Merge branch 'rickard/cpu-groups/OTP-8861' into dev

* rickard/cpu-groups/OTP-8861:
  Generalize reader groups
  Move cpu topology functionality into erl_cpu_topology.[ch]
  Do not use more reader groups for schedulers than schedulers

Conflicts:
	erts/emulator/beam/erl_init.c

1 files changed, 33 insertions, 2015 deletions

diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 799f1a3cc9..f252c2cbe2 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -24,7 +24,6 @@
 #endif
 
 #include <stddef.h> /* offsetof() */
-#include <ctype.h>
 #include "sys.h"
 #include "erl_vm.h"
 #include "global.h"
@@ -39,6 +38,7 @@
 #include "erl_threads.h"
 #include "erl_binary.h"
 #include "beam_bp.h"
+#include "erl_cpu_topology.h"
 
 #define ERTS_RUNQ_CHECK_BALANCE_REDS_PER_SCHED (2000*CONTEXT_REDS)
 #define ERTS_RUNQ_CALL_CHECK_BALANCE_REDS \
@@ -63,8 +63,6 @@
 #define ERTS_WAKEUP_OTHER_DEC 10
 #define ERTS_WAKEUP_OTHER_FIXED_INC (CONTEXT_REDS/10)
 
-#define ERTS_MAX_CPU_TOPOLOGY_ID ((int) 0xffff)
-
 #if 0 || defined(DEBUG)
 #define ERTS_FAKE_SCHED_BIND_PRINT_SORTED_CPU_DATA
 #endif
@@ -119,10 +117,6 @@ Uint erts_process_tab_index_mask;
 
 static int wakeup_other_limit;
 
-#ifdef ERTS_SMP
-Uint erts_max_main_threads;
-#endif
-
 int erts_sched_thread_suggested_stack_size = -1;
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
@@ -195,48 +189,6 @@ do {							\
 
 #endif
 
-/*
- * Cpu topology hierarchy.
- */
-#define ERTS_TOPOLOGY_NODE		0
-#define ERTS_TOPOLOGY_PROCESSOR		1
-#define ERTS_TOPOLOGY_PROCESSOR_NODE	2
-#define ERTS_TOPOLOGY_CORE		3
-#define ERTS_TOPOLOGY_THREAD		4
-#define ERTS_TOPOLOGY_LOGICAL		5
-
-#define ERTS_TOPOLOGY_MAX_DEPTH		6
-
-typedef struct {
-    int bind_id;
-    int bound_id;
-} ErtsCpuBindData;
-
-static ErtsCpuBindData *scheduler2cpu_map;
-erts_smp_rwmtx_t erts_cpu_bind_rwmtx;
-
-typedef enum {
-    ERTS_CPU_BIND_UNDEFINED,
-    ERTS_CPU_BIND_SPREAD,
-    ERTS_CPU_BIND_PROCESSOR_SPREAD,
-    ERTS_CPU_BIND_THREAD_SPREAD,
-    ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD,
-    ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD,
-    ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD,
-    ERTS_CPU_BIND_NO_SPREAD,
-    ERTS_CPU_BIND_NONE
-} ErtsCpuBindOrder;
-
-#define ERTS_CPU_BIND_DEFAULT_BIND \
-  ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD
-
-ErtsCpuBindOrder cpu_bind_order;
-
-static erts_cpu_topology_t *user_cpudata;
-static int user_cpudata_size;
-static erts_cpu_topology_t *system_cpudata;
-static int system_cpudata_size;
-
 erts_sched_stat_t erts_sched_stat;
 
 ErtsRunQueue *erts_common_run_queue;
@@ -259,11 +211,6 @@ ErtsSchedulerData *erts_scheduler_data;
 ErtsAlignedRunQueue *erts_aligned_run_queues;
 Uint erts_no_run_queues;
 
-typedef union {
-    ErtsSchedulerData esd;
-    char align[ERTS_ALC_CACHE_LINE_ALIGN_SIZE(sizeof(ErtsSchedulerData))];
-} ErtsAlignedSchedulerData;
-
 ErtsAlignedSchedulerData *erts_aligned_scheduler_data;
 
 #ifdef ERTS_SMP
@@ -334,12 +281,6 @@ ERTS_SCHED_PREF_QUICK_ALLOC_IMPL(proclist,
 				 200,
 				 ERTS_ALC_T_PROC_LIST)
 
-#define ERTS_RUNQ_IX(IX)						\
-  (ASSERT_EXPR(0 <= (IX) && (IX) < erts_no_run_queues),			\
-   &erts_aligned_run_queues[(IX)].runq)
-#define ERTS_SCHEDULER_IX(IX)						\
-  (ASSERT_EXPR(0 <= (IX) && (IX) < erts_no_schedulers),			\
-   &erts_aligned_scheduler_data[(IX)].esd)
 #define ERTS_SCHED_SLEEP_INFO_IX(IX)					\
   (ASSERT_EXPR(0 <= (IX) && (IX) < erts_no_schedulers),			\
    &aligned_sched_sleep_info[(IX)].ssi)
@@ -398,22 +339,8 @@ static int stack_element_dump(int to, void *to_arg, Process* p, Eterm* sp,
 #ifdef ERTS_SMP
 static void handle_pending_exiters(ErtsProcList *);
 
-static void cpu_bind_order_sort(erts_cpu_topology_t *cpudata,
-				int size,
-				ErtsCpuBindOrder bind_order,
-				int mk_seq);
-static void signal_schedulers_bind_change(erts_cpu_topology_t *cpudata, int size);
-
 #endif
 
-static int reader_group_lookup(int logical);
-static void create_tmp_cpu_topology_copy(erts_cpu_topology_t **cpudata,
-					 int *cpudata_size);
-static void destroy_tmp_cpu_topology_copy(erts_cpu_topology_t *cpudata);
-
-static void early_cpu_bind_init(void);
-static void late_cpu_bind_init(void);
-
 #if defined(ERTS_SMP) && defined(ERTS_ENABLE_LOCK_CHECK)
 int
 erts_smp_lc_runq_is_locked(ErtsRunQueue *runq)
@@ -469,13 +396,13 @@ erts_pre_init_process(void)
 
 /* initialize the scheduler */
 void
-erts_init_process(void)
+erts_init_process(int ncpu)
 {
     Uint proc_bits = ERTS_PROC_BITS;
 
 #ifdef ERTS_SMP
     erts_disable_proc_not_running_opt = 0;
-    erts_init_proc_lock();
+    erts_init_proc_lock(ncpu);
 #endif
 
     init_proclist_alloc();
@@ -1246,7 +1173,7 @@ wake_scheduler(ErtsRunQueue *rq, int incq, int one)
 	do {
 	    ErtsSchedulerSleepInfo *wake_ssi = ssi;
 	    ssi = ssi->next;
-	    erts_sched_finish_poke(ssi, ssi_flags_set_wake(wake_ssi));
+	    erts_sched_finish_poke(wake_ssi, ssi_flags_set_wake(wake_ssi));
 	} while (ssi);
     }
 }
@@ -1339,6 +1266,31 @@ erts_smp_notify_inc_runq(ErtsRunQueue *runq)
     smp_notify_inc_runq(runq);
 }
 
+void
+erts_sched_notify_check_cpu_bind(void)
+{
+#ifdef ERTS_SMP
+    int ix;
+    if (erts_common_run_queue) {
+	for (ix = 0; ix < erts_no_schedulers; ix++)
+	    erts_smp_atomic_set(&ERTS_SCHEDULER_IX(ix)->chk_cpu_bind, 1);
+	wake_all_schedulers();
+    }
+    else {
+	for (ix = 0; ix < erts_no_run_queues; ix++) {
+	    ErtsRunQueue *rq = ERTS_RUNQ_IX(ix);
+	    erts_smp_runq_lock(rq);
+	    rq->flags |= ERTS_RUNQ_FLG_CHK_CPU_BIND;
+	    erts_smp_runq_unlock(rq);
+	    wake_scheduler(rq, 0, 1);
+	};
+    }
+#else
+    erts_sched_check_cpu_bind(erts_get_scheduler_data());
+#endif
+}
+
+
 #ifdef ERTS_SMP
 
 ErtsRunQueue *
@@ -2383,7 +2335,6 @@ erts_debug_nbalance(void)
 void
 erts_early_init_scheduling(void)
 {
-    early_cpu_bind_init();
     wakeup_other_limit = ERTS_WAKEUP_OTHER_LIMIT_MEDIUM;
 }
 
@@ -2660,8 +2611,6 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
 
     /* init port tasks */
     erts_port_task_init();
-
-    late_cpu_bind_init();
 }
 
 ErtsRunQueue *
@@ -2887,12 +2836,10 @@ suspend_scheduler(ErtsSchedulerData *esdp)
     long flgs;
     int changing;
     long no = (long) esdp->no;
-    ErtsRunQueue *rq = esdp->run_queue;
     ErtsSchedulerSleepInfo *ssi = esdp->ssi;
     long active_schedulers;
     int curr_online = 1;
     int wake = 0;
-    int reset_read_group = 0;
 #if defined(ERTS_SCHED_NEED_NONBLOCKABLE_AUX_WORK) \
     || defined(ERTS_SCHED_NEED_BLOCKABLE_AUX_WORK)
     long aux_work;
@@ -2913,20 +2860,7 @@ suspend_scheduler(ErtsSchedulerData *esdp)
 
     erts_smp_runq_unlock(esdp->run_queue);
 
-    /* Unbind from cpu */
-    erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx);
-    if (scheduler2cpu_map[esdp->no].bound_id >= 0
-	&& erts_unbind_from_cpu(erts_cpuinfo) == 0) {
-	esdp->cpu_id = scheduler2cpu_map[esdp->no].bound_id = -1;
-	reset_read_group = 1;
-    }
-    erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx);
-
-    if (reset_read_group)
-	erts_smp_rwmtx_set_reader_group(0);
-
-    if (esdp->no <= erts_max_main_threads)
-	erts_thr_set_main_status(0, 0);
+    erts_sched_check_cpu_bind_prep_suspend(esdp);
 
     if (erts_system_profile_flags.scheduler)
     	profile_scheduler(make_small(esdp->no), am_inactive);
@@ -3060,17 +2994,10 @@ suspend_scheduler(ErtsSchedulerData *esdp)
     if (erts_system_profile_flags.scheduler)
     	profile_scheduler(make_small(esdp->no), am_active);
 
-    if (esdp->no <= erts_max_main_threads)
-	erts_thr_set_main_status(1, (int) esdp->no);
-
     erts_smp_runq_lock(esdp->run_queue);
     non_empty_runq(esdp->run_queue);
 
-    /* Make sure we check if we should bind to a cpu or not... */
-    if (rq->flags & ERTS_RUNQ_FLG_SHARED_RUNQ)
-	erts_smp_atomic_set(&esdp->chk_cpu_bind, 1);
-    else
-	rq->flags |= ERTS_RUNQ_FLG_CHK_CPU_BIND;
+    erts_sched_check_cpu_bind_post_suspend(esdp);
 }
 
 #define ERTS_RUNQ_RESET_SUSPEND_INFO(RQ, DBG_ID)			\
@@ -3587,15 +3514,7 @@ sched_thread_func(void *vesdp)
     erts_tsd_set(sched_data_key, vesdp);
 #ifdef ERTS_SMP
 
-    if (no <= erts_max_main_threads) {
-	erts_thr_set_main_status(1, (int) no);
-	if (erts_reader_groups) {
-	    int rg = (int) no;
-	    if (rg > erts_reader_groups)
-		rg = (((int) no) - 1) % erts_reader_groups + 1;
-	    erts_smp_rwmtx_set_reader_group(rg);
-	}
-    }
+    erts_sched_init_check_cpu_bind((ErtsSchedulerData *) vesdp);
 
     erts_proc_lock_prepare_proc_lock_waiter();
     ERTS_SCHED_SLEEP_INFO_IX(no - 1)->event = erts_tse_fetch();
@@ -3697,1907 +3616,6 @@ erts_start_schedulers(void)
 
