diff --git a/configure b/configure
index 9698e6d..4ac9904 100755
--- a/configure
+++ b/configure
@@ -3874,7 +3874,7 @@
 
 
 case "$target_arch2" in
-  alpha | arm* | cris | i386 | lm32 | m68k | microblaze* | or32 | s390x | sparc* | unicore32 | x86_64 | xtensa* | ppc*)
+  alpha | arm* | cris | i386 | lm32 | m68k | microblaze* | or32 | s390x | sh4* | sparc* | unicore32 | x86_64 | xtensa* | ppc*)
     echo "CONFIG_TCG_PASS_AREG0=y" >> $config_target_mak
   ;;
 esac
diff --git a/target-sh4/Makefile.objs b/target-sh4/Makefile.objs
index 2e0e093..ca20f21 100644
--- a/target-sh4/Makefile.objs
+++ b/target-sh4/Makefile.objs
@@ -1,4 +1,2 @@
 obj-y += translate.o op_helper.o helper.o cpu.o
 obj-$(CONFIG_SOFTMMU) += machine.o
-
-$(obj)/op_helper.o: QEMU_CFLAGS += $(HELPER_CFLAGS)
diff --git a/target-sh4/helper.h b/target-sh4/helper.h
index 95e3c7c..6e4f108 100644
--- a/target-sh4/helper.h
+++ b/target-sh4/helper.h
@@ -1,54 +1,54 @@
 #include "def-helper.h"
 
-DEF_HELPER_0(ldtlb, void)
-DEF_HELPER_0(raise_illegal_instruction, void)
-DEF_HELPER_0(raise_slot_illegal_instruction, void)
-DEF_HELPER_0(raise_fpu_disable, void)
-DEF_HELPER_0(raise_slot_fpu_disable, void)
-DEF_HELPER_0(debug, void)
-DEF_HELPER_1(sleep, void, i32)
-DEF_HELPER_1(trapa, void, i32)
+DEF_HELPER_1(ldtlb, void, env)
+DEF_HELPER_1(raise_illegal_instruction, void, env)
+DEF_HELPER_1(raise_slot_illegal_instruction, void, env)
+DEF_HELPER_1(raise_fpu_disable, void, env)
+DEF_HELPER_1(raise_slot_fpu_disable, void, env)
+DEF_HELPER_1(debug, void, env)
+DEF_HELPER_2(sleep, void, env, i32)
+DEF_HELPER_2(trapa, void, env, i32)
 
-DEF_HELPER_2(movcal, void, i32, i32)
-DEF_HELPER_0(discard_movcal_backup, void)
-DEF_HELPER_1(ocbi, void, i32)
+DEF_HELPER_3(movcal, void, env, i32, i32)
+DEF_HELPER_1(discard_movcal_backup, void, env)
+DEF_HELPER_2(ocbi, void, env, i32)
 
-DEF_HELPER_2(addv, i32, i32, i32)
-DEF_HELPER_2(addc, i32, i32, i32)
-DEF_HELPER_2(subv, i32, i32, i32)
-DEF_HELPER_2(subc, i32, i32, i32)
-DEF_HELPER_2(div1, i32, i32, i32)
-DEF_HELPER_2(macl, void, i32, i32)
-DEF_HELPER_2(macw, void, i32, i32)
+DEF_HELPER_3(addv, i32, env, i32, i32)
+DEF_HELPER_3(addc, i32, env, i32, i32)
+DEF_HELPER_3(subv, i32, env, i32, i32)
+DEF_HELPER_3(subc, i32, env, i32, i32)
+DEF_HELPER_3(div1, i32, env, i32, i32)
+DEF_HELPER_3(macl, void, env, i32, i32)
+DEF_HELPER_3(macw, void, env, i32, i32)
 
-DEF_HELPER_1(ld_fpscr, void, i32)
+DEF_HELPER_2(ld_fpscr, void, env, i32)
 
 DEF_HELPER_1(fabs_FT, f32, f32)
 DEF_HELPER_1(fabs_DT, f64, f64)
-DEF_HELPER_2(fadd_FT, f32, f32, f32)
-DEF_HELPER_2(fadd_DT, f64, f64, f64)
-DEF_HELPER_1(fcnvsd_FT_DT, f64, f32)
-DEF_HELPER_1(fcnvds_DT_FT, f32, f64)
+DEF_HELPER_3(fadd_FT, f32, env, f32, f32)
+DEF_HELPER_3(fadd_DT, f64, env, f64, f64)
+DEF_HELPER_2(fcnvsd_FT_DT, f64, env, f32)
+DEF_HELPER_2(fcnvds_DT_FT, f32, env, f64)
 
