diff --git a/target-i386/cpu.h b/target-i386/cpu.h
index 3549511106..f8373a1025 100644
--- a/target-i386/cpu.h
+++ b/target-i386/cpu.h
@@ -214,6 +214,12 @@
 #define MSR_IA32_SYSENTER_ESP           0x175
 #define MSR_IA32_SYSENTER_EIP           0x176
 
+#define MSR_MCG_CAP                     0x179
+#define MSR_MCG_STATUS                  0x17a
+#define MSR_MCG_CTL                     0x17b
+
+#define MSR_PAT                         0x277
+
 #define MSR_EFER                        0xc0000080
 
 #define MSR_EFER_SCE   (1 << 0)
@@ -246,6 +252,8 @@
 #define CPUID_PGE  (1 << 13)
 #define CPUID_MCA  (1 << 14)
 #define CPUID_CMOV (1 << 15)
+#define CPUID_PAT  (1 << 16)
+#define CPUID_CLFLUSH (1 << 19)
 /* ... */
 #define CPUID_MMX  (1 << 23)
 #define CPUID_FXSR (1 << 24)
@@ -474,6 +482,8 @@ typedef struct CPUX86State {
     target_ulong kernelgsbase;
 #endif
 
+    uint64_t pat;
+
     /* temporary data for USE_CODE_COPY mode */
 #ifdef USE_CODE_COPY
     uint32_t tmp0;
diff --git a/target-i386/exec.h b/target-i386/exec.h
index e4b6251a29..a3bbea9999 100644
--- a/target-i386/exec.h
+++ b/target-i386/exec.h
@@ -157,11 +157,11 @@ void helper_lldt_T0(void);
 void helper_ltr_T0(void);
 void helper_movl_crN_T0(int reg);
 void helper_movl_drN_T0(int reg);
-void helper_invlpg(unsigned int addr);
+void helper_invlpg(target_ulong addr);
 void cpu_x86_update_cr0(CPUX86State *env, uint32_t new_cr0);
 void cpu_x86_update_cr3(CPUX86State *env, target_ulong new_cr3);
 void cpu_x86_update_cr4(CPUX86State *env, uint32_t new_cr4);
-void cpu_x86_flush_tlb(CPUX86State *env, uint32_t addr);
+void cpu_x86_flush_tlb(CPUX86State *env, target_ulong addr);
 int cpu_x86_handle_mmu_fault(CPUX86State *env, target_ulong addr, 
                              int is_write, int is_user, int is_softmmu);
 void tlb_fill(target_ulong addr, int is_write, int is_user, 
@@ -190,6 +190,7 @@ void helper_idivq_EAX_T0(void);
 void helper_cmpxchg8b(void);
 void helper_cpuid(void);
 void helper_enter_level(int level, int data32);
+void helper_enter64_level(int level, int data64);
 void helper_sysenter(void);
 void helper_sysexit(void);
 void helper_syscall(int next_eip_addend);
diff --git a/target-i386/helper.c b/target-i386/helper.c
index 0d2cd74bc9..2d241a1431 100644
--- a/target-i386/helper.c
+++ b/target-i386/helper.c
@@ -1334,6 +1334,20 @@ void helper_cpuid(void)
         ECX = env->cpuid_model[(index - 0x80000002) * 4 + 2];
         EDX = env->cpuid_model[(index - 0x80000002) * 4 + 3];
         break;
+    case 0x80000005:
+        /* cache info (L1 cache) */
+        EAX = 0x01ff01ff;
+        EBX = 0x01ff01ff;
+        ECX = 0x40020140;
+        EDX = 0x40020140;
+        break;
+    case 0x80000006:
+        /* cache info (L2 cache) */
+        EAX = 0;
+        EBX = 0x42004200;
+        ECX = 0x02008140;
+        EDX = 0;
+        break;
     case 0x80000008:
         /* virtual & phys address size in low 2 bytes. */
         EAX = 0x00003028;
@@ -1383,6 +1397,37 @@ void helper_enter_level(int level, int data32)
     }
 }
 
+#ifdef TARGET_X86_64
+void helper_enter64_level(int level, int data64)
+{
+    target_ulong esp, ebp;
+    ebp = EBP;
+    esp = ESP;
+
+    if (data64) {
+        /* 64 bit */
+        esp -= 8;
+        while (--level) {
+            esp -= 8;
+            ebp -= 8;
+            stq(esp, ldq(ebp));
+        }
+        esp -= 8;
+        stq(esp, T1);
+    } else {
+        /* 16 bit */
+        esp -= 2;
+        while (--level) {
+            esp -= 2;
+            ebp -= 2;
+            stw(esp, lduw(ebp));
+        }
+        esp -= 2;
+        stw(esp, T1);
+    }
+}
+#endif
+
 void helper_lldt_T0(void)
 {
     int selector;
@@ -1963,6 +2008,7 @@ static inline void helper_ret_protected(int shift, int is_iret, int addend)
 #endif
         sp_mask = get_sp_mask(env->segs[R_SS].flags);
     sp = ESP;
+    /* XXX: ssp is zero in 64 bit ? */
     ssp = env->segs[R_SS].base;
     new_eflags = 0; /* avoid warning */
 #ifdef TARGET_X86_64
@@ -2271,7 +2317,7 @@ void helper_movl_drN_T0(int reg)
     env->dr[reg] = T0;
 }
 
