erts: Pass thread progress data where possible

The poll thread does a lot of waking up and then going
back to sleep. A large part of the waking up is managing
thread progress and a large part of that was using thread
specific data to get the thread progress data pointer.
With this refactor the tpd is passed to each of the functions
which greatly decreases the number of ethr_get_tsd calls
which in turn halves the CPU usage of the poller thread in
certain scenarios.

11 files changed, 109 insertions, 80 deletions

diff --git a/erts/emulator/beam/erl_async.c b/erts/emulator/beam/erl_async.c
index 605a2b3461..44655ad5df 100644
--- a/erts/emulator/beam/erl_async.c
+++ b/erts/emulator/beam/erl_async.c
@@ -336,7 +336,7 @@ static ERTS_INLINE ErtsAsync *async_get(ErtsThrQ_t *q,
 	    case ERTS_THR_Q_NEED_THR_PRGR:
 	    {
 		ErtsThrPrgrVal prgr = erts_thr_q_need_thr_progress(q);
-		erts_thr_progress_wakeup(NULL, prgr);
+		erts_thr_progress_wakeup(erts_thr_prgr_data(NULL), prgr);
 		/*
 		 * We do no dequeue finalizing in hope that a new async
 		 * job will arrive before we are woken due to thread
diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c
index 57c6c10c7f..ff8b9fd567 100644
--- a/erts/emulator/beam/erl_init.c
+++ b/erts/emulator/beam/erl_init.c
@@ -2358,8 +2358,8 @@ system_cleanup(int flush_async)
 	     * The exiting thread might be waiting for
 	     * us to block; need to update status...
 	     */
-	    erts_thr_progress_active(NULL, 0);
-	    erts_thr_progress_prepare_wait(NULL);
+	    erts_thr_progress_active(erts_thr_prgr_data(NULL), 0);
+	    erts_thr_progress_prepare_wait(erts_thr_prgr_data(NULL));
 	}
 	/* Wait forever... */
 	while (1)
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 0f7f1598fd..919d9f9edf 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -1646,7 +1646,7 @@ haw_thr_prgr_wakeup(ErtsAuxWorkData *awdp, ErtsThrPrgrVal val)
 	    awdp->latest_wakeup = val;
 	    haw_chk_later_cleanup_op_wakeup(awdp, val);
 	}
-	erts_thr_progress_wakeup(awdp->esdp, val);
+	erts_thr_progress_wakeup(erts_thr_prgr_data(awdp->esdp), val);
     }
 }
 
@@ -1656,7 +1656,7 @@ haw_thr_prgr_soft_wakeup(ErtsAuxWorkData *awdp, ErtsThrPrgrVal val)
     if (erts_thr_progress_cmp(val, awdp->latest_wakeup) > 0) {
 	awdp->latest_wakeup = val;
 	haw_chk_later_cleanup_op_wakeup(awdp, val);
-	erts_thr_progress_wakeup(awdp->esdp, val);
+	erts_thr_progress_wakeup(erts_thr_prgr_data(awdp->esdp), val);
     }
 }
 
@@ -1670,7 +1670,7 @@ haw_thr_prgr_later_cleanup_op_wakeup(ErtsAuxWorkData *awdp, ErtsThrPrgrVal val,
 	else {
 	    awdp->latest_wakeup = val;
 	    awdp->later_op.size = thr_prgr_later_cleanup_op_threshold;
-	    erts_thr_progress_wakeup(awdp->esdp, val);
+	    erts_thr_progress_wakeup(erts_thr_prgr_data(awdp->esdp), val);
 	}
     }
 }
@@ -3066,6 +3066,7 @@ aux_thread(void *unused)
     ErtsSchedulerSleepInfo *ssi = ERTS_SCHED_SLEEP_INFO_IX(-1);
     erts_aint32_t aux_work;
     ErtsThrPrgrCallbacks callbacks;
+    ErtsThrPrgrData *tpd;
     int thr_prgr_active = 1;
     ERTS_MSACC_DECLARE_CACHE();
 
@@ -3087,12 +3088,12 @@ aux_thread(void *unused)
     callbacks.wait = thr_prgr_wait;
     callbacks.finalize_wait = thr_prgr_fin_wait;
 
-    erts_thr_progress_register_managed_thread(NULL, &callbacks, 1);
+    tpd = erts_thr_progress_register_managed_thread(NULL, &callbacks, 1);
     init_aux_work_data(awdp, NULL, NULL);
     awdp->ssi = ssi;
 
 #if ERTS_POLL_USE_FALLBACK
-    ssi->psi = erts_create_pollset_thread(-1);
+    ssi->psi = erts_create_pollset_thread(-1, tpd);
 #endif
 
     sched_prep_spin_wait(ssi);
@@ -3105,11 +3106,11 @@ aux_thread(void *unused)
 	aux_work = erts_atomic32_read_acqb(&ssi->aux_work);
 	if (aux_work) {
 	    if (!thr_prgr_active)
-		erts_thr_progress_active(NULL, thr_prgr_active = 1);
+		erts_thr_progress_active(tpd, thr_prgr_active = 1);
 	    aux_work = handle_aux_work(awdp, aux_work, 1);
             ERTS_MSACC_UPDATE_CACHE();
-	    if (aux_work && erts_thr_progress_update(NULL))
-		erts_thr_progress_leader_update(NULL);
+	    if (aux_work && erts_thr_progress_update(tpd))
+		erts_thr_progress_leader_update(tpd);
 	}
 
 	if (!aux_work) {
@@ -3120,7 +3121,7 @@ aux_thread(void *unused)
 #endif
 
 	    if (thr_prgr_active)
-		erts_thr_progress_active(NULL, thr_prgr_active = 0);
+		erts_thr_progress_active(tpd, thr_prgr_active = 0);
 
 #if ERTS_POLL_USE_FALLBACK
 
@@ -3136,7 +3137,7 @@ aux_thread(void *unused)
 		}
             }
 #else
-            erts_thr_progress_prepare_wait(NULL);
+            erts_thr_progress_prepare_wait(tpd);
 
 	    flgs = sched_spin_wait(ssi, 0);
 
@@ -3153,7 +3154,7 @@ aux_thread(void *unused)
                     ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_OTHER);
 		}
             }
-            erts_thr_progress_finalize_wait(NULL);
+            erts_thr_progress_finalize_wait(tpd);
 #endif
 	}
 
@@ -3171,7 +3172,8 @@ poll_thread(void *arg)
     erts_aint32_t aux_work;
     ErtsThrPrgrCallbacks callbacks;
     int thr_prgr_active = 1;
-    struct erts_poll_thread *psi = erts_create_pollset_thread(id);
+    struct erts_poll_thread *psi;
+    ErtsThrPrgrData *tpd;
     ERTS_MSACC_DECLARE_CACHE();
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
@@ -3192,9 +3194,12 @@ poll_thread(void *arg)
     callbacks.wait = thr_prgr_wait;
     callbacks.finalize_wait = thr_prgr_fin_wait;
 
-    erts_thr_progress_register_managed_thread(NULL, &callbacks, 0);
+    tpd = erts_thr_progress_register_managed_thread(NULL, &callbacks, 0);
     init_aux_work_data(awdp, NULL, NULL);
     awdp->ssi = ssi;
+
+    psi = erts_create_pollset_thread(id, tpd);
+
     ssi->psi = psi;
 
     sched_prep_spin_wait(ssi);
@@ -3207,16 +3212,16 @@ poll_thread(void *arg)
 	aux_work = erts_atomic32_read_acqb(&ssi->aux_work);
 	if (aux_work) {
 	    if (!thr_prgr_active)
-		erts_thr_progress_active(NULL, thr_prgr_active = 1);
+		erts_thr_progress_active(tpd, thr_prgr_active = 1);
 	    aux_work = handle_aux_work(awdp, aux_work, 1);
             ERTS_MSACC_UPDATE_CACHE();
-	    if (aux_work && erts_thr_progress_update(NULL))
-		erts_thr_progress_leader_update(NULL);
+	    if (aux_work && erts_thr_progress_update(tpd))
+		erts_thr_progress_leader_update(tpd);
 	}
 
 	if (!aux_work) {
 	    if (thr_prgr_active)
-		erts_thr_progress_active(NULL, thr_prgr_active = 0);
+		erts_thr_progress_active(tpd, thr_prgr_active = 0);
 
 	    flgs = sched_spin_wait(ssi, 0);
 
@@ -3286,13 +3291,13 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
         aux_work = erts_atomic32_read_acqb(&ssi->aux_work);
         if (aux_work && !ERTS_SCHEDULER_IS_DIRTY(esdp)) {
             if (!thr_prgr_active) {
-                erts_thr_progress_active(esdp, thr_prgr_active = 1);
+                erts_thr_progress_active(erts_thr_prgr_data(esdp), thr_prgr_active = 1);
                 sched_wall_time_change(esdp, 1);
             }
             aux_work = handle_aux_work(&esdp->aux_work_data, aux_work, 1);
             ERTS_MSACC_UPDATE_CACHE();
-            if (aux_work && erts_thr_progress_update(esdp))
-                erts_thr_progress_leader_update(esdp);
+            if (aux_work && erts_thr_progress_update(erts_thr_prgr_data(esdp)))
+                erts_thr_progress_leader_update(erts_thr_prgr_data(esdp));
         }
 
         if (aux_work) {
@@ -3301,7 +3306,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
                 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);
+                        erts_thr_progress_active(erts_thr_prgr_data(esdp), thr_prgr_active = 1);
                         sched_wall_time_change(esdp, 1);
                     }
                     erts_bump_timers(esdp->timer_wheel, current_time);
@@ -3321,17 +3326,17 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
             }
             if (do_timeout) {
                 if (!thr_prgr_active) {
-                    erts_thr_progress_active(esdp, thr_prgr_active = 1);
+                    erts_thr_progress_active(erts_thr_prgr_data(esdp), thr_prgr_active = 1);
                     sched_wall_time_change(esdp, 1);
                 }
             }
             else {
                 if (!ERTS_SCHEDULER_IS_DIRTY(esdp)) {
                     if (thr_prgr_active) {
-                        erts_thr_progress_active(esdp, thr_prgr_active = 0);
+                        erts_thr_progress_active(erts_thr_prgr_data(esdp), thr_prgr_active = 0);
                         sched_wall_time_change(esdp, 0);
                     }
-                    erts_thr_progress_prepare_wait(esdp);
+                    erts_thr_progress_prepare_wait(erts_thr_prgr_data(esdp));
                 }
 
                 flgs = sched_spin_wait(ssi, spincount);
@@ -3363,7 +3368,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
                     }
                 }
                 if (!ERTS_SCHEDULER_IS_DIRTY(esdp))
-                    erts_thr_progress_finalize_wait(esdp);
+                    erts_thr_progress_finalize_wait(erts_thr_prgr_data(esdp));
             }
             if (!ERTS_SCHEDULER_IS_DIRTY(esdp) && current_time >= timeout_time)
                 erts_bump_timers(esdp->timer_wheel, current_time);
@@ -3392,7 +3397,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
     if (ERTS_SCHEDULER_IS_DIRTY(esdp))
         dirty_sched_wall_time_change(esdp, working = 1);
     else if (!thr_prgr_active) {
-        erts_thr_progress_active(esdp, thr_prgr_active = 1);
+        erts_thr_progress_active(erts_thr_prgr_data(esdp), thr_prgr_active = 1);
         sched_wall_time_change(esdp, 1);
     }
 
@@ -7565,7 +7570,8 @@ suspend_scheduler(ErtsSchedulerData *esdp)
 
 		    if (aux_work|evacuate) {
 			if (!thr_prgr_active) {
-			    erts_thr_progress_active(esdp, thr_prgr_active = 1);
+			    erts_thr_progress_active(erts_thr_prgr_data(esdp),
+                                                     thr_prgr_active = 1);
 			    sched_wall_time_change(esdp, 1);
 			}
 			if (aux_work)
@@ -7573,8 +7579,8 @@ suspend_scheduler(ErtsSchedulerData *esdp)
 						       aux_work,
 						       1);
 
-			if (aux_work && erts_thr_progress_update(esdp))
-			    erts_thr_progress_leader_update(esdp);
+			if (aux_work && erts_thr_progress_update(erts_thr_prgr_data(esdp)))
+			    erts_thr_progress_leader_update(erts_thr_prgr_data(esdp));
 			if (evacuate) {
 			    erts_runq_lock(esdp->run_queue);
 			    evacuate_run_queue(esdp->run_queue, &sbp);
@@ -7593,18 +7599,18 @@ suspend_scheduler(ErtsSchedulerData *esdp)
                     if (!aux_work && current_time < timeout_time) {
                         /* go to sleep... */
                         if (thr_prgr_active) {
-                            erts_thr_progress_active(esdp, thr_prgr_active = 0);
+                            erts_thr_progress_active(erts_thr_prgr_data(esdp), thr_prgr_active = 0);
                             sched_wall_time_change(esdp, 0);
                         }
-                        erts_thr_progress_prepare_wait(NULL);
+                        erts_thr_progress_prepare_wait(erts_thr_prgr_data(NULL));
                         suspend_normal_scheduler_sleep(esdp);
-                        erts_thr_progress_finalize_wait(NULL);
+                        erts_thr_progress_finalize_wait(erts_thr_prgr_data(NULL));
                         current_time = erts_get_monotonic_time(esdp);
                     }
 
                     if (current_time >= timeout_time) {
                         if (!thr_prgr_active) {
-                            erts_thr_progress_active(esdp, thr_prgr_active = 1);
+                            erts_thr_progress_active(erts_thr_prgr_data(esdp), thr_prgr_active = 1);
                             sched_wall_time_change(esdp, 1);
                         }
                         erts_bump_timers(esdp->timer_wheel, current_time);
@@ -7661,7 +7667,7 @@ suspend_scheduler(ErtsSchedulerData *esdp)
             profile_scheduler(make_small(esdp->no), am_active);
 
         if (!thr_prgr_active) {
-            erts_thr_progress_active(esdp, thr_prgr_active = 1);
+            erts_thr_progress_active(erts_thr_prgr_data(esdp), thr_prgr_active = 1);
             sched_wall_time_change(esdp, 1);
         }
     }
@@ -9313,12 +9319,12 @@ Process *erts_schedule(ErtsSchedulerData *esdp, Process *p, int calls)
 		}
 	    }
 
-	    leader_update = erts_thr_progress_update(esdp);
+	    leader_update = erts_thr_progress_update(erts_thr_prgr_data(esdp));
 	    aux_work = erts_atomic32_read_acqb(&esdp->ssi->aux_work);
 	    if (aux_work | leader_update) {
 		erts_runq_unlock(rq);
 		if (leader_update)
-		    erts_thr_progress_leader_update(esdp);
+		    erts_thr_progress_leader_update(erts_thr_prgr_data(esdp));
 		if (aux_work)
 		    handle_aux_work(&esdp->aux_work_data, aux_work, 0);
 		erts_runq_lock(rq);
diff --git a/erts/emulator/beam/erl_thr_progress.c b/erts/emulator/beam/erl_thr_progress.c
index aa08eb40ec..20b05dc942 100644
--- a/erts/emulator/beam/erl_thr_progress.c
+++ b/erts/emulator/beam/erl_thr_progress.c
@@ -508,6 +508,10 @@ init_wakeup_request_array(ErtsThrPrgrVal *w)
     }
 }
 
+ErtsThrPrgrData *erts_thr_progress_data(void) {
+    return erts_tsd_get(erts_thr_prgr_data_key__);
+}
+
 void
 erts_thr_progress_register_unmanaged_thread(ErtsThrPrgrCallbacks *callbacks)
 {
@@ -551,7 +555,7 @@ erts_thr_progress_register_unmanaged_thread(ErtsThrPrgrCallbacks *callbacks)
 }
 
 
-void
+ErtsThrPrgrData *
 erts_thr_progress_register_managed_thread(ErtsSchedulerData *esdp,
 					  ErtsThrPrgrCallbacks *callbacks,
 					  int pref_wakeup)
@@ -630,6 +634,7 @@ erts_thr_progress_register_managed_thread(ErtsSchedulerData *esdp,
 		wakeup_managed(id);
     }
     callbacks->finalize_wait(callbacks->arg);
+    return tpd;
 }
 
 static ERTS_INLINE int
@@ -849,23 +854,22 @@ update(ErtsThrPrgrData *tpd)
 }
 
 int
-erts_thr_progress_update(ErtsSchedulerData *esdp)
+erts_thr_progress_update(ErtsThrPrgrData *tpd)
 {
-    return update(thr_prgr_data(esdp));
+    return update(tpd);
 }
 
 
 int
-erts_thr_progress_leader_update(ErtsSchedulerData *esdp)
+erts_thr_progress_leader_update(ErtsThrPrgrData *tpd)
 {
-    return leader_update(thr_prgr_data(esdp));
+    return leader_update(tpd);
 }
 
 void
-erts_thr_progress_prepare_wait(ErtsSchedulerData *esdp)
+erts_thr_progress_prepare_wait(ErtsThrPrgrData *tpd)
 {
     erts_aint32_t lflgs;
-    ErtsThrPrgrData *tpd = thr_prgr_data(esdp);
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
     erts_lc_check_exact(NULL, 0);
@@ -889,9 +893,8 @@ erts_thr_progress_prepare_wait(ErtsSchedulerData *esdp)
 }
 
 void
-erts_thr_progress_finalize_wait(ErtsSchedulerData *esdp)
+erts_thr_progress_finalize_wait(ErtsThrPrgrData *tpd)
 {
-    ErtsThrPrgrData *tpd = thr_prgr_data(esdp);
     ErtsThrPrgrVal current, val;
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
@@ -921,9 +924,8 @@ erts_thr_progress_finalize_wait(ErtsSchedulerData *esdp)
 }
 
 void
-erts_thr_progress_active(ErtsSchedulerData *esdp, int on)
+erts_thr_progress_active(ErtsThrPrgrData *tpd, int on)
 {
-    ErtsThrPrgrData *tpd = thr_prgr_data(esdp);
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
     erts_lc_check_exact(NULL, 0);
@@ -1182,10 +1184,10 @@ request_wakeup_unmanaged(ErtsThrPrgrData *tpd, ErtsThrPrgrVal value)
 }
 
 void
-erts_thr_progress_wakeup(ErtsSchedulerData *esdp,
+erts_thr_progress_wakeup(ErtsThrPrgrData *tpd,
 			 ErtsThrPrgrVal value)
 {
-    ErtsThrPrgrData *tpd = thr_prgr_data(esdp);
+
     ASSERT(!tpd->is_temporary);
     if (tpd->is_managed)
 	request_wakeup_managed(tpd, value);
diff --git a/erts/emulator/beam/erl_thr_progress.h b/erts/emulator/beam/erl_thr_progress.h
index 8329995b24..00a9e61407 100644
--- a/erts/emulator/beam/erl_thr_progress.h
+++ b/erts/emulator/beam/erl_thr_progress.h
@@ -123,22 +123,24 @@ extern ErtsThrPrgr erts_thr_prgr__;
 
 void erts_thr_progress_pre_init(void);
 void erts_thr_progress_init(int no_schedulers, int managed, int unmanaged);
-void erts_thr_progress_register_managed_thread(ErtsSchedulerData *esdp,
-					       ErtsThrPrgrCallbacks *,
-					       int);
+ErtsThrPrgrData *erts_thr_progress_register_managed_thread(
+    ErtsSchedulerData *esdp, ErtsThrPrgrCallbacks *, int);
 void erts_thr_progress_register_unmanaged_thread(ErtsThrPrgrCallbacks *);
-void erts_thr_progress_active(ErtsSchedulerData *esdp, int on);
-void erts_thr_progress_wakeup(ErtsSchedulerData *esdp,
+void erts_thr_progress_active(ErtsThrPrgrData *, int on);
+void erts_thr_progress_wakeup(ErtsThrPrgrData *,
 			      ErtsThrPrgrVal value);
-int erts_thr_progress_update(ErtsSchedulerData *esdp);
-int erts_thr_progress_leader_update(ErtsSchedulerData *esdp);
-void erts_thr_progress_prepare_wait(ErtsSchedulerData *esdp);
-void erts_thr_progress_finalize_wait(ErtsSchedulerData *esdp);
+int erts_thr_progress_update(ErtsThrPrgrData *);
+int erts_thr_progress_leader_update(ErtsThrPrgrData *);
+void erts_thr_progress_prepare_wait(ErtsThrPrgrData *);
+void erts_thr_progress_finalize_wait(ErtsThrPrgrData *);
 ErtsThrPrgrDelayHandle erts_thr_progress_unmanaged_delay__(void);
 void erts_thr_progress_unmanaged_continue__(int umrefc_ix);
+ErtsThrPrgrData *erts_thr_progress_data(void);
 
 void erts_thr_progress_dbg_print_state(void);
 
+ERTS_GLB_INLINE ErtsThrPrgrData *erts_thr_prgr_data(ErtsSchedulerData *esdp);
+
 ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc);
 ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_acqb__(ERTS_THR_PRGR_ATOMIC *atmc);
 ERTS_GLB_INLINE ErtsThrPrgrVal erts_thr_prgr_read_mb__(ERTS_THR_PRGR_ATOMIC *atmc);
@@ -161,6 +163,15 @@ ERTS_GLB_INLINE int erts_thr_progress_has_reached(ErtsThrPrgrVal val);
 
 #if ERTS_GLB_INLINE_INCL_FUNC_DEF
 
+ERTS_GLB_INLINE ErtsThrPrgrData *
+erts_thr_prgr_data(ErtsSchedulerData *esdp) {
+    if (esdp) {
+        return &esdp->thr_progress_data;
+    } else {
+        return erts_thr_progress_data();
+    }
+}
+
 ERTS_GLB_INLINE ErtsThrPrgrVal
 erts_thr_prgr_read_nob__(ERTS_THR_PRGR_ATOMIC *atmc)
 {
diff --git a/erts/emulator/beam/erl_trace.c b/erts/emulator/beam/erl_trace.c
index 53a020e7a5..2350d4c02f 100644
--- a/erts/emulator/beam/erl_trace.c
+++ b/erts/emulator/beam/erl_trace.c
@@ -2177,6 +2177,7 @@ sys_msg_dispatcher_func(void *unused)
 {
     ErtsThrPrgrCallbacks callbacks;
     ErtsSysMsgQ *local_sys_message_queue = NULL;
+    ErtsThrPrgrData *tpd;
     int wait = 0;
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
@@ -2189,7 +2190,7 @@ sys_msg_dispatcher_func(void *unused)
     callbacks.wait = sys_msg_dispatcher_wait;
     callbacks.finalize_wait = sys_msg_dispatcher_fin_wait;
 
-    erts_thr_progress_register_managed_thread(NULL, &callbacks, 0);
+    tpd = erts_thr_progress_register_managed_thread(NULL, &callbacks, 0);
 
     while (1) {
 	int end_wait = 0;
@@ -2210,8 +2211,8 @@ sys_msg_dispatcher_func(void *unused)
 	if (!sys_message_queue) {
 	    erts_mtx_unlock(&smq_mtx);
 	    end_wait = 1;
-	    erts_thr_progress_active(NULL, 0);
-	    erts_thr_progress_prepare_wait(NULL);
+	    erts_thr_progress_active(tpd, 0);
+	    erts_thr_progress_prepare_wait(tpd);
 	    erts_mtx_lock(&smq_mtx);
 	}
 
@@ -2225,8 +2226,8 @@ sys_msg_dispatcher_func(void *unused)
 	erts_mtx_unlock(&smq_mtx);
 
 	if (end_wait) {
-	    erts_thr_progress_finalize_wait(NULL);
-	    erts_thr_progress_active(NULL, 1);
+	    erts_thr_progress_finalize_wait(tpd);
+	    erts_thr_progress_active(tpd, 1);
 	}
 
 	/* Send trace messages ... */
@@ -2239,8 +2240,8 @@ sys_msg_dispatcher_func(void *unused)
 	    Process *proc = NULL;
 	    Port *port = NULL;
 
-	    if (erts_thr_progress_update(NULL))
-		erts_thr_progress_leader_update(NULL);
+	    if (erts_thr_progress_update(tpd))
+		erts_thr_progress_leader_update(tpd);
 
 #ifdef DEBUG_PRINTOUTS
 	    print_msg_type(smqp);
diff --git a/erts/emulator/sys/common/erl_check_io.c b/erts/emulator/sys/common/erl_check_io.c
index 9f115706dc..748555165f 100644
--- a/erts/emulator/sys/common/erl_check_io.c
+++ b/erts/emulator/sys/common/erl_check_io.c
@@ -105,6 +105,7 @@ typedef struct erts_poll_thread
 {
     ErtsPollSet *ps;
     ErtsPollResFd *pollres;
+    ErtsThrPrgrData *tpd;
     int pollres_len;
 } ErtsPollThread;
 
@@ -1516,7 +1517,8 @@ erts_check_io_interrupt(ErtsPollThread *psi, int set)
 }
 
 ErtsPollThread *
-erts_create_pollset_thread(int id) {
+erts_create_pollset_thread(int id, ErtsThrPrgrData *tpd) {
+    psiv[id].tpd = tpd;
     return psiv+id;
 }
 
@@ -1538,12 +1540,12 @@ erts_check_io(ErtsPollThread *psi)
 #if ERTS_POLL_USE_FALLBACK
     if (psi->ps == get_fallback()) {
 
-        poll_ret = erts_poll_wait_flbk(psi->ps, psi->pollres, &pollres_len);
+        poll_ret = erts_poll_wait_flbk(psi->ps, psi->pollres, &pollres_len, psi->tpd);
 
     } else
 #endif
     {
-        poll_ret = erts_poll_wait(psi->ps, psi->pollres, &pollres_len);
+        poll_ret = erts_poll_wait(psi->ps, psi->pollres, &pollres_len, psi->tpd);
     }
 
 #ifdef ERTS_ENABLE_LOCK_CHECK
diff --git a/erts/emulator/sys/common/erl_check_io.h b/erts/emulator/sys/common/erl_check_io.h
index 443ef1264c..16aba8a5f3 100644
--- a/erts/emulator/sys/common/erl_check_io.h
+++ b/erts/emulator/sys/common/erl_check_io.h
@@ -90,8 +90,11 @@ void erts_check_io_interrupt(struct erts_poll_thread *pt, int set);
 /**
  * Create a new poll thread structure that is associated with the number no.
  * It is the callers responsibility that no is unique.
+ *
+ * @param no the id of the pollset thread, -2 = aux thread, -1 = scheduler
+ * @param tpd the thread progress data of the pollset thread
  */
-struct erts_poll_thread* erts_create_pollset_thread(int no);
+struct erts_poll_thread* erts_create_pollset_thread(int no, ErtsThrPrgrData *tpd);
 #ifdef ERTS_ENABLE_LOCK_COUNT
 /**
  * Toggle lock counting on all check io locks
diff --git a/erts/emulator/sys/common/erl_poll.c b/erts/emulator/sys/common/erl_poll.c
index b4d1575ee5..5a5874ddfa 100644
--- a/erts/emulator/sys/common/erl_poll.c
+++ b/erts/emulator/sys/common/erl_poll.c
@@ -75,6 +75,7 @@
 #  define WANT_NONBLOCKING
 #endif
 
+#include "erl_thr_progress.h"
 #include "erl_poll.h"
 #if ERTS_POLL_USE_KQUEUE
 #  include <sys/types.h>
@@ -95,7 +96,6 @@
 #    include <limits.h>
 #  endif
 #endif
-#include "erl_thr_progress.h"
 #include "erl_driver.h"
 #include "erl_alloc.h"
 #include "erl_msacc.h"
@@ -1629,7 +1629,8 @@ check_fd_events(ErtsPollSet *ps, ErtsPollResFd pr[], int do_wait, int max_res)
 int
 ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet *ps,
 				 ErtsPollResFd pr[],
-				 int *len)
+				 int *len,
+                                 ErtsThrPrgrData *tpd)
 {
     int res, no_fds, used_fds = 0;
     int ebadf = 0;
@@ -1659,7 +1660,7 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet *ps,
     DEBUG_PRINT_WAIT("Entering %s(), do_wait=%d", ps, __FUNCTION__, do_wait);
 
     if (do_wait) {
-        erts_thr_progress_prepare_wait(NULL);
+        erts_thr_progress_prepare_wait(tpd);
         ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_SLEEP);
     }
 
@@ -1703,7 +1704,7 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet *ps,
     }
 
     if (do_wait) {
-        erts_thr_progress_finalize_wait(NULL);
+        erts_thr_progress_finalize_wait(tpd);
         ERTS_MSACC_UPDATE_CACHE();
         ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_CHECK_IO);
     }
diff --git a/erts/emulator/sys/common/erl_poll_api.h b/erts/emulator/sys/common/erl_poll_api.h
index 1170a549b9..f35f64a9f3 100644
--- a/erts/emulator/sys/common/erl_poll_api.h
+++ b/erts/emulator/sys/common/erl_poll_api.h
@@ -72,11 +72,13 @@ ErtsPollEvents ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet *ps,
  * @param res an array of fd results that the ready fds are put in.
  * @param[in] length the length of the res array
  * @param[out] length the number of ready events returned in res
+ * @param tpd the thread progress data to note sleep state in
  * @return 0 on success, else the ERRNO of the error that happened.
  */
 int ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet *ps,
                                      ErtsPollResFd res[],
-                                     int *length);
+                                     int *length,
+                                     ErtsThrPrgrData *tpd);
 /**
  * Interrupt the thread waiting in the pollset. This function should be called
  * with set = 0 before any thread calls erts_poll_wait in order to clear any
diff --git a/erts/emulator/sys/win32/erl_poll.c b/erts/emulator/sys/win32/erl_poll.c
index 39bb4d515e..5d832f4d34 100644
--- a/erts/emulator/sys/win32/erl_poll.c
+++ b/erts/emulator/sys/win32/erl_poll.c
@@ -1017,7 +1017,8 @@ ErtsPollEvents erts_poll_control(ErtsPollSet *ps,
 
 int erts_poll_wait(ErtsPollSet *ps,
 		   ErtsPollResFd pr[],
-		   int *len)
+		   int *len,
+                   ErtsThrPrgrData *tpd)
 {
     int no_fds;
     DWORD timeout = INFINITE;
@@ -1056,10 +1057,10 @@ int erts_poll_wait(ErtsPollSet *ps,
 
 	HARDDEBUGF(("Start waiting %d [%d]",num_h, (int) timeout));
 	ERTS_POLLSET_UNLOCK(ps);
-	erts_thr_progress_prepare_wait(NULL);
+	erts_thr_progress_prepare_wait(tpd);
         ERTS_MSACC_SET_STATE_CACHED(ERTS_MSACC_STATE_SLEEP);
 	handle = WaitForMultipleObjects(num_h, harr, FALSE, timeout);
-	erts_thr_progress_finalize_wait(NULL);
+	erts_thr_progress_finalize_wait(tpd);
         ERTS_MSACC_POP_STATE();
 	ERTS_POLLSET_LOCK(ps);
 	HARDDEBUGF(("Stop waiting %d [%d]",num_h, (int) timeout));