-DEF_HELPER_2(fcmp_eq_FT, void, f32, f32)
-DEF_HELPER_2(fcmp_eq_DT, void, f64, f64)
-DEF_HELPER_2(fcmp_gt_FT, void, f32, f32)
-DEF_HELPER_2(fcmp_gt_DT, void, f64, f64)
-DEF_HELPER_2(fdiv_FT, f32, f32, f32)
-DEF_HELPER_2(fdiv_DT, f64, f64, f64)
-DEF_HELPER_1(float_FT, f32, i32)
-DEF_HELPER_1(float_DT, f64, i32)
-DEF_HELPER_3(fmac_FT, f32, f32, f32, f32)
-DEF_HELPER_2(fmul_FT, f32, f32, f32)
-DEF_HELPER_2(fmul_DT, f64, f64, f64)
+DEF_HELPER_3(fcmp_eq_FT, void, env, f32, f32)
+DEF_HELPER_3(fcmp_eq_DT, void, env, f64, f64)
+DEF_HELPER_3(fcmp_gt_FT, void, env, f32, f32)
+DEF_HELPER_3(fcmp_gt_DT, void, env, f64, f64)
+DEF_HELPER_3(fdiv_FT, f32, env, f32, f32)
+DEF_HELPER_3(fdiv_DT, f64, env, f64, f64)
+DEF_HELPER_2(float_FT, f32, env, i32)
+DEF_HELPER_2(float_DT, f64, env, i32)
+DEF_HELPER_4(fmac_FT, f32, env, f32, f32, f32)
+DEF_HELPER_3(fmul_FT, f32, env, f32, f32)
+DEF_HELPER_3(fmul_DT, f64, env, f64, f64)
 DEF_HELPER_1(fneg_T, f32, f32)
-DEF_HELPER_2(fsub_FT, f32, f32, f32)
-DEF_HELPER_2(fsub_DT, f64, f64, f64)
-DEF_HELPER_1(fsqrt_FT, f32, f32)
-DEF_HELPER_1(fsqrt_DT, f64, f64)
-DEF_HELPER_1(ftrc_FT, i32, f32)
-DEF_HELPER_1(ftrc_DT, i32, f64)
-DEF_HELPER_2(fipr, void, i32, i32)
-DEF_HELPER_1(ftrv, void, i32)
+DEF_HELPER_3(fsub_FT, f32, env, f32, f32)
+DEF_HELPER_3(fsub_DT, f64, env, f64, f64)
+DEF_HELPER_2(fsqrt_FT, f32, env, f32)
+DEF_HELPER_2(fsqrt_DT, f64, env, f64)
+DEF_HELPER_2(ftrc_FT, i32, env, f32)
+DEF_HELPER_2(ftrc_DT, i32, env, f64)
+DEF_HELPER_3(fipr, void, env, i32, i32)
+DEF_HELPER_2(ftrv, void, env, i32)
 
 #include "def-helper.h"
diff --git a/target-sh4/op_helper.c b/target-sh4/op_helper.c
index 4054791..9b4328d 100644
--- a/target-sh4/op_helper.c
+++ b/target-sh4/op_helper.c
@@ -19,10 +19,9 @@
 #include <assert.h>
 #include <stdlib.h>
 #include "cpu.h"
-#include "dyngen-exec.h"
 #include "helper.h"
 
-static void cpu_restore_state_from_retaddr(uintptr_t retaddr)
+static void cpu_restore_state_from_retaddr(CPUSH4State *env, uintptr_t retaddr)
 {
     TranslationBlock *tb;
 
@@ -53,26 +52,22 @@
 #define SHIFT 3
 #include "softmmu_template.h"
 
-void tlb_fill(CPUSH4State *env1, target_ulong addr, int is_write, int mmu_idx,
+void tlb_fill(CPUSH4State *env, target_ulong addr, int is_write, int mmu_idx,
               uintptr_t retaddr)
 {
-    CPUSH4State *saved_env;
     int ret;
 
-    saved_env = env;
-    env = env1;
     ret = cpu_sh4_handle_mmu_fault(env, addr, is_write, mmu_idx);
     if (ret) {
         /* now we have a real cpu fault */
-        cpu_restore_state_from_retaddr(retaddr);
+        cpu_restore_state_from_retaddr(env, retaddr);
         cpu_loop_exit(env);
     }
-    env = saved_env;
 }
 
 #endif
 
-void helper_ldtlb(void)
+void helper_ldtlb(CPUSH4State *env)
 {
 #ifdef CONFIG_USER_ONLY
     /* XXXXX */
@@ -82,40 +77,41 @@
 #endif
 }
 
-static inline void raise_exception(int index, uintptr_t retaddr)
+static inline void raise_exception(CPUSH4State *env, int index,
+                                   uintptr_t retaddr)
 {
     env->exception_index = index;
-    cpu_restore_state_from_retaddr(retaddr);
+    cpu_restore_state_from_retaddr(env, retaddr);
     cpu_loop_exit(env);
 }
 
-void helper_raise_illegal_instruction(void)
+void helper_raise_illegal_instruction(CPUSH4State *env)
 {
-    raise_exception(0x180, GETPC());
+    raise_exception(env, 0x180, GETPC());
 }
 
-void helper_raise_slot_illegal_instruction(void)
+void helper_raise_slot_illegal_instruction(CPUSH4State *env)
 {
-    raise_exception(0x1a0, GETPC());
+    raise_exception(env, 0x1a0, GETPC());
 }
 
-void helper_raise_fpu_disable(void)
+void helper_raise_fpu_disable(CPUSH4State *env)
 {
-    raise_exception(0x800, GETPC());
+    raise_exception(env, 0x800, GETPC());
 }
 
-void helper_raise_slot_fpu_disable(void)
+void helper_raise_slot_fpu_disable(CPUSH4State *env)
 {
-    raise_exception(0x820, GETPC());
+    raise_exception(env, 0x820, GETPC());
 }
 
