diff --git a/src/Makefile.dep b/src/Makefile.dep
index 05ef5748..5d1238f6 100644
--- a/src/Makefile.dep
+++ b/src/Makefile.dep
@@ -31,8 +31,8 @@ lib_io.o: lib_io.c lua.h luaconf.h lauxlib.h lualib.h lj_obj.h lj_def.h \
  lj_libdef.h
 lib_jit.o: lib_jit.c lua.h luaconf.h lauxlib.h lualib.h lj_arch.h \
  lj_obj.h lj_def.h lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_bc.h lj_ir.h \
- lj_jit.h lj_iropt.h lj_dispatch.h lj_vm.h lj_vmevent.h lj_lib.h luajit.h \
- lj_libdef.h
+ lj_jit.h lj_iropt.h lj_target.h lj_target_*.h lj_dispatch.h lj_vm.h \
+ lj_vmevent.h lj_lib.h luajit.h lj_libdef.h
 lib_math.o: lib_math.c lua.h luaconf.h lauxlib.h lualib.h lj_obj.h \
  lj_def.h lj_arch.h lj_lib.h lj_libdef.h
 lib_os.o: lib_os.c lua.h luaconf.h lauxlib.h lualib.h lj_obj.h lj_def.h \
diff --git a/src/lib_jit.c b/src/lib_jit.c
index e2fc8067..57837e1f 100644
--- a/src/lib_jit.c
+++ b/src/lib_jit.c
@@ -20,6 +20,7 @@
 #include "lj_ir.h"
 #include "lj_jit.h"
 #include "lj_iropt.h"
+#include "lj_target.h"
 #endif
 #include "lj_dispatch.h"
 #include "lj_vm.h"
@@ -368,16 +369,12 @@ LJLIB_CF(jit_util_tracemc)
 /* local addr = jit.util.traceexitstub(idx) */
 LJLIB_CF(jit_util_traceexitstub)
 {
-#ifdef EXITSTUBS_PER_GROUP
   ExitNo exitno = (ExitNo)lj_lib_checkint(L, 1);
   jit_State *J = L2J(L);
   if (exitno < EXITSTUBS_PER_GROUP*LJ_MAX_EXITSTUBGR) {
     setintptrV(L->top-1, (intptr_t)(void *)exitstub_addr(J, exitno));
     return 1;
   }
-#else
-  UNUSED(L);
-#endif
   return 0;
 }
 
diff --git a/src/lj_arch.h b/src/lj_arch.h
index 3c6d9e9c..978ccd1f 100644
--- a/src/lj_arch.h
+++ b/src/lj_arch.h
@@ -119,6 +119,7 @@
 #define LJ_TARGET_X64		1
 #define LJ_TARGET_X86ORX64	1
 #define LJ_TARGET_EHRETREG	0
+#define LJ_TARGET_JUMPRANGE	31	/* +-2^31 = +-2GB */
 #define LJ_TARGET_MASKSHIFT	1
 #define LJ_TARGET_MASKROT	1
 #define LJ_ARCH_DUALNUM		1
@@ -133,6 +134,7 @@
 #define LJ_ABI_EABI		1
 #define LJ_TARGET_ARM		1
 #define LJ_TARGET_EHRETREG	0
+#define LJ_TARGET_JUMPRANGE	25	/* +-2^25 = +-32MB */
 #define LJ_TARGET_MASKSHIFT	0
 #define LJ_TARGET_MASKROT	1
 #define LJ_ARCH_DUALNUM		2
@@ -153,6 +155,7 @@
 #define LJ_TARGET_PPC		1
 #define LJ_TARGET_PPCSPE	1
 #define LJ_TARGET_EHRETREG	3
+#define LJ_TARGET_JUMPRANGE	25	/* +-2^25 = +-32MB */
 #define LJ_TARGET_MASKSHIFT	0
 #define LJ_TARGET_MASKROT	1
 #define LJ_ARCH_DUALNUM		0
diff --git a/src/lj_asm.c b/src/lj_asm.c
index 8b984d8e..37b7e3f5 100644
--- a/src/lj_asm.c
+++ b/src/lj_asm.c
@@ -1439,10 +1439,10 @@ void lj_asm_trace(jit_State *J, GCtrace *T)
 
