Merge branch 'dev' into bmk/snmp/snmp419_integration/OTP-9068

16 files changed, 606 insertions, 385 deletions

diff --git a/erts/doc/src/escript.xml b/erts/doc/src/escript.xml
index 44c9a5ac68..588508aae6 100644
--- a/erts/doc/src/escript.xml
+++ b/erts/doc/src/escript.xml
@@ -153,7 +153,10 @@ halt(1).</pre>
         <p>Execution of interpreted code is slower than compiled code.
           If much of the execution takes place in interpreted code it
           may be worthwhile to compile it, even though the compilation
-          itself will take a little while.</p>
+          itself will take a little while. It is also possible to supply
+	  <c>native</c> instead of compile, this will compile the script 
+	  using the native flag, again depending on the characteristics 
+	  of the escript this could or could not be worth while.</p>
 
         <p>As mentioned earlier, it is possible to have a script which
           contains precompiled <c>beam</c> code. In a precompiled
@@ -397,6 +400,9 @@ ok
        Warnings and errors (if any) are written to the standard output, but
        the script will not be run. The exit status will be 0 if there were
        no errors, and 127 otherwise.</item>
+
+       <tag>-n</tag>
+      <item>Compile the escript using the +native flag.</item>
     </taglist>
   </section>
 </comref>
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 12a8022861..af7ed064d3 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -321,6 +321,7 @@ extern int count_instructions;
 #  define POST_BIF_GC_SWAPIN_0(_p, _res)				\
      ERTS_SMP_REQ_PROC_MAIN_LOCK((_p));					\
      PROCESS_MAIN_CHK_LOCKS((_p));					\
+     ERTS_VERIFY_UNUSED_TEMP_ALLOC((_p));				\
      if (((_p)->mbuf) || (MSO(_p).overhead >= BIN_VHEAP_SZ(_p)) ) {	\
        _res = erts_gc_after_bif_call((_p), (_res), NULL, 0);		\
        E = (_p)->stop;							\
@@ -328,6 +329,7 @@ extern int count_instructions;
      HTOP = HEAP_TOP((_p))
 
 #  define POST_BIF_GC_SWAPIN(_p, _res, _regs, _arity)			\
+     ERTS_VERIFY_UNUSED_TEMP_ALLOC((_p));				\
      ERTS_SMP_REQ_PROC_MAIN_LOCK((_p));					\
      PROCESS_MAIN_CHK_LOCKS((_p));					\
      if (((_p)->mbuf) || (MSO(_p).overhead >= BIN_VHEAP_SZ(_p)) ) {	\
@@ -367,6 +369,7 @@ extern int count_instructions;
            reg[0] = r(0); \
            PROCESS_MAIN_CHK_LOCKS(c_p); \
            FCALLS -= erts_garbage_collect(c_p, needed + (HeapNeed), reg, (M)); \
+           ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p); \
            PROCESS_MAIN_CHK_LOCKS(c_p); \
            r(0) = reg[0]; \
            SWAPIN; \
@@ -420,6 +423,7 @@ extern int count_instructions;
        reg[0] = r(0);                                           		\
        PROCESS_MAIN_CHK_LOCKS(c_p);                             		\
        FCALLS -= erts_garbage_collect(c_p, need, reg, (Live));  		\
+       ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);					\
        PROCESS_MAIN_CHK_LOCKS(c_p);                             		\
        r(0) = reg[0];                                           		\
        SWAPIN;                                                  		\
@@ -442,6 +446,7 @@ extern int count_instructions;
        reg[0] = r(0);                                           \
        PROCESS_MAIN_CHK_LOCKS(c_p);                             \
        FCALLS -= erts_garbage_collect(c_p, need, reg, (Live));  \
+       ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);			\
        PROCESS_MAIN_CHK_LOCKS(c_p);                             \
        r(0) = reg[0];                                           \
        SWAPIN;                                                  \
@@ -464,6 +469,7 @@ extern int count_instructions;
        reg[Live] = Extra;						\
        PROCESS_MAIN_CHK_LOCKS(c_p);					\
        FCALLS -= erts_garbage_collect(c_p, need, reg, (Live)+1);	\
+       ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);				\
        PROCESS_MAIN_CHK_LOCKS(c_p);					\
        if (Live > 0) {							\
 	   r(0) = reg[0];						\
@@ -1214,7 +1220,9 @@ void process_main(void)
  do_schedule1:
     PROCESS_MAIN_CHK_LOCKS(c_p);
     ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p);
+    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
     c_p = schedule(c_p, reds_used);
+    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 #ifdef DEBUG
     pid = c_p->id;
 #endif
@@ -1664,6 +1672,7 @@ void process_main(void)
 		 SWAPOUT;
 		 PROCESS_MAIN_CHK_LOCKS(c_p);
 		 FCALLS -= erts_garbage_collect(c_p, 3, reg+2, 1);
+		 ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 		 PROCESS_MAIN_CHK_LOCKS(c_p);
 		 SWAPIN;
 	     }
@@ -1782,6 +1791,7 @@ void process_main(void)
 				       PROCESS_MAIN_CHK_LOCKS(c_p);
 				   },
 				   {
+				       ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 				       PROCESS_MAIN_CHK_LOCKS(c_p);
 				       r(0) = reg[0];
 				       SWAPIN;
@@ -1840,6 +1850,7 @@ void process_main(void)
      CANCEL_TIMER(c_p);
      free_message(msgp);
 
+     ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
      PROCESS_MAIN_CHK_LOCKS(c_p);
 
      NextPF(0, next);
@@ -2230,6 +2241,7 @@ void process_main(void)
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p));
 	result = (*bf)(c_p, arg);
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(result));
+	ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	ERTS_HOLE_CHECK(c_p);
 	FCALLS = c_p->fcalls;
@@ -2258,6 +2270,7 @@ void process_main(void)
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p));
 	result = (*bf)(c_p, arg);
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(result));
+	ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	ERTS_HOLE_CHECK(c_p);
 	FCALLS = c_p->fcalls;
@@ -2287,6 +2300,7 @@ void process_main(void)
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p);
 	result = (*bf)(c_p, reg, live);
+	ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	SWAPIN;
@@ -2322,6 +2336,7 @@ void process_main(void)
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p);
 	result = (*bf)(c_p, reg, live);
+	ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	SWAPIN;
@@ -2360,6 +2375,7 @@ void process_main(void)
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p);
 	result = (*bf)(c_p, reg, live);
+	ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	SWAPIN;
@@ -2394,6 +2410,7 @@ void process_main(void)
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p));
 	result = (*bf)(c_p, tmp_arg1, tmp_arg2);
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(result));
+	ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	ERTS_HOLE_CHECK(c_p);
 	FCALLS = c_p->fcalls;
@@ -2417,6 +2434,7 @@ void process_main(void)
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p));
 	result = (*bf)(c_p, tmp_arg1, tmp_arg2);
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(result));
+	ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	ERTS_HOLE_CHECK(c_p);
 	if (is_value(result)) {
@@ -3287,6 +3305,7 @@ void process_main(void)
 	    PROCESS_MAIN_CHK_LOCKS(c_p);
 	    bif_nif_arity = I[-1];
 	    ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p);
+	    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 
 	    ASSERT(!ERTS_PROC_IS_EXITING(c_p));
 	    {
@@ -3300,6 +3319,7 @@ void process_main(void)
 	    }
 	    ASSERT(!ERTS_PROC_IS_EXITING(c_p) || is_non_value(nif_bif_result));
 	    PROCESS_MAIN_CHK_LOCKS(c_p);
+	    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	    goto apply_bif_or_nif_epilogue;
 	 
 	OpCase(apply_bif):
@@ -3326,6 +3346,7 @@ void process_main(void)
 	    bif_nif_arity = I[-1];
 	    ASSERT(bif_nif_arity <= 3);
 	    ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p);
+	    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	    switch (bif_nif_arity) {
 	    case 3:
 		{
@@ -3334,6 +3355,7 @@ void process_main(void)
 		    nif_bif_result = (*bf)(c_p, r(0), x(1), x(2), I);
 		    ASSERT(!ERTS_PROC_IS_EXITING(c_p) ||
 			   is_non_value(nif_bif_result));
+		    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 		    PROCESS_MAIN_CHK_LOCKS(c_p);
 		}
 		break;
@@ -3344,6 +3366,7 @@ void process_main(void)
 		    nif_bif_result = (*bf)(c_p, r(0), x(1), I);
 		    ASSERT(!ERTS_PROC_IS_EXITING(c_p) ||
 			   is_non_value(nif_bif_result));
+		    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 		    PROCESS_MAIN_CHK_LOCKS(c_p);
 		}
 		break;
@@ -3354,6 +3377,7 @@ void process_main(void)
 		    nif_bif_result = (*bf)(c_p, r(0), I);
 		    ASSERT(!ERTS_PROC_IS_EXITING(c_p) ||
 			   is_non_value(nif_bif_result));
+		    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 		    PROCESS_MAIN_CHK_LOCKS(c_p);
 		}
 		break;
@@ -3364,6 +3388,7 @@ void process_main(void)
 		    nif_bif_result = (*bf)(c_p, I);
 		    ASSERT(!ERTS_PROC_IS_EXITING(c_p) ||
 			   is_non_value(nif_bif_result));
+		    ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 		    PROCESS_MAIN_CHK_LOCKS(c_p);
 		    break;
 		}
@@ -4590,6 +4615,7 @@ void process_main(void)
 	 ERTS_SMP_UNREQ_PROC_MAIN_LOCK(c_p);
 	 flags = erts_call_trace(c_p, ep->code, ep->match_prog_set, reg,
 				 0, &c_p->tracer_proc);
+	 ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	 ERTS_SMP_REQ_PROC_MAIN_LOCK(c_p);
 	 PROCESS_MAIN_CHK_LOCKS(c_p);
 	 ASSERT(!ERTS_PROC_IS_EXITING(c_p));
@@ -4601,6 +4627,7 @@ void process_main(void)
 		 /* SWAPOUT, SWAPIN was done and r(0) was saved above */
 		 PROCESS_MAIN_CHK_LOCKS(c_p);
 		 FCALLS -= erts_garbage_collect(c_p, 3, reg, ep->code[2]);
+		 ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 		 PROCESS_MAIN_CHK_LOCKS(c_p);
 		 r(0) = reg[0];
 		 SWAPIN;
@@ -4690,6 +4717,7 @@ void process_main(void)
 		 reg[0] = r(0);
 		 PROCESS_MAIN_CHK_LOCKS(c_p);
 		 FCALLS -= erts_garbage_collect(c_p, 2, reg, I[-1]);
+		 ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 		 PROCESS_MAIN_CHK_LOCKS(c_p);
 		 r(0) = reg[0];
 	     }
@@ -4793,6 +4821,7 @@ void process_main(void)
 	     /* SWAPOUT was done and r(0) was saved above */
 	     PROCESS_MAIN_CHK_LOCKS(c_p);
 	     FCALLS -= erts_garbage_collect(c_p, need, reg, I[-1]);
+	     ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	     PROCESS_MAIN_CHK_LOCKS(c_p);
 	     r(0) = reg[0];
 	     SWAPIN;
@@ -6239,6 +6268,7 @@ hibernate(Process* c_p, Eterm module, Eterm function, Eterm args, Eterm* reg)
 	c_p->fvalue = NIL;
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	erts_garbage_collect_hibernate(c_p);
+	ERTS_VERIFY_UNUSED_TEMP_ALLOC(c_p);
 	PROCESS_MAIN_CHK_LOCKS(c_p);
 	erts_smp_proc_lock(c_p, ERTS_PROC_LOCK_MSGQ|ERTS_PROC_LOCK_STATUS);
 	ASSERT(!ERTS_PROC_IS_EXITING(c_p));
@@ -6484,6 +6514,7 @@ new_fun(Process* p, Eterm* reg, ErlFunEntry* fe, int num_free)
     if (HEAP_LIMIT(p) - HEAP_TOP(p) <= needed) {
 	PROCESS_MAIN_CHK_LOCKS(p);
 	erts_garbage_collect(p, needed, reg, num_free);
+	ERTS_VERIFY_UNUSED_TEMP_ALLOC(p);
 	PROCESS_MAIN_CHK_LOCKS(p);
     }
     hp = p->htop;
diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c
index e85e2d7e3f..323c422c6d 100644
--- a/erts/emulator/beam/erl_alloc.c
+++ b/erts/emulator/beam/erl_alloc.c
@@ -1411,6 +1411,33 @@ void erts_alloc_reg_scheduler_id(Uint id)
     erts_tsd_set(thr_ix_key, (void *)(long) ix);
 }
 
+static void
+no_verify(Allctr_t *allctr)
+{
+
+}
+
+erts_alloc_verify_func_t
+erts_alloc_get_verify_unused_temp_alloc(Allctr_t **allctr)
+{
+    if (erts_allctrs_info[ERTS_ALC_A_TEMPORARY].alloc_util
+	&& erts_allctrs_info[ERTS_ALC_A_TEMPORARY].thr_spec) {
+	ErtsAllocatorThrSpec_t *tspec;
+	tspec = &erts_allctr_thr_spec[ERTS_ALC_A_TEMPORARY];
+	if (!tspec->all_thr_safe) {
+	    int ix = erts_alc_get_thr_ix();
+
+	    if (ix < tspec->size) {
+		*allctr = tspec->allctr[ix];
+		return erts_alcu_verify_unused;
+	    }
+	}
+    }
+
+    *allctr = NULL;
+    return no_verify;
+}
+
 __decl_noreturn void
 erts_alc_fatal_error(int error, int func, ErtsAlcType_t n, ...)
 {
diff --git a/erts/emulator/beam/erl_alloc.h b/erts/emulator/beam/erl_alloc.h
index 3e96c76dbf..dd4cc22171 100644
--- a/erts/emulator/beam/erl_alloc.h
+++ b/erts/emulator/beam/erl_alloc.h
@@ -236,6 +236,11 @@ void *erts_realloc_fnf(ErtsAlcType_t type, void *ptr, Uint size)
 
 #endif /* #if ERTS_ALC_DO_INLINE || defined(ERTS_ALC_INTERNAL__) */
 
+typedef void (*erts_alloc_verify_func_t)(Allctr_t *);
+
+erts_alloc_verify_func_t
+erts_alloc_get_verify_unused_temp_alloc(Allctr_t **allctr);
+
 #ifndef ERTS_CACHE_LINE_SIZE
 /* Assume a cache line size of 64 bytes */
 #  define ERTS_CACHE_LINE_SIZE ((UWord) 64)
diff --git a/erts/emulator/beam/erl_alloc_util.c b/erts/emulator/beam/erl_alloc_util.c
index 8b184899c9..c09f0bbd77 100644
--- a/erts/emulator/beam/erl_alloc_util.c
+++ b/erts/emulator/beam/erl_alloc_util.c
@@ -3368,6 +3368,38 @@ erts_alcu_test(unsigned long op, unsigned long a1, unsigned long a2)
  * Debug functions                                                           *
 \*                                                                           */
 
+void
+erts_alcu_verify_unused(Allctr_t *allctr)
+{
+    UWord no;
+
+    no = allctr->sbcs.curr_mseg.no;
+    no += allctr->sbcs.curr_sys_alloc.no;
+    no += allctr->mbcs.blocks.curr.no;
+
+    if (no) {
+	UWord sz = allctr->sbcs.blocks.curr.size;
+	sz += allctr->mbcs.blocks.curr.size;
+	erl_exit(ERTS_ABORT_EXIT,
+		 "%salloc() used when expected to be unused!\n"
+		 "Total amount of blocks allocated: %bpu\n"
+		 "Total amount of bytes allocated: %bpu\n",
+		 allctr->name_prefix, no, sz);
+    }
+}
+
+void
+erts_alcu_verify_unused_ts(Allctr_t *allctr)
+{
+#ifdef USE_THREADS
+    erts_mtx_lock(&allctr->mutex);
+#endif
+    erts_alcu_verify_unused(allctr);
+#ifdef USE_THREADS
+    erts_mtx_unlock(&allctr->mutex);
+#endif
+}
+
 #ifdef ERTS_ALLOC_UTIL_HARD_DEBUG
 
 static void
diff --git a/erts/emulator/beam/erl_alloc_util.h b/erts/emulator/beam/erl_alloc_util.h
index f2b951bca6..5a8db5e29e 100644
--- a/erts/emulator/beam/erl_alloc_util.h
+++ b/erts/emulator/beam/erl_alloc_util.h
@@ -333,6 +333,9 @@ struct Allctr_t_ {
 int	erts_alcu_start(Allctr_t *, AllctrInit_t *);
 void	erts_alcu_stop(Allctr_t *);
 
+void	erts_alcu_verify_unused(Allctr_t *);
+void	erts_alcu_verify_unused_ts(Allctr_t *allctr);
+
 unsigned long	erts_alcu_test(unsigned long, unsigned long, unsigned long);
 
 
diff --git a/erts/emulator/beam/erl_cpu_topology.c b/erts/emulator/beam/erl_cpu_topology.c
index 8a6b4d8d6c..bcf8bcf270 100644
--- a/erts/emulator/beam/erl_cpu_topology.c
+++ b/erts/emulator/beam/erl_cpu_topology.c
@@ -1,7 +1,7 @@
 /*
  * %CopyrightBegin%
  *
- * Copyright Ericsson AB 2010. All Rights Reserved.
+ * Copyright Ericsson AB 2010-2011. 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
@@ -630,13 +630,13 @@ write_schedulers_bind_change(erts_cpu_topology_t *cpudata, int size)
 int
 erts_init_scheduler_bind_type_string(char *how)
 {
-    if (erts_bind_to_cpu(cpuinfo, -1) == -ENOTSUP)
+    if (sys_strcmp(how, "u") == 0)
+	cpu_bind_order = ERTS_CPU_BIND_NONE;
+    else if (erts_bind_to_cpu(cpuinfo, -1) == -ENOTSUP)
 	return ERTS_INIT_SCHED_BIND_TYPE_NOT_SUPPORTED;
-
-    if (!system_cpudata && !user_cpudata)
+    else if (!system_cpudata && !user_cpudata)
 	return ERTS_INIT_SCHED_BIND_TYPE_ERROR_NO_CPU_TOPOLOGY;
-
-    if (sys_strcmp(how, "db") == 0)
+    else 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;
@@ -652,11 +652,8 @@ erts_init_scheduler_bind_type_string(char *how)
 	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;
 }
 
@@ -724,6 +721,11 @@ erts_bind_schedulers(Process *c_p, Eterm how)
     erts_smp_rwmtx_rwlock(&cpuinfo_rwmtx);
 
     if (erts_bind_to_cpu(cpuinfo, -1) == -ENOTSUP) {
+	if (cpu_bind_order == ERTS_CPU_BIND_NONE
+	    && ERTS_IS_ATOM_STR("unbound", how)) {
+	    res = bound_schedulers_term(ERTS_CPU_BIND_NONE);
+	    goto done;
+	}
 	ERTS_BIF_PREP_ERROR(res, c_p, EXC_NOTSUP);
     }
     else {
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index ddfc27a93f..e4a871fd7e 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -2752,6 +2752,15 @@ erts_init_scheduling(int mrq, int no_schedulers, int no_schedulers_online)
 
     /* init port tasks */
     erts_port_task_init();
+
+#ifndef ERTS_SMP
+#ifdef ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC
+    erts_scheduler_data->verify_unused_temp_alloc
+	= erts_alloc_get_verify_unused_temp_alloc(
+	    &erts_scheduler_data->verify_unused_temp_alloc_data);
+    ERTS_VERIFY_UNUSED_TEMP_ALLOC(NULL);
+#endif
+#endif
 }
 
 ErtsRunQueue *
@@ -3711,6 +3720,13 @@ sched_thread_func(void *vesdp)
     }
     erts_smp_mtx_unlock(&schdlr_sspnd.mtx);
 
+#ifdef ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC
+    ((ErtsSchedulerData *) vesdp)->verify_unused_temp_alloc
+	= erts_alloc_get_verify_unused_temp_alloc(
+	    &((ErtsSchedulerData *) vesdp)->verify_unused_temp_alloc_data);
+    ERTS_VERIFY_UNUSED_TEMP_ALLOC(NULL);
+#endif
+
     process_main();
     /* No schedulers should *ever* terminate */
     erl_exit(ERTS_ABORT_EXIT, "Scheduler thread number %bpu terminated\n",
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index 8479e56710..e871a9834a 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -28,6 +28,12 @@
 #define ERTS_INCLUDE_SCHEDULER_INTERNALS
 #endif
 
+/* #define ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC */
+
+#if !defined(ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC) && defined(DEBUG)
+#  define ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC
+#endif
+
 typedef struct process Process;
 
 #include "sys.h"
@@ -421,6 +427,11 @@ struct ErtsSchedulerData_ {
     /* NOTE: These fields are modified under held mutexes by other threads */
     erts_smp_atomic32_t chk_cpu_bind; /* Only used when common run queue */
 #endif
+
+#ifdef ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC
+    erts_alloc_verify_func_t verify_unused_temp_alloc;
+    Allctr_t *verify_unused_temp_alloc_data;
+#endif
 };
 
 typedef union {
@@ -1145,6 +1156,20 @@ Uint erts_debug_nbalance(void);
 #  define ERTS_PROC_GET_SCHDATA(PROC) (erts_scheduler_data)
 #endif
 
+#ifdef ERTS_DO_VERIFY_UNUSED_TEMP_ALLOC
+#  define ERTS_VERIFY_UNUSED_TEMP_ALLOC(P)					\
+do {										\
+    ErtsSchedulerData *esdp__ = ((P)						\
+				 ? ERTS_PROC_GET_SCHDATA((Process *) (P))	\
+				 : erts_get_scheduler_data());			\
+    if (esdp__)									\
+	esdp__->verify_unused_temp_alloc(					\
+	    esdp__->verify_unused_temp_alloc_data);				\
+} while (0)
+#else
+#  define ERTS_VERIFY_UNUSED_TEMP_ALLOC(ESDP)
+#endif
+
 #if defined(ERTS_SMP) || defined(USE_THREADS)
 ErtsSchedulerData *erts_get_scheduler_data(void);
 #else
diff --git a/erts/emulator/beam/erl_threads.h b/erts/emulator/beam/erl_threads.h
index 84a20b51f2..8c9cace0c5 100644
--- a/erts/emulator/beam/erl_threads.h
+++ b/erts/emulator/beam/erl_threads.h
@@ -1,7 +1,7 @@
 /*
  * %CopyrightBegin%
  *
- * Copyright Ericsson AB 2001-2010. All Rights Reserved.
+ * Copyright Ericsson AB 2001-2011. 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
@@ -36,6 +36,16 @@
 #include "erl_lock_count.h"
 #include "erl_term.h"
 
+#if defined(__GLIBC__) && (__GLIBC__ << 16) + __GLIBC_MINOR__ < (2 << 16) + 4
+/*
+ * pthread_mutex_destroy() may return EBUSY when it shouldn't :( We have
+ * only seen this bug in glibc versions before 2.4. Note that condition
+ * variables, rwmutexes, spinlocks, and rwspinlocks also may be effected by
+ * this bug since these implementations may use mutexes internally.
+ */
+#  define ERTS_THR_HAVE_BUSY_DESTROY_BUG
+#endif
+
 #define ERTS_THR_MEMORY_BARRIER ETHR_MEMORY_BARRIER
 
 #ifdef ERTS_ENABLE_LOCK_COUNT
@@ -175,6 +185,9 @@ typedef struct { int gcc_is_buggy; } erts_rwlock_t;
 #endif /* #ifdef USE_THREADS */
 
 #define ERTS_AINT_T_MAX (~(((erts_aint_t) 1) << (sizeof(erts_aint_t)*8-1)))
+#define ERTS_AINT_T_MIN ((((erts_aint_t) 1) << (sizeof(erts_aint_t)*8-1)))
+#define ERTS_AINT32_T_MAX (~(((erts_aint32_t) 1) << (sizeof(erts_aint32_t)*8-1)))
+#define ERTS_AINT32_T_MIN ((((erts_aint32_t) 1) << (sizeof(erts_aint32_t)*8-1)))
 
 ERTS_GLB_INLINE void erts_thr_init(erts_thr_init_data_t *id);
 ERTS_GLB_INLINE void erts_thr_late_init(erts_thr_late_init_data_t *id);
@@ -551,8 +564,16 @@ erts_mtx_destroy(erts_mtx_t *mtx)
     erts_lcnt_destroy_lock(&mtx->lcnt);
 #endif
     res = ethr_mutex_destroy(&mtx->mtx);
-    if (res)
+    if (res != 0) {
+#ifdef ERTS_THR_HAVE_BUSY_DESTROY_BUG
+	if (res == EBUSY) {
+	    char *warn = "Ignoring busy mutex destroy. "
+		"Most likely a bug in pthread implementation.";
+	    erts_send_warning_to_logger_str_nogl(warn);
+	}
+#endif
 	erts_thr_fatal_error(res, "destroy mutex");
+    }
 #endif
 }
 
@@ -647,8 +668,16 @@ erts_cnd_destroy(erts_cnd_t *cnd)
 {
 #ifdef USE_THREADS
     int res = ethr_cond_destroy(cnd);
-    if (res)
+    if (res != 0) {
+#ifdef ERTS_THR_HAVE_BUSY_DESTROY_BUG
+	if (res == EBUSY) {
+	    char *warn = "Ignoring busy cond destroy. "
+		"Most likely a bug in pthread implementation.";
+	    erts_send_warning_to_logger_str_nogl(warn);
+	}
+#endif
 	erts_thr_fatal_error(res, "destroy condition variable");
+    }
 #endif
 }
 
@@ -774,8 +803,16 @@ erts_rwmtx_destroy(erts_rwmtx_t *rwmtx)
     erts_lcnt_destroy_lock(&rwmtx->lcnt);
 #endif
     res = ethr_rwmutex_destroy(&rwmtx->rwmtx);
-    if (res != 0)
+    if (res != 0) {
+#ifdef ERTS_THR_HAVE_BUSY_DESTROY_BUG
+	if (res == EBUSY) {
+	    char *warn = "Ignoring busy rwmutex destroy. "
+		"Most likely a bug in pthread implementation.";
+	    erts_send_warning_to_logger_str_nogl(warn);
+	}
+#endif
 	erts_thr_fatal_error(res, "destroy rwmutex");
+    }
 #endif
 }
 
@@ -1452,8 +1489,16 @@ erts_spinlock_destroy(erts_spinlock_t *lock)
     erts_lcnt_destroy_lock(&lock->lcnt);
 #endif
     res = ethr_spinlock_destroy(&lock->slck);
-    if (res)
-	erts_thr_fatal_error(res, "destroy spinlock");
+    if (res != 0) {
+#ifdef ERTS_THR_HAVE_BUSY_DESTROY_BUG
+	if (res == EBUSY) {
+	    char *warn = "Ignoring busy spinlock destroy. "
+		"Most likely a bug in pthread implementation.";
+	    erts_send_warning_to_logger_str_nogl(warn);
+	}
+#endif
+	erts_thr_fatal_error(res, "destroy rwlock");
+    }
 #else
     (void)lock;
 #endif
@@ -1562,8 +1607,16 @@ erts_rwlock_destroy(erts_rwlock_t *lock)
     erts_lcnt_destroy_lock(&lock->lcnt);
 #endif
     res = ethr_rwlock_destroy(&lock->rwlck);
-    if (res)
+    if (res != 0) {
+#ifdef ERTS_THR_HAVE_BUSY_DESTROY_BUG
+	if (res == EBUSY) {
+	    char *warn = "Ignoring busy rwlock destroy. "
+		"Most likely a bug in pthread implementation.";
+	    erts_send_warning_to_logger_str_nogl(warn);
+	}
+#endif
 	erts_thr_fatal_error(res, "destroy rwlock");
+    }
 #else
     (void)lock;
 #endif
diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h
index b1cd683b3b..ff828ae889 100644
--- a/erts/emulator/beam/sys.h
+++ b/erts/emulator/beam/sys.h
@@ -331,12 +331,16 @@ typedef unsigned char byte;
     (((size_t) 8) - (((size_t) (X)) & ((size_t) 7)))
 
 #include "erl_lock_check.h"
+
+/* needed by erl_smp.h */
+int erts_send_warning_to_logger_str_nogl(char *);
+
 #include "erl_smp.h"
 
 #ifdef ERTS_WANT_BREAK_HANDLING
 #  ifdef ERTS_SMP
-extern erts_smp_atomic_t erts_break_requested;
-#    define ERTS_BREAK_REQUESTED ((int) erts_smp_atomic_read(&erts_break_requested))
+extern erts_smp_atomic32_t erts_break_requested;
+#    define ERTS_BREAK_REQUESTED ((int) erts_smp_atomic32_read(&erts_break_requested))
 #  else
 extern volatile int erts_break_requested;
 #    define ERTS_BREAK_REQUESTED erts_break_requested
@@ -349,8 +353,8 @@ void erts_do_break_handling(void);
 #    define ERTS_GOT_SIGUSR1 0
 #  else
 #    ifdef ERTS_SMP
-extern erts_smp_atomic_t erts_got_sigusr1;
-#      define ERTS_GOT_SIGUSR1 ((int) erts_smp_atomic_read(&erts_got_sigusr1))
+extern erts_smp_atomic32_t erts_got_sigusr1;
+#      define ERTS_GOT_SIGUSR1 ((int) erts_smp_atomic32_read(&erts_got_sigusr1))
 #    else
 extern volatile int erts_got_sigusr1;
 #      define ERTS_GOT_SIGUSR1 erts_got_sigusr1
@@ -517,7 +521,8 @@ int erts_send_info_to_logger_nogl(erts_dsprintf_buf_t *);
 int erts_send_warning_to_logger_nogl(erts_dsprintf_buf_t *);
 int erts_send_error_to_logger_nogl(erts_dsprintf_buf_t *);
 int erts_send_info_to_logger_str_nogl(char *);
-int erts_send_warning_to_logger_str_nogl(char *);
+/* needed by erl_smp.h (declared above)
+   int erts_send_warning_to_logger_str_nogl(char *); */
 int erts_send_error_to_logger_str_nogl(char *);
 
 typedef struct preload {
diff --git a/erts/emulator/sys/common/erl_poll.c b/erts/emulator/sys/common/erl_poll.c
index 4d0ca97889..77ac2de5f6 100644
--- a/erts/emulator/sys/common/erl_poll.c
+++ b/erts/emulator/sys/common/erl_poll.c
@@ -1,7 +1,7 @@
 /*
  * %CopyrightBegin%
  *
- * Copyright Ericsson AB 2006-2010. All Rights Reserved.
+ * Copyright Ericsson AB 2006-2011. 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
@@ -124,25 +124,11 @@
   erts_smp_mtx_unlock(&(PS)->mtx)
 