-void helper_debug(void)
+void helper_debug(CPUSH4State *env)
 {
     env->exception_index = EXCP_DEBUG;
     cpu_loop_exit(env);
 }
 
-void helper_sleep(uint32_t next_pc)
+void helper_sleep(CPUSH4State *env, uint32_t next_pc)
 {
     env->halted = 1;
     env->in_sleep = 1;
@@ -124,13 +120,13 @@
     cpu_loop_exit(env);
 }
 
-void helper_trapa(uint32_t tra)
+void helper_trapa(CPUSH4State *env, uint32_t tra)
 {
     env->tra = tra << 2;
-    raise_exception(0x160, GETPC());
+    raise_exception(env, 0x160, GETPC());
 }
 
-void helper_movcal(uint32_t address, uint32_t value)
+void helper_movcal(CPUSH4State *env, uint32_t address, uint32_t value)
 {
     if (cpu_sh4_is_cached (env, address))
     {
@@ -144,7 +140,7 @@
     }
 }
 
-void helper_discard_movcal_backup(void)
+void helper_discard_movcal_backup(CPUSH4State *env)
 {
     memory_content *current = env->movcal_backup;
 
@@ -158,7 +154,7 @@
     } 
 }
 
-void helper_ocbi(uint32_t address)
+void helper_ocbi(CPUSH4State *env, uint32_t address)
 {
     memory_content **current = &(env->movcal_backup);
     while (*current)
@@ -167,7 +163,7 @@
 	if ((a & ~0x1F) == (address & ~0x1F))
 	{
 	    memory_content *next = (*current)->next;
-	    stl(a, (*current)->value);
+            cpu_stl_data(env, a, (*current)->value);
 	    
 	    if (next == NULL)
 	    {
@@ -181,7 +177,7 @@
     }
 }
 
-uint32_t helper_addc(uint32_t arg0, uint32_t arg1)
+uint32_t helper_addc(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
 {
     uint32_t tmp0, tmp1;
 
@@ -197,7 +193,7 @@
     return arg1;
 }
 
-uint32_t helper_addv(uint32_t arg0, uint32_t arg1)
+uint32_t helper_addv(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
 {
     uint32_t dest, src, ans;
 
@@ -236,7 +232,7 @@
 #define SETM env->sr |= SR_M
 #define CLRM env->sr &= ~SR_M
 
-uint32_t helper_div1(uint32_t arg0, uint32_t arg1)
+uint32_t helper_div1(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
 {
     uint32_t tmp0, tmp2;
     uint8_t old_q, tmp1 = 0xff;
@@ -344,7 +340,7 @@
     return arg1;
 }
 
-void helper_macl(uint32_t arg0, uint32_t arg1)
+void helper_macl(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
 {
     int64_t res;
 
@@ -360,7 +356,7 @@
     }
 }
 
-void helper_macw(uint32_t arg0, uint32_t arg1)
+void helper_macw(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
 {
     int64_t res;
 
@@ -379,7 +375,7 @@
     }
 }
 
-uint32_t helper_subc(uint32_t arg0, uint32_t arg1)
+uint32_t helper_subc(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
 {
     uint32_t tmp0, tmp1;
 
@@ -395,7 +391,7 @@
     return arg1;
 }
 
-uint32_t helper_subv(uint32_t arg0, uint32_t arg1)
+uint32_t helper_subv(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
 {
     int32_t dest, src, ans;
 
@@ -424,17 +420,17 @@
     return arg1;
 }
 
-static inline void set_t(void)
+static inline void set_t(CPUSH4State *env)
 {
     env->sr |= SR_T;
 }
 
-static inline void clr_t(void)
+static inline void clr_t(CPUSH4State *env)
 {
     env->sr &= ~SR_T;
 }
 
-void helper_ld_fpscr(uint32_t val)
+void helper_ld_fpscr(CPUSH4State *env, uint32_t val)
 {
     env->fpscr = val & FPSCR_MASK;
     if ((val & FPSCR_RM_MASK) == FPSCR_RM_ZERO) {
@@ -445,7 +441,7 @@
     set_flush_to_zero((val & FPSCR_DN) != 0, &env->fp_status);
 }
 
-static void update_fpscr(uintptr_t retaddr)
+static void update_fpscr(CPUSH4State *env, uintptr_t retaddr)
 {
     int xcpt, cause, enable;
 
@@ -479,7 +475,7 @@
         cause = (env->fpscr & FPSCR_CAUSE_MASK) >> FPSCR_CAUSE_SHIFT;
         enable = (env->fpscr & FPSCR_ENABLE_MASK) >> FPSCR_ENABLE_SHIFT;
         if (cause & enable) {
-            cpu_restore_state_from_retaddr(retaddr);
+            cpu_restore_state_from_retaddr(env, retaddr);
             env->exception_index = 0x120;
             cpu_loop_exit(env);
         }
@@ -496,156 +492,156 @@
     return float64_abs(t0);
 }
 
-float32 helper_fadd_FT(float32 t0, float32 t1)
+float32 helper_fadd_FT(CPUSH4State *env, float32 t0, float32 t1)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float32_add(t0, t1, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
-float64 helper_fadd_DT(float64 t0, float64 t1)
+float64 helper_fadd_DT(CPUSH4State *env, float64 t0, float64 t1)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float64_add(t0, t1, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
-void helper_fcmp_eq_FT(float32 t0, float32 t1)
+void helper_fcmp_eq_FT(CPUSH4State *env, float32 t0, float32 t1)
 {
     int relation;
 
     set_float_exception_flags(0, &env->fp_status);
     relation = float32_compare(t0, t1, &env->fp_status);
     if (unlikely(relation == float_relation_unordered)) {
-        update_fpscr(GETPC());
+        update_fpscr(env, GETPC());
     } else if (relation == float_relation_equal) {
-	set_t();
+        set_t(env);
     } else {
-	clr_t();
+        clr_t(env);
     }
 }
 
-void helper_fcmp_eq_DT(float64 t0, float64 t1)
+void helper_fcmp_eq_DT(CPUSH4State *env, float64 t0, float64 t1)
 {
     int relation;
 
     set_float_exception_flags(0, &env->fp_status);
     relation = float64_compare(t0, t1, &env->fp_status);
     if (unlikely(relation == float_relation_unordered)) {
-        update_fpscr(GETPC());
+        update_fpscr(env, GETPC());
     } else if (relation == float_relation_equal) {
-	set_t();
+        set_t(env);
     } else {
-	clr_t();
+        clr_t(env);
     }
 }
 
-void helper_fcmp_gt_FT(float32 t0, float32 t1)
+void helper_fcmp_gt_FT(CPUSH4State *env, float32 t0, float32 t1)
 {
     int relation;
 
     set_float_exception_flags(0, &env->fp_status);
     relation = float32_compare(t0, t1, &env->fp_status);
     if (unlikely(relation == float_relation_unordered)) {
-        update_fpscr(GETPC());
+        update_fpscr(env, GETPC());
     } else if (relation == float_relation_greater) {
-	set_t();
+        set_t(env);
     } else {
-	clr_t();
+        clr_t(env);
     }
 }
 
-void helper_fcmp_gt_DT(float64 t0, float64 t1)
+void helper_fcmp_gt_DT(CPUSH4State *env, float64 t0, float64 t1)
 {
     int relation;
 
     set_float_exception_flags(0, &env->fp_status);
     relation = float64_compare(t0, t1, &env->fp_status);
     if (unlikely(relation == float_relation_unordered)) {
-        update_fpscr(GETPC());
+        update_fpscr(env, GETPC());
     } else if (relation == float_relation_greater) {
-	set_t();
+        set_t(env);
     } else {
-	clr_t();
+        clr_t(env);
     }
 }
 
-float64 helper_fcnvsd_FT_DT(float32 t0)
+float64 helper_fcnvsd_FT_DT(CPUSH4State *env, float32 t0)
 {
     float64 ret;
     set_float_exception_flags(0, &env->fp_status);
     ret = float32_to_float64(t0, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return ret;
 }
 
-float32 helper_fcnvds_DT_FT(float64 t0)
+float32 helper_fcnvds_DT_FT(CPUSH4State *env, float64 t0)
 {
     float32 ret;
     set_float_exception_flags(0, &env->fp_status);
     ret = float64_to_float32(t0, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return ret;
 }
 
-float32 helper_fdiv_FT(float32 t0, float32 t1)
+float32 helper_fdiv_FT(CPUSH4State *env, float32 t0, float32 t1)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float32_div(t0, t1, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
-float64 helper_fdiv_DT(float64 t0, float64 t1)
+float64 helper_fdiv_DT(CPUSH4State *env, float64 t0, float64 t1)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float64_div(t0, t1, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
-float32 helper_float_FT(uint32_t t0)
+float32 helper_float_FT(CPUSH4State *env, uint32_t t0)
 {
     float32 ret;
     set_float_exception_flags(0, &env->fp_status);
     ret = int32_to_float32(t0, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return ret;
 }
 
-float64 helper_float_DT(uint32_t t0)
+float64 helper_float_DT(CPUSH4State *env, uint32_t t0)
 {
     float64 ret;
     set_float_exception_flags(0, &env->fp_status);
     ret = int32_to_float64(t0, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return ret;
 }
 
-float32 helper_fmac_FT(float32 t0, float32 t1, float32 t2)
+float32 helper_fmac_FT(CPUSH4State *env, float32 t0, float32 t1, float32 t2)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float32_mul(t0, t1, &env->fp_status);
     t0 = float32_add(t0, t2, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
-float32 helper_fmul_FT(float32 t0, float32 t1)
+float32 helper_fmul_FT(CPUSH4State *env, float32 t0, float32 t1)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float32_mul(t0, t1, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
-float64 helper_fmul_DT(float64 t0, float64 t1)
+float64 helper_fmul_DT(CPUSH4State *env, float64 t0, float64 t1)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float64_mul(t0, t1, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
@@ -654,57 +650,57 @@
     return float32_chs(t0);
 }
 
-float32 helper_fsqrt_FT(float32 t0)
+float32 helper_fsqrt_FT(CPUSH4State *env, float32 t0)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float32_sqrt(t0, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
-float64 helper_fsqrt_DT(float64 t0)
+float64 helper_fsqrt_DT(CPUSH4State *env, float64 t0)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float64_sqrt(t0, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
-float32 helper_fsub_FT(float32 t0, float32 t1)
+float32 helper_fsub_FT(CPUSH4State *env, float32 t0, float32 t1)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float32_sub(t0, t1, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
-float64 helper_fsub_DT(float64 t0, float64 t1)
+float64 helper_fsub_DT(CPUSH4State *env, float64 t0, float64 t1)
 {
     set_float_exception_flags(0, &env->fp_status);
     t0 = float64_sub(t0, t1, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return t0;
 }
 