-void helper_invlpg(unsigned int addr)
+void helper_invlpg(target_ulong addr)
 {
     cpu_x86_flush_tlb(env, addr);
 }
@@ -2332,6 +2378,9 @@ void helper_wrmsr(void)
     case MSR_STAR:
         env->star = val;
         break;
+    case MSR_PAT:
+        env->pat = val;
+        break;
 #ifdef TARGET_X86_64
     case MSR_LSTAR:
         env->lstar = val;
@@ -2380,6 +2429,9 @@ void helper_rdmsr(void)
     case MSR_STAR:
         val = env->star;
         break;
+    case MSR_PAT:
+        val = env->pat;
+        break;
 #ifdef TARGET_X86_64
     case MSR_LSTAR:
         val = env->lstar;
diff --git a/target-i386/helper2.c b/target-i386/helper2.c
index d571458f76..60335909a2 100644
--- a/target-i386/helper2.c
+++ b/target-i386/helper2.c
@@ -106,7 +106,9 @@ CPUX86State *cpu_x86_init(void)
         env->cpuid_version = (family << 8) | (model << 4) | stepping;
         env->cpuid_features = (CPUID_FP87 | CPUID_DE | CPUID_PSE |
                                CPUID_TSC | CPUID_MSR | CPUID_MCE |
-                               CPUID_CX8 | CPUID_PGE | CPUID_CMOV);
+                               CPUID_CX8 | CPUID_PGE | CPUID_CMOV |
+                               CPUID_PAT);
+        env->pat = 0x0007040600070406ULL;
         env->cpuid_ext_features = 0;
         env->cpuid_features |= CPUID_FXSR | CPUID_MMX | CPUID_SSE | CPUID_SSE2 | CPUID_PAE | CPUID_SEP;
         env->cpuid_xlevel = 0;
@@ -128,6 +130,9 @@ CPUX86State *cpu_x86_init(void)
         env->cpuid_ext2_features = (env->cpuid_features & 0x0183F3FF);
         env->cpuid_ext2_features |= CPUID_EXT2_LM | CPUID_EXT2_SYSCALL;
         env->cpuid_xlevel = 0x80000008;
+
+        /* these features are needed for Win64 and aren't fully implemented */
+        env->cpuid_features |= CPUID_MTRR | CPUID_CLFLUSH | CPUID_MCA;
 #endif
     }
     cpu_single_env = env;
@@ -546,7 +551,7 @@ void cpu_x86_update_cr4(CPUX86State *env, uint32_t new_cr4)
 }
 
 /* XXX: also flush 4MB pages */
-void cpu_x86_flush_tlb(CPUX86State *env, uint32_t addr)
+void cpu_x86_flush_tlb(CPUX86State *env, target_ulong addr)
 {
     tlb_flush_page(env, addr);
 }
diff --git a/target-i386/op.c b/target-i386/op.c
index 2af4e8eb37..3574576ab8 100644
--- a/target-i386/op.c
+++ b/target-i386/op.c
@@ -898,6 +898,11 @@ void op_addw_ESP_im(void)
 }
 
 #ifdef TARGET_X86_64
+void op_subq_A0_2(void)
+{
+    A0 -= 2;
+}
+
 void op_subq_A0_8(void)
 {
     A0 -= 8;
@@ -929,6 +934,13 @@ void OPPROTO op_enter_level(void)
     helper_enter_level(PARAM1, PARAM2);
 }
 
+#ifdef TARGET_X86_64
+void OPPROTO op_enter64_level(void)
+{
+    helper_enter64_level(PARAM1, PARAM2);
+}
+#endif
+
 void OPPROTO op_sysenter(void)
 {
     helper_sysenter();
diff --git a/target-i386/translate.c b/target-i386/translate.c
index c4584a0120..619522a635 100644
--- a/target-i386/translate.c
+++ b/target-i386/translate.c
@@ -1627,7 +1627,14 @@ static void gen_add_A0_ds_seg(DisasContext *s)
         override = R_DS;
     }
     if (must_add_seg) {
-        gen_op_addl_A0_seg(offsetof(CPUX86State,segs[override].base));
+#ifdef TARGET_X86_64
+        if (CODE64(s)) {
+            gen_op_addq_A0_seg(offsetof(CPUX86State,segs[override].base));
+        } else 
+#endif
+        {
+            gen_op_addl_A0_seg(offsetof(CPUX86State,segs[override].base));
+        }
     }
 }
 
