_PROC_LOCK_BUSY ((Process *) &erts_invalid_process)
 
@@ -928,11 +977,14 @@ erts_pid2proc_opt(Process *c_p_unused,
 		  int flags)
 {
     Process *proc = erts_proc_lookup_raw(pid);
-    return ((!(flags & ERTS_P2P_FLG_ALLOW_OTHER_X)
-	     && proc
-	     && ERTS_PROC_IS_EXITING(proc))
-	    ? NULL
-	    : proc);
+    if (!proc)
+	return NULL;
+    if (!(flags & ERTS_P2P_FLG_ALLOW_OTHER_X)
+	&& ERTS_PROC_IS_EXITING(proc))
+	return NULL;
+    if (flags & ERTS_P2P_FLG_INC_REFC)
+	erts_proc_inc_refc(proc);
+    return proc;
 }
 #endif /* !ERTS_SMP */
 
diff --git a/erts/emulator/beam/erl_ptab.c b/erts/emulator/beam/erl_ptab.c
index 02943ee683..c688db98d8 100644
--- a/erts/emulator/beam/erl_ptab.c
+++ b/erts/emulator/beam/erl_ptab.c
@@ -360,7 +360,8 @@ erts_ptab_init_table(ErtsPTab *ptab,
 		     int size,
 		     UWord element_size,
 		     char *name,
-		     int legacy)
+		     int legacy,
+		     int atomic_refc)
 {
     size_t tab_sz, alloc_sz;
     Uint32 bits, cl, cli, ix, ix_per_cache_line, tab_cache_lines; 
@@ -415,6 +416,8 @@ erts_ptab_init_table(ErtsPTab *ptab,
     ptab->r.o.invalid_data = erts_ptab_id2data(ptab, invalid_element->id);
     ptab->r.o.release_element = release_element;
 
+    ptab->r.o.atomic_refc = atomic_refc;
+
     if (legacy) {
 	ptab->r.o.free_id_data = NULL;
 	ptab->r.o.dix_cl_mask = 0;
@@ -533,9 +536,10 @@ erts_ptab_new_element(ErtsPTab *ptab,
 
 	init_ptab_el(init_arg, (Eterm) data);
 
-#ifdef ERTS_SMP
-	erts_smp_atomic32_init_nob(&ptab_el->refc, 1);
-#endif
+	if (ptab->r.o.atomic_refc)
+	    erts_atomic_init_nob(&ptab_el->refc.atmc, 1);
+	else
+	    ptab_el->refc.sint = 1;
 
 	pix = erts_ptab_data2pix(ptab, (Eterm) data);
 
@@ -608,9 +612,10 @@ erts_ptab_new_element(ErtsPTab *ptab,
 
 	init_ptab_el(init_arg, data);
 
-#ifdef ERTS_SMP
-	erts_smp_atomic32_init_nob(&ptab_el->refc, 1);
-#endif
+	if (ptab->r.o.atomic_refc)
+	    erts_atomic_init_nob(&ptab_el->refc.atmc, 1);
+	else
+	    ptab_el->refc.sint = 1;
 
 	/* Move into slot reserved */
 #ifdef DEBUG
diff --git a/erts/emulator/beam/erl_ptab.h b/erts/emulator/beam/erl_ptab.h
index 876241159b..102d41e07f 100644
--- a/erts/emulator/beam/erl_ptab.h
+++ b/erts/emulator/beam/erl_ptab.h
@@ -51,11 +51,13 @@
 
 typedef struct {
     Eterm id;
-#ifdef ERTS_SMP
-    erts_atomic32_t refc;
-#endif
+    union {
+	erts_atomic_t atmc;
+	Sint sint;
+    } refc;
     Eterm tracer_proc;
     Uint trace_flags;
+    erts_smp_atomic_t timer;
     union {
 	/* --- While being alive --- */
 	struct {
@@ -63,11 +65,6 @@ typedef struct {
 	    struct reg_proc *reg;
 	    ErtsLink *links;
 	    ErtsMonitor *monitors;
-#ifdef ERTS_SMP
-	    ErtsSmpPTimer *ptimer;
-#else
-	    ErlTimer tm;
-#endif
 	} alive;
 
 	/* --- While being released --- */
@@ -111,6 +108,7 @@ typedef struct {
     Eterm invalid_data;
     void (*release_element)(void *);
     UWord element_size;
+    int atomic_refc;
 } ErtsPTabReadOnlyData;
 
 typedef struct {
@@ -181,7 +179,8 @@ void erts_ptab_init_table(ErtsPTab *ptab,
 			  int size,
 			  UWord element_size,
 			  char *name,
-			  int legacy);
+			  int legacy,
+			  int atomic_refc);
 int erts_ptab_new_element(ErtsPTab *ptab,
 			  ErtsPTabElementCommon *ptab_el,
 			  void *init_arg,
@@ -206,9 +205,15 @@ ERTS_GLB_INLINE erts_aint_t erts_ptab_pix2intptr_ddrb(ErtsPTab *ptab, int ix);
 ERTS_GLB_INLINE erts_aint_t erts_ptab_pix2intptr_rb(ErtsPTab *ptab, int ix);
 ERTS_GLB_INLINE erts_aint_t erts_ptab_pix2intptr_acqb(ErtsPTab *ptab, int ix);
 ERTS_GLB_INLINE void erts_ptab_inc_refc(ErtsPTabElementCommon *ptab_el);
-ERTS_GLB_INLINE int erts_ptab_dec_test_refc(ErtsPTabElementCommon *ptab_el);
-ERTS_GLB_INLINE int erts_ptab_add_test_refc(ErtsPTabElementCommon *ptab_el,
-					    Sint32 add_refc);
+ERTS_GLB_INLINE Sint erts_ptab_dec_test_refc(ErtsPTabElementCommon *ptab_el);
+ERTS_GLB_INLINE Sint erts_ptab_add_test_refc(ErtsPTabElementCommon *ptab_el,
+					     Sint add_refc);
+ERTS_GLB_INLINE Sint erts_ptab_read_refc(ErtsPTabElementCommon *ptab_el);
+ERTS_GLB_INLINE void erts_ptab_atmc_inc_refc(ErtsPTabElementCommon *ptab_el);
+ERTS_GLB_INLINE Sint erts_ptab_atmc_dec_test_refc(ErtsPTabElementCommon *ptab_el);
+ERTS_GLB_INLINE Sint erts_ptab_atmc_add_test_refc(ErtsPTabElementCommon *ptab_el,
+						  Sint add_refc);
+ERTS_GLB_INLINE Sint erts_ptab_atmc_read_refc(ErtsPTabElementCommon *ptab_el);
 ERTS_GLB_INLINE void erts_ptab_rlock(ErtsPTab *ptab);
 ERTS_GLB_INLINE int erts_ptab_tryrlock(ErtsPTab *ptab);
 ERTS_GLB_INLINE void erts_ptab_runlock(ErtsPTab *ptab);
@@ -365,50 +370,65 @@ ERTS_GLB_INLINE erts_aint_t erts_ptab_pix2intptr_acqb(ErtsPTab *ptab, int ix)
     return erts_smp_atomic_read_acqb(&ptab->r.o.tab[ix]);
 }
 
-ERTS_GLB_INLINE void erts_ptab_inc_refc(ErtsPTabElementCommon *ptab_el)
+ERTS_GLB_INLINE void erts_ptab_atmc_inc_refc(ErtsPTabElementCommon *ptab_el)
 {
-#ifdef ERTS_SMP
 #ifdef ERTS_ENABLE_LOCK_CHECK
-    erts_aint32_t refc = erts_atomic32_inc_read_nob(&ptab_el->refc);
-    ERTS_SMP_LC_ASSERT(refc > 1);
+    erts_aint_t refc = erts_atomic_inc_read_nob(&ptab_el->refc.atmc);
+    ERTS_LC_ASSERT(refc > 1);
 #else
-    erts_atomic32_inc_nob(&ptab_el->refc);
-#endif
+    erts_atomic_inc_nob(&ptab_el->refc.atmc);
 #endif
 }
 
-ERTS_GLB_INLINE int erts_ptab_dec_test_refc(ErtsPTabElementCommon *ptab_el)
+ERTS_GLB_INLINE Sint erts_ptab_atmc_dec_test_refc(ErtsPTabElementCommon *ptab_el)
 {
-#ifdef ERTS_SMP
-    erts_aint32_t refc = erts_atomic32_dec_read_nob(&ptab_el->refc);
+    erts_aint_t refc = erts_atomic_dec_read_relb(&ptab_el->refc.atmc);
     ERTS_SMP_LC_ASSERT(refc >= 0);
-    return (int) refc;
-#else
-    return 0;
+#ifdef ERTS_SMP
+    if (refc == 0)
+	ETHR_MEMBAR(ETHR_LoadLoad|ETHR_LoadStore);
 #endif
+    return (Sint) refc;
 }
 
-ERTS_GLB_INLINE int erts_ptab_add_test_refc(ErtsPTabElementCommon *ptab_el,
-					    Sint32 add_refc)
+ERTS_GLB_INLINE Sint erts_ptab_atmc_add_test_refc(ErtsPTabElementCommon *ptab_el,
+						  Sint add_refc)
 {
-#ifdef ERTS_SMP
-    erts_aint32_t refc;
+    erts_aint_t refc = erts_atomic_add_read_mb(&ptab_el->refc.atmc,
+					       (erts_aint_t) add_refc);
+    ERTS_SMP_LC_ASSERT(refc >= 0);
+    return (Sint) refc;
+}
 
-#ifndef ERTS_ENABLE_LOCK_CHECK
-    if (add_refc >= 0) {
-	erts_atomic32_add_nob(&ptab_el->refc,
-			      (erts_aint32_t) add_refc);
-	return 1;
-    }
-#endif
+ERTS_GLB_INLINE Sint erts_ptab_atmc_read_refc(ErtsPTabElementCommon *ptab_el)
+{
+    return (Sint) erts_atomic_read_nob(&ptab_el->refc.atmc);
+}
+
+ERTS_GLB_INLINE void erts_ptab_inc_refc(ErtsPTabElementCommon *ptab_el)
+{
+    ptab_el->refc.sint++;
+    ASSERT(ptab_el->refc.sint > 1);
+}
 
-    refc = erts_atomic32_add_read_nob(&ptab_el->refc,
-				      (erts_aint32_t) add_refc);
+ERTS_GLB_INLINE Sint erts_ptab_dec_test_refc(ErtsPTabElementCommon *ptab_el)
+{
+    Sint refc = --ptab_el->refc.sint;
     ERTS_SMP_LC_ASSERT(refc >= 0);
-    return (int) refc;
-#else
-    return 0;
-#endif
+    return refc;
+}
+
+ERTS_GLB_INLINE Sint erts_ptab_add_test_refc(ErtsPTabElementCommon *ptab_el,
+					     Sint add_refc)
+{
+    ptab_el->refc.sint += add_refc;
+    ERTS_SMP_LC_ASSERT(ptab_el->refc.sint >= 0);
+    return (Sint) ptab_el->refc.sint;
+}
+
+ERTS_GLB_INLINE Sint erts_ptab_read_refc(ErtsPTabElementCommon *ptab_el)
+{
+    return ptab_el->refc.sint;
 }
 