-uint32_t helper_ftrc_FT(float32 t0)
+uint32_t helper_ftrc_FT(CPUSH4State *env, float32 t0)
 {
     uint32_t ret;
     set_float_exception_flags(0, &env->fp_status);
     ret = float32_to_int32_round_to_zero(t0, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return ret;
 }
 
-uint32_t helper_ftrc_DT(float64 t0)
+uint32_t helper_ftrc_DT(CPUSH4State *env, float64 t0)
 {
     uint32_t ret;
     set_float_exception_flags(0, &env->fp_status);
     ret = float64_to_int32_round_to_zero(t0, &env->fp_status);
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
     return ret;
 }
 
-void helper_fipr(uint32_t m, uint32_t n)
+void helper_fipr(CPUSH4State *env, uint32_t m, uint32_t n)
 {
     int bank, i;
     float32 r, p;
@@ -719,12 +715,12 @@
                         &env->fp_status);
         r = float32_add(r, p, &env->fp_status);
     }
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
 
     env->fregs[bank + n + 3] = r;
 }
 
-void helper_ftrv(uint32_t n)
+void helper_ftrv(CPUSH4State *env, uint32_t n)
 {
     int bank_matrix, bank_vector;
     int i, j;
@@ -743,7 +739,7 @@
             r[i] = float32_add(r[i], p, &env->fp_status);
         }
     }
-    update_fpscr(GETPC());
+    update_fpscr(env, GETPC());
 
     for (i = 0 ; i < 4 ; i++) {
         env->fregs[bank_vector + i] = r[i];
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index 6532ad2..d05c74c 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -276,7 +276,7 @@
     } else {
         tcg_gen_movi_i32(cpu_pc, dest);
         if (ctx->singlestep_enabled)
-            gen_helper_debug();
+            gen_helper_debug(cpu_env);
         tcg_gen_exit_tb(0);
     }
 }
@@ -288,7 +288,7 @@
 	   delayed jump as immediate jump are conditinal jumps */
 	tcg_gen_mov_i32(cpu_pc, cpu_delayed_pc);
 	if (ctx->singlestep_enabled)
-	    gen_helper_debug();
+            gen_helper_debug(cpu_env);
 	tcg_gen_exit_tb(0);
     } else {
 	gen_goto_tb(ctx, 0, ctx->delayed_pc);
@@ -437,7 +437,7 @@
 #define CHECK_NOT_DELAY_SLOT \
   if (ctx->flags & (DELAY_SLOT | DELAY_SLOT_CONDITIONAL))     \
   {                                                           \
-      gen_helper_raise_slot_illegal_instruction();            \
+      gen_helper_raise_slot_illegal_instruction(cpu_env);     \
       ctx->bstate = BS_EXCP;                                  \
       return;                                                 \
   }
@@ -445,9 +445,9 @@
 #define CHECK_PRIVILEGED                                        \
   if (IS_USER(ctx)) {                                           \
       if (ctx->flags & (DELAY_SLOT | DELAY_SLOT_CONDITIONAL)) { \
-         gen_helper_raise_slot_illegal_instruction();           \
+          gen_helper_raise_slot_illegal_instruction(cpu_env);   \
       } else {                                                  \
-         gen_helper_raise_illegal_instruction();                \
+          gen_helper_raise_illegal_instruction(cpu_env);        \
       }                                                         \
       ctx->bstate = BS_EXCP;                                    \
       return;                                                   \
@@ -456,9 +456,9 @@
 #define CHECK_FPU_ENABLED                                       \
   if (ctx->flags & SR_FD) {                                     \
       if (ctx->flags & (DELAY_SLOT | DELAY_SLOT_CONDITIONAL)) { \
-          gen_helper_raise_slot_fpu_disable();                  \
+          gen_helper_raise_slot_fpu_disable(cpu_env);           \
       } else {                                                  \
-          gen_helper_raise_fpu_disable();                       \
+          gen_helper_raise_fpu_disable(cpu_env);                \
       }                                                         \
       ctx->bstate = BS_EXCP;                                    \
       return;                                                   \
@@ -492,7 +492,7 @@
 	  if (opcode != 0x0093 /* ocbi */
 	      && opcode != 0x00c3 /* movca.l */)
 	      {
-		  gen_helper_discard_movcal_backup ();
+                  gen_helper_discard_movcal_backup(cpu_env);
 		  ctx->has_movcal = 0;
 	      }
 	}
@@ -523,7 +523,7 @@
 	return;
     case 0x0038:		/* ldtlb */
 	CHECK_PRIVILEGED
-	gen_helper_ldtlb();
+        gen_helper_ldtlb(cpu_env);
 	return;
     case 0x002b:		/* rte */
 	CHECK_PRIVILEGED
@@ -551,7 +551,7 @@
 	return;
     case 0x001b:		/* sleep */
 	CHECK_PRIVILEGED
-	gen_helper_sleep(tcg_const_i32(ctx->pc + 2));
+        gen_helper_sleep(cpu_env, tcg_const_i32(ctx->pc + 2));
 	return;
     }
 