   /* Set trace entry point before fixing up tail to allow link to self. */
   T->mcode = as->mcp;
-  T->mcloop = as->mcloop ? (MSize)(as->mcloop - as->mcp) : 0;
+  T->mcloop = as->mcloop ? (MSize)((char *)as->mcloop - (char *)as->mcp) : 0;
   if (!as->loopref)
     asm_tail_fixup(as, T->link);  /* Note: this may change as->mctop! */
-  T->szmcode = (MSize)(as->mctop - as->mcp);
+  T->szmcode = (MSize)((char *)as->mctop - (char *)as->mcp);
   VG_INVALIDATE(T->mcode, T->szmcode);
 }
 
diff --git a/src/lj_jit.h b/src/lj_jit.h
index 1cf63448..dd74dedb 100644
--- a/src/lj_jit.h
+++ b/src/lj_jit.h
@@ -258,13 +258,6 @@ enum {
 #define lj_resetsplit(J)	UNUSED(J)
 #endif
 
-/* Exit stubs. */
-#if LJ_TARGET_X86ORX64
-/* Limited by the range of a short fwd jump (127): (2+2)*(32-1)-2 = 122. */
-#define EXITSTUB_SPACING	(2+2)
-#define EXITSTUBS_PER_GROUP	32
-#endif
-
 /* Fold state is used to fold instructions on-the-fly. */
 typedef struct FoldState {
   IRIns ins;		/* Currently emitted instruction. */
@@ -331,9 +324,7 @@ typedef struct jit_State {
 
   int32_t param[JIT_P__MAX];  /* JIT engine parameters. */
 
-#if LJ_TARGET_X86ORX64
   MCode *exitstubgroup[LJ_MAX_EXITSTUBGR];  /* Exit stub group addresses. */
-#endif
 
   HotPenalty penalty[PENALTY_SLOTS];  /* Penalty slots. */
   uint32_t penaltyslot;	/* Round-robin index into penalty slots. */
@@ -369,14 +360,4 @@ static LJ_AINLINE uint32_t LJ_PRNG_BITS(jit_State *J, int bits)
   return J->prngstate >> (32-bits);
 }
 
-#ifdef EXITSTUBS_PER_GROUP
-/* Return the address of an exit stub. */
-static LJ_AINLINE MCode *exitstub_addr(jit_State *J, ExitNo exitno)
-{
-  lua_assert(J->exitstubgroup[exitno / EXITSTUBS_PER_GROUP] != NULL);
-  return (MCode *)((char *)J->exitstubgroup[exitno / EXITSTUBS_PER_GROUP] +
-		   EXITSTUB_SPACING*(exitno % EXITSTUBS_PER_GROUP));
-}
-#endif
-
 #endif
diff --git a/src/lj_mcode.c b/src/lj_mcode.c
index 8d0f7213..279854f8 100644
--- a/src/lj_mcode.c
+++ b/src/lj_mcode.c
@@ -162,20 +162,7 @@ static void mcode_protect(jit_State *J, int prot)
 #define mcode_validptr(p)	((p) && (uintptr_t)(p) < 0xffff0000)
 #endif
 
-#if LJ_TARGET_X64
-#define MCODE_JUMPRANGE		31
-#elif LJ_TARGET_ARM
-#define MCODE_JUMPRANGE		26
-#else
-#define MCODE_JUMPRANGE		32
-#endif
-
-#if MCODE_JUMPRANGE == 32
-
-/* All 32 bit memory addresses are reachable by relative jumps. */
-#define mcode_alloc(J, sz)	mcode_alloc_at((J), 0, (sz), MCPROT_GEN)
-
-#else
+#ifdef LJ_TARGET_JUMPRANGE
 
 /* Get memory within relative jump distance of our code in 64 bit mode. */
 static void *mcode_alloc(jit_State *J, size_t sz)
@@ -184,7 +171,7 @@ static void *mcode_alloc(jit_State *J, size_t sz)
   ** Try addresses within a distance of target-range/2+1MB..target+range/2-1MB.
   */
   uintptr_t target = (uintptr_t)(void *)lj_vm_exit_handler & ~(uintptr_t)0xffff;
-  const uintptr_t range = (1u << MCODE_JUMPRANGE) - (1u << 21);
+  const uintptr_t range = (1u << LJ_TARGET_JUMPRANGE) - (1u << 21);
   /* First try a contiguous area below the last one. */
   uintptr_t hint = J->mcarea ? (uintptr_t)J->mcarea - sz : 0;
   int i;
@@ -208,6 +195,11 @@ static void *mcode_alloc(jit_State *J, size_t sz)
   return NULL;
 }
 
+#else
+
+/* All memory addresses are reachable by relative jumps. */
+#define mcode_alloc(J, sz)	mcode_alloc_at((J), 0, (sz), MCPROT_GEN)
+
 #endif
 
 /* -- MCode area management ----------------------------------------------- */
diff --git a/src/lj_snap.c b/src/lj_snap.c
index e1791b51..dd70ece1 100644
--- a/src/lj_snap.c
+++ b/src/lj_snap.c
@@ -392,8 +392,10 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr)
 	  o->u32.lo = (uint32_t)*sps;
 	} else if (irt_isinteger(t)) {
 	  setintV(o, *sps);
-	} else if (!LJ_SOFTFP && irt_isnum(t)) {
+#if !LJ_SOFTFP
+	} else if (irt_isnum(t)) {
 	  o->u64 = *(uint64_t *)sps;
+#endif
 #if LJ_64
 	} else if (irt_islightud(t)) {
 	  /* 64 bit lightuserdata which may escape already has the tag bits. */
@@ -411,8 +413,10 @@ const BCIns *lj_snap_restore(jit_State *J, void *exptr)
 	  o->u32.lo = (uint32_t)ex->gpr[r-RID_MIN_GPR];
 	} else if (irt_isinteger(t)) {
 	  setintV(o, (int32_t)ex->gpr[r-RID_MIN_GPR]);
-	} else if (!LJ_SOFTFP && irt_isnum(t)) {
+#if !LJ_SOFTFP
+	} else if (irt_isnum(t)) {
 	  setnumV(o, ex->fpr[r-RID_MIN_FPR]);
+#endif
 #if LJ_64
 	} else if (irt_islightud(t)) {
 	  /* 64 bit lightuserdata which may escape already has the tag bits. */
diff --git a/src/lj_target.h b/src/lj_target.h
index 8c91bf7e..bfa1c9f0 100644
--- a/src/lj_target.h
+++ b/src/lj_target.h
@@ -129,4 +129,12 @@ typedef uint32_t RegCost;
 #error "Missing include for target CPU"
 #endif
 
+/* Return the address of an exit stub. */
+static LJ_AINLINE MCode *exitstub_addr(jit_State *J, ExitNo exitno)
+{
+  lua_assert(J->exitstubgroup[exitno / EXITSTUBS_PER_GROUP] != NULL);
+  return (MCode *)((char *)J->exitstubgroup[exitno / EXITSTUBS_PER_GROUP] +
+		   EXITSTUB_SPACING*(exitno % EXITSTUBS_PER_GROUP));
+}
+
 #endif
diff --git a/src/lj_target_x86.h b/src/lj_target_x86.h
index 5c24761e..fd2a32fe 100644
--- a/src/lj_target_x86.h
+++ b/src/lj_target_x86.h
@@ -147,6 +147,10 @@ typedef struct {
   int32_t spill[256];		/* Spill slots. */
 } ExitState;
 
+/* Limited by the range of a short fwd jump (127): (2+2)*(32-1)-2 = 122. */
+#define EXITSTUB_SPACING	(2+2)
+#define EXITSTUBS_PER_GROUP	32
+
 /* -- x86 ModRM operand encoding ------------------------------------------ */
 
 typedef enum {
diff --git a/src/lj_trace.c b/src/lj_trace.c
index ecfbfed6..7141de46 100644
--- a/src/lj_trace.c
+++ b/src/lj_trace.c
@@ -278,9 +278,7 @@ int lj_trace_flushall(lua_State *L)
   memset(J->penalty, 0, sizeof(J->penalty));
   /* Free the whole machine code and invalidate all exit stub groups. */
   lj_mcode_free(J);
-#ifdef EXITSTUBS_PER_GROUP
   memset(J->exitstubgroup, 0, sizeof(J->exitstubgroup));
-#endif
   lj_vmevent_send(L, TRACE,
     setstrV(L, L->top++, lj_str_newlit(L, "flush"));
   );
@@ -685,15 +683,58 @@ static TValue *trace_exit_cp(lua_State *L, lua_CFunction dummy, void *ud)
   return NULL;
 }
 
+#ifndef LUAJIT_DISABLE_VMEVENT
+/* Push all registers from exit state. */
+static void trace_exit_regs(lua_State *L, ExitState *ex)
+{
+  int32_t i;
+  setintV(L->top++, RID_NUM_GPR);
+  setintV(L->top++, RID_NUM_FPR);
+  for (i = 0; i < RID_NUM_GPR; i++) {
+    if (sizeof(ex->gpr[i]) == sizeof(int32_t))
+      setintV(L->top++, (int32_t)ex->gpr[i]);
+    else
+      setnumV(L->top++, (lua_Number)ex->gpr[i]);
+  }
+#if !LJ_SOFTFP
+  for (i = 0; i < RID_NUM_FPR; i++) {
+    setnumV(L->top, ex->fpr[i]);
+    if (LJ_UNLIKELY(tvisnan(L->top)))
+      setnanV(L->top);
+    L->top++;
+  }
+#endif
+}
+#endif
+
+#ifdef EXITSTATE_PCREG
+/* Determine trace number from pc of exit instruction. */
+static TraceNo trace_exit_find(jit_State *J, MCode *pc)
+{
+  TraceNo traceno;
+  for (traceno = 1; traceno < J->sizetrace; traceno++) {
+    GCtrace *T = traceref(J, traceno);
+    if (T && pc >= T->mcode && pc < (MCode *)((char *)T->mcode + T->szmcode))
+      return traceno;
+  }
+  lua_assert(0);
+  return 0;
+}
+#endif
+
 /* A trace exited. Restore interpreter state. */
 int LJ_FASTCALL lj_trace_exit(jit_State *J, void *exptr)
 {
   ERRNO_SAVE
   lua_State *L = J->L;
+  ExitState *ex = (ExitState *)exptr;
   ExitDataCP exd;
   int errcode;
   const BCIns *pc;
   void *cf;
+#ifdef EXITSTATE_PCREG
+  J->parent = trace_exit_find(J, (MCode *)(intptr_t)ex->gpr[EXITSTATE_PCREG]);
+#endif
   exd.J = J;
   exd.exptr = exptr;
   errcode = lj_vm_cpcall(L, NULL, &exd, trace_exit_cp);
@@ -701,25 +742,10 @@ int LJ_FASTCALL lj_trace_exit(jit_State *J, void *exptr)
     return -errcode;  /* Return negated error code. */
 
   lj_vmevent_send(L, TEXIT,
-    ExitState *ex = (ExitState *)exptr;
-    int32_t i;
     lj_state_checkstack(L, 4+RID_NUM_GPR+RID_NUM_FPR+LUA_MINSTACK);
     setintV(L->top++, J->parent);
     setintV(L->top++, J->exitno);
-    setintV(L->top++, RID_NUM_GPR);
-    setintV(L->top++, RID_NUM_FPR);
-    for (i = 0; i < RID_NUM_GPR; i++) {
-      if (sizeof(ex->gpr[i]) == sizeof(int32_t))
-	setintV(L->top++, (int32_t)ex->gpr[i]);
-      else
-	setnumV(L->top++, (lua_Number)ex->gpr[i]);
-    }
-    for (i = 0; i < RID_NUM_FPR; i++) {
-      setnumV(L->top, ex->fpr[i]);
-      if (LJ_UNLIKELY(tvisnan(L->top)))
-	setnanV(L->top);
-      L->top++;
-    }
+    trace_exit_regs(L, ex);
   );
 
   pc = exd.pc;