 ERTS_GLB_INLINE void erts_ptab_rlock(ErtsPTab *ptab)
diff --git a/erts/emulator/beam/erl_rbtree.h b/erts/emulator/beam/erl_rbtree.h
new file mode 100644
index 0000000000..ea0a8976bb
--- /dev/null
+++ b/erts/emulator/beam/erl_rbtree.h
@@ -0,0 +1,1740 @@
+/*
+ * %CopyrightBegin%
+ * 
+ * Copyright Ericsson AB 2015. All Rights Reserved.
+ * 
+ * The contents of this file are subject to the Erlang Public License,
+ * Version 1.1, (the "License"); you may not use this file except in
+ * compliance with the License. You should have received a copy of the
+ * Erlang Public License along with this software. If not, it can be
+ * retrieved online at http://www.erlang.org/.
+ * 
+ * Software distributed under the License is distributed on an "AS IS"
+ * basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
+ * the License for the specific language governing rights and limitations
+ * under the License.
+ * 
+ * %CopyrightEnd%
+ */
+
+/*
+ * Description:	A Red-Black (binary search) Tree implementation. The search,
+ *              insert, and delete operations are all O(log n) operations
+ *              on a Red-Black Tree. Red-Black Trees are described in
+ *              "Introduction to Algorithms", by Thomas H. Cormen, Charles
+ *              E. Leiserson, and Ronald L. Riverest.
+ *
+ *              Use by defining mandatory defines as well as defines for
+ *              API functions wanted, and include this header.
+ *
+ * Author: 	Rickard Green
+ *
+ *
+ * Mandatory defines:
+ * - ERTS_RBT_PREFIX - Prefix to use on functions.
+ * - ERTS_RBT_T - Type of a tree node.
+ * - ERTS_RBT_KEY_T - Type of key for a tree node.
+ * - ERTS_RBT_FLAGS_T - Type of flags for a tree node.
+ * - ERTS_RBT_INIT_EMPTY_TNODE(T) -Initialize an empty tree node.
+ * - ERTS_RBT_IS_RED(T) - Is tree node red?
+ * - ERTS_RBT_SET_RED(T) - Set tree node red.
+ * - ERTS_RBT_IS_BLACK(T) - Is tree node back?
+ * - ERTS_RBT_SET_BLACK(T) - Set tree node black.
+ * - ERTS_RBT_GET_FLAGS(T) - Get flags of tree node (incl colors).
+ * - ERTS_RBT_SET_FLAGS(T, F) - Set flags of tree note.
+ * - ERTS_RBT_GET_PARENT(T) - Get parent node.
+ * - ERTS_RBT_SET_PARENT(T, P) - Set parent node.
+ * - ERTS_RBT_GET_RIGHT(T) - Get right child node.
+ * - ERTS_RBT_SET_RIGHT(T, R) - Set right child node.
+ * - ERTS_RBT_GET_LEFT(T) - Get left child node.
+ * - ERTS_RBT_SET_LEFT(T, L) - Set left child node.
+ * - ERTS_RBT_GET_KEY(T) - Get key of node.
+ * - ERTS_RBT_IS_LT(KX, KY) - Is key KX less than key KY?
+ * - ERTS_RBT_IS_EQ(KX, KY) - Is key KX equal to key KY?
+ *
+ * Optional defines:
+ *
+ * - ERTS_RBT_UNDEF - Undefine all user defined ERTS_RBT_*
+ *     defines after use.
+ *
+ * - ERTS_RBT_NO_API_INLINE - Do not inline API functions.
+ *
+ * Attached data management:
+ * - ERTS_RBT_UPDATE_ATTACHED_DATA_ROTATE(L, OP, NP) - Called
+ *     when a rotate operation has been performed. If L (in int)
+ *     is a non zero, a left rotation was performed; otherwise,
+ *     a right rotation was performed. OR points to the old
+ *     parent node and NP points to the new parent node.
+ * - ERTS_RBT_UPDATE_ATTACHED_DATA_DMOD(F, T) - Called when
+ *     a delete operation modifies a tree node. A delete
+ *     modification is either a removal or replacement of a
+ *     node. F points to the parent of the tree node that was
+ *     modified. T points to the next ancestor that will be
+ *     modified. If T is NULL, no more removal and/or
+ *     replacements will be made. One typically wants to update
+ *     the attached data of each node between F and T. If T is
+ *     NULL all the way up to the root.
+ * - ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT(OR, NR) - Called
+ *     when the root node changes. OR points to the old
+ *     root node and NP points to the new root node.
+ *
+ * Request implementation of API functions:
+ * - ERTS_RBT_WANT_DELETE
+ * - ERTS_RBT_WANT_INSERT
+ * - ERTS_RBT_WANT_LOOKUP_INSERT
+ * - ERTS_RBT_WANT_REPLACE
+ * - ERTS_RBT_WANT_LOOKUP
+ * - ERTS_RBT_WANT_SMALLEST
+ * - ERTS_RBT_WANT_LARGEST
+ * - ERTS_RBT_WANT_FOREACH
+ * - ERTS_RBT_WANT_FOREACH_DESTROY
+ * - ERTS_RBT_WANT_FOREACH_YIELDING
+ * - ERTS_RBT_WANT_FOREACH_DESTROY_YIELDING
+ * - ERTS_RBT_WANT_FOREACH_SMALL
+ * - ERTS_RBT_WANT_FOREACH_LARGE
+ * - ERTS_RBT_WANT_FOREACH_SMALL_DESTROY
+ * - ERTS_RBT_WANT_FOREACH_LARGE_DESTROY
+ * - ERTS_RBT_WANT_FOREACH_SMALL_YIELDING
+ * - ERTS_RBT_WANT_FOREACH_LARGE_YIELDING
+ * - ERTS_RBT_WANT_FOREACH_SMALL_DESTROY_YIELDING
+ * - ERTS_RBT_WANT_FOREACH_LARGE_DESTROY_YIELDING
+ * - ERTS_RBT_WANT_DEBUG_PRINT
+ *
+ * The yield state data type will equal
+ * <ERTS_RBT_PREFIX>_rbt_yield_state_t.
+ *
+ * The yield state should be statically initialized by
+ * ERTS_RBT_YIELD_STAT_INITER.
+ *
+ *
+ * The following API functions are implemented if corresponding
+ * ERTS_RBT_WANT_<OPERATION> is defined:
+ *
+ * - void <ERTS_RBT_PREFIX>_rbt_delete(
+ *                         ERTS_RBT_T **tree,
+ *                         ERTS_RBT_T *element);
+ *         Delete element from tree.
+ *
+ * - void <ERTS_RBT_PREFIX>_rbt_insert(
+ *                         ERTS_RBT_T **tree,
+ *                         ERTS_RBT_T *element);
+ *         Insert element into tree.
+ *
+ * - ERTS_RBT_T * <ERTS_RBT_PREFIX>_rbt_lookup_insert(
+ *                         ERTS_RBT_T **tree,
+ *                         ERTS_RBT_T *element);
+ *         Look up an element in the tree that compares as equal to the
+ *         element passed as argument, and return the looked up element.
+ *         If no element compared as equal, insert the element passed as
+ *         argument into the tree, and return NULL.
+ *
+ * - void <ERTS_RBT_PREFIX>_rbt_replace(
+ *                         ERTS_RBT_T **tree,
+ *                         ERTS_RBT_T *old_element,
+ *                         ERTS_RBT_T *new_element);
+ *         Replace old_element in the tree with new_element. Both elements
+ *         *should* compare as equal.
+ *
+ * - ERTS_RBT_T * <ERTS_RBT_PREFIX>_rbt_lookup(
+ *                         ERTS_RBT_T *tree,
+ *                         ERTS_RBT_KEY_T key);
+ *         Look up an element with a key that compares as equal to
+ *         the key passed as argument.
+ *
+ * - ERTS_RBT_T * <ERTS_RBT_PREFIX>_rbt_smallest(
+ *                         ERTS_RBT_T *tree);
+ *         Look up the element with the smallest key.
+ *
+ * - ERTS_RBT_T * <ERTS_RBT_PREFIX>_rbt_largest(
+ *                         ERTS_RBT_T *tree);
+ *         Look up the element with the largest key.
+ *
+ * - void <ERTS_RBT_PREFIX>_rbt_foreach(
+ *                         ERTS_RBT_T *tree,
+ *                         void (*op)(ERTS_RBT_T *, void *),
+ *                         void *arg);
+ *         Operate by calling the operator 'op' on each element.
+ *         Order is undefined.
+ *
+ *         'arg' is passed as argument to 'op'.
+ *
+ * - void <ERTS_RBT_PREFIX>_rbt_foreach_destroy(
+ *                         ERTS_RBT_T *tree,
+ *                         void (*op)(ERTS_RBT_T *, void *),
+ *                         void *arg);
+ *         Operate by calling the operator 'op' on each element.
+ *         Order is undefined. Each element should be destroyed
+ *         by 'op'.
+ *
+ *         'arg' is passed as argument to 'op'.
+ *
+ * - int <ERTS_RBT_PREFIX>_rbt_foreach_yielding(
+ *                         ERTS_RBT_T *tree,
+ *                         void (*op)(ERTS_RBT_T *, void *),
+ *                         void *arg,
+ *                         <ERTS_RBT_PREFIX>_rbt_yield_state_t *ystate,
+ *                         Sint ylimit);
+ *         Operate by calling the operator 'op' on each element.
+ *         Order is undefined.
+ *
+ *         Yield when 'ylimit' elements has been processed. Zero is
+ *         returned when yielding, and a non-zero value is returned when
+ *         the whole tree has been processed. The tree should not be
+ *         modified until all of it has been processed.
+ *
+ *         'arg' is passed as argument to 'op'.
+ *
+ * - int <ERTS_RBT_PREFIX>_rbt_foreach_destroy_yielding(
+ *                         ERTS_RBT_T *tree,
+ *                         void (*op)(ERTS_RBT_T *, void *),
+ *                         void *arg,
+ *                         <ERTS_RBT_PREFIX>_rbt_yield_state_t *ystate,
+ *                         Sint ylimit);
+ *         Operate by calling the operator 'op' on each element.
+ *         Order is undefined. Each element should be destroyed
+ *         by 'op'.
+ *
+ *         Yield when 'ylimit' elements has been processed. Zero is
+ *         returned when yielding, and a non-zero value is returned when
+ *         the whole tree has been processed.
+ *
+ *         'arg' is passed as argument to 'op'.
+ *
+ * - void <ERTS_RBT_PREFIX>_rbt_foreach_small(
+ *                         ERTS_RBT_T *tree,
+ *                         void (*op)(ERTS_RBT_T *, void *),
+ *                         void *arg);
+ *         Operate by calling the operator 'op' on each element from
+ *         smallest towards larger elements.
+ *
+ *         'arg' is passed as argument to 'op'.
+ *
+ * - void <ERTS_RBT_PREFIX>_rbt_foreach_large(
+ *                         ERTS_RBT_T *tree,
+ *                         void (*op)(ERTS_RBT_T *, void *),
+ *                         void *arg);
+ *         Operate by calling the operator 'op' on each element from
+ *         largest towards smaller elements.
+ *
+ *         'arg' is passed as argument to 'op'.
+ *
+ * - int <ERTS_RBT_PREFIX>_rbt_foreach_small_yielding(
+ *                         ERTS_RBT_T *tree,
+ *                         void (*op)(ERTS_RBT_T *, void *),
+ *                         void *arg,
+ *                         <ERTS_RBT_PREFIX>_rbt_yield_state_t *ystate,
+ *                         Sint ylimit);
+ *         Operate by calling the operator 'op' on each element from
+ *         smallest towards larger elements.
+ *
+ *         Yield when 'ylimit' elements has been processed. Zero is
+ *         returned when yielding, and a non-zero value is returned when
+ *         the whole tree has been processed. The tree should not be
+ *         modified until all of it has been processed.
+ *
+ *         'arg' is passed as argument to 'op'.
+ *
+ * - int <ERTS_RBT_PREFIX>_rbt_foreach_large_yielding(
+ *                         ERTS_RBT_T *tree,
+ *                         void (*op)(ERTS_RBT_T *, void *),
+ *                         void *arg,
+ *                         <ERTS_RBT_PREFIX>_rbt_yield_state_t *ystate,
+ *                         Sint ylimit);
+ *         Operate by calling the operator 'op' on each element from
+ *         largest towards smaller elements.
+ *
+ *         Yield when 'ylimit' elements has been processed. Zero is
+ *         returned when yielding, and a non-zero value is returned when
+ *         the whole tree has been processed. The tree should not be
+ *         modified until all of it has been processed.
+ *
+ *         'arg' is passed as argument to 'op'.
+ *
+ * - void <ERTS_RBT_PREFIX>_rbt_foreach_small_destroy(
+ *                         ERTS_RBT_T **tree,
+ *                         void (*op)(ERTS_RBT_T *, void *),
+ *                         void (*destr)(ERTS_RBT_T *, void *),
+ *                         void *arg);
+ *         Operate by calling the operator 'op' on each element from
+ *         smallest towards larger elements.
+ *
+ *         Destroy elements by calling the destructor 'destr'. Elements
+ *         are destroyed when not needed by the tree structure anymore.
+ *         Note that elements are often *not* destroyed in another order
+ *         than the order that the elements are operated on.
+ *
+ *         'arg' is passed as argument to 'op' and 'destroy'.
+ *
+ * - void <ERTS_RBT_PREFIX>_rbt_foreach_large_destroy(
+ *                         ERTS_RBT_T **tree,
+ *                         void (*op)(ERTS_RBT_T *, void *),
+ *                         void (*destr)(ERTS_RBT_T *, void *),
+ *                         void *arg);
+ *         Operate by calling the operator 'op' on each element from
+ *         largest towards smaller elements.
+ *
+ *         Destroy elements by calling the destructor 'destr'. Elements
+ *         are destroyed when not needed by the tree structure anymore.
+ *         Note that elements are often destroyed in another order
+ *         than the order that the elements are operated on.
+ *
+ *         'arg' is passed as argument to 'op' and 'destroy'.
+ *
+ * - int <ERTS_RBT_PREFIX>_rbt_foreach_small_destroy_yielding(
+ *                        ERTS_RBT_T **tree,
+ *                        void (*op)(ERTS_RBT_T *, void *),
+ *                        void (*destr)(ERTS_RBT_T *, void *),
+ *                        void *arg,
+ *                        <ERTS_RBT_PREFIX>_rbt_yield_state_t *ystate,
+ *                        Sint ylimit);
+ *         Operate by calling the operator 'op' on each element from
+ *         smallest towards larger elements.
+ *
+ *         Destroy elements by calling the destructor 'destr'. Elements
+ *         are destroyed when not needed by the tree structure anymore.
+ *         Note that elements are often destroyed in another order
+ *         than the order that the elements are operated on.
+ *
+ *         Yield when 'ylimit' elements has been processed. Zero is
+ *         returned when yielding, and a non-zero value is returned when
+ *         the whole tree has been processed. The tree should not be
+ *         modified until all of it has been processed.
+ *
+ *         'arg' is passed as argument to 'op' and 'destroy'.
+ *
+ * - int <ERTS_RBT_PREFIX>_rbt_foreach_large_destroy_yielding(
+ *                        ERTS_RBT_T **tree,
+ *                        void (*op)(ERTS_RBT_T *, void *),
+ *                        void (*destr)(ERTS_RBT_T *, void *),
+ *                        void *arg,
+ *                        <ERTS_RBT_PREFIX>_rbt_yield_state_t *ystate,
+ *                        Sint ylimit);
+ *         Operate by calling the operator 'op' on each element from
+ *         largest towards smaller elements.
+ *
+ *         Destroy elements by calling the destructor 'destr'. Elements
+ *         are destroyed when not needed by the tree structure anymore.
+ *         Note that elements are often destroyed in another order
+ *         than the order that the elements are operated on.
+ *
+ *         Yield when 'ylimit' elements has been processed. Zero is
+ *         returned when yielding, and a non-zero value is returned when
+ *         the whole tree has been processed. The tree should not be
+ *         modified until all of it has been processed.
+ *
+ *         'arg' is passed as argument to 'op' and 'destroy'.
+ *
+ * - void <ERTS_RBT_PREFIX>_rbt_debug_print(
+ *                        FILE *filep,
+ *                        ERTS_RBT_T *x,
+ *                        int indent,
+ *                        (void)(*print_node)(ERTS_RBT_T *));
+ *         Prints the tree. Note that this function is recursive.
+ *         Should only be used for debuging.
+ */
+
+
+/*
+ * Check that we have all mandatory defines
+ */
+#ifndef ERTS_RBT_PREFIX
+#  error Missing definition of ERTS_RBT_PREFIX
+#endif
+#ifndef ERTS_RBT_T
+#  error Missing definition of ERTS_RBT_T
+#endif
+#ifndef ERTS_RBT_KEY_T
+#  error Missing definition of ERTS_RBT_KEY_T
+#endif
+#ifndef ERTS_RBT_FLAGS_T
+#  error Missing definition of ERTS_RBT_FLAGS_T
+#endif
+#ifndef ERTS_RBT_INIT_EMPTY_TNODE
+#  error Missing definition of ERTS_RBT_INIT_EMPTY_TNODE
+#endif
+#ifndef ERTS_RBT_IS_RED
+#  error Missing definition of ERTS_RBT_IS_RED
+#endif
+#ifndef ERTS_RBT_SET_RED
+#  error Missing definition of ERTS_RBT_SET_RED
+#endif
+#ifndef ERTS_RBT_IS_BLACK
+#  error Missing definition of ERTS_RBT_IS_BLACK
+#endif
+#ifndef ERTS_RBT_SET_BLACK
+#  error Missing definition of ERTS_RBT_SET_BLACK
+#endif
+#ifndef ERTS_RBT_GET_FLAGS
+#  error Missing definition of ERTS_RBT_GET_FLAGS
+#endif
+#ifndef ERTS_RBT_SET_FLAGS
+#  error Missing definition of ERTS_RBT_SET_FLAGS
+#endif
+#ifndef ERTS_RBT_GET_PARENT
+#  error Missing definition of ERTS_RBT_GET_PARENT
+#endif
+#ifndef ERTS_RBT_SET_PARENT
+#  error Missing definition of ERTS_RBT_SET_PARENT
+#endif
+#ifndef ERTS_RBT_GET_RIGHT
+#  error Missing definition of ERTS_RBT_GET_RIGHT
+#endif
+#ifndef ERTS_RBT_GET_LEFT
+#  error Missing definition of ERTS_RBT_GET_LEFT
+#endif
+#ifndef ERTS_RBT_IS_LT
+#  error Missing definition of ERTS_RBT_IS_LT
+#endif
+#ifndef ERTS_RBT_GET_KEY
+#  error Missing definition of ERTS_RBT_GET_KEY
+#endif
+#ifndef ERTS_RBT_IS_EQ
+#  error Missing definition of ERTS_RBT_IS_EQ
+#endif
+
+#if defined(ERTS_RBT_HARD_DEBUG) || defined(DEBUG)
+#  ifndef ERTS_RBT_DEBUG
+#    define ERTS_RBT_DEBUG 1
+#  endif
+#endif
+
+#if defined(ERTS_RBT_HARD_DEBUG) && defined(__GNUC__)
+#warning "* * * * * * * * * * * * * * * * * *"
+#warning "* ERTS_RBT_HARD_DEBUG IS ENABLED! *"
+#warning "* * * * * * * * * * * * * * * * * *"
+#endif
+
+#undef ERTS_RBT_ASSERT
+#if defined(ERTS_RBT_DEBUG)
+#define ERTS_RBT_ASSERT(E) ERTS_ASSERT(E)
+#else
+#define ERTS_RBT_ASSERT(E) ((void) 1)
+#endif
+
+#undef ERTS_RBT_API_INLINE__
+#if defined(ERTS_RBT_NO_API_INLINE) || defined(ERTS_RBT_DEBUG)
+#  define ERTS_RBT_API_INLINE__
+#else
+#  define ERTS_RBT_API_INLINE__ ERTS_INLINE
+#endif
+
+#ifndef ERTS_RBT_YIELD_STAT_INITER
+#  define ERTS_RBT_YIELD_STAT_INITER {NULL, 0}
+#endif
+
+#define ERTS_RBT_CONCAT_MACRO_VALUES___(X, Y) \
+    X ## Y
+#define ERTS_RBT_CONCAT_MACRO_VALUES__(X, Y) \
+    ERTS_RBT_CONCAT_MACRO_VALUES___(X, Y)
+
+#undef ERTS_RBT_YIELD_STATE_T__
+#define ERTS_RBT_YIELD_STATE_T__ \
+    ERTS_RBT_CONCAT_MACRO_VALUES__(ERTS_RBT_PREFIX, _rbt_yield_state_t)
+
+typedef struct {
+    ERTS_RBT_T *x;
+    int up;
+} ERTS_RBT_YIELD_STATE_T__;
+
+#define ERTS_RBT_FUNC__(Name) \
+    ERTS_RBT_CONCAT_MACRO_VALUES__(ERTS_RBT_PREFIX, _rbt_ ## Name)
+
+#undef ERTS_RBT_NEED_REPLACE__
+#undef ERTS_RBT_NEED_INSERT__
+#undef ERTS_RBT_NEED_ROTATE__
+#undef ERTS_RBT_NEED_FOREACH_UNORDERED__
+#undef ERTS_RBT_NEED_FOREACH_ORDERED__
+#undef ERTS_RBT_NEED_HDBG_CHECK_TREE__
+#undef ERTS_RBT_HDBG_CHECK_TREE__
+
+#if defined(ERTS_RBT_WANT_REPLACE) || defined(ERTS_RBT_WANT_DELETE)
+#  define ERTS_RBT_NEED_REPLACE__
+#endif
+#if defined(ERTS_RBT_WANT_INSERT) || defined(ERTS_RBT_WANT_LOOKUP_INSERT)
+#  define ERTS_RBT_NEED_INSERT__
+#endif
+#if defined(ERTS_RBT_WANT_DELETE) || defined(ERTS_RBT_NEED_INSERT__)
+#  define ERTS_RBT_NEED_ROTATE__
+#endif
+#if defined(ERTS_RBT_WANT_FOREACH) \
+    || defined(ERTS_RBT_WANT_FOREACH_YIELDING) \
+    || defined(ERTS_RBT_WANT_FOREACH_DESTROY) \
+    || defined(ERTS_RBT_WANT_FOREACH_DESTROY_YIELDING)
+#  define ERTS_RBT_NEED_FOREACH_UNORDERED__
+#endif
+#if defined(ERTS_RBT_WANT_FOREACH_SMALL) \
+    || defined(ERTS_RBT_WANT_FOREACH_LARGE) \
+    || defined(ERTS_RBT_WANT_FOREACH_SMALL_YIELDING) \
+    || defined(ERTS_RBT_WANT_FOREACH_LARGE_YIELDING) \
+    || defined(ERTS_RBT_WANT_FOREACH_SMALL_DESTROY) \
+    || defined(ERTS_RBT_WANT_FOREACH_LARGE_DESTROY) \
+    || defined(ERTS_RBT_WANT_FOREACH_SMALL_DESTROY_YIELDING) \
+    || defined(ERTS_RBT_WANT_FOREACH_LARGE_DESTROY_YIELDING)
+#  define ERTS_RBT_NEED_FOREACH_ORDERED__
+#endif
+#if defined(ERTS_RBT_HARD_DEBUG) \
+    && (defined(ERTS_RBT_WANT_DELETE) \
+	|| defined(ERTS_RBT_NEED_INSERT__))
+static void ERTS_RBT_FUNC__(hdbg_check_tree)(ERTS_RBT_T *root);
+#  define ERTS_RBT_NEED_HDBG_CHECK_TREE__
+#  define ERTS_RBT_HDBG_CHECK_TREE__(R) \
+    ERTS_RBT_FUNC__(hdbg_check_tree)((R))
+#else
+#  define ERTS_RBT_HDBG_CHECK_TREE__(R) ((void) 1)
+#endif
+
+#ifdef ERTS_RBT_NEED_ROTATE__
+
+static ERTS_INLINE void
+ERTS_RBT_FUNC__(left_rotate__)(ERTS_RBT_T **root, ERTS_RBT_T *x)
+{
+    ERTS_RBT_T *y, *l, *p;
+
+    y = ERTS_RBT_GET_RIGHT(x);
+    l = ERTS_RBT_GET_LEFT(y);
+    ERTS_RBT_SET_RIGHT(x, l);
+
+    if (l)
+	ERTS_RBT_SET_PARENT(l, x);
+
+    p = ERTS_RBT_GET_PARENT(x);
+    ERTS_RBT_SET_PARENT(y, p);
+
+    if (!p) {
+	ERTS_RBT_ASSERT(*root == x);
+	*root = y;
+#ifdef ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT
+	ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT(x, y);
+#endif
+    }
+    else if (x == ERTS_RBT_GET_LEFT(p))
+	ERTS_RBT_SET_LEFT(p, y);
+    else {
+	ERTS_RBT_ASSERT(x == ERTS_RBT_GET_RIGHT(p));
+	ERTS_RBT_SET_RIGHT(p, y);
+    }
+    ERTS_RBT_SET_LEFT(y, x);
+    ERTS_RBT_SET_PARENT(x, y);
+
+#ifdef ERTS_RBT_UPDATE_ATTACHED_DATA_ROTATE
+    ERTS_RBT_UPDATE_ATTACHED_DATA_ROTATE(!0, x, y);
+#endif
+
+}
+
+static ERTS_INLINE void
+ERTS_RBT_FUNC__(right_rotate__)(ERTS_RBT_T **root, ERTS_RBT_T *x)
+{
+    ERTS_RBT_T *y, *r, *p;
+
+    y = ERTS_RBT_GET_LEFT(x);
+    r = ERTS_RBT_GET_RIGHT(y);
+    ERTS_RBT_SET_LEFT(x, r);
+
+    if (r)
+	ERTS_RBT_SET_PARENT(r, x);
+
+    p = ERTS_RBT_GET_PARENT(x);
+    ERTS_RBT_SET_PARENT(y, p);
+
+    if (!p) {
+	ERTS_RBT_ASSERT(*root == x);
+	*root = y;
+#ifdef ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT
+	ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT(x, y);
+#endif
+    }
+    else if (x == ERTS_RBT_GET_RIGHT(p))
+	ERTS_RBT_SET_RIGHT(p, y);
+    else {
+	ERTS_RBT_ASSERT(x == ERTS_RBT_GET_LEFT(p));
+	ERTS_RBT_SET_LEFT(p, y);
+    }
+
+    ERTS_RBT_SET_RIGHT(y, x);
+    ERTS_RBT_SET_PARENT(x, y);
+
+#ifdef ERTS_RBT_UPDATE_ATTACHED_DATA_ROTATE
+    ERTS_RBT_UPDATE_ATTACHED_DATA_ROTATE(0, x, y);
+#endif
+
+}
+
+#endif /* ERTS_RBT_NEED_ROTATE__ */
+
+#ifdef ERTS_RBT_NEED_REPLACE__
+
+/*
+ * Replace node x with node y
+ */
+static ERTS_INLINE void
+ERTS_RBT_FUNC__(replace__)(ERTS_RBT_T **root, ERTS_RBT_T *x, ERTS_RBT_T *y)
+{
+    ERTS_RBT_T *p, *r, *l;
+    ERTS_RBT_FLAGS_T f;
+
+    p = ERTS_RBT_GET_PARENT(x);
+    if (!p) {
+	ERTS_RBT_ASSERT(*root == x);
+	*root = y;
+#ifdef ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT
+	ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT(x, y);
+#endif
+    }
+    else if (x == ERTS_RBT_GET_LEFT(p))
+	ERTS_RBT_SET_LEFT(p, y);
+    else {
+	ERTS_RBT_ASSERT(x == ERTS_RBT_GET_RIGHT(p));
+	ERTS_RBT_SET_RIGHT(p, y);
+    }
+    l = ERTS_RBT_GET_LEFT(x);
+    if (l) {
+	ERTS_RBT_ASSERT(ERTS_RBT_GET_PARENT(l) == x);
+	ERTS_RBT_SET_PARENT(l, y);
+    }
+    r = ERTS_RBT_GET_RIGHT(x);
+    if (r) {
+	ERTS_RBT_ASSERT(ERTS_RBT_GET_PARENT(r) == x);
+	ERTS_RBT_SET_PARENT(r, y);
+    }
+
+    f = ERTS_RBT_GET_FLAGS(x);
+    ERTS_RBT_SET_FLAGS(y, f);
+    ERTS_RBT_SET_PARENT(y, p);
+    ERTS_RBT_SET_RIGHT(y, r);
+    ERTS_RBT_SET_LEFT(y, l);
+}
+
+#endif /* ERTS_RBT_NEED_REPLACE__ */
+
+#ifdef ERTS_RBT_WANT_REPLACE
+
+static ERTS_RBT_API_INLINE__ void
+ERTS_RBT_FUNC__(replace)(ERTS_RBT_T **root, ERTS_RBT_T *x, ERTS_RBT_T *y)
+{
+    ERTS_RBT_ASSERT(ERTS_RBT_IS_EQ(ERTS_RBT_GET_KEY(x),
+				   ERTS_RBT_GET_KEY(y)));
+
+    ERTS_RBT_FUNC__(replace__)(root, x, y);
+}
+
+#endif /* ERTS_RBT_WANT_REPLACE */
+
+#ifdef ERTS_RBT_WANT_DELETE
+
+/*
+ * Delete a node.
+ */
+static ERTS_RBT_API_INLINE__ void
+ERTS_RBT_FUNC__(delete)(ERTS_RBT_T **root, ERTS_RBT_T *n)
+{
+    int spliced_is_black;
+    ERTS_RBT_T *p, *x, *y, *z = n;
+    ERTS_RBT_T null_x; /* null_x is used to get the fixup started when we
+			  splice out a node without children. */
+
+    ERTS_RBT_HDBG_CHECK_TREE__(*root);
+
+    ERTS_RBT_INIT_EMPTY_TNODE(&null_x);
+
+    /* Remove node from tree... */
+
+    /* Find node to splice out */
+    if (!ERTS_RBT_GET_LEFT(z) || !ERTS_RBT_GET_RIGHT(z))
+	y = z;
+    else {
+	/* Set y to z:s successor */
+	y = ERTS_RBT_GET_RIGHT(z);
+	while (1) {
+	    ERTS_RBT_T *t = ERTS_RBT_GET_LEFT(y);
+	    if (!t)
+		break;
+	    y = t;
+	}
+    }
+    /* splice out y */
+    x = ERTS_RBT_GET_LEFT(y);
+    if (!x)
+	x = ERTS_RBT_GET_RIGHT(y);
+    spliced_is_black = ERTS_RBT_IS_BLACK(y);
+    p = ERTS_RBT_GET_PARENT(y);
+    if (x)
+	ERTS_RBT_SET_PARENT(x, p);
+    else if (spliced_is_black) {
+	x = &null_x;
+	ERTS_RBT_SET_BLACK(x);
+	ERTS_RBT_SET_PARENT(x, p);
+	ERTS_RBT_SET_LEFT(y, x);
+    }
+
+    if (!p) {
+	ERTS_RBT_ASSERT(*root == y);
+	*root = x;
+#ifdef ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT
+	ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT(y, x);
+#endif
+    }
+    else {
+	if (y == ERTS_RBT_GET_LEFT(p))
+	    ERTS_RBT_SET_LEFT(p, x);
+	else {
+	    ERTS_RBT_ASSERT(y == ERTS_RBT_GET_RIGHT(p));
+	    ERTS_RBT_SET_RIGHT(p, x);
+	}
+#ifdef ERTS_RBT_UPDATE_ATTACHED_DATA_DMOD
+	if (p != z)
+	    ERTS_RBT_UPDATE_ATTACHED_DATA_DMOD(p, y == z ? NULL : z);
+#endif
+    }
+    if (y != z) {