@@ -1948,10 +1955,14 @@ static void gen_push_T0(DisasContext *s)
 {
 #ifdef TARGET_X86_64
     if (CODE64(s)) {
-        /* XXX: check 16 bit behaviour */
         gen_op_movq_A0_reg[R_ESP]();
-        gen_op_subq_A0_8();
-        gen_op_st_T0_A0[OT_QUAD + s->mem_index]();
+        if (s->dflag) {
+            gen_op_subq_A0_8();
+            gen_op_st_T0_A0[OT_QUAD + s->mem_index]();
+        } else {
+            gen_op_subq_A0_2();
+            gen_op_st_T0_A0[OT_WORD + s->mem_index]();
+        }
         gen_op_movq_ESP_A0();
     } else 
 #endif
@@ -1985,10 +1996,14 @@ static void gen_push_T1(DisasContext *s)
 {
 #ifdef TARGET_X86_64
     if (CODE64(s)) {
-        /* XXX: check 16 bit behaviour */
         gen_op_movq_A0_reg[R_ESP]();
-        gen_op_subq_A0_8();
-        gen_op_st_T1_A0[OT_QUAD + s->mem_index]();
+        if (s->dflag) {
+            gen_op_subq_A0_8();
+            gen_op_st_T1_A0[OT_QUAD + s->mem_index]();
+        } else {
+            gen_op_subq_A0_2();
+            gen_op_st_T0_A0[OT_WORD + s->mem_index]();
+        }
         gen_op_movq_ESP_A0();
     } else 
 #endif
@@ -2020,9 +2035,8 @@ static void gen_pop_T0(DisasContext *s)
 {
 #ifdef TARGET_X86_64
     if (CODE64(s)) {
-        /* XXX: check 16 bit behaviour */
         gen_op_movq_A0_reg[R_ESP]();
-        gen_op_ld_T0_A0[OT_QUAD + s->mem_index]();
+        gen_op_ld_T0_A0[(s->dflag ? OT_QUAD : OT_WORD) + s->mem_index]();
     } else 
 #endif
     {
@@ -2041,7 +2055,7 @@ static void gen_pop_T0(DisasContext *s)
 static void gen_pop_update(DisasContext *s)
 {
 #ifdef TARGET_X86_64
-    if (CODE64(s)) {
+    if (CODE64(s) && s->dflag) {
         gen_stack_update(s, 8);
     } else
 #endif
@@ -2105,26 +2119,48 @@ static void gen_enter(DisasContext *s, int esp_addend, int level)
 {
     int ot, opsize;
 
-    ot = s->dflag + OT_WORD;
     level &= 0x1f;
-    opsize = 2 << s->dflag;
+#ifdef TARGET_X86_64
+    if (CODE64(s)) {
+        ot = s->dflag ? OT_QUAD : OT_WORD;
+        opsize = 1 << ot;
+        
+        gen_op_movl_A0_ESP();
+        gen_op_addq_A0_im(-opsize);
+        gen_op_movl_T1_A0();
 
-    gen_op_movl_A0_ESP();
-    gen_op_addl_A0_im(-opsize);
-    if (!s->ss32)
-        gen_op_andl_A0_ffff();
-    gen_op_movl_T1_A0();
-    if (s->addseg)
-        gen_op_addl_A0_seg(offsetof(CPUX86State,segs[R_SS].base));
-    /* push bp */
-    gen_op_mov_TN_reg[OT_LONG][0][R_EBP]();
-    gen_op_st_T0_A0[ot + s->mem_index]();
-    if (level) {
-        gen_op_enter_level(level, s->dflag);
+        /* push bp */
+        gen_op_mov_TN_reg[OT_LONG][0][R_EBP]();
+        gen_op_st_T0_A0[ot + s->mem_index]();
+        if (level) {
+            gen_op_enter64_level(level, (ot == OT_QUAD));
+        }
+        gen_op_mov_reg_T1[ot][R_EBP]();
+        gen_op_addl_T1_im( -esp_addend + (-opsize * level) );
+        gen_op_mov_reg_T1[OT_QUAD][R_ESP]();
+    } else 
+#endif
+    {
+        ot = s->dflag + OT_WORD;
+        opsize = 2 << s->dflag;
+        
+        gen_op_movl_A0_ESP();
+        gen_op_addl_A0_im(-opsize);
+        if (!s->ss32)
+            gen_op_andl_A0_ffff();
+        gen_op_movl_T1_A0();
+        if (s->addseg)
+            gen_op_addl_A0_seg(offsetof(CPUX86State,segs[R_SS].base));
+        /* push bp */
+        gen_op_mov_TN_reg[OT_LONG][0][R_EBP]();
+        gen_op_st_T0_A0[ot + s->mem_index]();
+        if (level) {
+            gen_op_enter_level(level, s->dflag);
+        }
+        gen_op_mov_reg_T1[ot][R_EBP]();
+        gen_op_addl_T1_im( -esp_addend + (-opsize * level) );
+        gen_op_mov_reg_T1[OT_WORD + s->ss32][R_ESP]();
     }
-    gen_op_mov_reg_T1[ot][R_EBP]();
-    gen_op_addl_T1_im( -esp_addend + (-opsize * level) );
-    gen_op_mov_reg_T1[OT_WORD + s->ss32][R_ESP]();
 }
 
 static void gen_exception(DisasContext *s, int trapno, target_ulong cur_eip)
@@ -2901,7 +2937,7 @@ static void gen_sse(DisasContext *s, int b, target_ulong pc_start, int rex_r)
             if (mod != 3) 
                 goto illegal_op;
 #ifdef TARGET_X86_64
-            if (CODE64(s)) {
+            if (s->aflag == 2) {
                 gen_op_movq_A0_reg[R_EDI]();
             } else 
 #endif
@@ -3697,7 +3733,6 @@ static target_ulong disas_insn(DisasContext *s, target_ulong pc_start)
         break;
     case 0xc8: /* enter */
         {
-            /* XXX: long mode support */
             int level;
             val = lduw_code(s->pc);
             s->pc += 2;
@@ -3707,7 +3742,6 @@ static target_ulong disas_insn(DisasContext *s, target_ulong pc_start)
         break;
     case 0xc9: /* leave */
         /* XXX: exception not precise (ESP is updated before potential exception) */
-        /* XXX: may be invalid for 16 bit in long mode */
         if (CODE64(s)) {
             gen_op_mov_TN_reg[OT_QUAD][0][R_EBP]();
             gen_op_mov_reg_T0[OT_QUAD][R_ESP]();
@@ -3926,7 +3960,7 @@ static target_ulong disas_insn(DisasContext *s, target_ulong pc_start)
             else
                 ot = dflag + OT_WORD;
 #ifdef TARGET_X86_64
-            if (CODE64(s)) {
+            if (s->aflag == 2) {
                 offset_addr = ldq_code(s->pc);
                 s->pc += 8;
                 if (offset_addr == (int32_t)offset_addr)
@@ -3955,7 +3989,7 @@ static target_ulong disas_insn(DisasContext *s, target_ulong pc_start)
         break;
     case 0xd7: /* xlat */
 #ifdef TARGET_X86_64
-        if (CODE64(s)) {
+        if (s->aflag == 2) {
             gen_op_movq_A0_reg[R_EBX]();
             gen_op_addq_A0_AL();
         } else 
@@ -4779,6 +4813,8 @@ static target_ulong disas_insn(DisasContext *s, target_ulong pc_start)
         val = ldsw_code(s->pc);
         s->pc += 2;
         gen_pop_T0(s);
+        if (CODE64(s) && s->dflag)
+            s->dflag = 2;
         gen_stack_update(s, val + (2 << s->dflag));
         if (s->dflag == 0)
             gen_op_andl_T0_ffff();
@@ -5782,14 +5818,30 @@ static target_ulong disas_insn(DisasContext *s, target_ulong pc_start)
             break;
         case 5: /* lfence */
         case 6: /* mfence */
-        case 7: /* sfence */
             if ((modrm & 0xc7) != 0xc0 || !(s->cpuid_features & CPUID_SSE))
                 goto illegal_op;
             break;
+        case 7: /* sfence / clflush */
+            if ((modrm & 0xc7) == 0xc0) {
+                /* sfence */
+                if (!(s->cpuid_features & CPUID_SSE))
+                    goto illegal_op;
+            } else {
+                /* clflush */
+                if (!(s->cpuid_features & CPUID_CLFLUSH))
+                    goto illegal_op;
+                gen_lea_modrm(s, modrm, &reg_addr, &offset_addr);
+            }
+            break;
         default:
             goto illegal_op;
         }
         break;
+    case 0x10d: /* prefetch */
+        modrm = ldub_code(s->pc++);
+        gen_lea_modrm(s, modrm, &reg_addr, &offset_addr);
+        /* ignore for now */
+        break;
     case 0x110 ... 0x117:
     case 0x128 ... 0x12f:
     case 0x150 ... 0x177: