erts: Fix global tracing of beam function when called from hipe code

Make hipe to beam calls use export entry.

Makes it a lot easier to handle global tracing correctly
(breakpoints in export entry). A beam function should now be traced
correctly regardless how it is called.

This will also fix a SEGV crash when a hipe stub is made pointing into
a traced export entry and tracing is then stopped which clears the export
entry causing the hipe stub to execute beam instruction NULL.

This commit assumes that hipe code never calls local beam functions, which
should be the case nowadays as we only hipe compile entire modules.

16 files changed, 106 insertions, 108 deletions

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index f6524f36d2..9b251a6ad1 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -5034,7 +5034,10 @@ get_map_elements_fail:
        case HIPE_MODE_SWITCH_RES_RETURN:
 	 ASSERT(is_value(reg[0]));
 	 MoveReturn(reg[0], r(0));
-       case HIPE_MODE_SWITCH_RES_CALL:
+       case HIPE_MODE_SWITCH_RES_CALL_EXPORTED:
+	 c_p->i = c_p->hipe.u.callee_exp->addressv[erts_active_code_ix()];
+	 /*fall through*/
+       case HIPE_MODE_SWITCH_RES_CALL_BEAM:
 	 SET_I(c_p->i);
 	 r(0) = reg[0];
 	 Dispatch();
diff --git a/erts/emulator/hipe/hipe_amd64.c b/erts/emulator/hipe/hipe_amd64.c
index b5dff06987..16c597e7b4 100644
--- a/erts/emulator/hipe/hipe_amd64.c
+++ b/erts/emulator/hipe/hipe_amd64.c
@@ -224,18 +224,19 @@ void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process *
     return alloc_code(nrbytes);
 }
 
-/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2()
-   and hipe_bif0.c:hipe_make_stub() */
-void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
+
+/* Make stub for native code calling exported beam function.
+*/
+void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity)
 {
     /*
      * This creates a native code stub with the following contents:
      *
-     * movq $Address, P_BEAM_IP(%ebp)  %% Actually two movl
+     * movq $Address, P_CALLEE_EXP(%ebp)  %% Actually two movl
      * movb $Arity, P_ARITY(%ebp)
      * jmp callemu
      *
-     * The stub has variable size, depending on whether the P_BEAM_IP
+     * The stub has variable size, depending on whether the P_CALLEE_EXP
      * and P_ARITY offsets fit in 8-bit signed displacements or not.
      * The rel32 offset in the final jmp depends on its actual location,
      * which also depends on the size of the previous instructions.
@@ -248,49 +249,49 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
 
     codeSize =	/* 23, 26, 29, or 32 bytes */
       23 +	/* 23 when all offsets are 8-bit */
-      (P_BEAM_IP >= 128 ? 3 : 0) +
-      ((P_BEAM_IP + 4) >= 128 ? 3 : 0) +
+      (P_CALLEE_EXP >= 128 ? 3 : 0) +
+      ((P_CALLEE_EXP + 4) >= 128 ? 3 : 0) +
       (P_ARITY >= 128 ? 3 : 0);
     codep = code = alloc_code(codeSize);
 
-    /* movl $beamAddress, P_BEAM_IP(%ebp); 3 or 6 bytes, plus 4 */
+    /* movl $callee_exp, P_CALLEE_EXP(%ebp); 3 or 6 bytes, plus 4 */
     codep[0] = 0xc7;
-#if P_BEAM_IP >= 128
+#if P_CALLEE_EXP >= 128
     codep[1] = 0x85;	/* disp32[EBP] */
-    codep[2] =  P_BEAM_IP        & 0xFF;
-    codep[3] = (P_BEAM_IP >>  8) & 0xFF;
-    codep[4] = (P_BEAM_IP >> 16) & 0xFF;
-    codep[5] = (P_BEAM_IP >> 24) & 0xFF;
+    codep[2] =  P_CALLEE_EXP        & 0xFF;
+    codep[3] = (P_CALLEE_EXP >>  8) & 0xFF;
+    codep[4] = (P_CALLEE_EXP >> 16) & 0xFF;
+    codep[5] = (P_CALLEE_EXP >> 24) & 0xFF;
     codep += 6;
 #else
     codep[1] = 0x45;	/* disp8[EBP] */
-    codep[2] = P_BEAM_IP;
+    codep[2] = P_CALLEE_EXP;
     codep += 3;
 #endif
-    codep[0] = ((unsigned long)beamAddress      ) & 0xFF;
-    codep[1] = ((unsigned long)beamAddress >>  8) & 0xFF;
-    codep[2] = ((unsigned long)beamAddress >> 16) & 0xFF;
-    codep[3] = ((unsigned long)beamAddress >> 24) & 0xFF;
+    codep[0] = ((unsigned long)callee_exp      ) & 0xFF;
+    codep[1] = ((unsigned long)callee_exp >>  8) & 0xFF;
+    codep[2] = ((unsigned long)callee_exp >> 16) & 0xFF;
+    codep[3] = ((unsigned long)callee_exp >> 24) & 0xFF;
     codep += 4;
 
-    /* movl (shl 32 $beamAddress), P_BEAM_IP+4(%ebp); 3 or 6 bytes, plus 4 */
+    /* movl (shl 32 $callee_exp), P_CALLEE_EXP+4(%ebp); 3 or 6 bytes, plus 4 */
     codep[0] = 0xc7;
-#if P_BEAM_IP+4 >= 128
+#if P_CALLEE_EXP+4 >= 128
     codep[1] = 0x85;	/* disp32[EBP] */
-    codep[2] =  (P_BEAM_IP+4)        & 0xFF;
-    codep[3] = ((P_BEAM_IP+4) >>  8) & 0xFF;
-    codep[4] = ((P_BEAM_IP+4) >> 16) & 0xFF;
-    codep[5] = ((P_BEAM_IP+4) >> 24) & 0xFF;
+    codep[2] =  (P_CALLEE_EXP+4)        & 0xFF;
+    codep[3] = ((P_CALLEE_EXP+4) >>  8) & 0xFF;
+    codep[4] = ((P_CALLEE_EXP+4) >> 16) & 0xFF;
+    codep[5] = ((P_CALLEE_EXP+4) >> 24) & 0xFF;
     codep += 6;
 #else
     codep[1] = 0x45;	/* disp8[EBP] */
-    codep[2] =  (P_BEAM_IP+4);
+    codep[2] =  (P_CALLEE_EXP+4);
     codep += 3;
 #endif
-    codep[0] = ((unsigned long)beamAddress >> 32) & 0xFF;
-    codep[1] = ((unsigned long)beamAddress >> 40) & 0xFF;
-    codep[2] = ((unsigned long)beamAddress >> 48) & 0xFF;
-    codep[3] = ((unsigned long)beamAddress >> 56) & 0xFF;
+    codep[0] = ((unsigned long)callee_exp >> 32) & 0xFF;
+    codep[1] = ((unsigned long)callee_exp >> 40) & 0xFF;
+    codep[2] = ((unsigned long)callee_exp >> 48) & 0xFF;
+    codep[3] = ((unsigned long)callee_exp >> 56) & 0xFF;
     codep += 4;
 
     /* movb $beamArity, P_ARITY(%ebp); 3 or 6 bytes */
diff --git a/erts/emulator/hipe/hipe_amd64_glue.S b/erts/emulator/hipe/hipe_amd64_glue.S
index 8816906870..bebe0a8fd1 100644
--- a/erts/emulator/hipe/hipe_amd64_glue.S
+++ b/erts/emulator/hipe/hipe_amd64_glue.S
@@ -109,7 +109,7 @@ ASYM(nbif_return):
  * stub (hipe_x86_loader.erl) which should look as follows:
  *
  * stub for f/N:
- *	movq	$<f's BEAM code address>, P_BEAM_IP(P)
+ *	movq	$<f's export entry address>, P_CALLEE_EXP(P)
  *	movb	$<N>, P_ARITY(P)
  *	jmp	nbif_callemu
  *
@@ -119,7 +119,7 @@ ASYM(nbif_return):
 	GLOBAL(ASYM(nbif_callemu))
 ASYM(nbif_callemu):
 	STORE_ARG_REGS
-	movl	$HIPE_MODE_SWITCH_RES_CALL, %eax
+	movl	$HIPE_MODE_SWITCH_RES_CALL_EXPORTED, %eax
 	jmp	.suspend_exit
 
 /*
diff --git a/erts/emulator/hipe/hipe_arm.c b/erts/emulator/hipe/hipe_arm.c
index 3db3ffe9b1..165eb543c8 100644
--- a/erts/emulator/hipe/hipe_arm.c
+++ b/erts/emulator/hipe/hipe_arm.c
@@ -260,9 +260,9 @@ int hipe_patch_insn(void *address, Uint32 value, Eterm type)
     return 0;
 }
 
-/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2()
-   and hipe_bif0.c:hipe_make_stub() */
-void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
+/* Make stub for native code calling exported beam function
+*/
+void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity)
 {
     unsigned int *code;
     unsigned int *tramp_callemu;
@@ -272,9 +272,9 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
      * Native code calls BEAM via a stub looking as follows:
      *
      * mov r0, #beamArity
-     * ldr r8, [pc,#0] // beamAddress
+     * ldr r8, [pc,#0] // callee_exp
      * b nbif_callemu
-     * .long beamAddress
+     * .long callee_exp
      *
      * I'm using r0 and r8 since they aren't used for
      * parameter passing in native code. The branch to
@@ -292,12 +292,12 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
 
     /* mov r0, #beamArity */
     code[0] = 0xE3A00000 | (beamArity & 0xFF);
-    /* ldr r8, [pc,#0] // beamAddress */
+    /* ldr r8, [pc,#0] // callee_exp */
     code[1] = 0xE59F8000;
     /* b nbif_callemu */
     code[2] = 0xEA000000 | (callemu_offset & 0x00FFFFFF);
-    /* .long beamAddress */
-    code[3] = (unsigned int)beamAddress;
+    /* .long callee_exp */
+    code[3] = (unsigned int)callee_exp;
 
     hipe_flush_icache_range(code, 4*sizeof(int));
 
diff --git a/erts/emulator/hipe/hipe_arm_glue.S b/erts/emulator/hipe/hipe_arm_glue.S
index 2e2b8604a6..e58e112ca7 100644
--- a/erts/emulator/hipe/hipe_arm_glue.S
+++ b/erts/emulator/hipe/hipe_arm_glue.S
@@ -135,7 +135,7 @@ hipe_arm_throw_to_native:
  * which should look as follows:
  *
  * stub for f/N:
- *	<set r8 to f's BEAM code address>
+ *	<set r8 to f's export entry address>
  *	<set r0 to N>
  *	b nbif_callemu
  *
@@ -143,10 +143,10 @@ hipe_arm_throw_to_native:
  */
 	.global	nbif_callemu
 nbif_callemu:
-	str	r8, [P, #P_BEAM_IP]
+	str	r8, [P, #P_CALLEE_EXP]
 	str	r0, [P, #P_ARITY]
 	STORE_ARG_REGS
-	mov	r0, #HIPE_MODE_SWITCH_RES_CALL
+	mov	r0, #HIPE_MODE_SWITCH_RES_CALL_EXPORTED
 	b	.suspend_exit
 
 /*
diff --git a/erts/emulator/hipe/hipe_bif0.c b/erts/emulator/hipe/hipe_bif0.c
index 2497d51df1..b686717814 100644
--- a/erts/emulator/hipe/hipe_bif0.c
+++ b/erts/emulator/hipe/hipe_bif0.c
@@ -1490,18 +1490,13 @@ void hipe_mfa_save_orig_beam_op(Eterm mod, Eterm fun, unsigned int ari, Eterm *p
 
 static void *hipe_make_stub(Eterm m, Eterm f, unsigned int arity, int is_remote)
 {
-    void *BEAMAddress;
+    Export *export_entry;
     void *StubAddress;
 
-#if 0
-    if (is_not_atom(m) || is_not_atom(f) || arity > 255)
-	return NULL;
-#endif
-    BEAMAddress = hipe_get_emu_address(m, f, arity, is_remote);
-    StubAddress = hipe_make_native_stub(BEAMAddress, arity);
-#if 0
-    hipe_mfa_set_na(m, f, arity, StubAddress);
-#endif
+    ASSERT(is_remote);
+
+    export_entry = erts_export_get_or_make_stub(m, f, arity);
+    StubAddress = hipe_make_native_stub(export_entry, arity);
     return StubAddress;
 }
 
diff --git a/erts/emulator/hipe/hipe_mkliterals.c b/erts/emulator/hipe/hipe_mkliterals.c
index 23e9214cec..ed355ce264 100644
--- a/erts/emulator/hipe/hipe_mkliterals.c
+++ b/erts/emulator/hipe/hipe_mkliterals.c
@@ -524,6 +524,7 @@ static const struct rts_param rts_params[] = {
     },
     { 49, "P_MSG_FIRST", 1, offsetof(struct process, msg.first) },
     { 50, "P_MSG_SAVE", 1, offsetof(struct process, msg.save) },
+    { 51, "P_CALLEE_EXP", 1, offsetof(struct process, hipe.u.callee_exp) },
 };
 
 #define NR_PARAMS	ARRAY_SIZE(rts_params)
diff --git a/erts/emulator/hipe/hipe_mode_switch.c b/erts/emulator/hipe/hipe_mode_switch.c
index 7b9f2b5350..4dbba9da61 100644
--- a/erts/emulator/hipe/hipe_mode_switch.c
+++ b/erts/emulator/hipe/hipe_mode_switch.c
@@ -396,13 +396,13 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[])
 	  if (is_recursive)
 	      hipe_push_beam_trap_frame(p, reg, p->arity);
 	  
-	  result = HIPE_MODE_SWITCH_RES_CALL;
+	  result = HIPE_MODE_SWITCH_RES_CALL_BEAM;
 	  break;
       }
-      case HIPE_MODE_SWITCH_RES_CALL: {
+      case HIPE_MODE_SWITCH_RES_CALL_EXPORTED: {
 	  /* Native code calls or tailcalls BEAM.
 	   *
-	   * p->i is the callee's BEAM code
+	   * p->hipe.u.callee_exp is the callee's export entry
 	   * p->arity is the callee's arity
 	   * p->def_arg_reg[] contains the register parameters
 	   * p->hipe.nsp[] contains the stacked parameters
@@ -460,7 +460,7 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[])
 	      p->i = closure->fe->address;
 
 	      /* Change result code to the faster plain CALL type. */
-	      result = HIPE_MODE_SWITCH_RES_CALL;
+	      result = HIPE_MODE_SWITCH_RES_CALL_BEAM;
 	  }
 	  /* Append the closure as the last parameter. Don't increment arity. */
 	  reg[arity] = p->hipe.u.closure;
@@ -541,7 +541,7 @@ Process *hipe_mode_switch(Process *p, unsigned cmd, Eterm reg[])
 	      }
 	  }
 	  HIPE_CHECK_PCB(p);
-	  result = HIPE_MODE_SWITCH_RES_CALL;
+	  result = HIPE_MODE_SWITCH_RES_CALL_BEAM;
 	  p->def_arg_reg[3] = result;
 	  return p;
       }
diff --git a/erts/emulator/hipe/hipe_mode_switch.h b/erts/emulator/hipe/hipe_mode_switch.h
index 06721e3c04..6ec5da1ae9 100644
--- a/erts/emulator/hipe/hipe_mode_switch.h
+++ b/erts/emulator/hipe/hipe_mode_switch.h
@@ -31,7 +31,7 @@
 
 /* result codes for beam_emu <- hipe_mode_switch() return */
 #define HIPE_MODE_SWITCH_RES_RETURN		4
-#define HIPE_MODE_SWITCH_RES_CALL		5
+#define HIPE_MODE_SWITCH_RES_CALL_EXPORTED	5
 #define HIPE_MODE_SWITCH_RES_THROW		6
 
 /* additional result codes for hipe_mode_switch() <- native return */
@@ -45,6 +45,8 @@
 
 #define HIPE_MODE_SWITCH_RES_APPLY		13 /* mode_switch <- native */
 
+#define HIPE_MODE_SWITCH_RES_CALL_BEAM	        14
+
 #ifndef ASM
 
 #include "error.h"
diff --git a/erts/emulator/hipe/hipe_ppc.c b/erts/emulator/hipe/hipe_ppc.c
index 2d8fd61e1e..4dc26cdbc8 100644
--- a/erts/emulator/hipe/hipe_ppc.c
+++ b/erts/emulator/hipe/hipe_ppc.c
@@ -285,7 +285,7 @@ int hipe_patch_insn(void *address, Uint64 value, Eterm type)
     }
 }
 
-void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
+void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity)
 {
     unsigned int *code;
 
@@ -294,16 +294,16 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
 
     code = alloc_stub(7);
 
-    /* addis r12,0,beamAddress@highest */
-    code[0] = 0x3d800000 | (((unsigned long)beamAddress >> 48) & 0xffff);
-    /* ori r12,r12,beamAddress@higher */
-    code[1] = 0x618c0000 | (((unsigned long)beamAddress >> 32) & 0xffff);
+    /* addis r12,0,callee_exp@highest */
+    code[0] = 0x3d800000 | (((unsigned long)callee_exp >> 48) & 0xffff);
+    /* ori r12,r12,callee_exp@higher */
+    code[1] = 0x618c0000 | (((unsigned long)callee_exp >> 32) & 0xffff);
     /* sldi r12,r12,32 (rldicr r12,r12,32,31) */
     code[2] = 0x798c07c6;
-    /* oris r12,r12,beamAddress@h */
-    code[3] = 0x658c0000 | (((unsigned long)beamAddress >> 16) & 0xffff);
-    /* ori r12,r12,beamAddress@l */
-    code[4] = 0x618c0000 | ((unsigned long)beamAddress & 0xffff);
+    /* oris r12,r12,callee_exp@h */
+    code[3] = 0x658c0000 | (((unsigned long)callee_exp >> 16) & 0xffff);
+    /* ori r12,r12,callee_exp@l */
+    code[4] = 0x618c0000 | ((unsigned long)callee_exp & 0xffff);
     /* addi r0,0,beamArity */
     code[5] = 0x38000000 | (beamArity & 0x7FFF);
     /* ba nbif_callemu */
@@ -355,18 +355,16 @@ int hipe_patch_insn(void *address, Uint32 value, Eterm type)
     return 0;
 }
 
-/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2()
-   and hipe_bif0.c:hipe_make_stub() */
-void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
+void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity)
 {
     unsigned int *code;
 
     /*
      * Native code calls BEAM via a stub looking as follows:
      *
-     * addi r12,0,beamAddress@l
+     * addi r12,0,callee_exp@l
      * addi r0,0,beamArity
-     * addis r12,r12,beamAddress@ha
+     * addis r12,r12,callee_exp@ha
      * ba nbif_callemu
      *
      * I'm using r0 and r12 since the standard SVR4 ABI allows
@@ -384,12 +382,12 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
 
     code = alloc_stub(4);
 
-    /* addi r12,0,beamAddress@l */
-    code[0] = 0x39800000 | ((unsigned long)beamAddress & 0xFFFF);
+    /* addi r12,0,callee_exp@l */
+    code[0] = 0x39800000 | ((unsigned long)callee_exp & 0xFFFF);
     /* addi r0,0,beamArity */
     code[1] = 0x38000000 | (beamArity & 0x7FFF);
-    /* addis r12,r12,beamAddress@ha */
-    code[2] = 0x3D8C0000 | at_ha((unsigned long)beamAddress);
+    /* addis r12,r12,callee_exp@ha */
+    code[2] = 0x3D8C0000 | at_ha((unsigned long)callee_exp);
     /* ba nbif_callemu */
     code[3] = 0x48000002 | (unsigned long)&nbif_callemu;
 
diff --git a/erts/emulator/hipe/hipe_ppc_glue.S b/erts/emulator/hipe/hipe_ppc_glue.S
index 6f0217c738..0c337a14df 100644
--- a/erts/emulator/hipe/hipe_ppc_glue.S
+++ b/erts/emulator/hipe/hipe_ppc_glue.S
@@ -296,7 +296,7 @@ CSYM(hipe_ppc_throw_to_native):
  * which should look as follows:
  *
  * stub for f/N:
- *	<set r12 to f's BEAM code address>
+ *	<set r12 to f's export entry address>
  *	<set r0 to N>
  *	b nbif_callemu
  *
@@ -312,10 +312,10 @@ CSYM(hipe_ppc_throw_to_native):
  */
 	GLOBAL(ASYM(nbif_callemu))
 ASYM(nbif_callemu):
-	STORE	r12, P_BEAM_IP(P)
+	STORE	r12, P_CALLEE_EXP(P)
 	STORE	r0, P_ARITY(P)
 	STORE_ARG_REGS
-	li	r3, HIPE_MODE_SWITCH_RES_CALL
+	li	r3, HIPE_MODE_SWITCH_RES_CALL_EXPORTED
 	b	.suspend_exit
 
 /*
diff --git a/erts/emulator/hipe/hipe_process.h b/erts/emulator/hipe/hipe_process.h
index cd88a37f80..86655ad42c 100644
--- a/erts/emulator/hipe/hipe_process.h
+++ b/erts/emulator/hipe/hipe_process.h
@@ -23,6 +23,7 @@
 #define HIPE_PROCESS_H
 
 #include "erl_alloc.h"
+#include "export.h"
 
 struct hipe_process_state {
     Eterm *nsp;			/* Native stack pointer. */
@@ -31,6 +32,7 @@ struct hipe_process_state {
     union {
 	void (*ncallee)(void);	/* Native code callee (label) to invoke. */
 	Eterm closure;		/* Used to pass a closure from native code. */
+	Export* callee_exp;     /* Used to pass export entry from native code */
     }u;
     Eterm *nstgraylim;		/* Gray/white stack boundary. */
     Eterm *nstblacklim;		/* Black/gray stack boundary. Must exist if
diff --git a/erts/emulator/hipe/hipe_sparc.c b/erts/emulator/hipe/hipe_sparc.c
index 49d4da7bab..2052aa8498 100644
--- a/erts/emulator/hipe/hipe_sparc.c
+++ b/erts/emulator/hipe/hipe_sparc.c
@@ -204,9 +204,7 @@ void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process *
     return alloc_code(nrbytes);
 }
 
-/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2()
-   and hipe_bif0.c:hipe_make_stub() */
-void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
+void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity)
 {
     unsigned int *code;
     unsigned int callEmuOffset;
@@ -215,11 +213,11 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
     code = alloc_code(5*sizeof(int));
 
     /* sethi %hi(Address), %i4 */
-    code[0] = 0x39000000 | (((unsigned int)beamAddress >> 10) & 0x3FFFFF);
+    code[0] = 0x39000000 | (((unsigned int)callee_exp >> 10) & 0x3FFFFF);
     /* or %g0, %o7, %i3 ! mov %o7, %i3 */
     code[1] = 0xB610000F;
     /* or %i4, %lo(Address), %i4 */
-    code[2] = 0xB8172000 | ((unsigned int)beamAddress & 0x3FF);
+    code[2] = 0xB8172000 | ((unsigned int)callee_exp & 0x3FF);
     /* call callemu */
     callEmuOffset = (char*)nbif_callemu - (char*)&code[3];
     code[3] = (1 << 30) | ((callEmuOffset >> 2) & 0x3FFFFFFF);
diff --git a/erts/emulator/hipe/hipe_sparc_glue.S b/erts/emulator/hipe/hipe_sparc_glue.S
index 44bdf1bc7e..ab40a48ee7 100644
--- a/erts/emulator/hipe/hipe_sparc_glue.S
+++ b/erts/emulator/hipe/hipe_sparc_glue.S
@@ -155,9 +155,9 @@ hipe_sparc_throw_to_native:
  * which should look as follows:
  *
  * stub for f/N:
- *	sethi %hi(f's BEAM code address), TEMP_ARG0
+ *	sethi %hi(f's export entry address), TEMP_ARG0
  *	mov RA, TEMP_RA		! because the call below clobbers RA (%o7)
- *	or TEMP_ARG0, %lo(f's BEAM code address), TEMP_ARG0
+ *	or TEMP_ARG0, %lo(f's export entry address), TEMP_ARG0
  *	call nbif_callemu	! clobbers RA!
  *	mov N, TEMP_ARG1	! delay slot: TEMP_ARG1 := ARITY
  *
@@ -165,12 +165,12 @@ hipe_sparc_throw_to_native:
  */
 	.global nbif_callemu
 nbif_callemu:
-	st	TEMP_ARG0, [P+P_BEAM_IP]
+	st	TEMP_ARG0, [P+P_CALLEE_EXP]
 	st	TEMP_ARG1, [P+P_ARITY]
 	st	TEMP_RA, [P+P_NRA]
 	STORE_ARG_REGS
 	ba	.flush_exit
-	mov	HIPE_MODE_SWITCH_RES_CALL, %o0
+	mov	HIPE_MODE_SWITCH_RES_CALL_EXPORTED, %o0
 
 /*
  * nbif_apply
diff --git a/erts/emulator/hipe/hipe_x86.c b/erts/emulator/hipe/hipe_x86.c
index 327c74e9aa..314f6b597c 100644
--- a/erts/emulator/hipe/hipe_x86.c
+++ b/erts/emulator/hipe/hipe_x86.c
@@ -182,18 +182,16 @@ void *hipe_alloc_code(Uint nrbytes, Eterm callees, Eterm *trampolines, Process *
     return alloc_code(nrbytes);
 }
 
-/* called from hipe_bif0.c:hipe_bifs_make_native_stub_2()
-   and hipe_bif0.c:hipe_make_stub() */
-void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
+void *hipe_make_native_stub(void *callee_exp, unsigned int beamArity)
 {
     /*
      * This creates a native code stub with the following contents:
      *
-     * movl $Address, P_BEAM_IP(%ebp)
+     * movl $Address, P_CALLEE_EXP(%ebp)
      * movb $Arity, P_ARITY(%ebp)
      * jmp callemu
      *
-     * The stub has variable size, depending on whether the P_BEAM_IP
+     * The stub has variable size, depending on whether the P_CALLEE_EXP
      * and P_ARITY offsets fit in 8-bit signed displacements or not.
      * The rel32 offset in the final jmp depends on its actual location,
      * which also depends on the size of the previous instructions.
@@ -206,28 +204,28 @@ void *hipe_make_native_stub(void *beamAddress, unsigned int beamArity)
 
     codeSize =	/* 16, 19, or 22 bytes */
 	16 +	/* 16 when both offsets are 8-bit */
-	(P_BEAM_IP >= 128 ? 3 : 0) +
+	(P_CALLEE_EXP >= 128 ? 3 : 0) +
 	(P_ARITY >= 128 ? 3 : 0);
     codep = code = alloc_code(codeSize);
 
-    /* movl $beamAddress, P_BEAM_IP(%ebp); 3 or 6 bytes, plus 4 */
+    /* movl $beamAddress, P_CALLEE_EXP(%ebp); 3 or 6 bytes, plus 4 */
     codep[0] = 0xc7;
-#if P_BEAM_IP >= 128
+#if P_CALLEE_EXP >= 128
     codep[1] = 0x85;	/* disp32[EBP] */
-    codep[2] = P_BEAM_IP & 0xFF;
-    codep[3] = (P_BEAM_IP >> 8) & 0xFF;
-    codep[4] = (P_BEAM_IP >> 16) & 0xFF;
-    codep[5] = (P_BEAM_IP >> 24) & 0xFF;
+    codep[2] = P_CALLEE_EXP & 0xFF;
+    codep[3] = (P_CALLEE_EXP >> 8) & 0xFF;
+    codep[4] = (P_CALLEE_EXP >> 16) & 0xFF;
+    codep[5] = (P_CALLEE_EXP >> 24) & 0xFF;
     codep += 6;
 #else
     codep[1] = 0x45;	/* disp8[EBP] */
-    codep[2] = P_BEAM_IP;
+    codep[2] = P_CALLEE_EXP;
     codep += 3;
 #endif
-    codep[0] = ((unsigned int)beamAddress) & 0xFF;
-    codep[1] = ((unsigned int)beamAddress >> 8) & 0xFF;
-    codep[2] = ((unsigned int)beamAddress >> 16) & 0xFF;
-    codep[3] = ((unsigned int)beamAddress >> 24) & 0xFF;
+    codep[0] = ((unsigned int)callee_exp) & 0xFF;
+    codep[1] = ((unsigned int)callee_exp >> 8) & 0xFF;
+    codep[2] = ((unsigned int)callee_exp >> 16) & 0xFF;
+    codep[3] = ((unsigned int)callee_exp >> 24) & 0xFF;
     codep += 4;
 
     /* movb $beamArity, P_ARITY(%ebp); 3 or 6 bytes */
diff --git a/erts/emulator/hipe/hipe_x86_glue.S b/erts/emulator/hipe/hipe_x86_glue.S
index 88b86f4de7..638780156a 100644
--- a/erts/emulator/hipe/hipe_x86_glue.S
+++ b/erts/emulator/hipe/hipe_x86_glue.S
@@ -104,7 +104,7 @@ ASYM(nbif_return):
  * stub (hipe_x86_loader.erl) which should look as follows:
  *
  * stub for f/N:
- *	movl	$<f's BEAM code address>, P_BEAM_IP(P)
+ *	movl	$<f's export entry address>, P_CALLEE_EXP(P)
  *	movb	$<N>, P_ARITY(P)
  *	jmp	nbif_callemu
  *
@@ -114,7 +114,7 @@ ASYM(nbif_return):
 	GLOBAL(ASYM(nbif_callemu))
 ASYM(nbif_callemu):
 	STORE_ARG_REGS
-	movl	$HIPE_MODE_SWITCH_RES_CALL, %eax
+	movl	$HIPE_MODE_SWITCH_RES_CALL_EXPORTED, %eax
 	jmp	.suspend_exit
 
 /*