Merge branch 'sverk/bif-args/OTP-9662'

* sverk/bif-args/OTP-9662:
  erts,hipe: Limited support for hipe cross compilation
  erts-hipe: Change THE_NON_VALUE for HiPE enabled debug emulator
  erts-hipe: Enable debug compiled hipe-VM with lock checker
  erts-hipe: Rename fail_bif_interface_0 to standard_bif_interface_0
  erts-hipe: Deliberate leak of native fun entries
  erts-hipe: Fix new trap conventions for x86, amd64 and ppc
  Store the trap address in p->i
  Store the trap arguments in the X register array
  erts-hipe: Make some primops use new BIF calling convention
  erts-hipe: Adapt generated BIF wrappers for new calling convention
  erts-hipe: Remove obscuring macros in generated assembler code
  erts-hipe: Make hipe enabled emulator compile with new BIF calls
  Simplify the instructions for calling BIFs
  Change the calling convention for BIFs
  Use the proper macros in all BIFs

Conflicts:
	erts/emulator/beam/bif.h
	erts/emulator/beam/erl_bif_info.c

17 files changed, 380 insertions, 165 deletions

diff --git a/erts/emulator/hipe/hipe_abi.txt b/erts/emulator/hipe/hipe_abi.txt
index d0ec162342..9d4726de9d 100644
--- a/erts/emulator/hipe/hipe_abi.txt
+++ b/erts/emulator/hipe/hipe_abi.txt
@@ -62,7 +62,7 @@ exceptional condition, it puts an error code in p->freason
 and returns THE_NON_VALUE (zero, except in debug mode).
 
 If p->freason == TRAP, then the BIF redirects its call to some
-other function, given by p->def_arg_reg[].
+other function, given by p->i
 The BIF and the new callee may have different arities.
 
 The "hipe_${ARCH}_bifs.m4" macro files take care of these issues
diff --git a/erts/emulator/hipe/hipe_amd64_bifs.m4 b/erts/emulator/hipe/hipe_amd64_bifs.m4
index 0ba763cbea..97a8267647 100644
--- a/erts/emulator/hipe/hipe_amd64_bifs.m4
+++ b/erts/emulator/hipe/hipe_amd64_bifs.m4
@@ -20,24 +20,37 @@ changecom(`/*', `*/')dnl
 
 
 include(`hipe/hipe_amd64_asm.m4')
+#`include' "config.h"
 #`include' "hipe_literals.h"
 
+
 `#if THE_NON_VALUE == 0
 #define TEST_GOT_EXN	testq	%rax, %rax
 #else
 #define TEST_GOT_EXN	cmpq	$THE_NON_VALUE, %rax
 #endif'
 
-`#define TEST_GOT_MBUF		movq P_MBUF(P), %rdx; testq %rdx, %rdx; jnz 3f; 2:
-#define JOIN3(A,B,C)		A##B##C
-#define HANDLE_GOT_MBUF(ARITY)	3: call JOIN3(nbif_,ARITY,_gc_after_bif); jmp 2b'
+define(TEST_GOT_MBUF,`movq P_MBUF(P), %rdx	# `TEST_GOT_MBUF'
+	testq %rdx, %rdx
+	jnz 3f	
+2:')
+define(HANDLE_GOT_MBUF,`
+3:	call nbif_$1_gc_after_bif	# `HANDLE_GOT_MBUF'
+	jmp 2b')
+
+`#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP)
+#  define CALL_BIF(F)	movq $CSYM(F), P_BIF_CALLEE(P); call CSYM(hipe_debug_bif_wrapper) 
+#else
+#  define CALL_BIF(F)	call	CSYM(F)
+#endif'
 
 /*
  * standard_bif_interface_1(nbif_name, cbif_name)
  * standard_bif_interface_2(nbif_name, cbif_name)
  * standard_bif_interface_3(nbif_name, cbif_name)
+ * standard_bif_interface_0(nbif_name, cbif_name)
  *
- * Generate native interface for a BIF with 1-3 parameters and
+ * Generate native interface for a BIF with 0-3 parameters and
  * standard failure mode.
  */
 define(standard_bif_interface_1,
@@ -54,7 +67,11 @@ ASYM($1):
 
 	/* make the call on the C stack */
 	SWITCH_ERLANG_TO_C
-	call	CSYM($2)
+	pushq	%rsi
+	movq	%rsp, %rsi	/* Eterm* BIF__ARGS */
+	sub	$(8), %rsp	/* stack frame 16-byte alignment */
+	CALL_BIF($2)
+	add	$(1*8 + 8), %rsp
 	TEST_GOT_MBUF
 	SWITCH_C_TO_ERLANG
 
@@ -82,7 +99,11 @@ ASYM($1):
 
 	/* make the call on the C stack */
 	SWITCH_ERLANG_TO_C
-	call	CSYM($2)
+	pushq	%rdx
+	pushq 	%rsi
+	movq	%rsp, %rsi	/* Eterm* BIF__ARGS */
+	CALL_BIF($2)
+	add	$(2*8), %rsp
 	TEST_GOT_MBUF
 	SWITCH_C_TO_ERLANG
 
@@ -111,7 +132,13 @@ ASYM($1):
 
 	/* make the call on the C stack */
 	SWITCH_ERLANG_TO_C
-	call	CSYM($2)
+	pushq 	%rcx
+	pushq	%rdx
+	pushq	%rsi
+	movq	%rsp, %rsi	/* Eterm* BIF__ARGS */
+	sub	$(8), %rsp	/* stack frame 16-byte alignment */  
+	CALL_BIF($2)
+	add	$(3*8 + 8), %rsp
 	TEST_GOT_MBUF
 	SWITCH_C_TO_ERLANG
 
@@ -124,13 +151,7 @@ ASYM($1):
 	TYPE_FUNCTION(ASYM($1))
 #endif')
 
-/*
- * fail_bif_interface_0(nbif_name, cbif_name)
- *
- * Generate native interface for a BIF with 0 parameters and
- * standard failure mode.
- */
-define(fail_bif_interface_0,
+define(standard_bif_interface_0,
 `
 #ifndef HAVE_$1
 #`define' HAVE_$1
@@ -143,7 +164,7 @@ ASYM($1):
 
 	/* make the call on the C stack */
 	SWITCH_ERLANG_TO_C
-	call	CSYM($2)
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 	SWITCH_C_TO_ERLANG
 
diff --git a/erts/emulator/hipe/hipe_arm_bifs.m4 b/erts/emulator/hipe/hipe_arm_bifs.m4
index 3664fb6502..e0c6f09796 100644
--- a/erts/emulator/hipe/hipe_arm_bifs.m4
+++ b/erts/emulator/hipe/hipe_arm_bifs.m4
@@ -20,18 +20,27 @@ changecom(`/*', `*/')dnl
 
 
 include(`hipe/hipe_arm_asm.m4')
+#`include' "config.h"
 #`include' "hipe_literals.h"
 
 	.text
 	.p2align 2
 
-`#define JOIN3(A,B,C)		A##B##C
-#define TEST_GOT_MBUF(ARITY)	ldr r1, [P, #P_MBUF]; cmp r1, #0; blne JOIN3(nbif_,ARITY,_gc_after_bif)'
+`#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP)
+#  define CALL_BIF(F)	mov r14, #F; str r14, [r0, #P_BIF_CALLEE]; bl hipe_debug_bif_wrapper
+#else
+#  define CALL_BIF(F)	bl	F
+#endif'
+
+define(TEST_GOT_MBUF,`ldr r1, [P, #P_MBUF]	/* `TEST_GOT_MBUF' */
+	cmp r1, #0
+	blne nbif_$1_gc_after_bif')
 
 /*
  * standard_bif_interface_1(nbif_name, cbif_name)
  * standard_bif_interface_2(nbif_name, cbif_name)
  * standard_bif_interface_3(nbif_name, cbif_name)
+ * standard_bif_interface_0(nbif_name, cbif_name)
  *
  * Generate native interface for a BIF with 1-3 parameters and
  * standard failure mode.
@@ -48,7 +57,9 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	bl	$2
+	str	r1, [r0, #P_ARG0]	/* Store BIF__ARGS in def_arg_reg[] */
+	add	r1, r0, #P_ARG0
+	CALL_BIF($2)
 	TEST_GOT_MBUF(1)
 
 	/* Restore registers. Check for exception. */
@@ -73,7 +84,10 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	bl	$2
+	str	r1, [r0, #P_ARG0]	/* Store BIF__ARGS in def_arg_reg[] */
+	str	r2, [r0, #P_ARG1]
+	add	r1, r0, #P_ARG0
+	CALL_BIF($2)
 	TEST_GOT_MBUF(2)
 
 	/* Restore registers. Check for exception. */
@@ -99,7 +113,11 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	bl	$2
+	str	r1, [r0, #P_ARG0]	/* Store BIF__ARGS in def_arg_reg[] */
+	str	r2, [r0, #P_ARG1]
+	str	r3, [r0, #P_ARG2]
+	add	r1, r0, #P_ARG0
+	CALL_BIF($2)
 	TEST_GOT_MBUF(3)
 
 	/* Restore registers. Check for exception. */
@@ -111,13 +129,7 @@ $1:
 	.type	$1, %function
 #endif')
 
-/*
- * fail_bif_interface_0(nbif_name, cbif_name)
- *
- * Generate native interface for a BIF with 0 parameters and
- * standard failure mode.
- */
-define(fail_bif_interface_0,
+define(standard_bif_interface_0,
 `
 #ifndef HAVE_$1
 #`define' HAVE_$1
@@ -128,7 +140,8 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	bl	$2
+	/* ignore empty BIF__ARGS */
+	CALL_BIF($2)
 	TEST_GOT_MBUF(0)
 
 	/* Restore registers. Check for exception. */
diff --git a/erts/emulator/hipe/hipe_bif0.h b/erts/emulator/hipe/hipe_bif0.h
index f02e8862dc..c512d66f9d 100644
--- a/erts/emulator/hipe/hipe_bif0.h
+++ b/erts/emulator/hipe/hipe_bif0.h
@@ -29,7 +29,7 @@ extern Uint *hipe_bifs_find_pc_from_mfa(Eterm mfa);
 
 extern void hipe_mfa_info_table_init(void);
 extern void *hipe_get_remote_na(Eterm m, Eterm f, unsigned int a);
-extern Eterm hipe_find_na_or_make_stub(Process*, Eterm, Eterm, Eterm);
+extern BIF_RETTYPE hipe_find_na_or_make_stub(BIF_ALIST_3);
 extern int hipe_find_mfa_from_ra(const void *ra, Eterm *m, Eterm *f, unsigned int *a);
 #if defined(__powerpc__) || defined(__ppc__) || defined(__powerpc64__) || defined(__arm__)
 extern void *hipe_mfa_get_trampoline(Eterm m, Eterm f, unsigned int a);
diff --git a/erts/emulator/hipe/hipe_bif2.c b/erts/emulator/hipe/hipe_bif2.c
index 2660f74a82..ee97541e15 100644
--- a/erts/emulator/hipe/hipe_bif2.c
+++ b/erts/emulator/hipe/hipe_bif2.c
@@ -166,3 +166,26 @@ BIF_RETTYPE hipe_bifs_show_message_area_0(BIF_ALIST_0)
     BIF_RET(am_false);
 #endif
 }
+
+#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP)
+
+BIF_RETTYPE hipe_debug_bif_wrapper(BIF_ALIST_1);
+
+#    define ERTS_SMP_REQ_PROC_MAIN_LOCK(P) \
+        if ((P)) erts_proc_lc_require_lock((P), ERTS_PROC_LOCK_MAIN)
+#    define ERTS_SMP_UNREQ_PROC_MAIN_LOCK(P) \
+        if ((P)) erts_proc_lc_unrequire_lock((P), ERTS_PROC_LOCK_MAIN)
+
+BIF_RETTYPE hipe_debug_bif_wrapper(BIF_ALIST_1)
+{
+    typedef BIF_RETTYPE Bif(BIF_ALIST_1);
+    Bif* fp = (Bif*) (BIF_P->hipe.bif_callee);
+    BIF_RETTYPE res;
+    ERTS_SMP_UNREQ_PROC_MAIN_LOCK(BIF_P);
+    res = (*fp)(BIF_P, BIF__ARGS);
+    ERTS_SMP_REQ_PROC_MAIN_LOCK(BIF_P);
+    return res;
+}
+
+#endif /* ERTS_ENABLE_LOCK_CHECK && ERTS_SMP */
+
diff --git a/erts/emulator/hipe/hipe_bif_list.m4 b/erts/emulator/hipe/hipe_bif_list.m4
index 083788997b..14fb04f609 100644
--- a/erts/emulator/hipe/hipe_bif_list.m4
+++ b/erts/emulator/hipe/hipe_bif_list.m4
@@ -70,24 +70,18 @@
  ****************************************************************/
 
 /*
+ * standard_bif_interface_0(nbif_name, cbif_name)
  * standard_bif_interface_1(nbif_name, cbif_name)
  * standard_bif_interface_2(nbif_name, cbif_name)
  * standard_bif_interface_3(nbif_name, cbif_name)
  *
- * A BIF with implicit P parameter, 1-3 ordinary parameters,
+ * A BIF with implicit P parameter, 0-3 ordinary parameters,
  * which may fail.
  * HP and FCALLS may be read and updated.
  * HP_LIMIT, NSP, NSP_LIMIT, and NRA may not be accessed.
  */
 
 /*
- * fail_bif_interface_0(nbif_name, cbif_name)
- *
- * A zero-arity BIF which may fail, otherwise
- * identical to standard_bif_interface_N.
- */
-
-/*
  * nofail_primop_interface_0(nbif_name, cbif_name)
  * nofail_primop_interface_1(nbif_name, cbif_name)
  * nofail_primop_interface_2(nbif_name, cbif_name)
@@ -150,8 +144,8 @@
 /*
  * Zero-arity BIFs that can fail.
  */
-fail_bif_interface_0(nbif_memory_0, memory_0)
-fail_bif_interface_0(nbif_processes_0, processes_0)
+standard_bif_interface_0(nbif_memory_0, memory_0)
+standard_bif_interface_0(nbif_processes_0, processes_0)
 
 /*
  * BIFs and primops that may do a GC (change heap limit and walk the native stack).
@@ -176,10 +170,10 @@ gc_bif_interface_0(nbif_hipe_bifs_nstack_used_size_0, hipe_bifs_nstack_used_size
 /*
  * Arithmetic operators called indirectly by the HiPE compiler.
  */
-standard_bif_interface_2(nbif_add_2, erts_mixed_plus)
-standard_bif_interface_2(nbif_sub_2, erts_mixed_minus)
-standard_bif_interface_2(nbif_mul_2, erts_mixed_times)
-standard_bif_interface_2(nbif_div_2, erts_mixed_div)
+standard_bif_interface_2(nbif_add_2, splus_2)
+standard_bif_interface_2(nbif_sub_2, sminus_2)
+standard_bif_interface_2(nbif_mul_2, stimes_2)
+standard_bif_interface_2(nbif_div_2, div_2)
 standard_bif_interface_2(nbif_intdiv_2, intdiv_2)
 standard_bif_interface_2(nbif_rem_2, rem_2)
 standard_bif_interface_2(nbif_bsl_2, bsl_2)
@@ -261,11 +255,6 @@ noproc_primop_interface_1(nbif_atomic_inc, hipe_atomic_inc)
 ',)dnl
 
 /*
- * Implement standard_bif_interface_0 as nofail_primop_interface_0.
- */
-define(standard_bif_interface_0,`nofail_primop_interface_0($1, $2)')
-
-/*
  * Standard BIFs.
  * BIF_LIST(ModuleAtom,FunctionAtom,Arity,CFun,Index)
  */
diff --git a/erts/emulator/hipe/hipe_mkliterals.c b/erts/emulator/hipe/hipe_mkliterals.c
index bced90785d..61e15f1d58 100644
--- a/erts/emulator/hipe/hipe_mkliterals.c
+++ b/erts/emulator/hipe/hipe_mkliterals.c
@@ -1,9 +1,8 @@
 /*
  * %CopyrightBegin%
-
- *
+ * 
  * 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
  * compliance with the License. You should have received a copy of the
@@ -212,6 +211,11 @@ static const unsigned int CRCTABLE[256] = {
     0xB40BBE37, 0xC30C8EA1, 0x5A05DF1B, 0x2D02EF8D,
 };
 
+/* For hipe cross compiler. Hard code all values.
+   No calls by hipe compiler to query the running emulator.
+*/
+static int is_xcomp = 0;
+
 /*
  *  The algorithm for calculating the 32 bit CRC checksum is based upon
  *  documentation and algorithms provided by Dr. Ross N. Williams in the
@@ -243,7 +247,7 @@ crc_update_buf(unsigned int crc_value,
 }
 
 static unsigned int
-crc_update_int(unsigned int crc_value, const unsigned int *p)
+crc_update_int(unsigned int crc_value, const int *p)
 {
     return crc_update_buf(crc_value, p, sizeof *p);
 }
@@ -256,7 +260,7 @@ crc_update_int(unsigned int crc_value, const unsigned int *p)
  */
 static const struct literal {
     const char *name;
-    unsigned int value;
+    int value;
 } literals[] = {
     /* Field offsets in a process struct */
     { "P_HP", offsetof(struct process, htop) },
@@ -289,6 +293,9 @@ static const struct literal {
     { "P_NRA", offsetof(struct process, hipe.nra) },
 #endif
     { "P_NARITY", offsetof(struct process, hipe.narity) },
+# if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP)
+    { "P_BIF_CALLEE", offsetof(struct process, hipe.bif_callee) },
+# endif
 #endif /* HIPE */
 
     /* process flags bits */
@@ -298,7 +305,7 @@ static const struct literal {
     { "FREASON_TRAP", TRAP },
 
     /* special Erlang constants */
-    { "THE_NON_VALUE", THE_NON_VALUE },
+    { "THE_NON_VALUE", (int)THE_NON_VALUE },
 
     /* funs */
 #ifdef HIPE
@@ -452,7 +459,7 @@ static const struct rts_param {
     unsigned int nr;
     const char *name;
     unsigned int is_defined;
-    unsigned int value;
+    int value;
 } rts_params[] = {
     { 1, "P_OFF_HEAP_FUNS",
 #if !defined(HYBRID)
@@ -528,12 +535,12 @@ static void compute_crc(void)
 
 static void c_define_literal(FILE *fp, const struct literal *literal)
 {
-    fprintf(fp, "#define %s %u\n", literal->name, literal->value);
+    fprintf(fp, "#define %s %d\n", literal->name, literal->value);
 }
 
 static void e_define_literal(FILE *fp, const struct literal *literal)
 {
-    fprintf(fp, "-define(%s, %u).\n", literal->name, literal->value);
+    fprintf(fp, "-define(%s, %d).\n", literal->name, literal->value);
 }
 
 static void print_literals(FILE *fp, void (*print_literal)(FILE*, const struct literal*))
@@ -560,7 +567,7 @@ static void print_atom_literals(FILE *fp, void (*print_atom_literal)(FILE*, cons
 static void c_define_param(FILE *fp, const struct rts_param *param)
 {
     if (param->is_defined)
-	fprintf(fp, "#define %s %u\n", param->name, param->value);
+	fprintf(fp, "#define %s %d\n", param->name, param->value);
 }
 
 static void c_case_param(FILE *fp, const struct rts_param *param)
@@ -568,7 +575,7 @@ static void c_case_param(FILE *fp, const struct rts_param *param)
     fprintf(fp, " \\\n");
     fprintf(fp, "\tcase %u: ", param->nr);
     if (param->is_defined)
-	fprintf(fp, "value = %u", param->value);
+	fprintf(fp, "value = %d", param->value);
     else
 	fprintf(fp, "is_defined = 0");
     fprintf(fp, "; break;");
@@ -576,7 +583,15 @@ static void c_case_param(FILE *fp, const struct rts_param *param)
 
 static void e_define_param(FILE *fp, const struct rts_param *param)
 {
-    fprintf(fp, "-define(%s, hipe_bifs:get_rts_param(%u)).\n", param->name, param->nr);
+    if (is_xcomp) {
+	if (param->is_defined)
+	    fprintf(fp, "-define(%s, %d).\n", param->name, param->value);
+	else
+	    fprintf(fp, "-define(%s, []).\n", param->name);	
+    }
+    else {
+	fprintf(fp, "-define(%s, hipe_bifs:get_rts_param(%u)).\n", param->name, param->nr);
+    }    
 }
 
 static void print_params(FILE *fp, void (*print_param)(FILE*,const struct rts_param*))
@@ -613,19 +628,40 @@ static int do_e(FILE *fp, const char* this_exe)
     fprintf(fp, "\n");
     print_params(fp, e_define_param);
     fprintf(fp, "\n");
-    fprintf(fp, "-define(HIPE_SYSTEM_CRC, hipe_bifs:system_crc(%u)).\n", literals_crc);
+    if (is_xcomp) {
+	fprintf(fp, "-define(HIPE_SYSTEM_CRC, %u).\n", system_crc);
+    }
+    else {
+	fprintf(fp, "-define(HIPE_SYSTEM_CRC, hipe_bifs:system_crc(%u)).\n",
+		literals_crc);
+    }
     return 0;
 }
 
 int main(int argc, const char **argv)
 {
+    int i;
+    int (*do_func_ptr)(FILE *, const char*) = NULL;
+
     compute_crc();
-    if (argc == 2) {
-	if (strcmp(argv[1], "-c") == 0)
-	    return do_c(stdout, argv[0]);
-	if (strcmp(argv[1], "-e") == 0)
-	    return do_e(stdout, argv[0]);
+    for (i = 1; i < argc; i++) {
+	if      (strcmp(argv[i], "-c") == 0)
+	    do_func_ptr = &do_c;
+	else if (strcmp(argv[i], "-e") == 0)
+	    do_func_ptr = &do_e;
+	else if (strcmp(argv[i], "-x") == 0)
+	    is_xcomp = 1;
+	else
+	    goto error;
+    }
+    if (do_func_ptr) {
+	return do_func_ptr(stdout, argv[0]);
     }
-    fprintf(stderr, "usage: %s [-c | -e] > output-file\n", argv[0]);
+error:
+    fprintf(stderr, "usage: %s [-x] [-c | -e] > output-file\n"
+	    "\t-c\tC header file\n"
+	    "\t-e\tErlang header file\n"
+	    "\t-x\tCross compile. No dependencies to compiling emulator\n",	    
+	    argv[0]);
     return 1;
 }
diff --git a/erts/emulator/hipe/hipe_mode_switch.c b/erts/emulator/hipe/hipe_mode_switch.c
index e5b8cf8a19..4d75883fc5 100644
--- a/erts/emulator/hipe/hipe_mode_switch.c
+++ b/erts/emulator/hipe/hipe_mode_switch.c
@@ -35,6 +35,17 @@
 #include "hipe_stack.h"
 #include "hipe_bif0.h"	/* hipe_mfa_info_table_init() */
 
+#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP)
+#    define ERTS_SMP_REQ_PROC_MAIN_LOCK(P) \
+        if ((P)) erts_proc_lc_require_lock((P), ERTS_PROC_LOCK_MAIN)
+#    define ERTS_SMP_UNREQ_PROC_MAIN_LOCK(P) \
+        if ((P)) erts_proc_lc_unrequire_lock((P), ERTS_PROC_LOCK_MAIN)
+#else
+#    define ERTS_SMP_REQ_PROC_MAIN_LOCK(P)
+#    define ERTS_SMP_UNREQ_PROC_MAIN_LOCK(P)
+#endif
+
+
 /*
  * Internal debug support.
  * #define HIPE_DEBUG to the desired debug level:
@@ -318,8 +329,8 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[])
 	   * Native code called a BIF, which "failed" with a TRAP to BEAM.
 	   * Prior to returning, the BIF stored (see BIF_TRAP<N>):
 
-	   * the callee's address in p->def_arg_reg[3]
-	   * the callee's parameters in p->def_arg_reg[0..2]
+	   * the callee's address in p->i
+	   * the callee's parameters in reg[0..2]
 	   * the callee's arity in p->arity (for BEAM gc purposes)
 	   *
 	   * We need to remove the BIF's parameters from the native
@@ -331,19 +342,8 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[])
 	   */
 	  unsigned int i, is_recursive = 0;
 
-	  /* Save p->arity, then update it with the original BIF's arity.
-	     Get rid of any stacked parameters in that call. */
-	  /* XXX: hipe_call_from_native_is_recursive() copies data to
-	     reg[], which is useless in the TRAP case. Maybe write a
-	     specialised hipe_trap_from_native_is_recursive() later. */
           if (p->hipe.nsp != NULL) {
-	      unsigned int callee_arity;
-	      callee_arity = p->arity;
-	      p->arity = p->hipe.narity; /* caller's arity */
-	      is_recursive = hipe_call_from_native_is_recursive(p, reg);
-
-	      p->i = (Eterm *)(p->def_arg_reg[3]);
-	      p->arity = callee_arity;
+	      is_recursive = hipe_trap_from_native_is_recursive(p);
           }
 
 	  /* Schedule next process if current process was hibernated or is waiting
@@ -353,15 +353,14 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[])
 	      goto do_schedule;
 	  }
 	  if (p->status == P_WAITING) {
+	      for (i = 0; i < p->arity; ++i)
+		  p->arg_reg[i] = reg[i]; 	      
 	      goto do_schedule;
 	  }
 
-	  for (i = 0; i < p->arity; ++i)
-	      reg[i] = p->def_arg_reg[i];
-
 	  if (is_recursive)
 	      hipe_push_beam_trap_frame(p, reg, p->arity);
-
+	  
 	  result = HIPE_MODE_SWITCH_RES_CALL;
 	  break;
       }
@@ -470,7 +469,9 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[])
 #if !(NR_ARG_REGS > 5)
 	      int reds_in = p->def_arg_reg[5];
 #endif
+	      ERTS_SMP_UNREQ_PROC_MAIN_LOCK(p);
 	      p = schedule(p, reds_in - p->fcalls);
+	      ERTS_SMP_REQ_PROC_MAIN_LOCK(p);
 #ifdef ERTS_SMP
 	      p->hipe_smp.have_receive_locks = 0;
 	      reg = p->scheduler_data->x_reg_array;
diff --git a/erts/emulator/hipe/hipe_native_bif.c b/erts/emulator/hipe/hipe_native_bif.c
index dfb4ca794a..77dee6f9e9 100644
--- a/erts/emulator/hipe/hipe_native_bif.c
+++ b/erts/emulator/hipe/hipe_native_bif.c
@@ -41,9 +41,9 @@
  */
 
 /* for -Wmissing-prototypes :-( */
-extern Eterm hipe_check_process_code_2(Process*, Eterm, Eterm);
-extern Eterm hipe_garbage_collect_1(Process*, Eterm);
-extern Eterm hipe_show_nstack_1(Process*, Eterm);
+extern Eterm hipe_check_process_code_2(BIF_ALIST_2);
+extern Eterm hipe_garbage_collect_1(BIF_ALIST_1);
+extern Eterm hipe_show_nstack_1(BIF_ALIST_1);
 
 /* Used when a BIF can trigger a stack walk. */
 static __inline__ void hipe_set_narity(Process *p, unsigned int arity)
@@ -56,7 +56,7 @@ Eterm hipe_check_process_code_2(BIF_ALIST_2)
     Eterm ret;
 
     hipe_set_narity(BIF_P, 2);
-    ret = check_process_code_2(BIF_P, BIF_ARG_1, BIF_ARG_2);
+    ret = check_process_code_2(BIF_P, BIF__ARGS);
     hipe_set_narity(BIF_P, 0);
     return ret;
 }
@@ -66,7 +66,7 @@ Eterm hipe_garbage_collect_1(BIF_ALIST_1)
     Eterm ret;
 
     hipe_set_narity(BIF_P, 1);
-    ret = garbage_collect_1(BIF_P, BIF_ARG_1);
+    ret = garbage_collect_1(BIF_P, BIF__ARGS);
     hipe_set_narity(BIF_P, 0);
     return ret;
 }
@@ -76,7 +76,7 @@ Eterm hipe_show_nstack_1(BIF_ALIST_1)
     Eterm ret;
 
     hipe_set_narity(BIF_P, 1);
-    ret = hipe_bifs_show_nstack_1(BIF_P, BIF_ARG_1);
+    ret = hipe_bifs_show_nstack_1(BIF_P, BIF__ARGS);
     hipe_set_narity(BIF_P, 0);
     return ret;
 }
@@ -99,8 +99,10 @@ void hipe_gc(Process *p, Eterm need)
  *  has begun.
  * XXX: BUG: native code should check return status
  */
-Eterm hipe_set_timeout(Process *p, Eterm timeout_value)
+BIF_RETTYPE hipe_set_timeout(BIF_ALIST_1)
 {
+    Process* p = BIF_P;
+    Eterm timeout_value = BIF_ARG_1;
 #if !defined(ARCH_64)
     Uint time_val;
 #endif
@@ -286,8 +288,13 @@ static struct StackTrace *get_trace_from_exc(Eterm exc)
  * This does what the (misnamed) Beam instruction 'raise_ss' does,
  * namely, a proper re-throw of an exception that was caught by 'try'.
  */
-Eterm hipe_rethrow(Process *c_p, Eterm exc, Eterm value)
+
+BIF_RETTYPE hipe_rethrow(BIF_ALIST_2)
 {
+    Process* c_p = BIF_P;
+    Eterm exc = BIF_ARG_1;
+    Eterm value = BIF_ARG_2;
+
     c_p->fvalue = value;
     if (c_p->freason == EXC_NULL) {
 	/* a safety check for the R10-0 case; should not happen */
@@ -411,8 +418,12 @@ Eterm hipe_bs_utf8_size(Eterm arg)
 	return make_small(4);
 }
 
-Eterm hipe_bs_put_utf8(Process *p, Eterm arg, byte *base, unsigned int offset)
+BIF_RETTYPE hipe_bs_put_utf8(BIF_ALIST_3)
 {
+    Process* p = BIF_P;
+    Eterm arg = BIF_ARG_1;
+    byte* base = (byte*) BIF_ARG_2;
+    Uint offset = (Uint) BIF_ARG_3;
     byte *save_bin_buf;
     Uint save_bin_offset;
     int res;
@@ -468,13 +479,21 @@ Eterm hipe_bs_put_utf16(Process *p, Eterm arg, byte *base, unsigned int offset,
     return new_offset;
 }
 
-Eterm hipe_bs_put_utf16be(Process *p, Eterm arg, byte *base, unsigned int offset)
+BIF_RETTYPE hipe_bs_put_utf16be(BIF_ALIST_3)
 {
+    Process *p = BIF_P;
+    Eterm arg = BIF_ARG_1;
+    byte *base = (byte*) BIF_ARG_2; 
+    Uint offset = (Uint) BIF_ARG_3;
     return hipe_bs_put_utf16(p, arg, base, offset, 0);
 }
 
-Eterm hipe_bs_put_utf16le(Process *p, Eterm arg, byte *base, unsigned int offset)
+BIF_RETTYPE hipe_bs_put_utf16le(BIF_ALIST_3)
 {
+    Process *p = BIF_P;
+    Eterm arg = BIF_ARG_1;
+    byte *base = (byte*) BIF_ARG_2; 
+    Uint offset = (Uint) BIF_ARG_3;
     return hipe_bs_put_utf16(p, arg, base, offset, BSF_LITTLE);
 }
 
@@ -489,8 +508,10 @@ static int validate_unicode(Eterm arg)
     return 1;
 }
 
-Eterm hipe_bs_validate_unicode(Process *p, Eterm arg)
+BIF_RETTYPE hipe_bs_validate_unicode(BIF_ALIST_1)
 {
+    Process *p = BIF_P;
+    Eterm arg = BIF_ARG_1;
     if (!validate_unicode(arg))
 	BIF_ERROR(p, BADARG);
     return NIL;
diff --git a/erts/emulator/hipe/hipe_native_bif.h b/erts/emulator/hipe/hipe_native_bif.h
index 13a02b84a2..8c9dec180e 100644
--- a/erts/emulator/hipe/hipe_native_bif.h
+++ b/erts/emulator/hipe/hipe_native_bif.h
@@ -23,6 +23,7 @@
 #ifndef HIPE_NATIVE_BIF_H
 #define HIPE_NATIVE_BIF_H
 
+#include "bif.h"
 #include "hipe_arch.h"
 
 /*
@@ -71,24 +72,24 @@ AEXTERN(void,nbif_select_msg,(Process*));
 AEXTERN(Eterm,nbif_cmp_2,(void));
 AEXTERN(Eterm,nbif_eq_2,(void));
 
-Eterm hipe_nonclosure_address(Process*, Eterm, Uint);
-Eterm hipe_conv_big_to_float(Process*, Eterm);
+BIF_RETTYPE hipe_nonclosure_address(BIF_ALIST_2);
+BIF_RETTYPE hipe_conv_big_to_float(BIF_ALIST_1);
 void hipe_fclearerror_error(Process*);
 void hipe_select_msg(Process*);
 void hipe_gc(Process*, Eterm);
-Eterm hipe_set_timeout(Process*, Eterm);
+BIF_RETTYPE hipe_set_timeout(BIF_ALIST_1);
 void hipe_handle_exception(Process*);
-Eterm hipe_rethrow(Process *c_p, Eterm exc, Eterm value);
+BIF_RETTYPE hipe_rethrow(BIF_ALIST_2);
 char *hipe_bs_allocate(int);
 Binary *hipe_bs_reallocate(Binary*, int);
 int hipe_bs_put_small_float(Process*, Eterm, Uint, byte*, unsigned, unsigned);
 void hipe_bs_put_bits(Eterm, Uint, byte*, unsigned, unsigned);
 Eterm hipe_bs_utf8_size(Eterm);
-Eterm hipe_bs_put_utf8(Process*, Eterm, byte*, unsigned int);
+BIF_RETTYPE hipe_bs_put_utf8(BIF_ALIST_3);
 Eterm hipe_bs_utf16_size(Eterm);
-Eterm hipe_bs_put_utf16be(Process*, Eterm, byte*, unsigned int);
-Eterm hipe_bs_put_utf16le(Process*, Eterm, byte*, unsigned int);
-Eterm hipe_bs_validate_unicode(Process*, Eterm);
+BIF_RETTYPE hipe_bs_put_utf16be(BIF_ALIST_3);
+BIF_RETTYPE hipe_bs_put_utf16le(BIF_ALIST_3);
+BIF_RETTYPE hipe_bs_validate_unicode(BIF_ALIST_1);
 struct erl_bin_match_buffer;
 int hipe_bs_validate_unicode_retract(struct erl_bin_match_buffer*, Eterm);
 
diff --git a/erts/emulator/hipe/hipe_ppc_asm.m4 b/erts/emulator/hipe/hipe_ppc_asm.m4
index 0eb5c441e6..343402f9f0 100644
--- a/erts/emulator/hipe/hipe_ppc_asm.m4
+++ b/erts/emulator/hipe/hipe_ppc_asm.m4
@@ -31,12 +31,23 @@ define(LOAD,ld)dnl
 define(STORE,std)dnl
 define(CMPI,cmpdi)dnl
 define(WSIZE,8)dnl
+`#define STORE_IA(ADDR, DST, TMP)  \
+	addis TMP, 0, ADDR@highest 	SEMI\
+	ori TMP, TMP, ADDR@higher  	SEMI\
+	rldicr TMP, TMP, 32, 31 	SEMI\
+	oris TMP, TMP, ADDR@h 		SEMI\
+	ori TMP, TMP, ADDR@l		SEMI\
+	std TMP, DST'
 ',`
 /* 32-bit PowerPC */
 define(LOAD,lwz)dnl
 define(STORE,stw)dnl
 define(CMPI,cmpwi)dnl
 define(WSIZE,4)dnl
+`#define STORE_IA(ADDR, DST, TMP)   \
+	lis TMP, ADDR@ha 	SEMI\
+	addi TMP, TMP, ADDR@l	SEMI\
+	stw TMP, DST'
 ')dnl
 `#define LOAD	'LOAD
 `#define STORE	'STORE
diff --git a/erts/emulator/hipe/hipe_ppc_bifs.m4 b/erts/emulator/hipe/hipe_ppc_bifs.m4
index 203fefe1a1..d09551d10d 100644
--- a/erts/emulator/hipe/hipe_ppc_bifs.m4
+++ b/erts/emulator/hipe/hipe_ppc_bifs.m4
@@ -20,21 +20,34 @@ changecom(`/*', `*/')dnl
 
 
 include(`hipe/hipe_ppc_asm.m4')
+#`include' "config.h"
 #`include' "hipe_literals.h"
 
+`#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP)
+#  define CALL_BIF(F)	STORE_IA(CSYM(F), P_BIF_CALLEE(P), r29); bl CSYM(hipe_debug_bif_wrapper) 
+#else
+#  define CALL_BIF(F)	bl	CSYM(F)
+#endif'
+
 	.text
 	.p2align 2
 
-`#define TEST_GOT_MBUF		LOAD r4, P_MBUF(P) SEMI CMPI r4, 0 SEMI bne- 3f SEMI 2:
-#define JOIN3(A,B,C)		A##B##C
-#define HANDLE_GOT_MBUF(ARITY)	3: bl CSYM(JOIN3(nbif_,ARITY,_gc_after_bif)) SEMI b 2b'
+define(TEST_GOT_MBUF,`LOAD r4, P_MBUF(P)	# `TEST_GOT_MBUF'
+	CMPI r4, 0
+	bne- 3f
+2:')
+define(HANDLE_GOT_MBUF,`
+3:	bl CSYM(nbif_$1_gc_after_bif)	# `HANDLE_GOT_MBUF'
+	b 2b')
+
 
 /*
  * standard_bif_interface_1(nbif_name, cbif_name)
  * standard_bif_interface_2(nbif_name, cbif_name)
  * standard_bif_interface_3(nbif_name, cbif_name)
+ * standard_bif_interface_0(nbif_name, cbif_name)
  *
- * Generate native interface for a BIF with 1-3 parameters and
+ * Generate native interface for a BIF with 0-3 parameters and
  * standard failure mode.
  */
 define(standard_bif_interface_1,
@@ -49,7 +62,9 @@ ASYM($1):
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	bl	CSYM($2)
+	STORE	r4, P_ARG0(r3)		# Store BIF__ARGS in def_arg_reg[]
+	addi	r4, r3, P_ARG0
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* Restore registers. Check for exception. */
@@ -77,7 +92,10 @@ ASYM($1):
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	bl	CSYM($2)
+	STORE	r4, P_ARG0(r3)		# Store BIF__ARGS in def_arg_reg[]
+	STORE	r5, P_ARG1(r3)
+	addi	r4, r3, P_ARG0
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* Restore registers. Check for exception. */
@@ -106,7 +124,11 @@ ASYM($1):
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	bl	CSYM($2)
+	STORE	r4, P_ARG0(r3)		# Store BIF__ARGS in def_arg_reg[]
+	STORE	r5, P_ARG1(r3)
+	STORE	r6, P_ARG2(r3)
+	addi	r4, r3, P_ARG0
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* Restore registers. Check for exception. */
@@ -121,13 +143,7 @@ ASYM($1):
 	TYPE_FUNCTION(ASYM($1))
 #endif')
 
-/*
- * fail_bif_interface_0(nbif_name, cbif_name)
- *
- * Generate native interface for a BIF with 0 parameters and
- * standard failure mode.
- */
-define(fail_bif_interface_0,
+define(standard_bif_interface_0,
 `
 #ifndef HAVE_$1
 #`define' HAVE_$1
@@ -138,7 +154,8 @@ ASYM($1):
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	bl	CSYM($2)
+	/* ignore empty BIF__ARGS */
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* Restore registers. Check for exception. */
@@ -173,7 +190,8 @@ ASYM($1):
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_GC
-	bl	CSYM($2)
+	/* ignore empty BIF__ARGS */
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* Restore registers. */
@@ -196,7 +214,9 @@ ASYM($1):
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_GC
-	bl	CSYM($2)
+	STORE	r4, P_ARG0(r3)		# Store BIF__ARGS in def_arg_reg[]
+	addi	r4, r3, P_ARG0
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* Restore registers. Check for exception. */
@@ -224,7 +244,10 @@ ASYM($1):
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_GC
-	bl	CSYM($2)
+	STORE	r4, P_ARG0(r3)		# Store BIF__ARGS in def_arg_reg[]
+	STORE	r5, P_ARG1(r3)
+	addi	r4, r3, P_ARG0
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* Restore registers. Check for exception. */
diff --git a/erts/emulator/hipe/hipe_process.h b/erts/emulator/hipe/hipe_process.h
index 5effacb398..43f47d1a28 100644
--- a/erts/emulator/hipe/hipe_process.h
+++ b/erts/emulator/hipe/hipe_process.h
@@ -42,6 +42,9 @@ struct hipe_process_state {
     void (*nra)(void);		/* Native code return address. */
 #endif
     unsigned int narity;	/* Arity of BIF call, for stack walks. */
+#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP)
+    void (*bif_callee)(void);   /* When calling BIF's via debug wrapper */
+#endif
 };
 
 extern void hipe_arch_print_pcb(struct hipe_process_state *p);
diff --git a/erts/emulator/hipe/hipe_risc_glue.h b/erts/emulator/hipe/hipe_risc_glue.h
index e74023e3e9..cc2671c016 100644
--- a/erts/emulator/hipe/hipe_risc_glue.h
+++ b/erts/emulator/hipe/hipe_risc_glue.h
@@ -199,6 +199,22 @@ hipe_call_from_native_is_recursive(Process *p, Eterm reg[])
     return 0;
 }
 
+/* BEAM called native, which called BIF that returned trap
+ * Discard bif parameters.
+ * If tailcall, also clean up native stub continuation. */
+static __inline__ int
+hipe_trap_from_native_is_recursive(Process *p)
+{
+    if (p->hipe.narity > NR_ARG_REGS) {
+	p->hipe.nsp += (p->hipe.narity - NR_ARG_REGS);
+    }
+    if (p->hipe.nra != (void(*)(void))&nbif_return)
+	return 1;
+    hipe_pop_risc_nra_frame(p);
+    return 0;
+}
+
+
 /* Native makes a call which needs to unload the parameters.
    This differs from hipe_call_from_native_is_recursive() in
    that it doesn't check for or pop the BEAM-calls-native frame.
diff --git a/erts/emulator/hipe/hipe_sparc_bifs.m4 b/erts/emulator/hipe/hipe_sparc_bifs.m4
index 03db7f3413..ca5af45d58 100644
--- a/erts/emulator/hipe/hipe_sparc_bifs.m4
+++ b/erts/emulator/hipe/hipe_sparc_bifs.m4
@@ -20,27 +20,42 @@ changecom(`/*', `*/')dnl
 
 
 include(`hipe/hipe_sparc_asm.m4')
+#`include' "config.h"
 #`include' "hipe_literals.h"
 
 	.section ".text"
 	.align	4
 
+`#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP)
+#  define CALL_BIF(F)	set F, %o7; st %o7, [%o0+P_BIF_CALLEE]; call hipe_debug_bif_wrapper 
+#else
+#  define CALL_BIF(F)	call	F
+#endif'
+
 /*
  * Test for exception. This macro executes its delay slot.
  */
-`#define __TEST_GOT_EXN(LABEL)	cmp %o0, THE_NON_VALUE; bz,pn %icc, LABEL
-#define TEST_GOT_EXN(ARITY)	__TEST_GOT_EXN(JOIN3(nbif_,ARITY,_simple_exception))'
+define(TEST_GOT_EXN,`cmp %o0, THE_NON_VALUE	! `TEST_GOT_EXN'
+	bz,pn %icc, nbif_$1_simple_exception')
 
-`#define TEST_GOT_MBUF		ld [P+P_MBUF], %o1; cmp %o1, 0; bne 3f; nop; 2:
-#define JOIN3(A,B,C)		A##B##C
-#define HANDLE_GOT_MBUF(ARITY)	3: call JOIN3(nbif_,ARITY,_gc_after_bif); nop; b 2b; nop'
+define(TEST_GOT_MBUF,`ld [P+P_MBUF], %o1	! `TEST_GOT_MBUF'
+	cmp %o1, 0
+	bne 3f
+	nop
+2:')
+define(HANDLE_GOT_MBUF,`
+3:	call nbif_$1_gc_after_bif	! `HANDLE_GOT_MBUF'
+	nop
+	b 2b
+	nop')
 
 /*
  * standard_bif_interface_1(nbif_name, cbif_name)
  * standard_bif_interface_2(nbif_name, cbif_name)
  * standard_bif_interface_3(nbif_name, cbif_name)
+ * standard_bif_interface_0(nbif_name, cbif_name)
  *
- * Generate native interface for a BIF with 1-3 parameters and
+ * Generate native interface for a BIF with 0-3 parameters and
  * standard failure mode.
  */
 define(standard_bif_interface_1,
@@ -55,7 +70,9 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	call	$2
+	st 	%o1, [%o0+P_ARG0]	! Store BIF__ARGS in def_arg_reg
+	add	%o0, P_ARG0, %o1
+	CALL_BIF($2)
 	nop
 	TEST_GOT_MBUF
 
@@ -81,7 +98,10 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	call	$2
+	st 	%o1, [%o0+P_ARG0]	! Store BIF__ARGS in def_arg_reg
+	st 	%o2, [%o0+P_ARG1]
+	add	%o0, P_ARG0, %o1
+	CALL_BIF($2)
 	nop
 	TEST_GOT_MBUF
 
@@ -108,7 +128,11 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	call	$2
+	st 	%o1, [%o0+P_ARG0]	! Store BIF__ARGS in def_arg_reg
+	st 	%o2, [%o0+P_ARG1]
+	st 	%o3, [%o0+P_ARG2]
+	add	%o0, P_ARG0, %o1
+	CALL_BIF($2)
 	nop
 	TEST_GOT_MBUF
 
@@ -121,13 +145,7 @@ $1:
 	.type	$1, #function
 #endif')
 
-/*
- * fail_bif_interface_0(nbif_name, cbif_name)
- *
- * Generate native interface for a BIF with 0 parameters and
- * standard failure mode.
- */
-define(fail_bif_interface_0,
+define(standard_bif_interface_0,
 `
 #ifndef HAVE_$1
 #`define' HAVE_$1
@@ -138,7 +156,8 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_BIF
-	call	$2
+	/* ignore empty BIF__ARGS */
+	CALL_BIF($2)
 	nop
 	TEST_GOT_MBUF
 
@@ -171,7 +190,8 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_GC
-	call	$2
+	/* ignore empty BIF__ARGS */
+	CALL_BIF($2)
 	nop
 	TEST_GOT_MBUF
 
@@ -195,7 +215,9 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_GC
-	call	$2
+	st 	%o1, [%o0+P_ARG0]	! Store BIF__ARGS in def_arg_reg
+	add	%o0, P_ARG0, %o1
+	CALL_BIF($2)
 	nop
 	TEST_GOT_MBUF
 
@@ -221,7 +243,10 @@ $1:
 
 	/* Save caller-save registers and call the C function. */
 	SAVE_CONTEXT_GC
-	call	$2
+	st 	%o1, [%o0+P_ARG0]	! Store BIF__ARGS in def_arg_reg
+	st 	%o2, [%o0+P_ARG1]
+	add	%o0, P_ARG0, %o1
+	CALL_BIF($2)
 	nop
 	TEST_GOT_MBUF
 
diff --git a/erts/emulator/hipe/hipe_x86_bifs.m4 b/erts/emulator/hipe/hipe_x86_bifs.m4
index 1bb6488b00..2ea69bde3c 100644
--- a/erts/emulator/hipe/hipe_x86_bifs.m4
+++ b/erts/emulator/hipe/hipe_x86_bifs.m4
@@ -20,6 +20,7 @@ changecom(`/*', `*/')dnl
 
 
 include(`hipe/hipe_x86_asm.m4')
+#`include' "config.h"
 #`include' "hipe_literals.h"
 
 `#if THE_NON_VALUE == 0
@@ -28,16 +29,27 @@ include(`hipe/hipe_x86_asm.m4')
 #define TEST_GOT_EXN	cmpl	$THE_NON_VALUE,%eax
 #endif'
 
-`#define TEST_GOT_MBUF		movl P_MBUF(P), %edx; testl %edx, %edx; jnz 3f; 2:
-#define JOIN3(A,B,C)		A##B##C
-#define HANDLE_GOT_MBUF(ARITY)	3: call JOIN3(nbif_,ARITY,_gc_after_bif); jmp 2b'
+`#if defined(ERTS_ENABLE_LOCK_CHECK) && defined(ERTS_SMP)
+#  define CALL_BIF(F)	movl $CSYM(F), P_BIF_CALLEE(P); call CSYM(hipe_debug_bif_wrapper) 
+#else
+#  define CALL_BIF(F)	call	CSYM(F)
+#endif'
+
+define(TEST_GOT_MBUF,`movl P_MBUF(P), %edx	# `TEST_GOT_MBUF'
+	testl %edx, %edx
+	jnz 3f	
+2:')
+define(HANDLE_GOT_MBUF,`
+3:	call nbif_$1_gc_after_bif	# `HANDLE_GOT_MBUF'
+	jmp 2b')
 
 /*
  * standard_bif_interface_1(nbif_name, cbif_name)
  * standard_bif_interface_2(nbif_name, cbif_name)
  * standard_bif_interface_3(nbif_name, cbif_name)
+ * standard_bif_interface_0(nbif_name, cbif_name)
  *
- * Generate native interface for a BIF with 1-3 parameters and
+ * Generate native interface for a BIF with 0-3 parameters and
  * standard failure mode.
  */
 define(standard_bif_interface_1,
@@ -56,8 +68,10 @@ ASYM($1):
 
 	/* make the call on the C stack */
 	NBIF_ARG_REG(0,P)
-	NBIF_ARG(1,1,0)
-	call	CSYM($2)
+	NBIF_ARG(2,1,0)
+	lea 8(%esp), %eax
+	NBIF_ARG_REG(1,%eax)	# BIF__ARGS
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* switch to native stack */
@@ -88,9 +102,11 @@ ASYM($1):
 
 	/* make the call on the C stack */
 	NBIF_ARG_REG(0,P)
-	NBIF_ARG(1,2,0)
-	NBIF_ARG(2,2,1)
-	call	CSYM($2)
+	NBIF_ARG(2,2,0)
+	NBIF_ARG(3,2,1)
+	lea 8(%esp), %eax
+	NBIF_ARG_REG(1,%eax)	# BIF__ARGS
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* switch to native stack */
@@ -121,10 +137,12 @@ ASYM($1):
 
 	/* make the call on the C stack */
 	NBIF_ARG_REG(0,P)
-	NBIF_ARG(1,3,0)
-	NBIF_ARG(2,3,1)
-	NBIF_ARG(3,3,2)
-	call	CSYM($2)
+	NBIF_ARG(2,3,0)
+	NBIF_ARG(3,3,1)
+	NBIF_ARG(4,3,2)
+	lea 8(%esp), %eax
+	NBIF_ARG_REG(1,%eax)	# BIF__ARGS
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* switch to native stack */
@@ -139,13 +157,7 @@ ASYM($1):
 	TYPE_FUNCTION(ASYM($1))
 #endif')
 
-/*
- * fail_bif_interface_0(nbif_name, cbif_name)
- *
- * Generate native interface for a BIF with 0 parameters and
- * standard failure mode.
- */
-define(fail_bif_interface_0,
+define(standard_bif_interface_0,
 `
 #ifndef HAVE_$1
 #`define' HAVE_$1
@@ -158,7 +170,8 @@ ASYM($1):
 
 	/* make the call on the C stack */
 	NBIF_ARG_REG(0,P)
-	call	CSYM($2)
+	/* skip BIF__ARGS */	
+	CALL_BIF($2)
 	TEST_GOT_MBUF
 
 	/* switch to native stack */
diff --git a/erts/emulator/hipe/hipe_x86_glue.h b/erts/emulator/hipe/hipe_x86_glue.h
index a7b0f164be..b0db93267c 100644
--- a/erts/emulator/hipe/hipe_x86_glue.h
+++ b/erts/emulator/hipe/hipe_x86_glue.h
@@ -186,6 +186,25 @@ hipe_call_from_native_is_recursive(Process *p, Eterm reg[])
     return 0;
 }
 
+/* BEAM called native, which called BIF that returned trap
+ * Discard bif parameters.
+ * If tailcall, also clean up native stub continuation. */
+static __inline__ int
+hipe_trap_from_native_is_recursive(Process *p)
+{
+    Eterm nra = *(p->hipe.nsp++);
+
+    if (p->hipe.narity > NR_ARG_REGS) {
+	p->hipe.nsp += (p->hipe.narity - NR_ARG_REGS);
+    }
+    if (nra != (Eterm)nbif_return) {
+	*--(p->hipe.nsp) = nra;
+	return 1;
+    }
+    return 0;
+}
+
+
 /* Native makes a call which needs to unload the parameters.
    This differs from hipe_call_from_native_is_recursive() in
    that it doesn't check for or pop the BEAM-calls-native frame.