@@ -761,10 +761,10 @@
 	tcg_gen_add_i32(REG(B11_8), REG(B11_8), REG(B7_4));
 	return;
     case 0x300e:		/* addc Rm,Rn */
-	gen_helper_addc(REG(B11_8), REG(B7_4), REG(B11_8));
+        gen_helper_addc(REG(B11_8), cpu_env, REG(B7_4), REG(B11_8));
 	return;
     case 0x300f:		/* addv Rm,Rn */
-	gen_helper_addv(REG(B11_8), REG(B7_4), REG(B11_8));
+        gen_helper_addv(REG(B11_8), cpu_env, REG(B7_4), REG(B11_8));
 	return;
     case 0x2009:		/* and Rm,Rn */
 	tcg_gen_and_i32(REG(B11_8), REG(B11_8), REG(B7_4));
@@ -817,7 +817,7 @@
 	}
 	return;
     case 0x3004:		/* div1 Rm,Rn */
-	gen_helper_div1(REG(B11_8), REG(B7_4), REG(B11_8));
+        gen_helper_div1(REG(B11_8), cpu_env, REG(B7_4), REG(B11_8));
 	return;
     case 0x300d:		/* dmuls.l Rm,Rn */
 	{
@@ -870,7 +870,7 @@
 	    tcg_gen_qemu_ld32s(arg0, REG(B7_4), ctx->memidx);
 	    arg1 = tcg_temp_new();
 	    tcg_gen_qemu_ld32s(arg1, REG(B11_8), ctx->memidx);
-	    gen_helper_macl(arg0, arg1);
+            gen_helper_macl(cpu_env, arg0, arg1);
 	    tcg_temp_free(arg1);
 	    tcg_temp_free(arg0);
 	    tcg_gen_addi_i32(REG(B7_4), REG(B7_4), 4);
@@ -884,7 +884,7 @@
 	    tcg_gen_qemu_ld32s(arg0, REG(B7_4), ctx->memidx);
 	    arg1 = tcg_temp_new();
 	    tcg_gen_qemu_ld32s(arg1, REG(B11_8), ctx->memidx);
-	    gen_helper_macw(arg0, arg1);
+            gen_helper_macw(cpu_env, arg0, arg1);
 	    tcg_temp_free(arg1);
 	    tcg_temp_free(arg0);
 	    tcg_gen_addi_i32(REG(B11_8), REG(B11_8), 2);
@@ -1013,10 +1013,10 @@
 	tcg_gen_sub_i32(REG(B11_8), REG(B11_8), REG(B7_4));
 	return;
     case 0x300a:		/* subc Rm,Rn */
-	gen_helper_subc(REG(B11_8), REG(B7_4), REG(B11_8));
+        gen_helper_subc(REG(B11_8), cpu_env, REG(B7_4), REG(B11_8));
 	return;
     case 0x300b:		/* subv Rm,Rn */
-	gen_helper_subv(REG(B11_8), REG(B7_4), REG(B11_8));
+        gen_helper_subv(REG(B11_8), cpu_env, REG(B7_4), REG(B11_8));
 	return;
     case 0x2008:		/* tst Rm,Rn */
 	{
@@ -1152,22 +1152,22 @@
 		gen_load_fpr64(fp1, DREG(B7_4));
                 switch (ctx->opcode & 0xf00f) {
                 case 0xf000:		/* fadd Rm,Rn */
-                    gen_helper_fadd_DT(fp0, fp0, fp1);
+                    gen_helper_fadd_DT(fp0, cpu_env, fp0, fp1);
                     break;
                 case 0xf001:		/* fsub Rm,Rn */
-                    gen_helper_fsub_DT(fp0, fp0, fp1);
+                    gen_helper_fsub_DT(fp0, cpu_env, fp0, fp1);
                     break;
                 case 0xf002:		/* fmul Rm,Rn */
-                    gen_helper_fmul_DT(fp0, fp0, fp1);
+                    gen_helper_fmul_DT(fp0, cpu_env, fp0, fp1);
                     break;
                 case 0xf003:		/* fdiv Rm,Rn */
-                    gen_helper_fdiv_DT(fp0, fp0, fp1);
+                    gen_helper_fdiv_DT(fp0, cpu_env, fp0, fp1);
                     break;
                 case 0xf004:		/* fcmp/eq Rm,Rn */
-                    gen_helper_fcmp_eq_DT(fp0, fp1);
+                    gen_helper_fcmp_eq_DT(cpu_env, fp0, fp1);
                     return;
                 case 0xf005:		/* fcmp/gt Rm,Rn */
-                    gen_helper_fcmp_gt_DT(fp0, fp1);
+                    gen_helper_fcmp_gt_DT(cpu_env, fp0, fp1);
                     return;
                 }
 		gen_store_fpr64(fp0, DREG(B11_8));
@@ -1176,22 +1176,32 @@
 	    } else {
                 switch (ctx->opcode & 0xf00f) {
                 case 0xf000:		/* fadd Rm,Rn */
-                    gen_helper_fadd_FT(cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B7_4)]);
+                    gen_helper_fadd_FT(cpu_fregs[FREG(B11_8)], cpu_env,
+                                       cpu_fregs[FREG(B11_8)],
+                                       cpu_fregs[FREG(B7_4)]);
                     break;
                 case 0xf001:		/* fsub Rm,Rn */