+	/* We spliced out the successor of z; replace z by the successor */
+	ERTS_RBT_FUNC__(replace__)(root, z, y);
+#ifdef ERTS_RBT_UPDATE_ATTACHED_DATA_DMOD
+	ERTS_RBT_UPDATE_ATTACHED_DATA_DMOD(y, NULL);
+#endif
+    }
+
+    if (spliced_is_black) {
+	/* We removed a black node which makes the resulting tree
+	   violate the Red-Black Tree properties. Fixup tree... */
+
+	p = ERTS_RBT_GET_PARENT(x);
+	while (ERTS_RBT_IS_BLACK(x) && p) {
+	    ERTS_RBT_T *r, *l;
+
+	    /*
+	     * x has an "extra black" which we move up the tree
+	     * until we reach the root or until we can get rid of it.
+	     *
+	     * y is the sibbling of x, and p is their parent
+	     */
+
+	    if (x == ERTS_RBT_GET_LEFT(p)) {
+		y = ERTS_RBT_GET_RIGHT(p);
+
+		ERTS_RBT_ASSERT(y);
+
+		if (ERTS_RBT_IS_RED(y)) {
+		    ERTS_RBT_ASSERT(ERTS_RBT_GET_RIGHT(y));
+		    ERTS_RBT_ASSERT(ERTS_RBT_GET_LEFT(y));
+
+		    ERTS_RBT_SET_BLACK(y);
+
+		    ERTS_RBT_ASSERT(ERTS_RBT_IS_BLACK(p));
+
+		    ERTS_RBT_SET_RED(p);
+		    ERTS_RBT_FUNC__(left_rotate__)(root, p);
+		    p = ERTS_RBT_GET_PARENT(x);
+		    y = ERTS_RBT_GET_RIGHT(p);
+		}
+
+		ERTS_RBT_ASSERT(y);
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_BLACK(y));
+
+		l = ERTS_RBT_GET_LEFT(y);
+		r = ERTS_RBT_GET_RIGHT(y);
+		if ((!l || ERTS_RBT_IS_BLACK(l))
+		    && (!r || ERTS_RBT_IS_BLACK(r))) {
+		    ERTS_RBT_SET_RED(y);
+		    x = p;
+		    p = ERTS_RBT_GET_PARENT(x);
+		}
+		else {
+		    if (!r || ERTS_RBT_IS_BLACK(r)) {
+			ERTS_RBT_SET_BLACK(l);
+			ERTS_RBT_SET_RED(y);
+			ERTS_RBT_FUNC__(right_rotate__)(root, y);
+			p = ERTS_RBT_GET_PARENT(x);
+			y = ERTS_RBT_GET_RIGHT(p);
+		    }
+
+		    ERTS_RBT_ASSERT(y);
+
+		    if (p && ERTS_RBT_IS_RED(p)) {
+
+			ERTS_RBT_SET_BLACK(p);
+			ERTS_RBT_SET_RED(y);
+		    }
+
+		    ERTS_RBT_ASSERT(ERTS_RBT_GET_RIGHT(y));
+
+		    ERTS_RBT_SET_BLACK(ERTS_RBT_GET_RIGHT(y));
+		    ERTS_RBT_FUNC__(left_rotate__)(root, p);
+		    x = *root;
+		    break;
+		}
+	    }
+	    else {
+		ERTS_RBT_ASSERT(x == ERTS_RBT_GET_RIGHT(p));
+
+		y = ERTS_RBT_GET_LEFT(p);
+
+		ERTS_RBT_ASSERT(y);
+
+		if (ERTS_RBT_IS_RED(y)) {
+		    ERTS_RBT_ASSERT(ERTS_RBT_GET_RIGHT(y));
+		    ERTS_RBT_ASSERT(ERTS_RBT_GET_LEFT(y));
+
+		    ERTS_RBT_SET_BLACK(y);
+		    ERTS_RBT_ASSERT(ERTS_RBT_IS_BLACK(p));
+		    ERTS_RBT_SET_RED(p);
+		    ERTS_RBT_FUNC__(right_rotate__)(root, p);
+
+		    p = ERTS_RBT_GET_PARENT(x);
+		    y = ERTS_RBT_GET_LEFT(p);
+		}
+
+		ERTS_RBT_ASSERT(y);
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_BLACK(y));
+
+		l = ERTS_RBT_GET_LEFT(y);
+		r = ERTS_RBT_GET_RIGHT(y);
+
+		if ((!r || ERTS_RBT_IS_BLACK(r))
+		    && (!l || ERTS_RBT_IS_BLACK(l))) {
+		    ERTS_RBT_SET_RED(y);
+		    x = p;
+		    p = ERTS_RBT_GET_PARENT(x);
+		}
+		else {
+		    if (!l || ERTS_RBT_IS_BLACK(l)) {
+			ERTS_RBT_SET_BLACK(r);
+			ERTS_RBT_SET_RED(y);
+			ERTS_RBT_FUNC__(left_rotate__)(root, y);
+
+			p = ERTS_RBT_GET_PARENT(x);
+			y = ERTS_RBT_GET_LEFT(p);
+		    }
+
+		    ERTS_RBT_ASSERT(y);
+
+		    if (p && ERTS_RBT_IS_RED(p)) {
+			ERTS_RBT_SET_BLACK(p);
+			ERTS_RBT_SET_RED(y);
+		    }
+
+		    ERTS_RBT_ASSERT(ERTS_RBT_GET_LEFT(y));
+
+		    ERTS_RBT_SET_BLACK(ERTS_RBT_GET_LEFT(y));
+		    ERTS_RBT_FUNC__(right_rotate__)(root, p);
+		    x = *root;
+		    break;
+		}
+	    }
+	}
+
+	ERTS_RBT_SET_BLACK(x);
+
+	x = &null_x;
+	p = ERTS_RBT_GET_PARENT(x);
+
+	if (p) {
+	    if (ERTS_RBT_GET_LEFT(p) == x)
+		ERTS_RBT_SET_LEFT(p, NULL);
+	    else {
+		ERTS_RBT_ASSERT(ERTS_RBT_GET_RIGHT(p) == x);
+		ERTS_RBT_SET_RIGHT(p, NULL);
+	    }
+
+	    ERTS_RBT_ASSERT(!ERTS_RBT_GET_LEFT(x));
+	    ERTS_RBT_ASSERT(!ERTS_RBT_GET_RIGHT(x));
+	}
+	else if (*root == x) {
+	    *root = NULL;
+
+#ifdef ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT
+	    ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT(x, NULL);
+#endif
+
+	    ERTS_RBT_ASSERT(!ERTS_RBT_GET_LEFT(x));
+	    ERTS_RBT_ASSERT(!ERTS_RBT_GET_RIGHT(x));
+	}
+    }
+
+    ERTS_RBT_HDBG_CHECK_TREE__(*root);
+
+}
+
+#endif /* ERTS_RBT_WANT_DELETE */
+
+#ifdef ERTS_RBT_NEED_INSERT__
+
+static void
+ERTS_RBT_FUNC__(insert_fixup__)(ERTS_RBT_T **root, ERTS_RBT_T *n)
+{
+    ERTS_RBT_T *x, *y;
+
+    x = n;
+
+    /*
+     * Rearrange the tree so that it satisfies the Red-Black Tree properties
+     */
+
+    ERTS_RBT_ASSERT(x != *root && ERTS_RBT_IS_RED(ERTS_RBT_GET_PARENT(x)));
+    do {
+	ERTS_RBT_T *p, *pp;
+
+	/*
+	 * x and its parent are both red. Move the red pair up the tree
+	 * until we get to the root or until we can separate them.
+	 */
+
+	p = ERTS_RBT_GET_PARENT(x);
+	pp = ERTS_RBT_GET_PARENT(p);
+
+	ERTS_RBT_ASSERT(p && pp);
+	ERTS_RBT_ASSERT(ERTS_RBT_IS_RED(x));
+	ERTS_RBT_ASSERT(ERTS_RBT_IS_BLACK(pp));
+
+	if (p == ERTS_RBT_GET_LEFT(pp)) {
+	    y = ERTS_RBT_GET_RIGHT(pp);
+	    if (y && ERTS_RBT_IS_RED(y)) {
+		ERTS_RBT_SET_BLACK(y);
+		ERTS_RBT_SET_BLACK(p);
+		ERTS_RBT_SET_RED(pp);
+		x = pp;
+	    }
+	    else {
+
+		if (x == ERTS_RBT_GET_RIGHT(p)) {
+		    x = p;
+		    ERTS_RBT_FUNC__(left_rotate__)(root, x);
+		    p = ERTS_RBT_GET_PARENT(x);
+		    pp = ERTS_RBT_GET_PARENT(p);
+
+		    ERTS_RBT_ASSERT(p && pp);
+		}
+
+		ERTS_RBT_ASSERT(x == ERTS_RBT_GET_LEFT(ERTS_RBT_GET_LEFT(pp)));
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_RED(x));
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_RED(p));
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_BLACK(pp));
+		ERTS_RBT_ASSERT(!y || ERTS_RBT_IS_BLACK(y));
+
+
+		ERTS_RBT_SET_BLACK(p);
+		ERTS_RBT_SET_RED(pp);
+		ERTS_RBT_FUNC__(right_rotate__)(root, pp);
+
+
+		ERTS_RBT_ASSERT(ERTS_RBT_GET_LEFT(ERTS_RBT_GET_PARENT(x)) == x);
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_RED(x));
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_RED(
+				    ERTS_RBT_GET_RIGHT(
+					ERTS_RBT_GET_PARENT(x))));
+		ERTS_RBT_ASSERT(!ERTS_RBT_GET_PARENT(x)
+				|| ERTS_RBT_IS_BLACK(ERTS_RBT_GET_PARENT(x)));
+		break;
+	    }
+	}
+	else {
+	    ERTS_RBT_ASSERT(p == ERTS_RBT_GET_RIGHT(pp));
+
+	    y = ERTS_RBT_GET_LEFT(pp);
+	    if (y && ERTS_RBT_IS_RED(y)) {
+		ERTS_RBT_SET_BLACK(y);
+		ERTS_RBT_SET_BLACK(p);
+		ERTS_RBT_SET_RED(pp);
+		x = pp;
+	    }
+	    else {
+
+		if (x == ERTS_RBT_GET_LEFT(p)) {
+		    x = p;
+		    ERTS_RBT_FUNC__(right_rotate__)(root, x);
+		    p = ERTS_RBT_GET_PARENT(x);
+		    pp = ERTS_RBT_GET_PARENT(p);
+
+		    ERTS_RBT_ASSERT(p && pp);
+		}
+
+		ERTS_RBT_ASSERT(x == ERTS_RBT_GET_RIGHT(ERTS_RBT_GET_RIGHT(pp)));
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_RED(x));
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_RED(p));
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_BLACK(pp));
+		ERTS_RBT_ASSERT(!y || ERTS_RBT_IS_BLACK(y));
+
+
+		ERTS_RBT_SET_BLACK(p);
+		ERTS_RBT_SET_RED(pp);
+		ERTS_RBT_FUNC__(left_rotate__)(root, pp);
+
+
+		ERTS_RBT_ASSERT(ERTS_RBT_GET_RIGHT(ERTS_RBT_GET_PARENT(x)) == x);
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_RED(x));
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_RED(
+				    ERTS_RBT_GET_LEFT(
+					ERTS_RBT_GET_PARENT(x))));
+		ERTS_RBT_ASSERT(!ERTS_RBT_GET_PARENT(x)
+				|| ERTS_RBT_IS_BLACK(ERTS_RBT_GET_PARENT(x)));
+		break;
+	    }
+	}
+    } while (x != *root && ERTS_RBT_IS_RED(ERTS_RBT_GET_PARENT(x)));
+
+    ERTS_RBT_SET_BLACK(*root);
+
+}
+
+static ERTS_INLINE ERTS_RBT_T *
+ERTS_RBT_FUNC__(insert_aux__)(ERTS_RBT_T **root, ERTS_RBT_T *n, int lookup)
+{
+    ERTS_RBT_KEY_T kn = ERTS_RBT_GET_KEY(n);
+
+    ERTS_RBT_HDBG_CHECK_TREE__(*root);
+
+    ERTS_RBT_INIT_EMPTY_TNODE(n);	
+
+    if (!*root) {
+	ERTS_RBT_SET_BLACK(n);
+	*root = n;
+#ifdef ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT
+	ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT(NULL, n);
+#endif
+    }
+    else {
+	ERTS_RBT_T *p, *x = *root;
+
+	while (1) {
+	    ERTS_RBT_KEY_T kx;
+	    ERTS_RBT_T *c;
+
+	    kx = ERTS_RBT_GET_KEY(x);
+
+	    if (lookup && ERTS_RBT_IS_EQ(kn, kx)) {
+
+		ERTS_RBT_HDBG_CHECK_TREE__(*root);
+
+		return x;
+	    }
+
+	    if (ERTS_RBT_IS_LT(kn, kx)) {
+		c = ERTS_RBT_GET_LEFT(x);
+		if (!c) {
+		    ERTS_RBT_SET_PARENT(n, x);
+		    ERTS_RBT_SET_LEFT(x, n);
+		    p = x;
+		    break;
+		}
+	    }
+	    else {
+		c = ERTS_RBT_GET_RIGHT(x);
+		if (!c) {
+		    ERTS_RBT_SET_PARENT(n, x);
+		    ERTS_RBT_SET_RIGHT(x, n);
+		    p = x;
+		    break;
+		}
+	    }
+
+	    x = c;
+	}
+
+	ERTS_RBT_ASSERT(p);
+
+	ERTS_RBT_SET_RED(n);
+	if (ERTS_RBT_IS_RED(p))
+	    ERTS_RBT_FUNC__(insert_fixup__)(root, n);
+    }
+
+    ERTS_RBT_HDBG_CHECK_TREE__(*root);
+
+    return NULL;
+}
+
+#endif /* ERTS_RBT_NEED_INSERT__ */
+
+#ifdef ERTS_RBT_WANT_LOOKUP_INSERT
+
+static ERTS_RBT_API_INLINE__ ERTS_RBT_T *
+ERTS_RBT_FUNC__(lookup_insert)(ERTS_RBT_T **root, ERTS_RBT_T *n)
+{
+    return ERTS_RBT_FUNC__(insert_aux__)(root, n, !0);
+}
+
+#endif /* ERTS_RBT_WANT_LOOKUP_INSERT */
+
+#ifdef ERTS_RBT_WANT_INSERT
+
+static ERTS_RBT_API_INLINE__ void
+ERTS_RBT_FUNC__(insert)(ERTS_RBT_T **root, ERTS_RBT_T *n)
+{
+    (void) ERTS_RBT_FUNC__(insert_aux__)(root, n, 0);
+}
+
+#endif /* ERTS_RBT_WANT_INSERT */
+
+#ifdef ERTS_RBT_WANT_LOOKUP
+
+static ERTS_RBT_API_INLINE__ ERTS_RBT_T *
+ERTS_RBT_FUNC__(lookup)(ERTS_RBT_T *root, ERTS_RBT_KEY_T key)
+{
+    ERTS_RBT_T *x = root;
+
+    if (!x)
+	return NULL;
+
+    while (1) {
+	ERTS_RBT_KEY_T kx = ERTS_RBT_GET_KEY(x);
+	ERTS_RBT_T *c;
+
+	if (ERTS_RBT_IS_EQ(key, kx))
+	    return x;
+
+	if (ERTS_RBT_IS_LT(key, kx)) {
+	    c = ERTS_RBT_GET_LEFT(x);
+	    if (!c)
+		return NULL;
+	}
+	else {
+	    c = ERTS_RBT_GET_RIGHT(x);
+	    if (!c)
+		return NULL;
+	}
+
+	x = c;
+    }
+}
+
+#endif /* ERTS_RBT_WANT_LOOKUP */
+
+#ifdef ERTS_RBT_WANT_SMALLEST
+
+static ERTS_RBT_API_INLINE__ ERTS_RBT_T *
+ERTS_RBT_FUNC__(smallest)(ERTS_RBT_T *root)
+{
+    ERTS_RBT_T *x = root;
+
+    if (!x)
+	return NULL;
+
+    while (1) {
+	ERTS_RBT_T *c = ERTS_RBT_GET_LEFT(x);
+	if (!c)
+	    break;
+	x = c;
+    }
+
+    return x;
+}
+
+#endif /* ERTS_RBT_WANT_SMALLEST */
+
+#ifdef ERTS_RBT_WANT_LARGEST
+
+static ERTS_RBT_API_INLINE__ ERTS_RBT_T *
+ERTS_RBT_FUNC__(largest)(ERTS_RBT_T *root)
+{
+    ERTS_RBT_T *x = root;
+
+    if (!x)
+	return NULL;
+
+    while (1) {
+	ERTS_RBT_T *c = ERTS_RBT_GET_RIGHT(x);
+	if (!c)
+	    break;
+	x = c;
+    }
+
+    return x;
+}
+
+#endif /* ERTS_RBT_WANT_LARGEST */
+
+#ifdef ERTS_RBT_NEED_FOREACH_UNORDERED__
+
+static ERTS_INLINE int
+ERTS_RBT_FUNC__(foreach_unordered__)(ERTS_RBT_T **root,
+				     int destroying,
+				     void (*op)(ERTS_RBT_T *, void *),
+				     void *arg,
+				     int yielding,
+				     ERTS_RBT_YIELD_STATE_T__ *ystate,
+				     Sint ylimit)
+{
+    ERTS_RBT_T *c, *p, *x;
+
+    ERTS_RBT_ASSERT(!yielding || ystate);
+
+    if (yielding && ystate->x) {
+	x = ystate->x;
+	ERTS_RBT_ASSERT(ystate->up);
+	goto restart_up;
+    }
+    else {
+	x = *root;
+	if (!x)
+	    return 0;
+	if (destroying)
+	    *root = NULL;
+    }
+
+    while (1) {
+
+	while (1) {
+
+	    while (1) {
+		c = ERTS_RBT_GET_LEFT(x);
+		if (!c)
+		    break;
+		x = c;
+	    }
+
+	    c = ERTS_RBT_GET_RIGHT(x);
+	    if (!c)
+		break;
+	    x = c;
+	}
+
+	while (1) {
+#ifdef ERTS_RBT_DEBUG
+	    int cdir;
+#endif
+	    if (yielding && ylimit-- <= 0) {
+		ystate->x = x;
+		ystate->up = 1;
+		return 1;
+	    }
+
+	restart_up:
+
+	    p = ERTS_RBT_GET_PARENT(x);
+
+#ifdef ERTS_RBT_DEBUG
+	    ERTS_RBT_ASSERT(!destroying || !ERTS_RBT_GET_LEFT(x));
+	    ERTS_RBT_ASSERT(!destroying || !ERTS_RBT_GET_RIGHT(x));
+
+	    if (p) {
+		if (x == ERTS_RBT_GET_LEFT(p)) {
+		    cdir = -1;
+		    if (destroying)
+			ERTS_RBT_SET_LEFT(p, NULL);
+		}
+		else {
+		    ERTS_RBT_ASSERT(x == ERTS_RBT_GET_RIGHT(p));
+		    cdir = 1;
+		    if (destroying)
+			ERTS_RBT_SET_RIGHT(p, NULL);
+		}
+	    }
+#endif
+
+	    (*op)(x, arg);
+
+	    if (!p) {
+		if (yielding) {
+		    ystate->x = NULL;
+		    ystate->up = 0;
+		}
+		return 0; /* Done */
+	    }
+
+	    c = ERTS_RBT_GET_RIGHT(p);
+	    if (c && c != x) {
+		ERTS_RBT_ASSERT(cdir < 0);
+
+		/* Go down tree of x's sibling... */
+		x = c;
+		break;
+	    }
+
+	    x = p;
+	}
+    }
+}
+
+#endif /* ERTS_RBT_NEED_FOREACH_UNORDERED__ */
+
+#ifdef ERTS_RBT_NEED_FOREACH_ORDERED__
+
+static ERTS_INLINE int
+ERTS_RBT_FUNC__(foreach_ordered__)(ERTS_RBT_T **root,
+				   int from_small,
+				   int destroying,
+				   void (*op)(ERTS_RBT_T *, void *),
+				   void (*destroy)(ERTS_RBT_T *, void *),
+				   void *arg,
+				   int yielding,
+				   ERTS_RBT_YIELD_STATE_T__ *ystate,
+				   Sint ylimit)
+{
+    ERTS_RBT_T *c, *p, *x;
+
+    ERTS_RBT_ASSERT(!yielding || ystate);
+    ERTS_RBT_ASSERT(!destroying || destroy);
+
+    if (yielding && ystate->x) {
+	x = ystate->x;
+	if (ystate->up)
+	    goto restart_up;
+	else
+	    goto restart_down;
+    }
+    else {
+	x = *root;
+	if (!x)
+	    return 0;
+	if (destroying)
+	    *root = NULL;
+    }
+
+    while (1) {
+
+	while (1) {
+
+	    while (1) {
+		c = from_small ? ERTS_RBT_GET_LEFT(x) : ERTS_RBT_GET_RIGHT(x);
+		if (!c)
+		    break;
+		x = c;
+	    }
+
+	    (*op)(x, arg);
+
+	    if (yielding && --ylimit <= 0) {
+		ystate->x = x;
+		ystate->up = 0;
+		return 1;
+	    }
+
+	restart_down:
+
+	    c = from_small ? ERTS_RBT_GET_RIGHT(x) : ERTS_RBT_GET_LEFT(x);
+	    if (!c)
+		break;
+	    x = c;
+	}
+
+	while (1) {
+	    p = ERTS_RBT_GET_PARENT(x);
+
+	    if (p) {
+
+		c = from_small ? ERTS_RBT_GET_RIGHT(p) : ERTS_RBT_GET_LEFT(p);
+		if (!c || c != x) {
+		    ERTS_RBT_ASSERT((from_small
+				     ? ERTS_RBT_GET_LEFT(p)
+				     : ERTS_RBT_GET_RIGHT(p)) == x);
+
+		    (*op)(p, arg);
+
+		    if (yielding && --ylimit <= 0) {
+			ystate->x = x;
+			ystate->up = 1;
+			return 1;
+		    restart_up:
+			p = ERTS_RBT_GET_PARENT(x);
+		    }
+		}
+
+		if (c && c != x) {
+		    ERTS_RBT_ASSERT((from_small
+				     ? ERTS_RBT_GET_LEFT(p)
+				     : ERTS_RBT_GET_RIGHT(p)) == x);
+
+		    /* Go down tree of x's sibling... */
+		    x = c;
+		    break;
+		}
+	    }
+
+	    if (destroying) {
+
+#ifdef ERTS_RBT_DEBUG
+		ERTS_RBT_ASSERT(!ERTS_RBT_GET_LEFT(x)
+				&& !ERTS_RBT_GET_RIGHT(x));
+
+		if (p) {
+		    if (x == ERTS_RBT_GET_LEFT(p))
+			ERTS_RBT_SET_LEFT(p, NULL);
+		    else {
+			ERTS_RBT_ASSERT(x == ERTS_RBT_GET_RIGHT(p));
+			ERTS_RBT_SET_RIGHT(p, NULL);
+		    }
+		}
+#endif
+
+		(*destroy)(x, arg);
+	    }
+
+	    if (!p) {
+		if (yielding) {
+		    ystate->x = NULL;
+		    ystate->up = 0;
+		}
+		return 1; /* Done */
+	    }
+	    x = p;
+	}
+    }
+}
+
+#endif /* ERTS_RBT_NEED_FOREACH_ORDERED__ */
+
+#ifdef ERTS_RBT_WANT_FOREACH
+
+static ERTS_RBT_API_INLINE__ void
+ERTS_RBT_FUNC__(foreach)(ERTS_RBT_T *root,
+			 void (*op)(ERTS_RBT_T *, void *),
+			 void *arg)
+{
+    (void) ERTS_RBT_FUNC__(foreach_unordered__)(&root, 0, op, arg,
+						0, NULL, 0);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH */
+
+#ifdef ERTS_RBT_WANT_FOREACH_SMALL
+
+static ERTS_RBT_API_INLINE__ void
+ERTS_RBT_FUNC__(foreach_small)(ERTS_RBT_T *root,
+			       void (*op)(ERTS_RBT_T *, void *),
+			       void *arg)
+{
+    (void) ERTS_RBT_FUNC__(foreach_ordered__)(&root, 1, 0,
+					      op, NULL, arg,
+					      0, NULL, 0);
+}
+
+#endif /*  ERTS_RBT_WANT_FOREACH_SMALL */
+
+#ifdef ERTS_RBT_WANT_FOREACH_LARGE
+
+static ERTS_RBT_API_INLINE__ void
+ERTS_RBT_FUNC__(foreach_large)(ERTS_RBT_T *root,
+			       void (*op)(ERTS_RBT_T *, void *),
+			       void *arg)
+{
+    (void) ERTS_RBT_FUNC__(foreach_ordered__)(&root, 0, 0,
+					      op, NULL, arg,
+					      0, NULL, 0);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH_LARGE */
+
+#ifdef ERTS_RBT_WANT_FOREACH_YIELDING
+
+static ERTS_RBT_API_INLINE__ void
+ERTS_RBT_FUNC__(foreach_yielding)(ERTS_RBT_T *root,
+				  void (*op)(ERTS_RBT_T *, void *),
+				  void *arg,
+				  ERTS_RBT_YIELD_STATE_T__ *ystate,
+				  Sint ylimit)
+{
+    (void) ERTS_RBT_FUNC__(foreach_unordered__)(*root, 0, op, arg,
+						1, ystate, ylimit);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH_YIELDING */
+
+#ifdef ERTS_RBT_WANT_FOREACH_SMALL_YIELDING
+
+static ERTS_RBT_API_INLINE__ int
+ERTS_RBT_FUNC__(foreach_small_yielding)(ERTS_RBT_T *root,
+					void (*op)(ERTS_RBT_T *, void *),
+					void *arg,
+					ERTS_RBT_YIELD_STATE_T__ *ystate,
+					Sint ylimit)
+{
+    return ERTS_RBT_FUNC__(foreach_ordered__)(&root, 1, 0,
+					      op, NULL, arg,
+					      1, ystate, ylimit);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH_SMALL_YIELDING */
+
+#ifdef ERTS_RBT_WANT_FOREACH_LARGE_YIELDING
+
+static ERTS_RBT_API_INLINE__ int
+ERTS_RBT_FUNC__(foreach_large_yielding)(ERTS_RBT_T *root,
+					void (*op)(ERTS_RBT_T *, void *),
+					void *arg,
+					ERTS_RBT_YIELD_STATE_T__ *ystate,
+					Sint ylimit)
+{
+    return ERTS_RBT_FUNC__(foreach_ordered__)(&root, 0, 0,
+					      op, NULL, arg,
+					      1, ystate, ylimit);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH_LARGE_YIELDING */
+
+#ifdef ERTS_RBT_WANT_FOREACH_DESTROY
+
+static ERTS_RBT_API_INLINE__ void
+ERTS_RBT_FUNC__(foreach_destroy)(ERTS_RBT_T **root,
+				 void (*op)(ERTS_RBT_T *, void *),
+				 void *arg)
+{
+    (void) ERTS_RBT_FUNC__(foreach_unordered__)(root, 1, op, arg,
+						0, NULL, 0);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH_DESTROY */
+
+#ifdef ERTS_RBT_WANT_FOREACH_SMALL_DESTROY
+
+static ERTS_RBT_API_INLINE__ void
+ERTS_RBT_FUNC__(foreach_small_destroy)(ERTS_RBT_T **root,
+				       void (*op)(ERTS_RBT_T *, void *),
+				       void (*destr)(ERTS_RBT_T *, void *),
+				       void *arg)
+{
+    (void) ERTS_RBT_FUNC__(foreach_ordered__)(root, 1, 1,
+					      op, destr, arg,
+					      0, NULL, 0);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH_SMALL_DESTROY */
+
+#ifdef ERTS_RBT_WANT_FOREACH_LARGE_DESTROY
+
+static ERTS_RBT_API_INLINE__ void
+ERTS_RBT_FUNC__(foreach_large_destroy)(ERTS_RBT_T **root,
+				       void (*op)(ERTS_RBT_T *, void *),
+				       void (*destr)(ERTS_RBT_T *, void *),
+				       void *arg)
+{
+    (void) ERTS_RBT_FUNC__(foreach_ordered__)(root, 0, 1,
+					      op, destr, arg,
+					      0, NULL, 0);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH_LARGE_DESTROY */
+
+#ifdef ERTS_RBT_WANT_FOREACH_DESTROY_YIELDING
+
+static ERTS_RBT_API_INLINE__ int
+ERTS_RBT_FUNC__(foreach_destroy_yielding)(ERTS_RBT_T **root,
+					  void (*op)(ERTS_RBT_T *, void *),
+					  void *arg,
+					  ERTS_RBT_YIELD_STATE_T__ *ystate,
+					  Sint ylimit)
+{
+    return ERTS_RBT_FUNC__(foreach_unordered__)(root, 1, op, arg,
+						1, ystate, ylimit);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH_DESTROY_YIELDING */
+
+#ifdef ERTS_RBT_WANT_FOREACH_SMALL_DESTROY_YIELDING
+
+static ERTS_RBT_API_INLINE__ int
+ERTS_RBT_FUNC__(foreach_small_destroy_yielding)(ERTS_RBT_T **root,
+						void (*op)(ERTS_RBT_T *, void *),
+						void (*destr)(ERTS_RBT_T *, void *),
+						void *arg,
+						ERTS_RBT_YIELD_STATE_T__ *ystate,
+						Sint ylimit)
+{
+    return ERTS_RBT_FUNC__(foreach_ordered__)(root, 1, 1,
+					      op, destr, arg,
+					      1, ystate, ylimit);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH_SMALL_DESTROY_YIELDING */
+
+#ifdef ERTS_RBT_WANT_FOREACH_LARGE_DESTROY_YIELDING
+
+static ERTS_RBT_API_INLINE__ int
+ERTS_RBT_FUNC__(foreach_large_destroy_yielding)(ERTS_RBT_T **root,
+						void (*op)(ERTS_RBT_T *, void *),
+						void (*destr)(ERTS_RBT_T *, void *),
+						void *arg,
+						ERTS_RBT_YIELD_STATE_T__ *ystate,
+						Sint ylimit)
+{
+    return ERTS_RBT_FUNC__(foreach_ordered__)(root, 0, 1,
+					      op, destr, arg,
+					      1, ystate, ylimit);
+}
+
+#endif /* ERTS_RBT_WANT_FOREACH_LARGE_DESTROY_YIELDING */
+
+#ifdef ERTS_RBT_WANT_DEBUG_PRINT
+
+static void
+ERTS_RBT_FUNC__(debug_print)(FILE *filep, ERTS_RBT_T *x, int indent,
+			     void (*print_node)(ERTS_RBT_T *))
+{
+    if (x) {
+	ERTS_RBT_FUNC__(debug_print)(filep, ERTS_RBT_GET_RIGHT(x),
+				     indent+2, print_node);
+	erts_fprintf(filep,
+		     "%*s[%s:%p:",
+		     indent, "",
+		     ERTS_RBT_IS_BLACK(x) ? "Black" : "Red",
+		     x);
+	(*print_node)(x);
+	erts_fprintf(filep, "]\n");
+	ERTS_RBT_FUNC__(debug_print)(filep, ERTS_RBT_GET_LEFT(x),
+				     indent+2, print_node);
+    }
+}
+
+#endif /* ERTS_RBT_WANT_DEBUG_PRINT */
+
+#ifdef ERTS_RBT_NEED_HDBG_CHECK_TREE__
+
+static void
+ERTS_RBT_FUNC__(hdbg_check_tree)(ERTS_RBT_T *root)
+{
+    int black_depth = -1, no_black = 0;
+    ERTS_RBT_T *c, *p, *x = root;
+    ERTS_RBT_KEY_T kx;
+    ERTS_RBT_KEY_T kc;
+
+    if (!x)
+	return;
+
+    ERTS_RBT_ASSERT(!ERTS_RBT_GET_PARENT(x));
+
+    while (1) {
+
+	while (1) {
+
+	    while (1) {
+
+		if (ERTS_RBT_IS_BLACK(x))
+		    no_black++;
+		else {
+		    c = ERTS_RBT_GET_RIGHT(x);
+		    ERTS_RBT_ASSERT(!c || ERTS_RBT_IS_BLACK(c));
+		    c = ERTS_RBT_GET_LEFT(x);
+		    ERTS_RBT_ASSERT(!c || ERTS_RBT_IS_BLACK(c));
+		}
+
+		c = ERTS_RBT_GET_LEFT(x);
+		if (!c)
+		    break;
+
+		ERTS_RBT_ASSERT(x == ERTS_RBT_GET_PARENT(c));
+
+		kx = ERTS_RBT_GET_KEY(x);
+		kc = ERTS_RBT_GET_KEY(c);
+
+		ERTS_RBT_ASSERT(ERTS_RBT_IS_LT(kc, kx)
+				|| ERTS_RBT_IS_EQ(kc, kx));
+
+		x = c;
+	    }
+
+	    c = ERTS_RBT_GET_RIGHT(x);
+	    if (!c) {
+		if (black_depth < 0)
+		    black_depth = no_black;
+		ERTS_RBT_ASSERT(black_depth == no_black);
+		break;
+	    }
+
+	    ERTS_RBT_ASSERT(x == ERTS_RBT_GET_PARENT(c));
+
+	    kx = ERTS_RBT_GET_KEY(x);
+	    kc = ERTS_RBT_GET_KEY(c);
+
+	    ERTS_RBT_ASSERT(ERTS_RBT_IS_LT(kx, kc)
+			    || ERTS_RBT_IS_EQ(kx, kc));
+	    x = c;
+	}
+
+	while (1) {
+	    p = ERTS_RBT_GET_PARENT(x);
+
+	    if (ERTS_RBT_IS_BLACK(x))
+		no_black--;
+
+	    if (p) {
+
+		ERTS_RBT_ASSERT(x == ERTS_RBT_GET_LEFT(p)
+				|| x == ERTS_RBT_GET_RIGHT(p));
+
+		c = ERTS_RBT_GET_RIGHT(p);
+		if (c && c != x) {
+		    ERTS_RBT_ASSERT(ERTS_RBT_GET_LEFT(p) == x);
+
+		    kx = ERTS_RBT_GET_KEY(x);
+		    kc = ERTS_RBT_GET_KEY(c);
+
+		    ERTS_RBT_ASSERT(ERTS_RBT_IS_LT(kx, kc)
+				    || ERTS_RBT_IS_EQ(kx, kc));
+		    /* Go down tree of x's sibling... */
+		    x = c;
+		    break;
+		}
+	    }
+
+	    if (!p) {
+		ERTS_RBT_ASSERT(root == x);
+		ERTS_RBT_ASSERT(no_black == 0);
+		return; /* Done */
+	    }
+
+	    x = p;
+	}
+    }
+}
+
+#undef ERTS_RBT_PRINT_TREE__
+
+#endif /* ERTS_RBT_NEED_HDBG_CHECK_TREE__ */
+
+#undef ERTS_RBT_ASSERT
+#undef ERTS_RBT_DEBUG
+#undef ERTS_RBT_API_INLINE__
+#undef ERTS_RBT_YIELD_STATE_T__
+#undef ERTS_RBT_NEED_REPLACE__
+#undef ERTS_RBT_NEED_INSERT__
+#undef ERTS_RBT_NEED_ROTATE__
+#undef ERTS_RBT_NEED_FOREACH_UNORDERED__
+#undef ERTS_RBT_NEED_FOREACH_ORDERED__
+#undef ERTS_RBT_NEED_HDBG_CHECK_TREE__
+#undef ERTS_RBT_HDBG_CHECK_TREE__
+
+#ifdef ERTS_RBT_UNDEF
+#  undef ERTS_RBT_PREFIX
+#  undef ERTS_RBT_T
+#  undef ERTS_RBT_KEY_T
+#  undef ERTS_RBT_FLAGS_T
+#  undef ERTS_RBT_INIT_EMPTY_TNODE
+#  undef ERTS_RBT_IS_RED
+#  undef ERTS_RBT_SET_RED
+#  undef ERTS_RBT_IS_BLACK
+#  undef ERTS_RBT_SET_BLACK
+#  undef ERTS_RBT_GET_FLAGS
+#  undef ERTS_RBT_SET_FLAGS
+#  undef ERTS_RBT_GET_PARENT
+#  undef ERTS_RBT_SET_PARENT
+#  undef ERTS_RBT_GET_RIGHT
+#  undef ERTS_RBT_SET_RIGHT
+#  undef ERTS_RBT_GET_LEFT
+#  undef ERTS_RBT_SET_LEFT
+#  undef ERTS_RBT_GET_KEY
+#  undef ERTS_RBT_IS_LT
+#  undef ERTS_RBT_IS_EQ
+#  undef ERTS_RBT_UNDEF
+#  undef ERTS_RBT_NO_API_INLINE
+#  undef ERTS_RBT_UPDATE_ATTACHED_DATA_ROTATE
+#  undef ERTS_RBT_UPDATE_ATTACHED_DATA_DMOD
+#  undef ERTS_RBT_UPDATE_ATTACHED_DATA_CHGROOT
+#  undef ERTS_RBT_WANT_DELETE
+#  undef ERTS_RBT_WANT_INSERT
+#  undef ERTS_RBT_WANT_LOOKUP_INSERT
+#  undef ERTS_RBT_WANT_REPLACE
+#  undef ERTS_RBT_WANT_LOOKUP
+#  undef ERTS_RBT_WANT_SMALLEST
+#  undef ERTS_RBT_WANT_LARGEST
+#  undef ERTS_RBT_WANT_FOREACH
+#  undef ERTS_RBT_WANT_FOREACH_DESTROY
+#  undef ERTS_RBT_WANT_FOREACH_YIELDING
+#  undef ERTS_RBT_WANT_FOREACH_DESTROY_YIELDING
+#  undef ERTS_RBT_WANT_FOREACH_SMALL
+#  undef ERTS_RBT_WANT_FOREACH_LARGE
+#  undef ERTS_RBT_WANT_FOREACH_SMALL_DESTROY
+#  undef ERTS_RBT_WANT_FOREACH_LARGE_DESTROY
+#  undef ERTS_RBT_WANT_FOREACH_SMALL_YIELDING
+#  undef ERTS_RBT_WANT_FOREACH_LARGE_YIELDING
+#  undef ERTS_RBT_WANT_FOREACH_SMALL_DESTROY_YIELDING
+#  undef ERTS_RBT_WANT_FOREACH_LARGE_DESTROY_YIELDING
+#  undef ERTS_RBT_WANT_DEBUG_PRINT
+#endif
diff --git a/erts/emulator/beam/erl_thr_progress.c b/erts/emulator/beam/erl_thr_progress.c
index 4c9b00d2ee..78e0964e8b 100644
--- a/erts/emulator/beam/erl_thr_progress.c
+++ b/erts/emulator/beam/erl_thr_progress.c
@@ -1360,6 +1360,7 @@ erts_thr_progress_fatal_error_wait(SWord timeout) {
     erts_aint32_t bc;
     SWord time_left = timeout;
     ErtsMonotonicTime timeout_time;
+    ErtsSchedulerData *esdp = erts_get_scheduler_data();
 
     /*
      * Counting poll intervals may give us a too long timeout
@@ -1367,7 +1368,7 @@ erts_thr_progress_fatal_error_wait(SWord timeout) {
      * this. In case we havn't got time correction this may
      * however fail too...
      */
-    timeout_time = erts_get_monotonic_time();
+    timeout_time = erts_get_monotonic_time(esdp);
     timeout_time += ERTS_MSEC_TO_MONOTONIC((ErtsMonotonicTime) timeout);
 
     while (1) {
@@ -1378,7 +1379,7 @@ erts_thr_progress_fatal_error_wait(SWord timeout) {
 	    break; /* Succefully blocked all managed threads */
 	if (time_left <= 0)
 	    break; /* Timeout */
-	if (timeout_time <= erts_get_monotonic_time())
+	if (timeout_time <= erts_get_monotonic_time(esdp))
 	    break; /* Timeout */
     }
 }
diff --git a/erts/emulator/beam/erl_time.h b/erts/emulator/beam/erl_time.h
index 5af3c21d40..4560cd23af 100644
--- a/erts/emulator/beam/erl_time.h
+++ b/erts/emulator/beam/erl_time.h
@@ -20,6 +20,13 @@
 #ifndef ERL_TIME_H__
 #define ERL_TIME_H__
 
+/* timer wheel size NEED to be a power of 2 */
+#ifdef SMALL_MEMORY
+#define ERTS_TIW_SIZE (1 << 13)
+#else
+#define ERTS_TIW_SIZE (1 << 16)
+#endif
+
 #if defined(DEBUG) || 0
 #define ERTS_TIME_ASSERT(B) ERTS_ASSERT(B)
 #else
@@ -33,52 +40,10 @@ typedef enum {
 } ErtsTimeWarpMode;
 
 typedef struct ErtsTimerWheel_ ErtsTimerWheel;
-typedef erts_atomic64_t * ErtsNextTimeoutRef;
-extern ErtsTimerWheel *erts_default_timer_wheel;
+typedef ErtsMonotonicTime * ErtsNextTimeoutRef;
 
 extern SysTimeval erts_first_emu_time;
 
-/*
-** Timer entry:
-*/
-typedef struct erl_timer {
-    struct erl_timer* next;	/* next entry tiw slot or chain */
-    struct erl_timer* prev;	/* prev entry tiw slot or chain */
-    erts_smp_atomic_t wheel;
-    ErtsMonotonicTime timeout_pos; /* Timeout in absolute clock ticks */
-    /* called when timeout */
-    void (*timeout)(void*);
-    /* called when cancel (may be NULL) */
-    void (*cancel)(void*);
-    void* arg;        /* argument to timeout/cancel procs */
-    int slot;			/* slot in timer wheel */
-} ErlTimer;
-
-typedef void (*ErlTimeoutProc)(void*);
-typedef void (*ErlCancelProc)(void*);
-
-#ifdef ERTS_SMP
-/*
- * Process and port timer
- */
-typedef union ErtsSmpPTimer_ ErtsSmpPTimer;
-union ErtsSmpPTimer_ {
-    struct {
-	ErlTimer tm;
-	Eterm id;
-	void (*timeout_func)(void*);
-	ErtsSmpPTimer **timer_ref;
-	Uint32 flags;
-    } timer;
-    ErtsSmpPTimer *next;
-};
-
-void erts_create_smp_ptimer(ErtsSmpPTimer **timer_ref,
-			    Eterm id,
-			    ErlTimeoutProc timeout_func,
-			    Uint timeout);
-void erts_cancel_smp_ptimer(ErtsSmpPTimer *ptimer);
-#endif
 
 void erts_monitor_time_offset(Eterm id, Eterm ref);
 int erts_demonitor_time_offset(Eterm ref);
@@ -86,14 +51,8 @@ int erts_demonitor_time_offset(Eterm ref);
 int erts_init_time_sup(int, ErtsTimeWarpMode);
 void erts_late_init_time_sup(void);
 
-/* timer-wheel api */
-
-ErtsTimerWheel *erts_create_timer_wheel(int);
 ErtsNextTimeoutRef erts_get_next_timeout_reference(ErtsTimerWheel *);
 void erts_init_time(int time_correction, ErtsTimeWarpMode time_warp_mode);
-void erts_set_timer(ErlTimer*, ErlTimeoutProc, ErlCancelProc, void*, Uint);
-void erts_cancel_timer(ErlTimer*);
-Uint erts_time_left(ErlTimer *);
 void erts_bump_timers(ErtsTimerWheel *, ErtsMonotonicTime);
 Uint erts_timer_wheel_memory_size(void);
 
@@ -101,27 +60,6 @@ Uint erts_timer_wheel_memory_size(void);
 void erts_p_slpq(void);
 #endif
 
-ErtsMonotonicTime erts_check_next_timeout_time(ErtsTimerWheel *,
-					       ErtsMonotonicTime);
-
-ERTS_GLB_INLINE void erts_init_timer(ErlTimer *p);
-ERTS_GLB_INLINE ErtsMonotonicTime erts_next_timeout_time(ErtsNextTimeoutRef);
-
-#if ERTS_GLB_INLINE_INCL_FUNC_DEF
-
-ERTS_GLB_INLINE void erts_init_timer(ErlTimer *p)
-{
-    erts_smp_atomic_init_nob(&p->wheel, (erts_aint_t) NULL);
-}
-
-ERTS_GLB_INLINE ErtsMonotonicTime erts_next_timeout_time(ErtsNextTimeoutRef nxt_tmo_ref)
-{
-    return (ErtsMonotonicTime) erts_atomic64_read_acqb((erts_atomic64_t *) nxt_tmo_ref);
-}
-
-#endif /* #if ERTS_GLB_INLINE_INCL_FUNC_DEF */
-
-
 /* time_sup */
 
 #if (defined(HAVE_GETHRVTIME) || defined(HAVE_CLOCK_GETTIME_CPU_TIME))
@@ -154,6 +92,7 @@ ErtsTimeOffsetState erts_time_offset_state(void);
 ErtsTimeOffsetState erts_finalize_time_offset(void); 
 struct process;
 Eterm erts_get_monotonic_start_time(struct process *c_p);
+Eterm erts_get_monotonic_end_time(struct process *c_p);
 Eterm erts_monotonic_time_source(struct process*c_p);
 Eterm erts_system_time_source(struct process*c_p);
 
@@ -163,8 +102,20 @@ Eterm erts_system_time_source(struct process*c_p);
 #define ERTS_CLKTCK_RESOLUTION (erts_time_sup__.r.o.clktck_resolution)
 #endif
 
+#define ERTS_TIMER_WHEEL_MSEC (ERTS_TIW_SIZE/(ERTS_CLKTCK_RESOLUTION/1000))
+
 struct erts_time_sup_read_only__ {
     ErtsMonotonicTime monotonic_time_unit;
+#if !ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT
+    ErtsMonotonicTime start;
+    struct {
+	ErtsMonotonicTime native;
+	ErtsMonotonicTime nsec;
+	ErtsMonotonicTime usec;
+	ErtsMonotonicTime msec;
+	ErtsMonotonicTime sec;
+    } start_offset;
+#endif
 #ifndef SYS_CLOCK_RESOLUTION
     ErtsMonotonicTime clktck_resolution;
 #endif
@@ -220,6 +171,16 @@ erts_time_unit_conversion(Uint64 value,
 
 #endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */
 
+/*
+ * Range of monotonic time internally
+ */
+
+#define ERTS_MONOTONIC_BEGIN						\
+    ERTS_MONOTONIC_TIME_UNIT
+#define ERTS_MONOTONIC_END						\
+    ((ERTS_MONOTONIC_TIME_MAX / ERTS_MONOTONIC_TIME_UNIT)		\
+     * ERTS_MONOTONIC_TIME_UNIT)
+
 #if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT
 
 /*
@@ -231,9 +192,6 @@ erts_time_unit_conversion(Uint64 value,
 #  error Compile time time unit needs to be at least 1000000
 #endif
 
-#define ERTS_MONOTONIC_TIME_UNIT \
-    ((ErtsMonotonicTime) ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT)
-
 #if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT == 1000*1000*1000
 /* Nano-second time unit */
 
@@ -264,6 +222,66 @@ erts_time_unit_conversion(Uint64 value,
 #error Missing implementation for monotonic time unit
 #endif
 
+#define ERTS_MONOTONIC_TIME_UNIT \
+    ((ErtsMonotonicTime) ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT)
+
+/*
+ * NOTE! ERTS_MONOTONIC_TIME_START_EXTERNAL *need* to be a multiple
+ *       of ERTS_MONOTONIC_TIME_UNIT.
+ */
+
+#ifdef ARCH_32
+/*
+ * Want to use a big-num of arity 2 as long as possible (584 years
+ * in the nano-second time unit case).
+ */
+#define ERTS_MONOTONIC_TIME_START_EXTERNAL		\
+    (((((((ErtsMonotonicTime) 1) << 32)-1)	\
+       / ERTS_MONOTONIC_TIME_UNIT)		\
+      * ERTS_MONOTONIC_TIME_UNIT)		\
+     + ERTS_MONOTONIC_TIME_UNIT)
+
+#else /* ARCH_64 */
+
+#if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT <= 10*1000*1000
+
+/*
+ * Using micro second time unit or lower. Start at zero since
+ * time will remain an immediate for a very long time anyway
+ * (1827 years in the 10 micro second case)...
+ */
+#define ERTS_MONOTONIC_TIME_START_EXTERNAL ((ErtsMonotonicTime) 0)
+
+#else /* ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT > 10*1000*1000 */
+
+/*
+ * Want to use an immediate as long as possible (36 years in the
+ * nano-second time unit case).
+*/
+#define ERTS_MONOTONIC_TIME_START_EXTERNAL 		\
+    ((((ErtsMonotonicTime) MIN_SMALL)		\
+      / ERTS_MONOTONIC_TIME_UNIT)		\
+     * ERTS_MONOTONIC_TIME_UNIT)
+
+#endif /* ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT > 1000*1000 */
+
+#endif /* ARCH_64 */
+
+/*
+ * Offsets from internal monotonic time to external monotonic time
+ */
+
+#define ERTS_MONOTONIC_OFFSET_NATIVE \
+    (ERTS_MONOTONIC_TIME_START_EXTERNAL - ERTS_MONOTONIC_BEGIN)
+#define ERTS_MONOTONIC_OFFSET_NSEC					\
+    ERTS_MONOTONIC_TO_NSEC__(ERTS_MONOTONIC_OFFSET_NATIVE)
+#define ERTS_MONOTONIC_OFFSET_USEC					\
+    ERTS_MONOTONIC_TO_USEC__(ERTS_MONOTONIC_OFFSET_NATIVE)
+#define ERTS_MONOTONIC_OFFSET_MSEC					\
+    ERTS_MONOTONIC_TO_MSEC__(ERTS_MONOTONIC_OFFSET_NATIVE)
+#define ERTS_MONOTONIC_OFFSET_SEC					\
+    ERTS_MONOTONIC_TO_SEC__(ERTS_MONOTONIC_OFFSET_NATIVE)
+
 #define ERTS_MONOTONIC_TO_CLKTCKS__(MON) \
     ((MON) / (ERTS_MONOTONIC_TIME_UNIT/ERTS_CLKTCK_RESOLUTION))
 #define ERTS_CLKTCKS_TO_MONOTONIC__(TCKS) \
@@ -271,8 +289,23 @@ erts_time_unit_conversion(Uint64 value,
 
 #else /* !ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT */
 
+/*
+ * Initialized in erts_init_sys_time_sup()
+ */
 #define ERTS_MONOTONIC_TIME_UNIT (erts_time_sup__.r.o.monotonic_time_unit)
 
+/*
+ * Offsets from internal monotonic time to external monotonic time
+ *
+ * Initialized in erts_init_time_sup()...
+ */
+#define ERTS_MONOTONIC_TIME_START_EXTERNAL (erts_time_sup__.r.o.start)
+#define ERTS_MONOTONIC_OFFSET_NATIVE (erts_time_sup__.r.o.start_offset.native)
+#define ERTS_MONOTONIC_OFFSET_NSEC (erts_time_sup__.r.o.start_offset.nsec)
+#define ERTS_MONOTONIC_OFFSET_USEC (erts_time_sup__.r.o.start_offset.usec)
+#define ERTS_MONOTONIC_OFFSET_MSEC (erts_time_sup__.r.o.start_offset.msec)
+#define ERTS_MONOTONIC_OFFSET_SEC (erts_time_sup__.r.o.start_offset.sec)
+
 #define ERTS_CONV_FROM_MON_UNIT___(M, TO)				\
     ((ErtsMonotonicTime)						\
      erts_time_unit_conversion((Uint64) (M),				\
@@ -310,6 +343,10 @@ erts_time_unit_conversion(Uint64 value,
 
 #endif /* !ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT */
 
+#define ERTS_MONOTONIC_TIME_END_EXTERNAL				\
+    (ERTS_MONOTONIC_TIME_START_EXTERNAL					\
+     + (ERTS_MONOTONIC_END - ERTS_MONOTONIC_BEGIN))
+
 #define ERTS_MSEC_TO_CLKTCKS__(MON) \
     ((MON) * (ERTS_CLKTCK_RESOLUTION/1000))
 #define ERTS_CLKTCKS_TO_MSEC__(TCKS) \
@@ -355,3 +392,65 @@ erts_time_unit_conversion(Uint64 value,
      ERTS_CLKTCKS_TO_MSEC__((X)))
 
 #endif /* ERL_TIME_H__ */
+
+/* timer-wheel api */
+#if defined(ERTS_WANT_TIMER_WHEEL_API) && !defined(ERTS_GOT_TIMER_WHEEL_API)
+#define ERTS_GOT_TIMER_WHEEL_API
+
+#include "erl_thr_progress.h"
+#include "erl_process.h"
+
+void erts_sched_init_time_sup(ErtsSchedulerData *esdp);
+
+
+#define ERTS_TWHEEL_SLOT_AT_ONCE -1
+#define ERTS_TWHEEL_SLOT_INACTIVE -2
+
+/*
+** Timer entry:
+*/
+typedef struct erl_timer {
+    struct erl_timer* next;	/* next entry tiw slot or chain */
+    struct erl_timer* prev;	/* prev entry tiw slot or chain */
+    union {
+	struct {
+	    void (*timeout)(void*); /* called when timeout */
+	    void (*cancel)(void*);  /* called when cancel (may be NULL) */
+	    void* arg;              /* argument to timeout/cancel procs */
+	} func;
+	ErtsThrPrgrLaterOp cleanup;
+    } u;
+    ErtsMonotonicTime timeout_pos; /* Timeout in absolute clock ticks */
+    int slot;
+} ErtsTWheelTimer;
+
+typedef void (*ErlTimeoutProc)(void*);
+typedef void (*ErlCancelProc)(void*);
+
+void erts_twheel_set_timer(ErtsTimerWheel *tiw,
+			   ErtsTWheelTimer *p, ErlTimeoutProc timeout,
+			   ErlCancelProc cancel, void *arg,
+			   ErtsMonotonicTime timeout_pos);
+void erts_twheel_cancel_timer(ErtsTimerWheel *tiw, ErtsTWheelTimer *p);
+ErtsTimerWheel *erts_create_timer_wheel(ErtsSchedulerData *esdp);
+
+ErtsMonotonicTime erts_check_next_timeout_time(ErtsSchedulerData *);
+
+ERTS_GLB_INLINE void erts_twheel_init_timer(ErtsTWheelTimer *p);
+ERTS_GLB_INLINE ErtsMonotonicTime erts_next_timeout_time(ErtsNextTimeoutRef);
+
+#if ERTS_GLB_INLINE_INCL_FUNC_DEF
+
+ERTS_GLB_INLINE void erts_twheel_init_timer(ErtsTWheelTimer *p)
+{
+    p->slot = ERTS_TWHEEL_SLOT_INACTIVE;
+}
+
+ERTS_GLB_INLINE ErtsMonotonicTime erts_next_timeout_time(ErtsNextTimeoutRef nxt_tmo_ref)
+{
+    return *((ErtsMonotonicTime *) nxt_tmo_ref);
+}
+
+#endif /* ERTS_GLB_INLINE_INCL_FUNC_DEF */
+
+#endif /* timer wheel api */
diff --git a/erts/emulator/beam/erl_time_sup.c b/erts/emulator/beam/erl_time_sup.c
index 2b6abcda55..e550c999b8 100644
--- a/erts/emulator/beam/erl_time_sup.c
+++ b/erts/emulator/beam/erl_time_sup.c
@@ -30,6 +30,8 @@
 #include "sys.h"
 #include "erl_vm.h"
 #include "global.h"
+#define ERTS_WANT_TIMER_WHEEL_API
+#include "erl_time.h"
  
 static erts_smp_mtx_t erts_timeofday_mtx;
 static erts_smp_mtx_t erts_get_time_mtx;
@@ -57,76 +59,6 @@ static int time_sup_initialized = 0;
 static void
 schedule_send_time_offset_changed_notifications(ErtsMonotonicTime new_offset);
 
-/*
- * NOTE! ERTS_MONOTONIC_TIME_START *need* to be a multiple
- *       of ERTS_MONOTONIC_TIME_UNIT.
- */
-
-#if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT
-
-#ifdef ARCH_32
-/*
- * Want to use a big-num of arity 2 as long as possible (584 years
- * in the nano-second time unit case).
- */
-#define ERTS_MONOTONIC_TIME_START		\
-    (((((((ErtsMonotonicTime) 1) << 32)-1)	\
-       / ERTS_MONOTONIC_TIME_UNIT)		\
-      * ERTS_MONOTONIC_TIME_UNIT)		\
-     + ERTS_MONOTONIC_TIME_UNIT)
-
-#else /* ARCH_64 */
-
-#if ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT <= 10*1000*1000
-
-/*
- * Using micro second time unit or lower. Start at zero since
- * time will remain an immediate for a very long time anyway
- * (1827 years in the 10 micro second case)...
- */
-#define ERTS_MONOTONIC_TIME_START ((ErtsMonotonicTime) 0)
-
-#else /* ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT > 1000*1000 */
-
-/*
- * Want to use an immediate as long as possible (36 years in the
- * nano-second time unit case).
-*/
-#define ERTS_MONOTONIC_TIME_START 		\
-    ((((ErtsMonotonicTime) MIN_SMALL)		\
-      / ERTS_MONOTONIC_TIME_UNIT)		\
-     * ERTS_MONOTONIC_TIME_UNIT)
-
-#endif /* ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT > 1000*1000 */
-
-#endif /* ARCH_64 */
-
-#define ERTS_MONOTONIC_OFFSET_NATIVE \
-    (ERTS_MONOTONIC_TIME_START - ERTS_MONOTONIC_TIME_UNIT)
-#define ERTS_MONOTONIC_OFFSET_NSEC					\
-    ERTS_MONOTONIC_TO_NSEC__(ERTS_MONOTONIC_OFFSET_NATIVE)
-#define ERTS_MONOTONIC_OFFSET_USEC					\
-    ERTS_MONOTONIC_TO_USEC__(ERTS_MONOTONIC_OFFSET_NATIVE)
-#define ERTS_MONOTONIC_OFFSET_MSEC					\
-    ERTS_MONOTONIC_TO_MSEC__(ERTS_MONOTONIC_OFFSET_NATIVE)
-#define ERTS_MONOTONIC_OFFSET_SEC					\
-    ERTS_MONOTONIC_TO_SEC__(ERTS_MONOTONIC_OFFSET_NATIVE)
-
-#else /* ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT */
-
-/*
- * Initialized in erts_init_time_sup()...
- */
-
-#define ERTS_MONOTONIC_TIME_START (time_sup.r.o.start)
-#define ERTS_MONOTONIC_OFFSET_NATIVE (time_sup.r.o.start_offset.native)
-#define ERTS_MONOTONIC_OFFSET_NSEC (time_sup.r.o.start_offset.nsec)
-#define ERTS_MONOTONIC_OFFSET_USEC (time_sup.r.o.start_offset.usec)
-#define ERTS_MONOTONIC_OFFSET_MSEC (time_sup.r.o.start_offset.msec)
-#define ERTS_MONOTONIC_OFFSET_SEC (time_sup.r.o.start_offset.sec)
-
-#endif /* ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT */
-
 struct time_sup_read_only__ {
     ErtsMonotonicTime (*get_time)(void);
     int correction;
@@ -146,16 +78,6 @@ struct time_sup_read_only__ {
     int os_system_time_locked;
     Uint64 os_system_time_resolution;
     Uint64 os_system_time_extended;
-#if !ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT
-    ErtsMonotonicTime start;
-    struct {
-	ErtsMonotonicTime native;
-	ErtsMonotonicTime nsec;
-	ErtsMonotonicTime usec;
-	ErtsMonotonicTime msec;
-	ErtsMonotonicTime sec;
-    } start_offset;
-#endif
     struct {
 	ErtsMonotonicTime large_diff;
 	ErtsMonotonicTime small_diff;
@@ -211,7 +133,7 @@ struct time_sup_infrequently_changed__ {
 #ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT
     struct {
 	erts_smp_rwmtx_t rwmtx;
-	ErlTimer timer;
+	ErtsTWheelTimer timer;
 	ErtsMonotonicCorrectionData cdata;
     } parmon;
     ErtsMonotonicTime minit;
@@ -273,6 +195,17 @@ get_time_offset(void)
     return (ErtsMonotonicTime) erts_smp_atomic64_read_acqb(&time_sup.inf.c.offset);
 }
 
+static ERTS_INLINE void
+update_last_mtime(ErtsSchedulerData *esdp, ErtsMonotonicTime mtime)
+{
+    if (!esdp)
+	esdp = erts_get_scheduler_data();
+    if (esdp) {
+	ASSERT(mtime >= esdp->last_monotonic_time);
+	esdp->last_monotonic_time = mtime;
+	esdp->check_time_reds = 0;
+    }
+}
 
 #ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT
 
@@ -411,15 +344,26 @@ print_correction(int change,
 
 #endif
 
+static ERTS_INLINE ErtsMonotonicTime
+get_timeout_pos(ErtsMonotonicTime now, ErtsMonotonicTime tmo)
+{
+    ErtsMonotonicTime tpos;
+    tpos = ERTS_MONOTONIC_TO_CLKTCKS(now - 1);
+    tpos += ERTS_MSEC_TO_CLKTCKS(tmo);
+    tpos += 1;
+    return tpos;
+}
+
 static void
-check_time_correction(void *idap)
+check_time_correction(void *vesdp)
 {
-    UWord init_drift_adj = (UWord) idap;
+    int init_drift_adj = !vesdp;
+    ErtsSchedulerData *esdp = (ErtsSchedulerData *) vesdp;
     ErtsMonotonicCorrectionData cdata;
     ErtsMonotonicCorrection new_correction;
     ErtsMonotonicCorrectionInstance *cip;
     ErtsMonotonicTime mdiff, sdiff, os_mtime, erl_mtime, os_stime,
-	erl_stime, time_offset;
+	erl_stime, time_offset, timeout_pos;
     Uint timeout;
     int os_drift_corrected = time_sup.r.o.os_corrected_monotonic_time;
     int set_new_correction = 0, begin_short_intervals = 0;
@@ -681,6 +625,8 @@ check_time_correction(void *idap)
 	timeout = ERTS_MONOTONIC_TO_MSEC(ERTS_SHORT_TIME_CORRECTION_CHECK);
     }
 
+    timeout_pos = get_timeout_pos(erl_mtime, timeout);
+
 #ifdef ERTS_TIME_CORRECTION_PRINT
     print_correction(set_new_correction,
 		     sdiff,
@@ -723,11 +669,15 @@ check_time_correction(void *idap)
 	erts_smp_rwmtx_rwunlock(&time_sup.inf.c.parmon.rwmtx);
     }
 
-    erts_set_timer(&time_sup.inf.c.parmon.timer,
-		   check_time_correction,
-		   NULL,
-		   NULL,
-		   timeout);
+    if (!esdp)
+	esdp = erts_get_scheduler_data();
+
+    erts_twheel_set_timer(esdp->timer_wheel,
+			  &time_sup.inf.c.parmon.timer,
+			  check_time_correction,
+			  NULL,
+			  (void *) esdp,
+			  timeout_pos);
 }
 
 static ErtsMonotonicTime get_os_corrected_time(void)
@@ -737,10 +687,11 @@ static ErtsMonotonicTime get_os_corrected_time(void)
 }
 
 static void
-check_time_offset(void *unused)
+check_time_offset(void *vesdp)
 {
+    ErtsSchedulerData *esdp = (ErtsSchedulerData *) vesdp;
     ErtsMonotonicTime sdiff, os_mtime, erl_mtime, os_stime,
-	erl_stime, time_offset;
+	erl_stime, time_offset, timeout, timeout_pos;
 
     ASSERT(time_sup.r.o.warp_mode == ERTS_MULTI_TIME_WARP_MODE);
 
@@ -769,15 +720,19 @@ check_time_offset(void *unused)
 		      ERTS_MONOTONIC_TO_NSEC(sdiff));
 #endif
 
-    erts_set_timer(&time_sup.inf.c.parmon.timer,
-		   check_time_offset,
-		   NULL,
-		   NULL,
-		   ERTS_MONOTONIC_TO_MSEC(ERTS_LONG_TIME_CORRECTION_CHECK));
+    timeout = ERTS_MONOTONIC_TO_MSEC(ERTS_LONG_TIME_CORRECTION_CHECK);
+    timeout_pos = get_timeout_pos(erl_mtime, timeout);
+
+    erts_twheel_set_timer(esdp->timer_wheel,
+			  &time_sup.inf.c.parmon.timer,
+			  check_time_offset,
+			  NULL,
+			  vesdp,
+			  timeout_pos);
 }
 
 static void
-init_check_time_correction(void *quick_init_drift)
+init_check_time_correction(void *vesdp)
 {
     ErtsMonotonicDriftData *ddp;
     ErtsMonotonicTime old_mtime, old_stime, mtime, stime, mtime_diff,
@@ -821,7 +776,7 @@ init_check_time_correction(void *quick_init_drift)
     ddp->ix = 0;
     ddp->dirty_counter = time_sup.r.o.drift_adj.intervals;
 
-    check_time_correction(quick_init_drift);
+    check_time_correction(vesdp);
 }
 
 static ErtsMonotonicTime
@@ -850,14 +805,14 @@ finalize_corrected_time_offset(ErtsSystemTime *stimep)
 }
 
 static void
-late_init_time_correction(void)
+late_init_time_correction(ErtsSchedulerData *esdp)
 {
-    Uint timeout;
-    Uint quick_init_drift_adj;
+    int quick_init_drift_adj;
     void (*check_func)(void *);
+    ErtsMonotonicTime timeout, timeout_pos;
 
     quick_init_drift_adj =
-	(Uint) ERTS_MONOTONIC_TO_USEC(time_sup.r.o.drift_adj.error) == 0;
+	ERTS_MONOTONIC_TO_USEC(time_sup.r.o.drift_adj.error) == 0;
 
     if (quick_init_drift_adj)
 	timeout = ERTS_MONOTONIC_TO_MSEC(ERTS_SHORT_TIME_CORRECTION_CHECK/10);
@@ -866,17 +821,25 @@ late_init_time_correction(void)
 
     if (!time_sup.r.o.os_corrected_monotonic_time)
 	check_func = init_check_time_correction;
-    else if (time_sup.r.o.get_time == get_os_corrected_time)
+    else if (time_sup.r.o.get_time == get_os_corrected_time) {
+	quick_init_drift_adj = 0;
 	check_func = check_time_offset;
+    }
     else
 	check_func = check_time_correction;
 
-    erts_init_timer(&time_sup.inf.c.parmon.timer);
-    erts_set_timer(&time_sup.inf.c.parmon.timer,
-		   check_func,
-		   NULL,
-		   (void *) quick_init_drift_adj,
-		   timeout);
+    timeout_pos = get_timeout_pos(erts_get_monotonic_time(esdp),
+				  timeout);
+
+    erts_twheel_init_timer(&time_sup.inf.c.parmon.timer);
+    erts_twheel_set_timer(esdp->timer_wheel,
+			  &time_sup.inf.c.parmon.timer,
+			  check_func,
+			  NULL,
+			  (quick_init_drift_adj
+			   ? NULL
+			   : esdp),
+			  timeout_pos);
 }
 
 #endif /* ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT */
@@ -987,6 +950,8 @@ erts_init_time_sup(int time_correction, ErtsTimeWarpMode time_warp_mode)
     ErtsMonotonicTime abs_native_offset, native_offset;
 #endif
 
+    erts_hl_timer_init();
+
     ASSERT(ERTS_MONOTONIC_TIME_MIN < ERTS_MONOTONIC_TIME_MAX);
 
     erts_smp_mtx_init(&erts_timeofday_mtx, "timeofday");
@@ -1003,51 +968,55 @@ erts_init_time_sup(int time_correction, ErtsTimeWarpMode time_warp_mode)
 
 #if !ERTS_COMPILE_TIME_MONOTONIC_TIME_UNIT
 
+    /*
+     * NOTE! erts_time_sup__.r.o.start *need* to be a multiple
+     *       of ERTS_MONOTONIC_TIME_UNIT.
+     */
+
 #ifdef ARCH_32
-    time_sup.r.o.start = ((((ErtsMonotonicTime) 1) << 32)-1);
-    time_sup.r.o.start /= ERTS_MONOTONIC_TIME_UNIT;
-    time_sup.r.o.start *= ERTS_MONOTONIC_TIME_UNIT;
-    time_sup.r.o.start += ERTS_MONOTONIC_TIME_UNIT;
-    native_offset = time_sup.r.o.start - ERTS_MONOTONIC_TIME_UNIT;
+    erts_time_sup__.r.o.start = ((((ErtsMonotonicTime) 1) << 32)-1);
+    erts_time_sup__.r.o.start /= ERTS_MONOTONIC_TIME_UNIT;
+    erts_time_sup__.r.o.start *= ERTS_MONOTONIC_TIME_UNIT;
+    erts_time_sup__.r.o.start += ERTS_MONOTONIC_TIME_UNIT;
+    native_offset = erts_time_sup__.r.o.start - ERTS_MONOTONIC_BEGIN;
     abs_native_offset = native_offset;
 #else /* ARCH_64 */
     if (ERTS_MONOTONIC_TIME_UNIT <= 10*1000*1000) {
-	time_sup.r.o.start = 0;
-	native_offset = -ERTS_MONOTONIC_TIME_UNIT;
-	abs_native_offset = ERTS_MONOTONIC_TIME_UNIT;
+	erts_time_sup__.r.o.start = 0;
+	native_offset = -ERTS_MONOTONIC_BEGIN;
+	abs_native_offset = ERTS_MONOTONIC_BEGIN;
     }
     else {
-	time_sup.r.o.start = ((ErtsMonotonicTime) MIN_SMALL);
-	time_sup.r.o.start /= ERTS_MONOTONIC_TIME_UNIT;
-	time_sup.r.o.start *= ERTS_MONOTONIC_TIME_UNIT;
-	native_offset = time_sup.r.o.start - ERTS_MONOTONIC_TIME_UNIT;
+	erts_time_sup__.r.o.start = ((ErtsMonotonicTime) MIN_SMALL);
+	erts_time_sup__.r.o.start /= ERTS_MONOTONIC_TIME_UNIT;
+	erts_time_sup__.r.o.start *= ERTS_MONOTONIC_TIME_UNIT;
+	native_offset = erts_time_sup__.r.o.start - ERTS_MONOTONIC_BEGIN;
 	abs_native_offset = -1*native_offset;
     }
 #endif
 
-    time_sup.r.o.start_offset.native = (time_sup.r.o.start
-					- ERTS_MONOTONIC_TIME_UNIT);
-    time_sup.r.o.start_offset.nsec = (ErtsMonotonicTime)
+    erts_time_sup__.r.o.start_offset.native = native_offset;
+    erts_time_sup__.r.o.start_offset.nsec = (ErtsMonotonicTime)
 	erts_time_unit_conversion((Uint64) abs_native_offset,
 				  (Uint32) ERTS_MONOTONIC_TIME_UNIT,
 				  (Uint32) 1000*1000*1000);
-    time_sup.r.o.start_offset.usec = (ErtsMonotonicTime)
+    erts_time_sup__.r.o.start_offset.usec = (ErtsMonotonicTime)
 	erts_time_unit_conversion((Uint64) abs_native_offset,
 				  (Uint32) ERTS_MONOTONIC_TIME_UNIT,
 				  (Uint32) 1000*1000);
-    time_sup.r.o.start_offset.msec = (ErtsMonotonicTime)
+    erts_time_sup__.r.o.start_offset.msec = (ErtsMonotonicTime)
 	erts_time_unit_conversion((Uint64) abs_native_offset,
 				  (Uint32) ERTS_MONOTONIC_TIME_UNIT,
 				  (Uint32) 1000);
-    time_sup.r.o.start_offset.sec = (ErtsMonotonicTime)
+    erts_time_sup__.r.o.start_offset.sec = (ErtsMonotonicTime)
 	erts_time_unit_conversion((Uint64) abs_native_offset,
 				  (Uint32) ERTS_MONOTONIC_TIME_UNIT,
 				  (Uint32) 1);
     if (native_offset < 0) {
-	time_sup.r.o.start_offset.nsec *= -1;
-	time_sup.r.o.start_offset.usec *= -1;
-	time_sup.r.o.start_offset.msec *= -1;
-	time_sup.r.o.start_offset.sec *= -1;
+	erts_time_sup__.r.o.start_offset.nsec *= -1;
+	erts_time_sup__.r.o.start_offset.usec *= -1;
+	erts_time_sup__.r.o.start_offset.msec *= -1;
+	erts_time_sup__.r.o.start_offset.sec *= -1;
     }
 
 #endif
@@ -1143,9 +1112,9 @@ erts_init_time_sup(int time_correction, ErtsTimeWarpMode time_warp_mode)
 	erts_os_times(&time_sup.inf.c.minit,
 		      &time_sup.inf.c.sinit);
 	time_sup.r.o.moffset = -1*time_sup.inf.c.minit;
-	time_sup.r.o.moffset += ERTS_MONOTONIC_TIME_UNIT;
+	time_sup.r.o.moffset += ERTS_MONOTONIC_BEGIN;
 	offset = time_sup.inf.c.sinit;
-	offset -= ERTS_MONOTONIC_TIME_UNIT;
+	offset -= ERTS_MONOTONIC_BEGIN;
 	init_time_offset(offset);
 
 	rwmtx_opts.type = ERTS_SMP_RWMTX_TYPE_EXTREMELY_FREQUENT_READ;
@@ -1160,7 +1129,7 @@ erts_init_time_sup(int time_correction, ErtsTimeWarpMode time_warp_mode)
 	cdatap->drift.intervals[0].time.mon = time_sup.inf.c.minit;
 	cdatap->curr.correction.drift = 0;
 	cdatap->curr.correction.error = 0;
-	cdatap->curr.erl_mtime = ERTS_MONOTONIC_TIME_UNIT;
+	cdatap->curr.erl_mtime = ERTS_MONOTONIC_BEGIN;
 	cdatap->curr.os_mtime = time_sup.inf.c.minit;
 	cdatap->last_check = time_sup.inf.c.minit;
 	cdatap->short_check_interval = ERTS_INIT_SHORT_INTERVAL_COUNTER;
@@ -1179,7 +1148,7 @@ erts_init_time_sup(int time_correction, ErtsTimeWarpMode time_warp_mode)
 	ErtsMonotonicTime stime, offset;
 	time_sup.r.o.get_time = get_not_corrected_time;
 	stime = time_sup.inf.c.sinit = erts_os_system_time();
-	offset = stime - ERTS_MONOTONIC_TIME_UNIT;
+	offset = stime - ERTS_MONOTONIC_BEGIN;
 	time_sup.inf.c.not_corrected_moffset = offset;
 	init_time_offset(offset);
 	time_sup.f.c.last_not_corrected_time = 0;
@@ -1199,14 +1168,24 @@ erts_init_time_sup(int time_correction, ErtsTimeWarpMode time_warp_mode)
 void
 erts_late_init_time_sup(void)
 {
-#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT
-    /* Timer wheel must have beeen initialized */
-    if (time_sup.r.o.get_time != get_not_corrected_time)
-	late_init_time_correction();
-#endif
     erts_late_sys_init_time();
 }
 
+void
+erts_sched_init_time_sup(ErtsSchedulerData *esdp)
+{
+    esdp->timer_wheel = erts_create_timer_wheel(esdp);
+    esdp->next_tmo_ref = erts_get_next_timeout_reference(esdp->timer_wheel);
+    esdp->timer_service = erts_create_timer_service();
+#ifdef ERTS_HAVE_OS_MONOTONIC_TIME_SUPPORT
+    if (esdp->no == 1) {
+	/* A timer wheel to use must have beeen initialized */
+	if (time_sup.r.o.get_time != get_not_corrected_time)
+	    late_init_time_correction(esdp);
+    }
+#endif	
+}
+
 ErtsTimeWarpMode erts_time_warp_mode(void)
 {
     return time_sup.r.o.warp_mode;
@@ -1349,6 +1328,7 @@ wall_clock_elapsed_time_both(UWord *ms_total, UWord *ms_diff)
     erts_smp_mtx_lock(&erts_timeofday_mtx);
 
     now = time_sup.r.o.get_time();
+    update_last_mtime(NULL, now);
 
     elapsed = ERTS_MONOTONIC_TO_MSEC(now);
     *ms_total = (UWord) elapsed;
@@ -1737,6 +1717,7 @@ get_now(Uint* megasec, Uint* sec, Uint* microsec)
     
     mtime = time_sup.r.o.get_time();
     time_offset = get_time_offset();
+    update_last_mtime(NULL, mtime);
     now = ERTS_MONOTONIC_TO_USEC(mtime + time_offset);
 
     erts_smp_mtx_lock(&erts_timeofday_mtx);
@@ -1761,9 +1742,11 @@ get_now(Uint* megasec, Uint* sec, Uint* microsec)
 }
 
 ErtsMonotonicTime
-erts_get_monotonic_time(void)
+erts_get_monotonic_time(ErtsSchedulerData *esdp)
 {
-    return time_sup.r.o.get_time();
+    ErtsMonotonicTime mtime = time_sup.r.o.get_time();
+    update_last_mtime(esdp, mtime);
+    return mtime;    
 }
 
 void
@@ -1990,7 +1973,13 @@ make_time_val(Process *c_p, ErtsMonotonicTime time_val)
 Eterm
 erts_get_monotonic_start_time(struct process *c_p)
 {
-    return make_time_val(c_p, ERTS_MONOTONIC_TIME_START);
+    return make_time_val(c_p, ERTS_MONOTONIC_TIME_START_EXTERNAL);
+}
+
+Eterm
+erts_get_monotonic_end_time(struct process *c_p)
+{
+    return make_time_val(c_p, ERTS_MONOTONIC_TIME_END_EXTERNAL);
 }
 
 static Eterm
@@ -2182,13 +2171,16 @@ time_unit_conversion(Process *c_p, Eterm term, ErtsMonotonicTime val, ErtsMonoto
 BIF_RETTYPE monotonic_time_0(BIF_ALIST_0)
 {
     ErtsMonotonicTime mtime = time_sup.r.o.get_time();
+    update_last_mtime(ERTS_PROC_GET_SCHDATA(BIF_P), mtime);
     mtime += ERTS_MONOTONIC_OFFSET_NATIVE;
     BIF_RET(make_time_val(BIF_P, mtime));
 }
 
 BIF_RETTYPE monotonic_time_1(BIF_ALIST_1)
 {
-    BIF_RET(time_unit_conversion(BIF_P, BIF_ARG_1, time_sup.r.o.get_time(), 1));
+    ErtsMonotonicTime mtime = time_sup.r.o.get_time();
+    update_last_mtime(ERTS_PROC_GET_SCHDATA(BIF_P), mtime);
+    BIF_RET(time_unit_conversion(BIF_P, BIF_ARG_1, mtime, 1));
 }
 
 BIF_RETTYPE system_time_0(BIF_ALIST_0)
@@ -2196,6 +2188,7 @@ BIF_RETTYPE system_time_0(BIF_ALIST_0)
     ErtsMonotonicTime mtime, offset;
     mtime = time_sup.r.o.get_time();
     offset = get_time_offset();
+    update_last_mtime(ERTS_PROC_GET_SCHDATA(BIF_P), mtime);
     BIF_RET(make_time_val(BIF_P, mtime + offset));
 }
 
@@ -2204,6 +2197,7 @@ BIF_RETTYPE system_time_1(BIF_ALIST_0)
     ErtsMonotonicTime mtime, offset;
     mtime = time_sup.r.o.get_time();
     offset = get_time_offset();
+    update_last_mtime(ERTS_PROC_GET_SCHDATA(BIF_P), mtime);
     BIF_RET(time_unit_conversion(BIF_P, BIF_ARG_1, mtime + offset, 0));
 }
 
@@ -2233,6 +2227,7 @@ BIF_RETTYPE timestamp_0(BIF_ALIST_0)
 
     mtime = time_sup.r.o.get_time();
     offset = get_time_offset();
+    update_last_mtime(ERTS_PROC_GET_SCHDATA(BIF_P), mtime);
     stime = ERTS_MONOTONIC_TO_USEC(mtime + offset);
     all_sec = stime / ERTS_MONOTONIC_TIME_MEGA;
     mega_sec = (Uint) (stime / ERTS_MONOTONIC_TIME_TERA);
diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h
index 54fba9274f..340c7033ab 100644
--- a/erts/emulator/beam/global.h
+++ b/erts/emulator/beam/global.h
@@ -1309,8 +1309,7 @@ erts_alloc_message_heap_state(Uint size,
     state = erts_smp_atomic32_read_acqb(&receiver->state);
     if (statep)
 	*statep = state;
-    if (state & (ERTS_PSFLG_OFF_HEAP_MSGS
-		 | ERTS_PSFLG_EXITING
+    if (state & (ERTS_PSFLG_EXITING
 		 | ERTS_PSFLG_PENDING_EXIT))
 	goto allocate_in_mbuf;
 #endif
@@ -1331,8 +1330,7 @@ erts_alloc_message_heap_state(Uint size,
 	state = erts_smp_atomic32_read_nob(&receiver->state);
 	if (statep)
 	    *statep = state;
-	if ((state & (ERTS_PSFLG_OFF_HEAP_MSGS
-		      | ERTS_PSFLG_EXITING
+	if ((state & (ERTS_PSFLG_EXITING
 		      | ERTS_PSFLG_PENDING_EXIT))
 	    || (receiver->flags & F_DISABLE_GC)
 	    || HEAP_LIMIT(receiver) - HEAP_TOP(receiver) <= size) {
diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c
index dec92be40a..ccc7da265e 100644
--- a/erts/emulator/beam/io.c
+++ b/erts/emulator/beam/io.c
@@ -48,6 +48,7 @@
 #include "dtrace-wrapper.h"
 #include "erl_map.h"
 #include "erl_bif_unique.h"
+#include "erl_hl_timer.h"
 
 extern ErlDrvEntry fd_driver_entry;
 #ifndef __OSE__
@@ -379,11 +380,7 @@ static Port *create_port(char *name,
     prt->dist_entry = NULL;
     ERTS_PORT_INIT_CONNECTED(prt, pid);
     prt->common.u.alive.reg = NULL;
-#ifdef ERTS_SMP
-    prt->common.u.alive.ptimer = NULL;
-#else
-    sys_memset(&prt->common.u.alive.tm, 0, sizeof(ErlTimer));
-#endif
+    ERTS_PTMR_INIT(prt);
     erts_port_task_handle_init(&prt->timeout_task);
     prt->psd = NULL;
     prt->drv_data = (SWord) 0;
@@ -463,11 +460,7 @@ erts_port_free(Port *prt)
 			    | ERTS_PORT_SFLG_FREE));
     ASSERT(state & ERTS_PORT_SFLG_PORT_DEBUG);
 
-#ifdef ERTS_SMP
-    ERTS_LC_ASSERT(erts_atomic32_read_nob(&prt->common.refc) == 0);
-#else
-    ERTS_LC_ASSERT(erts_atomic32_read_nob(&prt->refc) == 0);
-#endif
+    ERTS_LC_ASSERT(erts_atomic_read_nob(&prt->common.refc.atmc) == 0);
 
     erts_port_task_fini_sched(&prt->sched);
 
@@ -736,11 +729,7 @@ erts_open_driver(erts_driver_t* driver,	/* Pointer to driver. */
 	/*
 	 * Must clean up the port.
 	 */
-#ifdef ERTS_SMP
-	erts_cancel_smp_ptimer(port->common.u.alive.ptimer);
-#else
-	erts_cancel_timer(&(port->common.u.alive.tm));
-#endif
+	erts_cancel_port_timer(port);
 	stopq(port);
 	if (port->linebuf != NULL) {
 	    erts_free(ERTS_ALC_T_LINEBUF,
@@ -2798,7 +2787,8 @@ void erts_init_io(int port_tab_size,
 			 port_tab_size,
 			 common_element_size, /* Doesn't need to be excact */
 			 "port_table",
-			 legacy_port_tab);
+			 legacy_port_tab,
+			 1);
 
     erts_smp_atomic_init_nob(&erts_bytes_out, 0);
     erts_smp_atomic_init_nob(&erts_bytes_in, 0);
@@ -3065,7 +3055,7 @@ deliver_result(Eterm sender, Eterm pid, Eterm res)
 
     rp = (scheduler
 	  ? erts_proc_lookup(pid)
-	  : erts_pid2proc_opt(NULL, 0, pid, 0, ERTS_P2P_FLG_SMP_INC_REFC));
+	  : erts_pid2proc_opt(NULL, 0, pid, 0, ERTS_P2P_FLG_INC_REFC));
 
     if (rp) {
 	Eterm tuple;
@@ -3083,7 +3073,7 @@ deliver_result(Eterm sender, Eterm pid, Eterm res)
 	if (rp_locks)
 	    erts_smp_proc_unlock(rp, rp_locks);
 	if (!scheduler)
-	    erts_smp_proc_dec_refc(rp);
+	    erts_proc_dec_refc(rp);
 
     }
 }
@@ -3127,7 +3117,7 @@ static void deliver_read_message(Port* prt, erts_aint32_t state, Eterm to,
 
     rp = (scheduler
 	  ? erts_proc_lookup(to)
-	  : erts_pid2proc_opt(NULL, 0, to, 0, ERTS_P2P_FLG_SMP_INC_REFC));
+	  : erts_pid2proc_opt(NULL, 0, to, 0, ERTS_P2P_FLG_INC_REFC));
 
     if (!rp)
 	return;
@@ -3178,7 +3168,7 @@ static void deliver_read_message(Port* prt, erts_aint32_t state, Eterm to,
     if (rp_locks)
 	erts_smp_proc_unlock(rp, rp_locks);
     if (!scheduler)
-	erts_smp_proc_dec_refc(rp);
+	erts_proc_dec_refc(rp);
 }
 
 /* 
@@ -3264,7 +3254,7 @@ deliver_vec_message(Port* prt,			/* Port */
 
     rp = (scheduler
 	  ? erts_proc_lookup(to)
-	  : erts_pid2proc_opt(NULL, 0, to, 0, ERTS_P2P_FLG_SMP_INC_REFC));
+	  : erts_pid2proc_opt(NULL, 0, to, 0, ERTS_P2P_FLG_INC_REFC));
     if (!rp)
 	return;
 
@@ -3344,7 +3334,7 @@ deliver_vec_message(Port* prt,			/* Port */
     erts_queue_message(rp, &rp_locks, bp, tuple, am_undefined);
     erts_smp_proc_unlock(rp, rp_locks);
     if (!scheduler)
-	erts_smp_proc_dec_refc(rp);
+	erts_proc_dec_refc(rp);
 }
 
 
@@ -3434,11 +3424,8 @@ terminate_port(Port *prt)
 	send_closed_port_id = NIL;
     }
 
-#ifdef ERTS_SMP
-    erts_cancel_smp_ptimer(prt->common.u.alive.ptimer);
-#else
-    erts_cancel_timer(&prt->common.u.alive.tm);
-#endif
+    if (ERTS_PTMR_IS_SET(prt))
+	erts_cancel_port_timer(prt);
 
     drv = prt->drv_ptr;
     if ((drv != NULL) && (drv->stop != NULL)) {
@@ -4985,24 +4972,6 @@ erts_free_port_names(ErtsPortNames *pnp)
     erts_free(ERTS_ALC_T_PORT_NAMES, pnp);
 }
 
-static void schedule_port_timeout(Port *p)
-{
-    /*
-     * Scheduling of port timeouts can be done without port locking, but
-     * since the task handle is stored in the port structure and the ptimer
-     * structure is protected by the port lock we require the port to be
-     * locked for now...
-     *
-     * TODO: Implement scheduling of port timeouts without locking
-     *       the port.
-     * /Rickard
-     */
-    ERTS_SMP_LC_ASSERT(erts_lc_is_port_locked(p));
-    erts_port_task_schedule(p->common.id,
-			    &p->timeout_task,
-			    ERTS_PORT_TASK_TIMEOUT);
-}
-
 ErlDrvTermData driver_mk_term_nil(void)
 {
     return driver_term_nil;
@@ -5031,7 +5000,7 @@ void driver_report_exit(ErlDrvPort ix, int status)
 
    rp = (scheduler
 	 ? erts_proc_lookup(pid)
-	 : erts_pid2proc_opt(NULL, 0, pid, 0, ERTS_P2P_FLG_SMP_INC_REFC));
+	 : erts_pid2proc_opt(NULL, 0, pid, 0, ERTS_P2P_FLG_INC_REFC));
    if (!rp)
        return;
 
@@ -5045,7 +5014,7 @@ void driver_report_exit(ErlDrvPort ix, int status)
 
    erts_smp_proc_unlock(rp, rp_locks);
    if (!scheduler)
-       erts_smp_proc_dec_refc(rp);
+       erts_proc_dec_refc(rp);
 }
 
 #define ERTS_B2T_STATES_DEF_STATES_SZ 5
@@ -5361,7 +5330,7 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len)
     scheduler = erts_get_scheduler_id() != 0;
     rp = (scheduler
 	  ? erts_proc_lookup(to)
-	  : erts_pid2proc_opt(NULL, 0, to, 0, ERTS_P2P_FLG_SMP_INC_REFC));
+	  : erts_pid2proc_opt(NULL, 0, to, 0, ERTS_P2P_FLG_INC_REFC));
     if (!rp) {
 	res = 0;
 	goto done;
@@ -5656,14 +5625,12 @@ driver_deliver_term(Eterm to, ErlDrvTermData* data, int len)
 	    HRelease(rp, hp_end, hp);
 	}
     }
-#ifdef ERTS_SMP
     if (rp) {
 	if (rp_locks)
 	    erts_smp_proc_unlock(rp, rp_locks);
 	if (!scheduler)
-	    erts_smp_proc_dec_refc(rp);
+	    erts_proc_dec_refc(rp);
     }
-#endif
     cleanup_b2t_states(&b2t);
     DESTROY_ESTACK(stack);
     return res;
@@ -6609,18 +6576,6 @@ int driver_pushq(ErlDrvPort ix, char* buffer, ErlDrvSizeT len)
     return code;
 }
 
-static ERTS_INLINE void
-drv_cancel_timer(Port *prt)
-{
-#ifdef ERTS_SMP
-    erts_cancel_smp_ptimer(prt->common.u.alive.ptimer);
-#else
-    erts_cancel_timer(&prt->common.u.alive.tm);
-#endif
-    if (erts_port_task_is_scheduled(&prt->timeout_task))
-	erts_port_task_abort(&prt->timeout_task);
-}
-
 int driver_set_timer(ErlDrvPort ix, unsigned long t)
 {
     Port* prt = erts_drvport2port(ix);
@@ -6632,19 +6587,8 @@ int driver_set_timer(ErlDrvPort ix, unsigned long t)
 
     if (prt->drv_ptr->timeout == NULL)
 	return -1;
-    drv_cancel_timer(prt);
-#ifdef ERTS_SMP
-    erts_create_smp_ptimer(&prt->common.u.alive.ptimer,
-			   prt->common.id,
-			   (ErlTimeoutProc) schedule_port_timeout,
-			   t);
-#else
-    erts_set_timer(&prt->common.u.alive.tm,
-		  (ErlTimeoutProc) schedule_port_timeout,
-		  NULL,
-		  prt,
-		  t);
-#endif
+
+    erts_set_port_timer(prt, (Sint64) t);
     return 0;
 }
 
@@ -6654,28 +6598,28 @@ int driver_cancel_timer(ErlDrvPort ix)
     if (prt == ERTS_INVALID_ERL_DRV_PORT)
 	return -1;
     ERTS_SMP_LC_ASSERT(erts_lc_is_port_locked(prt));
-    drv_cancel_timer(prt);
+    erts_cancel_port_timer(prt);
     return 0;
 }
 
-
 int
 driver_read_timer(ErlDrvPort ix, unsigned long* t)
 {
     Port* prt = erts_drvport2port(ix);
+    Sint64 left;
 
     ERTS_SMP_CHK_NO_PROC_LOCKS;
 
     if (prt == ERTS_INVALID_ERL_DRV_PORT)
 	return -1;
     ERTS_SMP_LC_ASSERT(erts_lc_is_port_locked(prt));
-#ifdef ERTS_SMP
-    *t = (prt->common.u.alive.ptimer
-	  ? erts_time_left(&prt->common.u.alive.ptimer->timer.tm)
-	  : 0);
-#else
-    *t = erts_time_left(&prt->common.u.alive.tm);
-#endif
+
+    left = erts_read_port_timer(prt);
+    if (left < 0)
+	left = 0;
+
+    *t = (unsigned long) left;
+
     return 0;
 }
 
diff --git a/erts/emulator/beam/register.c b/erts/emulator/beam/register.c
index c626cb2780..4d557b3a17 100644
--- a/erts/emulator/beam/register.c
+++ b/erts/emulator/beam/register.c
@@ -269,7 +269,10 @@ erts_whereis_name_to_id(Process *c_p, Eterm name)
 #ifdef ERTS_SMP
     ErtsProcLocks c_p_locks = c_p ? ERTS_PROC_LOCK_MAIN : 0;
 
-    ERTS_SMP_CHK_HAVE_ONLY_MAIN_PROC_LOCK(c_p);
+#ifdef ERTS_ENABLE_LOCK_CHECK
+    if (c_p) ERTS_SMP_CHK_HAVE_ONLY_MAIN_PROC_LOCK(c_p);
+#endif
+
     reg_safe_read_lock(c_p, &c_p_locks);
     if (c_p && !c_p_locks)
         erts_smp_proc_lock(c_p, ERTS_PROC_LOCK_MAIN);
@@ -380,8 +383,6 @@ erts_whereis_name(Process *c_p,
 			erts_smp_proc_unlock(rp->p, need_locks);
 		    *proc = NULL;
 		}
-		if (*proc && (flags & ERTS_P2P_FLG_SMP_INC_REFC))
-		    erts_smp_proc_inc_refc(rp->p);
 	    }
 #else
 	    if (rp->p
@@ -390,6 +391,8 @@ erts_whereis_name(Process *c_p,
 	    else
 		*proc = NULL;
 #endif
+	    if (*proc && (flags & ERTS_P2P_FLG_INC_REFC))
+		erts_proc_inc_refc(*proc);
 	}
     }
 
diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h
index f7a21406f3..54059ee9a5 100644
--- a/erts/emulator/beam/sys.h
+++ b/erts/emulator/beam/sys.h
@@ -776,8 +776,6 @@ extern char *erts_sys_ddll_error(int code);
 /*
  * System interfaces for startup.
  */
-#include "erl_time.h"
-
 void erts_sys_schedule_interrupt(int set);
 #ifdef ERTS_SMP
 void erts_sys_schedule_interrupt_timed(int, ErtsMonotonicTime);
@@ -819,7 +817,8 @@ int univ_to_local(
 int local_to_univ(Sint *year, Sint *month, Sint *day, 
 		  Sint *hour, Sint *minute, Sint *second, int isdst);
 void get_now(Uint*, Uint*, Uint*);
-ErtsMonotonicTime erts_get_monotonic_time(void);
+struct ErtsSchedulerData_;
+ErtsMonotonicTime erts_get_monotonic_time(struct ErtsSchedulerData_ *);
 void get_sys_now(Uint*, Uint*, Uint*);
 void set_break_quit(void (*)(void), void (*)(void));
 
diff --git a/erts/emulator/beam/time.c b/erts/emulator/beam/time.c
index 1fffb5f357..ea19d8b362 100644
--- a/erts/emulator/beam/time.c
+++ b/erts/emulator/beam/time.c
@@ -76,6 +76,11 @@
 #include "sys.h"
 #include "erl_vm.h"
 #include "global.h"
+#define ERTS_WANT_TIMER_WHEEL_API
+#include "erl_time.h"
+
+#define ERTS_MONOTONIC_DAY ERTS_SEC_TO_MONOTONIC(60*60*24)
+#define ERTS_CLKTCKS_DAY ERTS_MONOTONIC_TO_CLKTCKS(ERTS_MONOTONIC_DAY)
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
 #define ASSERT_NO_LOCKED_LOCKS		erts_lc_check_exact(NULL, 0)
@@ -83,20 +88,24 @@
 #define ASSERT_NO_LOCKED_LOCKS
 #endif
 
-#define ERTS_MONOTONIC_DAY ERTS_SEC_TO_MONOTONIC(60*60*24)
-#define ERTS_CLKTCKS_DAY ERTS_MONOTONIC_TO_CLKTCKS(ERTS_MONOTONIC_DAY)
-
+#if 0
+#  define ERTS_TW_DEBUG
+#endif
+#if defined(DEBUG) && !defined(ERTS_TW_DEBUG)
+#  define ERTS_TW_DEBUG
+#endif
 
-/* BEGIN tiw_lock protected variables 
-**
-** The individual timer cells in tiw are also protected by the same mutex.
-*/
+#undef ERTS_TW_ASSERT
+#if defined(ERTS_TW_DEBUG)
+#  define ERTS_TW_ASSERT(E) ERTS_ASSERT(E)
+#else
+#  define ERTS_TW_ASSERT(E) ((void) 1)
+#endif
 
-/* timing wheel size NEED to be a power of 2 */
-#ifdef SMALL_MEMORY
-#define TIW_SIZE (1 << 13)
+#ifdef ERTS_TW_DEBUG
+#  define ERTS_TWHEEL_BUMP_YIELD_LIMIT 5
 #else
-#define TIW_SIZE (1 << 20)
+#  define ERTS_TWHEEL_BUMP_YIELD_LIMIT 100
 #endif
 
 /* Actual interval time chosen by sys_init_time() */
@@ -110,54 +119,22 @@ static int tiw_itime; /* Constant after init */
 #endif
 
 struct ErtsTimerWheel_ {
-    ErlTimer *w[TIW_SIZE];
+    ErtsTWheelTimer *w[ERTS_TIW_SIZE];
     ErtsMonotonicTime pos;
     Uint nto;
     struct {
-	ErlTimer *head;
-	ErlTimer *tail;
+	ErtsTWheelTimer *head;
+	ErtsTWheelTimer *tail;
 	Uint nto;
     } at_once;
+    int yield_slot;
+    int yield_slots_left;
+    int yield_start_pos;
+    ErtsTWheelTimer sentinel;
     int true_next_timeout_time;
     ErtsMonotonicTime next_timeout_time;
-    erts_atomic64_t next_timeout;
-    erts_smp_atomic32_t is_bumping;
-    erts_smp_mtx_t lock;
 };
 
-ErtsTimerWheel *erts_default_timer_wheel; /* managed by aux thread */
-
-static ERTS_INLINE ErtsTimerWheel *
-get_timer_wheel(ErlTimer *p)
-{
-    return (ErtsTimerWheel *) erts_smp_atomic_read_acqb(&p->wheel);
-}
-
-static ERTS_INLINE void
-set_timer_wheel(ErlTimer *p, ErtsTimerWheel *tiw)
-{
-    erts_smp_atomic_set_relb(&p->wheel, (erts_aint_t) tiw);
-}
-
-static ERTS_INLINE void
-init_next_timeout(ErtsTimerWheel *tiw,
-		  ErtsMonotonicTime time)
-{
-    erts_atomic64_init_nob(&tiw->next_timeout,
-			   (erts_aint64_t) time);
-}
-
-static ERTS_INLINE void
-set_next_timeout(ErtsTimerWheel *tiw,
-		 ErtsMonotonicTime time,
-		 int true_timeout)
-{
-    tiw->true_next_timeout_time = true_timeout;
-    tiw->next_timeout_time = time;
-    erts_atomic64_set_relb(&tiw->next_timeout,
-			   (erts_aint64_t) time);
-}
-
 /* get the time (in units of TIW_ITIME) to the next timeout,
    or -1 if there are no timeouts                     */
 
@@ -167,35 +144,23 @@ find_next_timeout(ErtsTimerWheel *tiw,
 		  ErtsMonotonicTime max_search_time)
 {
     int start_ix, tiw_pos_ix;
-    ErlTimer *p;
+    ErtsTWheelTimer *p;
     int true_min_timeout;
-    ErtsMonotonicTime min_timeout, min_timeout_pos, slot_timeout_pos, timeout_limit;
-
-    ERTS_SMP_LC_ASSERT(erts_smp_lc_mtx_is_locked(&tiw->lock));
+    ErtsMonotonicTime min_timeout, min_timeout_pos, slot_timeout_pos;
 
     if (tiw->true_next_timeout_time)
 	return tiw->next_timeout_time;
 
-    /* We never set next timeout beyond timeout_limit */
-    timeout_limit = curr_time + ERTS_MONOTONIC_DAY;
-
     if (tiw->nto == 0) { /* no timeouts in wheel */
-	true_min_timeout = tiw->true_next_timeout_time = 0;
-	min_timeout_pos = ERTS_MONOTONIC_TO_CLKTCKS(timeout_limit);
+	true_min_timeout = 0;
+	min_timeout_pos = ERTS_MONOTONIC_TO_CLKTCKS(curr_time + ERTS_MONOTONIC_DAY);
 	goto found_next;
     }
 
-    /*
-     * Don't want others entering trying to bump
-     * timers while we are checking...
-     */
-    set_next_timeout(tiw, timeout_limit, 0);
-
-    true_min_timeout = 1;
     slot_timeout_pos = tiw->pos;
     min_timeout_pos = ERTS_MONOTONIC_TO_CLKTCKS(curr_time + max_search_time);
 
-    start_ix = tiw_pos_ix = (int) (tiw->pos & (TIW_SIZE-1));
+    start_ix = tiw_pos_ix = (int) (tiw->pos & (ERTS_TIW_SIZE-1));
 
     do {
 	slot_timeout_pos++;
@@ -206,93 +171,125 @@ find_next_timeout(ErtsTimerWheel *tiw,
 
 	p = tiw->w[tiw_pos_ix];
 
-	while (p) {
-	    ErtsMonotonicTime timeout_pos;
-	    ASSERT(p != p->next);
-	    timeout_pos = p->timeout_pos;
-	    if (min_timeout_pos > timeout_pos) {
-		min_timeout_pos = timeout_pos;
-		if (min_timeout_pos <= slot_timeout_pos)
-		    goto found_next;
-	    }
-	    p = p->next;
+	if (p) {
+	    ErtsTWheelTimer *end = p;
+
+	    do  {
+		ErtsMonotonicTime timeout_pos;
+		timeout_pos = p->timeout_pos;
+		if (min_timeout_pos > timeout_pos) {
+		    true_min_timeout = 1;
+		    min_timeout_pos = timeout_pos;
+		    if (min_timeout_pos <= slot_timeout_pos)
+			goto found_next;
+		}
+		p = p->next;
+	    } while (p != end);
 	}
 
 	tiw_pos_ix++;
-	if (tiw_pos_ix == TIW_SIZE)
+	if (tiw_pos_ix == ERTS_TIW_SIZE)
 	    tiw_pos_ix = 0;
     } while (start_ix != tiw_pos_ix);
 
 found_next:
 
     min_timeout = ERTS_CLKTCKS_TO_MONOTONIC(min_timeout_pos);
-    if (min_timeout != tiw->next_timeout_time)
-	set_next_timeout(tiw, min_timeout, true_min_timeout);
+    tiw->next_timeout_time = min_timeout;
+    tiw->true_next_timeout_time = true_min_timeout;
 
     return min_timeout;
 }
 
-static void
-remove_timer(ErtsTimerWheel *tiw, ErlTimer *p)
+static ERTS_INLINE void
+insert_timer_into_slot(ErtsTimerWheel *tiw, int slot, ErtsTWheelTimer *p)
 {
-    if (p->slot >= 0) {
-	/* Timer in wheel... */
-	ASSERT(p->slot < TIW_SIZE);
-	if (p->prev)
-	    p->prev->next = p->next;
-	else {
-	    ASSERT(tiw->w[p->slot] == p);
-	    tiw->w[p->slot] = p->next;
+    ERTS_TW_ASSERT(slot >= 0);
+    ERTS_TW_ASSERT(slot < ERTS_TIW_SIZE);
+    p->slot = slot;
+    if (!tiw->w[slot]) {
+	tiw->w[slot] = p;
+	p->next = p;
+	p->prev = p;
+    }
+    else {
+	ErtsTWheelTimer *next, *prev;
+	next = tiw->w[slot];
+	prev = next->prev;
+	p->next = next;
+	p->prev = prev;
+	prev->next = p;
+	next->prev = p;
+    }
+}
+
+static ERTS_INLINE void
+remove_timer(ErtsTimerWheel *tiw, ErtsTWheelTimer *p)
+{
+    int slot = p->slot;
+    ERTS_TW_ASSERT(slot != ERTS_TWHEEL_SLOT_INACTIVE);
+
+    if (slot >= 0) {
+	/*
+	 * Timer in wheel or in circular
+	 * list of timers currently beeing
+	 * triggered (referred by sentinel).
+	 */
+	ERTS_TW_ASSERT(slot < ERTS_TIW_SIZE);
+
+	if (p->next == p) {
+	    ERTS_TW_ASSERT(tiw->w[slot] == p);
+	    tiw->w[slot] = NULL;
 	}
-	if(p->next)
+	else {
+	    if (tiw->w[slot] == p)
+		tiw->w[slot] = p->next;
+	    p->prev->next = p->next;
 	    p->next->prev = p->prev;
+	}
     }
     else {
 	/* Timer in "at once" queue... */
-	ASSERT(p->slot == -1);
+	ERTS_TW_ASSERT(slot == ERTS_TWHEEL_SLOT_AT_ONCE);
 	if (p->prev)
 	    p->prev->next = p->next;
 	else {
-	    ASSERT(tiw->at_once.head == p);
+	    ERTS_TW_ASSERT(tiw->at_once.head == p);
 	    tiw->at_once.head = p->next;
 	}
 	if (p->next)
 	    p->next->prev = p->prev;
 	else {
-	    ASSERT(tiw->at_once.tail == p);
+	    ERTS_TW_ASSERT(tiw->at_once.tail == p);
 	    tiw->at_once.tail = p->prev;
 	}
-	ASSERT(tiw->at_once.nto > 0);
+	ERTS_TW_ASSERT(tiw->at_once.nto > 0);
 	tiw->at_once.nto--;
     }
 
-    p->slot = -2;
+    p->slot = ERTS_TWHEEL_SLOT_INACTIVE;
 
+#if 0
     p->next = NULL;
     p->prev = NULL;
+#endif
 
-    set_timer_wheel(p, NULL);
     tiw->nto--;
 }
 
 ErtsMonotonicTime
-erts_check_next_timeout_time(ErtsTimerWheel *tiw,
-			     ErtsMonotonicTime max_search_time)
+erts_check_next_timeout_time(ErtsSchedulerData *esdp)
 {
-    ErtsMonotonicTime next, curr;
-
-    curr = erts_get_monotonic_time();
-
-    erts_smp_mtx_lock(&tiw->lock);
-
-    next = find_next_timeout(tiw, curr, max_search_time);
-
-    erts_smp_mtx_unlock(&tiw->lock);
-
-    return next;
+    /*
+     * Called before a scheduler is about to wait. We wont
+     * check more than 10 minutes into the future.
+     */
+    return find_next_timeout(esdp->timer_wheel,
+			     erts_get_monotonic_time(esdp),
+			     ERTS_SEC_TO_MONOTONIC(10*60));
 }
 
-#ifndef DEBUG
+#ifndef ERTS_TW_DEBUG
 #define ERTS_DBG_CHK_SAFE_TO_SKIP_TO(TIW, TO) ((void) 0)
 #else
 #define ERTS_DBG_CHK_SAFE_TO_SKIP_TO(TIW, TO) debug_check_safe_to_skip_to((TIW), (TO))
@@ -300,198 +297,255 @@ static void
 debug_check_safe_to_skip_to(ErtsTimerWheel *tiw, ErtsMonotonicTime skip_to_pos)
 {
     int slots, ix;
-    ErlTimer *tmr;
+    ErtsTWheelTimer *tmr;
     ErtsMonotonicTime tmp;
 
-    ix = (int) (tiw->pos & (TIW_SIZE-1));
+    ix = (int) (tiw->pos & (ERTS_TIW_SIZE-1));
     tmp = skip_to_pos - tiw->pos;
-    ASSERT(tmp >= 0);
-    if (tmp < (ErtsMonotonicTime) TIW_SIZE)
+    ERTS_TW_ASSERT(tmp >= 0);
+    if (tmp < (ErtsMonotonicTime) ERTS_TIW_SIZE)
 	slots = (int) tmp;
     else
-	slots = TIW_SIZE;
+	slots = ERTS_TIW_SIZE;
 
      while (slots > 0) {
-	tmr = tiw->w[ix];
-	while (tmr) {
-	    ASSERT(tmr->timeout_pos > skip_to_pos);
-	    tmr = tmr->next;
-	}
-	ix++;
-	if (ix == TIW_SIZE)
-	    ix = 0;
-	slots--;
+	 tmr = tiw->w[ix];
+	 if (tmr) {
+	     ErtsTWheelTimer *end = tmr;
+	     do {
+		 ERTS_TW_ASSERT(tmr->timeout_pos > skip_to_pos);
+		 tmr = tmr->next;
+	     } while (tmr != end);
+	 }
+	 ix++;
+	 if (ix == ERTS_TIW_SIZE)
+	     ix = 0;
+	 slots--;
     }
 }
 #endif
 
+static ERTS_INLINE void
+timeout_timer(ErtsTWheelTimer *p)
+{
+    ErlTimeoutProc timeout;
+    void *arg;
+#if 0
+    p->next = NULL;
+    p->prev = NULL;
+#endif
+    p->slot = ERTS_TWHEEL_SLOT_INACTIVE;
+    timeout = p->u.func.timeout;
+    arg = p->u.func.arg;
+    (*timeout)(arg);
+}
+
 void
 erts_bump_timers(ErtsTimerWheel *tiw, ErtsMonotonicTime curr_time)
 {
-    int tiw_pos_ix, slots;
-    ErlTimer *p, *timeout_head, **timeout_tail;
-    ErtsMonotonicTime bump_to, tmp_slots;
-
-    if (erts_smp_atomic32_cmpxchg_nob(&tiw->is_bumping, 1, 0) != 0)
-	return; /* Another thread is currently bumping... */
+    int tiw_pos_ix, slots, yielded_slot_restarted, yield_count;
+    ErtsMonotonicTime bump_to, tmp_slots, old_pos;
 
-    bump_to = ERTS_MONOTONIC_TO_CLKTCKS(curr_time);
+    yield_count = ERTS_TWHEEL_BUMP_YIELD_LIMIT;
 
-    erts_smp_mtx_lock(&tiw->lock);
+    /*
+     * In order to be fair we always continue with work
+     * where we left off when restarting after a yield.
+     */
 
-    if (tiw->pos >= bump_to) {
-	timeout_head = NULL;
-	goto done;
+    if (tiw->yield_slot >= 0) {
+	yielded_slot_restarted = 1;
+	tiw_pos_ix = tiw->yield_slot;
+	slots = tiw->yield_slots_left;
+	bump_to = tiw->pos;
+	old_pos = tiw->yield_start_pos;
+	goto restart_yielded_slot;
     }
 
-    /* Don't want others here while we are bumping... */
-    set_next_timeout(tiw, curr_time + ERTS_MONOTONIC_DAY, 0);
+    do {
+
+	yielded_slot_restarted = 0;
+
+	bump_to = ERTS_MONOTONIC_TO_CLKTCKS(curr_time);
 
-    if (!tiw->at_once.head) {
-	timeout_head = NULL;
-	timeout_tail = &timeout_head;
-    }
-    else {
-	ASSERT(tiw->nto >= tiw->at_once.nto);
-	p = timeout_head = tiw->at_once.head;
 	while (1) {
-	    set_timer_wheel(p, NULL);
-	    if (!p->next) {
-		timeout_tail = &p->next;
-		break;
+	    ErtsTWheelTimer *p;
+
+	    old_pos = tiw->pos;
+
+	    if (tiw->nto == 0) {
+	    empty_wheel:
+		ERTS_DBG_CHK_SAFE_TO_SKIP_TO(tiw, bump_to);
+		tiw->true_next_timeout_time = 0;
+		tiw->next_timeout_time = curr_time + ERTS_MONOTONIC_DAY;
+		tiw->pos = bump_to;
+		tiw->yield_slot = ERTS_TWHEEL_SLOT_INACTIVE;
+		return;
 	    }
-	}
-	tiw->nto -= tiw->at_once.nto;
-	tiw->at_once.head = NULL;
-	tiw->at_once.tail = NULL;
-	tiw->at_once.nto = 0;
-    }
 
-    if (tiw->nto == 0) {
-	ERTS_DBG_CHK_SAFE_TO_SKIP_TO(tiw, bump_to);
-	tiw->pos = bump_to;
-	goto done;
-    }
+	    p = tiw->at_once.head;
+	    while (p) {
+		if (--yield_count <= 0) {
+		    ERTS_TW_ASSERT(tiw->nto > 0);
+		    ERTS_TW_ASSERT(tiw->at_once.nto > 0);
+		    tiw->yield_slot = ERTS_TWHEEL_SLOT_AT_ONCE;
+		    tiw->true_next_timeout_time = 1;
+		    tiw->next_timeout_time = ERTS_CLKTCKS_TO_MONOTONIC(old_pos);
+		    return;
+		}
+
+		ERTS_TW_ASSERT(tiw->nto > 0);
+		ERTS_TW_ASSERT(tiw->at_once.nto > 0);
+		tiw->nto--;
+		tiw->at_once.nto--;
+		tiw->at_once.head = p->next;
+		if (p->next)
+		    p->next->prev = NULL;
+		else
+		    tiw->at_once.tail = NULL;
+
+		timeout_timer(p);
+
+		p = tiw->at_once.head;
+	    }
 
-    if (tiw->true_next_timeout_time) {
-	ErtsMonotonicTime skip_until_pos;
-	/*
-	 * No need inspecting slots where we know no timeouts
-	 * to trigger should reside.
-	 */
+	    if (tiw->pos >= bump_to)
+		break;
 
-	skip_until_pos = ERTS_MONOTONIC_TO_CLKTCKS(tiw->next_timeout_time);
-	if (skip_until_pos > bump_to)
-	    skip_until_pos = bump_to;
+	    if (tiw->nto == 0)
+		goto empty_wheel;
 
-	ERTS_DBG_CHK_SAFE_TO_SKIP_TO(tiw, skip_until_pos);
-	ASSERT(skip_until_pos > tiw->pos);
+	    if (tiw->true_next_timeout_time) {
+		ErtsMonotonicTime skip_until_pos;
+		/*
+		 * No need inspecting slots where we know no timeouts
+		 * to trigger should reside.
+		 */
 
-	tiw->pos = skip_until_pos - 1;
-    }
+		skip_until_pos = ERTS_MONOTONIC_TO_CLKTCKS(tiw->next_timeout_time);
+		if (skip_until_pos > bump_to)
+		    skip_until_pos = bump_to;
 
-    tiw_pos_ix = (int) ((tiw->pos+1) & (TIW_SIZE-1));
-    tmp_slots = (bump_to - tiw->pos);
-    if (tmp_slots < (ErtsMonotonicTime) TIW_SIZE)
-	slots = (int) tmp_slots;
-    else
-	slots = TIW_SIZE;
- 
-    while (slots > 0) {
-	p = tiw->w[tiw_pos_ix];
-	while (p) {
-	    ErlTimer *next = p->next;
-	    ASSERT(p != next);
-	    if (p->timeout_pos <= bump_to) { /* we have a timeout */
-		/* Remove from list */
-		remove_timer(tiw, p);
-		*timeout_tail = p;	/* Insert in timeout queue */
-		timeout_tail = &p->next;
+		skip_until_pos--;
+
+		if (skip_until_pos > tiw->pos) {
+		    ERTS_DBG_CHK_SAFE_TO_SKIP_TO(tiw, skip_until_pos);
+
+		    tiw->pos = skip_until_pos;
+		}
+	    }
+
+	    tiw_pos_ix = (int) ((tiw->pos+1) & (ERTS_TIW_SIZE-1));
+	    tmp_slots = (bump_to - tiw->pos);
+	    if (tmp_slots < (ErtsMonotonicTime) ERTS_TIW_SIZE)
+		slots = (int) tmp_slots;
+	    else
+		slots = ERTS_TIW_SIZE;
+
+	    tiw->pos = bump_to;
+
+	    while (slots > 0) {
+
+		p = tiw->w[tiw_pos_ix];
+		if (p) {
+		    if (p->next == p) {
+			ERTS_TW_ASSERT(tiw->sentinel.next == &tiw->sentinel);
+			ERTS_TW_ASSERT(tiw->sentinel.prev == &tiw->sentinel);
+		    }
+		    else {
+			tiw->sentinel.next = p->next;
+			tiw->sentinel.prev = p->prev;
+			tiw->sentinel.next->prev = &tiw->sentinel;
+			tiw->sentinel.prev->next = &tiw->sentinel;
+		    }
+		    tiw->w[tiw_pos_ix] = NULL;
+
+		    while (1) {
+
+			if (p->timeout_pos > bump_to) {
+			    /* Very unusual case... */
+			    ++yield_count;
+			    insert_timer_into_slot(tiw, tiw_pos_ix, p);
+			}
+			else {
+			    /* Normal case... */
+			    timeout_timer(p);
+			    tiw->nto--;
+			}
+
+		    restart_yielded_slot:
+
+			p = tiw->sentinel.next;
+			if (p == &tiw->sentinel) {
+			    ERTS_TW_ASSERT(tiw->sentinel.prev == &tiw->sentinel);
+			    break;
+			}
+
+			if (--yield_count <= 0) {
+			    tiw->true_next_timeout_time = 1;
+			    tiw->next_timeout_time = ERTS_CLKTCKS_TO_MONOTONIC(old_pos);
+			    tiw->yield_slot = tiw_pos_ix;
+			    tiw->yield_slots_left = slots;
+			    tiw->yield_start_pos = old_pos;
+			    return; /* Yield! */
+			}
+
+			tiw->sentinel.next = p->next;
+			p->next->prev = &tiw->sentinel;
+		    }
+		}
+		tiw_pos_ix++;
+		if (tiw_pos_ix == ERTS_TIW_SIZE)
+		    tiw_pos_ix = 0;
+		slots--;
 	    }
-	    p = next;
 	}
-	tiw_pos_ix++;
-	if (tiw_pos_ix == TIW_SIZE)
-	    tiw_pos_ix = 0;
-	slots--;
-    }
 
-    ASSERT(tmp_slots >= (ErtsMonotonicTime) TIW_SIZE
-	   || tiw_pos_ix == (int) ((bump_to+1) & (TIW_SIZE-1)));
+    } while (yielded_slot_restarted);
 
-    tiw->pos = bump_to;
+    tiw->yield_slot = ERTS_TWHEEL_SLOT_INACTIVE;
+    tiw->true_next_timeout_time = 0;
+    tiw->next_timeout_time = curr_time + ERTS_MONOTONIC_DAY;
 
     /* Search at most two seconds ahead... */
     (void) find_next_timeout(tiw, curr_time, ERTS_SEC_TO_MONOTONIC(2));
-
-done:
-
-    erts_smp_mtx_unlock(&tiw->lock);
-    
-    erts_smp_atomic32_set_nob(&tiw->is_bumping, 0);
-
-    /* Call timedout timers callbacks */
-    while (timeout_head) {
-	ErlTimeoutProc timeout;
-	void *arg;
-	p = timeout_head;
-	timeout_head = p->next;
-	/* Here comes hairy use of the timer fields!
-	 * They are reset without having the lock.
-	 * It is assumed that no code but this will
-	 * accesses any field until the ->timeout
-	 * callback is called.
-	 */
-	ASSERT(p->timeout_pos <= bump_to);
-	p->next = NULL;
-	p->prev = NULL;
-	p->slot = 0;
-	timeout = p->timeout;
-	arg = p->arg;
-	(*timeout)(arg);
-    }
 }
 
 Uint
 erts_timer_wheel_memory_size(void)
 {
-#ifdef ERTS_SMP
-    return sizeof(ErtsTimerWheel)*(1 + erts_no_schedulers);
-#else
-    return sizeof(ErtsTimerWheel);
-#endif
+    return sizeof(ErtsTimerWheel)*erts_no_schedulers;
 }
 
 ErtsTimerWheel *
-erts_create_timer_wheel(int no)
+erts_create_timer_wheel(ErtsSchedulerData *esdp)
 {
     ErtsMonotonicTime mtime;
     int i;
     ErtsTimerWheel *tiw;
-    tiw = (ErtsTimerWheel *) erts_alloc(ERTS_ALC_T_TIMER_WHEEL,
-					sizeof(ErtsTimerWheel));
-    for(i = 0; i < TIW_SIZE; i++)
+    tiw = erts_alloc_permanent_cache_aligned(ERTS_ALC_T_TIMER_WHEEL,
+					     sizeof(ErtsTimerWheel));
+    for(i = 0; i < ERTS_TIW_SIZE; i++)
 	tiw->w[i] = NULL;
 
-    erts_smp_atomic32_init_nob(&tiw->is_bumping, 0);
-    erts_smp_mtx_init_x(&tiw->lock, "timer_wheel", make_small(no));
-
-    mtime = erts_get_monotonic_time();
+    mtime = erts_get_monotonic_time(esdp);
     tiw->pos = ERTS_MONOTONIC_TO_CLKTCKS(mtime);
     tiw->nto = 0;
     tiw->at_once.head = NULL;
     tiw->at_once.tail = NULL;
     tiw->at_once.nto = 0;
+    tiw->yield_slot = ERTS_TWHEEL_SLOT_INACTIVE;
     tiw->true_next_timeout_time = 0;
     tiw->next_timeout_time = mtime + ERTS_MONOTONIC_DAY;
-    init_next_timeout(tiw, mtime + ERTS_MONOTONIC_DAY);
+    tiw->sentinel.next = &tiw->sentinel;
+    tiw->sentinel.prev = &tiw->sentinel;
     return tiw;
 }
 
 ErtsNextTimeoutRef
 erts_get_next_timeout_reference(ErtsTimerWheel *tiw)
 {
-    return (ErtsNextTimeoutRef) &tiw->next_timeout;
+    return (ErtsNextTimeoutRef) &tiw->next_timeout_time;
 }
 
 
@@ -512,155 +566,83 @@ erts_init_time(int time_correction, ErtsTimeWarpMode time_warp_mode)
 #else
     tiw_itime = itime;
 #endif
-
-    erts_default_timer_wheel = erts_create_timer_wheel(0);
 }
 
 void
-erts_set_timer(ErlTimer *p, ErlTimeoutProc timeout,
-	       ErlCancelProc cancel, void *arg, Uint to)
+erts_twheel_set_timer(ErtsTimerWheel *tiw,
+		      ErtsTWheelTimer *p, ErlTimeoutProc timeout,
+		      ErlCancelProc cancel, void *arg,
+		      ErtsMonotonicTime timeout_pos)
 {
-    ErtsMonotonicTime timeout_time, timeout_pos;
-    ErtsMonotonicTime curr_time;
-    ErtsTimerWheel *tiw;
-    ErtsSchedulerData *esdp;
-
-    curr_time = erts_get_monotonic_time();
-    esdp = erts_get_scheduler_data();
-    if (esdp)
-	tiw = esdp->timer_wheel;
-    else
-	tiw = erts_default_timer_wheel;
+    ErtsMonotonicTime timeout_time;
 
-    erts_smp_mtx_lock(&tiw->lock);
+    p->u.func.timeout = timeout;
+    p->u.func.cancel = cancel;
+    p->u.func.arg = arg;
 
-    if (get_timer_wheel(p))
-	ERTS_INTERNAL_ERROR("Double set timer");
+    ERTS_TW_ASSERT(p->slot == ERTS_TWHEEL_SLOT_INACTIVE);
 
-    p->timeout = timeout;
-    p->cancel = cancel;
-    p->arg = arg;
-
-    if (to == 0) {
-	timeout_pos = ERTS_MONOTONIC_TO_CLKTCKS(curr_time);
+    if (timeout_pos <= tiw->pos) {
 	tiw->nto++;
 	tiw->at_once.nto++;
 	p->next = NULL;
 	p->prev = tiw->at_once.tail;
-	tiw->at_once.tail = p;
-	if (!tiw->at_once.head)
+	if (tiw->at_once.tail) {
+	    ERTS_TW_ASSERT(tiw->at_once.head);
+	    tiw->at_once.tail->next = p;
+	}
+	else {	
+	    ERTS_TW_ASSERT(!tiw->at_once.head);
 	    tiw->at_once.head = p;
-	p->timeout_pos = timeout_pos;
-	p->slot = -1;
-	timeout_time = ERTS_CLKTCKS_TO_MONOTONIC(timeout_pos);
+	}
+	tiw->at_once.tail = p;
+	p->timeout_pos = tiw->pos;
+	p->slot = ERTS_TWHEEL_SLOT_AT_ONCE;
+	timeout_time = ERTS_CLKTCKS_TO_MONOTONIC(tiw->pos);
     }
     else {
-	int tm;
-	ErtsMonotonicTime ticks;
-
-	ticks = ERTS_MSEC_TO_CLKTCKS(to);
-	timeout_pos = ERTS_MONOTONIC_TO_CLKTCKS(curr_time - 1) + 1 + ticks;
+	int slot;
 
 	/* calculate slot */
-	tm = (int) (timeout_pos & (TIW_SIZE-1));
-	p->slot = tm;
-  
-	/* insert at head of list at slot */
-	p->next = tiw->w[tm];
-	p->prev = NULL;
-	if (p->next != NULL)
-	    p->next->prev = p;
-	tiw->w[tm] = p;
+	slot = (int) (timeout_pos & (ERTS_TIW_SIZE-1));
+
+	insert_timer_into_slot(tiw, slot, p);
 
 	tiw->nto++;
 
 	timeout_time = ERTS_CLKTCKS_TO_MONOTONIC(timeout_pos);
 	p->timeout_pos = timeout_pos;
-
-	ASSERT(ERTS_MSEC_TO_MONOTONIC(to) <= timeout_time - curr_time);
-	ASSERT(timeout_time - curr_time
-	       < ERTS_MSEC_TO_MONOTONIC(to) + ERTS_CLKTCKS_TO_MONOTONIC(1));
     }
 
-    if (timeout_time < tiw->next_timeout_time)
-	set_next_timeout(tiw, timeout_time, 1);
-
-    set_timer_wheel(p, tiw);
-
-    erts_smp_mtx_unlock(&tiw->lock);
-
-#if defined(ERTS_SMP)
-    if (tiw == erts_default_timer_wheel)
-	erts_interupt_aux_thread_timed(timeout_time);
-#endif
-
+    if (timeout_time < tiw->next_timeout_time) {
+	tiw->true_next_timeout_time = 1;
+	tiw->next_timeout_time = timeout_time;
+    }
 }
 
 void
-erts_cancel_timer(ErlTimer *p)
+erts_twheel_cancel_timer(ErtsTimerWheel *tiw, ErtsTWheelTimer *p)
 {
-    ErtsTimerWheel *tiw;
-    ErlCancelProc cancel;
-    void *arg = NULL; /* Shut up faulty warning... */
-
-    tiw = get_timer_wheel(p);
-    if (!tiw)
-	return;
-    
-    erts_smp_mtx_lock(&tiw->lock);
-    if (tiw != get_timer_wheel(p))
-	cancel = NULL;
-    else {
+    if (p->slot != ERTS_TWHEEL_SLOT_INACTIVE) {
+	ErlCancelProc cancel;
+	void *arg;
 	remove_timer(tiw, p);
-
-	cancel = p->cancel;
-	arg = p->arg;
+	cancel = p->u.func.cancel;
+	arg = p->u.func.arg;
+	if (cancel)
+	    (*cancel)(arg);
     }
-    erts_smp_mtx_unlock(&tiw->lock);
-
-    if (cancel)
-	(*cancel)(arg);
 }
 
-/*
-  Returns the amount of time left in ms until the timer 'p' is triggered.
-  0 is returned if 'p' isn't active.
-  0 is returned also if the timer is overdue (i.e., would have triggered
-  immediately if it hadn't been cancelled).
-*/
-Uint
-erts_time_left(ErlTimer *p)
-{
-    ErtsTimerWheel *tiw;
-    ErtsMonotonicTime current_time, timeout_time;
-
-    tiw = get_timer_wheel(p);
-    if (!tiw)
-	return 0;
-
-    erts_smp_mtx_lock(&tiw->lock);
-    if (tiw != get_timer_wheel(p))
-	timeout_time = ERTS_MONOTONIC_TIME_MIN;
-    else
-	timeout_time = ERTS_CLKTCKS_TO_MONOTONIC(p->timeout_pos);
-    erts_smp_mtx_unlock(&tiw->lock);
-
-    current_time = erts_get_monotonic_time();
-    if (timeout_time <= current_time)
-	return 0;
-    return (Uint) ERTS_MONOTONIC_TO_MSEC(timeout_time - current_time);
-}
-
-#ifdef DEBUG
+#ifdef ERTS_TW_DEBUG
 void erts_p_slpq(void)
 {
-    ErtsTimerWheel *tiw = erts_default_timer_wheel;
-    ErtsMonotonicTime current_time = erts_get_monotonic_time();
+    erts_printf("Not yet implemented...\n");
+#if 0
+    ErtsMonotonicTime current_time = erts_get_monotonic_time(NULL);
     int i;
-    ErlTimer* p;
+    ErtsTWheelTimer* p;
   
-    erts_smp_mtx_lock(&tiw->lock);
-
     /* print the whole wheel, starting at the current position */
     erts_printf("\ncurrent time = %bps tiw_pos = %d tiw_nto %d\n",
 		current_time, tiw->pos, tiw->nto);
@@ -673,7 +655,7 @@ void erts_p_slpq(void)
 			p->slot);
 	}
     }
-    for(i = ((i+1) & (TIW_SIZE-1)); i != (tiw->pos & (TIW_SIZE-1)); i = ((i+1) & (TIW_SIZE-1))) {
+    for(i = ((i+1) & (ERTS_TIW_SIZE-1)); i != (tiw->pos & (ERTS_TIW_SIZE-1)); i = ((i+1) & (ERTS_TIW_SIZE-1))) {
 	if (tiw->w[i] != NULL) {
 	    erts_printf("%d:\n", i);
 	    for(p = tiw->w[i]; p != NULL; p = p->next) {
@@ -682,7 +664,6 @@ void erts_p_slpq(void)
 	    }
 	}
     }
-
-    erts_smp_mtx_unlock(&tiw->lock);
+#endif
 }
-#endif /* DEBUG */
+#endif /* ERTS_TW_DEBUG */
diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c
index 51de3528be..e21e916de5 100644
--- a/erts/emulator/beam/utils.c
+++ b/erts/emulator/beam/utils.c
@@ -50,6 +50,8 @@
 #include "erl_ptab.h"
 #include "erl_check_io.h"
 #include "erl_bif_unique.h"
+#define ERTS_WANT_TIMER_WHEEL_API
+#include "erl_time.h"
 #ifdef HIPE
 #  include "hipe_mode_switch.h"
 #endif
@@ -4441,145 +4443,6 @@ is_string(Eterm list)
     return 0;
 }
 
-#ifdef ERTS_SMP
-
-/*
- * Process and Port timers in smp case
- */
-
-ERTS_SCHED_PREF_PRE_ALLOC_IMPL(ptimer_pre, ErtsSmpPTimer, 1000)
-
-#define ERTS_PTMR_FLGS_ALLCD_SIZE \
-  2
-#define ERTS_PTMR_FLGS_ALLCD_MASK \
-  ((((Uint32) 1) << ERTS_PTMR_FLGS_ALLCD_SIZE) - 1)
-
-#define ERTS_PTMR_FLGS_PREALLCD	((Uint32) 1)
-#define ERTS_PTMR_FLGS_SLALLCD	((Uint32) 2)
-#define ERTS_PTMR_FLGS_LLALLCD	((Uint32) 3)
-#define ERTS_PTMR_FLG_CANCELLED	(((Uint32) 1) << (ERTS_PTMR_FLGS_ALLCD_SIZE+0))
-
-static void
-init_ptimers(void)
-{
-    init_ptimer_pre_alloc();
-}
-
-static ERTS_INLINE void
-free_ptimer(ErtsSmpPTimer *ptimer)
-{
-    switch (ptimer->timer.flags & ERTS_PTMR_FLGS_ALLCD_MASK) {
-    case ERTS_PTMR_FLGS_PREALLCD:
-	(void) ptimer_pre_free(ptimer);
-	break;
-    case ERTS_PTMR_FLGS_SLALLCD:
-	erts_free(ERTS_ALC_T_SL_PTIMER, (void *) ptimer);
-	break;
-    case ERTS_PTMR_FLGS_LLALLCD:
-	erts_free(ERTS_ALC_T_LL_PTIMER, (void *) ptimer);
-	break;
-    default:
-	erl_exit(ERTS_ABORT_EXIT,
-		 "Internal error: Bad ptimer alloc type\n");
-	break;
-    }
-}
-
-/* Callback for process timeout cancelled */
-static void
-ptimer_cancelled(ErtsSmpPTimer *ptimer)
-{
-    free_ptimer(ptimer);
-}
-
-/* Callback for process timeout */
-static void
-ptimer_timeout(ErtsSmpPTimer *ptimer)
-{
-    if (is_internal_pid(ptimer->timer.id)) {
-	Process *p;
-	p = erts_pid2proc_opt(NULL,
-			      0,
-			      ptimer->timer.id,
-			      ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS,
-			      ERTS_P2P_FLG_ALLOW_OTHER_X);
-	if (p) {
-	    if (!ERTS_PROC_IS_EXITING(p)
-		&& !(ptimer->timer.flags & ERTS_PTMR_FLG_CANCELLED)) {
-		ASSERT(*ptimer->timer.timer_ref == ptimer);
-		*ptimer->timer.timer_ref = NULL;
-		(*ptimer->timer.timeout_func)(p);
-	    }
-	    erts_smp_proc_unlock(p, ERTS_PROC_LOCK_MAIN|ERTS_PROC_LOCK_STATUS);
-	}
-    }
-    else {
-	Port *p;
-	ASSERT(is_internal_port(ptimer->timer.id));
-	p = erts_id2port_sflgs(ptimer->timer.id,
-			       NULL,
-			       0,
-			       ERTS_PORT_SFLGS_DEAD);
-	if (p) {
-	    if (!(ptimer->timer.flags & ERTS_PTMR_FLG_CANCELLED)) {
-		ASSERT(*ptimer->timer.timer_ref == ptimer);
-		*ptimer->timer.timer_ref = NULL;
-		(*ptimer->timer.timeout_func)(p);
-	    }
-	    erts_port_release(p);
-	}
-    }
-    free_ptimer(ptimer);
-}
-
-void
-erts_create_smp_ptimer(ErtsSmpPTimer **timer_ref,
-		       Eterm id,
-		       ErlTimeoutProc timeout_func,
-		       Uint timeout)
-{
-    ErtsSmpPTimer *res = ptimer_pre_alloc();
-    if (res)
-	res->timer.flags = ERTS_PTMR_FLGS_PREALLCD;
-    else {
-	if (timeout < ERTS_ALC_MIN_LONG_LIVED_TIME) {
-	    res = erts_alloc(ERTS_ALC_T_SL_PTIMER, sizeof(ErtsSmpPTimer));
-	    res->timer.flags = ERTS_PTMR_FLGS_SLALLCD;
-	}
-	else {
-	    res = erts_alloc(ERTS_ALC_T_LL_PTIMER, sizeof(ErtsSmpPTimer));
-	    res->timer.flags = ERTS_PTMR_FLGS_LLALLCD;
-	}
-    }
-    res->timer.timeout_func = timeout_func;
-    res->timer.timer_ref = timer_ref;
-    res->timer.id = id;
-    erts_init_timer(&res->timer.tm);
-
-    ASSERT(!*timer_ref);
-
-    *timer_ref = res;
-
-    erts_set_timer(&res->timer.tm,
-		  (ErlTimeoutProc) ptimer_timeout,
-		  (ErlCancelProc) ptimer_cancelled,
-		  (void*) res,
-		  timeout);
-}
-
-void
-erts_cancel_smp_ptimer(ErtsSmpPTimer *ptimer)
-{
-    if (ptimer) {
-	ASSERT(*ptimer->timer.timer_ref == ptimer);
-	*ptimer->timer.timer_ref = NULL;
-	ptimer->timer.flags |= ERTS_PTMR_FLG_CANCELLED;
-	erts_cancel_timer(&ptimer->timer.tm);
-    }
-}
-
-#endif
-
 static int trim_threshold;
 static int top_pad;
 static int mmap_threshold;
@@ -4589,9 +4452,7 @@ Uint tot_bin_allocated;
 
 void erts_init_utils(void)
 {
-#ifdef ERTS_SMP
-    init_ptimers();
-#endif
+
 }
 
 void erts_init_utils_mem(void) 
diff --git a/erts/emulator/hipe/hipe_debug.c b/erts/emulator/hipe/hipe_debug.c
index 61406b92af..2804d46249 100644
--- a/erts/emulator/hipe/hipe_debug.c
+++ b/erts/emulator/hipe/hipe_debug.c
@@ -213,9 +213,9 @@ void hipe_print_pcb(Process *p)
     U("seq..clock ", seq_trace_clock);
     U("seq..astcnt", seq_trace_lastcnt);
     U("seq..token ", seq_trace_token);
-    U("intial[0]  ", initial[0]);
-    U("intial[1]  ", initial[1]);
-    U("intial[2]  ", initial[2]);
+    U("intial[0]  ", u.initial[0]);
+    U("intial[1]  ", u.initial[1]);
+    U("intial[2]  ", u.initial[2]);
     P("current    ", current);
     P("cp         ", cp);
     P("i          ", i);
diff --git a/erts/emulator/hipe/hipe_native_bif.c b/erts/emulator/hipe/hipe_native_bif.c
index 7e8632b50d..85d945823e 100644
--- a/erts/emulator/hipe/hipe_native_bif.c
+++ b/erts/emulator/hipe/hipe_native_bif.c
@@ -102,7 +102,8 @@ BIF_RETTYPE hipe_set_timeout(BIF_ALIST_1)
      * p->def_arg_reg[0] and p->i are both defined and used.
      * If a message arrives, BEAM resumes at p->i.
      * If a timeout fires, BEAM resumes at p->def_arg_reg[0].
-     * (See set_timer() and timeout_proc() in erl_process.c.)
+     * (See erts_set_proc_timer() and proc_timeout_common() in
+     * erl_hl_timer.c.)
      *
      * Here we set p->def_arg_reg[0] to hipe_beam_pc_resume.
      * Assuming our caller invokes suspend immediately after
@@ -135,28 +136,21 @@ BIF_RETTYPE hipe_set_timeout(BIF_ALIST_1)
      */
     if (p->flags & (F_INSLPQUEUE | F_TIMO))
 	return NIL;	/* caller had better call nbif_suspend ASAP! */
-    if (is_small(timeout_value) && signed_val(timeout_value) >= 0 &&
-#if defined(ARCH_64)
-	(unsigned_val(timeout_value) >> 32) == 0
-#else
-	1
-#endif
-	) {
-	set_timer(p, unsigned_val(timeout_value));
-    } else if (timeout_value == am_infinity) {
+
+    if (timeout_value == am_infinity) {
 	/* p->flags |= F_TIMO; */	/* XXX: nbif_suspend_msg_timeout */
-#if !defined(ARCH_64)
-    } else if (term_to_Uint(timeout_value, &time_val)) {
-	set_timer(p, time_val);
-#endif
-    } else {
+    }
+    else {
+	int tres = erts_set_proc_timer_term(p, timeout_value);
+	if (tres != 0) { /* Wrong time */
 #ifdef ERTS_SMP
-	if (p->hipe_smp.have_receive_locks) {
-	    p->hipe_smp.have_receive_locks = 0;
-	    erts_smp_proc_unlock(p, ERTS_PROC_LOCKS_MSG_RECEIVE);
-	}
+	    if (p->hipe_smp.have_receive_locks) {
+		p->hipe_smp.have_receive_locks = 0;
+		erts_smp_proc_unlock(p, ERTS_PROC_LOCKS_MSG_RECEIVE);
+	    }
 #endif
-	BIF_ERROR(p, EXC_TIMEOUT_VALUE);
+	    BIF_ERROR(p, EXC_TIMEOUT_VALUE);
+	}
     }
     return NIL;	/* caller had better call nbif_suspend ASAP! */
 }
@@ -170,7 +164,7 @@ void hipe_select_msg(Process *p)
     msgp = PEEK_MESSAGE(p);
     UNLINK_MESSAGE(p, msgp);	/* decrements global 'erts_proc_tot_mem' variable */
     JOIN_MESSAGE(p);
-    CANCEL_TIMER(p);		/* calls erl_cancel_timer() */
+    CANCEL_TIMER(p);		/* calls erts_cancel_proc_timer() */
     free_message(msgp);
 }
 
diff --git a/erts/emulator/sys/common/erl_check_io.c b/erts/emulator/sys/common/erl_check_io.c
index 7be17d20bb..d1d6696090 100644
--- a/erts/emulator/sys/common/erl_check_io.c
+++ b/erts/emulator/sys/common/erl_check_io.c
@@ -38,6 +38,8 @@
 #include "erl_check_io.h"
 #include "erl_thr_progress.h"
 #include "dtrace-wrapper.h"
+#define ERTS_WANT_TIMER_WHEEL_API
+#include "erl_time.h"
 