 #endif /* ERTS_SMP */
 
-static int
-int_cmp(const void *vx, const void *vy)
-{
-    return *((int *) vx) - *((int *) vy);
-}
-
-static int
-cpu_spread_order_cmp(const void *vx, const void *vy)
-{
-    erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx;
-    erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy;
-
-    if (x->thread != y->thread)
-	return x->thread - y->thread;
-    if (x->core != y->core)
-	return x->core - y->core;
-    if (x->processor_node != y->processor_node)
-	return x->processor_node - y->processor_node;
-    if (x->processor != y->processor)
-	return x->processor - y->processor;
-    if (x->node != y->node)
-	return x->node - y->node;
-    return 0;
-}
-
-static int
-cpu_processor_spread_order_cmp(const void *vx, const void *vy)
-{
-    erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx;
-    erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy;
-
-    if (x->thread != y->thread)
-	return x->thread - y->thread;
-    if (x->processor_node != y->processor_node)
-	return x->processor_node - y->processor_node;
-    if (x->core != y->core)
-	return x->core - y->core;
-    if (x->node != y->node)
-	return x->node - y->node;
-    if (x->processor != y->processor)
-	return x->processor - y->processor;
-    return 0;
-}
-
-static int
-cpu_thread_spread_order_cmp(const void *vx, const void *vy)
-{
-    erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx;
-    erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy;
-
-    if (x->thread != y->thread)
-	return x->thread - y->thread;
-    if (x->node != y->node)
-	return x->node - y->node;
-    if (x->processor != y->processor)
-	return x->processor - y->processor;
-    if (x->processor_node != y->processor_node)
-	return x->processor_node - y->processor_node;
-    if (x->core != y->core)
-	return x->core - y->core;
-    return 0;
-}
-
-static int
-cpu_thread_no_node_processor_spread_order_cmp(const void *vx, const void *vy)
-{
-    erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx;
-    erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy;
-
-    if (x->thread != y->thread)
-	return x->thread - y->thread;
-    if (x->node != y->node)
-	return x->node - y->node;
-    if (x->core != y->core)
-	return x->core - y->core;
-    if (x->processor != y->processor)
-	return x->processor - y->processor;
-    return 0;
-}
-
-static int
-cpu_no_node_processor_spread_order_cmp(const void *vx, const void *vy)
-{
-    erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx;
-    erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy;
-
-    if (x->node != y->node)
-	return x->node - y->node;
-    if (x->thread != y->thread)
-	return x->thread - y->thread;
-    if (x->core != y->core)
-	return x->core - y->core;
-    if (x->processor != y->processor)
-	return x->processor - y->processor;
-    return 0;
-}
-
-static int
-cpu_no_node_thread_spread_order_cmp(const void *vx, const void *vy)
-{
-    erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx;
-    erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy;
-
-    if (x->node != y->node)
-	return x->node - y->node;
-    if (x->thread != y->thread)
-	return x->thread - y->thread;
-    if (x->processor != y->processor)
-	return x->processor - y->processor;
-    if (x->core != y->core)
-	return x->core - y->core;
-    return 0;
-}
-
-static int
-cpu_no_spread_order_cmp(const void *vx, const void *vy)
-{
-    erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx;
-    erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy;
-
-    if (x->node != y->node)
-	return x->node - y->node;
-    if (x->processor != y->processor)
-	return x->processor - y->processor;
-    if (x->processor_node != y->processor_node)
-	return x->processor_node - y->processor_node;
-    if (x->core != y->core)
-	return x->core - y->core;
-    if (x->thread != y->thread)
-	return x->thread - y->thread;
-    return 0;
-}
-
-static ERTS_INLINE void
-make_cpudata_id_seq(erts_cpu_topology_t *cpudata, int size, int no_node)
-{
-    int ix;
-    int node = -1;
-    int processor = -1;
-    int processor_node = -1;
-    int processor_node_node = -1;
-    int core = -1;
-    int thread = -1;
-    int old_node = -1;
-    int old_processor = -1;
-    int old_processor_node = -1;
-    int old_core = -1;
-    int old_thread = -1;
-
-    for (ix = 0; ix < size; ix++) {
-	if (!no_node || cpudata[ix].node >= 0) {
-	    if (old_node == cpudata[ix].node)
-		cpudata[ix].node = node;
-	    else {
-		old_node = cpudata[ix].node;
-		old_processor = processor = -1;
-		if (!no_node)
-		    old_processor_node = processor_node = -1;
-		old_core = core = -1;
-		old_thread = thread = -1;
-		if (no_node || cpudata[ix].node >= 0)
-		    cpudata[ix].node = ++node;
-	    }
-	}
-	if (old_processor == cpudata[ix].processor)
-	    cpudata[ix].processor = processor;
-	else {
-	    old_processor = cpudata[ix].processor;
-	    if (!no_node)
-		processor_node_node = old_processor_node = processor_node = -1;
-	    old_core = core = -1;
-	    old_thread = thread = -1;
-	    cpudata[ix].processor = ++processor;
-	}
-	if (no_node && cpudata[ix].processor_node < 0)
-	    old_processor_node = -1;
-	else {
-	    if (old_processor_node == cpudata[ix].processor_node) {
-		if (no_node)
-		    cpudata[ix].node = cpudata[ix].processor_node = node;
-		else {
-		    if (processor_node_node >= 0)
-			cpudata[ix].node = processor_node_node;
-		    cpudata[ix].processor_node = processor_node;
-		}
-	    }
-	    else {
-		old_processor_node = cpudata[ix].processor_node;
-		old_core = core = -1;
-		old_thread = thread = -1;
-		if (no_node)
-		    cpudata[ix].node = cpudata[ix].processor_node = ++node;
-		else {
-		    cpudata[ix].node = processor_node_node = ++node;
-		    cpudata[ix].processor_node = ++processor_node;
-		}
-	    }
-	}
-	if (!no_node && cpudata[ix].processor_node < 0)
-	    cpudata[ix].processor_node = 0;
-	if (old_core == cpudata[ix].core)
-	    cpudata[ix].core = core;
-	else {
-	    old_core = cpudata[ix].core;
-	    old_thread = thread = -1;
-	    cpudata[ix].core = ++core;
-	}
-	if (old_thread == cpudata[ix].thread)
-	    cpudata[ix].thread = thread;
-	else
-	    old_thread = cpudata[ix].thread = ++thread;
-    }
-}
-
-static void
-cpu_bind_order_sort(erts_cpu_topology_t *cpudata,
-		    int size,
-		    ErtsCpuBindOrder bind_order,
-		    int mk_seq)
-{
-    if (size > 1) {
-	int no_node = 0;
-	int (*cmp_func)(const void *, const void *);
-	switch (bind_order) {
-	case ERTS_CPU_BIND_SPREAD:
-	    cmp_func = cpu_spread_order_cmp;
-	    break;
-	case ERTS_CPU_BIND_PROCESSOR_SPREAD:
-	    cmp_func = cpu_processor_spread_order_cmp;
-	    break;
-	case ERTS_CPU_BIND_THREAD_SPREAD:
-	    cmp_func = cpu_thread_spread_order_cmp;
-	    break;
-	case ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD:
-	    no_node = 1;
-	    cmp_func = cpu_thread_no_node_processor_spread_order_cmp;
-	    break;
-	case ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD:
-	    no_node = 1;
-	    cmp_func = cpu_no_node_processor_spread_order_cmp;
-	    break;
-	case ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD:
-	    no_node = 1;
-	    cmp_func = cpu_no_node_thread_spread_order_cmp;
-	    break;
-	case ERTS_CPU_BIND_NO_SPREAD:
-	    cmp_func = cpu_no_spread_order_cmp;
-	    break;
-	default:
-	    cmp_func = NULL;
-	    erl_exit(ERTS_ABORT_EXIT,
-		     "Bad cpu bind type: %d\n",
-		     (int) cpu_bind_order);
-	    break;
-	}
-
-	if (mk_seq)
-	    make_cpudata_id_seq(cpudata, size, no_node);
-
-	qsort(cpudata, size, sizeof(erts_cpu_topology_t), cmp_func);
-    }
-}
-
-static int
-processor_order_cmp(const void *vx, const void *vy)
-{
-    erts_cpu_topology_t *x = (erts_cpu_topology_t *) vx;
-    erts_cpu_topology_t *y = (erts_cpu_topology_t *) vy;
-
-    if (x->processor != y->processor)
-	return x->processor - y->processor;
-    if (x->node != y->node)
-	return x->node - y->node;
-    if (x->processor_node != y->processor_node)
-	return x->processor_node - y->processor_node;
-    if (x->core != y->core)
-	return x->core - y->core;
-    if (x->thread != y->thread)
-	return x->thread - y->thread;
-    return 0;
-}
-
-static void
-check_cpu_bind(ErtsSchedulerData *esdp)
-{
-    int rg = 0;
-    int res;
-    int cpu_id;
-    erts_smp_runq_unlock(esdp->run_queue);
-    erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx);
-    cpu_id = scheduler2cpu_map[esdp->no].bind_id;
-    if (cpu_id >= 0 && cpu_id != scheduler2cpu_map[esdp->no].bound_id) {
-	res = erts_bind_to_cpu(erts_cpuinfo, cpu_id);
-	if (res == 0)
-	    esdp->cpu_id = scheduler2cpu_map[esdp->no].bound_id = cpu_id;
-	else {
-	    erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf();
-	    erts_dsprintf(dsbufp, "Scheduler %d failed to bind to cpu %d: %s\n",
-			  (int) esdp->no, cpu_id, erl_errno_id(-res));
-	    erts_send_error_to_logger_nogl(dsbufp);
-	    if (scheduler2cpu_map[esdp->no].bound_id >= 0)
-		goto unbind;
-	}
-    }
-    else if (cpu_id < 0) {
-    unbind:
-	/* Get rid of old binding */
-	res = erts_unbind_from_cpu(erts_cpuinfo);
-	if (res == 0)
-	    esdp->cpu_id = scheduler2cpu_map[esdp->no].bound_id = -1;
-	else if (res != -ENOTSUP) {
-	    erts_dsprintf_buf_t *dsbufp = erts_create_logger_dsbuf();
-	    erts_dsprintf(dsbufp, "Scheduler %d failed to unbind from cpu %d: %s\n",
-			  (int) esdp->no, cpu_id, erl_errno_id(-res));
-	    erts_send_error_to_logger_nogl(dsbufp);
-	}
-    }
-    if (erts_reader_groups) {
-	if (esdp->cpu_id >= 0)
-	    rg = reader_group_lookup(esdp->cpu_id);
-	else
-	    rg = (((int) esdp->no) - 1) % erts_reader_groups + 1;
-    }
-    erts_smp_runq_lock(esdp->run_queue);
-#ifdef ERTS_SMP
-    if (erts_common_run_queue)
-	erts_smp_atomic_set(&esdp->chk_cpu_bind, 0);
-    else {
-	esdp->run_queue->flags &= ~ERTS_RUNQ_FLG_CHK_CPU_BIND;
-    }
-#endif
-    erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx);
-
-    if (erts_reader_groups)
-	erts_smp_rwmtx_set_reader_group(rg);
-}
-
-static void
-signal_schedulers_bind_change(erts_cpu_topology_t *cpudata, int size)
-{
-    int s_ix = 1;
-    int cpu_ix;
-
-    if (cpu_bind_order != ERTS_CPU_BIND_NONE && size) {
-
-	cpu_bind_order_sort(cpudata, size, cpu_bind_order, 1);
-
-	for (cpu_ix = 0; cpu_ix < size && cpu_ix < erts_no_schedulers; cpu_ix++)
-	    if (erts_is_cpu_available(erts_cpuinfo, cpudata[cpu_ix].logical))
-		scheduler2cpu_map[s_ix++].bind_id = cpudata[cpu_ix].logical;
-    }
-
-    if (s_ix <= erts_no_schedulers)
-	for (; s_ix <= erts_no_schedulers; s_ix++)
-	    scheduler2cpu_map[s_ix].bind_id = -1;
-
-#ifdef ERTS_SMP
-    if (erts_common_run_queue) {
-	for (s_ix = 0; s_ix < erts_no_schedulers; s_ix++)
-	    erts_smp_atomic_set(&ERTS_SCHEDULER_IX(s_ix)->chk_cpu_bind, 1);
-	wake_all_schedulers();
-    }
-    else {
-	for (s_ix = 0; s_ix < erts_no_run_queues; s_ix++) {
-	    ErtsRunQueue *rq = ERTS_RUNQ_IX(s_ix);
-	    erts_smp_runq_lock(rq);
-	    rq->flags |= ERTS_RUNQ_FLG_CHK_CPU_BIND;
-	    erts_smp_runq_unlock(rq);
-	    wake_scheduler(rq, 0, 1);
-	};
-    }
-#else
-    check_cpu_bind(erts_get_scheduler_data());
-#endif
-}
-
-int
-erts_init_scheduler_bind_type(char *how)
-{
-    if (erts_bind_to_cpu(erts_cpuinfo, -1) == -ENOTSUP)
-	return ERTS_INIT_SCHED_BIND_TYPE_NOT_SUPPORTED;
-
-    if (!system_cpudata && !user_cpudata)
-	return ERTS_INIT_SCHED_BIND_TYPE_ERROR_NO_CPU_TOPOLOGY;
-
-    if (sys_strcmp(how, "db") == 0)
-	cpu_bind_order = ERTS_CPU_BIND_DEFAULT_BIND;
-    else if (sys_strcmp(how, "s") == 0)
-	cpu_bind_order = ERTS_CPU_BIND_SPREAD;
-    else if (sys_strcmp(how, "ps") == 0)
-	cpu_bind_order = ERTS_CPU_BIND_PROCESSOR_SPREAD;
-    else if (sys_strcmp(how, "ts") == 0)
-	cpu_bind_order = ERTS_CPU_BIND_THREAD_SPREAD;
-    else if (sys_strcmp(how, "tnnps") == 0)
-	cpu_bind_order = ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD;
-    else if (sys_strcmp(how, "nnps") == 0)
-	cpu_bind_order = ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD;
-    else if (sys_strcmp(how, "nnts") == 0)
-	cpu_bind_order = ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD;
-    else if (sys_strcmp(how, "ns") == 0)
-	cpu_bind_order = ERTS_CPU_BIND_NO_SPREAD;
-    else if (sys_strcmp(how, "u") == 0)
-	cpu_bind_order = ERTS_CPU_BIND_NONE;
-    else
-	return ERTS_INIT_SCHED_BIND_TYPE_ERROR_NO_BAD_TYPE;
-
-    return ERTS_INIT_SCHED_BIND_TYPE_SUCCESS;
-}
-
-/*
- * reader groups map
- */
-
-typedef struct {
-    int level[ERTS_TOPOLOGY_MAX_DEPTH+1];
-} erts_avail_cput;
-
-typedef struct {
-    int *map;
-    int size;
-    int groups;
-} erts_reader_groups_map_test;
-
-typedef struct {
-    int id;
-    int sub_levels;
-    int reader_groups;
-} erts_rg_count_t;
-
-typedef struct {
-    int logical;
-    int reader_group;
-} erts_reader_groups_map_t;
-
-typedef struct {
-    erts_reader_groups_map_t *map;
-    int map_size;
-    int logical_processors;
-    int groups;
-} erts_make_reader_groups_map_test;
-
-static int reader_groups_available_cpu_check;
-static int reader_groups_logical_processors;
-static int reader_groups_map_size;
-static erts_reader_groups_map_t *reader_groups_map;
-
-#define ERTS_TOPOLOGY_RG ERTS_TOPOLOGY_MAX_DEPTH
-
-static void
-make_reader_groups_map(erts_make_reader_groups_map_test *test);
-
-static Eterm
-get_reader_groups_map(Process *c_p,
-		      erts_reader_groups_map_t *map,
-		      int map_size,
-		      int logical_processors)
-{
-#ifdef DEBUG
-    Eterm *endp;
-#endif
-    Eterm res = NIL, tuple;
-    Eterm *hp;
-    int i;
-
-    hp = HAlloc(c_p, logical_processors*(2+3));
-#ifdef DEBUG
-    endp = hp + logical_processors*(2+3);
-#endif
-    for (i = map_size - 1; i >= 0; i--) {
-	if (map[i].logical >= 0) {
-	    tuple = TUPLE2(hp,
-			   make_small(map[i].logical),
-			   make_small(map[i].reader_group));
-	    hp += 3;
-	    res = CONS(hp, tuple, res);
-	    hp += 2;
-	}
-    }
-    ASSERT(hp == endp);
-    return res;
-}
-
-Eterm
-erts_debug_reader_groups_map(Process *c_p, int groups)
-{
-    Eterm res;
-    erts_make_reader_groups_map_test test;
-
-    test.groups = groups;
-    make_reader_groups_map(&test);
-    if (!test.map)
-	res = NIL;
-    else {
-	res = get_reader_groups_map(c_p,
-				    test.map,
-				    test.map_size,
-				    test.logical_processors);
-	erts_free(ERTS_ALC_T_TMP, test.map);
-    }
-    return res;
-}
-
-
-Eterm
-erts_get_reader_groups_map(Process *c_p)
-{
-    Eterm res;
-    erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx);
-    res = get_reader_groups_map(c_p,
-				reader_groups_map,
-				reader_groups_map_size,
-				reader_groups_logical_processors);
-    erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx);
-    return res;
-}
-
-static void
-make_available_cpu_topology(erts_avail_cput *no,
-			    erts_avail_cput *avail,
-			    erts_cpu_topology_t *cpudata,
-			    int *size,
-			    int test)
-{
-    int len = *size;
-    erts_cpu_topology_t last;
-    int a, i, j;
-
-    no->level[ERTS_TOPOLOGY_NODE] = -1;
-    no->level[ERTS_TOPOLOGY_PROCESSOR] = -1;
-    no->level[ERTS_TOPOLOGY_PROCESSOR_NODE] = -1;
-    no->level[ERTS_TOPOLOGY_CORE] = -1;
-    no->level[ERTS_TOPOLOGY_THREAD] = -1;
-    no->level[ERTS_TOPOLOGY_LOGICAL] = -1;
-
-    last.node = INT_MIN;
-    last.processor = INT_MIN;
-    last.processor_node = INT_MIN;
-    last.core = INT_MIN;
-    last.thread = INT_MIN;
-    last.logical = INT_MIN;
-
-    a = 0;
-
-    for (i = 0; i < len; i++) {
-
-	if (!test && !erts_is_cpu_available(erts_cpuinfo, cpudata[i].logical))
-	    continue;
-
-	if (last.node != cpudata[i].node)
-	    goto node;
-	if (last.processor != cpudata[i].processor)
-	    goto processor;
-	if (last.processor_node != cpudata[i].processor_node)
-	    goto processor_node;
-	if (last.core != cpudata[i].core)
-	    goto core;
-	ASSERT(last.thread != cpudata[i].thread);
-	goto thread;
-
-    node:
-	no->level[ERTS_TOPOLOGY_NODE]++;
-    processor:
-	no->level[ERTS_TOPOLOGY_PROCESSOR]++;
-    processor_node:
-	no->level[ERTS_TOPOLOGY_PROCESSOR_NODE]++;
-    core:
-	no->level[ERTS_TOPOLOGY_CORE]++;
-    thread:
-	no->level[ERTS_TOPOLOGY_THREAD]++;
-
-	no->level[ERTS_TOPOLOGY_LOGICAL]++;
-
-	for (j = 0; j < ERTS_TOPOLOGY_LOGICAL; j++)
-	    avail[a].level[j] = no->level[j];
-
-	avail[a].level[ERTS_TOPOLOGY_LOGICAL] = cpudata[i].logical;
-	avail[a].level[ERTS_TOPOLOGY_RG] = 0;
-
-	ASSERT(last.logical != cpudata[a].logical);
-
-	last = cpudata[i];
-	a++;
-    }
-
-    no->level[ERTS_TOPOLOGY_NODE]++;
-    no->level[ERTS_TOPOLOGY_PROCESSOR]++;
-    no->level[ERTS_TOPOLOGY_PROCESSOR_NODE]++;
-    no->level[ERTS_TOPOLOGY_CORE]++;
-    no->level[ERTS_TOPOLOGY_THREAD]++;
-    no->level[ERTS_TOPOLOGY_LOGICAL]++;
-
-    *size = a;
-}
-
-static int
-reader_group_lookup(int logical)
-{
-    int start = logical % reader_groups_map_size;
-    int ix = start;
-
-    do {
-	if (reader_groups_map[ix].logical == logical) {
-	    ASSERT(reader_groups_map[ix].reader_group > 0);
-	    return reader_groups_map[ix].reader_group;
-	}
-	ix++;
-	if (ix == reader_groups_map_size)
-	    ix = 0;
-    } while (ix != start);
-
-    erl_exit(ERTS_ABORT_EXIT, "Logical cpu id %d not found\n", logical);
-}
-
-static void
-reader_group_insert(erts_reader_groups_map_t *map, int map_size,
-		    int logical, int reader_group)
-{
-    int start = logical % map_size;
-    int ix = start;
-
-    do {
-	if (map[ix].logical < 0) {
-	    map[ix].logical = logical;
-	    map[ix].reader_group = reader_group;
-	    return;
-	}
-	ix++;
-	if (ix == map_size)
-	    ix = 0;
-    } while (ix != start);
-
-    erl_exit(ERTS_ABORT_EXIT, "Reader groups map full\n");
-}
-
-
-static int
-sub_levels(erts_rg_count_t *rgc, int level, int aix, int avail_sz, erts_avail_cput *avail)
-{
-    int sub_level = level+1;
-    int last = -1;
-    rgc->sub_levels = 0;
-
-    do {
-	if (last != avail[aix].level[sub_level]) {
-	    rgc->sub_levels++;
-	    last = avail[aix].level[sub_level];
-	}
-	aix++;
-    }
-    while (aix < avail_sz && rgc->id == avail[aix].level[level]);
-    rgc->reader_groups = 0;
-    return aix;
-}
-
-static int
-write_reader_groups(int *rgp, erts_rg_count_t *rgcp,
-		    int level, int a,
-		    int avail_sz, erts_avail_cput *avail)
-{
-    int rg = *rgp;
-    int sub_level = level+1;
-    int sl_per_gr = rgcp->sub_levels / rgcp->reader_groups;
-    int xsl = rgcp->sub_levels % rgcp->reader_groups;
-    int sls = 0;
-    int last = -1;
-    int xsl_rg_lim = (rgcp->reader_groups - xsl) + rg + 1;
-
-    ASSERT(level < 0 || avail[a].level[level] == rgcp->id)
-
-    do {
-	if (last != avail[a].level[sub_level]) {
-	    if (!sls) {
-		sls = sl_per_gr;
-		rg++;
-		if (rg >= xsl_rg_lim)
-		    sls++;
-	    }
-	    last = avail[a].level[sub_level];
-	    sls--;
-	}
-	avail[a].level[ERTS_TOPOLOGY_RG] = rg;
-	a++;
-    } while (a < avail_sz && (level < 0
-			      || avail[a].level[level] == rgcp->id));
-
-    ASSERT(rgcp->reader_groups == rg - *rgp);
-
-    *rgp = rg;
-
-    return a;
-}
-
-static int
-rg_count_sub_levels_compare(const void *vx, const void *vy)
-{
-    erts_rg_count_t *x = (erts_rg_count_t *) vx;
-    erts_rg_count_t *y = (erts_rg_count_t *) vy;
-    if (x->sub_levels != y->sub_levels)
-	return y->sub_levels - x->sub_levels;
-    return x->id - y->id;
-}
-
-static int
-rg_count_id_compare(const void *vx, const void *vy)
-{
-    erts_rg_count_t *x = (erts_rg_count_t *) vx;
-    erts_rg_count_t *y = (erts_rg_count_t *) vy;
-    return x->id - y->id;
-}
-
-static void
-make_reader_groups_map(erts_make_reader_groups_map_test *test)
-{
-    int i, spread_level, avail_sz;
-    erts_avail_cput no, *avail;
-    erts_cpu_topology_t *cpudata;
-    erts_reader_groups_map_t *map;
-    int map_sz;
-    int groups = erts_reader_groups;
-
-    if (test) {
-	test->map = NULL;
-	test->map_size = 0;
-	groups = test->groups;
-    }
-
-    if (!groups)
-	return;
-
-    if (!test) {
-	if (reader_groups_map)
-	    erts_free(ERTS_ALC_T_RDR_GRPS_MAP, reader_groups_map);
-
-	reader_groups_logical_processors = 0;
-	reader_groups_map_size = 0;
-	reader_groups_map = NULL;
-    }
-
-    create_tmp_cpu_topology_copy(&cpudata, &avail_sz);
-
-    if (!cpudata)
-	return;
-
-    cpu_bind_order_sort(cpudata,
-			avail_sz,
-			ERTS_CPU_BIND_NO_SPREAD,
-			1);
-
-    avail = erts_alloc(ERTS_ALC_T_TMP,
-		       sizeof(erts_avail_cput)*avail_sz);
-
-    make_available_cpu_topology(&no, avail, cpudata,
-				&avail_sz, test != NULL);
-
-    destroy_tmp_cpu_topology_copy(cpudata);
-
-    map_sz = avail_sz*2+1;
-
-    if (test) {
-	map = erts_alloc(ERTS_ALC_T_TMP,
-			 (sizeof(erts_reader_groups_map_t)
-			  * map_sz));
-	test->map = map;
-	test->map_size = map_sz;
-	test->logical_processors = avail_sz;
-    }
-    else {
-	map = erts_alloc(ERTS_ALC_T_RDR_GRPS_MAP,
-			 (sizeof(erts_reader_groups_map_t)
-			  * map_sz));
-	reader_groups_map = map;
-	reader_groups_logical_processors = avail_sz;
-	reader_groups_map_size = map_sz;
-
-    }
-
-    for (i = 0; i < map_sz; i++) {
-	map[i].logical = -1;
-	map[i].reader_group = 0;
-    }
-
-    spread_level = ERTS_TOPOLOGY_CORE;
-    for (i = ERTS_TOPOLOGY_NODE; i < ERTS_TOPOLOGY_THREAD; i++) {
-	if (no.level[i] > groups) {
-	    spread_level = i;
-	    break;
-	}
-    }
-
-    if (no.level[spread_level] <= groups) {
-	int a, rg, last = -1;
-	rg = 0;
-	ASSERT(spread_level == ERTS_TOPOLOGY_CORE);
-	for (a = 0; a < avail_sz; a++) {
-	    if (last != avail[a].level[spread_level]) {
-		rg++;
-		last = avail[a].level[spread_level];
-	    }
-	    reader_group_insert(map,
-				map_sz,
-				avail[a].level[ERTS_TOPOLOGY_LOGICAL],
-				rg);
-	}
-    }
-    else { /* groups < no.level[spread_level] */
-	erts_rg_count_t *rg_count;
-	int a, rg, tl, toplevels;
-
-	tl = spread_level-1;
-
-	if (spread_level == ERTS_TOPOLOGY_NODE)
-	    toplevels = 1;
-	else
-	    toplevels = no.level[tl];
-
-	rg_count = erts_alloc(ERTS_ALC_T_TMP,
-			      toplevels*sizeof(erts_rg_count_t));
-
-	if (toplevels == 1) {
-	    rg_count[0].id = 0;
-	    rg_count[0].sub_levels = no.level[spread_level];
-	    rg_count[0].reader_groups = groups;
-	}
-	else {
-	    int rgs_per_tl, rgs;
-	    rgs = groups;
-	    rgs_per_tl = rgs / toplevels;
-
-	    a = 0;
-	    for (i = 0; i < toplevels; i++) {
-		rg_count[i].id = avail[a].level[tl];
-		a = sub_levels(&rg_count[i], tl, a, avail_sz, avail);
-	    }
-
-	    qsort(rg_count,
-		  toplevels,
-		  sizeof(erts_rg_count_t),
-		  rg_count_sub_levels_compare);
-
-	    for (i = 0; i < toplevels; i++) {
-		if (rg_count[i].sub_levels < rgs_per_tl) {
-		    rg_count[i].reader_groups = rg_count[i].sub_levels;
-		    rgs -= rg_count[i].sub_levels;
-		}
-		else {
-		    rg_count[i].reader_groups = rgs_per_tl;
-		    rgs -= rgs_per_tl;
-		}
-	    }
-
-	    while (rgs > 0) {
-		for (i = 0; i < toplevels; i++) {
-		    if (rg_count[i].sub_levels == rg_count[i].reader_groups)
-			break;
-		    else {
-			rg_count[i].reader_groups++;
-			if (--rgs == 0)
-			    break;
-		    }
-		}
-	    }
-
-	    qsort(rg_count,
-		  toplevels,
-		  sizeof(erts_rg_count_t),
-		  rg_count_id_compare);
-	}
-
-	a = i = rg = 0;
-	while (a < avail_sz) {
-	    a = write_reader_groups(&rg, &rg_count[i], tl,
-				    a, avail_sz, avail);
-	    i++;
-	}
-
-	ASSERT(groups == rg);
-
-	for (a = 0; a < avail_sz; a++)
-	    reader_group_insert(map,
-				map_sz,
-				avail[a].level[ERTS_TOPOLOGY_LOGICAL],
-				avail[a].level[ERTS_TOPOLOGY_RG]);
-
-	erts_free(ERTS_ALC_T_TMP, rg_count);
-    }
-
-    erts_free(ERTS_ALC_T_TMP, avail);
-}
-
-/*
- * CPU topology
- */
-
-typedef struct {
-    int *id;
-    int used;
-    int size;
-} ErtsCpuTopIdSeq;
-
-typedef struct {
-    ErtsCpuTopIdSeq logical;
-    ErtsCpuTopIdSeq thread;
-    ErtsCpuTopIdSeq core;
-    ErtsCpuTopIdSeq processor_node;
-    ErtsCpuTopIdSeq processor;
-    ErtsCpuTopIdSeq node;
-} ErtsCpuTopEntry;
-
-static void
-init_cpu_top_entry(ErtsCpuTopEntry *cte)
-{
-    int size = 10;
-    cte->logical.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS,
-				 sizeof(int)*size);
-    cte->logical.size = size;
-    cte->thread.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS,
-				sizeof(int)*size);
-    cte->thread.size = size;
-    cte->core.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS,
-			      sizeof(int)*size);
-    cte->core.size = size;
-    cte->processor_node.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS,
-					sizeof(int)*size);
-    cte->processor_node.size = size;
-    cte->processor.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS,
-				   sizeof(int)*size);
-    cte->processor.size = size;
-    cte->node.id = erts_alloc(ERTS_ALC_T_TMP_CPU_IDS,
-			      sizeof(int)*size);
-    cte->node.size = size;
-}
-
-static void
-destroy_cpu_top_entry(ErtsCpuTopEntry *cte)
-{
-    erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->logical.id);
-    erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->thread.id);
-    erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->core.id);
-    erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->processor_node.id);
-    erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->processor.id);
-    erts_free(ERTS_ALC_T_TMP_CPU_IDS, cte->node.id);
-}
-
-static int
-get_cput_value_or_range(int *v, int *vr, char **str)
-{
-    long l;
-    char *c = *str;
-    errno = 0;
-    if (!isdigit((unsigned char)*c))
-	return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID;
-    l = strtol(c, &c, 10);
-    if (errno != 0 || l < 0 || ERTS_MAX_CPU_TOPOLOGY_ID < l)
-	return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID;
-    *v = (int) l;
-    if (*c == '-') {
-	c++;
-	if (!isdigit((unsigned char)*c))
-	    return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE;
-	l = strtol(c, &c, 10);
-	if (errno != 0 || l < 0 || ERTS_MAX_CPU_TOPOLOGY_ID < l)
-	    return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE;
-	*vr = (int) l;
-    }
-    *str = c;
-    return ERTS_INIT_CPU_TOPOLOGY_OK;
-}
-
-static int
-get_cput_id_seq(ErtsCpuTopIdSeq *idseq, char **str)
-{
-    int ix = 0;
-    int need_size = 0;
-    char *c = *str;
-
-    while (1) {
-	int res;
-	int val;
-	int nids;
-	int val_range = -1;
-	res = get_cput_value_or_range(&val, &val_range, &c);
-	if (res != ERTS_INIT_CPU_TOPOLOGY_OK)
-	    return res;
-	if (val_range < 0 || val_range == val)
-	    nids = 1;
-	else {
-	    if (val_range > val)
-		nids = val_range - val + 1;
-	    else
-		nids = val - val_range + 1;
-	}
-	need_size += nids;
-	if (need_size > idseq->size) {
-	    idseq->size = need_size + 10;
-	    idseq->id = erts_realloc(ERTS_ALC_T_TMP_CPU_IDS,
-				      idseq->id,
-				      sizeof(int)*idseq->size);
-	}
-	if (nids == 1)
-	    idseq->id[ix++] = val;
-	else if (val_range > val) {
-	    for (; val <= val_range; val++)
-		idseq->id[ix++] = val;
-	}
-	else {
-	    for (; val >= val_range; val--)
-		idseq->id[ix++] = val;
-	}
-	if (*c != ',')
-	    break;
-	c++;
-    }
-    *str = c;
-    idseq->used = ix;
-    return ERTS_INIT_CPU_TOPOLOGY_OK;
-}
-
-static int
-get_cput_entry(ErtsCpuTopEntry *cput, char **str)
-{
-    int h;
-    char *c = *str;
-
-    cput->logical.used = 0;
-    cput->thread.id[0] = 0;
-    cput->thread.used = 1;
-    cput->core.id[0] = 0;
-    cput->core.used = 1;
-    cput->processor_node.id[0] = -1;
-    cput->processor_node.used = 1;
-    cput->processor.id[0] = 0;
-    cput->processor.used = 1;
-    cput->node.id[0] = -1;
-    cput->node.used = 1;
-
-    h = ERTS_TOPOLOGY_MAX_DEPTH;
-    while (*c != ':' && *c != '\0') {
-	int res;
-	ErtsCpuTopIdSeq *idseqp;
-	switch (*c++) {
-	case 'L':
-	    if (h <= ERTS_TOPOLOGY_LOGICAL)
-		return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY;
-	    idseqp = &cput->logical;
-	    h = ERTS_TOPOLOGY_LOGICAL;
-	    break;
-	case 't':
-	case 'T':
-	    if (h <= ERTS_TOPOLOGY_THREAD)
-		return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY;
-	    idseqp = &cput->thread;
-	    h = ERTS_TOPOLOGY_THREAD;
-	    break;
-	case 'c':
-	case 'C':
-	    if (h <= ERTS_TOPOLOGY_CORE)
-		return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY;
-	    idseqp = &cput->core;
-	    h = ERTS_TOPOLOGY_CORE;
-	    break;
-	case 'p':
-	case 'P':
-	    if (h <= ERTS_TOPOLOGY_PROCESSOR)
-		return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY;
-	    idseqp = &cput->processor;
-	    h = ERTS_TOPOLOGY_PROCESSOR;
-	    break;
-	case 'n':
-	case 'N':
-	    if (h <= ERTS_TOPOLOGY_PROCESSOR) {
-	    do_node:
-		if (h <= ERTS_TOPOLOGY_NODE)
-		    return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY;
-		idseqp = &cput->node;
-		h = ERTS_TOPOLOGY_NODE;
-	    }
-	    else {
-		int p_node = 0;
-		char *p_chk = c;
-		while (*p_chk != '\0' && *p_chk != ':') {
-		    if (*p_chk == 'p' || *p_chk == 'P') {
-			p_node = 1;
-			break;
-		    }
-		    p_chk++;
-		}
-		if (!p_node)
-		    goto do_node;
-		if (h <= ERTS_TOPOLOGY_PROCESSOR_NODE)
-		    return ERTS_INIT_CPU_TOPOLOGY_INVALID_HIERARCHY;
-		idseqp = &cput->processor_node;
-		h = ERTS_TOPOLOGY_PROCESSOR_NODE;
-	    }
-	    break;
-	default:
-	    return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_TYPE;
-	}
-	res = get_cput_id_seq(idseqp, &c);
-	if (res != ERTS_INIT_CPU_TOPOLOGY_OK)
-		return res;
-    }
-
-    if (cput->logical.used < 1)
-	return ERTS_INIT_CPU_TOPOLOGY_MISSING_LID;
-
-    if (*c == ':') {
-	c++;
-    }
-
-    if (cput->thread.used != 1
-	&& cput->thread.used != cput->logical.used)
-	return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE;
-    if (cput->core.used != 1
-	&& cput->core.used != cput->logical.used)
-	return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE;
-    if (cput->processor_node.used != 1
-	&& cput->processor_node.used != cput->logical.used)
-	return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE;
-    if (cput->processor.used != 1
-	&& cput->processor.used != cput->logical.used)
-	return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE;
-    if (cput->node.used != 1
-	&& cput->node.used != cput->logical.used)
-	return ERTS_INIT_CPU_TOPOLOGY_INVALID_ID_RANGE;
-
-    *str = c;
-    return ERTS_INIT_CPU_TOPOLOGY_OK;
-}
-
-static int
-verify_topology(erts_cpu_topology_t *cpudata, int size)
-{
-    if (size > 0) {
-	int *logical;
-	int node, processor, no_nodes, i;
-
-	/* Verify logical ids */
-	logical = erts_alloc(ERTS_ALC_T_TMP, sizeof(int)*size);
-
-	for (i = 0; i < size; i++)
-	    logical[i] = cpudata[i].logical;
-
-	qsort(logical, size, sizeof(int), int_cmp);
-	for (i = 0; i < size-1; i++) {
-	    if (logical[i] == logical[i+1]) {
-		erts_free(ERTS_ALC_T_TMP, logical);
-		return ERTS_INIT_CPU_TOPOLOGY_NOT_UNIQUE_LIDS;
-	    }
-	}
-
-	erts_free(ERTS_ALC_T_TMP, logical);
-
-	qsort(cpudata, size, sizeof(erts_cpu_topology_t), processor_order_cmp);
-
-	/* Verify unique entities */
-
-	for (i = 1; i < size; i++) {
-	    if (cpudata[i-1].processor == cpudata[i].processor
-		&& cpudata[i-1].node == cpudata[i].node
-		&& (cpudata[i-1].processor_node
-		    == cpudata[i].processor_node)
-		&& cpudata[i-1].core == cpudata[i].core
-		&& cpudata[i-1].thread == cpudata[i].thread) {
-		return ERTS_INIT_CPU_TOPOLOGY_NOT_UNIQUE_ENTITIES;
-	    }
-	}
-
-	/* Verify numa nodes */
-	node = cpudata[0].node;
-	processor = cpudata[0].processor;
-	no_nodes = cpudata[0].node < 0 && cpudata[0].processor_node < 0;
-	for (i = 1; i < size; i++) {
-	    if (no_nodes) {
-		if (cpudata[i].node >= 0 || cpudata[i].processor_node >= 0)
-		    return ERTS_INIT_CPU_TOPOLOGY_INVALID_NODES;
-	    }
-	    else {
-		if (cpudata[i].processor == processor && cpudata[i].node != node)
-		    return ERTS_INIT_CPU_TOPOLOGY_INVALID_NODES;
-		node = cpudata[i].node;
-		processor = cpudata[i].processor;
-		if (node >= 0 && cpudata[i].processor_node >= 0)
-		    return ERTS_INIT_CPU_TOPOLOGY_INVALID_NODES;
-		if (node < 0 && cpudata[i].processor_node < 0)
-		    return ERTS_INIT_CPU_TOPOLOGY_INVALID_NODES;
-	    }
-	}
-    }
-
-    return ERTS_INIT_CPU_TOPOLOGY_OK;
-}
-
-int
-erts_init_cpu_topology(char *topology_str)
-{
-    ErtsCpuTopEntry cput;
-    int need_size;
-    char *c;
-    int ix;
-    int error = ERTS_INIT_CPU_TOPOLOGY_OK;
-
-    if (user_cpudata)
-	erts_free(ERTS_ALC_T_CPUDATA, user_cpudata);
-    user_cpudata_size = 10;
-
-    user_cpudata = erts_alloc(ERTS_ALC_T_CPUDATA,
-			      (sizeof(erts_cpu_topology_t)
-			       * user_cpudata_size));
-
-    init_cpu_top_entry(&cput);
-
-    ix = 0;
-    need_size = 0;
-
-    c = topology_str;
-    if (*c == '\0') {
-	error = ERTS_INIT_CPU_TOPOLOGY_MISSING;
-	goto fail;
-    }
-    do {
-	int r;
-	error = get_cput_entry(&cput, &c);
-	if (error != ERTS_INIT_CPU_TOPOLOGY_OK)
-	    goto fail;
-	need_size += cput.logical.used;
-	if (user_cpudata_size < need_size) {
-	    user_cpudata_size = need_size + 10;
-	    user_cpudata = erts_realloc(ERTS_ALC_T_CPUDATA,
-					user_cpudata,
-					(sizeof(erts_cpu_topology_t)
-					 * user_cpudata_size));
-	}
-
-	ASSERT(cput.thread.used == 1
-	       || cput.thread.used == cput.logical.used);
-	ASSERT(cput.core.used == 1
-	       || cput.core.used == cput.logical.used);
-	ASSERT(cput.processor_node.used == 1
-	       || cput.processor_node.used == cput.logical.used);
-	ASSERT(cput.processor.used == 1
-	       || cput.processor.used == cput.logical.used);
-	ASSERT(cput.node.used == 1
-	       || cput.node.used == cput.logical.used);
-
-	for (r = 0; r < cput.logical.used; r++) {
-	    user_cpudata[ix].logical = cput.logical.id[r];
-	    user_cpudata[ix].thread =
-		cput.thread.id[cput.thread.used == 1 ? 0 : r];
-	    user_cpudata[ix].core =
-		cput.core.id[cput.core.used == 1 ? 0 : r];
-	    user_cpudata[ix].processor_node =
-		cput.processor_node.id[cput.processor_node.used == 1 ? 0 : r];
-	    user_cpudata[ix].processor =
-		cput.processor.id[cput.processor.used == 1 ? 0 : r];
-	    user_cpudata[ix].node =
-		cput.node.id[cput.node.used == 1 ? 0 : r];
-	    ix++;
-	}
-    } while (*c != '\0');
-
-    if (user_cpudata_size != ix) {
-	user_cpudata_size = ix;
-	user_cpudata = erts_realloc(ERTS_ALC_T_CPUDATA,
-				    user_cpudata,
-				    (sizeof(erts_cpu_topology_t)
-				     * user_cpudata_size));
-    }
-
-    error = verify_topology(user_cpudata, user_cpudata_size);
-    if (error == ERTS_INIT_CPU_TOPOLOGY_OK) {
-	destroy_cpu_top_entry(&cput);
-	return ERTS_INIT_CPU_TOPOLOGY_OK;
-    }
-
- fail:
-    if (user_cpudata)
-	erts_free(ERTS_ALC_T_CPUDATA, user_cpudata);
-    user_cpudata_size = 0;
-    destroy_cpu_top_entry(&cput);
-    return error;
-}
-
-#define ERTS_GET_CPU_TOPOLOGY_ERROR		-1
-#define ERTS_GET_USED_CPU_TOPOLOGY		0
-#define ERTS_GET_DETECTED_CPU_TOPOLOGY		1
-#define ERTS_GET_DEFINED_CPU_TOPOLOGY		2
-
-static Eterm get_cpu_topology_term(Process *c_p, int type);
-
-Eterm
-erts_set_cpu_topology(Process *c_p, Eterm term)
-{
-    erts_cpu_topology_t *cpudata = NULL;
-    int cpudata_size = 0;
-    Eterm res;
-
-    erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx);
-    res = get_cpu_topology_term(c_p, ERTS_GET_USED_CPU_TOPOLOGY);
-    if (term == am_undefined) {
-	if (user_cpudata)
-	    erts_free(ERTS_ALC_T_CPUDATA, user_cpudata);
-	user_cpudata = NULL;
-	user_cpudata_size = 0;
-
-	if (cpu_bind_order != ERTS_CPU_BIND_NONE && system_cpudata) {
-	    cpudata_size = system_cpudata_size;
-	    cpudata = erts_alloc(ERTS_ALC_T_TMP,
-				 (sizeof(erts_cpu_topology_t)
-				  * cpudata_size));
-
-	    sys_memcpy((void *) cpudata,
-		       (void *) system_cpudata,
-		       sizeof(erts_cpu_topology_t)*cpudata_size);
-	}
-    }
-    else if (is_not_list(term)) {
-    error:
-	res = THE_NON_VALUE;
-	goto done;
-    }
-    else {
-	Eterm list = term;
-	int ix = 0;
-
-	cpudata_size = 100;
-	cpudata = erts_alloc(ERTS_ALC_T_TMP,
-			     (sizeof(erts_cpu_topology_t)
-			      * cpudata_size));
-
-	while (is_list(list)) {
-	    Eterm *lp = list_val(list);
-	    Eterm cpu = CAR(lp);
-	    Eterm* tp;
-	    Sint id;
-		
-	    if (is_not_tuple(cpu))
-		goto error;
-
-	    tp = tuple_val(cpu);
-
-	    if (arityval(tp[0]) != 7 || tp[1] != am_cpu)
-		goto error;
-
-	    if (ix >= cpudata_size) {
-		cpudata_size += 100;
-		cpudata = erts_realloc(ERTS_ALC_T_TMP,
-				       cpudata,
-				       (sizeof(erts_cpu_topology_t)
-					* cpudata_size));
-	    }
-
-	    id = signed_val(tp[2]);
-	    if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id)
-		goto error;
-	    cpudata[ix].node = (int) id;
-
-	    id = signed_val(tp[3]);
-	    if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id)
-		goto error;
-	    cpudata[ix].processor = (int) id;
-
-	    id = signed_val(tp[4]);
-	    if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id)
-		goto error;
-	    cpudata[ix].processor_node = (int) id;
-
-	    id = signed_val(tp[5]);
-	    if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id)
-		goto error;
-	    cpudata[ix].core = (int) id;
-
-	    id = signed_val(tp[6]);
-	    if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id)
-		goto error;
-	    cpudata[ix].thread = (int) id;
-
-	    id = signed_val(tp[7]);
-	    if (id < -1 || ERTS_MAX_CPU_TOPOLOGY_ID < id)
-		goto error;
-	    cpudata[ix].logical = (int) id;
-
-	    list = CDR(lp);
-	    ix++;
-	}
-
-	if (is_not_nil(list))
-	    goto error;
-	
-	cpudata_size = ix;
-
-	if (ERTS_INIT_CPU_TOPOLOGY_OK != verify_topology(cpudata, cpudata_size))
-	    goto error;
-
-	if (user_cpudata_size != cpudata_size) {
-	    if (user_cpudata)
-		erts_free(ERTS_ALC_T_CPUDATA, user_cpudata);
-	    user_cpudata = erts_alloc(ERTS_ALC_T_CPUDATA,
-				      sizeof(erts_cpu_topology_t)*cpudata_size);
-	    user_cpudata_size = cpudata_size;
-	}
-
-	sys_memcpy((void *) user_cpudata,
-		   (void *) cpudata,
-		   sizeof(erts_cpu_topology_t)*cpudata_size);
-    }
-
-    make_reader_groups_map(NULL);
-
-    signal_schedulers_bind_change(cpudata, cpudata_size);
-
- done:
-    erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx);
-
-    if (cpudata)
-	erts_free(ERTS_ALC_T_TMP, cpudata);
-
-    return res;
-}
-
-static Eterm
-bound_schedulers_term(ErtsCpuBindOrder order)
-{
-    switch (order) {
-    case ERTS_CPU_BIND_SPREAD: {
-	ERTS_DECL_AM(spread);
-	return AM_spread;
-    }
-    case ERTS_CPU_BIND_PROCESSOR_SPREAD: {
-	ERTS_DECL_AM(processor_spread);
-	return AM_processor_spread;
-    }
-    case ERTS_CPU_BIND_THREAD_SPREAD: {
-	ERTS_DECL_AM(thread_spread);
-	return AM_thread_spread;
-    }
-    case ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD: {
-	ERTS_DECL_AM(thread_no_node_processor_spread);
-	return AM_thread_no_node_processor_spread;
-    }
-    case ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD: {
-	ERTS_DECL_AM(no_node_processor_spread);
-	return AM_no_node_processor_spread;
-    }
-    case ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD: {
-	ERTS_DECL_AM(no_node_thread_spread);
-	return AM_no_node_thread_spread;
-    }
-    case ERTS_CPU_BIND_NO_SPREAD: {
-	ERTS_DECL_AM(no_spread);
-	return AM_no_spread;
-    }
-    case ERTS_CPU_BIND_NONE: {
-	ERTS_DECL_AM(unbound);
-	return AM_unbound;
-    }
-    default:
-	ASSERT(0);
-	return THE_NON_VALUE;
-    }
-}
-
-Eterm
-erts_bound_schedulers_term(Process *c_p)
-{
-    ErtsCpuBindOrder order;
-    erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx);
-    order = cpu_bind_order;
-    erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx);
-    return bound_schedulers_term(order);
-}
-
-static void
-create_tmp_cpu_topology_copy(erts_cpu_topology_t **cpudata, int *cpudata_size)
-{
-    if (user_cpudata) {
-	*cpudata_size = user_cpudata_size;
-	*cpudata = erts_alloc(ERTS_ALC_T_TMP,
-			      (sizeof(erts_cpu_topology_t)
-			       * (*cpudata_size)));
-	sys_memcpy((void *) *cpudata,
-		   (void *) user_cpudata,
-		   sizeof(erts_cpu_topology_t)*(*cpudata_size));
-    }
-    else if (system_cpudata) {
-	*cpudata_size = system_cpudata_size;
-	*cpudata = erts_alloc(ERTS_ALC_T_TMP,
-			      (sizeof(erts_cpu_topology_t)
-			       * (*cpudata_size)));
-	sys_memcpy((void *) *cpudata,
-		   (void *) system_cpudata,
-		   sizeof(erts_cpu_topology_t)*(*cpudata_size));
-    }
-    else {
-	*cpudata = NULL;
-	*cpudata_size = 0;
-    }
-}
-
-static void
-destroy_tmp_cpu_topology_copy(erts_cpu_topology_t *cpudata)
-{
-    if (cpudata)
-	erts_free(ERTS_ALC_T_TMP, cpudata);
-}
-
-Eterm
-erts_bind_schedulers(Process *c_p, Eterm how)
-{
-    Eterm res;
-    erts_cpu_topology_t *cpudata;
-    int cpudata_size;
-    ErtsCpuBindOrder old_cpu_bind_order;
-
-    erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx);
-
-    if (erts_bind_to_cpu(erts_cpuinfo, -1) == -ENOTSUP) {
-	ERTS_BIF_PREP_ERROR(res, c_p, EXC_NOTSUP);
-    }
-    else {
-
-	old_cpu_bind_order = cpu_bind_order;
-
-	if (ERTS_IS_ATOM_STR("default_bind", how))
-	    cpu_bind_order = ERTS_CPU_BIND_DEFAULT_BIND;
-	else if (ERTS_IS_ATOM_STR("spread", how))
-	    cpu_bind_order = ERTS_CPU_BIND_SPREAD;
-	else if (ERTS_IS_ATOM_STR("processor_spread", how))
-	    cpu_bind_order = ERTS_CPU_BIND_PROCESSOR_SPREAD;
-	else if (ERTS_IS_ATOM_STR("thread_spread", how))
-	    cpu_bind_order = ERTS_CPU_BIND_THREAD_SPREAD;
-	else if (ERTS_IS_ATOM_STR("thread_no_node_processor_spread", how))
-	    cpu_bind_order = ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD;
-	else if (ERTS_IS_ATOM_STR("no_node_processor_spread", how))
-	    cpu_bind_order = ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD;
-	else if (ERTS_IS_ATOM_STR("no_node_thread_spread", how))
-	    cpu_bind_order = ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD;
-	else if (ERTS_IS_ATOM_STR("no_spread", how))
-	    cpu_bind_order = ERTS_CPU_BIND_NO_SPREAD;
-	else if (ERTS_IS_ATOM_STR("unbound", how))
-	    cpu_bind_order = ERTS_CPU_BIND_NONE;
-	else {
-	    cpu_bind_order = old_cpu_bind_order;
-	    ERTS_BIF_PREP_ERROR(res, c_p, BADARG);
-	    goto done;
-	}
-
-	create_tmp_cpu_topology_copy(&cpudata, &cpudata_size);
-
-	if (!cpudata) {
-	    cpu_bind_order = old_cpu_bind_order;
-	    ERTS_BIF_PREP_ERROR(res, c_p, BADARG);
-	    goto done;
-	}
-
-	signal_schedulers_bind_change(cpudata, cpudata_size);
-
-	destroy_tmp_cpu_topology_copy(cpudata);
-    
-	res = bound_schedulers_term(old_cpu_bind_order);
-    }
-
- done:
-
-    erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx);
-
-    return res;
-}
-
-Eterm
-erts_fake_scheduler_bindings(Process *p, Eterm how)
-{
-    ErtsCpuBindOrder fake_cpu_bind_order;
-    erts_cpu_topology_t *cpudata;
-    int cpudata_size;
-    Eterm res;
-
-    if (ERTS_IS_ATOM_STR("default_bind", how))
-	fake_cpu_bind_order = ERTS_CPU_BIND_DEFAULT_BIND;
-    else if (ERTS_IS_ATOM_STR("spread", how))
-	fake_cpu_bind_order = ERTS_CPU_BIND_SPREAD;
-    else if (ERTS_IS_ATOM_STR("processor_spread", how))
-	fake_cpu_bind_order = ERTS_CPU_BIND_PROCESSOR_SPREAD;
-    else if (ERTS_IS_ATOM_STR("thread_spread", how))
-	fake_cpu_bind_order = ERTS_CPU_BIND_THREAD_SPREAD;
-    else if (ERTS_IS_ATOM_STR("thread_no_node_processor_spread", how))
-	fake_cpu_bind_order = ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD;
-    else if (ERTS_IS_ATOM_STR("no_node_processor_spread", how))
-	fake_cpu_bind_order = ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD;
-    else if (ERTS_IS_ATOM_STR("no_node_thread_spread", how))
-	fake_cpu_bind_order = ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD;
-    else if (ERTS_IS_ATOM_STR("no_spread", how))
-	fake_cpu_bind_order = ERTS_CPU_BIND_NO_SPREAD;
-    else if (ERTS_IS_ATOM_STR("unbound", how))
-	fake_cpu_bind_order = ERTS_CPU_BIND_NONE;
-    else {
-	ERTS_BIF_PREP_ERROR(res, p, BADARG);
-	return res;
-    }
-
-    erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx);
-    create_tmp_cpu_topology_copy(&cpudata, &cpudata_size);
-    erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx);
-
-    if (!cpudata || fake_cpu_bind_order == ERTS_CPU_BIND_NONE)
-	ERTS_BIF_PREP_RET(res, am_false);
-    else {
-	int i;
-	Eterm *hp;
-	
-	cpu_bind_order_sort(cpudata, cpudata_size, fake_cpu_bind_order, 1);
-
-#ifdef ERTS_FAKE_SCHED_BIND_PRINT_SORTED_CPU_DATA
-
-	erts_fprintf(stderr, "node:          ");
-	for (i = 0; i < cpudata_size; i++)
-	    erts_fprintf(stderr, " %2d", cpudata[i].node);
-	erts_fprintf(stderr, "\n");
-	erts_fprintf(stderr, "processor:     ");
-	for (i = 0; i < cpudata_size; i++)
-	    erts_fprintf(stderr, " %2d", cpudata[i].processor);
-	erts_fprintf(stderr, "\n");
-	if (fake_cpu_bind_order != ERTS_CPU_BIND_THREAD_NO_NODE_PROCESSOR_SPREAD
-	    && fake_cpu_bind_order != ERTS_CPU_BIND_NO_NODE_PROCESSOR_SPREAD
-	    && fake_cpu_bind_order != ERTS_CPU_BIND_NO_NODE_THREAD_SPREAD) {
-	    erts_fprintf(stderr, "processor_node:");
-	    for (i = 0; i < cpudata_size; i++)
-		erts_fprintf(stderr, " %2d", cpudata[i].processor_node);
-	    erts_fprintf(stderr, "\n");
-	}
-	erts_fprintf(stderr, "core:          ");
-	for (i = 0; i < cpudata_size; i++)
-	    erts_fprintf(stderr, " %2d", cpudata[i].core);
-	erts_fprintf(stderr, "\n");
-	erts_fprintf(stderr, "thread:        ");
-	for (i = 0; i < cpudata_size; i++)
-	    erts_fprintf(stderr, " %2d", cpudata[i].thread);
-	erts_fprintf(stderr, "\n");
-	erts_fprintf(stderr, "logical:       ");
-	for (i = 0; i < cpudata_size; i++)
-	    erts_fprintf(stderr, " %2d", cpudata[i].logical);
-	erts_fprintf(stderr, "\n");
-#endif
-
-	hp = HAlloc(p, cpudata_size+1);
-	ERTS_BIF_PREP_RET(res, make_tuple(hp));
-	*hp++ = make_arityval((Uint) cpudata_size);
-	for (i = 0; i < cpudata_size; i++)
-	    *hp++ = make_small((Uint) cpudata[i].logical);
-    }
-
-    destroy_tmp_cpu_topology_copy(cpudata);
-
-    return res;
-}
-
-Eterm
-erts_get_schedulers_binds(Process *c_p)
-{
-    int ix;
-    ERTS_DECL_AM(unbound);
-    Eterm *hp = HAlloc(c_p, erts_no_schedulers+1);
-    Eterm res = make_tuple(hp);
-
-    *(hp++) = make_arityval(erts_no_schedulers);
-    erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx);
-    for (ix = 1; ix <= erts_no_schedulers; ix++)
-	*(hp++) = (scheduler2cpu_map[ix].bound_id >= 0
-		   ? make_small(scheduler2cpu_map[ix].bound_id)
-		   : AM_unbound);
-    erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx);
-    return res;
-}
-
-static Eterm
-bld_topology_term(Eterm **hpp,
-		  Uint *hszp,
-		  erts_cpu_topology_t *cpudata,
-		  int size)
-{
-    Eterm res = NIL;
-    int i;
-
-    if (size == 0)
-	return am_undefined;
-
-    for (i = size-1; i >= 0; i--) {
-	res = erts_bld_cons(hpp,
-			    hszp,
-			    erts_bld_tuple(hpp,
-					   hszp,
-					   7,
-					   am_cpu,
-					   make_small(cpudata[i].node),
-					   make_small(cpudata[i].processor),
-					   make_small(cpudata[i].processor_node),
-					   make_small(cpudata[i].core),
-					   make_small(cpudata[i].thread),
-					   make_small(cpudata[i].logical)),
-			    res);
-    }
-    return res;
-}
-
-static Eterm
-get_cpu_topology_term(Process *c_p, int type)
-{
-#ifdef DEBUG
-    Eterm *hp_end;
-#endif
-    Eterm *hp;
-    Uint hsz;
-    Eterm res = THE_NON_VALUE;
-    erts_cpu_topology_t *cpudata = NULL;
-    int size = 0;
-
-    switch (type) {
-    case ERTS_GET_USED_CPU_TOPOLOGY:
-	if (user_cpudata)
-	    goto defined;
-	else
-	    goto detected;
-    case ERTS_GET_DETECTED_CPU_TOPOLOGY:
-    detected:
-	if (!system_cpudata)
-	    res = am_undefined;
-	else {
-	    size = system_cpudata_size;
-	    cpudata = erts_alloc(ERTS_ALC_T_TMP,
-				 (sizeof(erts_cpu_topology_t)
-				  * size));
-	    sys_memcpy((void *) cpudata,
-		       (void *) system_cpudata,
-		       sizeof(erts_cpu_topology_t)*size);
-	}
-	break;
-    case ERTS_GET_DEFINED_CPU_TOPOLOGY:
-    defined:
-	if (!user_cpudata)
-	    res = am_undefined;
-	else {
-	    size = user_cpudata_size;
-	    cpudata = user_cpudata;
-	}
-	break;
-    default:
-	erl_exit(ERTS_ABORT_EXIT, "Bad cpu topology type: %d\n", type);
-	break;
-    }
-
-    if (res == am_undefined) {
-	ASSERT(!cpudata);
-	return res;
-    }
-
-    hsz = 0;
-
-    bld_topology_term(NULL, &hsz,
-		      cpudata, size);
-
-    hp = HAlloc(c_p, hsz);
-
-#ifdef DEBUG
-    hp_end = hp + hsz;
-#endif
-
-    res = bld_topology_term(&hp, NULL,
-			    cpudata, size);
-
-    ASSERT(hp_end == hp);
-
-    if (cpudata && cpudata != system_cpudata && cpudata != user_cpudata)
-	erts_free(ERTS_ALC_T_TMP, cpudata);
-
-    return res;
-}
-
-Eterm
-erts_get_cpu_topology_term(Process *c_p, Eterm which)
-{
-    Eterm res;
-    int type;
-    erts_smp_rwmtx_rlock(&erts_cpu_bind_rwmtx);
-    if (ERTS_IS_ATOM_STR("used", which))
-	type = ERTS_GET_USED_CPU_TOPOLOGY;
-    else if (ERTS_IS_ATOM_STR("detected", which))
-	type = ERTS_GET_DETECTED_CPU_TOPOLOGY;
-    else if (ERTS_IS_ATOM_STR("defined", which))
-	type = ERTS_GET_DEFINED_CPU_TOPOLOGY;
-    else
-	type = ERTS_GET_CPU_TOPOLOGY_ERROR;
-    if (type == ERTS_GET_CPU_TOPOLOGY_ERROR)
-	res = THE_NON_VALUE;
-    else
-	res = get_cpu_topology_term(c_p, type);
-    erts_smp_rwmtx_runlock(&erts_cpu_bind_rwmtx);
-    return res;
-}
-
-static void
-early_cpu_bind_init(void)
-{
-    user_cpudata = NULL;
-    user_cpudata_size = 0;
-
-    system_cpudata_size = erts_get_cpu_topology_size(erts_cpuinfo);
-    system_cpudata = erts_alloc(ERTS_ALC_T_CPUDATA,
-				(sizeof(erts_cpu_topology_t)
-				 * system_cpudata_size));
-
-    cpu_bind_order = ERTS_CPU_BIND_UNDEFINED;
-
-    reader_groups_available_cpu_check = 1;
-    reader_groups_logical_processors = 0;
-    reader_groups_map_size = 0;
-    reader_groups_map = NULL;
-
-    if (!erts_get_cpu_topology(erts_cpuinfo, system_cpudata)
-	|| ERTS_INIT_CPU_TOPOLOGY_OK != verify_topology(system_cpudata,
-							system_cpudata_size)) {
-	erts_free(ERTS_ALC_T_CPUDATA, system_cpudata);
-	system_cpudata = NULL;
-	system_cpudata_size = 0;
-    }
-}
-
-static void
-late_cpu_bind_init(void)
-{
-    int ix;
-
-    erts_smp_rwmtx_init(&erts_cpu_bind_rwmtx, "cpu_bind");
-
-    scheduler2cpu_map = erts_alloc(ERTS_ALC_T_CPUDATA,
-				   (sizeof(ErtsCpuBindData)
-				    * (erts_no_schedulers+1)));
-    for (ix = 1; ix <= erts_no_schedulers; ix++) {
-	scheduler2cpu_map[ix].bind_id = -1;
-	scheduler2cpu_map[ix].bound_id = -1;
-    }
-
-    if (cpu_bind_order == ERTS_CPU_BIND_UNDEFINED) {
-	int ncpus = erts_get_cpu_configured(erts_cpuinfo);
-	if (ncpus < 1 || erts_no_schedulers < ncpus)
-	    cpu_bind_order = ERTS_CPU_BIND_NONE;
-	else
-	    cpu_bind_order = ((system_cpudata || user_cpudata)
-			      && (erts_bind_to_cpu(erts_cpuinfo, -1) != -ENOTSUP)
-			      ? ERTS_CPU_BIND_DEFAULT_BIND
-			      : ERTS_CPU_BIND_NONE);
-    }
-
-    make_reader_groups_map(NULL);
-
-    if (cpu_bind_order != ERTS_CPU_BIND_NONE) {
-	erts_cpu_topology_t *cpudata;
-	int cpudata_size;
-	create_tmp_cpu_topology_copy(&cpudata, &cpudata_size);
-	signal_schedulers_bind_change(cpudata, cpudata_size);
-	destroy_tmp_cpu_topology_copy(cpudata);
-    }
-}
-
-int
-erts_update_cpu_info(void)
-{
-    int changed;
-    erts_smp_rwmtx_rwlock(&erts_cpu_bind_rwmtx);
-    changed = erts_cpu_info_update(erts_cpuinfo);
-    if (changed) {
-	erts_cpu_topology_t *cpudata;
-	int cpudata_size;
-
-	if (system_cpudata)
-	    erts_free(ERTS_ALC_T_CPUDATA, system_cpudata);
-
-	system_cpudata_size = erts_get_cpu_topology_size(erts_cpuinfo);
-	if (!system_cpudata_size)
-	    system_cpudata = NULL;
-	else {
-	    system_cpudata = erts_alloc(ERTS_ALC_T_CPUDATA,
-					(sizeof(erts_cpu_topology_t)
-					 * system_cpudata_size));
-
-	    if (!erts_get_cpu_topology(erts_cpuinfo, system_cpudata)
-		|| (ERTS_INIT_CPU_TOPOLOGY_OK
-		    != verify_topology(system_cpudata,
-				       system_cpudata_size))) {
-		erts_free(ERTS_ALC_T_CPUDATA, system_cpudata);
-		system_cpudata = NULL;
-		system_cpudata_size = 0;
-	    }
-	}
-
-	create_tmp_cpu_topology_copy(&cpudata, &cpudata_size);
-	signal_schedulers_bind_change(cpudata, cpudata_size);
-	destroy_tmp_cpu_topology_copy(cpudata);
-    }
-    erts_smp_rwmtx_rwunlock(&erts_cpu_bind_rwmtx);
-    return changed;
-}
-
 #ifdef ERTS_SMP
 
 static void
@@ -7073,7 +5091,7 @@ Process *schedule(Process *p, int calls)
 	    }
 	    if ((rq->flags & ERTS_RUNQ_FLG_CHK_CPU_BIND)
 		|| erts_smp_atomic_read(&esdp->chk_cpu_bind)) {
-		check_cpu_bind(esdp);
+		erts_sched_check_cpu_bind(esdp);
 	    }
 	}