-                    gen_helper_fsub_FT(cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B7_4)]);
+                    gen_helper_fsub_FT(cpu_fregs[FREG(B11_8)], cpu_env,
+                                       cpu_fregs[FREG(B11_8)],
+                                       cpu_fregs[FREG(B7_4)]);
                     break;
                 case 0xf002:		/* fmul Rm,Rn */
-                    gen_helper_fmul_FT(cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B7_4)]);
+                    gen_helper_fmul_FT(cpu_fregs[FREG(B11_8)], cpu_env,
+                                       cpu_fregs[FREG(B11_8)],
+                                       cpu_fregs[FREG(B7_4)]);
                     break;
                 case 0xf003:		/* fdiv Rm,Rn */
-                    gen_helper_fdiv_FT(cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B7_4)]);
+                    gen_helper_fdiv_FT(cpu_fregs[FREG(B11_8)], cpu_env,
+                                       cpu_fregs[FREG(B11_8)],
+                                       cpu_fregs[FREG(B7_4)]);
                     break;
                 case 0xf004:		/* fcmp/eq Rm,Rn */
-                    gen_helper_fcmp_eq_FT(cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B7_4)]);
+                    gen_helper_fcmp_eq_FT(cpu_env, cpu_fregs[FREG(B11_8)],
+                                          cpu_fregs[FREG(B7_4)]);
                     return;
                 case 0xf005:		/* fcmp/gt Rm,Rn */
-                    gen_helper_fcmp_gt_FT(cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B7_4)]);
+                    gen_helper_fcmp_gt_FT(cpu_env, cpu_fregs[FREG(B11_8)],
+                                          cpu_fregs[FREG(B7_4)]);
                     return;
                 }
 	    }
@@ -1203,8 +1213,9 @@
             if (ctx->fpscr & FPSCR_PR) {
                 break; /* illegal instruction */
             } else {
-                gen_helper_fmac_FT(cpu_fregs[FREG(B11_8)],
-                                   cpu_fregs[FREG(0)], cpu_fregs[FREG(B7_4)], cpu_fregs[FREG(B11_8)]);
+                gen_helper_fmac_FT(cpu_fregs[FREG(B11_8)], cpu_env,
+                                   cpu_fregs[FREG(0)], cpu_fregs[FREG(B7_4)],
+                                   cpu_fregs[FREG(B11_8)]);
                 return;
             }
         }
@@ -1356,7 +1367,7 @@
 	    TCGv imm;
 	    CHECK_NOT_DELAY_SLOT
 	    imm = tcg_const_i32(B7_0);
-	    gen_helper_trapa(imm);
+            gen_helper_trapa(cpu_env, imm);
 	    tcg_temp_free(imm);
 	    ctx->bstate = BS_BRANCH;
 	}
@@ -1531,7 +1542,7 @@
 	LDST(fpul, 0x405a, 0x4056, 0x005a, 0x4052, {CHECK_FPU_ENABLED})
     case 0x406a:		/* lds Rm,FPSCR */
 	CHECK_FPU_ENABLED
-	gen_helper_ld_fpscr(REG(B11_8));
+        gen_helper_ld_fpscr(cpu_env, REG(B11_8));
 	ctx->bstate = BS_STOP;
 	return;
     case 0x4066:		/* lds.l @Rm+,FPSCR */
@@ -1540,7 +1551,7 @@
 	    TCGv addr = tcg_temp_new();
 	    tcg_gen_qemu_ld32s(addr, REG(B11_8), ctx->memidx);
 	    tcg_gen_addi_i32(REG(B11_8), REG(B11_8), 4);
-	    gen_helper_ld_fpscr(addr);
+            gen_helper_ld_fpscr(cpu_env, addr);
 	    tcg_temp_free(addr);
 	    ctx->bstate = BS_STOP;
 	}
@@ -1567,7 +1578,7 @@
         {
             TCGv val = tcg_temp_new();
             tcg_gen_qemu_ld32u(val, REG(B11_8), ctx->memidx);
-            gen_helper_movcal (REG(B11_8), val);            
+            gen_helper_movcal(cpu_env, REG(B11_8), val);
             tcg_gen_qemu_st32(REG(0), REG(B11_8), ctx->memidx);
         }
         ctx->has_movcal = 1;
@@ -1619,7 +1630,7 @@
 	    break;
     case 0x0093:		/* ocbi @Rn */
 	{
-	    gen_helper_ocbi (REG(B11_8));
+            gen_helper_ocbi(cpu_env, REG(B11_8));
 	}
 	return;
     case 0x00a3:		/* ocbp @Rn */
@@ -1733,12 +1744,12 @@
 	    if (ctx->opcode & 0x0100)
 		break; /* illegal instruction */
 	    fp = tcg_temp_new_i64();
-	    gen_helper_float_DT(fp, cpu_fpul);
+            gen_helper_float_DT(fp, cpu_env, cpu_fpul);
 	    gen_store_fpr64(fp, DREG(B11_8));
 	    tcg_temp_free_i64(fp);
 	}
 	else {
-	    gen_helper_float_FT(cpu_fregs[FREG(B11_8)], cpu_fpul);
+            gen_helper_float_FT(cpu_fregs[FREG(B11_8)], cpu_env, cpu_fpul);
 	}
 	return;
     case 0xf03d: /* ftrc FRm/DRm,FPUL - FPSCR: R[PR,Enable.V]/W[Cause,Flag] */
@@ -1749,11 +1760,11 @@
 		break; /* illegal instruction */
 	    fp = tcg_temp_new_i64();
 	    gen_load_fpr64(fp, DREG(B11_8));
-	    gen_helper_ftrc_DT(cpu_fpul, fp);
+            gen_helper_ftrc_DT(cpu_fpul, cpu_env, fp);
 	    tcg_temp_free_i64(fp);
 	}
 	else {
-	    gen_helper_ftrc_FT(cpu_fpul, cpu_fregs[FREG(B11_8)]);
+            gen_helper_ftrc_FT(cpu_fpul, cpu_env, cpu_fregs[FREG(B11_8)]);
 	}
 	return;
     case 0xf04d: /* fneg FRn/DRn - FPSCR: Nothing */
@@ -1783,11 +1794,12 @@
 		break; /* illegal instruction */
 	    TCGv_i64 fp = tcg_temp_new_i64();
 	    gen_load_fpr64(fp, DREG(B11_8));
-	    gen_helper_fsqrt_DT(fp, fp);
+            gen_helper_fsqrt_DT(fp, cpu_env, fp);
 	    gen_store_fpr64(fp, DREG(B11_8));
 	    tcg_temp_free_i64(fp);
 	} else {
-	    gen_helper_fsqrt_FT(cpu_fregs[FREG(B11_8)], cpu_fregs[FREG(B11_8)]);
+            gen_helper_fsqrt_FT(cpu_fregs[FREG(B11_8)], cpu_env,
+                                cpu_fregs[FREG(B11_8)]);
 	}
 	return;
     case 0xf07d: /* fsrra FRn */
@@ -1809,7 +1821,7 @@
 	CHECK_FPU_ENABLED
 	{
 	    TCGv_i64 fp = tcg_temp_new_i64();
-	    gen_helper_fcnvsd_FT_DT(fp, cpu_fpul);
+            gen_helper_fcnvsd_FT_DT(fp, cpu_env, cpu_fpul);
 	    gen_store_fpr64(fp, DREG(B11_8));
 	    tcg_temp_free_i64(fp);
 	}
@@ -1819,7 +1831,7 @@
 	{
 	    TCGv_i64 fp = tcg_temp_new_i64();
 	    gen_load_fpr64(fp, DREG(B11_8));
-	    gen_helper_fcnvds_DT_FT(cpu_fpul, fp);
+            gen_helper_fcnvds_DT_FT(cpu_fpul, cpu_env, fp);
 	    tcg_temp_free_i64(fp);
 	}
 	return;
@@ -1829,7 +1841,7 @@
             TCGv m, n;
             m = tcg_const_i32((ctx->opcode >> 8) & 3);
             n = tcg_const_i32((ctx->opcode >> 10) & 3);
-            gen_helper_fipr(m, n);
+            gen_helper_fipr(cpu_env, m, n);
             tcg_temp_free(m);
             tcg_temp_free(n);
             return;
@@ -1841,7 +1853,7 @@
             (ctx->fpscr & FPSCR_PR) == 0) {
             TCGv n;
             n = tcg_const_i32((ctx->opcode >> 10) & 3);
-            gen_helper_ftrv(n);
+            gen_helper_ftrv(cpu_env, n);
             tcg_temp_free(n);
             return;
         }
@@ -1853,9 +1865,9 @@
     fflush(stderr);
 #endif
     if (ctx->flags & (DELAY_SLOT | DELAY_SLOT_CONDITIONAL)) {
-       gen_helper_raise_slot_illegal_instruction();
+        gen_helper_raise_slot_illegal_instruction(cpu_env);
     } else {
-       gen_helper_raise_illegal_instruction();
+        gen_helper_raise_illegal_instruction(cpu_env);
     }
     ctx->bstate = BS_EXCP;
 }
@@ -1934,7 +1946,7 @@
                 if (ctx.pc == bp->pc) {
 		    /* We have hit a breakpoint - make sure PC is up-to-date */
 		    tcg_gen_movi_i32(cpu_pc, ctx.pc);
-		    gen_helper_debug();
+                    gen_helper_debug(cpu_env);
 		    ctx.bstate = BS_EXCP;
 		    break;
 		}
@@ -1958,7 +1970,7 @@
 	fprintf(stderr, "Loading opcode at address 0x%08x\n", ctx.pc);
 	fflush(stderr);
 #endif
-	ctx.opcode = lduw_code(ctx.pc);
+        ctx.opcode = cpu_lduw_code(env, ctx.pc);
 	decode_opc(&ctx);
         num_insns++;
 	ctx.pc += 2;
@@ -1975,7 +1987,7 @@
         gen_io_end();
     if (env->singlestep_enabled) {
         tcg_gen_movi_i32(cpu_pc, ctx.pc);
-        gen_helper_debug();
+        gen_helper_debug(cpu_env);
     } else {
 	switch (ctx.bstate) {
         case BS_STOP: