erts: Add 'exec_alloc' for hipe code

that uses its own super carrier (erts_exec_mmapper)
to guarantee low addressed and executable memory (PROT_EXEC).

Currently only used on x86_64 that needs low memory
for HiPE/AMD64's small code model.

By initializing erts_exec_mapper early we secure
its low memory area before erts_literal_mmapper might
steal it.

11 files changed, 188 insertions, 117 deletions

diff --git a/erts/configure.in b/erts/configure.in
index 0257079c3b..462a13269e 100644
--- a/erts/configure.in
+++ b/erts/configure.in
@@ -3351,6 +3351,13 @@ if test X${enable_hipe} = Xyes; then
      AC_DEFINE(HIPE,[1],[Define to enable HiPE])
      HIPE_HELPERS="xmerl syntax_tools edoc"
      ENABLE_ALLOC_TYPE_VARS="$ENABLE_ALLOC_TYPE_VARS hipe"
+     case "$ARCH" in
+	  amd64)
+		# For now exec_alloc is only used for hipe on amd64
+		AC_MSG_NOTICE([Enable exec_alloc for hipe code allocation])
+		ENABLE_ALLOC_TYPE_VARS="$ENABLE_ALLOC_TYPE_VARS exec_alloc"
+		;;
+     esac
   fi
 fi
 AC_SUBST(HIPE_HELPERS)
diff --git a/erts/doc/src/erts_alloc.xml b/erts/doc/src/erts_alloc.xml
index 0965f9b49c..c54082f8c8 100644
--- a/erts/doc/src/erts_alloc.xml
+++ b/erts/doc/src/erts_alloc.xml
@@ -63,6 +63,9 @@
       <tag><c>fix_alloc</c></tag>
       <item>A fast allocator used for some frequently used
        fixed size data types.</item>
+      <tag><c>exec_alloc</c></tag>
+      <item>Allocator used by hipe for native executable code
+      on specific architectures (x86_64).</item>
       <tag><c>std_alloc</c></tag>
       <item>Allocator used for most memory blocks not allocated via any of
        the other allocators described above.</item>
@@ -80,7 +83,8 @@
        the number of system calls made.</item>
     </taglist>
     <p><c>sys_alloc</c> and <c>literal_alloc</c> are always enabled and
-      cannot be disabled. <c>mseg_alloc</c> is always enabled if it is
+      cannot be disabled. <c>exec_alloc</c> is only available if it is needed
+      and cannot be disabled. <c>mseg_alloc</c> is always enabled if it is
       available and an allocator that uses it is enabled. All other
       allocators can be <seealso marker="#M_e">enabled or disabled</seealso>.
       By default all allocators are enabled.
@@ -248,16 +252,17 @@
       the currently present allocators:</p>
     <list type="bulleted">
       <item><c>B: binary_alloc</c></item>
-      <item><c>I: literal_alloc</c></item>
       <item><c>D: std_alloc</c></item>
       <item><c>E: ets_alloc</c></item>
       <item><c>F: fix_alloc</c></item>
       <item><c>H: eheap_alloc</c></item>
+      <item><c>I: literal_alloc</c></item>
       <item><c>L: ll_alloc</c></item>
       <item><c>M: mseg_alloc</c></item>
       <item><c>R: driver_alloc</c></item>
       <item><c>S: sl_alloc</c></item>
       <item><c>T: temp_alloc</c></item>
+      <item><c>X: exec_alloc</c></item>
       <item><c>Y: sys_alloc</c></item>
     </list>
     <p>The following flags are available for configuration of
@@ -576,6 +581,15 @@
       and is usually sufficient. The flag is ignored on 32-bit
       architectures.</item>
     </taglist>
+    <p>The following flag is special for <c>exec_alloc</c>:</p>
+    <taglist>
+      <tag><marker id="MIscs"/><c><![CDATA[+MXscs <size in MB>]]></c></tag>
+      <item>
+      <c>exec_alloc</c> super carrier size (in MB). The amount of
+      <em>virtual</em> address space reserved for native executable code
+      used by hipe on specific architectures (x86_64). The default is 512 MB.
+      </item>
+    </taglist>
     <p>Instrumentation flags:</p>
     <taglist>
       <tag><marker id="Mim"/><c>+Mim true|false</c></tag>
diff --git a/erts/emulator/beam/erl_alloc.c b/erts/emulator/beam/erl_alloc.c
index efe31193b9..00911ff027 100644
--- a/erts/emulator/beam/erl_alloc.c
+++ b/erts/emulator/beam/erl_alloc.c
@@ -30,6 +30,7 @@
 #endif
 #define ERTS_ALLOC_C__
 #define ERTS_ALC_INTERNAL__
+#define ERTS_WANT_MEM_MAPPERS
 #include "sys.h"
 #define ERL_THREADS_EMU_INTERNAL__
 #include "erl_threads.h"
@@ -131,6 +132,9 @@ static ErtsAllocatorState_t ets_alloc_state;
 static ErtsAllocatorState_t driver_alloc_state;
 static ErtsAllocatorState_t fix_alloc_state;
 static ErtsAllocatorState_t literal_alloc_state;
+#ifdef ERTS_ALC_A_EXEC
+static ErtsAllocatorState_t exec_alloc_state;
+#endif
 static ErtsAllocatorState_t test_alloc_state;
 
 typedef struct {
@@ -216,6 +220,7 @@ typedef struct {
     struct au_init driver_alloc;
     struct au_init fix_alloc;
     struct au_init literal_alloc;
+    struct au_init exec_alloc;
     struct au_init test_alloc;
 } erts_alc_hndl_args_init_t;
 
@@ -339,6 +344,38 @@ set_default_literal_alloc_opts(struct au_init *ip)
 #endif
 }
 
+#ifdef ERTS_ALC_A_EXEC
+static void
+set_default_exec_alloc_opts(struct au_init *ip)
+{
+    SET_DEFAULT_ALLOC_OPTS(ip);
+    ip->enable			= 1;
+    ip->thr_spec		= 0;
+    ip->disable_allowed         = 0;
+    ip->thr_spec_allowed        = 0;
+    ip->carrier_migration_allowed = 0;
+    ip->atype			= BESTFIT;
+    ip->init.bf.ao		= 1;
+    ip->init.util.ramv		= 0;
+    ip->init.util.mmsbc		= 0;
+    ip->init.util.sbct		= ~((UWord) 0);
+    ip->init.util.name_prefix	= "exec_";
+    ip->init.util.alloc_no	= ERTS_ALC_A_EXEC;
+    ip->init.util.mmbcs 	= 0; /* No main carrier */
+    ip->init.util.ts 		= ERTS_ALC_MTA_EXEC;
+    ip->init.util.asbcst	= 0;
+    ip->init.util.rsbcst	= 0;
+    ip->init.util.rsbcmt	= 0;
+    ip->init.util.rmbcmt	= 0;
+    ip->init.util.acul		= 0;
+
+    ip->init.util.mseg_alloc    = &erts_alcu_mmapper_mseg_alloc;
+    ip->init.util.mseg_realloc  = &erts_alcu_mmapper_mseg_realloc;
+    ip->init.util.mseg_dealloc  = &erts_alcu_mmapper_mseg_dealloc;
+    ip->init.util.mseg_mmapper  = &erts_exec_mmapper;
+}
+#endif /* ERTS_ALC_A_EXEC */
+
 static void
 set_default_temp_alloc_opts(struct au_init *ip)
 {
@@ -660,6 +697,9 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
     set_default_fix_alloc_opts(&init.fix_alloc,
 			       fix_type_sizes);
     set_default_literal_alloc_opts(&init.literal_alloc);
+#ifdef ERTS_ALC_A_EXEC
+    set_default_exec_alloc_opts(&init.exec_alloc);
+#endif
     set_default_test_alloc_opts(&init.test_alloc);
 
     if (argc && argv)
@@ -689,6 +729,9 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
     init.driver_alloc.thr_spec = 0;
     init.fix_alloc.thr_spec = 0;
     init.literal_alloc.thr_spec = 0;
+#ifdef ERTS_ALC_A_EXEC
+    init.exec_alloc.thr_spec = 0;
+#endif
 #endif
 
     /* Make adjustments for carrier migration support */
@@ -702,6 +745,9 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
     adjust_carrier_migration_support(&init.driver_alloc);
     adjust_carrier_migration_support(&init.fix_alloc);
     adjust_carrier_migration_support(&init.literal_alloc);
+#ifdef ERTS_ALC_A_EXEC
+    adjust_carrier_migration_support(&init.exec_alloc);
+#endif
 
     if (init.erts_alloc_config) {
 	/* Adjust flags that erts_alloc_config won't like */
@@ -717,6 +763,9 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
 	init.driver_alloc.thr_spec = 0;
 	init.fix_alloc.thr_spec = 0;
         init.literal_alloc.thr_spec = 0;
+#ifdef ERTS_ALC_A_EXEC
+        init.exec_alloc.thr_spec = 0;
+#endif
 
 	/* No carrier migration */
 	init.temp_alloc.init.util.acul = 0;
@@ -729,6 +778,9 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
 	init.driver_alloc.init.util.acul = 0;
 	init.fix_alloc.init.util.acul = 0;
         init.literal_alloc.init.util.acul = 0;
+#ifdef ERTS_ALC_A_EXEC
+        init.exec_alloc.init.util.acul = 0;
+#endif
     }
 
 #ifdef ERTS_SMP
@@ -746,6 +798,9 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
     adjust_tpref(&init.driver_alloc, erts_no_schedulers);
     adjust_tpref(&init.fix_alloc, erts_no_schedulers);
     adjust_tpref(&init.literal_alloc, erts_no_schedulers);
+#ifdef ERTS_ALC_A_EXEC
+    adjust_tpref(&init.exec_alloc, erts_no_schedulers);
+#endif
 
 #else
     /* No thread specific if not smp */
@@ -765,6 +820,9 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
     refuse_af_strategy(&init.driver_alloc);
     refuse_af_strategy(&init.fix_alloc);
     refuse_af_strategy(&init.literal_alloc);
+#ifdef ERTS_ALC_A_EXEC
+    refuse_af_strategy(&init.exec_alloc);
+#endif
 
 #ifdef ERTS_SMP 
     if (!init.temp_alloc.thr_spec)
@@ -809,6 +867,9 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
     set_au_allocator(ERTS_ALC_A_DRIVER, &init.driver_alloc, ncpu);
     set_au_allocator(ERTS_ALC_A_FIXED_SIZE, &init.fix_alloc, ncpu);
     set_au_allocator(ERTS_ALC_A_LITERAL, &init.literal_alloc, ncpu);
+#ifdef ERTS_ALC_A_EXEC
+    set_au_allocator(ERTS_ALC_A_EXEC, &init.exec_alloc, ncpu);
+#endif
     set_au_allocator(ERTS_ALC_A_TEST, &init.test_alloc, ncpu);
 
     for (i = ERTS_ALC_A_MIN; i <= ERTS_ALC_A_MAX; i++) {
@@ -865,7 +926,11 @@ erts_alloc_init(int *argc, char **argv, ErtsAllocInitOpts *eaiop)
     start_au_allocator(ERTS_ALC_A_LITERAL,
                        &init.literal_alloc,
                        &literal_alloc_state);
-
+#ifdef ERTS_ALC_A_EXEC
+    start_au_allocator(ERTS_ALC_A_EXEC,
+                       &init.exec_alloc,
+                       &exec_alloc_state);
+#endif
     start_au_allocator(ERTS_ALC_A_TEST,
 		       &init.test_alloc,
 		       &test_alloc_state);
@@ -1500,6 +1565,16 @@ handle_args(int *argc, char **argv, erts_alc_hndl_args_init_t *init)
                     else
                         handle_au_arg(&init->literal_alloc, &argv[i][3], argv, &i, 0);
 		    break;
+                case 'X':
+                    if (has_prefix("scs", argv[i]+3)) {
+#ifdef ERTS_ALC_A_EXEC
+                        init->mseg.exec_mmap.scs =
+#endif
+                            get_mb_value(argv[i]+6, argv, &i);
+                    }
+                    else
+                        handle_au_arg(&init->exec_alloc, &argv[i][3], argv, &i, 0);
+                    break;
 		case 'D':
 		    handle_au_arg(&init->std_alloc, &argv[i][3], argv, &i, 0);
 		    break;
diff --git a/erts/emulator/beam/erl_alloc.types b/erts/emulator/beam/erl_alloc.types
index 14067283bd..0649ed903a 100644
--- a/erts/emulator/beam/erl_alloc.types
+++ b/erts/emulator/beam/erl_alloc.types
@@ -87,6 +87,9 @@ allocator	EHEAP		true		eheap_alloc
 allocator	ETS		true		ets_alloc
 allocator	FIXED_SIZE	true		fix_alloc
 allocator	LITERAL 	true		literal_alloc
++if exec_alloc
+allocator	EXEC     	true	        exec_alloc
++endif
 
 +else # Non smp build
 
@@ -98,6 +101,9 @@ allocator	EHEAP		false		eheap_alloc
 allocator	ETS		false		ets_alloc
 allocator	FIXED_SIZE	false		fix_alloc
 allocator	LITERAL 	false		literal_alloc
++if exec_alloc
+allocator	EXEC     	false	        exec_alloc
++endif
 
 +endif
 
@@ -335,8 +341,14 @@ type	SL_MPATHS	SHORT_LIVED	SYSTEM		sl_migration_paths
 # Currently most hipe code use this type.
 type	HIPE		SYSTEM		SYSTEM		hipe_data
 
++if exec_alloc
+type	HIPE_EXEC	EXEC            CODE		hipe_code
++endif
+
 +endif
 
+
+
 +if heap_frag_elim_test
 
 type	SSB		SHORT_LIVED	PROCESSES	ssb
diff --git a/erts/emulator/beam/erl_bif_info.c b/erts/emulator/beam/erl_bif_info.c
index 0b2c26a548..163070e85a 100644
--- a/erts/emulator/beam/erl_bif_info.c
+++ b/erts/emulator/beam/erl_bif_info.c
@@ -22,6 +22,7 @@
 #  include "config.h"
 #endif
 
+#define ERTS_WANT_MEM_MAPPERS
 #include "sys.h"
 #include "erl_vm.h"
 #include "global.h"
diff --git a/erts/emulator/hipe/hipe_amd64.c b/erts/emulator/hipe/hipe_amd64.c
index 1bb336637d..9d7ada18bc 100644
--- a/erts/emulator/hipe/hipe_amd64.c
+++ b/erts/emulator/hipe/hipe_amd64.c
@@ -83,21 +83,6 @@ int hipe_patch_call(void *callAddress, void *destAddress, void *trampoline)
     return 0;
 }
 
-/*
- * Memory allocator for executable code.
- *
- * This is required on AMD64 because some Linux kernels
- * (including 2.6.10-rc1 and newer www.kernel.org ones)
- * default to non-executable memory mappings, causing
- * ordinary malloc() memory to be non-executable.
- *
- * Implementing this properly also allows us to ensure that
- * executable code ends up in the low 2GB of the address space,
- * as required by HiPE/AMD64's small code model.
- */
-static unsigned int code_bytes;
-static char *code_next;
-
 #if 0	/* change to non-zero to get allocation statistics at exit() */
 static unsigned int total_mapped, nr_joins, nr_splits, total_alloc, nr_allocs, nr_large, total_lost;
 static unsigned int atexit_done;
@@ -121,101 +106,20 @@ static void atexit_alloc_code_stats(void)
 #define ALLOC_CODE_STATS(X)	do{}while(0)
 #endif
 
-/* FreeBSD 6.1 breakage */
-#if !defined(MAP_ANONYMOUS) && defined(MAP_ANON)
-#define MAP_ANONYMOUS MAP_ANON
-#endif
-
-static int morecore(unsigned int alloc_bytes)
-{
-    unsigned int map_bytes;
-    char *map_hint, *map_start;
-
-    /* Page-align the amount to allocate. */
-    map_bytes = (alloc_bytes + 4095) & ~4095;
-
-    /* Round up small allocations. */
-    if (map_bytes < 1024*1024)
-	map_bytes = 1024*1024;
-    else
-	ALLOC_CODE_STATS(++nr_large);
-
-    /* Create a new memory mapping, ensuring it is executable
-       and in the low 2GB of the address space. Also attempt
-       to make it adjacent to the previous mapping. */
-    map_hint = code_next + code_bytes;
-#if !defined(MAP_32BIT)
-    /* FreeBSD doesn't have MAP_32BIT, and it doesn't respect
-       a plain map_hint (returns high mappings even though the
-       hint refers to a free area), so we have to use both map_hint
-       and MAP_FIXED to get addresses below the 2GB boundary.
-       This is even worse than the Linux/ppc64 case.
-       Similarly, Solaris 10 doesn't have MAP_32BIT,
-       and it doesn't respect a plain map_hint. */
-    if (!map_hint) /* first call */
-	map_hint = (char*)(512*1024*1024); /* 0.5GB */
-#endif
-    if ((unsigned long)map_hint & 4095)
-	abort();
-    map_start = mmap(map_hint, map_bytes,
-		     PROT_EXEC|PROT_READ|PROT_WRITE,
-		     MAP_PRIVATE|MAP_ANONYMOUS
-#if defined(MAP_32BIT)
-		     |MAP_32BIT
-#elif defined(__FreeBSD__) || defined(__sun__)
-		     |MAP_FIXED
-#endif
-		     ,
-		     -1, 0);
-    ALLOC_CODE_STATS(fprintf(stderr, "%s: mmap(%p,%u,...) == %p\r\n", __FUNCTION__, map_hint, map_bytes, map_start));
-#if !defined(MAP_32BIT)
-    if (map_start != MAP_FAILED &&
-	(((unsigned long)map_start + (map_bytes-1)) & ~0x7FFFFFFFUL)) {
-	fprintf(stderr, "mmap with hint %p returned code memory %p\r\n", map_hint, map_start);
-	abort();
-    }
-#endif
-    if (map_start == MAP_FAILED)
-	return -1;
-
-    ALLOC_CODE_STATS(total_mapped += map_bytes);
-
-    /* Merge adjacent mappings, so the trailing portion of the previous
-       mapping isn't lost. In practice this is quite successful. */
-    if (map_start == map_hint) {
-	ALLOC_CODE_STATS(++nr_joins);
-	code_bytes += map_bytes;
-#if !defined(MAP_32BIT)
-	if (!code_next) /* first call */
-	    code_next = map_start;
-#endif
-    } else {
-	ALLOC_CODE_STATS(++nr_splits);
-	ALLOC_CODE_STATS(total_lost += code_bytes);
-	code_next = map_start;
-	code_bytes = map_bytes;
-    }
-
-    ALLOC_CODE_STATS(atexit_alloc_code_stats());
-
-    return 0;
-}
-
+/*
+ * Memory allocator for executable code.
+ *
+ * We use a dedicated allocator for executable code (from OTP 19.0)
+ * to make sure the memory we get is executable (PROT_EXEC)
+ * and to ensure that executable code ends up in the low 2GB
+ * of the address space, as required by HiPE/AMD64's small code model.
+ */
 static void *alloc_code(unsigned int alloc_bytes)
 {
-    void *res;
-
-    /* Align function entries. */
-    alloc_bytes = (alloc_bytes + 3) & ~3;
-
-    if (code_bytes < alloc_bytes && morecore(alloc_bytes) != 0)
-	return NULL;
     ALLOC_CODE_STATS(++nr_allocs);
     ALLOC_CODE_STATS(total_alloc += alloc_bytes);
-    res = code_next;
-    code_next += alloc_bytes;
-    code_bytes -= alloc_bytes;
-    return res;
+
+    return erts_alloc(ERTS_ALC_T_HIPE_EXEC, alloc_bytes);
 }
 
 void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process *p)
diff --git a/erts/emulator/sys/common/erl_mmap.c b/erts/emulator/sys/common/erl_mmap.c
index f5f40b35a8..38970bd24e 100644
--- a/erts/emulator/sys/common/erl_mmap.c
+++ b/erts/emulator/sys/common/erl_mmap.c
@@ -356,6 +356,12 @@ char* erts_literals_start;
 UWord erts_literals_size;
 #endif
 
+#ifdef ERTS_ALC_A_EXEC
+ErtsMemMapper erts_exec_mmapper;
+#endif
+
+
+
 #define ERTS_MMAP_SIZE_SC_SA_INC(SZ) 						\
     do {									\
 	mm->size.supercarrier.used.total += (SZ);			\
@@ -1351,10 +1357,39 @@ os_unreserve_physical(char *ptr, UWord size)
 }
 
 static void *
-os_mmap_virtual(char *ptr, UWord size)
+os_mmap_virtual(char *ptr, UWord size, int exec)
 {
-    void *res = mmap((void *) ptr, (size_t) size, ERTS_MMAP_VIRTUAL_PROT,
-		     ERTS_MMAP_VIRTUAL_FLAGS, ERTS_MMAP_FD, 0);
+    int flags = ERTS_MMAP_VIRTUAL_FLAGS;
+    void* res;
+
+#ifdef ERTS_ALC_A_EXEC
+    if (exec) {
+        ASSERT(!ptr);
+        /* OTP-19.0: Nice hack below cut-and-pasted from hipe_amd64.c */
+
+#  ifdef MAP_32BIT
+        /* If we got MAP_32BIT (Linux), then use that to ask for low memory */
+        flags |= MAP_32BIT;
+#  else
+        /* FreeBSD doesn't have MAP_32BIT, and it doesn't respect
+           a plain map_hint (returns high mappings even though the
+           hint refers to a free area), so we have to use both map_hint
+           and MAP_FIXED to get addresses below the 2GB boundary.
+           This is even worse than the Linux/ppc64 case.
+           Similarly, Solaris 10 doesn't have MAP_32BIT,
+           and it doesn't respect a plain map_hint. */
+        ptr = (char*)(512*1024*1024); /* 0.5GB */
+
+#    if defined(__FreeBSD__) || defined(__sun__)
+        flags |= MAP_FIXED;
+#    endif
+#  endif /* !MAP_32BIT */
+    }
+#else /* !ERTS_ALC_A_EXEC */
+    ASSERT(!exec);
+#endif
+    res = mmap((void *) ptr, (size_t) size, ERTS_MMAP_VIRTUAL_PROT,
+               flags, ERTS_MMAP_FD, 0);
     if (res == (void *) MAP_FAILED)
 	return NULL;
     return res;
@@ -2177,7 +2212,7 @@ erts_mmap_init(ErtsMemMapper* mm, ErtsMMapInit *init, int executable)
 	ptr = (char *) ERTS_PAGEALIGNED_CEILING(init->virtual_range.start);
 	end = (char *) ERTS_PAGEALIGNED_FLOOR(init->virtual_range.end);
 	sz = end - ptr;
-	start = os_mmap_virtual(ptr, sz);
+	start = os_mmap_virtual(ptr, sz, executable);
 	if (!start || start > ptr || start >= end)
 	    erts_exit(1,
 		     "erts_mmap: Failed to create virtual range for super carrier\n");
@@ -2202,7 +2237,7 @@ erts_mmap_init(ErtsMemMapper* mm, ErtsMMapInit *init, int executable)
 	sz = ERTS_PAGEALIGNED_CEILING(init->scs);
 #ifdef ERTS_HAVE_OS_PHYSICAL_MEMORY_RESERVATION
 	if (!init->scrpm) {
-	    start = os_mmap_virtual(NULL, sz);
+	    start = os_mmap_virtual(NULL, sz, executable);
 	    mm->reserve_physical = os_reserve_physical;
 	    mm->unreserve_physical = os_unreserve_physical;
 	    virtual_map = 1;
diff --git a/erts/emulator/sys/common/erl_mmap.h b/erts/emulator/sys/common/erl_mmap.h
index 30548bd0b0..8a7530d943 100644
--- a/erts/emulator/sys/common/erl_mmap.h
+++ b/erts/emulator/sys/common/erl_mmap.h
@@ -55,6 +55,11 @@ typedef struct {
 #define ERTS_MMAP_INIT_LITERAL_INITER \
     {{NULL, NULL}, {NULL, NULL}, ERTS_LITERAL_VIRTUAL_AREA_SIZE, 1, (1 << 10), 0}
 
+#define ERTS_HIPE_EXEC_VIRTUAL_AREA_SIZE (UWORD_CONSTANT(512)*1024*1024)
+
+#define ERTS_MMAP_INIT_HIPE_EXEC_INITER \
+    {{NULL, NULL}, {NULL, NULL}, ERTS_HIPE_EXEC_VIRTUAL_AREA_SIZE, 1, (1 << 10), 0}
+
 typedef struct ErtsMemMapper_ ErtsMemMapper;
 
 void *erts_mmap(ErtsMemMapper*, Uint32 flags, UWord *sizep);
@@ -126,10 +131,17 @@ Eterm erts_mmap_debug_info(ErtsMemMapper*, struct process*);
 #  define ERTS_HAVE_OS_MMAP 1
 #endif
 
+#ifdef ERTS_WANT_MEM_MAPPERS
+#  include "erl_alloc_types.h"
+
 extern ErtsMemMapper erts_dflt_mmapper;
-#if defined(ARCH_64) && defined(ERTS_HAVE_OS_PHYSICAL_MEMORY_RESERVATION)
+#  if defined(ARCH_64) && defined(ERTS_HAVE_OS_PHYSICAL_MEMORY_RESERVATION)
 extern ErtsMemMapper erts_literal_mmapper;
-#endif
+#  endif
+#  ifdef ERTS_ALC_A_EXEC
+extern ErtsMemMapper erts_exec_mmapper;
+#  endif
+#endif /* ERTS_WANT_MEM_MAPPERS */
 
 /*#define HARD_DEBUG_MSEG*/
 #ifdef HARD_DEBUG_MSEG
diff --git a/erts/emulator/sys/common/erl_mseg.c b/erts/emulator/sys/common/erl_mseg.c
index f0fbb682b0..f373302e65 100644
--- a/erts/emulator/sys/common/erl_mseg.c
+++ b/erts/emulator/sys/common/erl_mseg.c
@@ -31,6 +31,7 @@
 #  include "config.h"
 #endif
 
+#define ERTS_WANT_MEM_MAPPERS
 #include "sys.h"
 #include "erl_mseg.h"
 #include "global.h"
@@ -1402,6 +1403,12 @@ erts_mseg_init(ErtsMsegInit_t *init)
 
     erts_mtx_init(&init_atoms_mutex, "mseg_init_atoms");
 
+#ifdef ERTS_ALC_A_EXEC
+    /* Initialize erts_exec_mapper *FIRST*, to increase probability
+     * of getting low memory for HiPE AMD64's small code model.
+     */
+    erts_mmap_init(&erts_exec_mmapper, &init->exec_mmap, 1);
+#endif
     erts_mmap_init(&erts_dflt_mmapper, &init->dflt_mmap, 0);
 #if defined(ARCH_64) && defined(ERTS_HAVE_OS_PHYSICAL_MEMORY_RESERVATION)
     erts_mmap_init(&erts_literal_mmapper, &init->literal_mmap, 0);
diff --git a/erts/emulator/sys/common/erl_mseg.h b/erts/emulator/sys/common/erl_mseg.h
index 2acd8f8505..556a135798 100644
--- a/erts/emulator/sys/common/erl_mseg.h
+++ b/erts/emulator/sys/common/erl_mseg.h
@@ -61,6 +61,7 @@ typedef struct {
     Uint nos;
     ErtsMMapInit dflt_mmap;
     ErtsMMapInit literal_mmap;
+    ErtsMMapInit exec_mmap;
 } ErtsMsegInit_t;
 
 #define ERTS_MSEG_INIT_DEFAULT_INITIALIZER				\
@@ -70,7 +71,8 @@ typedef struct {
     10,			/* mcs:   Max cache size		*/	\
     1000,		/* cci:   Cache check interval		*/	\
     ERTS_MMAP_INIT_DEFAULT_INITER,					\
-    ERTS_MMAP_INIT_LITERAL_INITER                                       \
+    ERTS_MMAP_INIT_LITERAL_INITER,                                      \
+    ERTS_MMAP_INIT_HIPE_EXEC_INITER                                     \
 }
 
 typedef struct {
diff --git a/erts/etc/common/erlexec.c b/erts/etc/common/erlexec.c
index 54da59e50d..96259f8fa6 100644
--- a/erts/etc/common/erlexec.c
+++ b/erts/etc/common/erlexec.c
@@ -74,6 +74,7 @@ static const char plusM_au_allocs[]= {
     'R',	/* driver_alloc		*/
     'S',	/* sl_alloc		*/
     'T',	/* temp_alloc		*/
+    'X',	/* exec_alloc		*/
     'Z',        /* test_alloc           */
     '\0'
 };
@@ -123,6 +124,7 @@ static char *plusM_other_switches[] = {
     "Ytp",
     "Ytt",
     "Iscs",
+    "Xscs",
     NULL
 };