diff --git a/cpu-exec.c b/cpu-exec.c
index 9d5c35e782..926093afbb 100644
--- a/cpu-exec.c
+++ b/cpu-exec.c
@@ -47,7 +47,7 @@ void cpu_loop_exit(void)
     longjmp(env->jmp_env, 1);
 }
 #endif
-#ifndef TARGET_SPARC
+#if !(defined(TARGET_SPARC) || defined(TARGET_SH4))
 #define reg_T2
 #endif
 
diff --git a/gdbstub.c b/gdbstub.c
index 37241ce589..3dcaa4c9cd 100644
--- a/gdbstub.c
+++ b/gdbstub.c
@@ -504,7 +504,12 @@ static int cpu_gdb_read_registers(CPUState *env, uint8_t *mem_buf)
   int i;
 
 #define SAVE(x) *ptr++=tswapl(x)
-  for (i = 0; i < 16; i++) SAVE(env->gregs[i]);
+  if ((env->sr & (SR_MD | SR_RB)) == (SR_MD | SR_RB)) {
+      for (i = 0; i < 8; i++) SAVE(env->gregs[i + 16]);
+  } else {
+      for (i = 0; i < 8; i++) SAVE(env->gregs[i]);
+  }
+  for (i = 8; i < 16; i++) SAVE(env->gregs[i]);
   SAVE (env->pc);
   SAVE (env->pr);
   SAVE (env->gbr);
@@ -527,7 +532,12 @@ static void cpu_gdb_write_registers(CPUState *env, uint8_t *mem_buf, int size)
   int i;
 
 #define LOAD(x) (x)=*ptr++;
-  for (i = 0; i < 16; i++) LOAD(env->gregs[i]);
+  if ((env->sr & (SR_MD | SR_RB)) == (SR_MD | SR_RB)) {
+      for (i = 0; i < 8; i++) LOAD(env->gregs[i + 16]);
+  } else {
+      for (i = 0; i < 8; i++) LOAD(env->gregs[i]);
+  }
+  for (i = 8; i < 16; i++) LOAD(env->gregs[i]);
   LOAD (env->pc);
   LOAD (env->pr);
   LOAD (env->gbr);
diff --git a/linux-user/main.c b/linux-user/main.c
index 74642cc361..a73919ec3c 100644
--- a/linux-user/main.c
+++ b/linux-user/main.c
@@ -1370,14 +1370,14 @@ void cpu_loop (CPUState *env)
         switch (trapnr) {
         case 0x160:
             ret = do_syscall(env, 
-                             env->gregs[0x13], 
-                             env->gregs[0x14], 
-                             env->gregs[0x15], 
-                             env->gregs[0x16], 
-                             env->gregs[0x17], 
-                             env->gregs[0x10], 
+                             env->gregs[3], 
+                             env->gregs[4], 
+                             env->gregs[5], 
+                             env->gregs[6], 
+                             env->gregs[7], 
+                             env->gregs[0], 
                              0);
-            env->gregs[0x10] = ret;
+            env->gregs[0] = ret;
             env->pc += 2;
             break;
         case EXCP_DEBUG:
diff --git a/target-sh4/cpu.h b/target-sh4/cpu.h
index c71af74373..ef818fdf25 100644
--- a/target-sh4/cpu.h
+++ b/target-sh4/cpu.h
@@ -45,7 +45,9 @@
 #define FPSCR_PR (1 << 19)
 #define FPSCR_DN (1 << 18)
 
-#define DELAY_SLOT             (1 << 0)
+#define DELAY_SLOT             (1 << 0) /* Must be the same as SR_T.  */
+/* This flag is set if the next insn is a delay slot for a conditional jump.
+   The dynamic value of the DELAY_SLOT determines whether the jup is taken. */
 #define DELAY_SLOT_CONDITIONAL (1 << 1)
 /* Those are used in contexts only */
 #define BRANCH                 (1 << 2)
diff --git a/target-sh4/exec.h b/target-sh4/exec.h
index b828206121..3563300307 100644
--- a/target-sh4/exec.h
+++ b/target-sh4/exec.h
@@ -26,7 +26,7 @@
 register struct CPUSH4State *env asm(AREG0);
 register uint32_t T0 asm(AREG1);
 register uint32_t T1 asm(AREG2);
-register uint32_t T2 asm(AREG3);
+//register uint32_t T2 asm(AREG3);
 
 #define FT0 (env->ft0)
 #define FT1 (env->ft1)
diff --git a/target-sh4/op.c b/target-sh4/op.c
index 64f952fc54..d3b68bc665 100644
--- a/target-sh4/op.c
+++ b/target-sh4/op.c
@@ -109,17 +109,15 @@ void OPPROTO op_not_T0(void)
 
 void OPPROTO op_bf_s(void)
 {
-    T2 = ~env->sr;
     env->delayed_pc = PARAM1;
-    set_flag(DELAY_SLOT_CONDITIONAL);
+    set_flag(DELAY_SLOT_CONDITIONAL | ((~env->sr) & SR_T));
     RETURN();
 }
 
 void OPPROTO op_bt_s(void)
 {
-    T2 = env->sr;
     env->delayed_pc = PARAM1;
-    set_flag(DELAY_SLOT_CONDITIONAL);
+    set_flag(DELAY_SLOT_CONDITIONAL | (env->sr & SR_T));
     RETURN();
 }
 
@@ -888,9 +886,12 @@ void OPPROTO op_jT(void)
     RETURN();
 }
 
-void OPPROTO op_jTT2(void)
+void OPPROTO op_jdelayed(void)
 {
-    if (T2 & SR_T)
+    uint32_t flags;
+    flags = env->flags;
+    env->flags &= ~(DELAY_SLOT | DELAY_SLOT_CONDITIONAL);
+    if (flags & DELAY_SLOT)
 	GOTO_LABEL_PARAM(1);
     RETURN();
 }
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index 0dca45ee9c..358f975e0c 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -124,7 +124,11 @@ void cpu_dump_state(CPUState * env, FILE * f,
 
 void cpu_sh4_reset(CPUSH4State * env)
 {
+#if defined(CONFIG_USER_ONLY)
+    env->sr = 0x00000000;
+#else
     env->sr = 0x700000F0;	/* MD, RB, BL, I3-I0 */
+#endif
     env->vbr = 0;
     env->pc = 0xA0000000;
     env->fpscr = 0x00040001;
@@ -209,10 +213,10 @@ static void gen_delayed_conditional_jump(DisasContext * ctx)
     int l1;
 
     l1 = gen_new_label();
-    gen_op_jTT2(l1);
-    gen_goto_tb(ctx, 0, ctx->pc);
+    gen_op_jdelayed(l1);
+    gen_goto_tb(ctx, 1, ctx->pc);
     gen_set_label(l1);
-    gen_goto_tb(ctx, 1, ctx->delayed_pc);
+    gen_jump(ctx);
 }
 
 #define B3_0 (ctx->opcode & 0xf)
@@ -1160,25 +1164,15 @@ int gen_intermediate_code_internal(CPUState * env, TranslationBlock * tb,
 #endif
     }
 
-    switch (old_flags & (DELAY_SLOT_CONDITIONAL | DELAY_SLOT)) {
-    case DELAY_SLOT_CONDITIONAL:
-	gen_op_clr_delay_slot_conditional();
+    if (old_flags & DELAY_SLOT_CONDITIONAL) {
 	gen_delayed_conditional_jump(&ctx);
-	break;
-    case DELAY_SLOT:
+    } else if (old_flags & DELAY_SLOT) {
 	gen_op_clr_delay_slot();
 	gen_jump(&ctx);
-	break;
-    case 0:
-	if (ctx.flags & BRANCH_EXCEPTION) {
-	    gen_jump_exception(&ctx);
-	} else if ((ctx.flags & (BRANCH | BRANCH_CONDITIONAL)) == 0) {
-	    gen_goto_tb(&ctx, 0, ctx.pc);
-	}
-	break;
-    default:
-	/* Both cannot be set at the same time */
-	assert(0);
+    } else if (ctx.flags & BRANCH_EXCEPTION) {
+        gen_jump_exception(&ctx);
+    } else if ((ctx.flags & (BRANCH | BRANCH_CONDITIONAL)) == 0) {
+        gen_goto_tb(&ctx, 0, ctx.pc);
     }
 
     if (env->singlestep_enabled) {