 #define ERTS_POLLSET_SET_POLLED_CHK(PS) \
-  ((int) erts_smp_atomic_xchg(&(PS)->polled, (erts_aint_t) 1))
+  ((int) erts_atomic32_xchg(&(PS)->polled, (erts_aint32_t) 1))
 #define ERTS_POLLSET_UNSET_POLLED(PS) \
-  erts_smp_atomic_set(&(PS)->polled, (erts_aint_t) 0)
+  erts_atomic32_set(&(PS)->polled, (erts_aint32_t) 0)
 #define ERTS_POLLSET_IS_POLLED(PS) \
-  ((int) erts_smp_atomic_read(&(PS)->polled))
-
-#define ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) set_poller_woken_chk((PS))
-#define ERTS_POLLSET_SET_POLLER_WOKEN(PS) 				\
-do {									\
-      ERTS_THR_MEMORY_BARRIER;						\
-      erts_smp_atomic_set(&(PS)->woken, (erts_aint_t) 1);		\
-} while (0)
-#define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS)				\
-do {									\
-    erts_smp_atomic_set(&(PS)->woken, (erts_aint_t) 0);			\
-    ERTS_THR_MEMORY_BARRIER;						\
-} while (0)
-#define ERTS_POLLSET_IS_POLLER_WOKEN(PS)				\
-  ((int) erts_smp_atomic_read(&(PS)->woken))
+  ((int) erts_atomic32_read(&(PS)->polled))
 
 #else
 
@@ -152,69 +138,21 @@ do {									\
 #define ERTS_POLLSET_UNSET_POLLED(PS)
 #define ERTS_POLLSET_IS_POLLED(PS) 0
 
-#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT
-
-/*
- * Ideally, the ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) operation would
- * be atomic. This operation isn't, but we will do okay anyway. The
- * "woken check" is only an optimization. The only requirement we have:
- * If (PS)->woken is set to a value != 0 when interrupting, we have to
- * write on the the wakeup pipe at least once. Multiple writes are okay.
- */
-#define ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) ((PS)->woken++)
-#define ERTS_POLLSET_SET_POLLER_WOKEN(PS) ((PS)->woken = 1, (void) 0)
-#define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS) ((PS)->woken = 0, (void) 0)
-#define ERTS_POLLSET_IS_POLLER_WOKEN(PS) ((PS)->woken)
-
-#else
-
-#define ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) 1
-#define ERTS_POLLSET_SET_POLLER_WOKEN(PS)
-#define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS)
-#define ERTS_POLLSET_IS_POLLER_WOKEN(PS) 1
-
-#endif
-
 #endif
 
 #if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE
 #define ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(PS) \
-  erts_smp_atomic_set(&(PS)->have_update_requests, (erts_aint_t) 1)
+  erts_smp_atomic32_set(&(PS)->have_update_requests, (erts_aint32_t) 1)
 #define ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(PS) \
-  erts_smp_atomic_set(&(PS)->have_update_requests, (erts_aint_t) 0)
+  erts_smp_atomic32_set(&(PS)->have_update_requests, (erts_aint32_t) 0)
 #define ERTS_POLLSET_HAVE_UPDATE_REQUESTS(PS) \
-  ((int) erts_smp_atomic_read(&(PS)->have_update_requests))
+  ((int) erts_smp_atomic32_read(&(PS)->have_update_requests))
 #else
 #define ERTS_POLLSET_SET_HAVE_UPDATE_REQUESTS(PS)
 #define ERTS_POLLSET_UNSET_HAVE_UPDATE_REQUESTS(PS)
 #define ERTS_POLLSET_HAVE_UPDATE_REQUESTS(PS) 0
 #endif
 
-#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT && !defined(ERTS_SMP)
-
-#define ERTS_POLLSET_UNSET_INTERRUPTED_CHK(PS) unset_interrupted_chk((PS))
-#define ERTS_POLLSET_UNSET_INTERRUPTED(PS) ((PS)->interrupt = 0, (void) 0)
-#define ERTS_POLLSET_SET_INTERRUPTED(PS) ((PS)->interrupt = 1, (void) 0)
-#define ERTS_POLLSET_IS_INTERRUPTED(PS) ((PS)->interrupt)
-
-#else
-
-#define ERTS_POLLSET_UNSET_INTERRUPTED_CHK(PS) unset_interrupted_chk((PS))
-#define ERTS_POLLSET_UNSET_INTERRUPTED(PS)				\
-do {									\
-    erts_smp_atomic_set(&(PS)->interrupt, (erts_aint_t) 0);		\
-    ERTS_THR_MEMORY_BARRIER;						\
-} while (0)
-#define ERTS_POLLSET_SET_INTERRUPTED(PS) 				\
-do {									\
-      ERTS_THR_MEMORY_BARRIER;						\
-      erts_smp_atomic_set(&(PS)->interrupt, (erts_aint_t) 1);		\
-} while (0)
-#define ERTS_POLLSET_IS_INTERRUPTED(PS)					\
-  ((int) erts_smp_atomic_read(&(PS)->interrupt))
-
-#endif
-
 #if ERTS_POLL_USE_FALLBACK
 #  if ERTS_POLL_USE_POLL
 #    define ERTS_POLL_NEED_FALLBACK(PS) ((PS)->no_poll_fds > 1)
@@ -318,14 +256,12 @@ struct ErtsPollSet_ {
 #if ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE
     ErtsPollSetUpdateRequestsBlock update_requests;
     ErtsPollSetUpdateRequestsBlock *curr_upd_req_block;
-    erts_smp_atomic_t have_update_requests;
+    erts_smp_atomic32_t have_update_requests;
 #endif
 #ifdef ERTS_SMP
-    erts_smp_atomic_t polled;
-    erts_smp_atomic_t woken;
+    erts_atomic32_t polled;
     erts_smp_mtx_t mtx;
 #elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT
-    volatile int woken;
 #endif
 #if ERTS_POLL_USE_WAKEUP_PIPE
     int wake_fds[2];
@@ -333,12 +269,12 @@ struct ErtsPollSet_ {
 #if ERTS_POLL_USE_FALLBACK
     int fallback_used;
 #endif
-#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT && !defined(ERTS_SMP)
-    volatile int interrupt;
-#else
-    erts_smp_atomic_t interrupt;
+#ifdef ERTS_SMP
+    erts_atomic32_t wakeup_state;
+#elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT
+    volatile int wakeup_state;
 #endif
-    erts_smp_atomic_t timeout;
+    erts_smp_atomic32_t timeout;
 #ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS
     erts_smp_atomic_t no_avoided_wakeups;
     erts_smp_atomic_t no_avoided_interrupts;
@@ -346,34 +282,6 @@ struct ErtsPollSet_ {
 #endif
 };
 
-static ERTS_INLINE int
-unset_interrupted_chk(ErtsPollSet ps)
-{
-    int res;
-#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT && !defined(ERTS_SMP)
-    /* This operation isn't atomic, but we have no need at all for an
-       atomic operation here... */
-    res = ps->interrupt;
-    ps->interrupt = 0;
-#else
-    res = (int) erts_smp_atomic_xchg(&ps->interrupt, (erts_aint_t) 0);
-    ERTS_THR_MEMORY_BARRIER;
-#endif
-    return res;
-
-}
-
-#ifdef ERTS_SMP
-
-static ERTS_INLINE int
-set_poller_woken_chk(ErtsPollSet ps)
-{
-    ERTS_THR_MEMORY_BARRIER;
-    return (int) erts_smp_atomic_xchg(&ps->woken, (erts_aint_t) 1);
-}
-
-#endif
-
 void erts_silence_warn_unused_result(long unused);
 static void fatal_error(char *format, ...);
 static void fatal_error_async_signal_safe(char *error_str);
@@ -430,6 +338,63 @@ static void check_poll_status(ErtsPollSet ps);
 static void print_misc_debug_info(void);
 #endif
 
+#define ERTS_POLL_NOT_WOKEN	0
+#define ERTS_POLL_WOKEN		-1
+#define ERTS_POLL_WOKEN_INTR	1
+
+static ERTS_INLINE void
+reset_wakeup_state(ErtsPollSet ps)
+{
+#ifdef ERTS_SMP
+    erts_atomic32_set(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN);
+#elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT
+    ps->wakeup_state = 0;
+#endif
+}
+
+static ERTS_INLINE int
+is_woken(ErtsPollSet ps)
+{
+#ifdef ERTS_SMP
+    return erts_atomic32_read_acqb(&ps->wakeup_state) != ERTS_POLL_NOT_WOKEN;
+#elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT
+    return ps->wakeup_state != ERTS_POLL_NOT_WOKEN;
+#else
+    return 0;
+#endif
+}
+
+static ERTS_INLINE int
+is_interrupted_reset(ErtsPollSet ps)
+{
+#ifdef ERTS_SMP
+    return (erts_atomic32_xchg(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN)
+	    == ERTS_POLL_WOKEN_INTR);
+#elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT
+    int res = ps->wakeup_state == ERTS_POLL_WOKEN_INTR;
+    ps->wakeup_state = ERTS_POLL_NOT_WOKEN;
+    return res;
+#else
+    return 0;
+#endif
+}
+
+static ERTS_INLINE void
+woke_up(ErtsPollSet ps)
+{
+#ifdef ERTS_SMP
+    erts_aint32_t wakeup_state = erts_atomic32_read(&ps->wakeup_state);
+    if (wakeup_state == ERTS_POLL_NOT_WOKEN)
+	(void) erts_atomic32_cmpxchg(&ps->wakeup_state,
+				     ERTS_POLL_WOKEN,
+				     ERTS_POLL_NOT_WOKEN);
+    ASSERT(erts_atomic32_read(&ps->wakeup_state) != ERTS_POLL_NOT_WOKEN);
+#elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT
+    if (ps->wakeup_state == ERTS_POLL_NOT_WOKEN)
+	ps->wakeup_state = ERTS_POLL_WOKEN;
+#endif
+}
+
 /*
  * --- Wakeup pipe -----------------------------------------------------------
  */
@@ -437,14 +402,34 @@ static void print_misc_debug_info(void);
 #if ERTS_POLL_USE_WAKEUP_PIPE
 
 static ERTS_INLINE void
-wake_poller(ErtsPollSet ps)
+wake_poller(ErtsPollSet ps, int interrupted)
 {
+    int wake;
+#ifdef ERTS_SMP
+    erts_aint32_t wakeup_state;
+    if (!interrupted)
+	wakeup_state = erts_atomic32_cmpxchg_relb(&ps->wakeup_state,
+						  ERTS_POLL_WOKEN,
+						  ERTS_POLL_NOT_WOKEN);
+    else {
+	/*
+	 * We might unnecessarily write to the pipe, however,
+	 * that isn't problematic.
+	 */
+	wakeup_state = erts_atomic32_read(&ps->wakeup_state);
+	erts_atomic32_set_relb(&ps->wakeup_state, ERTS_POLL_WOKEN_INTR);
+    }
+    wake = wakeup_state == ERTS_POLL_NOT_WOKEN;
+#elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT
+    wake = ps->wakeup_state == ERTS_POLL_NOT_WOKEN;
+    ps->wakeup_state = interrupted ? ERTS_POLL_WOKEN_INTR : ERTS_POLL_NOT_WOKEN;
+#endif
     /*
      * NOTE: This function might be called from signal handlers in the
      *       non-smp case; therefore, it has to be async-signal safe in
      *       the non-smp case.
      */
-    if (!ERTS_POLLSET_SET_POLLER_WOKEN_CHK(ps)) {
+    if (wake) {
 	ssize_t res;
 	if (ps->wake_fds[1] < 0)
 	    return; /* Not initialized yet */
@@ -1387,9 +1372,7 @@ handle_update_requests(ErtsPollSet ps)
 #endif /* ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE */
 
 static ERTS_INLINE ErtsPollEvents
-poll_control(ErtsPollSet ps, int fd, ErtsPollEvents events, int on,
-	     int *have_set_have_update_requests,
-	     int *do_wake)
+poll_control(ErtsPollSet ps, int fd, ErtsPollEvents events, int on, int *do_wake)
 {
     ErtsPollEvents new_events;
 
@@ -1493,7 +1476,6 @@ ERTS_POLL_EXPORT(erts_poll_controlv)(ErtsPollSet ps,
 				     int len)
 {
     int i;
-    int hshur = 0;
     int do_wake;
     int final_do_wake = 0;
 
@@ -1505,17 +1487,17 @@ ERTS_POLL_EXPORT(erts_poll_controlv)(ErtsPollSet ps,
 				      pcev[i].fd,
 				      pcev[i].events,
 				      pcev[i].on,
-				      &hshur,
 				      &do_wake);
 	final_do_wake |= do_wake;
     }
 
+    ERTS_POLLSET_UNLOCK(ps);
+
 #ifdef ERTS_SMP
     if (final_do_wake)
-	wake_poller(ps);
+	wake_poller(ps, 0);
 #endif /* ERTS_SMP */
 
-    ERTS_POLLSET_UNLOCK(ps);
 }
 
 ErtsPollEvents
@@ -1526,20 +1508,20 @@ ERTS_POLL_EXPORT(erts_poll_control)(ErtsPollSet ps,
 				    int* do_wake) /* In: Wake up polling thread */
 				                  /* Out: Poller is woken */
 {
-    int hshur = 0;
     ErtsPollEvents res;
 
     ERTS_POLLSET_LOCK(ps);
 
-    res = poll_control(ps, fd, events, on, &hshur, do_wake);
+    res = poll_control(ps, fd, events, on, do_wake);
+
+    ERTS_POLLSET_UNLOCK(ps);
 
 #ifdef ERTS_SMP
     if (*do_wake) {
-	wake_poller(ps);
+	wake_poller(ps, 0);
     }
 #endif /* ERTS_SMP */
 
-    ERTS_POLLSET_UNLOCK(ps);
     return res;
 }
 
@@ -1918,9 +1900,11 @@ check_fd_events(ErtsPollSet ps, SysTimeval *tv, int max_res, int *ps_locked)
 	return 0;
     }
     else {
-	erts_aint_t timeout = tv->tv_sec*1000 + tv->tv_usec/1000;
+	long timeout = tv->tv_sec*1000 + tv->tv_usec/1000;
+	if (timeout > ERTS_AINT32_T_MAX)
+	    timeout = ERTS_AINT32_T_MAX;
 	ASSERT(timeout >= 0);
-	erts_smp_atomic_set(&ps->timeout, timeout);
+	erts_smp_atomic32_set_relb(&ps->timeout, (erts_aint32_t) timeout);
 #if ERTS_POLL_USE_FALLBACK
 	if (!(ps->fallback_used = ERTS_POLL_NEED_FALLBACK(ps))) {
 
@@ -2042,15 +2026,14 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps,
 		 (int) tv->tv_sec*1000 + tv->tv_usec/1000);
 #endif
 
-    ERTS_POLLSET_UNSET_POLLER_WOKEN(ps);
     if (ERTS_POLLSET_SET_POLLED_CHK(ps)) {
 	res = EINVAL; /* Another thread is in erts_poll_wait()
 			 on this pollset... */
 	goto done;
     }
 
-    if (ERTS_POLLSET_IS_INTERRUPTED(ps)) {
-	/* Interrupt use zero timeout */
+    if (is_woken(ps)) {
+	/* Use zero timeout */
 	itv.tv_sec = 0;
 	itv.tv_usec = 0;
 	tvp = &itv;
@@ -2067,7 +2050,7 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps,
     ps_locked = 0;
     res = check_fd_events(ps, tvp, no_fds, &ps_locked);
 
-    ERTS_POLLSET_SET_POLLER_WOKEN(ps);
+    woke_up(ps);
 
     if (res == 0) {
 	res = ETIMEDOUT;
@@ -2099,9 +2082,7 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps,
 	check_poll_result(pr, no_fds);
 #endif
 
-	res = (no_fds == 0
-	       ? (ERTS_POLLSET_UNSET_INTERRUPTED_CHK(ps) ? EINTR : EAGAIN)
-	       : 0);
+	res = (no_fds == 0 ? (is_interrupted_reset(ps) ? EINTR : EAGAIN) : 0);
 	*len = no_fds;
     }
 
@@ -2112,7 +2093,7 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps,
 #endif
 
  done:
-    erts_smp_atomic_set(&ps->timeout, ERTS_AINT_T_MAX);
+    erts_smp_atomic32_set_relb(&ps->timeout, ERTS_AINT32_T_MAX);
 #ifdef ERTS_POLL_DEBUG_PRINT
     erts_printf("Leaving %s = erts_poll_wait()\n",
 		 res == 0 ? "0" : erl_errno_id(res));
@@ -2128,20 +2109,17 @@ ERTS_POLL_EXPORT(erts_poll_wait)(ErtsPollSet ps,
 void
 ERTS_POLL_EXPORT(erts_poll_interrupt)(ErtsPollSet ps, int set)
 {
+#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT || defined(ERTS_SMP)
     /*
      * NOTE: This function might be called from signal handlers in the
      *       non-smp case; therefore, it has to be async-signal safe in
      *       the non-smp case.
      */
-    if (set) {
-	ERTS_POLLSET_SET_INTERRUPTED(ps);
-#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT || defined(ERTS_SMP)
-	wake_poller(ps);
+    if (!set)
+	reset_wakeup_state(ps);
+    else
+	wake_poller(ps, 1);
 #endif
-    }
-    else {
-	ERTS_POLLSET_UNSET_INTERRUPTED(ps);
-    }
 }
 
 /*
@@ -2154,13 +2132,12 @@ ERTS_POLL_EXPORT(erts_poll_interrupt_timed)(ErtsPollSet ps,
 					    int set,
 					    long msec)
 {
-    if (set) {
-	if (erts_smp_atomic_read(&ps->timeout) > (erts_aint_t) msec) {
-	    ERTS_POLLSET_SET_INTERRUPTED(ps);
 #if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT || defined(ERTS_SMP)
-	    wake_poller(ps);
-#endif
-	}
+    if (!set)
+	reset_wakeup_state(ps);
+    else {
+	if (erts_smp_atomic32_read_acqb(&ps->timeout) > (erts_aint32_t) msec)
+	    wake_poller(ps, 1);
 #ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS
 	else {
 	    if (ERTS_POLLSET_IS_POLLED(ps))
@@ -2170,9 +2147,7 @@ ERTS_POLL_EXPORT(erts_poll_interrupt_timed)(ErtsPollSet ps,
 	erts_smp_atomic_inc(&ps->no_interrupt_timed);
 #endif
     }
-    else {
-	ERTS_POLLSET_UNSET_INTERRUPTED(ps);
-    }
+#endif
 }
 
 int
@@ -2283,14 +2258,16 @@ ERTS_POLL_EXPORT(erts_poll_create_pollset)(void)
     ps->update_requests.next = NULL;
     ps->update_requests.len = 0;
     ps->curr_upd_req_block = &ps->update_requests;
-    erts_smp_atomic_init(&ps->have_update_requests, 0);
+    erts_smp_atomic32_init(&ps->have_update_requests, 0);
 #endif
 #ifdef ERTS_SMP
-    erts_smp_atomic_init(&ps->polled, 0);
-    erts_smp_atomic_init(&ps->woken, 0);
+    erts_atomic32_init(&ps->polled, 0);
     erts_smp_mtx_init(&ps->mtx, "pollset");
+#endif
+#ifdef ERTS_SMP
+    erts_atomic32_init(&ps->wakeup_state, (erts_aint32_t) 0);
 #elif ERTS_POLL_ASYNC_INTERRUPT_SUPPORT
-    ps->woken = 0;
+    ps->wakeup_state = 0;
 #endif
 #if ERTS_POLL_USE_WAKEUP_PIPE
     create_wakeup_pipe(ps);
@@ -2312,12 +2289,7 @@ ERTS_POLL_EXPORT(erts_poll_create_pollset)(void)
 	ps->internal_fd_limit = kp_fd + 1;
     ps->kp_fd = kp_fd;
 #endif
-#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT && !defined(ERTS_SMP)
-    ps->interrupt = 0;
-#else
-    erts_smp_atomic_init(&ps->interrupt, 0);
-#endif
-    erts_smp_atomic_init(&ps->timeout, ERTS_AINT_T_MAX);
+    erts_smp_atomic32_init(&ps->timeout, ERTS_AINT32_T_MAX);
 #ifdef ERTS_POLL_COUNT_AVOIDED_WAKEUPS
     erts_smp_atomic_init(&ps->no_avoided_wakeups, 0);
     erts_smp_atomic_init(&ps->no_avoided_interrupts, 0);
diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c
index bfc04faa45..bc940d2084 100644
--- a/erts/emulator/sys/unix/sys.c
+++ b/erts/emulator/sys/unix/sys.c
@@ -1,7 +1,7 @@
 /*
  * %CopyrightBegin%
  *
- * Copyright Ericsson AB 1996-2010. All Rights Reserved.
+ * Copyright Ericsson AB 1996-2011. 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
@@ -160,14 +160,14 @@ static int debug_log = 0;
 #endif
 
 #ifdef ERTS_SMP
-erts_smp_atomic_t erts_got_sigusr1;
+erts_smp_atomic32_t erts_got_sigusr1;
 #define ERTS_SET_GOT_SIGUSR1 \
-  erts_smp_atomic_set(&erts_got_sigusr1, 1)
+  erts_smp_atomic32_set(&erts_got_sigusr1, 1)
 #define ERTS_UNSET_GOT_SIGUSR1 \
-  erts_smp_atomic_set(&erts_got_sigusr1, 0)
-static erts_smp_atomic_t have_prepared_crash_dump;
+  erts_smp_atomic32_set(&erts_got_sigusr1, 0)
+static erts_smp_atomic32_t have_prepared_crash_dump;
 #define ERTS_PREPARED_CRASH_DUMP \
-  ((int) erts_smp_atomic_xchg(&have_prepared_crash_dump, 1))
+  ((int) erts_smp_atomic32_xchg(&have_prepared_crash_dump, 1))
 #else
 volatile int erts_got_sigusr1;
 #define ERTS_SET_GOT_SIGUSR1 (erts_got_sigusr1 = 1)
@@ -235,11 +235,11 @@ static int max_files = -1;
  * a few variables used by the break handler 
  */
 #ifdef ERTS_SMP
-erts_smp_atomic_t erts_break_requested;
+erts_smp_atomic32_t erts_break_requested;
 #define ERTS_SET_BREAK_REQUESTED \
-  erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 1)
+  erts_smp_atomic32_set(&erts_break_requested, (erts_aint32_t) 1)
 #define ERTS_UNSET_BREAK_REQUESTED \
-  erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 0)
+  erts_smp_atomic32_set(&erts_break_requested, (erts_aint32_t) 0)
 #else
 volatile int erts_break_requested = 0;
 #define ERTS_SET_BREAK_REQUESTED (erts_break_requested = 1)
@@ -504,9 +504,9 @@ erts_sys_pre_init(void)
 #endif
     }
 #ifdef ERTS_SMP
-    erts_smp_atomic_init(&erts_break_requested, 0);
-    erts_smp_atomic_init(&erts_got_sigusr1, 0);
-    erts_smp_atomic_init(&have_prepared_crash_dump, 0);
+    erts_smp_atomic32_init(&erts_break_requested, 0);
+    erts_smp_atomic32_init(&erts_got_sigusr1, 0);
+    erts_smp_atomic32_init(&have_prepared_crash_dump, 0);
 #else
     erts_break_requested = 0;
     erts_got_sigusr1 = 0;
diff --git a/erts/emulator/sys/win32/erl_poll.c b/erts/emulator/sys/win32/erl_poll.c
index d84ae2ede2..1f2877b682 100644
--- a/erts/emulator/sys/win32/erl_poll.c
+++ b/erts/emulator/sys/win32/erl_poll.c
@@ -1,7 +1,7 @@
 /*
  * %CopyrightBegin%
  *
- * Copyright Ericsson AB 2007-2010. All Rights Reserved.
+ * Copyright Ericsson AB 2007-2011. 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
@@ -274,7 +274,6 @@ struct ErtsPollSet_ {
     Waiter** waiter;
     int allocated_waiters;  /* Size ow waiter array */ 
     int num_waiters;	    /* Number of waiter threads. */
-    erts_atomic_t sys_io_ready; /* Tells us there is I/O ready (already). */
     int restore_events;        /* Tells us to restore waiters events 
 				  next time around */
     HANDLE event_io_ready;     /* To be used when waiting for io */
@@ -282,12 +281,11 @@ struct ErtsPollSet_ {
     volatile int standby_wait_counter; /* Number of threads to wait for */
     CRITICAL_SECTION standby_crit;     /* CS to guard the counter */
     HANDLE standby_wait_event;         /* Event signalled when counte == 0 */
+    erts_atomic32_t wakeup_state;
 #ifdef ERTS_SMP
-    erts_smp_atomic_t woken;
     erts_smp_mtx_t mtx;
-    erts_smp_atomic_t interrupt;
 #endif
-    erts_smp_atomic_t timeout;
+    erts_smp_atomic32_t timeout;
 };
 
 #ifdef ERTS_SMP
@@ -296,126 +294,24 @@ struct ErtsPollSet_ {
   erts_smp_mtx_lock(&(PS)->mtx)
 #define ERTS_POLLSET_UNLOCK(PS) \
   erts_smp_mtx_unlock(&(PS)->mtx)
-#define ERTS_POLLSET_SET_POLLED_CHK(PS) \
-  ((int) erts_smp_atomic_xchg(&(PS)->polled, (erts_aint_t) 1))
-#define ERTS_POLLSET_SET_POLLED(PS) \
-  erts_smp_atomic_set(&(PS)->polled, (erts_aint_t) 1)
-#define ERTS_POLLSET_UNSET_POLLED(PS) \
-  erts_smp_atomic_set(&(PS)->polled, (erts_aint_t) 0)
-#define ERTS_POLLSET_IS_POLLED(PS) \
-  ((int) erts_smp_atomic_read(&(PS)->polled))
-
-#define ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) set_poller_woken_chk((PS))
-#define ERTS_POLLSET_SET_POLLER_WOKEN(PS) 				\
-do {									\
-      ERTS_THR_MEMORY_BARRIER;						\
-      erts_smp_atomic_set(&(PS)->woken, (erts_aint_t) 1);			\
-} while (0)
-#define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS)				\
-do {									\
-    erts_smp_atomic_set(&(PS)->woken, (erts_aint_t) 0);			\
-    ERTS_THR_MEMORY_BARRIER;						\
-} while (0)
-#define ERTS_POLLSET_IS_POLLER_WOKEN(PS)				\
-  ((int) erts_smp_atomic_read(&(PS)->woken))
-
-#define ERTS_POLLSET_UNSET_INTERRUPTED_CHK(PS) unset_interrupted_chk((PS))
-#define ERTS_POLLSET_UNSET_INTERRUPTED(PS)				\
-do {									\
-    erts_smp_atomic_set(&(PS)->interrupt, (erts_aint_t) 0);			\
-    ERTS_THR_MEMORY_BARRIER;						\
-} while (0)
-#define ERTS_POLLSET_SET_INTERRUPTED(PS) 				\
-do {									\
-      ERTS_THR_MEMORY_BARRIER;						\
-      erts_smp_atomic_set(&(PS)->interrupt, (erts_aint_t) 1);			\
-} while (0)
-#define ERTS_POLLSET_IS_INTERRUPTED(PS)					\
-  ((int) erts_smp_atomic_read(&(PS)->interrupt))
-
-static ERTS_INLINE int
-unset_interrupted_chk(ErtsPollSet ps)
-{
-    int res = (int) erts_smp_atomic_xchg(&ps->interrupt, (erts_aint_t) 0);
-    ERTS_THR_MEMORY_BARRIER;
-    return res;
-
-}
-
-static ERTS_INLINE int
-set_poller_woken_chk(ErtsPollSet ps)
-{
-    ERTS_THR_MEMORY_BARRIER;
-    return (int) erts_smp_atomic_xchg(&ps->woken, (erts_aint_t) 1);
-}
 
 #else
 
 #define ERTS_POLLSET_LOCK(PS)
 #define ERTS_POLLSET_UNLOCK(PS)
-#define ERTS_POLLSET_SET_POLLED_CHK(PS) 0
-#define ERTS_POLLSET_UNSET_POLLED(PS)
-#define ERTS_POLLSET_IS_POLLED(PS) 0
-#define ERTS_POLLSET_SET_POLLER_WOKEN_CHK(PS) 1
-#define ERTS_POLLSET_SET_POLLER_WOKEN(PS)
-#define ERTS_POLLSET_UNSET_POLLER_WOKEN(PS)
-#define ERTS_POLLSET_IS_POLLER_WOKEN(PS) 1
-
 
 #endif
 
 /*
- * While atomics are not yet implemented for windows in the common library...
- *
- * MSDN doc states that SMP machines and old compilers require
- * InterLockedExchange to properly read and write interlocked
- * variables, otherwise the processors might reschedule
- * the access and order of atomics access is destroyed... 
- * While they only mention it in white-papers, the problem
- * in VS2003 is due to the IA64 arch, so we can still count
- * on the CPU not rescheduling the access to volatile in X86 arch using 
- * even the slightly older compiler...
- *
- * So here's (hopefully) a subset of the generally working atomic 
- * variable access...
- */
-
-#if defined(__GNUC__)
-#  if defined(__i386__) || defined(__x86_64__)
-#    define VOLATILE_IN_SEQUENCE 1
-#  else
-#    define VOLATILE_IN_SEQUENCE 0
-#  endif 
-#elif defined(_MSC_VER)
-#  if _MSC_VER < 1300
-#    define VOLATILE_IN_SEQUENCE 0 /* Dont trust really old compilers */
-#  else
-#    if defined(_M_IX86)
-#      define VOLATILE_IN_SEQUENCE 1
-#    else /* I.e. IA64 */
-#      if _MSC_VER >= 1400 
-#        define VOLATILE_IN_SEQUENCE 1
-#      else
-#        define VOLATILE_IN_SEQUENCE 0
-#      endif
-#    endif
-#  endif
-#else
-# define VOLATILE_IN_SEQUENCE 0
-#endif
- 
-
-
-/*
  * Communication with sys_interrupt
  */
 
 #ifdef ERTS_SMP
-extern erts_smp_atomic_t erts_break_requested;
+extern erts_smp_atomic32_t erts_break_requested;
 #define ERTS_SET_BREAK_REQUESTED \
-  erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 1)
+  erts_smp_atomic32_set(&erts_break_requested, (erts_aint32_t) 1)
 #define ERTS_UNSET_BREAK_REQUESTED \
-  erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 0)
+  erts_smp_atomic32_set(&erts_break_requested, (erts_aint32_t) 0)
 #else
 extern volatile int erts_break_requested;
 #define ERTS_SET_BREAK_REQUESTED (erts_break_requested = 1)
@@ -424,7 +320,7 @@ extern volatile int erts_break_requested;
 
 static erts_mtx_t break_waiter_lock;
 static HANDLE break_happened_event;
-static erts_atomic_t break_waiter_state;
+static erts_atomic32_t break_waiter_state;
 #define BREAK_WAITER_GOT_BREAK 1
 #define BREAK_WAITER_GOT_HALT 2
 
@@ -467,29 +363,168 @@ do { \
     wait_standby(PS); \
  } while(0)
 
-#if ERTS_POLL_ASYNC_INTERRUPT_SUPPORT && !defined(ERTS_SMP)
+#define ERTS_POLL_NOT_WOKEN		((erts_aint32_t) 0)
+#define ERTS_POLL_WOKEN_IO_READY	((erts_aint32_t) 1)
+#define ERTS_POLL_WOKEN_INTR		((erts_aint32_t) 2)
+#define ERTS_POLL_WOKEN_TIMEDOUT	((erts_aint32_t) 3)
 
 static ERTS_INLINE int
-unset_interrupted_chk(ErtsPollSet ps)
+is_io_ready(ErtsPollSet ps)
 {
-    /* This operation isn't atomic, but we have no need at all for an
-       atomic operation here... */
-    int res = ps->interrupt;
-    ps->interrupt = 0;
-    return res;
+    return erts_atomic32_read(&ps->wakeup_state) == ERTS_POLL_WOKEN_IO_READY;
 }
 
+static ERTS_INLINE void
+woke_up(ErtsPollSet ps)
+{
+    if (erts_atomic32_read(&ps->wakeup_state) == ERTS_POLL_NOT_WOKEN)
+	erts_atomic32_cmpxchg(&ps->wakeup_state,
+			      ERTS_POLL_WOKEN_TIMEDOUT,
+			      ERTS_POLL_NOT_WOKEN);
+#ifdef DEBUG
+    {
+	erts_aint32_t wakeup_state = erts_atomic32_read(&ps->wakeup_state);
+	switch (wakeup_state) {
+	case ERTS_POLL_WOKEN_IO_READY:
+	case ERTS_POLL_WOKEN_INTR:
+	case ERTS_POLL_WOKEN_TIMEDOUT:
+	    break;
+	default:
+	    ASSERT(0);
+	    break;
+	}
+    }
 #endif
+}
+
+static ERTS_INLINE int
+wakeup_cause(ErtsPollSet ps)
+{
+    int res;
+    erts_aint32_t wakeup_state = erts_atomic32_read(&ps->wakeup_state);
+    switch (wakeup_state) {
+    case ERTS_POLL_WOKEN_IO_READY:
+	res = 0;
+	break;
+    case ERTS_POLL_WOKEN_INTR:
+	res = EINTR;
+	break;
+    case ERTS_POLL_WOKEN_TIMEDOUT:
+	res = ETIMEDOUT;
+	break;
+    default:
+	res = 0;
+	erl_exit(ERTS_ABORT_EXIT,
+		 "%s:%d: Internal error: Invalid wakeup_state=%d\n",
+		 __FILE__, __LINE__, (int) wakeup_state);
+    }
+    return res;
+}
+
+static ERTS_INLINE DWORD
+poll_wait_timeout(ErtsPollSet ps, SysTimeval *tvp)
+{
+    time_t timeout = tvp->tv_sec * 1000 + tvp->tv_usec / 1000;
+
+    if (timeout <= 0) {
+	woke_up(ps);
+	return (DWORD) 0;
+    }
+
+    ResetEvent(ps->event_io_ready);
+    /*
+     * Since we don't know the internals of ResetEvent() we issue
+     * a memory barrier as a safety precaution ensuring that
+     * the load of wakeup_state wont be reordered with stores made
+     * by ResetEvent().
+     */
+    ERTS_THR_MEMORY_BARRIER;
+    if (erts_atomic32_read(&ps->wakeup_state) != ERTS_POLL_NOT_WOKEN)
+	return (DWORD) 0;
+
+    if (timeout > ERTS_AINT32_T_MAX) /* Also prevents DWORD overflow */
+	timeout = ERTS_AINT32_T_MAX;
+
+    erts_smp_atomic32_set_relb(&ps->timeout, (erts_aint32_t) timeout);
+    return (DWORD) timeout;
+}
 
-#ifdef ERTS_SMP
 static ERTS_INLINE void
-wake_poller(ErtsPollSet ps)
+wake_poller(ErtsPollSet ps, int io_ready)
 {
-    if (!ERTS_POLLSET_SET_POLLER_WOKEN_CHK(ps)) {
+    erts_aint32_t wakeup_state = erts_atomic32_read(&ps->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);
+    }
+    else {
+	while (wakeup_state != ERTS_POLL_WOKEN_IO_READY
+	       && wakeup_state != ERTS_POLL_WOKEN_INTR) {
+	    erts_aint32_t act = erts_atomic32_cmpxchg(&ps->wakeup_state,
+						      ERTS_POLL_WOKEN_INTR,
+						      wakeup_state);
+	    if (act == wakeup_state) {
+		wakeup_state = act;
+		break;
+	    }
+	    wakeup_state = act;
+	}
+    }
+    if (wakeup_state == ERTS_POLL_NOT_WOKEN) {
+	/*
+	 * Since we don't know the internals of SetEvent() we issue
+	 * a memory barrier as a safety precaution ensuring that
+	 * the store we just made to wakeup_state wont be reordered
+	 * with loads in SetEvent().
+	 */
+	ERTS_THR_MEMORY_BARRIER;
 	SetEvent(ps->event_io_ready);
     }
 }
-#endif
+
+static ERTS_INLINE void
+reset_io_ready(ErtsPollSet ps)
+{
+    erts_atomic32_set(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN);
+}
+
+static ERTS_INLINE void
+restore_io_ready(ErtsPollSet ps)
+{
+    erts_atomic32_set(&ps->wakeup_state, ERTS_POLL_WOKEN_IO_READY);
+}
+
+/*
+ * notify_io_ready() is used by threads waiting for events, when
+ * notifying a poller thread about I/O ready.
+ */
+static ERTS_INLINE void
+notify_io_ready(ErtsPollSet ps)
+{
+    wake_poller(ps, 1);
+}
+
+static ERTS_INLINE void
+reset_interrupt(ErtsPollSet ps)
+{
+    /* We need to keep io-ready if set */
+    erts_aint32_t wakeup_state = erts_atomic32_read(&ps->wakeup_state);
+    while (wakeup_state != ERTS_POLL_WOKEN_IO_READY
+	   && wakeup_state != ERTS_POLL_NOT_WOKEN) {
+	erts_aint32_t act = erts_atomic32_cmpxchg(&ps->wakeup_state,
+						  ERTS_POLL_NOT_WOKEN,
+						  wakeup_state);
+	if (wakeup_state == act)
+	    break;
+	wakeup_state = act;
+    }
+}
+
+static ERTS_INLINE void
+set_interrupt(ErtsPollSet ps)
+{
+    wake_poller(ps, 0);
+}
 
 static void setup_standby_wait(ErtsPollSet ps, int num_threads)
 {
@@ -653,14 +688,14 @@ static void *break_waiter(void *param)
 	case WAIT_OBJECT_0:
 	    ResetEvent(harr[0]);
 	    erts_mtx_lock(&break_waiter_lock);
-	    erts_atomic_set(&break_waiter_state,BREAK_WAITER_GOT_BREAK);
+	    erts_atomic32_set(&break_waiter_state,BREAK_WAITER_GOT_BREAK);
 	    SetEvent(break_happened_event);
 	    erts_mtx_unlock(&break_waiter_lock);
 	    break;
 	case (WAIT_OBJECT_0+1):
 	    ResetEvent(harr[1]);
 	    erts_mtx_lock(&break_waiter_lock);
-	    erts_atomic_set(&break_waiter_state,BREAK_WAITER_GOT_HALT);
+	    erts_atomic32_set(&break_waiter_state,BREAK_WAITER_GOT_HALT);
 	    SetEvent(break_happened_event);
 	    erts_mtx_unlock(&break_waiter_lock);
 	    break;
@@ -767,12 +802,7 @@ event_happened:
 	    consistency_check(w);
 #endif
 	    ASSERT(WAIT_OBJECT_0 < i && i < WAIT_OBJECT_0+w->active_events);
-	    if (!erts_atomic_xchg(&ps->sys_io_ready,1)) { 
-		HARDDEBUGF(("SET EventIoReady (%d)",erts_atomic_read(&ps->sys_io_ready)));
-		SetEvent(ps->event_io_ready);
-	    } else {
-		HARDDEBUGF(("DONT SET EventIoReady"));
-	    }
+	    notify_io_ready(ps);
 
 	    /* 
 	     * The main thread wont start working on our arrays untill we're
@@ -967,15 +997,10 @@ static int cancel_driver_select(ErtsPollSet ps, HANDLE event)
 void  erts_poll_interrupt(ErtsPollSet ps, int set /* bool */)
 {
     HARDTRACEF(("In erts_poll_interrupt(%d)",set));
-#ifdef ERTS_SMP
-    if (set) {
-	ERTS_POLLSET_SET_INTERRUPTED(ps);
-	wake_poller(ps);
-    }
-    else {
-	ERTS_POLLSET_UNSET_INTERRUPTED(ps);
-    }
-#endif
+    if (!set)
+	reset_interrupt(ps);
+    else
+	set_interrupt(ps);
     HARDTRACEF(("Out erts_poll_interrupt(%d)",set));
 }
 
@@ -984,17 +1009,10 @@ void erts_poll_interrupt_timed(ErtsPollSet ps,
 			       long msec)
 {
     HARDTRACEF(("In erts_poll_interrupt_timed(%d,%ld)",set,msec));
-#ifdef ERTS_SMP
-    if (set) {
-	if (erts_smp_atomic_read(&ps->timeout) > (erts_aint_t) msec) {
-	    ERTS_POLLSET_SET_INTERRUPTED(ps);
-	    wake_poller(ps);
-	}
-    }
-    else {
-	ERTS_POLLSET_UNSET_INTERRUPTED(ps);
-    }
-#endif
+    if (!set)
+	reset_interrupt(ps);
+    else if (erts_smp_atomic32_read_acqb(&ps->timeout) > (erts_aint32_t) msec)
+	set_interrupt(ps);
     HARDTRACEF(("Out erts_poll_interrupt_timed"));
 }
 
@@ -1068,10 +1086,8 @@ void erts_poll_controlv(ErtsPollSet ps,
 int erts_poll_wait(ErtsPollSet ps,
 		   ErtsPollResFd pr[],
 		   int *len,
-		   SysTimeval *utvp)
+		   SysTimeval *tvp)
 {
-    SysTimeval *tvp = utvp;
-    SysTimeval itv;
     int no_fds;
     DWORD timeout;
     EventData* ev;
@@ -1084,7 +1100,7 @@ int erts_poll_wait(ErtsPollSet ps,
     HARDTRACEF(("In erts_poll_wait"));
     ERTS_POLLSET_LOCK(ps);
 
-    if (!erts_atomic_read(&ps->sys_io_ready) && ps->restore_events) {
+    if (!is_io_ready(ps) && ps->restore_events) {
 	HARDDEBUGF(("Restore events: %d",ps->num_waiters));
 	ps->restore_events = 0;
 	for (i = 0; i < ps->num_waiters; ++i) {
@@ -1102,7 +1118,7 @@ int erts_poll_wait(ErtsPollSet ps,
 	       if (w->highwater != w->active_events) {
 		   HARDDEBUGF(("Oups!"));
 		   /* Oups, got signalled before we took the lock, can't reset */
-		   if(erts_atomic_read(&ps->sys_io_ready) == 0) {
+		   if(!is_io_ready(ps)) {
 		       erl_exit(1,"Internal error: "
 				"Inconsistent io structures in erl_poll.\n");
 		   }
@@ -1127,39 +1143,27 @@ int erts_poll_wait(ErtsPollSet ps,
 	no_fds = ERTS_POLL_MAX_RES;
 #endif
 
+    timeout = poll_wait_timeout(ps, tvp);
 
-    ResetEvent(ps->event_io_ready);
-    ERTS_POLLSET_UNSET_POLLER_WOKEN(ps);
-
-#ifdef ERTS_SMP
-    if (ERTS_POLLSET_IS_INTERRUPTED(ps)) {
-	/* Interrupt use zero timeout */
-	itv.tv_sec = 0;
-	itv.tv_usec = 0;
-	tvp = &itv;
-    }
-#endif
-
-    timeout = tvp->tv_sec * 1000 + tvp->tv_usec / 1000;
     /*HARDDEBUGF(("timeout = %ld",(long) timeout));*/
-    erts_smp_atomic_set(&ps->timeout, timeout);
 
-    if (timeout > 0 && ! erts_atomic_read(&ps->sys_io_ready) && ! erts_atomic_read(&break_waiter_state)) {
+    if (timeout > 0 && !erts_atomic32_read(&break_waiter_state)) {
 	HANDLE harr[2] = {ps->event_io_ready, break_happened_event};
 	int num_h = 2;
 
-	HARDDEBUGF(("Start waiting %d [%d]",num_h, (long) timeout));
+	HARDDEBUGF(("Start waiting %d [%d]",num_h, (int) timeout));
 	ERTS_POLLSET_UNLOCK(ps);
 	WaitForMultipleObjects(num_h, harr, FALSE, timeout);
 	ERTS_POLLSET_LOCK(ps);
-	HARDDEBUGF(("Stop waiting %d [%d]",num_h, (long) timeout));
+	HARDDEBUGF(("Stop waiting %d [%d]",num_h, (int) timeout));
+	woke_up(ps);
     }
 
     ERTS_UNSET_BREAK_REQUESTED;
-    if(erts_atomic_read(&break_waiter_state)) {
+    if(erts_atomic32_read(&break_waiter_state)) {
 	erts_mtx_lock(&break_waiter_lock);
-	break_state = erts_atomic_read(&break_waiter_state);
-	erts_atomic_set(&break_waiter_state,0);
+	break_state = erts_atomic32_read(&break_waiter_state);
+	erts_atomic32_set(&break_waiter_state,0);
 	ResetEvent(break_happened_event);
 	erts_mtx_unlock(&break_waiter_lock);
 	switch (break_state) {
@@ -1174,15 +1178,13 @@ int erts_poll_wait(ErtsPollSet ps,
 	}
     }
 
-    ERTS_POLLSET_SET_POLLER_WOKEN(ps);
-
-    if (!erts_atomic_read(&ps->sys_io_ready)) {
-	res = EINTR;
-	HARDDEBUGF(("EINTR!"));
-	goto done; 
+    res = wakeup_cause(ps);
+    if (res != 0) {
+	HARDDEBUGF(("%s!", res == EINTR ? "EINTR" : "ETIMEDOUT"));
+	goto done;
     }
 
-    erts_atomic_set(&ps->sys_io_ready,0);
+    reset_io_ready(ps);
 
     n = ps->num_waiters;	
 
@@ -1204,9 +1206,9 @@ int erts_poll_wait(ErtsPollSet ps,
 	    if (num >= no_fds) {
 		w->highwater=j+1;
 		erts_mtx_unlock(&w->mtx);
-		/* This might mean we still have data to report, set
-		   back the global flag! */
-		erts_atomic_set(&ps->sys_io_ready,1);
+		/* This might mean we still have data to report,
+		   restore flag indicating I/O ready! */
+		restore_io_ready(ps);
 		HARDDEBUGF(("To many FD's to report!"));
 		goto done;
 	    }
@@ -1228,7 +1230,7 @@ int erts_poll_wait(ErtsPollSet ps,
 	erts_mtx_unlock(&w->mtx);
     }
  done:
-    erts_smp_atomic_set(&ps->timeout, ERTS_AINT_T_MAX);
+    erts_smp_atomic32_set(&ps->timeout, ERTS_AINT32_T_MAX);
     *len = num;
     ERTS_POLLSET_UNLOCK(ps);
     HARDTRACEF(("Out erts_poll_wait"));
@@ -1306,15 +1308,13 @@ ErtsPollSet erts_poll_create_pollset(void)
     ps->standby_wait_counter = 0;
     ps->event_io_ready = CreateManualEvent(FALSE);
     ps->standby_wait_event = CreateManualEvent(FALSE); 
-    erts_atomic_init(&ps->sys_io_ready,0);
     ps->restore_events = 0;
 
+    erts_atomic32_init(&ps->wakeup_state, ERTS_POLL_NOT_WOKEN);
 #ifdef ERTS_SMP
-    erts_smp_atomic_init(&ps->woken, 0);
     erts_smp_mtx_init(&ps->mtx, "pollset");
-    erts_smp_atomic_init(&ps->interrupt, 0);
 #endif
-    erts_smp_atomic_init(&ps->timeout, ERTS_AINT_T_MAX);
+    erts_smp_atomic32_init(&ps->timeout, ERTS_AINT32_T_MAX);
 
     HARDTRACEF(("Out erts_poll_create_pollset"));
     return ps;
@@ -1366,7 +1366,7 @@ void  erts_poll_init(void)
 
     erts_mtx_init(&break_waiter_lock,"break_waiter_lock");
     break_happened_event = CreateManualEvent(FALSE);
-    erts_atomic_init(&break_waiter_state, 0); 
+    erts_atomic32_init(&break_waiter_state, 0); 
 
     erts_thr_create(&thread, &break_waiter, NULL, NULL);
     ERTS_UNSET_BREAK_REQUESTED;
diff --git a/erts/emulator/sys/win32/sys_interrupt.c b/erts/emulator/sys/win32/sys_interrupt.c
index 262f84babc..943c338794 100644
--- a/erts/emulator/sys/win32/sys_interrupt.c
+++ b/erts/emulator/sys/win32/sys_interrupt.c
@@ -1,7 +1,7 @@
 /*
  * %CopyrightBegin%
  * 
- * Copyright Ericsson AB 1997-2010. All Rights Reserved.
+ * Copyright Ericsson AB 1997-2011. 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
@@ -31,11 +31,11 @@
 #endif
 
 #ifdef ERTS_SMP
-erts_smp_atomic_t erts_break_requested;
+erts_smp_atomic32_t erts_break_requested;
 #define ERTS_SET_BREAK_REQUESTED \
-  erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 1)
+  erts_smp_atomic32_set(&erts_break_requested, (erts_aint32_t) 1)
 #define ERTS_UNSET_BREAK_REQUESTED \
-  erts_smp_atomic_set(&erts_break_requested, (erts_aint_t) 0)
+  erts_smp_atomic32_set(&erts_break_requested, (erts_aint32_t) 0)
 #else
 volatile int erts_break_requested = 0;
 #define ERTS_SET_BREAK_REQUESTED (erts_break_requested = 1)
diff --git a/erts/emulator/utils/loaded b/erts/emulator/utils/loaded
new file mode 100644
index 0000000000..99a66e7fdb
--- /dev/null
+++ b/erts/emulator/utils/loaded
@@ -0,0 +1,44 @@
+%% -*- erlang -*-
+%%
+%% %CopyrightBegin%
+%%
+%% Copyright Ericsson AB 1998-2010. 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%
+%%
+
+%% Run like:
+%%   $ERL_TOP/bin/escript erts/emulator/utils/loaded
+
+-mode(compile).
+
+main(_) ->
+    LibDir = code:lib_dir(),
+    io:format("Library root is ~s\n", [LibDir]),
+    Wc = filename:join(LibDir, "*/ebin/*.beam"),
+    Beams = filelib:wildcard(Wc),
+    BeamFileSize = lists:sum([filelib:file_size(Beam) || Beam <- Beams]),
+    io:format("~w BEAM files containing ~w bytes\n",
+	      [length(Beams),BeamFileSize]),
+    Ms = [list_to_atom(filename:rootname(filename:basename(Beam))) ||
+	     Beam <- Beams],
+    [{module,_} = code:ensure_loaded(M) || M <- Ms],
+    <<"Current code: ",T/binary>> = erlang:system_info(loaded),
+    Digits = grab_digits(T),
+    io:format("~w modules comprising ~s words when loaded\n",
+	      [length(Ms),Digits]).
+
+grab_digits(<<H,T/binary>>) when $0 =< H, H =< $9 ->
+    [H|grab_digits(T)];
+grab_digits(<<$\n,_/binary>>) -> [].