17 files changed, 332 insertions, 146 deletions

diff --git a/erts/emulator/beam/beam_debug.c b/erts/emulator/beam/beam_debug.c
index 8326d348af..3f4a1f9e80 100644
--- a/erts/emulator/beam/beam_debug.c
+++ b/erts/emulator/beam/beam_debug.c
@@ -308,7 +308,7 @@ erts_debug_disassemble_1(BIF_ALIST_1)
 		BIF_RET(am_undef);
 	    }
 	}
-        code_ptr = erts_codemfa_to_code(cmfa);
+        code_ptr = (BeamInstr*)erts_code_to_codeinfo(erts_codemfa_to_code(cmfa));
     } else {
 	goto error;
     }
@@ -853,15 +853,23 @@ dirty_test(Process *c_p, Eterm type, Eterm arg1, Eterm arg2, UWord *I)
 	    goto badarg;
 	esdp = erts_proc_sched_data(c_p);
 	if (!esdp)
-	    ERTS_BIF_PREP_RET(ret, am_error);
-	else if (!ERTS_SCHEDULER_IS_DIRTY(esdp))
+            goto scheduler_type_error;
+      
+        switch (esdp->type) {
+        case ERTS_SCHED_NORMAL:
 	    ERTS_BIF_PREP_RET(ret, am_normal);
-	else if (ERTS_SCHEDULER_IS_DIRTY_CPU(esdp))
+            break;
+        case ERTS_SCHED_DIRTY_CPU:
 	    ERTS_BIF_PREP_RET(ret, am_dirty_cpu);
-	else if (ERTS_SCHEDULER_IS_DIRTY_IO(esdp))
+            break;
+        case ERTS_SCHED_DIRTY_IO:
 	    ERTS_BIF_PREP_RET(ret, am_dirty_io);
-	else
+            break;
+        default:
+        scheduler_type_error:
 	    ERTS_BIF_PREP_RET(ret, am_error);
+            break;
+        }
     }
     else if (am_error == arg1) {
 	switch (arg2) {
@@ -1118,3 +1126,48 @@ ms_wait(Process *c_p, Eterm etimeout, int busy)
 }
 
 #endif /* ERTS_DIRTY_SCHEDULERS */
+
+#ifdef ERTS_SMP
+#  define ERTS_STACK_LIMIT ((char *) ethr_get_stacklimit())
+#else
+#  define ERTS_STACK_LIMIT ((char *) erts_scheduler_stack_limit)
+#endif
+
+/*
+ * The below functions is for testing of the stack
+ * limit functionality. They are intentionally
+ * written body recursive in order to prevent
+ * last call optimization...
+ */
+
+UWord
+erts_check_stack_recursion_downwards(char *start_c)
+{
+    char *limit = ERTS_STACK_LIMIT;
+    char c;
+    UWord res;
+    if (erts_check_below_limit(&c, limit + 1024))
+        return (char *) erts_ptr_id(start_c) - (char *) erts_ptr_id(&c);
+    res = erts_check_stack_recursion_downwards(start_c);
+    erts_ptr_id(&c);
+    return res;
+}
+
+UWord
+erts_check_stack_recursion_upwards(char *start_c)
+{
+    char *limit = ERTS_STACK_LIMIT;
+    char c;
+    UWord res;
+    if (erts_check_above_limit(&c, limit - 1024))
+        return (char *) erts_ptr_id(&c) - (char *) erts_ptr_id(start_c);
+    res = erts_check_stack_recursion_upwards(start_c);
+    erts_ptr_id(&c);
+    return res;
+}
+
+int
+erts_is_above_stack_limit(char *ptr)
+{
+    return (char *) ptr > ERTS_STACK_LIMIT;
+}
diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 9a91fdce08..6010c17c17 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -3044,10 +3044,12 @@ do {						\
      GetArg2(2, Op1, Op2);
      if (is_both_small(Op1, Op2)) {
 	 /*
-	  * We could extract the tag from one argument, but a tag extraction
-	  * could mean a shift.  Therefore, play it safe here.
+          * TAG ^ TAG == 0.
+          *
+          * Therefore, we perform the XOR operation on the tagged values,
+          * and OR in the tag bits.
 	  */
-	 Eterm result = make_small(signed_val(Op1) ^ signed_val(Op2));
+	 Eterm result = (Op1 ^ Op2) | make_small(0);
 	 StoreBifResult(4, result);
      }
      DO_OUTLINED_ARITH_2(bxor, Op1, Op2);
diff --git a/erts/emulator/beam/bif.c b/erts/emulator/beam/bif.c
index 2d37f977c0..214de3652f 100644
--- a/erts/emulator/beam/bif.c
+++ b/erts/emulator/beam/bif.c
@@ -4506,41 +4506,37 @@ BIF_RETTYPE system_flag_2(BIF_ALIST_2)
 			 || BIF_ARG_2 == am_block_normal);
 	    int normal = (BIF_ARG_2 == am_block_normal
 			  || BIF_ARG_2 == am_unblock_normal);
-	    if (erts_no_schedulers == 1)
-		BIF_RET(am_disabled);
-	    else {
-		switch (erts_block_multi_scheduling(BIF_P,
-						    ERTS_PROC_LOCK_MAIN,
-						    block,
-						    normal,
-						    0)) {
-		case ERTS_SCHDLR_SSPND_DONE_MSCHED_BLOCKED:
-		    BIF_RET(am_blocked);
-		case ERTS_SCHDLR_SSPND_DONE_NMSCHED_BLOCKED:
-		    BIF_RET(am_blocked_normal);
-		case ERTS_SCHDLR_SSPND_YIELD_DONE_MSCHED_BLOCKED:
-		    ERTS_BIF_YIELD_RETURN_X(BIF_P, am_blocked,
-					    am_multi_scheduling);
-		case ERTS_SCHDLR_SSPND_YIELD_DONE_NMSCHED_BLOCKED:
-		    ERTS_BIF_YIELD_RETURN_X(BIF_P, am_blocked_normal,
-					    am_multi_scheduling);
-		case ERTS_SCHDLR_SSPND_DONE:
-		    BIF_RET(am_enabled);
-		case ERTS_SCHDLR_SSPND_YIELD_RESTART:
-		    ERTS_VBUMP_ALL_REDS(BIF_P);
-		    BIF_TRAP2(bif_export[BIF_system_flag_2],
-			      BIF_P, BIF_ARG_1, BIF_ARG_2);
-		case ERTS_SCHDLR_SSPND_YIELD_DONE:
-		    ERTS_BIF_YIELD_RETURN_X(BIF_P, am_enabled,
-					    am_multi_scheduling);
-		case ERTS_SCHDLR_SSPND_EINVAL:
-		    goto error;
-		default:
-		    ASSERT(0);
-		    BIF_ERROR(BIF_P, EXC_INTERNAL_ERROR);
-		    break;
-		}
-	    }
+            switch (erts_block_multi_scheduling(BIF_P,
+                                                ERTS_PROC_LOCK_MAIN,
+                                                block,
+                                                normal,
+                                                0)) {
+            case ERTS_SCHDLR_SSPND_DONE_MSCHED_BLOCKED:
+                BIF_RET(am_blocked);
+            case ERTS_SCHDLR_SSPND_DONE_NMSCHED_BLOCKED:
+                BIF_RET(am_blocked_normal);
+            case ERTS_SCHDLR_SSPND_YIELD_DONE_MSCHED_BLOCKED:
+                ERTS_BIF_YIELD_RETURN_X(BIF_P, am_blocked,
+                                        am_multi_scheduling);
+            case ERTS_SCHDLR_SSPND_YIELD_DONE_NMSCHED_BLOCKED:
+                ERTS_BIF_YIELD_RETURN_X(BIF_P, am_blocked_normal,
+                                        am_multi_scheduling);
+            case ERTS_SCHDLR_SSPND_DONE:
+                BIF_RET(am_enabled);
+            case ERTS_SCHDLR_SSPND_YIELD_RESTART:
+                ERTS_VBUMP_ALL_REDS(BIF_P);
+                BIF_TRAP2(bif_export[BIF_system_flag_2],
+                          BIF_P, BIF_ARG_1, BIF_ARG_2);
+            case ERTS_SCHDLR_SSPND_YIELD_DONE:
+                ERTS_BIF_YIELD_RETURN_X(BIF_P, am_enabled,
+                                        am_multi_scheduling);
+            case ERTS_SCHDLR_SSPND_EINVAL:
+                goto error;
+            default:
+                ASSERT(0);
+                BIF_ERROR(BIF_P, EXC_INTERNAL_ERROR);
+                break;
+            }
 #endif
 	}
     } else if (BIF_ARG_1 == am_schedulers_online) {
diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab
index 4140938210..66e5dc2988 100644
--- a/erts/emulator/beam/bif.tab
+++ b/erts/emulator/beam/bif.tab
@@ -393,6 +393,7 @@ bif erl_ddll:demonitor/1
 #
 # Bifs in the re module 
 #
+bif re:version/0
 bif re:compile/1
 bif re:compile/2
 bif re:run/2
diff --git a/erts/emulator/beam/dist.h b/erts/emulator/beam/dist.h
index 8f6be1061a..6ed36a478e 100644
--- a/erts/emulator/beam/dist.h
+++ b/erts/emulator/beam/dist.h
@@ -53,6 +53,7 @@
 			       | DFLAG_EXPORT_PTR_TAG		\
 			       | DFLAG_BIT_BINARIES             \
 			       | DFLAG_MAP_TAG                  \
+                               | DFLAG_UTF8_ATOMS               \
                                | DFLAG_BIG_CREATION)
 
 /* opcodes used in distribution messages */
diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c
index 025f99330a..3a8687dc59 100644
--- a/erts/emulator/beam/erl_bif_info.c
+++ b/erts/emulator/beam/erl_bif_info.c
@@ -3743,6 +3743,21 @@ BIF_RETTYPE erts_debug_get_internal_state_1(BIF_ALIST_1)
 		BIF_RET(erts_sint64_to_big(value, &hp));
 	    }
 	}
+        else if (ERTS_IS_ATOM_STR("stack_check", BIF_ARG_1)) {
+            UWord size;
+            char c;
+            if (erts_is_above_stack_limit(&c))
+                size = erts_check_stack_recursion_downwards(&c);
+            else
+                size = erts_check_stack_recursion_upwards(&c);
+	    if (IS_SSMALL(size))
+		BIF_RET(make_small(size));
+	    else {
+		Uint hsz = BIG_UWORD_HEAP_SIZE(size);
+		Eterm *hp = HAlloc(BIF_P, hsz);
+		BIF_RET(uword_to_big(size, hp));
+	    }
+        }
     }
     else if (is_tuple(BIF_ARG_1)) {
 	Eterm* tp = tuple_val(BIF_ARG_1);
diff --git a/erts/emulator/beam/erl_bif_re.c b/erts/emulator/beam/erl_bif_re.c
index a66b05c6ff..a72697eb80 100644
--- a/erts/emulator/beam/erl_bif_re.c
+++ b/erts/emulator/beam/erl_bif_re.c
@@ -64,12 +64,47 @@ static void erts_erts_pcre_stack_free(void *ptr) {
     erts_free(ERTS_ALC_T_RE_STACK,ptr);
 }
 
+#define ERTS_PCRE_STACK_MARGIN (10*1024)
+
+#ifdef ERTS_SMP
+#  define ERTS_STACK_LIMIT ((char *) ethr_get_stacklimit())
+#else
+#  define ERTS_STACK_LIMIT ((char *) erts_scheduler_stack_limit)
+#endif
+
+static int
+stack_guard_downwards(void)
+{
+    char *limit = ERTS_STACK_LIMIT;
+    char c;
+
+    ASSERT(limit);
+
+    return erts_check_below_limit(&c, limit + ERTS_PCRE_STACK_MARGIN);
+}
+
+static int
+stack_guard_upwards(void)
+{
+    char *limit = ERTS_STACK_LIMIT;
+    char c;
+
+    ASSERT(limit);
+
+    return erts_check_above_limit(&c, limit - ERTS_PCRE_STACK_MARGIN);
+}
+
 void erts_init_bif_re(void)
 {
+    char c;
     erts_pcre_malloc = &erts_erts_pcre_malloc;
     erts_pcre_free = &erts_erts_pcre_free;
     erts_pcre_stack_malloc = &erts_erts_pcre_stack_malloc;
     erts_pcre_stack_free = &erts_erts_pcre_stack_free;
+    if ((char *) erts_ptr_id(&c) > ERTS_STACK_LIMIT)
+        erts_pcre_stack_guard = stack_guard_downwards;
+    else
+        erts_pcre_stack_guard = stack_guard_upwards;
     default_table = NULL; /* ISO8859-1 default, forced into pcre */
     max_loop_limit = CONTEXT_REDS * LOOP_FACTOR;
 
@@ -477,6 +512,17 @@ build_compile_result(Process *p, Eterm error_tag, pcre *result, int errcode, con
  * Compile BIFs
  */
 
+BIF_RETTYPE
+re_version_0(BIF_ALIST_0)
+{
+    Eterm ret;
+    size_t version_size = 0;
+    byte *version = (byte *) erts_pcre_version();
+    version_size = strlen((const char *) version);
+    ret = new_binary(BIF_P, version, version_size);
+    BIF_RET(ret);
+}
+
 static BIF_RETTYPE
 re_compile(Process* p, Eterm arg1, Eterm arg2)
 {
diff --git a/erts/emulator/beam/erl_gc.c b/erts/emulator/beam/erl_gc.c
index 50805d9cd9..d51d4fff45 100644
--- a/erts/emulator/beam/erl_gc.c
+++ b/erts/emulator/beam/erl_gc.c
@@ -479,9 +479,15 @@ delay_garbage_collection(Process *p, ErlHeapFragment *live_hf_end, int need, int
 	p->live_hf_end = live_hf_end;
     }
 
-    if (need == 0)
+    if (need == 0) {
+#ifdef ERTS_DIRTY_SCHEDULERS
+        if (p->flags & (F_DIRTY_MAJOR_GC|F_DIRTY_MINOR_GC)) {
+            ASSERT(!ERTS_SCHEDULER_IS_DIRTY(erts_proc_sched_data(p)));
+            goto force_reschedule;
+        }
+#endif
 	return 1;
-
+    }
     /*
      * Satisfy need in a heap fragment...
      */
@@ -534,6 +540,10 @@ delay_garbage_collection(Process *p, ErlHeapFragment *live_hf_end, int need, int
     p->heap_hfrag = hfrag;
 #endif
 
+#ifdef ERTS_DIRTY_SCHEDULERS
+force_reschedule:
+#endif
+
     /* Make sure that we do a proper GC as soon as possible... */
     p->flags |= F_FORCE_GC;
     reds_left = ERTS_REDS_LEFT(p, fcalls);
diff --git a/erts/emulator/beam/erl_init.c b/erts/emulator/beam/erl_init.c
index 61477af316..541bfec532 100644
--- a/erts/emulator/beam/erl_init.c
+++ b/erts/emulator/beam/erl_init.c
@@ -61,6 +61,9 @@
 
 #define ERTS_DEFAULT_NO_ASYNC_THREADS	10
 
+#define ERTS_DEFAULT_SCHED_STACK_SIZE   256
+#define ERTS_MIN_SCHED_STACK_SIZE       20
+
 /*
  * The variables below (prefixed with etp_) are for erts/etc/unix/etp-commands
  * only. Do not remove even though they aren't used elsewhere in the emulator!
@@ -125,6 +128,8 @@ const Eterm etp_hole_marker = ERTS_HOLE_MARKER;
 const Eterm etp_hole_marker = 0;
 #endif
 
+static int modified_sched_thread_suggested_stack_size = 0;
+
 /*
  * Note about VxWorks: All variables must be initialized by executable code,
  * not by an initializer. Otherwise a new instance of the emulator will
@@ -1133,8 +1138,12 @@ early_init(int *argc, char **argv) /*
 	}
 	if (dirty_cpu_scheds > schdlrs)
 	    dirty_cpu_scheds = schdlrs;
+        if (dirty_cpu_scheds < 1)
+            dirty_cpu_scheds = 1;
 	if (dirty_cpu_scheds_online > schdlrs_onln)
 	    dirty_cpu_scheds_online = schdlrs_onln;
+	if (dirty_cpu_scheds_online < 1)
+	    dirty_cpu_scheds_online = 1;
 #endif
     }
 
@@ -1237,24 +1246,38 @@ early_init(int *argc, char **argv) /*
 }
 
 #ifndef ERTS_SMP
+
+void *erts_scheduler_stack_limit;
+
+
 static void set_main_stack_size(void)
 {
-    if (erts_sched_thread_suggested_stack_size > 0) {
+    char c;
+    UWord stacksize;
 # if HAVE_DECL_GETRLIMIT && HAVE_DECL_SETRLIMIT && HAVE_DECL_RLIMIT_STACK
-	struct rlimit rl;
-	int bytes = erts_sched_thread_suggested_stack_size * sizeof(Uint) * 1024;
-	if (getrlimit(RLIMIT_STACK, &rl) != 0 ||
-	    (rl.rlim_cur = bytes, setrlimit(RLIMIT_STACK, &rl) != 0)) {
-	    erts_fprintf(stderr, "failed to set stack size for scheduler "
-				 "thread to %d bytes\n", bytes);
-	    erts_usage();
-	}	    
+    struct rlimit rl;
+    int bytes;
+    stacksize = erts_sched_thread_suggested_stack_size * sizeof(Uint) * 1024;
+    /* Add some extra pages... neede by some systems... */
+    bytes = (int) stacksize + 3*erts_sys_get_page_size();
+    if (getrlimit(RLIMIT_STACK, &rl) != 0 ||
+        (rl.rlim_cur = bytes, setrlimit(RLIMIT_STACK, &rl) != 0)) {
+        erts_fprintf(stderr, "failed to set stack size for scheduler "
+                     "thread to %d bytes\n", bytes);
+        erts_usage();
+    }
 # else
+    if (modified_sched_thread_suggested_stack_size) {
 	erts_fprintf(stderr, "no OS support for dynamic stack size limit\n");
-	erts_usage();    
-# endif
+	erts_usage();
     }
+    /* Be conservative and hope it is not more than 64 kWords... */
+    stacksize = 64*1024*sizeof(void *);
+# endif
+
+    erts_scheduler_stack_limit = erts_calc_stacklimit(&c, stacksize);
 }
+
 #endif
 
 void
@@ -1299,12 +1322,11 @@ erl_start(int argc, char **argv)
 	port_tab_sz_ignore_files = 1;
     }
 
-#if (defined(__APPLE__) && defined(__MACH__)) || defined(__DARWIN__)
     /*
-     * The default stack size on MacOS X is too small for pcre.
+     * A default stack size suitable for pcre which might use quite
+     * a lot of stack.
      */
-    erts_sched_thread_suggested_stack_size = 256;
-#endif
+    erts_sched_thread_suggested_stack_size = ERTS_DEFAULT_SCHED_STACK_SIZE;
 
 #ifdef DEBUG
     verbose = DEBUG_DEFAULT;
@@ -1927,6 +1949,7 @@ erl_start(int argc, char **argv)
 		/* suggested stack size (Kilo Words) for scheduler threads */
 		arg = get_arg(sub_param+2, argv[i+1], &i);
 		erts_sched_thread_suggested_stack_size = atoi(arg);
+                modified_sched_thread_suggested_stack_size = 1;
 
 		if ((erts_sched_thread_suggested_stack_size
 		     < ERTS_SCHED_THREAD_MIN_STACK_SIZE)
@@ -2236,6 +2259,9 @@ erl_start(int argc, char **argv)
     boot_argc = argc - i;  /* Number of arguments to init */
     boot_argv = &argv[i];
 
+    if (erts_sched_thread_suggested_stack_size < ERTS_MIN_SCHED_STACK_SIZE)
+        erts_sched_thread_suggested_stack_size = ERTS_MIN_SCHED_STACK_SIZE;
+
     erl_init(ncpu,
 	     proc_tab_sz,
 	     legacy_proc_tab,
diff --git a/erts/emulator/beam/erl_nif.c b/erts/emulator/beam/erl_nif.c
index 63a4a997da..f86b9739fa 100644
--- a/erts/emulator/beam/erl_nif.c
+++ b/erts/emulator/beam/erl_nif.c
@@ -2818,14 +2818,19 @@ enif_thread_type(void)
     if (!esdp)
 	return ERL_NIF_THR_UNDEFINED;
 
-    if (!ERTS_SCHEDULER_IS_DIRTY(esdp))
+    switch (esdp->type) {
+    case ERTS_SCHED_NORMAL:
 	return ERL_NIF_THR_NORMAL_SCHEDULER;
-
-    if (ERTS_SCHEDULER_IS_DIRTY_CPU(esdp))
+#ifdef ERTS_DIRTY_SCHEDULERS
+    case ERTS_SCHED_DIRTY_CPU:
 	return ERL_NIF_THR_DIRTY_CPU_SCHEDULER;
-
-    ASSERT(ERTS_SCHEDULER_IS_DIRTY_IO(esdp));
-    return ERL_NIF_THR_DIRTY_IO_SCHEDULER;
+    case ERTS_SCHED_DIRTY_IO:
+        return ERL_NIF_THR_DIRTY_IO_SCHEDULER;
+#endif
+    default:
+        ERTS_INTERNAL_ERROR("Invalid scheduler type");
+	return -1;
+    }
 }
 
 /* Maps */
diff --git a/erts/emulator/beam/erl_process.c b/erts/emulator/beam/erl_process.c
index 88fae30845..894e0ee582 100644
--- a/erts/emulator/beam/erl_process.c
+++ b/erts/emulator/beam/erl_process.c
@@ -280,6 +280,7 @@ schdlr_sspnd_get_nscheds(ErtsSchedTypeCounters *valp,
     }
 }
 
+#ifdef DEBUG
 static ERTS_INLINE Uint32
 schdlr_sspnd_get_nscheds_tot(ErtsSchedTypeCounters *valp)
 {
@@ -290,6 +291,7 @@ schdlr_sspnd_get_nscheds_tot(ErtsSchedTypeCounters *valp)
 #endif
     return res;
 }
+#endif
 
 static ERTS_INLINE void
 schdlr_sspnd_dec_nscheds(ErtsSchedTypeCounters *valp,
@@ -5947,14 +5949,30 @@ erts_sched_set_wake_cleanup_threshold(char *str)
 static void
 init_aux_work_data(ErtsAuxWorkData *awdp, ErtsSchedulerData *esdp, char *dawwp)
 {
-    if (!esdp)
-	awdp->sched_id = 0;
+    int id = 0;
+    if (esdp) {
+        switch (esdp->type) {
+        case ERTS_SCHED_NORMAL:
+            id = (int) esdp->no;
+            break;
 #ifdef ERTS_DIRTY_SCHEDULERS
-    else if (ERTS_SCHEDULER_IS_DIRTY(esdp))
-	awdp->sched_id = (int) ERTS_DIRTY_SCHEDULER_NO(esdp);
+        case ERTS_SCHED_DIRTY_CPU:
+            id = (int) erts_no_schedulers;
+            id += (int) esdp->dirty_no;
+            break;
+        case ERTS_SCHED_DIRTY_IO:
+            id = (int) erts_no_schedulers;
+            id += (int) erts_no_dirty_cpu_schedulers;
+            id += (int) esdp->dirty_no;
+            break;
 #endif
-    else
-	awdp->sched_id = (int) esdp->no;
+        default:
+            ERTS_INTERNAL_ERROR("Invalid scheduler type");
+            break;
+        }
+    }
+
+    awdp->sched_id = id;
     awdp->esdp = esdp;
     awdp->ssi = esdp ? esdp->ssi : NULL;
 #ifdef ERTS_SMP
@@ -6027,7 +6045,7 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num,
             ASSERT(runq == ERTS_DIRTY_IO_RUNQ);
             esdp->type = ERTS_SCHED_DIRTY_IO;
         }
-	ERTS_DIRTY_SCHEDULER_NO(esdp) = (Uint) num;
+        esdp->dirty_no = (Uint) num;
         if (num == 1) {
             /*
              * Multi-scheduling block functionality depends
@@ -6039,7 +6057,7 @@ init_scheduler_data(ErtsSchedulerData* esdp, int num,
     else {
         esdp->type = ERTS_SCHED_NORMAL;
 	esdp->no = (Uint) num;
-	ERTS_DIRTY_SCHEDULER_NO(esdp) = 0;
+	esdp->dirty_no = 0;
         runq->scheduler = esdp;
     }
     esdp->dirty_shadow_process = shadow_proc;
@@ -7776,11 +7794,11 @@ suspend_scheduler(ErtsSchedulerData *esdp)
         break;
     case ERTS_SCHED_DIRTY_CPU:
         online_flag = ERTS_SCHDLR_SSPND_CHNG_DCPU_ONLN;
-        no = ERTS_DIRTY_SCHEDULER_NO(esdp);
+        no = esdp->dirty_no;
         break;
     case ERTS_SCHED_DIRTY_IO:
         online_flag = 0;
-	no = ERTS_DIRTY_SCHEDULER_NO(esdp);
+	no = esdp->dirty_no;
         break;
     default:
         ERTS_INTERNAL_ERROR("Invalid scheduler type");
@@ -8515,8 +8533,9 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int normal
 	    p->flags |= have_blckd_flg;
 	    goto wait_until_msb;
 	}
-	else if (msbp->blckrs) {
-	    ASSERT(msbp->ongoing);
+	else if (msbp->blckrs || (normal && erts_no_schedulers == 1)) {
+	    ASSERT(!msbp->blckrs || msbp->ongoing);
+	    msbp->ongoing = 1;
 	    plp = proclist_create(p);
 	    erts_proclist_store_last(&msbp->blckrs, plp);
 	    p->flags |= have_blckd_flg;
@@ -8530,7 +8549,7 @@ erts_block_multi_scheduling(Process *p, ErtsProcLocks plocks, int on, int normal
 	    else
 		res = ERTS_SCHDLR_SSPND_DONE_NMSCHED_BLOCKED;
 	}
-	else {
+        else {
 	    int online = (int) schdlr_sspnd_get_nscheds(&schdlr_sspnd.online,
 							ERTS_SCHED_NORMAL);
 	    ASSERT(!msbp->ongoing);
@@ -8804,8 +8823,7 @@ sched_dirty_cpu_thread_func(void *vesdp)
 {
     ErtsThrPrgrCallbacks callbacks;
     ErtsSchedulerData *esdp = vesdp;
-    Uint no = ERTS_DIRTY_SCHEDULER_NO(esdp);
-    ERTS_DIRTY_SCHEDULER_TYPE(esdp) = ERTS_DIRTY_CPU_SCHEDULER;
+    Uint no = esdp->dirty_no;
     ASSERT(no != 0);
     ERTS_DIRTY_CPU_SCHED_SLEEP_INFO_IX(no-1)->event = erts_tse_fetch();
     callbacks.arg = (void *) esdp->ssi;
@@ -8853,8 +8871,7 @@ sched_dirty_io_thread_func(void *vesdp)
 {
     ErtsThrPrgrCallbacks callbacks;
     ErtsSchedulerData *esdp = vesdp;
-    Uint no = ERTS_DIRTY_SCHEDULER_NO(esdp);
-    ERTS_DIRTY_SCHEDULER_TYPE(esdp) = ERTS_DIRTY_IO_SCHEDULER;
+    Uint no = esdp->dirty_no;
     ASSERT(no != 0);
     ERTS_DIRTY_IO_SCHED_SLEEP_INFO_IX(no-1)->event = erts_tse_fetch();
     callbacks.arg = (void *) esdp->ssi;
@@ -11359,7 +11376,7 @@ erts_execute_dirty_system_task(Process *c_p)
 
 	switch (st->type) {
 	case ERTS_PSTT_CLA:
-	    ASSERT(is_value(st_res));
+	    ASSERT(is_value(cla_res));
 	    st_res = cla_res;
 	    break;
 	case ERTS_PSTT_GC_MAJOR:
diff --git a/erts/emulator/beam/erl_process.h b/erts/emulator/beam/erl_process.h
index baf830615d..883d9f2a4c 100644
--- a/erts/emulator/beam/erl_process.h
+++ b/erts/emulator/beam/erl_process.h
@@ -641,13 +641,6 @@ typedef enum {
     ERTS_DIRTY_IO_SCHEDULER
 } ErtsDirtySchedulerType;
 
-typedef union {
-    struct {
-	ErtsDirtySchedulerType type: 1;
-	Uint num: sizeof(Uint)*8 - 1;
-    } s;
-    Uint no;
-} ErtsDirtySchedId;
 #endif
 
 struct ErtsSchedulerData_ {
@@ -674,7 +667,7 @@ struct ErtsSchedulerData_ {
     ErtsSchedType type;
     Uint no;			/* Scheduler number for normal schedulers */
 #ifdef ERTS_DIRTY_SCHEDULERS
-    ErtsDirtySchedId dirty_no;  /* Scheduler number for dirty schedulers */
+    Uint dirty_no;  /* Scheduler number for dirty schedulers */
     Process *dirty_shadow_process;
 #endif
     Port *current_port;
@@ -1578,23 +1571,13 @@ extern int erts_system_profile_ts_type;
 #define ERTS_DIRTY_IO_SCHEDULER_IX(IX)					\
   (ASSERT(0 <= (IX) && (IX) < erts_no_dirty_io_schedulers),		\
    &erts_aligned_dirty_io_scheduler_data[(IX)].esd)
-#define ERTS_DIRTY_SCHEDULER_NO(ESDP)					\
-  ((ESDP)->dirty_no.s.num)
-#define ERTS_DIRTY_SCHEDULER_TYPE(ESDP)					\
-  ((ESDP)->dirty_no.s.type)
-#ifdef ERTS_SMP
 #define ERTS_SCHEDULER_IS_DIRTY(ESDP)					\
-  ((ESDP)->dirty_no.s.num != 0)
+  ((ESDP)->type != ERTS_SCHED_NORMAL)
 #define ERTS_SCHEDULER_IS_DIRTY_CPU(ESDP)				\
-    (ERTS_SCHEDULER_IS_DIRTY((ESDP)) & ((ESDP)->dirty_no.s.type == 0))
+    ((ESDP)->type == ERTS_SCHED_DIRTY_CPU)
 #define ERTS_SCHEDULER_IS_DIRTY_IO(ESDP)				\
-    (ERTS_SCHEDULER_IS_DIRTY((ESDP)) & ((ESDP)->dirty_no.s.type == 1))
-#else
-#define ERTS_SCHEDULER_IS_DIRTY(ESDP) 0
-#define ERTS_SCHEDULER_IS_DIRTY_CPU(ESDP) 0
-#define ERTS_SCHEDULER_IS_DIRTY_IO(ESDP) 0
-#endif
-#else
+    ((ESDP)->type == ERTS_SCHED_DIRTY_IO)
+#else /* !ERTS_DIRTY_SCHEDULERS */
 #define ERTS_RUNQ_IX_IS_DIRTY(IX) 0
 #define ERTS_SCHEDULER_IS_DIRTY(ESDP) 0
 #define ERTS_SCHEDULER_IS_DIRTY_CPU(ESDP) 0
diff --git a/erts/emulator/beam/erl_process_dict.c b/erts/emulator/beam/erl_process_dict.c
index 8311fde025..7cfdf20341 100644
--- a/erts/emulator/beam/erl_process_dict.c
+++ b/erts/emulator/beam/erl_process_dict.c
@@ -54,9 +54,9 @@
 #define HASH_RANGE(PDict) ((PDict)->usedSlots)
 
 #define MAKE_HASH(Term)                                \
-    ((is_small(Term)) ? unsigned_val(Term) :           \
+    ((is_small(Term)) ? (Uint32) unsigned_val(Term) :  \
      ((is_atom(Term)) ?                                \
-      atom_val(Term) :								   \
+      (Uint32) atom_val(Term) :                        \
       make_internal_hash(Term)))
 
 #define PD_SZ2BYTES(Sz) (sizeof(ProcDict) + ((Sz) - 1)*sizeof(Eterm))
diff --git a/erts/emulator/beam/external.c b/erts/emulator/beam/external.c
index 205a7711ec..285ae4ac78 100644
--- a/erts/emulator/beam/external.c
+++ b/erts/emulator/beam/external.c
@@ -2093,7 +2093,6 @@ enc_atom(ErtsAtomCacheMap *acmp, Eterm atom, byte *ep, Uint32 dflags)
 {
     int iix;
     int len;
-    int utf8_atoms = (int) (dflags & DFLAG_UTF8_ATOMS);
 
     ASSERT(is_atom(atom));
 
@@ -2122,8 +2121,8 @@ enc_atom(ErtsAtomCacheMap *acmp, Eterm atom, byte *ep, Uint32 dflags)
     if (iix < 0) {
 	Atom *a = atom_tab(atom_val(atom));
 	len = a->len;
-	if (utf8_atoms || a->latin1_chars < 0) {
-	    if (len > 255) {
+        {
+            if (len > 255) {
 		*ep++ = ATOM_UTF8_EXT;
 		put_int16(len, ep);
 		ep += 2;
@@ -2135,32 +2134,6 @@ enc_atom(ErtsAtomCacheMap *acmp, Eterm atom, byte *ep, Uint32 dflags)
 	    }
 	    sys_memcpy((char *) ep, (char *) a->name, len);
 	}
-	else {
-	    if (a->latin1_chars <= 255 && (dflags & DFLAG_SMALL_ATOM_TAGS)) {
-		*ep++ = SMALL_ATOM_EXT;
-		if (len == a->latin1_chars) {
-		    sys_memcpy(ep+1, a->name, len);
-		}
-		else {
-		    len = erts_utf8_to_latin1(ep+1, a->name, len);
-		    ASSERT(len == a->latin1_chars);
-		}
-		put_int8(len, ep);
-		ep++;
-	    }
-	    else {
-		*ep++ = ATOM_EXT;
-		if (len == a->latin1_chars) {
-		    sys_memcpy(ep+2, a->name, len);
-		}
-		else {
-		    len = erts_utf8_to_latin1(ep+2, a->name, len);
-		    ASSERT(len == a->latin1_chars);
-		}
-		put_int16(len, ep);
-		ep += 2;
-	    }	    
-	}
 	ep += len;
 	return ep;
     }
@@ -4010,7 +3983,7 @@ error:
 	    factory->hp = hp;   /* the largest must be the freshest */
 	}
     }
-    else ASSERT(factory->hp == hp);
+    else ASSERT(!factory->hp || factory->hp == hp);
 
 error_hamt:
     erts_factory_undo(factory);
@@ -4085,19 +4058,13 @@ encode_size_struct_int(TTBSizeContext* ctx, ErtsAtomCacheMap *acmp, Eterm obj,
 	    else {
 		Atom *a = atom_tab(atom_val(obj));
 		int alen;
-		if ((dflags & DFLAG_UTF8_ATOMS) || a->latin1_chars < 0) {
+                {
 		    alen = a->len;
 		    result += 1 + 1 + alen;
 		    if (alen > 255) {
 			result++; /* ATOM_UTF8_EXT (not small) */
 		    }
 		}
-		else {
-		    alen = a->latin1_chars;
-		    result += 1 + 1 + alen;
-		    if (alen > 255 || !(dflags & DFLAG_SMALL_ATOM_TAGS))
-			result++; /* ATOM_EXT (not small) */
-		}
 		insert_acache_map(acmp, obj, dflags);
 	    }
 	    break;
diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h
index c4c848f49f..139394680a 100644
--- a/erts/emulator/beam/global.h
+++ b/erts/emulator/beam/global.h
@@ -1136,6 +1136,11 @@ void erts_short_init(void);
 void erl_start(int, char**);
 void erts_usage(void);
 Eterm erts_preloaded(Process* p);
+
+#ifndef ERTS_SMP
+extern void *erts_scheduler_stack_limit;
+#endif
+
 /* erl_md5.c */
 
 typedef struct {
@@ -1198,6 +1203,11 @@ Uint64 erts_timestamp_millis(void);
 
 Export* erts_find_function(Eterm, Eterm, unsigned int, ErtsCodeIndex);
 
+void *erts_calc_stacklimit(char *prev_c, UWord stacksize);
+int erts_check_below_limit(char *ptr, char *limit);
+int erts_check_above_limit(char *ptr, char *limit);
+void *erts_ptr_id(void *ptr);
+
 Eterm store_external_or_ref_in_proc_(Process *, Eterm);
 Eterm store_external_or_ref_(Uint **, ErlOffHeap*, Eterm);
 
@@ -1233,6 +1243,11 @@ void erts_init_external(void);
 /* erl_map.c */
 void erts_init_map(void);
 
+/* beam_debug.c */
+UWord erts_check_stack_recursion_downwards(char *start_c);
+UWord erts_check_stack_recursion_upwards(char *start_c);
+int erts_is_above_stack_limit(char *ptr);
+
 /* erl_unicode.c */
 void erts_init_unicode(void);
 Sint erts_unicode_set_loop_limit(Sint limit);
diff --git a/erts/emulator/beam/sys.h b/erts/emulator/beam/sys.h
index 144dd60d21..24de35696c 100644
--- a/erts/emulator/beam/sys.h
+++ b/erts/emulator/beam/sys.h
@@ -602,6 +602,8 @@ __decl_noreturn void __noreturn erts_exit(int n, char*, ...);
 
 Eterm erts_check_io_info(void *p);
 
+UWord erts_sys_get_page_size(void);
+
 /* Size of misc memory allocated from system dependent code */
 Uint erts_sys_misc_mem_sz(void);
 
diff --git a/erts/emulator/beam/utils.c b/erts/emulator/beam/utils.c
index 092a5320ba..8f3f48f38f 100644
--- a/erts/emulator/beam/utils.c
+++ b/erts/emulator/beam/utils.c
@@ -4835,6 +4835,53 @@ Uint64 erts_timestamp_millis(void)
 #endif
 }
 
+void *
+erts_calc_stacklimit(char *prev_c, UWord stacksize)
+{
+    /*
+     * We *don't* want this function inlined, i.e., it is
+     * risky to call this function from another function
+     * in utils.c
+     */
+
+    UWord pagesize = erts_sys_get_page_size();
+    char c;
+    char *start;
+    if (&c > prev_c) {
+        start = (char *) ((((UWord) prev_c) / pagesize) * pagesize);
+        return (void *) (start + stacksize);
+    }
+    else {
+        start = (char *) (((((UWord) prev_c) - 1) / pagesize + 1) * pagesize);
+        return (void *) (start - stacksize);
+    }
+}
+
+/*
+ * erts_check_below_limit() and
+ * erts_check_above_limit() are put
+ * in utils.c in order to prevent
+ * inlining.
+ */
+
+int
+erts_check_below_limit(char *ptr, char *limit)
+{
+    return ptr < limit;
+}
+
+int
+erts_check_above_limit(char *ptr, char *limit)
+{
+    return ptr > limit;
+}
+
+void *
+erts_ptr_id(void *ptr)
+{
+    return ptr;
+}
+
 #ifdef DEBUG
 /*
  * Handy functions when using a debugger - don't use in the code!