 #ifdef ERTS_SYS_CONTINOUS_FD_NUMBERS
 #else
@@ -1616,8 +1618,7 @@ ERTS_CIO_EXPORT(erts_check_io)(int do_wait)
 
     /* Figure out timeout value */
     timeout_time = (do_wait
-		    ? erts_check_next_timeout_time(esdp->timer_wheel,
-						   ERTS_SEC_TO_MONOTONIC(10*60))
+		    ? erts_check_next_timeout_time(esdp)
 		    : ERTS_POLL_NO_TIMEOUT /* poll only */);
 
     /*
diff --git a/erts/emulator/sys/common/erl_poll.c b/erts/emulator/sys/common/erl_poll.c
index f4d4a85ca4..446cca77ff 100644
--- a/erts/emulator/sys/common/erl_poll.c
+++ b/erts/emulator/sys/common/erl_poll.c
@@ -2026,7 +2026,7 @@ get_timeout(ErtsPollSet ps,
     }
     else {
 	ErtsMonotonicTime diff_time, current_time;
-	current_time = erts_get_monotonic_time();
+	current_time = erts_get_monotonic_time(NULL);
 	diff_time = timeout_time - current_time;
 	if (diff_time <= 0) {
 	    save_timeout_time = ERTS_MONOTONIC_TIME_MIN;
diff --git a/erts/emulator/sys/ose/erl_poll.c b/erts/emulator/sys/ose/erl_poll.c
index 36ee2557e8..3d4ac0365f 100644
--- a/erts/emulator/sys/ose/erl_poll.c
+++ b/erts/emulator/sys/ose/erl_poll.c
@@ -506,7 +506,7 @@ int erts_poll_wait(ErtsPollSet ps,
     }
     else {
 	ErtsMonotonicTime current_time, diff_time;
-	current_time = erts_get_monotonic_time();
+	current_time = erts_get_monotonic_time(NULL);
 	diff_time = timeout_time - current_time;
 	if (diff_time <= 0)
 	    goto no_timeout;
diff --git a/erts/emulator/sys/win32/erl_poll.c b/erts/emulator/sys/win32/erl_poll.c
index 937a678702..9196561944 100644
--- a/erts/emulator/sys/win32/erl_poll.c
+++ b/erts/emulator/sys/win32/erl_poll.c
@@ -454,7 +454,7 @@ poll_wait_timeout(ErtsPollSet ps, ErtsMonotonicTime timeout_time)
 	return (DWORD) 0;
     }
 
-    current_time = erts_get_monotonic_time();
+    current_time = erts_get_monotonic_time(NULL);
     diff_time = timeout_time - current_time;
     if (diff_time <= 0)
 	goto no_timeout;
diff --git a/erts/preloaded/ebin/erlang.beam b/erts/preloaded/ebin/erlang.beam
index 3478a80dd4..ce61199567 100644
--- a/erts/preloaded/ebin/erlang.beam
+++ b/erts/preloaded/ebin/erlang.beam
diff --git a/erts/preloaded/ebin/erts_internal.beam b/erts/preloaded/ebin/erts_internal.beam
index 9ed45b34bf..bee3477b2e 100644
--- a/erts/preloaded/ebin/erts_internal.beam
+++ b/erts/preloaded/ebin/erts_internal.beam
diff --git a/erts/preloaded/ebin/init.beam b/erts/preloaded/ebin/init.beam
index 7361139cde..f3abb5c1c7 100644
--- a/erts/preloaded/ebin/init.beam
+++ b/erts/preloaded/ebin/init.beam
diff --git a/erts/preloaded/src/erlang.erl b/erts/preloaded/src/erlang.erl
index 629702fc1c..4ef4288ead 100644
--- a/erts/preloaded/src/erlang.erl
+++ b/erts/preloaded/src/erlang.erl
@@ -129,10 +129,11 @@
 -export([process_display/2]).
 -export([process_flag/3, process_info/1, processes/0, purge_module/1]).
 -export([put/2, raise/3, read_timer/1, read_timer/2, ref_to_list/1, register/2]).
--export([registered/0, resume_process/1, round/1, self/0, send_after/3]).
+-export([send_after/3, send_after/4, start_timer/3, start_timer/4]).
+-export([registered/0, resume_process/1, round/1, self/0]).
 -export([seq_trace/2, seq_trace_print/1, seq_trace_print/2, setnode/2]).
 -export([setnode/3, size/1, spawn/3, spawn_link/3, split_binary/2]).
--export([start_timer/3, suspend_process/2, system_monitor/0]).
+-export([suspend_process/2, system_monitor/0]).
 -export([system_monitor/1, system_monitor/2, system_profile/0]).
 -export([system_profile/2, throw/1, time/0, trace/3, trace_delivered/1]).
 -export([trace_info/2, trunc/1, tuple_size/1, universaltime/0]).
@@ -427,23 +428,9 @@ call_on_load_function(_P1) ->
 -spec erlang:cancel_timer(TimerRef) -> Time | false when
       TimerRef :: reference(),
       Time :: non_neg_integer().
-cancel_timer(TimerRef) ->
-    try
-	case erts_internal:access_bif_timer(TimerRef) of
-	    undefined ->
-		false;
-	    {BTR, TSrv} ->
-		Req = erlang:make_ref(),
-		TSrv ! {cancel_timeout, BTR, erlang:self(),
-			true, Req, TimerRef},
-		receive
-		    {cancel_timer, Req, Result} ->
-			Result
-		end
-	end
-    catch
-	_:_ -> erlang:error(badarg, [TimerRef])
-    end.
+
+cancel_timer(_TimerRef) ->
+    erlang:nif_error(undefined).
 
 %% cancel_timer/2
 -spec erlang:cancel_timer(TimerRef, Options) -> Time | false | ok when
@@ -451,53 +438,9 @@ cancel_timer(TimerRef) ->
       Option :: {async, boolean()} | {info, boolean()},
       Options :: [Option],
       Time :: non_neg_integer().
-cancel_timer(TimerRef, Options) ->
-    try
-	{Async, Info} = get_cancel_timer_options(Options, false, true),
-	case erts_internal:access_bif_timer(TimerRef) of
-	    undefined ->
-		case {Async, Info} of
-		    {true, true} ->
-			erlang:self() ! {cancel_timer, TimerRef, false}, ok;
-		    {false, true} ->
-			false;
-		    _ ->
-			ok
-		end;
-	    {BTR, TSrv} ->
-		case Async of
-		    true ->
-			TSrv ! {cancel_timeout, BTR, erlang:self(),
-				Info, TimerRef, TimerRef},
-			ok;
-		    false ->
-			Req = erlang:make_ref(),
-			TSrv ! {cancel_timeout, BTR, erlang:self(),
-				true, Req, TimerRef},
-			receive
-			    {cancel_timer, Req, Result} ->
-				case Info of
-				    true -> Result;
-				    false -> ok
-				end
-			end
-		end
-	end
-    catch
-	_:_ -> erlang:error(badarg, [TimerRef, Options])
-    end.
 
-get_cancel_timer_options([], Async, Info) ->
-    {Async, Info};
-get_cancel_timer_options([{async, Bool} | Opts],
-			 _Async, Info) when Bool == true;
-					    Bool == false ->
-    get_cancel_timer_options(Opts, Bool, Info);
-get_cancel_timer_options([{info, Bool} | Opts],
-			 Async, _Info) when Bool == true;
-					    Bool == false ->
-    get_cancel_timer_options(Opts, Async, Bool).
-    
+cancel_timer(_TimerRef, _Options) ->
+    erlang:nif_error(undefined).
 
 %% check_old_code/1
 -spec check_old_code(Module) -> boolean() when
@@ -1538,8 +1481,8 @@ raise(_Class, _Reason, _Stacktrace) ->
 -spec erlang:read_timer(TimerRef) -> non_neg_integer() | false when
       TimerRef :: reference().
 
-read_timer(TimerRef) ->
-    read_timer(TimerRef, []).
+read_timer(_TimerRef) ->
+    erlang:nif_error(undefined).
 
 %% read_timer/2
 -spec erlang:read_timer(TimerRef, Options) -> non_neg_integer() | false | ok when
@@ -1547,43 +1490,8 @@ read_timer(TimerRef) ->
       Option :: {async, boolean()},
       Options :: [Option].
 
-read_timer(TimerRef, Options) ->
-    try
-	Async = get_read_timer_options(Options, false),
-	case erts_internal:access_bif_timer(TimerRef) of
-	    undefined ->
-		case Async of
-		    true ->
-			erlang:self() ! {read_timer, TimerRef, false},
-			ok;
-		    false ->
-			false
-		end;
-	    {BTR, TSrv} ->
-		case Async of
-		    true ->
-			TSrv ! {read_timeout, BTR, erlang:self(),
-				TimerRef, TimerRef},
-			ok;
-		    false ->
-			Req = erlang:make_ref(),
-			TSrv ! {read_timeout, BTR, erlang:self(),
-				Req, TimerRef},
-			receive
-			    {read_timer, Req, Result} ->
-				Result
-			end
-		end
-	end
-    catch
-	_:_ -> erlang:error(badarg, [TimerRef])
-    end.
-
-get_read_timer_options([], Async) ->
-    Async;
-get_read_timer_options([{async, Bool} | Opts],
-		       _Async) when Bool == true; Bool == false ->
-    get_read_timer_options(Opts, Bool).
+read_timer(_TimerRef, _Options) ->
+    erlang:nif_error(undefined).
 
 %% ref_to_list/1
 -spec erlang:ref_to_list(Ref) -> string() when
@@ -1630,35 +1538,20 @@ self() ->
       Msg :: term(),
       TimerRef :: reference().
 
-send_after(0, Dest, Msg) ->
-    try
-	true = ((erlang:is_pid(Dest)
-		 andalso erlang:node(Dest) == erlang:node())
-		orelse (erlang:is_atom(Dest)
-			andalso Dest /= undefined)),
-	try Dest ! Msg catch _:_ -> ok end,
-	erlang:make_ref()
-    catch
-	_:_ ->
-	    erlang:error(badarg, [0, Dest, Msg])
-    end;
-send_after(Time, Dest, Msg) ->
-    Now = erlang:monotonic_time(),
-    try
-	true = ((erlang:is_pid(Dest)
-		 andalso erlang:node(Dest) == erlang:node())
-		orelse (erlang:is_atom(Dest)
-			andalso Dest /= undefined)),
-	true = Time > 0,
-	true = Time < (1 bsl 32), % Maybe lift this restriction...
-	TO = Now + (erts_internal:time_unit()*Time) div 1000,
-	{BTR, TSrv, TRef} = erts_internal:create_bif_timer(),
-	TSrv ! {set_timeout, BTR, Dest, TO, TRef, Msg},
-	TRef
-    catch
-	_:_ ->
-	    erlang:error(badarg, [Time, Dest, Msg])
-    end.
+send_after(_Time, _Dest, _Msg) ->
+    erlang:nif_error(undefined).
+
+%% send_after/4
+-spec erlang:send_after(Time, Dest, Msg, Options) -> TimerRef when
+      Time :: integer(),
+      Dest :: pid() | atom(),
+      Msg :: term(),
+      Options :: [Option],
+      Option :: {abs, boolean()} | {accessor, pid()},
+      TimerRef :: reference().
+
+send_after(_Time, _Dest, _Msg, _Options) ->
+    erlang:nif_error(undefined).
 
 %% seq_trace/2
 -spec erlang:seq_trace(P1, P2) -> seq_trace_info_returns() | {term(), term(), term(), term(), term()} when
@@ -1731,37 +1624,21 @@ split_binary(_Bin, _Pos) ->
       Dest :: pid() | atom(),
       Msg :: term(),
       TimerRef :: reference().
-start_timer(0, Dest, Msg) ->
-    try
-	true = ((erlang:is_pid(Dest)
-		 andalso erlang:node(Dest) == erlang:node())
-		orelse (erlang:is_atom(Dest)
-			andalso Dest /= undefined)),
-	TimerRef = erlang:make_ref(),
-	_ = try Dest ! {timeout, TimerRef, Msg} catch _:_ -> ok end,
-	TimerRef
-    catch
-	_:_ ->
-	    erlang:error(badarg, [0, Dest, Msg])
-    end;
-start_timer(Time, Dest, Msg) ->
-    Now = erlang:monotonic_time(),
-    try
-	true = ((erlang:is_pid(Dest)
-		 andalso erlang:node(Dest) == erlang:node())
-		orelse (erlang:is_atom(Dest)
-			andalso Dest /= undefined)),
-	true = Time > 0,
-	true = Time < (1 bsl 32), % Maybe lift this restriction...
-	TO = Now + (erts_internal:time_unit()*Time) div 1000,
-	{BTR, TSrv, TimerRef} = erts_internal:create_bif_timer(),
-	TSrv ! {set_timeout, BTR, Dest, TO, TimerRef,
-		{timeout, TimerRef, Msg}},
-	TimerRef
-    catch
-	_:_ ->
-	    erlang:error(badarg, [Time, Dest, Msg])
-    end.
+
+start_timer(_Time, _Dest, _Msg) ->
+    erlang:nif_error(undefined).
+
+%% start_timer/4
+-spec erlang:start_timer(Time, Dest, Msg, Options) -> TimerRef when
+      Time :: integer(),
+      Dest :: pid() | atom(),
+      Msg :: term(),
+      Options :: [Option],
+      Option :: {abs, boolean()} | {accessor, pid()},
+      TimerRef :: reference().
+
+start_timer(_Time, _Dest, _Msg, _Options) ->
+    erlang:nif_error(undefined).
 
 %% suspend_process/2
 -spec erlang:suspend_process(Suspendee, OptList) -> boolean() when
diff --git a/erts/preloaded/src/erts_internal.erl b/erts/preloaded/src/erts_internal.erl
index e489001532..65a1f1ed3a 100644
--- a/erts/preloaded/src/erts_internal.erl
+++ b/erts/preloaded/src/erts_internal.erl
@@ -40,13 +40,9 @@
 
 -export([flush_monitor_messages/3]).
 
--export([time_unit/0]).
-
--export([bif_timer_server/2]).
-
--export([get_bif_timer_servers/0, create_bif_timer/0, access_bif_timer/1]).
+-export([await_result/1]).
 
--export([monitor_process/2]).
+-export([time_unit/0]).
 
 -export([is_system_process/1]).
 
@@ -61,6 +57,16 @@ await_port_send_result(Ref, Busy, Ok) ->
     end.
 
 %%
+%% Await result...
+%%
+
+await_result(Ref) when is_reference(Ref) ->
+    receive
+	{Ref, Result} ->
+	    Result
+    end.
+
+%%
 %% Statically linked port NIFs
 %%
 
@@ -234,245 +240,8 @@ flush_monitor_messages(Ref, Multi, Res) when is_reference(Ref) ->
 time_unit() ->
     erlang:nif_error(undefined).
 
--spec erts_internal:get_bif_timer_servers() -> Pids when
-      Pid :: pid(),
-      Pids :: [Pid].
-
-get_bif_timer_servers() ->
-    erlang:nif_error(undefined).
-
--spec erts_internal:create_bif_timer() -> Res when
-      Res :: {reference(), pid(), reference()}.
-
-create_bif_timer() ->
-    erlang:nif_error(undefined).
-
--spec erts_internal:access_bif_timer(Ref) -> Res when
-      Ref :: reference(),
-      Res :: {reference(), pid()} | 'undefined'.
-
-access_bif_timer(_Ref) ->
-    erlang:nif_error(undefined).
-
--spec erts_internal:monitor_process(Pid, Ref) -> boolean() when
-      Pid :: pid(),
-      Ref :: reference().
-
-monitor_process(_Pid, _Ref) ->
-    erlang:nif_error(undefined).
-
 -spec erts_internal:is_system_process(Pid) -> boolean() when
       Pid :: pid().
 
 is_system_process(_Pid) ->
     erlang:nif_error(undefined).
-
-%%
-%% BIF timer servers
-%%
-
--record(tsrv_state, {rtab,
-		     ttab,
-		     btr,
-		     unit,
-		     next}).
-
-bif_timer_server(N, BTR) ->
-    try
-	tsrv_loop(tsrv_init_static_state(N, BTR), infinity)
-    catch
-	Type:Reason ->
-	    erlang:display({'BIF_timer_server',
-			    {Type, Reason},
-			    erlang:get_stacktrace()}),
-	    exit(Reason)
-    end.
-
-tsrv_init_static_state(N, BTR) ->
-    process_flag(trap_exit, true),
-    NList = integer_to_list(N),
-    RTabName = list_to_atom("BIF_timer_reference_table_" ++ NList),
-    TTabName = list_to_atom("BIF_timer_time_table_" ++ NList),
-    #tsrv_state{rtab = ets:new(RTabName,
-			       [set, private, {keypos, 2}]),
-		ttab = ets:new(TTabName,
-			       [ordered_set, private, {keypos, 1}]),
-		btr = BTR,
-		unit = erts_internal:time_unit(),
-		next = infinity}.
-    
-
-tsrv_loop(#tsrv_state{unit = Unit} = StaticState, Nxt) ->
-    CallTime = erlang:monotonic_time(),
-    %% 'infinity' is greater than all integers...
-    NewNxt = case CallTime >= Nxt of
-		 true ->
-		     tsrv_handle_timeout(CallTime, StaticState);
-		 false ->
-		     TMO = try
-			       (1000*(Nxt - CallTime - 1)) div Unit + 1
-			   catch
-			       error:badarith when Nxt == infinity -> infinity
-			   end,
-		     receive
-			 Msg ->
-			     tsrv_handle_msg(Msg, StaticState, Nxt)
-		     after TMO ->
-			     Nxt
-		     end
-	     end,
-    tsrv_loop(StaticState, NewNxt).
-
-tsrv_handle_msg({set_timeout, BTR, Proc, Time, TRef, Msg},
-		#tsrv_state{rtab = RTab,
-			    ttab = TTab,
-			    btr = BTR},
-		Nxt) when erlang:is_integer(Time) ->
-    RcvTime = erlang:monotonic_time(),
-    case Time =< RcvTime of
-	true ->
-	    try Proc ! Msg catch _:_ -> ok end,
-	    Nxt;
-	false ->
-	    Ins = case erlang:is_atom(Proc) of
-		      true ->
-			  true;
-		      false ->
-			  try
-			      erts_internal:monitor_process(Proc, TRef)
-			  catch
-			      _:_ -> false
-			  end
-		  end,
-	    case Ins of
-		false ->
-		    Nxt;
-		true ->
-		    TKey = {Time, TRef},
-		    true = ets:insert(RTab, TKey),
-		    true = ets:insert(TTab, {TKey, Proc, Msg}),
-		    case Time < Nxt of
-			true -> Time;
-			false -> Nxt
-		    end
-	    end
-    end;
-tsrv_handle_msg({cancel_timeout, BTR, From, Reply, Req, TRef},
-		#tsrv_state{rtab = RTab,
-			    ttab = TTab,
-			    unit = Unit,
-			    btr = BTR},
-		Nxt) ->
-    case ets:lookup(RTab, TRef) of
-	[] ->
-	    case Reply of
-		false ->
-		    ok;
-		_ ->
-		    _ = try From ! {cancel_timer, Req, false} catch _:_ -> ok end
-	    end,
-	    Nxt;
-	[{Time, TRef} = TKey] ->
-	    ets:delete(RTab, TRef),
-	    ets:delete(TTab, TKey),
-	    erlang:demonitor(TRef),
-	    case Reply of
-		false ->
-		    ok;
-		_ ->
-		    RcvTime = erlang:monotonic_time(),
-		    RT = case Time =< RcvTime of
-			     true ->
-				 0;
-			     false ->
-				 ((1000*(Time - RcvTime)) div Unit)
-			 end,
-		    _ = try From ! {cancel_timer, Req, RT} catch _:_ -> ok end
-	    end,
-	    case Time =:= Nxt of
-		false ->
-		    Nxt;
-		true ->
-		    case ets:first(TTab) of
-			'$end_of_table' -> infinity;
-			{NextTime, _TRef} -> NextTime
-		    end
-	    end
-    end;
-tsrv_handle_msg({read_timeout, BTR, From, Req, TRef},
-		#tsrv_state{rtab = RTab,
-			    unit = Unit,
-			    btr = BTR},
-		Nxt) ->
-    case ets:lookup(RTab, TRef) of
-	[] ->
-	    _ = try From ! {read_timer, Req, false} catch _:_ -> ok end;
-	[{Time, TRef}] ->
-	    RcvTime = erlang:monotonic_time(),
-	    RT = case Time =< RcvTime of
-		     true -> 0;
-		     false -> (1000*(Time - RcvTime)) div Unit
-		 end,
-	    _ = try From ! {read_timer, Req, RT} catch _:_ -> ok end
-    end,
-    Nxt;
-tsrv_handle_msg({'DOWN', TRef, process, _, _},
-		#tsrv_state{rtab = RTab,
-			    ttab = TTab},
-		Nxt) ->
-    case ets:lookup(RTab, TRef) of
-	[] ->
-	    Nxt;
-	[{Time, TRef} = TKey] ->
-	    ets:delete(RTab, TRef),
-	    ets:delete(TTab, TKey),
-	    case Time =:= Nxt of
-		false ->
-		    Nxt;
-		true ->
-		    case ets:first(TTab) of
-			'$end_of_table' -> infinity;
-			{NextTime, _} -> NextTime
-		    end
-	    end
-    end;
-tsrv_handle_msg({cancel_all_timeouts, BTR, From, Ref},
-		#tsrv_state{rtab = RTab,
-			    ttab = TTab,
-			    btr = BTR},
-		_Nxt) ->
-    tsrv_delete_monitor_objects(RTab),
-    ets:delete_all_objects(TTab),
-    try From ! {canceled_all_timeouts, Ref} catch _:_ -> ok end,
-    infinity;
-tsrv_handle_msg(_GarbageMsg, _StaticState, Nxt) ->
-    Nxt.
-
-tsrv_delete_monitor_objects(RTab) ->
-    case ets:first(RTab) of
-	'$end_of_table' ->
-	    ok;
-	TRef ->
-	    erlang:demonitor(TRef),
-	    ets:delete(RTab, TRef),
-	    tsrv_delete_monitor_objects(RTab)
-    end.
-
-tsrv_handle_timeout(CallTime, #tsrv_state{rtab = RTab,
-					  ttab = TTab} = S) ->
-    case ets:first(TTab) of
-	'$end_of_table' ->
-	    infinity;
-	{Time, _TRef} when Time > CallTime ->
-	    Time;
-	{_Time, TRef} = TKey ->
-	    [{TKey, Proc, Msg}] = ets:lookup(TTab, TKey),
-	    case erlang:is_pid(Proc) of
-		false -> ok;
-		_ -> erlang:demonitor(TRef)
-	    end,
-	    ets:delete(TTab, TKey),
-	    ets:delete(RTab, TRef),
-	    _ = try Proc ! Msg catch _:_ -> ok end,
-	    tsrv_handle_timeout(CallTime, S)
-    end.
diff --git a/erts/preloaded/src/init.erl b/erts/preloaded/src/init.erl
index 48c5c37717..bb56c9ff73 100644
--- a/erts/preloaded/src/init.erl
+++ b/erts/preloaded/src/init.erl
@@ -522,7 +522,6 @@ shutdown_pids(Heart,BootPid,State) ->
     Timer = shutdown_timer(State#state.flags),
     catch shutdown(State#state.kernel,BootPid,Timer,State),
     kill_all_pids(Heart), % Even the shutdown timer.
-    cancel_all_bif_timeouts(),
     kill_all_ports(Heart),
     flush_timout(Timer).
 
@@ -581,30 +580,6 @@ resend([ExitMsg|Exits]) ->
 resend(_) ->
     ok.
 
-
-cancel_all_bif_timeouts() ->
-    TSrvs = erts_internal:get_bif_timer_servers(),
-    Ref = make_ref(),
-    {BTR, _TSrv} = erts_internal:access_bif_timer(Ref), %% Cheat...
-    request_cancel_all_bif_timeouts(Ref, BTR, TSrvs),
-    wait_response_cancel_all_bif_timeouts(Ref, BTR, TSrvs),
-    ok.
-
-request_cancel_all_bif_timeouts(_Ref, _BTR, []) ->
-    ok;
-request_cancel_all_bif_timeouts(Ref, BTR, [TSrv|TSrvs]) ->
-    TSrv ! {cancel_all_timeouts, BTR, self(), {Ref, TSrv}},
-    request_cancel_all_bif_timeouts(Ref, BTR, TSrvs).
-
-wait_response_cancel_all_bif_timeouts(_Ref, _BTR, []) ->
-    ok;
-wait_response_cancel_all_bif_timeouts(Ref, BTR, [TSrv|TSrvs]) ->
-    receive
-	{canceled_all_timeouts, {Ref, TSrv}} ->
-	    wait_response_cancel_all_bif_timeouts(Ref, BTR, TSrvs)
-    end.
-
-
 %%
 %% Kill all existing pids in the system (except init and heart).
 kill_all_pids(Heart) ->