Merge branch 'rickard/scheduler-wait/OTP-9105' into dev

* rickard/scheduler-wait/OTP-9105:
  The emulator could get into a state where it didn't check for I/O.

6 files changed, 38 insertions, 11 deletions

diff --git a/erts/emulator/beam/erl_port_task.h b/erts/emulator/beam/erl_port_task.h
index 714b4ea7dd..49a0b4c63a 100644
--- a/erts/emulator/beam/erl_port_task.h
+++ b/erts/emulator/beam/erl_port_task.h
@@ -102,6 +102,7 @@ erts_port_task_init_sched(ErtsPortTaskSched *ptsp)
 ERTS_GLB_INLINE int
 erts_port_task_have_outstanding_io_tasks(void)
 {
+    ERTS_THR_MEMORY_BARRIER;
     return erts_smp_atomic_read(&erts_port_task_outstanding_io_tasks) != 0;
 }
 
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 4d6e982325..fd2bd4469c 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -815,15 +815,31 @@ erts_active_schedulers(void)
     return as;
 }
 
+#ifdef ERTS_SMP
+
+static ERTS_INLINE void
+clear_sys_scheduling(void)
+{
+    erts_smp_atomic32_set_relb(&doing_sys_schedule, 0);
+}
+
+static ERTS_INLINE int
+try_set_sys_scheduling(void)
+{
+    return 0 == erts_smp_atomic32_cmpxchg_acqb(&doing_sys_schedule, 1, 0);
+}
+
+#endif
+
 static ERTS_INLINE int
 prepare_for_sys_schedule(void)
 {
 #ifdef ERTS_SMP
     while (!erts_port_task_have_outstanding_io_tasks()
-	   && !erts_smp_atomic32_xchg(&doing_sys_schedule, 1)) {
+	   && try_set_sys_scheduling()) {
 	if (!erts_port_task_have_outstanding_io_tasks())
 	    return 1;
-	erts_smp_atomic32_set(&doing_sys_schedule, 0);
+	clear_sys_scheduling();
     }
     return 0;
 #else
@@ -1153,7 +1169,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 	     * call erl_sys_schedule() until it is handled.
 	     */
 	    if (erts_port_task_have_outstanding_io_tasks()) {
-		erts_smp_atomic32_set(&doing_sys_schedule, 0);
+		clear_sys_scheduling();
 		/*
 		 * Got to check that we still got I/O tasks; otherwise
 		 * we have to continue checking for I/O...
@@ -1172,7 +1188,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 	 * sleep in erl_sys_schedule().
 	 */
 	if (erts_port_task_have_outstanding_io_tasks()) {
-	    erts_smp_atomic32_set(&doing_sys_schedule, 0);
+	    clear_sys_scheduling();
 
 	    /*
 	     * Got to check that we still got I/O tasks; otherwise
@@ -1226,7 +1242,7 @@ scheduler_wait(int *fcalls, ErtsSchedulerData *esdp, ErtsRunQueue *rq)
 	sys_woken:
 	    erts_smp_runq_lock(rq);
 	sys_locked_woken:
-	    erts_smp_atomic32_set(&doing_sys_schedule, 0);
+	    clear_sys_scheduling();
 	    if (flgs & ~ERTS_SSI_FLG_SUSPENDED)
 		erts_smp_atomic32_band(&ssi->flags, ERTS_SSI_FLG_SUSPENDED);
 	    sched_active_sys(esdp->no, rq);
@@ -1289,6 +1305,7 @@ wake_scheduler(ErtsRunQueue *rq, int incq, int one)
 	res = sl->list != NULL;
 	erts_smp_spin_unlock(&sl->lock);
 
+	ERTS_THR_MEMORY_BARRIER;
 	flgs = ssi_flags_set_wake(ssi);
 	erts_sched_finish_poke(ssi, flgs);
 
@@ -1298,6 +1315,8 @@ wake_scheduler(ErtsRunQueue *rq, int incq, int one)
     else {
 	sl->list = NULL;
 	erts_smp_spin_unlock(&sl->lock);
+
+	ERTS_THR_MEMORY_BARRIER;
 	do {
 	    ErtsSchedulerSleepInfo *wake_ssi = ssi;
 	    ssi = ssi->next;
@@ -5371,7 +5390,7 @@ Process *schedule(Process *p, int calls)
 	    if (dt) erts_bump_timer(dt);
 #ifdef ERTS_SMP
 	    erts_smp_runq_lock(rq);
-	    erts_smp_atomic32_set(&doing_sys_schedule, 0);
+	    clear_sys_scheduling();
 	    goto continue_check_activities_to_run;
 #else
 	    if (!runnable)
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index e871a9834a..8f78a7d76e 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -1592,7 +1592,9 @@ ERTS_GLB_INLINE void erts_sched_poke(ErtsSchedulerSleepInfo *ssi);
 ERTS_GLB_INLINE void
 erts_sched_poke(ErtsSchedulerSleepInfo *ssi)
 {
-    erts_aint32_t flags = erts_smp_atomic32_read(&ssi->flags);
+    erts_aint32_t flags;
+    ERTS_THR_MEMORY_BARRIER;
+    flags = erts_smp_atomic32_read(&ssi->flags);
     ASSERT(!(flags & ERTS_SSI_FLG_SLEEPING)
 	   || (flags & ERTS_SSI_FLG_WAITING));
     if (flags & ERTS_SSI_FLG_SLEEPING) {
diff --git a/erts/emulator/sys/common/erl_poll.c b/erts/emulator/sys/common/erl_poll.c
index 77ac2de5f6..3ae5b8d747 100644
--- a/erts/emulator/sys/common/erl_poll.c
+++ b/erts/emulator/sys/common/erl_poll.c
@@ -347,6 +347,7 @@ reset_wakeup_state(ErtsPollSet ps)
 {
 #ifdef ERTS_SMP
     erts_atomic32_set(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN);
+    ERTS_THR_MEMORY_BARRIER;
 #elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT
     ps->wakeup_state = 0;
 #endif
diff --git a/erts/emulator/sys/win32/erl_poll.c b/erts/emulator/sys/win32/erl_poll.c
index 1f2877b682..7662f190ef 100644
--- a/erts/emulator/sys/win32/erl_poll.c
+++ b/erts/emulator/sys/win32/erl_poll.c
@@ -452,12 +452,15 @@ poll_wait_timeout(ErtsPollSet ps, SysTimeval *tvp)
 static ERTS_INLINE void
 wake_poller(ErtsPollSet ps, int io_ready)
 {
-    erts_aint32_t wakeup_state = erts_atomic32_read(&ps->wakeup_state);
+    erts_aint32_t wakeup_state;
     if (io_ready) {
 	/* We may set the event multiple times. This is, however, harmless. */
-	erts_atomic32_set(&ps->wakeup_state, ERTS_POLL_WOKEN_IO_READY);
+	wakeup_state = erts_atomic32_read(&ps->wakeup_state);
+	erts_atomic32_set_relb(&ps->wakeup_state, ERTS_POLL_WOKEN_IO_READY);
     }
     else {
+	ERTS_THR_MEMORY_BARRIER;
+	wakeup_state = erts_atomic32_read(&ps->wakeup_state);
 	while (wakeup_state != ERTS_POLL_WOKEN_IO_READY
 	       && wakeup_state != ERTS_POLL_WOKEN_INTR) {
 	    erts_aint32_t act = erts_atomic32_cmpxchg(&ps->wakeup_state,
@@ -518,6 +521,7 @@ reset_interrupt(ErtsPollSet ps)
 	    break;
 	wakeup_state = act;
     }
+    ERTS_THR_MEMORY_BARRIER;
 }
 
 static ERTS_INLINE void
diff --git a/erts/include/internal/pthread/ethr_event.h b/erts/include/internal/pthread/ethr_event.h
index 93da8a0429..b74b76a443 100644
--- a/erts/include/internal/pthread/ethr_event.h
+++ b/erts/include/internal/pthread/ethr_event.h
@@ -62,7 +62,7 @@ static void ETHR_INLINE
 ETHR_INLINE_FUNC_NAME_(ethr_event_set)(ethr_event *e)
 {
     ethr_sint32_t val;
-    ETHR_WRITE_MEMORY_BARRIER;
+    ETHR_MEMORY_BARRIER;
     val = ethr_atomic32_xchg(&e->futex, ETHR_EVENT_ON__);
     if (val == ETHR_EVENT_OFF_WAITER__) {
 	int res = ETHR_FUTEX__(&e->futex, ETHR_FUTEX_WAKE__, 1);
@@ -99,7 +99,7 @@ static void ETHR_INLINE
 ETHR_INLINE_FUNC_NAME_(ethr_event_set)(ethr_event *e)
 {
     ethr_sint32_t val;
-    ETHR_WRITE_MEMORY_BARRIER;
+    ETHR_MEMORY_BARRIER;
     val = ethr_atomic32_xchg(&e->state, ETHR_EVENT_ON__);
     if (val == ETHR_EVENT_OFF_WAITER__) {
 	int res = pthread_mutex_lock(&e->mtx);